Browse Source

Merge branch 'add_tests'

# Conflicts:
#	Include/RmlUi/Core/Element.h
Michael Ragazzon 5 years ago
parent
commit
138c9c86e1
100 changed files with 21593 additions and 1122 deletions
  1. 3 4
      .appveyor.yml
  2. 3 4
      .travis.yml
  3. 2 0
      CMake/FileList.cmake
  4. 64 24
      CMakeLists.txt
  5. 2 0
      Include/RmlUi/Core/ComputedValues.h
  6. 9 0
      Include/RmlUi/Core/Element.h
  7. 2 2
      Include/RmlUi/Core/ElementUtilities.h
  8. 1 0
      Include/RmlUi/Core/ID.h
  9. 9 0
      Include/RmlUi/Core/Platform.h
  10. 2 2
      Include/RmlUi/Core/StringUtilities.h
  11. 12 0
      Include/RmlUi/Core/TypeConverter.inl
  12. 4 0
      Include/RmlUi/Core/XMLParser.h
  13. 20 57
      Samples/basic/treeview/src/FileSystem.cpp
  14. 2 3
      Samples/basic/treeview/src/main.cpp
  15. 5 0
      Samples/shell/include/Shell.h
  16. 9 0
      Samples/shell/include/ShellRenderInterfaceOpenGL.h
  17. 87 0
      Samples/shell/src/Shell.cpp
  18. 26 0
      Samples/shell/src/ShellRenderInterfaceOpenGL.cpp
  19. 1 1
      Samples/shell/src/win32/ShellWin32.cpp
  20. 1 1
      Samples/shell/src/x11/ShellX11.cpp
  21. 9 122
      Source/Core/DataExpression.cpp
  22. 0 92
      Source/Core/DataModel.cpp
  23. 8 1
      Source/Core/Element.cpp
  24. 31 19
      Source/Core/ElementDocument.cpp
  25. 3 2
      Source/Core/ElementScroll.cpp
  26. 4 0
      Source/Core/ElementStyle.cpp
  27. 49 9
      Source/Core/ElementTextDefault.cpp
  28. 6 6
      Source/Core/ElementUtilities.cpp
  29. 1 1
      Source/Core/Elements/DataFormatter.cpp
  30. 1 1
      Source/Core/Elements/DataSource.cpp
  31. 1 1
      Source/Core/Elements/ElementDataGridRow.cpp
  32. 1 0
      Source/Core/Factory.cpp
  33. 29 49
      Source/Core/GeometryDatabase.cpp
  34. 5 0
      Source/Core/GeometryDatabase.h
  35. 144 99
      Source/Core/LayoutBlockBox.cpp
  36. 24 18
      Source/Core/LayoutBlockBox.h
  37. 4 2
      Source/Core/LayoutBlockBoxSpace.cpp
  38. 1 1
      Source/Core/LayoutBlockBoxSpace.h
  39. 490 0
      Source/Core/LayoutDetails.cpp
  40. 101 0
      Source/Core/LayoutDetails.h
  41. 32 482
      Source/Core/LayoutEngine.cpp
  42. 16 64
      Source/Core/LayoutEngine.h
  43. 2 3
      Source/Core/LayoutInlineBox.cpp
  44. 1 1
      Source/Core/LayoutInlineBox.h
  45. 2 2
      Source/Core/LayoutInlineBoxText.cpp
  46. 1 1
      Source/Core/LayoutInlineBoxText.h
  47. 33 19
      Source/Core/LayoutLineBox.cpp
  48. 7 5
      Source/Core/LayoutLineBox.h
  49. 1 1
      Source/Core/PropertyParserColour.cpp
  50. 3 3
      Source/Core/Spritesheet.cpp
  51. 10 6
      Source/Core/StyleSheetParser.cpp
  52. 3 5
      Source/Core/StyleSheetParser.h
  53. 1 0
      Source/Core/StyleSheetSpecification.cpp
  54. 5 5
      Source/Core/WidgetScroll.cpp
  55. 9 0
      Source/Core/XMLParser.cpp
  56. 3 1
      Source/Debugger/CommonSource.h
  57. 1 0
      Source/Debugger/ElementContextHook.cpp
  58. 9 2
      Source/Debugger/ElementInfo.cpp
  59. 1 1
      Source/Debugger/LogSource.h
  60. 104 0
      Tests/CMakeLists.txt
  61. 28 0
      Tests/Data/VisualTests/LICENSE.txt
  62. 174 0
      Tests/Data/VisualTests/acid1.rml
  63. 50 0
      Tests/Data/VisualTests/css1_clear.rml
  64. 52 0
      Tests/Data/VisualTests/float_basic.rml
  65. 43 0
      Tests/Data/VisualTests/float_overflow.rml
  66. 46 0
      Tests/Data/VisualTests/inline_block.rml
  67. 71 0
      Tests/Data/VisualTests/overflow_hidden.rml
  68. 42 0
      Tests/Data/VisualTests/overflow_nested.rml
  69. 29 0
      Tests/Data/VisualTests/position_01_normal_flow.rml
  70. 29 0
      Tests/Data/VisualTests/position_02_relative_positioning.rml
  71. 29 0
      Tests/Data/VisualTests/position_03_floating_a_box.rml
  72. 30 0
      Tests/Data/VisualTests/position_04_floating_a_box_sibling.rml
  73. 30 0
      Tests/Data/VisualTests/position_05_floating_a_box_clear.rml
  74. 34 0
      Tests/Data/VisualTests/position_06_absolute_positioning.rml
  75. 37 0
      Tests/Data/VisualTests/position_07_absolute_positioning_relative.rml
  76. 34 0
      Tests/Data/VisualTests/position_08_absolute_positioning_no_relative.rml
  77. 27 0
      Tests/Data/VisualTests/position_09_absolute_positioning_change_bars.rml
  78. 60 0
      Tests/Data/VisualTests/reference/overflow_hidden-ref.rml
  79. 58 0
      Tests/Data/VisualTests/shrink_to_fit_01.rml
  80. 97 0
      Tests/Data/VisualTests/shrink_to_fit_02.rml
  81. 53 0
      Tests/Data/VisualTests/word_break.rml
  82. 96 0
      Tests/Data/description.rml
  83. 96 0
      Tests/Data/style.rcss
  84. 31 0
      Tests/Data/view_source.rml
  85. 78 0
      Tests/Data/visual_tests_help.rml
  86. 6 0
      Tests/Dependencies/doctest/cmake/Config.cmake.in
  87. 13 0
      Tests/Dependencies/doctest/cmake/assemble_single_header.cmake
  88. 207 0
      Tests/Dependencies/doctest/cmake/common.cmake
  89. 175 0
      Tests/Dependencies/doctest/cmake/doctest.cmake
  90. 81 0
      Tests/Dependencies/doctest/cmake/doctestAddTests.cmake
  91. 70 0
      Tests/Dependencies/doctest/cmake/exec_test.cmake
  92. 6205 0
      Tests/Dependencies/doctest/doctest.h
  93. 6410 0
      Tests/Dependencies/lodepng/lodepng.cpp
  94. 1945 0
      Tests/Dependencies/lodepng/lodepng.h
  95. 3225 0
      Tests/Dependencies/nanobench/nanobench.h
  96. 2 0
      Tests/Output/.gitignore
  97. 3 0
      Tests/Output/Readme.txt
  98. 129 0
      Tests/Source/Benchmarks/DataExpression.cpp
  99. 293 0
      Tests/Source/Benchmarks/Element.cpp
  100. 55 0
      Tests/Source/Benchmarks/main.cpp

+ 3 - 4
.appveyor.yml

@@ -4,7 +4,6 @@ matrix:
   fast_finish: true
 environment:
   FREETYPE_VER: 2.10.1
-  VS_CXXFLAGS: /DWIN32 /D_WINDOWS /W4 /GR /EHsc /permissive- /w44062
   RUN_MINGW: false
   matrix:
     - RUN_MINGW: true
@@ -25,10 +24,10 @@ install:
     mkdir Build-Dynamic, Build-Static
     
     cd Build-Dynamic
-    cmake -G "%VS_GENERATOR%" -DBUILD_SHARED_LIBS=ON -DBUILD_SAMPLES=ON -DCMAKE_CXX_FLAGS="%VS_CXXFLAGS%" ..
+    cmake -G "%VS_GENERATOR%" -DBUILD_SHARED_LIBS=ON -DBUILD_SAMPLES=ON -DWARNINGS_AS_ERRORS=ON ..
 
     cd ../Build-Static
-    cmake -G "%VS_GENERATOR%" -DBUILD_SHARED_LIBS=OFF -DBUILD_SAMPLES=OFF ..
+    cmake -G "%VS_GENERATOR%" -DBUILD_SHARED_LIBS=OFF -DBUILD_SAMPLES=OFF -DWARNINGS_AS_ERRORS=ON ..
     
     cd ..
     
@@ -97,7 +96,7 @@ for:
       set GLEW_INCLUDE_DIR=C:/msys64/mingw64/include/GL
       set GLEW_LIBRARIES=C:/msys64/mingw64/lib/libglew32.dll.a
       
-      cmake -G "MinGW Makefiles" -DBUILD_SHARED_LIBS=ON -DENABLE_PRECOMPILED_HEADERS=OFF -DBUILD_SAMPLES=ON -DBUILD_LUA_BINDINGS=ON -DCMAKE_C_COMPILER=gcc.exe -DCMAKE_CXX_COMPILER=g++.exe -DCMAKE_MAKE_PROGRAM=mingw32-make.exe -DGLEW_INCLUDE_DIR=%GLEW_INCLUDE_DIR% -DGLEW_LIBRARIES=%GLEW_LIBRARIES% ..
+      cmake -G "MinGW Makefiles" -DBUILD_SHARED_LIBS=ON -DENABLE_PRECOMPILED_HEADERS=OFF -DBUILD_SAMPLES=ON -DBUILD_LUA_BINDINGS=ON -DWARNINGS_AS_ERRORS=ON -DCMAKE_C_COMPILER=gcc.exe -DCMAKE_CXX_COMPILER=g++.exe -DCMAKE_MAKE_PROGRAM=mingw32-make.exe -DGLEW_INCLUDE_DIR=%GLEW_INCLUDE_DIR% -DGLEW_LIBRARIES=%GLEW_LIBRARIES% ..
 
       set CHERE_INVOKING=yes
       set MSYSTEM=MINGW64

+ 3 - 4
.travis.yml

@@ -1,9 +1,6 @@
 sudo: false
 dist: bionic
 language: c++
-env:
-  global:
-    - CXXFLAGS="-pedantic -Wall -Wextra"
 
 cache: 
   apt: true
@@ -17,6 +14,7 @@ matrix:
       compiler: clang
     - os: linux
       compiler: clang
+      env: BUILD_TESTING="ON"
       addons:
         apt:
           packages:
@@ -90,7 +88,7 @@ install:
   - cmake --version
   - cd "$TRAVIS_BUILD_DIR"
   - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then cmake -DNO_THIRDPARTY_CONTAINERS=ON -DENABLE_PRECOMPILED_HEADERS=OFF -G Xcode .; fi
-  - if [[ "$TRAVIS_OS_NAME" != "osx" ]]; then cmake -DBUILD_LUA_BINDINGS=ON -DBUILD_SAMPLES=ON -DDISABLE_RTTI_AND_EXCEPTIONS=${DISABLE_RTTI_AND_EXCEPTIONS:-OFF} -DNO_THIRDPARTY_CONTAINERS=${NO_THIRDPARTY_CONTAINERS:-OFF} -DNO_FONT_INTERFACE_DEFAULT=${NO_FONT_INTERFACE_DEFAULT:-OFF} -G Ninja .; fi
+  - if [[ "$TRAVIS_OS_NAME" != "osx" ]]; then cmake -DBUILD_LUA_BINDINGS=ON -DBUILD_SAMPLES=ON -DBUILD_TESTING=${BUILD_TESTING:-OFF} -DWARNINGS_AS_ERRORS=ON -DDISABLE_RTTI_AND_EXCEPTIONS=${DISABLE_RTTI_AND_EXCEPTIONS:-OFF} -DNO_THIRDPARTY_CONTAINERS=${NO_THIRDPARTY_CONTAINERS:-OFF} -DNO_FONT_INTERFACE_DEFAULT=${NO_FONT_INTERFACE_DEFAULT:-OFF} -G Ninja .; fi
 
 before_script:
   - if [[ "$VALGRIND_SAMPLES" == "1" ]] && [[ "$TRAVIS_OS_NAME" == "linux" ]]; then export DISPLAY=:99.0; fi
@@ -98,6 +96,7 @@ before_script:
 script:
   - if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then xcodebuild -project rmlui.xcodeproj/ -jobs 4 -configuration Release -scheme ALL_BUILD | xcpretty --color; test ${PIPESTATUS[0]} -eq 0; fi
   - if [[ "$TRAVIS_OS_NAME" != "osx" ]]; then ninja; fi
+  - if [[ "$BUILD_TESTING" == "ON" ]]; then ctest; fi
   - if [[ "$VALGRIND_SAMPLES" == "1" ]]; then mkdir build && DESTDIR=build ninja install; fi
   - |
     if [[ "$VALGRIND_SAMPLES" == "1" ]]; then

+ 2 - 0
CMake/FileList.cmake

@@ -59,6 +59,7 @@ set(Core_HDR_FILES
     ${PROJECT_SOURCE_DIR}/Source/Core/IdNameMap.h
     ${PROJECT_SOURCE_DIR}/Source/Core/LayoutBlockBox.h
     ${PROJECT_SOURCE_DIR}/Source/Core/LayoutBlockBoxSpace.h
+    ${PROJECT_SOURCE_DIR}/Source/Core/LayoutDetails.h
     ${PROJECT_SOURCE_DIR}/Source/Core/LayoutEngine.h
     ${PROJECT_SOURCE_DIR}/Source/Core/LayoutInlineBox.h
     ${PROJECT_SOURCE_DIR}/Source/Core/LayoutInlineBoxText.h
@@ -328,6 +329,7 @@ set(Core_SRC_FILES
     ${PROJECT_SOURCE_DIR}/Source/Core/GeometryUtilities.cpp
     ${PROJECT_SOURCE_DIR}/Source/Core/LayoutBlockBox.cpp
     ${PROJECT_SOURCE_DIR}/Source/Core/LayoutBlockBoxSpace.cpp
+    ${PROJECT_SOURCE_DIR}/Source/Core/LayoutDetails.cpp
     ${PROJECT_SOURCE_DIR}/Source/Core/LayoutEngine.cpp
     ${PROJECT_SOURCE_DIR}/Source/Core/LayoutInlineBox.cpp
     ${PROJECT_SOURCE_DIR}/Source/Core/LayoutInlineBoxText.cpp

+ 64 - 24
CMakeLists.txt

@@ -69,6 +69,18 @@ endif()
 
 set(RMLUI_VERSION_SHORT ${PROJECT_VERSION_MAJOR}.${PROJECT_VERSION_MINOR}${RMLUI_VERSION_SUFFIX})
 
+if(CMAKE_PROJECT_NAME STREQUAL PROJECT_NAME)
+	option(BUILD_TESTING "" OFF)
+    include(CTest)
+	
+	if(BUILD_TESTING)
+		set(RMLUI_TESTS_ENABLED ON)
+		set(VISUAL_TESTS_RML_DIRECTORIES "" CACHE PATH "Specify additional directories containing *.rml test documents for VisualTests. Separate multiple directories by semicolon.")
+		set(VISUAL_TESTS_COMPARE_DIRECTORY "" CACHE PATH "Set the input directory for screenshot comparison performed by VisualTests.")
+		set(VISUAL_TESTS_CAPTURE_DIRECTORY "" CACHE PATH "Set the output directory for screenshots generated by VisualTests.")
+	endif()
+endif()
+
 # paths
 include(GNUInstallDirs)
 
@@ -248,6 +260,26 @@ elseif (ENABLE_PRECOMPILED_HEADERS)
 	set(PRECOMPILED_HEADERS_ENABLED ON)
 endif()
 
+
+option(WARNINGS_AS_ERRORS "Treat compiler warnings as errors." OFF)
+mark_as_advanced(WARNINGS_AS_ERRORS)
+
+macro(add_common_target_options NAME)
+	if(CMAKE_CXX_COMPILER_ID MATCHES "Clang" OR CMAKE_CXX_COMPILER_ID MATCHES "GNU")
+		target_compile_options(${NAME} PRIVATE -Wall -pedantic -Wextra)
+		
+		if(WARNINGS_AS_ERRORS)
+			target_compile_options(${NAME} PRIVATE -Werror)
+		endif()
+	elseif(MSVC)
+		target_compile_options(${NAME} PRIVATE /MP /W4 /w44062 /permissive-)
+		
+		if(WARNINGS_AS_ERRORS)
+			target_compile_options(${NAME} PRIVATE /WX)
+		endif()
+	endif()
+endmacro()
+
 #===================================
 # Find dependencies ================
 #===================================
@@ -306,17 +338,7 @@ foreach(library ${LIBRARIES})
 	   SOVERSION ${PROJECT_VERSION_MAJOR}
 	)
 	
-	if (MSVC)
-		target_compile_options(${NAME} PUBLIC "/MP")
-	endif(MSVC)
-
-	if( MATRIX_ROW_MAJOR )
-		target_compile_definitions(${NAME} -DRMLUI_MATRIX_ROW_MAJOR=1)
-	endif ()
-
-	if( CUSTOM_CONFIGURATION AND CUSTOM_LINK_LIBRARIES )
-		target_link_libraries(${NAME} ${CUSTOM_LINK_LIBRARIES})
-	endif ()
+	add_common_target_options(${NAME})
 
 	install(TARGETS ${NAME}
 		EXPORT RmlUiTargets
@@ -342,6 +364,13 @@ endif()
 
 target_compile_definitions(RmlCore PRIVATE RMLUI_VERSION="${RMLUI_VERSION_SHORT}")
 
+if( MATRIX_ROW_MAJOR )
+	target_compile_definitions(RmlCore PUBLIC -DRMLUI_MATRIX_ROW_MAJOR)
+endif ()
+if( CUSTOM_CONFIGURATION AND CUSTOM_LINK_LIBRARIES )
+	target_link_libraries(RmlCore PUBLIC ${CUSTOM_LINK_LIBRARIES})
+endif ()
+
 if (PRECOMPILED_HEADERS_ENABLED)
 	target_precompile_headers(RmlCore PRIVATE ${PROJECT_SOURCE_DIR}/Source/Core/precompiled.h)
 endif()
@@ -418,6 +447,8 @@ if(BUILD_LUA_BINDINGS)
 	   VERSION ${PROJECT_VERSION}
 	   SOVERSION ${PROJECT_VERSION_MAJOR}
 	)
+	
+	add_common_target_options(${NAME})
 
 	install(TARGETS ${NAME}
 		EXPORT RmlUiTargets
@@ -455,18 +486,18 @@ endif()
 #===================================
 
 if(NOT BUILD_FRAMEWORK)
-target_link_libraries(RmlCore ${CORE_LINK_LIBS})
-target_link_libraries(RmlDebugger RmlCore)
-else(NOT BUILD_FRAMEWORK)
-target_link_libraries(RmlUi ${CORE_LINK_LIBS})
-endif(NOT BUILD_FRAMEWORK)
+	target_link_libraries(RmlCore ${CORE_LINK_LIBS})
+	target_link_libraries(RmlDebugger RmlCore)
+else()
+	target_link_libraries(RmlUi ${CORE_LINK_LIBS})
+endif()
 
 if(BUILD_LUA_BINDINGS)
 	if(NOT BUILD_FRAMEWORK)
 		target_link_libraries(RmlLua RmlCore ${LUA_BINDINGS_LINK_LIBS})
-	else(NOT BUILD_FRAMEWORK)
+	else()
 		target_link_libraries(RmlLua RmlUi ${LUA_BINDINGS_LINK_LIBS})
-	endif(NOT BUILD_FRAMEWORK)
+	endif()
 endif()
 
 
@@ -491,6 +522,8 @@ macro(bl_sample NAME)
 	else()
 		add_executable(${NAME} ${${NAME}_SRC_FILES} ${${NAME}_HDR_FILES} )
 	endif()
+	
+	add_common_target_options(${NAME})
 
 	target_link_libraries(${NAME} ${ARGN})
 endmacro()
@@ -518,8 +551,8 @@ endif(NOT BUILD_FRAMEWORK)
 	find_package(OpenGL REQUIRED)
 		   
 	if(OPENGL_FOUND)
-	include_directories(${OPENGL_INCLUDE_DIR})
-	list(APPEND sample_LIBRARIES ${OPENGL_LIBRARIES})
+		include_directories(${OPENGL_INCLUDE_DIR})
+		list(APPEND sample_LIBRARIES ${OPENGL_LIBRARIES})
 	endif()
 	
 	# Set up required system libraries
@@ -563,10 +596,10 @@ endif(NOT BUILD_FRAMEWORK)
 	if (WIN32)
 		target_link_libraries(shell PUBLIC shlwapi)
 	endif()
-
-	if( CUSTOM_CONFIGURATION AND CUSTOM_LINK_LIBRARIES )
-		target_link_libraries(shell PUBLIC ${CUSTOM_LINK_LIBRARIES})
-	endif ()
+	
+	target_link_libraries(shell PUBLIC RmlCore)
+	
+	add_common_target_options(shell)
 
 	# Build and install the basic samples
 	foreach(sample ${samples})
@@ -663,6 +696,13 @@ endif(NOT BUILD_FRAMEWORK)
 	endif()
 endif()
 
+#===================================
+# Add tests ========================
+#===================================
+
+if(RMLUI_TESTS_ENABLED)
+    add_subdirectory(Tests)
+endif()
 
 #===================================
 # Installation =====================

+ 2 - 0
Include/RmlUi/Core/ComputedValues.h

@@ -113,6 +113,7 @@ enum class TextAlign : uint8_t { Left, Right, Center, Justify };
 enum class TextDecoration : uint8_t { None, Underline, Overline, LineThrough };
 enum class TextTransform : uint8_t { None, Capitalize, Uppercase, Lowercase };
 enum class WhiteSpace : uint8_t { Normal, Pre, Nowrap, Prewrap, Preline };
+enum class WordBreak : uint8_t { Normal, BreakAll, BreakWord };
 
 enum class Drag : uint8_t { None, Drag, DragDrop, Block, Clone };
 enum class TabIndex : uint8_t { None, Auto };
@@ -184,6 +185,7 @@ struct ComputedValues
 	TextDecoration text_decoration = TextDecoration::None;
 	TextTransform text_transform = TextTransform::None;
 	WhiteSpace white_space = WhiteSpace::Normal;
+	WordBreak word_break = WordBreak::Normal;
 
 	String cursor;
 

+ 9 - 0
Include/RmlUi/Core/Element.h

@@ -57,6 +57,7 @@ class ElementScroll;
 class ElementStyle;
 class LayoutEngine;
 class LayoutInlineBox;
+class LayoutBlockBox;
 class PropertiesIteratorView;
 class PropertyDictionary;
 class RenderInterface;
@@ -467,8 +468,12 @@ public:
 	/// @param[in] event Event to attach to.
 	/// @param[in] listener The listener object to be attached.
 	/// @param[in] in_capture_phase True to attach in the capture phase, false in bubble phase.
+	/// @lifetime The added listener must stay alive until after the dispatched call from EventListener::OnDetach(). This occurs
+	///     eg. when the element is destroyed or when RemoveEventListener() is called with the same parameters passed here.
 	void AddEventListener(const String& event, EventListener* listener, bool in_capture_phase = false);
 	/// Adds an event listener to this element by id.
+	/// @lifetime The added listener must stay alive until after the dispatched call from EventListener::OnDetach(). This occurs
+	///     eg. when the element is destroyed or when RemoveEventListener() is called with the same parameters passed here.
 	void AddEventListener(EventId id, EventListener* listener, bool in_capture_phase = false);
 	/// Removes an event listener from this element.
 	/// @param[in] event Event to detach from.
@@ -636,6 +641,7 @@ private:
 
 	void DirtyOffset();
 	void UpdateOffset();
+	void SetBaseline(float baseline);
 
 	void BuildLocalStackingContext();
 	void BuildStackingContext(ElementList* stacking_context);
@@ -712,6 +718,8 @@ private:
 	// Defines what box area represents the element's client area; this is usually padding, but may be content.
 	Box::Area client_area;
 
+	float baseline;
+
 	// True if the element is visible and active.
 	bool visible;
 
@@ -748,6 +756,7 @@ private:
 	friend class Rml::Context;
 	friend class Rml::ElementStyle;
 	friend class Rml::LayoutEngine;
+	friend class Rml::LayoutBlockBox;
 	friend class Rml::LayoutInlineBox;
 	friend class Rml::ElementScroll;
 };

+ 2 - 2
Include/RmlUi/Core/ElementUtilities.h

@@ -117,14 +117,14 @@ public:
 	/// @param[in] containing_block The dimensions of the content area of the block containing the element.
 	/// @param[in] element The element to build the box for.
 	/// @param[in] inline_element True if the element is placed in an inline context, false if not.
-	static void BuildBox(Box& box, const Vector2f& containing_block, Element* element, bool inline_element = false);
+	static void BuildBox(Box& box, Vector2f containing_block, Element* element, bool inline_element = false);
 
 	/// Sizes an element, and positions it within its parent offset from the borders of its content area. Any relative
 	/// values will be evaluated against the size of the element parent's content area.
 	/// @param element[in] The element to size and position.
 	/// @param offset[in] The offset from the parent's borders.
 	/// @param anchor[in] Defines which corner or edge the border is to be positioned relative to.
-	static bool PositionElement(Element* element, const Vector2f& offset, PositionAnchor anchor);
+	static bool PositionElement(Element* element, Vector2f offset, PositionAnchor anchor);
 
 	/// Applies an element's accumulated transform matrix, determined from its and ancestor's `perspective' and `transform' properties.
 	/// Note: All calls to RenderInterface::SetTransform must go through here.

+ 1 - 0
Include/RmlUi/Core/ID.h

@@ -123,6 +123,7 @@ enum class PropertyId : uint8_t
 	TextDecoration,
 	TextTransform,
 	WhiteSpace,
+	WordBreak,
 	Cursor,
 	Drag,
 	TabIndex,

+ 9 - 0
Include/RmlUi/Core/Platform.h

@@ -126,4 +126,13 @@
     RMLUI_ERRORMSG("Switch case for unhandled ENUM has been hit!  This shouldn't happen!  ENUM Name: " # x); \
     break;
 
+// Tell the compiler of printf-like functions, warns on incorrect usage.
+#if defined __MINGW32__
+#  define RMLUI_ATTRIBUTE_FORMAT_PRINTF(i, f) __attribute__((format (__MINGW_PRINTF_FORMAT, i, f)))
+#elif defined __GNUC__ || defined __clang__
+#  define RMLUI_ATTRIBUTE_FORMAT_PRINTF(i, f) __attribute__((format (printf, i, f)))
+#else
+#  define RMLUI_ATTRIBUTE_FORMAT_PRINTF(i, f)
+#endif
+
 #endif

+ 2 - 2
Include/RmlUi/Core/StringUtilities.h

@@ -42,10 +42,10 @@ namespace Rml {
 class StringView;
 
 /// Construct a string using sprintf-style syntax.
-RMLUICORE_API String CreateString(size_t max_size, const char* format, ...);
+RMLUICORE_API String CreateString(size_t max_size, const char* format, ...) RMLUI_ATTRIBUTE_FORMAT_PRINTF(2,3);
 
 /// Format to a string using sprintf-style syntax.
-RMLUICORE_API int FormatString(String& string, size_t max_size, const char* format, ...);
+RMLUICORE_API int FormatString(String& string, size_t max_size, const char* format, ...) RMLUI_ATTRIBUTE_FORMAT_PRINTF(3,4);
 
 
 namespace StringUtilities

+ 12 - 0
Include/RmlUi/Core/TypeConverter.inl

@@ -35,6 +35,13 @@ bool TypeConverter<SourceType, DestType>::Convert(const SourceType& /*src*/, Des
 	return false;
 }
 
+#if defined(RMLUI_PLATFORM_WIN32) && defined(__MINGW32__)
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wformat"
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wformat-extra-args"
+#endif
+
 ///
 /// Full Specialisations
 ///
@@ -377,4 +384,9 @@ VECTOR_STRING_CONVERTER(Colourb, byte, 4);
 #undef STRING_VECTOR_CONVERTER
 #undef VECTOR_STRING_CONVERTER
 
+#if defined(RMLUI_PLATFORM_WIN32) && defined(__MINGW32__)
+#pragma GCC diagnostic pop
+#pragma GCC diagnostic pop
+#endif
+
 } // namespace Rml

+ 4 - 0
Include/RmlUi/Core/XMLParser.h

@@ -57,6 +57,10 @@ public:
 	/// @param[in] handler The custom handler.
 	/// @return The registered XML node handler.
 	static XMLNodeHandler* RegisterNodeHandler(const String& tag, SharedPtr<XMLNodeHandler> handler);
+	/// Retrieve a registered node handler.
+	/// @param[in] tag The tag the custom parser handles.
+	/// @return The registered XML node handler or nullptr if it does not exist for the given tag.
+	static XMLNodeHandler* GetNodeHandler(const String& tag);
 	/// Releases all registered node handlers. This is called internally.
 	static void ReleaseHandlers();
 

+ 20 - 57
Samples/basic/treeview/src/FileSystem.cpp

@@ -31,19 +31,14 @@
 #include <cstdlib>
 #include <cstdio>
 #include <string.h>
-
-#ifdef RMLUI_PLATFORM_WIN32
-#include <io.h>
-#else
-#include <dirent.h>
-#endif
+#include <Shell.h>
 
 struct FileSystemNode;
 
-typedef Rml::UnorderedMap< Rml::String, FileSystemNode* > NodeMap;
+using NodeMap = Rml::UnorderedMap< Rml::String, FileSystemNode* >;
 
-FileSystemNode* file_system_root = nullptr;
-NodeMap node_map;
+static Rml::UniquePtr<FileSystemNode> file_system_root;
+static NodeMap node_map;
 
 
 /**
@@ -55,7 +50,7 @@ struct FileSystemNode
 {
 	FileSystemNode(const Rml::String _name, bool _directory, int _depth = -1) : name(_name)
 	{
-		id = Rml::CreateString(16, "%x", this);
+		id = Rml::CreateString(32, "%p", (void*)this);
 
 		directory = _directory;
 		depth = _depth;
@@ -64,60 +59,29 @@ struct FileSystemNode
 	}
 
 	~FileSystemNode()
-	{
-		for (size_t i = 0; i < child_nodes.size(); ++i)
-			delete child_nodes[i];
-	}
+	{}
 
 	// Build the list of files and directories within this directory.
 	void BuildTree(const Rml::String& root = "")
 	{
-#ifdef RMLUI_PLATFORM_WIN32
-		_finddata_t find_data;
-		intptr_t find_handle = _findfirst((root + name + "/*.*").c_str(), &find_data);
-		if (find_handle != -1)
-		{
-			do
-			{
-				if (strcmp(find_data.name, ".") == 0 ||
-					strcmp(find_data.name, "..") == 0)
-					continue;
+		const Rml::String current_directory = root + name + '/';
 
-				child_nodes.push_back(new FileSystemNode(find_data.name, (find_data.attrib & _A_SUBDIR) == _A_SUBDIR, depth + 1));
+		const Rml::StringList directories = Shell::ListDirectories(current_directory);
 
-			} while (_findnext(find_handle, &find_data) == 0);
-
-			_findclose(find_handle);
+		for (const Rml::String& directory : directories)
+		{
+			child_nodes.push_back(Rml::MakeUnique<FileSystemNode>(directory, true, depth + 1));
+			child_nodes.back()->BuildTree(current_directory);
 		}
-#else
-			struct dirent** file_list = nullptr;
-			int file_count = -1;
-			file_count = scandir((root + name).c_str(), &file_list, 0, alphasort);
-			if (file_count == -1)
-				return;
-
-			while (file_count--)
-			{
-				if (strcmp(file_list[file_count]->d_name, ".") == 0 ||
-					strcmp(file_list[file_count]->d_name, "..") == 0)
-					continue;
-
-				child_nodes.push_back(new FileSystemNode(file_list[file_count]->d_name, (file_list[file_count]->d_type & DT_DIR) == DT_DIR, depth + 1));
-
-				free(file_list[file_count]);
-			}
-			free(file_list);
-#endif
-
-		// Generate the trees of all of our subdirectories.
-		for (size_t i = 0; i < child_nodes.size(); ++i)
+
+		const Rml::StringList files = Shell::ListFiles(current_directory);
+		for (const Rml::String& file : files)
 		{
-			if (child_nodes[i]->directory)
-				child_nodes[i]->BuildTree(root + name + "/");
+			child_nodes.push_back(Rml::MakeUnique<FileSystemNode>(file, false, depth + 1));
 		}
 	}
 
-	typedef Rml::Vector< FileSystemNode* > NodeList;
+	using NodeList = Rml::Vector< Rml::UniquePtr<FileSystemNode> >;
 
 	Rml::String id;
 	Rml::String name;
@@ -131,14 +95,13 @@ struct FileSystemNode
 FileSystem::FileSystem(const Rml::String& root) : Rml::DataSource("file")
 {
 	// Generate the file system nodes starting at the RmlUi's root directory.
-	file_system_root = new FileSystemNode(".", true);
+	file_system_root = Rml::MakeUnique<FileSystemNode>(".", true);
 	file_system_root->BuildTree(root);
 }
 
 FileSystem::~FileSystem()
 {
-	delete file_system_root;
-	file_system_root = nullptr;
+	file_system_root.reset();
 }
 
 void FileSystem::GetRow(Rml::StringList& row, const Rml::String& table, int row_index, const Rml::StringList& columns)
@@ -183,7 +146,7 @@ FileSystemNode* FileSystem::GetNode(const Rml::String& table)
 {
 	// Determine which node the row is being requested from.
 	if (table == "root")
-		return file_system_root;
+		return file_system_root.get();
 	else
 	{
 		NodeMap::iterator i = node_map.find(table);

+ 2 - 3
Samples/basic/treeview/src/main.cpp

@@ -107,8 +107,8 @@ int main(int RMLUI_UNUSED_PARAMETER(argc), char** RMLUI_UNUSED_PARAMETER(argv))
 	Shell::LoadFonts("assets/");
 
 	// Create the file data source and formatter.
-	Rml::String root = Shell::FindSamplesRoot();
-	FileSystem file_system(root + "basic/");
+	const Rml::String root = Shell::FindSamplesRoot();
+	FileSystem file_system(root);
 	FileFormatter file_formatter;
 
 	// Load and show the demo document.
@@ -121,7 +121,6 @@ int main(int RMLUI_UNUSED_PARAMETER(argc), char** RMLUI_UNUSED_PARAMETER(argv))
 
 	Shell::EventLoop(GameLoop);
 
-	// Shutdown RmlUi.
 	Rml::Shutdown();
 
 	Shell::CloseWindow();

+ 5 - 0
Samples/shell/include/Shell.h

@@ -53,6 +53,11 @@ public:
 	/// Loads the default fonts from the given path.
 	static void LoadFonts(const char* directory);
 
+	/// List files in the given directory. An initial forward slash '/' makes it relative to the samples root.
+	static Rml::StringList ListFiles(const Rml::String& in_directory, const Rml::String& extension = Rml::String());
+	/// List subdirectories in the given directory. An initial forward slash '/' makes it relative to the samples root.
+	static Rml::StringList ListDirectories(const Rml::String& in_directory);
+
 	/// Open a platform specific window, optionally initialising an OpenGL context on it.
 	/// @param[in] title Title of the window.
 	/// @param[in] srie Provides the interface for attaching a renderer to the window and performing related bits of interface.

+ 9 - 0
Samples/shell/include/ShellRenderInterfaceOpenGL.h

@@ -68,6 +68,15 @@ public:
 	/// Called by RmlUi when it wants to set the current transform matrix to a new matrix.
 	void SetTransform(const Rml::Matrix4f* transform) override;
 
+	// Extensions used by the test suite
+	struct Image {
+		int width = 0;
+		int height = 0;
+		int num_components = 0;
+		Rml::UniquePtr<Rml::byte[]> data;
+	};
+	Image CaptureScreen();
+
 	// ShellRenderInterfaceExtensions
 	void SetViewport(int width, int height) override;
 	void SetContext(void *context) override;

+ 87 - 0
Samples/shell/src/Shell.cpp

@@ -28,6 +28,13 @@
 
 #include "Shell.h"
 #include <RmlUi/Core/Core.h>
+#include <string.h>
+
+#ifdef RMLUI_PLATFORM_WIN32
+#include <io.h>
+#else
+#include <dirent.h>
+#endif
 
 /// Loads the default fonts from the given path.
 void Shell::LoadFonts(const char* directory)
@@ -47,3 +54,83 @@ void Shell::LoadFonts(const char* directory)
 	}
 }
 
+
+enum class ListType { Files, Directories };
+
+static Rml::StringList ListFilesOrDirectories(ListType type, const Rml::String& directory, const Rml::String& extension)
+{
+	if (directory.empty())
+		return Rml::StringList();
+
+	Rml::StringList result;
+
+#ifdef RMLUI_PLATFORM_WIN32
+	const Rml::String find_path = directory + "/*." + (extension.empty() ? Rml::String("*") : extension);
+
+	_finddata_t find_data;
+	intptr_t find_handle = _findfirst(find_path.c_str(), &find_data);
+	if (find_handle != -1)
+	{
+		do
+		{
+			if (strcmp(find_data.name, ".") == 0 ||
+				strcmp(find_data.name, "..") == 0)
+				continue;
+
+			bool is_directory = ((find_data.attrib & _A_SUBDIR) == _A_SUBDIR);
+			bool is_file = (!is_directory && ((find_data.attrib & _A_NORMAL) == _A_NORMAL));
+
+			if (((type == ListType::Files) && is_file) ||
+				((type == ListType::Directories) && is_directory))
+			{
+				result.push_back(find_data.name);
+			}
+
+		} while (_findnext(find_handle, &find_data) == 0);
+
+		_findclose(find_handle);
+	}
+#else
+	struct dirent** file_list = nullptr;
+	const int file_count = scandir(directory.c_str(), &file_list, 0, alphasort);
+	if (file_count == -1)
+		return Rml::StringList();
+
+	for (int i = 0; i < file_count; i++)
+	{
+		if (strcmp(file_list[i]->d_name, ".") == 0 ||
+			strcmp(file_list[i]->d_name, "..") == 0)
+			continue;
+
+		bool is_directory = ((file_list[i]->d_type & DT_DIR) == DT_DIR);
+		bool is_file = ((file_list[i]->d_type & DT_REG) == DT_REG);
+
+		if (!extension.empty())
+		{
+			const char* last_dot = strrchr(file_list[i]->d_name, '.');
+			if (!last_dot || strcmp(last_dot + 1, extension.c_str()) != 0)
+				continue;
+		}
+
+		if ((type == ListType::Files && is_file) ||
+			(type == ListType::Directories && is_directory))
+		{
+			result.push_back(file_list[i]->d_name);
+		}
+	}
+	free(file_list);
+#endif
+
+	return result;
+}
+
+Rml::StringList Shell::ListDirectories(const Rml::String& in_directory)
+{
+	return ListFilesOrDirectories(ListType::Directories, in_directory, Rml::String());
+}
+
+Rml::StringList Shell::ListFiles(const Rml::String& in_directory, const Rml::String& extension)
+{
+	return ListFilesOrDirectories(ListType::Files, in_directory, extension);
+}
+

+ 26 - 0
Samples/shell/src/ShellRenderInterfaceOpenGL.cpp

@@ -299,3 +299,29 @@ void ShellRenderInterfaceOpenGL::SetTransform(const Rml::Matrix4f* transform)
 		glLoadIdentity();
 }
 
+
+ShellRenderInterfaceOpenGL::Image ShellRenderInterfaceOpenGL::CaptureScreen()
+{
+	Image image;
+	image.num_components = 3;
+	image.width = m_width;
+	image.height = m_height;
+
+	const int byte_size = image.width * image.height * image.num_components;
+	image.data = Rml::UniquePtr<Rml::byte[]>(new Rml::byte[byte_size]);
+
+	glReadPixels(0, 0, image.width, image.height, GL_RGB, GL_UNSIGNED_BYTE, image.data.get());
+
+	bool result = true;
+	GLenum err;
+	while ((err = glGetError()) != GL_NO_ERROR)
+	{
+		result = false;
+		Rml::Log::Message(Rml::Log::LT_ERROR, "Could not capture screenshot, got GL error: 0x%x", err);
+	}
+
+	if (!result)
+		return Image();
+
+	return image;
+}

+ 1 - 1
Samples/shell/src/win32/ShellWin32.cpp

@@ -95,7 +95,7 @@ void Shell::Shutdown()
 
 Rml::String Shell::FindSamplesRoot()
 {
-	const char* candidate_paths[] = { "", "..\\..\\Samples\\", "..\\Samples\\", "..\\..\\..\\Samples\\" };
+	const char* candidate_paths[] = { "", "..\\Samples\\", "..\\..\\Samples\\", "..\\..\\..\\Samples\\", "..\\..\\..\\..\\Samples\\" };
 	
 	// Fetch the path of the executable, test the candidate paths appended to that.
 	char executable_file_name[MAX_PATH];

+ 1 - 1
Samples/shell/src/x11/ShellX11.cpp

@@ -107,7 +107,7 @@ Rml::String Shell::FindSamplesRoot()
 	// For "../Samples/" to be valid we must be in the Build directory.
 	// If "../" is valid we are probably in the installation directory.
 	// Some build setups may nest the executables deeper in a build directory, try them last.
-	const char* candidate_paths[] = { "../Samples/", "../", "", "../../Samples/", "../../../Samples/"};
+	const char* candidate_paths[] = { "", "../", "../Samples/", "../../Samples/", "../../../Samples/", "../../../../Samples/" };
 
 	for (const char* relative_path : candidate_paths)
 	{

+ 9 - 122
Source/Core/DataExpression.cpp

@@ -42,7 +42,7 @@ namespace Rml {
 class DataParser;
 
 /*
-	The abstract machine for RmlUi data scripts.
+	The abstract machine for RmlUi data expressions.
 
 	The machine can execute a program which contains a list of instructions listed below.
 
@@ -62,7 +62,8 @@ class DataParser;
 		S+  Push to stack S.
 		S-  Pop stack S (returns the popped value).
 */
-enum class Instruction { // Assignment (register/stack) = Read (register R/L/C, instruction data D, or stack)
+enum class Instruction {    
+	                        // Assignment (register/stack) = Read (register R/L/C, instruction data D, or stack)
 	Push         = 'P',     //      S+ = R
 	Pop          = 'o',     // <R/L/C> = S-  (D determines R/L/C)
 	Literal      = 'D',     //       R = D
@@ -86,6 +87,7 @@ enum class Instruction { // Assignment (register/stack) = Read (register R/L/C,
 	EventFnc     = 'E',     //       DataModel.EventCallback(D, A); A.Clear();
 	Assign       = 'A',     //       DataModel.SetVariable(D, R)
 };
+
 enum class Register {
 	R,
 	L,
@@ -791,7 +793,7 @@ public:
 		for (size_t i = 0; i < program.size(); i++)
 		{
 			String instruction_str = program[i].data.Get<String>();
-			str += CreateString(50 + instruction_str.size(), "  %4d  '%c'  %s\n", i, char(program[i].instruction), instruction_str.c_str());
+			str += CreateString(50 + instruction_str.size(), "  %4zu  '%c'  %s\n", i, char(program[i].instruction), instruction_str.c_str());
 		}
 		return str;
 	}
@@ -902,7 +904,7 @@ private:
 			if (num_arguments < 0)
 				return Error("Invalid number of arguments.");
 			if (stack.size() < size_t(num_arguments))
-				return Error(CreateString(100, "Cannot pop %d arguments, stack contains only %d elements.", num_arguments, stack.size()));
+				return Error(CreateString(100, "Cannot pop %d arguments, stack contains only %zu elements.", num_arguments, stack.size()));
 
 			arguments.resize(num_arguments);
 			for (int i = num_arguments - 1; i >= 0; i--)
@@ -970,126 +972,11 @@ private:
 };
 
 
-
-#ifdef RMLUI_TESTS_ENABLED
-
-struct TestParser {
-	TestParser() : model(type_register.GetTransformFuncRegister())
-	{
-		DataModelConstructor handle(&model, &type_register);
-		handle.Bind("radius", &radius);
-		handle.Bind("color_name", &color_name);
-		handle.BindFunc("color_value", [this](Variant& variant) {
-			variant = ToString(color_value);
-		});
-
-		String result;
-		result = TestExpression("!!10 - 1 ? 'hello' : 'world' | to_upper",         "WORLD");
-		result = TestExpression("(color_name) + (': rgba(' + color_value + ')')",  "color: rgba(180, 100, 255, 255)");
-		result = TestExpression("'hello world' | to_upper(5 + 12 == 17 ? 'yes' : 'no', 9*2)",  "HELLO WORLD");
-		result = TestExpression("true == false",  "0");
-		result = TestExpression("true != false",  "1");
-		result = TestExpression("true",           "1");
-
-		result = TestExpression("true || false ? true && 3==1+2 ? 'Absolutely!' : 'well..' : 'no'",  "Absolutely!");
-		result = TestExpression(R"('It\'s a fit')",  R"(It's a fit)");
-		result = TestExpression("2 * 2",           "4");
-		result = TestExpression("50000 / 1500",    "33.333");
-		result = TestExpression("5*1+2",           "7");
-		result = TestExpression("5*(1+2)",         "15");
-		result = TestExpression("2*(-2)/4",        "-1");
-		result = TestExpression("5.2 + 19 + 'px'", "24.2px");
-
-		result = TestExpression("(radius | format(2)) + 'm'",    "8.70m");
-		result = TestExpression("radius < 10.5 ? 'smaller' : 'larger'",  "smaller");
-		TestAssignment("radius = 15");
-		result = TestExpression("radius < 10.5 ? 'smaller' : 'larger'",  "larger");
-		TestAssignment("radius = 4; color_name = 'image-color'");
-		result = TestExpression("radius == 4 && color_name == 'image-color'",  "1");
-
-		result = TestExpression("5 == 1 + 2*2 || 8 == 1 + 4  ? 'yes' : 'no'",  "yes");
-		result = TestExpression("!!('fa' + 'lse')", "0");
-		result = TestExpression("!!('tr' + 'ue')", "1");
-		result = TestExpression("'fox' + 'dog' ? 'FoxyDog' : 'hot' + 'dog' | to_upper", "HOTDOG");
-
-		result = TestExpression("3.62345 | round", "4");
-		result = TestExpression("3.62345 | format(0)", "4");
-		result = TestExpression("3.62345 | format(2)", "3.62");
-		result = TestExpression("3.62345 | format(10)", "3.6234500000");
-		result = TestExpression("3.62345 | format(10, true)", "3.62345");
-		result = TestExpression("3.62345 | round | format(2)", "4.00");
-		result = TestExpression("3.0001 | format(2, false)", "3.00");
-		result = TestExpression("3.0001 | format(2, true)", "3");
-
-		result = TestExpression("0.2 + 3.42345 | round", "4");
-		result = TestExpression("(3.42345 | round) + 0.2", "3.2");
-		result = TestExpression("(3.42345 | format(0)) + 0.2", "30.2"); // Here, format(0) returns a string, so the + means string concatenation.
-	}
-
-	String TestExpression(String expression, String expected = String())
-	{
-		String result;
-		DataExpressionInterface interface(&model, nullptr);
-		DataParser parser(expression, interface);
-		if (parser.Parse(false))
-		{
-			Program program = parser.ReleaseProgram();
-			AddressList addresses = parser.ReleaseAddresses();
-
-			DataInterpreter interpreter(program, addresses, interface);
-			if (interpreter.Run())
-				result = interpreter.Result().Get<String>();
-
-			if (!expected.empty() && result != expected)
-			{
-				String program_str = interpreter.DumpProgram();
-				Log::Message(Log::LT_WARNING, "%s", program_str.c_str());
-				RMLUI_ERRORMSG("Got unexpected data parser result.");
-			}
-		}
-		else
-		{
-			RMLUI_ERRORMSG("Could not parse expression.");
-		}
-
-		return result;
-	};
-
-	bool TestAssignment(String expression)
-	{
-		bool result = false;
-		DataExpressionInterface interface(&model, nullptr);
-		DataParser parser(expression, interface);
-		if (parser.Parse(true))
-		{
-			Program program = parser.ReleaseProgram();
-			AddressList addresses = parser.ReleaseAddresses();
-
-			DataInterpreter interpreter(program, addresses, interface);
-			result = interpreter.Run();
-		}
-		RMLUI_ASSERT(result);
-		return result;
-	};
-
-	DataTypeRegister type_register;
-	DataModel model;
-
-	float radius = 8.7f;
-	String color_name = "color";
-	Colourb color_value = Colourb(180, 100, 255);
-};
-
-static TestParser test_parser;
-
-#endif
-
-
-DataExpression::DataExpression(String expression) : expression(expression) {}
+DataExpression::DataExpression(String expression) : expression(expression)
+{}
 
 DataExpression::~DataExpression()
-{
-}
+{}
 
 bool DataExpression::Parse(const DataExpressionInterface& expression_interface, bool is_assignment_expression)
 {

+ 0 - 92
Source/Core/DataModel.cpp

@@ -372,96 +372,4 @@ bool DataModel::Update()
 	return result;
 }
 
-
-
-#ifdef RMLUI_DEBUG
-
-static struct TestDataVariables {
-	TestDataVariables() 
-	{
-		using IntVector = Vector<int>;
-
-		struct FunData {
-			int i = 99;
-			String x = "hello";
-			IntVector magic = { 3, 5, 7, 11, 13 };
-		};
-
-		using FunArray = Array<FunData, 3>;
-
-		struct SmartData {
-			bool valid = true;
-			FunData fun;
-			FunArray more_fun;
-		};
-
-		DataModel model;
-		DataTypeRegister types;
-
-		DataModelConstructor handle(&model, &types);
-
-		{
-			handle.RegisterArray<IntVector>();
-
-			if (auto fun_handle = handle.RegisterStruct<FunData>())
-			{
-				fun_handle.RegisterMember("i", &FunData::i);
-				fun_handle.RegisterMember("x", &FunData::x);
-				fun_handle.RegisterMember("magic", &FunData::magic);
-			}
-
-			handle.RegisterArray<FunArray>();
-
-			if (auto smart_handle = handle.RegisterStruct<SmartData>())
-			{
-				smart_handle.RegisterMember("valid", &SmartData::valid);
-				smart_handle.RegisterMember("fun", &SmartData::fun);
-				smart_handle.RegisterMember("more_fun", &SmartData::more_fun);
-			}
-		}
-
-		SmartData data;
-		data.fun.x = "Hello, we're in SmartData!";
-		
-		handle.Bind("data", &data);
-
-		{
-			Vector<String> test_addresses = { "data.more_fun[1].magic[3]", "data.more_fun[1].magic.size", "data.fun.x", "data.valid" };
-			Vector<String> expected_results = { ToString(data.more_fun[1].magic[3]), ToString(int(data.more_fun[1].magic.size())), ToString(data.fun.x), ToString(data.valid) };
-
-			Vector<String> results;
-
-			for (auto& str_address : test_addresses)
-			{
-				DataAddress address = ParseAddress(str_address);
-
-				Variant result;
-				if(model.GetVariableInto(address, result))
-					results.push_back(result.Get<String>());
-			}
-
-			RMLUI_ASSERT(results == expected_results);
-
-			bool success = true;
-			success &= model.GetVariable(ParseAddress("data.more_fun[1].magic[1]")).Set(Variant(String("199")));
-			RMLUI_ASSERT(success && data.more_fun[1].magic[1] == 199);
-
-			data.fun.magic = { 99, 190, 55, 2000, 50, 60, 70, 80, 90 };
-
-			Variant get_result;
-
-			const int magic_size = int(data.fun.magic.size());
-			success &= model.GetVariable(ParseAddress("data.fun.magic.size")).Get(get_result);
-			RMLUI_ASSERT(success && get_result.Get<String>() == ToString(magic_size));
-			RMLUI_ASSERT(model.GetVariable(ParseAddress("data.fun.magic")).Size() == magic_size);
-
-			success &= model.GetVariable(ParseAddress("data.fun.magic[8]")).Get(get_result);
-			RMLUI_ASSERT(success && get_result.Get<String>() == "90");
-		}
-	}
-} test_data_variables;
-
-
-#endif
-
 } // namespace Rml

+ 8 - 1
Source/Core/Element.cpp

@@ -130,6 +130,8 @@ transform_state(), dirty_transform(false), dirty_perspective(false), dirty_anima
 
 	client_area = Box::PADDING;
 
+	baseline = 0.0f;
+
 	num_non_dom_children = 0;
 
 	visible = true;
@@ -542,7 +544,7 @@ int Element::GetNumBoxes()
 // Returns the baseline of the element, in pixels offset from the bottom of the element's content area.
 float Element::GetBaseline() const
 {
-	return 0;
+	return baseline;
 }
 
 // Gets the intrinsic dimensions of this element, if it is of a type that has an inherent size.
@@ -2165,6 +2167,11 @@ void Element::UpdateOffset()
 	}
 }
 
+void Element::SetBaseline(float in_baseline)
+{
+	baseline = in_baseline;
+}
+
 void Element::BuildLocalStackingContext()
 {
 	stacking_context_dirty = false;

+ 31 - 19
Source/Core/ElementDocument.cpp

@@ -373,8 +373,7 @@ void ElementDocument::UpdateLayout()
 		if (GetParentNode() != nullptr)
 			containing_block = GetParentNode()->GetBox().GetSize();
 
-		LayoutEngine layout_engine;
-		layout_engine.FormatElement(this, containing_block);
+		LayoutEngine::FormatElement(this, containing_block);
 	}
 }
 
@@ -491,6 +490,20 @@ void ElementDocument::OnResize()
 	DirtyPosition();
 }
 
+enum class CanFocus { Yes, No, NoAndNoChildren };
+static CanFocus CanFocusElement(Element* element)
+{
+	if (element->IsPseudoClassSet("disabled"))
+		return CanFocus::NoAndNoChildren;
+
+	if (!element->IsVisible())
+		return CanFocus::NoAndNoChildren;
+
+	if (element->GetComputedValues().tab_index == Style::TabIndex::Auto)
+		return CanFocus::Yes;
+
+	return CanFocus::No;
+}
 
 // Find the next element to focus, starting at the current element
 //
@@ -533,7 +546,7 @@ Element* ElementDocument::FindNextTabElement(Element* current_element, bool forw
 				if(Element* result = SearchFocusSubtree(search_child, forward))
 					return result;
 
-			// If we find the child, enable searching
+			// Enable searching when we reach the child.
 			if (search_child == child)
 				search_enabled = true;
 		}
@@ -542,11 +555,20 @@ Element* ElementDocument::FindNextTabElement(Element* current_element, bool forw
 		child = parent;
 		parent = parent->GetParentNode();
 
-		// If we hit the top, enable searching the entire tree
 		if (parent == document)
+		{
+			// When we hit the top, see if we can focus the document first.
+			if (CanFocusElement(document) == CanFocus::Yes)
+				return document;
+			
+			// Otherwise, search the entire tree to loop back around.
 			search_enabled = true;
-		else // otherwise enable searching if we're going backward and disable if we're going forward
+		}
+		else
+		{
+			// Prepare for the next iteration by disabling searching.
 			search_enabled = false;
+		}
 	}
 
 	return nullptr;
@@ -554,21 +576,11 @@ Element* ElementDocument::FindNextTabElement(Element* current_element, bool forw
 
 Element* ElementDocument::SearchFocusSubtree(Element* element, bool forward)
 {
-	// Skip disabled elements
-	if (element->IsPseudoClassSet("disabled"))
-	{
-		return nullptr;
-	}
-	if (!element->IsVisible())
-	{
-		return nullptr;
-	}
-
-	// Check if this is the node we're looking for
-	if (element->GetComputedValues().tab_index == Style::TabIndex::Auto)
-	{
+	CanFocus can_focus = CanFocusElement(element);
+	if (can_focus == CanFocus::Yes)
 		return element;
-	}
+	else if (can_focus == CanFocus::NoAndNoChildren)
+		return nullptr;
 
 	// Check all children
 	for (int i = 0; i < element->GetNumChildren(); i++)

+ 3 - 2
Source/Core/ElementScroll.cpp

@@ -27,7 +27,7 @@
  */
 
 #include "../../Include/RmlUi/Core/ElementScroll.h"
-#include "LayoutEngine.h"
+#include "LayoutDetails.h"
 #include "WidgetScroll.h"
 #include "../../Include/RmlUi/Core/Element.h"
 #include "../../Include/RmlUi/Core/ElementUtilities.h"
@@ -86,7 +86,7 @@ void ElementScroll::EnableScrollbar(Orientation orientation, float element_width
 
 	// Determine the size of the scrollbar.
 	Box box;
-	LayoutEngine::BuildBox(box, Vector2f(element_width, element_width), scrollbars[orientation].element);
+	LayoutDetails::BuildBox(box, Vector2f(element_width, element_width), scrollbars[orientation].element);
 
 	if (orientation == VERTICAL)
 		scrollbars[orientation].size = box.GetSize(Box::MARGIN).x;
@@ -215,6 +215,7 @@ void ElementScroll::FormatScrollbars()
 		corner_box.SetContent(Vector2f(scrollbars[VERTICAL].size, scrollbars[HORIZONTAL].size));
 		corner->SetBox(corner_box);
 		corner->SetOffset(containing_block - Vector2f(scrollbars[VERTICAL].size, scrollbars[HORIZONTAL].size), element, true);
+		corner->SetProperty(PropertyId::Clip, Property(1, Property::NUMBER));
 
 		corner->SetProperty(PropertyId::Visibility, Property(Style::Visibility::Visible));
 	}

+ 4 - 0
Source/Core/ElementStyle.cpp

@@ -598,6 +598,7 @@ PropertyIdSet ElementStyle::ComputeValues(Style::ComputedValues& values, const S
 		values.text_decoration = parent_values->text_decoration;
 		values.text_transform = parent_values->text_transform;
 		values.white_space = parent_values->white_space;
+		values.word_break = parent_values->word_break;
 
 		values.cursor = parent_values->cursor;
 		values.focus = parent_values->focus;
@@ -786,6 +787,9 @@ PropertyIdSet ElementStyle::ComputeValues(Style::ComputedValues& values, const S
 		case PropertyId::WhiteSpace:
 			values.white_space = (WhiteSpace)p->Get< int >();
 			break;
+		case PropertyId::WordBreak:
+			values.word_break = (WordBreak)p->Get< int >();
+			break;
 
 		case PropertyId::Cursor:
 			values.cursor = p->Get< String >();

+ 49 - 9
Source/Core/ElementTextDefault.cpp

@@ -202,7 +202,7 @@ bool ElementTextDefault::GenerateLine(String& line, int& line_length, float& lin
 	bool collapse_white_space = white_space_property == WhiteSpace::Normal ||
 								white_space_property == WhiteSpace::Nowrap ||
 								white_space_property == WhiteSpace::Preline;
-	bool break_at_line = maximum_line_width >= 0 && 
+	bool break_at_line = (maximum_line_width >= 0) && 
 		                   (white_space_property == WhiteSpace::Normal ||
 							white_space_property == WhiteSpace::Prewrap ||
 							white_space_property == WhiteSpace::Preline);
@@ -210,8 +210,10 @@ bool ElementTextDefault::GenerateLine(String& line, int& line_length, float& lin
 							white_space_property == WhiteSpace::Prewrap ||
 							white_space_property == WhiteSpace::Preline;
 
-	// Determine what (if any) text transformation we are putting the characters through.
 	TextTransform text_transform_property = computed.text_transform;
+	WordBreak word_break = computed.word_break;
+
+	FontEngineInterface* font_engine_interface = GetFontEngineInterface();
 
 	// Starting at the line_begin character, we generate sections of the text (we'll call them tokens) depending on the
 	// white-space parsing parameters. Each section is then appended to the line if it can fit. If not, or if an
@@ -228,21 +230,59 @@ bool ElementTextDefault::GenerateLine(String& line, int& line_length, float& lin
 
 		// Generate the next token and determine its pixel-length.
 		bool break_line = BuildToken(token, next_token_begin, string_end, line.empty() && trim_whitespace_prefix, collapse_white_space, break_at_endline, text_transform_property, decode_escape_characters);
-		int token_width = GetFontEngineInterface()->GetStringWidth(font_face_handle, token, previous_codepoint);
+		int token_width = font_engine_interface->GetStringWidth(font_face_handle, token, previous_codepoint);
 
 		// If we're breaking to fit a line box, check if the token can fit on the line before we add it.
 		if (break_at_line)
 		{
-			if (!line.empty() &&
-				(line_width + token_width > maximum_line_width ||
-				 (LastToken(next_token_begin, string_end, collapse_white_space, break_at_endline) && line_width + token_width > maximum_line_width - right_spacing_width)))
+			const bool is_last_token = LastToken(next_token_begin, string_end, collapse_white_space, break_at_endline);
+			int max_token_width = int(maximum_line_width - (is_last_token ? line_width + right_spacing_width : line_width));
+
+			if (token_width > max_token_width)
 			{
-				return false;
+				if (word_break == WordBreak::BreakAll || (word_break == WordBreak::BreakWord && line.empty()))
+				{
+					// Try to break up the word
+					max_token_width = int(maximum_line_width - line_width);
+					const int token_max_size = int(next_token_begin - token_begin);
+					bool force_loop_break_after_next = false;
+
+					// @performance: Can be made much faster. Use string width heuristics and logarithmic search.
+					for (int i = token_max_size - 1; i > 0; --i)
+					{
+						token.clear();
+						next_token_begin = token_begin;
+						const char* partial_string_end = StringUtilities::SeekBackwardUTF8(token_begin + i, token_begin);
+						break_line = BuildToken(token, next_token_begin, partial_string_end, line.empty() && trim_whitespace_prefix, collapse_white_space, break_at_endline, text_transform_property, decode_escape_characters);
+						token_width = font_engine_interface->GetStringWidth(font_face_handle, token, previous_codepoint);
+
+						if (force_loop_break_after_next || token_width <= max_token_width)
+						{
+							break;
+						}
+						else if (next_token_begin == token_begin)
+						{
+							// This means the first character of the token doesn't fit. Let it overflow into the next line if we can.
+							if (!line.empty())
+								return false;
+
+							// Not even the first character of the line fits. Go back to consume the first character even though it will overflow.
+							i += 2;
+							force_loop_break_after_next = true;
+						}
+					}
+
+					break_line = true;
+				}
+				else if (!line.empty())
+				{
+					// Let the token overflow into the next line.
+					return false;
+				}
 			}
 		}
 
-		// The token can fit on the end of the line, so add it onto the end and increment our width and length
-		// counters.
+		// The token can fit on the end of the line, so add it onto the end and increment our width and length counters.
 		line += token;
 		line_length += (int)(next_token_begin - token_begin);
 		line_width += token_width;

+ 6 - 6
Source/Core/ElementUtilities.cpp

@@ -38,6 +38,7 @@
 #include "../../Include/RmlUi/Core/FontEngineInterface.h"
 #include "../../Include/RmlUi/Core/RenderInterface.h"
 #include "ElementStyle.h"
+#include "LayoutDetails.h"
 #include "LayoutEngine.h"
 #include "TransformState.h"
 #include <limits>
@@ -55,7 +56,7 @@ static void SetBox(Element* element)
 	containing_block.y -= parent->GetElementScroll()->GetScrollbarSize(ElementScroll::HORIZONTAL);
 
 	Box box;
-	LayoutEngine::BuildBox(box, containing_block, element);
+	LayoutDetails::BuildBox(box, containing_block, element);
 
 	if (element->GetComputedValues().height.type != Style::Height::Auto)
 		box.SetContent(Vector2f(box.GetSize().x, containing_block.y));
@@ -303,18 +304,17 @@ void ElementUtilities::ApplyActiveClipRegion(Context* context, RenderInterface*
 // Formats the contents of an element.
 bool ElementUtilities::FormatElement(Element* element, const Vector2f& containing_block)
 {
-	LayoutEngine layout_engine;
-	return layout_engine.FormatElement(element, containing_block);
+	return LayoutEngine::FormatElement(element, containing_block);
 }
 
 // Generates the box for an element.
-void ElementUtilities::BuildBox(Box& box, const Vector2f& containing_block, Element* element, bool inline_element)
+void ElementUtilities::BuildBox(Box& box, Vector2f containing_block, Element* element, bool inline_element)
 {
-	LayoutEngine::BuildBox(box, containing_block, element, inline_element);
+	LayoutDetails::BuildBox(box, containing_block, element, inline_element);
 }
 
 // Sizes an element, and positions it within its parent offset from the borders of its content area.
-bool ElementUtilities::PositionElement(Element* element, const Vector2f& offset, PositionAnchor anchor)
+bool ElementUtilities::PositionElement(Element* element, Vector2f offset, PositionAnchor anchor)
 {
 	Element* parent = element->GetParentNode();
 	if (parent == nullptr)

+ 1 - 1
Source/Core/Elements/DataFormatter.cpp

@@ -42,7 +42,7 @@ DataFormatter::DataFormatter(const String& _name)
 	}
 	else
 	{
-		name = CreateString(64, "%x", this);
+		name = CreateString(64, "%p", (void*)this);
 	}
 	data_formatters[name] = this;
 }

+ 1 - 1
Source/Core/Elements/DataSource.cpp

@@ -49,7 +49,7 @@ DataSource::DataSource(const String& _name)
 	}
 	else
 	{
-		name = CreateString(64, "%x", this);
+		name = CreateString(64, "%p", (void*)this);
 	}
 	data_sources[name] = this;
 }

+ 1 - 1
Source/Core/Elements/ElementDataGridRow.cpp

@@ -505,7 +505,7 @@ void ElementDataGridRow::Load(const DataQuery& row_information)
 				}
 				else if (column->fields[j] == DataSource::NUM_CHILDREN)
 				{
-					raw_data.push_back(CreateString(8, "%d", children.size()));
+					raw_data.push_back(CreateString(8, "%zu", children.size()));
 					raw_data_total_len += raw_data.back().length();
 				}
 				else

+ 1 - 0
Source/Core/Factory.cpp

@@ -294,6 +294,7 @@ void Factory::Shutdown()
 
 	font_effect_instancers.clear();
 
+	data_controller_instancers.clear();
 	data_view_instancers.clear();
 	structural_data_view_instancers.clear();
 	structural_data_view_attribute_names.clear();

+ 29 - 49
Source/Core/GeometryDatabase.cpp

@@ -30,8 +30,8 @@
 #include "../../Include/RmlUi/Core/Geometry.h"
 #include <algorithm>
 
-namespace Rml {
 
+namespace Rml {
 namespace GeometryDatabase {
 
 class Database {
@@ -70,6 +70,15 @@ public:
 		return handle;
 	}
 
+	int size() const
+	{
+		return (int)geometry_list.size() - (int)free_list.size();
+	}
+
+	void clear() {
+		geometry_list.clear();
+		free_list.clear();
+	}
 	void erase(GeometryDatabaseHandle handle)
 	{
 		free_list.push_back(handle);
@@ -125,59 +134,30 @@ void ReleaseAll()
 }
 
 
-
 #ifdef RMLUI_TESTS_ENABLED
 
-static class TestGeometryDatabase {
-private:
-	Vector<Geometry> geometry_list;
-
-	bool list_database_equivalent()
-	{
-		int i = 0;
-		bool result = true;
-		GetDatabase().for_each([this, &i, &result](Geometry* geometry) {
-			result &= (geometry == &geometry_list[i++]);
-		});
-		return result;
-	}
-
-public:
-	TestGeometryDatabase() : geometry_list(10)
-	{
-		bool result = true;
-
-		int i = 0;
-		for (auto& geometry : geometry_list)
-			geometry.GetIndices().push_back(i++);
-
-		result &= list_database_equivalent();
-
-		geometry_list.reserve(2000);
-		result &= list_database_equivalent();
-
-		geometry_list.erase(geometry_list.begin() + 5);
-		result &= list_database_equivalent();
-
-		std::swap(geometry_list.front(), geometry_list.back());
-		geometry_list.pop_back();
-		result &= list_database_equivalent();
-
-		std::swap(geometry_list.front(), geometry_list.back());
-		result &= list_database_equivalent();
-
-		geometry_list.emplace_back();
-		result &= list_database_equivalent();
+bool PrepareForTests()
+{
+	if (geometry_database.size() > 0)
+		return false;
 
-		geometry_list.clear();
-		result &= list_database_equivalent();
+	// Even with size()==0 we can have items in the geometry list which should all be duplicated by the free list. We want to clear them for the tests.
+	geometry_database.clear();
 
-		RMLUI_ASSERT(result);
-	}
+	return true;
+}
 
-} test_geometry_database;
+bool ListMatchesDatabase(const Vector<Geometry>& geometry_list)
+{
+	int i = 0;
+	bool result = true;
+	geometry_database.for_each([&geometry_list, &i, &result](Geometry* geometry) {
+		result &= (geometry == &geometry_list[i++]);
+		});
+	return result;
+}
 
-#endif
+#endif // RMLUI_TESTS_ENABLED
 
-}
+} // namespace GeometryDatabase
 } // namespace Rml

+ 5 - 0
Source/Core/GeometryDatabase.h

@@ -30,6 +30,7 @@
 #define RMLUI_CORE_GEOMETRYDATABASE_H
 
 #include <stdint.h>
+#include "../../Include/RmlUi/Core/Types.h"
 
 namespace Rml {
 
@@ -52,6 +53,10 @@ namespace GeometryDatabase {
 
     void ReleaseAll();
 
+#ifdef RMLUI_TESTS_ENABLED
+    bool PrepareForTests();
+    bool ListMatchesDatabase(const Vector<Geometry>& geometry_list);
+#endif
 }
 
 } // namespace Rml

+ 144 - 99
Source/Core/LayoutBlockBox.cpp

@@ -29,6 +29,7 @@
 #include "LayoutBlockBox.h"
 #include "LayoutBlockBoxSpace.h"
 #include "LayoutEngine.h"
+#include "LayoutDetails.h"
 #include "../../Include/RmlUi/Core/Element.h"
 #include "../../Include/RmlUi/Core/ElementUtilities.h"
 #include "../../Include/RmlUi/Core/ElementScroll.h"
@@ -39,13 +40,13 @@
 namespace Rml {
 
 // Creates a new block box for rendering a block element.
-LayoutBlockBox::LayoutBlockBox(LayoutEngine* _layout_engine, LayoutBlockBox* _parent, Element* _element) : position(0, 0)
+LayoutBlockBox::LayoutBlockBox(LayoutBlockBox* _parent, Element* _element, float override_shrink_to_fit_width) : position(0), visible_outer_width(0)
 {
 	RMLUI_ZoneScoped;
 
-	space = new LayoutBlockBoxSpace(this);
+	space_owner = MakeUnique<LayoutBlockBoxSpace>(this);
+	space = space_owner.get();
 
-	layout_engine = _layout_engine;
 	parent = _parent;
 
 	context = BLOCK;
@@ -63,7 +64,7 @@ LayoutBlockBox::LayoutBlockBox(LayoutEngine* _layout_engine, LayoutBlockBox* _pa
 		offset_root = this;
 
 	// Determine the offset parent for this element.
-	LayoutBlockBox* self_offset_parent;
+	const LayoutBlockBox* self_offset_parent;
 	if (parent != nullptr &&
 		parent->offset_parent->GetElement() != nullptr)
 		self_offset_parent = parent->offset_parent;
@@ -84,7 +85,7 @@ LayoutBlockBox::LayoutBlockBox(LayoutEngine* _layout_engine, LayoutBlockBox* _pa
 		space->ImportSpace(*parent->space);
 
 		// Build our box if possible; if not, it will have to be set up manually.
-		layout_engine->BuildBox(box, min_height, max_height, parent, element);
+		LayoutDetails::BuildBox(box, min_height, max_height, parent, element, false, override_shrink_to_fit_width);
 
 		// Position ourselves within our containing block (if we have a valid offset parent).
 		if (parent->GetElement() != nullptr)
@@ -128,9 +129,8 @@ LayoutBlockBox::LayoutBlockBox(LayoutEngine* _layout_engine, LayoutBlockBox* _pa
 }
 
 // Creates a new block box in an inline context.
-LayoutBlockBox::LayoutBlockBox(LayoutEngine* _layout_engine, LayoutBlockBox* _parent) : position(-1, -1)
+LayoutBlockBox::LayoutBlockBox(LayoutBlockBox* _parent) : position(-1, -1)
 {
-	layout_engine = _layout_engine;
 	parent = _parent;
 	offset_parent = parent->offset_parent;
 	offset_root = parent->offset_root;
@@ -138,7 +138,7 @@ LayoutBlockBox::LayoutBlockBox(LayoutEngine* _layout_engine, LayoutBlockBox* _pa
 	space = _parent->space;
 
 	context = INLINE;
-	line_boxes.push_back(new LayoutLineBox(this));
+	line_boxes.push_back(MakeUnique<LayoutLineBox>(this));
 	wrap_content = parent->wrap_content;
 
 	element = nullptr;
@@ -147,7 +147,7 @@ LayoutBlockBox::LayoutBlockBox(LayoutEngine* _layout_engine, LayoutBlockBox* _pa
 	box_cursor = 0;
 	vertical_overflow = false;
 
-	layout_engine->BuildBox(box, min_height, max_height, parent, nullptr);
+	LayoutDetails::BuildBox(box, min_height, max_height, parent, nullptr, false);
 	parent->PositionBlockBox(position, box, Style::Clear::None);
 	box.SetContent(Vector2f(box.GetSize(Box::CONTENT).x, -1));
 
@@ -159,14 +159,6 @@ LayoutBlockBox::LayoutBlockBox(LayoutEngine* _layout_engine, LayoutBlockBox* _pa
 // Releases the block box.
 LayoutBlockBox::~LayoutBlockBox()
 {
-	for (size_t i = 0; i < block_boxes.size(); i++)
-		delete block_boxes[i];
-
-	for (size_t i = 0; i < line_boxes.size(); i++)
-		delete line_boxes[i];
-
-	if (context == BLOCK)
-		delete space;
 }
 
 // Closes the box.
@@ -203,72 +195,125 @@ LayoutBlockBox::CloseResult LayoutBlockBox::Close()
 
 		box.SetContent(content_area);
 	}
+	
+	visible_outer_width = 0;
+	RMLUI_ASSERTMSG(!(context == INLINE && element), "The following assumes inline contexts do not represent a particular element.");
 
 	// Set the computed box on the element.
-	if (element != nullptr)
+	if (context == BLOCK && element)
 	{
-		if (context == BLOCK)
-		{
-			// Calculate the dimensions of the box's *internal* content; this is the tightest-fitting box around all of the
-			// internal elements, plus this element's padding.
-			Vector2f content_box(0, 0);
+		// Calculate the dimensions of the box's *internal* content; this is the tightest-fitting box around all of the
+		// internal elements, plus this element's padding.
+		Vector2f content_box(0, 0);
 
-			for (size_t i = 0; i < block_boxes.size(); i++)
-				content_box.x = Math::Max(content_box.x, block_boxes[i]->GetBox().GetSize(Box::MARGIN).x);
+		for (size_t i = 0; i < block_boxes.size(); i++)
+		{
+			// TODO: Only if the containing block is not an ancestor of us (ie. we are the containing block?).
+			content_box.x = Math::Max(content_box.x, block_boxes[i]->visible_outer_width);
+		}
 
-			// Check how big our floated area is.
-			Vector2f space_box = space->GetDimensions();
-			content_box.x = Math::Max(content_box.x, space_box.x);
+		// Check how big our floated area is.
+		Vector2f space_box = space->GetDimensions();
+		content_box.x = Math::Max(content_box.x, space_box.x);
 
-			// If our content is larger than our window, we can enable the horizontal scrollbar if
-			// we're set to auto-scrollbars. If we're set to always use scrollbars, then the horiontal
-			// scrollbar will already have been enabled in the constructor.
-			if (content_box.x > box.GetSize().x)
+		// If our content is larger than our window, we can enable the horizontal scrollbar if
+		// we're set to auto-scrollbars. If we're set to always use scrollbars, then the horiontal
+		// scrollbar will already have been enabled in the constructor.
+		if (content_box.x > box.GetSize().x)
+		{
+			if (overflow_x_property == Style::Overflow::Auto)
 			{
-				if (overflow_x_property == Style::Overflow::Auto)
-				{
-					element->GetElementScroll()->EnableScrollbar(ElementScroll::HORIZONTAL, box.GetSize(Box::PADDING).x);
+				element->GetElementScroll()->EnableScrollbar(ElementScroll::HORIZONTAL, box.GetSize(Box::PADDING).x);
 
-					if (!CatchVerticalOverflow())
-						return LAYOUT_SELF;
-				}
+				if (!CatchVerticalOverflow())
+					return LAYOUT_SELF;
 			}
+		}
 
-			content_box.x += (box.GetEdge(Box::PADDING, Box::LEFT) + box.GetEdge(Box::PADDING, Box::RIGHT));
+		content_box.x += (box.GetEdge(Box::PADDING, Box::LEFT) + box.GetEdge(Box::PADDING, Box::RIGHT));
 
-			content_box.y = box_cursor;
-			content_box.y = Math::Max(content_box.y, space_box.y);
-			if (!CatchVerticalOverflow(content_box.y))
-				return LAYOUT_SELF;
+		content_box.y = box_cursor;
+		content_box.y = Math::Max(content_box.y, space_box.y);
+		if (!CatchVerticalOverflow(content_box.y))
+			return LAYOUT_SELF;
 
-			content_box.y += (box.GetEdge(Box::PADDING, Box::TOP) + box.GetEdge(Box::PADDING, Box::BOTTOM));
+		content_box.y += (box.GetEdge(Box::PADDING, Box::TOP) + box.GetEdge(Box::PADDING, Box::BOTTOM));
 
-			element->SetBox(box);
-			element->SetContentBox(space->GetOffset(), content_box);
+		element->SetBox(box);
+		element->SetContentBox(space->GetOffset(), content_box);
 
-			// Format any scrollbars which were enabled on this element.
-			element->GetElementScroll()->FormatScrollbars();
-		}
+		const float margin_width = box.GetSize(Box::MARGIN).x;
+
+		// Set the visible outer width so that ancestors can catch any overflow produced by us. That is, hiding it or providing a scrolling mechanism.
+		// If we catch our own overflow here, then just use the normal margin box as that will effectively remove the overflow from our ancestor's perspective.
+		if (overflow_x_property != Style::Overflow::Visible)
+			visible_outer_width = margin_width;
 		else
-			element->SetBox(box);
+			visible_outer_width = Math::Max(margin_width, space->GetOffset().x + content_box.x + box.GetEdge(Box::MARGIN, Box::LEFT) + box.GetEdge(Box::MARGIN, Box::RIGHT));
+
+		// Format any scrollbars which were enabled on this element.
+		element->GetElementScroll()->FormatScrollbars();
+	}
+	else if (context == INLINE)
+	{
+		// Find the largest line in this layout block
+		for (size_t i = 0; i < line_boxes.size(); i++)
+		{
+			LayoutLineBox* line_box = line_boxes[i].get();
+			visible_outer_width = Math::Max(visible_outer_width, line_box->GetBoxCursor());
+		}
 	}
 
 	// Increment the parent's cursor.
 	if (parent != nullptr)
 	{
-		// If this close fails, it means this block box has caused our parent block box to generate an automatic
-		// vertical scrollbar.
+		// If this close fails, it means this block box has caused our parent block box to generate an automatic vertical scrollbar.
 		if (!parent->CloseBlockBox(this))
 			return LAYOUT_PARENT;
 	}
 
-	// If we represent a positioned element, then we can now (as we've been sized) act as the containing block for all
-	// the absolutely-positioned elements of our descendants.
-	if (context == BLOCK &&
-		element != nullptr)
+	if (context == BLOCK && element)
 	{
+		// If we represent a positioned element, then we can now (as we've been sized) act as the containing block for all
+		// the absolutely-positioned elements of our descendants.
 		if (element->GetPosition() != Style::Position::Static)
+		{
 			CloseAbsoluteElements();
+		}
+
+		// Set the baseline for inline-block elements to the baseline of the last line of the element.
+		// This is a special rule for inline-blocks (see CSS 2.1 §10.8.1).
+		if (element->GetDisplay() == Style::Display::InlineBlock)
+		{
+			bool found_baseline = false;
+			float baseline = 0;
+
+			for (int i = (int)block_boxes.size() - 1; i >= 0; i--)
+			{
+				if (block_boxes[i]->context == INLINE)
+				{
+					const LineBoxList& line_boxes = block_boxes[i]->line_boxes;
+					for (int j = (int)line_boxes.size() - 1; j >= 0; j--)
+					{
+						found_baseline = line_boxes[j]->GetBaselineOfLastLine(baseline);
+						if (found_baseline)
+							break;
+					}
+					if (found_baseline)
+						break;
+				}
+			}
+
+			if (found_baseline)
+			{
+				if (baseline < 0 && (overflow_x_property != Style::Overflow::Visible || overflow_x_property != Style::Overflow::Visible))
+				{
+					baseline = 0;
+				}
+
+				element->SetBaseline(baseline);
+			}
+		}
 	}
 
 	return OK;
@@ -284,7 +329,7 @@ bool LayoutBlockBox::CloseBlockBox(LayoutBlockBox* child)
 }
 
 // Called by a closing line box child.
-LayoutInlineBox* LayoutBlockBox::CloseLineBox(LayoutLineBox* child, LayoutInlineBox* overflow, LayoutInlineBox* overflow_chain)
+LayoutInlineBox* LayoutBlockBox::CloseLineBox(LayoutLineBox* child, UniquePtr<LayoutInlineBox> overflow, LayoutInlineBox* overflow_chain)
 {
 	RMLUI_ZoneScoped;
 
@@ -302,19 +347,19 @@ LayoutInlineBox* LayoutBlockBox::CloseLineBox(LayoutLineBox* child, LayoutInline
 	}
 
 	// Add a new line box.
-	line_boxes.push_back(new LayoutLineBox(this));
+	line_boxes.push_back(MakeUnique<LayoutLineBox>(this));
 
-	if (overflow_chain != nullptr)
+	if (overflow_chain)
 		line_boxes.back()->AddChainedBox(overflow_chain);
 
-	if (overflow != nullptr)
-		return line_boxes.back()->AddBox(overflow);
+	if (overflow)
+		return line_boxes.back()->AddBox(std::move(overflow));
 
 	return nullptr;
 }
 
 // Adds a new block element to this block box.
-LayoutBlockBox* LayoutBlockBox::AddBlockElement(Element* element)
+LayoutBlockBox* LayoutBlockBox::AddBlockElement(Element* element, float override_shrink_to_fit_width)
 {
 	RMLUI_ZoneScoped;
 
@@ -324,7 +369,7 @@ LayoutBlockBox* LayoutBlockBox::AddBlockElement(Element* element)
 	if (!block_boxes.empty() &&
 		block_boxes.back()->context == INLINE)
 	{
-		LayoutBlockBox* inline_block_box = block_boxes.back();
+		LayoutBlockBox* inline_block_box = block_boxes.back().get();
 		LayoutInlineBox* open_inline_box = inline_block_box->line_boxes.back()->GetOpenInlineBox();
 		if (open_inline_box != nullptr)
 		{
@@ -347,8 +392,8 @@ LayoutBlockBox* LayoutBlockBox::AddBlockElement(Element* element)
 		}
 	}
 
-	block_boxes.push_back(new LayoutBlockBox(layout_engine, this, element));
-	return block_boxes.back();
+	block_boxes.push_back(MakeUnique<LayoutBlockBox>(this, element, override_shrink_to_fit_width));
+	return block_boxes.back().get();
 }
 
 // Adds a new inline element to this inline box.
@@ -368,7 +413,7 @@ LayoutInlineBox* LayoutBlockBox::AddInlineElement(Element* element, const Box& b
 		// No dice! Ah well, nothing for it but to open a new inline context block box.
 		else
 		{
-			block_boxes.push_back(new LayoutBlockBox(layout_engine, this));
+			block_boxes.push_back(MakeUnique<LayoutBlockBox>(this));
 
 			if (interrupted_chain != nullptr)
 			{
@@ -396,10 +441,10 @@ void LayoutBlockBox::AddBreak()
 	// Check for an inline box as our last child; if so, we can simply end its line and bail.
 	if (!block_boxes.empty())
 	{
-		LayoutBlockBox* block_box = block_boxes.back();
+		LayoutBlockBox* block_box = block_boxes.back().get();
 		if (block_box->context == INLINE)
 		{
-			LayoutLineBox* last_line = block_box->line_boxes.back();
+			LayoutLineBox* last_line = block_box->line_boxes.back().get();
 			if (last_line->GetDimensions().y < 0)
 				block_box->box_cursor += line_height;
 			else
@@ -443,7 +488,7 @@ void LayoutBlockBox::AddAbsoluteElement(Element* element)
 	if (!block_boxes.empty()
 		&& block_boxes.back()->context == INLINE)
 	{
-		LayoutBlockBox* inline_context_box = block_boxes.back();
+		LayoutBlockBox* inline_context_box = block_boxes.back().get();
 		float last_line_height = inline_context_box->line_boxes.back()->GetDimensions().y;
 
 		absolute_element.position.y += (inline_context_box->box_cursor + Math::Max(0.0f, last_line_height));
@@ -472,8 +517,7 @@ void LayoutBlockBox::CloseAbsoluteElements()
 			absolute_position -= position - offset_root->GetPosition();
 
 			// Lay out the element.
-			LayoutEngine layout_engine;
-			layout_engine.FormatElement(absolute_element, containing_block);
+			LayoutEngine::FormatElement(absolute_element, containing_block);
 
 			// Now that the element's box has been built, we can offset the position we determined was appropriate for
 			// it by the element's margin. This is necessary because the coordinate system for the box begins at the
@@ -534,44 +578,47 @@ void LayoutBlockBox::PositionLineBox(Vector2f& box_position, float& box_width, b
 }
 
 
-// Calculate the dimensions of the box's internal width; i.e. the size of the largest line, plus this element's padding.
-float LayoutBlockBox::InternalContentWidth() const
+// Calculate the dimensions of the box's internal content width; i.e. the size of the largest line.
+float LayoutBlockBox::GetShrinkToFitWidth() const
 {
 	float content_width = 0.0f;
 
 	if (context == BLOCK)
 	{
+		auto get_content_width_from_children = [this, &content_width]() {
+			for (size_t i = 0; i < block_boxes.size(); i++)
+			{
+				const Box& box = block_boxes[i]->GetBox();
+				const float edge_size = box.GetCumulativeEdge(Box::PADDING, Box::LEFT) + box.GetCumulativeEdge(Box::PADDING, Box::RIGHT);
+				content_width = Math::Max(content_width, block_boxes[i]->GetShrinkToFitWidth() + edge_size);
+			}
+		};
 
-		for (size_t i = 0; i < block_boxes.size(); i++)
-		{
-			content_width = Math::Max(content_width, block_boxes[i]->InternalContentWidth());
-		}
-
-		// Work-around for supporting 'width' specification of 'display:block' elements inside 'display:inline-block'.
-		//  Alternative solution: Add some 'intrinsic_width' property to  every 'LayoutBlockBox' and have that propagate up to the nearest 'inline-block'.
+		// Block boxes with definite sizes should use that size. Otherwise, find the maximum content width of our children.
+		//  Alternative solution: Add some 'intrinsic_width' property to every 'LayoutBlockBox' and have that propagate up.
 		if (element)
 		{
 			auto& computed = element->GetComputedValues();
 			const float block_width = box.GetSize(Box::CONTENT).x;
 
-			if(computed.width.type != Style::Width::Auto)
+			if(computed.width.type == Style::Width::Auto)
 			{
-				float w_value = ResolveValue(computed.width, block_width);
-				content_width = Math::Max(content_width, w_value);
+				get_content_width_from_children();
 			}
-
-			float min_width = ResolveValue(computed.min_width, block_width);
-			content_width = Math::Max(content_width, min_width);
-			
-			if (computed.max_width.value >= 0.f)
+			else
 			{
-				float value = ResolveValue(computed.max_width, block_width);
-				content_width = Math::Min(content_width, value);
+				float width_value = ResolveValue(computed.width, block_width);
+				content_width = Math::Max(content_width, width_value);
 			}
+
+			content_width = LayoutDetails::ClampWidth(content_width, computed, block_width);
+		}
+		else
+		{
+			get_content_width_from_children();
 		}
 
-		content_width += (box.GetEdge(Box::PADDING, Box::LEFT) + box.GetEdge(Box::PADDING, Box::RIGHT));
-		content_width += (box.GetEdge(Box::MARGIN, Box::LEFT) + box.GetEdge(Box::MARGIN, Box::RIGHT));
+		// Can add the dimensions of floating elements here if we want to support that.
 	}
 	else
 	{
@@ -580,7 +627,7 @@ float LayoutBlockBox::InternalContentWidth() const
 		{
 			// Perhaps a more robust solution is to modify how we set the line box dimension on 'line_box->close()'
 			// and use that, or add another value in the line_box ... but seems to work for now.
-			LayoutLineBox* line_box = line_boxes[i];
+			LayoutLineBox* line_box = line_boxes[i].get();
 			content_width = Math::Max(content_width, line_box->GetBoxCursor());
 		}
 		content_width = Math::Min(content_width, box.GetSize(Box::CONTENT).x);
@@ -609,13 +656,13 @@ const Vector2f& LayoutBlockBox::GetPosition() const
 }
 
 // Returns the element against which all positions of boxes in the hierarchy are calculated relative to.
-LayoutBlockBox* LayoutBlockBox::GetOffsetParent() const
+const LayoutBlockBox* LayoutBlockBox::GetOffsetParent() const
 {
 	return offset_parent;
 }
 
 // Returns the block box against which all positions of boxes in the hierarchy are calculated relative to.
-LayoutBlockBox* LayoutBlockBox::GetOffsetRoot() const
+const LayoutBlockBox* LayoutBlockBox::GetOffsetRoot() const
 {
 	return offset_root;
 }
@@ -683,12 +730,10 @@ bool LayoutBlockBox::CatchVerticalOverflow(float cursor)
 			vertical_overflow = true;
 			element->GetElementScroll()->EnableScrollbar(ElementScroll::VERTICAL, box.GetSize(Box::PADDING).x);
 
-			for (size_t i = 0; i < block_boxes.size(); i++)
-				delete block_boxes[i];
 			block_boxes.clear();
 
-			delete space;
-			space = new LayoutBlockBoxSpace(this);
+			space_owner = MakeUnique<LayoutBlockBoxSpace>(this);
+			space = space_owner.get();
 
 			box_cursor = 0;
 			interrupted_chain = nullptr;

+ 24 - 18
Source/Core/LayoutBlockBox.h

@@ -59,14 +59,13 @@ public:
 	};
 
 	/// Creates a new block box for rendering a block element.
-	/// @param layout_engine[in] The layout engine that created this block box.
 	/// @param parent[in] The parent of this block box. This will be nullptr for the root element.
 	/// @param element[in] The element this block box is laying out.
-	LayoutBlockBox(LayoutEngine* layout_engine, LayoutBlockBox* parent, Element* element);
+	/// @param override_shrink_to_fit_width[in] Provide a fixed shrink-to-fit width instead of formatting the element when its properties allow shrinking.
+	LayoutBlockBox(LayoutBlockBox* parent, Element* element, float override_shrink_to_fit_width = -1);
 	/// Creates a new block box in an inline context.
-	/// @param layout_engine[in] The layout engine that created this block box.
 	/// @param parent[in] The parent of this block box.
-	LayoutBlockBox(LayoutEngine* layout_engine, LayoutBlockBox* parent);
+	LayoutBlockBox(LayoutBlockBox* parent);
 	/// Releases the block box.
 	~LayoutBlockBox();
 
@@ -84,13 +83,14 @@ public:
 	/// @param overflow[in] The overflow from the closing line box. May be nullptr if there was no overflow.
 	/// @param overflow_chain[in] The end of the chained hierarchy to be spilled over to the new line, as the parent to the overflow box (if one exists).
 	/// @return If the line box had overflow, this will be the last inline box created by the overflow.
-	LayoutInlineBox* CloseLineBox(LayoutLineBox* child, LayoutInlineBox* overflow, LayoutInlineBox* overflow_chain);
+	LayoutInlineBox* CloseLineBox(LayoutLineBox* child, UniquePtr<LayoutInlineBox> overflow, LayoutInlineBox* overflow_chain);
 
 	/// Adds a new block element to this block-context box.
 	/// @param element[in] The new block element.
 	/// @param placed[in] True if the element is to be placed, false otherwise.
+	/// @param override_shrink_to_fit_width[in] Provide a fixed shrink-to-fit width instead of formatting the element when its properties allow shrinking.
 	/// @return The block box representing the element. Once the element's children have been positioned, Close() must be called on it.
-	LayoutBlockBox* AddBlockElement(Element* element);
+	LayoutBlockBox* AddBlockElement(Element* element, float override_shrink_to_fit_width = -1);
 	/// Adds a new inline element to this inline-context box.
 	/// @param element[in] The new inline element.
 	/// @param box[in] The box defining the element's bounds.
@@ -107,8 +107,6 @@ public:
 	/// rendering in a block-context.
 	/// @param element[in] The element to be positioned absolutely within this block box.
 	void AddAbsoluteElement(Element* element);
-	/// Formats, sizes, and positions all absolute elements in this block.
-	void CloseAbsoluteElements();
 
 	/// Returns the offset from the top-left corner of this box's offset element the next child box will be
 	/// positioned at.
@@ -129,7 +127,8 @@ public:
 	/// @param dimensions[in] The minimum dimensions of the line.
 	void PositionLineBox(Vector2f& box_position, float& box_width, bool& wrap_content, const Vector2f& dimensions) const;
 
-	float InternalContentWidth() const;
+	/// Calculate the dimensions of the box's internal content width; i.e. the size used to calculate the shrink-to-fit width.
+	float GetShrinkToFitWidth() const;
 
 	/// Returns the block box's element.
 	/// @return The block box's element.
@@ -145,10 +144,10 @@ public:
 
 	/// Returns the block box against which all positions of boxes in the hierarchy are set relative to.
 	/// @return This box's offset parent.
-	LayoutBlockBox* GetOffsetParent() const;
+	const LayoutBlockBox* GetOffsetParent() const;
 	/// Returns the block box against which all positions of boxes in the hierarchy are calculated relative to.
 	/// @return This box's offset root.
-	LayoutBlockBox* GetOffsetRoot() const;
+	const LayoutBlockBox* GetOffsetRoot() const;
 
 
 	/// Returns the block box's dimension box.
@@ -168,6 +167,9 @@ private:
 		Vector2f position;
 	};
 
+	/// Formats, sizes, and positions all absolute elements in this block.
+	void CloseAbsoluteElements();
+
 	// Closes our last block box, if it is an open inline block box.
 	CloseResult CloseInlineBlockBox();
 
@@ -179,22 +181,20 @@ private:
 	// overflow occured, false if it did.
 	bool CatchVerticalOverflow(float cursor = -1);
 
-	typedef Vector< AbsoluteElement > AbsoluteElementList;
-	typedef Vector< LayoutBlockBox* > BlockBoxList;
-	typedef Vector< LayoutLineBox* > LineBoxList;
+	using AbsoluteElementList = Vector< AbsoluteElement >;
+	using BlockBoxList = Vector< UniquePtr<LayoutBlockBox> >;
+	using LineBoxList = Vector< UniquePtr<LayoutLineBox> >;
 
 	// The object managing our space, as occupied by floating elements of this box and our ancestors.
 	LayoutBlockBoxSpace* space;
 
-	// The box's layout engine.
-	LayoutEngine* layout_engine;
 	// The element this box represents. This will be nullptr for boxes rendering in an inline context.
 	Element* element;
 
 	// The element we'll be computing our offset relative to during layout.
-	LayoutBlockBox* offset_root;
+	const LayoutBlockBox* offset_root;
 	// The element this block box's children are to be offset from.
-	LayoutBlockBox* offset_parent;
+	const LayoutBlockBox* offset_parent;
 
 	// The box's block parent. This will be nullptr for the root of the box tree.
 	LayoutBlockBox* parent;
@@ -208,6 +208,7 @@ private:
 	Box box;
 	float min_height;
 	float max_height;
+
 	// Used by inline contexts only; set to true if the block box's line boxes should stretch to fit their inline content instead of wrapping.
 	bool wrap_content;
 
@@ -218,12 +219,17 @@ private:
 	BlockBoxList block_boxes;
 	// Used by block contexts only; stores any elements that are to be absolutely positioned within this block box.
 	AbsoluteElementList absolute_elements;
+	// Used by block contexts only; stores the block box space pointed to by the 'space' member.
+	UniquePtr<LayoutBlockBoxSpace> space_owner;
 	// Used by block contexts only; stores an inline element hierarchy that was interrupted by a child block box.
 	// The hierarchy will be resumed in an inline-context box once the intervening block box is completed.
 	LayoutInlineBox* interrupted_chain;
 	// Used by block contexts only; stores the value of the overflow property for the element.
 	Style::Overflow overflow_x_property;
 	Style::Overflow overflow_y_property;
+	//  Used by block contexts only; the content width as visible from the parent. Similar to scroll width, but shrinked if overflow is caught here. 
+	//   This can be wider than the box if we are overflowing. Only available after the box has been closed. 
+	float visible_outer_width;
 	// Used by block contexts only; if true, we've enabled our vertical scrollbar.
 	bool vertical_overflow;
 

+ 4 - 2
Source/Core/LayoutBlockBoxSpace.cpp

@@ -132,8 +132,10 @@ float LayoutBlockBoxSpace::PositionBox(Vector2f& box_position, float cursor, con
 	AnchorEdge box_edge = float_property == Style::Float::Right ? RIGHT : LEFT;
 
 	box_position.y = cursor;
-	box_position.x = box_edge == LEFT ? 0 : (parent->GetBox().GetSize().x - dimensions.x) - parent_scrollbar_width;
-	box_position.x += parent_origin;
+	box_position.x = parent_origin;
+
+	if (box_edge == RIGHT)
+		box_position.x += parent->GetBox().GetSize().x - dimensions.x - parent_scrollbar_width;
 
 	float next_cursor = FLT_MAX;
 

+ 1 - 1
Source/Core/LayoutBlockBoxSpace.h

@@ -110,7 +110,7 @@ private:
 		Vector2f dimensions;
 	};
 
-	typedef Vector< SpaceBox > SpaceBoxList;
+	using SpaceBoxList = Vector< SpaceBox >;
 
 	// Our block-box parent.
 	LayoutBlockBox* parent;

+ 490 - 0
Source/Core/LayoutDetails.cpp

@@ -0,0 +1,490 @@
+/*
+ * This source file is part of RmlUi, the HTML/CSS Interface Middleware
+ *
+ * For the latest information, see http://github.com/mikke89/RmlUi
+ *
+ * Copyright (c) 2008-2010 CodePoint Ltd, Shift Technology Ltd
+ * Copyright (c) 2019 The RmlUi Team, and contributors
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ */
+
+#include "LayoutDetails.h"
+#include "LayoutEngine.h"
+#include "../../Include/RmlUi/Core/Element.h"
+#include "../../Include/RmlUi/Core/ElementScroll.h"
+#include "../../Include/RmlUi/Core/Math.h"
+#include "../../Include/RmlUi/Core/Profiling.h"
+#include <float.h>
+
+namespace Rml {
+
+// Generates the box for an element.
+void LayoutDetails::BuildBox(Box& box, Vector2f containing_block, Element* element, bool inline_element, float override_shrink_to_fit_width)
+{
+	if (element == nullptr)
+	{
+		box.SetContent(containing_block);
+		return;
+	}
+
+	const ComputedValues& computed = element->GetComputedValues();
+
+	// Calculate the padding area.
+	float padding = ResolveValue(computed.padding_top, containing_block.x);
+	box.SetEdge(Box::PADDING, Box::TOP, Math::Max(0.0f, padding));
+	padding = ResolveValue(computed.padding_right, containing_block.x);
+	box.SetEdge(Box::PADDING, Box::RIGHT, Math::Max(0.0f, padding));
+	padding = ResolveValue(computed.padding_bottom, containing_block.x);
+	box.SetEdge(Box::PADDING, Box::BOTTOM, Math::Max(0.0f, padding));
+	padding = ResolveValue(computed.padding_left, containing_block.x);
+	box.SetEdge(Box::PADDING, Box::LEFT, Math::Max(0.0f, padding));
+
+	// Calculate the border area.
+	box.SetEdge(Box::BORDER, Box::TOP, Math::Max(0.0f, computed.border_top_width));
+	box.SetEdge(Box::BORDER, Box::RIGHT, Math::Max(0.0f, computed.border_right_width));
+	box.SetEdge(Box::BORDER, Box::BOTTOM, Math::Max(0.0f, computed.border_bottom_width));
+	box.SetEdge(Box::BORDER, Box::LEFT, Math::Max(0.0f, computed.border_left_width));
+
+	// Calculate the size of the content area.
+	Vector2f content_area(-1, -1);
+	bool replaced_element = false;
+
+	// If the element has intrinsic dimensions, then we use those as the basis for the content area and only adjust
+	// them if a non-auto style has been applied to them.
+	if (element->GetIntrinsicDimensions(content_area))
+	{
+		replaced_element = true;
+
+		Vector2f original_content_area = content_area;
+
+		// The element has resized itself, so we only resize it if a RCSS width or height was set explicitly. A value of
+		// 'auto' (or 'auto-fit', ie, both keywords) means keep (or adjust) the intrinsic dimensions.
+		bool auto_width = false, auto_height = false;
+
+		if (computed.width.type != Style::Width::Auto)
+			content_area.x = ResolveValue(computed.width, containing_block.x);
+		else
+			auto_width = true;
+
+		if (computed.height.type != Style::Height::Auto)
+			content_area.y = ResolveValue(computed.height, containing_block.y);
+		else
+			auto_height = true;
+
+		// If one of the dimensions is 'auto' then we need to scale it such that the original ratio is preserved.
+		if (auto_width && !auto_height)
+			content_area.x = (content_area.y / original_content_area.y) * original_content_area.x;
+		else if (auto_height && !auto_width)
+			content_area.y = (content_area.x / original_content_area.x) * original_content_area.y;
+
+		// Reduce the width and height to make up for borders and padding.
+		content_area.x -= (box.GetEdge(Box::BORDER, Box::LEFT) +
+						   box.GetEdge(Box::PADDING, Box::LEFT) +
+						   box.GetEdge(Box::BORDER, Box::RIGHT) +
+						   box.GetEdge(Box::PADDING, Box::RIGHT));
+		content_area.y -= (box.GetEdge(Box::BORDER, Box::TOP) +
+						   box.GetEdge(Box::PADDING, Box::TOP) +
+						   box.GetEdge(Box::BORDER, Box::BOTTOM) +
+						   box.GetEdge(Box::PADDING, Box::BOTTOM));
+
+		content_area.x = Math::Max(content_area.x, 0.0f);
+		content_area.y = Math::Max(content_area.y, 0.0f);
+	}
+
+	// If the element is inline, then its calculations are much more straightforward (no worrying about auto margins
+	// and dimensions, etc). All we do is calculate the margins, set the content area and bail.
+	if (inline_element)
+	{
+		if (replaced_element)
+		{
+			content_area.x = ClampWidth(content_area.x, computed, containing_block.x);
+			content_area.y = ClampHeight(content_area.y, computed, containing_block.y);
+		}
+
+		// If the element was not replaced, then we leave its dimension as unsized (-1, -1) and ignore the width and
+		// height properties.
+		box.SetContent(content_area);
+
+		// Evaluate the margins. Any declared as 'auto' will resolve to 0.
+		box.SetEdge(Box::MARGIN, Box::TOP, ResolveValue(computed.margin_top, containing_block.x));
+		box.SetEdge(Box::MARGIN, Box::RIGHT, ResolveValue(computed.margin_right, containing_block.x));
+		box.SetEdge(Box::MARGIN, Box::BOTTOM, ResolveValue(computed.margin_bottom, containing_block.x));
+		box.SetEdge(Box::MARGIN, Box::LEFT, ResolveValue(computed.margin_left, containing_block.x));
+	}
+
+	// The element is block, so we need to run the box through the ringer to potentially evaluate auto margins and
+	// dimensions.
+	else
+	{
+		box.SetContent(content_area);
+		BuildBoxWidth(box, computed, containing_block, element, replaced_element, override_shrink_to_fit_width);
+		BuildBoxHeight(box, computed, containing_block.y);
+	}
+}
+
+// Generates the box for an element placed in a block box.
+void LayoutDetails::BuildBox(Box& box, float& min_height, float& max_height, LayoutBlockBox* containing_box, Element* element, bool inline_element, float override_shrink_to_fit_width)
+{
+	Vector2f containing_block = GetContainingBlock(containing_box);
+	BuildBox(box, containing_block, element, inline_element, override_shrink_to_fit_width);
+
+	float box_height = box.GetSize().y;
+	if (box_height < 0 && element)
+	{
+		auto& computed = element->GetComputedValues();
+		min_height = ResolveValue(computed.min_height, containing_block.y);
+		max_height = (computed.max_height.value < 0.f ? FLT_MAX : ResolveValue(computed.max_height, containing_block.y));
+	}
+	else
+	{
+		min_height = box_height;
+		max_height = box_height;
+	}
+}
+
+// Clamps the width of an element based from its min-width and max-width properties.
+float LayoutDetails::ClampWidth(float width, const ComputedValues& computed, float containing_block_width)
+{
+	float min_width = ResolveValue(computed.min_width, containing_block_width);
+	float max_width = (computed.max_width.value < 0.f ? FLT_MAX : ResolveValue(computed.max_width, containing_block_width));
+
+	return Math::Clamp(width, min_width, max_width);
+}
+
+// Clamps the height of an element based from its min-height and max-height properties.
+float LayoutDetails::ClampHeight(float height, const ComputedValues& computed, float containing_block_height)
+{
+	float min_height = ResolveValue(computed.min_height, containing_block_height);
+	float max_height = (computed.max_height.value < 0.f ? FLT_MAX : ResolveValue(computed.max_height, containing_block_height));
+
+	return Math::Clamp(height, min_height, max_height);
+}
+
+// Returns the fully-resolved, fixed-width and -height containing block from a block box.
+Vector2f LayoutDetails::GetContainingBlock(const LayoutBlockBox* containing_box)
+{
+	Vector2f containing_block;
+
+	containing_block.x = containing_box->GetBox().GetSize(Box::CONTENT).x;
+	if (containing_box->GetElement() != nullptr)
+		containing_block.x -= containing_box->GetElement()->GetElementScroll()->GetScrollbarSize(ElementScroll::VERTICAL);
+
+	while ((containing_block.y = containing_box->GetBox().GetSize(Box::CONTENT).y) < 0)
+	{
+		containing_box = containing_box->GetParent();
+		if (containing_box == nullptr)
+		{
+			RMLUI_ERROR;
+			containing_block.y = 0;
+		}
+	}
+	if (containing_box != nullptr &&
+		containing_box->GetElement() != nullptr)
+		containing_block.y -= containing_box->GetElement()->GetElementScroll()->GetScrollbarSize(ElementScroll::HORIZONTAL);
+
+	containing_block.x = Math::Max(0.0f, containing_block.x);
+	containing_block.y = Math::Max(0.0f, containing_block.y);
+
+	return containing_block;
+}
+
+
+float LayoutDetails::GetShrinkToFitWidth(Element* element, Vector2f containing_block)
+{
+	// First we need to format the element, then we get the shrink-to-fit width based on the largest line or box.
+
+	LayoutBlockBox containing_block_box(nullptr, nullptr);
+	containing_block_box.GetBox().SetContent(containing_block);
+
+	// Here we fix the element's width to its containing block so that any content is wrapped at this width.
+	// We can consider to instead set this to infinity and clamp it to the available width later after formatting,
+	// but right now the formatting procedure doesn't work well with such numbers.
+	LayoutBlockBox* block_context_box = containing_block_box.AddBlockElement(element, containing_block.x);
+
+	// @performance. Some formatting can be simplified, eg. absolute elements do not contribute to the shrink-to-fit width.
+	// Also, children of elements with a fixed width and height don't need to be formatted further.
+	for (int i = 0; i < element->GetNumChildren(); i++)
+	{
+		if (!LayoutEngine::FormatElement(block_context_box, element->GetChild(i)))
+			i = -1;
+	}
+
+	// We only do layouting to get the fit-to-shrink width here, and for this purpose we may get
+	// away with not closing the boxes. This is avoided for performance reasons.
+	//block_context_box->Close();
+
+	return Math::Min(containing_block.x, block_context_box->GetShrinkToFitWidth());
+}
+
+// Builds the block-specific width and horizontal margins of a Box.
+void LayoutDetails::BuildBoxWidth(Box& box, const ComputedValues& computed, Vector2f containing_block, Element* element, bool replaced_element, float override_shrink_to_fit_width)
+{
+	RMLUI_ZoneScoped;
+
+	Vector2f content_area = box.GetSize();
+
+	// Determine if the element has an automatic width, and if not calculate it.
+	bool width_auto;
+	if (content_area.x >= 0)
+	{
+		width_auto = false;
+	}
+	else
+	{
+		if (computed.width.type == Style::Width::Auto)
+		{
+			width_auto = true;
+		}
+		else
+		{
+			width_auto = false;
+			content_area.x = ResolveValue(computed.width, containing_block.x);
+		}
+	}
+
+	// Determine if the element has automatic margins.
+	bool margins_auto[2];
+	int num_auto_margins = 0;
+
+	for (int i = 0; i < 2; ++i)
+	{
+		auto* margin_value = (i == 0 ? &computed.margin_left : &computed.margin_right);
+		if (margin_value->type == Style::Margin::Auto)
+		{
+			margins_auto[i] = true;
+			num_auto_margins++;
+		}
+		else
+		{
+			margins_auto[i] = false;
+			box.SetEdge(Box::MARGIN, i == 0 ? Box::LEFT : Box::RIGHT, ResolveValue(*margin_value, containing_block.x));
+		}
+	}
+
+	// If the width is set to auto, we need to calculate the width
+	if (width_auto)
+	{
+		// Apply the shrink-to-fit algorithm here to find the width of the element.
+		// See CSS 2.1 section 10.3.7 for when this should be applied.
+		const bool shrink_to_fit = !replaced_element &&
+			(
+				(computed.float_ != Style::Float::None) ||
+				((computed.position == Style::Position::Absolute || computed.position == Style::Position::Fixed) && (computed.left.type == Style::Left::Auto || computed.right.type == Style::Right::Auto)) ||
+				(computed.display == Style::Display::InlineBlock)
+			);
+
+		
+		float left = 0.0f, right = 0.0f;
+		// If we are dealing with an absolutely positioned element we need to
+		// consider if the left and right properties are set, since the width can be affected.
+		if (computed.position == Style::Position::Absolute || computed.position == Style::Position::Fixed)
+		{
+			if (computed.left.type != Style::Left::Auto)
+				left = ResolveValue(computed.left, containing_block.x);
+			if (computed.right.type != Style::Right::Auto)
+				right = ResolveValue(computed.right, containing_block.x);
+		}
+
+		if (margins_auto[0])
+			box.SetEdge(Box::MARGIN, Box::LEFT, 0);
+		if (margins_auto[1])
+			box.SetEdge(Box::MARGIN, Box::RIGHT, 0);
+
+		if (shrink_to_fit && override_shrink_to_fit_width < 0)
+		{
+			content_area.x = GetShrinkToFitWidth(element, containing_block);
+			override_shrink_to_fit_width = content_area.x;
+		}
+		else if (shrink_to_fit)
+		{
+			content_area.x = override_shrink_to_fit_width;
+		}
+		else
+		{
+			// We resolve any auto margins to 0 and the width is set to whatever is left of the containing block.
+			content_area.x = containing_block.x - (left +
+				box.GetCumulativeEdge(Box::CONTENT, Box::LEFT) +
+				box.GetCumulativeEdge(Box::CONTENT, Box::RIGHT) +
+				right);
+			content_area.x = Math::Max(0.0f, content_area.x);
+		}
+	}
+	// Otherwise, the margins that are set to auto will pick up the remaining width of the containing block.
+	else if (num_auto_margins > 0)
+	{
+		float margin = (containing_block.x - (box.GetCumulativeEdge(Box::CONTENT, Box::LEFT) +
+												  box.GetCumulativeEdge(Box::CONTENT, Box::RIGHT) +
+												  content_area.x)) / num_auto_margins;
+
+		if (margins_auto[0])
+			box.SetEdge(Box::MARGIN, Box::LEFT, margin);
+		if (margins_auto[1])
+			box.SetEdge(Box::MARGIN, Box::RIGHT, margin);
+	}
+
+	// Clamp the calculated width; if the width is changed by the clamp, then the margins need to be recalculated if
+	// they were set to auto.
+	float clamped_width = ClampWidth(content_area.x, computed, containing_block.x);
+	if (clamped_width != content_area.x)
+	{
+		content_area.x = clamped_width;
+		box.SetContent(content_area);
+
+		if (num_auto_margins > 0)
+		{
+			// Reset the automatic margins.
+			if (margins_auto[0])
+				box.SetEdge(Box::MARGIN, Box::LEFT, 0);
+			if (margins_auto[1])
+				box.SetEdge(Box::MARGIN, Box::RIGHT, 0);
+
+			BuildBoxWidth(box, computed, containing_block, element, replaced_element, override_shrink_to_fit_width);
+		}
+	}
+	else
+		box.SetContent(content_area);
+}
+
+// Builds the block-specific height and vertical margins of a Box.
+void LayoutDetails::BuildBoxHeight(Box& box, const ComputedValues& computed, float containing_block_height)
+{
+	RMLUI_ZoneScoped;
+
+	Vector2f content_area = box.GetSize();
+
+	// Determine if the element has an automatic height, and if not calculate it.
+	bool height_auto;
+	if (content_area.y >= 0)
+	{
+		height_auto = false;
+	}
+	else
+	{
+		if (computed.height.type == Style::Height::Auto)
+		{
+			height_auto = true;
+		}
+		else
+		{
+			height_auto = false;
+			content_area.y = ResolveValue(computed.height, containing_block_height);
+		}
+	}
+
+	// Determine if the element has automatic margins.
+	bool margins_auto[2];
+	int num_auto_margins = 0;
+
+	for (int i = 0; i < 2; ++i)
+	{
+		auto* margin_value = (i == 0 ? &computed.margin_top : &computed.margin_bottom);
+		if (margin_value->type == Style::Margin::Auto)
+		{
+			margins_auto[i] = true;
+			num_auto_margins++;
+		}
+		else
+		{
+			margins_auto[i] = false;
+			box.SetEdge(Box::MARGIN, i == 0 ? Box::TOP : Box::BOTTOM, ResolveValue(*margin_value, containing_block_height));
+		}
+	}
+
+	// If the height is set to auto, we need to calculate the height
+	if (height_auto)
+	{
+		// We resolve any auto margins to 0
+		if (margins_auto[0])
+			box.SetEdge(Box::MARGIN, Box::TOP, 0);
+		if (margins_auto[1])
+			box.SetEdge(Box::MARGIN, Box::BOTTOM, 0);
+
+		// If the height is set to auto for a box in normal flow, the height is set to -1.
+		content_area.y = -1;
+
+		// But if we are dealing with an absolutely positioned element we need to
+		// consider if the top and bottom properties are set, since the height can be affected.
+		if (computed.position == Style::Position::Absolute || computed.position == Style::Position::Fixed)
+		{
+			float top = 0.0f, bottom = 0.0f;
+
+			if (computed.top.type != Style::Top::Auto && computed.bottom.type != Style::Bottom::Auto)
+			{
+				top = ResolveValue(computed.top, containing_block_height );
+				bottom = ResolveValue(computed.bottom, containing_block_height );
+
+				// The height gets resolved to whatever is left of the containing block
+				content_area.y = containing_block_height - (top +
+				                                            box.GetCumulativeEdge(Box::CONTENT, Box::TOP) +
+				                                            box.GetCumulativeEdge(Box::CONTENT, Box::BOTTOM) +
+				                                            bottom);
+				content_area.y = Math::Max(0.0f, content_area.y);
+			}
+		}
+	}
+	// Otherwise, the margins that are set to auto will pick up the remaining width of the containing block.
+	else if (num_auto_margins > 0)
+	{
+		float margin;
+		if (content_area.y >= 0)
+		{
+			margin = (containing_block_height - (box.GetCumulativeEdge(Box::CONTENT, Box::TOP) +
+												 box.GetCumulativeEdge(Box::CONTENT, Box::BOTTOM) +
+												 content_area.y)) / num_auto_margins;
+		}
+		else
+			margin = 0;
+
+		if (margins_auto[0])
+			box.SetEdge(Box::MARGIN, Box::TOP, margin);
+		if (margins_auto[1])
+			box.SetEdge(Box::MARGIN, Box::BOTTOM, margin);
+	}
+
+	if (content_area.y >= 0)
+	{
+		// Clamp the calculated height; if the height is changed by the clamp, then the margins need to be recalculated if
+		// they were set to auto.
+		float clamped_height = ClampHeight(content_area.y, computed, containing_block_height);
+		if (clamped_height != content_area.y)
+		{
+			content_area.y = clamped_height;
+			box.SetContent(content_area);
+
+			if (num_auto_margins > 0)
+			{
+				// Reset the automatic margins.
+				if (margins_auto[0])
+					box.SetEdge(Box::MARGIN, Box::TOP, 0);
+				if (margins_auto[1])
+					box.SetEdge(Box::MARGIN, Box::BOTTOM, 0);
+
+				BuildBoxHeight(box, computed, containing_block_height);
+			}
+
+			return;
+		}
+	}
+
+	box.SetContent(content_area);
+}
+
+} // namespace Rml

+ 101 - 0
Source/Core/LayoutDetails.h

@@ -0,0 +1,101 @@
+/*
+ * This source file is part of RmlUi, the HTML/CSS Interface Middleware
+ *
+ * For the latest information, see http://github.com/mikke89/RmlUi
+ *
+ * Copyright (c) 2008-2010 CodePoint Ltd, Shift Technology Ltd
+ * Copyright (c) 2019 The RmlUi Team, and contributors
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ */
+
+#ifndef RMLUI_CORE_LAYOUTDETAILS_H
+#define RMLUI_CORE_LAYOUTDETAILS_H
+
+#include "LayoutBlockBox.h"
+
+namespace Rml {
+
+class Box;
+
+/**
+	Layout functions for sizing elements.
+	
+	Corresponds to the CSS 2.1 specification, 'Section 10. Visual formatting model details'.
+ */
+
+class LayoutDetails
+{
+public:
+	/// Generates the box for an element.
+	/// @param[out] box The box to be built.
+	/// @param[in] containing_block The dimensions of the content area of the block containing the element.
+	/// @param[in] element The element to build the box for.
+	/// @param[in] inline_element True if the element is placed in an inline context, false if not.
+	/// @param[in] override_shrink_to_fit_width Provide a fixed shrink-to-fit width instead of formatting the element when its properties allow shrinking.
+	static void BuildBox(Box& box, Vector2f containing_block, Element* element, bool inline_element = false, float override_shrink_to_fit_width = -1);
+	/// Generates the box for an element placed in a block box.
+	/// @param[out] box The box to be built.
+	/// @param[out] min_height The minimum height of the element's box.
+	/// @param[out] max_height The maximum height of the element's box.
+	/// @param[in] containing_box The block box containing the element.
+	/// @param[in] element The element to build the box for.
+	/// @param[in] inline_element True if the element is placed in an inline context, false if not.
+	/// @param[in] override_shrink_to_fit_width Provide a fixed shrink-to-fit width instead of formatting the element when its properties allow shrinking.
+	static void BuildBox(Box& box, float& min_height, float& max_height, LayoutBlockBox* containing_box, Element* element, bool inline_element, float override_shrink_to_fit_width = -1);
+
+	/// Clamps the width of an element based from its min-width and max-width properties.
+	/// @param[in] width The width to clamp.
+	/// @param[in] element The element to read the properties from.
+	/// @param[in] containing_block_width The width of the element's containing block.
+	/// @return The clamped width.
+	static float ClampWidth(float width, const ComputedValues& computed, float containing_block_width);
+	/// Clamps the height of an element based from its min-height and max-height properties.
+	/// @param[in] height The height to clamp.
+	/// @param[in] element The element to read the properties from.
+	/// @param[in] containing_block_height The height of the element's containing block.
+	/// @return The clamped height.
+	static float ClampHeight(float height, const ComputedValues& computed, float containing_block_height);
+
+	/// Returns the fully-resolved, fixed-width and -height containing block from a block box.
+	/// @param[in] containing_box The leaf box.
+	/// @return The dimensions of the content area, using the latest fixed dimensions for width and height in the hierarchy.
+	static Vector2f GetContainingBlock(const LayoutBlockBox* containing_box);
+
+private:
+	/// Formats the element and returns the width of its contents.
+	static float GetShrinkToFitWidth(Element* element, Vector2f containing_block);
+
+	/// Builds the block-specific width and horizontal margins of a Box.
+	/// @param[in,out] box The box to generate. The padding and borders must be set on the box already. If the content area is sized, then it will be used instead of the width property.
+	/// @param[in] element The element the box is being generated for.
+	/// @param[in] containing_block_width The width of the containing block.
+	/// @param[in] replaced_element True when the element is a replaced element.
+	/// @param[in] override_shrink_to_fit_width Provide a fixed shrink-to-fit width instead of formatting the element when its properties allow shrinking.
+	static void BuildBoxWidth(Box& box, const ComputedValues& computed, Vector2f containing_block_width, Element* element, bool replaced_element, float override_shrink_to_fit_width = -1);
+	/// Builds the block-specific height and vertical margins of a Box.
+	/// @param[in,out] box The box to generate. The padding and borders must be set on the box already. If the content area is sized, then it will be used instead of the height property.
+	/// @param[in] element The element the box is being generated for.
+	/// @param[in] containing_block_height The height of the containing block.
+	static void BuildBoxHeight(Box& box, const ComputedValues& computed, float containing_block_height);
+};
+
+} // namespace Rml
+#endif

+ 32 - 482
Source/Core/LayoutEngine.cpp

@@ -27,17 +27,13 @@
  */
 
 #include "LayoutEngine.h"
-#include "../../Include/RmlUi/Core/Math.h"
-#include "Pool.h"
 #include "LayoutBlockBoxSpace.h"
+#include "LayoutDetails.h"
 #include "LayoutInlineBoxText.h"
+#include "Pool.h"
 #include "../../Include/RmlUi/Core/Element.h"
-#include "../../Include/RmlUi/Core/ElementScroll.h"
-#include "../../Include/RmlUi/Core/ElementText.h"
-#include "../../Include/RmlUi/Core/Property.h"
 #include "../../Include/RmlUi/Core/Profiling.h"
 #include "../../Include/RmlUi/Core/Types.h"
-#include <float.h>
 #include <cstddef>
 
 namespace Rml {
@@ -52,18 +48,8 @@ struct LayoutChunk
 
 static Pool< LayoutChunk > layout_chunk_pool(200, true);
 
-LayoutEngine::LayoutEngine()
-{
-	block_box = nullptr;
-	block_context_box = nullptr;
-}
-
-LayoutEngine::~LayoutEngine()
-{
-}
-
 // Formats the contents for a root-level element (usually a document or floating element).
-bool LayoutEngine::FormatElement(Element* element, const Vector2f& containing_block, bool shrink_to_fit)
+bool LayoutEngine::FormatElement(Element* element, Vector2f containing_block)
 {
 #ifdef RMLUI_ENABLE_PROFILING
 	RMLUI_ZoneScopedC(0xB22222);
@@ -71,193 +57,28 @@ bool LayoutEngine::FormatElement(Element* element, const Vector2f& containing_bl
 	RMLUI_ZoneName(name.c_str(), name.size());
 #endif
 
-	block_box = new LayoutBlockBox(this, nullptr, nullptr);
-	block_box->GetBox().SetContent(containing_block);
+	LayoutBlockBox containing_block_box(nullptr, nullptr);
+	containing_block_box.GetBox().SetContent(containing_block);
 
-	block_context_box = block_box->AddBlockElement(element);
+	LayoutBlockBox* block_context_box = containing_block_box.AddBlockElement(element);
 
-	for (int i = 0; i < element->GetNumChildren(); i++)
+	for (int layout_iteration = 0; layout_iteration < 2; layout_iteration++)
 	{
-		if (!FormatElement(element->GetChild(i)))
-			i = -1;
-	}
-
-	if (shrink_to_fit)
-	{
-		// For inline blocks with 'auto' width, we want to shrink the box back to its inner content width, recreating the LayoutBlockBox.
-		float content_width = block_box->InternalContentWidth();
-
-		if (content_width < containing_block.x)
+		for (int i = 0; i < element->GetNumChildren(); i++)
 		{
-			RMLUI_ZoneScopedNC("shrink_to_fit", 0xB27222);
-
-			Vector2f shrinked_block_size(content_width, containing_block.y);
-			
-			delete block_box;
-			block_box = new LayoutBlockBox(this, nullptr, nullptr);
-			block_box->GetBox().SetContent(shrinked_block_size);
-
-			block_context_box = block_box->AddBlockElement(element);
-
-			for (int i = 0; i < element->GetNumChildren(); i++)
-			{
-				if (!FormatElement(element->GetChild(i)))
-					i = -1;
-			}
+			if (!FormatElement(block_context_box, element->GetChild(i)))
+				i = -1;
 		}
-	}
 
-	block_context_box->Close();
-	block_context_box->CloseAbsoluteElements();
+		if (block_context_box->Close() == LayoutBlockBox::OK)
+			break;
+	}
 
 	element->OnLayout();
 
-	delete block_box;
 	return true;
 }
 
-// Generates the box for an element.
-void LayoutEngine::BuildBox(Box& box, const Vector2f& containing_block, Element* element, bool inline_element)
-{
-	if (element == nullptr)
-	{
-		box.SetContent(containing_block);
-		return;
-	}
-
-	const ComputedValues& computed = element->GetComputedValues();
-
-	// Calculate the padding area.
-	float padding = ResolveValue(computed.padding_top, containing_block.x);
-	box.SetEdge(Box::PADDING, Box::TOP, Math::Max(0.0f, padding));
-	padding = ResolveValue(computed.padding_right, containing_block.x);
-	box.SetEdge(Box::PADDING, Box::RIGHT, Math::Max(0.0f, padding));
-	padding = ResolveValue(computed.padding_bottom, containing_block.x);
-	box.SetEdge(Box::PADDING, Box::BOTTOM, Math::Max(0.0f, padding));
-	padding = ResolveValue(computed.padding_left, containing_block.x);
-	box.SetEdge(Box::PADDING, Box::LEFT, Math::Max(0.0f, padding));
-
-	// Calculate the border area.
-	box.SetEdge(Box::BORDER, Box::TOP, Math::Max(0.0f, computed.border_top_width));
-	box.SetEdge(Box::BORDER, Box::RIGHT, Math::Max(0.0f, computed.border_right_width));
-	box.SetEdge(Box::BORDER, Box::BOTTOM, Math::Max(0.0f, computed.border_bottom_width));
-	box.SetEdge(Box::BORDER, Box::LEFT, Math::Max(0.0f, computed.border_left_width));
-
-	// Calculate the size of the content area.
-	Vector2f content_area(-1, -1);
-	bool replaced_element = false;
-
-	// If the element has intrinsic dimensions, then we use those as the basis for the content area and only adjust
-	// them if a non-auto style has been applied to them.
-	if (element->GetIntrinsicDimensions(content_area))
-	{
-		replaced_element = true;
-
-		Vector2f original_content_area = content_area;
-
-		// The element has resized itself, so we only resize it if a RCSS width or height was set explicitly. A value of
-		// 'auto' (or 'auto-fit', ie, both keywords) means keep (or adjust) the intrinsic dimensions.
-		bool auto_width = false, auto_height = false;
-
-		if (computed.width.type != Style::Width::Auto)
-			content_area.x = ResolveValue(computed.width, containing_block.x);
-		else
-			auto_width = true;
-
-		if (computed.height.type != Style::Height::Auto)
-			content_area.y = ResolveValue(computed.height, containing_block.y);
-		else
-			auto_height = true;
-
-		// If one of the dimensions is 'auto' then we need to scale it such that the original ratio is preserved.
-		if (auto_width && !auto_height)
-			content_area.x = (content_area.y / original_content_area.y) * original_content_area.x;
-		else if (auto_height && !auto_width)
-			content_area.y = (content_area.x / original_content_area.x) * original_content_area.y;
-
-		// Reduce the width and height to make up for borders and padding.
-		content_area.x -= (box.GetEdge(Box::BORDER, Box::LEFT) +
-						   box.GetEdge(Box::PADDING, Box::LEFT) +
-						   box.GetEdge(Box::BORDER, Box::RIGHT) +
-						   box.GetEdge(Box::PADDING, Box::RIGHT));
-		content_area.y -= (box.GetEdge(Box::BORDER, Box::TOP) +
-						   box.GetEdge(Box::PADDING, Box::TOP) +
-						   box.GetEdge(Box::BORDER, Box::BOTTOM) +
-						   box.GetEdge(Box::PADDING, Box::BOTTOM));
-
-		content_area.x = Math::Max(content_area.x, 0.0f);
-		content_area.y = Math::Max(content_area.y, 0.0f);
-	}
-
-	// If the element is inline, then its calculations are much more straightforward (no worrying about auto margins
-	// and dimensions, etc). All we do is calculate the margins, set the content area and bail.
-	if (inline_element)
-	{
-		if (replaced_element)
-		{
-			content_area.x = ClampWidth(content_area.x, computed, containing_block.x);
-			content_area.y = ClampHeight(content_area.y, computed, containing_block.y);
-		}
-
-		// If the element was not replaced, then we leave its dimension as unsized (-1, -1) and ignore the width and
-		// height properties.
-		box.SetContent(content_area);
-
-		// Evaluate the margins. Any declared as 'auto' will resolve to 0.
-		box.SetEdge(Box::MARGIN, Box::TOP, ResolveValue(computed.margin_top, containing_block.x));
-		box.SetEdge(Box::MARGIN, Box::RIGHT, ResolveValue(computed.margin_right, containing_block.x));
-		box.SetEdge(Box::MARGIN, Box::BOTTOM, ResolveValue(computed.margin_bottom, containing_block.x));
-		box.SetEdge(Box::MARGIN, Box::LEFT, ResolveValue(computed.margin_left, containing_block.x));
-	}
-
-	// The element is block, so we need to run the box through the ringer to potentially evaluate auto margins and
-	// dimensions.
-	else
-	{
-		box.SetContent(content_area);
-		BuildBoxWidth(box, computed, containing_block.x);
-		BuildBoxHeight(box, computed, containing_block.y);
-	}
-}
-
-// Generates the box for an element placed in a block box.
-void LayoutEngine::BuildBox(Box& box, float& min_height, float& max_height, LayoutBlockBox* containing_box, Element* element, bool inline_element)
-{
-	Vector2f containing_block = GetContainingBlock(containing_box);
-	BuildBox(box, containing_block, element, inline_element);
-
-	float box_height = box.GetSize().y;
-	if (box_height < 0)
-	{
-		auto& computed = element->GetComputedValues();
-		min_height = ResolveValue(computed.min_height, containing_block.y);
-		max_height = (computed.max_height.value < 0.f ? FLT_MAX : ResolveValue(computed.max_height, containing_block.y));
-	}
-	else
-	{
-		min_height = box_height;
-		max_height = box_height;
-	}
-}
-
-// Clamps the width of an element based from its min-width and max-width properties.
-float LayoutEngine::ClampWidth(float width, const ComputedValues& computed, float containing_block_width)
-{
-	float min_width = ResolveValue(computed.min_width, containing_block_width);
-	float max_width = (computed.max_width.value < 0.f ? FLT_MAX : ResolveValue(computed.max_width, containing_block_width));
-
-	return Math::Clamp(width, min_width, max_width);
-}
-
-// Clamps the height of an element based from its min-height and max-height properties.
-float LayoutEngine::ClampHeight(float height, const ComputedValues& computed, float containing_block_height)
-{
-	float min_height = ResolveValue(computed.min_height, containing_block_height);
-	float max_height = (computed.max_height.value < 0.f ? FLT_MAX : ResolveValue(computed.max_height, containing_block_height));
-
-	return Math::Clamp(height, min_height, max_height);
-}
-
 void* LayoutEngine::AllocateLayoutChunk(size_t size)
 {
 	RMLUI_ASSERT(size <= LayoutChunk::size);
@@ -272,7 +93,7 @@ void LayoutEngine::DeallocateLayoutChunk(void* chunk)
 }
 
 // Positions a single element and its children within this layout.
-bool LayoutEngine::FormatElement(Element* element)
+bool LayoutEngine::FormatElement(LayoutBlockBox* block_context_box, Element* element)
 {
 #ifdef RMLUI_ENABLE_PROFILING
 	RMLUI_ZoneScoped;
@@ -283,7 +104,7 @@ bool LayoutEngine::FormatElement(Element* element)
 	auto& computed = element->GetComputedValues();
 
 	// Check if we have to do any special formatting for any elements that don't fit into the standard layout scheme.
-	if (FormatElementSpecial(element))
+	if (FormatElementSpecial(block_context_box, element))
 		return true;
 
 	// Fetch the display property, and don't lay this element out if it is set to a display type of none.
@@ -300,22 +121,18 @@ bool LayoutEngine::FormatElement(Element* element)
 	}
 
 	// If the element is floating, we remove it from the flow.
-	Style::Float float_property = element->GetFloat();
-	if (float_property != Style::Float::None)
+	if (computed.float_ != Style::Float::None)
 	{
-		// Format the element as a block element.
-		LayoutEngine layout_engine;
-		layout_engine.FormatElement(element, GetContainingBlock(block_context_box));
-
+		LayoutEngine::FormatElement(element, LayoutDetails::GetContainingBlock(block_context_box));
 		return block_context_box->AddFloatElement(element);
 	}
 
 	// The element is nothing exceptional, so we treat it as a normal block, inline or replaced element.
 	switch (computed.display)
 	{
-		case Style::Display::Block:       return FormatElementBlock(element); break;
-		case Style::Display::Inline:      return FormatElementInline(element); break;
-		case Style::Display::InlineBlock: return FormatElementReplaced(element); break;
+		case Style::Display::Block:       return FormatElementBlock(block_context_box, element); break;
+		case Style::Display::Inline:      return FormatElementInline(block_context_box, element); break;
+		case Style::Display::InlineBlock: return FormatElementInlineBlock(block_context_box, element); break;
 		default: RMLUI_ERROR;
 	}
 
@@ -323,7 +140,7 @@ bool LayoutEngine::FormatElement(Element* element)
 }
 
 // Formats and positions an element as a block element.
-bool LayoutEngine::FormatElementBlock(Element* element)
+bool LayoutEngine::FormatElementBlock(LayoutBlockBox* block_context_box, Element* element)
 {
 	RMLUI_ZoneScopedC(0x2F4F4F);
 
@@ -331,27 +148,24 @@ bool LayoutEngine::FormatElementBlock(Element* element)
 	if (new_block_context_box == nullptr)
 		return false;
 
-	block_context_box = new_block_context_box;
-
 	// Format the element's children.
 	for (int i = 0; i < element->GetNumChildren(); i++)
 	{
-		if (!FormatElement(element->GetChild(i)))
+		if (!FormatElement(new_block_context_box, element->GetChild(i)))
 			i = -1;
 	}
 
 	// Close the block box, and check the return code; we may have overflowed either this element or our parent.
-	new_block_context_box = block_context_box->GetParent();
-	switch (block_context_box->Close())
+	switch (new_block_context_box->Close())
 	{
 		// We need to reformat ourself; format all of our children again and close the box. No need to check for error
 		// codes, as we already have our vertical slider bar.
 		case LayoutBlockBox::LAYOUT_SELF:
 		{
 			for (int i = 0; i < element->GetNumChildren(); i++)
-				FormatElement(element->GetChild(i));
+				FormatElement(new_block_context_box, element->GetChild(i));
 
-			if (block_context_box->Close() == LayoutBlockBox::OK)
+			if (new_block_context_box->Close() == LayoutBlockBox::OK)
 			{
 				element->OnLayout();
 				break;
@@ -361,7 +175,6 @@ bool LayoutEngine::FormatElementBlock(Element* element)
 		// We caused our parent to add a vertical scrollbar; bail out!
 		case LayoutBlockBox::LAYOUT_PARENT:
 		{
-			block_context_box = new_block_context_box;
 			return false;
 		}
 		break;
@@ -370,44 +183,40 @@ bool LayoutEngine::FormatElementBlock(Element* element)
 			element->OnLayout();
 	}
 
-	block_context_box = new_block_context_box;
 	return true;
 }
 
 // Formats and positions an element as an inline element.
-bool LayoutEngine::FormatElementInline(Element* element)
+bool LayoutEngine::FormatElementInline(LayoutBlockBox* block_context_box, Element* element)
 {
 	RMLUI_ZoneScopedC(0x3F6F6F);
 
 	Box box;
 	float min_height, max_height;
-	BuildBox(box, min_height, max_height, block_context_box, element, true);
+	LayoutDetails::BuildBox(box, min_height, max_height, block_context_box, element, true);
 	LayoutInlineBox* inline_box = block_context_box->AddInlineElement(element, box);
 
 	// Format the element's children.
 	for (int i = 0; i < element->GetNumChildren(); i++)
 	{
-		if (!FormatElement(element->GetChild(i)))
+		if (!FormatElement(block_context_box, element->GetChild(i)))
 			return false;
 	}
 
 	inline_box->Close();
-//	element->OnLayout();
 
 	return true;
 }
 
 // Positions an element as a sized inline element, formatting its internal hierarchy as a block element.
-bool LayoutEngine::FormatElementReplaced(Element* element)
+bool LayoutEngine::FormatElementInlineBlock(LayoutBlockBox* block_context_box, Element* element)
 {
 	RMLUI_ZoneScopedC(0x1F2F2F);
 
 	// Format the element separately as a block element, then position it inside our own layout as an inline element.
-	Vector2f containing_block_size = GetContainingBlock(block_context_box);
+	Vector2f containing_block_size = LayoutDetails::GetContainingBlock(block_context_box);
 
-	LayoutEngine layout_engine;
-	bool shrink_to_width = element->GetComputedValues().width.type == Style::Width::Auto;
-	layout_engine.FormatElement(element, containing_block_size, shrink_to_width);
+	FormatElement(element, containing_block_size);
 
 	block_context_box->AddInlineElement(element, element->GetBox())->Close();
 
@@ -415,7 +224,7 @@ bool LayoutEngine::FormatElementReplaced(Element* element)
 }
 
 // Executes any special formatting for special elements.
-bool LayoutEngine::FormatElementSpecial(Element* element)
+bool LayoutEngine::FormatElementSpecial(LayoutBlockBox* block_context_box, Element* element)
 {
 	static const String br("br");
 	
@@ -430,263 +239,4 @@ bool LayoutEngine::FormatElementSpecial(Element* element)
 	return false;
 }
 
-// Returns the fully-resolved, fixed-width and -height containing block from a block box.
-Vector2f LayoutEngine::GetContainingBlock(const LayoutBlockBox* containing_box)
-{
-	Vector2f containing_block;
-
-	containing_block.x = containing_box->GetBox().GetSize(Box::CONTENT).x;
-	if (containing_box->GetElement() != nullptr)
-		containing_block.x -= containing_box->GetElement()->GetElementScroll()->GetScrollbarSize(ElementScroll::VERTICAL);
-
-	while ((containing_block.y = containing_box->GetBox().GetSize(Box::CONTENT).y) < 0)
-	{
-		containing_box = containing_box->GetParent();
-		if (containing_box == nullptr)
-		{
-			RMLUI_ERROR;
-			containing_block.y = 0;
-		}
-	}
-	if (containing_box != nullptr &&
-		containing_box->GetElement() != nullptr)
-		containing_block.y -= containing_box->GetElement()->GetElementScroll()->GetScrollbarSize(ElementScroll::HORIZONTAL);
-
-	containing_block.x = Math::Max(0.0f, containing_block.x);
-	containing_block.y = Math::Max(0.0f, containing_block.y);
-
-	return containing_block;
-}
-
-// Builds the block-specific width and horizontal margins of a Box.
-void LayoutEngine::BuildBoxWidth(Box& box, const ComputedValues& computed, float containing_block_width)
-{
-	RMLUI_ZoneScoped;
-
-	Vector2f content_area = box.GetSize();
-
-	// Determine if the element has an automatic width, and if not calculate it.
-	bool width_auto;
-	if (content_area.x >= 0)
-	{
-		width_auto = false;
-	}
-	else
-	{
-		if (computed.width.type == Style::Width::Auto)
-		{
-			width_auto = true;
-		}
-		else
-		{
-			width_auto = false;
-			content_area.x = ResolveValue(computed.width, containing_block_width);
-		}
-	}
-
-	// Determine if the element has automatic margins.
-	bool margins_auto[2];
-	int num_auto_margins = 0;
-
-	for (int i = 0; i < 2; ++i)
-	{
-		auto* margin_value = (i == 0 ? &computed.margin_left : &computed.margin_right);
-		if (margin_value->type == Style::Margin::Auto)
-		{
-			margins_auto[i] = true;
-			num_auto_margins++;
-		}
-		else
-		{
-			margins_auto[i] = false;
-			box.SetEdge(Box::MARGIN, i == 0 ? Box::LEFT : Box::RIGHT, ResolveValue(*margin_value, containing_block_width));
-		}
-	}
-
-	// If the width is set to auto, we need to calculate the width
-	if (width_auto)
-	{
-		float left = 0.0f, right = 0.0f;
-		// If we are dealing with an absolutely positioned element we need to
-		// consider if the left and right properties are set, since the width can be affected.
-		if (computed.position == Style::Position::Absolute || computed.position == Style::Position::Fixed)
-		{
-			if (computed.left.type != Style::Left::Auto)
-				left = ResolveValue(computed.left, containing_block_width );
-			if (computed.right.type != Style::Right::Auto)
-				right = ResolveValue(computed.right, containing_block_width);
-		}
-
-		// We resolve any auto margins to 0 and the width is set to whatever is left of the containing block.
-		if (margins_auto[0])
-			box.SetEdge(Box::MARGIN, Box::LEFT, 0);
-		if (margins_auto[1])
-			box.SetEdge(Box::MARGIN, Box::RIGHT, 0);
-
-		content_area.x = containing_block_width - (left +
-		                                           box.GetCumulativeEdge(Box::CONTENT, Box::LEFT) +
-		                                           box.GetCumulativeEdge(Box::CONTENT, Box::RIGHT) +
-		                                           right);
-		content_area.x = Math::Max(0.0f, content_area.x);
-	}
-	// Otherwise, the margins that are set to auto will pick up the remaining width of the containing block.
-	else if (num_auto_margins > 0)
-	{
-		float margin = (containing_block_width - (box.GetCumulativeEdge(Box::CONTENT, Box::LEFT) +
-												  box.GetCumulativeEdge(Box::CONTENT, Box::RIGHT) +
-												  content_area.x)) / num_auto_margins;
-
-		if (margins_auto[0])
-			box.SetEdge(Box::MARGIN, Box::LEFT, margin);
-		if (margins_auto[1])
-			box.SetEdge(Box::MARGIN, Box::RIGHT, margin);
-	}
-
-	// Clamp the calculated width; if the width is changed by the clamp, then the margins need to be recalculated if
-	// they were set to auto.
-	float clamped_width = ClampWidth(content_area.x, computed, containing_block_width);
-	if (clamped_width != content_area.x)
-	{
-		content_area.x = clamped_width;
-		box.SetContent(content_area);
-
-		if (num_auto_margins > 0)
-		{
-			// Reset the automatic margins.
-			if (margins_auto[0])
-				box.SetEdge(Box::MARGIN, Box::LEFT, 0);
-			if (margins_auto[1])
-				box.SetEdge(Box::MARGIN, Box::RIGHT, 0);
-
-			BuildBoxWidth(box, computed, containing_block_width);
-		}
-	}
-	else
-		box.SetContent(content_area);
-}
-
-// Builds the block-specific height and vertical margins of a Box.
-void LayoutEngine::BuildBoxHeight(Box& box, const ComputedValues& computed, float containing_block_height)
-{
-	RMLUI_ZoneScoped;
-
-	Vector2f content_area = box.GetSize();
-
-	// Determine if the element has an automatic height, and if not calculate it.
-	bool height_auto;
-	if (content_area.y >= 0)
-	{
-		height_auto = false;
-	}
-	else
-	{
-		if (computed.height.type == Style::Height::Auto)
-		{
-			height_auto = true;
-		}
-		else
-		{
-			height_auto = false;
-			content_area.y = ResolveValue(computed.height, containing_block_height);
-		}
-	}
-
-	// Determine if the element has automatic margins.
-	bool margins_auto[2];
-	int num_auto_margins = 0;
-
-	for (int i = 0; i < 2; ++i)
-	{
-		auto* margin_value = (i == 0 ? &computed.margin_top : &computed.margin_bottom);
-		if (margin_value->type == Style::Margin::Auto)
-		{
-			margins_auto[i] = true;
-			num_auto_margins++;
-		}
-		else
-		{
-			margins_auto[i] = false;
-			box.SetEdge(Box::MARGIN, i == 0 ? Box::TOP : Box::BOTTOM, ResolveValue(*margin_value, containing_block_height));
-		}
-	}
-
-	// If the height is set to auto, we need to calculate the height
-	if (height_auto)
-	{
-		// We resolve any auto margins to 0
-		if (margins_auto[0])
-			box.SetEdge(Box::MARGIN, Box::TOP, 0);
-		if (margins_auto[1])
-			box.SetEdge(Box::MARGIN, Box::BOTTOM, 0);
-
-		// If the height is set to auto for a box in normal flow, the height is set to -1.
-		content_area.y = -1;
-
-		// But if we are dealing with an absolutely positioned element we need to
-		// consider if the top and bottom properties are set, since the height can be affected.
-		if (computed.position == Style::Position::Absolute || computed.position == Style::Position::Fixed)
-		{
-			float top = 0.0f, bottom = 0.0f;
-
-			if (computed.top.type != Style::Top::Auto && computed.bottom.type != Style::Bottom::Auto)
-			{
-				top = ResolveValue(computed.top, containing_block_height );
-				bottom = ResolveValue(computed.bottom, containing_block_height );
-
-				// The height gets resolved to whatever is left of the containing block
-				content_area.y = containing_block_height - (top +
-				                                            box.GetCumulativeEdge(Box::CONTENT, Box::TOP) +
-				                                            box.GetCumulativeEdge(Box::CONTENT, Box::BOTTOM) +
-				                                            bottom);
-				content_area.y = Math::Max(0.0f, content_area.y);
-			}
-		}
-	}
-	// Otherwise, the margins that are set to auto will pick up the remaining width of the containing block.
-	else if (num_auto_margins > 0)
-	{
-		float margin;
-		if (content_area.y >= 0)
-		{
-			margin = (containing_block_height - (box.GetCumulativeEdge(Box::CONTENT, Box::TOP) +
-												 box.GetCumulativeEdge(Box::CONTENT, Box::BOTTOM) +
-												 content_area.y)) / num_auto_margins;
-		}
-		else
-			margin = 0;
-
-		if (margins_auto[0])
-			box.SetEdge(Box::MARGIN, Box::TOP, margin);
-		if (margins_auto[1])
-			box.SetEdge(Box::MARGIN, Box::BOTTOM, margin);
-	}
-
-	if (content_area.y >= 0)
-	{
-		// Clamp the calculated height; if the height is changed by the clamp, then the margins need to be recalculated if
-		// they were set to auto.
-		float clamped_height = ClampHeight(content_area.y, computed, containing_block_height);
-		if (clamped_height != content_area.y)
-		{
-			content_area.y = clamped_height;
-			box.SetContent(content_area);
-
-			if (num_auto_margins > 0)
-			{
-				// Reset the automatic margins.
-				if (margins_auto[0])
-					box.SetEdge(Box::MARGIN, Box::TOP, 0);
-				if (margins_auto[1])
-					box.SetEdge(Box::MARGIN, Box::BOTTOM, 0);
-
-				BuildBoxHeight(box, computed, containing_block_height);
-			}
-
-			return;
-		}
-	}
-
-	box.SetContent(content_area);
-}
-
 } // namespace Rml

+ 16 - 64
Source/Core/LayoutEngine.h

@@ -42,86 +42,38 @@ class Box;
 class LayoutEngine
 {
 public:
-	/// Constructs a new layout engine.
-	LayoutEngine();
-	~LayoutEngine();
-
-	/// Formats the contents for a root-level element (usually a document, floating or replaced element).
+	/// Formats the contents for a root-level element (usually a document, floating or replaced element). Establishes a new block formatting context.
 	/// @param element[in] The element to lay out.
 	/// @param containing_block[in] The size of the containing block.
-	bool FormatElement(Element* element, const Vector2f& containing_block, bool shrink_to_fit = false);
-
-	/// Generates the box for an element.
-	/// @param[out] box The box to be built.
-	/// @param[in] containing_block The dimensions of the content area of the block containing the element.
-	/// @param[in] element The element to build the box for.
-	/// @param[in] inline_element True if the element is placed in an inline context, false if not.
-	static void BuildBox(Box& box, const Vector2f& containing_block, Element* element, bool inline_element = false);
-	/// Generates the box for an element placed in a block box.
-	/// @param[out] box The box to be built.
-	/// @param[out] min_height The minimum height of the element's box.
-	/// @param[out] max_height The maximum height of the element's box.
-	/// @param[in] containing_box The block box containing the element.
-	/// @param[in] element The element to build the box for.
-	/// @param[in] inline_element True if the element is placed in an inline context, false if not.
-	static void BuildBox(Box& box, float& min_height, float& max_height, LayoutBlockBox* containing_box, Element* element, bool inline_element = false);
+	/// @param shrink_to_fit[in] True to shrink the element to the width of its contents.
+	static bool FormatElement(Element* element, Vector2f containing_block);
 
-	/// Clamps the width of an element based from its min-width and max-width properties.
-	/// @param[in] width The width to clamp.
-	/// @param[in] element The element to read the properties from.
-	/// @param[in] containing_block_width The width of the element's containing block.
-	/// @return The clamped width.
-	static float ClampWidth(float width, const ComputedValues& computed, float containing_block_width);
-	/// Clamps the height of an element based from its min-height and max-height properties.
-	/// @param[in] height The height to clamp.
-	/// @param[in] element The element to read the properties from.
-	/// @param[in] containing_block_height The height of the element's containing block.
-	/// @return The clamped height.
-	static float ClampHeight(float height, const ComputedValues& computed, float containing_block_height);
+	/// Positions a single element and its children within a block formatting context.
+	/// @param[in] block_context_box The open block box to layout the element in.
+	/// @param[in] element The element to lay out.
+	static bool FormatElement(LayoutBlockBox* block_context_box, Element* element);
 
 	static void* AllocateLayoutChunk(size_t size);
 	static void DeallocateLayoutChunk(void* chunk);
 
 private:
-	/// Positions a single element and its children within this layout.
-	/// @param[in] element The element to lay out.
-	bool FormatElement(Element* element);
-
 	/// Formats and positions an element as a block element.
+	/// @param[in] block_context_box The open block box to layout the element in.
 	/// @param[in] element The block element.
-	bool FormatElementBlock(Element* element);
+	static bool FormatElementBlock(LayoutBlockBox* block_context_box, Element* element);
 	/// Formats and positions an element as an inline element.
+	/// @param[in] block_context_box The open block box to layout the element in.
 	/// @param[in] element The inline element.
-	bool FormatElementInline(Element* element);
+	static bool FormatElementInline(LayoutBlockBox* block_context_box, Element* element);
 	/// Positions an element as a sized inline element, formatting its internal hierarchy as a block element.
-	/// @param[in] element The replaced element.
-	bool FormatElementReplaced(Element* element);
+	/// @param[in] block_context_box The open block box to layout the element in.
+	/// @param[in] element The inline-block element.
+	static bool FormatElementInlineBlock(LayoutBlockBox* block_context_box, Element* element);
 	/// Executes any special formatting for special elements.
+	/// @param[in] block_context_box The open block box to layout the element in.
 	/// @param[in] element The element to parse.
 	/// @return True if the element was parsed as a special element, false otherwise.
-	bool FormatElementSpecial(Element* element);
-
-	/// Returns the fully-resolved, fixed-width and -height containing block from a block box.
-	/// @param[in] containing_box The leaf box.
-	/// @return The dimensions of the content area, using the latest fixed dimensions for width and height in the hierarchy.
-	static Vector2f GetContainingBlock(const LayoutBlockBox* containing_box);
-
-	/// Builds the block-specific width and horizontal margins of a Box.
-	/// @param[in,out] box The box to generate. The padding and borders must be set on the box already. If the content area is sized, then it will be used instead of the width property.
-	/// @param[in] element The element the box is being generated for.
-	/// @param[in] containing_block_width The width of the containing block.
-	static void BuildBoxWidth(Box& box, const ComputedValues& computed, float containing_block_width);
-	/// Builds the block-specific height and vertical margins of a Box.
-	/// @param[in,out] box The box to generate. The padding and borders must be set on the box already. If the content area is sized, then it will be used instead of the height property.
-	/// @param[in] element The element the box is being generated for.
-	/// @param[in] containing_block_height The height of the containing block.
-	static void BuildBoxHeight(Box& box, const ComputedValues& computed, float containing_block_height);
-
-	// The root block box, representing the document.
-	LayoutBlockBox* block_box;
-
-	// The open block box containing displaying in a block-context.
-	LayoutBlockBox* block_context_box;
+	static bool FormatElementSpecial(LayoutBlockBox* block_context_box, Element* element);
 };
 
 } // namespace Rml

+ 2 - 3
Source/Core/LayoutInlineBox.cpp

@@ -47,8 +47,7 @@ LayoutInlineBox::LayoutInlineBox(Element* _element, const Box& _box) : position(
 
 	width = 0;
 
-	// If this box has intrinsic dimensions, then we set our height to the total height of the element; otherwise, it
-	// is zero height.
+	// If this box has intrinsic dimensions, then we set our height to the total height of the element; otherwise, it is zero height.
 	if (box.GetSize().y > 0)
 	{
 		height = box.GetSize(Box::MARGIN).y;
@@ -146,7 +145,7 @@ bool LayoutInlineBox::IsLastChild() const
 }
 
 // Flows the inline box's content into its parent line.
-LayoutInlineBox* LayoutInlineBox::FlowContent(bool RMLUI_UNUSED_PARAMETER(first_box), float RMLUI_UNUSED_PARAMETER(available_width), float RMLUI_UNUSED_PARAMETER(right_spacing_width))
+UniquePtr<LayoutInlineBox> LayoutInlineBox::FlowContent(bool RMLUI_UNUSED_PARAMETER(first_box), float RMLUI_UNUSED_PARAMETER(available_width), float RMLUI_UNUSED_PARAMETER(right_spacing_width))
 {
 	RMLUI_UNUSED(first_box);
 	RMLUI_UNUSED(available_width);

+ 1 - 1
Source/Core/LayoutInlineBox.h

@@ -71,7 +71,7 @@ public:
 	/// @param[in] available_width The width available for flowing this box's content. This is measured from the left side of this box's content area.
 	/// @param[in] right_spacing_width The width of the spacing that must be left on the right of the element if no overflow occurs. If overflow occurs, then the entire width can be used.
 	/// @return The overflow box containing any content that spilled over from the flow. This must be nullptr if no overflow occured.
-	virtual LayoutInlineBox* FlowContent(bool first_box, float available_width, float right_spacing_width);
+	virtual UniquePtr<LayoutInlineBox> FlowContent(bool first_box, float available_width, float right_spacing_width);
 
 	/// Computes and sets the vertical position of this element, relative to its parent inline box (or block box,
 	/// for an un-nested inline box).

+ 2 - 2
Source/Core/LayoutInlineBoxText.cpp

@@ -58,7 +58,7 @@ bool LayoutInlineBoxText::CanOverflow() const
 }
 
 // Flows the inline box's content into its parent line.
-LayoutInlineBox* LayoutInlineBoxText::FlowContent(bool first_box, float available_width, float right_spacing_width)
+UniquePtr<LayoutInlineBox> LayoutInlineBoxText::FlowContent(bool first_box, float available_width, float right_spacing_width)
 {
 	ElementText* text_element = GetTextElement();
 	RMLUI_ASSERT(text_element != nullptr);
@@ -76,7 +76,7 @@ LayoutInlineBox* LayoutInlineBoxText::FlowContent(bool first_box, float availabl
 	LayoutInlineBox::FlowContent(first_box, available_width, right_spacing_width);
 
 	if (overflow)
-		return new LayoutInlineBoxText(element, line_begin + line_length);
+		return MakeUnique<LayoutInlineBoxText>(element, line_begin + line_length);
 
 	return nullptr;
 }

+ 1 - 1
Source/Core/LayoutInlineBoxText.h

@@ -55,7 +55,7 @@ public:
 	/// @param available_width[in] The width available for flowing this box's content. This is measured from the left side of this box's content area.
 	/// @param right_spacing_width[in] The width of the spacing that must be left on the right of the element if no overflow occurs. If overflow occurs, then the entire width can be used.
 	/// @return The overflow box containing any content that spilled over from the flow. This must be nullptr if no overflow occured.
-	LayoutInlineBox* FlowContent(bool first_box, float available_width, float right_spacing_width) override;
+	UniquePtr<LayoutInlineBox> FlowContent(bool first_box, float available_width, float right_spacing_width) override;
 
 	/// Computes and sets the vertical position of this element, relative to its parent inline box (or block box,
 	/// for an un-nested inline box).

+ 33 - 19
Source/Core/LayoutLineBox.cpp

@@ -58,12 +58,10 @@ LayoutLineBox::LayoutLineBox(LayoutBlockBox* _parent) : position(-1, -1), dimens
 
 LayoutLineBox::~LayoutLineBox()
 {
-	for (size_t i = 0; i < inline_boxes.size(); i++)
-		delete inline_boxes[i];
 }
 
 // Closes the line box, positioning all inline elements within it.
-LayoutInlineBox* LayoutLineBox::Close(LayoutInlineBox* overflow)
+LayoutInlineBox* LayoutLineBox::Close(UniquePtr<LayoutInlineBox> overflow)
 {
 	RMLUI_ZoneScoped;
 
@@ -90,7 +88,7 @@ LayoutInlineBox* LayoutLineBox::Close(LayoutInlineBox* overflow)
 
 	for (size_t i = 0; i < inline_boxes.size(); ++i)
 	{
-		LayoutInlineBox* inline_box = inline_boxes[i];
+		LayoutInlineBox* inline_box = inline_boxes[i].get();
 
 		// Check if we've got an element aligned to the line box rather than a baseline.
 		if (inline_box->GetVerticalAlignProperty().type == Style::VerticalAlign::Top ||
@@ -121,7 +119,7 @@ LayoutInlineBox* LayoutLineBox::Close(LayoutInlineBox* overflow)
 	// And from that, we can now set the final baseline of each box.
 	for (size_t i = 0; i < inline_boxes.size(); ++i)
 	{
-		LayoutInlineBox* inline_box = inline_boxes[i];
+		LayoutInlineBox* inline_box = inline_boxes[i].get();
 
 		// Check again if this element is aligned to the line box. We don't need to worry about offsetting an element
 		// tied to the top of the line box, as its position will always stay at exactly 0.
@@ -170,7 +168,7 @@ LayoutInlineBox* LayoutLineBox::Close(LayoutInlineBox* overflow)
 		while (open_box != nullptr &&
 			   !inline_box_open)
 		{
-			if (inline_boxes[i] == open_box)
+			if (inline_boxes[i].get() == open_box)
 				inline_box_open = true;
 
 			open_box = open_box->GetParent();
@@ -179,7 +177,7 @@ LayoutInlineBox* LayoutLineBox::Close(LayoutInlineBox* overflow)
 		inline_boxes[i]->SizeElement(inline_box_open);
 	}
 
-	return parent->CloseLineBox(this, overflow, open_inline_box);
+	return parent->CloseLineBox(this, std::move(overflow), open_inline_box);
 }
 
 // Closes one of the line box's inline boxes.
@@ -197,13 +195,13 @@ LayoutInlineBox* LayoutLineBox::AddElement(Element* element, const Box& box)
 	RMLUI_ZoneScoped;
 
 	if (rmlui_dynamic_cast< ElementText* >(element) != nullptr)
-		return AddBox(new LayoutInlineBoxText(element));
+		return AddBox(MakeUnique<LayoutInlineBoxText>(element));
 	else
-		return AddBox(new LayoutInlineBox(element, box));
+		return AddBox(MakeUnique<LayoutInlineBox>(element, box));
 }
 
 // Attempts to add a new inline box to this line.
-LayoutInlineBox* LayoutLineBox::AddBox(LayoutInlineBox* box)
+LayoutInlineBox* LayoutLineBox::AddBox(UniquePtr<LayoutInlineBox> box_ptr)
 {
 	RMLUI_ZoneScoped;
 
@@ -219,7 +217,7 @@ LayoutInlineBox* LayoutLineBox::AddBox(LayoutInlineBox* box)
 	{
 		// Add the new box to the list of boxes in the line box. As this line box has not been placed, we don't have to
 		// check if it can fit yet.
-		AppendBox(box);
+		LayoutInlineBox* box = AppendBox(std::move(box_ptr));
 
 		// If the new box has a physical prescence, then we must place this line once we've figured out how wide it has to
 		// be.
@@ -258,6 +256,8 @@ LayoutInlineBox* LayoutLineBox::AddBox(LayoutInlineBox* box)
 	// This line has already been placed and sized, so we'll check if we can fit this new inline box on the line.
 	else
 	{
+		LayoutInlineBox* box = box_ptr.get();
+
 		// Build up the spacing required on the right side of this element. This consists of the right spacing on the
 		// new element, and the right spacing on all parent element that will close next.
 		right_spacing = GetSpacing(box->GetBox(), Box::RIGHT);
@@ -288,25 +288,29 @@ LayoutInlineBox* LayoutLineBox::AddBox(LayoutInlineBox* box)
 		{
 			// We can't fit the new inline element into this box! So we'll close this line box, and send the inline box
 			// onto the next line.
-			return Close(box);
+			return Close(std::move(box_ptr));
 		}
 		else
 		{
 			// We can fit the new inline element into this box.
-			AppendBox(box);
+			AppendBox(std::move(box_ptr));
 		}
 	}
 
+	float available_width = -1;
+	if (wrap_content)
+		available_width = dimensions.x - (open_inline_box->GetPosition().x + open_inline_box->GetBox().GetPosition(Box::CONTENT).x);
+
 	// Flow the box's content into the line.
-	LayoutInlineBox* overflow_box = open_inline_box->FlowContent(first_box, wrap_content ? dimensions.x - (open_inline_box->GetPosition().x + open_inline_box->GetBox().GetPosition(Box::CONTENT).x) : -1, right_spacing);
+	UniquePtr<LayoutInlineBox> overflow_box = open_inline_box->FlowContent(first_box, available_width, right_spacing);
 	box_cursor += open_inline_box->GetBox().GetSize().x;
 
 	// If our box overflowed, then we'll close this line (as no more content will fit onto it) and tell our block box
 	// to make a new line.
-	if (overflow_box != nullptr)
+	if (overflow_box)
 	{
 		open_inline_box = open_inline_box->GetParent();
-		return Close(overflow_box);
+		return Close(std::move(overflow_box));
 	}
 
 	return open_inline_box;
@@ -325,7 +329,7 @@ void LayoutLineBox::AddChainedBox(LayoutInlineBox* chained_box)
 
 	while (!hierarchy.empty())
 	{
-		AddBox(new LayoutInlineBox(hierarchy.top()));
+		AddBox(MakeUnique<LayoutInlineBox>(hierarchy.top()));
 		hierarchy.pop();
 	}
 }
@@ -365,6 +369,14 @@ float LayoutLineBox::GetBoxCursor() const
 	return box_cursor; 
 }
 
+bool LayoutLineBox::GetBaselineOfLastLine(float& baseline) const
+{
+	if (inline_boxes.empty())
+		return false;
+	baseline = inline_boxes.back()->GetBaseline();
+	return true;
+}
+
 void* LayoutLineBox::operator new(size_t size)
 {
 	return LayoutEngine::AllocateLayoutChunk(size);
@@ -376,9 +388,10 @@ void LayoutLineBox::operator delete(void* chunk)
 }
 
 // Appends an inline box to the end of the line box's list of inline boxes.
-void LayoutLineBox::AppendBox(LayoutInlineBox* box)
+LayoutInlineBox* LayoutLineBox::AppendBox(UniquePtr<LayoutInlineBox> box_ptr)
 {
-	inline_boxes.push_back(box);
+	LayoutInlineBox* box = box_ptr.get();
+	inline_boxes.push_back(std::move(box_ptr));
 
 	box->SetParent(open_inline_box);
 	box->SetLine(this);
@@ -386,6 +399,7 @@ void LayoutLineBox::AppendBox(LayoutInlineBox* box)
 	box_cursor += GetSpacing(box->GetBox(), Box::LEFT);
 
 	open_inline_box = box;
+	return box;
 }
 
 } // namespace Rml

+ 7 - 5
Source/Core/LayoutLineBox.h

@@ -48,7 +48,7 @@ public:
 	/// Closes the line box, positioning all inline elements within it.
 	/// @param overflow[in] The overflow box from a split inline box that caused this line to close. Leave this as nullptr if we closed naturally.
 	/// @return If there was any overflow, this will be the last element generated by the spilling content. Otherwise, this will be nullptr.
-	LayoutInlineBox* Close(LayoutInlineBox* overflow = nullptr);
+	LayoutInlineBox* Close(UniquePtr<LayoutInlineBox> overflow = nullptr);
 
 	/// Closes one of the line box's inline boxes.
 	/// @param inline_box[in] The inline box to close. This should always be the line box's open box.
@@ -65,7 +65,7 @@ public:
 	/// be created. The inline box for the final section of the element will be returned.
 	/// @param box[in] The inline box to be added to the line.
 	/// @return The final inline box.
-	LayoutInlineBox* AddBox(LayoutInlineBox* box);
+	LayoutInlineBox* AddBox(UniquePtr<LayoutInlineBox> box);
 	/// Adds an inline box as a chained hierarchy overflowing to this line. The chain will be extended into
 	/// this line box.
 	/// @param split_box[in] The box overflowed from a previous line.
@@ -90,14 +90,16 @@ public:
 
 	float GetBoxCursor() const;
 
+	bool GetBaselineOfLastLine(float& baseline) const;
+
 	void* operator new(size_t size);
 	void operator delete(void* chunk);
 
 private:
-	/// Appends an inline box to the end of the line box's list of inline boxes.
-	void AppendBox(LayoutInlineBox* box);
+	/// Appends an inline box to the end of the line box's list of inline boxes. Returns a pointer to the appended box.
+	LayoutInlineBox* AppendBox(UniquePtr<LayoutInlineBox> box);
 
-	typedef Vector< LayoutInlineBox* > InlineBoxList;
+	using InlineBoxList = Vector< UniquePtr<LayoutInlineBox> >;
 
 	// The block box containing this line.
 	LayoutBlockBox* parent;

+ 1 - 1
Source/Core/PropertyParserColour.cpp

@@ -42,7 +42,7 @@ PropertyParserColour::PropertyParserColour()
 	html_colours["red"] = Colourb(255, 0, 0);
 	html_colours["orange"] = Colourb(255, 165, 0);
 	html_colours["purple"] = Colourb(128, 0, 128);
-	html_colours["fuschia"] =  Colourb(255, 0, 255);
+	html_colours["fuchsia"] =  Colourb(255, 0, 255);
 	html_colours["green"] =  Colourb(0, 128, 0);
 	html_colours["lime"] =  Colourb(0, 255, 0);
 	html_colours["olive"] =  Colourb(128, 128, 0);

+ 3 - 3
Source/Core/Spritesheet.cpp

@@ -136,14 +136,14 @@ size_t SpritesheetList::NumSprites() const
 
 String SpritesheetList::ToString() const
 {
-	String result = CreateString(100, "#SpriteSheets: %d\n", spritesheet_map.size());
+	String result = CreateString(100, "#SpriteSheets: %zu\n", spritesheet_map.size());
 
 	for (auto& sheet : spritesheet_map)
 	{
-		result += CreateString(100, "  Sheet '%s'.   #Sprites %d.\n", sheet.first.c_str(), sheet.second->sprite_names.size());
+		result += CreateString(100, "  Sheet '%s'.   #Sprites %zu.\n", sheet.first.c_str(), sheet.second->sprite_names.size());
 	}
 
-	result += CreateString(100, "\n#Sprites: %d\n", sprite_map.size());
+	result += CreateString(100, "\n#Sprites: %zu\n", sprite_map.size());
 	for (auto& sprite : sprite_map)
 	{
 		result += CreateString(100, "  In '%s': %s\n", sprite.second.sprite_sheet->name.c_str(), sprite.first.c_str());

+ 10 - 6
Source/Core/StyleSheetParser.cpp

@@ -55,7 +55,10 @@ public:
 */
 class PropertySpecificationParser : public AbstractPropertyParser {
 private:
+	// The dictionary to store the properties in.
 	PropertyDictionary& properties;
+
+	// The specification used to parse the values. Normally the default stylesheet specification, but not for e.g. all at-rules such as decorators.
 	const PropertySpecification& specification;
 
 public:
@@ -499,7 +502,7 @@ bool StyleSheetParser::ParseProperties(PropertyDictionary& parsed_properties, co
 	StreamMemory stream_owner((const byte*)properties.c_str(), properties.size());
 	stream = &stream_owner;
 	PropertySpecificationParser parser(parsed_properties, StyleSheetSpecification::GetPropertySpecification());
-	bool success = ReadProperties(parser, false);
+	bool success = ReadProperties(parser);
 	stream = nullptr;
 	return success;
 }
@@ -525,7 +528,7 @@ StyleSheetNodeListRaw StyleSheetParser::ConstructNodes(StyleSheetNode& root_node
 }
 
 
-bool StyleSheetParser::ReadProperties(AbstractPropertyParser& property_parser, bool require_end_semicolon)
+bool StyleSheetParser::ReadProperties(AbstractPropertyParser& property_parser)
 {
 	String name;
 	String value;
@@ -555,7 +558,7 @@ bool StyleSheetParser::ReadProperties(AbstractPropertyParser& property_parser, b
 				else if (character == '}')
 				{
 					name = StringUtilities::StripWhitespace(name);
-					if (!StringUtilities::StripWhitespace(name).empty())
+					if (!name.empty())
 						Log::Message(Log::LT_WARNING, "End of rule encountered while parsing property declaration '%s' at %s:%d", name.c_str(), stream_file_name.c_str(), line_number);
 					return true;
 				}
@@ -584,8 +587,7 @@ bool StyleSheetParser::ReadProperties(AbstractPropertyParser& property_parser, b
 				}
 				else if (character == '}')
 				{
-					Log::Message(Log::LT_WARNING, "End of rule encountered while parsing property declaration '%s: %s;' in %s: %d.", name.c_str(), value.c_str(), stream_file_name.c_str(), line_number);
-					return true;
+					break;
 				}
 				else
 				{
@@ -605,10 +607,12 @@ bool StyleSheetParser::ReadProperties(AbstractPropertyParser& property_parser, b
 			break;
 		}
 
+		if (character == '}')
+			break;
 		previous_character = character;
 	}
 
-	if (!require_end_semicolon && !name.empty() && !value.empty())
+	if (state == VALUE && !name.empty() && !value.empty())
 	{
 		value = StringUtilities::StripWhitespace(value);
 

+ 3 - 5
Source/Core/StyleSheetParser.h

@@ -84,11 +84,9 @@ private:
 	// Current line number we're parsing.
 	size_t line_number;
 
-	// Parses properties from the parse buffer into the dictionary
-	// @param properties The dictionary to store the properties in
-	// @param property_specification The specification used to parse the values. Normally the default stylesheet specification, but not for e.g. all at-rules such as decorators.
-	// @param require_end_semicolon True to require a ';' character after the final property.
-	bool ReadProperties(AbstractPropertyParser& property_parser, bool require_end_semicolon = true);
+	// Parses properties from the parse buffer.
+	// @param property_parser An abstract parser which specifies how the properties are parsed and stored.
+	bool ReadProperties(AbstractPropertyParser& property_parser);
 
 	// Import properties into the stylesheet node
 	// @param node Node to import into

+ 1 - 0
Source/Core/StyleSheetSpecification.cpp

@@ -366,6 +366,7 @@ void StyleSheetSpecification::RegisterDefaultProperties()
 	RegisterProperty(PropertyId::TextDecoration, "text-decoration", "none", true, false).AddParser("keyword", "none, underline, overline, line-through");
 	RegisterProperty(PropertyId::TextTransform, "text-transform", "none", true, true).AddParser("keyword", "none, capitalize, uppercase, lowercase");
 	RegisterProperty(PropertyId::WhiteSpace, "white-space", "normal", true, true).AddParser("keyword", "normal, pre, nowrap, pre-wrap, pre-line");
+	RegisterProperty(PropertyId::WordBreak, "word-break", "normal", true, true).AddParser("keyword", "normal, break-all, break-word");
 
 	RegisterProperty(PropertyId::Cursor, "cursor", "", true, false).AddParser("string");
 

+ 5 - 5
Source/Core/WidgetScroll.cpp

@@ -28,7 +28,7 @@
 
 #include "WidgetScroll.h"
 #include "Clock.h"
-#include "LayoutEngine.h"
+#include "LayoutDetails.h"
 #include "../../Include/RmlUi/Core/Element.h"
 #include "../../Include/RmlUi/Core/Event.h"
 #include "../../Include/RmlUi/Core/Factory.h"
@@ -217,7 +217,7 @@ void WidgetScroll::FormatElements(const Vector2f& containing_block, bool resize_
 	// Build the box for the containing slider element. As the containing block is not guaranteed to have a defined
 	// height, we must use the width for both axes.
 	Box parent_box;
-	LayoutEngine::BuildBox(parent_box, Vector2f(containing_block.x, containing_block.x), parent);
+	LayoutDetails::BuildBox(parent_box, Vector2f(containing_block.x, containing_block.x), parent);
 	slider_length -= orientation == VERTICAL ? (parent_box.GetCumulativeEdge(Box::CONTENT, Box::TOP) + parent_box.GetCumulativeEdge(Box::CONTENT, Box::BOTTOM)) :
 											   (parent_box.GetCumulativeEdge(Box::CONTENT, Box::LEFT) + parent_box.GetCumulativeEdge(Box::CONTENT, Box::RIGHT));
 
@@ -231,7 +231,7 @@ void WidgetScroll::FormatElements(const Vector2f& containing_block, bool resize_
 
 	// Generate the initial dimensions for the track. It'll need to be cut down to fit the arrows.
 	Box track_box;
-	LayoutEngine::BuildBox(track_box, parent_box.GetSize(), track);
+	LayoutDetails::BuildBox(track_box, parent_box.GetSize(), track);
 	content = track_box.GetSize();
 	content[length_axis] = slider_length -= orientation == VERTICAL ? (track_box.GetCumulativeEdge(Box::CONTENT, Box::TOP) + track_box.GetCumulativeEdge(Box::CONTENT, Box::BOTTOM)) :
 																	  (track_box.GetCumulativeEdge(Box::CONTENT, Box::LEFT) + track_box.GetCumulativeEdge(Box::CONTENT, Box::RIGHT));
@@ -245,7 +245,7 @@ void WidgetScroll::FormatElements(const Vector2f& containing_block, bool resize_
 	for (int i = 0; i < 2; i++)
 	{
 		Box arrow_box;
-		LayoutEngine::BuildBox(arrow_box, parent_box.GetSize(), arrows[i]);
+		LayoutDetails::BuildBox(arrow_box, parent_box.GetSize(), arrows[i]);
 
 		// Clamp the size to (0, 0).
 		Vector2f arrow_size = arrow_box.GetSize();
@@ -297,7 +297,7 @@ void WidgetScroll::FormatElements(const Vector2f& containing_block, bool resize_
 void WidgetScroll::FormatBar(float bar_length)
 {
 	Box bar_box;
-	LayoutEngine::BuildBox(bar_box, parent->GetBox().GetSize(), bar);
+	LayoutDetails::BuildBox(bar_box, parent->GetBox().GetSize(), bar);
 
 	const auto& computed = bar->GetComputedValues();
 

+ 9 - 0
Source/Core/XMLParser.cpp

@@ -79,6 +79,15 @@ XMLNodeHandler* XMLParser::RegisterNodeHandler(const String& _tag, SharedPtr<XML
 	return result;
 }
 
+XMLNodeHandler* XMLParser::GetNodeHandler(const String& tag)
+{
+	auto it = node_handlers.find(tag);
+	if (it != node_handlers.end())
+		return it->second.get();
+	
+	return nullptr;
+}
+
 // Releases all registered node handlers. This is called internally.
 void XMLParser::ReleaseHandlers()
 {

+ 3 - 1
Source/Debugger/CommonSource.h

@@ -47,7 +47,9 @@ em
 h1
 {
 	position: absolute;
-	top: 0px;
+	top: 0;
+	right: 0;
+	left: 0;
 	height: 22dp;
 	padding: 4dp 5dp;
 	color: white;

+ 1 - 0
Source/Debugger/ElementContextHook.cpp

@@ -44,6 +44,7 @@ ElementContextHook::~ElementContextHook()
 void ElementContextHook::Initialise(DebuggerPlugin* _debugger)
 {
 	SetId("rmlui-debug-hook");
+	SetProperty(PropertyId::ZIndex, Property(999'999, Property::NUMBER));
 	debugger = _debugger;
 }
 

+ 9 - 2
Source/Debugger/ElementInfo.cpp

@@ -36,6 +36,7 @@
 #include "../../Include/RmlUi/Core/StyleSheet.h"
 #include "../../Include/RmlUi/Core/StyleSheetSpecification.h"
 #include "../../Include/RmlUi/Core/SystemInterface.h"
+#include "../../Include/RmlUi/Core/PropertyDefinition.h"
 #include "Geometry.h"
 #include "CommonSource.h"
 #include "InfoSource.h"
@@ -603,7 +604,13 @@ void ElementInfo::BuildElementPropertiesRML(String& property_rml, Element* eleme
 		[](const NamedProperty& a, const NamedProperty& b) {
 			if (a.second->source && !b.second->source) return false;
 			if (!a.second->source && b.second->source) return true;
-			return a.second->specificity > b.second->specificity; 
+			if (a.second->specificity < b.second->specificity) return false;
+			if (a.second->specificity > b.second->specificity) return true;
+			if (a.second->definition && !b.second->definition) return false;
+			if (!a.second->definition && b.second->definition) return true;
+			const String& a_name = StyleSheetSpecification::GetPropertyName(a.second->definition->GetId());
+			const String& b_name = StyleSheetSpecification::GetPropertyName(b.second->definition->GetId());
+			return a_name < b_name;
 		}
 	);
 
@@ -627,7 +634,7 @@ void ElementInfo::BuildElementPropertiesRML(String& property_rml, Element* eleme
 				first_iteration = false;
 
 				// Print the rule name header.
-				if(source)
+				if (source)
 				{
 					String str_line_number;
 					TypeConverter<int, String>::Convert(source->line_number, str_line_number);

+ 1 - 1
Source/Debugger/LogSource.h

@@ -30,7 +30,7 @@ static const char* log_rcss = R"RCSS(body
 {
 	width: 400dp;
 	height: 300dp;
-	min-width: 230dp;
+	min-width: 250dp;
 	min-height: 150dp;
 	top: 42dp;
 	left: 20dp;

+ 104 - 0
Tests/CMakeLists.txt

@@ -0,0 +1,104 @@
+#===================================
+# RmlUi tests definitions ==========
+#===================================
+target_compile_definitions(RmlCore PUBLIC RMLUI_TESTS_ENABLED)
+
+
+#===================================
+# Include dependencies =============
+#===================================
+set(DOCTEST_INCLUDE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/Dependencies/doctest)
+add_library(doctest::doctest IMPORTED INTERFACE)
+set_property(TARGET doctest::doctest PROPERTY INTERFACE_INCLUDE_DIRECTORIES "${DOCTEST_INCLUDE_DIR}")
+
+# Include doctest's discovery module
+include(${DOCTEST_INCLUDE_DIR}/cmake/doctest.cmake)
+
+set(NANOBENCH_INCLUDE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/Dependencies/nanobench)
+add_library(nanobench::nanobench IMPORTED INTERFACE)
+set_property(TARGET nanobench::nanobench PROPERTY INTERFACE_INCLUDE_DIRECTORIES "${NANOBENCH_INCLUDE_DIR}")
+
+set(LODEPNG_INCLUDE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/Dependencies/lodepng)
+add_library(lodepng::lodepng IMPORTED INTERFACE)
+set_property(TARGET lodepng::lodepng PROPERTY INTERFACE_INCLUDE_DIRECTORIES "${LODEPNG_INCLUDE_DIR}")
+
+#===================================
+# Common source files ==============
+#===================================
+
+file(GLOB TestsCommon_HDR_FILES ${CMAKE_CURRENT_SOURCE_DIR}/Source/Common/*.h )
+file(GLOB TestsCommon_SRC_FILES ${CMAKE_CURRENT_SOURCE_DIR}/Source/Common/*.cpp )
+
+#===================================
+# Unit Tests =======================
+#===================================
+
+file(GLOB UnitTests_HDR_FILES ${CMAKE_CURRENT_SOURCE_DIR}/Source/UnitTests/*.h )
+file(GLOB UnitTests_SRC_FILES ${CMAKE_CURRENT_SOURCE_DIR}/Source/UnitTests/*.cpp )
+
+add_executable(UnitTests ${UnitTests_HDR_FILES} ${UnitTests_SRC_FILES})
+target_link_libraries(UnitTests RmlCore doctest::doctest)
+set_target_properties(UnitTests PROPERTIES CXX_STANDARD 14)
+
+add_common_target_options(UnitTests)
+
+if(MSVC)
+	target_compile_definitions(UnitTests PUBLIC DOCTEST_CONFIG_USE_STD_HEADERS)
+endif()
+
+doctest_discover_tests(UnitTests)
+
+
+
+#===================================
+# Visual Tests =====================
+#===================================
+
+file(GLOB VisualTests_HDR_FILES ${CMAKE_CURRENT_SOURCE_DIR}/Source/VisualTests/*.h )
+file(GLOB VisualTests_SRC_FILES ${CMAKE_CURRENT_SOURCE_DIR}/Source/VisualTests/*.cpp )
+
+if(WIN32)
+	add_executable(VisualTests WIN32 ${VisualTests_HDR_FILES} ${VisualTests_SRC_FILES})
+else()
+	add_executable(VisualTests ${VisualTests_HDR_FILES} ${VisualTests_SRC_FILES})
+endif()
+
+target_link_libraries(VisualTests RmlCore RmlDebugger lodepng::lodepng ${sample_LIBRARIES})
+set_target_properties(VisualTests PROPERTIES CXX_STANDARD 14)
+
+# Enable compiler warnings
+add_common_target_options(VisualTests)
+
+if(VISUAL_TESTS_RML_DIRECTORIES)
+	target_compile_definitions(VisualTests PRIVATE RMLUI_VISUAL_TESTS_RML_DIRECTORIES="${VISUAL_TESTS_RML_DIRECTORIES}")
+endif()
+if(VISUAL_TESTS_COMPARE_DIRECTORY)
+	target_compile_definitions(VisualTests PRIVATE RMLUI_VISUAL_TESTS_COMPARE_DIRECTORY="${VISUAL_TESTS_COMPARE_DIRECTORY}")
+endif()
+if(VISUAL_TESTS_CAPTURE_DIRECTORY)
+	target_compile_definitions(VisualTests PRIVATE RMLUI_VISUAL_TESTS_CAPTURE_DIRECTORY="${VISUAL_TESTS_CAPTURE_DIRECTORY}")
+endif()
+
+if(MSVC)
+	target_compile_definitions(VisualTests PUBLIC DOCTEST_CONFIG_USE_STD_HEADERS)
+endif()
+
+
+
+#===================================
+# Benchmarks =======================
+#===================================
+
+file(GLOB Benchmarks_HDR_FILES ${CMAKE_CURRENT_SOURCE_DIR}/Source/Benchmarks/*.h )
+file(GLOB Benchmarks_SRC_FILES ${CMAKE_CURRENT_SOURCE_DIR}/Source/Benchmarks/*.cpp )
+
+add_executable(Benchmarks ${Benchmarks_HDR_FILES} ${Benchmarks_SRC_FILES} ${TestsCommon_HDR_FILES} ${TestsCommon_SRC_FILES})
+target_link_libraries(Benchmarks RmlCore RmlDebugger doctest::doctest nanobench::nanobench ${sample_LIBRARIES})
+set_target_properties(Benchmarks PROPERTIES CXX_STANDARD 14)
+
+# Enable compiler warnings
+add_common_target_options(Benchmarks)
+
+if(MSVC)
+	target_compile_definitions(Benchmarks PUBLIC DOCTEST_CONFIG_USE_STD_HEADERS)
+endif()

+ 28 - 0
Tests/Data/VisualTests/LICENSE.txt

@@ -0,0 +1,28 @@
+This software or document includes material copied and modified from the CSS specifications [1], in particular from examples there-in. In addition, parts of the CSS test suites [2] have been modified and included with this software. The material is located in some of the *.rml files in the same directory as this file, and have been tagged with an XML 'link' node with a 'rel=source' attribute pointing to the exact place on the 'w3.org' or 'csswg.org' websites the material has been copied and modified from. The license [3] of this material is restated below. Copyright © 2020 W3C® (MIT, ERCIM, Keio, Beihang).
+
+[1] https://drafts.csswg.org/
+[2] https://www.w3.org/Style/CSS/Test/
+[3] https://www.w3.org/Consortium/Legal/2015/copyright-software-and-document
+
+--------------------------------------------
+
+W3C SOFTWARE AND DOCUMENT NOTICE AND LICENSE
+Status: This license takes effect 13 May, 2015.
+
+This work is being provided by the copyright holders under the following license.
+
+License
+By obtaining and/or copying this work, you (the licensee) agree that you have read, understood, and will comply with the following terms and conditions.
+
+Permission to copy, modify, and distribute this work, with or without modification, for any purpose and without fee or royalty is hereby granted, provided that you include the following on ALL copies of the work or portions thereof, including modifications:
+
+- The full text of this NOTICE in a location viewable to users of the redistributed or derivative work.
+- Any pre-existing intellectual property disclaimers, notices, or terms and conditions. If none exist, the W3C Software and Document Short Notice should be included.
+- Notice of any changes or modifications, through a copyright statement on the new code or document such as "This software or document includes material copied from or derived from [title and URI of the W3C document]. Copyright © [YEAR] W3C® (MIT, ERCIM, Keio, Beihang)."
+
+Disclaimers
+THIS WORK IS PROVIDED "AS IS," AND COPYRIGHT HOLDERS MAKE NO REPRESENTATIONS OR WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO, WARRANTIES OF MERCHANTABILITY OR FITNESS FOR ANY PARTICULAR PURPOSE OR THAT THE USE OF THE SOFTWARE OR DOCUMENT WILL NOT INFRINGE ANY THIRD PARTY PATENTS, COPYRIGHTS, TRADEMARKS OR OTHER RIGHTS.
+
+COPYRIGHT HOLDERS WILL NOT BE LIABLE FOR ANY DIRECT, INDIRECT, SPECIAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF ANY USE OF THE SOFTWARE OR DOCUMENT.
+
+The name and trademarks of copyright holders may NOT be used in advertising or publicity pertaining to the work without specific, written prior permission. Title to copyright in this work will at all times remain with copyright holders.

+ 174 - 0
Tests/Data/VisualTests/acid1.rml

@@ -0,0 +1,174 @@
+<rml>
+<head>
+<title>display/box/float/clear test</title>
+<link type="text/rcss" href="../style.rcss"/>
+<link rel="source" href="https://www.w3.org/Style/CSS/Test/CSS1/current/test5526c.htm" />
+<link rel="help" href="http://www.w3.org/TR/REC-CSS1#clear" />
+<link rel="reference" href="https://www.w3.org/Style/CSS/Test/CSS1/current/sec5526c.gif" />
+<meta name="description" content="This is the ACID1 test. Some minor modifications have been applied such as matching fonts and borders to the RmlUi syntax." />
+<style type="text/css">
+/* RmlUi does not accept styling of the html/rml element, they have been moved into body. */
+body {
+font-size: 10px;
+color: white;
+margin: 1.5em;
+border: .5em black;
+padding: 0;
+width: 48em;
+background-color: white;
+min-height: -1px;
+}
+
+dl {
+margin: 0;
+border: 0;
+padding: .5em;
+}
+
+dt { 
+background-color: rgb(204,0,0);
+margin: 0; 
+padding: 1em;
+width: 10.638%; /* refers to parent element's width of 47em. = 5em or 50px */
+height: 28em;
+border: .5em black;
+float: left;
+}
+
+dd {
+float: right;
+margin: 0 0 0 1em;
+border: 1em black;
+padding: 1em;
+width: 34em;
+height: 27em;
+}
+
+ul {
+margin: 0;
+border: 0;
+padding: 0;
+}
+
+li {
+display: block; /* i.e., suppress marker */
+color: black;
+height: 9em;
+width: 5em;
+margin: 0;
+border: .5em black;
+padding: 1em;
+float: left;
+background-color: #FC0;
+}
+
+#bar {
+background-color: black;
+color: white;
+width: 41.17%; /* = 14em */
+border: 0;
+margin: 0 1em;
+}
+
+#baz {
+margin: 1em 0;
+border: 0;
+padding: 1em;
+width: 10em;
+height: 10em;
+background-color: black;
+color: white;
+}
+
+form { 
+margin: 0;
+display: inline;
+}
+
+p { 
+margin: 0;
+}
+
+form p {
+line-height: 1.9;
+}
+
+blockquote {
+margin: 1em 1em 1em 2em;
+border-width: 1em 1.5em 2em .5em;
+border-color: black;
+padding: 1em 0;
+width: 5em;
+height: 9em;
+float: left;
+background-color: #FC0;
+color: black;
+}
+
+address {
+font-style: normal;
+}
+
+h1 {
+background-color: black;
+color: white;
+float: left;
+margin: 1em 0;
+border: 0;
+padding: 1em;
+width: 10em;
+height: 10em;
+font-weight: normal;
+font-size: 1em;
+}
+</style>
+</head>
+<body>
+	<dl>
+		<dt>
+		 toggle 
+		</dt>
+		<dd>
+		<ul>
+			<li>
+			 the way 
+			</li>
+			<li id="bar">
+			<p>
+			 the world ends 
+			</p>
+			<form action="./" method="get">
+				<p>
+				 bang 
+				<input type="radio" name="foo" value="off"/>
+				</p>
+				<p>
+				 whimper 
+				<input type="radio" name="foo2" value="on"/>
+				</p>
+			</form>
+			</li>
+			<li>
+			 i grow old 
+			</li>
+			<li id="baz">
+			 pluot? 
+			</li>
+		</ul>
+		<blockquote>
+			<address>
+				 bar maids, 
+			</address>
+		</blockquote>
+		<h1>
+			 sing to me, erbarme dich 
+		</h1>
+		</dd>
+	</dl>
+	<p style="color: black; font-size: 1em; line-height: 1.3em; clear: both">
+	 This is a nonsensical document, but syntactically valid HTML 4.0. All 100%-conformant CSS1 agents should be able to render the document elements above this paragraph indistinguishably (to the pixel) from this 
+		<a href="sec5526c.gif">reference rendering,</a>
+	 (except font rasterization and form widgets). All discrepancies should be traceable to CSS1 implementation shortcomings. Once you have finished evaluating this test, you can return to the <a href="sec5526c.htm">parent page</a>. 
+	</p>
+</body>
+</rml>

+ 50 - 0
Tests/Data/VisualTests/css1_clear.rml

@@ -0,0 +1,50 @@
+<rml>
+<head>
+<title>CSS1 Test Suite: 5.5.26 clear</title>
+<link type="text/rcss" href="../style.rcss"/>
+<link rel="source" href="https://www.w3.org/Style/CSS/Test/CSS1/current/sec5526.htm"/>
+<link rel="help" href="http://www.w3.org/TR/REC-CSS1#clear"/>
+<meta name="description" content="Basic CSS1 test of the 'clear' property." />
+<style type="text/css">
+@spritesheet theme 
+{
+	src: /assets/invader.tga;
+	vblank.gif: 68px 158px  9px 30px;
+}
+
+.one {clear: left;}
+.two {clear: right;}
+.three {clear: both;}
+.four {clear: none;}
+</style>
+</head>
+
+<body>
+<img sprite="vblank.gif" height="50" style="float: left" alt="[Image]"/>
+<p>
+This text should be flowing past a tall orange rectangle on the left side of the browser window.
+</p>
+<hr/>
+<img sprite="vblank.gif" height="50" style="float: left" alt="[Image]"/>
+<p class="one">
+This paragraph should appear below the tall orange rectangle above and to the left, and not flow past it. 
+</p>
+<hr/>
+<img sprite="vblank.gif" height="50" style="float: right" alt="[Image]"/>
+<p class="two">
+This paragraph should appear below the tall orange rectangle above and to the right, and not flow past it. 
+</p>
+<hr/>
+<img sprite="vblank.gif" height="50" style="float: left" alt="[Image]"/>
+<img sprite="vblank.gif" height="50" style="float: right" alt="[Image]"/>
+<p class="three">
+This paragraph should appear below the two tall orange rectangles, and not flow between them. 
+</p>
+<img sprite="vblank.gif" height="50" style="float: left" alt="[Image]"/>
+<img sprite="vblank.gif" height="50" style="float: right" alt="[Image]"/>
+<p class="four">
+This paragraph should be between both tall orange rectangles.
+</p>
+<hr/>
+</body>
+</rml>

+ 52 - 0
Tests/Data/VisualTests/float_basic.rml

@@ -0,0 +1,52 @@
+<rml>
+<head>
+    <title>Floats, block formatting contexts</title>
+    <link type="text/rcss" href="../style.rcss"/>
+	<link rel="help" href="https://www.w3.org/TR/CSS21/visufx.html#propdef-overflow" />
+	<meta name="Description" content="Nesting divs should still hide overflow. Elements whose containing block is located above the 'overflow: hidden' element should be visible." />
+	<meta name="See also" content="CSS 2.1 'clipping-' and 'overflow-' tests." />
+	<style>
+		body {
+			background: #ddd;
+			color: #444;
+		}
+		#content {
+			width: 230px;
+			margin: 0 auto;
+		}
+		div.float {
+			float: left;
+			background-color: #bbb;
+			border: 1px #333;
+			width: 100px;
+			height: 100px;
+			margin: 5px;
+		}
+		.red {
+			color: red;
+		}
+	</style>
+</head>
+
+<body>
+<p>This test makes assumptions about the font properties, that is, it depends on the size of the layed-out text.</p>
+<div id="content">
+	<p>Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod.
+	<div class="float">float: left</div>
+	<span class="red">This</span> is the first word after the float and should flow next to the float.</p>
+	<p>This paragraph should flow next to the float.</p>
+
+	<hr/>
+	<p>Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod.
+	<div class="float">float: left</div>
+	<span class="red">This</span> is the first word after the float and should flow next to the float.</p>
+	<p style="clear: left;">This paragraph should be below the float as it clears the float.</p>
+
+	<hr/>
+	<p>Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod.
+	<div class="float">float: left</div>
+	<span class="red">This</span> is the first word after the float and should flow next to the float.</p>
+	<p style="overflow: auto;">This paragraph should establish a new block formatting context. This element's size and position should not overlap with the float, but still be located next to it.</p>
+</div>
+</body>
+</rml>

+ 43 - 0
Tests/Data/VisualTests/float_overflow.rml

@@ -0,0 +1,43 @@
+<rml>
+<head>
+    <title>Floats: overflow</title>
+    <link type="text/rcss" href="../style.rcss"/>
+	<link rel="help" href="http://www.w3.org/TR/CSS21/visuren.html#floats" />
+	<meta name="Description" content="Floating boxes" />
+	<style>
+		body {
+			background: #ddd;
+			color: #444;
+		}
+		#content {
+			width: 200px;
+			margin: 0 auto;
+		}
+		.box {
+			background-color: #cce;
+			border: 5px #77b;
+		}
+		.float {
+			float: left;
+			background-color: #ddda;
+			border: 1px #333;
+			width: 200px;
+			height: 130px;
+			margin: 5px;
+		}
+		
+	</style>
+</head>
+
+<body>
+<div class="box">
+	<div class="float">float: left</div>
+	<p>The float to the left should extend past the background of the containing '.box' element.</p>
+</div>
+<hr/>
+<div class="box" style="overflow:auto">
+	<div class="float">float: left</div>
+	<p>Using 'overflow: auto' on the containing '.box' element. This should establish a new block formatting context (thereby resolving all floats) so that the background wraps around the entire float.</p>
+</div>
+</body>
+</rml>

+ 46 - 0
Tests/Data/VisualTests/inline_block.rml

@@ -0,0 +1,46 @@
+<rml>
+<head>
+    <title>Inline-block</title>
+    <link type="text/rcss" href="../style.rcss"/>
+	<link rel="help" href="https://www.w3.org/TR/2011/REC-CSS2-20110607/visudet.html#inlineblock-width" />
+	<link rel="help" href="https://www.w3.org/TR/2011/REC-CSS2-20110607/visudet.html#leading" />
+	<meta name="Description" content="Elements with 'display: inline-block' should shrink-to-fit when no width is specified. The baseline of the element should be the baseline of its last line." />
+	<style>
+		body {
+			background: #ddd;
+			color: #444;
+		}
+		.float {
+			color: #393;
+		}
+		.left {
+			float: left;
+		}
+		.right {
+			float: right;
+		}
+		.iblock {
+			display: inline-block;
+			color: #33c;
+		}
+		.iblock p {
+			text-align: right;
+		}
+	</style>
+</head>
+
+<body>
+<div>The following text <div class="iblock">is an inline-block</div> and should flow naturally.</div>
+<hr/>
+<div>Left filler text <div class="iblock">meets an inline-block<br/>with a line-break</div> right filler text.</div>
+<hr/>
+<div>Left filler text. <div class="iblock">An inline-block. <p>A paragraph</p> Filler text.</div> Right filler text.</div>
+<hr/>
+<div>Left filler text. <div class="iblock">An inline-block. <div class="float right">Float right</div><div class="float left">Float left</div> Filler text.</div> Right filler text.</div>
+<hr/>
+<div class="iblock">Nesting inline-blocks: 
+	<div class="iblock">A<div class="iblock">B<div class="iblock">C</div></div></div>
+	<div class="iblock">D<div class="iblock">E<div class="iblock">F</div></div></div>
+</div>
+</body>
+</rml>

+ 71 - 0
Tests/Data/VisualTests/overflow_hidden.rml

@@ -0,0 +1,71 @@
+<rml>
+<head>
+    <title>Overflow: hidden</title>
+    <link type="text/rcss" href="../style.rcss"/>
+    <link rel="match" href="reference/overflow_hidden-ref.rml"/>
+	<link rel="help" href="https://www.w3.org/TR/CSS21/visufx.html#propdef-overflow" />
+	<link rel="GitHub issue" href="https://github.com/mikke89/RmlUi/issues/116" />
+	<meta name="Description" content="Nesting divs should still hide overflow. Elements whose containing block is located above the 'overflow: hidden' element should be visible." />
+	<meta name="See also" content="CSS 2.1 'clipping-' and 'overflow-' tests." />
+	<style>
+		body {
+			display: block;
+			background: #ddd;
+			color: #444;
+		}
+		div.overflow {
+			border: 1px black;
+			overflow: hidden;
+			width: 200px;
+			height: 200px;
+		}
+		div.relative {
+			position: relative;
+		}
+		div.wide {
+			width: 300px;
+			height: 20px;
+			color: #f00;
+			text-align: right;
+			background-color: #aaa;
+		}
+		div.absolute {
+			position: absolute;
+			top: 100px;
+			left: 250px;
+			width: 80px;
+			height: 80px;
+		}
+		.green { background-color: #7f7; }
+		.red { background-color: #f77; }
+	</style>
+</head>
+
+<body>
+<p>There should be a green box, no red visible, and the word 'fail' should not appear.</p>
+<div class="overflow">
+	<div>
+		<div class="wide">FAIL</div>
+		LONG_WOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOORD&nbsp;FAIL
+		<!-- This should be visible as the containing block should be 'body' which is above 'div.overflow'. -->
+		<div class="absolute green">Should be visible</div>
+	</div>
+</div>
+<div class="overflow relative">
+	<div>
+		<div class="wide">FAIL</div>
+		LONG_WOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOORD&nbsp;FAIL
+		<!-- This should not be visible as the containing block should now be 'div.overflow' since it uses position: relative. -->
+		<div class="absolute red">Should not be visible</div>
+	</div>
+</div>
+<div class="overflow">
+	<span>
+		LONG_WOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOORD&nbsp;FAIL
+	</span>
+	<div>
+		LONG_WOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOORD&nbsp;FAIL
+	</div>
+</div>
+</body>
+</rml>

+ 42 - 0
Tests/Data/VisualTests/overflow_nested.rml

@@ -0,0 +1,42 @@
+<rml>
+<head>
+    <title>Nested overflow</title>
+    <link type="text/rcss" href="../style.rcss"/>
+	<meta name="Description" content="The deepest element in the tree should catch the overflow." />
+	<style>
+		body {
+			display: block;
+			background: #ddd;
+			color: #444;
+		}
+		div.outer {
+			overflow: auto;
+			width: 200px;
+			height: 200px;
+		}
+		div.overflow {
+			border: 1px black;
+			overflow: auto;
+			width: 150px;
+			height: 150px;
+		}
+		div.wide {
+			width: 300px;
+			height: 20px;
+			text-align: center;
+			border: 1px #0a0;
+			margin: 5px;
+			background-color: #eee;
+		}
+	</style>
+</head>
+
+<body>
+<p>There should should only be one scroll bar visible, inside the black border.</p>
+<div class="outer">
+	<div class="overflow">
+		<div class="wide">Wide element</div>
+	</div>
+</div>
+</body>
+</rml>

+ 29 - 0
Tests/Data/VisualTests/position_01_normal_flow.rml

@@ -0,0 +1,29 @@
+<rml>
+<head>
+	<title>CSS Position: Normal flow</title>
+	<link type="text/rcss" href="../style.rcss"/>
+	<link rel="source" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<link rel="help" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<meta name="description" content="Position demo" />
+	<style>
+		body {
+			font-size: 20px;
+			display: block;
+			background: #ddd;
+			color: #444;
+			line-height: 200%;
+		}
+		#outer { color: red }
+		#inner { color: blue }
+	</style>
+</head>
+<body>
+	<p>
+		Beginning of p contents.
+		<span id="outer"> Start of outer contents.
+		<span id="inner"> Inner contents.</span>
+		End of outer contents.</span>
+		End of p contents.
+	</p>
+</body>
+</rml>

+ 29 - 0
Tests/Data/VisualTests/position_02_relative_positioning.rml

@@ -0,0 +1,29 @@
+<rml>
+<head>
+	<title>CSS Position: Relative positioning</title>
+	<link type="text/rcss" href="../style.rcss"/>
+	<link rel="source" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<link rel="help" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<meta name="description" content='The result is identical to normal flow, except that the "outer" text is shifted 12px upward, without affecting the flow of the "body" or "inner" text.' />
+	<style>
+		body {
+			font-size: 20px;
+			display: block;
+			background: #ddd;
+			color: #444;
+			line-height: 200%;
+		}
+		#outer { position: relative; top: -12px; color: red }
+		#inner { position: relative; top: 12px; color: blue }
+	</style>
+</head>
+<body>
+	<p>
+		Beginning of p contents.
+		<span id="outer"> Start of outer contents.
+		<span id="inner"> Inner contents.</span>
+		End of outer contents.</span>
+		End of p contents.
+	</p>
+</body>
+</rml>

+ 29 - 0
Tests/Data/VisualTests/position_03_floating_a_box.rml

@@ -0,0 +1,29 @@
+<rml>
+<head>
+	<title>CSS Position: Floating a box</title>
+	<link type="text/rcss" href="../style.rcss"/>
+	<link rel="source" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<link rel="help" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<meta name="description" content='The "inner" text lays out in an independent box on the right, causing the remaining "body" and "outer" text to flow around it.' />
+	<style>
+		body {
+			font-size: 20px;
+			display: block;
+			background: #ddd;
+			color: #444;
+			line-height: 200%;
+		}
+		#outer { color: red }
+		#inner { float: right; width: 130px; color: blue }
+	</style>
+</head>
+<body>
+	<p>
+		Beginning of p contents.
+		<span id="outer"> Start of outer contents.
+		<span id="inner"> Inner contents.</span>
+		End of outer contents.</span>
+		End of p contents.
+	</p>
+</body>
+</rml>

+ 30 - 0
Tests/Data/VisualTests/position_04_floating_a_box_sibling.rml

@@ -0,0 +1,30 @@
+<rml>
+<head>
+	<title>CSS Position: Floating a box - Sibling</title>
+	<link type="text/rcss" href="../style.rcss"/>
+	<link rel="source" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<link rel="help" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<meta name="description" content='Identical to the previous example, save that there is now "sibling" text flowing with the "body" and "outer" text.' />
+	<style>
+		body {
+			font-size: 20px;
+			display: block;
+			background: #ddd;
+			color: #444;
+			line-height: 200%;
+		}
+		#inner { float: right; width: 130px; color: blue }
+		#sibling { color: red }
+	</style>
+</head>
+<body>
+	<p>
+		Beginning of p contents.
+		<span id="outer"> Start of outer contents.
+		<span id="inner"> Inner contents.</span>
+		<span id="sibling"> Sibling contents.</span>
+		End of outer contents.</span>
+		End of p contents.
+	</p>
+</body>
+</rml>

+ 30 - 0
Tests/Data/VisualTests/position_05_floating_a_box_clear.rml

@@ -0,0 +1,30 @@
+<rml>
+<head>
+	<title>CSS Position: Floating a box - Clear</title>
+	<link type="text/rcss" href="../style.rcss"/>
+	<link rel="source" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<link rel="help" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<meta name="description" content='Now the "sibling" text moves down to below the "inner" text’s box, leaving blank space behind. The text following the "sibling" text flows after it as normal.' />
+	<style>
+		body {
+			font-size: 20px;
+			display: block;
+			background: #ddd;
+			color: #444;
+			line-height: 200%;
+		}
+		#inner { float: right; width: 130px; color: blue }
+		#sibling { clear: right; color: red }
+	</style>
+</head>
+<body>
+	<p>
+		Beginning of p contents.
+		<span id="outer"> Start of outer contents.
+		<span id="inner"> Inner contents.</span>
+		<span id="sibling"> Sibling contents.</span>
+		End of outer contents.</span>
+		End of p contents.
+	</p>
+</body>
+</rml>

+ 34 - 0
Tests/Data/VisualTests/position_06_absolute_positioning.rml

@@ -0,0 +1,34 @@
+<rml>
+<head>
+	<title>CSS Position: Absolute positioning</title>
+	<link type="text/rcss" href="../style.rcss"/>
+	<link rel="source" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<link rel="help" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<meta name="description" content='All of the text within #outer (the "outer" and "inner" text) moves down to an independent box in the lower right corner. The two halves of "body" text flow together.' />
+	<style>
+		body {
+			font-size: 20px;
+			display: block;
+			background: #ddd;
+			color: #444;
+			line-height: 200%;
+		}
+		#outer {
+			position: absolute;
+			top: 200px; left: 200px;
+			width: 200px;
+			color: red;
+		}
+		#inner { color: blue }
+	</style>
+</head>
+<body>
+	<p>
+		Beginning of p contents.
+		<span id="outer"> Start of outer contents.
+		<span id="inner"> Inner contents.</span>
+		End of outer contents.</span>
+		End of p contents.
+	</p>
+</body>
+</rml>

+ 37 - 0
Tests/Data/VisualTests/position_07_absolute_positioning_relative.rml

@@ -0,0 +1,37 @@
+<rml>
+<head>
+	<title>CSS Position: Absolute positioning - Relative</title>
+	<link type="text/rcss" href="../style.rcss"/>
+	<link rel="source" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<link rel="help" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<meta name="description" content='The "inner" text is positioned in an independent box, relative to the top-left corner of the start of the "outer" text.' />
+	<style>
+		body {
+			font-size: 20px;
+			display: block;
+			background: #ddd;
+			color: #444;
+			line-height: 200%;
+		}
+		#outer {
+			position: relative;
+			color: red
+		}
+		#inner {
+			position: absolute;
+			top: 200px; left: -100px;
+			height: 130px; width: 130px;
+			color: blue;
+		}
+	</style>
+</head>
+<body>
+	<p>
+		Beginning of p contents.
+		<span id="outer"> Start of outer contents.
+		<span id="inner"> Inner contents.</span>
+		End of outer contents.</span>
+		End of p contents.
+	</p>
+</body>
+</rml>

+ 34 - 0
Tests/Data/VisualTests/position_08_absolute_positioning_no_relative.rml

@@ -0,0 +1,34 @@
+<rml>
+<head>
+	<title>CSS Position: Absolute positioning - No relative</title>
+	<link type="text/rcss" href="../style.rcss"/>
+	<link rel="source" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<link rel="help" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<meta name="description" content='Same as before, except now the "inner text" is positioned relative to the top-left corner of the page itself.' />
+	<style>
+		body {
+			font-size: 20px;
+			display: block;
+			background: #ddd;
+			color: #444;
+			line-height: 200%;
+		}
+		#outer { color: red }
+		#inner {
+			position: absolute;
+			top: 200px; left: -100px;
+			height: 130px; width: 130px;
+			color: blue;
+		}
+	</style>
+</head>
+<body>
+	<p>
+		Beginning of p contents.
+		<span id="outer"> Start of outer contents.
+		<span id="inner"> Inner contents.</span>
+		End of outer contents.</span>
+		End of p contents.
+	</p>
+</body>
+</rml>

+ 27 - 0
Tests/Data/VisualTests/position_09_absolute_positioning_change_bars.rml

@@ -0,0 +1,27 @@
+<rml>
+<head>
+	<title>CSS Position: Absolute positioning - Change bars</title>
+	<link type="text/rcss" href="../style.rcss"/>
+	<link rel="source" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<link rel="help" href="https://drafts.csswg.org/css-position-3/#comparison" />
+	<meta name="description" content='The two red hyphens, indicating a change, sit in the left margin of the page on the line containing the word "THIS", regardless of what line that ends up being.' />
+	<style>
+		body {
+			font-size: 20px;
+			display: block;
+			background: #ddd;
+			color: #444;
+			line-height: 200%;
+		}
+		body { padding: 30px; }
+	</style>
+</head>
+<body>
+	<p style="position: relative; margin-right: 10px; left: 10px;">
+	  I used two red hyphens to serve as a change bar. They
+	  will "float" to the left of the line containing THIS
+	  <span style="position: absolute; top: auto; left: -1em; color: red;">--</span>
+	  word.
+	</p>
+</body>
+</rml>

+ 60 - 0
Tests/Data/VisualTests/reference/overflow_hidden-ref.rml

@@ -0,0 +1,60 @@
+<rml>
+<head>
+    <title>Overflow: hidden</title>
+    <link type="text/rcss" href="../../style.rcss"/>
+	<style>
+		body {
+			display: block;
+			background: #ddd;
+			color: #444;
+		}
+		div.overflow {
+			border: 1px black;
+			overflow: hidden;
+			width: 200px;
+			height: 200px;
+		}
+		div.relative {
+			position: relative;
+		}
+		div.wide {
+			width: 300px;
+			height: 20px;
+			color: #f00;
+			text-align: right;
+			background-color: #aaa;
+		}
+		div.absolute {
+			position: absolute;
+			top: 100px;
+			left: 250px;
+			width: 80px;
+			height: 80px;
+		}
+		.green { background-color: #7f7; }
+		.red { background-color: #f77; }
+	</style>
+</head>
+
+<body>
+<p>There should be a green box, no red visible, and the word 'fail' should not appear.</p>
+<div class="overflow">
+	<div class="wide">FAIL</div>
+	LONG_WOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOORD&nbsp;FAIL
+</div>
+<div class="absolute green">Should be visible</div>
+<div class="overflow relative">
+	<div class="wide">FAIL</div>
+	LONG_WOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOORD&nbsp;FAIL
+	<div class="absolute red">Should not be visible</div>
+</div>
+<div class="overflow">
+	<span>
+		LONG_WOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOORD&nbsp;FAIL
+	</span>
+	<div>
+		LONG_WOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOORD&nbsp;FAIL
+	</div>
+</div>
+</body>
+</rml>

+ 58 - 0
Tests/Data/VisualTests/shrink_to_fit_01.rml

@@ -0,0 +1,58 @@
+<rml>
+<head>
+    <title>Shrink-to-fit</title>
+    <link type="text/rcss" href="../style.rcss"/>
+	<link rel="help" href="https://www.w3.org/TR/CSS21/visudet.html#shrink-to-fit-float" />
+	<meta name="Description" content="Floating elements, absolutely positioned elements, and inline-block elements should shrink-to-fit when their width is auto." />
+	<style>
+		body {
+			background: #ddd;
+			color: #444;
+		}
+		body > div {
+			clear: both;
+			position: relative;
+			height: 50px;
+		}
+		body > div > * {
+			border: 2px #888;
+			color: black;
+			background: #ccf;
+		}
+		.absolute {
+			position: absolute;
+		}
+		.float {
+			float: left;
+		}
+		.iblock {
+			display: inline-block;
+		}
+		.right {
+			text-align: right;
+		}
+	</style>
+</head>
+
+<body>
+<p>The following boxes should all appear the same, with the background and border tightly packed around the words 'A BOX'.</p>
+<div>
+	<span>A BOX</span>
+</div>
+<div>
+	<div class="absolute">A BOX</div>
+</div>
+<div>
+	<div class="float">A BOX</div>
+</div>
+<div>
+	<span class="float">A BOX</span>
+</div>
+<div>
+	<div class="iblock">A BOX</div>
+</div>
+<div>
+	<div class="iblock right">A BOX</div>
+</div>
+</body>
+</rml>

+ 97 - 0
Tests/Data/VisualTests/shrink_to_fit_02.rml

@@ -0,0 +1,97 @@
+<rml>
+<head>
+    <title>Shrink-to-fit 2</title>
+    <link type="text/rcss" href="../style.rcss"/>
+	<link rel="help" href="https://www.w3.org/TR/CSS21/visudet.html#shrink-to-fit-float" />
+	<meta name="Description" content="Shrink-to-fit width with differently positioned and floated boxes." />
+	<style>
+		body {
+			background: #ddd;
+			color: #444;
+		}
+		hr {
+			margin-top: 15px;
+			margin-bottom: 15px;
+			background-color: #ccc;
+		}
+		div {
+			border: 2px #444;
+			background: #ccf;
+		}
+		body > div {
+			display: inline-block;
+			position: relative;
+			/*overflow: auto;*/
+		}
+		body > div > div {
+			width: 150px;
+			height: 30px;
+			background: #cfc;
+		}
+		.absolute {
+			position: absolute;
+			top: 20px;
+			left: 50px;
+		}
+		.relative {
+			position: relative;
+			top: 20px;
+			left: 50px;
+		}
+		.float-left {
+			float: left;
+		}
+		.float-right {
+			float: right;
+		}
+	</style>
+</head>
+
+<body>
+<div>
+  Some text
+  <div>Block</div>
+  Text after
+</div>
+<hr/>
+<div>
+  Some text
+  <div class="absolute">Absolute</div>
+  Text after
+</div>
+<hr style="margin-top: 55px;"/>
+<div>
+  Some text
+  <div class="relative">Relative</div>
+  Text after
+</div>
+<hr/>
+<div>
+  Some text
+  <div class="float-left">Float left</div>
+  Text after
+</div>
+<hr/>
+<div>
+  Some text
+  <div class="float-right">Float right</div>
+  Text after
+</div>
+<hr/>
+<div>
+  <div class="float-left">Float left</div>
+  Text after
+</div>
+<hr/>
+<div>
+  <div class="float-right">Float right</div>
+  Text after
+</div>
+<hr/>
+<div>
+  Some text
+  <div class="float-left relative">Float relative</div>
+  Text after
+</div>
+</body>
+</rml>

+ 53 - 0
Tests/Data/VisualTests/word_break.rml

@@ -0,0 +1,53 @@
+<rml>
+<head>
+    <title>Word-break property</title>
+    <link type="text/rcss" href="../style.rcss"/>
+	<link rel="help" href="https://drafts.csswg.org/css-text-3/#word-break-property" />
+	<meta name="Description" content="Word-break property." />
+	<style>
+		body {
+			background: #ddd;
+			color: #444;
+		}
+		h1 {
+			margin-top: 0.5em;
+			font-size: 1.1em;
+		}
+		p {
+			color: #45e;
+		}
+		.box {
+			width: 60px;
+			border: 2px #aaa;
+		}
+		.zero {
+			width: 0px;
+		}
+		.break-all {
+			word-break: break-all;
+		}
+		.break-word {
+			word-break: break-word;
+		}
+		
+	</style>
+</head>
+
+<body>
+	<h1>Fixed-width box</h1>
+	<p>word-break: normal</p>
+	<div class="box">A very veeery veeeeeeeeeeeery long word.</div>
+	<p>word-break: break-all</p>
+	<div class="box break-all">A very veeery veeeeeeeeeeeery long word.</div>
+	<p>word-break: break-word</p>
+	<div class="box break-word">A very veeery veeeeeeeeeeeery long word.</div>
+	<hr/>
+	<h1>Zero-width box</h1>
+	<p>word-break: normal</p>
+	<div class="box zero">A WORD</div>
+	<p>word-break: break-all</p>
+	<div class="box zero break-all">A WORD</div>
+	<p>word-break: break-word</p>
+	<div class="box zero break-word">A WORD</div>
+</body>
+</rml>

+ 96 - 0
Tests/Data/description.rml

@@ -0,0 +1,96 @@
+<rml>
+<head>
+<title>RCSS Test description</title>
+<link type="text/rcss" href="style.rcss"/>
+<style>
+	body {
+		font-family: Delicious;
+		font-weight: normal;
+		font-style: normal;
+		font-size: 17px;
+		color: #444;
+		position: absolute;
+		top: 0; bottom: 0;
+		right: 0;
+		width: 400px;
+		background: #333;
+		color: #ccc;
+		padding: 20px 20px;
+		z-index: 100;
+		tab-index: auto;
+	}
+	h1, h2, h3 { padding-top: 0.8em; }
+	code {
+		display: block;
+		white-space: pre-wrap;
+		font-size: 0.9em;
+		color: #aaa;
+	}
+	h1   { color: white; font-size: 1.3em; }
+	h3   { color: white; font-size: 1.15em; }
+	
+	p.links a { margin: 0 0.7em; }
+	#header {
+		color: #ddb;
+	}
+	#filter {
+		position: absolute;
+		right: 20px;
+		top: 20px;
+		width: 50%;
+		text-align: right;
+	}
+	input.text {
+		font-size: 0.85em;
+		padding: 3px 4px;
+		width: 80%;
+		background-color: #444;
+		color: #ffe;
+		border: 1px #555;
+		line-height: 1.2;
+		height: 1.8em;
+		cursor: text;
+		tab-index: auto;
+	}
+	input.text:focus {
+		border-color: #dda;
+	}
+	#filter_text {
+		color: #ffc;
+	}
+	#content {
+		overflow: hidden auto;	
+	}
+	#goto {
+		margin: 0 20px 10px 20px;
+		color: #ddb;
+	}
+	#bottom {
+		position: absolute;
+		left: 0;
+		bottom: 0;
+		right: 0;
+		text-align: left;
+		word-break: break-all;
+	}
+	#hovertext {
+		margin: 0px 20px 10px 20px;
+		font-size: 0.9em;
+		color: #aaa;
+		min-height: 1.5em;
+	}
+	#hovertext.confirmation {
+		color: #afa;
+	}
+</style>
+</head>
+<body>
+<div id="header"/>
+<div id="filter"><input id="filterinput" type="text" value=""/><br/><span id="filter_text">Filtered 2 of 63.</span></div>
+<div id="content"/>
+<div id="bottom">
+	<div id="goto"/>
+	<div id="hovertext"/>
+</div>
+</body>
+</rml>

+ 96 - 0
Tests/Data/style.rcss

@@ -0,0 +1,96 @@
+dl,dt,dd,ul,li,blockquote,address,h1,h2,h3,h4,h5,h6,p,pre,div,section {
+	display: block;
+}
+
+p, pre { margin: 0.8em 0; }
+pre { white-space: pre; }
+hr {
+	display: block;
+	clear: both;
+	padding: 1px;
+	background-color: #999;
+	margin: 5px 0;
+}
+
+input.radio {
+	width: 10px;
+	height: 10px;
+	border: 1px #666;
+	background: #fff;
+}
+
+body {
+	font-family: Delicious;
+	font-weight: normal;
+	font-style: normal;
+	font-size: 14px;
+	padding: 8px;
+	width: 484px;
+	min-height: 484px;
+	max-height: 100%;
+	color: black;
+	background: #ccc;
+	overflow: auto;
+}
+strong {
+	font-weight: bold;
+}
+em {
+	font-style: italic;
+}
+* {
+	border-color: black;
+}
+a {
+	color: #9ab7ef;
+}
+a:hover {
+	color: #5285e6;
+}
+a:active {
+	color: #4275e0;
+}
+scrollbarvertical
+{
+	width: 16dp;
+}
+scrollbarhorizontal
+{
+	height: 16dp;
+}
+scrollbarvertical slidertrack,
+scrollbarhorizontal slidertrack
+{
+	background: #aaa;
+	border-color: #888;
+}
+scrollbarvertical slidertrack
+{
+	border-left-width: 1px;
+}
+scrollbarhorizontal slidertrack
+{
+	height: 15dp;
+	border-top-width: 1px;
+}
+scrollbarvertical sliderbar,
+scrollbarhorizontal sliderbar
+{
+	background: #ddd;
+	border-color: #888;
+}
+scrollbarvertical sliderbar
+{
+	border-width: 1px 0px;
+	margin-left: 1dp;
+}
+scrollbarhorizontal sliderbar
+{
+	height: 15dp;
+	border-width: 0px 1px;
+	margin-top: 1dp;
+}
+scrollbarcorner
+{
+	background: #888;
+}

+ 31 - 0
Tests/Data/view_source.rml

@@ -0,0 +1,31 @@
+<rml>
+<head>
+<title>RCSS Test description</title>
+<link type="text/rcss" href="style.rcss"/>
+<style>
+	body {
+		font-family: rmlui-debugger-font;
+		font-weight: normal;
+		font-style: normal;
+		font-size: 16px;
+		width: auto;
+		color: #444;
+		position: absolute;
+		top: 0px; right: 0px;
+		bottom: 0px; left: 0px;
+		background-color: #222222f9;
+		color: #ddd;
+		z-index: 200;
+	}
+	#code {
+		display: block;
+		white-space: pre-wrap;
+		font-size: 0.9em;
+		padding: 20px 30px;
+	}
+</style>
+</head>
+<body>
+<div id="code"/>
+</body>
+</rml>

+ 78 - 0
Tests/Data/visual_tests_help.rml

@@ -0,0 +1,78 @@
+<rml>
+<head>
+<title>RCSS Test value</title>
+<link type="text/rcss" href="style.rcss"/>
+<style>
+	body {
+		width: auto;
+		color: #444;
+		position: absolute;
+		top: 0px; right: 0px;
+		bottom: 0px; left: 0px;
+		background-color: #222222f9;
+		color: #ddd;
+		z-index: 300;
+	}
+	#content {
+		display: block;
+		white-space: pre-wrap;
+		font-size: 17px;
+		width: 700px;
+		margin: 0 auto;
+		padding-top: 30px;
+	}
+	p {
+		clear: both;
+		padding-top: 1em;
+	}
+	h1 {
+		font-size: 1.2em;
+		padding: 1em 0;
+		color: #fff;
+		clear: both;
+	}
+	key, value {
+		display: block;
+		line-height: 1.5;
+	}
+	key {
+		width: 40%;
+		float: left;
+		clear: both;
+		text-align: right;
+	}
+	value {
+		width: 52%;
+		float: right;
+		clear: right;
+	}
+</style>
+</head>
+<body>
+<div id="content">
+<h1>Keyboard shortcuts</h1>
+<key>Left / Right</key> <value>Navigate tests</value>
+<key>Home / End</key> <value>Go to first / last test</value>
+<key>Up / Down</key> <value>Navigate test suites</value>
+<key>F5</key> <value>Compare test to previous capture</value>
+<key>Ctrl+Shift+F5</key> <value>Compare all tests</value>
+<key>F7</key> <value>Capture screenshot of test</value>
+<key>Ctrl+Shift+F7</key> <value>Capture all tests</value>
+<key>F8</key> <value>Open debugger</value>
+<key>Ctrl+G</key> <value>Go To test (followed by number, enter)</value>
+<key>Ctrl+F</key> <value>Filter tests</value>
+<key>Ctrl+R</key> <value>Reload test</value>
+<key>Ctrl+S</key> <value>View source of test</value>
+<key>Ctrl+Shift+S</key> <value>View source of reference</value>
+<key>Left Mouse Button on Link</key> <value>Copy link to clipboard (on supported platforms)</value>
+<key>F1</key> <value>Show / hide help</value>
+
+<p>Results of comparisons and captures are written to the log output.</p>
+
+<h1>Directories</h1>
+<key>Test suite(s)</key> <span id="test_directories"><value>C:\</value> <value>E:\</value> <value>D:\</value></span>
+<key>Compare input</key> <value id="compare_input">C:\</value>
+<key>Capture output</key> <value id="capture_output">C:\</value>
+</div>
+</body>
+</rml>

+ 6 - 0
Tests/Dependencies/doctest/cmake/Config.cmake.in

@@ -0,0 +1,6 @@
+if(NOT TARGET doctest::doctest)
+    # Provide path for scripts
+    list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_LIST_DIR}")
+
+    include("${CMAKE_CURRENT_LIST_DIR}/@[email protected]")
+endif()

+ 13 - 0
Tests/Dependencies/doctest/cmake/assemble_single_header.cmake

@@ -0,0 +1,13 @@
+set(doctest_include_folder "${CMAKE_CURRENT_LIST_DIR}/../../doctest/")
+
+file(READ ${doctest_include_folder}/parts/doctest_fwd.h fwd)
+file(READ ${doctest_include_folder}/parts/doctest.cpp impl)
+
+file(WRITE  ${doctest_include_folder}/doctest.h "// ====================================================================== lgtm [cpp/missing-header-guard]\n")
+file(APPEND ${doctest_include_folder}/doctest.h "// == DO NOT MODIFY THIS FILE BY HAND - IT IS AUTO GENERATED BY CMAKE! ==\n")
+file(APPEND ${doctest_include_folder}/doctest.h "// ======================================================================\n")
+file(APPEND ${doctest_include_folder}/doctest.h "${fwd}\n")
+file(APPEND ${doctest_include_folder}/doctest.h "#ifndef DOCTEST_SINGLE_HEADER\n")
+file(APPEND ${doctest_include_folder}/doctest.h "#define DOCTEST_SINGLE_HEADER\n")
+file(APPEND ${doctest_include_folder}/doctest.h "#endif // DOCTEST_SINGLE_HEADER\n")
+file(APPEND ${doctest_include_folder}/doctest.h "\n${impl}")

+ 207 - 0
Tests/Dependencies/doctest/cmake/common.cmake

@@ -0,0 +1,207 @@
+include(CMakeParseArguments)
+
+# cache this for use inside of the function
+set(CURRENT_LIST_DIR_CACHED ${CMAKE_CURRENT_LIST_DIR})
+
+set_property(GLOBAL PROPERTY USE_FOLDERS ON)
+
+enable_testing()
+
+find_package(Threads)
+
+set(DOCTEST_TEST_MODE "COMPARE" CACHE STRING "Test mode - normal/run through valgrind/collect output/compare with output")
+set_property(CACHE DOCTEST_TEST_MODE PROPERTY STRINGS "NORMAL;VALGRIND;COLLECT;COMPARE")
+
+function(doctest_add_test_impl)
+    cmake_parse_arguments(ARG "NO_VALGRIND;NO_OUTPUT;XML_OUTPUT;JUNIT_OUTPUT" "NAME" "COMMAND" ${ARGN})
+    if(NOT "${ARG_UNPARSED_ARGUMENTS}" STREQUAL "" OR "${ARG_NAME}" STREQUAL "" OR "${ARG_COMMAND}" STREQUAL "")
+        message(FATAL_ERROR "doctest_add_test() called with wrong options!")
+    endif()
+
+    set(the_test_mode NORMAL)
+
+    # construct the command that will be called by the exec_test.cmake script
+    set(the_command "")
+    if(${DOCTEST_TEST_MODE} STREQUAL "VALGRIND" AND NOT ARG_NO_VALGRIND)
+        set(the_test_mode VALGRIND)
+        set(the_command "valgrind -v --leak-check=full --track-origins=yes --error-exitcode=1")
+    endif()
+    foreach(cur ${ARG_COMMAND})
+        set(the_command "${the_command} ${cur}")
+    endforeach()
+    if(ARG_XML_OUTPUT)
+        set(the_command "${the_command} --reporters=xml")
+        set(ARG_NAME ${ARG_NAME}_xml)
+    endif()
+    if(ARG_JUNIT_OUTPUT)
+        set(the_command "${the_command} --reporters=junit")
+        set(ARG_NAME ${ARG_NAME}_junit)
+    endif()
+
+    # append the argument for removing paths from filenames in the output so tests give the same output everywhere
+    set(the_command "${the_command} --dt-no-path-filenames=1")
+    # append the argument for substituting source line numbers with 0 in the output so tests give the same output when lines change a bit
+    set(the_command "${the_command} --dt-no-line-numbers=1")
+    # append the argument for ignoring the exit code of the test programs because some are intended to have failing tests
+    set(the_command "${the_command} --dt-no-exitcode=1")
+    # append the argument for using the same line format in the output - so gcc/non-gcc builds have the same output
+    set(the_command "${the_command} --dt-gnu-file-line=0")
+    # append the argument for skipping any time-related output so that the reference output from reporters is stable on CI
+    set(the_command "${the_command} --dt-no-time-in-output=1")
+
+    string(STRIP ${the_command} the_command)
+
+    if(${DOCTEST_TEST_MODE} STREQUAL "COLLECT" OR ${DOCTEST_TEST_MODE} STREQUAL "COMPARE")
+        if(NOT ARG_NO_OUTPUT)
+            file(MAKE_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/test_output/)
+            set(the_test_mode ${DOCTEST_TEST_MODE})
+            list(APPEND ADDITIONAL_FLAGS -DTEST_OUTPUT_FILE=${CMAKE_CURRENT_SOURCE_DIR}/test_output/${ARG_NAME}.txt)
+            list(APPEND ADDITIONAL_FLAGS -DTEST_TEMP_FILE=${CMAKE_CURRENT_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/temp_test_output_${ARG_NAME}.txt)
+        endif()
+    endif()
+
+    list(APPEND ADDITIONAL_FLAGS -DTEST_MODE=${the_test_mode})
+
+    add_test(NAME ${ARG_NAME} COMMAND ${CMAKE_COMMAND} -DCOMMAND=${the_command} ${ADDITIONAL_FLAGS} -P ${CURRENT_LIST_DIR_CACHED}/exec_test.cmake)
+endfunction()
+
+# a custom version of add_test() to suite my needs
+function(doctest_add_test)
+    doctest_add_test_impl(${ARGN})
+    doctest_add_test_impl(${ARGN} XML_OUTPUT)
+    doctest_add_test_impl(${ARGN} JUNIT_OUTPUT)
+endfunction()
+
+macro(add_compiler_flags)
+    foreach(flag ${ARGV})
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${flag}")
+    endforeach()
+endmacro()
+
+if(CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang")
+    add_compiler_flags(-Werror)
+    add_compiler_flags(-fstrict-aliasing)
+
+    # The following options are not valid when clang-cl is used.
+    if(NOT MSVC)
+        add_compiler_flags(-pedantic)
+        add_compiler_flags(-pedantic-errors)
+        add_compiler_flags(-fvisibility=hidden)
+    endif()
+endif()
+
+if(CMAKE_CXX_COMPILER_ID MATCHES "GNU")
+    #add_compiler_flags(-Wno-unknown-pragmas)
+    add_compiler_flags(-Wall)
+    add_compiler_flags(-Wextra)
+    add_compiler_flags(-fdiagnostics-show-option)
+    add_compiler_flags(-Wconversion)
+    add_compiler_flags(-Wold-style-cast)
+    add_compiler_flags(-Wfloat-equal)
+    add_compiler_flags(-Wlogical-op)
+    add_compiler_flags(-Wundef)
+    add_compiler_flags(-Wredundant-decls)
+    add_compiler_flags(-Wshadow)
+    add_compiler_flags(-Wstrict-overflow=5)
+    add_compiler_flags(-Wwrite-strings)
+    add_compiler_flags(-Wpointer-arith)
+    add_compiler_flags(-Wcast-qual)
+    add_compiler_flags(-Wformat=2)
+    add_compiler_flags(-Wswitch-default)
+    add_compiler_flags(-Wmissing-include-dirs)
+    add_compiler_flags(-Wcast-align)
+    add_compiler_flags(-Wswitch-enum)
+    add_compiler_flags(-Wnon-virtual-dtor)
+    add_compiler_flags(-Wctor-dtor-privacy)
+    add_compiler_flags(-Wsign-conversion)
+    add_compiler_flags(-Wdisabled-optimization)
+    add_compiler_flags(-Weffc++)
+    add_compiler_flags(-Winvalid-pch)
+    add_compiler_flags(-Wmissing-declarations)
+    add_compiler_flags(-Woverloaded-virtual)
+    add_compiler_flags(-Wunused-but-set-variable)
+    add_compiler_flags(-Wunused-result)
+
+    # add_compiler_flags(-Wsuggest-override)
+    # add_compiler_flags(-Wmultiple-inheritance)
+    # add_compiler_flags(-Wcatch-value)
+    # add_compiler_flags(-Wsuggest-attribute=cold)
+    # add_compiler_flags(-Wsuggest-attribute=const)
+    # add_compiler_flags(-Wsuggest-attribute=format)
+    # add_compiler_flags(-Wsuggest-attribute=malloc)
+    # add_compiler_flags(-Wsuggest-attribute=noreturn)
+    # add_compiler_flags(-Wsuggest-attribute=pure)
+    # add_compiler_flags(-Wsuggest-final-methods)
+    # add_compiler_flags(-Wsuggest-final-types)
+
+    if(NOT CMAKE_CXX_COMPILER_VERSION VERSION_LESS 4.6)
+        add_compiler_flags(-Wnoexcept)
+    endif()
+
+    # no way to silence it in the expression decomposition macros: _Pragma() in macros doesn't work for the c++ front-end of g++
+    # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55578
+    # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69543
+    # Also the warning is completely worthless nowadays - https://stackoverflow.com/questions/14016993
+    #add_compiler_flags(-Waggregate-return)
+
+    if(NOT CMAKE_CXX_COMPILER_VERSION VERSION_LESS 5.0)
+        add_compiler_flags(-Wdouble-promotion)
+        add_compiler_flags(-Wtrampolines)
+        add_compiler_flags(-Wzero-as-null-pointer-constant)
+        add_compiler_flags(-Wuseless-cast)
+        add_compiler_flags(-Wvector-operation-performance)
+    endif()
+
+    if(NOT CMAKE_CXX_COMPILER_VERSION VERSION_LESS 6.0)
+        add_compiler_flags(-Wshift-overflow=2)
+        add_compiler_flags(-Wnull-dereference)
+        add_compiler_flags(-Wduplicated-cond)
+    endif()
+
+    if(NOT CMAKE_CXX_COMPILER_VERSION VERSION_LESS 7.0)
+        add_compiler_flags(-Walloc-zero)
+        add_compiler_flags(-Walloca)
+        add_compiler_flags(-Wduplicated-branches)
+    endif()
+
+    if(NOT CMAKE_CXX_COMPILER_VERSION VERSION_LESS 8.0)
+        add_compiler_flags(-Wcast-align=strict)
+    endif()
+endif()
+
+# necessary for some older compilers which don't default to C++11
+set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+
+if(CMAKE_CXX_COMPILER_ID MATCHES "Clang")
+    add_compiler_flags(-Weverything)
+    add_compiler_flags(-Wno-c++98-compat)
+    add_compiler_flags(-Wno-c++98-compat-pedantic)
+    add_compiler_flags(-Wno-c++98-compat-bind-to-temporary-copy)
+    add_compiler_flags(-Wno-c++98-compat-local-type-template-args)
+    add_compiler_flags(-Qunused-arguments -fcolor-diagnostics) # needed for ccache integration on travis
+endif()
+
+if(MSVC)
+    add_compiler_flags(/std:c++latest) # for post c++14 updates in MSVC
+    add_compiler_flags(/permissive-)   # force standard conformance - this is the better flag than /Za
+    add_compiler_flags(/WX)
+    add_compiler_flags(/Wall) # turns on warnings from levels 1 through 4 which are off by default - https://msdn.microsoft.com/en-us/library/23k5d385.aspx
+
+    add_compiler_flags(
+        /wd4514 # unreferenced inline function has been removed
+        /wd4571 # SEH related
+        /wd4710 # function not inlined
+        /wd4711 # function 'x' selected for automatic inline expansion
+
+        /wd4616 # invalid compiler warnings - https://msdn.microsoft.com/en-us/library/t7ab6xtd.aspx
+        /wd4619 # invalid compiler warnings - https://msdn.microsoft.com/en-us/library/tacee08d.aspx
+
+        #/wd4820 # padding in structs
+        #/wd4625 # copy constructor was implicitly defined as deleted
+        #/wd4626 # assignment operator was implicitly defined as deleted
+        #/wd5027 # move assignment operator was implicitly defined as deleted
+        #/wd5026 # move constructor was implicitly defined as deleted
+        #/wd4623 # default constructor was implicitly defined as deleted
+    )
+endif()

+ 175 - 0
Tests/Dependencies/doctest/cmake/doctest.cmake

@@ -0,0 +1,175 @@
+# Distributed under the OSI-approved BSD 3-Clause License.  See accompanying
+# file Copyright.txt or https://cmake.org/licensing for details.
+
+#[=======================================================================[.rst:
+doctest
+-----
+
+This module defines a function to help use the doctest test framework.
+
+The :command:`doctest_discover_tests` discovers tests by asking the compiled test
+executable to enumerate its tests.  This does not require CMake to be re-run
+when tests change.  However, it may not work in a cross-compiling environment,
+and setting test properties is less convenient.
+
+This command is intended to replace use of :command:`add_test` to register
+tests, and will create a separate CTest test for each doctest test case.  Note
+that this is in some cases less efficient, as common set-up and tear-down logic
+cannot be shared by multiple test cases executing in the same instance.
+However, it provides more fine-grained pass/fail information to CTest, which is
+usually considered as more beneficial.  By default, the CTest test name is the
+same as the doctest name; see also ``TEST_PREFIX`` and ``TEST_SUFFIX``.
+
+.. command:: doctest_discover_tests
+
+  Automatically add tests with CTest by querying the compiled test executable
+  for available tests::
+
+    doctest_discover_tests(target
+                         [TEST_SPEC arg1...]
+                         [EXTRA_ARGS arg1...]
+                         [WORKING_DIRECTORY dir]
+                         [TEST_PREFIX prefix]
+                         [TEST_SUFFIX suffix]
+                         [PROPERTIES name1 value1...]
+                         [TEST_LIST var]
+    )
+
+  ``doctest_discover_tests`` sets up a post-build command on the test executable
+  that generates the list of tests by parsing the output from running the test
+  with the ``--list-test-cases`` argument.  This ensures that the full
+  list of tests is obtained.  Since test discovery occurs at build time, it is
+  not necessary to re-run CMake when the list of tests changes.
+  However, it requires that :prop_tgt:`CROSSCOMPILING_EMULATOR` is properly set
+  in order to function in a cross-compiling environment.
+
+  Additionally, setting properties on tests is somewhat less convenient, since
+  the tests are not available at CMake time.  Additional test properties may be
+  assigned to the set of tests as a whole using the ``PROPERTIES`` option.  If
+  more fine-grained test control is needed, custom content may be provided
+  through an external CTest script using the :prop_dir:`TEST_INCLUDE_FILES`
+  directory property.  The set of discovered tests is made accessible to such a
+  script via the ``<target>_TESTS`` variable.
+
+  The options are:
+
+  ``target``
+    Specifies the doctest executable, which must be a known CMake executable
+    target.  CMake will substitute the location of the built executable when
+    running the test.
+
+  ``TEST_SPEC arg1...``
+    Specifies test cases, wildcarded test cases, tags and tag expressions to
+    pass to the doctest executable with the ``--list-test-cases`` argument.
+
+  ``EXTRA_ARGS arg1...``
+    Any extra arguments to pass on the command line to each test case.
+
+  ``WORKING_DIRECTORY dir``
+    Specifies the directory in which to run the discovered test cases.  If this
+    option is not provided, the current binary directory is used.
+
+  ``TEST_PREFIX prefix``
+    Specifies a ``prefix`` to be prepended to the name of each discovered test
+    case.  This can be useful when the same test executable is being used in
+    multiple calls to ``doctest_discover_tests()`` but with different
+    ``TEST_SPEC`` or ``EXTRA_ARGS``.
+
+  ``TEST_SUFFIX suffix``
+    Similar to ``TEST_PREFIX`` except the ``suffix`` is appended to the name of
+    every discovered test case.  Both ``TEST_PREFIX`` and ``TEST_SUFFIX`` may
+    be specified.
+
+  ``PROPERTIES name1 value1...``
+    Specifies additional properties to be set on all tests discovered by this
+    invocation of ``doctest_discover_tests``.
+
+  ``TEST_LIST var``
+    Make the list of tests available in the variable ``var``, rather than the
+    default ``<target>_TESTS``.  This can be useful when the same test
+    executable is being used in multiple calls to ``doctest_discover_tests()``.
+    Note that this variable is only available in CTest.
+
+#]=======================================================================]
+
+#------------------------------------------------------------------------------
+function(doctest_discover_tests TARGET)
+  cmake_parse_arguments(
+    ""
+    ""
+    "TEST_PREFIX;TEST_SUFFIX;WORKING_DIRECTORY;TEST_LIST"
+    "TEST_SPEC;EXTRA_ARGS;PROPERTIES"
+    ${ARGN}
+  )
+
+  if(NOT _WORKING_DIRECTORY)
+    set(_WORKING_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}")
+  endif()
+  if(NOT _TEST_LIST)
+    set(_TEST_LIST ${TARGET}_TESTS)
+  endif()
+
+  ## Generate a unique name based on the extra arguments
+  string(SHA1 args_hash "${_TEST_SPEC} ${_EXTRA_ARGS}")
+  string(SUBSTRING ${args_hash} 0 7 args_hash)
+
+  # Define rule to generate test list for aforementioned test executable
+  set(ctest_include_file "${CMAKE_CURRENT_BINARY_DIR}/${TARGET}_include-${args_hash}.cmake")
+  set(ctest_tests_file "${CMAKE_CURRENT_BINARY_DIR}/${TARGET}_tests-${args_hash}.cmake")
+  get_property(crosscompiling_emulator
+    TARGET ${TARGET}
+    PROPERTY CROSSCOMPILING_EMULATOR
+  )
+  add_custom_command(
+    TARGET ${TARGET} POST_BUILD
+    BYPRODUCTS "${ctest_tests_file}"
+    COMMAND "${CMAKE_COMMAND}"
+            -D "TEST_TARGET=${TARGET}"
+            -D "TEST_EXECUTABLE=$<TARGET_FILE:${TARGET}>"
+            -D "TEST_EXECUTOR=${crosscompiling_emulator}"
+            -D "TEST_WORKING_DIR=${_WORKING_DIRECTORY}"
+            -D "TEST_SPEC=${_TEST_SPEC}"
+            -D "TEST_EXTRA_ARGS=${_EXTRA_ARGS}"
+            -D "TEST_PROPERTIES=${_PROPERTIES}"
+            -D "TEST_PREFIX=${_TEST_PREFIX}"
+            -D "TEST_SUFFIX=${_TEST_SUFFIX}"
+            -D "TEST_LIST=${_TEST_LIST}"
+            -D "CTEST_FILE=${ctest_tests_file}"
+            -P "${_DOCTEST_DISCOVER_TESTS_SCRIPT}"
+    VERBATIM
+  )
+
+  file(WRITE "${ctest_include_file}"
+    "if(EXISTS \"${ctest_tests_file}\")\n"
+    "  include(\"${ctest_tests_file}\")\n"
+    "else()\n"
+    "  add_test(${TARGET}_NOT_BUILT-${args_hash} ${TARGET}_NOT_BUILT-${args_hash})\n"
+    "endif()\n"
+  )
+
+  if(NOT CMAKE_VERSION VERSION_LESS 3.10)
+    # Add discovered tests to directory TEST_INCLUDE_FILES
+    set_property(DIRECTORY
+      APPEND PROPERTY TEST_INCLUDE_FILES "${ctest_include_file}"
+    )
+  else()
+    # Add discovered tests as directory TEST_INCLUDE_FILE if possible
+    get_property(test_include_file_set DIRECTORY PROPERTY TEST_INCLUDE_FILE SET)
+    if(NOT ${test_include_file_set})
+      set_property(DIRECTORY
+        PROPERTY TEST_INCLUDE_FILE "${ctest_include_file}"
+      )
+    else()
+      message(FATAL_ERROR
+        "Cannot set more than one TEST_INCLUDE_FILE"
+      )
+    endif()
+  endif()
+
+endfunction()
+
+###############################################################################
+
+set(_DOCTEST_DISCOVER_TESTS_SCRIPT
+  ${CMAKE_CURRENT_LIST_DIR}/doctestAddTests.cmake
+)

+ 81 - 0
Tests/Dependencies/doctest/cmake/doctestAddTests.cmake

@@ -0,0 +1,81 @@
+# Distributed under the OSI-approved BSD 3-Clause License.  See accompanying
+# file Copyright.txt or https://cmake.org/licensing for details.
+
+set(prefix "${TEST_PREFIX}")
+set(suffix "${TEST_SUFFIX}")
+set(spec ${TEST_SPEC})
+set(extra_args ${TEST_EXTRA_ARGS})
+set(properties ${TEST_PROPERTIES})
+set(script)
+set(suite)
+set(tests)
+
+function(add_command NAME)
+  set(_args "")
+  foreach(_arg ${ARGN})
+    if(_arg MATCHES "[^-./:a-zA-Z0-9_]")
+      set(_args "${_args} [==[${_arg}]==]") # form a bracket_argument
+    else()
+      set(_args "${_args} ${_arg}")
+    endif()
+  endforeach()
+  set(script "${script}${NAME}(${_args})\n" PARENT_SCOPE)
+endfunction()
+
+# Run test executable to get list of available tests
+if(NOT EXISTS "${TEST_EXECUTABLE}")
+  message(FATAL_ERROR
+    "Specified test executable '${TEST_EXECUTABLE}' does not exist"
+  )
+endif()
+
+if("${spec}" MATCHES .)
+  set(spec "--test-case=${spec}")
+endif()
+
+execute_process(
+  COMMAND ${TEST_EXECUTOR} "${TEST_EXECUTABLE}" ${spec} --list-test-cases
+  OUTPUT_VARIABLE output
+  RESULT_VARIABLE result
+)
+if(NOT ${result} EQUAL 0)
+  message(FATAL_ERROR
+    "Error running test executable '${TEST_EXECUTABLE}':\n"
+    "  Result: ${result}\n"
+    "  Output: ${output}\n"
+  )
+endif()
+
+string(REPLACE "\n" ";" output "${output}")
+
+# Parse output
+foreach(line ${output})
+  if("${line}" STREQUAL "===============================================================================" OR "${line}" MATCHES [==[^\[doctest\] ]==])
+    continue()
+  endif()
+  set(test ${line})
+  # use escape commas to handle properly test cases with commas inside the name
+  string(REPLACE "," "\\," test_name ${test})
+  # ...and add to script
+  add_command(add_test
+    "${prefix}${test}${suffix}"
+    ${TEST_EXECUTOR}
+    "${TEST_EXECUTABLE}"
+    "--test-case=${test_name}"
+    ${extra_args}
+  )
+  add_command(set_tests_properties
+    "${prefix}${test}${suffix}"
+    PROPERTIES
+    WORKING_DIRECTORY "${TEST_WORKING_DIR}"
+    ${properties}
+  )
+  list(APPEND tests "${prefix}${test}${suffix}")
+endforeach()
+
+# Create a list of all discovered tests, which users may use to e.g. set
+# properties on the tests
+add_command(set ${TEST_LIST} ${tests})
+
+# Write CTest script
+file(WRITE "${CTEST_FILE}" "${script}")

+ 70 - 0
Tests/Dependencies/doctest/cmake/exec_test.cmake

@@ -0,0 +1,70 @@
+# Arguments:
+# - COMMAND: the command to run with all it's arguments
+# - TEST_MODE: NORMAL/VALGRIND/COLLECT/COMPARE
+# - TEST_OUTPUT_FILE: the file to/from which to write/read the output of the test
+# - TEST_TEMP_FILE: the temp file for the current test output used in COMPARE mode
+# To run something through this script use cmake like this:
+# cmake -DCOMMAND=path/to/my.exe -arg1 -arg2 -DTEST_MODE=VALGRIND -P path/to/exec_test.cmake
+
+#message("COMMAND: ${COMMAND}")
+#message("TEST_MODE: ${TEST_MODE}")
+#message("TEST_OUTPUT_FILE: ${TEST_OUTPUT_FILE}")
+#message("TEST_TEMP_FILE: ${TEST_TEMP_FILE}")
+
+string(REPLACE " " ";" COMMAND_LIST ${COMMAND})
+set(cmd COMMAND ${COMMAND_LIST} RESULT_VARIABLE CMD_RESULT)
+if("${TEST_MODE}" STREQUAL "COLLECT")
+    list(APPEND cmd OUTPUT_FILE ${TEST_OUTPUT_FILE} ERROR_FILE ${TEST_OUTPUT_FILE})
+elseif("${TEST_MODE}" STREQUAL "COMPARE")
+    list(APPEND cmd OUTPUT_FILE ${TEST_TEMP_FILE} ERROR_FILE ${TEST_TEMP_FILE})
+endif()
+
+execute_process(${cmd})
+
+# fix line endings
+if("${TEST_MODE}" STREQUAL "COLLECT" AND NOT CMAKE_HOST_UNIX)
+    execute_process(COMMAND dos2unix ${TEST_OUTPUT_FILE})
+endif()
+
+if("${TEST_MODE}" STREQUAL "COMPARE")
+    # fix line endings
+    if(NOT CMAKE_HOST_UNIX)
+        execute_process(COMMAND dos2unix ${TEST_TEMP_FILE})
+    endif()
+
+    if(CMAKE_VERSION VERSION_GREATER_EQUAL "3.14.0")
+        set(IGNORE_EOL --ignore-eol)
+    endif()
+        
+    execute_process(COMMAND ${CMAKE_COMMAND} -E compare_files ${IGNORE_EOL} ${TEST_OUTPUT_FILE} ${TEST_TEMP_FILE} RESULT_VARIABLE cmp_result)
+
+    if(cmp_result)
+        find_package(Git)
+        if(GIT_FOUND)
+            set(cmd ${GIT_EXECUTABLE} diff --no-index ${TEST_OUTPUT_FILE} ${TEST_TEMP_FILE})
+            execute_process(COMMAND ${GIT_EXECUTABLE} diff --no-index ${TEST_OUTPUT_FILE} ${TEST_TEMP_FILE} OUTPUT_VARIABLE DIFF)
+            message("${DIFF}")
+        endif()
+
+        # file(READ ${TEST_OUTPUT_FILE} orig)
+        # file(READ ${TEST_TEMP_FILE} temp)
+
+        # message("==========================================================================")
+        # message("== CONTENTS OF ${TEST_OUTPUT_FILE}")
+        # message("==========================================================================")
+        # message("${orig}")
+        # message("==========================================================================")
+        # message("== CONTENTS OF ${TEST_TEMP_FILE}")
+        # message("==========================================================================")
+        # message("${temp}")
+        # message("==========================================================================")
+        # message("== CONTENTS END")
+        # message("==========================================================================")
+
+        set(CMD_RESULT "Output is different from reference file!")
+    endif()
+endif()
+
+if(CMD_RESULT)
+    message(FATAL_ERROR "Running '${COMMAND}' ended with code '${CMD_RESULT}'")
+endif()

+ 6205 - 0
Tests/Dependencies/doctest/doctest.h

@@ -0,0 +1,6205 @@
+// ====================================================================== lgtm [cpp/missing-header-guard]
+// == DO NOT MODIFY THIS FILE BY HAND - IT IS AUTO GENERATED BY CMAKE! ==
+// ======================================================================
+//
+// doctest.h - the lightest feature-rich C++ single-header testing framework for unit tests and TDD
+//
+// Copyright (c) 2016-2019 Viktor Kirilov
+//
+// Distributed under the MIT Software License
+// See accompanying file LICENSE.txt or copy at
+// https://opensource.org/licenses/MIT
+//
+// The documentation can be found at the library's page:
+// https://github.com/onqtam/doctest/blob/master/doc/markdown/readme.md
+//
+// =================================================================================================
+// =================================================================================================
+// =================================================================================================
+//
+// The library is heavily influenced by Catch - https://github.com/catchorg/Catch2
+// which uses the Boost Software License - Version 1.0
+// see here - https://github.com/catchorg/Catch2/blob/master/LICENSE.txt
+//
+// The concept of subcases (sections in Catch) and expression decomposition are from there.
+// Some parts of the code are taken directly:
+// - stringification - the detection of "ostream& operator<<(ostream&, const T&)" and StringMaker<>
+// - the Approx() helper class for floating point comparison
+// - colors in the console
+// - breaking into a debugger
+// - signal / SEH handling
+// - timer
+// - XmlWriter class - thanks to Phil Nash for allowing the direct reuse (AKA copy/paste)
+//
+// The expression decomposing templates are taken from lest - https://github.com/martinmoene/lest
+// which uses the Boost Software License - Version 1.0
+// see here - https://github.com/martinmoene/lest/blob/master/LICENSE.txt
+//
+// =================================================================================================
+// =================================================================================================
+// =================================================================================================
+
+#ifndef DOCTEST_LIBRARY_INCLUDED
+#define DOCTEST_LIBRARY_INCLUDED
+
+// =================================================================================================
+// == VERSION ======================================================================================
+// =================================================================================================
+
+#define DOCTEST_VERSION_MAJOR 2
+#define DOCTEST_VERSION_MINOR 4
+#define DOCTEST_VERSION_PATCH 0
+#define DOCTEST_VERSION_STR "2.4.0"
+
+#define DOCTEST_VERSION                                                                            \
+    (DOCTEST_VERSION_MAJOR * 10000 + DOCTEST_VERSION_MINOR * 100 + DOCTEST_VERSION_PATCH)
+
+// =================================================================================================
+// == COMPILER VERSION =============================================================================
+// =================================================================================================
+
+// ideas for the version stuff are taken from here: https://github.com/cxxstuff/cxx_detect
+
+#define DOCTEST_COMPILER(MAJOR, MINOR, PATCH) ((MAJOR)*10000000 + (MINOR)*100000 + (PATCH))
+
+// GCC/Clang and GCC/MSVC are mutually exclusive, but Clang/MSVC are not because of clang-cl...
+#if defined(_MSC_VER) && defined(_MSC_FULL_VER)
+#if _MSC_VER == _MSC_FULL_VER / 10000
+#define DOCTEST_MSVC DOCTEST_COMPILER(_MSC_VER / 100, _MSC_VER % 100, _MSC_FULL_VER % 10000)
+#else // MSVC
+#define DOCTEST_MSVC                                                                               \
+    DOCTEST_COMPILER(_MSC_VER / 100, (_MSC_FULL_VER / 100000) % 100, _MSC_FULL_VER % 100000)
+#endif // MSVC
+#endif // MSVC
+#if defined(__clang__) && defined(__clang_minor__)
+#define DOCTEST_CLANG DOCTEST_COMPILER(__clang_major__, __clang_minor__, __clang_patchlevel__)
+#elif defined(__GNUC__) && defined(__GNUC_MINOR__) && defined(__GNUC_PATCHLEVEL__) &&              \
+        !defined(__INTEL_COMPILER)
+#define DOCTEST_GCC DOCTEST_COMPILER(__GNUC__, __GNUC_MINOR__, __GNUC_PATCHLEVEL__)
+#endif // GCC
+
+#ifndef DOCTEST_MSVC
+#define DOCTEST_MSVC 0
+#endif // DOCTEST_MSVC
+#ifndef DOCTEST_CLANG
+#define DOCTEST_CLANG 0
+#endif // DOCTEST_CLANG
+#ifndef DOCTEST_GCC
+#define DOCTEST_GCC 0
+#endif // DOCTEST_GCC
+
+// =================================================================================================
+// == COMPILER WARNINGS HELPERS ====================================================================
+// =================================================================================================
+
+#if DOCTEST_CLANG
+#define DOCTEST_PRAGMA_TO_STR(x) _Pragma(#x)
+#define DOCTEST_CLANG_SUPPRESS_WARNING_PUSH _Pragma("clang diagnostic push")
+#define DOCTEST_CLANG_SUPPRESS_WARNING(w) DOCTEST_PRAGMA_TO_STR(clang diagnostic ignored w)
+#define DOCTEST_CLANG_SUPPRESS_WARNING_POP _Pragma("clang diagnostic pop")
+#define DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH(w)                                                \
+    DOCTEST_CLANG_SUPPRESS_WARNING_PUSH DOCTEST_CLANG_SUPPRESS_WARNING(w)
+#else // DOCTEST_CLANG
+#define DOCTEST_CLANG_SUPPRESS_WARNING_PUSH
+#define DOCTEST_CLANG_SUPPRESS_WARNING(w)
+#define DOCTEST_CLANG_SUPPRESS_WARNING_POP
+#define DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH(w)
+#endif // DOCTEST_CLANG
+
+#if DOCTEST_GCC
+#define DOCTEST_PRAGMA_TO_STR(x) _Pragma(#x)
+#define DOCTEST_GCC_SUPPRESS_WARNING_PUSH _Pragma("GCC diagnostic push")
+#define DOCTEST_GCC_SUPPRESS_WARNING(w) DOCTEST_PRAGMA_TO_STR(GCC diagnostic ignored w)
+#define DOCTEST_GCC_SUPPRESS_WARNING_POP _Pragma("GCC diagnostic pop")
+#define DOCTEST_GCC_SUPPRESS_WARNING_WITH_PUSH(w)                                                  \
+    DOCTEST_GCC_SUPPRESS_WARNING_PUSH DOCTEST_GCC_SUPPRESS_WARNING(w)
+#else // DOCTEST_GCC
+#define DOCTEST_GCC_SUPPRESS_WARNING_PUSH
+#define DOCTEST_GCC_SUPPRESS_WARNING(w)
+#define DOCTEST_GCC_SUPPRESS_WARNING_POP
+#define DOCTEST_GCC_SUPPRESS_WARNING_WITH_PUSH(w)
+#endif // DOCTEST_GCC
+
+#if DOCTEST_MSVC
+#define DOCTEST_MSVC_SUPPRESS_WARNING_PUSH __pragma(warning(push))
+#define DOCTEST_MSVC_SUPPRESS_WARNING(w) __pragma(warning(disable : w))
+#define DOCTEST_MSVC_SUPPRESS_WARNING_POP __pragma(warning(pop))
+#define DOCTEST_MSVC_SUPPRESS_WARNING_WITH_PUSH(w)                                                 \
+    DOCTEST_MSVC_SUPPRESS_WARNING_PUSH DOCTEST_MSVC_SUPPRESS_WARNING(w)
+#else // DOCTEST_MSVC
+#define DOCTEST_MSVC_SUPPRESS_WARNING_PUSH
+#define DOCTEST_MSVC_SUPPRESS_WARNING(w)
+#define DOCTEST_MSVC_SUPPRESS_WARNING_POP
+#define DOCTEST_MSVC_SUPPRESS_WARNING_WITH_PUSH(w)
+#endif // DOCTEST_MSVC
+
+// =================================================================================================
+// == COMPILER WARNINGS ============================================================================
+// =================================================================================================
+
+DOCTEST_CLANG_SUPPRESS_WARNING_PUSH
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wunknown-pragmas")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wnon-virtual-dtor")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wweak-vtables")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wpadded")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wdeprecated")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wmissing-prototypes")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wunused-local-typedef")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wc++98-compat")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wc++98-compat-pedantic")
+
+DOCTEST_GCC_SUPPRESS_WARNING_PUSH
+DOCTEST_GCC_SUPPRESS_WARNING("-Wunknown-pragmas")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wpragmas")
+DOCTEST_GCC_SUPPRESS_WARNING("-Weffc++")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wstrict-overflow")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wstrict-aliasing")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wctor-dtor-privacy")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wmissing-declarations")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wnon-virtual-dtor")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wunused-local-typedefs")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wuseless-cast")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wnoexcept")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wsign-promo")
+
+DOCTEST_MSVC_SUPPRESS_WARNING_PUSH
+DOCTEST_MSVC_SUPPRESS_WARNING(4616) // invalid compiler warning
+DOCTEST_MSVC_SUPPRESS_WARNING(4619) // invalid compiler warning
+DOCTEST_MSVC_SUPPRESS_WARNING(4996) // The compiler encountered a deprecated declaration
+DOCTEST_MSVC_SUPPRESS_WARNING(4706) // assignment within conditional expression
+DOCTEST_MSVC_SUPPRESS_WARNING(4512) // 'class' : assignment operator could not be generated
+DOCTEST_MSVC_SUPPRESS_WARNING(4127) // conditional expression is constant
+DOCTEST_MSVC_SUPPRESS_WARNING(4820) // padding
+DOCTEST_MSVC_SUPPRESS_WARNING(4625) // copy constructor was implicitly defined as deleted
+DOCTEST_MSVC_SUPPRESS_WARNING(4626) // assignment operator was implicitly defined as deleted
+DOCTEST_MSVC_SUPPRESS_WARNING(5027) // move assignment operator was implicitly defined as deleted
+DOCTEST_MSVC_SUPPRESS_WARNING(5026) // move constructor was implicitly defined as deleted
+DOCTEST_MSVC_SUPPRESS_WARNING(4623) // default constructor was implicitly defined as deleted
+DOCTEST_MSVC_SUPPRESS_WARNING(4640) // construction of local static object is not thread-safe
+// static analysis
+DOCTEST_MSVC_SUPPRESS_WARNING(26439) // This kind of function may not throw. Declare it 'noexcept'
+DOCTEST_MSVC_SUPPRESS_WARNING(26495) // Always initialize a member variable
+DOCTEST_MSVC_SUPPRESS_WARNING(26451) // Arithmetic overflow ...
+DOCTEST_MSVC_SUPPRESS_WARNING(26444) // Avoid unnamed objects with custom construction and dtr...
+DOCTEST_MSVC_SUPPRESS_WARNING(26812) // Prefer 'enum class' over 'enum'
+
+// 4548 - expression before comma has no effect; expected expression with side - effect
+// 4265 - class has virtual functions, but destructor is not virtual
+// 4986 - exception specification does not match previous declaration
+// 4350 - behavior change: 'member1' called instead of 'member2'
+// 4668 - 'x' is not defined as a preprocessor macro, replacing with '0' for '#if/#elif'
+// 4365 - conversion from 'int' to 'unsigned long', signed/unsigned mismatch
+// 4774 - format string expected in argument 'x' is not a string literal
+// 4820 - padding in structs
+
+// only 4 should be disabled globally:
+// - 4514 # unreferenced inline function has been removed
+// - 4571 # SEH related
+// - 4710 # function not inlined
+// - 4711 # function 'x' selected for automatic inline expansion
+
+#define DOCTEST_MAKE_STD_HEADERS_CLEAN_FROM_WARNINGS_ON_WALL_BEGIN                                 \
+    DOCTEST_MSVC_SUPPRESS_WARNING_PUSH                                                             \
+    DOCTEST_MSVC_SUPPRESS_WARNING(4548)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(4265)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(4986)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(4350)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(4668)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(4365)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(4774)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(4820)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(4625)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(4626)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(5027)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(5026)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(4623)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(5039)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(5045)                                                            \
+    DOCTEST_MSVC_SUPPRESS_WARNING(5105)
+
+#define DOCTEST_MAKE_STD_HEADERS_CLEAN_FROM_WARNINGS_ON_WALL_END DOCTEST_MSVC_SUPPRESS_WARNING_POP
+
+// =================================================================================================
+// == FEATURE DETECTION ============================================================================
+// =================================================================================================
+
+// general compiler feature support table: https://en.cppreference.com/w/cpp/compiler_support
+// MSVC C++11 feature support table: https://msdn.microsoft.com/en-us/library/hh567368.aspx
+// GCC C++11 feature support table: https://gcc.gnu.org/projects/cxx-status.html
+// MSVC version table:
+// https://en.wikipedia.org/wiki/Microsoft_Visual_C%2B%2B#Internal_version_numbering
+// MSVC++ 14.2 (16) _MSC_VER == 1920 (Visual Studio 2019)
+// MSVC++ 14.1 (15) _MSC_VER == 1910 (Visual Studio 2017)
+// MSVC++ 14.0      _MSC_VER == 1900 (Visual Studio 2015)
+// MSVC++ 12.0      _MSC_VER == 1800 (Visual Studio 2013)
+// MSVC++ 11.0      _MSC_VER == 1700 (Visual Studio 2012)
+// MSVC++ 10.0      _MSC_VER == 1600 (Visual Studio 2010)
+// MSVC++ 9.0       _MSC_VER == 1500 (Visual Studio 2008)
+// MSVC++ 8.0       _MSC_VER == 1400 (Visual Studio 2005)
+
+#if DOCTEST_MSVC && !defined(DOCTEST_CONFIG_WINDOWS_SEH)
+#define DOCTEST_CONFIG_WINDOWS_SEH
+#endif // MSVC
+#if defined(DOCTEST_CONFIG_NO_WINDOWS_SEH) && defined(DOCTEST_CONFIG_WINDOWS_SEH)
+#undef DOCTEST_CONFIG_WINDOWS_SEH
+#endif // DOCTEST_CONFIG_NO_WINDOWS_SEH
+
+#if !defined(_WIN32) && !defined(__QNX__) && !defined(DOCTEST_CONFIG_POSIX_SIGNALS) &&             \
+        !defined(__EMSCRIPTEN__)
+#define DOCTEST_CONFIG_POSIX_SIGNALS
+#endif // _WIN32
+#if defined(DOCTEST_CONFIG_NO_POSIX_SIGNALS) && defined(DOCTEST_CONFIG_POSIX_SIGNALS)
+#undef DOCTEST_CONFIG_POSIX_SIGNALS
+#endif // DOCTEST_CONFIG_NO_POSIX_SIGNALS
+
+#ifndef DOCTEST_CONFIG_NO_EXCEPTIONS
+#if !defined(__cpp_exceptions) && !defined(__EXCEPTIONS) && !defined(_CPPUNWIND)
+#define DOCTEST_CONFIG_NO_EXCEPTIONS
+#endif // no exceptions
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS
+
+#ifdef DOCTEST_CONFIG_NO_EXCEPTIONS_BUT_WITH_ALL_ASSERTS
+#ifndef DOCTEST_CONFIG_NO_EXCEPTIONS
+#define DOCTEST_CONFIG_NO_EXCEPTIONS
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS_BUT_WITH_ALL_ASSERTS
+
+#if defined(DOCTEST_CONFIG_NO_EXCEPTIONS) && !defined(DOCTEST_CONFIG_NO_TRY_CATCH_IN_ASSERTS)
+#define DOCTEST_CONFIG_NO_TRY_CATCH_IN_ASSERTS
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS && !DOCTEST_CONFIG_NO_TRY_CATCH_IN_ASSERTS
+
+#if defined(DOCTEST_CONFIG_IMPLEMENT_WITH_MAIN) && !defined(DOCTEST_CONFIG_IMPLEMENT)
+#define DOCTEST_CONFIG_IMPLEMENT
+#endif // DOCTEST_CONFIG_IMPLEMENT_WITH_MAIN
+
+#if defined(_WIN32) || defined(__CYGWIN__)
+#if DOCTEST_MSVC
+#define DOCTEST_SYMBOL_EXPORT __declspec(dllexport)
+#define DOCTEST_SYMBOL_IMPORT __declspec(dllimport)
+#else // MSVC
+#define DOCTEST_SYMBOL_EXPORT __attribute__((dllexport))
+#define DOCTEST_SYMBOL_IMPORT __attribute__((dllimport))
+#endif // MSVC
+#else  // _WIN32
+#define DOCTEST_SYMBOL_EXPORT __attribute__((visibility("default")))
+#define DOCTEST_SYMBOL_IMPORT
+#endif // _WIN32
+
+#ifdef DOCTEST_CONFIG_IMPLEMENTATION_IN_DLL
+#ifdef DOCTEST_CONFIG_IMPLEMENT
+#define DOCTEST_INTERFACE DOCTEST_SYMBOL_EXPORT
+#else // DOCTEST_CONFIG_IMPLEMENT
+#define DOCTEST_INTERFACE DOCTEST_SYMBOL_IMPORT
+#endif // DOCTEST_CONFIG_IMPLEMENT
+#else  // DOCTEST_CONFIG_IMPLEMENTATION_IN_DLL
+#define DOCTEST_INTERFACE
+#endif // DOCTEST_CONFIG_IMPLEMENTATION_IN_DLL
+
+#define DOCTEST_EMPTY
+
+#if DOCTEST_MSVC
+#define DOCTEST_NOINLINE __declspec(noinline)
+#define DOCTEST_UNUSED
+#define DOCTEST_ALIGNMENT(x)
+#else // MSVC
+#define DOCTEST_NOINLINE __attribute__((noinline))
+#define DOCTEST_UNUSED __attribute__((unused))
+#define DOCTEST_ALIGNMENT(x) __attribute__((aligned(x)))
+#endif // MSVC
+
+#ifndef DOCTEST_NORETURN
+#define DOCTEST_NORETURN [[noreturn]]
+#endif // DOCTEST_NORETURN
+
+#ifndef DOCTEST_NOEXCEPT
+#define DOCTEST_NOEXCEPT noexcept
+#endif // DOCTEST_NOEXCEPT
+
+// =================================================================================================
+// == FEATURE DETECTION END ========================================================================
+// =================================================================================================
+
+// internal macros for string concatenation and anonymous variable name generation
+#define DOCTEST_CAT_IMPL(s1, s2) s1##s2
+#define DOCTEST_CAT(s1, s2) DOCTEST_CAT_IMPL(s1, s2)
+#ifdef __COUNTER__ // not standard and may be missing for some compilers
+#define DOCTEST_ANONYMOUS(x) DOCTEST_CAT(x, __COUNTER__)
+#else // __COUNTER__
+#define DOCTEST_ANONYMOUS(x) DOCTEST_CAT(x, __LINE__)
+#endif // __COUNTER__
+
+#define DOCTEST_TOSTR(x) #x
+
+#ifndef DOCTEST_CONFIG_ASSERTION_PARAMETERS_BY_VALUE
+#define DOCTEST_REF_WRAP(x) x&
+#else // DOCTEST_CONFIG_ASSERTION_PARAMETERS_BY_VALUE
+#define DOCTEST_REF_WRAP(x) x
+#endif // DOCTEST_CONFIG_ASSERTION_PARAMETERS_BY_VALUE
+
+// not using __APPLE__ because... this is how Catch does it
+#ifdef __MAC_OS_X_VERSION_MIN_REQUIRED
+#define DOCTEST_PLATFORM_MAC
+#elif defined(__IPHONE_OS_VERSION_MIN_REQUIRED)
+#define DOCTEST_PLATFORM_IPHONE
+#elif defined(_WIN32)
+#define DOCTEST_PLATFORM_WINDOWS
+#else // DOCTEST_PLATFORM
+#define DOCTEST_PLATFORM_LINUX
+#endif // DOCTEST_PLATFORM
+
+#define DOCTEST_GLOBAL_NO_WARNINGS(var)                                                            \
+    DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Wglobal-constructors")                              \
+    DOCTEST_CLANG_SUPPRESS_WARNING("-Wunused-variable")                                            \
+    static int var DOCTEST_UNUSED // NOLINT(fuchsia-statically-constructed-objects,cert-err58-cpp)
+#define DOCTEST_GLOBAL_NO_WARNINGS_END() DOCTEST_CLANG_SUPPRESS_WARNING_POP
+
+#ifndef DOCTEST_BREAK_INTO_DEBUGGER
+// should probably take a look at https://github.com/scottt/debugbreak
+#ifdef DOCTEST_PLATFORM_MAC
+#define DOCTEST_BREAK_INTO_DEBUGGER() __asm__("int $3\n" : :)
+#elif DOCTEST_MSVC
+#define DOCTEST_BREAK_INTO_DEBUGGER() __debugbreak()
+#elif defined(__MINGW32__)
+DOCTEST_GCC_SUPPRESS_WARNING_WITH_PUSH("-Wredundant-decls")
+extern "C" __declspec(dllimport) void __stdcall DebugBreak();
+DOCTEST_GCC_SUPPRESS_WARNING_POP
+#define DOCTEST_BREAK_INTO_DEBUGGER() ::DebugBreak()
+#else // linux
+#define DOCTEST_BREAK_INTO_DEBUGGER() ((void)0)
+#endif // linux
+#endif // DOCTEST_BREAK_INTO_DEBUGGER
+
+// this is kept here for backwards compatibility since the config option was changed
+#ifdef DOCTEST_CONFIG_USE_IOSFWD
+#define DOCTEST_CONFIG_USE_STD_HEADERS
+#endif // DOCTEST_CONFIG_USE_IOSFWD
+
+#ifdef DOCTEST_CONFIG_USE_STD_HEADERS
+#include <iosfwd>
+#include <cstddef>
+#include <ostream>
+#else // DOCTEST_CONFIG_USE_STD_HEADERS
+
+#if DOCTEST_CLANG
+// to detect if libc++ is being used with clang (the _LIBCPP_VERSION identifier)
+#include <ciso646>
+#endif // clang
+
+#ifdef _LIBCPP_VERSION
+#define DOCTEST_STD_NAMESPACE_BEGIN _LIBCPP_BEGIN_NAMESPACE_STD
+#define DOCTEST_STD_NAMESPACE_END _LIBCPP_END_NAMESPACE_STD
+#else // _LIBCPP_VERSION
+#define DOCTEST_STD_NAMESPACE_BEGIN namespace std {
+#define DOCTEST_STD_NAMESPACE_END }
+#endif // _LIBCPP_VERSION
+
+// Forward declaring 'X' in namespace std is not permitted by the C++ Standard.
+DOCTEST_MSVC_SUPPRESS_WARNING_WITH_PUSH(4643)
+
+DOCTEST_STD_NAMESPACE_BEGIN // NOLINT (cert-dcl58-cpp)
+typedef decltype(nullptr) nullptr_t;
+template <class charT>
+struct char_traits;
+template <>
+struct char_traits<char>;
+template <class charT, class traits>
+class basic_ostream;
+typedef basic_ostream<char, char_traits<char>> ostream;
+template <class... Types>
+class tuple;
+#if DOCTEST_MSVC >= DOCTEST_COMPILER(19, 20, 0)
+// see this issue on why this is needed: https://github.com/onqtam/doctest/issues/183
+template <class _Ty>
+class allocator;
+template <class _Elem, class _Traits, class _Alloc>
+class basic_string;
+using string = basic_string<char, char_traits<char>, allocator<char>>;
+#endif // VS 2019
+DOCTEST_STD_NAMESPACE_END
+
+DOCTEST_MSVC_SUPPRESS_WARNING_POP
+
+#endif // DOCTEST_CONFIG_USE_STD_HEADERS
+
+#ifdef DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS
+#include <type_traits>
+#endif // DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS
+
+namespace doctest {
+
+DOCTEST_INTERFACE extern bool is_running_in_test;
+
+// A 24 byte string class (can be as small as 17 for x64 and 13 for x86) that can hold strings with length
+// of up to 23 chars on the stack before going on the heap - the last byte of the buffer is used for:
+// - "is small" bit - the highest bit - if "0" then it is small - otherwise its "1" (128)
+// - if small - capacity left before going on the heap - using the lowest 5 bits
+// - if small - 2 bits are left unused - the second and third highest ones
+// - if small - acts as a null terminator if strlen() is 23 (24 including the null terminator)
+//              and the "is small" bit remains "0" ("as well as the capacity left") so its OK
+// Idea taken from this lecture about the string implementation of facebook/folly - fbstring
+// https://www.youtube.com/watch?v=kPR8h4-qZdk
+// TODO:
+// - optimizations - like not deleting memory unnecessarily in operator= and etc.
+// - resize/reserve/clear
+// - substr
+// - replace
+// - back/front
+// - iterator stuff
+// - find & friends
+// - push_back/pop_back
+// - assign/insert/erase
+// - relational operators as free functions - taking const char* as one of the params
+class DOCTEST_INTERFACE String
+{
+    static const unsigned len  = 24;      //!OCLINT avoid private static members
+    static const unsigned last = len - 1; //!OCLINT avoid private static members
+
+    struct view // len should be more than sizeof(view) - because of the final byte for flags
+    {
+        char*    ptr;
+        unsigned size;
+        unsigned capacity;
+    };
+
+    union
+    {
+        char buf[len];
+        view data;
+    };
+
+    bool isOnStack() const { return (buf[last] & 128) == 0; }
+    void setOnHeap();
+    void setLast(unsigned in = last);
+
+    void copy(const String& other);
+
+public:
+    String();
+    ~String();
+
+    // cppcheck-suppress noExplicitConstructor
+    String(const char* in);
+    String(const char* in, unsigned in_size);
+
+    String(const String& other);
+    String& operator=(const String& other);
+
+    String& operator+=(const String& other);
+    String  operator+(const String& other) const;
+
+    String(String&& other);
+    String& operator=(String&& other);
+
+    char  operator[](unsigned i) const;
+    char& operator[](unsigned i);
+
+    // the only functions I'm willing to leave in the interface - available for inlining
+    const char* c_str() const { return const_cast<String*>(this)->c_str(); } // NOLINT
+    char*       c_str() {
+        if(isOnStack())
+            return reinterpret_cast<char*>(buf);
+        return data.ptr;
+    }
+
+    unsigned size() const;
+    unsigned capacity() const;
+
+    int compare(const char* other, bool no_case = false) const;
+    int compare(const String& other, bool no_case = false) const;
+};
+
+DOCTEST_INTERFACE bool operator==(const String& lhs, const String& rhs);
+DOCTEST_INTERFACE bool operator!=(const String& lhs, const String& rhs);
+DOCTEST_INTERFACE bool operator<(const String& lhs, const String& rhs);
+DOCTEST_INTERFACE bool operator>(const String& lhs, const String& rhs);
+DOCTEST_INTERFACE bool operator<=(const String& lhs, const String& rhs);
+DOCTEST_INTERFACE bool operator>=(const String& lhs, const String& rhs);
+
+DOCTEST_INTERFACE std::ostream& operator<<(std::ostream& s, const String& in);
+
+namespace Color {
+    enum Enum
+    {
+        None = 0,
+        White,
+        Red,
+        Green,
+        Blue,
+        Cyan,
+        Yellow,
+        Grey,
+
+        Bright = 0x10,
+
+        BrightRed   = Bright | Red,
+        BrightGreen = Bright | Green,
+        LightGrey   = Bright | Grey,
+        BrightWhite = Bright | White
+    };
+
+    DOCTEST_INTERFACE std::ostream& operator<<(std::ostream& s, Color::Enum code);
+} // namespace Color
+
+namespace assertType {
+    enum Enum
+    {
+        // macro traits
+
+        is_warn    = 1,
+        is_check   = 2 * is_warn,
+        is_require = 2 * is_check,
+
+        is_normal      = 2 * is_require,
+        is_throws      = 2 * is_normal,
+        is_throws_as   = 2 * is_throws,
+        is_throws_with = 2 * is_throws_as,
+        is_nothrow     = 2 * is_throws_with,
+
+        is_false = 2 * is_nothrow,
+        is_unary = 2 * is_false, // not checked anywhere - used just to distinguish the types
+
+        is_eq = 2 * is_unary,
+        is_ne = 2 * is_eq,
+
+        is_lt = 2 * is_ne,
+        is_gt = 2 * is_lt,
+
+        is_ge = 2 * is_gt,
+        is_le = 2 * is_ge,
+
+        // macro types
+
+        DT_WARN    = is_normal | is_warn,
+        DT_CHECK   = is_normal | is_check,
+        DT_REQUIRE = is_normal | is_require,
+
+        DT_WARN_FALSE    = is_normal | is_false | is_warn,
+        DT_CHECK_FALSE   = is_normal | is_false | is_check,
+        DT_REQUIRE_FALSE = is_normal | is_false | is_require,
+
+        DT_WARN_THROWS    = is_throws | is_warn,
+        DT_CHECK_THROWS   = is_throws | is_check,
+        DT_REQUIRE_THROWS = is_throws | is_require,
+
+        DT_WARN_THROWS_AS    = is_throws_as | is_warn,
+        DT_CHECK_THROWS_AS   = is_throws_as | is_check,
+        DT_REQUIRE_THROWS_AS = is_throws_as | is_require,
+
+        DT_WARN_THROWS_WITH    = is_throws_with | is_warn,
+        DT_CHECK_THROWS_WITH   = is_throws_with | is_check,
+        DT_REQUIRE_THROWS_WITH = is_throws_with | is_require,
+        
+        DT_WARN_THROWS_WITH_AS    = is_throws_with | is_throws_as | is_warn,
+        DT_CHECK_THROWS_WITH_AS   = is_throws_with | is_throws_as | is_check,
+        DT_REQUIRE_THROWS_WITH_AS = is_throws_with | is_throws_as | is_require,
+
+        DT_WARN_NOTHROW    = is_nothrow | is_warn,
+        DT_CHECK_NOTHROW   = is_nothrow | is_check,
+        DT_REQUIRE_NOTHROW = is_nothrow | is_require,
+
+        DT_WARN_EQ    = is_normal | is_eq | is_warn,
+        DT_CHECK_EQ   = is_normal | is_eq | is_check,
+        DT_REQUIRE_EQ = is_normal | is_eq | is_require,
+
+        DT_WARN_NE    = is_normal | is_ne | is_warn,
+        DT_CHECK_NE   = is_normal | is_ne | is_check,
+        DT_REQUIRE_NE = is_normal | is_ne | is_require,
+
+        DT_WARN_GT    = is_normal | is_gt | is_warn,
+        DT_CHECK_GT   = is_normal | is_gt | is_check,
+        DT_REQUIRE_GT = is_normal | is_gt | is_require,
+
+        DT_WARN_LT    = is_normal | is_lt | is_warn,
+        DT_CHECK_LT   = is_normal | is_lt | is_check,
+        DT_REQUIRE_LT = is_normal | is_lt | is_require,
+
+        DT_WARN_GE    = is_normal | is_ge | is_warn,
+        DT_CHECK_GE   = is_normal | is_ge | is_check,
+        DT_REQUIRE_GE = is_normal | is_ge | is_require,
+
+        DT_WARN_LE    = is_normal | is_le | is_warn,
+        DT_CHECK_LE   = is_normal | is_le | is_check,
+        DT_REQUIRE_LE = is_normal | is_le | is_require,
+
+        DT_WARN_UNARY    = is_normal | is_unary | is_warn,
+        DT_CHECK_UNARY   = is_normal | is_unary | is_check,
+        DT_REQUIRE_UNARY = is_normal | is_unary | is_require,
+
+        DT_WARN_UNARY_FALSE    = is_normal | is_false | is_unary | is_warn,
+        DT_CHECK_UNARY_FALSE   = is_normal | is_false | is_unary | is_check,
+        DT_REQUIRE_UNARY_FALSE = is_normal | is_false | is_unary | is_require,
+    };
+} // namespace assertType
+
+DOCTEST_INTERFACE const char* assertString(assertType::Enum at);
+DOCTEST_INTERFACE const char* failureString(assertType::Enum at);
+DOCTEST_INTERFACE const char* skipPathFromFilename(const char* file);
+
+struct DOCTEST_INTERFACE TestCaseData
+{
+    String      m_file;       // the file in which the test was registered
+    unsigned    m_line;       // the line where the test was registered
+    const char* m_name;       // name of the test case
+    const char* m_test_suite; // the test suite in which the test was added
+    const char* m_description;
+    bool        m_skip;
+    bool        m_may_fail;
+    bool        m_should_fail;
+    int         m_expected_failures;
+    double      m_timeout;
+};
+
+struct DOCTEST_INTERFACE AssertData
+{
+    // common - for all asserts
+    const TestCaseData* m_test_case;
+    assertType::Enum    m_at;
+    const char*         m_file;
+    int                 m_line;
+    const char*         m_expr;
+    bool                m_failed;
+
+    // exception-related - for all asserts
+    bool   m_threw;
+    String m_exception;
+
+    // for normal asserts
+    String m_decomp;
+
+    // for specific exception-related asserts
+    bool        m_threw_as;
+    const char* m_exception_type;
+    const char* m_exception_string;
+};
+
+struct DOCTEST_INTERFACE MessageData
+{
+    String           m_string;
+    const char*      m_file;
+    int              m_line;
+    assertType::Enum m_severity;
+};
+
+struct DOCTEST_INTERFACE SubcaseSignature
+{
+    String      m_name;
+    const char* m_file;
+    int         m_line;
+
+    bool operator<(const SubcaseSignature& other) const;
+};
+
+struct DOCTEST_INTERFACE IContextScope
+{
+    IContextScope();
+    virtual ~IContextScope();
+    virtual void stringify(std::ostream*) const = 0;
+};
+
+struct ContextOptions //!OCLINT too many fields
+{
+    std::ostream* cout;        // stdout stream - std::cout by default
+    std::ostream* cerr;        // stderr stream - std::cerr by default
+    String        binary_name; // the test binary name
+
+    // == parameters from the command line
+    String   out;       // output filename
+    String   order_by;  // how tests should be ordered
+    unsigned rand_seed; // the seed for rand ordering
+
+    unsigned first; // the first (matching) test to be executed
+    unsigned last;  // the last (matching) test to be executed
+
+    int abort_after;           // stop tests after this many failed assertions
+    int subcase_filter_levels; // apply the subcase filters for the first N levels
+
+    bool success;              // include successful assertions in output
+    bool case_sensitive;       // if filtering should be case sensitive
+    bool exit;                 // if the program should be exited after the tests are ran/whatever
+    bool duration;             // print the time duration of each test case
+    bool no_throw;             // to skip exceptions-related assertion macros
+    bool no_exitcode;          // if the framework should return 0 as the exitcode
+    bool no_run;               // to not run the tests at all (can be done with an "*" exclude)
+    bool no_version;           // to not print the version of the framework
+    bool no_colors;            // if output to the console should be colorized
+    bool force_colors;         // forces the use of colors even when a tty cannot be detected
+    bool no_breaks;            // to not break into the debugger
+    bool no_skip;              // don't skip test cases which are marked to be skipped
+    bool gnu_file_line;        // if line numbers should be surrounded with :x: and not (x):
+    bool no_path_in_filenames; // if the path to files should be removed from the output
+    bool no_line_numbers;      // if source code line numbers should be omitted from the output
+    bool no_skipped_summary;   // don't print "skipped" in the summary !!! UNDOCUMENTED !!!
+    bool no_time_in_output;    // omit any time/timestamps from output !!! UNDOCUMENTED !!!
+
+    bool help;             // to print the help
+    bool version;          // to print the version
+    bool count;            // if only the count of matching tests is to be retrieved
+    bool list_test_cases;  // to list all tests matching the filters
+    bool list_test_suites; // to list all suites matching the filters
+    bool list_reporters;   // lists all registered reporters
+};
+
+namespace detail {
+#if defined(DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING) || defined(DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS)
+    template <bool CONDITION, typename TYPE = void>
+    struct enable_if
+    {};
+
+    template <typename TYPE>
+    struct enable_if<true, TYPE>
+    { typedef TYPE type; };
+#endif // DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING) || DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS
+
+    // clang-format off
+    template<class T> struct remove_reference      { typedef T type; };
+    template<class T> struct remove_reference<T&>  { typedef T type; };
+    template<class T> struct remove_reference<T&&> { typedef T type; };
+
+    template<class T> struct remove_const          { typedef T type; };
+    template<class T> struct remove_const<const T> { typedef T type; };
+    // clang-format on
+
+    template <typename T>
+    struct deferred_false
+    // cppcheck-suppress unusedStructMember
+    { static const bool value = false; };
+
+    namespace has_insertion_operator_impl {
+        std::ostream &os();
+        template<class T>
+        DOCTEST_REF_WRAP(T) val();
+
+        template<class, class = void>
+        struct check {
+            static constexpr auto value = false;
+        };
+
+        template<class T>
+        struct check<T, decltype(os() << val<T>(), void())> {
+            static constexpr auto value = true;
+        };
+    } // namespace has_insertion_operator_impl
+
+    template<class T>
+    using has_insertion_operator = has_insertion_operator_impl::check<T>;
+
+    DOCTEST_INTERFACE void my_memcpy(void* dest, const void* src, unsigned num);
+
+    DOCTEST_INTERFACE std::ostream* getTlsOss(); // returns a thread-local ostringstream
+    DOCTEST_INTERFACE String getTlsOssResult();
+
+    template <bool C>
+    struct StringMakerBase
+    {
+        template <typename T>
+        static String convert(const DOCTEST_REF_WRAP(T)) {
+            return "{?}";
+        }
+    };
+
+    template <>
+    struct StringMakerBase<true>
+    {
+        template <typename T>
+        static String convert(const DOCTEST_REF_WRAP(T) in) {
+            *getTlsOss() << in;
+            return getTlsOssResult();
+        }
+    };
+
+    DOCTEST_INTERFACE String rawMemoryToString(const void* object, unsigned size);
+
+    template <typename T>
+    String rawMemoryToString(const DOCTEST_REF_WRAP(T) object) {
+        return rawMemoryToString(&object, sizeof(object));
+    }
+
+    template <typename T>
+    const char* type_to_string() {
+        return "<>";
+    }
+} // namespace detail
+
+template <typename T>
+struct StringMaker : public detail::StringMakerBase<detail::has_insertion_operator<T>::value>
+{};
+
+template <typename T>
+struct StringMaker<T*>
+{
+    template <typename U>
+    static String convert(U* p) {
+        if(p)
+            return detail::rawMemoryToString(p);
+        return "NULL";
+    }
+};
+
+template <typename R, typename C>
+struct StringMaker<R C::*>
+{
+    static String convert(R C::*p) {
+        if(p)
+            return detail::rawMemoryToString(p);
+        return "NULL";
+    }
+};
+
+template <typename T>
+String toString(const DOCTEST_REF_WRAP(T) value) {
+    return StringMaker<T>::convert(value);
+}
+
+#ifdef DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+DOCTEST_INTERFACE String toString(char* in);
+DOCTEST_INTERFACE String toString(const char* in);
+#endif // DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+DOCTEST_INTERFACE String toString(bool in);
+DOCTEST_INTERFACE String toString(float in);
+DOCTEST_INTERFACE String toString(double in);
+DOCTEST_INTERFACE String toString(double long in);
+
+DOCTEST_INTERFACE String toString(char in);
+DOCTEST_INTERFACE String toString(char signed in);
+DOCTEST_INTERFACE String toString(char unsigned in);
+DOCTEST_INTERFACE String toString(int short in);
+DOCTEST_INTERFACE String toString(int short unsigned in);
+DOCTEST_INTERFACE String toString(int in);
+DOCTEST_INTERFACE String toString(int unsigned in);
+DOCTEST_INTERFACE String toString(int long in);
+DOCTEST_INTERFACE String toString(int long unsigned in);
+DOCTEST_INTERFACE String toString(int long long in);
+DOCTEST_INTERFACE String toString(int long long unsigned in);
+DOCTEST_INTERFACE String toString(std::nullptr_t in);
+
+#if DOCTEST_MSVC >= DOCTEST_COMPILER(19, 20, 0)
+// see this issue on why this is needed: https://github.com/onqtam/doctest/issues/183
+DOCTEST_INTERFACE String toString(const std::string& in);
+#endif // VS 2019
+
+class DOCTEST_INTERFACE Approx
+{
+public:
+    explicit Approx(double value);
+
+    Approx operator()(double value) const;
+
+#ifdef DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS
+    template <typename T>
+    explicit Approx(const T& value,
+                    typename detail::enable_if<std::is_constructible<double, T>::value>::type* =
+                            static_cast<T*>(nullptr)) {
+        *this = Approx(static_cast<double>(value));
+    }
+#endif // DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS
+
+    Approx& epsilon(double newEpsilon);
+
+#ifdef DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS
+    template <typename T>
+    typename detail::enable_if<std::is_constructible<double, T>::value, Approx&>::type epsilon(
+            const T& newEpsilon) {
+        m_epsilon = static_cast<double>(newEpsilon);
+        return *this;
+    }
+#endif //  DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS
+
+    Approx& scale(double newScale);
+
+#ifdef DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS
+    template <typename T>
+    typename detail::enable_if<std::is_constructible<double, T>::value, Approx&>::type scale(
+            const T& newScale) {
+        m_scale = static_cast<double>(newScale);
+        return *this;
+    }
+#endif // DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS
+
+    // clang-format off
+    DOCTEST_INTERFACE friend bool operator==(double lhs, const Approx & rhs);
+    DOCTEST_INTERFACE friend bool operator==(const Approx & lhs, double rhs);
+    DOCTEST_INTERFACE friend bool operator!=(double lhs, const Approx & rhs);
+    DOCTEST_INTERFACE friend bool operator!=(const Approx & lhs, double rhs);
+    DOCTEST_INTERFACE friend bool operator<=(double lhs, const Approx & rhs);
+    DOCTEST_INTERFACE friend bool operator<=(const Approx & lhs, double rhs);
+    DOCTEST_INTERFACE friend bool operator>=(double lhs, const Approx & rhs);
+    DOCTEST_INTERFACE friend bool operator>=(const Approx & lhs, double rhs);
+    DOCTEST_INTERFACE friend bool operator< (double lhs, const Approx & rhs);
+    DOCTEST_INTERFACE friend bool operator< (const Approx & lhs, double rhs);
+    DOCTEST_INTERFACE friend bool operator> (double lhs, const Approx & rhs);
+    DOCTEST_INTERFACE friend bool operator> (const Approx & lhs, double rhs);
+
+    DOCTEST_INTERFACE friend String toString(const Approx& in);
+
+#ifdef DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS
+#define DOCTEST_APPROX_PREFIX \
+    template <typename T> friend typename detail::enable_if<std::is_constructible<double, T>::value, bool>::type
+
+    DOCTEST_APPROX_PREFIX operator==(const T& lhs, const Approx& rhs) { return operator==(double(lhs), rhs); }
+    DOCTEST_APPROX_PREFIX operator==(const Approx& lhs, const T& rhs) { return operator==(rhs, lhs); }
+    DOCTEST_APPROX_PREFIX operator!=(const T& lhs, const Approx& rhs) { return !operator==(lhs, rhs); }
+    DOCTEST_APPROX_PREFIX operator!=(const Approx& lhs, const T& rhs) { return !operator==(rhs, lhs); }
+    DOCTEST_APPROX_PREFIX operator<=(const T& lhs, const Approx& rhs) { return double(lhs) < rhs.m_value || lhs == rhs; }
+    DOCTEST_APPROX_PREFIX operator<=(const Approx& lhs, const T& rhs) { return lhs.m_value < double(rhs) || lhs == rhs; }
+    DOCTEST_APPROX_PREFIX operator>=(const T& lhs, const Approx& rhs) { return double(lhs) > rhs.m_value || lhs == rhs; }
+    DOCTEST_APPROX_PREFIX operator>=(const Approx& lhs, const T& rhs) { return lhs.m_value > double(rhs) || lhs == rhs; }
+    DOCTEST_APPROX_PREFIX operator< (const T& lhs, const Approx& rhs) { return double(lhs) < rhs.m_value && lhs != rhs; }
+    DOCTEST_APPROX_PREFIX operator< (const Approx& lhs, const T& rhs) { return lhs.m_value < double(rhs) && lhs != rhs; }
+    DOCTEST_APPROX_PREFIX operator> (const T& lhs, const Approx& rhs) { return double(lhs) > rhs.m_value && lhs != rhs; }
+    DOCTEST_APPROX_PREFIX operator> (const Approx& lhs, const T& rhs) { return lhs.m_value > double(rhs) && lhs != rhs; }
+#undef DOCTEST_APPROX_PREFIX
+#endif // DOCTEST_CONFIG_INCLUDE_TYPE_TRAITS
+
+    // clang-format on
+
+private:
+    double m_epsilon;
+    double m_scale;
+    double m_value;
+};
+
+DOCTEST_INTERFACE String toString(const Approx& in);
+
+DOCTEST_INTERFACE const ContextOptions* getContextOptions();
+
+#if !defined(DOCTEST_CONFIG_DISABLE)
+
+namespace detail {
+    // clang-format off
+#ifdef DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+    template<class T>               struct decay_array       { typedef T type; };
+    template<class T, unsigned N>   struct decay_array<T[N]> { typedef T* type; };
+    template<class T>               struct decay_array<T[]>  { typedef T* type; };
+
+    template<class T>   struct not_char_pointer              { enum { value = 1 }; };
+    template<>          struct not_char_pointer<char*>       { enum { value = 0 }; };
+    template<>          struct not_char_pointer<const char*> { enum { value = 0 }; };
+
+    template<class T> struct can_use_op : public not_char_pointer<typename decay_array<T>::type> {};
+#endif // DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+    // clang-format on
+
+    struct DOCTEST_INTERFACE TestFailureException
+    {
+    };
+
+    DOCTEST_INTERFACE bool checkIfShouldThrow(assertType::Enum at);
+
+#ifndef DOCTEST_CONFIG_NO_EXCEPTIONS
+    DOCTEST_NORETURN
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS
+    DOCTEST_INTERFACE void throwException();
+
+    struct DOCTEST_INTERFACE Subcase
+    {
+        SubcaseSignature m_signature;
+        bool             m_entered = false;
+
+        Subcase(const String& name, const char* file, int line);
+        ~Subcase();
+
+        operator bool() const;
+    };
+
+    template <typename L, typename R>
+    String stringifyBinaryExpr(const DOCTEST_REF_WRAP(L) lhs, const char* op,
+                               const DOCTEST_REF_WRAP(R) rhs) {
+        return toString(lhs) + op + toString(rhs);
+    }
+
+#define DOCTEST_DO_BINARY_EXPRESSION_COMPARISON(op, op_str, op_macro)                              \
+    template <typename R>                                                                          \
+    DOCTEST_NOINLINE Result operator op(const DOCTEST_REF_WRAP(R) rhs) {                           \
+        bool res = op_macro(lhs, rhs);                                                             \
+        if(m_at & assertType::is_false)                                                            \
+            res = !res;                                                                            \
+        if(!res || doctest::getContextOptions()->success)                                          \
+            return Result(res, stringifyBinaryExpr(lhs, op_str, rhs));                             \
+        return Result(res);                                                                        \
+    }
+
+    // more checks could be added - like in Catch:
+    // https://github.com/catchorg/Catch2/pull/1480/files
+    // https://github.com/catchorg/Catch2/pull/1481/files
+#define DOCTEST_FORBIT_EXPRESSION(rt, op)                                                          \
+    template <typename R>                                                                          \
+    rt& operator op(const R&) {                                                                    \
+        static_assert(deferred_false<R>::value,                                                    \
+                      "Expression Too Complex Please Rewrite As Binary Comparison!");              \
+        return *this;                                                                              \
+    }
+
+    struct DOCTEST_INTERFACE Result
+    {
+        bool   m_passed;
+        String m_decomp;
+
+        Result(bool passed, const String& decomposition = String());
+
+        // forbidding some expressions based on this table: https://en.cppreference.com/w/cpp/language/operator_precedence
+        DOCTEST_FORBIT_EXPRESSION(Result, &)
+        DOCTEST_FORBIT_EXPRESSION(Result, ^)
+        DOCTEST_FORBIT_EXPRESSION(Result, |)
+        DOCTEST_FORBIT_EXPRESSION(Result, &&)
+        DOCTEST_FORBIT_EXPRESSION(Result, ||)
+        DOCTEST_FORBIT_EXPRESSION(Result, ==)
+        DOCTEST_FORBIT_EXPRESSION(Result, !=)
+        DOCTEST_FORBIT_EXPRESSION(Result, <)
+        DOCTEST_FORBIT_EXPRESSION(Result, >)
+        DOCTEST_FORBIT_EXPRESSION(Result, <=)
+        DOCTEST_FORBIT_EXPRESSION(Result, >=)
+        DOCTEST_FORBIT_EXPRESSION(Result, =)
+        DOCTEST_FORBIT_EXPRESSION(Result, +=)
+        DOCTEST_FORBIT_EXPRESSION(Result, -=)
+        DOCTEST_FORBIT_EXPRESSION(Result, *=)
+        DOCTEST_FORBIT_EXPRESSION(Result, /=)
+        DOCTEST_FORBIT_EXPRESSION(Result, %=)
+        DOCTEST_FORBIT_EXPRESSION(Result, <<=)
+        DOCTEST_FORBIT_EXPRESSION(Result, >>=)
+        DOCTEST_FORBIT_EXPRESSION(Result, &=)
+        DOCTEST_FORBIT_EXPRESSION(Result, ^=)
+        DOCTEST_FORBIT_EXPRESSION(Result, |=)
+    };
+
+#ifndef DOCTEST_CONFIG_NO_COMPARISON_WARNING_SUPPRESSION
+
+    DOCTEST_CLANG_SUPPRESS_WARNING_PUSH
+    DOCTEST_CLANG_SUPPRESS_WARNING("-Wsign-conversion")
+    DOCTEST_CLANG_SUPPRESS_WARNING("-Wsign-compare")
+    //DOCTEST_CLANG_SUPPRESS_WARNING("-Wdouble-promotion")
+    //DOCTEST_CLANG_SUPPRESS_WARNING("-Wconversion")
+    //DOCTEST_CLANG_SUPPRESS_WARNING("-Wfloat-equal")
+
+    DOCTEST_GCC_SUPPRESS_WARNING_PUSH
+    DOCTEST_GCC_SUPPRESS_WARNING("-Wsign-conversion")
+    DOCTEST_GCC_SUPPRESS_WARNING("-Wsign-compare")
+    //DOCTEST_GCC_SUPPRESS_WARNING("-Wdouble-promotion")
+    //DOCTEST_GCC_SUPPRESS_WARNING("-Wconversion")
+    //DOCTEST_GCC_SUPPRESS_WARNING("-Wfloat-equal")
+
+    DOCTEST_MSVC_SUPPRESS_WARNING_PUSH
+    // https://stackoverflow.com/questions/39479163 what's the difference between 4018 and 4389
+    DOCTEST_MSVC_SUPPRESS_WARNING(4388) // signed/unsigned mismatch
+    DOCTEST_MSVC_SUPPRESS_WARNING(4389) // 'operator' : signed/unsigned mismatch
+    DOCTEST_MSVC_SUPPRESS_WARNING(4018) // 'expression' : signed/unsigned mismatch
+    //DOCTEST_MSVC_SUPPRESS_WARNING(4805) // 'operation' : unsafe mix of type 'type' and type 'type' in operation
+
+#endif // DOCTEST_CONFIG_NO_COMPARISON_WARNING_SUPPRESSION
+
+    // clang-format off
+#ifndef DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+#define DOCTEST_COMPARISON_RETURN_TYPE bool
+#else // DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+#define DOCTEST_COMPARISON_RETURN_TYPE typename enable_if<can_use_op<L>::value || can_use_op<R>::value, bool>::type
+    inline bool eq(const char* lhs, const char* rhs) { return String(lhs) == String(rhs); }
+    inline bool ne(const char* lhs, const char* rhs) { return String(lhs) != String(rhs); }
+    inline bool lt(const char* lhs, const char* rhs) { return String(lhs) <  String(rhs); }
+    inline bool gt(const char* lhs, const char* rhs) { return String(lhs) >  String(rhs); }
+    inline bool le(const char* lhs, const char* rhs) { return String(lhs) <= String(rhs); }
+    inline bool ge(const char* lhs, const char* rhs) { return String(lhs) >= String(rhs); }
+#endif // DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+    // clang-format on
+
+#define DOCTEST_RELATIONAL_OP(name, op)                                                            \
+    template <typename L, typename R>                                                              \
+    DOCTEST_COMPARISON_RETURN_TYPE name(const DOCTEST_REF_WRAP(L) lhs,                             \
+                                        const DOCTEST_REF_WRAP(R) rhs) {                           \
+        return lhs op rhs;                                                                         \
+    }
+
+    DOCTEST_RELATIONAL_OP(eq, ==)
+    DOCTEST_RELATIONAL_OP(ne, !=)
+    DOCTEST_RELATIONAL_OP(lt, <)
+    DOCTEST_RELATIONAL_OP(gt, >)
+    DOCTEST_RELATIONAL_OP(le, <=)
+    DOCTEST_RELATIONAL_OP(ge, >=)
+
+#ifndef DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+#define DOCTEST_CMP_EQ(l, r) l == r
+#define DOCTEST_CMP_NE(l, r) l != r
+#define DOCTEST_CMP_GT(l, r) l > r
+#define DOCTEST_CMP_LT(l, r) l < r
+#define DOCTEST_CMP_GE(l, r) l >= r
+#define DOCTEST_CMP_LE(l, r) l <= r
+#else // DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+#define DOCTEST_CMP_EQ(l, r) eq(l, r)
+#define DOCTEST_CMP_NE(l, r) ne(l, r)
+#define DOCTEST_CMP_GT(l, r) gt(l, r)
+#define DOCTEST_CMP_LT(l, r) lt(l, r)
+#define DOCTEST_CMP_GE(l, r) ge(l, r)
+#define DOCTEST_CMP_LE(l, r) le(l, r)
+#endif // DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+
+    template <typename L>
+    // cppcheck-suppress copyCtorAndEqOperator
+    struct Expression_lhs
+    {
+        L                lhs;
+        assertType::Enum m_at;
+
+        explicit Expression_lhs(L in, assertType::Enum at)
+                : lhs(in)
+                , m_at(at) {}
+
+        DOCTEST_NOINLINE operator Result() {
+            bool res = !!lhs;
+            if(m_at & assertType::is_false) //!OCLINT bitwise operator in conditional
+                res = !res;
+
+            if(!res || getContextOptions()->success)
+                return Result(res, toString(lhs));
+            return Result(res);
+        }
+
+        // clang-format off
+        DOCTEST_DO_BINARY_EXPRESSION_COMPARISON(==, " == ", DOCTEST_CMP_EQ) //!OCLINT bitwise operator in conditional
+        DOCTEST_DO_BINARY_EXPRESSION_COMPARISON(!=, " != ", DOCTEST_CMP_NE) //!OCLINT bitwise operator in conditional
+        DOCTEST_DO_BINARY_EXPRESSION_COMPARISON(>,  " >  ", DOCTEST_CMP_GT) //!OCLINT bitwise operator in conditional
+        DOCTEST_DO_BINARY_EXPRESSION_COMPARISON(<,  " <  ", DOCTEST_CMP_LT) //!OCLINT bitwise operator in conditional
+        DOCTEST_DO_BINARY_EXPRESSION_COMPARISON(>=, " >= ", DOCTEST_CMP_GE) //!OCLINT bitwise operator in conditional
+        DOCTEST_DO_BINARY_EXPRESSION_COMPARISON(<=, " <= ", DOCTEST_CMP_LE) //!OCLINT bitwise operator in conditional
+        // clang-format on
+
+        // forbidding some expressions based on this table: https://en.cppreference.com/w/cpp/language/operator_precedence
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, &)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, ^)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, |)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, &&)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, ||)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, =)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, +=)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, -=)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, *=)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, /=)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, %=)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, <<=)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, >>=)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, &=)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, ^=)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, |=)
+        // these 2 are unfortunate because they should be allowed - they have higher precedence over the comparisons, but the
+        // ExpressionDecomposer class uses the left shift operator to capture the left operand of the binary expression...
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, <<)
+        DOCTEST_FORBIT_EXPRESSION(Expression_lhs, >>)
+    };
+
+#ifndef DOCTEST_CONFIG_NO_COMPARISON_WARNING_SUPPRESSION
+
+    DOCTEST_CLANG_SUPPRESS_WARNING_POP
+    DOCTEST_MSVC_SUPPRESS_WARNING_POP
+    DOCTEST_GCC_SUPPRESS_WARNING_POP
+
+#endif // DOCTEST_CONFIG_NO_COMPARISON_WARNING_SUPPRESSION
+
+    struct DOCTEST_INTERFACE ExpressionDecomposer
+    {
+        assertType::Enum m_at;
+
+        ExpressionDecomposer(assertType::Enum at);
+
+        // The right operator for capturing expressions is "<=" instead of "<<" (based on the operator precedence table)
+        // but then there will be warnings from GCC about "-Wparentheses" and since "_Pragma()" is problematic this will stay for now...
+        // https://github.com/catchorg/Catch2/issues/870
+        // https://github.com/catchorg/Catch2/issues/565
+        template <typename L>
+        Expression_lhs<const DOCTEST_REF_WRAP(L)> operator<<(const DOCTEST_REF_WRAP(L) operand) {
+            return Expression_lhs<const DOCTEST_REF_WRAP(L)>(operand, m_at);
+        }
+    };
+
+    struct DOCTEST_INTERFACE TestSuite
+    {
+        const char* m_test_suite;
+        const char* m_description;
+        bool        m_skip;
+        bool        m_may_fail;
+        bool        m_should_fail;
+        int         m_expected_failures;
+        double      m_timeout;
+
+        TestSuite& operator*(const char* in);
+
+        template <typename T>
+        TestSuite& operator*(const T& in) {
+            in.fill(*this);
+            return *this;
+        }
+    };
+
+    typedef void (*funcType)();
+
+    struct DOCTEST_INTERFACE TestCase : public TestCaseData
+    {
+        funcType m_test; // a function pointer to the test case
+
+        const char* m_type; // for templated test cases - gets appended to the real name
+        int m_template_id; // an ID used to distinguish between the different versions of a templated test case
+        String m_full_name; // contains the name (only for templated test cases!) + the template type
+
+        TestCase(funcType test, const char* file, unsigned line, const TestSuite& test_suite,
+                 const char* type = "", int template_id = -1);
+
+        TestCase(const TestCase& other);
+
+        DOCTEST_MSVC_SUPPRESS_WARNING_WITH_PUSH(26434) // hides a non-virtual function
+        TestCase& operator=(const TestCase& other);
+        DOCTEST_MSVC_SUPPRESS_WARNING_POP
+
+        TestCase& operator*(const char* in);
+
+        template <typename T>
+        TestCase& operator*(const T& in) {
+            in.fill(*this);
+            return *this;
+        }
+
+        bool operator<(const TestCase& other) const;
+    };
+
+    // forward declarations of functions used by the macros
+    DOCTEST_INTERFACE int  regTest(const TestCase& tc);
+    DOCTEST_INTERFACE int  setTestSuite(const TestSuite& ts);
+    DOCTEST_INTERFACE bool isDebuggerActive();
+
+    template<typename T>
+    int instantiationHelper(const T&) { return 0; }
+
+    namespace binaryAssertComparison {
+        enum Enum
+        {
+            eq = 0,
+            ne,
+            gt,
+            lt,
+            ge,
+            le
+        };
+    } // namespace binaryAssertComparison
+
+    // clang-format off
+    template <int, class L, class R> struct RelationalComparator     { bool operator()(const DOCTEST_REF_WRAP(L),     const DOCTEST_REF_WRAP(R)    ) const { return false;        } };
+
+#define DOCTEST_BINARY_RELATIONAL_OP(n, op) \
+    template <class L, class R> struct RelationalComparator<n, L, R> { bool operator()(const DOCTEST_REF_WRAP(L) lhs, const DOCTEST_REF_WRAP(R) rhs) const { return op(lhs, rhs); } };
+    // clang-format on
+
+    DOCTEST_BINARY_RELATIONAL_OP(0, eq)
+    DOCTEST_BINARY_RELATIONAL_OP(1, ne)
+    DOCTEST_BINARY_RELATIONAL_OP(2, gt)
+    DOCTEST_BINARY_RELATIONAL_OP(3, lt)
+    DOCTEST_BINARY_RELATIONAL_OP(4, ge)
+    DOCTEST_BINARY_RELATIONAL_OP(5, le)
+
+    struct DOCTEST_INTERFACE ResultBuilder : public AssertData
+    {
+        ResultBuilder(assertType::Enum at, const char* file, int line, const char* expr,
+                      const char* exception_type = "", const char* exception_string = "");
+
+        void setResult(const Result& res);
+
+        template <int comparison, typename L, typename R>
+        DOCTEST_NOINLINE void binary_assert(const DOCTEST_REF_WRAP(L) lhs,
+                                            const DOCTEST_REF_WRAP(R) rhs) {
+            m_failed = !RelationalComparator<comparison, L, R>()(lhs, rhs);
+            if(m_failed || getContextOptions()->success)
+                m_decomp = stringifyBinaryExpr(lhs, ", ", rhs);
+        }
+
+        template <typename L>
+        DOCTEST_NOINLINE void unary_assert(const DOCTEST_REF_WRAP(L) val) {
+            m_failed = !val;
+
+            if(m_at & assertType::is_false) //!OCLINT bitwise operator in conditional
+                m_failed = !m_failed;
+
+            if(m_failed || getContextOptions()->success)
+                m_decomp = toString(val);
+        }
+
+        void translateException();
+
+        bool log();
+        void react() const;
+    };
+
+    namespace assertAction {
+        enum Enum
+        {
+            nothing     = 0,
+            dbgbreak    = 1,
+            shouldthrow = 2
+        };
+    } // namespace assertAction
+
+    DOCTEST_INTERFACE void failed_out_of_a_testing_context(const AssertData& ad);
+
+    DOCTEST_INTERFACE void decomp_assert(assertType::Enum at, const char* file, int line,
+                                         const char* expr, Result result);
+
+#define DOCTEST_ASSERT_OUT_OF_TESTS(decomp)                                                        \
+    do {                                                                                           \
+        if(!is_running_in_test) {                                                                  \
+            if(failed) {                                                                           \
+                ResultBuilder rb(at, file, line, expr);                                            \
+                rb.m_failed = failed;                                                              \
+                rb.m_decomp = decomp;                                                              \
+                failed_out_of_a_testing_context(rb);                                               \
+                if(isDebuggerActive() && !getContextOptions()->no_breaks)                          \
+                    DOCTEST_BREAK_INTO_DEBUGGER();                                                 \
+                if(checkIfShouldThrow(at))                                                         \
+                    throwException();                                                              \
+            }                                                                                      \
+            return;                                                                                \
+        }                                                                                          \
+    } while(false)
+
+#define DOCTEST_ASSERT_IN_TESTS(decomp)                                                            \
+    ResultBuilder rb(at, file, line, expr);                                                        \
+    rb.m_failed = failed;                                                                          \
+    if(rb.m_failed || getContextOptions()->success)                                                \
+        rb.m_decomp = decomp;                                                                      \
+    if(rb.log())                                                                                   \
+        DOCTEST_BREAK_INTO_DEBUGGER();                                                             \
+    if(rb.m_failed && checkIfShouldThrow(at))                                                      \
+    throwException()
+
+    template <int comparison, typename L, typename R>
+    DOCTEST_NOINLINE void binary_assert(assertType::Enum at, const char* file, int line,
+                                        const char* expr, const DOCTEST_REF_WRAP(L) lhs,
+                                        const DOCTEST_REF_WRAP(R) rhs) {
+        bool failed = !RelationalComparator<comparison, L, R>()(lhs, rhs);
+
+        // ###################################################################################
+        // IF THE DEBUGGER BREAKS HERE - GO 1 LEVEL UP IN THE CALLSTACK FOR THE FAILING ASSERT
+        // THIS IS THE EFFECT OF HAVING 'DOCTEST_CONFIG_SUPER_FAST_ASSERTS' DEFINED
+        // ###################################################################################
+        DOCTEST_ASSERT_OUT_OF_TESTS(stringifyBinaryExpr(lhs, ", ", rhs));
+        DOCTEST_ASSERT_IN_TESTS(stringifyBinaryExpr(lhs, ", ", rhs));
+    }
+
+    template <typename L>
+    DOCTEST_NOINLINE void unary_assert(assertType::Enum at, const char* file, int line,
+                                       const char* expr, const DOCTEST_REF_WRAP(L) val) {
+        bool failed = !val;
+
+        if(at & assertType::is_false) //!OCLINT bitwise operator in conditional
+            failed = !failed;
+
+        // ###################################################################################
+        // IF THE DEBUGGER BREAKS HERE - GO 1 LEVEL UP IN THE CALLSTACK FOR THE FAILING ASSERT
+        // THIS IS THE EFFECT OF HAVING 'DOCTEST_CONFIG_SUPER_FAST_ASSERTS' DEFINED
+        // ###################################################################################
+        DOCTEST_ASSERT_OUT_OF_TESTS(toString(val));
+        DOCTEST_ASSERT_IN_TESTS(toString(val));
+    }
+
+    struct DOCTEST_INTERFACE IExceptionTranslator
+    {
+        IExceptionTranslator();
+        virtual ~IExceptionTranslator();
+        virtual bool translate(String&) const = 0;
+    };
+
+    template <typename T>
+    class ExceptionTranslator : public IExceptionTranslator //!OCLINT destructor of virtual class
+    {
+    public:
+        explicit ExceptionTranslator(String (*translateFunction)(T))
+                : m_translateFunction(translateFunction) {}
+
+        bool translate(String& res) const override {
+#ifndef DOCTEST_CONFIG_NO_EXCEPTIONS
+            try {
+                throw; // lgtm [cpp/rethrow-no-exception]
+                // cppcheck-suppress catchExceptionByValue
+            } catch(T ex) {                    // NOLINT
+                res = m_translateFunction(ex); //!OCLINT parameter reassignment
+                return true;
+            } catch(...) {} //!OCLINT -  empty catch statement
+#endif                      // DOCTEST_CONFIG_NO_EXCEPTIONS
+            ((void)res);    // to silence -Wunused-parameter
+            return false;
+        }
+
+    private:
+        String (*m_translateFunction)(T);
+    };
+
+    DOCTEST_INTERFACE void registerExceptionTranslatorImpl(const IExceptionTranslator* et);
+
+    template <bool C>
+    struct StringStreamBase
+    {
+        template <typename T>
+        static void convert(std::ostream* s, const T& in) {
+            *s << toString(in);
+        }
+
+        // always treat char* as a string in this context - no matter
+        // if DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING is defined
+        static void convert(std::ostream* s, const char* in) { *s << String(in); }
+    };
+
+    template <>
+    struct StringStreamBase<true>
+    {
+        template <typename T>
+        static void convert(std::ostream* s, const T& in) {
+            *s << in;
+        }
+    };
+
+    template <typename T>
+    struct StringStream : public StringStreamBase<has_insertion_operator<T>::value>
+    {};
+
+    template <typename T>
+    void toStream(std::ostream* s, const T& value) {
+        StringStream<T>::convert(s, value);
+    }
+
+#ifdef DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+    DOCTEST_INTERFACE void toStream(std::ostream* s, char* in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, const char* in);
+#endif // DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+    DOCTEST_INTERFACE void toStream(std::ostream* s, bool in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, float in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, double in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, double long in);
+
+    DOCTEST_INTERFACE void toStream(std::ostream* s, char in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, char signed in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, char unsigned in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, int short in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, int short unsigned in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, int in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, int unsigned in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, int long in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, int long unsigned in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, int long long in);
+    DOCTEST_INTERFACE void toStream(std::ostream* s, int long long unsigned in);
+
+    // ContextScope base class used to allow implementing methods of ContextScope 
+    // that don't depend on the template parameter in doctest.cpp.
+    class DOCTEST_INTERFACE ContextScopeBase : public IContextScope {
+    protected:
+        ContextScopeBase();
+
+        void destroy();
+    };
+
+    template <typename L> class ContextScope : public ContextScopeBase
+    {
+        const L &lambda_;
+
+    public:
+        explicit ContextScope(const L &lambda) : lambda_(lambda) {}
+
+        ContextScope(ContextScope &&other) : lambda_(other.lambda_) {}
+
+        void stringify(std::ostream* s) const override { lambda_(s); }
+
+        ~ContextScope() override { destroy(); }
+    };
+
+    struct DOCTEST_INTERFACE MessageBuilder : public MessageData
+    {
+        std::ostream* m_stream;
+
+        MessageBuilder(const char* file, int line, assertType::Enum severity);
+        MessageBuilder() = delete;
+        ~MessageBuilder();
+
+        template <typename T>
+        MessageBuilder& operator<<(const T& in) {
+            toStream(m_stream, in);
+            return *this;
+        }
+
+        bool log();
+        void react();
+    };
+    
+    template <typename L>
+    ContextScope<L> MakeContextScope(const L &lambda) {
+        return ContextScope<L>(lambda);
+    }
+} // namespace detail
+
+#define DOCTEST_DEFINE_DECORATOR(name, type, def)                                                  \
+    struct name                                                                                    \
+    {                                                                                              \
+        type data;                                                                                 \
+        name(type in = def)                                                                        \
+                : data(in) {}                                                                      \
+        void fill(detail::TestCase& state) const { state.DOCTEST_CAT(m_, name) = data; }           \
+        void fill(detail::TestSuite& state) const { state.DOCTEST_CAT(m_, name) = data; }          \
+    }
+
+DOCTEST_DEFINE_DECORATOR(test_suite, const char*, "");
+DOCTEST_DEFINE_DECORATOR(description, const char*, "");
+DOCTEST_DEFINE_DECORATOR(skip, bool, true);
+DOCTEST_DEFINE_DECORATOR(timeout, double, 0);
+DOCTEST_DEFINE_DECORATOR(may_fail, bool, true);
+DOCTEST_DEFINE_DECORATOR(should_fail, bool, true);
+DOCTEST_DEFINE_DECORATOR(expected_failures, int, 0);
+
+template <typename T>
+int registerExceptionTranslator(String (*translateFunction)(T)) {
+    DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Wexit-time-destructors")
+    static detail::ExceptionTranslator<T> exceptionTranslator(translateFunction);
+    DOCTEST_CLANG_SUPPRESS_WARNING_POP
+    detail::registerExceptionTranslatorImpl(&exceptionTranslator);
+    return 0;
+}
+
+} // namespace doctest
+
+// in a separate namespace outside of doctest because the DOCTEST_TEST_SUITE macro
+// introduces an anonymous namespace in which getCurrentTestSuite gets overridden
+namespace doctest_detail_test_suite_ns {
+DOCTEST_INTERFACE doctest::detail::TestSuite& getCurrentTestSuite();
+} // namespace doctest_detail_test_suite_ns
+
+namespace doctest {
+#else  // DOCTEST_CONFIG_DISABLE
+template <typename T>
+int registerExceptionTranslator(String (*)(T)) {
+    return 0;
+}
+#endif // DOCTEST_CONFIG_DISABLE
+
+namespace detail {
+    typedef void (*assert_handler)(const AssertData&);
+    struct ContextState;
+} // namespace detail
+
+class DOCTEST_INTERFACE Context
+{
+    detail::ContextState* p;
+
+    void parseArgs(int argc, const char* const* argv, bool withDefaults = false);
+
+public:
+    explicit Context(int argc = 0, const char* const* argv = nullptr);
+
+    ~Context();
+
+    void applyCommandLine(int argc, const char* const* argv);
+
+    void addFilter(const char* filter, const char* value);
+    void clearFilters();
+    void setOption(const char* option, int value);
+    void setOption(const char* option, const char* value);
+
+    bool shouldExit();
+
+    void setAsDefaultForAssertsOutOfTestCases();
+
+    void setAssertHandler(detail::assert_handler ah);
+
+    int run();
+};
+
+namespace TestCaseFailureReason {
+    enum Enum
+    {
+        None                     = 0,
+        AssertFailure            = 1,   // an assertion has failed in the test case
+        Exception                = 2,   // test case threw an exception
+        Crash                    = 4,   // a crash...
+        TooManyFailedAsserts     = 8,   // the abort-after option
+        Timeout                  = 16,  // see the timeout decorator
+        ShouldHaveFailedButDidnt = 32,  // see the should_fail decorator
+        ShouldHaveFailedAndDid   = 64,  // see the should_fail decorator
+        DidntFailExactlyNumTimes = 128, // see the expected_failures decorator
+        FailedExactlyNumTimes    = 256, // see the expected_failures decorator
+        CouldHaveFailedAndDid    = 512  // see the may_fail decorator
+    };
+} // namespace TestCaseFailureReason
+
+struct DOCTEST_INTERFACE CurrentTestCaseStats
+{
+    int    numAssertsCurrentTest;
+    int    numAssertsFailedCurrentTest;
+    double seconds;
+    int    failure_flags; // use TestCaseFailureReason::Enum
+};
+
+struct DOCTEST_INTERFACE TestCaseException
+{
+    String error_string;
+    bool   is_crash;
+};
+
+struct DOCTEST_INTERFACE TestRunStats
+{
+    unsigned numTestCases;
+    unsigned numTestCasesPassingFilters;
+    unsigned numTestSuitesPassingFilters;
+    unsigned numTestCasesFailed;
+    int      numAsserts;
+    int      numAssertsFailed;
+};
+
+struct QueryData
+{
+    const TestRunStats*  run_stats = nullptr;
+    const TestCaseData** data      = nullptr;
+    unsigned             num_data  = 0;
+};
+
+struct DOCTEST_INTERFACE IReporter
+{
+    // The constructor has to accept "const ContextOptions&" as a single argument
+    // which has most of the options for the run + a pointer to the stdout stream
+    // Reporter(const ContextOptions& in)
+
+    // called when a query should be reported (listing test cases, printing the version, etc.)
+    virtual void report_query(const QueryData&) = 0;
+
+    // called when the whole test run starts
+    virtual void test_run_start() = 0;
+    // called when the whole test run ends (caching a pointer to the input doesn't make sense here)
+    virtual void test_run_end(const TestRunStats&) = 0;
+
+    // called when a test case is started (safe to cache a pointer to the input)
+    virtual void test_case_start(const TestCaseData&) = 0;
+    // called when a test case is reentered because of unfinished subcases (safe to cache a pointer to the input)
+    virtual void test_case_reenter(const TestCaseData&) = 0;
+    // called when a test case has ended
+    virtual void test_case_end(const CurrentTestCaseStats&) = 0;
+
+    // called when an exception is thrown from the test case (or it crashes)
+    virtual void test_case_exception(const TestCaseException&) = 0;
+
+    // called whenever a subcase is entered (don't cache pointers to the input)
+    virtual void subcase_start(const SubcaseSignature&) = 0;
+    // called whenever a subcase is exited (don't cache pointers to the input)
+    virtual void subcase_end() = 0;
+
+    // called for each assert (don't cache pointers to the input)
+    virtual void log_assert(const AssertData&) = 0;
+    // called for each message (don't cache pointers to the input)
+    virtual void log_message(const MessageData&) = 0;
+
+    // called when a test case is skipped either because it doesn't pass the filters, has a skip decorator
+    // or isn't in the execution range (between first and last) (safe to cache a pointer to the input)
+    virtual void test_case_skipped(const TestCaseData&) = 0;
+
+    // doctest will not be managing the lifetimes of reporters given to it but this would still be nice to have
+    virtual ~IReporter();
+
+    // can obtain all currently active contexts and stringify them if one wishes to do so
+    static int                         get_num_active_contexts();
+    static const IContextScope* const* get_active_contexts();
+
+    // can iterate through contexts which have been stringified automatically in their destructors when an exception has been thrown
+    static int           get_num_stringified_contexts();
+    static const String* get_stringified_contexts();
+};
+
+namespace detail {
+    typedef IReporter* (*reporterCreatorFunc)(const ContextOptions&);
+
+    DOCTEST_INTERFACE void registerReporterImpl(const char* name, int prio, reporterCreatorFunc c, bool isReporter);
+
+    template <typename Reporter>
+    IReporter* reporterCreator(const ContextOptions& o) {
+        return new Reporter(o);
+    }
+} // namespace detail
+
+template <typename Reporter>
+int registerReporter(const char* name, int priority, bool isReporter) {
+    detail::registerReporterImpl(name, priority, detail::reporterCreator<Reporter>, isReporter);
+    return 0;
+}
+} // namespace doctest
+
+// if registering is not disabled
+#if !defined(DOCTEST_CONFIG_DISABLE)
+
+// common code in asserts - for convenience
+#define DOCTEST_ASSERT_LOG_AND_REACT(b)                                                            \
+    if(b.log())                                                                                    \
+        DOCTEST_BREAK_INTO_DEBUGGER();                                                             \
+    b.react()
+
+#ifdef DOCTEST_CONFIG_NO_TRY_CATCH_IN_ASSERTS
+#define DOCTEST_WRAP_IN_TRY(x) x;
+#else // DOCTEST_CONFIG_NO_TRY_CATCH_IN_ASSERTS
+#define DOCTEST_WRAP_IN_TRY(x)                                                                     \
+    try {                                                                                          \
+        x;                                                                                         \
+    } catch(...) { _DOCTEST_RB.translateException(); }
+#endif // DOCTEST_CONFIG_NO_TRY_CATCH_IN_ASSERTS
+
+#ifdef DOCTEST_CONFIG_VOID_CAST_EXPRESSIONS
+#define DOCTEST_CAST_TO_VOID(...)                                                                  \
+    DOCTEST_GCC_SUPPRESS_WARNING_WITH_PUSH("-Wuseless-cast")                                       \
+    static_cast<void>(__VA_ARGS__);                                                                \
+    DOCTEST_GCC_SUPPRESS_WARNING_POP
+#else // DOCTEST_CONFIG_VOID_CAST_EXPRESSIONS
+#define DOCTEST_CAST_TO_VOID(...) __VA_ARGS__;
+#endif // DOCTEST_CONFIG_VOID_CAST_EXPRESSIONS
+
+// registers the test by initializing a dummy var with a function
+#define DOCTEST_REGISTER_FUNCTION(global_prefix, f, decorators)                                    \
+    global_prefix DOCTEST_GLOBAL_NO_WARNINGS(DOCTEST_ANONYMOUS(_DOCTEST_ANON_VAR_)) =              \
+            doctest::detail::regTest(                                                              \
+                    doctest::detail::TestCase(                                                     \
+                            f, __FILE__, __LINE__,                                                 \
+                            doctest_detail_test_suite_ns::getCurrentTestSuite()) *                 \
+                    decorators);                                                                   \
+    DOCTEST_GLOBAL_NO_WARNINGS_END()
+
+#define DOCTEST_IMPLEMENT_FIXTURE(der, base, func, decorators)                                     \
+    namespace {                                                                                    \
+        struct der : public base                                                                   \
+        {                                                                                          \
+            void f();                                                                              \
+        };                                                                                         \
+        static void func() {                                                                       \
+            der v;                                                                                 \
+            v.f();                                                                                 \
+        }                                                                                          \
+        DOCTEST_REGISTER_FUNCTION(DOCTEST_EMPTY, func, decorators)                                 \
+    }                                                                                              \
+    inline DOCTEST_NOINLINE void der::f()
+
+#define DOCTEST_CREATE_AND_REGISTER_FUNCTION(f, decorators)                                        \
+    static void f();                                                                               \
+    DOCTEST_REGISTER_FUNCTION(DOCTEST_EMPTY, f, decorators)                                        \
+    static void f()
+
+#define DOCTEST_CREATE_AND_REGISTER_FUNCTION_IN_CLASS(f, proxy, decorators)                        \
+    static doctest::detail::funcType proxy() { return f; }                                         \
+    DOCTEST_REGISTER_FUNCTION(inline const, proxy(), decorators)                                   \
+    static void f()
+
+// for registering tests
+#define DOCTEST_TEST_CASE(decorators)                                                              \
+    DOCTEST_CREATE_AND_REGISTER_FUNCTION(DOCTEST_ANONYMOUS(_DOCTEST_ANON_FUNC_), decorators)
+
+// for registering tests in classes - requires C++17 for inline variables!
+#if __cplusplus >= 201703L || (DOCTEST_MSVC >= DOCTEST_COMPILER(19, 12, 0) && _MSVC_LANG >= 201703L)
+#define DOCTEST_TEST_CASE_CLASS(decorators)                                                        \
+    DOCTEST_CREATE_AND_REGISTER_FUNCTION_IN_CLASS(DOCTEST_ANONYMOUS(_DOCTEST_ANON_FUNC_),          \
+                                                  DOCTEST_ANONYMOUS(_DOCTEST_ANON_PROXY_),         \
+                                                  decorators)
+#else // DOCTEST_TEST_CASE_CLASS
+#define DOCTEST_TEST_CASE_CLASS(...)                                                               \
+    TEST_CASES_CAN_BE_REGISTERED_IN_CLASSES_ONLY_IN_CPP17_MODE_OR_WITH_VS_2017_OR_NEWER
+#endif // DOCTEST_TEST_CASE_CLASS
+
+// for registering tests with a fixture
+#define DOCTEST_TEST_CASE_FIXTURE(c, decorators)                                                   \
+    DOCTEST_IMPLEMENT_FIXTURE(DOCTEST_ANONYMOUS(_DOCTEST_ANON_CLASS_), c,                          \
+                              DOCTEST_ANONYMOUS(_DOCTEST_ANON_FUNC_), decorators)
+
+// for converting types to strings without the <typeinfo> header and demangling
+#define DOCTEST_TYPE_TO_STRING_IMPL(...)                                                           \
+    template <>                                                                                    \
+    inline const char* type_to_string<__VA_ARGS__>() {                                             \
+        return "<" #__VA_ARGS__ ">";                                                               \
+    }
+#define DOCTEST_TYPE_TO_STRING(...)                                                                \
+    namespace doctest { namespace detail {                                                         \
+            DOCTEST_TYPE_TO_STRING_IMPL(__VA_ARGS__)                                               \
+        }                                                                                          \
+    }                                                                                              \
+    typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+
+#define DOCTEST_TEST_CASE_TEMPLATE_DEFINE_IMPL(dec, T, iter, func)                                 \
+    template <typename T>                                                                          \
+    static void func();                                                                            \
+    namespace {                                                                                    \
+        template <typename Tuple>                                                                  \
+        struct iter;                                                                               \
+        template <typename Type, typename... Rest>                                                 \
+        struct iter<std::tuple<Type, Rest...>>                                                     \
+        {                                                                                          \
+            iter(const char* file, unsigned line, int index) {                                     \
+                doctest::detail::regTest(doctest::detail::TestCase(func<Type>, file, line,         \
+                                            doctest_detail_test_suite_ns::getCurrentTestSuite(),   \
+                                            doctest::detail::type_to_string<Type>(),               \
+                                            int(line) * 1000 + index)                              \
+                                         * dec);                                                   \
+                iter<std::tuple<Rest...>>(file, line, index + 1);                                  \
+            }                                                                                      \
+        };                                                                                         \
+        template <>                                                                                \
+        struct iter<std::tuple<>>                                                                  \
+        {                                                                                          \
+            iter(const char*, unsigned, int) {}                                                    \
+        };                                                                                         \
+    }                                                                                              \
+    template <typename T>                                                                          \
+    static void func()
+
+#define DOCTEST_TEST_CASE_TEMPLATE_DEFINE(dec, T, id)                                              \
+    DOCTEST_TEST_CASE_TEMPLATE_DEFINE_IMPL(dec, T, DOCTEST_CAT(id, ITERATOR),                      \
+                                           DOCTEST_ANONYMOUS(_DOCTEST_ANON_TMP_))
+
+#define DOCTEST_TEST_CASE_TEMPLATE_INSTANTIATE_IMPL(id, anon, ...)                                 \
+    DOCTEST_GLOBAL_NO_WARNINGS(DOCTEST_CAT(anon, DUMMY)) =                                         \
+        doctest::detail::instantiationHelper(DOCTEST_CAT(id, ITERATOR)<__VA_ARGS__>(__FILE__, __LINE__, 0));\
+    DOCTEST_GLOBAL_NO_WARNINGS_END()
+
+#define DOCTEST_TEST_CASE_TEMPLATE_INVOKE(id, ...)                                                 \
+    DOCTEST_TEST_CASE_TEMPLATE_INSTANTIATE_IMPL(id, DOCTEST_ANONYMOUS(_DOCTEST_ANON_TMP_), std::tuple<__VA_ARGS__>) \
+    typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+
+#define DOCTEST_TEST_CASE_TEMPLATE_APPLY(id, ...)                                                  \
+    DOCTEST_TEST_CASE_TEMPLATE_INSTANTIATE_IMPL(id, DOCTEST_ANONYMOUS(_DOCTEST_ANON_TMP_), __VA_ARGS__) \
+    typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+
+#define DOCTEST_TEST_CASE_TEMPLATE_IMPL(dec, T, anon, ...)                                         \
+    DOCTEST_TEST_CASE_TEMPLATE_DEFINE_IMPL(dec, T, DOCTEST_CAT(anon, ITERATOR), anon);             \
+    DOCTEST_TEST_CASE_TEMPLATE_INSTANTIATE_IMPL(anon, anon, std::tuple<__VA_ARGS__>)               \
+    template <typename T>                                                                          \
+    static void anon()
+
+#define DOCTEST_TEST_CASE_TEMPLATE(dec, T, ...)                                                    \
+    DOCTEST_TEST_CASE_TEMPLATE_IMPL(dec, T, DOCTEST_ANONYMOUS(_DOCTEST_ANON_TMP_), __VA_ARGS__)
+
+// for subcases
+#define DOCTEST_SUBCASE(name)                                                                      \
+    if(const doctest::detail::Subcase & DOCTEST_ANONYMOUS(_DOCTEST_ANON_SUBCASE_) DOCTEST_UNUSED = \
+               doctest::detail::Subcase(name, __FILE__, __LINE__))
+
+// for grouping tests in test suites by using code blocks
+#define DOCTEST_TEST_SUITE_IMPL(decorators, ns_name)                                               \
+    namespace ns_name { namespace doctest_detail_test_suite_ns {                                   \
+            static DOCTEST_NOINLINE doctest::detail::TestSuite& getCurrentTestSuite() {            \
+                DOCTEST_MSVC_SUPPRESS_WARNING_WITH_PUSH(4640)                                      \
+                DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Wexit-time-destructors")                \
+                static doctest::detail::TestSuite data;                                            \
+                static bool                       inited = false;                                  \
+                DOCTEST_MSVC_SUPPRESS_WARNING_POP                                                  \
+                DOCTEST_CLANG_SUPPRESS_WARNING_POP                                                 \
+                if(!inited) {                                                                      \
+                    data* decorators;                                                              \
+                    inited = true;                                                                 \
+                }                                                                                  \
+                return data;                                                                       \
+            }                                                                                      \
+        }                                                                                          \
+    }                                                                                              \
+    namespace ns_name
+
+#define DOCTEST_TEST_SUITE(decorators)                                                             \
+    DOCTEST_TEST_SUITE_IMPL(decorators, DOCTEST_ANONYMOUS(_DOCTEST_ANON_SUITE_))
+
+// for starting a testsuite block
+#define DOCTEST_TEST_SUITE_BEGIN(decorators)                                                       \
+    DOCTEST_GLOBAL_NO_WARNINGS(DOCTEST_ANONYMOUS(_DOCTEST_ANON_VAR_)) =                            \
+            doctest::detail::setTestSuite(doctest::detail::TestSuite() * decorators);              \
+    DOCTEST_GLOBAL_NO_WARNINGS_END()                                                               \
+    typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+
+// for ending a testsuite block
+#define DOCTEST_TEST_SUITE_END                                                                     \
+    DOCTEST_GLOBAL_NO_WARNINGS(DOCTEST_ANONYMOUS(_DOCTEST_ANON_VAR_)) =                            \
+            doctest::detail::setTestSuite(doctest::detail::TestSuite() * "");                      \
+    DOCTEST_GLOBAL_NO_WARNINGS_END()                                                               \
+    typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+
+// for registering exception translators
+#define DOCTEST_REGISTER_EXCEPTION_TRANSLATOR_IMPL(translatorName, signature)                      \
+    inline doctest::String translatorName(signature);                                              \
+    DOCTEST_GLOBAL_NO_WARNINGS(DOCTEST_ANONYMOUS(_DOCTEST_ANON_TRANSLATOR_)) =                     \
+            doctest::registerExceptionTranslator(translatorName);                                  \
+    DOCTEST_GLOBAL_NO_WARNINGS_END()                                                               \
+    doctest::String translatorName(signature)
+
+#define DOCTEST_REGISTER_EXCEPTION_TRANSLATOR(signature)                                           \
+    DOCTEST_REGISTER_EXCEPTION_TRANSLATOR_IMPL(DOCTEST_ANONYMOUS(_DOCTEST_ANON_TRANSLATOR_),       \
+                                               signature)
+
+// for registering reporters
+#define DOCTEST_REGISTER_REPORTER(name, priority, reporter)                                        \
+    DOCTEST_GLOBAL_NO_WARNINGS(DOCTEST_ANONYMOUS(_DOCTEST_ANON_REPORTER_)) =                       \
+            doctest::registerReporter<reporter>(name, priority, true);                             \
+    DOCTEST_GLOBAL_NO_WARNINGS_END() typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+
+// for registering listeners
+#define DOCTEST_REGISTER_LISTENER(name, priority, reporter)                                        \
+    DOCTEST_GLOBAL_NO_WARNINGS(DOCTEST_ANONYMOUS(_DOCTEST_ANON_REPORTER_)) =                       \
+            doctest::registerReporter<reporter>(name, priority, false);                            \
+    DOCTEST_GLOBAL_NO_WARNINGS_END() typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+
+// for logging
+#define DOCTEST_INFO(expression)                                                                   \
+    DOCTEST_INFO_IMPL(DOCTEST_ANONYMOUS(_DOCTEST_CAPTURE_), DOCTEST_ANONYMOUS(_DOCTEST_CAPTURE_),  \
+                      DOCTEST_ANONYMOUS(_DOCTEST_CAPTURE_), expression)
+
+#define DOCTEST_INFO_IMPL(lambda_name, mb_name, s_name, expression)                                \
+    DOCTEST_MSVC_SUPPRESS_WARNING_WITH_PUSH(4626)                                                  \
+    auto lambda_name = [&](std::ostream* s_name) {                                                 \
+        doctest::detail::MessageBuilder mb_name(__FILE__, __LINE__, doctest::assertType::is_warn); \
+        mb_name.m_stream = s_name;                                                                 \
+        mb_name << expression;                                                                     \
+    };                                                                                             \
+    DOCTEST_MSVC_SUPPRESS_WARNING_POP                                                              \
+    auto DOCTEST_ANONYMOUS(_DOCTEST_CAPTURE_) = doctest::detail::MakeContextScope(lambda_name)
+
+#define DOCTEST_CAPTURE(x) DOCTEST_INFO(#x " := " << x)
+
+#define DOCTEST_ADD_AT_IMPL(type, file, line, mb, x)                                               \
+    do {                                                                                           \
+        doctest::detail::MessageBuilder mb(file, line, doctest::assertType::type);                 \
+        mb << x;                                                                                   \
+        DOCTEST_ASSERT_LOG_AND_REACT(mb);                                                          \
+    } while(false)
+
+// clang-format off
+#define DOCTEST_ADD_MESSAGE_AT(file, line, x) DOCTEST_ADD_AT_IMPL(is_warn, file, line, DOCTEST_ANONYMOUS(_DOCTEST_MESSAGE_), x)
+#define DOCTEST_ADD_FAIL_CHECK_AT(file, line, x) DOCTEST_ADD_AT_IMPL(is_check, file, line, DOCTEST_ANONYMOUS(_DOCTEST_MESSAGE_), x)
+#define DOCTEST_ADD_FAIL_AT(file, line, x) DOCTEST_ADD_AT_IMPL(is_require, file, line, DOCTEST_ANONYMOUS(_DOCTEST_MESSAGE_), x)
+// clang-format on
+
+#define DOCTEST_MESSAGE(x) DOCTEST_ADD_MESSAGE_AT(__FILE__, __LINE__, x)
+#define DOCTEST_FAIL_CHECK(x) DOCTEST_ADD_FAIL_CHECK_AT(__FILE__, __LINE__, x)
+#define DOCTEST_FAIL(x) DOCTEST_ADD_FAIL_AT(__FILE__, __LINE__, x)
+
+#define DOCTEST_TO_LVALUE(...) __VA_ARGS__ // Not removed to keep backwards compatibility.
+
+#ifndef DOCTEST_CONFIG_SUPER_FAST_ASSERTS
+
+#define DOCTEST_ASSERT_IMPLEMENT_2(assert_type, ...)                                               \
+    DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Woverloaded-shift-op-parentheses")                  \
+    doctest::detail::ResultBuilder _DOCTEST_RB(doctest::assertType::assert_type, __FILE__,         \
+                                               __LINE__, #__VA_ARGS__);                            \
+    DOCTEST_WRAP_IN_TRY(_DOCTEST_RB.setResult(                                                     \
+            doctest::detail::ExpressionDecomposer(doctest::assertType::assert_type)                \
+            << __VA_ARGS__))                                                                       \
+    DOCTEST_ASSERT_LOG_AND_REACT(_DOCTEST_RB)                                                      \
+    DOCTEST_CLANG_SUPPRESS_WARNING_POP
+
+#define DOCTEST_ASSERT_IMPLEMENT_1(assert_type, ...)                                               \
+    do {                                                                                           \
+        DOCTEST_ASSERT_IMPLEMENT_2(assert_type, __VA_ARGS__);                                      \
+    } while(false)
+
+#else // DOCTEST_CONFIG_SUPER_FAST_ASSERTS
+
+// necessary for <ASSERT>_MESSAGE
+#define DOCTEST_ASSERT_IMPLEMENT_2 DOCTEST_ASSERT_IMPLEMENT_1
+
+#define DOCTEST_ASSERT_IMPLEMENT_1(assert_type, ...)                                               \
+    DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Woverloaded-shift-op-parentheses")                  \
+    doctest::detail::decomp_assert(                                                                \
+            doctest::assertType::assert_type, __FILE__, __LINE__, #__VA_ARGS__,                    \
+            doctest::detail::ExpressionDecomposer(doctest::assertType::assert_type)                \
+                    << __VA_ARGS__) DOCTEST_CLANG_SUPPRESS_WARNING_POP
+
+#endif // DOCTEST_CONFIG_SUPER_FAST_ASSERTS
+
+#define DOCTEST_WARN(...) DOCTEST_ASSERT_IMPLEMENT_1(DT_WARN, __VA_ARGS__)
+#define DOCTEST_CHECK(...) DOCTEST_ASSERT_IMPLEMENT_1(DT_CHECK, __VA_ARGS__)
+#define DOCTEST_REQUIRE(...) DOCTEST_ASSERT_IMPLEMENT_1(DT_REQUIRE, __VA_ARGS__)
+#define DOCTEST_WARN_FALSE(...) DOCTEST_ASSERT_IMPLEMENT_1(DT_WARN_FALSE, __VA_ARGS__)
+#define DOCTEST_CHECK_FALSE(...) DOCTEST_ASSERT_IMPLEMENT_1(DT_CHECK_FALSE, __VA_ARGS__)
+#define DOCTEST_REQUIRE_FALSE(...) DOCTEST_ASSERT_IMPLEMENT_1(DT_REQUIRE_FALSE, __VA_ARGS__)
+
+// clang-format off
+#define DOCTEST_WARN_MESSAGE(cond, msg) do { DOCTEST_INFO(msg); DOCTEST_ASSERT_IMPLEMENT_2(DT_WARN, cond); } while(false)
+#define DOCTEST_CHECK_MESSAGE(cond, msg) do { DOCTEST_INFO(msg); DOCTEST_ASSERT_IMPLEMENT_2(DT_CHECK, cond); } while(false)
+#define DOCTEST_REQUIRE_MESSAGE(cond, msg) do { DOCTEST_INFO(msg); DOCTEST_ASSERT_IMPLEMENT_2(DT_REQUIRE, cond); } while(false)
+#define DOCTEST_WARN_FALSE_MESSAGE(cond, msg) do { DOCTEST_INFO(msg); DOCTEST_ASSERT_IMPLEMENT_2(DT_WARN_FALSE, cond); } while(false)
+#define DOCTEST_CHECK_FALSE_MESSAGE(cond, msg) do { DOCTEST_INFO(msg); DOCTEST_ASSERT_IMPLEMENT_2(DT_CHECK_FALSE, cond); } while(false)
+#define DOCTEST_REQUIRE_FALSE_MESSAGE(cond, msg) do { DOCTEST_INFO(msg); DOCTEST_ASSERT_IMPLEMENT_2(DT_REQUIRE_FALSE, cond); } while(false)
+// clang-format on
+
+#define DOCTEST_ASSERT_THROWS_AS(expr, assert_type, message, ...)                                  \
+    do {                                                                                           \
+        if(!doctest::getContextOptions()->no_throw) {                                              \
+            doctest::detail::ResultBuilder _DOCTEST_RB(doctest::assertType::assert_type, __FILE__, \
+                                                       __LINE__, #expr, #__VA_ARGS__, message);    \
+            try {                                                                                  \
+                DOCTEST_CAST_TO_VOID(expr)                                                         \
+            } catch(const doctest::detail::remove_const<                                           \
+                    doctest::detail::remove_reference<__VA_ARGS__>::type>::type&) {                \
+                _DOCTEST_RB.translateException();                                                  \
+                _DOCTEST_RB.m_threw_as = true;                                                     \
+            } catch(...) { _DOCTEST_RB.translateException(); }                                     \
+            DOCTEST_ASSERT_LOG_AND_REACT(_DOCTEST_RB);                                             \
+        }                                                                                          \
+    } while(false)
+
+#define DOCTEST_ASSERT_THROWS_WITH(expr, expr_str, assert_type, ...)                               \
+    do {                                                                                           \
+        if(!doctest::getContextOptions()->no_throw) {                                              \
+            doctest::detail::ResultBuilder _DOCTEST_RB(doctest::assertType::assert_type, __FILE__, \
+                                                       __LINE__, expr_str, "", __VA_ARGS__);       \
+            try {                                                                                  \
+                DOCTEST_CAST_TO_VOID(expr)                                                         \
+            } catch(...) { _DOCTEST_RB.translateException(); }                                     \
+            DOCTEST_ASSERT_LOG_AND_REACT(_DOCTEST_RB);                                             \
+        }                                                                                          \
+    } while(false)
+
+#define DOCTEST_ASSERT_NOTHROW(assert_type, ...)                                                   \
+    do {                                                                                           \
+        doctest::detail::ResultBuilder _DOCTEST_RB(doctest::assertType::assert_type, __FILE__,     \
+                                                   __LINE__, #__VA_ARGS__);                        \
+        try {                                                                                      \
+            DOCTEST_CAST_TO_VOID(__VA_ARGS__)                                                      \
+        } catch(...) { _DOCTEST_RB.translateException(); }                                         \
+        DOCTEST_ASSERT_LOG_AND_REACT(_DOCTEST_RB);                                                 \
+    } while(false)
+
+// clang-format off
+#define DOCTEST_WARN_THROWS(...) DOCTEST_ASSERT_THROWS_WITH((__VA_ARGS__), #__VA_ARGS__, DT_WARN_THROWS, "")
+#define DOCTEST_CHECK_THROWS(...) DOCTEST_ASSERT_THROWS_WITH((__VA_ARGS__), #__VA_ARGS__, DT_CHECK_THROWS, "")
+#define DOCTEST_REQUIRE_THROWS(...) DOCTEST_ASSERT_THROWS_WITH((__VA_ARGS__), #__VA_ARGS__, DT_REQUIRE_THROWS, "")
+
+#define DOCTEST_WARN_THROWS_AS(expr, ...) DOCTEST_ASSERT_THROWS_AS(expr, DT_WARN_THROWS_AS, "", __VA_ARGS__)
+#define DOCTEST_CHECK_THROWS_AS(expr, ...) DOCTEST_ASSERT_THROWS_AS(expr, DT_CHECK_THROWS_AS, "", __VA_ARGS__)
+#define DOCTEST_REQUIRE_THROWS_AS(expr, ...) DOCTEST_ASSERT_THROWS_AS(expr, DT_REQUIRE_THROWS_AS, "", __VA_ARGS__)
+
+#define DOCTEST_WARN_THROWS_WITH(expr, ...) DOCTEST_ASSERT_THROWS_WITH(expr, #expr, DT_WARN_THROWS_WITH, __VA_ARGS__)
+#define DOCTEST_CHECK_THROWS_WITH(expr, ...) DOCTEST_ASSERT_THROWS_WITH(expr, #expr, DT_CHECK_THROWS_WITH, __VA_ARGS__)
+#define DOCTEST_REQUIRE_THROWS_WITH(expr, ...) DOCTEST_ASSERT_THROWS_WITH(expr, #expr, DT_REQUIRE_THROWS_WITH, __VA_ARGS__)
+
+#define DOCTEST_WARN_THROWS_WITH_AS(expr, message, ...) DOCTEST_ASSERT_THROWS_AS(expr, DT_WARN_THROWS_WITH_AS, message, __VA_ARGS__)
+#define DOCTEST_CHECK_THROWS_WITH_AS(expr, message, ...) DOCTEST_ASSERT_THROWS_AS(expr, DT_CHECK_THROWS_WITH_AS, message, __VA_ARGS__)
+#define DOCTEST_REQUIRE_THROWS_WITH_AS(expr, message, ...) DOCTEST_ASSERT_THROWS_AS(expr, DT_REQUIRE_THROWS_WITH_AS, message, __VA_ARGS__)
+
+#define DOCTEST_WARN_NOTHROW(...) DOCTEST_ASSERT_NOTHROW(DT_WARN_NOTHROW, __VA_ARGS__)
+#define DOCTEST_CHECK_NOTHROW(...) DOCTEST_ASSERT_NOTHROW(DT_CHECK_NOTHROW, __VA_ARGS__)
+#define DOCTEST_REQUIRE_NOTHROW(...) DOCTEST_ASSERT_NOTHROW(DT_REQUIRE_NOTHROW, __VA_ARGS__)
+
+#define DOCTEST_WARN_THROWS_MESSAGE(expr, msg) do { DOCTEST_INFO(msg); DOCTEST_WARN_THROWS(expr); } while(false)
+#define DOCTEST_CHECK_THROWS_MESSAGE(expr, msg) do { DOCTEST_INFO(msg); DOCTEST_CHECK_THROWS(expr); } while(false)
+#define DOCTEST_REQUIRE_THROWS_MESSAGE(expr, msg) do { DOCTEST_INFO(msg); DOCTEST_REQUIRE_THROWS(expr); } while(false)
+#define DOCTEST_WARN_THROWS_AS_MESSAGE(expr, ex, msg) do { DOCTEST_INFO(msg); DOCTEST_WARN_THROWS_AS(expr, ex); } while(false)
+#define DOCTEST_CHECK_THROWS_AS_MESSAGE(expr, ex, msg) do { DOCTEST_INFO(msg); DOCTEST_CHECK_THROWS_AS(expr, ex); } while(false)
+#define DOCTEST_REQUIRE_THROWS_AS_MESSAGE(expr, ex, msg) do { DOCTEST_INFO(msg); DOCTEST_REQUIRE_THROWS_AS(expr, ex); } while(false)
+#define DOCTEST_WARN_THROWS_WITH_MESSAGE(expr, with, msg) do { DOCTEST_INFO(msg); DOCTEST_WARN_THROWS_WITH(expr, with); } while(false)
+#define DOCTEST_CHECK_THROWS_WITH_MESSAGE(expr, with, msg) do { DOCTEST_INFO(msg); DOCTEST_CHECK_THROWS_WITH(expr, with); } while(false)
+#define DOCTEST_REQUIRE_THROWS_WITH_MESSAGE(expr, with, msg) do { DOCTEST_INFO(msg); DOCTEST_REQUIRE_THROWS_WITH(expr, with); } while(false)
+#define DOCTEST_WARN_THROWS_WITH_AS_MESSAGE(expr, with, ex, msg) do { DOCTEST_INFO(msg); DOCTEST_WARN_THROWS_WITH_AS(expr, with, ex); } while(false)
+#define DOCTEST_CHECK_THROWS_WITH_AS_MESSAGE(expr, with, ex, msg) do { DOCTEST_INFO(msg); DOCTEST_CHECK_THROWS_WITH_AS(expr, with, ex); } while(false)
+#define DOCTEST_REQUIRE_THROWS_WITH_AS_MESSAGE(expr, with, ex, msg) do { DOCTEST_INFO(msg); DOCTEST_REQUIRE_THROWS_WITH_AS(expr, with, ex); } while(false)
+#define DOCTEST_WARN_NOTHROW_MESSAGE(expr, msg) do { DOCTEST_INFO(msg); DOCTEST_WARN_NOTHROW(expr); } while(false)
+#define DOCTEST_CHECK_NOTHROW_MESSAGE(expr, msg) do { DOCTEST_INFO(msg); DOCTEST_CHECK_NOTHROW(expr); } while(false)
+#define DOCTEST_REQUIRE_NOTHROW_MESSAGE(expr, msg) do { DOCTEST_INFO(msg); DOCTEST_REQUIRE_NOTHROW(expr); } while(false)
+// clang-format on
+
+#ifndef DOCTEST_CONFIG_SUPER_FAST_ASSERTS
+
+#define DOCTEST_BINARY_ASSERT(assert_type, comp, ...)                                              \
+    do {                                                                                           \
+        doctest::detail::ResultBuilder _DOCTEST_RB(doctest::assertType::assert_type, __FILE__,     \
+                                                   __LINE__, #__VA_ARGS__);                        \
+        DOCTEST_WRAP_IN_TRY(                                                                       \
+                _DOCTEST_RB.binary_assert<doctest::detail::binaryAssertComparison::comp>(          \
+                        __VA_ARGS__))                                                              \
+        DOCTEST_ASSERT_LOG_AND_REACT(_DOCTEST_RB);                                                 \
+    } while(false)
+
+#define DOCTEST_UNARY_ASSERT(assert_type, ...)                                                     \
+    do {                                                                                           \
+        doctest::detail::ResultBuilder _DOCTEST_RB(doctest::assertType::assert_type, __FILE__,     \
+                                                   __LINE__, #__VA_ARGS__);                        \
+        DOCTEST_WRAP_IN_TRY(_DOCTEST_RB.unary_assert(__VA_ARGS__))                                 \
+        DOCTEST_ASSERT_LOG_AND_REACT(_DOCTEST_RB);                                                 \
+    } while(false)
+
+#else // DOCTEST_CONFIG_SUPER_FAST_ASSERTS
+
+#define DOCTEST_BINARY_ASSERT(assert_type, comparison, ...)                                        \
+    doctest::detail::binary_assert<doctest::detail::binaryAssertComparison::comparison>(           \
+            doctest::assertType::assert_type, __FILE__, __LINE__, #__VA_ARGS__, __VA_ARGS__)
+
+#define DOCTEST_UNARY_ASSERT(assert_type, ...)                                                     \
+    doctest::detail::unary_assert(doctest::assertType::assert_type, __FILE__, __LINE__,            \
+                                  #__VA_ARGS__, __VA_ARGS__)
+
+#endif // DOCTEST_CONFIG_SUPER_FAST_ASSERTS
+
+#define DOCTEST_WARN_EQ(...) DOCTEST_BINARY_ASSERT(DT_WARN_EQ, eq, __VA_ARGS__)
+#define DOCTEST_CHECK_EQ(...) DOCTEST_BINARY_ASSERT(DT_CHECK_EQ, eq, __VA_ARGS__)
+#define DOCTEST_REQUIRE_EQ(...) DOCTEST_BINARY_ASSERT(DT_REQUIRE_EQ, eq, __VA_ARGS__)
+#define DOCTEST_WARN_NE(...) DOCTEST_BINARY_ASSERT(DT_WARN_NE, ne, __VA_ARGS__)
+#define DOCTEST_CHECK_NE(...) DOCTEST_BINARY_ASSERT(DT_CHECK_NE, ne, __VA_ARGS__)
+#define DOCTEST_REQUIRE_NE(...) DOCTEST_BINARY_ASSERT(DT_REQUIRE_NE, ne, __VA_ARGS__)
+#define DOCTEST_WARN_GT(...) DOCTEST_BINARY_ASSERT(DT_WARN_GT, gt, __VA_ARGS__)
+#define DOCTEST_CHECK_GT(...) DOCTEST_BINARY_ASSERT(DT_CHECK_GT, gt, __VA_ARGS__)
+#define DOCTEST_REQUIRE_GT(...) DOCTEST_BINARY_ASSERT(DT_REQUIRE_GT, gt, __VA_ARGS__)
+#define DOCTEST_WARN_LT(...) DOCTEST_BINARY_ASSERT(DT_WARN_LT, lt, __VA_ARGS__)
+#define DOCTEST_CHECK_LT(...) DOCTEST_BINARY_ASSERT(DT_CHECK_LT, lt, __VA_ARGS__)
+#define DOCTEST_REQUIRE_LT(...) DOCTEST_BINARY_ASSERT(DT_REQUIRE_LT, lt, __VA_ARGS__)
+#define DOCTEST_WARN_GE(...) DOCTEST_BINARY_ASSERT(DT_WARN_GE, ge, __VA_ARGS__)
+#define DOCTEST_CHECK_GE(...) DOCTEST_BINARY_ASSERT(DT_CHECK_GE, ge, __VA_ARGS__)
+#define DOCTEST_REQUIRE_GE(...) DOCTEST_BINARY_ASSERT(DT_REQUIRE_GE, ge, __VA_ARGS__)
+#define DOCTEST_WARN_LE(...) DOCTEST_BINARY_ASSERT(DT_WARN_LE, le, __VA_ARGS__)
+#define DOCTEST_CHECK_LE(...) DOCTEST_BINARY_ASSERT(DT_CHECK_LE, le, __VA_ARGS__)
+#define DOCTEST_REQUIRE_LE(...) DOCTEST_BINARY_ASSERT(DT_REQUIRE_LE, le, __VA_ARGS__)
+
+#define DOCTEST_WARN_UNARY(...) DOCTEST_UNARY_ASSERT(DT_WARN_UNARY, __VA_ARGS__)
+#define DOCTEST_CHECK_UNARY(...) DOCTEST_UNARY_ASSERT(DT_CHECK_UNARY, __VA_ARGS__)
+#define DOCTEST_REQUIRE_UNARY(...) DOCTEST_UNARY_ASSERT(DT_REQUIRE_UNARY, __VA_ARGS__)
+#define DOCTEST_WARN_UNARY_FALSE(...) DOCTEST_UNARY_ASSERT(DT_WARN_UNARY_FALSE, __VA_ARGS__)
+#define DOCTEST_CHECK_UNARY_FALSE(...) DOCTEST_UNARY_ASSERT(DT_CHECK_UNARY_FALSE, __VA_ARGS__)
+#define DOCTEST_REQUIRE_UNARY_FALSE(...) DOCTEST_UNARY_ASSERT(DT_REQUIRE_UNARY_FALSE, __VA_ARGS__)
+
+#ifdef DOCTEST_CONFIG_NO_EXCEPTIONS
+
+#undef DOCTEST_WARN_THROWS
+#undef DOCTEST_CHECK_THROWS
+#undef DOCTEST_REQUIRE_THROWS
+#undef DOCTEST_WARN_THROWS_AS
+#undef DOCTEST_CHECK_THROWS_AS
+#undef DOCTEST_REQUIRE_THROWS_AS
+#undef DOCTEST_WARN_THROWS_WITH
+#undef DOCTEST_CHECK_THROWS_WITH
+#undef DOCTEST_REQUIRE_THROWS_WITH
+#undef DOCTEST_WARN_THROWS_WITH_AS
+#undef DOCTEST_CHECK_THROWS_WITH_AS
+#undef DOCTEST_REQUIRE_THROWS_WITH_AS
+#undef DOCTEST_WARN_NOTHROW
+#undef DOCTEST_CHECK_NOTHROW
+#undef DOCTEST_REQUIRE_NOTHROW
+
+#undef DOCTEST_WARN_THROWS_MESSAGE
+#undef DOCTEST_CHECK_THROWS_MESSAGE
+#undef DOCTEST_REQUIRE_THROWS_MESSAGE
+#undef DOCTEST_WARN_THROWS_AS_MESSAGE
+#undef DOCTEST_CHECK_THROWS_AS_MESSAGE
+#undef DOCTEST_REQUIRE_THROWS_AS_MESSAGE
+#undef DOCTEST_WARN_THROWS_WITH_MESSAGE
+#undef DOCTEST_CHECK_THROWS_WITH_MESSAGE
+#undef DOCTEST_REQUIRE_THROWS_WITH_MESSAGE
+#undef DOCTEST_WARN_THROWS_WITH_AS_MESSAGE
+#undef DOCTEST_CHECK_THROWS_WITH_AS_MESSAGE
+#undef DOCTEST_REQUIRE_THROWS_WITH_AS_MESSAGE
+#undef DOCTEST_WARN_NOTHROW_MESSAGE
+#undef DOCTEST_CHECK_NOTHROW_MESSAGE
+#undef DOCTEST_REQUIRE_NOTHROW_MESSAGE
+
+#ifdef DOCTEST_CONFIG_NO_EXCEPTIONS_BUT_WITH_ALL_ASSERTS
+
+#define DOCTEST_WARN_THROWS(...) ((void)0)
+#define DOCTEST_CHECK_THROWS(...) ((void)0)
+#define DOCTEST_REQUIRE_THROWS(...) ((void)0)
+#define DOCTEST_WARN_THROWS_AS(expr, ...) ((void)0)
+#define DOCTEST_CHECK_THROWS_AS(expr, ...) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_AS(expr, ...) ((void)0)
+#define DOCTEST_WARN_THROWS_WITH(expr, ...) ((void)0)
+#define DOCTEST_CHECK_THROWS_WITH(expr, ...) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_WITH(expr, ...) ((void)0)
+#define DOCTEST_WARN_THROWS_WITH_AS(expr, with, ...) ((void)0)
+#define DOCTEST_CHECK_THROWS_WITH_AS(expr, with, ...) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_WITH_AS(expr, with, ...) ((void)0)
+#define DOCTEST_WARN_NOTHROW(...) ((void)0)
+#define DOCTEST_CHECK_NOTHROW(...) ((void)0)
+#define DOCTEST_REQUIRE_NOTHROW(...) ((void)0)
+
+#define DOCTEST_WARN_THROWS_MESSAGE(expr, msg) ((void)0)
+#define DOCTEST_CHECK_THROWS_MESSAGE(expr, msg) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_MESSAGE(expr, msg) ((void)0)
+#define DOCTEST_WARN_THROWS_AS_MESSAGE(expr, ex, msg) ((void)0)
+#define DOCTEST_CHECK_THROWS_AS_MESSAGE(expr, ex, msg) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_AS_MESSAGE(expr, ex, msg) ((void)0)
+#define DOCTEST_WARN_THROWS_WITH_MESSAGE(expr, with, msg) ((void)0)
+#define DOCTEST_CHECK_THROWS_WITH_MESSAGE(expr, with, msg) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_WITH_MESSAGE(expr, with, msg) ((void)0)
+#define DOCTEST_WARN_THROWS_WITH_AS_MESSAGE(expr, with, ex, msg) ((void)0)
+#define DOCTEST_CHECK_THROWS_WITH_AS_MESSAGE(expr, with, ex, msg) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_WITH_AS_MESSAGE(expr, with, ex, msg) ((void)0)
+#define DOCTEST_WARN_NOTHROW_MESSAGE(expr, msg) ((void)0)
+#define DOCTEST_CHECK_NOTHROW_MESSAGE(expr, msg) ((void)0)
+#define DOCTEST_REQUIRE_NOTHROW_MESSAGE(expr, msg) ((void)0)
+
+#else // DOCTEST_CONFIG_NO_EXCEPTIONS_BUT_WITH_ALL_ASSERTS
+
+#undef DOCTEST_REQUIRE
+#undef DOCTEST_REQUIRE_FALSE
+#undef DOCTEST_REQUIRE_MESSAGE
+#undef DOCTEST_REQUIRE_FALSE_MESSAGE
+#undef DOCTEST_REQUIRE_EQ
+#undef DOCTEST_REQUIRE_NE
+#undef DOCTEST_REQUIRE_GT
+#undef DOCTEST_REQUIRE_LT
+#undef DOCTEST_REQUIRE_GE
+#undef DOCTEST_REQUIRE_LE
+#undef DOCTEST_REQUIRE_UNARY
+#undef DOCTEST_REQUIRE_UNARY_FALSE
+
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS_BUT_WITH_ALL_ASSERTS
+
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS
+
+// =================================================================================================
+// == WHAT FOLLOWS IS VERSIONS OF THE MACROS THAT DO NOT DO ANY REGISTERING!                      ==
+// == THIS CAN BE ENABLED BY DEFINING DOCTEST_CONFIG_DISABLE GLOBALLY!                            ==
+// =================================================================================================
+#else // DOCTEST_CONFIG_DISABLE
+
+#define DOCTEST_IMPLEMENT_FIXTURE(der, base, func, name)                                           \
+    namespace {                                                                                    \
+        template <typename DOCTEST_UNUSED_TEMPLATE_TYPE>                                           \
+        struct der : public base                                                                   \
+        { void f(); };                                                                             \
+    }                                                                                              \
+    template <typename DOCTEST_UNUSED_TEMPLATE_TYPE>                                               \
+    inline void der<DOCTEST_UNUSED_TEMPLATE_TYPE>::f()
+
+#define DOCTEST_CREATE_AND_REGISTER_FUNCTION(f, name)                                              \
+    template <typename DOCTEST_UNUSED_TEMPLATE_TYPE>                                               \
+    static inline void f()
+
+// for registering tests
+#define DOCTEST_TEST_CASE(name)                                                                    \
+    DOCTEST_CREATE_AND_REGISTER_FUNCTION(DOCTEST_ANONYMOUS(_DOCTEST_ANON_FUNC_), name)
+
+// for registering tests in classes
+#define DOCTEST_TEST_CASE_CLASS(name)                                                              \
+    DOCTEST_CREATE_AND_REGISTER_FUNCTION(DOCTEST_ANONYMOUS(_DOCTEST_ANON_FUNC_), name)
+
+// for registering tests with a fixture
+#define DOCTEST_TEST_CASE_FIXTURE(x, name)                                                         \
+    DOCTEST_IMPLEMENT_FIXTURE(DOCTEST_ANONYMOUS(_DOCTEST_ANON_CLASS_), x,                          \
+                              DOCTEST_ANONYMOUS(_DOCTEST_ANON_FUNC_), name)
+
+// for converting types to strings without the <typeinfo> header and demangling
+#define DOCTEST_TYPE_TO_STRING(...) typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+#define DOCTEST_TYPE_TO_STRING_IMPL(...)
+
+// for typed tests
+#define DOCTEST_TEST_CASE_TEMPLATE(name, type, ...)                                                \
+    template <typename type>                                                                       \
+    inline void DOCTEST_ANONYMOUS(_DOCTEST_ANON_TMP_)()
+
+#define DOCTEST_TEST_CASE_TEMPLATE_DEFINE(name, type, id)                                          \
+    template <typename type>                                                                       \
+    inline void DOCTEST_ANONYMOUS(_DOCTEST_ANON_TMP_)()
+
+#define DOCTEST_TEST_CASE_TEMPLATE_INVOKE(id, ...)                                                 \
+    typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+
+#define DOCTEST_TEST_CASE_TEMPLATE_APPLY(id, ...)                                                  \
+    typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+
+// for subcases
+#define DOCTEST_SUBCASE(name)
+
+// for a testsuite block
+#define DOCTEST_TEST_SUITE(name) namespace
+
+// for starting a testsuite block
+#define DOCTEST_TEST_SUITE_BEGIN(name) typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+
+// for ending a testsuite block
+#define DOCTEST_TEST_SUITE_END typedef int DOCTEST_ANONYMOUS(_DOCTEST_ANON_FOR_SEMICOLON_)
+
+#define DOCTEST_REGISTER_EXCEPTION_TRANSLATOR(signature)                                           \
+    template <typename DOCTEST_UNUSED_TEMPLATE_TYPE>                                               \
+    static inline doctest::String DOCTEST_ANONYMOUS(_DOCTEST_ANON_TRANSLATOR_)(signature)
+
+#define DOCTEST_REGISTER_REPORTER(name, priority, reporter)
+#define DOCTEST_REGISTER_LISTENER(name, priority, reporter)
+
+#define DOCTEST_INFO(x) ((void)0)
+#define DOCTEST_CAPTURE(x) ((void)0)
+#define DOCTEST_ADD_MESSAGE_AT(file, line, x) ((void)0)
+#define DOCTEST_ADD_FAIL_CHECK_AT(file, line, x) ((void)0)
+#define DOCTEST_ADD_FAIL_AT(file, line, x) ((void)0)
+#define DOCTEST_MESSAGE(x) ((void)0)
+#define DOCTEST_FAIL_CHECK(x) ((void)0)
+#define DOCTEST_FAIL(x) ((void)0)
+
+#define DOCTEST_WARN(...) ((void)0)
+#define DOCTEST_CHECK(...) ((void)0)
+#define DOCTEST_REQUIRE(...) ((void)0)
+#define DOCTEST_WARN_FALSE(...) ((void)0)
+#define DOCTEST_CHECK_FALSE(...) ((void)0)
+#define DOCTEST_REQUIRE_FALSE(...) ((void)0)
+
+#define DOCTEST_WARN_MESSAGE(cond, msg) ((void)0)
+#define DOCTEST_CHECK_MESSAGE(cond, msg) ((void)0)
+#define DOCTEST_REQUIRE_MESSAGE(cond, msg) ((void)0)
+#define DOCTEST_WARN_FALSE_MESSAGE(cond, msg) ((void)0)
+#define DOCTEST_CHECK_FALSE_MESSAGE(cond, msg) ((void)0)
+#define DOCTEST_REQUIRE_FALSE_MESSAGE(cond, msg) ((void)0)
+
+#define DOCTEST_WARN_THROWS(...) ((void)0)
+#define DOCTEST_CHECK_THROWS(...) ((void)0)
+#define DOCTEST_REQUIRE_THROWS(...) ((void)0)
+#define DOCTEST_WARN_THROWS_AS(expr, ...) ((void)0)
+#define DOCTEST_CHECK_THROWS_AS(expr, ...) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_AS(expr, ...) ((void)0)
+#define DOCTEST_WARN_THROWS_WITH(expr, ...) ((void)0)
+#define DOCTEST_CHECK_THROWS_WITH(expr, ...) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_WITH(expr, ...) ((void)0)
+#define DOCTEST_WARN_THROWS_WITH_AS(expr, with, ...) ((void)0)
+#define DOCTEST_CHECK_THROWS_WITH_AS(expr, with, ...) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_WITH_AS(expr, with, ...) ((void)0)
+#define DOCTEST_WARN_NOTHROW(...) ((void)0)
+#define DOCTEST_CHECK_NOTHROW(...) ((void)0)
+#define DOCTEST_REQUIRE_NOTHROW(...) ((void)0)
+
+#define DOCTEST_WARN_THROWS_MESSAGE(expr, msg) ((void)0)
+#define DOCTEST_CHECK_THROWS_MESSAGE(expr, msg) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_MESSAGE(expr, msg) ((void)0)
+#define DOCTEST_WARN_THROWS_AS_MESSAGE(expr, ex, msg) ((void)0)
+#define DOCTEST_CHECK_THROWS_AS_MESSAGE(expr, ex, msg) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_AS_MESSAGE(expr, ex, msg) ((void)0)
+#define DOCTEST_WARN_THROWS_WITH_MESSAGE(expr, with, msg) ((void)0)
+#define DOCTEST_CHECK_THROWS_WITH_MESSAGE(expr, with, msg) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_WITH_MESSAGE(expr, with, msg) ((void)0)
+#define DOCTEST_WARN_THROWS_WITH_AS_MESSAGE(expr, with, ex, msg) ((void)0)
+#define DOCTEST_CHECK_THROWS_WITH_AS_MESSAGE(expr, with, ex, msg) ((void)0)
+#define DOCTEST_REQUIRE_THROWS_WITH_AS_MESSAGE(expr, with, ex, msg) ((void)0)
+#define DOCTEST_WARN_NOTHROW_MESSAGE(expr, msg) ((void)0)
+#define DOCTEST_CHECK_NOTHROW_MESSAGE(expr, msg) ((void)0)
+#define DOCTEST_REQUIRE_NOTHROW_MESSAGE(expr, msg) ((void)0)
+
+#define DOCTEST_WARN_EQ(...) ((void)0)
+#define DOCTEST_CHECK_EQ(...) ((void)0)
+#define DOCTEST_REQUIRE_EQ(...) ((void)0)
+#define DOCTEST_WARN_NE(...) ((void)0)
+#define DOCTEST_CHECK_NE(...) ((void)0)
+#define DOCTEST_REQUIRE_NE(...) ((void)0)
+#define DOCTEST_WARN_GT(...) ((void)0)
+#define DOCTEST_CHECK_GT(...) ((void)0)
+#define DOCTEST_REQUIRE_GT(...) ((void)0)
+#define DOCTEST_WARN_LT(...) ((void)0)
+#define DOCTEST_CHECK_LT(...) ((void)0)
+#define DOCTEST_REQUIRE_LT(...) ((void)0)
+#define DOCTEST_WARN_GE(...) ((void)0)
+#define DOCTEST_CHECK_GE(...) ((void)0)
+#define DOCTEST_REQUIRE_GE(...) ((void)0)
+#define DOCTEST_WARN_LE(...) ((void)0)
+#define DOCTEST_CHECK_LE(...) ((void)0)
+#define DOCTEST_REQUIRE_LE(...) ((void)0)
+
+#define DOCTEST_WARN_UNARY(...) ((void)0)
+#define DOCTEST_CHECK_UNARY(...) ((void)0)
+#define DOCTEST_REQUIRE_UNARY(...) ((void)0)
+#define DOCTEST_WARN_UNARY_FALSE(...) ((void)0)
+#define DOCTEST_CHECK_UNARY_FALSE(...) ((void)0)
+#define DOCTEST_REQUIRE_UNARY_FALSE(...) ((void)0)
+
+#endif // DOCTEST_CONFIG_DISABLE
+
+// clang-format off
+// KEPT FOR BACKWARDS COMPATIBILITY - FORWARDING TO THE RIGHT MACROS
+#define DOCTEST_FAST_WARN_EQ             DOCTEST_WARN_EQ
+#define DOCTEST_FAST_CHECK_EQ            DOCTEST_CHECK_EQ
+#define DOCTEST_FAST_REQUIRE_EQ          DOCTEST_REQUIRE_EQ
+#define DOCTEST_FAST_WARN_NE             DOCTEST_WARN_NE
+#define DOCTEST_FAST_CHECK_NE            DOCTEST_CHECK_NE
+#define DOCTEST_FAST_REQUIRE_NE          DOCTEST_REQUIRE_NE
+#define DOCTEST_FAST_WARN_GT             DOCTEST_WARN_GT
+#define DOCTEST_FAST_CHECK_GT            DOCTEST_CHECK_GT
+#define DOCTEST_FAST_REQUIRE_GT          DOCTEST_REQUIRE_GT
+#define DOCTEST_FAST_WARN_LT             DOCTEST_WARN_LT
+#define DOCTEST_FAST_CHECK_LT            DOCTEST_CHECK_LT
+#define DOCTEST_FAST_REQUIRE_LT          DOCTEST_REQUIRE_LT
+#define DOCTEST_FAST_WARN_GE             DOCTEST_WARN_GE
+#define DOCTEST_FAST_CHECK_GE            DOCTEST_CHECK_GE
+#define DOCTEST_FAST_REQUIRE_GE          DOCTEST_REQUIRE_GE
+#define DOCTEST_FAST_WARN_LE             DOCTEST_WARN_LE
+#define DOCTEST_FAST_CHECK_LE            DOCTEST_CHECK_LE
+#define DOCTEST_FAST_REQUIRE_LE          DOCTEST_REQUIRE_LE
+
+#define DOCTEST_FAST_WARN_UNARY          DOCTEST_WARN_UNARY
+#define DOCTEST_FAST_CHECK_UNARY         DOCTEST_CHECK_UNARY
+#define DOCTEST_FAST_REQUIRE_UNARY       DOCTEST_REQUIRE_UNARY
+#define DOCTEST_FAST_WARN_UNARY_FALSE    DOCTEST_WARN_UNARY_FALSE
+#define DOCTEST_FAST_CHECK_UNARY_FALSE   DOCTEST_CHECK_UNARY_FALSE
+#define DOCTEST_FAST_REQUIRE_UNARY_FALSE DOCTEST_REQUIRE_UNARY_FALSE
+
+#define DOCTEST_TEST_CASE_TEMPLATE_INSTANTIATE DOCTEST_TEST_CASE_TEMPLATE_INVOKE
+// clang-format on
+
+// BDD style macros
+// clang-format off
+#define DOCTEST_SCENARIO(name) DOCTEST_TEST_CASE("  Scenario: " name)
+#define DOCTEST_SCENARIO_CLASS(name) DOCTEST_TEST_CASE_CLASS("  Scenario: " name)
+#define DOCTEST_SCENARIO_TEMPLATE(name, T, ...)  DOCTEST_TEST_CASE_TEMPLATE("  Scenario: " name, T, __VA_ARGS__)
+#define DOCTEST_SCENARIO_TEMPLATE_DEFINE(name, T, id) DOCTEST_TEST_CASE_TEMPLATE_DEFINE("  Scenario: " name, T, id)
+
+#define DOCTEST_GIVEN(name)     DOCTEST_SUBCASE("   Given: " name)
+#define DOCTEST_WHEN(name)      DOCTEST_SUBCASE("    When: " name)
+#define DOCTEST_AND_WHEN(name)  DOCTEST_SUBCASE("And when: " name)
+#define DOCTEST_THEN(name)      DOCTEST_SUBCASE("    Then: " name)
+#define DOCTEST_AND_THEN(name)  DOCTEST_SUBCASE("     And: " name)
+// clang-format on
+
+// == SHORT VERSIONS OF THE MACROS
+#if !defined(DOCTEST_CONFIG_NO_SHORT_MACRO_NAMES)
+
+#define TEST_CASE DOCTEST_TEST_CASE
+#define TEST_CASE_CLASS DOCTEST_TEST_CASE_CLASS
+#define TEST_CASE_FIXTURE DOCTEST_TEST_CASE_FIXTURE
+#define TYPE_TO_STRING DOCTEST_TYPE_TO_STRING
+#define TEST_CASE_TEMPLATE DOCTEST_TEST_CASE_TEMPLATE
+#define TEST_CASE_TEMPLATE_DEFINE DOCTEST_TEST_CASE_TEMPLATE_DEFINE
+#define TEST_CASE_TEMPLATE_INVOKE DOCTEST_TEST_CASE_TEMPLATE_INVOKE
+#define TEST_CASE_TEMPLATE_APPLY DOCTEST_TEST_CASE_TEMPLATE_APPLY
+#define SUBCASE DOCTEST_SUBCASE
+#define TEST_SUITE DOCTEST_TEST_SUITE
+#define TEST_SUITE_BEGIN DOCTEST_TEST_SUITE_BEGIN
+#define TEST_SUITE_END DOCTEST_TEST_SUITE_END
+#define REGISTER_EXCEPTION_TRANSLATOR DOCTEST_REGISTER_EXCEPTION_TRANSLATOR
+#define REGISTER_REPORTER DOCTEST_REGISTER_REPORTER
+#define REGISTER_LISTENER DOCTEST_REGISTER_LISTENER
+#define INFO DOCTEST_INFO
+#define CAPTURE DOCTEST_CAPTURE
+#define ADD_MESSAGE_AT DOCTEST_ADD_MESSAGE_AT
+#define ADD_FAIL_CHECK_AT DOCTEST_ADD_FAIL_CHECK_AT
+#define ADD_FAIL_AT DOCTEST_ADD_FAIL_AT
+#define MESSAGE DOCTEST_MESSAGE
+#define FAIL_CHECK DOCTEST_FAIL_CHECK
+#define FAIL DOCTEST_FAIL
+#define TO_LVALUE DOCTEST_TO_LVALUE
+
+#define WARN DOCTEST_WARN
+#define WARN_FALSE DOCTEST_WARN_FALSE
+#define WARN_THROWS DOCTEST_WARN_THROWS
+#define WARN_THROWS_AS DOCTEST_WARN_THROWS_AS
+#define WARN_THROWS_WITH DOCTEST_WARN_THROWS_WITH
+#define WARN_THROWS_WITH_AS DOCTEST_WARN_THROWS_WITH_AS
+#define WARN_NOTHROW DOCTEST_WARN_NOTHROW
+#define CHECK DOCTEST_CHECK
+#define CHECK_FALSE DOCTEST_CHECK_FALSE
+#define CHECK_THROWS DOCTEST_CHECK_THROWS
+#define CHECK_THROWS_AS DOCTEST_CHECK_THROWS_AS
+#define CHECK_THROWS_WITH DOCTEST_CHECK_THROWS_WITH
+#define CHECK_THROWS_WITH_AS DOCTEST_CHECK_THROWS_WITH_AS
+#define CHECK_NOTHROW DOCTEST_CHECK_NOTHROW
+#define REQUIRE DOCTEST_REQUIRE
+#define REQUIRE_FALSE DOCTEST_REQUIRE_FALSE
+#define REQUIRE_THROWS DOCTEST_REQUIRE_THROWS
+#define REQUIRE_THROWS_AS DOCTEST_REQUIRE_THROWS_AS
+#define REQUIRE_THROWS_WITH DOCTEST_REQUIRE_THROWS_WITH
+#define REQUIRE_THROWS_WITH_AS DOCTEST_REQUIRE_THROWS_WITH_AS
+#define REQUIRE_NOTHROW DOCTEST_REQUIRE_NOTHROW
+
+#define WARN_MESSAGE DOCTEST_WARN_MESSAGE
+#define WARN_FALSE_MESSAGE DOCTEST_WARN_FALSE_MESSAGE
+#define WARN_THROWS_MESSAGE DOCTEST_WARN_THROWS_MESSAGE
+#define WARN_THROWS_AS_MESSAGE DOCTEST_WARN_THROWS_AS_MESSAGE
+#define WARN_THROWS_WITH_MESSAGE DOCTEST_WARN_THROWS_WITH_MESSAGE
+#define WARN_THROWS_WITH_AS_MESSAGE DOCTEST_WARN_THROWS_WITH_AS_MESSAGE
+#define WARN_NOTHROW_MESSAGE DOCTEST_WARN_NOTHROW_MESSAGE
+#define CHECK_MESSAGE DOCTEST_CHECK_MESSAGE
+#define CHECK_FALSE_MESSAGE DOCTEST_CHECK_FALSE_MESSAGE
+#define CHECK_THROWS_MESSAGE DOCTEST_CHECK_THROWS_MESSAGE
+#define CHECK_THROWS_AS_MESSAGE DOCTEST_CHECK_THROWS_AS_MESSAGE
+#define CHECK_THROWS_WITH_MESSAGE DOCTEST_CHECK_THROWS_WITH_MESSAGE
+#define CHECK_THROWS_WITH_AS_MESSAGE DOCTEST_CHECK_THROWS_WITH_AS_MESSAGE
+#define CHECK_NOTHROW_MESSAGE DOCTEST_CHECK_NOTHROW_MESSAGE
+#define REQUIRE_MESSAGE DOCTEST_REQUIRE_MESSAGE
+#define REQUIRE_FALSE_MESSAGE DOCTEST_REQUIRE_FALSE_MESSAGE
+#define REQUIRE_THROWS_MESSAGE DOCTEST_REQUIRE_THROWS_MESSAGE
+#define REQUIRE_THROWS_AS_MESSAGE DOCTEST_REQUIRE_THROWS_AS_MESSAGE
+#define REQUIRE_THROWS_WITH_MESSAGE DOCTEST_REQUIRE_THROWS_WITH_MESSAGE
+#define REQUIRE_THROWS_WITH_AS_MESSAGE DOCTEST_REQUIRE_THROWS_WITH_AS_MESSAGE
+#define REQUIRE_NOTHROW_MESSAGE DOCTEST_REQUIRE_NOTHROW_MESSAGE
+
+#define SCENARIO DOCTEST_SCENARIO
+#define SCENARIO_CLASS DOCTEST_SCENARIO_CLASS
+#define SCENARIO_TEMPLATE DOCTEST_SCENARIO_TEMPLATE
+#define SCENARIO_TEMPLATE_DEFINE DOCTEST_SCENARIO_TEMPLATE_DEFINE
+#define GIVEN DOCTEST_GIVEN
+#define WHEN DOCTEST_WHEN
+#define AND_WHEN DOCTEST_AND_WHEN
+#define THEN DOCTEST_THEN
+#define AND_THEN DOCTEST_AND_THEN
+
+#define WARN_EQ DOCTEST_WARN_EQ
+#define CHECK_EQ DOCTEST_CHECK_EQ
+#define REQUIRE_EQ DOCTEST_REQUIRE_EQ
+#define WARN_NE DOCTEST_WARN_NE
+#define CHECK_NE DOCTEST_CHECK_NE
+#define REQUIRE_NE DOCTEST_REQUIRE_NE
+#define WARN_GT DOCTEST_WARN_GT
+#define CHECK_GT DOCTEST_CHECK_GT
+#define REQUIRE_GT DOCTEST_REQUIRE_GT
+#define WARN_LT DOCTEST_WARN_LT
+#define CHECK_LT DOCTEST_CHECK_LT
+#define REQUIRE_LT DOCTEST_REQUIRE_LT
+#define WARN_GE DOCTEST_WARN_GE
+#define CHECK_GE DOCTEST_CHECK_GE
+#define REQUIRE_GE DOCTEST_REQUIRE_GE
+#define WARN_LE DOCTEST_WARN_LE
+#define CHECK_LE DOCTEST_CHECK_LE
+#define REQUIRE_LE DOCTEST_REQUIRE_LE
+#define WARN_UNARY DOCTEST_WARN_UNARY
+#define CHECK_UNARY DOCTEST_CHECK_UNARY
+#define REQUIRE_UNARY DOCTEST_REQUIRE_UNARY
+#define WARN_UNARY_FALSE DOCTEST_WARN_UNARY_FALSE
+#define CHECK_UNARY_FALSE DOCTEST_CHECK_UNARY_FALSE
+#define REQUIRE_UNARY_FALSE DOCTEST_REQUIRE_UNARY_FALSE
+
+// KEPT FOR BACKWARDS COMPATIBILITY
+#define FAST_WARN_EQ DOCTEST_FAST_WARN_EQ
+#define FAST_CHECK_EQ DOCTEST_FAST_CHECK_EQ
+#define FAST_REQUIRE_EQ DOCTEST_FAST_REQUIRE_EQ
+#define FAST_WARN_NE DOCTEST_FAST_WARN_NE
+#define FAST_CHECK_NE DOCTEST_FAST_CHECK_NE
+#define FAST_REQUIRE_NE DOCTEST_FAST_REQUIRE_NE
+#define FAST_WARN_GT DOCTEST_FAST_WARN_GT
+#define FAST_CHECK_GT DOCTEST_FAST_CHECK_GT
+#define FAST_REQUIRE_GT DOCTEST_FAST_REQUIRE_GT
+#define FAST_WARN_LT DOCTEST_FAST_WARN_LT
+#define FAST_CHECK_LT DOCTEST_FAST_CHECK_LT
+#define FAST_REQUIRE_LT DOCTEST_FAST_REQUIRE_LT
+#define FAST_WARN_GE DOCTEST_FAST_WARN_GE
+#define FAST_CHECK_GE DOCTEST_FAST_CHECK_GE
+#define FAST_REQUIRE_GE DOCTEST_FAST_REQUIRE_GE
+#define FAST_WARN_LE DOCTEST_FAST_WARN_LE
+#define FAST_CHECK_LE DOCTEST_FAST_CHECK_LE
+#define FAST_REQUIRE_LE DOCTEST_FAST_REQUIRE_LE
+
+#define FAST_WARN_UNARY DOCTEST_FAST_WARN_UNARY
+#define FAST_CHECK_UNARY DOCTEST_FAST_CHECK_UNARY
+#define FAST_REQUIRE_UNARY DOCTEST_FAST_REQUIRE_UNARY
+#define FAST_WARN_UNARY_FALSE DOCTEST_FAST_WARN_UNARY_FALSE
+#define FAST_CHECK_UNARY_FALSE DOCTEST_FAST_CHECK_UNARY_FALSE
+#define FAST_REQUIRE_UNARY_FALSE DOCTEST_FAST_REQUIRE_UNARY_FALSE
+
+#define TEST_CASE_TEMPLATE_INSTANTIATE DOCTEST_TEST_CASE_TEMPLATE_INSTANTIATE
+
+#endif // DOCTEST_CONFIG_NO_SHORT_MACRO_NAMES
+
+#if !defined(DOCTEST_CONFIG_DISABLE)
+
+// this is here to clear the 'current test suite' for the current translation unit - at the top
+DOCTEST_TEST_SUITE_END();
+
+// add stringification for primitive/fundamental types
+namespace doctest { namespace detail {
+    DOCTEST_TYPE_TO_STRING_IMPL(bool)
+    DOCTEST_TYPE_TO_STRING_IMPL(float)
+    DOCTEST_TYPE_TO_STRING_IMPL(double)
+    DOCTEST_TYPE_TO_STRING_IMPL(long double)
+    DOCTEST_TYPE_TO_STRING_IMPL(char)
+    DOCTEST_TYPE_TO_STRING_IMPL(signed char)
+    DOCTEST_TYPE_TO_STRING_IMPL(unsigned char)
+#if !DOCTEST_MSVC || defined(_NATIVE_WCHAR_T_DEFINED)
+    DOCTEST_TYPE_TO_STRING_IMPL(wchar_t)
+#endif // not MSVC or wchar_t support enabled
+    DOCTEST_TYPE_TO_STRING_IMPL(short int)
+    DOCTEST_TYPE_TO_STRING_IMPL(unsigned short int)
+    DOCTEST_TYPE_TO_STRING_IMPL(int)
+    DOCTEST_TYPE_TO_STRING_IMPL(unsigned int)
+    DOCTEST_TYPE_TO_STRING_IMPL(long int)
+    DOCTEST_TYPE_TO_STRING_IMPL(unsigned long int)
+    DOCTEST_TYPE_TO_STRING_IMPL(long long int)
+    DOCTEST_TYPE_TO_STRING_IMPL(unsigned long long int)
+}} // namespace doctest::detail
+
+#endif // DOCTEST_CONFIG_DISABLE
+
+DOCTEST_CLANG_SUPPRESS_WARNING_POP
+DOCTEST_MSVC_SUPPRESS_WARNING_POP
+DOCTEST_GCC_SUPPRESS_WARNING_POP
+
+#endif // DOCTEST_LIBRARY_INCLUDED
+
+#ifndef DOCTEST_SINGLE_HEADER
+#define DOCTEST_SINGLE_HEADER
+#endif // DOCTEST_SINGLE_HEADER
+
+#if defined(DOCTEST_CONFIG_IMPLEMENT) || !defined(DOCTEST_SINGLE_HEADER)
+
+#ifndef DOCTEST_SINGLE_HEADER
+#include "doctest_fwd.h"
+#endif // DOCTEST_SINGLE_HEADER
+
+DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Wunused-macros")
+
+#ifndef DOCTEST_LIBRARY_IMPLEMENTATION
+#define DOCTEST_LIBRARY_IMPLEMENTATION
+
+DOCTEST_CLANG_SUPPRESS_WARNING_POP
+
+DOCTEST_CLANG_SUPPRESS_WARNING_PUSH
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wunknown-pragmas")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wpadded")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wweak-vtables")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wglobal-constructors")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wexit-time-destructors")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wmissing-prototypes")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wsign-conversion")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wshorten-64-to-32")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wmissing-variable-declarations")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wswitch")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wswitch-enum")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wcovered-switch-default")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wmissing-noreturn")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wunused-local-typedef")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wdisabled-macro-expansion")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wmissing-braces")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wmissing-field-initializers")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wc++98-compat")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wc++98-compat-pedantic")
+DOCTEST_CLANG_SUPPRESS_WARNING("-Wunused-member-function")
+
+DOCTEST_GCC_SUPPRESS_WARNING_PUSH
+DOCTEST_GCC_SUPPRESS_WARNING("-Wunknown-pragmas")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wpragmas")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wconversion")
+DOCTEST_GCC_SUPPRESS_WARNING("-Weffc++")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wsign-conversion")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wstrict-overflow")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wstrict-aliasing")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wmissing-field-initializers")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wmissing-braces")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wmissing-declarations")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wswitch")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wswitch-enum")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wswitch-default")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wunsafe-loop-optimizations")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wold-style-cast")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wunused-local-typedefs")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wuseless-cast")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wunused-function")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wmultiple-inheritance")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wnoexcept")
+DOCTEST_GCC_SUPPRESS_WARNING("-Wsuggest-attribute")
+
+DOCTEST_MSVC_SUPPRESS_WARNING_PUSH
+DOCTEST_MSVC_SUPPRESS_WARNING(4616) // invalid compiler warning
+DOCTEST_MSVC_SUPPRESS_WARNING(4619) // invalid compiler warning
+DOCTEST_MSVC_SUPPRESS_WARNING(4996) // The compiler encountered a deprecated declaration
+DOCTEST_MSVC_SUPPRESS_WARNING(4267) // 'var' : conversion from 'x' to 'y', possible loss of data
+DOCTEST_MSVC_SUPPRESS_WARNING(4706) // assignment within conditional expression
+DOCTEST_MSVC_SUPPRESS_WARNING(4512) // 'class' : assignment operator could not be generated
+DOCTEST_MSVC_SUPPRESS_WARNING(4127) // conditional expression is constant
+DOCTEST_MSVC_SUPPRESS_WARNING(4530) // C++ exception handler used, but unwind semantics not enabled
+DOCTEST_MSVC_SUPPRESS_WARNING(4577) // 'noexcept' used with no exception handling mode specified
+DOCTEST_MSVC_SUPPRESS_WARNING(4774) // format string expected in argument is not a string literal
+DOCTEST_MSVC_SUPPRESS_WARNING(4365) // conversion from 'int' to 'unsigned', signed/unsigned mismatch
+DOCTEST_MSVC_SUPPRESS_WARNING(4820) // padding in structs
+DOCTEST_MSVC_SUPPRESS_WARNING(4640) // construction of local static object is not thread-safe
+DOCTEST_MSVC_SUPPRESS_WARNING(5039) // pointer to potentially throwing function passed to extern C
+DOCTEST_MSVC_SUPPRESS_WARNING(5045) // Spectre mitigation stuff
+DOCTEST_MSVC_SUPPRESS_WARNING(4626) // assignment operator was implicitly defined as deleted
+DOCTEST_MSVC_SUPPRESS_WARNING(5027) // move assignment operator was implicitly defined as deleted
+DOCTEST_MSVC_SUPPRESS_WARNING(5026) // move constructor was implicitly defined as deleted
+DOCTEST_MSVC_SUPPRESS_WARNING(4625) // copy constructor was implicitly defined as deleted
+DOCTEST_MSVC_SUPPRESS_WARNING(4800) // forcing value to bool 'true' or 'false' (performance warning)
+// static analysis
+DOCTEST_MSVC_SUPPRESS_WARNING(26439) // This kind of function may not throw. Declare it 'noexcept'
+DOCTEST_MSVC_SUPPRESS_WARNING(26495) // Always initialize a member variable
+DOCTEST_MSVC_SUPPRESS_WARNING(26451) // Arithmetic overflow ...
+DOCTEST_MSVC_SUPPRESS_WARNING(26444) // Avoid unnamed objects with custom construction and dtor...
+DOCTEST_MSVC_SUPPRESS_WARNING(26812) // Prefer 'enum class' over 'enum'
+
+DOCTEST_MAKE_STD_HEADERS_CLEAN_FROM_WARNINGS_ON_WALL_BEGIN
+
+// required includes - will go only in one translation unit!
+#include <ctime>
+#include <cmath>
+#include <climits>
+// borland (Embarcadero) compiler requires math.h and not cmath - https://github.com/onqtam/doctest/pull/37
+#ifdef __BORLANDC__
+#include <math.h>
+#endif // __BORLANDC__
+#include <new>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+#include <limits>
+#include <utility>
+#include <fstream>
+#include <sstream>
+#include <iostream>
+#include <algorithm>
+#include <iomanip>
+#include <vector>
+#include <atomic>
+#include <mutex>
+#include <set>
+#include <map>
+#include <exception>
+#include <stdexcept>
+#ifdef DOCTEST_CONFIG_POSIX_SIGNALS
+#include <csignal>
+#endif // DOCTEST_CONFIG_POSIX_SIGNALS
+#include <cfloat>
+#include <cctype>
+#include <cstdint>
+
+#ifdef DOCTEST_PLATFORM_MAC
+#include <sys/types.h>
+#include <unistd.h>
+#include <sys/sysctl.h>
+#endif // DOCTEST_PLATFORM_MAC
+
+#ifdef DOCTEST_PLATFORM_WINDOWS
+
+// defines for a leaner windows.h
+#ifndef WIN32_LEAN_AND_MEAN
+#define WIN32_LEAN_AND_MEAN
+#endif // WIN32_LEAN_AND_MEAN
+#ifndef NOMINMAX
+#define NOMINMAX
+#endif // NOMINMAX
+
+// not sure what AfxWin.h is for - here I do what Catch does
+#ifdef __AFXDLL
+#include <AfxWin.h>
+#else
+#if defined(__MINGW32__) || defined(__MINGW64__)
+#include <windows.h>
+#else // MINGW
+#include <Windows.h>
+#endif // MINGW
+#endif
+#include <io.h>
+
+#else // DOCTEST_PLATFORM_WINDOWS
+
+#include <sys/time.h>
+#include <unistd.h>
+
+#endif // DOCTEST_PLATFORM_WINDOWS
+
+// this is a fix for https://github.com/onqtam/doctest/issues/348
+// https://mail.gnome.org/archives/xml/2012-January/msg00000.html
+#if !defined(HAVE_UNISTD_H) && !defined(STDOUT_FILENO)
+#define STDOUT_FILENO fileno(stdout)
+#endif // HAVE_UNISTD_H
+
+DOCTEST_MAKE_STD_HEADERS_CLEAN_FROM_WARNINGS_ON_WALL_END
+
+// counts the number of elements in a C array
+#define DOCTEST_COUNTOF(x) (sizeof(x) / sizeof(x[0]))
+
+#ifdef DOCTEST_CONFIG_DISABLE
+#define DOCTEST_BRANCH_ON_DISABLED(if_disabled, if_not_disabled) if_disabled
+#else // DOCTEST_CONFIG_DISABLE
+#define DOCTEST_BRANCH_ON_DISABLED(if_disabled, if_not_disabled) if_not_disabled
+#endif // DOCTEST_CONFIG_DISABLE
+
+#ifndef DOCTEST_CONFIG_OPTIONS_PREFIX
+#define DOCTEST_CONFIG_OPTIONS_PREFIX "dt-"
+#endif
+
+#ifndef DOCTEST_THREAD_LOCAL
+#define DOCTEST_THREAD_LOCAL thread_local
+#endif
+
+#ifdef DOCTEST_CONFIG_NO_UNPREFIXED_OPTIONS
+#define DOCTEST_OPTIONS_PREFIX_DISPLAY DOCTEST_CONFIG_OPTIONS_PREFIX
+#else
+#define DOCTEST_OPTIONS_PREFIX_DISPLAY ""
+#endif
+
+namespace doctest {
+
+bool is_running_in_test = false;
+
+namespace {
+    using namespace detail;
+    // case insensitive strcmp
+    int stricmp(const char* a, const char* b) {
+        for(;; a++, b++) {
+            const int d = tolower(*a) - tolower(*b);
+            if(d != 0 || !*a)
+                return d;
+        }
+    }
+
+    template <typename T>
+    String fpToString(T value, int precision) {
+        std::ostringstream oss;
+        oss << std::setprecision(precision) << std::fixed << value;
+        std::string d = oss.str();
+        size_t      i = d.find_last_not_of('0');
+        if(i != std::string::npos && i != d.size() - 1) {
+            if(d[i] == '.')
+                i++;
+            d = d.substr(0, i + 1);
+        }
+        return d.c_str();
+    }
+
+    struct Endianness
+    {
+        enum Arch
+        {
+            Big,
+            Little
+        };
+
+        static Arch which() {
+            int x = 1;
+            // casting any data pointer to char* is allowed
+            auto ptr = reinterpret_cast<char*>(&x);
+            if(*ptr)
+                return Little;
+            return Big;
+        }
+    };
+} // namespace
+
+namespace detail {
+    void my_memcpy(void* dest, const void* src, unsigned num) { memcpy(dest, src, num); }
+
+    String rawMemoryToString(const void* object, unsigned size) {
+        // Reverse order for little endian architectures
+        int i = 0, end = static_cast<int>(size), inc = 1;
+        if(Endianness::which() == Endianness::Little) {
+            i   = end - 1;
+            end = inc = -1;
+        }
+
+        unsigned const char* bytes = static_cast<unsigned const char*>(object);
+        std::ostringstream   oss;
+        oss << "0x" << std::setfill('0') << std::hex;
+        for(; i != end; i += inc)
+            oss << std::setw(2) << static_cast<unsigned>(bytes[i]);
+        return oss.str().c_str();
+    }
+
+    DOCTEST_THREAD_LOCAL std::ostringstream g_oss; // NOLINT(cert-err58-cpp)
+
+    std::ostream* getTlsOss() {
+        g_oss.clear(); // there shouldn't be anything worth clearing in the flags
+        g_oss.str(""); // the slow way of resetting a string stream
+        //g_oss.seekp(0); // optimal reset - as seen here: https://stackoverflow.com/a/624291/3162383
+        return &g_oss;
+    }
+
+    String getTlsOssResult() {
+        //g_oss << std::ends; // needed - as shown here: https://stackoverflow.com/a/624291/3162383
+        return g_oss.str().c_str();
+    }
+
+#ifndef DOCTEST_CONFIG_DISABLE
+
+namespace timer_large_integer
+{
+    
+#if defined(DOCTEST_PLATFORM_WINDOWS)
+    typedef ULONGLONG type;
+#else // DOCTEST_PLATFORM_WINDOWS
+    using namespace std;
+    typedef uint64_t type;
+#endif // DOCTEST_PLATFORM_WINDOWS
+}
+
+typedef timer_large_integer::type ticks_t;
+
+#ifdef DOCTEST_CONFIG_GETCURRENTTICKS
+    ticks_t getCurrentTicks() { return DOCTEST_CONFIG_GETCURRENTTICKS(); }
+#elif defined(DOCTEST_PLATFORM_WINDOWS)
+    ticks_t getCurrentTicks() {
+        static LARGE_INTEGER hz = {0}, hzo = {0};
+        if(!hz.QuadPart) {
+            QueryPerformanceFrequency(&hz);
+            QueryPerformanceCounter(&hzo);
+        }
+        LARGE_INTEGER t;
+        QueryPerformanceCounter(&t);
+        return ((t.QuadPart - hzo.QuadPart) * LONGLONG(1000000)) / hz.QuadPart;
+    }
+#else  // DOCTEST_PLATFORM_WINDOWS
+    ticks_t getCurrentTicks() {
+        timeval t;
+        gettimeofday(&t, nullptr);
+        return static_cast<ticks_t>(t.tv_sec) * 1000000 + static_cast<ticks_t>(t.tv_usec);
+    }
+#endif // DOCTEST_PLATFORM_WINDOWS
+
+    struct Timer
+    {
+        void         start() { m_ticks = getCurrentTicks(); }
+        unsigned int getElapsedMicroseconds() const {
+            return static_cast<unsigned int>(getCurrentTicks() - m_ticks);
+        }
+        //unsigned int getElapsedMilliseconds() const {
+        //    return static_cast<unsigned int>(getElapsedMicroseconds() / 1000);
+        //}
+        double getElapsedSeconds() const { return static_cast<double>(getCurrentTicks() - m_ticks) / 1000000.0; }
+
+    private:
+        ticks_t m_ticks = 0;
+    };
+
+    // this holds both parameters from the command line and runtime data for tests
+    struct ContextState : ContextOptions, TestRunStats, CurrentTestCaseStats
+    {
+        std::atomic<int> numAssertsCurrentTest_atomic;
+        std::atomic<int> numAssertsFailedCurrentTest_atomic;
+
+        std::vector<std::vector<String>> filters = decltype(filters)(9); // 9 different filters
+
+        std::vector<IReporter*> reporters_currently_used;
+
+        const TestCase* currentTest = nullptr;
+
+        assert_handler ah = nullptr;
+
+        Timer timer;
+
+        std::vector<String> stringifiedContexts; // logging from INFO() due to an exception
+
+        // stuff for subcases
+        std::vector<SubcaseSignature>     subcasesStack;
+        std::set<decltype(subcasesStack)> subcasesPassed;
+        int                               subcasesCurrentMaxLevel;
+        bool                              should_reenter;
+        std::atomic<bool>                 shouldLogCurrentException;
+
+        void resetRunData() {
+            numTestCases                = 0;
+            numTestCasesPassingFilters  = 0;
+            numTestSuitesPassingFilters = 0;
+            numTestCasesFailed          = 0;
+            numAsserts                  = 0;
+            numAssertsFailed            = 0;
+            numAssertsCurrentTest       = 0;
+            numAssertsFailedCurrentTest = 0;
+        }
+
+        void finalizeTestCaseData() {
+            seconds = timer.getElapsedSeconds();
+
+            // update the non-atomic counters
+            numAsserts += numAssertsCurrentTest_atomic;
+            numAssertsFailed += numAssertsFailedCurrentTest_atomic;
+            numAssertsCurrentTest       = numAssertsCurrentTest_atomic;
+            numAssertsFailedCurrentTest = numAssertsFailedCurrentTest_atomic;
+
+            if(numAssertsFailedCurrentTest)
+                failure_flags |= TestCaseFailureReason::AssertFailure;
+
+            if(Approx(currentTest->m_timeout).epsilon(DBL_EPSILON) != 0 &&
+               Approx(seconds).epsilon(DBL_EPSILON) > currentTest->m_timeout)
+                failure_flags |= TestCaseFailureReason::Timeout;
+
+            if(currentTest->m_should_fail) {
+                if(failure_flags) {
+                    failure_flags |= TestCaseFailureReason::ShouldHaveFailedAndDid;
+                } else {
+                    failure_flags |= TestCaseFailureReason::ShouldHaveFailedButDidnt;
+                }
+            } else if(failure_flags && currentTest->m_may_fail) {
+                failure_flags |= TestCaseFailureReason::CouldHaveFailedAndDid;
+            } else if(currentTest->m_expected_failures > 0) {
+                if(numAssertsFailedCurrentTest == currentTest->m_expected_failures) {
+                    failure_flags |= TestCaseFailureReason::FailedExactlyNumTimes;
+                } else {
+                    failure_flags |= TestCaseFailureReason::DidntFailExactlyNumTimes;
+                }
+            }
+
+            bool ok_to_fail = (TestCaseFailureReason::ShouldHaveFailedAndDid & failure_flags) ||
+                              (TestCaseFailureReason::CouldHaveFailedAndDid & failure_flags) ||
+                              (TestCaseFailureReason::FailedExactlyNumTimes & failure_flags);
+
+            // if any subcase has failed - the whole test case has failed
+            if(failure_flags && !ok_to_fail)
+                numTestCasesFailed++;
+        }
+    };
+
+    ContextState* g_cs = nullptr;
+
+    // used to avoid locks for the debug output
+    // TODO: figure out if this is indeed necessary/correct - seems like either there still
+    // could be a race or that there wouldn't be a race even if using the context directly
+    DOCTEST_THREAD_LOCAL bool g_no_colors;
+
+#endif // DOCTEST_CONFIG_DISABLE
+} // namespace detail
+
+void String::setOnHeap() { *reinterpret_cast<unsigned char*>(&buf[last]) = 128; }
+void String::setLast(unsigned in) { buf[last] = char(in); }
+
+void String::copy(const String& other) {
+    using namespace std;
+    if(other.isOnStack()) {
+        memcpy(buf, other.buf, len);
+    } else {
+        setOnHeap();
+        data.size     = other.data.size;
+        data.capacity = data.size + 1;
+        data.ptr      = new char[data.capacity];
+        memcpy(data.ptr, other.data.ptr, data.size + 1);
+    }
+}
+
+String::String() {
+    buf[0] = '\0';
+    setLast();
+}
+
+String::~String() {
+    if(!isOnStack())
+        delete[] data.ptr;
+}
+
+String::String(const char* in)
+        : String(in, strlen(in)) {}
+
+String::String(const char* in, unsigned in_size) {
+    using namespace std;
+    if(in_size <= last) {
+        memcpy(buf, in, in_size + 1);
+        setLast(last - in_size);
+    } else {
+        setOnHeap();
+        data.size     = in_size;
+        data.capacity = data.size + 1;
+        data.ptr      = new char[data.capacity];
+        memcpy(data.ptr, in, in_size + 1);
+    }
+}
+
+String::String(const String& other) { copy(other); }
+
+String& String::operator=(const String& other) {
+    if(this != &other) {
+        if(!isOnStack())
+            delete[] data.ptr;
+
+        copy(other);
+    }
+
+    return *this;
+}
+
+String& String::operator+=(const String& other) {
+    const unsigned my_old_size = size();
+    const unsigned other_size  = other.size();
+    const unsigned total_size  = my_old_size + other_size;
+    using namespace std;
+    if(isOnStack()) {
+        if(total_size < len) {
+            // append to the current stack space
+            memcpy(buf + my_old_size, other.c_str(), other_size + 1);
+            setLast(last - total_size);
+        } else {
+            // alloc new chunk
+            char* temp = new char[total_size + 1];
+            // copy current data to new location before writing in the union
+            memcpy(temp, buf, my_old_size); // skip the +1 ('\0') for speed
+            // update data in union
+            setOnHeap();
+            data.size     = total_size;
+            data.capacity = data.size + 1;
+            data.ptr      = temp;
+            // transfer the rest of the data
+            memcpy(data.ptr + my_old_size, other.c_str(), other_size + 1);
+        }
+    } else {
+        if(data.capacity > total_size) {
+            // append to the current heap block
+            data.size = total_size;
+            memcpy(data.ptr + my_old_size, other.c_str(), other_size + 1);
+        } else {
+            // resize
+            data.capacity *= 2;
+            if(data.capacity <= total_size)
+                data.capacity = total_size + 1;
+            // alloc new chunk
+            char* temp = new char[data.capacity];
+            // copy current data to new location before releasing it
+            memcpy(temp, data.ptr, my_old_size); // skip the +1 ('\0') for speed
+            // release old chunk
+            delete[] data.ptr;
+            // update the rest of the union members
+            data.size = total_size;
+            data.ptr  = temp;
+            // transfer the rest of the data
+            memcpy(data.ptr + my_old_size, other.c_str(), other_size + 1);
+        }
+    }
+
+    return *this;
+}
+
+String String::operator+(const String& other) const { return String(*this) += other; }
+
+String::String(String&& other) {
+    using namespace std;
+    memcpy(buf, other.buf, len);
+    other.buf[0] = '\0';
+    other.setLast();
+}
+
+String& String::operator=(String&& other) {
+    using namespace std;
+    if(this != &other) {
+        if(!isOnStack())
+            delete[] data.ptr;
+        memcpy(buf, other.buf, len);
+        other.buf[0] = '\0';
+        other.setLast();
+    }
+    return *this;
+}
+
+char String::operator[](unsigned i) const {
+    return const_cast<String*>(this)->operator[](i); // NOLINT
+}
+
+char& String::operator[](unsigned i) {
+    if(isOnStack())
+        return reinterpret_cast<char*>(buf)[i];
+    return data.ptr[i];
+}
+
+DOCTEST_GCC_SUPPRESS_WARNING_WITH_PUSH("-Wmaybe-uninitialized")
+unsigned String::size() const {
+    if(isOnStack())
+        return last - (unsigned(buf[last]) & 31); // using "last" would work only if "len" is 32
+    return data.size;
+}
+DOCTEST_GCC_SUPPRESS_WARNING_POP
+
+unsigned String::capacity() const {
+    if(isOnStack())
+        return len;
+    return data.capacity;
+}
+
+int String::compare(const char* other, bool no_case) const {
+    if(no_case)
+        return doctest::stricmp(c_str(), other);
+    return std::strcmp(c_str(), other);
+}
+
+int String::compare(const String& other, bool no_case) const {
+    return compare(other.c_str(), no_case);
+}
+
+// clang-format off
+bool operator==(const String& lhs, const String& rhs) { return lhs.compare(rhs) == 0; }
+bool operator!=(const String& lhs, const String& rhs) { return lhs.compare(rhs) != 0; }
+bool operator< (const String& lhs, const String& rhs) { return lhs.compare(rhs) < 0; }
+bool operator> (const String& lhs, const String& rhs) { return lhs.compare(rhs) > 0; }
+bool operator<=(const String& lhs, const String& rhs) { return (lhs != rhs) ? lhs.compare(rhs) < 0 : true; }
+bool operator>=(const String& lhs, const String& rhs) { return (lhs != rhs) ? lhs.compare(rhs) > 0 : true; }
+// clang-format on
+
+std::ostream& operator<<(std::ostream& s, const String& in) { return s << in.c_str(); }
+
+namespace {
+    void color_to_stream(std::ostream&, Color::Enum) DOCTEST_BRANCH_ON_DISABLED({}, ;)
+} // namespace
+
+namespace Color {
+    std::ostream& operator<<(std::ostream& s, Color::Enum code) {
+        color_to_stream(s, code);
+        return s;
+    }
+} // namespace Color
+
+// clang-format off
+const char* assertString(assertType::Enum at) {
+    DOCTEST_MSVC_SUPPRESS_WARNING_WITH_PUSH(4062) // enum 'x' in switch of enum 'y' is not handled
+    switch(at) {  //!OCLINT missing default in switch statements
+        case assertType::DT_WARN                    : return "WARN";
+        case assertType::DT_CHECK                   : return "CHECK";
+        case assertType::DT_REQUIRE                 : return "REQUIRE";
+
+        case assertType::DT_WARN_FALSE              : return "WARN_FALSE";
+        case assertType::DT_CHECK_FALSE             : return "CHECK_FALSE";
+        case assertType::DT_REQUIRE_FALSE           : return "REQUIRE_FALSE";
+
+        case assertType::DT_WARN_THROWS             : return "WARN_THROWS";
+        case assertType::DT_CHECK_THROWS            : return "CHECK_THROWS";
+        case assertType::DT_REQUIRE_THROWS          : return "REQUIRE_THROWS";
+
+        case assertType::DT_WARN_THROWS_AS          : return "WARN_THROWS_AS";
+        case assertType::DT_CHECK_THROWS_AS         : return "CHECK_THROWS_AS";
+        case assertType::DT_REQUIRE_THROWS_AS       : return "REQUIRE_THROWS_AS";
+
+        case assertType::DT_WARN_THROWS_WITH        : return "WARN_THROWS_WITH";
+        case assertType::DT_CHECK_THROWS_WITH       : return "CHECK_THROWS_WITH";
+        case assertType::DT_REQUIRE_THROWS_WITH     : return "REQUIRE_THROWS_WITH";
+
+        case assertType::DT_WARN_THROWS_WITH_AS     : return "WARN_THROWS_WITH_AS";
+        case assertType::DT_CHECK_THROWS_WITH_AS    : return "CHECK_THROWS_WITH_AS";
+        case assertType::DT_REQUIRE_THROWS_WITH_AS  : return "REQUIRE_THROWS_WITH_AS";
+
+        case assertType::DT_WARN_NOTHROW            : return "WARN_NOTHROW";
+        case assertType::DT_CHECK_NOTHROW           : return "CHECK_NOTHROW";
+        case assertType::DT_REQUIRE_NOTHROW         : return "REQUIRE_NOTHROW";
+
+        case assertType::DT_WARN_EQ                 : return "WARN_EQ";
+        case assertType::DT_CHECK_EQ                : return "CHECK_EQ";
+        case assertType::DT_REQUIRE_EQ              : return "REQUIRE_EQ";
+        case assertType::DT_WARN_NE                 : return "WARN_NE";
+        case assertType::DT_CHECK_NE                : return "CHECK_NE";
+        case assertType::DT_REQUIRE_NE              : return "REQUIRE_NE";
+        case assertType::DT_WARN_GT                 : return "WARN_GT";
+        case assertType::DT_CHECK_GT                : return "CHECK_GT";
+        case assertType::DT_REQUIRE_GT              : return "REQUIRE_GT";
+        case assertType::DT_WARN_LT                 : return "WARN_LT";
+        case assertType::DT_CHECK_LT                : return "CHECK_LT";
+        case assertType::DT_REQUIRE_LT              : return "REQUIRE_LT";
+        case assertType::DT_WARN_GE                 : return "WARN_GE";
+        case assertType::DT_CHECK_GE                : return "CHECK_GE";
+        case assertType::DT_REQUIRE_GE              : return "REQUIRE_GE";
+        case assertType::DT_WARN_LE                 : return "WARN_LE";
+        case assertType::DT_CHECK_LE                : return "CHECK_LE";
+        case assertType::DT_REQUIRE_LE              : return "REQUIRE_LE";
+
+        case assertType::DT_WARN_UNARY              : return "WARN_UNARY";
+        case assertType::DT_CHECK_UNARY             : return "CHECK_UNARY";
+        case assertType::DT_REQUIRE_UNARY           : return "REQUIRE_UNARY";
+        case assertType::DT_WARN_UNARY_FALSE        : return "WARN_UNARY_FALSE";
+        case assertType::DT_CHECK_UNARY_FALSE       : return "CHECK_UNARY_FALSE";
+        case assertType::DT_REQUIRE_UNARY_FALSE     : return "REQUIRE_UNARY_FALSE";
+    }
+    DOCTEST_MSVC_SUPPRESS_WARNING_POP
+    return "";
+}
+// clang-format on
+
+const char* failureString(assertType::Enum at) {
+    if(at & assertType::is_warn) //!OCLINT bitwise operator in conditional
+        return "WARNING";
+    if(at & assertType::is_check) //!OCLINT bitwise operator in conditional
+        return "ERROR";
+    if(at & assertType::is_require) //!OCLINT bitwise operator in conditional
+        return "FATAL ERROR";
+    return "";
+}
+
+DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Wnull-dereference")
+DOCTEST_GCC_SUPPRESS_WARNING_WITH_PUSH("-Wnull-dereference")
+// depending on the current options this will remove the path of filenames
+const char* skipPathFromFilename(const char* file) {
+    if(getContextOptions()->no_path_in_filenames) {
+        auto back    = std::strrchr(file, '\\');
+        auto forward = std::strrchr(file, '/');
+        if(back || forward) {
+            if(back > forward)
+                forward = back;
+            return forward + 1;
+        }
+    }
+    return file;
+}
+DOCTEST_CLANG_SUPPRESS_WARNING_POP
+DOCTEST_GCC_SUPPRESS_WARNING_POP
+
+bool SubcaseSignature::operator<(const SubcaseSignature& other) const {
+    if(m_line != other.m_line)
+        return m_line < other.m_line;
+    if(std::strcmp(m_file, other.m_file) != 0)
+        return std::strcmp(m_file, other.m_file) < 0;
+    return m_name.compare(other.m_name) < 0;
+}
+
+IContextScope::IContextScope()  = default;
+IContextScope::~IContextScope() = default;
+
+#ifdef DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+String toString(char* in) { return toString(static_cast<const char*>(in)); }
+String toString(const char* in) { return String("\"") + (in ? in : "{null string}") + "\""; }
+#endif // DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+String toString(bool in) { return in ? "true" : "false"; }
+String toString(float in) { return fpToString(in, 5) + "f"; }
+String toString(double in) { return fpToString(in, 10); }
+String toString(double long in) { return fpToString(in, 15); }
+
+#define DOCTEST_TO_STRING_OVERLOAD(type, fmt)                                                      \
+    String toString(type in) {                                                                     \
+        char buf[64];                                                                              \
+        std::sprintf(buf, fmt, in);                                                                \
+        return buf;                                                                                \
+    }
+
+DOCTEST_TO_STRING_OVERLOAD(char, "%d")
+DOCTEST_TO_STRING_OVERLOAD(char signed, "%d")
+DOCTEST_TO_STRING_OVERLOAD(char unsigned, "%u")
+DOCTEST_TO_STRING_OVERLOAD(int short, "%d")
+DOCTEST_TO_STRING_OVERLOAD(int short unsigned, "%u")
+DOCTEST_TO_STRING_OVERLOAD(int, "%d")
+DOCTEST_TO_STRING_OVERLOAD(unsigned, "%u")
+DOCTEST_TO_STRING_OVERLOAD(int long, "%ld")
+DOCTEST_TO_STRING_OVERLOAD(int long unsigned, "%lu")
+DOCTEST_TO_STRING_OVERLOAD(int long long, "%lld")
+DOCTEST_TO_STRING_OVERLOAD(int long long unsigned, "%llu")
+
+String toString(std::nullptr_t) { return "NULL"; }
+
+#if DOCTEST_MSVC >= DOCTEST_COMPILER(19, 20, 0)
+// see this issue on why this is needed: https://github.com/onqtam/doctest/issues/183
+String toString(const std::string& in) { return in.c_str(); }
+#endif // VS 2019
+
+Approx::Approx(double value)
+        : m_epsilon(static_cast<double>(std::numeric_limits<float>::epsilon()) * 100)
+        , m_scale(1.0)
+        , m_value(value) {}
+
+Approx Approx::operator()(double value) const {
+    Approx approx(value);
+    approx.epsilon(m_epsilon);
+    approx.scale(m_scale);
+    return approx;
+}
+
+Approx& Approx::epsilon(double newEpsilon) {
+    m_epsilon = newEpsilon;
+    return *this;
+}
+Approx& Approx::scale(double newScale) {
+    m_scale = newScale;
+    return *this;
+}
+
+bool operator==(double lhs, const Approx& rhs) {
+    // Thanks to Richard Harris for his help refining this formula
+    return std::fabs(lhs - rhs.m_value) <
+           rhs.m_epsilon * (rhs.m_scale + std::max<double>(std::fabs(lhs), std::fabs(rhs.m_value)));
+}
+bool operator==(const Approx& lhs, double rhs) { return operator==(rhs, lhs); }
+bool operator!=(double lhs, const Approx& rhs) { return !operator==(lhs, rhs); }
+bool operator!=(const Approx& lhs, double rhs) { return !operator==(rhs, lhs); }
+bool operator<=(double lhs, const Approx& rhs) { return lhs < rhs.m_value || lhs == rhs; }
+bool operator<=(const Approx& lhs, double rhs) { return lhs.m_value < rhs || lhs == rhs; }
+bool operator>=(double lhs, const Approx& rhs) { return lhs > rhs.m_value || lhs == rhs; }
+bool operator>=(const Approx& lhs, double rhs) { return lhs.m_value > rhs || lhs == rhs; }
+bool operator<(double lhs, const Approx& rhs) { return lhs < rhs.m_value && lhs != rhs; }
+bool operator<(const Approx& lhs, double rhs) { return lhs.m_value < rhs && lhs != rhs; }
+bool operator>(double lhs, const Approx& rhs) { return lhs > rhs.m_value && lhs != rhs; }
+bool operator>(const Approx& lhs, double rhs) { return lhs.m_value > rhs && lhs != rhs; }
+
+String toString(const Approx& in) {
+    return String("Approx( ") + doctest::toString(in.m_value) + " )";
+}
+const ContextOptions* getContextOptions() { return DOCTEST_BRANCH_ON_DISABLED(nullptr, g_cs); }
+
+} // namespace doctest
+
+#ifdef DOCTEST_CONFIG_DISABLE
+namespace doctest {
+Context::Context(int, const char* const*) {}
+Context::~Context() = default;
+void Context::applyCommandLine(int, const char* const*) {}
+void Context::addFilter(const char*, const char*) {}
+void Context::clearFilters() {}
+void Context::setOption(const char*, int) {}
+void Context::setOption(const char*, const char*) {}
+bool Context::shouldExit() { return false; }
+void Context::setAsDefaultForAssertsOutOfTestCases() {}
+void Context::setAssertHandler(detail::assert_handler) {}
+int  Context::run() { return 0; }
+
+IReporter::~IReporter() = default;
+
+int                         IReporter::get_num_active_contexts() { return 0; }
+const IContextScope* const* IReporter::get_active_contexts() { return nullptr; }
+int                         IReporter::get_num_stringified_contexts() { return 0; }
+const String*               IReporter::get_stringified_contexts() { return nullptr; }
+
+int registerReporter(const char*, int, IReporter*) { return 0; }
+
+} // namespace doctest
+#else // DOCTEST_CONFIG_DISABLE
+
+#if !defined(DOCTEST_CONFIG_COLORS_NONE)
+#if !defined(DOCTEST_CONFIG_COLORS_WINDOWS) && !defined(DOCTEST_CONFIG_COLORS_ANSI)
+#ifdef DOCTEST_PLATFORM_WINDOWS
+#define DOCTEST_CONFIG_COLORS_WINDOWS
+#else // linux
+#define DOCTEST_CONFIG_COLORS_ANSI
+#endif // platform
+#endif // DOCTEST_CONFIG_COLORS_WINDOWS && DOCTEST_CONFIG_COLORS_ANSI
+#endif // DOCTEST_CONFIG_COLORS_NONE
+
+namespace doctest_detail_test_suite_ns {
+// holds the current test suite
+doctest::detail::TestSuite& getCurrentTestSuite() {
+    static doctest::detail::TestSuite data;
+    return data;
+}
+} // namespace doctest_detail_test_suite_ns
+
+namespace doctest {
+namespace {
+    // the int (priority) is part of the key for automatic sorting - sadly one can register a
+    // reporter with a duplicate name and a different priority but hopefully that won't happen often :|
+    typedef std::map<std::pair<int, String>, reporterCreatorFunc> reporterMap;
+
+    reporterMap& getReporters() {
+        static reporterMap data;
+        return data;
+    }
+    reporterMap& getListeners() {
+        static reporterMap data;
+        return data;
+    }
+} // namespace
+namespace detail {
+#define DOCTEST_ITERATE_THROUGH_REPORTERS(function, ...)                                           \
+    for(auto& curr_rep : g_cs->reporters_currently_used)                                           \
+    curr_rep->function(__VA_ARGS__)
+
+    bool checkIfShouldThrow(assertType::Enum at) {
+        if(at & assertType::is_require) //!OCLINT bitwise operator in conditional
+            return true;
+
+        if((at & assertType::is_check) //!OCLINT bitwise operator in conditional
+           && getContextOptions()->abort_after > 0 &&
+           (g_cs->numAssertsFailed + g_cs->numAssertsFailedCurrentTest_atomic) >=
+                   getContextOptions()->abort_after)
+            return true;
+
+        return false;
+    }
+
+#ifndef DOCTEST_CONFIG_NO_EXCEPTIONS
+    DOCTEST_NORETURN void throwException() {
+        g_cs->shouldLogCurrentException = false;
+        throw TestFailureException();
+    } // NOLINT(cert-err60-cpp)
+#else // DOCTEST_CONFIG_NO_EXCEPTIONS
+    void throwException() {}
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS
+} // namespace detail
+
+namespace {
+    using namespace detail;
+    // matching of a string against a wildcard mask (case sensitivity configurable) taken from
+    // https://www.codeproject.com/Articles/1088/Wildcard-string-compare-globbing
+    int wildcmp(const char* str, const char* wild, bool caseSensitive) {
+        const char* cp = str;
+        const char* mp = wild;
+
+        while((*str) && (*wild != '*')) {
+            if((caseSensitive ? (*wild != *str) : (tolower(*wild) != tolower(*str))) &&
+               (*wild != '?')) {
+                return 0;
+            }
+            wild++;
+            str++;
+        }
+
+        while(*str) {
+            if(*wild == '*') {
+                if(!*++wild) {
+                    return 1;
+                }
+                mp = wild;
+                cp = str + 1;
+            } else if((caseSensitive ? (*wild == *str) : (tolower(*wild) == tolower(*str))) ||
+                      (*wild == '?')) {
+                wild++;
+                str++;
+            } else {
+                wild = mp;   //!OCLINT parameter reassignment
+                str  = cp++; //!OCLINT parameter reassignment
+            }
+        }
+
+        while(*wild == '*') {
+            wild++;
+        }
+        return !*wild;
+    }
+
+    //// C string hash function (djb2) - taken from http://www.cse.yorku.ca/~oz/hash.html
+    //unsigned hashStr(unsigned const char* str) {
+    //    unsigned long hash = 5381;
+    //    char          c;
+    //    while((c = *str++))
+    //        hash = ((hash << 5) + hash) + c; // hash * 33 + c
+    //    return hash;
+    //}
+
+    // checks if the name matches any of the filters (and can be configured what to do when empty)
+    bool matchesAny(const char* name, const std::vector<String>& filters, bool matchEmpty,
+                    bool caseSensitive) {
+        if(filters.empty() && matchEmpty)
+            return true;
+        for(auto& curr : filters)
+            if(wildcmp(name, curr.c_str(), caseSensitive))
+                return true;
+        return false;
+    }
+} // namespace
+namespace detail {
+
+    Subcase::Subcase(const String& name, const char* file, int line)
+            : m_signature({name, file, line}) {
+        ContextState* s = g_cs;
+
+        // check subcase filters
+        if(s->subcasesStack.size() < size_t(s->subcase_filter_levels)) {
+            if(!matchesAny(m_signature.m_name.c_str(), s->filters[6], true, s->case_sensitive))
+                return;
+            if(matchesAny(m_signature.m_name.c_str(), s->filters[7], false, s->case_sensitive))
+                return;
+        }
+        
+        // if a Subcase on the same level has already been entered
+        if(s->subcasesStack.size() < size_t(s->subcasesCurrentMaxLevel)) {
+            s->should_reenter = true;
+            return;
+        }
+
+        // push the current signature to the stack so we can check if the
+        // current stack + the current new subcase have been traversed
+        s->subcasesStack.push_back(m_signature);
+        if(s->subcasesPassed.count(s->subcasesStack) != 0) {
+            // pop - revert to previous stack since we've already passed this
+            s->subcasesStack.pop_back();
+            return;
+        }
+
+        s->subcasesCurrentMaxLevel = s->subcasesStack.size();
+        m_entered = true;
+
+        DOCTEST_ITERATE_THROUGH_REPORTERS(subcase_start, m_signature);
+    }
+
+    DOCTEST_MSVC_SUPPRESS_WARNING_WITH_PUSH(4996) // std::uncaught_exception is deprecated in C++17	
+    DOCTEST_GCC_SUPPRESS_WARNING_WITH_PUSH("-Wdeprecated-declarations")	
+    DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Wdeprecated-declarations")
+
+    Subcase::~Subcase() {
+        if(m_entered) {
+            // only mark the subcase stack as passed if no subcases have been skipped
+            if(g_cs->should_reenter == false)
+                g_cs->subcasesPassed.insert(g_cs->subcasesStack);
+            g_cs->subcasesStack.pop_back();
+
+#if defined(__cpp_lib_uncaught_exceptions) && __cpp_lib_uncaught_exceptions >= 201411L
+            if(std::uncaught_exceptions() > 0
+#else
+            if(std::uncaught_exception()
+#endif
+            && g_cs->shouldLogCurrentException) {
+                DOCTEST_ITERATE_THROUGH_REPORTERS(
+                        test_case_exception, {"exception thrown in subcase - will translate later "
+                                              "when the whole test case has been exited (cannot "
+                                              "translate while there is an active exception)",
+                                              false});
+                g_cs->shouldLogCurrentException = false;
+            }
+            DOCTEST_ITERATE_THROUGH_REPORTERS(subcase_end, DOCTEST_EMPTY);
+        }
+    }
+
+    DOCTEST_CLANG_SUPPRESS_WARNING_POP	
+    DOCTEST_GCC_SUPPRESS_WARNING_POP	
+    DOCTEST_MSVC_SUPPRESS_WARNING_POP
+
+    Subcase::operator bool() const { return m_entered; }
+
+    Result::Result(bool passed, const String& decomposition)
+            : m_passed(passed)
+            , m_decomp(decomposition) {}
+
+    ExpressionDecomposer::ExpressionDecomposer(assertType::Enum at)
+            : m_at(at) {}
+
+    TestSuite& TestSuite::operator*(const char* in) {
+        m_test_suite = in;
+        // clear state
+        m_description       = nullptr;
+        m_skip              = false;
+        m_may_fail          = false;
+        m_should_fail       = false;
+        m_expected_failures = 0;
+        m_timeout           = 0;
+        return *this;
+    }
+
+    TestCase::TestCase(funcType test, const char* file, unsigned line, const TestSuite& test_suite,
+                       const char* type, int template_id) {
+        m_file              = file;
+        m_line              = line;
+        m_name              = nullptr; // will be later overridden in operator*
+        m_test_suite        = test_suite.m_test_suite;
+        m_description       = test_suite.m_description;
+        m_skip              = test_suite.m_skip;
+        m_may_fail          = test_suite.m_may_fail;
+        m_should_fail       = test_suite.m_should_fail;
+        m_expected_failures = test_suite.m_expected_failures;
+        m_timeout           = test_suite.m_timeout;
+
+        m_test        = test;
+        m_type        = type;
+        m_template_id = template_id;
+    }
+
+    TestCase::TestCase(const TestCase& other)
+            : TestCaseData() {
+        *this = other;
+    }
+
+    DOCTEST_MSVC_SUPPRESS_WARNING_WITH_PUSH(26434) // hides a non-virtual function
+    DOCTEST_MSVC_SUPPRESS_WARNING(26437)           // Do not slice
+    TestCase& TestCase::operator=(const TestCase& other) {
+        static_cast<TestCaseData&>(*this) = static_cast<const TestCaseData&>(other);
+
+        m_test        = other.m_test;
+        m_type        = other.m_type;
+        m_template_id = other.m_template_id;
+        m_full_name   = other.m_full_name;
+
+        if(m_template_id != -1)
+            m_name = m_full_name.c_str();
+        return *this;
+    }
+    DOCTEST_MSVC_SUPPRESS_WARNING_POP
+
+    TestCase& TestCase::operator*(const char* in) {
+        m_name = in;
+        // make a new name with an appended type for templated test case
+        if(m_template_id != -1) {
+            m_full_name = String(m_name) + m_type;
+            // redirect the name to point to the newly constructed full name
+            m_name = m_full_name.c_str();
+        }
+        return *this;
+    }
+
+    bool TestCase::operator<(const TestCase& other) const {
+        if(m_line != other.m_line)
+            return m_line < other.m_line;
+        const int file_cmp = m_file.compare(other.m_file);
+        if(file_cmp != 0)
+            return file_cmp < 0;
+        return m_template_id < other.m_template_id;
+    }
+} // namespace detail
+namespace {
+    using namespace detail;
+    // for sorting tests by file/line
+    bool fileOrderComparator(const TestCase* lhs, const TestCase* rhs) {
+        // this is needed because MSVC gives different case for drive letters
+        // for __FILE__ when evaluated in a header and a source file
+        const int res = lhs->m_file.compare(rhs->m_file, bool(DOCTEST_MSVC));
+        if(res != 0)
+            return res < 0;
+        if(lhs->m_line != rhs->m_line)
+            return lhs->m_line < rhs->m_line;
+        return lhs->m_template_id < rhs->m_template_id;
+    }
+
+    // for sorting tests by suite/file/line
+    bool suiteOrderComparator(const TestCase* lhs, const TestCase* rhs) {
+        const int res = std::strcmp(lhs->m_test_suite, rhs->m_test_suite);
+        if(res != 0)
+            return res < 0;
+        return fileOrderComparator(lhs, rhs);
+    }
+
+    // for sorting tests by name/suite/file/line
+    bool nameOrderComparator(const TestCase* lhs, const TestCase* rhs) {
+        const int res = std::strcmp(lhs->m_name, rhs->m_name);
+        if(res != 0)
+            return res < 0;
+        return suiteOrderComparator(lhs, rhs);
+    }
+
+    // all the registered tests
+    std::set<TestCase>& getRegisteredTests() {
+        static std::set<TestCase> data;
+        return data;
+    }
+
+#ifdef DOCTEST_CONFIG_COLORS_WINDOWS
+    HANDLE g_stdoutHandle;
+    WORD   g_origFgAttrs;
+    WORD   g_origBgAttrs;
+    bool   g_attrsInitted = false;
+
+    int colors_init() {
+        if(!g_attrsInitted) {
+            g_stdoutHandle = GetStdHandle(STD_OUTPUT_HANDLE);
+            g_attrsInitted = true;
+            CONSOLE_SCREEN_BUFFER_INFO csbiInfo;
+            GetConsoleScreenBufferInfo(g_stdoutHandle, &csbiInfo);
+            g_origFgAttrs = csbiInfo.wAttributes & ~(BACKGROUND_GREEN | BACKGROUND_RED |
+                                                     BACKGROUND_BLUE | BACKGROUND_INTENSITY);
+            g_origBgAttrs = csbiInfo.wAttributes & ~(FOREGROUND_GREEN | FOREGROUND_RED |
+                                                     FOREGROUND_BLUE | FOREGROUND_INTENSITY);
+        }
+        return 0;
+    }
+
+    int dumy_init_console_colors = colors_init();
+#endif // DOCTEST_CONFIG_COLORS_WINDOWS
+
+    DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Wdeprecated-declarations")
+    void color_to_stream(std::ostream& s, Color::Enum code) {
+        ((void)s);    // for DOCTEST_CONFIG_COLORS_NONE or DOCTEST_CONFIG_COLORS_WINDOWS
+        ((void)code); // for DOCTEST_CONFIG_COLORS_NONE
+#ifdef DOCTEST_CONFIG_COLORS_ANSI
+        if(g_no_colors ||
+           (isatty(STDOUT_FILENO) == false && getContextOptions()->force_colors == false))
+            return;
+
+        auto col = "";
+        // clang-format off
+            switch(code) { //!OCLINT missing break in switch statement / unnecessary default statement in covered switch statement
+                case Color::Red:         col = "[0;31m"; break;
+                case Color::Green:       col = "[0;32m"; break;
+                case Color::Blue:        col = "[0;34m"; break;
+                case Color::Cyan:        col = "[0;36m"; break;
+                case Color::Yellow:      col = "[0;33m"; break;
+                case Color::Grey:        col = "[1;30m"; break;
+                case Color::LightGrey:   col = "[0;37m"; break;
+                case Color::BrightRed:   col = "[1;31m"; break;
+                case Color::BrightGreen: col = "[1;32m"; break;
+                case Color::BrightWhite: col = "[1;37m"; break;
+                case Color::Bright: // invalid
+                case Color::None:
+                case Color::White:
+                default:                 col = "[0m";
+            }
+        // clang-format on
+        s << "\033" << col;
+#endif // DOCTEST_CONFIG_COLORS_ANSI
+
+#ifdef DOCTEST_CONFIG_COLORS_WINDOWS
+        if(g_no_colors ||
+           (isatty(fileno(stdout)) == false && getContextOptions()->force_colors == false))
+            return;
+
+#define DOCTEST_SET_ATTR(x) SetConsoleTextAttribute(g_stdoutHandle, x | g_origBgAttrs)
+
+        // clang-format off
+        switch (code) {
+            case Color::White:       DOCTEST_SET_ATTR(FOREGROUND_GREEN | FOREGROUND_RED | FOREGROUND_BLUE); break;
+            case Color::Red:         DOCTEST_SET_ATTR(FOREGROUND_RED);                                      break;
+            case Color::Green:       DOCTEST_SET_ATTR(FOREGROUND_GREEN);                                    break;
+            case Color::Blue:        DOCTEST_SET_ATTR(FOREGROUND_BLUE);                                     break;
+            case Color::Cyan:        DOCTEST_SET_ATTR(FOREGROUND_BLUE | FOREGROUND_GREEN);                  break;
+            case Color::Yellow:      DOCTEST_SET_ATTR(FOREGROUND_RED | FOREGROUND_GREEN);                   break;
+            case Color::Grey:        DOCTEST_SET_ATTR(0);                                                   break;
+            case Color::LightGrey:   DOCTEST_SET_ATTR(FOREGROUND_INTENSITY);                                break;
+            case Color::BrightRed:   DOCTEST_SET_ATTR(FOREGROUND_INTENSITY | FOREGROUND_RED);               break;
+            case Color::BrightGreen: DOCTEST_SET_ATTR(FOREGROUND_INTENSITY | FOREGROUND_GREEN);             break;
+            case Color::BrightWhite: DOCTEST_SET_ATTR(FOREGROUND_INTENSITY | FOREGROUND_GREEN | FOREGROUND_RED | FOREGROUND_BLUE); break;
+            case Color::None:
+            case Color::Bright: // invalid
+            default:                 DOCTEST_SET_ATTR(g_origFgAttrs);
+        }
+            // clang-format on
+#endif // DOCTEST_CONFIG_COLORS_WINDOWS
+    }
+    DOCTEST_CLANG_SUPPRESS_WARNING_POP
+
+    std::vector<const IExceptionTranslator*>& getExceptionTranslators() {
+        static std::vector<const IExceptionTranslator*> data;
+        return data;
+    }
+
+    String translateActiveException() {
+#ifndef DOCTEST_CONFIG_NO_EXCEPTIONS
+        String res;
+        auto&  translators = getExceptionTranslators();
+        for(auto& curr : translators)
+            if(curr->translate(res))
+                return res;
+        // clang-format off
+        DOCTEST_GCC_SUPPRESS_WARNING_WITH_PUSH("-Wcatch-value")
+        try {
+            throw;
+        } catch(std::exception& ex) {
+            return ex.what();
+        } catch(std::string& msg) {
+            return msg.c_str();
+        } catch(const char* msg) {
+            return msg;
+        } catch(...) {
+            return "unknown exception";
+        }
+        DOCTEST_GCC_SUPPRESS_WARNING_POP
+// clang-format on
+#else  // DOCTEST_CONFIG_NO_EXCEPTIONS
+        return "";
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS
+    }
+} // namespace
+
+namespace detail {
+    // used by the macros for registering tests
+    int regTest(const TestCase& tc) {
+        getRegisteredTests().insert(tc);
+        return 0;
+    }
+
+    // sets the current test suite
+    int setTestSuite(const TestSuite& ts) {
+        doctest_detail_test_suite_ns::getCurrentTestSuite() = ts;
+        return 0;
+    }
+
+#ifdef DOCTEST_IS_DEBUGGER_ACTIVE
+    bool isDebuggerActive() { return DOCTEST_IS_DEBUGGER_ACTIVE(); }
+#else // DOCTEST_IS_DEBUGGER_ACTIVE
+#ifdef DOCTEST_PLATFORM_MAC
+    // The following function is taken directly from the following technical note:
+    // https://developer.apple.com/library/archive/qa/qa1361/_index.html
+    // Returns true if the current process is being debugged (either
+    // running under the debugger or has a debugger attached post facto).
+    bool isDebuggerActive() {
+        int        mib[4];
+        kinfo_proc info;
+        size_t     size;
+        // Initialize the flags so that, if sysctl fails for some bizarre
+        // reason, we get a predictable result.
+        info.kp_proc.p_flag = 0;
+        // Initialize mib, which tells sysctl the info we want, in this case
+        // we're looking for information about a specific process ID.
+        mib[0] = CTL_KERN;
+        mib[1] = KERN_PROC;
+        mib[2] = KERN_PROC_PID;
+        mib[3] = getpid();
+        // Call sysctl.
+        size = sizeof(info);
+        if(sysctl(mib, DOCTEST_COUNTOF(mib), &info, &size, 0, 0) != 0) {
+            std::cerr << "\nCall to sysctl failed - unable to determine if debugger is active **\n";
+            return false;
+        }
+        // We're being debugged if the P_TRACED flag is set.
+        return ((info.kp_proc.p_flag & P_TRACED) != 0);
+    }
+#elif DOCTEST_MSVC || defined(__MINGW32__) || defined(__MINGW64__)
+    bool isDebuggerActive() { return ::IsDebuggerPresent() != 0; }
+#else
+    bool isDebuggerActive() { return false; }
+#endif // Platform
+#endif // DOCTEST_IS_DEBUGGER_ACTIVE
+
+    void registerExceptionTranslatorImpl(const IExceptionTranslator* et) {
+        if(std::find(getExceptionTranslators().begin(), getExceptionTranslators().end(), et) ==
+           getExceptionTranslators().end())
+            getExceptionTranslators().push_back(et);
+    }
+
+#ifdef DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+    void toStream(std::ostream* s, char* in) { *s << in; }
+    void toStream(std::ostream* s, const char* in) { *s << in; }
+#endif // DOCTEST_CONFIG_TREAT_CHAR_STAR_AS_STRING
+    void toStream(std::ostream* s, bool in) { *s << std::boolalpha << in << std::noboolalpha; }
+    void toStream(std::ostream* s, float in) { *s << in; }
+    void toStream(std::ostream* s, double in) { *s << in; }
+    void toStream(std::ostream* s, double long in) { *s << in; }
+
+    void toStream(std::ostream* s, char in) { *s << in; }
+    void toStream(std::ostream* s, char signed in) { *s << in; }
+    void toStream(std::ostream* s, char unsigned in) { *s << in; }
+    void toStream(std::ostream* s, int short in) { *s << in; }
+    void toStream(std::ostream* s, int short unsigned in) { *s << in; }
+    void toStream(std::ostream* s, int in) { *s << in; }
+    void toStream(std::ostream* s, int unsigned in) { *s << in; }
+    void toStream(std::ostream* s, int long in) { *s << in; }
+    void toStream(std::ostream* s, int long unsigned in) { *s << in; }
+    void toStream(std::ostream* s, int long long in) { *s << in; }
+    void toStream(std::ostream* s, int long long unsigned in) { *s << in; }
+
+    DOCTEST_THREAD_LOCAL std::vector<IContextScope*> g_infoContexts; // for logging with INFO()
+
+    ContextScopeBase::ContextScopeBase() {
+        g_infoContexts.push_back(this);
+    }
+
+    DOCTEST_MSVC_SUPPRESS_WARNING_WITH_PUSH(4996) // std::uncaught_exception is deprecated in C++17	
+    DOCTEST_GCC_SUPPRESS_WARNING_WITH_PUSH("-Wdeprecated-declarations")	
+    DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Wdeprecated-declarations")
+
+    // destroy cannot be inlined into the destructor because that would mean calling stringify after
+    // ContextScope has been destroyed (base class destructors run after derived class destructors).
+    // Instead, ContextScope calls this method directly from its destructor.
+    void ContextScopeBase::destroy() {
+#if defined(__cpp_lib_uncaught_exceptions) && __cpp_lib_uncaught_exceptions >= 201411L
+        if(std::uncaught_exceptions() > 0) {
+#else
+        if(std::uncaught_exception()) {
+#endif
+            std::ostringstream s;
+            this->stringify(&s);
+            g_cs->stringifiedContexts.push_back(s.str().c_str());
+        }
+        g_infoContexts.pop_back();
+    }
+
+    DOCTEST_CLANG_SUPPRESS_WARNING_POP	
+    DOCTEST_GCC_SUPPRESS_WARNING_POP	
+    DOCTEST_MSVC_SUPPRESS_WARNING_POP
+} // namespace detail
+namespace {
+    using namespace detail;
+
+#if !defined(DOCTEST_CONFIG_POSIX_SIGNALS) && !defined(DOCTEST_CONFIG_WINDOWS_SEH)
+    struct FatalConditionHandler
+    {
+        void reset() {}
+    };
+#else // DOCTEST_CONFIG_POSIX_SIGNALS || DOCTEST_CONFIG_WINDOWS_SEH
+
+    void reportFatal(const std::string&);
+
+#ifdef DOCTEST_PLATFORM_WINDOWS
+
+    struct SignalDefs
+    {
+        DWORD id;
+        const char* name;
+    };
+    // There is no 1-1 mapping between signals and windows exceptions.
+    // Windows can easily distinguish between SO and SigSegV,
+    // but SigInt, SigTerm, etc are handled differently.
+    SignalDefs signalDefs[] = {
+            {EXCEPTION_ILLEGAL_INSTRUCTION, "SIGILL - Illegal instruction signal"},
+            {EXCEPTION_STACK_OVERFLOW, "SIGSEGV - Stack overflow"},
+            {EXCEPTION_ACCESS_VIOLATION, "SIGSEGV - Segmentation violation signal"},
+            {EXCEPTION_INT_DIVIDE_BY_ZERO, "Divide by zero error"},
+    };
+
+    struct FatalConditionHandler
+    {
+        static LONG CALLBACK handleException(PEXCEPTION_POINTERS ExceptionInfo) {
+            for(size_t i = 0; i < DOCTEST_COUNTOF(signalDefs); ++i) {
+                if(ExceptionInfo->ExceptionRecord->ExceptionCode == signalDefs[i].id) {
+                    reportFatal(signalDefs[i].name);
+                    break;
+                }
+            }
+            // If its not an exception we care about, pass it along.
+            // This stops us from eating debugger breaks etc.
+            return EXCEPTION_CONTINUE_SEARCH;
+        }
+
+        FatalConditionHandler() {
+            isSet = true;
+            // 32k seems enough for doctest to handle stack overflow,
+            // but the value was found experimentally, so there is no strong guarantee
+            guaranteeSize = 32 * 1024;
+            // Register an unhandled exception filter
+            previousTop = SetUnhandledExceptionFilter(handleException);
+            // Pass in guarantee size to be filled
+            SetThreadStackGuarantee(&guaranteeSize);
+        }
+
+        static void reset() {
+            if(isSet) {
+                // Unregister handler and restore the old guarantee
+                SetUnhandledExceptionFilter(previousTop);
+                SetThreadStackGuarantee(&guaranteeSize);
+                previousTop = nullptr;
+                isSet = false;
+            }
+        }
+
+        ~FatalConditionHandler() { reset(); }
+
+    private:
+        static bool isSet;
+        static ULONG guaranteeSize;
+        static LPTOP_LEVEL_EXCEPTION_FILTER previousTop;
+    };
+
+    bool FatalConditionHandler::isSet = false;
+    ULONG FatalConditionHandler::guaranteeSize = 0;
+    LPTOP_LEVEL_EXCEPTION_FILTER FatalConditionHandler::previousTop = nullptr;
+
+#else // DOCTEST_PLATFORM_WINDOWS
+
+    struct SignalDefs
+    {
+        int         id;
+        const char* name;
+    };
+    SignalDefs signalDefs[] = {{SIGINT, "SIGINT - Terminal interrupt signal"},
+                               {SIGILL, "SIGILL - Illegal instruction signal"},
+                               {SIGFPE, "SIGFPE - Floating point error signal"},
+                               {SIGSEGV, "SIGSEGV - Segmentation violation signal"},
+                               {SIGTERM, "SIGTERM - Termination request signal"},
+                               {SIGABRT, "SIGABRT - Abort (abnormal termination) signal"}};
+
+    struct FatalConditionHandler
+    {
+        static bool             isSet;
+        static struct sigaction oldSigActions[DOCTEST_COUNTOF(signalDefs)];
+        static stack_t          oldSigStack;
+        static char             altStackMem[4 * SIGSTKSZ];
+
+        static void handleSignal(int sig) {
+            const char* name = "<unknown signal>";
+            for(std::size_t i = 0; i < DOCTEST_COUNTOF(signalDefs); ++i) {
+                SignalDefs& def = signalDefs[i];
+                if(sig == def.id) {
+                    name = def.name;
+                    break;
+                }
+            }
+            reset();
+            reportFatal(name);
+            raise(sig);
+        }
+
+        FatalConditionHandler() {
+            isSet = true;
+            stack_t sigStack;
+            sigStack.ss_sp    = altStackMem;
+            sigStack.ss_size  = sizeof(altStackMem);
+            sigStack.ss_flags = 0;
+            sigaltstack(&sigStack, &oldSigStack);
+            struct sigaction sa = {};
+            sa.sa_handler       = handleSignal; // NOLINT
+            sa.sa_flags         = SA_ONSTACK;
+            for(std::size_t i = 0; i < DOCTEST_COUNTOF(signalDefs); ++i) {
+                sigaction(signalDefs[i].id, &sa, &oldSigActions[i]);
+            }
+        }
+
+        ~FatalConditionHandler() { reset(); }
+        static void reset() {
+            if(isSet) {
+                // Set signals back to previous values -- hopefully nobody overwrote them in the meantime
+                for(std::size_t i = 0; i < DOCTEST_COUNTOF(signalDefs); ++i) {
+                    sigaction(signalDefs[i].id, &oldSigActions[i], nullptr);
+                }
+                // Return the old stack
+                sigaltstack(&oldSigStack, nullptr);
+                isSet = false;
+            }
+        }
+    };
+
+    bool             FatalConditionHandler::isSet                                      = false;
+    struct sigaction FatalConditionHandler::oldSigActions[DOCTEST_COUNTOF(signalDefs)] = {};
+    stack_t          FatalConditionHandler::oldSigStack                                = {};
+    char             FatalConditionHandler::altStackMem[]                              = {};
+
+#endif // DOCTEST_PLATFORM_WINDOWS
+#endif // DOCTEST_CONFIG_POSIX_SIGNALS || DOCTEST_CONFIG_WINDOWS_SEH
+
+} // namespace
+
+namespace {
+    using namespace detail;
+
+#ifdef DOCTEST_PLATFORM_WINDOWS
+#define DOCTEST_OUTPUT_DEBUG_STRING(text) ::OutputDebugStringA(text)
+#else
+    // TODO: integration with XCode and other IDEs
+#define DOCTEST_OUTPUT_DEBUG_STRING(text) // NOLINT(clang-diagnostic-unused-macros)
+#endif // Platform
+
+    void addAssert(assertType::Enum at) {
+        if((at & assertType::is_warn) == 0) //!OCLINT bitwise operator in conditional
+            g_cs->numAssertsCurrentTest_atomic++;
+    }
+
+    void addFailedAssert(assertType::Enum at) {
+        if((at & assertType::is_warn) == 0) //!OCLINT bitwise operator in conditional
+            g_cs->numAssertsFailedCurrentTest_atomic++;
+    }
+
+#if defined(DOCTEST_CONFIG_POSIX_SIGNALS) || defined(DOCTEST_CONFIG_WINDOWS_SEH)
+    void reportFatal(const std::string& message) {
+        g_cs->failure_flags |= TestCaseFailureReason::Crash;
+
+        DOCTEST_ITERATE_THROUGH_REPORTERS(test_case_exception, {message.c_str(), true});
+
+        while(g_cs->subcasesStack.size()) {
+            g_cs->subcasesStack.pop_back();
+            DOCTEST_ITERATE_THROUGH_REPORTERS(subcase_end, DOCTEST_EMPTY);
+        }
+
+        g_cs->finalizeTestCaseData();
+
+        DOCTEST_ITERATE_THROUGH_REPORTERS(test_case_end, *g_cs);
+
+        DOCTEST_ITERATE_THROUGH_REPORTERS(test_run_end, *g_cs);
+    }
+#endif // DOCTEST_CONFIG_POSIX_SIGNALS || DOCTEST_CONFIG_WINDOWS_SEH
+} // namespace
+namespace detail {
+
+    ResultBuilder::ResultBuilder(assertType::Enum at, const char* file, int line, const char* expr,
+                                 const char* exception_type, const char* exception_string) {
+        m_test_case        = g_cs->currentTest;
+        m_at               = at;
+        m_file             = file;
+        m_line             = line;
+        m_expr             = expr;
+        m_failed           = true;
+        m_threw            = false;
+        m_threw_as         = false;
+        m_exception_type   = exception_type;
+        m_exception_string = exception_string;
+#if DOCTEST_MSVC
+        if(m_expr[0] == ' ') // this happens when variadic macros are disabled under MSVC
+            ++m_expr;
+#endif // MSVC
+    }
+
+    void ResultBuilder::setResult(const Result& res) {
+        m_decomp = res.m_decomp;
+        m_failed = !res.m_passed;
+    }
+
+    void ResultBuilder::translateException() {
+        m_threw     = true;
+        m_exception = translateActiveException();
+    }
+
+    bool ResultBuilder::log() {
+        if(m_at & assertType::is_throws) { //!OCLINT bitwise operator in conditional
+            m_failed = !m_threw;
+        } else if((m_at & assertType::is_throws_as) && (m_at & assertType::is_throws_with)) { //!OCLINT
+            m_failed = !m_threw_as || (m_exception != m_exception_string);
+        } else if(m_at & assertType::is_throws_as) { //!OCLINT bitwise operator in conditional
+            m_failed = !m_threw_as;
+        } else if(m_at & assertType::is_throws_with) { //!OCLINT bitwise operator in conditional
+            m_failed = m_exception != m_exception_string;
+        } else if(m_at & assertType::is_nothrow) { //!OCLINT bitwise operator in conditional
+            m_failed = m_threw;
+        }
+
+        if(m_exception.size())
+            m_exception = String("\"") + m_exception + "\"";
+
+        if(is_running_in_test) {
+            addAssert(m_at);
+            DOCTEST_ITERATE_THROUGH_REPORTERS(log_assert, *this);
+
+            if(m_failed)
+                addFailedAssert(m_at);
+        } else if(m_failed) {
+            failed_out_of_a_testing_context(*this);
+        }
+
+        return m_failed && isDebuggerActive() &&
+               !getContextOptions()->no_breaks; // break into debugger
+    }
+
+    void ResultBuilder::react() const {
+        if(m_failed && checkIfShouldThrow(m_at))
+            throwException();
+    }
+
+    void failed_out_of_a_testing_context(const AssertData& ad) {
+        if(g_cs->ah)
+            g_cs->ah(ad);
+        else
+            std::abort();
+    }
+
+    void decomp_assert(assertType::Enum at, const char* file, int line, const char* expr,
+                       Result result) {
+        bool failed = !result.m_passed;
+
+        // ###################################################################################
+        // IF THE DEBUGGER BREAKS HERE - GO 1 LEVEL UP IN THE CALLSTACK FOR THE FAILING ASSERT
+        // THIS IS THE EFFECT OF HAVING 'DOCTEST_CONFIG_SUPER_FAST_ASSERTS' DEFINED
+        // ###################################################################################
+        DOCTEST_ASSERT_OUT_OF_TESTS(result.m_decomp);
+        DOCTEST_ASSERT_IN_TESTS(result.m_decomp);
+    }
+
+    MessageBuilder::MessageBuilder(const char* file, int line, assertType::Enum severity) {
+        m_stream   = getTlsOss();
+        m_file     = file;
+        m_line     = line;
+        m_severity = severity;
+    }
+
+    IExceptionTranslator::IExceptionTranslator()  = default;
+    IExceptionTranslator::~IExceptionTranslator() = default;
+
+    bool MessageBuilder::log() {
+        m_string = getTlsOssResult();
+        DOCTEST_ITERATE_THROUGH_REPORTERS(log_message, *this);
+
+        const bool isWarn = m_severity & assertType::is_warn;
+
+        // warn is just a message in this context so we don't treat it as an assert
+        if(!isWarn) {
+            addAssert(m_severity);
+            addFailedAssert(m_severity);
+        }
+
+        return isDebuggerActive() && !getContextOptions()->no_breaks && !isWarn; // break
+    }
+
+    void MessageBuilder::react() {
+        if(m_severity & assertType::is_require) //!OCLINT bitwise operator in conditional
+            throwException();
+    }
+
+    MessageBuilder::~MessageBuilder() = default;
+} // namespace detail
+namespace {
+    using namespace detail;
+
+    template <typename Ex>
+    DOCTEST_NORETURN void throw_exception(Ex const& e) {
+#ifndef DOCTEST_CONFIG_NO_EXCEPTIONS
+        throw e;
+#else  // DOCTEST_CONFIG_NO_EXCEPTIONS
+        std::cerr << "doctest will terminate because it needed to throw an exception.\n"
+                  << "The message was: " << e.what() << '\n';
+        std::terminate();
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS
+    }
+
+#ifndef DOCTEST_INTERNAL_ERROR
+#define DOCTEST_INTERNAL_ERROR(msg)                                                                \
+    throw_exception(std::logic_error(                                                              \
+            __FILE__ ":" DOCTEST_TOSTR(__LINE__) ": Internal doctest error: " msg))
+#endif // DOCTEST_INTERNAL_ERROR
+
+    // clang-format off
+
+// =================================================================================================
+// The following code has been taken verbatim from Catch2/include/internal/catch_xmlwriter.h/cpp
+// This is done so cherry-picking bug fixes is trivial - even the style/formatting is untouched.
+// =================================================================================================
+
+    class XmlEncode {
+    public:
+        enum ForWhat { ForTextNodes, ForAttributes };
+
+        XmlEncode( std::string const& str, ForWhat forWhat = ForTextNodes );
+
+        void encodeTo( std::ostream& os ) const;
+
+        friend std::ostream& operator << ( std::ostream& os, XmlEncode const& xmlEncode );
+
+    private:
+        std::string m_str;
+        ForWhat m_forWhat;
+    };
+
+    class XmlWriter {
+    public:
+
+        class ScopedElement {
+        public:
+            ScopedElement( XmlWriter* writer );
+
+            ScopedElement( ScopedElement&& other ) DOCTEST_NOEXCEPT;
+            ScopedElement& operator=( ScopedElement&& other ) DOCTEST_NOEXCEPT;
+
+            ~ScopedElement();
+
+            ScopedElement& writeText( std::string const& text, bool indent = true );
+
+            template<typename T>
+            ScopedElement& writeAttribute( std::string const& name, T const& attribute ) {
+                m_writer->writeAttribute( name, attribute );
+                return *this;
+            }
+
+        private:
+            mutable XmlWriter* m_writer = nullptr;
+        };
+
+        XmlWriter( std::ostream& os = std::cout );
+        ~XmlWriter();
+
+        XmlWriter( XmlWriter const& ) = delete;
+        XmlWriter& operator=( XmlWriter const& ) = delete;
+
+        XmlWriter& startElement( std::string const& name );
+
+        ScopedElement scopedElement( std::string const& name );
+
+        XmlWriter& endElement();
+
+        XmlWriter& writeAttribute( std::string const& name, std::string const& attribute );
+
+        XmlWriter& writeAttribute( std::string const& name, const char* attribute );
+
+        XmlWriter& writeAttribute( std::string const& name, bool attribute );
+
+        template<typename T>
+        XmlWriter& writeAttribute( std::string const& name, T const& attribute ) {
+        std::stringstream rss;
+            rss << attribute;
+            return writeAttribute( name, rss.str() );
+        }
+
+        XmlWriter& writeText( std::string const& text, bool indent = true );
+
+        //XmlWriter& writeComment( std::string const& text );
+
+        //void writeStylesheetRef( std::string const& url );
+
+        //XmlWriter& writeBlankLine();
+
+        void ensureTagClosed();
+
+    private:
+
+        void writeDeclaration();
+
+        void newlineIfNecessary();
+
+        bool m_tagIsOpen = false;
+        bool m_needsNewline = false;
+        std::vector<std::string> m_tags;
+        std::string m_indent;
+        std::ostream& m_os;
+    };
+
+// =================================================================================================
+// The following code has been taken verbatim from Catch2/include/internal/catch_xmlwriter.h/cpp
+// This is done so cherry-picking bug fixes is trivial - even the style/formatting is untouched.
+// =================================================================================================
+
+using uchar = unsigned char;
+
+namespace {
+
+    size_t trailingBytes(unsigned char c) {
+        if ((c & 0xE0) == 0xC0) {
+            return 2;
+        }
+        if ((c & 0xF0) == 0xE0) {
+            return 3;
+        }
+        if ((c & 0xF8) == 0xF0) {
+            return 4;
+        }
+        DOCTEST_INTERNAL_ERROR("Invalid multibyte utf-8 start byte encountered");
+    }
+
+    uint32_t headerValue(unsigned char c) {
+        if ((c & 0xE0) == 0xC0) {
+            return c & 0x1F;
+        }
+        if ((c & 0xF0) == 0xE0) {
+            return c & 0x0F;
+        }
+        if ((c & 0xF8) == 0xF0) {
+            return c & 0x07;
+        }
+        DOCTEST_INTERNAL_ERROR("Invalid multibyte utf-8 start byte encountered");
+    }
+
+    void hexEscapeChar(std::ostream& os, unsigned char c) {
+        std::ios_base::fmtflags f(os.flags());
+        os << "\\x"
+            << std::uppercase << std::hex << std::setfill('0') << std::setw(2)
+            << static_cast<int>(c);
+        os.flags(f);
+    }
+
+} // anonymous namespace
+
+    XmlEncode::XmlEncode( std::string const& str, ForWhat forWhat )
+    :   m_str( str ),
+        m_forWhat( forWhat )
+    {}
+
+    void XmlEncode::encodeTo( std::ostream& os ) const {
+        // Apostrophe escaping not necessary if we always use " to write attributes
+        // (see: https://www.w3.org/TR/xml/#syntax)
+
+        for( std::size_t idx = 0; idx < m_str.size(); ++ idx ) {
+            uchar c = m_str[idx];
+            switch (c) {
+            case '<':   os << "&lt;"; break;
+            case '&':   os << "&amp;"; break;
+
+            case '>':
+                // See: https://www.w3.org/TR/xml/#syntax
+                if (idx > 2 && m_str[idx - 1] == ']' && m_str[idx - 2] == ']')
+                    os << "&gt;";
+                else
+                    os << c;
+                break;
+
+            case '\"':
+                if (m_forWhat == ForAttributes)
+                    os << "&quot;";
+                else
+                    os << c;
+                break;
+
+            default:
+                // Check for control characters and invalid utf-8
+
+                // Escape control characters in standard ascii
+                // see https://stackoverflow.com/questions/404107/why-are-control-characters-illegal-in-xml-1-0
+                if (c < 0x09 || (c > 0x0D && c < 0x20) || c == 0x7F) {
+                    hexEscapeChar(os, c);
+                    break;
+                }
+
+                // Plain ASCII: Write it to stream
+                if (c < 0x7F) {
+                    os << c;
+                    break;
+                }
+
+                // UTF-8 territory
+                // Check if the encoding is valid and if it is not, hex escape bytes.
+                // Important: We do not check the exact decoded values for validity, only the encoding format
+                // First check that this bytes is a valid lead byte:
+                // This means that it is not encoded as 1111 1XXX
+                // Or as 10XX XXXX
+                if (c <  0xC0 ||
+                    c >= 0xF8) {
+                    hexEscapeChar(os, c);
+                    break;
+                }
+
+                auto encBytes = trailingBytes(c);
+                // Are there enough bytes left to avoid accessing out-of-bounds memory?
+                if (idx + encBytes - 1 >= m_str.size()) {
+                    hexEscapeChar(os, c);
+                    break;
+                }
+                // The header is valid, check data
+                // The next encBytes bytes must together be a valid utf-8
+                // This means: bitpattern 10XX XXXX and the extracted value is sane (ish)
+                bool valid = true;
+                uint32_t value = headerValue(c);
+                for (std::size_t n = 1; n < encBytes; ++n) {
+                    uchar nc = m_str[idx + n];
+                    valid &= ((nc & 0xC0) == 0x80);
+                    value = (value << 6) | (nc & 0x3F);
+                }
+
+                if (
+                    // Wrong bit pattern of following bytes
+                    (!valid) ||
+                    // Overlong encodings
+                    (value < 0x80) ||
+                    (                 value < 0x800   && encBytes > 2) || // removed "0x80 <= value &&" because redundant
+                    (0x800 < value && value < 0x10000 && encBytes > 3) ||
+                    // Encoded value out of range
+                    (value >= 0x110000)
+                    ) {
+                    hexEscapeChar(os, c);
+                    break;
+                }
+
+                // If we got here, this is in fact a valid(ish) utf-8 sequence
+                for (std::size_t n = 0; n < encBytes; ++n) {
+                    os << m_str[idx + n];
+                }
+                idx += encBytes - 1;
+                break;
+            }
+        }
+    }
+
+    std::ostream& operator << ( std::ostream& os, XmlEncode const& xmlEncode ) {
+        xmlEncode.encodeTo( os );
+        return os;
+    }
+
+    XmlWriter::ScopedElement::ScopedElement( XmlWriter* writer )
+    :   m_writer( writer )
+    {}
+
+    XmlWriter::ScopedElement::ScopedElement( ScopedElement&& other ) DOCTEST_NOEXCEPT
+    :   m_writer( other.m_writer ){
+        other.m_writer = nullptr;
+    }
+    XmlWriter::ScopedElement& XmlWriter::ScopedElement::operator=( ScopedElement&& other ) DOCTEST_NOEXCEPT {
+        if ( m_writer ) {
+            m_writer->endElement();
+        }
+        m_writer = other.m_writer;
+        other.m_writer = nullptr;
+        return *this;
+    }
+
+
+    XmlWriter::ScopedElement::~ScopedElement() {
+        if( m_writer )
+            m_writer->endElement();
+    }
+
+    XmlWriter::ScopedElement& XmlWriter::ScopedElement::writeText( std::string const& text, bool indent ) {
+        m_writer->writeText( text, indent );
+        return *this;
+    }
+
+    XmlWriter::XmlWriter( std::ostream& os ) : m_os( os )
+    {
+        writeDeclaration();
+    }
+
+    XmlWriter::~XmlWriter() {
+        while( !m_tags.empty() )
+            endElement();
+    }
+
+    XmlWriter& XmlWriter::startElement( std::string const& name ) {
+        ensureTagClosed();
+        newlineIfNecessary();
+        m_os << m_indent << '<' << name;
+        m_tags.push_back( name );
+        m_indent += "  ";
+        m_tagIsOpen = true;
+        return *this;
+    }
+
+    XmlWriter::ScopedElement XmlWriter::scopedElement( std::string const& name ) {
+        ScopedElement scoped( this );
+        startElement( name );
+        return scoped;
+    }
+
+    XmlWriter& XmlWriter::endElement() {
+        newlineIfNecessary();
+        m_indent = m_indent.substr( 0, m_indent.size()-2 );
+        if( m_tagIsOpen ) {
+            m_os << "/>";
+            m_tagIsOpen = false;
+        }
+        else {
+            m_os << m_indent << "</" << m_tags.back() << ">";
+        }
+        m_os << std::endl;
+        m_tags.pop_back();
+        return *this;
+    }
+
+    XmlWriter& XmlWriter::writeAttribute( std::string const& name, std::string const& attribute ) {
+        if( !name.empty() && !attribute.empty() )
+            m_os << ' ' << name << "=\"" << XmlEncode( attribute, XmlEncode::ForAttributes ) << '"';
+        return *this;
+    }
+
+    XmlWriter& XmlWriter::writeAttribute( std::string const& name, const char* attribute ) {
+        if( !name.empty() && attribute && attribute[0] != '\0' )
+            m_os << ' ' << name << "=\"" << XmlEncode( attribute, XmlEncode::ForAttributes ) << '"';
+        return *this;
+    }
+
+    XmlWriter& XmlWriter::writeAttribute( std::string const& name, bool attribute ) {
+        m_os << ' ' << name << "=\"" << ( attribute ? "true" : "false" ) << '"';
+        return *this;
+    }
+
+    XmlWriter& XmlWriter::writeText( std::string const& text, bool indent ) {
+        if( !text.empty() ){
+            bool tagWasOpen = m_tagIsOpen;
+            ensureTagClosed();
+            if( tagWasOpen && indent )
+                m_os << m_indent;
+            m_os << XmlEncode( text );
+            m_needsNewline = true;
+        }
+        return *this;
+    }
+
+    //XmlWriter& XmlWriter::writeComment( std::string const& text ) {
+    //    ensureTagClosed();
+    //    m_os << m_indent << "<!--" << text << "-->";
+    //    m_needsNewline = true;
+    //    return *this;
+    //}
+
+    //void XmlWriter::writeStylesheetRef( std::string const& url ) {
+    //    m_os << "<?xml-stylesheet type=\"text/xsl\" href=\"" << url << "\"?>\n";
+    //}
+
+    //XmlWriter& XmlWriter::writeBlankLine() {
+    //    ensureTagClosed();
+    //    m_os << '\n';
+    //    return *this;
+    //}
+
+    void XmlWriter::ensureTagClosed() {
+        if( m_tagIsOpen ) {
+            m_os << ">" << std::endl;
+            m_tagIsOpen = false;
+        }
+    }
+
+    void XmlWriter::writeDeclaration() {
+        m_os << "<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n";
+    }
+
+    void XmlWriter::newlineIfNecessary() {
+        if( m_needsNewline ) {
+            m_os << std::endl;
+            m_needsNewline = false;
+        }
+    }
+
+// =================================================================================================
+// End of copy-pasted code from Catch
+// =================================================================================================
+
+    // clang-format on
+
+    struct XmlReporter : public IReporter
+    {
+        XmlWriter  xml;
+        std::mutex mutex;
+
+        // caching pointers/references to objects of these types - safe to do
+        const ContextOptions& opt;
+        const TestCaseData*   tc = nullptr;
+
+        XmlReporter(const ContextOptions& co)
+                : xml(*co.cout)
+                , opt(co) {}
+
+        void log_contexts() {
+            int num_contexts = get_num_active_contexts();
+            if(num_contexts) {
+                auto              contexts = get_active_contexts();
+                std::stringstream ss;
+                for(int i = 0; i < num_contexts; ++i) {
+                    contexts[i]->stringify(&ss);
+                    xml.scopedElement("Info").writeText(ss.str());
+                    ss.str("");
+                }
+            }
+        }
+
+        unsigned line(unsigned l) const { return opt.no_line_numbers ? 0 : l; }
+
+        void test_case_start_impl(const TestCaseData& in) {
+            bool open_ts_tag = false;
+            if(tc != nullptr) { // we have already opened a test suite
+                if(std::strcmp(tc->m_test_suite, in.m_test_suite) != 0) {
+                    xml.endElement();
+                    open_ts_tag = true;
+                }
+            }
+            else {
+                open_ts_tag = true; // first test case ==> first test suite
+            }
+
+            if(open_ts_tag) {
+                xml.startElement("TestSuite");
+                xml.writeAttribute("name", in.m_test_suite);
+            }
+
+            tc = &in;
+            xml.startElement("TestCase")
+                    .writeAttribute("name", in.m_name)
+                    .writeAttribute("filename", skipPathFromFilename(in.m_file.c_str()))
+                    .writeAttribute("line", line(in.m_line))
+                    .writeAttribute("description", in.m_description);
+
+            if(Approx(in.m_timeout) != 0)
+                xml.writeAttribute("timeout", in.m_timeout);
+            if(in.m_may_fail)
+                xml.writeAttribute("may_fail", true);
+            if(in.m_should_fail)
+                xml.writeAttribute("should_fail", true);
+        }
+
+        // =========================================================================================
+        // WHAT FOLLOWS ARE OVERRIDES OF THE VIRTUAL METHODS OF THE REPORTER INTERFACE
+        // =========================================================================================
+
+        void report_query(const QueryData& in) override {
+            test_run_start();
+            if(opt.list_reporters) {
+                for(auto& curr : getListeners())
+                    xml.scopedElement("Listener")
+                            .writeAttribute("priority", curr.first.first)
+                            .writeAttribute("name", curr.first.second);
+                for(auto& curr : getReporters())
+                    xml.scopedElement("Reporter")
+                            .writeAttribute("priority", curr.first.first)
+                            .writeAttribute("name", curr.first.second);
+            } else if(opt.count || opt.list_test_cases) {
+                for(unsigned i = 0; i < in.num_data; ++i) {
+                    xml.scopedElement("TestCase").writeAttribute("name", in.data[i]->m_name)
+                        .writeAttribute("testsuite", in.data[i]->m_test_suite)
+                        .writeAttribute("filename", skipPathFromFilename(in.data[i]->m_file.c_str()))
+                        .writeAttribute("line", line(in.data[i]->m_line));
+                }
+                xml.scopedElement("OverallResultsTestCases")
+                        .writeAttribute("unskipped", in.run_stats->numTestCasesPassingFilters);
+            } else if(opt.list_test_suites) {
+                for(unsigned i = 0; i < in.num_data; ++i)
+                    xml.scopedElement("TestSuite").writeAttribute("name", in.data[i]->m_test_suite);
+                xml.scopedElement("OverallResultsTestCases")
+                        .writeAttribute("unskipped", in.run_stats->numTestCasesPassingFilters);
+                xml.scopedElement("OverallResultsTestSuites")
+                        .writeAttribute("unskipped", in.run_stats->numTestSuitesPassingFilters);
+            }
+            xml.endElement();
+        }
+
+        void test_run_start() override {
+            // remove .exe extension - mainly to have the same output on UNIX and Windows
+            std::string binary_name = skipPathFromFilename(opt.binary_name.c_str());
+#ifdef DOCTEST_PLATFORM_WINDOWS
+            if(binary_name.rfind(".exe") != std::string::npos)
+                binary_name = binary_name.substr(0, binary_name.length() - 4);
+#endif // DOCTEST_PLATFORM_WINDOWS
+
+            xml.startElement("doctest").writeAttribute("binary", binary_name);
+            if(opt.no_version == false)
+                xml.writeAttribute("version", DOCTEST_VERSION_STR);
+
+            // only the consequential ones (TODO: filters)
+            xml.scopedElement("Options")
+                    .writeAttribute("order_by", opt.order_by.c_str())
+                    .writeAttribute("rand_seed", opt.rand_seed)
+                    .writeAttribute("first", opt.first)
+                    .writeAttribute("last", opt.last)
+                    .writeAttribute("abort_after", opt.abort_after)
+                    .writeAttribute("subcase_filter_levels", opt.subcase_filter_levels)
+                    .writeAttribute("case_sensitive", opt.case_sensitive)
+                    .writeAttribute("no_throw", opt.no_throw)
+                    .writeAttribute("no_skip", opt.no_skip);
+        }
+
+        void test_run_end(const TestRunStats& p) override {
+            if(tc) // the TestSuite tag - only if there has been at least 1 test case
+                xml.endElement();
+
+            xml.scopedElement("OverallResultsAsserts")
+                    .writeAttribute("successes", p.numAsserts - p.numAssertsFailed)
+                    .writeAttribute("failures", p.numAssertsFailed);
+
+            xml.startElement("OverallResultsTestCases")
+                    .writeAttribute("successes",
+                                    p.numTestCasesPassingFilters - p.numTestCasesFailed)
+                    .writeAttribute("failures", p.numTestCasesFailed);
+            if(opt.no_skipped_summary == false)
+                xml.writeAttribute("skipped", p.numTestCases - p.numTestCasesPassingFilters);
+            xml.endElement();
+
+            xml.endElement();
+        }
+
+        void test_case_start(const TestCaseData& in) override {
+            test_case_start_impl(in);
+            xml.ensureTagClosed();
+        }
+        
+        void test_case_reenter(const TestCaseData&) override {}
+
+        void test_case_end(const CurrentTestCaseStats& st) override {
+            xml.startElement("OverallResultsAsserts")
+                    .writeAttribute("successes",
+                                    st.numAssertsCurrentTest - st.numAssertsFailedCurrentTest)
+                    .writeAttribute("failures", st.numAssertsFailedCurrentTest);
+            if(opt.duration)
+                xml.writeAttribute("duration", st.seconds);
+            if(tc->m_expected_failures)
+                xml.writeAttribute("expected_failures", tc->m_expected_failures);
+            xml.endElement();
+
+            xml.endElement();
+        }
+
+        void test_case_exception(const TestCaseException& e) override {
+            std::lock_guard<std::mutex> lock(mutex);
+
+            xml.scopedElement("Exception")
+                    .writeAttribute("crash", e.is_crash)
+                    .writeText(e.error_string.c_str());
+        }
+
+        void subcase_start(const SubcaseSignature& in) override {
+            std::lock_guard<std::mutex> lock(mutex);
+
+            xml.startElement("SubCase")
+                    .writeAttribute("name", in.m_name)
+                    .writeAttribute("filename", skipPathFromFilename(in.m_file))
+                    .writeAttribute("line", line(in.m_line));
+            xml.ensureTagClosed();
+        }
+
+        void subcase_end() override { xml.endElement(); }
+
+        void log_assert(const AssertData& rb) override {
+            if(!rb.m_failed && !opt.success)
+                return;
+
+            std::lock_guard<std::mutex> lock(mutex);
+
+            xml.startElement("Expression")
+                    .writeAttribute("success", !rb.m_failed)
+                    .writeAttribute("type", assertString(rb.m_at))
+                    .writeAttribute("filename", skipPathFromFilename(rb.m_file))
+                    .writeAttribute("line", line(rb.m_line));
+
+            xml.scopedElement("Original").writeText(rb.m_expr);
+
+            if(rb.m_threw)
+                xml.scopedElement("Exception").writeText(rb.m_exception.c_str());
+
+            if(rb.m_at & assertType::is_throws_as)
+                xml.scopedElement("ExpectedException").writeText(rb.m_exception_type);
+            if(rb.m_at & assertType::is_throws_with)
+                xml.scopedElement("ExpectedExceptionString").writeText(rb.m_exception_string);
+            if((rb.m_at & assertType::is_normal) && !rb.m_threw)
+                xml.scopedElement("Expanded").writeText(rb.m_decomp.c_str());
+
+            log_contexts();
+
+            xml.endElement();
+        }
+
+        void log_message(const MessageData& mb) override {
+            std::lock_guard<std::mutex> lock(mutex);
+
+            xml.startElement("Message")
+                    .writeAttribute("type", failureString(mb.m_severity))
+                    .writeAttribute("filename", skipPathFromFilename(mb.m_file))
+                    .writeAttribute("line", line(mb.m_line));
+
+            xml.scopedElement("Text").writeText(mb.m_string.c_str());
+
+            log_contexts();
+
+            xml.endElement();
+        }
+
+        void test_case_skipped(const TestCaseData& in) override {
+            if(opt.no_skipped_summary == false) {
+                test_case_start_impl(in);
+                xml.writeAttribute("skipped", "true");
+                xml.endElement();
+            }
+        }
+    };
+
+    DOCTEST_REGISTER_REPORTER("xml", 0, XmlReporter);
+
+    void fulltext_log_assert_to_stream(std::ostream& s, const AssertData& rb) {
+        if((rb.m_at & (assertType::is_throws_as | assertType::is_throws_with)) ==
+            0) //!OCLINT bitwise operator in conditional
+            s << Color::Cyan << assertString(rb.m_at) << "( " << rb.m_expr << " ) "
+                << Color::None;
+
+        if(rb.m_at & assertType::is_throws) { //!OCLINT bitwise operator in conditional
+            s << (rb.m_threw ? "threw as expected!" : "did NOT throw at all!") << "\n";
+        } else if((rb.m_at & assertType::is_throws_as) &&
+                    (rb.m_at & assertType::is_throws_with)) { //!OCLINT
+            s << Color::Cyan << assertString(rb.m_at) << "( " << rb.m_expr << ", \""
+                << rb.m_exception_string << "\", " << rb.m_exception_type << " ) " << Color::None;
+            if(rb.m_threw) {
+                if(!rb.m_failed) {
+                    s << "threw as expected!\n";
+                } else {
+                    s << "threw a DIFFERENT exception! (contents: " << rb.m_exception << ")\n";
+                }
+            } else {
+                s << "did NOT throw at all!\n";
+            }
+        } else if(rb.m_at &
+                    assertType::is_throws_as) { //!OCLINT bitwise operator in conditional
+            s << Color::Cyan << assertString(rb.m_at) << "( " << rb.m_expr << ", "
+                << rb.m_exception_type << " ) " << Color::None
+                << (rb.m_threw ? (rb.m_threw_as ? "threw as expected!" :
+                                                "threw a DIFFERENT exception: ") :
+                                "did NOT throw at all!")
+                << Color::Cyan << rb.m_exception << "\n";
+        } else if(rb.m_at &
+                    assertType::is_throws_with) { //!OCLINT bitwise operator in conditional
+            s << Color::Cyan << assertString(rb.m_at) << "( " << rb.m_expr << ", \""
+                << rb.m_exception_string << "\" ) " << Color::None
+                << (rb.m_threw ? (!rb.m_failed ? "threw as expected!" :
+                                                "threw a DIFFERENT exception: ") :
+                                "did NOT throw at all!")
+                << Color::Cyan << rb.m_exception << "\n";
+        } else if(rb.m_at & assertType::is_nothrow) { //!OCLINT bitwise operator in conditional
+            s << (rb.m_threw ? "THREW exception: " : "didn't throw!") << Color::Cyan
+                << rb.m_exception << "\n";
+        } else {
+            s << (rb.m_threw ? "THREW exception: " :
+                                (!rb.m_failed ? "is correct!\n" : "is NOT correct!\n"));
+            if(rb.m_threw)
+                s << rb.m_exception << "\n";
+            else
+                s << "  values: " << assertString(rb.m_at) << "( " << rb.m_decomp << " )\n";
+        }
+    }
+
+    // TODO:
+    // - log_contexts()
+    // - log_message()
+    // - respond to queries
+    // - honor remaining options
+    // - more attributes in tags
+    struct JUnitReporter : public IReporter
+    {
+        XmlWriter  xml;
+        std::mutex mutex;
+        Timer timer;
+        std::vector<String> deepestSubcaseStackNames;
+
+        struct JUnitTestCaseData
+        {
+DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Wdeprecated-declarations") // gmtime
+            static std::string getCurrentTimestamp() {
+                // Beware, this is not reentrant because of backward compatibility issues
+                // Also, UTC only, again because of backward compatibility (%z is C++11)
+                time_t rawtime;
+                std::time(&rawtime);
+                auto const timeStampSize = sizeof("2017-01-16T17:06:45Z");
+
+                std::tm* timeInfo;
+                timeInfo = std::gmtime(&rawtime);
+
+                char timeStamp[timeStampSize];
+                const char* const fmt = "%Y-%m-%dT%H:%M:%SZ";
+
+                std::strftime(timeStamp, timeStampSize, fmt, timeInfo);
+                return std::string(timeStamp);
+            }
+DOCTEST_CLANG_SUPPRESS_WARNING_POP
+
+            struct JUnitTestMessage
+            {
+                JUnitTestMessage(const std::string& _message, const std::string& _type, const std::string& _details)
+                    : message(_message), type(_type), details(_details) {}
+
+                JUnitTestMessage(const std::string& _message, const std::string& _details)
+                    : message(_message), type(), details(_details) {}
+
+                std::string message, type, details;
+            };
+
+            struct JUnitTestCase
+            {
+                JUnitTestCase(const std::string& _classname, const std::string& _name)
+                    : classname(_classname), name(_name), time(0), failures() {}
+
+                std::string classname, name;
+                double time;
+                std::vector<JUnitTestMessage> failures, errors;
+            };
+
+            void add(const std::string& classname, const std::string& name) {
+                testcases.emplace_back(classname, name);
+            }
+
+            void appendSubcaseNamesToLastTestcase(std::vector<String> nameStack) {
+                for(auto& curr: nameStack)
+                    if(curr.size())
+                        testcases.back().name += std::string("/") + curr.c_str();
+            }
+
+            void addTime(double time) {
+                if(time < 1e-4)
+                    time = 0;
+                testcases.back().time = time;
+                totalSeconds += time;
+            }
+
+            void addFailure(const std::string& message, const std::string& type, const std::string& details) {
+                testcases.back().failures.emplace_back(message, type, details);
+                ++totalFailures;
+            }
+
+            void addError(const std::string& message, const std::string& details) {
+                testcases.back().errors.emplace_back(message, details);
+                ++totalErrors;
+            }
+
+            std::vector<JUnitTestCase> testcases;
+            double totalSeconds = 0;
+            int totalErrors = 0, totalFailures = 0;
+        };
+
+        JUnitTestCaseData testCaseData;
+
+        // caching pointers/references to objects of these types - safe to do
+        const ContextOptions& opt;
+        const TestCaseData*   tc = nullptr;
+
+        JUnitReporter(const ContextOptions& co)
+                : xml(*co.cout)
+                , opt(co) {}
+
+        unsigned line(unsigned l) const { return opt.no_line_numbers ? 0 : l; }
+
+        // =========================================================================================
+        // WHAT FOLLOWS ARE OVERRIDES OF THE VIRTUAL METHODS OF THE REPORTER INTERFACE
+        // =========================================================================================
+
+        void report_query(const QueryData&) override {}
+
+        void test_run_start() override {}
+
+        void test_run_end(const TestRunStats& p) override {
+            // remove .exe extension - mainly to have the same output on UNIX and Windows
+            std::string binary_name = skipPathFromFilename(opt.binary_name.c_str());
+#ifdef DOCTEST_PLATFORM_WINDOWS
+            if(binary_name.rfind(".exe") != std::string::npos)
+                binary_name = binary_name.substr(0, binary_name.length() - 4);
+#endif // DOCTEST_PLATFORM_WINDOWS
+            xml.startElement("testsuites");
+            xml.startElement("testsuite").writeAttribute("name", binary_name)
+                    .writeAttribute("errors", testCaseData.totalErrors)
+                    .writeAttribute("failures", testCaseData.totalFailures)
+                    .writeAttribute("tests", p.numAsserts);
+            if(opt.no_time_in_output == false) {
+                xml.writeAttribute("time", testCaseData.totalSeconds);
+                xml.writeAttribute("timestamp", JUnitTestCaseData::getCurrentTimestamp());
+            }
+            if(opt.no_version == false)
+                xml.writeAttribute("doctest_version", DOCTEST_VERSION_STR);
+
+            for(const auto& testCase : testCaseData.testcases) {
+                xml.startElement("testcase")
+                    .writeAttribute("classname", testCase.classname)
+                    .writeAttribute("name", testCase.name);
+                if(opt.no_time_in_output == false)
+                    xml.writeAttribute("time", testCase.time);
+                // This is not ideal, but it should be enough to mimic gtest's junit output.
+                xml.writeAttribute("status", "run");
+
+                for(const auto& failure : testCase.failures) {
+                    xml.scopedElement("failure")
+                        .writeAttribute("message", failure.message)
+                        .writeAttribute("type", failure.type)
+                        .writeText(failure.details, false);
+                }
+
+                for(const auto& error : testCase.errors) {
+                    xml.scopedElement("error")
+                        .writeAttribute("message", error.message)
+                        .writeText(error.details);
+                }
+
+                xml.endElement();
+            }
+            xml.endElement();
+            xml.endElement();
+        }
+
+        void test_case_start(const TestCaseData& in) override {
+            testCaseData.add(skipPathFromFilename(in.m_file.c_str()), in.m_name);
+            timer.start();
+        }
+
+        void test_case_reenter(const TestCaseData& in) override {
+            testCaseData.addTime(timer.getElapsedSeconds());
+            testCaseData.appendSubcaseNamesToLastTestcase(deepestSubcaseStackNames);
+            deepestSubcaseStackNames.clear();
+
+            timer.start();
+            testCaseData.add(skipPathFromFilename(in.m_file.c_str()), in.m_name);
+        }
+
+        void test_case_end(const CurrentTestCaseStats&) override {
+            testCaseData.addTime(timer.getElapsedSeconds());
+            testCaseData.appendSubcaseNamesToLastTestcase(deepestSubcaseStackNames);
+            deepestSubcaseStackNames.clear();
+        }
+
+        void test_case_exception(const TestCaseException& e) override {
+            std::lock_guard<std::mutex> lock(mutex);
+            testCaseData.addError("exception", e.error_string.c_str());
+        }
+
+        void subcase_start(const SubcaseSignature& in) override {
+            std::lock_guard<std::mutex> lock(mutex);
+            deepestSubcaseStackNames.push_back(in.m_name);
+        }
+
+        void subcase_end() override {}
+
+        void log_assert(const AssertData& rb) override {
+            if(!rb.m_failed) // report only failures & ignore the `success` option
+                return;
+
+            std::lock_guard<std::mutex> lock(mutex);
+
+            std::ostringstream os;
+            os << skipPathFromFilename(rb.m_file) << (opt.gnu_file_line ? ":" : "(")
+              << line(rb.m_line) << (opt.gnu_file_line ? ":" : "):") << std::endl;
+
+            fulltext_log_assert_to_stream(os, rb);
+            testCaseData.addFailure(rb.m_decomp.c_str(), assertString(rb.m_at), os.str());
+        }
+
+        void log_message(const MessageData&) override {}
+
+        void test_case_skipped(const TestCaseData&) override {}
+    };
+
+    DOCTEST_REGISTER_REPORTER("junit", 0, JUnitReporter);
+
+    struct Whitespace
+    {
+        int nrSpaces;
+        explicit Whitespace(int nr)
+                : nrSpaces(nr) {}
+    };
+
+    std::ostream& operator<<(std::ostream& out, const Whitespace& ws) {
+        if(ws.nrSpaces != 0)
+            out << std::setw(ws.nrSpaces) << ' ';
+        return out;
+    }
+
+    struct ConsoleReporter : public IReporter
+    {
+        std::ostream&                 s;
+        bool                          hasLoggedCurrentTestStart;
+        std::vector<SubcaseSignature> subcasesStack;
+        size_t                        currentSubcaseLevel;
+        std::mutex                    mutex;
+
+        // caching pointers/references to objects of these types - safe to do
+        const ContextOptions& opt;
+        const TestCaseData*   tc;
+
+        ConsoleReporter(const ContextOptions& co)
+                : s(*co.cout)
+                , opt(co) {}
+
+        ConsoleReporter(const ContextOptions& co, std::ostream& ostr)
+                : s(ostr)
+                , opt(co) {}
+
+        // =========================================================================================
+        // WHAT FOLLOWS ARE HELPERS USED BY THE OVERRIDES OF THE VIRTUAL METHODS OF THE INTERFACE
+        // =========================================================================================
+
+        void separator_to_stream() {
+            s << Color::Yellow
+              << "==============================================================================="
+                 "\n";
+        }
+
+        const char* getSuccessOrFailString(bool success, assertType::Enum at,
+                                           const char* success_str) {
+            if(success)
+                return success_str;
+            return failureString(at);
+        }
+
+        Color::Enum getSuccessOrFailColor(bool success, assertType::Enum at) {
+            return success ? Color::BrightGreen :
+                             (at & assertType::is_warn) ? Color::Yellow : Color::Red;
+        }
+
+        void successOrFailColoredStringToStream(bool success, assertType::Enum at,
+                                                const char* success_str = "SUCCESS") {
+            s << getSuccessOrFailColor(success, at)
+              << getSuccessOrFailString(success, at, success_str) << ": ";
+        }
+
+        void log_contexts() {
+            int num_contexts = get_num_active_contexts();
+            if(num_contexts) {
+                auto contexts = get_active_contexts();
+
+                s << Color::None << "  logged: ";
+                for(int i = 0; i < num_contexts; ++i) {
+                    s << (i == 0 ? "" : "          ");
+                    contexts[i]->stringify(&s);
+                    s << "\n";
+                }
+            }
+
+            s << "\n";
+        }
+
+        // this was requested to be made virtual so users could override it
+        virtual void file_line_to_stream(const char* file, int line,
+                                        const char* tail = "") {
+            s << Color::LightGrey << skipPathFromFilename(file) << (opt.gnu_file_line ? ":" : "(")
+            << (opt.no_line_numbers ? 0 : line) // 0 or the real num depending on the option
+            << (opt.gnu_file_line ? ":" : "):") << tail;
+        }
+
+        void logTestStart() {
+            if(hasLoggedCurrentTestStart)
+                return;
+
+            separator_to_stream();
+            file_line_to_stream(tc->m_file.c_str(), tc->m_line, "\n");
+            if(tc->m_description)
+                s << Color::Yellow << "DESCRIPTION: " << Color::None << tc->m_description << "\n";
+            if(tc->m_test_suite && tc->m_test_suite[0] != '\0')
+                s << Color::Yellow << "TEST SUITE: " << Color::None << tc->m_test_suite << "\n";
+            if(strncmp(tc->m_name, "  Scenario:", 11) != 0)
+                s << Color::Yellow << "TEST CASE:  ";
+            s << Color::None << tc->m_name << "\n";
+
+            for(size_t i = 0; i < currentSubcaseLevel; ++i) {
+                if(subcasesStack[i].m_name[0] != '\0')
+                    s << "  " << subcasesStack[i].m_name << "\n";
+            }
+
+            if(currentSubcaseLevel != subcasesStack.size()) {
+                s << Color::Yellow << "\nDEEPEST SUBCASE STACK REACHED (DIFFERENT FROM THE CURRENT ONE):\n" << Color::None;
+                for(size_t i = 0; i < subcasesStack.size(); ++i) {
+                    if(subcasesStack[i].m_name[0] != '\0')
+                        s << "  " << subcasesStack[i].m_name << "\n";
+                }
+            }
+
+            s << "\n";
+
+            hasLoggedCurrentTestStart = true;
+        }
+
+        void printVersion() {
+            if(opt.no_version == false)
+                s << Color::Cyan << "[doctest] " << Color::None << "doctest version is \""
+                  << DOCTEST_VERSION_STR << "\"\n";
+        }
+
+        void printIntro() {
+            printVersion();
+            s << Color::Cyan << "[doctest] " << Color::None
+              << "run with \"--" DOCTEST_OPTIONS_PREFIX_DISPLAY "help\" for options\n";
+        }
+
+        void printHelp() {
+            int sizePrefixDisplay = static_cast<int>(strlen(DOCTEST_OPTIONS_PREFIX_DISPLAY));
+            printVersion();
+            // clang-format off
+            s << Color::Cyan << "[doctest]\n" << Color::None;
+            s << Color::Cyan << "[doctest] " << Color::None;
+            s << "boolean values: \"1/on/yes/true\" or \"0/off/no/false\"\n";
+            s << Color::Cyan << "[doctest] " << Color::None;
+            s << "filter  values: \"str1,str2,str3\" (comma separated strings)\n";
+            s << Color::Cyan << "[doctest]\n" << Color::None;
+            s << Color::Cyan << "[doctest] " << Color::None;
+            s << "filters use wildcards for matching strings\n";
+            s << Color::Cyan << "[doctest] " << Color::None;
+            s << "something passes a filter if any of the strings in a filter matches\n";
+#ifndef DOCTEST_CONFIG_NO_UNPREFIXED_OPTIONS
+            s << Color::Cyan << "[doctest]\n" << Color::None;
+            s << Color::Cyan << "[doctest] " << Color::None;
+            s << "ALL FLAGS, OPTIONS AND FILTERS ALSO AVAILABLE WITH A \"" DOCTEST_CONFIG_OPTIONS_PREFIX "\" PREFIX!!!\n";
+#endif
+            s << Color::Cyan << "[doctest]\n" << Color::None;
+            s << Color::Cyan << "[doctest] " << Color::None;
+            s << "Query flags - the program quits after them. Available:\n\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "?,   --" DOCTEST_OPTIONS_PREFIX_DISPLAY "help, -" DOCTEST_OPTIONS_PREFIX_DISPLAY "h                      "
+              << Whitespace(sizePrefixDisplay*0) <<  "prints this message\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "v,   --" DOCTEST_OPTIONS_PREFIX_DISPLAY "version                       "
+              << Whitespace(sizePrefixDisplay*1) << "prints the version\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "c,   --" DOCTEST_OPTIONS_PREFIX_DISPLAY "count                         "
+              << Whitespace(sizePrefixDisplay*1) << "prints the number of matching tests\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "ltc, --" DOCTEST_OPTIONS_PREFIX_DISPLAY "list-test-cases               "
+              << Whitespace(sizePrefixDisplay*1) << "lists all matching tests by name\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "lts, --" DOCTEST_OPTIONS_PREFIX_DISPLAY "list-test-suites              "
+              << Whitespace(sizePrefixDisplay*1) << "lists all matching test suites\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "lr,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "list-reporters                "
+              << Whitespace(sizePrefixDisplay*1) << "lists all registered reporters\n\n";
+            // ================================================================================== << 79
+            s << Color::Cyan << "[doctest] " << Color::None;
+            s << "The available <int>/<string> options/filters are:\n\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "tc,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "test-case=<filters>           "
+              << Whitespace(sizePrefixDisplay*1) << "filters     tests by their name\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "tce, --" DOCTEST_OPTIONS_PREFIX_DISPLAY "test-case-exclude=<filters>   "
+              << Whitespace(sizePrefixDisplay*1) << "filters OUT tests by their name\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "sf,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "source-file=<filters>         "
+              << Whitespace(sizePrefixDisplay*1) << "filters     tests by their file\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "sfe, --" DOCTEST_OPTIONS_PREFIX_DISPLAY "source-file-exclude=<filters> "
+              << Whitespace(sizePrefixDisplay*1) << "filters OUT tests by their file\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "ts,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "test-suite=<filters>          "
+              << Whitespace(sizePrefixDisplay*1) << "filters     tests by their test suite\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "tse, --" DOCTEST_OPTIONS_PREFIX_DISPLAY "test-suite-exclude=<filters>  "
+              << Whitespace(sizePrefixDisplay*1) << "filters OUT tests by their test suite\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "sc,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "subcase=<filters>             "
+              << Whitespace(sizePrefixDisplay*1) << "filters     subcases by their name\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "sce, --" DOCTEST_OPTIONS_PREFIX_DISPLAY "subcase-exclude=<filters>     "
+              << Whitespace(sizePrefixDisplay*1) << "filters OUT subcases by their name\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "r,   --" DOCTEST_OPTIONS_PREFIX_DISPLAY "reporters=<filters>           "
+              << Whitespace(sizePrefixDisplay*1) << "reporters to use (console is default)\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "o,   --" DOCTEST_OPTIONS_PREFIX_DISPLAY "out=<string>                  "
+              << Whitespace(sizePrefixDisplay*1) << "output filename\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "ob,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "order-by=<string>             "
+              << Whitespace(sizePrefixDisplay*1) << "how the tests should be ordered\n";
+            s << Whitespace(sizePrefixDisplay*3) << "                                       <string> - by [file/suite/name/rand]\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "rs,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "rand-seed=<int>               "
+              << Whitespace(sizePrefixDisplay*1) << "seed for random ordering\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "f,   --" DOCTEST_OPTIONS_PREFIX_DISPLAY "first=<int>                   "
+              << Whitespace(sizePrefixDisplay*1) << "the first test passing the filters to\n";
+            s << Whitespace(sizePrefixDisplay*3) << "                                       execute - for range-based execution\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "l,   --" DOCTEST_OPTIONS_PREFIX_DISPLAY "last=<int>                    "
+              << Whitespace(sizePrefixDisplay*1) << "the last test passing the filters to\n";
+            s << Whitespace(sizePrefixDisplay*3) << "                                       execute - for range-based execution\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "aa,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "abort-after=<int>             "
+              << Whitespace(sizePrefixDisplay*1) << "stop after <int> failed assertions\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "scfl,--" DOCTEST_OPTIONS_PREFIX_DISPLAY "subcase-filter-levels=<int>   "
+              << Whitespace(sizePrefixDisplay*1) << "apply filters for the first <int> levels\n";
+            s << Color::Cyan << "\n[doctest] " << Color::None;
+            s << "Bool options - can be used like flags and true is assumed. Available:\n\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "s,   --" DOCTEST_OPTIONS_PREFIX_DISPLAY "success=<bool>                "
+              << Whitespace(sizePrefixDisplay*1) << "include successful assertions in output\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "cs,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "case-sensitive=<bool>         "
+              << Whitespace(sizePrefixDisplay*1) << "filters being treated as case sensitive\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "e,   --" DOCTEST_OPTIONS_PREFIX_DISPLAY "exit=<bool>                   "
+              << Whitespace(sizePrefixDisplay*1) << "exits after the tests finish\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "d,   --" DOCTEST_OPTIONS_PREFIX_DISPLAY "duration=<bool>               "
+              << Whitespace(sizePrefixDisplay*1) << "prints the time duration of each test\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "nt,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "no-throw=<bool>               "
+              << Whitespace(sizePrefixDisplay*1) << "skips exceptions-related assert checks\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "ne,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "no-exitcode=<bool>            "
+              << Whitespace(sizePrefixDisplay*1) << "returns (or exits) always with success\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "nr,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "no-run=<bool>                 "
+              << Whitespace(sizePrefixDisplay*1) << "skips all runtime doctest operations\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "nv,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "no-version=<bool>             "
+              << Whitespace(sizePrefixDisplay*1) << "omit the framework version in the output\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "nc,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "no-colors=<bool>              "
+              << Whitespace(sizePrefixDisplay*1) << "disables colors in output\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "fc,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "force-colors=<bool>           "
+              << Whitespace(sizePrefixDisplay*1) << "use colors even when not in a tty\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "nb,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "no-breaks=<bool>              "
+              << Whitespace(sizePrefixDisplay*1) << "disables breakpoints in debuggers\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "ns,  --" DOCTEST_OPTIONS_PREFIX_DISPLAY "no-skip=<bool>                "
+              << Whitespace(sizePrefixDisplay*1) << "don't skip test cases marked as skip\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "gfl, --" DOCTEST_OPTIONS_PREFIX_DISPLAY "gnu-file-line=<bool>          "
+              << Whitespace(sizePrefixDisplay*1) << ":n: vs (n): for line numbers in output\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "npf, --" DOCTEST_OPTIONS_PREFIX_DISPLAY "no-path-filenames=<bool>      "
+              << Whitespace(sizePrefixDisplay*1) << "only filenames and no paths in output\n";
+            s << " -" DOCTEST_OPTIONS_PREFIX_DISPLAY "nln, --" DOCTEST_OPTIONS_PREFIX_DISPLAY "no-line-numbers=<bool>        "
+              << Whitespace(sizePrefixDisplay*1) << "0 instead of real line numbers in output\n";
+            // ================================================================================== << 79
+            // clang-format on
+
+            s << Color::Cyan << "\n[doctest] " << Color::None;
+            s << "for more information visit the project documentation\n\n";
+        }
+
+        void printRegisteredReporters() {
+            printVersion();
+            auto printReporters = [this] (const reporterMap& reporters, const char* type) {
+                if(reporters.size()) {
+                    s << Color::Cyan << "[doctest] " << Color::None << "listing all registered " << type << "\n";
+                    for(auto& curr : reporters)
+                        s << "priority: " << std::setw(5) << curr.first.first
+                          << " name: " << curr.first.second << "\n";
+                }
+            };
+            printReporters(getListeners(), "listeners");
+            printReporters(getReporters(), "reporters");
+        }
+
+        void list_query_results() {
+            separator_to_stream();
+            if(opt.count || opt.list_test_cases) {
+                s << Color::Cyan << "[doctest] " << Color::None
+                  << "unskipped test cases passing the current filters: "
+                  << g_cs->numTestCasesPassingFilters << "\n";
+            } else if(opt.list_test_suites) {
+                s << Color::Cyan << "[doctest] " << Color::None
+                  << "unskipped test cases passing the current filters: "
+                  << g_cs->numTestCasesPassingFilters << "\n";
+                s << Color::Cyan << "[doctest] " << Color::None
+                  << "test suites with unskipped test cases passing the current filters: "
+                  << g_cs->numTestSuitesPassingFilters << "\n";
+            }
+        }
+
+        // =========================================================================================
+        // WHAT FOLLOWS ARE OVERRIDES OF THE VIRTUAL METHODS OF THE REPORTER INTERFACE
+        // =========================================================================================
+
+        void report_query(const QueryData& in) override {
+            if(opt.version) {
+                printVersion();
+            } else if(opt.help) {
+                printHelp();
+            } else if(opt.list_reporters) {
+                printRegisteredReporters();
+            } else if(opt.count || opt.list_test_cases) {
+                if(opt.list_test_cases) {
+                    s << Color::Cyan << "[doctest] " << Color::None
+                      << "listing all test case names\n";
+                    separator_to_stream();
+                }
+
+                for(unsigned i = 0; i < in.num_data; ++i)
+                    s << Color::None << in.data[i]->m_name << "\n";
+
+                separator_to_stream();
+
+                s << Color::Cyan << "[doctest] " << Color::None
+                  << "unskipped test cases passing the current filters: "
+                  << g_cs->numTestCasesPassingFilters << "\n";
+
+            } else if(opt.list_test_suites) {
+                s << Color::Cyan << "[doctest] " << Color::None << "listing all test suites\n";
+                separator_to_stream();
+
+                for(unsigned i = 0; i < in.num_data; ++i)
+                    s << Color::None << in.data[i]->m_test_suite << "\n";
+
+                separator_to_stream();
+
+                s << Color::Cyan << "[doctest] " << Color::None
+                  << "unskipped test cases passing the current filters: "
+                  << g_cs->numTestCasesPassingFilters << "\n";
+                s << Color::Cyan << "[doctest] " << Color::None
+                  << "test suites with unskipped test cases passing the current filters: "
+                  << g_cs->numTestSuitesPassingFilters << "\n";
+            }
+        }
+
+        void test_run_start() override { printIntro(); }
+
+        void test_run_end(const TestRunStats& p) override {
+            separator_to_stream();
+            s << std::dec;
+
+            const bool anythingFailed = p.numTestCasesFailed > 0 || p.numAssertsFailed > 0;
+            s << Color::Cyan << "[doctest] " << Color::None << "test cases: " << std::setw(6)
+              << p.numTestCasesPassingFilters << " | "
+              << ((p.numTestCasesPassingFilters == 0 || anythingFailed) ? Color::None :
+                                                                          Color::Green)
+              << std::setw(6) << p.numTestCasesPassingFilters - p.numTestCasesFailed << " passed"
+              << Color::None << " | " << (p.numTestCasesFailed > 0 ? Color::Red : Color::None)
+              << std::setw(6) << p.numTestCasesFailed << " failed" << Color::None << " | ";
+            if(opt.no_skipped_summary == false) {
+                const int numSkipped = p.numTestCases - p.numTestCasesPassingFilters;
+                s << (numSkipped == 0 ? Color::None : Color::Yellow) << std::setw(6) << numSkipped
+                  << " skipped" << Color::None;
+            }
+            s << "\n";
+            s << Color::Cyan << "[doctest] " << Color::None << "assertions: " << std::setw(6)
+              << p.numAsserts << " | "
+              << ((p.numAsserts == 0 || anythingFailed) ? Color::None : Color::Green)
+              << std::setw(6) << (p.numAsserts - p.numAssertsFailed) << " passed" << Color::None
+              << " | " << (p.numAssertsFailed > 0 ? Color::Red : Color::None) << std::setw(6)
+              << p.numAssertsFailed << " failed" << Color::None << " |\n";
+            s << Color::Cyan << "[doctest] " << Color::None
+              << "Status: " << (p.numTestCasesFailed > 0 ? Color::Red : Color::Green)
+              << ((p.numTestCasesFailed > 0) ? "FAILURE!" : "SUCCESS!") << Color::None << std::endl;
+        }
+
+        void test_case_start(const TestCaseData& in) override {
+            hasLoggedCurrentTestStart = false;
+            tc                        = &in;
+            subcasesStack.clear();
+            currentSubcaseLevel = 0;
+        }
+        
+        void test_case_reenter(const TestCaseData&) override {
+            subcasesStack.clear();
+        }
+
+        void test_case_end(const CurrentTestCaseStats& st) override {
+            // log the preamble of the test case only if there is something
+            // else to print - something other than that an assert has failed
+            if(opt.duration ||
+               (st.failure_flags && st.failure_flags != TestCaseFailureReason::AssertFailure))
+                logTestStart();
+
+            if(opt.duration)
+                s << Color::None << std::setprecision(6) << std::fixed << st.seconds
+                  << " s: " << tc->m_name << "\n";
+
+            if(st.failure_flags & TestCaseFailureReason::Timeout)
+                s << Color::Red << "Test case exceeded time limit of " << std::setprecision(6)
+                  << std::fixed << tc->m_timeout << "!\n";
+
+            if(st.failure_flags & TestCaseFailureReason::ShouldHaveFailedButDidnt) {
+                s << Color::Red << "Should have failed but didn't! Marking it as failed!\n";
+            } else if(st.failure_flags & TestCaseFailureReason::ShouldHaveFailedAndDid) {
+                s << Color::Yellow << "Failed as expected so marking it as not failed\n";
+            } else if(st.failure_flags & TestCaseFailureReason::CouldHaveFailedAndDid) {
+                s << Color::Yellow << "Allowed to fail so marking it as not failed\n";
+            } else if(st.failure_flags & TestCaseFailureReason::DidntFailExactlyNumTimes) {
+                s << Color::Red << "Didn't fail exactly " << tc->m_expected_failures
+                  << " times so marking it as failed!\n";
+            } else if(st.failure_flags & TestCaseFailureReason::FailedExactlyNumTimes) {
+                s << Color::Yellow << "Failed exactly " << tc->m_expected_failures
+                  << " times as expected so marking it as not failed!\n";
+            }
+            if(st.failure_flags & TestCaseFailureReason::TooManyFailedAsserts) {
+                s << Color::Red << "Aborting - too many failed asserts!\n";
+            }
+            s << Color::None; // lgtm [cpp/useless-expression]
+        }
+
+        void test_case_exception(const TestCaseException& e) override {
+            logTestStart();
+
+            file_line_to_stream(tc->m_file.c_str(), tc->m_line, " ");
+            successOrFailColoredStringToStream(false, e.is_crash ? assertType::is_require :
+                                                                   assertType::is_check);
+            s << Color::Red << (e.is_crash ? "test case CRASHED: " : "test case THREW exception: ")
+              << Color::Cyan << e.error_string << "\n";
+
+            int num_stringified_contexts = get_num_stringified_contexts();
+            if(num_stringified_contexts) {
+                auto stringified_contexts = get_stringified_contexts();
+                s << Color::None << "  logged: ";
+                for(int i = num_stringified_contexts; i > 0; --i) {
+                    s << (i == num_stringified_contexts ? "" : "          ")
+                      << stringified_contexts[i - 1] << "\n";
+                }
+            }
+            s << "\n" << Color::None;
+        }
+
+        void subcase_start(const SubcaseSignature& subc) override {
+            std::lock_guard<std::mutex> lock(mutex);
+            subcasesStack.push_back(subc);
+            ++currentSubcaseLevel;
+            hasLoggedCurrentTestStart = false;
+        }
+
+        void subcase_end() override {
+            std::lock_guard<std::mutex> lock(mutex);
+            --currentSubcaseLevel;
+            hasLoggedCurrentTestStart = false;
+        }
+
+        void log_assert(const AssertData& rb) override {
+            if(!rb.m_failed && !opt.success)
+                return;
+
+            std::lock_guard<std::mutex> lock(mutex);
+
+            logTestStart();
+
+            file_line_to_stream(rb.m_file, rb.m_line, " ");
+            successOrFailColoredStringToStream(!rb.m_failed, rb.m_at);
+
+            fulltext_log_assert_to_stream(s, rb);
+
+            log_contexts();
+        }
+
+        void log_message(const MessageData& mb) override {
+            std::lock_guard<std::mutex> lock(mutex);
+
+            logTestStart();
+
+            file_line_to_stream(mb.m_file, mb.m_line, " ");
+            s << getSuccessOrFailColor(false, mb.m_severity)
+              << getSuccessOrFailString(mb.m_severity & assertType::is_warn, mb.m_severity,
+                                        "MESSAGE") << ": ";
+            s << Color::None << mb.m_string << "\n";
+            log_contexts();
+        }
+
+        void test_case_skipped(const TestCaseData&) override {}
+    };
+
+    DOCTEST_REGISTER_REPORTER("console", 0, ConsoleReporter);
+
+#ifdef DOCTEST_PLATFORM_WINDOWS
+    struct DebugOutputWindowReporter : public ConsoleReporter
+    {
+        DOCTEST_THREAD_LOCAL static std::ostringstream oss;
+
+        DebugOutputWindowReporter(const ContextOptions& co)
+                : ConsoleReporter(co, oss) {}
+
+#define DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(func, type, arg)                                    \
+    void func(type arg) override {                                                                 \
+        bool with_col = g_no_colors;                                                               \
+        g_no_colors   = false;                                                                     \
+        ConsoleReporter::func(arg);                                                                \
+        DOCTEST_OUTPUT_DEBUG_STRING(oss.str().c_str());                                            \
+        oss.str("");                                                                               \
+        g_no_colors = with_col;                                                                    \
+    }
+
+        DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(test_run_start, DOCTEST_EMPTY, DOCTEST_EMPTY)
+        DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(test_run_end, const TestRunStats&, in)
+        DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(test_case_start, const TestCaseData&, in)
+        DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(test_case_reenter, const TestCaseData&, in)
+        DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(test_case_end, const CurrentTestCaseStats&, in)
+        DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(test_case_exception, const TestCaseException&, in)
+        DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(subcase_start, const SubcaseSignature&, in)
+        DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(subcase_end, DOCTEST_EMPTY, DOCTEST_EMPTY)
+        DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(log_assert, const AssertData&, in)
+        DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(log_message, const MessageData&, in)
+        DOCTEST_DEBUG_OUTPUT_REPORTER_OVERRIDE(test_case_skipped, const TestCaseData&, in)
+    };
+
+    DOCTEST_THREAD_LOCAL std::ostringstream DebugOutputWindowReporter::oss;
+#endif // DOCTEST_PLATFORM_WINDOWS
+
+    // the implementation of parseOption()
+    bool parseOptionImpl(int argc, const char* const* argv, const char* pattern, String* value) {
+        // going from the end to the beginning and stopping on the first occurrence from the end
+        for(int i = argc; i > 0; --i) {
+            auto index = i - 1;
+            auto temp = std::strstr(argv[index], pattern);
+            if(temp && (value || strlen(temp) == strlen(pattern))) { //!OCLINT prefer early exits and continue
+                // eliminate matches in which the chars before the option are not '-'
+                bool noBadCharsFound = true;
+                auto curr            = argv[index];
+                while(curr != temp) {
+                    if(*curr++ != '-') {
+                        noBadCharsFound = false;
+                        break;
+                    }
+                }
+                if(noBadCharsFound && argv[index][0] == '-') {
+                    if(value) {
+                        // parsing the value of an option
+                        temp += strlen(pattern);
+                        const unsigned len = strlen(temp);
+                        if(len) {
+                            *value = temp;
+                            return true;
+                        }
+                    } else {
+                        // just a flag - no value
+                        return true;
+                    }
+                }
+            }
+        }
+        return false;
+    }
+
+    // parses an option and returns the string after the '=' character
+    bool parseOption(int argc, const char* const* argv, const char* pattern, String* value = nullptr,
+                     const String& defaultVal = String()) {
+        if(value)
+            *value = defaultVal;
+#ifndef DOCTEST_CONFIG_NO_UNPREFIXED_OPTIONS
+        // offset (normally 3 for "dt-") to skip prefix
+        if(parseOptionImpl(argc, argv, pattern + strlen(DOCTEST_CONFIG_OPTIONS_PREFIX), value))
+            return true;
+#endif // DOCTEST_CONFIG_NO_UNPREFIXED_OPTIONS
+        return parseOptionImpl(argc, argv, pattern, value);
+    }
+
+    // locates a flag on the command line
+    bool parseFlag(int argc, const char* const* argv, const char* pattern) {
+        return parseOption(argc, argv, pattern);
+    }
+
+    // parses a comma separated list of words after a pattern in one of the arguments in argv
+    bool parseCommaSepArgs(int argc, const char* const* argv, const char* pattern,
+                           std::vector<String>& res) {
+        String filtersString;
+        if(parseOption(argc, argv, pattern, &filtersString)) {
+            // tokenize with "," as a separator
+            // cppcheck-suppress strtokCalled
+            DOCTEST_CLANG_SUPPRESS_WARNING_WITH_PUSH("-Wdeprecated-declarations")
+            auto pch = std::strtok(filtersString.c_str(), ","); // modifies the string
+            while(pch != nullptr) {
+                if(strlen(pch))
+                    res.push_back(pch);
+                // uses the strtok() internal state to go to the next token
+                // cppcheck-suppress strtokCalled
+                pch = std::strtok(nullptr, ",");
+            }
+            DOCTEST_CLANG_SUPPRESS_WARNING_POP
+            return true;
+        }
+        return false;
+    }
+
+    enum optionType
+    {
+        option_bool,
+        option_int
+    };
+
+    // parses an int/bool option from the command line
+    bool parseIntOption(int argc, const char* const* argv, const char* pattern, optionType type,
+                        int& res) {
+        String parsedValue;
+        if(!parseOption(argc, argv, pattern, &parsedValue))
+            return false;
+
+        if(type == 0) {
+            // boolean
+            const char positive[][5] = {"1", "true", "on", "yes"};  // 5 - strlen("true") + 1
+            const char negative[][6] = {"0", "false", "off", "no"}; // 6 - strlen("false") + 1
+
+            // if the value matches any of the positive/negative possibilities
+            for(unsigned i = 0; i < 4; i++) {
+                if(parsedValue.compare(positive[i], true) == 0) {
+                    res = 1; //!OCLINT parameter reassignment
+                    return true;
+                }
+                if(parsedValue.compare(negative[i], true) == 0) {
+                    res = 0; //!OCLINT parameter reassignment
+                    return true;
+                }
+            }
+        } else {
+            // integer
+            // TODO: change this to use std::stoi or something else! currently it uses undefined behavior - assumes '0' on failed parse...
+            int theInt = std::atoi(parsedValue.c_str()); // NOLINT
+            if(theInt != 0) {
+                res = theInt; //!OCLINT parameter reassignment
+                return true;
+            }
+        }
+        return false;
+    }
+} // namespace
+
+Context::Context(int argc, const char* const* argv)
+        : p(new detail::ContextState) {
+    parseArgs(argc, argv, true);
+    if(argc)
+        p->binary_name = argv[0];
+}
+
+Context::~Context() {
+    if(g_cs == p)
+        g_cs = nullptr;
+    delete p;
+}
+
+void Context::applyCommandLine(int argc, const char* const* argv) {
+    parseArgs(argc, argv);
+    if(argc)
+        p->binary_name = argv[0];
+}
+
+// parses args
+void Context::parseArgs(int argc, const char* const* argv, bool withDefaults) {
+    using namespace detail;
+
+    // clang-format off
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "source-file=",        p->filters[0]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "sf=",                 p->filters[0]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "source-file-exclude=",p->filters[1]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "sfe=",                p->filters[1]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "test-suite=",         p->filters[2]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "ts=",                 p->filters[2]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "test-suite-exclude=", p->filters[3]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "tse=",                p->filters[3]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "test-case=",          p->filters[4]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "tc=",                 p->filters[4]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "test-case-exclude=",  p->filters[5]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "tce=",                p->filters[5]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "subcase=",            p->filters[6]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "sc=",                 p->filters[6]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "subcase-exclude=",    p->filters[7]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "sce=",                p->filters[7]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "reporters=",          p->filters[8]);
+    parseCommaSepArgs(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "r=",                  p->filters[8]);
+    // clang-format on
+
+    int    intRes = 0;
+    String strRes;
+
+#define DOCTEST_PARSE_AS_BOOL_OR_FLAG(name, sname, var, default)                                   \
+    if(parseIntOption(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX name "=", option_bool, intRes) ||  \
+       parseIntOption(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX sname "=", option_bool, intRes))   \
+        p->var = !!intRes;                                                                         \
+    else if(parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX name) ||                           \
+            parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX sname))                            \
+        p->var = true;                                                                             \
+    else if(withDefaults)                                                                          \
+    p->var = default
+
+#define DOCTEST_PARSE_INT_OPTION(name, sname, var, default)                                        \
+    if(parseIntOption(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX name "=", option_int, intRes) ||   \
+       parseIntOption(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX sname "=", option_int, intRes))    \
+        p->var = intRes;                                                                           \
+    else if(withDefaults)                                                                          \
+    p->var = default
+
+#define DOCTEST_PARSE_STR_OPTION(name, sname, var, default)                                        \
+    if(parseOption(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX name "=", &strRes, default) ||        \
+       parseOption(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX sname "=", &strRes, default) ||       \
+       withDefaults)                                                                               \
+    p->var = strRes
+
+    // clang-format off
+    DOCTEST_PARSE_STR_OPTION("out", "o", out, "");
+    DOCTEST_PARSE_STR_OPTION("order-by", "ob", order_by, "file");
+    DOCTEST_PARSE_INT_OPTION("rand-seed", "rs", rand_seed, 0);
+
+    DOCTEST_PARSE_INT_OPTION("first", "f", first, 0);
+    DOCTEST_PARSE_INT_OPTION("last", "l", last, UINT_MAX);
+
+    DOCTEST_PARSE_INT_OPTION("abort-after", "aa", abort_after, 0);
+    DOCTEST_PARSE_INT_OPTION("subcase-filter-levels", "scfl", subcase_filter_levels, INT_MAX);
+
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("success", "s", success, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("case-sensitive", "cs", case_sensitive, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("exit", "e", exit, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("duration", "d", duration, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("no-throw", "nt", no_throw, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("no-exitcode", "ne", no_exitcode, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("no-run", "nr", no_run, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("no-version", "nv", no_version, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("no-colors", "nc", no_colors, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("force-colors", "fc", force_colors, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("no-breaks", "nb", no_breaks, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("no-skip", "ns", no_skip, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("gnu-file-line", "gfl", gnu_file_line, !bool(DOCTEST_MSVC));
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("no-path-filenames", "npf", no_path_in_filenames, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("no-line-numbers", "nln", no_line_numbers, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("no-skipped-summary", "nss", no_skipped_summary, false);
+    DOCTEST_PARSE_AS_BOOL_OR_FLAG("no-time-in-output", "ntio", no_time_in_output, false);
+    // clang-format on
+
+    if(withDefaults) {
+        p->help             = false;
+        p->version          = false;
+        p->count            = false;
+        p->list_test_cases  = false;
+        p->list_test_suites = false;
+        p->list_reporters   = false;
+    }
+    if(parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "help") ||
+       parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "h") ||
+       parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "?")) {
+        p->help = true;
+        p->exit = true;
+    }
+    if(parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "version") ||
+       parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "v")) {
+        p->version = true;
+        p->exit    = true;
+    }
+    if(parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "count") ||
+       parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "c")) {
+        p->count = true;
+        p->exit  = true;
+    }
+    if(parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "list-test-cases") ||
+       parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "ltc")) {
+        p->list_test_cases = true;
+        p->exit            = true;
+    }
+    if(parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "list-test-suites") ||
+       parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "lts")) {
+        p->list_test_suites = true;
+        p->exit             = true;
+    }
+    if(parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "list-reporters") ||
+       parseFlag(argc, argv, DOCTEST_CONFIG_OPTIONS_PREFIX "lr")) {
+        p->list_reporters = true;
+        p->exit           = true;
+    }
+}
+
+// allows the user to add procedurally to the filters from the command line
+void Context::addFilter(const char* filter, const char* value) { setOption(filter, value); }
+
+// allows the user to clear all filters from the command line
+void Context::clearFilters() {
+    for(auto& curr : p->filters)
+        curr.clear();
+}
+
+// allows the user to override procedurally the int/bool options from the command line
+void Context::setOption(const char* option, int value) {
+    setOption(option, toString(value).c_str());
+}
+
+// allows the user to override procedurally the string options from the command line
+void Context::setOption(const char* option, const char* value) {
+    auto argv   = String("-") + option + "=" + value;
+    auto lvalue = argv.c_str();
+    parseArgs(1, &lvalue);
+}
+
+// users should query this in their main() and exit the program if true
+bool Context::shouldExit() { return p->exit; }
+
+void Context::setAsDefaultForAssertsOutOfTestCases() { g_cs = p; }
+
+void Context::setAssertHandler(detail::assert_handler ah) { p->ah = ah; }
+
+// the main function that does all the filtering and test running
+int Context::run() {
+    using namespace detail;
+
+    // save the old context state in case such was setup - for using asserts out of a testing context
+    auto old_cs = g_cs;
+    // this is the current contest
+    g_cs               = p;
+    is_running_in_test = true;
+
+    g_no_colors = p->no_colors;
+    p->resetRunData();
+
+    // stdout by default
+    p->cout = &std::cout;
+    p->cerr = &std::cerr;
+
+    // or to a file if specified
+    std::fstream fstr;
+    if(p->out.size()) {
+        fstr.open(p->out.c_str(), std::fstream::out);
+        p->cout = &fstr;
+    }
+
+    auto cleanup_and_return = [&]() {
+        if(fstr.is_open())
+            fstr.close();
+
+        // restore context
+        g_cs               = old_cs;
+        is_running_in_test = false;
+
+        // we have to free the reporters which were allocated when the run started
+        for(auto& curr : p->reporters_currently_used)
+            delete curr;
+        p->reporters_currently_used.clear();
+
+        if(p->numTestCasesFailed && !p->no_exitcode)
+            return EXIT_FAILURE;
+        return EXIT_SUCCESS;
+    };
+
+    // setup default reporter if none is given through the command line
+    if(p->filters[8].empty())
+        p->filters[8].push_back("console");
+
+    // check to see if any of the registered reporters has been selected
+    for(auto& curr : getReporters()) {
+        if(matchesAny(curr.first.second.c_str(), p->filters[8], false, p->case_sensitive))
+            p->reporters_currently_used.push_back(curr.second(*g_cs));
+    }
+
+    // TODO: check if there is nothing in reporters_currently_used
+
+    // prepend all listeners
+    for(auto& curr : getListeners())
+        p->reporters_currently_used.insert(p->reporters_currently_used.begin(), curr.second(*g_cs));
+
+#ifdef DOCTEST_PLATFORM_WINDOWS
+    if(isDebuggerActive())
+        p->reporters_currently_used.push_back(new DebugOutputWindowReporter(*g_cs));
+#endif // DOCTEST_PLATFORM_WINDOWS
+
+    // handle version, help and no_run
+    if(p->no_run || p->version || p->help || p->list_reporters) {
+        DOCTEST_ITERATE_THROUGH_REPORTERS(report_query, QueryData());
+
+        return cleanup_and_return();
+    }
+
+    std::vector<const TestCase*> testArray;
+    for(auto& curr : getRegisteredTests())
+        testArray.push_back(&curr);
+    p->numTestCases = testArray.size();
+
+    // sort the collected records
+    if(!testArray.empty()) {
+        if(p->order_by.compare("file", true) == 0) {
+            std::sort(testArray.begin(), testArray.end(), fileOrderComparator);
+        } else if(p->order_by.compare("suite", true) == 0) {
+            std::sort(testArray.begin(), testArray.end(), suiteOrderComparator);
+        } else if(p->order_by.compare("name", true) == 0) {
+            std::sort(testArray.begin(), testArray.end(), nameOrderComparator);
+        } else if(p->order_by.compare("rand", true) == 0) {
+            std::srand(p->rand_seed);
+
+            // random_shuffle implementation
+            const auto first = &testArray[0];
+            for(size_t i = testArray.size() - 1; i > 0; --i) {
+                int idxToSwap = std::rand() % (i + 1); // NOLINT
+
+                const auto temp = first[i];
+
+                first[i]         = first[idxToSwap];
+                first[idxToSwap] = temp;
+            }
+        }
+    }
+
+    std::set<String> testSuitesPassingFilt;
+
+    bool                             query_mode = p->count || p->list_test_cases || p->list_test_suites;
+    std::vector<const TestCaseData*> queryResults;
+
+    if(!query_mode)
+        DOCTEST_ITERATE_THROUGH_REPORTERS(test_run_start, DOCTEST_EMPTY);
+
+    // invoke the registered functions if they match the filter criteria (or just count them)
+    for(auto& curr : testArray) {
+        const auto& tc = *curr;
+
+        bool skip_me = false;
+        if(tc.m_skip && !p->no_skip)
+            skip_me = true;
+
+        if(!matchesAny(tc.m_file.c_str(), p->filters[0], true, p->case_sensitive))
+            skip_me = true;
+        if(matchesAny(tc.m_file.c_str(), p->filters[1], false, p->case_sensitive))
+            skip_me = true;
+        if(!matchesAny(tc.m_test_suite, p->filters[2], true, p->case_sensitive))
+            skip_me = true;
+        if(matchesAny(tc.m_test_suite, p->filters[3], false, p->case_sensitive))
+            skip_me = true;
+        if(!matchesAny(tc.m_name, p->filters[4], true, p->case_sensitive))
+            skip_me = true;
+        if(matchesAny(tc.m_name, p->filters[5], false, p->case_sensitive))
+            skip_me = true;
+
+        if(!skip_me)
+            p->numTestCasesPassingFilters++;
+
+        // skip the test if it is not in the execution range
+        if((p->last < p->numTestCasesPassingFilters && p->first <= p->last) ||
+           (p->first > p->numTestCasesPassingFilters))
+            skip_me = true;
+
+        if(skip_me) {
+            if(!query_mode)
+                DOCTEST_ITERATE_THROUGH_REPORTERS(test_case_skipped, tc);
+            continue;
+        }
+
+        // do not execute the test if we are to only count the number of filter passing tests
+        if(p->count)
+            continue;
+
+        // print the name of the test and don't execute it
+        if(p->list_test_cases) {
+            queryResults.push_back(&tc);
+            continue;
+        }
+
+        // print the name of the test suite if not done already and don't execute it
+        if(p->list_test_suites) {
+            if((testSuitesPassingFilt.count(tc.m_test_suite) == 0) && tc.m_test_suite[0] != '\0') {
+                queryResults.push_back(&tc);
+                testSuitesPassingFilt.insert(tc.m_test_suite);
+                p->numTestSuitesPassingFilters++;
+            }
+            continue;
+        }
+
+        // execute the test if it passes all the filtering
+        {
+            p->currentTest = &tc;
+
+            p->failure_flags = TestCaseFailureReason::None;
+            p->seconds       = 0;
+
+            // reset atomic counters
+            p->numAssertsFailedCurrentTest_atomic = 0;
+            p->numAssertsCurrentTest_atomic       = 0;
+
+            p->subcasesPassed.clear();
+
+            DOCTEST_ITERATE_THROUGH_REPORTERS(test_case_start, tc);
+
+            p->timer.start();
+            
+            bool run_test = true;
+
+            do {
+                // reset some of the fields for subcases (except for the set of fully passed ones)
+                p->should_reenter          = false;
+                p->subcasesCurrentMaxLevel = 0;
+                p->subcasesStack.clear();
+
+                p->shouldLogCurrentException = true;
+
+                // reset stuff for logging with INFO()
+                p->stringifiedContexts.clear();
+
+#ifndef DOCTEST_CONFIG_NO_EXCEPTIONS
+                try {
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS
+                    FatalConditionHandler fatalConditionHandler; // Handle signals
+                    // execute the test
+                    tc.m_test();
+                    fatalConditionHandler.reset();
+#ifndef DOCTEST_CONFIG_NO_EXCEPTIONS
+                } catch(const TestFailureException&) {
+                    p->failure_flags |= TestCaseFailureReason::AssertFailure;
+                } catch(...) {
+                    DOCTEST_ITERATE_THROUGH_REPORTERS(test_case_exception,
+                                                      {translateActiveException(), false});
+                    p->failure_flags |= TestCaseFailureReason::Exception;
+                }
+#endif // DOCTEST_CONFIG_NO_EXCEPTIONS
+
+                // exit this loop if enough assertions have failed - even if there are more subcases
+                if(p->abort_after > 0 &&
+                   p->numAssertsFailed + p->numAssertsFailedCurrentTest_atomic >= p->abort_after) {
+                    run_test = false;
+                    p->failure_flags |= TestCaseFailureReason::TooManyFailedAsserts;
+                }
+                
+                if(p->should_reenter && run_test)
+                    DOCTEST_ITERATE_THROUGH_REPORTERS(test_case_reenter, tc);
+                if(!p->should_reenter)
+                    run_test = false;
+            } while(run_test);
+
+            p->finalizeTestCaseData();
+
+            DOCTEST_ITERATE_THROUGH_REPORTERS(test_case_end, *g_cs);
+
+            p->currentTest = nullptr;
+
+            // stop executing tests if enough assertions have failed
+            if(p->abort_after > 0 && p->numAssertsFailed >= p->abort_after)
+                break;
+        }
+    }
+
+    if(!query_mode) {
+        DOCTEST_ITERATE_THROUGH_REPORTERS(test_run_end, *g_cs);
+    } else {
+        QueryData qdata;
+        qdata.run_stats = g_cs;
+        qdata.data      = queryResults.data();
+        qdata.num_data  = unsigned(queryResults.size());
+        DOCTEST_ITERATE_THROUGH_REPORTERS(report_query, qdata);
+    }
+
+    // see these issues on the reasoning for this:
+    // - https://github.com/onqtam/doctest/issues/143#issuecomment-414418903
+    // - https://github.com/onqtam/doctest/issues/126
+    auto DOCTEST_FIX_FOR_MACOS_LIBCPP_IOSFWD_STRING_LINK_ERRORS = []() DOCTEST_NOINLINE
+        { std::cout << std::string(); };
+    DOCTEST_FIX_FOR_MACOS_LIBCPP_IOSFWD_STRING_LINK_ERRORS();
+
+    return cleanup_and_return();
+}
+
+IReporter::~IReporter() = default;
+
+int IReporter::get_num_active_contexts() { return detail::g_infoContexts.size(); }
+const IContextScope* const* IReporter::get_active_contexts() {
+    return get_num_active_contexts() ? &detail::g_infoContexts[0] : nullptr;
+}
+
+int IReporter::get_num_stringified_contexts() { return detail::g_cs->stringifiedContexts.size(); }
+const String* IReporter::get_stringified_contexts() {
+    return get_num_stringified_contexts() ? &detail::g_cs->stringifiedContexts[0] : nullptr;
+}
+
+namespace detail {
+    void registerReporterImpl(const char* name, int priority, reporterCreatorFunc c, bool isReporter) {
+        if(isReporter)
+            getReporters().insert(reporterMap::value_type(reporterMap::key_type(priority, name), c));
+        else
+            getListeners().insert(reporterMap::value_type(reporterMap::key_type(priority, name), c));
+    }
+} // namespace detail
+
+} // namespace doctest
+
+#endif // DOCTEST_CONFIG_DISABLE
+
+#ifdef DOCTEST_CONFIG_IMPLEMENT_WITH_MAIN
+DOCTEST_MSVC_SUPPRESS_WARNING_WITH_PUSH(4007) // 'function' : must be 'attribute' - see issue #182
+int main(int argc, char** argv) { return doctest::Context(argc, argv).run(); }
+DOCTEST_MSVC_SUPPRESS_WARNING_POP
+#endif // DOCTEST_CONFIG_IMPLEMENT_WITH_MAIN
+
+DOCTEST_CLANG_SUPPRESS_WARNING_POP
+DOCTEST_MSVC_SUPPRESS_WARNING_POP
+DOCTEST_GCC_SUPPRESS_WARNING_POP
+
+#endif // DOCTEST_LIBRARY_IMPLEMENTATION
+#endif // DOCTEST_CONFIG_IMPLEMENT

+ 6410 - 0
Tests/Dependencies/lodepng/lodepng.cpp

@@ -0,0 +1,6410 @@
+/*
+LodePNG version 20200306
+
+Copyright (c) 2005-2020 Lode Vandevenne
+
+This software is provided 'as-is', without any express or implied
+warranty. In no event will the authors be held liable for any damages
+arising from the use of this software.
+
+Permission is granted to anyone to use this software for any purpose,
+including commercial applications, and to alter it and redistribute it
+freely, subject to the following restrictions:
+
+    1. The origin of this software must not be misrepresented; you must not
+    claim that you wrote the original software. If you use this software
+    in a product, an acknowledgment in the product documentation would be
+    appreciated but is not required.
+
+    2. Altered source versions must be plainly marked as such, and must not be
+    misrepresented as being the original software.
+
+    3. This notice may not be removed or altered from any source
+    distribution.
+*/
+
+/*
+The manual and changelog are in the header file "lodepng.h"
+Rename this file to lodepng.cpp to use it for C++, or to lodepng.c to use it for C.
+*/
+
+#include "lodepng.h"
+
+#ifdef LODEPNG_COMPILE_DISK
+#include <limits.h> /* LONG_MAX */
+#include <stdio.h> /* file handling */
+#endif /* LODEPNG_COMPILE_DISK */
+
+#ifdef LODEPNG_COMPILE_ALLOCATORS
+#include <stdlib.h> /* allocations */
+#endif /* LODEPNG_COMPILE_ALLOCATORS */
+
+#if defined(_MSC_VER) && (_MSC_VER >= 1310) /*Visual Studio: A few warning types are not desired here.*/
+#pragma warning( disable : 4244 ) /*implicit conversions: not warned by gcc -Wall -Wextra and requires too much casts*/
+#pragma warning( disable : 4996 ) /*VS does not like fopen, but fopen_s is not standard C so unusable here*/
+#endif /*_MSC_VER */
+
+const char* LODEPNG_VERSION_STRING = "20200306";
+
+/*
+This source file is built up in the following large parts. The code sections
+with the "LODEPNG_COMPILE_" #defines divide this up further in an intermixed way.
+-Tools for C and common code for PNG and Zlib
+-C Code for Zlib (huffman, deflate, ...)
+-C Code for PNG (file format chunks, adam7, PNG filters, color conversions, ...)
+-The C++ wrapper around all of the above
+*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* // Tools for C, and common code for PNG and Zlib.                       // */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*The malloc, realloc and free functions defined here with "lodepng_" in front
+of the name, so that you can easily change them to others related to your
+platform if needed. Everything else in the code calls these. Pass
+-DLODEPNG_NO_COMPILE_ALLOCATORS to the compiler, or comment out
+#define LODEPNG_COMPILE_ALLOCATORS in the header, to disable the ones here and
+define them in your own project's source files without needing to change
+lodepng source code. Don't forget to remove "static" if you copypaste them
+from here.*/
+
+#ifdef LODEPNG_COMPILE_ALLOCATORS
+static void* lodepng_malloc(size_t size) {
+#ifdef LODEPNG_MAX_ALLOC
+  if(size > LODEPNG_MAX_ALLOC) return 0;
+#endif
+  return malloc(size);
+}
+
+/* NOTE: when realloc returns NULL, it leaves the original memory untouched */
+static void* lodepng_realloc(void* ptr, size_t new_size) {
+#ifdef LODEPNG_MAX_ALLOC
+  if(new_size > LODEPNG_MAX_ALLOC) return 0;
+#endif
+  return realloc(ptr, new_size);
+}
+
+static void lodepng_free(void* ptr) {
+  free(ptr);
+}
+#else /*LODEPNG_COMPILE_ALLOCATORS*/
+/* TODO: support giving additional void* payload to the custom allocators */
+void* lodepng_malloc(size_t size);
+void* lodepng_realloc(void* ptr, size_t new_size);
+void lodepng_free(void* ptr);
+#endif /*LODEPNG_COMPILE_ALLOCATORS*/
+
+/* convince the compiler to inline a function, for use when this measurably improves performance */
+/* inline is not available in C90, but use it when supported by the compiler */
+#if (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)) || (defined(__cplusplus) && (__cplusplus >= 199711L))
+#define LODEPNG_INLINE inline
+#else
+#define LODEPNG_INLINE /* not available */
+#endif
+
+/* restrict is not available in C90, but use it when supported by the compiler */
+#if (defined(__GNUC__) && (__GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))) ||\
+    (defined(_MSC_VER) && (_MSC_VER >= 1400)) || \
+    (defined(__WATCOMC__) && (__WATCOMC__ >= 1250) && !defined(__cplusplus))
+#define LODEPNG_RESTRICT __restrict
+#else
+#define LODEPNG_RESTRICT /* not available */
+#endif
+
+/* Replacements for C library functions such as memcpy and strlen, to support platforms
+where a full C library is not available. The compiler can recognize them and compile
+to something as fast. */
+
+static void lodepng_memcpy(void* LODEPNG_RESTRICT dst,
+                           const void* LODEPNG_RESTRICT src, size_t size) {
+  size_t i;
+  for(i = 0; i < size; i++) ((char*)dst)[i] = ((const char*)src)[i];
+}
+
+static void lodepng_memset(void* LODEPNG_RESTRICT dst,
+                           int value, size_t num) {
+  size_t i;
+  for(i = 0; i < num; i++) ((char*)dst)[i] = (char)value;
+}
+
+/* does not check memory out of bounds, do not use on untrusted data */
+static size_t lodepng_strlen(const char* a) {
+  const char* orig = a;
+  /* avoid warning about unused function in case of disabled COMPILE... macros */
+  (void)(&lodepng_strlen);
+  while(*a) a++;
+  return (size_t)(a - orig);
+}
+
+#define LODEPNG_MAX(a, b) (((a) > (b)) ? (a) : (b))
+#define LODEPNG_MIN(a, b) (((a) < (b)) ? (a) : (b))
+#define LODEPNG_ABS(x) ((x) < 0 ? -(x) : (x))
+
+#if defined(LODEPNG_COMPILE_PNG) || defined(LODEPNG_COMPILE_DECODER)
+/* Safely check if adding two integers will overflow (no undefined
+behavior, compiler removing the code, etc...) and output result. */
+static int lodepng_addofl(size_t a, size_t b, size_t* result) {
+  *result = a + b; /* Unsigned addition is well defined and safe in C90 */
+  return *result < a;
+}
+#endif /*defined(LODEPNG_COMPILE_PNG) || defined(LODEPNG_COMPILE_DECODER)*/
+
+#ifdef LODEPNG_COMPILE_DECODER
+/* Safely check if multiplying two integers will overflow (no undefined
+behavior, compiler removing the code, etc...) and output result. */
+static int lodepng_mulofl(size_t a, size_t b, size_t* result) {
+  *result = a * b; /* Unsigned multiplication is well defined and safe in C90 */
+  return (a != 0 && *result / a != b);
+}
+
+#ifdef LODEPNG_COMPILE_ZLIB
+/* Safely check if a + b > c, even if overflow could happen. */
+static int lodepng_gtofl(size_t a, size_t b, size_t c) {
+  size_t d;
+  if(lodepng_addofl(a, b, &d)) return 1;
+  return d > c;
+}
+#endif /*LODEPNG_COMPILE_ZLIB*/
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+
+/*
+Often in case of an error a value is assigned to a variable and then it breaks
+out of a loop (to go to the cleanup phase of a function). This macro does that.
+It makes the error handling code shorter and more readable.
+
+Example: if(!uivector_resize(&lz77_encoded, datasize)) ERROR_BREAK(83);
+*/
+#define CERROR_BREAK(errorvar, code){\
+  errorvar = code;\
+  break;\
+}
+
+/*version of CERROR_BREAK that assumes the common case where the error variable is named "error"*/
+#define ERROR_BREAK(code) CERROR_BREAK(error, code)
+
+/*Set error var to the error code, and return it.*/
+#define CERROR_RETURN_ERROR(errorvar, code){\
+  errorvar = code;\
+  return code;\
+}
+
+/*Try the code, if it returns error, also return the error.*/
+#define CERROR_TRY_RETURN(call){\
+  unsigned error = call;\
+  if(error) return error;\
+}
+
+/*Set error var to the error code, and return from the void function.*/
+#define CERROR_RETURN(errorvar, code){\
+  errorvar = code;\
+  return;\
+}
+
+/*
+About uivector, ucvector and string:
+-All of them wrap dynamic arrays or text strings in a similar way.
+-LodePNG was originally written in C++. The vectors replace the std::vectors that were used in the C++ version.
+-The string tools are made to avoid problems with compilers that declare things like strncat as deprecated.
+-They're not used in the interface, only internally in this file as static functions.
+-As with many other structs in this file, the init and cleanup functions serve as ctor and dtor.
+*/
+
+#ifdef LODEPNG_COMPILE_ZLIB
+#ifdef LODEPNG_COMPILE_ENCODER
+/*dynamic vector of unsigned ints*/
+typedef struct uivector {
+  unsigned* data;
+  size_t size; /*size in number of unsigned longs*/
+  size_t allocsize; /*allocated size in bytes*/
+} uivector;
+
+static void uivector_cleanup(void* p) {
+  ((uivector*)p)->size = ((uivector*)p)->allocsize = 0;
+  lodepng_free(((uivector*)p)->data);
+  ((uivector*)p)->data = NULL;
+}
+
+/*returns 1 if success, 0 if failure ==> nothing done*/
+static unsigned uivector_resize(uivector* p, size_t size) {
+  size_t allocsize = size * sizeof(unsigned);
+  if(allocsize > p->allocsize) {
+    size_t newsize = allocsize + (p->allocsize >> 1u);
+    void* data = lodepng_realloc(p->data, newsize);
+    if(data) {
+      p->allocsize = newsize;
+      p->data = (unsigned*)data;
+    }
+    else return 0; /*error: not enough memory*/
+  }
+  p->size = size;
+  return 1; /*success*/
+}
+
+static void uivector_init(uivector* p) {
+  p->data = NULL;
+  p->size = p->allocsize = 0;
+}
+
+/*returns 1 if success, 0 if failure ==> nothing done*/
+static unsigned uivector_push_back(uivector* p, unsigned c) {
+  if(!uivector_resize(p, p->size + 1)) return 0;
+  p->data[p->size - 1] = c;
+  return 1;
+}
+#endif /*LODEPNG_COMPILE_ENCODER*/
+#endif /*LODEPNG_COMPILE_ZLIB*/
+
+/* /////////////////////////////////////////////////////////////////////////// */
+
+/*dynamic vector of unsigned chars*/
+typedef struct ucvector {
+  unsigned char* data;
+  size_t size; /*used size*/
+  size_t allocsize; /*allocated size*/
+} ucvector;
+
+/*returns 1 if success, 0 if failure ==> nothing done*/
+static unsigned ucvector_resize(ucvector* p, size_t size) {
+  if(size > p->allocsize) {
+    size_t newsize = size + (p->allocsize >> 1u);
+    void* data = lodepng_realloc(p->data, newsize);
+    if(data) {
+      p->allocsize = newsize;
+      p->data = (unsigned char*)data;
+    }
+    else return 0; /*error: not enough memory*/
+  }
+  p->size = size;
+  return 1; /*success*/
+}
+
+static ucvector ucvector_init(unsigned char* buffer, size_t size) {
+  ucvector v;
+  v.data = buffer;
+  v.allocsize = v.size = size;
+  return v;
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef LODEPNG_COMPILE_PNG
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+
+/*free string pointer and set it to NULL*/
+static void string_cleanup(char** out) {
+  lodepng_free(*out);
+  *out = NULL;
+}
+
+static char* alloc_string_sized(const char* in, size_t insize) {
+  char* out = (char*)lodepng_malloc(insize + 1);
+  if(out) {
+    lodepng_memcpy(out, in, insize);
+    out[insize] = 0;
+  }
+  return out;
+}
+
+/* dynamically allocates a new string with a copy of the null terminated input text */
+static char* alloc_string(const char* in) {
+  return alloc_string_sized(in, lodepng_strlen(in));
+}
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+#endif /*LODEPNG_COMPILE_PNG*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#if defined(LODEPNG_COMPILE_DECODER) || defined(LODEPNG_COMPILE_PNG)
+static unsigned lodepng_read32bitInt(const unsigned char* buffer) {
+  return (((unsigned)buffer[0] << 24u) | ((unsigned)buffer[1] << 16u) |
+         ((unsigned)buffer[2] << 8u) | (unsigned)buffer[3]);
+}
+#endif /*defined(LODEPNG_COMPILE_DECODER) || defined(LODEPNG_COMPILE_PNG)*/
+
+#if defined(LODEPNG_COMPILE_PNG) || defined(LODEPNG_COMPILE_ENCODER)
+/*buffer must have at least 4 allocated bytes available*/
+static void lodepng_set32bitInt(unsigned char* buffer, unsigned value) {
+  buffer[0] = (unsigned char)((value >> 24) & 0xff);
+  buffer[1] = (unsigned char)((value >> 16) & 0xff);
+  buffer[2] = (unsigned char)((value >>  8) & 0xff);
+  buffer[3] = (unsigned char)((value      ) & 0xff);
+}
+#endif /*defined(LODEPNG_COMPILE_PNG) || defined(LODEPNG_COMPILE_ENCODER)*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / File IO                                                                / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef LODEPNG_COMPILE_DISK
+
+/* returns negative value on error. This should be pure C compatible, so no fstat. */
+static long lodepng_filesize(const char* filename) {
+  FILE* file;
+  long size;
+  file = fopen(filename, "rb");
+  if(!file) return -1;
+
+  if(fseek(file, 0, SEEK_END) != 0) {
+    fclose(file);
+    return -1;
+  }
+
+  size = ftell(file);
+  /* It may give LONG_MAX as directory size, this is invalid for us. */
+  if(size == LONG_MAX) size = -1;
+
+  fclose(file);
+  return size;
+}
+
+/* load file into buffer that already has the correct allocated size. Returns error code.*/
+static unsigned lodepng_buffer_file(unsigned char* out, size_t size, const char* filename) {
+  FILE* file;
+  size_t readsize;
+  file = fopen(filename, "rb");
+  if(!file) return 78;
+
+  readsize = fread(out, 1, size, file);
+  fclose(file);
+
+  if(readsize != size) return 78;
+  return 0;
+}
+
+unsigned lodepng_load_file(unsigned char** out, size_t* outsize, const char* filename) {
+  long size = lodepng_filesize(filename);
+  if(size < 0) return 78;
+  *outsize = (size_t)size;
+
+  *out = (unsigned char*)lodepng_malloc((size_t)size);
+  if(!(*out) && size > 0) return 83; /*the above malloc failed*/
+
+  return lodepng_buffer_file(*out, (size_t)size, filename);
+}
+
+/*write given buffer to the file, overwriting the file, it doesn't append to it.*/
+unsigned lodepng_save_file(const unsigned char* buffer, size_t buffersize, const char* filename) {
+  FILE* file;
+  file = fopen(filename, "wb" );
+  if(!file) return 79;
+  fwrite(buffer, 1, buffersize, file);
+  fclose(file);
+  return 0;
+}
+
+#endif /*LODEPNG_COMPILE_DISK*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* // End of common code and tools. Begin of Zlib related code.            // */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef LODEPNG_COMPILE_ZLIB
+#ifdef LODEPNG_COMPILE_ENCODER
+
+typedef struct {
+  ucvector* data;
+  unsigned char bp; /*ok to overflow, indicates bit pos inside byte*/
+} LodePNGBitWriter;
+
+static void LodePNGBitWriter_init(LodePNGBitWriter* writer, ucvector* data) {
+  writer->data = data;
+  writer->bp = 0;
+}
+
+/*TODO: this ignores potential out of memory errors*/
+#define WRITEBIT(writer, bit){\
+  /* append new byte */\
+  if(((writer->bp) & 7u) == 0) {\
+    if(!ucvector_resize(writer->data, writer->data->size + 1)) return;\
+    writer->data->data[writer->data->size - 1] = 0;\
+  }\
+  (writer->data->data[writer->data->size - 1]) |= (bit << ((writer->bp) & 7u));\
+  ++writer->bp;\
+}
+
+/* LSB of value is written first, and LSB of bytes is used first */
+static void writeBits(LodePNGBitWriter* writer, unsigned value, size_t nbits) {
+  if(nbits == 1) { /* compiler should statically compile this case if nbits == 1 */
+    WRITEBIT(writer, value);
+  } else {
+    /* TODO: increase output size only once here rather than in each WRITEBIT */
+    size_t i;
+    for(i = 0; i != nbits; ++i) {
+      WRITEBIT(writer, (unsigned char)((value >> i) & 1));
+    }
+  }
+}
+
+/* This one is to use for adding huffman symbol, the value bits are written MSB first */
+static void writeBitsReversed(LodePNGBitWriter* writer, unsigned value, size_t nbits) {
+  size_t i;
+  for(i = 0; i != nbits; ++i) {
+    /* TODO: increase output size only once here rather than in each WRITEBIT */
+    WRITEBIT(writer, (unsigned char)((value >> (nbits - 1u - i)) & 1u));
+  }
+}
+#endif /*LODEPNG_COMPILE_ENCODER*/
+
+#ifdef LODEPNG_COMPILE_DECODER
+
+typedef struct {
+  const unsigned char* data;
+  size_t size; /*size of data in bytes*/
+  size_t bitsize; /*size of data in bits, end of valid bp values, should be 8*size*/
+  size_t bp;
+  unsigned buffer; /*buffer for reading bits. NOTE: 'unsigned' must support at least 32 bits*/
+} LodePNGBitReader;
+
+/* data size argument is in bytes. Returns error if size too large causing overflow */
+static unsigned LodePNGBitReader_init(LodePNGBitReader* reader, const unsigned char* data, size_t size) {
+  size_t temp;
+  reader->data = data;
+  reader->size = size;
+  /* size in bits, return error if overflow (if size_t is 32 bit this supports up to 500MB)  */
+  if(lodepng_mulofl(size, 8u, &reader->bitsize)) return 105;
+  /*ensure incremented bp can be compared to bitsize without overflow even when it would be incremented 32 too much and
+  trying to ensure 32 more bits*/
+  if(lodepng_addofl(reader->bitsize, 64u, &temp)) return 105;
+  reader->bp = 0;
+  reader->buffer = 0;
+  return 0; /*ok*/
+}
+
+/*
+ensureBits functions:
+Ensures the reader can at least read nbits bits in one or more readBits calls,
+safely even if not enough bits are available.
+Returns 1 if there are enough bits available, 0 if not.
+*/
+
+/*See ensureBits documentation above. This one ensures exactly 1 bit */
+/*static unsigned ensureBits1(LodePNGBitReader* reader) {
+  if(reader->bp >= reader->bitsize) return 0;
+  reader->buffer = (unsigned)reader->data[reader->bp >> 3u] >> (reader->bp & 7u);
+  return 1;
+}*/
+
+/*See ensureBits documentation above. This one ensures up to 9 bits */
+static unsigned ensureBits9(LodePNGBitReader* reader, size_t nbits) {
+  size_t start = reader->bp >> 3u;
+  size_t size = reader->size;
+  if(start + 1u < size) {
+    reader->buffer = (unsigned)reader->data[start + 0] | ((unsigned)reader->data[start + 1] << 8u);
+    reader->buffer >>= (reader->bp & 7u);
+    return 1;
+  } else {
+    reader->buffer = 0;
+    if(start + 0u < size) reader->buffer |= reader->data[start + 0];
+    reader->buffer >>= (reader->bp & 7u);
+    return reader->bp + nbits <= reader->bitsize;
+  }
+}
+
+/*See ensureBits documentation above. This one ensures up to 17 bits */
+static unsigned ensureBits17(LodePNGBitReader* reader, size_t nbits) {
+  size_t start = reader->bp >> 3u;
+  size_t size = reader->size;
+  if(start + 2u < size) {
+    reader->buffer = (unsigned)reader->data[start + 0] | ((unsigned)reader->data[start + 1] << 8u) |
+                     ((unsigned)reader->data[start + 2] << 16u);
+    reader->buffer >>= (reader->bp & 7u);
+    return 1;
+  } else {
+    reader->buffer = 0;
+    if(start + 0u < size) reader->buffer |= reader->data[start + 0];
+    if(start + 1u < size) reader->buffer |= ((unsigned)reader->data[start + 1] << 8u);
+    reader->buffer >>= (reader->bp & 7u);
+    return reader->bp + nbits <= reader->bitsize;
+  }
+}
+
+/*See ensureBits documentation above. This one ensures up to 25 bits */
+static LODEPNG_INLINE unsigned ensureBits25(LodePNGBitReader* reader, size_t nbits) {
+  size_t start = reader->bp >> 3u;
+  size_t size = reader->size;
+  if(start + 3u < size) {
+    reader->buffer = (unsigned)reader->data[start + 0] | ((unsigned)reader->data[start + 1] << 8u) |
+                     ((unsigned)reader->data[start + 2] << 16u) | ((unsigned)reader->data[start + 3] << 24u);
+    reader->buffer >>= (reader->bp & 7u);
+    return 1;
+  } else {
+    reader->buffer = 0;
+    if(start + 0u < size) reader->buffer |= reader->data[start + 0];
+    if(start + 1u < size) reader->buffer |= ((unsigned)reader->data[start + 1] << 8u);
+    if(start + 2u < size) reader->buffer |= ((unsigned)reader->data[start + 2] << 16u);
+    reader->buffer >>= (reader->bp & 7u);
+    return reader->bp + nbits <= reader->bitsize;
+  }
+}
+
+/*See ensureBits documentation above. This one ensures up to 32 bits */
+static LODEPNG_INLINE unsigned ensureBits32(LodePNGBitReader* reader, size_t nbits) {
+  size_t start = reader->bp >> 3u;
+  size_t size = reader->size;
+  if(start + 4u < size) {
+    reader->buffer = (unsigned)reader->data[start + 0] | ((unsigned)reader->data[start + 1] << 8u) |
+                     ((unsigned)reader->data[start + 2] << 16u) | ((unsigned)reader->data[start + 3] << 24u);
+    reader->buffer >>= (reader->bp & 7u);
+    reader->buffer |= (((unsigned)reader->data[start + 4] << 24u) << (8u - (reader->bp & 7u)));
+    return 1;
+  } else {
+    reader->buffer = 0;
+    if(start + 0u < size) reader->buffer |= reader->data[start + 0];
+    if(start + 1u < size) reader->buffer |= ((unsigned)reader->data[start + 1] << 8u);
+    if(start + 2u < size) reader->buffer |= ((unsigned)reader->data[start + 2] << 16u);
+    if(start + 3u < size) reader->buffer |= ((unsigned)reader->data[start + 3] << 24u);
+    reader->buffer >>= (reader->bp & 7u);
+    return reader->bp + nbits <= reader->bitsize;
+  }
+}
+
+/* Get bits without advancing the bit pointer. Must have enough bits available with ensureBits. Max nbits is 31. */
+static unsigned peekBits(LodePNGBitReader* reader, size_t nbits) {
+  /* The shift allows nbits to be only up to 31. */
+  return reader->buffer & ((1u << nbits) - 1u);
+}
+
+/* Must have enough bits available with ensureBits */
+static void advanceBits(LodePNGBitReader* reader, size_t nbits) {
+  reader->buffer >>= nbits;
+  reader->bp += nbits;
+}
+
+/* Must have enough bits available with ensureBits */
+static unsigned readBits(LodePNGBitReader* reader, size_t nbits) {
+  unsigned result = peekBits(reader, nbits);
+  advanceBits(reader, nbits);
+  return result;
+}
+
+/* Public for testing only. steps and result must have numsteps values. */
+unsigned lode_png_test_bitreader(const unsigned char* data, size_t size,
+                                 size_t numsteps, const size_t* steps, unsigned* result) {
+  size_t i;
+  LodePNGBitReader reader;
+  unsigned error = LodePNGBitReader_init(&reader, data, size);
+  if(error) return 0;
+  for(i = 0; i < numsteps; i++) {
+    size_t step = steps[i];
+    unsigned ok;
+    if(step > 25) ok = ensureBits32(&reader, step);
+    else if(step > 17) ok = ensureBits25(&reader, step);
+    else if(step > 9) ok = ensureBits17(&reader, step);
+    else ok = ensureBits9(&reader, step);
+    if(!ok) return 0;
+    result[i] = readBits(&reader, step);
+  }
+  return 1;
+}
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+static unsigned reverseBits(unsigned bits, unsigned num) {
+  /*TODO: implement faster lookup table based version when needed*/
+  unsigned i, result = 0;
+  for(i = 0; i < num; i++) result |= ((bits >> (num - i - 1u)) & 1u) << i;
+  return result;
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Deflate - Huffman                                                      / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#define FIRST_LENGTH_CODE_INDEX 257
+#define LAST_LENGTH_CODE_INDEX 285
+/*256 literals, the end code, some length codes, and 2 unused codes*/
+#define NUM_DEFLATE_CODE_SYMBOLS 288
+/*the distance codes have their own symbols, 30 used, 2 unused*/
+#define NUM_DISTANCE_SYMBOLS 32
+/*the code length codes. 0-15: code lengths, 16: copy previous 3-6 times, 17: 3-10 zeros, 18: 11-138 zeros*/
+#define NUM_CODE_LENGTH_CODES 19
+
+/*the base lengths represented by codes 257-285*/
+static const unsigned LENGTHBASE[29]
+  = {3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, 35, 43, 51, 59,
+     67, 83, 99, 115, 131, 163, 195, 227, 258};
+
+/*the extra bits used by codes 257-285 (added to base length)*/
+static const unsigned LENGTHEXTRA[29]
+  = {0, 0, 0, 0, 0, 0, 0,  0,  1,  1,  1,  1,  2,  2,  2,  2,  3,  3,  3,  3,
+      4,  4,  4,   4,   5,   5,   5,   5,   0};
+
+/*the base backwards distances (the bits of distance codes appear after length codes and use their own huffman tree)*/
+static const unsigned DISTANCEBASE[30]
+  = {1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, 257, 385, 513,
+     769, 1025, 1537, 2049, 3073, 4097, 6145, 8193, 12289, 16385, 24577};
+
+/*the extra bits of backwards distances (added to base)*/
+static const unsigned DISTANCEEXTRA[30]
+  = {0, 0, 0, 0, 1, 1, 2,  2,  3,  3,  4,  4,  5,  5,   6,   6,   7,   7,   8,
+       8,    9,    9,   10,   10,   11,   11,   12,    12,    13,    13};
+
+/*the order in which "code length alphabet code lengths" are stored as specified by deflate, out of this the huffman
+tree of the dynamic huffman tree lengths is generated*/
+static const unsigned CLCL_ORDER[NUM_CODE_LENGTH_CODES]
+  = {16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15};
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*
+Huffman tree struct, containing multiple representations of the tree
+*/
+typedef struct HuffmanTree {
+  unsigned* codes; /*the huffman codes (bit patterns representing the symbols)*/
+  unsigned* lengths; /*the lengths of the huffman codes*/
+  unsigned maxbitlen; /*maximum number of bits a single code can get*/
+  unsigned numcodes; /*number of symbols in the alphabet = number of codes*/
+  /* for reading only */
+  unsigned char* table_len; /*length of symbol from lookup table, or max length if secondary lookup needed*/
+  unsigned short* table_value; /*value of symbol from lookup table, or pointer to secondary table if needed*/
+} HuffmanTree;
+
+static void HuffmanTree_init(HuffmanTree* tree) {
+  tree->codes = 0;
+  tree->lengths = 0;
+  tree->table_len = 0;
+  tree->table_value = 0;
+}
+
+static void HuffmanTree_cleanup(HuffmanTree* tree) {
+  lodepng_free(tree->codes);
+  lodepng_free(tree->lengths);
+  lodepng_free(tree->table_len);
+  lodepng_free(tree->table_value);
+}
+
+/* amount of bits for first huffman table lookup (aka root bits), see HuffmanTree_makeTable and huffmanDecodeSymbol.*/
+/* values 8u and 9u work the fastest */
+#define FIRSTBITS 9u
+
+/* a symbol value too big to represent any valid symbol, to indicate reading disallowed huffman bits combination,
+which is possible in case of only 0 or 1 present symbols. */
+#define INVALIDSYMBOL 65535u
+
+/* make table for huffman decoding */
+static unsigned HuffmanTree_makeTable(HuffmanTree* tree) {
+  static const unsigned headsize = 1u << FIRSTBITS; /*size of the first table*/
+  static const unsigned mask = (1u << FIRSTBITS) /*headsize*/ - 1u;
+  size_t i, numpresent, pointer, size; /*total table size*/
+  unsigned* maxlens = (unsigned*)lodepng_malloc(headsize * sizeof(unsigned));
+  if(!maxlens) return 83; /*alloc fail*/
+
+  /* compute maxlens: max total bit length of symbols sharing prefix in the first table*/
+  lodepng_memset(maxlens, 0, headsize * sizeof(*maxlens));
+  for(i = 0; i < tree->numcodes; i++) {
+    unsigned symbol = tree->codes[i];
+    unsigned l = tree->lengths[i];
+    unsigned index;
+    if(l <= FIRSTBITS) continue; /*symbols that fit in first table don't increase secondary table size*/
+    /*get the FIRSTBITS MSBs, the MSBs of the symbol are encoded first. See later comment about the reversing*/
+    index = reverseBits(symbol >> (l - FIRSTBITS), FIRSTBITS);
+    maxlens[index] = LODEPNG_MAX(maxlens[index], l);
+  }
+  /* compute total table size: size of first table plus all secondary tables for symbols longer than FIRSTBITS */
+  size = headsize;
+  for(i = 0; i < headsize; ++i) {
+    unsigned l = maxlens[i];
+    if(l > FIRSTBITS) size += (1u << (l - FIRSTBITS));
+  }
+  tree->table_len = (unsigned char*)lodepng_malloc(size * sizeof(*tree->table_len));
+  tree->table_value = (unsigned short*)lodepng_malloc(size * sizeof(*tree->table_value));
+  if(!tree->table_len || !tree->table_value) {
+    lodepng_free(maxlens);
+    /* freeing tree->table values is done at a higher scope */
+    return 83; /*alloc fail*/
+  }
+  /*initialize with an invalid length to indicate unused entries*/
+  for(i = 0; i < size; ++i) tree->table_len[i] = 16;
+
+  /*fill in the first table for long symbols: max prefix size and pointer to secondary tables*/
+  pointer = headsize;
+  for(i = 0; i < headsize; ++i) {
+    unsigned l = maxlens[i];
+    if(l <= FIRSTBITS) continue;
+    tree->table_len[i] = l;
+    tree->table_value[i] = pointer;
+    pointer += (1u << (l - FIRSTBITS));
+  }
+  lodepng_free(maxlens);
+
+  /*fill in the first table for short symbols, or secondary table for long symbols*/
+  numpresent = 0;
+  for(i = 0; i < tree->numcodes; ++i) {
+    unsigned l = tree->lengths[i];
+    unsigned symbol = tree->codes[i]; /*the huffman bit pattern. i itself is the value.*/
+    /*reverse bits, because the huffman bits are given in MSB first order but the bit reader reads LSB first*/
+    unsigned reverse = reverseBits(symbol, l);
+    if(l == 0) continue;
+    numpresent++;
+
+    if(l <= FIRSTBITS) {
+      /*short symbol, fully in first table, replicated num times if l < FIRSTBITS*/
+      unsigned num = 1u << (FIRSTBITS - l);
+      unsigned j;
+      for(j = 0; j < num; ++j) {
+        /*bit reader will read the l bits of symbol first, the remaining FIRSTBITS - l bits go to the MSB's*/
+        unsigned index = reverse | (j << l);
+        if(tree->table_len[index] != 16) return 55; /*invalid tree: long symbol shares prefix with short symbol*/
+        tree->table_len[index] = l;
+        tree->table_value[index] = i;
+      }
+    } else {
+      /*long symbol, shares prefix with other long symbols in first lookup table, needs second lookup*/
+      /*the FIRSTBITS MSBs of the symbol are the first table index*/
+      unsigned index = reverse & mask;
+      unsigned maxlen = tree->table_len[index];
+      /*log2 of secondary table length, should be >= l - FIRSTBITS*/
+      unsigned tablelen = maxlen - FIRSTBITS;
+      unsigned start = tree->table_value[index]; /*starting index in secondary table*/
+      unsigned num = 1u << (tablelen - (l - FIRSTBITS)); /*amount of entries of this symbol in secondary table*/
+      unsigned j;
+      if(maxlen < l) return 55; /*invalid tree: long symbol shares prefix with short symbol*/
+      for(j = 0; j < num; ++j) {
+        unsigned reverse2 = reverse >> FIRSTBITS; /* l - FIRSTBITS bits */
+        unsigned index2 = start + (reverse2 | (j << (l - FIRSTBITS)));
+        tree->table_len[index2] = l;
+        tree->table_value[index2] = i;
+      }
+    }
+  }
+
+  if(numpresent < 2) {
+    /* In case of exactly 1 symbol, in theory the huffman symbol needs 0 bits,
+    but deflate uses 1 bit instead. In case of 0 symbols, no symbols can
+    appear at all, but such huffman tree could still exist (e.g. if distance
+    codes are never used). In both cases, not all symbols of the table will be
+    filled in. Fill them in with an invalid symbol value so returning them from
+    huffmanDecodeSymbol will cause error. */
+    for(i = 0; i < size; ++i) {
+      if(tree->table_len[i] == 16) {
+        /* As length, use a value smaller than FIRSTBITS for the head table,
+        and a value larger than FIRSTBITS for the secondary table, to ensure
+        valid behavior for advanceBits when reading this symbol. */
+        tree->table_len[i] = (i < headsize) ? 1 : (FIRSTBITS + 1);
+        tree->table_value[i] = INVALIDSYMBOL;
+      }
+    }
+  } else {
+    /* A good huffman tree has N * 2 - 1 nodes, of which N - 1 are internal nodes.
+    If that is not the case (due to too long length codes), the table will not
+    have been fully used, and this is an error (not all bit combinations can be
+    decoded): an oversubscribed huffman tree, indicated by error 55. */
+    for(i = 0; i < size; ++i) {
+      if(tree->table_len[i] == 16) return 55;
+    }
+  }
+
+  return 0;
+}
+
+/*
+Second step for the ...makeFromLengths and ...makeFromFrequencies functions.
+numcodes, lengths and maxbitlen must already be filled in correctly. return
+value is error.
+*/
+static unsigned HuffmanTree_makeFromLengths2(HuffmanTree* tree) {
+  unsigned* blcount;
+  unsigned* nextcode;
+  unsigned error = 0;
+  unsigned bits, n;
+
+  tree->codes = (unsigned*)lodepng_malloc(tree->numcodes * sizeof(unsigned));
+  blcount = (unsigned*)lodepng_malloc((tree->maxbitlen + 1) * sizeof(unsigned));
+  nextcode = (unsigned*)lodepng_malloc((tree->maxbitlen + 1) * sizeof(unsigned));
+  if(!tree->codes || !blcount || !nextcode) error = 83; /*alloc fail*/
+
+  if(!error) {
+    for(n = 0; n != tree->maxbitlen + 1; n++) blcount[n] = nextcode[n] = 0;
+    /*step 1: count number of instances of each code length*/
+    for(bits = 0; bits != tree->numcodes; ++bits) ++blcount[tree->lengths[bits]];
+    /*step 2: generate the nextcode values*/
+    for(bits = 1; bits <= tree->maxbitlen; ++bits) {
+      nextcode[bits] = (nextcode[bits - 1] + blcount[bits - 1]) << 1u;
+    }
+    /*step 3: generate all the codes*/
+    for(n = 0; n != tree->numcodes; ++n) {
+      if(tree->lengths[n] != 0) {
+        tree->codes[n] = nextcode[tree->lengths[n]]++;
+        /*remove superfluous bits from the code*/
+        tree->codes[n] &= ((1u << tree->lengths[n]) - 1u);
+      }
+    }
+  }
+
+  lodepng_free(blcount);
+  lodepng_free(nextcode);
+
+  if(!error) error = HuffmanTree_makeTable(tree);
+  return error;
+}
+
+/*
+given the code lengths (as stored in the PNG file), generate the tree as defined
+by Deflate. maxbitlen is the maximum bits that a code in the tree can have.
+return value is error.
+*/
+static unsigned HuffmanTree_makeFromLengths(HuffmanTree* tree, const unsigned* bitlen,
+                                            size_t numcodes, unsigned maxbitlen) {
+  unsigned i;
+  tree->lengths = (unsigned*)lodepng_malloc(numcodes * sizeof(unsigned));
+  if(!tree->lengths) return 83; /*alloc fail*/
+  for(i = 0; i != numcodes; ++i) tree->lengths[i] = bitlen[i];
+  tree->numcodes = (unsigned)numcodes; /*number of symbols*/
+  tree->maxbitlen = maxbitlen;
+  return HuffmanTree_makeFromLengths2(tree);
+}
+
+#ifdef LODEPNG_COMPILE_ENCODER
+
+/*BPM: Boundary Package Merge, see "A Fast and Space-Economical Algorithm for Length-Limited Coding",
+Jyrki Katajainen, Alistair Moffat, Andrew Turpin, 1995.*/
+
+/*chain node for boundary package merge*/
+typedef struct BPMNode {
+  int weight; /*the sum of all weights in this chain*/
+  unsigned index; /*index of this leaf node (called "count" in the paper)*/
+  struct BPMNode* tail; /*the next nodes in this chain (null if last)*/
+  int in_use;
+} BPMNode;
+
+/*lists of chains*/
+typedef struct BPMLists {
+  /*memory pool*/
+  unsigned memsize;
+  BPMNode* memory;
+  unsigned numfree;
+  unsigned nextfree;
+  BPMNode** freelist;
+  /*two heads of lookahead chains per list*/
+  unsigned listsize;
+  BPMNode** chains0;
+  BPMNode** chains1;
+} BPMLists;
+
+/*creates a new chain node with the given parameters, from the memory in the lists */
+static BPMNode* bpmnode_create(BPMLists* lists, int weight, unsigned index, BPMNode* tail) {
+  unsigned i;
+  BPMNode* result;
+
+  /*memory full, so garbage collect*/
+  if(lists->nextfree >= lists->numfree) {
+    /*mark only those that are in use*/
+    for(i = 0; i != lists->memsize; ++i) lists->memory[i].in_use = 0;
+    for(i = 0; i != lists->listsize; ++i) {
+      BPMNode* node;
+      for(node = lists->chains0[i]; node != 0; node = node->tail) node->in_use = 1;
+      for(node = lists->chains1[i]; node != 0; node = node->tail) node->in_use = 1;
+    }
+    /*collect those that are free*/
+    lists->numfree = 0;
+    for(i = 0; i != lists->memsize; ++i) {
+      if(!lists->memory[i].in_use) lists->freelist[lists->numfree++] = &lists->memory[i];
+    }
+    lists->nextfree = 0;
+  }
+
+  result = lists->freelist[lists->nextfree++];
+  result->weight = weight;
+  result->index = index;
+  result->tail = tail;
+  return result;
+}
+
+/*sort the leaves with stable mergesort*/
+static void bpmnode_sort(BPMNode* leaves, size_t num) {
+  BPMNode* mem = (BPMNode*)lodepng_malloc(sizeof(*leaves) * num);
+  size_t width, counter = 0;
+  for(width = 1; width < num; width *= 2) {
+    BPMNode* a = (counter & 1) ? mem : leaves;
+    BPMNode* b = (counter & 1) ? leaves : mem;
+    size_t p;
+    for(p = 0; p < num; p += 2 * width) {
+      size_t q = (p + width > num) ? num : (p + width);
+      size_t r = (p + 2 * width > num) ? num : (p + 2 * width);
+      size_t i = p, j = q, k;
+      for(k = p; k < r; k++) {
+        if(i < q && (j >= r || a[i].weight <= a[j].weight)) b[k] = a[i++];
+        else b[k] = a[j++];
+      }
+    }
+    counter++;
+  }
+  if(counter & 1) lodepng_memcpy(leaves, mem, sizeof(*leaves) * num);
+  lodepng_free(mem);
+}
+
+/*Boundary Package Merge step, numpresent is the amount of leaves, and c is the current chain.*/
+static void boundaryPM(BPMLists* lists, BPMNode* leaves, size_t numpresent, int c, int num) {
+  unsigned lastindex = lists->chains1[c]->index;
+
+  if(c == 0) {
+    if(lastindex >= numpresent) return;
+    lists->chains0[c] = lists->chains1[c];
+    lists->chains1[c] = bpmnode_create(lists, leaves[lastindex].weight, lastindex + 1, 0);
+  } else {
+    /*sum of the weights of the head nodes of the previous lookahead chains.*/
+    int sum = lists->chains0[c - 1]->weight + lists->chains1[c - 1]->weight;
+    lists->chains0[c] = lists->chains1[c];
+    if(lastindex < numpresent && sum > leaves[lastindex].weight) {
+      lists->chains1[c] = bpmnode_create(lists, leaves[lastindex].weight, lastindex + 1, lists->chains1[c]->tail);
+      return;
+    }
+    lists->chains1[c] = bpmnode_create(lists, sum, lastindex, lists->chains1[c - 1]);
+    /*in the end we are only interested in the chain of the last list, so no
+    need to recurse if we're at the last one (this gives measurable speedup)*/
+    if(num + 1 < (int)(2 * numpresent - 2)) {
+      boundaryPM(lists, leaves, numpresent, c - 1, num);
+      boundaryPM(lists, leaves, numpresent, c - 1, num);
+    }
+  }
+}
+
+unsigned lodepng_huffman_code_lengths(unsigned* lengths, const unsigned* frequencies,
+                                      size_t numcodes, unsigned maxbitlen) {
+  unsigned error = 0;
+  unsigned i;
+  size_t numpresent = 0; /*number of symbols with non-zero frequency*/
+  BPMNode* leaves; /*the symbols, only those with > 0 frequency*/
+
+  if(numcodes == 0) return 80; /*error: a tree of 0 symbols is not supposed to be made*/
+  if((1u << maxbitlen) < (unsigned)numcodes) return 80; /*error: represent all symbols*/
+
+  leaves = (BPMNode*)lodepng_malloc(numcodes * sizeof(*leaves));
+  if(!leaves) return 83; /*alloc fail*/
+
+  for(i = 0; i != numcodes; ++i) {
+    if(frequencies[i] > 0) {
+      leaves[numpresent].weight = (int)frequencies[i];
+      leaves[numpresent].index = i;
+      ++numpresent;
+    }
+  }
+
+  lodepng_memset(lengths, 0, numcodes * sizeof(*lengths));
+
+  /*ensure at least two present symbols. There should be at least one symbol
+  according to RFC 1951 section 3.2.7. Some decoders incorrectly require two. To
+  make these work as well ensure there are at least two symbols. The
+  Package-Merge code below also doesn't work correctly if there's only one
+  symbol, it'd give it the theoretical 0 bits but in practice zlib wants 1 bit*/
+  if(numpresent == 0) {
+    lengths[0] = lengths[1] = 1; /*note that for RFC 1951 section 3.2.7, only lengths[0] = 1 is needed*/
+  } else if(numpresent == 1) {
+    lengths[leaves[0].index] = 1;
+    lengths[leaves[0].index == 0 ? 1 : 0] = 1;
+  } else {
+    BPMLists lists;
+    BPMNode* node;
+
+    bpmnode_sort(leaves, numpresent);
+
+    lists.listsize = maxbitlen;
+    lists.memsize = 2 * maxbitlen * (maxbitlen + 1);
+    lists.nextfree = 0;
+    lists.numfree = lists.memsize;
+    lists.memory = (BPMNode*)lodepng_malloc(lists.memsize * sizeof(*lists.memory));
+    lists.freelist = (BPMNode**)lodepng_malloc(lists.memsize * sizeof(BPMNode*));
+    lists.chains0 = (BPMNode**)lodepng_malloc(lists.listsize * sizeof(BPMNode*));
+    lists.chains1 = (BPMNode**)lodepng_malloc(lists.listsize * sizeof(BPMNode*));
+    if(!lists.memory || !lists.freelist || !lists.chains0 || !lists.chains1) error = 83; /*alloc fail*/
+
+    if(!error) {
+      for(i = 0; i != lists.memsize; ++i) lists.freelist[i] = &lists.memory[i];
+
+      bpmnode_create(&lists, leaves[0].weight, 1, 0);
+      bpmnode_create(&lists, leaves[1].weight, 2, 0);
+
+      for(i = 0; i != lists.listsize; ++i) {
+        lists.chains0[i] = &lists.memory[0];
+        lists.chains1[i] = &lists.memory[1];
+      }
+
+      /*each boundaryPM call adds one chain to the last list, and we need 2 * numpresent - 2 chains.*/
+      for(i = 2; i != 2 * numpresent - 2; ++i) boundaryPM(&lists, leaves, numpresent, (int)maxbitlen - 1, (int)i);
+
+      for(node = lists.chains1[maxbitlen - 1]; node; node = node->tail) {
+        for(i = 0; i != node->index; ++i) ++lengths[leaves[i].index];
+      }
+    }
+
+    lodepng_free(lists.memory);
+    lodepng_free(lists.freelist);
+    lodepng_free(lists.chains0);
+    lodepng_free(lists.chains1);
+  }
+
+  lodepng_free(leaves);
+  return error;
+}
+
+/*Create the Huffman tree given the symbol frequencies*/
+static unsigned HuffmanTree_makeFromFrequencies(HuffmanTree* tree, const unsigned* frequencies,
+                                                size_t mincodes, size_t numcodes, unsigned maxbitlen) {
+  unsigned error = 0;
+  while(!frequencies[numcodes - 1] && numcodes > mincodes) --numcodes; /*trim zeroes*/
+  tree->lengths = (unsigned*)lodepng_malloc(numcodes * sizeof(unsigned));
+  if(!tree->lengths) return 83; /*alloc fail*/
+  tree->maxbitlen = maxbitlen;
+  tree->numcodes = (unsigned)numcodes; /*number of symbols*/
+
+  error = lodepng_huffman_code_lengths(tree->lengths, frequencies, numcodes, maxbitlen);
+  if(!error) error = HuffmanTree_makeFromLengths2(tree);
+  return error;
+}
+#endif /*LODEPNG_COMPILE_ENCODER*/
+
+/*get the literal and length code tree of a deflated block with fixed tree, as per the deflate specification*/
+static unsigned generateFixedLitLenTree(HuffmanTree* tree) {
+  unsigned i, error = 0;
+  unsigned* bitlen = (unsigned*)lodepng_malloc(NUM_DEFLATE_CODE_SYMBOLS * sizeof(unsigned));
+  if(!bitlen) return 83; /*alloc fail*/
+
+  /*288 possible codes: 0-255=literals, 256=endcode, 257-285=lengthcodes, 286-287=unused*/
+  for(i =   0; i <= 143; ++i) bitlen[i] = 8;
+  for(i = 144; i <= 255; ++i) bitlen[i] = 9;
+  for(i = 256; i <= 279; ++i) bitlen[i] = 7;
+  for(i = 280; i <= 287; ++i) bitlen[i] = 8;
+
+  error = HuffmanTree_makeFromLengths(tree, bitlen, NUM_DEFLATE_CODE_SYMBOLS, 15);
+
+  lodepng_free(bitlen);
+  return error;
+}
+
+/*get the distance code tree of a deflated block with fixed tree, as specified in the deflate specification*/
+static unsigned generateFixedDistanceTree(HuffmanTree* tree) {
+  unsigned i, error = 0;
+  unsigned* bitlen = (unsigned*)lodepng_malloc(NUM_DISTANCE_SYMBOLS * sizeof(unsigned));
+  if(!bitlen) return 83; /*alloc fail*/
+
+  /*there are 32 distance codes, but 30-31 are unused*/
+  for(i = 0; i != NUM_DISTANCE_SYMBOLS; ++i) bitlen[i] = 5;
+  error = HuffmanTree_makeFromLengths(tree, bitlen, NUM_DISTANCE_SYMBOLS, 15);
+
+  lodepng_free(bitlen);
+  return error;
+}
+
+#ifdef LODEPNG_COMPILE_DECODER
+
+/*
+returns the code. The bit reader must already have been ensured at least 15 bits
+*/
+static unsigned huffmanDecodeSymbol(LodePNGBitReader* reader, const HuffmanTree* codetree) {
+  unsigned short code = peekBits(reader, FIRSTBITS);
+  unsigned short l = codetree->table_len[code];
+  unsigned short value = codetree->table_value[code];
+  if(l <= FIRSTBITS) {
+    advanceBits(reader, l);
+    return value;
+  } else {
+    unsigned index2;
+    advanceBits(reader, FIRSTBITS);
+    index2 = value + peekBits(reader, l - FIRSTBITS);
+    advanceBits(reader, codetree->table_len[index2] - FIRSTBITS);
+    return codetree->table_value[index2];
+  }
+}
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+#ifdef LODEPNG_COMPILE_DECODER
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Inflator (Decompressor)                                                / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*get the tree of a deflated block with fixed tree, as specified in the deflate specification
+Returns error code.*/
+static unsigned getTreeInflateFixed(HuffmanTree* tree_ll, HuffmanTree* tree_d) {
+  unsigned error = generateFixedLitLenTree(tree_ll);
+  if(error) return error;
+  return generateFixedDistanceTree(tree_d);
+}
+
+/*get the tree of a deflated block with dynamic tree, the tree itself is also Huffman compressed with a known tree*/
+static unsigned getTreeInflateDynamic(HuffmanTree* tree_ll, HuffmanTree* tree_d,
+                                      LodePNGBitReader* reader) {
+  /*make sure that length values that aren't filled in will be 0, or a wrong tree will be generated*/
+  unsigned error = 0;
+  unsigned n, HLIT, HDIST, HCLEN, i;
+
+  /*see comments in deflateDynamic for explanation of the context and these variables, it is analogous*/
+  unsigned* bitlen_ll = 0; /*lit,len code lengths*/
+  unsigned* bitlen_d = 0; /*dist code lengths*/
+  /*code length code lengths ("clcl"), the bit lengths of the huffman tree used to compress bitlen_ll and bitlen_d*/
+  unsigned* bitlen_cl = 0;
+  HuffmanTree tree_cl; /*the code tree for code length codes (the huffman tree for compressed huffman trees)*/
+
+  if(!ensureBits17(reader, 14)) return 49; /*error: the bit pointer is or will go past the memory*/
+
+  /*number of literal/length codes + 257. Unlike the spec, the value 257 is added to it here already*/
+  HLIT =  readBits(reader, 5) + 257;
+  /*number of distance codes. Unlike the spec, the value 1 is added to it here already*/
+  HDIST = readBits(reader, 5) + 1;
+  /*number of code length codes. Unlike the spec, the value 4 is added to it here already*/
+  HCLEN = readBits(reader, 4) + 4;
+
+  bitlen_cl = (unsigned*)lodepng_malloc(NUM_CODE_LENGTH_CODES * sizeof(unsigned));
+  if(!bitlen_cl) return 83 /*alloc fail*/;
+
+  HuffmanTree_init(&tree_cl);
+
+  while(!error) {
+    /*read the code length codes out of 3 * (amount of code length codes) bits*/
+    if(lodepng_gtofl(reader->bp, HCLEN * 3, reader->bitsize)) {
+      ERROR_BREAK(50); /*error: the bit pointer is or will go past the memory*/
+    }
+    for(i = 0; i != HCLEN; ++i) {
+      ensureBits9(reader, 3); /*out of bounds already checked above */
+      bitlen_cl[CLCL_ORDER[i]] = readBits(reader, 3);
+    }
+    for(i = HCLEN; i != NUM_CODE_LENGTH_CODES; ++i) {
+      bitlen_cl[CLCL_ORDER[i]] = 0;
+    }
+
+    error = HuffmanTree_makeFromLengths(&tree_cl, bitlen_cl, NUM_CODE_LENGTH_CODES, 7);
+    if(error) break;
+
+    /*now we can use this tree to read the lengths for the tree that this function will return*/
+    bitlen_ll = (unsigned*)lodepng_malloc(NUM_DEFLATE_CODE_SYMBOLS * sizeof(unsigned));
+    bitlen_d = (unsigned*)lodepng_malloc(NUM_DISTANCE_SYMBOLS * sizeof(unsigned));
+    if(!bitlen_ll || !bitlen_d) ERROR_BREAK(83 /*alloc fail*/);
+    lodepng_memset(bitlen_ll, 0, NUM_DEFLATE_CODE_SYMBOLS * sizeof(*bitlen_ll));
+    lodepng_memset(bitlen_d, 0, NUM_DISTANCE_SYMBOLS * sizeof(*bitlen_d));
+
+    /*i is the current symbol we're reading in the part that contains the code lengths of lit/len and dist codes*/
+    i = 0;
+    while(i < HLIT + HDIST) {
+      unsigned code;
+      ensureBits25(reader, 22); /* up to 15 bits for huffman code, up to 7 extra bits below*/
+      code = huffmanDecodeSymbol(reader, &tree_cl);
+      if(code <= 15) /*a length code*/ {
+        if(i < HLIT) bitlen_ll[i] = code;
+        else bitlen_d[i - HLIT] = code;
+        ++i;
+      } else if(code == 16) /*repeat previous*/ {
+        unsigned replength = 3; /*read in the 2 bits that indicate repeat length (3-6)*/
+        unsigned value; /*set value to the previous code*/
+
+        if(i == 0) ERROR_BREAK(54); /*can't repeat previous if i is 0*/
+
+        replength += readBits(reader, 2);
+
+        if(i < HLIT + 1) value = bitlen_ll[i - 1];
+        else value = bitlen_d[i - HLIT - 1];
+        /*repeat this value in the next lengths*/
+        for(n = 0; n < replength; ++n) {
+          if(i >= HLIT + HDIST) ERROR_BREAK(13); /*error: i is larger than the amount of codes*/
+          if(i < HLIT) bitlen_ll[i] = value;
+          else bitlen_d[i - HLIT] = value;
+          ++i;
+        }
+      } else if(code == 17) /*repeat "0" 3-10 times*/ {
+        unsigned replength = 3; /*read in the bits that indicate repeat length*/
+        replength += readBits(reader, 3);
+
+        /*repeat this value in the next lengths*/
+        for(n = 0; n < replength; ++n) {
+          if(i >= HLIT + HDIST) ERROR_BREAK(14); /*error: i is larger than the amount of codes*/
+
+          if(i < HLIT) bitlen_ll[i] = 0;
+          else bitlen_d[i - HLIT] = 0;
+          ++i;
+        }
+      } else if(code == 18) /*repeat "0" 11-138 times*/ {
+        unsigned replength = 11; /*read in the bits that indicate repeat length*/
+        replength += readBits(reader, 7);
+
+        /*repeat this value in the next lengths*/
+        for(n = 0; n < replength; ++n) {
+          if(i >= HLIT + HDIST) ERROR_BREAK(15); /*error: i is larger than the amount of codes*/
+
+          if(i < HLIT) bitlen_ll[i] = 0;
+          else bitlen_d[i - HLIT] = 0;
+          ++i;
+        }
+      } else /*if(code == INVALIDSYMBOL)*/ {
+        ERROR_BREAK(16); /*error: tried to read disallowed huffman symbol*/
+      }
+      /*check if any of the ensureBits above went out of bounds*/
+      if(reader->bp > reader->bitsize) {
+        /*return error code 10 or 11 depending on the situation that happened in huffmanDecodeSymbol
+        (10=no endcode, 11=wrong jump outside of tree)*/
+        /* TODO: revise error codes 10,11,50: the above comment is no longer valid */
+        ERROR_BREAK(50); /*error, bit pointer jumps past memory*/
+      }
+    }
+    if(error) break;
+
+    if(bitlen_ll[256] == 0) ERROR_BREAK(64); /*the length of the end code 256 must be larger than 0*/
+
+    /*now we've finally got HLIT and HDIST, so generate the code trees, and the function is done*/
+    error = HuffmanTree_makeFromLengths(tree_ll, bitlen_ll, NUM_DEFLATE_CODE_SYMBOLS, 15);
+    if(error) break;
+    error = HuffmanTree_makeFromLengths(tree_d, bitlen_d, NUM_DISTANCE_SYMBOLS, 15);
+
+    break; /*end of error-while*/
+  }
+
+  lodepng_free(bitlen_cl);
+  lodepng_free(bitlen_ll);
+  lodepng_free(bitlen_d);
+  HuffmanTree_cleanup(&tree_cl);
+
+  return error;
+}
+
+/*inflate a block with dynamic of fixed Huffman tree. btype must be 1 or 2.*/
+static unsigned inflateHuffmanBlock(ucvector* out, LodePNGBitReader* reader,
+                                    unsigned btype) {
+  unsigned error = 0;
+  HuffmanTree tree_ll; /*the huffman tree for literal and length codes*/
+  HuffmanTree tree_d; /*the huffman tree for distance codes*/
+
+  HuffmanTree_init(&tree_ll);
+  HuffmanTree_init(&tree_d);
+
+  if(btype == 1) error = getTreeInflateFixed(&tree_ll, &tree_d);
+  else /*if(btype == 2)*/ error = getTreeInflateDynamic(&tree_ll, &tree_d, reader);
+
+  while(!error) /*decode all symbols until end reached, breaks at end code*/ {
+    /*code_ll is literal, length or end code*/
+    unsigned code_ll;
+    ensureBits25(reader, 20); /* up to 15 for the huffman symbol, up to 5 for the length extra bits */
+    code_ll = huffmanDecodeSymbol(reader, &tree_ll);
+    if(code_ll <= 255) /*literal symbol*/ {
+      if(!ucvector_resize(out, out->size + 1)) ERROR_BREAK(83 /*alloc fail*/);
+      out->data[out->size - 1] = (unsigned char)code_ll;
+    } else if(code_ll >= FIRST_LENGTH_CODE_INDEX && code_ll <= LAST_LENGTH_CODE_INDEX) /*length code*/ {
+      unsigned code_d, distance;
+      unsigned numextrabits_l, numextrabits_d; /*extra bits for length and distance*/
+      size_t start, backward, length;
+
+      /*part 1: get length base*/
+      length = LENGTHBASE[code_ll - FIRST_LENGTH_CODE_INDEX];
+
+      /*part 2: get extra bits and add the value of that to length*/
+      numextrabits_l = LENGTHEXTRA[code_ll - FIRST_LENGTH_CODE_INDEX];
+      if(numextrabits_l != 0) {
+        /* bits already ensured above */
+        length += readBits(reader, numextrabits_l);
+      }
+
+      /*part 3: get distance code*/
+      ensureBits32(reader, 28); /* up to 15 for the huffman symbol, up to 13 for the extra bits */
+      code_d = huffmanDecodeSymbol(reader, &tree_d);
+      if(code_d > 29) {
+        if(code_d <= 31) {
+          ERROR_BREAK(18); /*error: invalid distance code (30-31 are never used)*/
+        } else /* if(code_d == INVALIDSYMBOL) */{
+          ERROR_BREAK(16); /*error: tried to read disallowed huffman symbol*/
+        }
+      }
+      distance = DISTANCEBASE[code_d];
+
+      /*part 4: get extra bits from distance*/
+      numextrabits_d = DISTANCEEXTRA[code_d];
+      if(numextrabits_d != 0) {
+        /* bits already ensured above */
+        distance += readBits(reader, numextrabits_d);
+      }
+
+      /*part 5: fill in all the out[n] values based on the length and dist*/
+      start = out->size;
+      if(distance > start) ERROR_BREAK(52); /*too long backward distance*/
+      backward = start - distance;
+
+      if(!ucvector_resize(out, out->size + length)) ERROR_BREAK(83 /*alloc fail*/);
+      if(distance < length) {
+        size_t forward;
+        lodepng_memcpy(out->data + start, out->data + backward, distance);
+        start += distance;
+        for(forward = distance; forward < length; ++forward) {
+          out->data[start++] = out->data[backward++];
+        }
+      } else {
+        lodepng_memcpy(out->data + start, out->data + backward, length);
+      }
+    } else if(code_ll == 256) {
+      break; /*end code, break the loop*/
+    } else /*if(code_ll == INVALIDSYMBOL)*/ {
+      ERROR_BREAK(16); /*error: tried to read disallowed huffman symbol*/
+    }
+    /*check if any of the ensureBits above went out of bounds*/
+    if(reader->bp > reader->bitsize) {
+      /*return error code 10 or 11 depending on the situation that happened in huffmanDecodeSymbol
+      (10=no endcode, 11=wrong jump outside of tree)*/
+      /* TODO: revise error codes 10,11,50: the above comment is no longer valid */
+      ERROR_BREAK(51); /*error, bit pointer jumps past memory*/
+    }
+  }
+
+  HuffmanTree_cleanup(&tree_ll);
+  HuffmanTree_cleanup(&tree_d);
+
+  return error;
+}
+
+static unsigned inflateNoCompression(ucvector* out, LodePNGBitReader* reader,
+                                     const LodePNGDecompressSettings* settings) {
+  size_t bytepos;
+  size_t size = reader->size;
+  unsigned LEN, NLEN, error = 0;
+
+  /*go to first boundary of byte*/
+  bytepos = (reader->bp + 7u) >> 3u;
+
+  /*read LEN (2 bytes) and NLEN (2 bytes)*/
+  if(bytepos + 4 >= size) return 52; /*error, bit pointer will jump past memory*/
+  LEN = (unsigned)reader->data[bytepos] + ((unsigned)reader->data[bytepos + 1] << 8u); bytepos += 2;
+  NLEN = (unsigned)reader->data[bytepos] + ((unsigned)reader->data[bytepos + 1] << 8u); bytepos += 2;
+
+  /*check if 16-bit NLEN is really the one's complement of LEN*/
+  if(!settings->ignore_nlen && LEN + NLEN != 65535) {
+    return 21; /*error: NLEN is not one's complement of LEN*/
+  }
+
+  if(!ucvector_resize(out, out->size + LEN)) return 83; /*alloc fail*/
+
+  /*read the literal data: LEN bytes are now stored in the out buffer*/
+  if(bytepos + LEN > size) return 23; /*error: reading outside of in buffer*/
+
+  lodepng_memcpy(out->data + out->size - LEN, reader->data + bytepos, LEN);
+  bytepos += LEN;
+
+  reader->bp = bytepos << 3u;
+
+  return error;
+}
+
+static unsigned lodepng_inflatev(ucvector* out,
+                                 const unsigned char* in, size_t insize,
+                                 const LodePNGDecompressSettings* settings) {
+  unsigned BFINAL = 0;
+  LodePNGBitReader reader;
+  unsigned error = LodePNGBitReader_init(&reader, in, insize);
+
+  if(error) return error;
+
+  while(!BFINAL) {
+    unsigned BTYPE;
+    if(!ensureBits9(&reader, 3)) return 52; /*error, bit pointer will jump past memory*/
+    BFINAL = readBits(&reader, 1);
+    BTYPE = readBits(&reader, 2);
+
+    if(BTYPE == 3) return 20; /*error: invalid BTYPE*/
+    else if(BTYPE == 0) error = inflateNoCompression(out, &reader, settings); /*no compression*/
+    else error = inflateHuffmanBlock(out, &reader, BTYPE); /*compression, BTYPE 01 or 10*/
+
+    if(error) return error;
+  }
+
+  return error;
+}
+
+unsigned lodepng_inflate(unsigned char** out, size_t* outsize,
+                         const unsigned char* in, size_t insize,
+                         const LodePNGDecompressSettings* settings) {
+  ucvector v = ucvector_init(*out, *outsize);
+  unsigned error = lodepng_inflatev(&v, in, insize, settings);
+  *out = v.data;
+  *outsize = v.size;
+  return error;
+}
+
+static unsigned inflatev(ucvector* out, const unsigned char* in, size_t insize,
+                        const LodePNGDecompressSettings* settings) {
+  if(settings->custom_inflate) {
+    unsigned error = settings->custom_inflate(&out->data, &out->size, in, insize, settings);
+    out->allocsize = out->size;
+    return error;
+  } else {
+    return lodepng_inflatev(out, in, insize, settings);
+  }
+}
+
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+#ifdef LODEPNG_COMPILE_ENCODER
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Deflator (Compressor)                                                  / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+static const size_t MAX_SUPPORTED_DEFLATE_LENGTH = 258;
+
+/*search the index in the array, that has the largest value smaller than or equal to the given value,
+given array must be sorted (if no value is smaller, it returns the size of the given array)*/
+static size_t searchCodeIndex(const unsigned* array, size_t array_size, size_t value) {
+  /*binary search (only small gain over linear). TODO: use CPU log2 instruction for getting symbols instead*/
+  size_t left = 1;
+  size_t right = array_size - 1;
+
+  while(left <= right) {
+    size_t mid = (left + right) >> 1;
+    if(array[mid] >= value) right = mid - 1;
+    else left = mid + 1;
+  }
+  if(left >= array_size || array[left] > value) left--;
+  return left;
+}
+
+static void addLengthDistance(uivector* values, size_t length, size_t distance) {
+  /*values in encoded vector are those used by deflate:
+  0-255: literal bytes
+  256: end
+  257-285: length/distance pair (length code, followed by extra length bits, distance code, extra distance bits)
+  286-287: invalid*/
+
+  unsigned length_code = (unsigned)searchCodeIndex(LENGTHBASE, 29, length);
+  unsigned extra_length = (unsigned)(length - LENGTHBASE[length_code]);
+  unsigned dist_code = (unsigned)searchCodeIndex(DISTANCEBASE, 30, distance);
+  unsigned extra_distance = (unsigned)(distance - DISTANCEBASE[dist_code]);
+
+  size_t pos = values->size;
+  /*TODO: return error when this fails (out of memory)*/
+  unsigned ok = uivector_resize(values, values->size + 4);
+  if(ok) {
+    values->data[pos + 0] = length_code + FIRST_LENGTH_CODE_INDEX;
+    values->data[pos + 1] = extra_length;
+    values->data[pos + 2] = dist_code;
+    values->data[pos + 3] = extra_distance;
+  }
+}
+
+/*3 bytes of data get encoded into two bytes. The hash cannot use more than 3
+bytes as input because 3 is the minimum match length for deflate*/
+static const unsigned HASH_NUM_VALUES = 65536;
+static const unsigned HASH_BIT_MASK = 65535; /*HASH_NUM_VALUES - 1, but C90 does not like that as initializer*/
+
+typedef struct Hash {
+  int* head; /*hash value to head circular pos - can be outdated if went around window*/
+  /*circular pos to prev circular pos*/
+  unsigned short* chain;
+  int* val; /*circular pos to hash value*/
+
+  /*TODO: do this not only for zeros but for any repeated byte. However for PNG
+  it's always going to be the zeros that dominate, so not important for PNG*/
+  int* headz; /*similar to head, but for chainz*/
+  unsigned short* chainz; /*those with same amount of zeros*/
+  unsigned short* zeros; /*length of zeros streak, used as a second hash chain*/
+} Hash;
+
+static unsigned hash_init(Hash* hash, unsigned windowsize) {
+  unsigned i;
+  hash->head = (int*)lodepng_malloc(sizeof(int) * HASH_NUM_VALUES);
+  hash->val = (int*)lodepng_malloc(sizeof(int) * windowsize);
+  hash->chain = (unsigned short*)lodepng_malloc(sizeof(unsigned short) * windowsize);
+
+  hash->zeros = (unsigned short*)lodepng_malloc(sizeof(unsigned short) * windowsize);
+  hash->headz = (int*)lodepng_malloc(sizeof(int) * (MAX_SUPPORTED_DEFLATE_LENGTH + 1));
+  hash->chainz = (unsigned short*)lodepng_malloc(sizeof(unsigned short) * windowsize);
+
+  if(!hash->head || !hash->chain || !hash->val  || !hash->headz|| !hash->chainz || !hash->zeros) {
+    return 83; /*alloc fail*/
+  }
+
+  /*initialize hash table*/
+  for(i = 0; i != HASH_NUM_VALUES; ++i) hash->head[i] = -1;
+  for(i = 0; i != windowsize; ++i) hash->val[i] = -1;
+  for(i = 0; i != windowsize; ++i) hash->chain[i] = i; /*same value as index indicates uninitialized*/
+
+  for(i = 0; i <= MAX_SUPPORTED_DEFLATE_LENGTH; ++i) hash->headz[i] = -1;
+  for(i = 0; i != windowsize; ++i) hash->chainz[i] = i; /*same value as index indicates uninitialized*/
+
+  return 0;
+}
+
+static void hash_cleanup(Hash* hash) {
+  lodepng_free(hash->head);
+  lodepng_free(hash->val);
+  lodepng_free(hash->chain);
+
+  lodepng_free(hash->zeros);
+  lodepng_free(hash->headz);
+  lodepng_free(hash->chainz);
+}
+
+
+
+static unsigned getHash(const unsigned char* data, size_t size, size_t pos) {
+  unsigned result = 0;
+  if(pos + 2 < size) {
+    /*A simple shift and xor hash is used. Since the data of PNGs is dominated
+    by zeroes due to the filters, a better hash does not have a significant
+    effect on speed in traversing the chain, and causes more time spend on
+    calculating the hash.*/
+    result ^= ((unsigned)data[pos + 0] << 0u);
+    result ^= ((unsigned)data[pos + 1] << 4u);
+    result ^= ((unsigned)data[pos + 2] << 8u);
+  } else {
+    size_t amount, i;
+    if(pos >= size) return 0;
+    amount = size - pos;
+    for(i = 0; i != amount; ++i) result ^= ((unsigned)data[pos + i] << (i * 8u));
+  }
+  return result & HASH_BIT_MASK;
+}
+
+static unsigned countZeros(const unsigned char* data, size_t size, size_t pos) {
+  const unsigned char* start = data + pos;
+  const unsigned char* end = start + MAX_SUPPORTED_DEFLATE_LENGTH;
+  if(end > data + size) end = data + size;
+  data = start;
+  while(data != end && *data == 0) ++data;
+  /*subtracting two addresses returned as 32-bit number (max value is MAX_SUPPORTED_DEFLATE_LENGTH)*/
+  return (unsigned)(data - start);
+}
+
+/*wpos = pos & (windowsize - 1)*/
+static void updateHashChain(Hash* hash, size_t wpos, unsigned hashval, unsigned short numzeros) {
+  hash->val[wpos] = (int)hashval;
+  if(hash->head[hashval] != -1) hash->chain[wpos] = hash->head[hashval];
+  hash->head[hashval] = (int)wpos;
+
+  hash->zeros[wpos] = numzeros;
+  if(hash->headz[numzeros] != -1) hash->chainz[wpos] = hash->headz[numzeros];
+  hash->headz[numzeros] = (int)wpos;
+}
+
+/*
+LZ77-encode the data. Return value is error code. The input are raw bytes, the output
+is in the form of unsigned integers with codes representing for example literal bytes, or
+length/distance pairs.
+It uses a hash table technique to let it encode faster. When doing LZ77 encoding, a
+sliding window (of windowsize) is used, and all past bytes in that window can be used as
+the "dictionary". A brute force search through all possible distances would be slow, and
+this hash technique is one out of several ways to speed this up.
+*/
+static unsigned encodeLZ77(uivector* out, Hash* hash,
+                           const unsigned char* in, size_t inpos, size_t insize, unsigned windowsize,
+                           unsigned minmatch, unsigned nicematch, unsigned lazymatching) {
+  size_t pos;
+  unsigned i, error = 0;
+  /*for large window lengths, assume the user wants no compression loss. Otherwise, max hash chain length speedup.*/
+  unsigned maxchainlength = windowsize >= 8192 ? windowsize : windowsize / 8u;
+  unsigned maxlazymatch = windowsize >= 8192 ? MAX_SUPPORTED_DEFLATE_LENGTH : 64;
+
+  unsigned usezeros = 1; /*not sure if setting it to false for windowsize < 8192 is better or worse*/
+  unsigned numzeros = 0;
+
+  unsigned offset; /*the offset represents the distance in LZ77 terminology*/
+  unsigned length;
+  unsigned lazy = 0;
+  unsigned lazylength = 0, lazyoffset = 0;
+  unsigned hashval;
+  unsigned current_offset, current_length;
+  unsigned prev_offset;
+  const unsigned char *lastptr, *foreptr, *backptr;
+  unsigned hashpos;
+
+  if(windowsize == 0 || windowsize > 32768) return 60; /*error: windowsize smaller/larger than allowed*/
+  if((windowsize & (windowsize - 1)) != 0) return 90; /*error: must be power of two*/
+
+  if(nicematch > MAX_SUPPORTED_DEFLATE_LENGTH) nicematch = MAX_SUPPORTED_DEFLATE_LENGTH;
+
+  for(pos = inpos; pos < insize; ++pos) {
+    size_t wpos = pos & (windowsize - 1); /*position for in 'circular' hash buffers*/
+    unsigned chainlength = 0;
+
+    hashval = getHash(in, insize, pos);
+
+    if(usezeros && hashval == 0) {
+      if(numzeros == 0) numzeros = countZeros(in, insize, pos);
+      else if(pos + numzeros > insize || in[pos + numzeros - 1] != 0) --numzeros;
+    } else {
+      numzeros = 0;
+    }
+
+    updateHashChain(hash, wpos, hashval, numzeros);
+
+    /*the length and offset found for the current position*/
+    length = 0;
+    offset = 0;
+
+    hashpos = hash->chain[wpos];
+
+    lastptr = &in[insize < pos + MAX_SUPPORTED_DEFLATE_LENGTH ? insize : pos + MAX_SUPPORTED_DEFLATE_LENGTH];
+
+    /*search for the longest string*/
+    prev_offset = 0;
+    for(;;) {
+      if(chainlength++ >= maxchainlength) break;
+      current_offset = (unsigned)(hashpos <= wpos ? wpos - hashpos : wpos - hashpos + windowsize);
+
+      if(current_offset < prev_offset) break; /*stop when went completely around the circular buffer*/
+      prev_offset = current_offset;
+      if(current_offset > 0) {
+        /*test the next characters*/
+        foreptr = &in[pos];
+        backptr = &in[pos - current_offset];
+
+        /*common case in PNGs is lots of zeros. Quickly skip over them as a speedup*/
+        if(numzeros >= 3) {
+          unsigned skip = hash->zeros[hashpos];
+          if(skip > numzeros) skip = numzeros;
+          backptr += skip;
+          foreptr += skip;
+        }
+
+        while(foreptr != lastptr && *backptr == *foreptr) /*maximum supported length by deflate is max length*/ {
+          ++backptr;
+          ++foreptr;
+        }
+        current_length = (unsigned)(foreptr - &in[pos]);
+
+        if(current_length > length) {
+          length = current_length; /*the longest length*/
+          offset = current_offset; /*the offset that is related to this longest length*/
+          /*jump out once a length of max length is found (speed gain). This also jumps
+          out if length is MAX_SUPPORTED_DEFLATE_LENGTH*/
+          if(current_length >= nicematch) break;
+        }
+      }
+
+      if(hashpos == hash->chain[hashpos]) break;
+
+      if(numzeros >= 3 && length > numzeros) {
+        hashpos = hash->chainz[hashpos];
+        if(hash->zeros[hashpos] != numzeros) break;
+      } else {
+        hashpos = hash->chain[hashpos];
+        /*outdated hash value, happens if particular value was not encountered in whole last window*/
+        if(hash->val[hashpos] != (int)hashval) break;
+      }
+    }
+
+    if(lazymatching) {
+      if(!lazy && length >= 3 && length <= maxlazymatch && length < MAX_SUPPORTED_DEFLATE_LENGTH) {
+        lazy = 1;
+        lazylength = length;
+        lazyoffset = offset;
+        continue; /*try the next byte*/
+      }
+      if(lazy) {
+        lazy = 0;
+        if(pos == 0) ERROR_BREAK(81);
+        if(length > lazylength + 1) {
+          /*push the previous character as literal*/
+          if(!uivector_push_back(out, in[pos - 1])) ERROR_BREAK(83 /*alloc fail*/);
+        } else {
+          length = lazylength;
+          offset = lazyoffset;
+          hash->head[hashval] = -1; /*the same hashchain update will be done, this ensures no wrong alteration*/
+          hash->headz[numzeros] = -1; /*idem*/
+          --pos;
+        }
+      }
+    }
+    if(length >= 3 && offset > windowsize) ERROR_BREAK(86 /*too big (or overflown negative) offset*/);
+
+    /*encode it as length/distance pair or literal value*/
+    if(length < 3) /*only lengths of 3 or higher are supported as length/distance pair*/ {
+      if(!uivector_push_back(out, in[pos])) ERROR_BREAK(83 /*alloc fail*/);
+    } else if(length < minmatch || (length == 3 && offset > 4096)) {
+      /*compensate for the fact that longer offsets have more extra bits, a
+      length of only 3 may be not worth it then*/
+      if(!uivector_push_back(out, in[pos])) ERROR_BREAK(83 /*alloc fail*/);
+    } else {
+      addLengthDistance(out, length, offset);
+      for(i = 1; i < length; ++i) {
+        ++pos;
+        wpos = pos & (windowsize - 1);
+        hashval = getHash(in, insize, pos);
+        if(usezeros && hashval == 0) {
+          if(numzeros == 0) numzeros = countZeros(in, insize, pos);
+          else if(pos + numzeros > insize || in[pos + numzeros - 1] != 0) --numzeros;
+        } else {
+          numzeros = 0;
+        }
+        updateHashChain(hash, wpos, hashval, numzeros);
+      }
+    }
+  } /*end of the loop through each character of input*/
+
+  return error;
+}
+
+/* /////////////////////////////////////////////////////////////////////////// */
+
+static unsigned deflateNoCompression(ucvector* out, const unsigned char* data, size_t datasize) {
+  /*non compressed deflate block data: 1 bit BFINAL,2 bits BTYPE,(5 bits): it jumps to start of next byte,
+  2 bytes LEN, 2 bytes NLEN, LEN bytes literal DATA*/
+
+  size_t i, numdeflateblocks = (datasize + 65534u) / 65535u;
+  unsigned datapos = 0;
+  for(i = 0; i != numdeflateblocks; ++i) {
+    unsigned BFINAL, BTYPE, LEN, NLEN;
+    unsigned char firstbyte;
+    size_t pos = out->size;
+
+    BFINAL = (i == numdeflateblocks - 1);
+    BTYPE = 0;
+
+    LEN = 65535;
+    if(datasize - datapos < 65535u) LEN = (unsigned)datasize - datapos;
+    NLEN = 65535 - LEN;
+
+    if(!ucvector_resize(out, out->size + LEN + 5)) return 83; /*alloc fail*/
+
+    firstbyte = (unsigned char)(BFINAL + ((BTYPE & 1u) << 1u) + ((BTYPE & 2u) << 1u));
+    out->data[pos + 0] = firstbyte;
+    out->data[pos + 1] = (unsigned char)(LEN & 255);
+    out->data[pos + 2] = (unsigned char)(LEN >> 8u);
+    out->data[pos + 3] = (unsigned char)(NLEN & 255);
+    out->data[pos + 4] = (unsigned char)(NLEN >> 8u);
+    lodepng_memcpy(out->data + pos + 5, data + datapos, LEN);
+    datapos += LEN;
+  }
+
+  return 0;
+}
+
+/*
+write the lz77-encoded data, which has lit, len and dist codes, to compressed stream using huffman trees.
+tree_ll: the tree for lit and len codes.
+tree_d: the tree for distance codes.
+*/
+static void writeLZ77data(LodePNGBitWriter* writer, const uivector* lz77_encoded,
+                          const HuffmanTree* tree_ll, const HuffmanTree* tree_d) {
+  size_t i = 0;
+  for(i = 0; i != lz77_encoded->size; ++i) {
+    unsigned val = lz77_encoded->data[i];
+    writeBitsReversed(writer, tree_ll->codes[val], tree_ll->lengths[val]);
+    if(val > 256) /*for a length code, 3 more things have to be added*/ {
+      unsigned length_index = val - FIRST_LENGTH_CODE_INDEX;
+      unsigned n_length_extra_bits = LENGTHEXTRA[length_index];
+      unsigned length_extra_bits = lz77_encoded->data[++i];
+
+      unsigned distance_code = lz77_encoded->data[++i];
+
+      unsigned distance_index = distance_code;
+      unsigned n_distance_extra_bits = DISTANCEEXTRA[distance_index];
+      unsigned distance_extra_bits = lz77_encoded->data[++i];
+
+      writeBits(writer, length_extra_bits, n_length_extra_bits);
+      writeBitsReversed(writer, tree_d->codes[distance_code], tree_d->lengths[distance_code]);
+      writeBits(writer, distance_extra_bits, n_distance_extra_bits);
+    }
+  }
+}
+
+/*Deflate for a block of type "dynamic", that is, with freely, optimally, created huffman trees*/
+static unsigned deflateDynamic(LodePNGBitWriter* writer, Hash* hash,
+                               const unsigned char* data, size_t datapos, size_t dataend,
+                               const LodePNGCompressSettings* settings, unsigned final) {
+  unsigned error = 0;
+
+  /*
+  A block is compressed as follows: The PNG data is lz77 encoded, resulting in
+  literal bytes and length/distance pairs. This is then huffman compressed with
+  two huffman trees. One huffman tree is used for the lit and len values ("ll"),
+  another huffman tree is used for the dist values ("d"). These two trees are
+  stored using their code lengths, and to compress even more these code lengths
+  are also run-length encoded and huffman compressed. This gives a huffman tree
+  of code lengths "cl". The code lengths used to describe this third tree are
+  the code length code lengths ("clcl").
+  */
+
+  /*The lz77 encoded data, represented with integers since there will also be length and distance codes in it*/
+  uivector lz77_encoded;
+  HuffmanTree tree_ll; /*tree for lit,len values*/
+  HuffmanTree tree_d; /*tree for distance codes*/
+  HuffmanTree tree_cl; /*tree for encoding the code lengths representing tree_ll and tree_d*/
+  unsigned* frequencies_ll = 0; /*frequency of lit,len codes*/
+  unsigned* frequencies_d = 0; /*frequency of dist codes*/
+  unsigned* frequencies_cl = 0; /*frequency of code length codes*/
+  unsigned* bitlen_lld = 0; /*lit,len,dist code lengths (int bits), literally (without repeat codes).*/
+  unsigned* bitlen_lld_e = 0; /*bitlen_lld encoded with repeat codes (this is a rudimentary run length compression)*/
+  size_t datasize = dataend - datapos;
+
+  /*
+  If we could call "bitlen_cl" the the code length code lengths ("clcl"), that is the bit lengths of codes to represent
+  tree_cl in CLCL_ORDER, then due to the huffman compression of huffman tree representations ("two levels"), there are
+  some analogies:
+  bitlen_lld is to tree_cl what data is to tree_ll and tree_d.
+  bitlen_lld_e is to bitlen_lld what lz77_encoded is to data.
+  bitlen_cl is to bitlen_lld_e what bitlen_lld is to lz77_encoded.
+  */
+
+  unsigned BFINAL = final;
+  size_t i;
+  size_t numcodes_ll, numcodes_d, numcodes_lld, numcodes_lld_e, numcodes_cl;
+  unsigned HLIT, HDIST, HCLEN;
+
+  uivector_init(&lz77_encoded);
+  HuffmanTree_init(&tree_ll);
+  HuffmanTree_init(&tree_d);
+  HuffmanTree_init(&tree_cl);
+  /* could fit on stack, but >1KB is on the larger side so allocate instead */
+  frequencies_ll = (unsigned*)lodepng_malloc(286 * sizeof(*frequencies_ll));
+  frequencies_d = (unsigned*)lodepng_malloc(30 * sizeof(*frequencies_d));
+  frequencies_cl = (unsigned*)lodepng_malloc(NUM_CODE_LENGTH_CODES * sizeof(*frequencies_cl));
+
+  if(!frequencies_ll || !frequencies_d || !frequencies_cl) error = 83; /*alloc fail*/
+
+  /*This while loop never loops due to a break at the end, it is here to
+  allow breaking out of it to the cleanup phase on error conditions.*/
+  while(!error) {
+    lodepng_memset(frequencies_ll, 0, 286 * sizeof(*frequencies_ll));
+    lodepng_memset(frequencies_d, 0, 30 * sizeof(*frequencies_d));
+    lodepng_memset(frequencies_cl, 0, NUM_CODE_LENGTH_CODES * sizeof(*frequencies_cl));
+
+    if(settings->use_lz77) {
+      error = encodeLZ77(&lz77_encoded, hash, data, datapos, dataend, settings->windowsize,
+                         settings->minmatch, settings->nicematch, settings->lazymatching);
+      if(error) break;
+    } else {
+      if(!uivector_resize(&lz77_encoded, datasize)) ERROR_BREAK(83 /*alloc fail*/);
+      for(i = datapos; i < dataend; ++i) lz77_encoded.data[i - datapos] = data[i]; /*no LZ77, but still will be Huffman compressed*/
+    }
+
+    /*Count the frequencies of lit, len and dist codes*/
+    for(i = 0; i != lz77_encoded.size; ++i) {
+      unsigned symbol = lz77_encoded.data[i];
+      ++frequencies_ll[symbol];
+      if(symbol > 256) {
+        unsigned dist = lz77_encoded.data[i + 2];
+        ++frequencies_d[dist];
+        i += 3;
+      }
+    }
+    frequencies_ll[256] = 1; /*there will be exactly 1 end code, at the end of the block*/
+
+    /*Make both huffman trees, one for the lit and len codes, one for the dist codes*/
+    error = HuffmanTree_makeFromFrequencies(&tree_ll, frequencies_ll, 257, 286, 15);
+    if(error) break;
+    /*2, not 1, is chosen for mincodes: some buggy PNG decoders require at least 2 symbols in the dist tree*/
+    error = HuffmanTree_makeFromFrequencies(&tree_d, frequencies_d, 2, 30, 15);
+    if(error) break;
+
+    numcodes_ll = LODEPNG_MIN(tree_ll.numcodes, 286);
+    numcodes_d = LODEPNG_MIN(tree_d.numcodes, 30);
+    /*store the code lengths of both generated trees in bitlen_lld*/
+    numcodes_lld = numcodes_ll + numcodes_d;
+    bitlen_lld = (unsigned*)lodepng_malloc(numcodes_lld * sizeof(*bitlen_lld));
+    /*numcodes_lld_e never needs more size than bitlen_lld*/
+    bitlen_lld_e = (unsigned*)lodepng_malloc(numcodes_lld * sizeof(*bitlen_lld_e));
+    if(!bitlen_lld || !bitlen_lld_e) ERROR_BREAK(83); /*alloc fail*/
+    numcodes_lld_e = 0;
+
+    for(i = 0; i != numcodes_ll; ++i) bitlen_lld[i] = tree_ll.lengths[i];
+    for(i = 0; i != numcodes_d; ++i) bitlen_lld[numcodes_ll + i] = tree_d.lengths[i];
+
+    /*run-length compress bitlen_ldd into bitlen_lld_e by using repeat codes 16 (copy length 3-6 times),
+    17 (3-10 zeroes), 18 (11-138 zeroes)*/
+    for(i = 0; i != numcodes_lld; ++i) {
+      unsigned j = 0; /*amount of repetitions*/
+      while(i + j + 1 < numcodes_lld && bitlen_lld[i + j + 1] == bitlen_lld[i]) ++j;
+
+      if(bitlen_lld[i] == 0 && j >= 2) /*repeat code for zeroes*/ {
+        ++j; /*include the first zero*/
+        if(j <= 10) /*repeat code 17 supports max 10 zeroes*/ {
+          bitlen_lld_e[numcodes_lld_e++] = 17;
+          bitlen_lld_e[numcodes_lld_e++] = j - 3;
+        } else /*repeat code 18 supports max 138 zeroes*/ {
+          if(j > 138) j = 138;
+          bitlen_lld_e[numcodes_lld_e++] = 18;
+          bitlen_lld_e[numcodes_lld_e++] = j - 11;
+        }
+        i += (j - 1);
+      } else if(j >= 3) /*repeat code for value other than zero*/ {
+        size_t k;
+        unsigned num = j / 6u, rest = j % 6u;
+        bitlen_lld_e[numcodes_lld_e++] = bitlen_lld[i];
+        for(k = 0; k < num; ++k) {
+          bitlen_lld_e[numcodes_lld_e++] = 16;
+          bitlen_lld_e[numcodes_lld_e++] = 6 - 3;
+        }
+        if(rest >= 3) {
+          bitlen_lld_e[numcodes_lld_e++] = 16;
+          bitlen_lld_e[numcodes_lld_e++] = rest - 3;
+        }
+        else j -= rest;
+        i += j;
+      } else /*too short to benefit from repeat code*/ {
+        bitlen_lld_e[numcodes_lld_e++] = bitlen_lld[i];
+      }
+    }
+
+    /*generate tree_cl, the huffmantree of huffmantrees*/
+    for(i = 0; i != numcodes_lld_e; ++i) {
+      ++frequencies_cl[bitlen_lld_e[i]];
+      /*after a repeat code come the bits that specify the number of repetitions,
+      those don't need to be in the frequencies_cl calculation*/
+      if(bitlen_lld_e[i] >= 16) ++i;
+    }
+
+    error = HuffmanTree_makeFromFrequencies(&tree_cl, frequencies_cl,
+                                            NUM_CODE_LENGTH_CODES, NUM_CODE_LENGTH_CODES, 7);
+    if(error) break;
+
+    /*compute amount of code-length-code-lengths to output*/
+    numcodes_cl = NUM_CODE_LENGTH_CODES;
+    /*trim zeros at the end (using CLCL_ORDER), but minimum size must be 4 (see HCLEN below)*/
+    while(numcodes_cl > 4u && tree_cl.lengths[CLCL_ORDER[numcodes_cl - 1u]] == 0) {
+      numcodes_cl--;
+    }
+
+    /*
+    Write everything into the output
+
+    After the BFINAL and BTYPE, the dynamic block consists out of the following:
+    - 5 bits HLIT, 5 bits HDIST, 4 bits HCLEN
+    - (HCLEN+4)*3 bits code lengths of code length alphabet
+    - HLIT + 257 code lengths of lit/length alphabet (encoded using the code length
+      alphabet, + possible repetition codes 16, 17, 18)
+    - HDIST + 1 code lengths of distance alphabet (encoded using the code length
+      alphabet, + possible repetition codes 16, 17, 18)
+    - compressed data
+    - 256 (end code)
+    */
+
+    /*Write block type*/
+    writeBits(writer, BFINAL, 1);
+    writeBits(writer, 0, 1); /*first bit of BTYPE "dynamic"*/
+    writeBits(writer, 1, 1); /*second bit of BTYPE "dynamic"*/
+
+    /*write the HLIT, HDIST and HCLEN values*/
+    /*all three sizes take trimmed ending zeroes into account, done either by HuffmanTree_makeFromFrequencies
+    or in the loop for numcodes_cl above, which saves space. */
+    HLIT = (unsigned)(numcodes_ll - 257);
+    HDIST = (unsigned)(numcodes_d - 1);
+    HCLEN = (unsigned)(numcodes_cl - 4);
+    writeBits(writer, HLIT, 5);
+    writeBits(writer, HDIST, 5);
+    writeBits(writer, HCLEN, 4);
+
+    /*write the code lengths of the code length alphabet ("bitlen_cl")*/
+    for(i = 0; i != numcodes_cl; ++i) writeBits(writer, tree_cl.lengths[CLCL_ORDER[i]], 3);
+
+    /*write the lengths of the lit/len AND the dist alphabet*/
+    for(i = 0; i != numcodes_lld_e; ++i) {
+      writeBitsReversed(writer, tree_cl.codes[bitlen_lld_e[i]], tree_cl.lengths[bitlen_lld_e[i]]);
+      /*extra bits of repeat codes*/
+      if(bitlen_lld_e[i] == 16) writeBits(writer, bitlen_lld_e[++i], 2);
+      else if(bitlen_lld_e[i] == 17) writeBits(writer, bitlen_lld_e[++i], 3);
+      else if(bitlen_lld_e[i] == 18) writeBits(writer, bitlen_lld_e[++i], 7);
+    }
+
+    /*write the compressed data symbols*/
+    writeLZ77data(writer, &lz77_encoded, &tree_ll, &tree_d);
+    /*error: the length of the end code 256 must be larger than 0*/
+    if(tree_ll.lengths[256] == 0) ERROR_BREAK(64);
+
+    /*write the end code*/
+    writeBitsReversed(writer, tree_ll.codes[256], tree_ll.lengths[256]);
+
+    break; /*end of error-while*/
+  }
+
+  /*cleanup*/
+  uivector_cleanup(&lz77_encoded);
+  HuffmanTree_cleanup(&tree_ll);
+  HuffmanTree_cleanup(&tree_d);
+  HuffmanTree_cleanup(&tree_cl);
+  lodepng_free(frequencies_ll);
+  lodepng_free(frequencies_d);
+  lodepng_free(frequencies_cl);
+  lodepng_free(bitlen_lld);
+  lodepng_free(bitlen_lld_e);
+
+  return error;
+}
+
+static unsigned deflateFixed(LodePNGBitWriter* writer, Hash* hash,
+                             const unsigned char* data,
+                             size_t datapos, size_t dataend,
+                             const LodePNGCompressSettings* settings, unsigned final) {
+  HuffmanTree tree_ll; /*tree for literal values and length codes*/
+  HuffmanTree tree_d; /*tree for distance codes*/
+
+  unsigned BFINAL = final;
+  unsigned error = 0;
+  size_t i;
+
+  HuffmanTree_init(&tree_ll);
+  HuffmanTree_init(&tree_d);
+
+  error = generateFixedLitLenTree(&tree_ll);
+  if(!error) error = generateFixedDistanceTree(&tree_d);
+
+  if(!error) {
+    writeBits(writer, BFINAL, 1);
+    writeBits(writer, 1, 1); /*first bit of BTYPE*/
+    writeBits(writer, 0, 1); /*second bit of BTYPE*/
+
+    if(settings->use_lz77) /*LZ77 encoded*/ {
+      uivector lz77_encoded;
+      uivector_init(&lz77_encoded);
+      error = encodeLZ77(&lz77_encoded, hash, data, datapos, dataend, settings->windowsize,
+                         settings->minmatch, settings->nicematch, settings->lazymatching);
+      if(!error) writeLZ77data(writer, &lz77_encoded, &tree_ll, &tree_d);
+      uivector_cleanup(&lz77_encoded);
+    } else /*no LZ77, but still will be Huffman compressed*/ {
+      for(i = datapos; i < dataend; ++i) {
+        writeBitsReversed(writer, tree_ll.codes[data[i]], tree_ll.lengths[data[i]]);
+      }
+    }
+    /*add END code*/
+    if(!error) writeBitsReversed(writer,tree_ll.codes[256], tree_ll.lengths[256]);
+  }
+
+  /*cleanup*/
+  HuffmanTree_cleanup(&tree_ll);
+  HuffmanTree_cleanup(&tree_d);
+
+  return error;
+}
+
+static unsigned lodepng_deflatev(ucvector* out, const unsigned char* in, size_t insize,
+                                 const LodePNGCompressSettings* settings) {
+  unsigned error = 0;
+  size_t i, blocksize, numdeflateblocks;
+  Hash hash;
+  LodePNGBitWriter writer;
+
+  LodePNGBitWriter_init(&writer, out);
+
+  if(settings->btype > 2) return 61;
+  else if(settings->btype == 0) return deflateNoCompression(out, in, insize);
+  else if(settings->btype == 1) blocksize = insize;
+  else /*if(settings->btype == 2)*/ {
+    /*on PNGs, deflate blocks of 65-262k seem to give most dense encoding*/
+    blocksize = insize / 8u + 8;
+    if(blocksize < 65536) blocksize = 65536;
+    if(blocksize > 262144) blocksize = 262144;
+  }
+
+  numdeflateblocks = (insize + blocksize - 1) / blocksize;
+  if(numdeflateblocks == 0) numdeflateblocks = 1;
+
+  error = hash_init(&hash, settings->windowsize);
+
+  if(!error) {
+    for(i = 0; i != numdeflateblocks && !error; ++i) {
+      unsigned final = (i == numdeflateblocks - 1);
+      size_t start = i * blocksize;
+      size_t end = start + blocksize;
+      if(end > insize) end = insize;
+
+      if(settings->btype == 1) error = deflateFixed(&writer, &hash, in, start, end, settings, final);
+      else if(settings->btype == 2) error = deflateDynamic(&writer, &hash, in, start, end, settings, final);
+    }
+  }
+
+  hash_cleanup(&hash);
+
+  return error;
+}
+
+unsigned lodepng_deflate(unsigned char** out, size_t* outsize,
+                         const unsigned char* in, size_t insize,
+                         const LodePNGCompressSettings* settings) {
+  ucvector v = ucvector_init(*out, *outsize);
+  unsigned error = lodepng_deflatev(&v, in, insize, settings);
+  *out = v.data;
+  *outsize = v.size;
+  return error;
+}
+
+static unsigned deflate(unsigned char** out, size_t* outsize,
+                        const unsigned char* in, size_t insize,
+                        const LodePNGCompressSettings* settings) {
+  if(settings->custom_deflate) {
+    return settings->custom_deflate(out, outsize, in, insize, settings);
+  } else {
+    return lodepng_deflate(out, outsize, in, insize, settings);
+  }
+}
+
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Adler32                                                                / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+static unsigned update_adler32(unsigned adler, const unsigned char* data, unsigned len) {
+  unsigned s1 = adler & 0xffffu;
+  unsigned s2 = (adler >> 16u) & 0xffffu;
+
+  while(len != 0u) {
+    unsigned i;
+    /*at least 5552 sums can be done before the sums overflow, saving a lot of module divisions*/
+    unsigned amount = len > 5552u ? 5552u : len;
+    len -= amount;
+    for(i = 0; i != amount; ++i) {
+      s1 += (*data++);
+      s2 += s1;
+    }
+    s1 %= 65521u;
+    s2 %= 65521u;
+  }
+
+  return (s2 << 16u) | s1;
+}
+
+/*Return the adler32 of the bytes data[0..len-1]*/
+static unsigned adler32(const unsigned char* data, unsigned len) {
+  return update_adler32(1u, data, len);
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Zlib                                                                   / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef LODEPNG_COMPILE_DECODER
+
+static unsigned lodepng_zlib_decompressv(ucvector* out,
+                                         const unsigned char* in, size_t insize,
+                                         const LodePNGDecompressSettings* settings) {
+  unsigned error = 0;
+  unsigned CM, CINFO, FDICT;
+
+  if(insize < 2) return 53; /*error, size of zlib data too small*/
+  /*read information from zlib header*/
+  if((in[0] * 256 + in[1]) % 31 != 0) {
+    /*error: 256 * in[0] + in[1] must be a multiple of 31, the FCHECK value is supposed to be made that way*/
+    return 24;
+  }
+
+  CM = in[0] & 15;
+  CINFO = (in[0] >> 4) & 15;
+  /*FCHECK = in[1] & 31;*/ /*FCHECK is already tested above*/
+  FDICT = (in[1] >> 5) & 1;
+  /*FLEVEL = (in[1] >> 6) & 3;*/ /*FLEVEL is not used here*/
+
+  if(CM != 8 || CINFO > 7) {
+    /*error: only compression method 8: inflate with sliding window of 32k is supported by the PNG spec*/
+    return 25;
+  }
+  if(FDICT != 0) {
+    /*error: the specification of PNG says about the zlib stream:
+      "The additional flags shall not specify a preset dictionary."*/
+    return 26;
+  }
+
+  error = inflatev(out, in + 2, insize - 2, settings);
+  if(error) return error;
+
+  if(!settings->ignore_adler32) {
+    unsigned ADLER32 = lodepng_read32bitInt(&in[insize - 4]);
+    unsigned checksum = adler32(out->data, (unsigned)(out->size));
+    if(checksum != ADLER32) return 58; /*error, adler checksum not correct, data must be corrupted*/
+  }
+
+  return 0; /*no error*/
+}
+
+
+unsigned lodepng_zlib_decompress(unsigned char** out, size_t* outsize, const unsigned char* in,
+                                 size_t insize, const LodePNGDecompressSettings* settings) {
+  ucvector v = ucvector_init(*out, *outsize);
+  unsigned error = lodepng_zlib_decompressv(&v, in, insize, settings);
+  *out = v.data;
+  *outsize = v.size;
+  return error;
+}
+
+/*expected_size is expected output size, to avoid intermediate allocations. Set to 0 if not known. */
+static unsigned zlib_decompress(unsigned char** out, size_t* outsize, size_t expected_size,
+                                const unsigned char* in, size_t insize, const LodePNGDecompressSettings* settings) {
+  if(settings->custom_zlib) {
+    return settings->custom_zlib(out, outsize, in, insize, settings);
+  } else {
+    unsigned error;
+    ucvector v = ucvector_init(*out, *outsize);
+    if(expected_size) {
+      /*reserve the memory to avoid intermediate reallocations*/
+      ucvector_resize(&v, *outsize + expected_size);
+      v.size = *outsize;
+    }
+    error = lodepng_zlib_decompressv(&v, in, insize, settings);
+    *out = v.data;
+    *outsize = v.size;
+    return error;
+  }
+}
+
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+#ifdef LODEPNG_COMPILE_ENCODER
+
+unsigned lodepng_zlib_compress(unsigned char** out, size_t* outsize, const unsigned char* in,
+                               size_t insize, const LodePNGCompressSettings* settings) {
+  size_t i;
+  unsigned error;
+  unsigned char* deflatedata = 0;
+  size_t deflatesize = 0;
+
+  error = deflate(&deflatedata, &deflatesize, in, insize, settings);
+
+  *out = NULL;
+  *outsize = 0;
+  if(!error) {
+    *outsize = deflatesize + 6;
+    *out = (unsigned char*)lodepng_malloc(*outsize);
+    if(!*out) error = 83; /*alloc fail*/
+  }
+
+  if(!error) {
+    unsigned ADLER32 = adler32(in, (unsigned)insize);
+    /*zlib data: 1 byte CMF (CM+CINFO), 1 byte FLG, deflate data, 4 byte ADLER32 checksum of the Decompressed data*/
+    unsigned CMF = 120; /*0b01111000: CM 8, CINFO 7. With CINFO 7, any window size up to 32768 can be used.*/
+    unsigned FLEVEL = 0;
+    unsigned FDICT = 0;
+    unsigned CMFFLG = 256 * CMF + FDICT * 32 + FLEVEL * 64;
+    unsigned FCHECK = 31 - CMFFLG % 31;
+    CMFFLG += FCHECK;
+
+    (*out)[0] = (unsigned char)(CMFFLG >> 8);
+    (*out)[1] = (unsigned char)(CMFFLG & 255);
+    for(i = 0; i != deflatesize; ++i) (*out)[i + 2] = deflatedata[i];
+    lodepng_set32bitInt(&(*out)[*outsize - 4], ADLER32);
+  }
+
+  lodepng_free(deflatedata);
+  return error;
+}
+
+/* compress using the default or custom zlib function */
+static unsigned zlib_compress(unsigned char** out, size_t* outsize, const unsigned char* in,
+                              size_t insize, const LodePNGCompressSettings* settings) {
+  if(settings->custom_zlib) {
+    return settings->custom_zlib(out, outsize, in, insize, settings);
+  } else {
+    return lodepng_zlib_compress(out, outsize, in, insize, settings);
+  }
+}
+
+#endif /*LODEPNG_COMPILE_ENCODER*/
+
+#else /*no LODEPNG_COMPILE_ZLIB*/
+
+#ifdef LODEPNG_COMPILE_DECODER
+static unsigned zlib_decompress(unsigned char** out, size_t* outsize, size_t expected_size,
+                                const unsigned char* in, size_t insize, const LodePNGDecompressSettings* settings) {
+  if(!settings->custom_zlib) return 87; /*no custom zlib function provided */
+  (void)expected_size;
+  return settings->custom_zlib(out, outsize, in, insize, settings);
+}
+#endif /*LODEPNG_COMPILE_DECODER*/
+#ifdef LODEPNG_COMPILE_ENCODER
+static unsigned zlib_compress(unsigned char** out, size_t* outsize, const unsigned char* in,
+                              size_t insize, const LodePNGCompressSettings* settings) {
+  if(!settings->custom_zlib) return 87; /*no custom zlib function provided */
+  return settings->custom_zlib(out, outsize, in, insize, settings);
+}
+#endif /*LODEPNG_COMPILE_ENCODER*/
+
+#endif /*LODEPNG_COMPILE_ZLIB*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef LODEPNG_COMPILE_ENCODER
+
+/*this is a good tradeoff between speed and compression ratio*/
+#define DEFAULT_WINDOWSIZE 2048
+
+void lodepng_compress_settings_init(LodePNGCompressSettings* settings) {
+  /*compress with dynamic huffman tree (not in the mathematical sense, just not the predefined one)*/
+  settings->btype = 2;
+  settings->use_lz77 = 1;
+  settings->windowsize = DEFAULT_WINDOWSIZE;
+  settings->minmatch = 3;
+  settings->nicematch = 128;
+  settings->lazymatching = 1;
+
+  settings->custom_zlib = 0;
+  settings->custom_deflate = 0;
+  settings->custom_context = 0;
+}
+
+const LodePNGCompressSettings lodepng_default_compress_settings = {2, 1, DEFAULT_WINDOWSIZE, 3, 128, 1, 0, 0, 0};
+
+
+#endif /*LODEPNG_COMPILE_ENCODER*/
+
+#ifdef LODEPNG_COMPILE_DECODER
+
+void lodepng_decompress_settings_init(LodePNGDecompressSettings* settings) {
+  settings->ignore_adler32 = 0;
+  settings->ignore_nlen = 0;
+
+  settings->custom_zlib = 0;
+  settings->custom_inflate = 0;
+  settings->custom_context = 0;
+}
+
+const LodePNGDecompressSettings lodepng_default_decompress_settings = {0, 0, 0, 0, 0};
+
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* // End of Zlib related code. Begin of PNG related code.                 // */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef LODEPNG_COMPILE_PNG
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / CRC32                                                                  / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+
+#ifndef LODEPNG_NO_COMPILE_CRC
+/* CRC polynomial: 0xedb88320 */
+static unsigned lodepng_crc32_table[256] = {
+           0u, 1996959894u, 3993919788u, 2567524794u,  124634137u, 1886057615u, 3915621685u, 2657392035u,
+   249268274u, 2044508324u, 3772115230u, 2547177864u,  162941995u, 2125561021u, 3887607047u, 2428444049u,
+   498536548u, 1789927666u, 4089016648u, 2227061214u,  450548861u, 1843258603u, 4107580753u, 2211677639u,
+   325883990u, 1684777152u, 4251122042u, 2321926636u,  335633487u, 1661365465u, 4195302755u, 2366115317u,
+   997073096u, 1281953886u, 3579855332u, 2724688242u, 1006888145u, 1258607687u, 3524101629u, 2768942443u,
+   901097722u, 1119000684u, 3686517206u, 2898065728u,  853044451u, 1172266101u, 3705015759u, 2882616665u,
+   651767980u, 1373503546u, 3369554304u, 3218104598u,  565507253u, 1454621731u, 3485111705u, 3099436303u,
+   671266974u, 1594198024u, 3322730930u, 2970347812u,  795835527u, 1483230225u, 3244367275u, 3060149565u,
+  1994146192u,   31158534u, 2563907772u, 4023717930u, 1907459465u,  112637215u, 2680153253u, 3904427059u,
+  2013776290u,  251722036u, 2517215374u, 3775830040u, 2137656763u,  141376813u, 2439277719u, 3865271297u,
+  1802195444u,  476864866u, 2238001368u, 4066508878u, 1812370925u,  453092731u, 2181625025u, 4111451223u,
+  1706088902u,  314042704u, 2344532202u, 4240017532u, 1658658271u,  366619977u, 2362670323u, 4224994405u,
+  1303535960u,  984961486u, 2747007092u, 3569037538u, 1256170817u, 1037604311u, 2765210733u, 3554079995u,
+  1131014506u,  879679996u, 2909243462u, 3663771856u, 1141124467u,  855842277u, 2852801631u, 3708648649u,
+  1342533948u,  654459306u, 3188396048u, 3373015174u, 1466479909u,  544179635u, 3110523913u, 3462522015u,
+  1591671054u,  702138776u, 2966460450u, 3352799412u, 1504918807u,  783551873u, 3082640443u, 3233442989u,
+  3988292384u, 2596254646u,   62317068u, 1957810842u, 3939845945u, 2647816111u,   81470997u, 1943803523u,
+  3814918930u, 2489596804u,  225274430u, 2053790376u, 3826175755u, 2466906013u,  167816743u, 2097651377u,
+  4027552580u, 2265490386u,  503444072u, 1762050814u, 4150417245u, 2154129355u,  426522225u, 1852507879u,
+  4275313526u, 2312317920u,  282753626u, 1742555852u, 4189708143u, 2394877945u,  397917763u, 1622183637u,
+  3604390888u, 2714866558u,  953729732u, 1340076626u, 3518719985u, 2797360999u, 1068828381u, 1219638859u,
+  3624741850u, 2936675148u,  906185462u, 1090812512u, 3747672003u, 2825379669u,  829329135u, 1181335161u,
+  3412177804u, 3160834842u,  628085408u, 1382605366u, 3423369109u, 3138078467u,  570562233u, 1426400815u,
+  3317316542u, 2998733608u,  733239954u, 1555261956u, 3268935591u, 3050360625u,  752459403u, 1541320221u,
+  2607071920u, 3965973030u, 1969922972u,   40735498u, 2617837225u, 3943577151u, 1913087877u,   83908371u,
+  2512341634u, 3803740692u, 2075208622u,  213261112u, 2463272603u, 3855990285u, 2094854071u,  198958881u,
+  2262029012u, 4057260610u, 1759359992u,  534414190u, 2176718541u, 4139329115u, 1873836001u,  414664567u,
+  2282248934u, 4279200368u, 1711684554u,  285281116u, 2405801727u, 4167216745u, 1634467795u,  376229701u,
+  2685067896u, 3608007406u, 1308918612u,  956543938u, 2808555105u, 3495958263u, 1231636301u, 1047427035u,
+  2932959818u, 3654703836u, 1088359270u,  936918000u, 2847714899u, 3736837829u, 1202900863u,  817233897u,
+  3183342108u, 3401237130u, 1404277552u,  615818150u, 3134207493u, 3453421203u, 1423857449u,  601450431u,
+  3009837614u, 3294710456u, 1567103746u,  711928724u, 3020668471u, 3272380065u, 1510334235u,  755167117u
+};
+
+/*Return the CRC of the bytes buf[0..len-1].*/
+unsigned lodepng_crc32(const unsigned char* data, size_t length) {
+  unsigned r = 0xffffffffu;
+  size_t i;
+  for(i = 0; i < length; ++i) {
+    r = lodepng_crc32_table[(r ^ data[i]) & 0xffu] ^ (r >> 8u);
+  }
+  return r ^ 0xffffffffu;
+}
+#else /* !LODEPNG_NO_COMPILE_CRC */
+unsigned lodepng_crc32(const unsigned char* data, size_t length);
+#endif /* !LODEPNG_NO_COMPILE_CRC */
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Reading and writing PNG color channel bits                             / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/* The color channel bits of less-than-8-bit pixels are read with the MSB of bytes first,
+so LodePNGBitWriter and LodePNGBitReader can't be used for those. */
+
+static unsigned char readBitFromReversedStream(size_t* bitpointer, const unsigned char* bitstream) {
+  unsigned char result = (unsigned char)((bitstream[(*bitpointer) >> 3] >> (7 - ((*bitpointer) & 0x7))) & 1);
+  ++(*bitpointer);
+  return result;
+}
+
+/* TODO: make this faster */
+static unsigned readBitsFromReversedStream(size_t* bitpointer, const unsigned char* bitstream, size_t nbits) {
+  unsigned result = 0;
+  size_t i;
+  for(i = 0 ; i < nbits; ++i) {
+    result <<= 1u;
+    result |= (unsigned)readBitFromReversedStream(bitpointer, bitstream);
+  }
+  return result;
+}
+
+static void setBitOfReversedStream(size_t* bitpointer, unsigned char* bitstream, unsigned char bit) {
+  /*the current bit in bitstream may be 0 or 1 for this to work*/
+  if(bit == 0) bitstream[(*bitpointer) >> 3u] &=  (unsigned char)(~(1u << (7u - ((*bitpointer) & 7u))));
+  else         bitstream[(*bitpointer) >> 3u] |=  (1u << (7u - ((*bitpointer) & 7u)));
+  ++(*bitpointer);
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / PNG chunks                                                             / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+unsigned lodepng_chunk_length(const unsigned char* chunk) {
+  return lodepng_read32bitInt(&chunk[0]);
+}
+
+void lodepng_chunk_type(char type[5], const unsigned char* chunk) {
+  unsigned i;
+  for(i = 0; i != 4; ++i) type[i] = (char)chunk[4 + i];
+  type[4] = 0; /*null termination char*/
+}
+
+unsigned char lodepng_chunk_type_equals(const unsigned char* chunk, const char* type) {
+  if(lodepng_strlen(type) != 4) return 0;
+  return (chunk[4] == type[0] && chunk[5] == type[1] && chunk[6] == type[2] && chunk[7] == type[3]);
+}
+
+unsigned char lodepng_chunk_ancillary(const unsigned char* chunk) {
+  return((chunk[4] & 32) != 0);
+}
+
+unsigned char lodepng_chunk_private(const unsigned char* chunk) {
+  return((chunk[6] & 32) != 0);
+}
+
+unsigned char lodepng_chunk_safetocopy(const unsigned char* chunk) {
+  return((chunk[7] & 32) != 0);
+}
+
+unsigned char* lodepng_chunk_data(unsigned char* chunk) {
+  return &chunk[8];
+}
+
+const unsigned char* lodepng_chunk_data_const(const unsigned char* chunk) {
+  return &chunk[8];
+}
+
+unsigned lodepng_chunk_check_crc(const unsigned char* chunk) {
+  unsigned length = lodepng_chunk_length(chunk);
+  unsigned CRC = lodepng_read32bitInt(&chunk[length + 8]);
+  /*the CRC is taken of the data and the 4 chunk type letters, not the length*/
+  unsigned checksum = lodepng_crc32(&chunk[4], length + 4);
+  if(CRC != checksum) return 1;
+  else return 0;
+}
+
+void lodepng_chunk_generate_crc(unsigned char* chunk) {
+  unsigned length = lodepng_chunk_length(chunk);
+  unsigned CRC = lodepng_crc32(&chunk[4], length + 4);
+  lodepng_set32bitInt(chunk + 8 + length, CRC);
+}
+
+unsigned char* lodepng_chunk_next(unsigned char* chunk, unsigned char* end) {
+  if(chunk >= end || end - chunk < 12) return end; /*too small to contain a chunk*/
+  if(chunk[0] == 0x89 && chunk[1] == 0x50 && chunk[2] == 0x4e && chunk[3] == 0x47
+    && chunk[4] == 0x0d && chunk[5] == 0x0a && chunk[6] == 0x1a && chunk[7] == 0x0a) {
+    /* Is PNG magic header at start of PNG file. Jump to first actual chunk. */
+    return chunk + 8;
+  } else {
+    size_t total_chunk_length;
+    unsigned char* result;
+    if(lodepng_addofl(lodepng_chunk_length(chunk), 12, &total_chunk_length)) return end;
+    result = chunk + total_chunk_length;
+    if(result < chunk) return end; /*pointer overflow*/
+    return result;
+  }
+}
+
+const unsigned char* lodepng_chunk_next_const(const unsigned char* chunk, const unsigned char* end) {
+  if(chunk >= end || end - chunk < 12) return end; /*too small to contain a chunk*/
+  if(chunk[0] == 0x89 && chunk[1] == 0x50 && chunk[2] == 0x4e && chunk[3] == 0x47
+    && chunk[4] == 0x0d && chunk[5] == 0x0a && chunk[6] == 0x1a && chunk[7] == 0x0a) {
+    /* Is PNG magic header at start of PNG file. Jump to first actual chunk. */
+    return chunk + 8;
+  } else {
+    size_t total_chunk_length;
+    const unsigned char* result;
+    if(lodepng_addofl(lodepng_chunk_length(chunk), 12, &total_chunk_length)) return end;
+    result = chunk + total_chunk_length;
+    if(result < chunk) return end; /*pointer overflow*/
+    return result;
+  }
+}
+
+unsigned char* lodepng_chunk_find(unsigned char* chunk, unsigned char* end, const char type[5]) {
+  for(;;) {
+    if(chunk >= end || end - chunk < 12) return 0; /* past file end: chunk + 12 > end */
+    if(lodepng_chunk_type_equals(chunk, type)) return chunk;
+    chunk = lodepng_chunk_next(chunk, end);
+  }
+}
+
+const unsigned char* lodepng_chunk_find_const(const unsigned char* chunk, const unsigned char* end, const char type[5]) {
+  for(;;) {
+    if(chunk >= end || end - chunk < 12) return 0; /* past file end: chunk + 12 > end */
+    if(lodepng_chunk_type_equals(chunk, type)) return chunk;
+    chunk = lodepng_chunk_next_const(chunk, end);
+  }
+}
+
+unsigned lodepng_chunk_append(unsigned char** out, size_t* outsize, const unsigned char* chunk) {
+  unsigned i;
+  size_t total_chunk_length, new_length;
+  unsigned char *chunk_start, *new_buffer;
+
+  if(lodepng_addofl(lodepng_chunk_length(chunk), 12, &total_chunk_length)) return 77;
+  if(lodepng_addofl(*outsize, total_chunk_length, &new_length)) return 77;
+
+  new_buffer = (unsigned char*)lodepng_realloc(*out, new_length);
+  if(!new_buffer) return 83; /*alloc fail*/
+  (*out) = new_buffer;
+  (*outsize) = new_length;
+  chunk_start = &(*out)[new_length - total_chunk_length];
+
+  for(i = 0; i != total_chunk_length; ++i) chunk_start[i] = chunk[i];
+
+  return 0;
+}
+
+/*Sets length and name and allocates the space for data and crc but does not
+set data or crc yet. Returns the start of the chunk in chunk. The start of
+the data is at chunk + 8. To finalize chunk, add the data, then use
+lodepng_chunk_generate_crc */
+static unsigned lodepng_chunk_init(unsigned char** chunk,
+                                   ucvector* out,
+                                   unsigned length, const char* type) {
+  size_t new_length = out->size;
+  if(lodepng_addofl(new_length, length, &new_length)) return 77;
+  if(lodepng_addofl(new_length, 12, &new_length)) return 77;
+  if(!ucvector_resize(out, new_length)) return 83; /*alloc fail*/
+  *chunk = out->data + new_length - length - 12u;
+
+  /*1: length*/
+  lodepng_set32bitInt(*chunk, length);
+
+  /*2: chunk name (4 letters)*/
+  lodepng_memcpy(*chunk + 4, type, 4);
+
+  return 0;
+}
+
+/* like lodepng_chunk_create but with custom allocsize */
+static unsigned lodepng_chunk_createv(ucvector* out,
+                                      unsigned length, const char* type, const unsigned char* data) {
+  unsigned char* chunk;
+  CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, length, type));
+
+  /*3: the data*/
+  lodepng_memcpy(chunk + 8, data, length);
+
+  /*4: CRC (of the chunkname characters and the data)*/
+  lodepng_chunk_generate_crc(chunk);
+
+  return 0;
+}
+
+unsigned lodepng_chunk_create(unsigned char** out, size_t* outsize,
+                              unsigned length, const char* type, const unsigned char* data) {
+  ucvector v = ucvector_init(*out, *outsize);
+  unsigned error = lodepng_chunk_createv(&v, length, type, data);
+  *out = v.data;
+  *outsize = v.size;
+  return error;
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Color types, channels, bits                                            / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*checks if the colortype is valid and the bitdepth bd is allowed for this colortype.
+Return value is a LodePNG error code.*/
+static unsigned checkColorValidity(LodePNGColorType colortype, unsigned bd) {
+  switch(colortype) {
+    case LCT_GREY:       if(!(bd == 1 || bd == 2 || bd == 4 || bd == 8 || bd == 16)) return 37; break;
+    case LCT_RGB:        if(!(                                 bd == 8 || bd == 16)) return 37; break;
+    case LCT_PALETTE:    if(!(bd == 1 || bd == 2 || bd == 4 || bd == 8            )) return 37; break;
+    case LCT_GREY_ALPHA: if(!(                                 bd == 8 || bd == 16)) return 37; break;
+    case LCT_RGBA:       if(!(                                 bd == 8 || bd == 16)) return 37; break;
+    case LCT_MAX_OCTET_VALUE: return 31; /* invalid color type */
+    default: return 31; /* invalid color type */
+  }
+  return 0; /*allowed color type / bits combination*/
+}
+
+static unsigned getNumColorChannels(LodePNGColorType colortype) {
+  switch(colortype) {
+    case LCT_GREY: return 1;
+    case LCT_RGB: return 3;
+    case LCT_PALETTE: return 1;
+    case LCT_GREY_ALPHA: return 2;
+    case LCT_RGBA: return 4;
+    case LCT_MAX_OCTET_VALUE: return 0; /* invalid color type */
+    default: return 0; /*invalid color type*/
+  }
+}
+
+static unsigned lodepng_get_bpp_lct(LodePNGColorType colortype, unsigned bitdepth) {
+  /*bits per pixel is amount of channels * bits per channel*/
+  return getNumColorChannels(colortype) * bitdepth;
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+void lodepng_color_mode_init(LodePNGColorMode* info) {
+  info->key_defined = 0;
+  info->key_r = info->key_g = info->key_b = 0;
+  info->colortype = LCT_RGBA;
+  info->bitdepth = 8;
+  info->palette = 0;
+  info->palettesize = 0;
+}
+
+/*allocates palette memory if needed, and initializes all colors to black*/
+static void lodepng_color_mode_alloc_palette(LodePNGColorMode* info) {
+  size_t i;
+  /*if the palette is already allocated, it will have size 1024 so no reallocation needed in that case*/
+  /*the palette must have room for up to 256 colors with 4 bytes each.*/
+  if(!info->palette) info->palette = (unsigned char*)lodepng_malloc(1024);
+  if(!info->palette) return; /*alloc fail*/
+  for(i = 0; i != 256; ++i) {
+    /*Initialize all unused colors with black, the value used for invalid palette indices.
+    This is an error according to the PNG spec, but common PNG decoders make it black instead.
+    That makes color conversion slightly faster due to no error handling needed.*/
+    info->palette[i * 4 + 0] = 0;
+    info->palette[i * 4 + 1] = 0;
+    info->palette[i * 4 + 2] = 0;
+    info->palette[i * 4 + 3] = 255;
+  }
+}
+
+void lodepng_color_mode_cleanup(LodePNGColorMode* info) {
+  lodepng_palette_clear(info);
+}
+
+unsigned lodepng_color_mode_copy(LodePNGColorMode* dest, const LodePNGColorMode* source) {
+  lodepng_color_mode_cleanup(dest);
+  lodepng_memcpy(dest, source, sizeof(LodePNGColorMode));
+  if(source->palette) {
+    dest->palette = (unsigned char*)lodepng_malloc(1024);
+    if(!dest->palette && source->palettesize) return 83; /*alloc fail*/
+    lodepng_memcpy(dest->palette, source->palette, source->palettesize * 4);
+  }
+  return 0;
+}
+
+LodePNGColorMode lodepng_color_mode_make(LodePNGColorType colortype, unsigned bitdepth) {
+  LodePNGColorMode result;
+  lodepng_color_mode_init(&result);
+  result.colortype = colortype;
+  result.bitdepth = bitdepth;
+  return result;
+}
+
+static int lodepng_color_mode_equal(const LodePNGColorMode* a, const LodePNGColorMode* b) {
+  size_t i;
+  if(a->colortype != b->colortype) return 0;
+  if(a->bitdepth != b->bitdepth) return 0;
+  if(a->key_defined != b->key_defined) return 0;
+  if(a->key_defined) {
+    if(a->key_r != b->key_r) return 0;
+    if(a->key_g != b->key_g) return 0;
+    if(a->key_b != b->key_b) return 0;
+  }
+  if(a->palettesize != b->palettesize) return 0;
+  for(i = 0; i != a->palettesize * 4; ++i) {
+    if(a->palette[i] != b->palette[i]) return 0;
+  }
+  return 1;
+}
+
+void lodepng_palette_clear(LodePNGColorMode* info) {
+  if(info->palette) lodepng_free(info->palette);
+  info->palette = 0;
+  info->palettesize = 0;
+}
+
+unsigned lodepng_palette_add(LodePNGColorMode* info,
+                             unsigned char r, unsigned char g, unsigned char b, unsigned char a) {
+  if(!info->palette) /*allocate palette if empty*/ {
+    lodepng_color_mode_alloc_palette(info);
+    if(!info->palette) return 83; /*alloc fail*/
+  }
+  if(info->palettesize >= 256) {
+    return 108; /*too many palette values*/
+  }
+  info->palette[4 * info->palettesize + 0] = r;
+  info->palette[4 * info->palettesize + 1] = g;
+  info->palette[4 * info->palettesize + 2] = b;
+  info->palette[4 * info->palettesize + 3] = a;
+  ++info->palettesize;
+  return 0;
+}
+
+/*calculate bits per pixel out of colortype and bitdepth*/
+unsigned lodepng_get_bpp(const LodePNGColorMode* info) {
+  return lodepng_get_bpp_lct(info->colortype, info->bitdepth);
+}
+
+unsigned lodepng_get_channels(const LodePNGColorMode* info) {
+  return getNumColorChannels(info->colortype);
+}
+
+unsigned lodepng_is_greyscale_type(const LodePNGColorMode* info) {
+  return info->colortype == LCT_GREY || info->colortype == LCT_GREY_ALPHA;
+}
+
+unsigned lodepng_is_alpha_type(const LodePNGColorMode* info) {
+  return (info->colortype & 4) != 0; /*4 or 6*/
+}
+
+unsigned lodepng_is_palette_type(const LodePNGColorMode* info) {
+  return info->colortype == LCT_PALETTE;
+}
+
+unsigned lodepng_has_palette_alpha(const LodePNGColorMode* info) {
+  size_t i;
+  for(i = 0; i != info->palettesize; ++i) {
+    if(info->palette[i * 4 + 3] < 255) return 1;
+  }
+  return 0;
+}
+
+unsigned lodepng_can_have_alpha(const LodePNGColorMode* info) {
+  return info->key_defined
+      || lodepng_is_alpha_type(info)
+      || lodepng_has_palette_alpha(info);
+}
+
+static size_t lodepng_get_raw_size_lct(unsigned w, unsigned h, LodePNGColorType colortype, unsigned bitdepth) {
+  size_t bpp = lodepng_get_bpp_lct(colortype, bitdepth);
+  size_t n = (size_t)w * (size_t)h;
+  return ((n / 8u) * bpp) + ((n & 7u) * bpp + 7u) / 8u;
+}
+
+size_t lodepng_get_raw_size(unsigned w, unsigned h, const LodePNGColorMode* color) {
+  return lodepng_get_raw_size_lct(w, h, color->colortype, color->bitdepth);
+}
+
+
+#ifdef LODEPNG_COMPILE_PNG
+
+/*in an idat chunk, each scanline is a multiple of 8 bits, unlike the lodepng output buffer,
+and in addition has one extra byte per line: the filter byte. So this gives a larger
+result than lodepng_get_raw_size. Set h to 1 to get the size of 1 row including filter byte. */
+static size_t lodepng_get_raw_size_idat(unsigned w, unsigned h, unsigned bpp) {
+  /* + 1 for the filter byte, and possibly plus padding bits per line. */
+  /* Ignoring casts, the expression is equal to (w * bpp + 7) / 8 + 1, but avoids overflow of w * bpp */
+  size_t line = ((size_t)(w / 8u) * bpp) + 1u + ((w & 7u) * bpp + 7u) / 8u;
+  return (size_t)h * line;
+}
+
+#ifdef LODEPNG_COMPILE_DECODER
+/*Safely checks whether size_t overflow can be caused due to amount of pixels.
+This check is overcautious rather than precise. If this check indicates no overflow,
+you can safely compute in a size_t (but not an unsigned):
+-(size_t)w * (size_t)h * 8
+-amount of bytes in IDAT (including filter, padding and Adam7 bytes)
+-amount of bytes in raw color model
+Returns 1 if overflow possible, 0 if not.
+*/
+static int lodepng_pixel_overflow(unsigned w, unsigned h,
+                                  const LodePNGColorMode* pngcolor, const LodePNGColorMode* rawcolor) {
+  size_t bpp = LODEPNG_MAX(lodepng_get_bpp(pngcolor), lodepng_get_bpp(rawcolor));
+  size_t numpixels, total;
+  size_t line; /* bytes per line in worst case */
+
+  if(lodepng_mulofl((size_t)w, (size_t)h, &numpixels)) return 1;
+  if(lodepng_mulofl(numpixels, 8, &total)) return 1; /* bit pointer with 8-bit color, or 8 bytes per channel color */
+
+  /* Bytes per scanline with the expression "(w / 8u) * bpp) + ((w & 7u) * bpp + 7u) / 8u" */
+  if(lodepng_mulofl((size_t)(w / 8u), bpp, &line)) return 1;
+  if(lodepng_addofl(line, ((w & 7u) * bpp + 7u) / 8u, &line)) return 1;
+
+  if(lodepng_addofl(line, 5, &line)) return 1; /* 5 bytes overhead per line: 1 filterbyte, 4 for Adam7 worst case */
+  if(lodepng_mulofl(line, h, &total)) return 1; /* Total bytes in worst case */
+
+  return 0; /* no overflow */
+}
+#endif /*LODEPNG_COMPILE_DECODER*/
+#endif /*LODEPNG_COMPILE_PNG*/
+
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+
+static void LodePNGUnknownChunks_init(LodePNGInfo* info) {
+  unsigned i;
+  for(i = 0; i != 3; ++i) info->unknown_chunks_data[i] = 0;
+  for(i = 0; i != 3; ++i) info->unknown_chunks_size[i] = 0;
+}
+
+static void LodePNGUnknownChunks_cleanup(LodePNGInfo* info) {
+  unsigned i;
+  for(i = 0; i != 3; ++i) lodepng_free(info->unknown_chunks_data[i]);
+}
+
+static unsigned LodePNGUnknownChunks_copy(LodePNGInfo* dest, const LodePNGInfo* src) {
+  unsigned i;
+
+  LodePNGUnknownChunks_cleanup(dest);
+
+  for(i = 0; i != 3; ++i) {
+    size_t j;
+    dest->unknown_chunks_size[i] = src->unknown_chunks_size[i];
+    dest->unknown_chunks_data[i] = (unsigned char*)lodepng_malloc(src->unknown_chunks_size[i]);
+    if(!dest->unknown_chunks_data[i] && dest->unknown_chunks_size[i]) return 83; /*alloc fail*/
+    for(j = 0; j < src->unknown_chunks_size[i]; ++j) {
+      dest->unknown_chunks_data[i][j] = src->unknown_chunks_data[i][j];
+    }
+  }
+
+  return 0;
+}
+
+/******************************************************************************/
+
+static void LodePNGText_init(LodePNGInfo* info) {
+  info->text_num = 0;
+  info->text_keys = NULL;
+  info->text_strings = NULL;
+}
+
+static void LodePNGText_cleanup(LodePNGInfo* info) {
+  size_t i;
+  for(i = 0; i != info->text_num; ++i) {
+    string_cleanup(&info->text_keys[i]);
+    string_cleanup(&info->text_strings[i]);
+  }
+  lodepng_free(info->text_keys);
+  lodepng_free(info->text_strings);
+}
+
+static unsigned LodePNGText_copy(LodePNGInfo* dest, const LodePNGInfo* source) {
+  size_t i = 0;
+  dest->text_keys = 0;
+  dest->text_strings = 0;
+  dest->text_num = 0;
+  for(i = 0; i != source->text_num; ++i) {
+    CERROR_TRY_RETURN(lodepng_add_text(dest, source->text_keys[i], source->text_strings[i]));
+  }
+  return 0;
+}
+
+static unsigned lodepng_add_text_sized(LodePNGInfo* info, const char* key, const char* str, size_t size) {
+  char** new_keys = (char**)(lodepng_realloc(info->text_keys, sizeof(char*) * (info->text_num + 1)));
+  char** new_strings = (char**)(lodepng_realloc(info->text_strings, sizeof(char*) * (info->text_num + 1)));
+
+  if(new_keys) info->text_keys = new_keys;
+  if(new_strings) info->text_strings = new_strings;
+
+  if(!new_keys || !new_strings) return 83; /*alloc fail*/
+
+  ++info->text_num;
+  info->text_keys[info->text_num - 1] = alloc_string(key);
+  info->text_strings[info->text_num - 1] = alloc_string_sized(str, size);
+  if(!info->text_keys[info->text_num - 1] || !info->text_strings[info->text_num - 1]) return 83; /*alloc fail*/
+
+  return 0;
+}
+
+unsigned lodepng_add_text(LodePNGInfo* info, const char* key, const char* str) {
+  return lodepng_add_text_sized(info, key, str, lodepng_strlen(str));
+}
+
+void lodepng_clear_text(LodePNGInfo* info) {
+  LodePNGText_cleanup(info);
+}
+
+/******************************************************************************/
+
+static void LodePNGIText_init(LodePNGInfo* info) {
+  info->itext_num = 0;
+  info->itext_keys = NULL;
+  info->itext_langtags = NULL;
+  info->itext_transkeys = NULL;
+  info->itext_strings = NULL;
+}
+
+static void LodePNGIText_cleanup(LodePNGInfo* info) {
+  size_t i;
+  for(i = 0; i != info->itext_num; ++i) {
+    string_cleanup(&info->itext_keys[i]);
+    string_cleanup(&info->itext_langtags[i]);
+    string_cleanup(&info->itext_transkeys[i]);
+    string_cleanup(&info->itext_strings[i]);
+  }
+  lodepng_free(info->itext_keys);
+  lodepng_free(info->itext_langtags);
+  lodepng_free(info->itext_transkeys);
+  lodepng_free(info->itext_strings);
+}
+
+static unsigned LodePNGIText_copy(LodePNGInfo* dest, const LodePNGInfo* source) {
+  size_t i = 0;
+  dest->itext_keys = 0;
+  dest->itext_langtags = 0;
+  dest->itext_transkeys = 0;
+  dest->itext_strings = 0;
+  dest->itext_num = 0;
+  for(i = 0; i != source->itext_num; ++i) {
+    CERROR_TRY_RETURN(lodepng_add_itext(dest, source->itext_keys[i], source->itext_langtags[i],
+                                        source->itext_transkeys[i], source->itext_strings[i]));
+  }
+  return 0;
+}
+
+void lodepng_clear_itext(LodePNGInfo* info) {
+  LodePNGIText_cleanup(info);
+}
+
+static unsigned lodepng_add_itext_sized(LodePNGInfo* info, const char* key, const char* langtag,
+                                        const char* transkey, const char* str, size_t size) {
+  char** new_keys = (char**)(lodepng_realloc(info->itext_keys, sizeof(char*) * (info->itext_num + 1)));
+  char** new_langtags = (char**)(lodepng_realloc(info->itext_langtags, sizeof(char*) * (info->itext_num + 1)));
+  char** new_transkeys = (char**)(lodepng_realloc(info->itext_transkeys, sizeof(char*) * (info->itext_num + 1)));
+  char** new_strings = (char**)(lodepng_realloc(info->itext_strings, sizeof(char*) * (info->itext_num + 1)));
+
+  if(new_keys) info->itext_keys = new_keys;
+  if(new_langtags) info->itext_langtags = new_langtags;
+  if(new_transkeys) info->itext_transkeys = new_transkeys;
+  if(new_strings) info->itext_strings = new_strings;
+
+  if(!new_keys || !new_langtags || !new_transkeys || !new_strings) return 83; /*alloc fail*/
+
+  ++info->itext_num;
+
+  info->itext_keys[info->itext_num - 1] = alloc_string(key);
+  info->itext_langtags[info->itext_num - 1] = alloc_string(langtag);
+  info->itext_transkeys[info->itext_num - 1] = alloc_string(transkey);
+  info->itext_strings[info->itext_num - 1] = alloc_string_sized(str, size);
+
+  return 0;
+}
+
+unsigned lodepng_add_itext(LodePNGInfo* info, const char* key, const char* langtag,
+                           const char* transkey, const char* str) {
+  return lodepng_add_itext_sized(info, key, langtag, transkey, str, lodepng_strlen(str));
+}
+
+/* same as set but does not delete */
+static unsigned lodepng_assign_icc(LodePNGInfo* info, const char* name, const unsigned char* profile, unsigned profile_size) {
+  if(profile_size == 0) return 100; /*invalid ICC profile size*/
+
+  info->iccp_name = alloc_string(name);
+  info->iccp_profile = (unsigned char*)lodepng_malloc(profile_size);
+
+  if(!info->iccp_name || !info->iccp_profile) return 83; /*alloc fail*/
+
+  lodepng_memcpy(info->iccp_profile, profile, profile_size);
+  info->iccp_profile_size = profile_size;
+
+  return 0; /*ok*/
+}
+
+unsigned lodepng_set_icc(LodePNGInfo* info, const char* name, const unsigned char* profile, unsigned profile_size) {
+  if(info->iccp_name) lodepng_clear_icc(info);
+  info->iccp_defined = 1;
+
+  return lodepng_assign_icc(info, name, profile, profile_size);
+}
+
+void lodepng_clear_icc(LodePNGInfo* info) {
+  string_cleanup(&info->iccp_name);
+  lodepng_free(info->iccp_profile);
+  info->iccp_profile = NULL;
+  info->iccp_profile_size = 0;
+  info->iccp_defined = 0;
+}
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+
+void lodepng_info_init(LodePNGInfo* info) {
+  lodepng_color_mode_init(&info->color);
+  info->interlace_method = 0;
+  info->compression_method = 0;
+  info->filter_method = 0;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+  info->background_defined = 0;
+  info->background_r = info->background_g = info->background_b = 0;
+
+  LodePNGText_init(info);
+  LodePNGIText_init(info);
+
+  info->time_defined = 0;
+  info->phys_defined = 0;
+
+  info->gama_defined = 0;
+  info->chrm_defined = 0;
+  info->srgb_defined = 0;
+  info->iccp_defined = 0;
+  info->iccp_name = NULL;
+  info->iccp_profile = NULL;
+
+  LodePNGUnknownChunks_init(info);
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+}
+
+void lodepng_info_cleanup(LodePNGInfo* info) {
+  lodepng_color_mode_cleanup(&info->color);
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+  LodePNGText_cleanup(info);
+  LodePNGIText_cleanup(info);
+
+  lodepng_clear_icc(info);
+
+  LodePNGUnknownChunks_cleanup(info);
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+}
+
+unsigned lodepng_info_copy(LodePNGInfo* dest, const LodePNGInfo* source) {
+  lodepng_info_cleanup(dest);
+  lodepng_memcpy(dest, source, sizeof(LodePNGInfo));
+  lodepng_color_mode_init(&dest->color);
+  CERROR_TRY_RETURN(lodepng_color_mode_copy(&dest->color, &source->color));
+
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+  CERROR_TRY_RETURN(LodePNGText_copy(dest, source));
+  CERROR_TRY_RETURN(LodePNGIText_copy(dest, source));
+  if(source->iccp_defined) {
+    CERROR_TRY_RETURN(lodepng_assign_icc(dest, source->iccp_name, source->iccp_profile, source->iccp_profile_size));
+  }
+
+  LodePNGUnknownChunks_init(dest);
+  CERROR_TRY_RETURN(LodePNGUnknownChunks_copy(dest, source));
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+  return 0;
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*index: bitgroup index, bits: bitgroup size(1, 2 or 4), in: bitgroup value, out: octet array to add bits to*/
+static void addColorBits(unsigned char* out, size_t index, unsigned bits, unsigned in) {
+  unsigned m = bits == 1 ? 7 : bits == 2 ? 3 : 1; /*8 / bits - 1*/
+  /*p = the partial index in the byte, e.g. with 4 palettebits it is 0 for first half or 1 for second half*/
+  unsigned p = index & m;
+  in &= (1u << bits) - 1u; /*filter out any other bits of the input value*/
+  in = in << (bits * (m - p));
+  if(p == 0) out[index * bits / 8u] = in;
+  else out[index * bits / 8u] |= in;
+}
+
+typedef struct ColorTree ColorTree;
+
+/*
+One node of a color tree
+This is the data structure used to count the number of unique colors and to get a palette
+index for a color. It's like an octree, but because the alpha channel is used too, each
+node has 16 instead of 8 children.
+*/
+struct ColorTree {
+  ColorTree* children[16]; /*up to 16 pointers to ColorTree of next level*/
+  int index; /*the payload. Only has a meaningful value if this is in the last level*/
+};
+
+static void color_tree_init(ColorTree* tree) {
+  lodepng_memset(tree->children, 0, 16 * sizeof(*tree->children));
+  tree->index = -1;
+}
+
+static void color_tree_cleanup(ColorTree* tree) {
+  int i;
+  for(i = 0; i != 16; ++i) {
+    if(tree->children[i]) {
+      color_tree_cleanup(tree->children[i]);
+      lodepng_free(tree->children[i]);
+    }
+  }
+}
+
+/*returns -1 if color not present, its index otherwise*/
+static int color_tree_get(ColorTree* tree, unsigned char r, unsigned char g, unsigned char b, unsigned char a) {
+  int bit = 0;
+  for(bit = 0; bit < 8; ++bit) {
+    int i = 8 * ((r >> bit) & 1) + 4 * ((g >> bit) & 1) + 2 * ((b >> bit) & 1) + 1 * ((a >> bit) & 1);
+    if(!tree->children[i]) return -1;
+    else tree = tree->children[i];
+  }
+  return tree ? tree->index : -1;
+}
+
+#ifdef LODEPNG_COMPILE_ENCODER
+static int color_tree_has(ColorTree* tree, unsigned char r, unsigned char g, unsigned char b, unsigned char a) {
+  return color_tree_get(tree, r, g, b, a) >= 0;
+}
+#endif /*LODEPNG_COMPILE_ENCODER*/
+
+/*color is not allowed to already exist.
+Index should be >= 0 (it's signed to be compatible with using -1 for "doesn't exist")
+Returns error code, or 0 if ok*/
+static unsigned color_tree_add(ColorTree* tree,
+                               unsigned char r, unsigned char g, unsigned char b, unsigned char a, unsigned index) {
+  int bit;
+  for(bit = 0; bit < 8; ++bit) {
+    int i = 8 * ((r >> bit) & 1) + 4 * ((g >> bit) & 1) + 2 * ((b >> bit) & 1) + 1 * ((a >> bit) & 1);
+    if(!tree->children[i]) {
+      tree->children[i] = (ColorTree*)lodepng_malloc(sizeof(ColorTree));
+      if(!tree->children[i]) return 83; /*alloc fail*/
+      color_tree_init(tree->children[i]);
+    }
+    tree = tree->children[i];
+  }
+  tree->index = (int)index;
+  return 0;
+}
+
+/*put a pixel, given its RGBA color, into image of any color type*/
+static unsigned rgba8ToPixel(unsigned char* out, size_t i,
+                             const LodePNGColorMode* mode, ColorTree* tree /*for palette*/,
+                             unsigned char r, unsigned char g, unsigned char b, unsigned char a) {
+  if(mode->colortype == LCT_GREY) {
+    unsigned char gray = r; /*((unsigned short)r + g + b) / 3u;*/
+    if(mode->bitdepth == 8) out[i] = gray;
+    else if(mode->bitdepth == 16) out[i * 2 + 0] = out[i * 2 + 1] = gray;
+    else {
+      /*take the most significant bits of gray*/
+      gray = ((unsigned)gray >> (8u - mode->bitdepth)) & ((1u << mode->bitdepth) - 1u);
+      addColorBits(out, i, mode->bitdepth, gray);
+    }
+  } else if(mode->colortype == LCT_RGB) {
+    if(mode->bitdepth == 8) {
+      out[i * 3 + 0] = r;
+      out[i * 3 + 1] = g;
+      out[i * 3 + 2] = b;
+    } else {
+      out[i * 6 + 0] = out[i * 6 + 1] = r;
+      out[i * 6 + 2] = out[i * 6 + 3] = g;
+      out[i * 6 + 4] = out[i * 6 + 5] = b;
+    }
+  } else if(mode->colortype == LCT_PALETTE) {
+    int index = color_tree_get(tree, r, g, b, a);
+    if(index < 0) return 82; /*color not in palette*/
+    if(mode->bitdepth == 8) out[i] = index;
+    else addColorBits(out, i, mode->bitdepth, (unsigned)index);
+  } else if(mode->colortype == LCT_GREY_ALPHA) {
+    unsigned char gray = r; /*((unsigned short)r + g + b) / 3u;*/
+    if(mode->bitdepth == 8) {
+      out[i * 2 + 0] = gray;
+      out[i * 2 + 1] = a;
+    } else if(mode->bitdepth == 16) {
+      out[i * 4 + 0] = out[i * 4 + 1] = gray;
+      out[i * 4 + 2] = out[i * 4 + 3] = a;
+    }
+  } else if(mode->colortype == LCT_RGBA) {
+    if(mode->bitdepth == 8) {
+      out[i * 4 + 0] = r;
+      out[i * 4 + 1] = g;
+      out[i * 4 + 2] = b;
+      out[i * 4 + 3] = a;
+    } else {
+      out[i * 8 + 0] = out[i * 8 + 1] = r;
+      out[i * 8 + 2] = out[i * 8 + 3] = g;
+      out[i * 8 + 4] = out[i * 8 + 5] = b;
+      out[i * 8 + 6] = out[i * 8 + 7] = a;
+    }
+  }
+
+  return 0; /*no error*/
+}
+
+/*put a pixel, given its RGBA16 color, into image of any color 16-bitdepth type*/
+static void rgba16ToPixel(unsigned char* out, size_t i,
+                         const LodePNGColorMode* mode,
+                         unsigned short r, unsigned short g, unsigned short b, unsigned short a) {
+  if(mode->colortype == LCT_GREY) {
+    unsigned short gray = r; /*((unsigned)r + g + b) / 3u;*/
+    out[i * 2 + 0] = (gray >> 8) & 255;
+    out[i * 2 + 1] = gray & 255;
+  } else if(mode->colortype == LCT_RGB) {
+    out[i * 6 + 0] = (r >> 8) & 255;
+    out[i * 6 + 1] = r & 255;
+    out[i * 6 + 2] = (g >> 8) & 255;
+    out[i * 6 + 3] = g & 255;
+    out[i * 6 + 4] = (b >> 8) & 255;
+    out[i * 6 + 5] = b & 255;
+  } else if(mode->colortype == LCT_GREY_ALPHA) {
+    unsigned short gray = r; /*((unsigned)r + g + b) / 3u;*/
+    out[i * 4 + 0] = (gray >> 8) & 255;
+    out[i * 4 + 1] = gray & 255;
+    out[i * 4 + 2] = (a >> 8) & 255;
+    out[i * 4 + 3] = a & 255;
+  } else if(mode->colortype == LCT_RGBA) {
+    out[i * 8 + 0] = (r >> 8) & 255;
+    out[i * 8 + 1] = r & 255;
+    out[i * 8 + 2] = (g >> 8) & 255;
+    out[i * 8 + 3] = g & 255;
+    out[i * 8 + 4] = (b >> 8) & 255;
+    out[i * 8 + 5] = b & 255;
+    out[i * 8 + 6] = (a >> 8) & 255;
+    out[i * 8 + 7] = a & 255;
+  }
+}
+
+/*Get RGBA8 color of pixel with index i (y * width + x) from the raw image with given color type.*/
+static void getPixelColorRGBA8(unsigned char* r, unsigned char* g,
+                               unsigned char* b, unsigned char* a,
+                               const unsigned char* in, size_t i,
+                               const LodePNGColorMode* mode) {
+  if(mode->colortype == LCT_GREY) {
+    if(mode->bitdepth == 8) {
+      *r = *g = *b = in[i];
+      if(mode->key_defined && *r == mode->key_r) *a = 0;
+      else *a = 255;
+    } else if(mode->bitdepth == 16) {
+      *r = *g = *b = in[i * 2 + 0];
+      if(mode->key_defined && 256U * in[i * 2 + 0] + in[i * 2 + 1] == mode->key_r) *a = 0;
+      else *a = 255;
+    } else {
+      unsigned highest = ((1U << mode->bitdepth) - 1U); /*highest possible value for this bit depth*/
+      size_t j = i * mode->bitdepth;
+      unsigned value = readBitsFromReversedStream(&j, in, mode->bitdepth);
+      *r = *g = *b = (value * 255) / highest;
+      if(mode->key_defined && value == mode->key_r) *a = 0;
+      else *a = 255;
+    }
+  } else if(mode->colortype == LCT_RGB) {
+    if(mode->bitdepth == 8) {
+      *r = in[i * 3 + 0]; *g = in[i * 3 + 1]; *b = in[i * 3 + 2];
+      if(mode->key_defined && *r == mode->key_r && *g == mode->key_g && *b == mode->key_b) *a = 0;
+      else *a = 255;
+    } else {
+      *r = in[i * 6 + 0];
+      *g = in[i * 6 + 2];
+      *b = in[i * 6 + 4];
+      if(mode->key_defined && 256U * in[i * 6 + 0] + in[i * 6 + 1] == mode->key_r
+         && 256U * in[i * 6 + 2] + in[i * 6 + 3] == mode->key_g
+         && 256U * in[i * 6 + 4] + in[i * 6 + 5] == mode->key_b) *a = 0;
+      else *a = 255;
+    }
+  } else if(mode->colortype == LCT_PALETTE) {
+    unsigned index;
+    if(mode->bitdepth == 8) index = in[i];
+    else {
+      size_t j = i * mode->bitdepth;
+      index = readBitsFromReversedStream(&j, in, mode->bitdepth);
+    }
+    /*out of bounds of palette not checked: see lodepng_color_mode_alloc_palette.*/
+    *r = mode->palette[index * 4 + 0];
+    *g = mode->palette[index * 4 + 1];
+    *b = mode->palette[index * 4 + 2];
+    *a = mode->palette[index * 4 + 3];
+  } else if(mode->colortype == LCT_GREY_ALPHA) {
+    if(mode->bitdepth == 8) {
+      *r = *g = *b = in[i * 2 + 0];
+      *a = in[i * 2 + 1];
+    } else {
+      *r = *g = *b = in[i * 4 + 0];
+      *a = in[i * 4 + 2];
+    }
+  } else if(mode->colortype == LCT_RGBA) {
+    if(mode->bitdepth == 8) {
+      *r = in[i * 4 + 0];
+      *g = in[i * 4 + 1];
+      *b = in[i * 4 + 2];
+      *a = in[i * 4 + 3];
+    } else {
+      *r = in[i * 8 + 0];
+      *g = in[i * 8 + 2];
+      *b = in[i * 8 + 4];
+      *a = in[i * 8 + 6];
+    }
+  }
+}
+
+/*Similar to getPixelColorRGBA8, but with all the for loops inside of the color
+mode test cases, optimized to convert the colors much faster, when converting
+to the common case of RGBA with 8 bit per channel. buffer must be RGBA with
+enough memory.*/
+static void getPixelColorsRGBA8(unsigned char* LODEPNG_RESTRICT buffer, size_t numpixels,
+                                const unsigned char* LODEPNG_RESTRICT in,
+                                const LodePNGColorMode* mode) {
+  unsigned num_channels = 4;
+  size_t i;
+  if(mode->colortype == LCT_GREY) {
+    if(mode->bitdepth == 8) {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = buffer[1] = buffer[2] = in[i];
+        buffer[3] = 255;
+      }
+      if(mode->key_defined) {
+        buffer -= numpixels * num_channels;
+        for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+          if(buffer[0] == mode->key_r) buffer[3] = 0;
+        }
+      }
+    } else if(mode->bitdepth == 16) {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = buffer[1] = buffer[2] = in[i * 2];
+        buffer[3] = mode->key_defined && 256U * in[i * 2 + 0] + in[i * 2 + 1] == mode->key_r ? 0 : 255;
+      }
+    } else {
+      unsigned highest = ((1U << mode->bitdepth) - 1U); /*highest possible value for this bit depth*/
+      size_t j = 0;
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        unsigned value = readBitsFromReversedStream(&j, in, mode->bitdepth);
+        buffer[0] = buffer[1] = buffer[2] = (value * 255) / highest;
+        buffer[3] = mode->key_defined && value == mode->key_r ? 0 : 255;
+      }
+    }
+  } else if(mode->colortype == LCT_RGB) {
+    if(mode->bitdepth == 8) {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        lodepng_memcpy(buffer, &in[i * 3], 3);
+        buffer[3] = 255;
+      }
+      if(mode->key_defined) {
+        buffer -= numpixels * num_channels;
+        for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+          if(buffer[0] == mode->key_r && buffer[1]== mode->key_g && buffer[2] == mode->key_b) buffer[3] = 0;
+        }
+      }
+    } else {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = in[i * 6 + 0];
+        buffer[1] = in[i * 6 + 2];
+        buffer[2] = in[i * 6 + 4];
+        buffer[3] = mode->key_defined
+           && 256U * in[i * 6 + 0] + in[i * 6 + 1] == mode->key_r
+           && 256U * in[i * 6 + 2] + in[i * 6 + 3] == mode->key_g
+           && 256U * in[i * 6 + 4] + in[i * 6 + 5] == mode->key_b ? 0 : 255;
+      }
+    }
+  } else if(mode->colortype == LCT_PALETTE) {
+    if(mode->bitdepth == 8) {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        unsigned index = in[i];
+        /*out of bounds of palette not checked: see lodepng_color_mode_alloc_palette.*/
+        lodepng_memcpy(buffer, &mode->palette[index * 4], 4);
+      }
+    } else {
+      size_t j = 0;
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        unsigned index = readBitsFromReversedStream(&j, in, mode->bitdepth);
+        /*out of bounds of palette not checked: see lodepng_color_mode_alloc_palette.*/
+        lodepng_memcpy(buffer, &mode->palette[index * 4], 4);
+      }
+    }
+  } else if(mode->colortype == LCT_GREY_ALPHA) {
+    if(mode->bitdepth == 8) {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = buffer[1] = buffer[2] = in[i * 2 + 0];
+        buffer[3] = in[i * 2 + 1];
+      }
+    } else {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = buffer[1] = buffer[2] = in[i * 4 + 0];
+        buffer[3] = in[i * 4 + 2];
+      }
+    }
+  } else if(mode->colortype == LCT_RGBA) {
+    if(mode->bitdepth == 8) {
+      lodepng_memcpy(buffer, in, numpixels * 4);
+    } else {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = in[i * 8 + 0];
+        buffer[1] = in[i * 8 + 2];
+        buffer[2] = in[i * 8 + 4];
+        buffer[3] = in[i * 8 + 6];
+      }
+    }
+  }
+}
+
+/*Similar to getPixelColorsRGBA8, but with 3-channel RGB output.*/
+static void getPixelColorsRGB8(unsigned char* LODEPNG_RESTRICT buffer, size_t numpixels,
+                               const unsigned char* LODEPNG_RESTRICT in,
+                               const LodePNGColorMode* mode) {
+  const unsigned num_channels = 3;
+  size_t i;
+  if(mode->colortype == LCT_GREY) {
+    if(mode->bitdepth == 8) {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = buffer[1] = buffer[2] = in[i];
+      }
+    } else if(mode->bitdepth == 16) {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = buffer[1] = buffer[2] = in[i * 2];
+      }
+    } else {
+      unsigned highest = ((1U << mode->bitdepth) - 1U); /*highest possible value for this bit depth*/
+      size_t j = 0;
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        unsigned value = readBitsFromReversedStream(&j, in, mode->bitdepth);
+        buffer[0] = buffer[1] = buffer[2] = (value * 255) / highest;
+      }
+    }
+  } else if(mode->colortype == LCT_RGB) {
+    if(mode->bitdepth == 8) {
+      lodepng_memcpy(buffer, in, numpixels * 3);
+    } else {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = in[i * 6 + 0];
+        buffer[1] = in[i * 6 + 2];
+        buffer[2] = in[i * 6 + 4];
+      }
+    }
+  } else if(mode->colortype == LCT_PALETTE) {
+    if(mode->bitdepth == 8) {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        unsigned index = in[i];
+        /*out of bounds of palette not checked: see lodepng_color_mode_alloc_palette.*/
+        lodepng_memcpy(buffer, &mode->palette[index * 4], 3);
+      }
+    } else {
+      size_t j = 0;
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        unsigned index = readBitsFromReversedStream(&j, in, mode->bitdepth);
+        /*out of bounds of palette not checked: see lodepng_color_mode_alloc_palette.*/
+        lodepng_memcpy(buffer, &mode->palette[index * 4], 3);
+      }
+    }
+  } else if(mode->colortype == LCT_GREY_ALPHA) {
+    if(mode->bitdepth == 8) {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = buffer[1] = buffer[2] = in[i * 2 + 0];
+      }
+    } else {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = buffer[1] = buffer[2] = in[i * 4 + 0];
+      }
+    }
+  } else if(mode->colortype == LCT_RGBA) {
+    if(mode->bitdepth == 8) {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        lodepng_memcpy(buffer, &in[i * 4], 3);
+      }
+    } else {
+      for(i = 0; i != numpixels; ++i, buffer += num_channels) {
+        buffer[0] = in[i * 8 + 0];
+        buffer[1] = in[i * 8 + 2];
+        buffer[2] = in[i * 8 + 4];
+      }
+    }
+  }
+}
+
+/*Get RGBA16 color of pixel with index i (y * width + x) from the raw image with
+given color type, but the given color type must be 16-bit itself.*/
+static void getPixelColorRGBA16(unsigned short* r, unsigned short* g, unsigned short* b, unsigned short* a,
+                                const unsigned char* in, size_t i, const LodePNGColorMode* mode) {
+  if(mode->colortype == LCT_GREY) {
+    *r = *g = *b = 256 * in[i * 2 + 0] + in[i * 2 + 1];
+    if(mode->key_defined && 256U * in[i * 2 + 0] + in[i * 2 + 1] == mode->key_r) *a = 0;
+    else *a = 65535;
+  } else if(mode->colortype == LCT_RGB) {
+    *r = 256u * in[i * 6 + 0] + in[i * 6 + 1];
+    *g = 256u * in[i * 6 + 2] + in[i * 6 + 3];
+    *b = 256u * in[i * 6 + 4] + in[i * 6 + 5];
+    if(mode->key_defined
+       && 256u * in[i * 6 + 0] + in[i * 6 + 1] == mode->key_r
+       && 256u * in[i * 6 + 2] + in[i * 6 + 3] == mode->key_g
+       && 256u * in[i * 6 + 4] + in[i * 6 + 5] == mode->key_b) *a = 0;
+    else *a = 65535;
+  } else if(mode->colortype == LCT_GREY_ALPHA) {
+    *r = *g = *b = 256u * in[i * 4 + 0] + in[i * 4 + 1];
+    *a = 256u * in[i * 4 + 2] + in[i * 4 + 3];
+  } else if(mode->colortype == LCT_RGBA) {
+    *r = 256u * in[i * 8 + 0] + in[i * 8 + 1];
+    *g = 256u * in[i * 8 + 2] + in[i * 8 + 3];
+    *b = 256u * in[i * 8 + 4] + in[i * 8 + 5];
+    *a = 256u * in[i * 8 + 6] + in[i * 8 + 7];
+  }
+}
+
+unsigned lodepng_convert(unsigned char* out, const unsigned char* in,
+                         const LodePNGColorMode* mode_out, const LodePNGColorMode* mode_in,
+                         unsigned w, unsigned h) {
+  size_t i;
+  ColorTree tree;
+  size_t numpixels = (size_t)w * (size_t)h;
+  unsigned error = 0;
+
+  if(mode_in->colortype == LCT_PALETTE && !mode_in->palette) {
+    return 107; /* error: must provide palette if input mode is palette */
+  }
+
+  if(lodepng_color_mode_equal(mode_out, mode_in)) {
+    size_t numbytes = lodepng_get_raw_size(w, h, mode_in);
+    lodepng_memcpy(out, in, numbytes);
+    return 0;
+  }
+
+  if(mode_out->colortype == LCT_PALETTE) {
+    size_t palettesize = mode_out->palettesize;
+    const unsigned char* palette = mode_out->palette;
+    size_t palsize = (size_t)1u << mode_out->bitdepth;
+    /*if the user specified output palette but did not give the values, assume
+    they want the values of the input color type (assuming that one is palette).
+    Note that we never create a new palette ourselves.*/
+    if(palettesize == 0) {
+      palettesize = mode_in->palettesize;
+      palette = mode_in->palette;
+      /*if the input was also palette with same bitdepth, then the color types are also
+      equal, so copy literally. This to preserve the exact indices that were in the PNG
+      even in case there are duplicate colors in the palette.*/
+      if(mode_in->colortype == LCT_PALETTE && mode_in->bitdepth == mode_out->bitdepth) {
+        size_t numbytes = lodepng_get_raw_size(w, h, mode_in);
+        lodepng_memcpy(out, in, numbytes);
+        return 0;
+      }
+    }
+    if(palettesize < palsize) palsize = palettesize;
+    color_tree_init(&tree);
+    for(i = 0; i != palsize; ++i) {
+      const unsigned char* p = &palette[i * 4];
+      error = color_tree_add(&tree, p[0], p[1], p[2], p[3], (unsigned)i);
+      if(error) break;
+    }
+  }
+
+  if(!error) {
+    if(mode_in->bitdepth == 16 && mode_out->bitdepth == 16) {
+      for(i = 0; i != numpixels; ++i) {
+        unsigned short r = 0, g = 0, b = 0, a = 0;
+        getPixelColorRGBA16(&r, &g, &b, &a, in, i, mode_in);
+        rgba16ToPixel(out, i, mode_out, r, g, b, a);
+      }
+    } else if(mode_out->bitdepth == 8 && mode_out->colortype == LCT_RGBA) {
+      getPixelColorsRGBA8(out, numpixels, in, mode_in);
+    } else if(mode_out->bitdepth == 8 && mode_out->colortype == LCT_RGB) {
+      getPixelColorsRGB8(out, numpixels, in, mode_in);
+    } else {
+      unsigned char r = 0, g = 0, b = 0, a = 0;
+      for(i = 0; i != numpixels; ++i) {
+        getPixelColorRGBA8(&r, &g, &b, &a, in, i, mode_in);
+        error = rgba8ToPixel(out, i, mode_out, &tree, r, g, b, a);
+        if(error) break;
+      }
+    }
+  }
+
+  if(mode_out->colortype == LCT_PALETTE) {
+    color_tree_cleanup(&tree);
+  }
+
+  return error;
+}
+
+
+/* Converts a single rgb color without alpha from one type to another, color bits truncated to
+their bitdepth. In case of single channel (gray or palette), only the r channel is used. Slow
+function, do not use to process all pixels of an image. Alpha channel not supported on purpose:
+this is for bKGD, supporting alpha may prevent it from finding a color in the palette, from the
+specification it looks like bKGD should ignore the alpha values of the palette since it can use
+any palette index but doesn't have an alpha channel. Idem with ignoring color key. */
+unsigned lodepng_convert_rgb(
+    unsigned* r_out, unsigned* g_out, unsigned* b_out,
+    unsigned r_in, unsigned g_in, unsigned b_in,
+    const LodePNGColorMode* mode_out, const LodePNGColorMode* mode_in) {
+  unsigned r = 0, g = 0, b = 0;
+  unsigned mul = 65535 / ((1u << mode_in->bitdepth) - 1u); /*65535, 21845, 4369, 257, 1*/
+  unsigned shift = 16 - mode_out->bitdepth;
+
+  if(mode_in->colortype == LCT_GREY || mode_in->colortype == LCT_GREY_ALPHA) {
+    r = g = b = r_in * mul;
+  } else if(mode_in->colortype == LCT_RGB || mode_in->colortype == LCT_RGBA) {
+    r = r_in * mul;
+    g = g_in * mul;
+    b = b_in * mul;
+  } else if(mode_in->colortype == LCT_PALETTE) {
+    if(r_in >= mode_in->palettesize) return 82;
+    r = mode_in->palette[r_in * 4 + 0] * 257u;
+    g = mode_in->palette[r_in * 4 + 1] * 257u;
+    b = mode_in->palette[r_in * 4 + 2] * 257u;
+  } else {
+    return 31;
+  }
+
+  /* now convert to output format */
+  if(mode_out->colortype == LCT_GREY || mode_out->colortype == LCT_GREY_ALPHA) {
+    *r_out = r >> shift ;
+  } else if(mode_out->colortype == LCT_RGB || mode_out->colortype == LCT_RGBA) {
+    *r_out = r >> shift ;
+    *g_out = g >> shift ;
+    *b_out = b >> shift ;
+  } else if(mode_out->colortype == LCT_PALETTE) {
+    unsigned i;
+    /* a 16-bit color cannot be in the palette */
+    if((r >> 8) != (r & 255) || (g >> 8) != (g & 255) || (b >> 8) != (b & 255)) return 82;
+    for(i = 0; i < mode_out->palettesize; i++) {
+      unsigned j = i * 4;
+      if((r >> 8) == mode_out->palette[j + 0] && (g >> 8) == mode_out->palette[j + 1] &&
+          (b >> 8) == mode_out->palette[j + 2]) {
+        *r_out = i;
+        return 0;
+      }
+    }
+    return 82;
+  } else {
+    return 31;
+  }
+
+  return 0;
+}
+
+#ifdef LODEPNG_COMPILE_ENCODER
+
+void lodepng_color_stats_init(LodePNGColorStats* stats) {
+  /*stats*/
+  stats->colored = 0;
+  stats->key = 0;
+  stats->key_r = stats->key_g = stats->key_b = 0;
+  stats->alpha = 0;
+  stats->numcolors = 0;
+  stats->bits = 1;
+  stats->numpixels = 0;
+  /*settings*/
+  stats->allow_palette = 1;
+  stats->allow_greyscale = 1;
+}
+
+/*function used for debug purposes with C++*/
+/*void printColorStats(LodePNGColorStats* p) {
+  std::cout << "colored: " << (int)p->colored << ", ";
+  std::cout << "key: " << (int)p->key << ", ";
+  std::cout << "key_r: " << (int)p->key_r << ", ";
+  std::cout << "key_g: " << (int)p->key_g << ", ";
+  std::cout << "key_b: " << (int)p->key_b << ", ";
+  std::cout << "alpha: " << (int)p->alpha << ", ";
+  std::cout << "numcolors: " << (int)p->numcolors << ", ";
+  std::cout << "bits: " << (int)p->bits << std::endl;
+}*/
+
+/*Returns how many bits needed to represent given value (max 8 bit)*/
+static unsigned getValueRequiredBits(unsigned char value) {
+  if(value == 0 || value == 255) return 1;
+  /*The scaling of 2-bit and 4-bit values uses multiples of 85 and 17*/
+  if(value % 17 == 0) return value % 85 == 0 ? 2 : 4;
+  return 8;
+}
+
+/*stats must already have been inited. */
+unsigned lodepng_compute_color_stats(LodePNGColorStats* stats,
+                                     const unsigned char* in, unsigned w, unsigned h,
+                                     const LodePNGColorMode* mode_in) {
+  size_t i;
+  ColorTree tree;
+  size_t numpixels = (size_t)w * (size_t)h;
+  unsigned error = 0;
+
+  /* mark things as done already if it would be impossible to have a more expensive case */
+  unsigned colored_done = lodepng_is_greyscale_type(mode_in) ? 1 : 0;
+  unsigned alpha_done = lodepng_can_have_alpha(mode_in) ? 0 : 1;
+  unsigned numcolors_done = 0;
+  unsigned bpp = lodepng_get_bpp(mode_in);
+  unsigned bits_done = (stats->bits == 1 && bpp == 1) ? 1 : 0;
+  unsigned sixteen = 0; /* whether the input image is 16 bit */
+  unsigned maxnumcolors = 257;
+  if(bpp <= 8) maxnumcolors = LODEPNG_MIN(257, stats->numcolors + (1u << bpp));
+
+  stats->numpixels += numpixels;
+
+  /*if palette not allowed, no need to compute numcolors*/
+  if(!stats->allow_palette) numcolors_done = 1;
+
+  color_tree_init(&tree);
+
+  /*If the stats was already filled in from previous data, fill its palette in tree
+  and mark things as done already if we know they are the most expensive case already*/
+  if(stats->alpha) alpha_done = 1;
+  if(stats->colored) colored_done = 1;
+  if(stats->bits == 16) numcolors_done = 1;
+  if(stats->bits >= bpp) bits_done = 1;
+  if(stats->numcolors >= maxnumcolors) numcolors_done = 1;
+
+  if(!numcolors_done) {
+    for(i = 0; i < stats->numcolors; i++) {
+      const unsigned char* color = &stats->palette[i * 4];
+      error = color_tree_add(&tree, color[0], color[1], color[2], color[3], i);
+      if(error) goto cleanup;
+    }
+  }
+
+  /*Check if the 16-bit input is truly 16-bit*/
+  if(mode_in->bitdepth == 16 && !sixteen) {
+    unsigned short r = 0, g = 0, b = 0, a = 0;
+    for(i = 0; i != numpixels; ++i) {
+      getPixelColorRGBA16(&r, &g, &b, &a, in, i, mode_in);
+      if((r & 255) != ((r >> 8) & 255) || (g & 255) != ((g >> 8) & 255) ||
+         (b & 255) != ((b >> 8) & 255) || (a & 255) != ((a >> 8) & 255)) /*first and second byte differ*/ {
+        stats->bits = 16;
+        sixteen = 1;
+        bits_done = 1;
+        numcolors_done = 1; /*counting colors no longer useful, palette doesn't support 16-bit*/
+        break;
+      }
+    }
+  }
+
+  if(sixteen) {
+    unsigned short r = 0, g = 0, b = 0, a = 0;
+
+    for(i = 0; i != numpixels; ++i) {
+      getPixelColorRGBA16(&r, &g, &b, &a, in, i, mode_in);
+
+      if(!colored_done && (r != g || r != b)) {
+        stats->colored = 1;
+        colored_done = 1;
+      }
+
+      if(!alpha_done) {
+        unsigned matchkey = (r == stats->key_r && g == stats->key_g && b == stats->key_b);
+        if(a != 65535 && (a != 0 || (stats->key && !matchkey))) {
+          stats->alpha = 1;
+          stats->key = 0;
+          alpha_done = 1;
+        } else if(a == 0 && !stats->alpha && !stats->key) {
+          stats->key = 1;
+          stats->key_r = r;
+          stats->key_g = g;
+          stats->key_b = b;
+        } else if(a == 65535 && stats->key && matchkey) {
+          /* Color key cannot be used if an opaque pixel also has that RGB color. */
+          stats->alpha = 1;
+          stats->key = 0;
+          alpha_done = 1;
+        }
+      }
+      if(alpha_done && numcolors_done && colored_done && bits_done) break;
+    }
+
+    if(stats->key && !stats->alpha) {
+      for(i = 0; i != numpixels; ++i) {
+        getPixelColorRGBA16(&r, &g, &b, &a, in, i, mode_in);
+        if(a != 0 && r == stats->key_r && g == stats->key_g && b == stats->key_b) {
+          /* Color key cannot be used if an opaque pixel also has that RGB color. */
+          stats->alpha = 1;
+          stats->key = 0;
+          alpha_done = 1;
+        }
+      }
+    }
+  } else /* < 16-bit */ {
+    unsigned char r = 0, g = 0, b = 0, a = 0;
+    for(i = 0; i != numpixels; ++i) {
+      getPixelColorRGBA8(&r, &g, &b, &a, in, i, mode_in);
+
+      if(!bits_done && stats->bits < 8) {
+        /*only r is checked, < 8 bits is only relevant for grayscale*/
+        unsigned bits = getValueRequiredBits(r);
+        if(bits > stats->bits) stats->bits = bits;
+      }
+      bits_done = (stats->bits >= bpp);
+
+      if(!colored_done && (r != g || r != b)) {
+        stats->colored = 1;
+        colored_done = 1;
+        if(stats->bits < 8) stats->bits = 8; /*PNG has no colored modes with less than 8-bit per channel*/
+      }
+
+      if(!alpha_done) {
+        unsigned matchkey = (r == stats->key_r && g == stats->key_g && b == stats->key_b);
+        if(a != 255 && (a != 0 || (stats->key && !matchkey))) {
+          stats->alpha = 1;
+          stats->key = 0;
+          alpha_done = 1;
+          if(stats->bits < 8) stats->bits = 8; /*PNG has no alphachannel modes with less than 8-bit per channel*/
+        } else if(a == 0 && !stats->alpha && !stats->key) {
+          stats->key = 1;
+          stats->key_r = r;
+          stats->key_g = g;
+          stats->key_b = b;
+        } else if(a == 255 && stats->key && matchkey) {
+          /* Color key cannot be used if an opaque pixel also has that RGB color. */
+          stats->alpha = 1;
+          stats->key = 0;
+          alpha_done = 1;
+          if(stats->bits < 8) stats->bits = 8; /*PNG has no alphachannel modes with less than 8-bit per channel*/
+        }
+      }
+
+      if(!numcolors_done) {
+        if(!color_tree_has(&tree, r, g, b, a)) {
+          error = color_tree_add(&tree, r, g, b, a, stats->numcolors);
+          if(error) goto cleanup;
+          if(stats->numcolors < 256) {
+            unsigned char* p = stats->palette;
+            unsigned n = stats->numcolors;
+            p[n * 4 + 0] = r;
+            p[n * 4 + 1] = g;
+            p[n * 4 + 2] = b;
+            p[n * 4 + 3] = a;
+          }
+          ++stats->numcolors;
+          numcolors_done = stats->numcolors >= maxnumcolors;
+        }
+      }
+
+      if(alpha_done && numcolors_done && colored_done && bits_done) break;
+    }
+
+    if(stats->key && !stats->alpha) {
+      for(i = 0; i != numpixels; ++i) {
+        getPixelColorRGBA8(&r, &g, &b, &a, in, i, mode_in);
+        if(a != 0 && r == stats->key_r && g == stats->key_g && b == stats->key_b) {
+          /* Color key cannot be used if an opaque pixel also has that RGB color. */
+          stats->alpha = 1;
+          stats->key = 0;
+          alpha_done = 1;
+          if(stats->bits < 8) stats->bits = 8; /*PNG has no alphachannel modes with less than 8-bit per channel*/
+        }
+      }
+    }
+
+    /*make the stats's key always 16-bit for consistency - repeat each byte twice*/
+    stats->key_r += (stats->key_r << 8);
+    stats->key_g += (stats->key_g << 8);
+    stats->key_b += (stats->key_b << 8);
+  }
+
+cleanup:
+  color_tree_cleanup(&tree);
+  return error;
+}
+
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+/*Adds a single color to the color stats. The stats must already have been inited. The color must be given as 16-bit
+(with 2 bytes repeating for 8-bit and 65535 for opaque alpha channel). This function is expensive, do not call it for
+all pixels of an image but only for a few additional values. */
+static unsigned lodepng_color_stats_add(LodePNGColorStats* stats,
+                                        unsigned r, unsigned g, unsigned b, unsigned a) {
+  unsigned error = 0;
+  unsigned char image[8];
+  LodePNGColorMode mode;
+  lodepng_color_mode_init(&mode);
+  image[0] = r >> 8; image[1] = r; image[2] = g >> 8; image[3] = g;
+  image[4] = b >> 8; image[5] = b; image[6] = a >> 8; image[7] = a;
+  mode.bitdepth = 16;
+  mode.colortype = LCT_RGBA;
+  error = lodepng_compute_color_stats(stats, image, 1, 1, &mode);
+  lodepng_color_mode_cleanup(&mode);
+  return error;
+}
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+
+/*Computes a minimal PNG color model that can contain all colors as indicated by the stats.
+The stats should be computed with lodepng_compute_color_stats.
+mode_in is raw color profile of the image the stats were computed on, to copy palette order from when relevant.
+Minimal PNG color model means the color type and bit depth that gives smallest amount of bits in the output image,
+e.g. gray if only grayscale pixels, palette if less than 256 colors, color key if only single transparent color, ...
+This is used if auto_convert is enabled (it is by default).
+*/
+static unsigned auto_choose_color(LodePNGColorMode* mode_out,
+                                  const LodePNGColorMode* mode_in,
+                                  const LodePNGColorStats* stats) {
+  unsigned error = 0;
+  unsigned palettebits;
+  size_t i, n;
+  size_t numpixels = stats->numpixels;
+  unsigned palette_ok, gray_ok;
+
+  unsigned alpha = stats->alpha;
+  unsigned key = stats->key;
+  unsigned bits = stats->bits;
+
+  mode_out->key_defined = 0;
+
+  if(key && numpixels <= 16) {
+    alpha = 1; /*too few pixels to justify tRNS chunk overhead*/
+    key = 0;
+    if(bits < 8) bits = 8; /*PNG has no alphachannel modes with less than 8-bit per channel*/
+  }
+
+  gray_ok = !stats->colored;
+  if(!stats->allow_greyscale) gray_ok = 0;
+  if(!gray_ok && bits < 8) bits = 8;
+
+  n = stats->numcolors;
+  palettebits = n <= 2 ? 1 : (n <= 4 ? 2 : (n <= 16 ? 4 : 8));
+  palette_ok = n <= 256 && bits <= 8 && n != 0; /*n==0 means likely numcolors wasn't computed*/
+  if(numpixels < n * 2) palette_ok = 0; /*don't add palette overhead if image has only a few pixels*/
+  if(gray_ok && !alpha && bits <= palettebits) palette_ok = 0; /*gray is less overhead*/
+  if(!stats->allow_palette) palette_ok = 0;
+
+  if(palette_ok) {
+    const unsigned char* p = stats->palette;
+    lodepng_palette_clear(mode_out); /*remove potential earlier palette*/
+    for(i = 0; i != stats->numcolors; ++i) {
+      error = lodepng_palette_add(mode_out, p[i * 4 + 0], p[i * 4 + 1], p[i * 4 + 2], p[i * 4 + 3]);
+      if(error) break;
+    }
+
+    mode_out->colortype = LCT_PALETTE;
+    mode_out->bitdepth = palettebits;
+
+    if(mode_in->colortype == LCT_PALETTE && mode_in->palettesize >= mode_out->palettesize
+        && mode_in->bitdepth == mode_out->bitdepth) {
+      /*If input should have same palette colors, keep original to preserve its order and prevent conversion*/
+      lodepng_color_mode_cleanup(mode_out);
+      lodepng_color_mode_copy(mode_out, mode_in);
+    }
+  } else /*8-bit or 16-bit per channel*/ {
+    mode_out->bitdepth = bits;
+    mode_out->colortype = alpha ? (gray_ok ? LCT_GREY_ALPHA : LCT_RGBA)
+                                : (gray_ok ? LCT_GREY : LCT_RGB);
+    if(key) {
+      unsigned mask = (1u << mode_out->bitdepth) - 1u; /*stats always uses 16-bit, mask converts it*/
+      mode_out->key_r = stats->key_r & mask;
+      mode_out->key_g = stats->key_g & mask;
+      mode_out->key_b = stats->key_b & mask;
+      mode_out->key_defined = 1;
+    }
+  }
+
+  return error;
+}
+
+#endif /* #ifdef LODEPNG_COMPILE_ENCODER */
+
+/*
+Paeth predictor, used by PNG filter type 4
+The parameters are of type short, but should come from unsigned chars, the shorts
+are only needed to make the paeth calculation correct.
+*/
+static unsigned char paethPredictor(short a, short b, short c) {
+  short pa = LODEPNG_ABS(b - c);
+  short pb = LODEPNG_ABS(a - c);
+  short pc = LODEPNG_ABS(a + b - c - c);
+  /* return input value associated with smallest of pa, pb, pc (with certain priority if equal) */
+  if(pb < pa) { a = b; pa = pb; }
+  return (pc < pa) ? c : a;
+}
+
+/*shared values used by multiple Adam7 related functions*/
+
+static const unsigned ADAM7_IX[7] = { 0, 4, 0, 2, 0, 1, 0 }; /*x start values*/
+static const unsigned ADAM7_IY[7] = { 0, 0, 4, 0, 2, 0, 1 }; /*y start values*/
+static const unsigned ADAM7_DX[7] = { 8, 8, 4, 4, 2, 2, 1 }; /*x delta values*/
+static const unsigned ADAM7_DY[7] = { 8, 8, 8, 4, 4, 2, 2 }; /*y delta values*/
+
+/*
+Outputs various dimensions and positions in the image related to the Adam7 reduced images.
+passw: output containing the width of the 7 passes
+passh: output containing the height of the 7 passes
+filter_passstart: output containing the index of the start and end of each
+ reduced image with filter bytes
+padded_passstart output containing the index of the start and end of each
+ reduced image when without filter bytes but with padded scanlines
+passstart: output containing the index of the start and end of each reduced
+ image without padding between scanlines, but still padding between the images
+w, h: width and height of non-interlaced image
+bpp: bits per pixel
+"padded" is only relevant if bpp is less than 8 and a scanline or image does not
+ end at a full byte
+*/
+static void Adam7_getpassvalues(unsigned passw[7], unsigned passh[7], size_t filter_passstart[8],
+                                size_t padded_passstart[8], size_t passstart[8], unsigned w, unsigned h, unsigned bpp) {
+  /*the passstart values have 8 values: the 8th one indicates the byte after the end of the 7th (= last) pass*/
+  unsigned i;
+
+  /*calculate width and height in pixels of each pass*/
+  for(i = 0; i != 7; ++i) {
+    passw[i] = (w + ADAM7_DX[i] - ADAM7_IX[i] - 1) / ADAM7_DX[i];
+    passh[i] = (h + ADAM7_DY[i] - ADAM7_IY[i] - 1) / ADAM7_DY[i];
+    if(passw[i] == 0) passh[i] = 0;
+    if(passh[i] == 0) passw[i] = 0;
+  }
+
+  filter_passstart[0] = padded_passstart[0] = passstart[0] = 0;
+  for(i = 0; i != 7; ++i) {
+    /*if passw[i] is 0, it's 0 bytes, not 1 (no filtertype-byte)*/
+    filter_passstart[i + 1] = filter_passstart[i]
+                            + ((passw[i] && passh[i]) ? passh[i] * (1u + (passw[i] * bpp + 7u) / 8u) : 0);
+    /*bits padded if needed to fill full byte at end of each scanline*/
+    padded_passstart[i + 1] = padded_passstart[i] + passh[i] * ((passw[i] * bpp + 7u) / 8u);
+    /*only padded at end of reduced image*/
+    passstart[i + 1] = passstart[i] + (passh[i] * passw[i] * bpp + 7u) / 8u;
+  }
+}
+
+#ifdef LODEPNG_COMPILE_DECODER
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / PNG Decoder                                                            / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*read the information from the header and store it in the LodePNGInfo. return value is error*/
+unsigned lodepng_inspect(unsigned* w, unsigned* h, LodePNGState* state,
+                         const unsigned char* in, size_t insize) {
+  unsigned width, height;
+  LodePNGInfo* info = &state->info_png;
+  if(insize == 0 || in == 0) {
+    CERROR_RETURN_ERROR(state->error, 48); /*error: the given data is empty*/
+  }
+  if(insize < 33) {
+    CERROR_RETURN_ERROR(state->error, 27); /*error: the data length is smaller than the length of a PNG header*/
+  }
+
+  /*when decoding a new PNG image, make sure all parameters created after previous decoding are reset*/
+  /* TODO: remove this. One should use a new LodePNGState for new sessions */
+  lodepng_info_cleanup(info);
+  lodepng_info_init(info);
+
+  if(in[0] != 137 || in[1] != 80 || in[2] != 78 || in[3] != 71
+     || in[4] != 13 || in[5] != 10 || in[6] != 26 || in[7] != 10) {
+    CERROR_RETURN_ERROR(state->error, 28); /*error: the first 8 bytes are not the correct PNG signature*/
+  }
+  if(lodepng_chunk_length(in + 8) != 13) {
+    CERROR_RETURN_ERROR(state->error, 94); /*error: header size must be 13 bytes*/
+  }
+  if(!lodepng_chunk_type_equals(in + 8, "IHDR")) {
+    CERROR_RETURN_ERROR(state->error, 29); /*error: it doesn't start with a IHDR chunk!*/
+  }
+
+  /*read the values given in the header*/
+  width = lodepng_read32bitInt(&in[16]);
+  height = lodepng_read32bitInt(&in[20]);
+  /*TODO: remove the undocumented feature that allows to give null pointers to width or height*/
+  if(w) *w = width;
+  if(h) *h = height;
+  info->color.bitdepth = in[24];
+  info->color.colortype = (LodePNGColorType)in[25];
+  info->compression_method = in[26];
+  info->filter_method = in[27];
+  info->interlace_method = in[28];
+
+  /*errors returned only after the parsing so other values are still output*/
+
+  /*error: invalid image size*/
+  if(width == 0 || height == 0) CERROR_RETURN_ERROR(state->error, 93);
+  /*error: invalid colortype or bitdepth combination*/
+  state->error = checkColorValidity(info->color.colortype, info->color.bitdepth);
+  if(state->error) return state->error;
+  /*error: only compression method 0 is allowed in the specification*/
+  if(info->compression_method != 0) CERROR_RETURN_ERROR(state->error, 32);
+  /*error: only filter method 0 is allowed in the specification*/
+  if(info->filter_method != 0) CERROR_RETURN_ERROR(state->error, 33);
+  /*error: only interlace methods 0 and 1 exist in the specification*/
+  if(info->interlace_method > 1) CERROR_RETURN_ERROR(state->error, 34);
+
+  if(!state->decoder.ignore_crc) {
+    unsigned CRC = lodepng_read32bitInt(&in[29]);
+    unsigned checksum = lodepng_crc32(&in[12], 17);
+    if(CRC != checksum) {
+      CERROR_RETURN_ERROR(state->error, 57); /*invalid CRC*/
+    }
+  }
+
+  return state->error;
+}
+
+static unsigned unfilterScanline(unsigned char* recon, const unsigned char* scanline, const unsigned char* precon,
+                                 size_t bytewidth, unsigned char filterType, size_t length) {
+  /*
+  For PNG filter method 0
+  unfilter a PNG image scanline by scanline. when the pixels are smaller than 1 byte,
+  the filter works byte per byte (bytewidth = 1)
+  precon is the previous unfiltered scanline, recon the result, scanline the current one
+  the incoming scanlines do NOT include the filtertype byte, that one is given in the parameter filterType instead
+  recon and scanline MAY be the same memory address! precon must be disjoint.
+  */
+
+  size_t i;
+  switch(filterType) {
+    case 0:
+      for(i = 0; i != length; ++i) recon[i] = scanline[i];
+      break;
+    case 1:
+      for(i = 0; i != bytewidth; ++i) recon[i] = scanline[i];
+      for(i = bytewidth; i < length; ++i) recon[i] = scanline[i] + recon[i - bytewidth];
+      break;
+    case 2:
+      if(precon) {
+        for(i = 0; i != length; ++i) recon[i] = scanline[i] + precon[i];
+      } else {
+        for(i = 0; i != length; ++i) recon[i] = scanline[i];
+      }
+      break;
+    case 3:
+      if(precon) {
+        for(i = 0; i != bytewidth; ++i) recon[i] = scanline[i] + (precon[i] >> 1u);
+        for(i = bytewidth; i < length; ++i) recon[i] = scanline[i] + ((recon[i - bytewidth] + precon[i]) >> 1u);
+      } else {
+        for(i = 0; i != bytewidth; ++i) recon[i] = scanline[i];
+        for(i = bytewidth; i < length; ++i) recon[i] = scanline[i] + (recon[i - bytewidth] >> 1u);
+      }
+      break;
+    case 4:
+      if(precon) {
+        for(i = 0; i != bytewidth; ++i) {
+          recon[i] = (scanline[i] + precon[i]); /*paethPredictor(0, precon[i], 0) is always precon[i]*/
+        }
+
+        /* Unroll independent paths of the paeth predictor. A 6x and 8x version would also be possible but that
+        adds too much code. Whether this actually speeds anything up at all depends on compiler and settings. */
+        if(bytewidth >= 4) {
+          for(; i + 3 < length; i += 4) {
+            size_t j = i - bytewidth;
+            unsigned char s0 = scanline[i + 0], s1 = scanline[i + 1], s2 = scanline[i + 2], s3 = scanline[i + 3];
+            unsigned char r0 = recon[j + 0], r1 = recon[j + 1], r2 = recon[j + 2], r3 = recon[j + 3];
+            unsigned char p0 = precon[i + 0], p1 = precon[i + 1], p2 = precon[i + 2], p3 = precon[i + 3];
+            unsigned char q0 = precon[j + 0], q1 = precon[j + 1], q2 = precon[j + 2], q3 = precon[j + 3];
+            recon[i + 0] = s0 + paethPredictor(r0, p0, q0);
+            recon[i + 1] = s1 + paethPredictor(r1, p1, q1);
+            recon[i + 2] = s2 + paethPredictor(r2, p2, q2);
+            recon[i + 3] = s3 + paethPredictor(r3, p3, q3);
+          }
+        } else if(bytewidth >= 3) {
+          for(; i + 2 < length; i += 3) {
+            size_t j = i - bytewidth;
+            unsigned char s0 = scanline[i + 0], s1 = scanline[i + 1], s2 = scanline[i + 2];
+            unsigned char r0 = recon[j + 0], r1 = recon[j + 1], r2 = recon[j + 2];
+            unsigned char p0 = precon[i + 0], p1 = precon[i + 1], p2 = precon[i + 2];
+            unsigned char q0 = precon[j + 0], q1 = precon[j + 1], q2 = precon[j + 2];
+            recon[i + 0] = s0 + paethPredictor(r0, p0, q0);
+            recon[i + 1] = s1 + paethPredictor(r1, p1, q1);
+            recon[i + 2] = s2 + paethPredictor(r2, p2, q2);
+          }
+        } else if(bytewidth >= 2) {
+          for(; i + 1 < length; i += 2) {
+            size_t j = i - bytewidth;
+            unsigned char s0 = scanline[i + 0], s1 = scanline[i + 1];
+            unsigned char r0 = recon[j + 0], r1 = recon[j + 1];
+            unsigned char p0 = precon[i + 0], p1 = precon[i + 1];
+            unsigned char q0 = precon[j + 0], q1 = precon[j + 1];
+            recon[i + 0] = s0 + paethPredictor(r0, p0, q0);
+            recon[i + 1] = s1 + paethPredictor(r1, p1, q1);
+          }
+        }
+
+        for(; i != length; ++i) {
+          recon[i] = (scanline[i] + paethPredictor(recon[i - bytewidth], precon[i], precon[i - bytewidth]));
+        }
+      } else {
+        for(i = 0; i != bytewidth; ++i) {
+          recon[i] = scanline[i];
+        }
+        for(i = bytewidth; i < length; ++i) {
+          /*paethPredictor(recon[i - bytewidth], 0, 0) is always recon[i - bytewidth]*/
+          recon[i] = (scanline[i] + recon[i - bytewidth]);
+        }
+      }
+      break;
+    default: return 36; /*error: invalid filter type given*/
+  }
+  return 0;
+}
+
+static unsigned unfilter(unsigned char* out, const unsigned char* in, unsigned w, unsigned h, unsigned bpp) {
+  /*
+  For PNG filter method 0
+  this function unfilters a single image (e.g. without interlacing this is called once, with Adam7 seven times)
+  out must have enough bytes allocated already, in must have the scanlines + 1 filtertype byte per scanline
+  w and h are image dimensions or dimensions of reduced image, bpp is bits per pixel
+  in and out are allowed to be the same memory address (but aren't the same size since in has the extra filter bytes)
+  */
+
+  unsigned y;
+  unsigned char* prevline = 0;
+
+  /*bytewidth is used for filtering, is 1 when bpp < 8, number of bytes per pixel otherwise*/
+  size_t bytewidth = (bpp + 7u) / 8u;
+  /*the width of a scanline in bytes, not including the filter type*/
+  size_t linebytes = lodepng_get_raw_size_idat(w, 1, bpp) - 1u;
+
+  for(y = 0; y < h; ++y) {
+    size_t outindex = linebytes * y;
+    size_t inindex = (1 + linebytes) * y; /*the extra filterbyte added to each row*/
+    unsigned char filterType = in[inindex];
+
+    CERROR_TRY_RETURN(unfilterScanline(&out[outindex], &in[inindex + 1], prevline, bytewidth, filterType, linebytes));
+
+    prevline = &out[outindex];
+  }
+
+  return 0;
+}
+
+/*
+in: Adam7 interlaced image, with no padding bits between scanlines, but between
+ reduced images so that each reduced image starts at a byte.
+out: the same pixels, but re-ordered so that they're now a non-interlaced image with size w*h
+bpp: bits per pixel
+out has the following size in bits: w * h * bpp.
+in is possibly bigger due to padding bits between reduced images.
+out must be big enough AND must be 0 everywhere if bpp < 8 in the current implementation
+(because that's likely a little bit faster)
+NOTE: comments about padding bits are only relevant if bpp < 8
+*/
+static void Adam7_deinterlace(unsigned char* out, const unsigned char* in, unsigned w, unsigned h, unsigned bpp) {
+  unsigned passw[7], passh[7];
+  size_t filter_passstart[8], padded_passstart[8], passstart[8];
+  unsigned i;
+
+  Adam7_getpassvalues(passw, passh, filter_passstart, padded_passstart, passstart, w, h, bpp);
+
+  if(bpp >= 8) {
+    for(i = 0; i != 7; ++i) {
+      unsigned x, y, b;
+      size_t bytewidth = bpp / 8u;
+      for(y = 0; y < passh[i]; ++y)
+      for(x = 0; x < passw[i]; ++x) {
+        size_t pixelinstart = passstart[i] + (y * passw[i] + x) * bytewidth;
+        size_t pixeloutstart = ((ADAM7_IY[i] + (size_t)y * ADAM7_DY[i]) * (size_t)w
+                             + ADAM7_IX[i] + (size_t)x * ADAM7_DX[i]) * bytewidth;
+        for(b = 0; b < bytewidth; ++b) {
+          out[pixeloutstart + b] = in[pixelinstart + b];
+        }
+      }
+    }
+  } else /*bpp < 8: Adam7 with pixels < 8 bit is a bit trickier: with bit pointers*/ {
+    for(i = 0; i != 7; ++i) {
+      unsigned x, y, b;
+      unsigned ilinebits = bpp * passw[i];
+      unsigned olinebits = bpp * w;
+      size_t obp, ibp; /*bit pointers (for out and in buffer)*/
+      for(y = 0; y < passh[i]; ++y)
+      for(x = 0; x < passw[i]; ++x) {
+        ibp = (8 * passstart[i]) + (y * ilinebits + x * bpp);
+        obp = (ADAM7_IY[i] + (size_t)y * ADAM7_DY[i]) * olinebits + (ADAM7_IX[i] + (size_t)x * ADAM7_DX[i]) * bpp;
+        for(b = 0; b < bpp; ++b) {
+          unsigned char bit = readBitFromReversedStream(&ibp, in);
+          setBitOfReversedStream(&obp, out, bit);
+        }
+      }
+    }
+  }
+}
+
+static void removePaddingBits(unsigned char* out, const unsigned char* in,
+                              size_t olinebits, size_t ilinebits, unsigned h) {
+  /*
+  After filtering there are still padding bits if scanlines have non multiple of 8 bit amounts. They need
+  to be removed (except at last scanline of (Adam7-reduced) image) before working with pure image buffers
+  for the Adam7 code, the color convert code and the output to the user.
+  in and out are allowed to be the same buffer, in may also be higher but still overlapping; in must
+  have >= ilinebits*h bits, out must have >= olinebits*h bits, olinebits must be <= ilinebits
+  also used to move bits after earlier such operations happened, e.g. in a sequence of reduced images from Adam7
+  only useful if (ilinebits - olinebits) is a value in the range 1..7
+  */
+  unsigned y;
+  size_t diff = ilinebits - olinebits;
+  size_t ibp = 0, obp = 0; /*input and output bit pointers*/
+  for(y = 0; y < h; ++y) {
+    size_t x;
+    for(x = 0; x < olinebits; ++x) {
+      unsigned char bit = readBitFromReversedStream(&ibp, in);
+      setBitOfReversedStream(&obp, out, bit);
+    }
+    ibp += diff;
+  }
+}
+
+/*out must be buffer big enough to contain full image, and in must contain the full decompressed data from
+the IDAT chunks (with filter index bytes and possible padding bits)
+return value is error*/
+static unsigned postProcessScanlines(unsigned char* out, unsigned char* in,
+                                     unsigned w, unsigned h, const LodePNGInfo* info_png) {
+  /*
+  This function converts the filtered-padded-interlaced data into pure 2D image buffer with the PNG's colortype.
+  Steps:
+  *) if no Adam7: 1) unfilter 2) remove padding bits (= possible extra bits per scanline if bpp < 8)
+  *) if adam7: 1) 7x unfilter 2) 7x remove padding bits 3) Adam7_deinterlace
+  NOTE: the in buffer will be overwritten with intermediate data!
+  */
+  unsigned bpp = lodepng_get_bpp(&info_png->color);
+  if(bpp == 0) return 31; /*error: invalid colortype*/
+
+  if(info_png->interlace_method == 0) {
+    if(bpp < 8 && w * bpp != ((w * bpp + 7u) / 8u) * 8u) {
+      CERROR_TRY_RETURN(unfilter(in, in, w, h, bpp));
+      removePaddingBits(out, in, w * bpp, ((w * bpp + 7u) / 8u) * 8u, h);
+    }
+    /*we can immediately filter into the out buffer, no other steps needed*/
+    else CERROR_TRY_RETURN(unfilter(out, in, w, h, bpp));
+  } else /*interlace_method is 1 (Adam7)*/ {
+    unsigned passw[7], passh[7]; size_t filter_passstart[8], padded_passstart[8], passstart[8];
+    unsigned i;
+
+    Adam7_getpassvalues(passw, passh, filter_passstart, padded_passstart, passstart, w, h, bpp);
+
+    for(i = 0; i != 7; ++i) {
+      CERROR_TRY_RETURN(unfilter(&in[padded_passstart[i]], &in[filter_passstart[i]], passw[i], passh[i], bpp));
+      /*TODO: possible efficiency improvement: if in this reduced image the bits fit nicely in 1 scanline,
+      move bytes instead of bits or move not at all*/
+      if(bpp < 8) {
+        /*remove padding bits in scanlines; after this there still may be padding
+        bits between the different reduced images: each reduced image still starts nicely at a byte*/
+        removePaddingBits(&in[passstart[i]], &in[padded_passstart[i]], passw[i] * bpp,
+                          ((passw[i] * bpp + 7u) / 8u) * 8u, passh[i]);
+      }
+    }
+
+    Adam7_deinterlace(out, in, w, h, bpp);
+  }
+
+  return 0;
+}
+
+static unsigned readChunk_PLTE(LodePNGColorMode* color, const unsigned char* data, size_t chunkLength) {
+  unsigned pos = 0, i;
+  color->palettesize = chunkLength / 3u;
+  if(color->palettesize == 0 || color->palettesize > 256) return 38; /*error: palette too small or big*/
+  lodepng_color_mode_alloc_palette(color);
+  if(!color->palette && color->palettesize) {
+    color->palettesize = 0;
+    return 83; /*alloc fail*/
+  }
+
+  for(i = 0; i != color->palettesize; ++i) {
+    color->palette[4 * i + 0] = data[pos++]; /*R*/
+    color->palette[4 * i + 1] = data[pos++]; /*G*/
+    color->palette[4 * i + 2] = data[pos++]; /*B*/
+    color->palette[4 * i + 3] = 255; /*alpha*/
+  }
+
+  return 0; /* OK */
+}
+
+static unsigned readChunk_tRNS(LodePNGColorMode* color, const unsigned char* data, size_t chunkLength) {
+  unsigned i;
+  if(color->colortype == LCT_PALETTE) {
+    /*error: more alpha values given than there are palette entries*/
+    if(chunkLength > color->palettesize) return 39;
+
+    for(i = 0; i != chunkLength; ++i) color->palette[4 * i + 3] = data[i];
+  } else if(color->colortype == LCT_GREY) {
+    /*error: this chunk must be 2 bytes for grayscale image*/
+    if(chunkLength != 2) return 30;
+
+    color->key_defined = 1;
+    color->key_r = color->key_g = color->key_b = 256u * data[0] + data[1];
+  } else if(color->colortype == LCT_RGB) {
+    /*error: this chunk must be 6 bytes for RGB image*/
+    if(chunkLength != 6) return 41;
+
+    color->key_defined = 1;
+    color->key_r = 256u * data[0] + data[1];
+    color->key_g = 256u * data[2] + data[3];
+    color->key_b = 256u * data[4] + data[5];
+  }
+  else return 42; /*error: tRNS chunk not allowed for other color models*/
+
+  return 0; /* OK */
+}
+
+
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+/*background color chunk (bKGD)*/
+static unsigned readChunk_bKGD(LodePNGInfo* info, const unsigned char* data, size_t chunkLength) {
+  if(info->color.colortype == LCT_PALETTE) {
+    /*error: this chunk must be 1 byte for indexed color image*/
+    if(chunkLength != 1) return 43;
+
+    /*error: invalid palette index, or maybe this chunk appeared before PLTE*/
+    if(data[0] >= info->color.palettesize) return 103;
+
+    info->background_defined = 1;
+    info->background_r = info->background_g = info->background_b = data[0];
+  } else if(info->color.colortype == LCT_GREY || info->color.colortype == LCT_GREY_ALPHA) {
+    /*error: this chunk must be 2 bytes for grayscale image*/
+    if(chunkLength != 2) return 44;
+
+    /*the values are truncated to bitdepth in the PNG file*/
+    info->background_defined = 1;
+    info->background_r = info->background_g = info->background_b = 256u * data[0] + data[1];
+  } else if(info->color.colortype == LCT_RGB || info->color.colortype == LCT_RGBA) {
+    /*error: this chunk must be 6 bytes for grayscale image*/
+    if(chunkLength != 6) return 45;
+
+    /*the values are truncated to bitdepth in the PNG file*/
+    info->background_defined = 1;
+    info->background_r = 256u * data[0] + data[1];
+    info->background_g = 256u * data[2] + data[3];
+    info->background_b = 256u * data[4] + data[5];
+  }
+
+  return 0; /* OK */
+}
+
+/*text chunk (tEXt)*/
+static unsigned readChunk_tEXt(LodePNGInfo* info, const unsigned char* data, size_t chunkLength) {
+  unsigned error = 0;
+  char *key = 0, *str = 0;
+
+  while(!error) /*not really a while loop, only used to break on error*/ {
+    unsigned length, string2_begin;
+
+    length = 0;
+    while(length < chunkLength && data[length] != 0) ++length;
+    /*even though it's not allowed by the standard, no error is thrown if
+    there's no null termination char, if the text is empty*/
+    if(length < 1 || length > 79) CERROR_BREAK(error, 89); /*keyword too short or long*/
+
+    key = (char*)lodepng_malloc(length + 1);
+    if(!key) CERROR_BREAK(error, 83); /*alloc fail*/
+
+    lodepng_memcpy(key, data, length);
+    key[length] = 0;
+
+    string2_begin = length + 1; /*skip keyword null terminator*/
+
+    length = (unsigned)(chunkLength < string2_begin ? 0 : chunkLength - string2_begin);
+    str = (char*)lodepng_malloc(length + 1);
+    if(!str) CERROR_BREAK(error, 83); /*alloc fail*/
+
+    lodepng_memcpy(str, data + string2_begin, length);
+    str[length] = 0;
+
+    error = lodepng_add_text(info, key, str);
+
+    break;
+  }
+
+  lodepng_free(key);
+  lodepng_free(str);
+
+  return error;
+}
+
+/*compressed text chunk (zTXt)*/
+static unsigned readChunk_zTXt(LodePNGInfo* info, const LodePNGDecompressSettings* zlibsettings,
+                               const unsigned char* data, size_t chunkLength) {
+  unsigned error = 0;
+
+  unsigned length, string2_begin;
+  char *key = 0;
+  unsigned char* str = 0;
+  size_t size = 0;
+
+  while(!error) /*not really a while loop, only used to break on error*/ {
+    for(length = 0; length < chunkLength && data[length] != 0; ++length) ;
+    if(length + 2 >= chunkLength) CERROR_BREAK(error, 75); /*no null termination, corrupt?*/
+    if(length < 1 || length > 79) CERROR_BREAK(error, 89); /*keyword too short or long*/
+
+    key = (char*)lodepng_malloc(length + 1);
+    if(!key) CERROR_BREAK(error, 83); /*alloc fail*/
+
+    lodepng_memcpy(key, data, length);
+    key[length] = 0;
+
+    if(data[length + 1] != 0) CERROR_BREAK(error, 72); /*the 0 byte indicating compression must be 0*/
+
+    string2_begin = length + 2;
+    if(string2_begin > chunkLength) CERROR_BREAK(error, 75); /*no null termination, corrupt?*/
+
+    length = (unsigned)chunkLength - string2_begin;
+    /*will fail if zlib error, e.g. if length is too small*/
+    error = zlib_decompress(&str, &size, 0, &data[string2_begin],
+                            length, zlibsettings);
+    if(error) break;
+    error = lodepng_add_text_sized(info, key, (char*)str, size);
+
+    break;
+  }
+
+  lodepng_free(key);
+  lodepng_free(str);
+
+  return error;
+}
+
+/*international text chunk (iTXt)*/
+static unsigned readChunk_iTXt(LodePNGInfo* info, const LodePNGDecompressSettings* zlibsettings,
+                               const unsigned char* data, size_t chunkLength) {
+  unsigned error = 0;
+  unsigned i;
+
+  unsigned length, begin, compressed;
+  char *key = 0, *langtag = 0, *transkey = 0;
+
+  while(!error) /*not really a while loop, only used to break on error*/ {
+    /*Quick check if the chunk length isn't too small. Even without check
+    it'd still fail with other error checks below if it's too short. This just gives a different error code.*/
+    if(chunkLength < 5) CERROR_BREAK(error, 30); /*iTXt chunk too short*/
+
+    /*read the key*/
+    for(length = 0; length < chunkLength && data[length] != 0; ++length) ;
+    if(length + 3 >= chunkLength) CERROR_BREAK(error, 75); /*no null termination char, corrupt?*/
+    if(length < 1 || length > 79) CERROR_BREAK(error, 89); /*keyword too short or long*/
+
+    key = (char*)lodepng_malloc(length + 1);
+    if(!key) CERROR_BREAK(error, 83); /*alloc fail*/
+
+    lodepng_memcpy(key, data, length);
+    key[length] = 0;
+
+    /*read the compression method*/
+    compressed = data[length + 1];
+    if(data[length + 2] != 0) CERROR_BREAK(error, 72); /*the 0 byte indicating compression must be 0*/
+
+    /*even though it's not allowed by the standard, no error is thrown if
+    there's no null termination char, if the text is empty for the next 3 texts*/
+
+    /*read the langtag*/
+    begin = length + 3;
+    length = 0;
+    for(i = begin; i < chunkLength && data[i] != 0; ++i) ++length;
+
+    langtag = (char*)lodepng_malloc(length + 1);
+    if(!langtag) CERROR_BREAK(error, 83); /*alloc fail*/
+
+    lodepng_memcpy(langtag, data + begin, length);
+    langtag[length] = 0;
+
+    /*read the transkey*/
+    begin += length + 1;
+    length = 0;
+    for(i = begin; i < chunkLength && data[i] != 0; ++i) ++length;
+
+    transkey = (char*)lodepng_malloc(length + 1);
+    if(!transkey) CERROR_BREAK(error, 83); /*alloc fail*/
+
+    lodepng_memcpy(transkey, data + begin, length);
+    transkey[length] = 0;
+
+    /*read the actual text*/
+    begin += length + 1;
+
+    length = (unsigned)chunkLength < begin ? 0 : (unsigned)chunkLength - begin;
+
+    if(compressed) {
+      unsigned char* str = 0;
+      size_t size = 0;
+      /*will fail if zlib error, e.g. if length is too small*/
+      error = zlib_decompress(&str, &size, 0, &data[begin],
+                              length, zlibsettings);
+      if(!error) error = lodepng_add_itext_sized(info, key, langtag, transkey, (char*)str, size);
+      lodepng_free(str);
+    } else {
+      error = lodepng_add_itext_sized(info, key, langtag, transkey, (char*)(data + begin), length);
+    }
+
+    break;
+  }
+
+  lodepng_free(key);
+  lodepng_free(langtag);
+  lodepng_free(transkey);
+
+  return error;
+}
+
+static unsigned readChunk_tIME(LodePNGInfo* info, const unsigned char* data, size_t chunkLength) {
+  if(chunkLength != 7) return 73; /*invalid tIME chunk size*/
+
+  info->time_defined = 1;
+  info->time.year = 256u * data[0] + data[1];
+  info->time.month = data[2];
+  info->time.day = data[3];
+  info->time.hour = data[4];
+  info->time.minute = data[5];
+  info->time.second = data[6];
+
+  return 0; /* OK */
+}
+
+static unsigned readChunk_pHYs(LodePNGInfo* info, const unsigned char* data, size_t chunkLength) {
+  if(chunkLength != 9) return 74; /*invalid pHYs chunk size*/
+
+  info->phys_defined = 1;
+  info->phys_x = 16777216u * data[0] + 65536u * data[1] + 256u * data[2] + data[3];
+  info->phys_y = 16777216u * data[4] + 65536u * data[5] + 256u * data[6] + data[7];
+  info->phys_unit = data[8];
+
+  return 0; /* OK */
+}
+
+static unsigned readChunk_gAMA(LodePNGInfo* info, const unsigned char* data, size_t chunkLength) {
+  if(chunkLength != 4) return 96; /*invalid gAMA chunk size*/
+
+  info->gama_defined = 1;
+  info->gama_gamma = 16777216u * data[0] + 65536u * data[1] + 256u * data[2] + data[3];
+
+  return 0; /* OK */
+}
+
+static unsigned readChunk_cHRM(LodePNGInfo* info, const unsigned char* data, size_t chunkLength) {
+  if(chunkLength != 32) return 97; /*invalid cHRM chunk size*/
+
+  info->chrm_defined = 1;
+  info->chrm_white_x = 16777216u * data[ 0] + 65536u * data[ 1] + 256u * data[ 2] + data[ 3];
+  info->chrm_white_y = 16777216u * data[ 4] + 65536u * data[ 5] + 256u * data[ 6] + data[ 7];
+  info->chrm_red_x   = 16777216u * data[ 8] + 65536u * data[ 9] + 256u * data[10] + data[11];
+  info->chrm_red_y   = 16777216u * data[12] + 65536u * data[13] + 256u * data[14] + data[15];
+  info->chrm_green_x = 16777216u * data[16] + 65536u * data[17] + 256u * data[18] + data[19];
+  info->chrm_green_y = 16777216u * data[20] + 65536u * data[21] + 256u * data[22] + data[23];
+  info->chrm_blue_x  = 16777216u * data[24] + 65536u * data[25] + 256u * data[26] + data[27];
+  info->chrm_blue_y  = 16777216u * data[28] + 65536u * data[29] + 256u * data[30] + data[31];
+
+  return 0; /* OK */
+}
+
+static unsigned readChunk_sRGB(LodePNGInfo* info, const unsigned char* data, size_t chunkLength) {
+  if(chunkLength != 1) return 98; /*invalid sRGB chunk size (this one is never ignored)*/
+
+  info->srgb_defined = 1;
+  info->srgb_intent = data[0];
+
+  return 0; /* OK */
+}
+
+static unsigned readChunk_iCCP(LodePNGInfo* info, const LodePNGDecompressSettings* zlibsettings,
+                               const unsigned char* data, size_t chunkLength) {
+  unsigned error = 0;
+  unsigned i;
+  size_t size = 0;
+
+  unsigned length, string2_begin;
+
+  info->iccp_defined = 1;
+  if(info->iccp_name) lodepng_clear_icc(info);
+
+  for(length = 0; length < chunkLength && data[length] != 0; ++length) ;
+  if(length + 2 >= chunkLength) return 75; /*no null termination, corrupt?*/
+  if(length < 1 || length > 79) return 89; /*keyword too short or long*/
+
+  info->iccp_name = (char*)lodepng_malloc(length + 1);
+  if(!info->iccp_name) return 83; /*alloc fail*/
+
+  info->iccp_name[length] = 0;
+  for(i = 0; i != length; ++i) info->iccp_name[i] = (char)data[i];
+
+  if(data[length + 1] != 0) return 72; /*the 0 byte indicating compression must be 0*/
+
+  string2_begin = length + 2;
+  if(string2_begin > chunkLength) return 75; /*no null termination, corrupt?*/
+
+  length = (unsigned)chunkLength - string2_begin;
+  error = zlib_decompress(&info->iccp_profile, &size, 0,
+                          &data[string2_begin],
+                          length, zlibsettings);
+  info->iccp_profile_size = size;
+  if(!error && !info->iccp_profile_size) error = 100; /*invalid ICC profile size*/
+  return error;
+}
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+
+unsigned lodepng_inspect_chunk(LodePNGState* state, size_t pos,
+                               const unsigned char* in, size_t insize) {
+  const unsigned char* chunk = in + pos;
+  unsigned chunkLength;
+  const unsigned char* data;
+  unsigned unhandled = 0;
+  unsigned error = 0;
+
+  if(pos + 4 > insize) return 30;
+  chunkLength = lodepng_chunk_length(chunk);
+  if(chunkLength > 2147483647) return 63;
+  data = lodepng_chunk_data_const(chunk);
+  if(data + chunkLength + 4 > in + insize) return 30;
+
+  if(lodepng_chunk_type_equals(chunk, "PLTE")) {
+    error = readChunk_PLTE(&state->info_png.color, data, chunkLength);
+  } else if(lodepng_chunk_type_equals(chunk, "tRNS")) {
+    error = readChunk_tRNS(&state->info_png.color, data, chunkLength);
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+  } else if(lodepng_chunk_type_equals(chunk, "bKGD")) {
+    error = readChunk_bKGD(&state->info_png, data, chunkLength);
+  } else if(lodepng_chunk_type_equals(chunk, "tEXt")) {
+    error = readChunk_tEXt(&state->info_png, data, chunkLength);
+  } else if(lodepng_chunk_type_equals(chunk, "zTXt")) {
+    error = readChunk_zTXt(&state->info_png, &state->decoder.zlibsettings, data, chunkLength);
+  } else if(lodepng_chunk_type_equals(chunk, "iTXt")) {
+    error = readChunk_iTXt(&state->info_png, &state->decoder.zlibsettings, data, chunkLength);
+  } else if(lodepng_chunk_type_equals(chunk, "tIME")) {
+    error = readChunk_tIME(&state->info_png, data, chunkLength);
+  } else if(lodepng_chunk_type_equals(chunk, "pHYs")) {
+    error = readChunk_pHYs(&state->info_png, data, chunkLength);
+  } else if(lodepng_chunk_type_equals(chunk, "gAMA")) {
+    error = readChunk_gAMA(&state->info_png, data, chunkLength);
+  } else if(lodepng_chunk_type_equals(chunk, "cHRM")) {
+    error = readChunk_cHRM(&state->info_png, data, chunkLength);
+  } else if(lodepng_chunk_type_equals(chunk, "sRGB")) {
+    error = readChunk_sRGB(&state->info_png, data, chunkLength);
+  } else if(lodepng_chunk_type_equals(chunk, "iCCP")) {
+    error = readChunk_iCCP(&state->info_png, &state->decoder.zlibsettings, data, chunkLength);
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+  } else {
+    /* unhandled chunk is ok (is not an error) */
+    unhandled = 1;
+  }
+
+  if(!error && !unhandled && !state->decoder.ignore_crc) {
+    if(lodepng_chunk_check_crc(chunk)) return 57; /*invalid CRC*/
+  }
+
+  return error;
+}
+
+/*read a PNG, the result will be in the same color type as the PNG (hence "generic")*/
+static void decodeGeneric(unsigned char** out, unsigned* w, unsigned* h,
+                          LodePNGState* state,
+                          const unsigned char* in, size_t insize) {
+  unsigned char IEND = 0;
+  const unsigned char* chunk;
+  unsigned char* idat; /*the data from idat chunks, zlib compressed*/
+  size_t idatsize = 0;
+  unsigned char* scanlines = 0;
+  size_t scanlines_size = 0, expected_size = 0;
+  size_t outsize = 0;
+
+  /*for unknown chunk order*/
+  unsigned unknown = 0;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+  unsigned critical_pos = 1; /*1 = after IHDR, 2 = after PLTE, 3 = after IDAT*/
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+
+
+  /* safe output values in case error happens */
+  *out = 0;
+  *w = *h = 0;
+
+  state->error = lodepng_inspect(w, h, state, in, insize); /*reads header and resets other parameters in state->info_png*/
+  if(state->error) return;
+
+  if(lodepng_pixel_overflow(*w, *h, &state->info_png.color, &state->info_raw)) {
+    CERROR_RETURN(state->error, 92); /*overflow possible due to amount of pixels*/
+  }
+
+  /*the input filesize is a safe upper bound for the sum of idat chunks size*/
+  idat = (unsigned char*)lodepng_malloc(insize);
+  if(!idat) CERROR_RETURN(state->error, 83); /*alloc fail*/
+
+  chunk = &in[33]; /*first byte of the first chunk after the header*/
+
+  /*loop through the chunks, ignoring unknown chunks and stopping at IEND chunk.
+  IDAT data is put at the start of the in buffer*/
+  while(!IEND && !state->error) {
+    unsigned chunkLength;
+    const unsigned char* data; /*the data in the chunk*/
+
+    /*error: size of the in buffer too small to contain next chunk*/
+    if((size_t)((chunk - in) + 12) > insize || chunk < in) {
+      if(state->decoder.ignore_end) break; /*other errors may still happen though*/
+      CERROR_BREAK(state->error, 30);
+    }
+
+    /*length of the data of the chunk, excluding the length bytes, chunk type and CRC bytes*/
+    chunkLength = lodepng_chunk_length(chunk);
+    /*error: chunk length larger than the max PNG chunk size*/
+    if(chunkLength > 2147483647) {
+      if(state->decoder.ignore_end) break; /*other errors may still happen though*/
+      CERROR_BREAK(state->error, 63);
+    }
+
+    if((size_t)((chunk - in) + chunkLength + 12) > insize || (chunk + chunkLength + 12) < in) {
+      CERROR_BREAK(state->error, 64); /*error: size of the in buffer too small to contain next chunk*/
+    }
+
+    data = lodepng_chunk_data_const(chunk);
+
+    unknown = 0;
+
+    /*IDAT chunk, containing compressed image data*/
+    if(lodepng_chunk_type_equals(chunk, "IDAT")) {
+      size_t newsize;
+      if(lodepng_addofl(idatsize, chunkLength, &newsize)) CERROR_BREAK(state->error, 95);
+      if(newsize > insize) CERROR_BREAK(state->error, 95);
+      lodepng_memcpy(idat + idatsize, data, chunkLength);
+      idatsize += chunkLength;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+      critical_pos = 3;
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+    } else if(lodepng_chunk_type_equals(chunk, "IEND")) {
+      /*IEND chunk*/
+      IEND = 1;
+    } else if(lodepng_chunk_type_equals(chunk, "PLTE")) {
+      /*palette chunk (PLTE)*/
+      state->error = readChunk_PLTE(&state->info_png.color, data, chunkLength);
+      if(state->error) break;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+      critical_pos = 2;
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+    } else if(lodepng_chunk_type_equals(chunk, "tRNS")) {
+      /*palette transparency chunk (tRNS). Even though this one is an ancillary chunk , it is still compiled
+      in without 'LODEPNG_COMPILE_ANCILLARY_CHUNKS' because it contains essential color information that
+      affects the alpha channel of pixels. */
+      state->error = readChunk_tRNS(&state->info_png.color, data, chunkLength);
+      if(state->error) break;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+      /*background color chunk (bKGD)*/
+    } else if(lodepng_chunk_type_equals(chunk, "bKGD")) {
+      state->error = readChunk_bKGD(&state->info_png, data, chunkLength);
+      if(state->error) break;
+    } else if(lodepng_chunk_type_equals(chunk, "tEXt")) {
+      /*text chunk (tEXt)*/
+      if(state->decoder.read_text_chunks) {
+        state->error = readChunk_tEXt(&state->info_png, data, chunkLength);
+        if(state->error) break;
+      }
+    } else if(lodepng_chunk_type_equals(chunk, "zTXt")) {
+      /*compressed text chunk (zTXt)*/
+      if(state->decoder.read_text_chunks) {
+        state->error = readChunk_zTXt(&state->info_png, &state->decoder.zlibsettings, data, chunkLength);
+        if(state->error) break;
+      }
+    } else if(lodepng_chunk_type_equals(chunk, "iTXt")) {
+      /*international text chunk (iTXt)*/
+      if(state->decoder.read_text_chunks) {
+        state->error = readChunk_iTXt(&state->info_png, &state->decoder.zlibsettings, data, chunkLength);
+        if(state->error) break;
+      }
+    } else if(lodepng_chunk_type_equals(chunk, "tIME")) {
+      state->error = readChunk_tIME(&state->info_png, data, chunkLength);
+      if(state->error) break;
+    } else if(lodepng_chunk_type_equals(chunk, "pHYs")) {
+      state->error = readChunk_pHYs(&state->info_png, data, chunkLength);
+      if(state->error) break;
+    } else if(lodepng_chunk_type_equals(chunk, "gAMA")) {
+      state->error = readChunk_gAMA(&state->info_png, data, chunkLength);
+      if(state->error) break;
+    } else if(lodepng_chunk_type_equals(chunk, "cHRM")) {
+      state->error = readChunk_cHRM(&state->info_png, data, chunkLength);
+      if(state->error) break;
+    } else if(lodepng_chunk_type_equals(chunk, "sRGB")) {
+      state->error = readChunk_sRGB(&state->info_png, data, chunkLength);
+      if(state->error) break;
+    } else if(lodepng_chunk_type_equals(chunk, "iCCP")) {
+      state->error = readChunk_iCCP(&state->info_png, &state->decoder.zlibsettings, data, chunkLength);
+      if(state->error) break;
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+    } else /*it's not an implemented chunk type, so ignore it: skip over the data*/ {
+      /*error: unknown critical chunk (5th bit of first byte of chunk type is 0)*/
+      if(!state->decoder.ignore_critical && !lodepng_chunk_ancillary(chunk)) {
+        CERROR_BREAK(state->error, 69);
+      }
+
+      unknown = 1;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+      if(state->decoder.remember_unknown_chunks) {
+        state->error = lodepng_chunk_append(&state->info_png.unknown_chunks_data[critical_pos - 1],
+                                            &state->info_png.unknown_chunks_size[critical_pos - 1], chunk);
+        if(state->error) break;
+      }
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+    }
+
+    if(!state->decoder.ignore_crc && !unknown) /*check CRC if wanted, only on known chunk types*/ {
+      if(lodepng_chunk_check_crc(chunk)) CERROR_BREAK(state->error, 57); /*invalid CRC*/
+    }
+
+    if(!IEND) chunk = lodepng_chunk_next_const(chunk, in + insize);
+  }
+
+  if(state->info_png.color.colortype == LCT_PALETTE && !state->info_png.color.palette) {
+    state->error = 106; /* error: PNG file must have PLTE chunk if color type is palette */
+  }
+
+  if(!state->error) {
+    /*predict output size, to allocate exact size for output buffer to avoid more dynamic allocation.
+    If the decompressed size does not match the prediction, the image must be corrupt.*/
+    if(state->info_png.interlace_method == 0) {
+      size_t bpp = lodepng_get_bpp(&state->info_png.color);
+      expected_size = lodepng_get_raw_size_idat(*w, *h, bpp);
+    } else {
+      size_t bpp = lodepng_get_bpp(&state->info_png.color);
+      /*Adam-7 interlaced: expected size is the sum of the 7 sub-images sizes*/
+      expected_size = 0;
+      expected_size += lodepng_get_raw_size_idat((*w + 7) >> 3, (*h + 7) >> 3, bpp);
+      if(*w > 4) expected_size += lodepng_get_raw_size_idat((*w + 3) >> 3, (*h + 7) >> 3, bpp);
+      expected_size += lodepng_get_raw_size_idat((*w + 3) >> 2, (*h + 3) >> 3, bpp);
+      if(*w > 2) expected_size += lodepng_get_raw_size_idat((*w + 1) >> 2, (*h + 3) >> 2, bpp);
+      expected_size += lodepng_get_raw_size_idat((*w + 1) >> 1, (*h + 1) >> 2, bpp);
+      if(*w > 1) expected_size += lodepng_get_raw_size_idat((*w + 0) >> 1, (*h + 1) >> 1, bpp);
+      expected_size += lodepng_get_raw_size_idat((*w + 0), (*h + 0) >> 1, bpp);
+    }
+
+    state->error = zlib_decompress(&scanlines, &scanlines_size, expected_size, idat, idatsize, &state->decoder.zlibsettings);
+  }
+  if(!state->error && scanlines_size != expected_size) state->error = 91; /*decompressed size doesn't match prediction*/
+  lodepng_free(idat);
+
+  if(!state->error) {
+    outsize = lodepng_get_raw_size(*w, *h, &state->info_png.color);
+    *out = (unsigned char*)lodepng_malloc(outsize);
+    if(!*out) state->error = 83; /*alloc fail*/
+  }
+  if(!state->error) {
+    lodepng_memset(*out, 0, outsize);
+    state->error = postProcessScanlines(*out, scanlines, *w, *h, &state->info_png);
+  }
+  lodepng_free(scanlines);
+}
+
+unsigned lodepng_decode(unsigned char** out, unsigned* w, unsigned* h,
+                        LodePNGState* state,
+                        const unsigned char* in, size_t insize) {
+  *out = 0;
+  decodeGeneric(out, w, h, state, in, insize);
+  if(state->error) return state->error;
+  if(!state->decoder.color_convert || lodepng_color_mode_equal(&state->info_raw, &state->info_png.color)) {
+    /*same color type, no copying or converting of data needed*/
+    /*store the info_png color settings on the info_raw so that the info_raw still reflects what colortype
+    the raw image has to the end user*/
+    if(!state->decoder.color_convert) {
+      state->error = lodepng_color_mode_copy(&state->info_raw, &state->info_png.color);
+      if(state->error) return state->error;
+    }
+  } else { /*color conversion needed*/
+    unsigned char* data = *out;
+    size_t outsize;
+
+    /*TODO: check if this works according to the statement in the documentation: "The converter can convert
+    from grayscale input color type, to 8-bit grayscale or grayscale with alpha"*/
+    if(!(state->info_raw.colortype == LCT_RGB || state->info_raw.colortype == LCT_RGBA)
+       && !(state->info_raw.bitdepth == 8)) {
+      return 56; /*unsupported color mode conversion*/
+    }
+
+    outsize = lodepng_get_raw_size(*w, *h, &state->info_raw);
+    *out = (unsigned char*)lodepng_malloc(outsize);
+    if(!(*out)) {
+      state->error = 83; /*alloc fail*/
+    }
+    else state->error = lodepng_convert(*out, data, &state->info_raw,
+                                        &state->info_png.color, *w, *h);
+    lodepng_free(data);
+  }
+  return state->error;
+}
+
+unsigned lodepng_decode_memory(unsigned char** out, unsigned* w, unsigned* h, const unsigned char* in,
+                               size_t insize, LodePNGColorType colortype, unsigned bitdepth) {
+  unsigned error;
+  LodePNGState state;
+  lodepng_state_init(&state);
+  state.info_raw.colortype = colortype;
+  state.info_raw.bitdepth = bitdepth;
+  error = lodepng_decode(out, w, h, &state, in, insize);
+  lodepng_state_cleanup(&state);
+  return error;
+}
+
+unsigned lodepng_decode32(unsigned char** out, unsigned* w, unsigned* h, const unsigned char* in, size_t insize) {
+  return lodepng_decode_memory(out, w, h, in, insize, LCT_RGBA, 8);
+}
+
+unsigned lodepng_decode24(unsigned char** out, unsigned* w, unsigned* h, const unsigned char* in, size_t insize) {
+  return lodepng_decode_memory(out, w, h, in, insize, LCT_RGB, 8);
+}
+
+#ifdef LODEPNG_COMPILE_DISK
+unsigned lodepng_decode_file(unsigned char** out, unsigned* w, unsigned* h, const char* filename,
+                             LodePNGColorType colortype, unsigned bitdepth) {
+  unsigned char* buffer = 0;
+  size_t buffersize;
+  unsigned error;
+  /* safe output values in case error happens */
+  *out = 0;
+  *w = *h = 0;
+  error = lodepng_load_file(&buffer, &buffersize, filename);
+  if(!error) error = lodepng_decode_memory(out, w, h, buffer, buffersize, colortype, bitdepth);
+  lodepng_free(buffer);
+  return error;
+}
+
+unsigned lodepng_decode32_file(unsigned char** out, unsigned* w, unsigned* h, const char* filename) {
+  return lodepng_decode_file(out, w, h, filename, LCT_RGBA, 8);
+}
+
+unsigned lodepng_decode24_file(unsigned char** out, unsigned* w, unsigned* h, const char* filename) {
+  return lodepng_decode_file(out, w, h, filename, LCT_RGB, 8);
+}
+#endif /*LODEPNG_COMPILE_DISK*/
+
+void lodepng_decoder_settings_init(LodePNGDecoderSettings* settings) {
+  settings->color_convert = 1;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+  settings->read_text_chunks = 1;
+  settings->remember_unknown_chunks = 0;
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+  settings->ignore_crc = 0;
+  settings->ignore_critical = 0;
+  settings->ignore_end = 0;
+  lodepng_decompress_settings_init(&settings->zlibsettings);
+}
+
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+#if defined(LODEPNG_COMPILE_DECODER) || defined(LODEPNG_COMPILE_ENCODER)
+
+void lodepng_state_init(LodePNGState* state) {
+#ifdef LODEPNG_COMPILE_DECODER
+  lodepng_decoder_settings_init(&state->decoder);
+#endif /*LODEPNG_COMPILE_DECODER*/
+#ifdef LODEPNG_COMPILE_ENCODER
+  lodepng_encoder_settings_init(&state->encoder);
+#endif /*LODEPNG_COMPILE_ENCODER*/
+  lodepng_color_mode_init(&state->info_raw);
+  lodepng_info_init(&state->info_png);
+  state->error = 1;
+}
+
+void lodepng_state_cleanup(LodePNGState* state) {
+  lodepng_color_mode_cleanup(&state->info_raw);
+  lodepng_info_cleanup(&state->info_png);
+}
+
+void lodepng_state_copy(LodePNGState* dest, const LodePNGState* source) {
+  lodepng_state_cleanup(dest);
+  *dest = *source;
+  lodepng_color_mode_init(&dest->info_raw);
+  lodepng_info_init(&dest->info_png);
+  dest->error = lodepng_color_mode_copy(&dest->info_raw, &source->info_raw); if(dest->error) return;
+  dest->error = lodepng_info_copy(&dest->info_png, &source->info_png); if(dest->error) return;
+}
+
+#endif /* defined(LODEPNG_COMPILE_DECODER) || defined(LODEPNG_COMPILE_ENCODER) */
+
+#ifdef LODEPNG_COMPILE_ENCODER
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / PNG Encoder                                                            / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+
+static unsigned writeSignature(ucvector* out) {
+  size_t pos = out->size;
+  const unsigned char signature[] = {137, 80, 78, 71, 13, 10, 26, 10};
+  /*8 bytes PNG signature, aka the magic bytes*/
+  if(!ucvector_resize(out, out->size + 8)) return 83; /*alloc fail*/
+  lodepng_memcpy(out->data + pos, signature, 8);
+  return 0;
+}
+
+static unsigned addChunk_IHDR(ucvector* out, unsigned w, unsigned h,
+                              LodePNGColorType colortype, unsigned bitdepth, unsigned interlace_method) {
+  unsigned char *chunk, *data;
+  CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, 13, "IHDR"));
+  data = chunk + 8;
+
+  lodepng_set32bitInt(data + 0, w); /*width*/
+  lodepng_set32bitInt(data + 4, h); /*height*/
+  data[8] = (unsigned char)bitdepth; /*bit depth*/
+  data[9] = (unsigned char)colortype; /*color type*/
+  data[10] = 0; /*compression method*/
+  data[11] = 0; /*filter method*/
+  data[12] = interlace_method; /*interlace method*/
+
+  lodepng_chunk_generate_crc(chunk);
+  return 0;
+}
+
+/* only adds the chunk if needed (there is a key or palette with alpha) */
+static unsigned addChunk_PLTE(ucvector* out, const LodePNGColorMode* info) {
+  unsigned char* chunk;
+  size_t i, j = 8;
+
+  CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, info->palettesize * 3, "PLTE"));
+
+  for(i = 0; i != info->palettesize; ++i) {
+    /*add all channels except alpha channel*/
+    chunk[j++] = info->palette[i * 4 + 0];
+    chunk[j++] = info->palette[i * 4 + 1];
+    chunk[j++] = info->palette[i * 4 + 2];
+  }
+
+  lodepng_chunk_generate_crc(chunk);
+  return 0;
+}
+
+static unsigned addChunk_tRNS(ucvector* out, const LodePNGColorMode* info) {
+  unsigned char* chunk = 0;
+
+  if(info->colortype == LCT_PALETTE) {
+    size_t i, amount = info->palettesize;
+    /*the tail of palette values that all have 255 as alpha, does not have to be encoded*/
+    for(i = info->palettesize; i != 0; --i) {
+      if(info->palette[4 * (i - 1) + 3] != 255) break;
+      --amount;
+    }
+    if(amount) {
+      CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, amount, "tRNS"));
+      /*add the alpha channel values from the palette*/
+      for(i = 0; i != amount; ++i) chunk[8 + i] = info->palette[4 * i + 3];
+    }
+  } else if(info->colortype == LCT_GREY) {
+    if(info->key_defined) {
+      CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, 2, "tRNS"));
+      chunk[8] = (unsigned char)(info->key_r >> 8);
+      chunk[9] = (unsigned char)(info->key_r & 255);
+    }
+  } else if(info->colortype == LCT_RGB) {
+    if(info->key_defined) {
+      CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, 6, "tRNS"));
+      chunk[8] = (unsigned char)(info->key_r >> 8);
+      chunk[9] = (unsigned char)(info->key_r & 255);
+      chunk[10] = (unsigned char)(info->key_g >> 8);
+      chunk[11] = (unsigned char)(info->key_g & 255);
+      chunk[12] = (unsigned char)(info->key_b >> 8);
+      chunk[13] = (unsigned char)(info->key_b & 255);
+    }
+  }
+
+  if(chunk) lodepng_chunk_generate_crc(chunk);
+  return 0;
+}
+
+static unsigned addChunk_IDAT(ucvector* out, const unsigned char* data, size_t datasize,
+                              LodePNGCompressSettings* zlibsettings) {
+  unsigned error = 0;
+  unsigned char* zlib = 0;
+  size_t zlibsize = 0;
+
+  error = zlib_compress(&zlib, &zlibsize, data, datasize, zlibsettings);
+  if(!error) {
+    error = lodepng_chunk_createv(out, zlibsize, "IDAT", zlib);
+  }
+  lodepng_free(zlib);
+  return error;
+}
+
+static unsigned addChunk_IEND(ucvector* out) {
+  return lodepng_chunk_createv(out, 0, "IEND", 0);
+}
+
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+
+static unsigned addChunk_tEXt(ucvector* out, const char* keyword, const char* textstring) {
+  unsigned char* chunk = 0;
+  size_t keysize = lodepng_strlen(keyword), textsize = lodepng_strlen(textstring);
+  size_t size = keysize + 1 + textsize;
+  if(keysize < 1 || keysize > 79) return 89; /*error: invalid keyword size*/
+  CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, size, "tEXt"));
+  lodepng_memcpy(chunk + 8, keyword, keysize);
+  chunk[8 + keysize] = 0; /*null termination char*/
+  lodepng_memcpy(chunk + 9 + keysize, textstring, textsize);
+  lodepng_chunk_generate_crc(chunk);
+  return 0;
+}
+
+static unsigned addChunk_zTXt(ucvector* out, const char* keyword, const char* textstring,
+                              LodePNGCompressSettings* zlibsettings) {
+  unsigned error = 0;
+  unsigned char* chunk = 0;
+  unsigned char* compressed = 0;
+  size_t compressedsize = 0;
+  size_t textsize = lodepng_strlen(textstring);
+  size_t keysize = lodepng_strlen(keyword);
+  if(keysize < 1 || keysize > 79) return 89; /*error: invalid keyword size*/
+
+  error = zlib_compress(&compressed, &compressedsize,
+                        (const unsigned char*)textstring, textsize, zlibsettings);
+  if(!error) {
+    size_t size = keysize + 2 + compressedsize;
+    error = lodepng_chunk_init(&chunk, out, size, "zTXt");
+  }
+  if(!error) {
+    lodepng_memcpy(chunk + 8, keyword, keysize);
+    chunk[8 + keysize] = 0; /*null termination char*/
+    chunk[9 + keysize] = 0; /*compression method: 0*/
+    lodepng_memcpy(chunk + 10 + keysize, compressed, compressedsize);
+    lodepng_chunk_generate_crc(chunk);
+  }
+
+  lodepng_free(compressed);
+  return error;
+}
+
+static unsigned addChunk_iTXt(ucvector* out, unsigned compress, const char* keyword, const char* langtag,
+                              const char* transkey, const char* textstring, LodePNGCompressSettings* zlibsettings) {
+  unsigned error = 0;
+  unsigned char* chunk = 0;
+  unsigned char* compressed = 0;
+  size_t compressedsize = 0;
+  size_t textsize = lodepng_strlen(textstring);
+  size_t keysize = lodepng_strlen(keyword), langsize = lodepng_strlen(langtag), transsize = lodepng_strlen(transkey);
+
+  if(keysize < 1 || keysize > 79) return 89; /*error: invalid keyword size*/
+
+  if(compress) {
+    error = zlib_compress(&compressed, &compressedsize,
+                          (const unsigned char*)textstring, textsize, zlibsettings);
+  }
+  if(!error) {
+    size_t size = keysize + 3 + langsize + 1 + transsize + 1 + (compress ? compressedsize : textsize);
+    error = lodepng_chunk_init(&chunk, out, size, "iTXt");
+  }
+  if(!error) {
+    size_t pos = 8;
+    lodepng_memcpy(chunk + pos, keyword, keysize);
+    pos += keysize;
+    chunk[pos++] = 0; /*null termination char*/
+    chunk[pos++] = (compress ? 1 : 0); /*compression flag*/
+    chunk[pos++] = 0; /*compression method: 0*/
+    lodepng_memcpy(chunk + pos, langtag, langsize);
+    pos += langsize;
+    chunk[pos++] = 0; /*null termination char*/
+    lodepng_memcpy(chunk + pos, transkey, transsize);
+    pos += transsize;
+    chunk[pos++] = 0; /*null termination char*/
+    if(compress) {
+      lodepng_memcpy(chunk + pos, compressed, compressedsize);
+    } else {
+      lodepng_memcpy(chunk + pos, textstring, textsize);
+    }
+    lodepng_chunk_generate_crc(chunk);
+  }
+
+  lodepng_free(compressed);
+  return error;
+}
+
+static unsigned addChunk_bKGD(ucvector* out, const LodePNGInfo* info) {
+  unsigned char* chunk = 0;
+  if(info->color.colortype == LCT_GREY || info->color.colortype == LCT_GREY_ALPHA) {
+    CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, 2, "bKGD"));
+    chunk[8] = (unsigned char)(info->background_r >> 8);
+    chunk[9] = (unsigned char)(info->background_r & 255);
+  } else if(info->color.colortype == LCT_RGB || info->color.colortype == LCT_RGBA) {
+    CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, 6, "bKGD"));
+    chunk[8] = (unsigned char)(info->background_r >> 8);
+    chunk[9] = (unsigned char)(info->background_r & 255);
+    chunk[10] = (unsigned char)(info->background_g >> 8);
+    chunk[11] = (unsigned char)(info->background_g & 255);
+    chunk[12] = (unsigned char)(info->background_b >> 8);
+    chunk[13] = (unsigned char)(info->background_b & 255);
+  } else if(info->color.colortype == LCT_PALETTE) {
+    CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, 1, "bKGD"));
+    chunk[8] = (unsigned char)(info->background_r & 255); /*palette index*/
+  }
+  if(chunk) lodepng_chunk_generate_crc(chunk);
+  return 0;
+}
+
+static unsigned addChunk_tIME(ucvector* out, const LodePNGTime* time) {
+  unsigned char* chunk;
+  CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, 7, "tIME"));
+  chunk[8] = (unsigned char)(time->year >> 8);
+  chunk[9] = (unsigned char)(time->year & 255);
+  chunk[10] = (unsigned char)time->month;
+  chunk[11] = (unsigned char)time->day;
+  chunk[12] = (unsigned char)time->hour;
+  chunk[13] = (unsigned char)time->minute;
+  chunk[14] = (unsigned char)time->second;
+  lodepng_chunk_generate_crc(chunk);
+  return 0;
+}
+
+static unsigned addChunk_pHYs(ucvector* out, const LodePNGInfo* info) {
+  unsigned char* chunk;
+  CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, 9, "pHYs"));
+  lodepng_set32bitInt(chunk + 8, info->phys_x);
+  lodepng_set32bitInt(chunk + 12, info->phys_y);
+  chunk[16] = info->phys_unit;
+  lodepng_chunk_generate_crc(chunk);
+  return 0;
+}
+
+static unsigned addChunk_gAMA(ucvector* out, const LodePNGInfo* info) {
+  unsigned char* chunk;
+  CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, 4, "gAMA"));
+  lodepng_set32bitInt(chunk + 8, info->gama_gamma);
+  lodepng_chunk_generate_crc(chunk);
+  return 0;
+}
+
+static unsigned addChunk_cHRM(ucvector* out, const LodePNGInfo* info) {
+  unsigned char* chunk;
+  CERROR_TRY_RETURN(lodepng_chunk_init(&chunk, out, 32, "cHRM"));
+  lodepng_set32bitInt(chunk + 8, info->chrm_white_x);
+  lodepng_set32bitInt(chunk + 12, info->chrm_white_y);
+  lodepng_set32bitInt(chunk + 16, info->chrm_red_x);
+  lodepng_set32bitInt(chunk + 20, info->chrm_red_y);
+  lodepng_set32bitInt(chunk + 24, info->chrm_green_x);
+  lodepng_set32bitInt(chunk + 28, info->chrm_green_y);
+  lodepng_set32bitInt(chunk + 32, info->chrm_blue_x);
+  lodepng_set32bitInt(chunk + 36, info->chrm_blue_y);
+  lodepng_chunk_generate_crc(chunk);
+  return 0;
+}
+
+static unsigned addChunk_sRGB(ucvector* out, const LodePNGInfo* info) {
+  unsigned char data = info->srgb_intent;
+  return lodepng_chunk_createv(out, 1, "sRGB", &data);
+}
+
+static unsigned addChunk_iCCP(ucvector* out, const LodePNGInfo* info, LodePNGCompressSettings* zlibsettings) {
+  unsigned error = 0;
+  unsigned char* chunk = 0;
+  unsigned char* compressed = 0;
+  size_t compressedsize = 0;
+  size_t keysize = lodepng_strlen(info->iccp_name);
+
+  if(keysize < 1 || keysize > 79) return 89; /*error: invalid keyword size*/
+  error = zlib_compress(&compressed, &compressedsize,
+                        info->iccp_profile, info->iccp_profile_size, zlibsettings);
+  if(!error) {
+    size_t size = keysize + 2 + compressedsize;
+    error = lodepng_chunk_init(&chunk, out, size, "iCCP");
+  }
+  if(!error) {
+    lodepng_memcpy(chunk + 8, info->iccp_name, keysize);
+    chunk[8 + keysize] = 0; /*null termination char*/
+    chunk[9 + keysize] = 0; /*compression method: 0*/
+    lodepng_memcpy(chunk + 10 + keysize, compressed, compressedsize);
+    lodepng_chunk_generate_crc(chunk);
+  }
+
+  lodepng_free(compressed);
+  return error;
+}
+
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+
+static void filterScanline(unsigned char* out, const unsigned char* scanline, const unsigned char* prevline,
+                           size_t length, size_t bytewidth, unsigned char filterType) {
+  size_t i;
+  switch(filterType) {
+    case 0: /*None*/
+      for(i = 0; i != length; ++i) out[i] = scanline[i];
+      break;
+    case 1: /*Sub*/
+      for(i = 0; i != bytewidth; ++i) out[i] = scanline[i];
+      for(i = bytewidth; i < length; ++i) out[i] = scanline[i] - scanline[i - bytewidth];
+      break;
+    case 2: /*Up*/
+      if(prevline) {
+        for(i = 0; i != length; ++i) out[i] = scanline[i] - prevline[i];
+      } else {
+        for(i = 0; i != length; ++i) out[i] = scanline[i];
+      }
+      break;
+    case 3: /*Average*/
+      if(prevline) {
+        for(i = 0; i != bytewidth; ++i) out[i] = scanline[i] - (prevline[i] >> 1);
+        for(i = bytewidth; i < length; ++i) out[i] = scanline[i] - ((scanline[i - bytewidth] + prevline[i]) >> 1);
+      } else {
+        for(i = 0; i != bytewidth; ++i) out[i] = scanline[i];
+        for(i = bytewidth; i < length; ++i) out[i] = scanline[i] - (scanline[i - bytewidth] >> 1);
+      }
+      break;
+    case 4: /*Paeth*/
+      if(prevline) {
+        /*paethPredictor(0, prevline[i], 0) is always prevline[i]*/
+        for(i = 0; i != bytewidth; ++i) out[i] = (scanline[i] - prevline[i]);
+        for(i = bytewidth; i < length; ++i) {
+          out[i] = (scanline[i] - paethPredictor(scanline[i - bytewidth], prevline[i], prevline[i - bytewidth]));
+        }
+      } else {
+        for(i = 0; i != bytewidth; ++i) out[i] = scanline[i];
+        /*paethPredictor(scanline[i - bytewidth], 0, 0) is always scanline[i - bytewidth]*/
+        for(i = bytewidth; i < length; ++i) out[i] = (scanline[i] - scanline[i - bytewidth]);
+      }
+      break;
+    default: return; /*invalid filter type given*/
+  }
+}
+
+/* integer binary logarithm, max return value is 31 */
+static size_t ilog2(size_t i) {
+  size_t result = 0;
+  if(i >= 65536) { result += 16; i >>= 16; }
+  if(i >= 256) { result += 8; i >>= 8; }
+  if(i >= 16) { result += 4; i >>= 4; }
+  if(i >= 4) { result += 2; i >>= 2; }
+  if(i >= 2) { result += 1; /*i >>= 1;*/ }
+  return result;
+}
+
+/* integer approximation for i * log2(i), helper function for LFS_ENTROPY */
+static size_t ilog2i(size_t i) {
+  size_t l;
+  if(i == 0) return 0;
+  l = ilog2(i);
+  /* approximate i*log2(i): l is integer logarithm, ((i - (1u << l)) << 1u)
+  linearly approximates the missing fractional part multiplied by i */
+  return i * l + ((i - (1u << l)) << 1u);
+}
+
+static unsigned filter(unsigned char* out, const unsigned char* in, unsigned w, unsigned h,
+                       const LodePNGColorMode* color, const LodePNGEncoderSettings* settings) {
+  /*
+  For PNG filter method 0
+  out must be a buffer with as size: h + (w * h * bpp + 7u) / 8u, because there are
+  the scanlines with 1 extra byte per scanline
+  */
+
+  unsigned bpp = lodepng_get_bpp(color);
+  /*the width of a scanline in bytes, not including the filter type*/
+  size_t linebytes = lodepng_get_raw_size_idat(w, 1, bpp) - 1u;
+
+  /*bytewidth is used for filtering, is 1 when bpp < 8, number of bytes per pixel otherwise*/
+  size_t bytewidth = (bpp + 7u) / 8u;
+  const unsigned char* prevline = 0;
+  unsigned x, y;
+  unsigned error = 0;
+  LodePNGFilterStrategy strategy = settings->filter_strategy;
+
+  /*
+  There is a heuristic called the minimum sum of absolute differences heuristic, suggested by the PNG standard:
+   *  If the image type is Palette, or the bit depth is smaller than 8, then do not filter the image (i.e.
+      use fixed filtering, with the filter None).
+   * (The other case) If the image type is Grayscale or RGB (with or without Alpha), and the bit depth is
+     not smaller than 8, then use adaptive filtering heuristic as follows: independently for each row, apply
+     all five filters and select the filter that produces the smallest sum of absolute values per row.
+  This heuristic is used if filter strategy is LFS_MINSUM and filter_palette_zero is true.
+
+  If filter_palette_zero is true and filter_strategy is not LFS_MINSUM, the above heuristic is followed,
+  but for "the other case", whatever strategy filter_strategy is set to instead of the minimum sum
+  heuristic is used.
+  */
+  if(settings->filter_palette_zero &&
+     (color->colortype == LCT_PALETTE || color->bitdepth < 8)) strategy = LFS_ZERO;
+
+  if(bpp == 0) return 31; /*error: invalid color type*/
+
+  if(strategy >= LFS_ZERO && strategy <= LFS_FOUR) {
+    unsigned char type = (unsigned char)strategy;
+    for(y = 0; y != h; ++y) {
+      size_t outindex = (1 + linebytes) * y; /*the extra filterbyte added to each row*/
+      size_t inindex = linebytes * y;
+      out[outindex] = type; /*filter type byte*/
+      filterScanline(&out[outindex + 1], &in[inindex], prevline, linebytes, bytewidth, type);
+      prevline = &in[inindex];
+    }
+  } else if(strategy == LFS_MINSUM) {
+    /*adaptive filtering*/
+    unsigned char* attempt[5]; /*five filtering attempts, one for each filter type*/
+    size_t smallest = 0;
+    unsigned char type, bestType = 0;
+
+    for(type = 0; type != 5; ++type) {
+      attempt[type] = (unsigned char*)lodepng_malloc(linebytes);
+      if(!attempt[type]) error = 83; /*alloc fail*/
+    }
+
+    if(!error) {
+      for(y = 0; y != h; ++y) {
+        /*try the 5 filter types*/
+        for(type = 0; type != 5; ++type) {
+          size_t sum = 0;
+          filterScanline(attempt[type], &in[y * linebytes], prevline, linebytes, bytewidth, type);
+
+          /*calculate the sum of the result*/
+          if(type == 0) {
+            for(x = 0; x != linebytes; ++x) sum += (unsigned char)(attempt[type][x]);
+          } else {
+            for(x = 0; x != linebytes; ++x) {
+              /*For differences, each byte should be treated as signed, values above 127 are negative
+              (converted to signed char). Filtertype 0 isn't a difference though, so use unsigned there.
+              This means filtertype 0 is almost never chosen, but that is justified.*/
+              unsigned char s = attempt[type][x];
+              sum += s < 128 ? s : (255U - s);
+            }
+          }
+
+          /*check if this is smallest sum (or if type == 0 it's the first case so always store the values)*/
+          if(type == 0 || sum < smallest) {
+            bestType = type;
+            smallest = sum;
+          }
+        }
+
+        prevline = &in[y * linebytes];
+
+        /*now fill the out values*/
+        out[y * (linebytes + 1)] = bestType; /*the first byte of a scanline will be the filter type*/
+        for(x = 0; x != linebytes; ++x) out[y * (linebytes + 1) + 1 + x] = attempt[bestType][x];
+      }
+    }
+
+    for(type = 0; type != 5; ++type) lodepng_free(attempt[type]);
+  } else if(strategy == LFS_ENTROPY) {
+    unsigned char* attempt[5]; /*five filtering attempts, one for each filter type*/
+    size_t bestSum = 0;
+    unsigned type, bestType = 0;
+    unsigned count[256];
+
+    for(type = 0; type != 5; ++type) {
+      attempt[type] = (unsigned char*)lodepng_malloc(linebytes);
+      if(!attempt[type]) error = 83; /*alloc fail*/
+    }
+
+    if(!error) {
+      for(y = 0; y != h; ++y) {
+        /*try the 5 filter types*/
+        for(type = 0; type != 5; ++type) {
+          size_t sum = 0;
+          filterScanline(attempt[type], &in[y * linebytes], prevline, linebytes, bytewidth, type);
+          lodepng_memset(count, 0, 256 * sizeof(*count));
+          for(x = 0; x != linebytes; ++x) ++count[attempt[type][x]];
+          ++count[type]; /*the filter type itself is part of the scanline*/
+          for(x = 0; x != 256; ++x) {
+            sum += ilog2i(count[x]);
+          }
+          /*check if this is smallest sum (or if type == 0 it's the first case so always store the values)*/
+          if(type == 0 || sum > bestSum) {
+            bestType = type;
+            bestSum = sum;
+          }
+        }
+
+        prevline = &in[y * linebytes];
+
+        /*now fill the out values*/
+        out[y * (linebytes + 1)] = bestType; /*the first byte of a scanline will be the filter type*/
+        for(x = 0; x != linebytes; ++x) out[y * (linebytes + 1) + 1 + x] = attempt[bestType][x];
+      }
+    }
+
+    for(type = 0; type != 5; ++type) lodepng_free(attempt[type]);
+  } else if(strategy == LFS_PREDEFINED) {
+    for(y = 0; y != h; ++y) {
+      size_t outindex = (1 + linebytes) * y; /*the extra filterbyte added to each row*/
+      size_t inindex = linebytes * y;
+      unsigned char type = settings->predefined_filters[y];
+      out[outindex] = type; /*filter type byte*/
+      filterScanline(&out[outindex + 1], &in[inindex], prevline, linebytes, bytewidth, type);
+      prevline = &in[inindex];
+    }
+  } else if(strategy == LFS_BRUTE_FORCE) {
+    /*brute force filter chooser.
+    deflate the scanline after every filter attempt to see which one deflates best.
+    This is very slow and gives only slightly smaller, sometimes even larger, result*/
+    size_t size[5];
+    unsigned char* attempt[5]; /*five filtering attempts, one for each filter type*/
+    size_t smallest = 0;
+    unsigned type = 0, bestType = 0;
+    unsigned char* dummy;
+    LodePNGCompressSettings zlibsettings;
+    lodepng_memcpy(&zlibsettings, &settings->zlibsettings, sizeof(LodePNGCompressSettings));
+    /*use fixed tree on the attempts so that the tree is not adapted to the filtertype on purpose,
+    to simulate the true case where the tree is the same for the whole image. Sometimes it gives
+    better result with dynamic tree anyway. Using the fixed tree sometimes gives worse, but in rare
+    cases better compression. It does make this a bit less slow, so it's worth doing this.*/
+    zlibsettings.btype = 1;
+    /*a custom encoder likely doesn't read the btype setting and is optimized for complete PNG
+    images only, so disable it*/
+    zlibsettings.custom_zlib = 0;
+    zlibsettings.custom_deflate = 0;
+    for(type = 0; type != 5; ++type) {
+      attempt[type] = (unsigned char*)lodepng_malloc(linebytes);
+      if(!attempt[type]) error = 83; /*alloc fail*/
+    }
+    if(!error) {
+      for(y = 0; y != h; ++y) /*try the 5 filter types*/ {
+        for(type = 0; type != 5; ++type) {
+          unsigned testsize = (unsigned)linebytes;
+          /*if(testsize > 8) testsize /= 8;*/ /*it already works good enough by testing a part of the row*/
+
+          filterScanline(attempt[type], &in[y * linebytes], prevline, linebytes, bytewidth, type);
+          size[type] = 0;
+          dummy = 0;
+          zlib_compress(&dummy, &size[type], attempt[type], testsize, &zlibsettings);
+          lodepng_free(dummy);
+          /*check if this is smallest size (or if type == 0 it's the first case so always store the values)*/
+          if(type == 0 || size[type] < smallest) {
+            bestType = type;
+            smallest = size[type];
+          }
+        }
+        prevline = &in[y * linebytes];
+        out[y * (linebytes + 1)] = bestType; /*the first byte of a scanline will be the filter type*/
+        for(x = 0; x != linebytes; ++x) out[y * (linebytes + 1) + 1 + x] = attempt[bestType][x];
+      }
+    }
+    for(type = 0; type != 5; ++type) lodepng_free(attempt[type]);
+  }
+  else return 88; /* unknown filter strategy */
+
+  return error;
+}
+
+static void addPaddingBits(unsigned char* out, const unsigned char* in,
+                           size_t olinebits, size_t ilinebits, unsigned h) {
+  /*The opposite of the removePaddingBits function
+  olinebits must be >= ilinebits*/
+  unsigned y;
+  size_t diff = olinebits - ilinebits;
+  size_t obp = 0, ibp = 0; /*bit pointers*/
+  for(y = 0; y != h; ++y) {
+    size_t x;
+    for(x = 0; x < ilinebits; ++x) {
+      unsigned char bit = readBitFromReversedStream(&ibp, in);
+      setBitOfReversedStream(&obp, out, bit);
+    }
+    /*obp += diff; --> no, fill in some value in the padding bits too, to avoid
+    "Use of uninitialised value of size ###" warning from valgrind*/
+    for(x = 0; x != diff; ++x) setBitOfReversedStream(&obp, out, 0);
+  }
+}
+
+/*
+in: non-interlaced image with size w*h
+out: the same pixels, but re-ordered according to PNG's Adam7 interlacing, with
+ no padding bits between scanlines, but between reduced images so that each
+ reduced image starts at a byte.
+bpp: bits per pixel
+there are no padding bits, not between scanlines, not between reduced images
+in has the following size in bits: w * h * bpp.
+out is possibly bigger due to padding bits between reduced images
+NOTE: comments about padding bits are only relevant if bpp < 8
+*/
+static void Adam7_interlace(unsigned char* out, const unsigned char* in, unsigned w, unsigned h, unsigned bpp) {
+  unsigned passw[7], passh[7];
+  size_t filter_passstart[8], padded_passstart[8], passstart[8];
+  unsigned i;
+
+  Adam7_getpassvalues(passw, passh, filter_passstart, padded_passstart, passstart, w, h, bpp);
+
+  if(bpp >= 8) {
+    for(i = 0; i != 7; ++i) {
+      unsigned x, y, b;
+      size_t bytewidth = bpp / 8u;
+      for(y = 0; y < passh[i]; ++y)
+      for(x = 0; x < passw[i]; ++x) {
+        size_t pixelinstart = ((ADAM7_IY[i] + y * ADAM7_DY[i]) * w + ADAM7_IX[i] + x * ADAM7_DX[i]) * bytewidth;
+        size_t pixeloutstart = passstart[i] + (y * passw[i] + x) * bytewidth;
+        for(b = 0; b < bytewidth; ++b) {
+          out[pixeloutstart + b] = in[pixelinstart + b];
+        }
+      }
+    }
+  } else /*bpp < 8: Adam7 with pixels < 8 bit is a bit trickier: with bit pointers*/ {
+    for(i = 0; i != 7; ++i) {
+      unsigned x, y, b;
+      unsigned ilinebits = bpp * passw[i];
+      unsigned olinebits = bpp * w;
+      size_t obp, ibp; /*bit pointers (for out and in buffer)*/
+      for(y = 0; y < passh[i]; ++y)
+      for(x = 0; x < passw[i]; ++x) {
+        ibp = (ADAM7_IY[i] + y * ADAM7_DY[i]) * olinebits + (ADAM7_IX[i] + x * ADAM7_DX[i]) * bpp;
+        obp = (8 * passstart[i]) + (y * ilinebits + x * bpp);
+        for(b = 0; b < bpp; ++b) {
+          unsigned char bit = readBitFromReversedStream(&ibp, in);
+          setBitOfReversedStream(&obp, out, bit);
+        }
+      }
+    }
+  }
+}
+
+/*out must be buffer big enough to contain uncompressed IDAT chunk data, and in must contain the full image.
+return value is error**/
+static unsigned preProcessScanlines(unsigned char** out, size_t* outsize, const unsigned char* in,
+                                    unsigned w, unsigned h,
+                                    const LodePNGInfo* info_png, const LodePNGEncoderSettings* settings) {
+  /*
+  This function converts the pure 2D image with the PNG's colortype, into filtered-padded-interlaced data. Steps:
+  *) if no Adam7: 1) add padding bits (= possible extra bits per scanline if bpp < 8) 2) filter
+  *) if adam7: 1) Adam7_interlace 2) 7x add padding bits 3) 7x filter
+  */
+  unsigned bpp = lodepng_get_bpp(&info_png->color);
+  unsigned error = 0;
+
+  if(info_png->interlace_method == 0) {
+    *outsize = h + (h * ((w * bpp + 7u) / 8u)); /*image size plus an extra byte per scanline + possible padding bits*/
+    *out = (unsigned char*)lodepng_malloc(*outsize);
+    if(!(*out) && (*outsize)) error = 83; /*alloc fail*/
+
+    if(!error) {
+      /*non multiple of 8 bits per scanline, padding bits needed per scanline*/
+      if(bpp < 8 && w * bpp != ((w * bpp + 7u) / 8u) * 8u) {
+        unsigned char* padded = (unsigned char*)lodepng_malloc(h * ((w * bpp + 7u) / 8u));
+        if(!padded) error = 83; /*alloc fail*/
+        if(!error) {
+          addPaddingBits(padded, in, ((w * bpp + 7u) / 8u) * 8u, w * bpp, h);
+          error = filter(*out, padded, w, h, &info_png->color, settings);
+        }
+        lodepng_free(padded);
+      } else {
+        /*we can immediately filter into the out buffer, no other steps needed*/
+        error = filter(*out, in, w, h, &info_png->color, settings);
+      }
+    }
+  } else /*interlace_method is 1 (Adam7)*/ {
+    unsigned passw[7], passh[7];
+    size_t filter_passstart[8], padded_passstart[8], passstart[8];
+    unsigned char* adam7;
+
+    Adam7_getpassvalues(passw, passh, filter_passstart, padded_passstart, passstart, w, h, bpp);
+
+    *outsize = filter_passstart[7]; /*image size plus an extra byte per scanline + possible padding bits*/
+    *out = (unsigned char*)lodepng_malloc(*outsize);
+    if(!(*out)) error = 83; /*alloc fail*/
+
+    adam7 = (unsigned char*)lodepng_malloc(passstart[7]);
+    if(!adam7 && passstart[7]) error = 83; /*alloc fail*/
+
+    if(!error) {
+      unsigned i;
+
+      Adam7_interlace(adam7, in, w, h, bpp);
+      for(i = 0; i != 7; ++i) {
+        if(bpp < 8) {
+          unsigned char* padded = (unsigned char*)lodepng_malloc(padded_passstart[i + 1] - padded_passstart[i]);
+          if(!padded) ERROR_BREAK(83); /*alloc fail*/
+          addPaddingBits(padded, &adam7[passstart[i]],
+                         ((passw[i] * bpp + 7u) / 8u) * 8u, passw[i] * bpp, passh[i]);
+          error = filter(&(*out)[filter_passstart[i]], padded,
+                         passw[i], passh[i], &info_png->color, settings);
+          lodepng_free(padded);
+        } else {
+          error = filter(&(*out)[filter_passstart[i]], &adam7[padded_passstart[i]],
+                         passw[i], passh[i], &info_png->color, settings);
+        }
+
+        if(error) break;
+      }
+    }
+
+    lodepng_free(adam7);
+  }
+
+  return error;
+}
+
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+static unsigned addUnknownChunks(ucvector* out, unsigned char* data, size_t datasize) {
+  unsigned char* inchunk = data;
+  while((size_t)(inchunk - data) < datasize) {
+    CERROR_TRY_RETURN(lodepng_chunk_append(&out->data, &out->size, inchunk));
+    out->allocsize = out->size; /*fix the allocsize again*/
+    inchunk = lodepng_chunk_next(inchunk, data + datasize);
+  }
+  return 0;
+}
+
+static unsigned isGrayICCProfile(const unsigned char* profile, unsigned size) {
+  /*
+  It is a gray profile if bytes 16-19 are "GRAY", rgb profile if bytes 16-19
+  are "RGB ". We do not perform any full parsing of the ICC profile here, other
+  than check those 4 bytes to grayscale profile. Other than that, validity of
+  the profile is not checked. This is needed only because the PNG specification
+  requires using a non-gray color model if there is an ICC profile with "RGB "
+  (sadly limiting compression opportunities if the input data is grayscale RGB
+  data), and requires using a gray color model if it is "GRAY".
+  */
+  if(size < 20) return 0;
+  return profile[16] == 'G' &&  profile[17] == 'R' &&  profile[18] == 'A' &&  profile[19] == 'Y';
+}
+
+static unsigned isRGBICCProfile(const unsigned char* profile, unsigned size) {
+  /* See comment in isGrayICCProfile*/
+  if(size < 20) return 0;
+  return profile[16] == 'R' &&  profile[17] == 'G' &&  profile[18] == 'B' &&  profile[19] == ' ';
+}
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+
+unsigned lodepng_encode(unsigned char** out, size_t* outsize,
+                        const unsigned char* image, unsigned w, unsigned h,
+                        LodePNGState* state) {
+  unsigned char* data = 0; /*uncompressed version of the IDAT chunk data*/
+  size_t datasize = 0;
+  ucvector outv = ucvector_init(NULL, 0);
+  LodePNGInfo info;
+  const LodePNGInfo* info_png = &state->info_png;
+
+  lodepng_info_init(&info);
+
+  /*provide some proper output values if error will happen*/
+  *out = 0;
+  *outsize = 0;
+  state->error = 0;
+
+  /*check input values validity*/
+  if((info_png->color.colortype == LCT_PALETTE || state->encoder.force_palette)
+      && (info_png->color.palettesize == 0 || info_png->color.palettesize > 256)) {
+    state->error = 68; /*invalid palette size, it is only allowed to be 1-256*/
+    goto cleanup;
+  }
+  if(state->encoder.zlibsettings.btype > 2) {
+    state->error = 61; /*error: invalid btype*/
+    goto cleanup;
+  }
+  if(info_png->interlace_method > 1) {
+    state->error = 71; /*error: invalid interlace mode*/
+    goto cleanup;
+  }
+  state->error = checkColorValidity(info_png->color.colortype, info_png->color.bitdepth);
+  if(state->error) goto cleanup; /*error: invalid color type given*/
+  state->error = checkColorValidity(state->info_raw.colortype, state->info_raw.bitdepth);
+  if(state->error) goto cleanup; /*error: invalid color type given*/
+
+  /* color convert and compute scanline filter types */
+  lodepng_info_copy(&info, &state->info_png);
+  if(state->encoder.auto_convert) {
+    LodePNGColorStats stats;
+    lodepng_color_stats_init(&stats);
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+    if(info_png->iccp_defined &&
+        isGrayICCProfile(info_png->iccp_profile, info_png->iccp_profile_size)) {
+      /*the PNG specification does not allow to use palette with a GRAY ICC profile, even
+      if the palette has only gray colors, so disallow it.*/
+      stats.allow_palette = 0;
+    }
+    if(info_png->iccp_defined &&
+        isRGBICCProfile(info_png->iccp_profile, info_png->iccp_profile_size)) {
+      /*the PNG specification does not allow to use grayscale color with RGB ICC profile, so disallow gray.*/
+      stats.allow_greyscale = 0;
+    }
+#endif /* LODEPNG_COMPILE_ANCILLARY_CHUNKS */
+    state->error = lodepng_compute_color_stats(&stats, image, w, h, &state->info_raw);
+    if(state->error) goto cleanup;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+    if(info_png->background_defined) {
+      /*the background chunk's color must be taken into account as well*/
+      unsigned r = 0, g = 0, b = 0;
+      LodePNGColorMode mode16 = lodepng_color_mode_make(LCT_RGB, 16);
+      lodepng_convert_rgb(&r, &g, &b, info_png->background_r, info_png->background_g, info_png->background_b, &mode16, &info_png->color);
+      state->error = lodepng_color_stats_add(&stats, r, g, b, 65535);
+      if(state->error) goto cleanup;
+    }
+#endif /* LODEPNG_COMPILE_ANCILLARY_CHUNKS */
+    state->error = auto_choose_color(&info.color, &state->info_raw, &stats);
+    if(state->error) goto cleanup;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+    /*also convert the background chunk*/
+    if(info_png->background_defined) {
+      if(lodepng_convert_rgb(&info.background_r, &info.background_g, &info.background_b,
+          info_png->background_r, info_png->background_g, info_png->background_b, &info.color, &info_png->color)) {
+        state->error = 104;
+        goto cleanup;
+      }
+    }
+#endif /* LODEPNG_COMPILE_ANCILLARY_CHUNKS */
+  }
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+  if(info_png->iccp_defined) {
+    unsigned gray_icc = isGrayICCProfile(info_png->iccp_profile, info_png->iccp_profile_size);
+    unsigned rgb_icc = isRGBICCProfile(info_png->iccp_profile, info_png->iccp_profile_size);
+    unsigned gray_png = info.color.colortype == LCT_GREY || info.color.colortype == LCT_GREY_ALPHA;
+    if(!gray_icc && !rgb_icc) {
+      state->error = 100; /* Disallowed profile color type for PNG */
+      goto cleanup;
+    }
+    if(gray_icc != gray_png) {
+      /*Not allowed to use RGB/RGBA/palette with GRAY ICC profile or vice versa,
+      or in case of auto_convert, it wasn't possible to find appropriate model*/
+      state->error = state->encoder.auto_convert ? 102 : 101;
+      goto cleanup;
+    }
+  }
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+  if(!lodepng_color_mode_equal(&state->info_raw, &info.color)) {
+    unsigned char* converted;
+    size_t size = ((size_t)w * (size_t)h * (size_t)lodepng_get_bpp(&info.color) + 7u) / 8u;
+
+    converted = (unsigned char*)lodepng_malloc(size);
+    if(!converted && size) state->error = 83; /*alloc fail*/
+    if(!state->error) {
+      state->error = lodepng_convert(converted, image, &info.color, &state->info_raw, w, h);
+    }
+    if(!state->error) {
+      state->error = preProcessScanlines(&data, &datasize, converted, w, h, &info, &state->encoder);
+    }
+    lodepng_free(converted);
+    if(state->error) goto cleanup;
+  } else {
+    state->error = preProcessScanlines(&data, &datasize, image, w, h, &info, &state->encoder);
+    if(state->error) goto cleanup;
+  }
+
+  /* output all PNG chunks */ {
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+    size_t i;
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+    /*write signature and chunks*/
+    state->error = writeSignature(&outv);
+    if(state->error) goto cleanup;
+    /*IHDR*/
+    state->error = addChunk_IHDR(&outv, w, h, info.color.colortype, info.color.bitdepth, info.interlace_method);
+    if(state->error) goto cleanup;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+    /*unknown chunks between IHDR and PLTE*/
+    if(info.unknown_chunks_data[0]) {
+      state->error = addUnknownChunks(&outv, info.unknown_chunks_data[0], info.unknown_chunks_size[0]);
+      if(state->error) goto cleanup;
+    }
+    /*color profile chunks must come before PLTE */
+    if(info.iccp_defined) {
+      state->error = addChunk_iCCP(&outv, &info, &state->encoder.zlibsettings);
+      if(state->error) goto cleanup;
+    }
+    if(info.srgb_defined) {
+      state->error = addChunk_sRGB(&outv, &info);
+      if(state->error) goto cleanup;
+    }
+    if(info.gama_defined) {
+      state->error = addChunk_gAMA(&outv, &info);
+      if(state->error) goto cleanup;
+    }
+    if(info.chrm_defined) {
+      state->error = addChunk_cHRM(&outv, &info);
+      if(state->error) goto cleanup;
+    }
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+    /*PLTE*/
+    if(info.color.colortype == LCT_PALETTE) {
+      state->error = addChunk_PLTE(&outv, &info.color);
+      if(state->error) goto cleanup;
+    }
+    if(state->encoder.force_palette && (info.color.colortype == LCT_RGB || info.color.colortype == LCT_RGBA)) {
+      /*force_palette means: write suggested palette for truecolor in PLTE chunk*/
+      state->error = addChunk_PLTE(&outv, &info.color);
+      if(state->error) goto cleanup;
+    }
+    /*tRNS (this will only add if when necessary) */
+    state->error = addChunk_tRNS(&outv, &info.color);
+    if(state->error) goto cleanup;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+    /*bKGD (must come between PLTE and the IDAt chunks*/
+    if(info.background_defined) {
+      state->error = addChunk_bKGD(&outv, &info);
+      if(state->error) goto cleanup;
+    }
+    /*pHYs (must come before the IDAT chunks)*/
+    if(info.phys_defined) {
+      state->error = addChunk_pHYs(&outv, &info);
+      if(state->error) goto cleanup;
+    }
+
+    /*unknown chunks between PLTE and IDAT*/
+    if(info.unknown_chunks_data[1]) {
+      state->error = addUnknownChunks(&outv, info.unknown_chunks_data[1], info.unknown_chunks_size[1]);
+      if(state->error) goto cleanup;
+    }
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+    /*IDAT (multiple IDAT chunks must be consecutive)*/
+    state->error = addChunk_IDAT(&outv, data, datasize, &state->encoder.zlibsettings);
+    if(state->error) goto cleanup;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+    /*tIME*/
+    if(info.time_defined) {
+      state->error = addChunk_tIME(&outv, &info.time);
+      if(state->error) goto cleanup;
+    }
+    /*tEXt and/or zTXt*/
+    for(i = 0; i != info.text_num; ++i) {
+      if(lodepng_strlen(info.text_keys[i]) > 79) {
+        state->error = 66; /*text chunk too large*/
+        goto cleanup;
+      }
+      if(lodepng_strlen(info.text_keys[i]) < 1) {
+        state->error = 67; /*text chunk too small*/
+        goto cleanup;
+      }
+      if(state->encoder.text_compression) {
+        state->error = addChunk_zTXt(&outv, info.text_keys[i], info.text_strings[i], &state->encoder.zlibsettings);
+        if(state->error) goto cleanup;
+      } else {
+        state->error = addChunk_tEXt(&outv, info.text_keys[i], info.text_strings[i]);
+        if(state->error) goto cleanup;
+      }
+    }
+    /*LodePNG version id in text chunk*/
+    if(state->encoder.add_id) {
+      unsigned already_added_id_text = 0;
+      for(i = 0; i != info.text_num; ++i) {
+        const char* k = info.text_keys[i];
+        /* Could use strcmp, but we're not calling or reimplementing this C library function for this use only */
+        if(k[0] == 'L' && k[1] == 'o' && k[2] == 'd' && k[3] == 'e' &&
+           k[4] == 'P' && k[5] == 'N' && k[6] == 'G' && k[7] == '\0') {
+          already_added_id_text = 1;
+          break;
+        }
+      }
+      if(already_added_id_text == 0) {
+        state->error = addChunk_tEXt(&outv, "LodePNG", LODEPNG_VERSION_STRING); /*it's shorter as tEXt than as zTXt chunk*/
+        if(state->error) goto cleanup;
+      }
+    }
+    /*iTXt*/
+    for(i = 0; i != info.itext_num; ++i) {
+      if(lodepng_strlen(info.itext_keys[i]) > 79) {
+        state->error = 66; /*text chunk too large*/
+        goto cleanup;
+      }
+      if(lodepng_strlen(info.itext_keys[i]) < 1) {
+        state->error = 67; /*text chunk too small*/
+        goto cleanup;
+      }
+      state->error = addChunk_iTXt(
+          &outv, state->encoder.text_compression,
+          info.itext_keys[i], info.itext_langtags[i], info.itext_transkeys[i], info.itext_strings[i],
+          &state->encoder.zlibsettings);
+      if(state->error) goto cleanup;
+    }
+
+    /*unknown chunks between IDAT and IEND*/
+    if(info.unknown_chunks_data[2]) {
+      state->error = addUnknownChunks(&outv, info.unknown_chunks_data[2], info.unknown_chunks_size[2]);
+      if(state->error) goto cleanup;
+    }
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+    state->error = addChunk_IEND(&outv);
+    if(state->error) goto cleanup;
+  }
+
+cleanup:
+  lodepng_info_cleanup(&info);
+  lodepng_free(data);
+
+  /*instead of cleaning the vector up, give it to the output*/
+  *out = outv.data;
+  *outsize = outv.size;
+
+  return state->error;
+}
+
+unsigned lodepng_encode_memory(unsigned char** out, size_t* outsize, const unsigned char* image,
+                               unsigned w, unsigned h, LodePNGColorType colortype, unsigned bitdepth) {
+  unsigned error;
+  LodePNGState state;
+  lodepng_state_init(&state);
+  state.info_raw.colortype = colortype;
+  state.info_raw.bitdepth = bitdepth;
+  state.info_png.color.colortype = colortype;
+  state.info_png.color.bitdepth = bitdepth;
+  lodepng_encode(out, outsize, image, w, h, &state);
+  error = state.error;
+  lodepng_state_cleanup(&state);
+  return error;
+}
+
+unsigned lodepng_encode32(unsigned char** out, size_t* outsize, const unsigned char* image, unsigned w, unsigned h) {
+  return lodepng_encode_memory(out, outsize, image, w, h, LCT_RGBA, 8);
+}
+
+unsigned lodepng_encode24(unsigned char** out, size_t* outsize, const unsigned char* image, unsigned w, unsigned h) {
+  return lodepng_encode_memory(out, outsize, image, w, h, LCT_RGB, 8);
+}
+
+#ifdef LODEPNG_COMPILE_DISK
+unsigned lodepng_encode_file(const char* filename, const unsigned char* image, unsigned w, unsigned h,
+                             LodePNGColorType colortype, unsigned bitdepth) {
+  unsigned char* buffer;
+  size_t buffersize;
+  unsigned error = lodepng_encode_memory(&buffer, &buffersize, image, w, h, colortype, bitdepth);
+  if(!error) error = lodepng_save_file(buffer, buffersize, filename);
+  lodepng_free(buffer);
+  return error;
+}
+
+unsigned lodepng_encode32_file(const char* filename, const unsigned char* image, unsigned w, unsigned h) {
+  return lodepng_encode_file(filename, image, w, h, LCT_RGBA, 8);
+}
+
+unsigned lodepng_encode24_file(const char* filename, const unsigned char* image, unsigned w, unsigned h) {
+  return lodepng_encode_file(filename, image, w, h, LCT_RGB, 8);
+}
+#endif /*LODEPNG_COMPILE_DISK*/
+
+void lodepng_encoder_settings_init(LodePNGEncoderSettings* settings) {
+  lodepng_compress_settings_init(&settings->zlibsettings);
+  settings->filter_palette_zero = 1;
+  settings->filter_strategy = LFS_MINSUM;
+  settings->auto_convert = 1;
+  settings->force_palette = 0;
+  settings->predefined_filters = 0;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+  settings->add_id = 0;
+  settings->text_compression = 1;
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+}
+
+#endif /*LODEPNG_COMPILE_ENCODER*/
+#endif /*LODEPNG_COMPILE_PNG*/
+
+#ifdef LODEPNG_COMPILE_ERROR_TEXT
+/*
+This returns the description of a numerical error code in English. This is also
+the documentation of all the error codes.
+*/
+const char* lodepng_error_text(unsigned code) {
+  switch(code) {
+    case 0: return "no error, everything went ok";
+    case 1: return "nothing done yet"; /*the Encoder/Decoder has done nothing yet, error checking makes no sense yet*/
+    case 10: return "end of input memory reached without huffman end code"; /*while huffman decoding*/
+    case 11: return "error in code tree made it jump outside of huffman tree"; /*while huffman decoding*/
+    case 13: return "problem while processing dynamic deflate block";
+    case 14: return "problem while processing dynamic deflate block";
+    case 15: return "problem while processing dynamic deflate block";
+    /*this error could happen if there are only 0 or 1 symbols present in the huffman code:*/
+    case 16: return "invalid code while processing dynamic deflate block";
+    case 17: return "end of out buffer memory reached while inflating";
+    case 18: return "invalid distance code while inflating";
+    case 19: return "end of out buffer memory reached while inflating";
+    case 20: return "invalid deflate block BTYPE encountered while decoding";
+    case 21: return "NLEN is not ones complement of LEN in a deflate block";
+
+    /*end of out buffer memory reached while inflating:
+    This can happen if the inflated deflate data is longer than the amount of bytes required to fill up
+    all the pixels of the image, given the color depth and image dimensions. Something that doesn't
+    happen in a normal, well encoded, PNG image.*/
+    case 22: return "end of out buffer memory reached while inflating";
+    case 23: return "end of in buffer memory reached while inflating";
+    case 24: return "invalid FCHECK in zlib header";
+    case 25: return "invalid compression method in zlib header";
+    case 26: return "FDICT encountered in zlib header while it's not used for PNG";
+    case 27: return "PNG file is smaller than a PNG header";
+    /*Checks the magic file header, the first 8 bytes of the PNG file*/
+    case 28: return "incorrect PNG signature, it's no PNG or corrupted";
+    case 29: return "first chunk is not the header chunk";
+    case 30: return "chunk length too large, chunk broken off at end of file";
+    case 31: return "illegal PNG color type or bpp";
+    case 32: return "illegal PNG compression method";
+    case 33: return "illegal PNG filter method";
+    case 34: return "illegal PNG interlace method";
+    case 35: return "chunk length of a chunk is too large or the chunk too small";
+    case 36: return "illegal PNG filter type encountered";
+    case 37: return "illegal bit depth for this color type given";
+    case 38: return "the palette is too small or too big"; /*0, or more than 256 colors*/
+    case 39: return "tRNS chunk before PLTE or has more entries than palette size";
+    case 40: return "tRNS chunk has wrong size for grayscale image";
+    case 41: return "tRNS chunk has wrong size for RGB image";
+    case 42: return "tRNS chunk appeared while it was not allowed for this color type";
+    case 43: return "bKGD chunk has wrong size for palette image";
+    case 44: return "bKGD chunk has wrong size for grayscale image";
+    case 45: return "bKGD chunk has wrong size for RGB image";
+    case 48: return "empty input buffer given to decoder. Maybe caused by non-existing file?";
+    case 49: return "jumped past memory while generating dynamic huffman tree";
+    case 50: return "jumped past memory while generating dynamic huffman tree";
+    case 51: return "jumped past memory while inflating huffman block";
+    case 52: return "jumped past memory while inflating";
+    case 53: return "size of zlib data too small";
+    case 54: return "repeat symbol in tree while there was no value symbol yet";
+    /*jumped past tree while generating huffman tree, this could be when the
+    tree will have more leaves than symbols after generating it out of the
+    given lengths. They call this an oversubscribed dynamic bit lengths tree in zlib.*/
+    case 55: return "jumped past tree while generating huffman tree";
+    case 56: return "given output image colortype or bitdepth not supported for color conversion";
+    case 57: return "invalid CRC encountered (checking CRC can be disabled)";
+    case 58: return "invalid ADLER32 encountered (checking ADLER32 can be disabled)";
+    case 59: return "requested color conversion not supported";
+    case 60: return "invalid window size given in the settings of the encoder (must be 0-32768)";
+    case 61: return "invalid BTYPE given in the settings of the encoder (only 0, 1 and 2 are allowed)";
+    /*LodePNG leaves the choice of RGB to grayscale conversion formula to the user.*/
+    case 62: return "conversion from color to grayscale not supported";
+    /*(2^31-1)*/
+    case 63: return "length of a chunk too long, max allowed for PNG is 2147483647 bytes per chunk";
+    /*this would result in the inability of a deflated block to ever contain an end code. It must be at least 1.*/
+    case 64: return "the length of the END symbol 256 in the Huffman tree is 0";
+    case 66: return "the length of a text chunk keyword given to the encoder is longer than the maximum of 79 bytes";
+    case 67: return "the length of a text chunk keyword given to the encoder is smaller than the minimum of 1 byte";
+    case 68: return "tried to encode a PLTE chunk with a palette that has less than 1 or more than 256 colors";
+    case 69: return "unknown chunk type with 'critical' flag encountered by the decoder";
+    case 71: return "invalid interlace mode given to encoder (must be 0 or 1)";
+    case 72: return "while decoding, invalid compression method encountering in zTXt or iTXt chunk (it must be 0)";
+    case 73: return "invalid tIME chunk size";
+    case 74: return "invalid pHYs chunk size";
+    /*length could be wrong, or data chopped off*/
+    case 75: return "no null termination char found while decoding text chunk";
+    case 76: return "iTXt chunk too short to contain required bytes";
+    case 77: return "integer overflow in buffer size";
+    case 78: return "failed to open file for reading"; /*file doesn't exist or couldn't be opened for reading*/
+    case 79: return "failed to open file for writing";
+    case 80: return "tried creating a tree of 0 symbols";
+    case 81: return "lazy matching at pos 0 is impossible";
+    case 82: return "color conversion to palette requested while a color isn't in palette, or index out of bounds";
+    case 83: return "memory allocation failed";
+    case 84: return "given image too small to contain all pixels to be encoded";
+    case 86: return "impossible offset in lz77 encoding (internal bug)";
+    case 87: return "must provide custom zlib function pointer if LODEPNG_COMPILE_ZLIB is not defined";
+    case 88: return "invalid filter strategy given for LodePNGEncoderSettings.filter_strategy";
+    case 89: return "text chunk keyword too short or long: must have size 1-79";
+    /*the windowsize in the LodePNGCompressSettings. Requiring POT(==> & instead of %) makes encoding 12% faster.*/
+    case 90: return "windowsize must be a power of two";
+    case 91: return "invalid decompressed idat size";
+    case 92: return "integer overflow due to too many pixels";
+    case 93: return "zero width or height is invalid";
+    case 94: return "header chunk must have a size of 13 bytes";
+    case 95: return "integer overflow with combined idat chunk size";
+    case 96: return "invalid gAMA chunk size";
+    case 97: return "invalid cHRM chunk size";
+    case 98: return "invalid sRGB chunk size";
+    case 99: return "invalid sRGB rendering intent";
+    case 100: return "invalid ICC profile color type, the PNG specification only allows RGB or GRAY";
+    case 101: return "PNG specification does not allow RGB ICC profile on gray color types and vice versa";
+    case 102: return "not allowed to set grayscale ICC profile with colored pixels by PNG specification";
+    case 103: return "invalid palette index in bKGD chunk. Maybe it came before PLTE chunk?";
+    case 104: return "invalid bKGD color while encoding (e.g. palette index out of range)";
+    case 105: return "integer overflow of bitsize";
+    case 106: return "PNG file must have PLTE chunk if color type is palette";
+    case 107: return "color convert from palette mode requested without setting the palette data in it";
+    case 108: return "tried to add more than 256 values to a palette";
+  }
+  return "unknown error code";
+}
+#endif /*LODEPNG_COMPILE_ERROR_TEXT*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* // C++ Wrapper                                                          // */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef LODEPNG_COMPILE_CPP
+namespace lodepng {
+
+#ifdef LODEPNG_COMPILE_DISK
+unsigned load_file(std::vector<unsigned char>& buffer, const std::string& filename) {
+  long size = lodepng_filesize(filename.c_str());
+  if(size < 0) return 78;
+  buffer.resize((size_t)size);
+  return size == 0 ? 0 : lodepng_buffer_file(&buffer[0], (size_t)size, filename.c_str());
+}
+
+/*write given buffer to the file, overwriting the file, it doesn't append to it.*/
+unsigned save_file(const std::vector<unsigned char>& buffer, const std::string& filename) {
+  return lodepng_save_file(buffer.empty() ? 0 : &buffer[0], buffer.size(), filename.c_str());
+}
+#endif /* LODEPNG_COMPILE_DISK */
+
+#ifdef LODEPNG_COMPILE_ZLIB
+#ifdef LODEPNG_COMPILE_DECODER
+unsigned decompress(std::vector<unsigned char>& out, const unsigned char* in, size_t insize,
+                    const LodePNGDecompressSettings& settings) {
+  unsigned char* buffer = 0;
+  size_t buffersize = 0;
+  unsigned error = zlib_decompress(&buffer, &buffersize, 0, in, insize, &settings);
+  if(buffer) {
+    out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+    lodepng_free(buffer);
+  }
+  return error;
+}
+
+unsigned decompress(std::vector<unsigned char>& out, const std::vector<unsigned char>& in,
+                    const LodePNGDecompressSettings& settings) {
+  return decompress(out, in.empty() ? 0 : &in[0], in.size(), settings);
+}
+#endif /* LODEPNG_COMPILE_DECODER */
+
+#ifdef LODEPNG_COMPILE_ENCODER
+unsigned compress(std::vector<unsigned char>& out, const unsigned char* in, size_t insize,
+                  const LodePNGCompressSettings& settings) {
+  unsigned char* buffer = 0;
+  size_t buffersize = 0;
+  unsigned error = zlib_compress(&buffer, &buffersize, in, insize, &settings);
+  if(buffer) {
+    out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+    lodepng_free(buffer);
+  }
+  return error;
+}
+
+unsigned compress(std::vector<unsigned char>& out, const std::vector<unsigned char>& in,
+                  const LodePNGCompressSettings& settings) {
+  return compress(out, in.empty() ? 0 : &in[0], in.size(), settings);
+}
+#endif /* LODEPNG_COMPILE_ENCODER */
+#endif /* LODEPNG_COMPILE_ZLIB */
+
+
+#ifdef LODEPNG_COMPILE_PNG
+
+State::State() {
+  lodepng_state_init(this);
+}
+
+State::State(const State& other) {
+  lodepng_state_init(this);
+  lodepng_state_copy(this, &other);
+}
+
+State::~State() {
+  lodepng_state_cleanup(this);
+}
+
+State& State::operator=(const State& other) {
+  lodepng_state_copy(this, &other);
+  return *this;
+}
+
+#ifdef LODEPNG_COMPILE_DECODER
+
+unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h, const unsigned char* in,
+                size_t insize, LodePNGColorType colortype, unsigned bitdepth) {
+  unsigned char* buffer = 0;
+  unsigned error = lodepng_decode_memory(&buffer, &w, &h, in, insize, colortype, bitdepth);
+  if(buffer && !error) {
+    State state;
+    state.info_raw.colortype = colortype;
+    state.info_raw.bitdepth = bitdepth;
+    size_t buffersize = lodepng_get_raw_size(w, h, &state.info_raw);
+    out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+  }
+  lodepng_free(buffer);
+  return error;
+}
+
+unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
+                const std::vector<unsigned char>& in, LodePNGColorType colortype, unsigned bitdepth) {
+  return decode(out, w, h, in.empty() ? 0 : &in[0], (unsigned)in.size(), colortype, bitdepth);
+}
+
+unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
+                State& state,
+                const unsigned char* in, size_t insize) {
+  unsigned char* buffer = NULL;
+  unsigned error = lodepng_decode(&buffer, &w, &h, &state, in, insize);
+  if(buffer && !error) {
+    size_t buffersize = lodepng_get_raw_size(w, h, &state.info_raw);
+    out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+  }
+  lodepng_free(buffer);
+  return error;
+}
+
+unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
+                State& state,
+                const std::vector<unsigned char>& in) {
+  return decode(out, w, h, state, in.empty() ? 0 : &in[0], in.size());
+}
+
+#ifdef LODEPNG_COMPILE_DISK
+unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h, const std::string& filename,
+                LodePNGColorType colortype, unsigned bitdepth) {
+  std::vector<unsigned char> buffer;
+  /* safe output values in case error happens */
+  w = h = 0;
+  unsigned error = load_file(buffer, filename);
+  if(error) return error;
+  return decode(out, w, h, buffer, colortype, bitdepth);
+}
+#endif /* LODEPNG_COMPILE_DECODER */
+#endif /* LODEPNG_COMPILE_DISK */
+
+#ifdef LODEPNG_COMPILE_ENCODER
+unsigned encode(std::vector<unsigned char>& out, const unsigned char* in, unsigned w, unsigned h,
+                LodePNGColorType colortype, unsigned bitdepth) {
+  unsigned char* buffer;
+  size_t buffersize;
+  unsigned error = lodepng_encode_memory(&buffer, &buffersize, in, w, h, colortype, bitdepth);
+  if(buffer) {
+    out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+    lodepng_free(buffer);
+  }
+  return error;
+}
+
+unsigned encode(std::vector<unsigned char>& out,
+                const std::vector<unsigned char>& in, unsigned w, unsigned h,
+                LodePNGColorType colortype, unsigned bitdepth) {
+  if(lodepng_get_raw_size_lct(w, h, colortype, bitdepth) > in.size()) return 84;
+  return encode(out, in.empty() ? 0 : &in[0], w, h, colortype, bitdepth);
+}
+
+unsigned encode(std::vector<unsigned char>& out,
+                const unsigned char* in, unsigned w, unsigned h,
+                State& state) {
+  unsigned char* buffer;
+  size_t buffersize;
+  unsigned error = lodepng_encode(&buffer, &buffersize, in, w, h, &state);
+  if(buffer) {
+    out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+    lodepng_free(buffer);
+  }
+  return error;
+}
+
+unsigned encode(std::vector<unsigned char>& out,
+                const std::vector<unsigned char>& in, unsigned w, unsigned h,
+                State& state) {
+  if(lodepng_get_raw_size(w, h, &state.info_raw) > in.size()) return 84;
+  return encode(out, in.empty() ? 0 : &in[0], w, h, state);
+}
+
+#ifdef LODEPNG_COMPILE_DISK
+unsigned encode(const std::string& filename,
+                const unsigned char* in, unsigned w, unsigned h,
+                LodePNGColorType colortype, unsigned bitdepth) {
+  std::vector<unsigned char> buffer;
+  unsigned error = encode(buffer, in, w, h, colortype, bitdepth);
+  if(!error) error = save_file(buffer, filename);
+  return error;
+}
+
+unsigned encode(const std::string& filename,
+                const std::vector<unsigned char>& in, unsigned w, unsigned h,
+                LodePNGColorType colortype, unsigned bitdepth) {
+  if(lodepng_get_raw_size_lct(w, h, colortype, bitdepth) > in.size()) return 84;
+  return encode(filename, in.empty() ? 0 : &in[0], w, h, colortype, bitdepth);
+}
+#endif /* LODEPNG_COMPILE_DISK */
+#endif /* LODEPNG_COMPILE_ENCODER */
+#endif /* LODEPNG_COMPILE_PNG */
+} /* namespace lodepng */
+#endif /*LODEPNG_COMPILE_CPP*/

+ 1945 - 0
Tests/Dependencies/lodepng/lodepng.h

@@ -0,0 +1,1945 @@
+/*
+LodePNG version 20200306
+
+Copyright (c) 2005-2020 Lode Vandevenne
+
+This software is provided 'as-is', without any express or implied
+warranty. In no event will the authors be held liable for any damages
+arising from the use of this software.
+
+Permission is granted to anyone to use this software for any purpose,
+including commercial applications, and to alter it and redistribute it
+freely, subject to the following restrictions:
+
+    1. The origin of this software must not be misrepresented; you must not
+    claim that you wrote the original software. If you use this software
+    in a product, an acknowledgment in the product documentation would be
+    appreciated but is not required.
+
+    2. Altered source versions must be plainly marked as such, and must not be
+    misrepresented as being the original software.
+
+    3. This notice may not be removed or altered from any source
+    distribution.
+*/
+
+#ifndef LODEPNG_H
+#define LODEPNG_H
+
+#include <string.h> /*for size_t*/
+
+extern const char* LODEPNG_VERSION_STRING;
+
+/*
+The following #defines are used to create code sections. They can be disabled
+to disable code sections, which can give faster compile time and smaller binary.
+The "NO_COMPILE" defines are designed to be used to pass as defines to the
+compiler command to disable them without modifying this header, e.g.
+-DLODEPNG_NO_COMPILE_ZLIB for gcc.
+In addition to those below, you can also define LODEPNG_NO_COMPILE_CRC to
+allow implementing a custom lodepng_crc32.
+*/
+/*deflate & zlib. If disabled, you must specify alternative zlib functions in
+the custom_zlib field of the compress and decompress settings*/
+#ifndef LODEPNG_NO_COMPILE_ZLIB
+#define LODEPNG_COMPILE_ZLIB
+#endif
+
+/*png encoder and png decoder*/
+#ifndef LODEPNG_NO_COMPILE_PNG
+#define LODEPNG_COMPILE_PNG
+#endif
+
+/*deflate&zlib decoder and png decoder*/
+#ifndef LODEPNG_NO_COMPILE_DECODER
+#define LODEPNG_COMPILE_DECODER
+#endif
+
+/*deflate&zlib encoder and png encoder*/
+#ifndef LODEPNG_NO_COMPILE_ENCODER
+#define LODEPNG_COMPILE_ENCODER
+#endif
+
+/*the optional built in harddisk file loading and saving functions*/
+#ifndef LODEPNG_NO_COMPILE_DISK
+#define LODEPNG_COMPILE_DISK
+#endif
+
+/*support for chunks other than IHDR, IDAT, PLTE, tRNS, IEND: ancillary and unknown chunks*/
+#ifndef LODEPNG_NO_COMPILE_ANCILLARY_CHUNKS
+#define LODEPNG_COMPILE_ANCILLARY_CHUNKS
+#endif
+
+/*ability to convert error numerical codes to English text string*/
+#ifndef LODEPNG_NO_COMPILE_ERROR_TEXT
+#define LODEPNG_COMPILE_ERROR_TEXT
+#endif
+
+/*Compile the default allocators (C's free, malloc and realloc). If you disable this,
+you can define the functions lodepng_free, lodepng_malloc and lodepng_realloc in your
+source files with custom allocators.*/
+#ifndef LODEPNG_NO_COMPILE_ALLOCATORS
+#define LODEPNG_COMPILE_ALLOCATORS
+#endif
+
+/*compile the C++ version (you can disable the C++ wrapper here even when compiling for C++)*/
+#ifdef __cplusplus
+#ifndef LODEPNG_NO_COMPILE_CPP
+#define LODEPNG_COMPILE_CPP
+#endif
+#endif
+
+#ifdef LODEPNG_COMPILE_CPP
+#include <vector>
+#include <string>
+#endif /*LODEPNG_COMPILE_CPP*/
+
+#ifdef LODEPNG_COMPILE_PNG
+/*The PNG color types (also used for raw image).*/
+typedef enum LodePNGColorType {
+  LCT_GREY = 0, /*grayscale: 1,2,4,8,16 bit*/
+  LCT_RGB = 2, /*RGB: 8,16 bit*/
+  LCT_PALETTE = 3, /*palette: 1,2,4,8 bit*/
+  LCT_GREY_ALPHA = 4, /*grayscale with alpha: 8,16 bit*/
+  LCT_RGBA = 6, /*RGB with alpha: 8,16 bit*/
+  /*LCT_MAX_OCTET_VALUE lets the compiler allow this enum to represent any invalid
+  byte value from 0 to 255 that could be present in an invalid PNG file header. Do
+  not use, compare with or set the name LCT_MAX_OCTET_VALUE, instead either use
+  the valid color type names above, or numeric values like 1 or 7 when checking for
+  particular disallowed color type byte values, or cast to integer to print it.*/
+  LCT_MAX_OCTET_VALUE = 255
+} LodePNGColorType;
+
+#ifdef LODEPNG_COMPILE_DECODER
+/*
+Converts PNG data in memory to raw pixel data.
+out: Output parameter. Pointer to buffer that will contain the raw pixel data.
+     After decoding, its size is w * h * (bytes per pixel) bytes larger than
+     initially. Bytes per pixel depends on colortype and bitdepth.
+     Must be freed after usage with free(*out).
+     Note: for 16-bit per channel colors, uses big endian format like PNG does.
+w: Output parameter. Pointer to width of pixel data.
+h: Output parameter. Pointer to height of pixel data.
+in: Memory buffer with the PNG file.
+insize: size of the in buffer.
+colortype: the desired color type for the raw output image. See explanation on PNG color types.
+bitdepth: the desired bit depth for the raw output image. See explanation on PNG color types.
+Return value: LodePNG error code (0 means no error).
+*/
+unsigned lodepng_decode_memory(unsigned char** out, unsigned* w, unsigned* h,
+                               const unsigned char* in, size_t insize,
+                               LodePNGColorType colortype, unsigned bitdepth);
+
+/*Same as lodepng_decode_memory, but always decodes to 32-bit RGBA raw image*/
+unsigned lodepng_decode32(unsigned char** out, unsigned* w, unsigned* h,
+                          const unsigned char* in, size_t insize);
+
+/*Same as lodepng_decode_memory, but always decodes to 24-bit RGB raw image*/
+unsigned lodepng_decode24(unsigned char** out, unsigned* w, unsigned* h,
+                          const unsigned char* in, size_t insize);
+
+#ifdef LODEPNG_COMPILE_DISK
+/*
+Load PNG from disk, from file with given name.
+Same as the other decode functions, but instead takes a filename as input.
+*/
+unsigned lodepng_decode_file(unsigned char** out, unsigned* w, unsigned* h,
+                             const char* filename,
+                             LodePNGColorType colortype, unsigned bitdepth);
+
+/*Same as lodepng_decode_file, but always decodes to 32-bit RGBA raw image.*/
+unsigned lodepng_decode32_file(unsigned char** out, unsigned* w, unsigned* h,
+                               const char* filename);
+
+/*Same as lodepng_decode_file, but always decodes to 24-bit RGB raw image.*/
+unsigned lodepng_decode24_file(unsigned char** out, unsigned* w, unsigned* h,
+                               const char* filename);
+#endif /*LODEPNG_COMPILE_DISK*/
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+
+#ifdef LODEPNG_COMPILE_ENCODER
+/*
+Converts raw pixel data into a PNG image in memory. The colortype and bitdepth
+  of the output PNG image cannot be chosen, they are automatically determined
+  by the colortype, bitdepth and content of the input pixel data.
+  Note: for 16-bit per channel colors, needs big endian format like PNG does.
+out: Output parameter. Pointer to buffer that will contain the PNG image data.
+     Must be freed after usage with free(*out).
+outsize: Output parameter. Pointer to the size in bytes of the out buffer.
+image: The raw pixel data to encode. The size of this buffer should be
+       w * h * (bytes per pixel), bytes per pixel depends on colortype and bitdepth.
+w: width of the raw pixel data in pixels.
+h: height of the raw pixel data in pixels.
+colortype: the color type of the raw input image. See explanation on PNG color types.
+bitdepth: the bit depth of the raw input image. See explanation on PNG color types.
+Return value: LodePNG error code (0 means no error).
+*/
+unsigned lodepng_encode_memory(unsigned char** out, size_t* outsize,
+                               const unsigned char* image, unsigned w, unsigned h,
+                               LodePNGColorType colortype, unsigned bitdepth);
+
+/*Same as lodepng_encode_memory, but always encodes from 32-bit RGBA raw image.*/
+unsigned lodepng_encode32(unsigned char** out, size_t* outsize,
+                          const unsigned char* image, unsigned w, unsigned h);
+
+/*Same as lodepng_encode_memory, but always encodes from 24-bit RGB raw image.*/
+unsigned lodepng_encode24(unsigned char** out, size_t* outsize,
+                          const unsigned char* image, unsigned w, unsigned h);
+
+#ifdef LODEPNG_COMPILE_DISK
+/*
+Converts raw pixel data into a PNG file on disk.
+Same as the other encode functions, but instead takes a filename as output.
+NOTE: This overwrites existing files without warning!
+*/
+unsigned lodepng_encode_file(const char* filename,
+                             const unsigned char* image, unsigned w, unsigned h,
+                             LodePNGColorType colortype, unsigned bitdepth);
+
+/*Same as lodepng_encode_file, but always encodes from 32-bit RGBA raw image.*/
+unsigned lodepng_encode32_file(const char* filename,
+                               const unsigned char* image, unsigned w, unsigned h);
+
+/*Same as lodepng_encode_file, but always encodes from 24-bit RGB raw image.*/
+unsigned lodepng_encode24_file(const char* filename,
+                               const unsigned char* image, unsigned w, unsigned h);
+#endif /*LODEPNG_COMPILE_DISK*/
+#endif /*LODEPNG_COMPILE_ENCODER*/
+
+
+#ifdef LODEPNG_COMPILE_CPP
+namespace lodepng {
+#ifdef LODEPNG_COMPILE_DECODER
+/*Same as lodepng_decode_memory, but decodes to an std::vector. The colortype
+is the format to output the pixels to. Default is RGBA 8-bit per channel.*/
+unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
+                const unsigned char* in, size_t insize,
+                LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
+unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
+                const std::vector<unsigned char>& in,
+                LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
+#ifdef LODEPNG_COMPILE_DISK
+/*
+Converts PNG file from disk to raw pixel data in memory.
+Same as the other decode functions, but instead takes a filename as input.
+*/
+unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
+                const std::string& filename,
+                LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
+#endif /* LODEPNG_COMPILE_DISK */
+#endif /* LODEPNG_COMPILE_DECODER */
+
+#ifdef LODEPNG_COMPILE_ENCODER
+/*Same as lodepng_encode_memory, but encodes to an std::vector. colortype
+is that of the raw input data. The output PNG color type will be auto chosen.*/
+unsigned encode(std::vector<unsigned char>& out,
+                const unsigned char* in, unsigned w, unsigned h,
+                LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
+unsigned encode(std::vector<unsigned char>& out,
+                const std::vector<unsigned char>& in, unsigned w, unsigned h,
+                LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
+#ifdef LODEPNG_COMPILE_DISK
+/*
+Converts 32-bit RGBA raw pixel data into a PNG file on disk.
+Same as the other encode functions, but instead takes a filename as output.
+NOTE: This overwrites existing files without warning!
+*/
+unsigned encode(const std::string& filename,
+                const unsigned char* in, unsigned w, unsigned h,
+                LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
+unsigned encode(const std::string& filename,
+                const std::vector<unsigned char>& in, unsigned w, unsigned h,
+                LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
+#endif /* LODEPNG_COMPILE_DISK */
+#endif /* LODEPNG_COMPILE_ENCODER */
+} /* namespace lodepng */
+#endif /*LODEPNG_COMPILE_CPP*/
+#endif /*LODEPNG_COMPILE_PNG*/
+
+#ifdef LODEPNG_COMPILE_ERROR_TEXT
+/*Returns an English description of the numerical error code.*/
+const char* lodepng_error_text(unsigned code);
+#endif /*LODEPNG_COMPILE_ERROR_TEXT*/
+
+#ifdef LODEPNG_COMPILE_DECODER
+/*Settings for zlib decompression*/
+typedef struct LodePNGDecompressSettings LodePNGDecompressSettings;
+struct LodePNGDecompressSettings {
+  /* Check LodePNGDecoderSettings for more ignorable errors such as ignore_crc */
+  unsigned ignore_adler32; /*if 1, continue and don't give an error message if the Adler32 checksum is corrupted*/
+  unsigned ignore_nlen; /*ignore complement of len checksum in uncompressed blocks*/
+
+  /*use custom zlib decoder instead of built in one (default: null)*/
+  unsigned (*custom_zlib)(unsigned char**, size_t*,
+                          const unsigned char*, size_t,
+                          const LodePNGDecompressSettings*);
+  /*use custom deflate decoder instead of built in one (default: null)
+  if custom_zlib is not null, custom_inflate is ignored (the zlib format uses deflate)*/
+  unsigned (*custom_inflate)(unsigned char**, size_t*,
+                             const unsigned char*, size_t,
+                             const LodePNGDecompressSettings*);
+
+  const void* custom_context; /*optional custom settings for custom functions*/
+};
+
+extern const LodePNGDecompressSettings lodepng_default_decompress_settings;
+void lodepng_decompress_settings_init(LodePNGDecompressSettings* settings);
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+#ifdef LODEPNG_COMPILE_ENCODER
+/*
+Settings for zlib compression. Tweaking these settings tweaks the balance
+between speed and compression ratio.
+*/
+typedef struct LodePNGCompressSettings LodePNGCompressSettings;
+struct LodePNGCompressSettings /*deflate = compress*/ {
+  /*LZ77 related settings*/
+  unsigned btype; /*the block type for LZ (0, 1, 2 or 3, see zlib standard). Should be 2 for proper compression.*/
+  unsigned use_lz77; /*whether or not to use LZ77. Should be 1 for proper compression.*/
+  unsigned windowsize; /*must be a power of two <= 32768. higher compresses more but is slower. Default value: 2048.*/
+  unsigned minmatch; /*minimum lz77 length. 3 is normally best, 6 can be better for some PNGs. Default: 0*/
+  unsigned nicematch; /*stop searching if >= this length found. Set to 258 for best compression. Default: 128*/
+  unsigned lazymatching; /*use lazy matching: better compression but a bit slower. Default: true*/
+
+  /*use custom zlib encoder instead of built in one (default: null)*/
+  unsigned (*custom_zlib)(unsigned char**, size_t*,
+                          const unsigned char*, size_t,
+                          const LodePNGCompressSettings*);
+  /*use custom deflate encoder instead of built in one (default: null)
+  if custom_zlib is used, custom_deflate is ignored since only the built in
+  zlib function will call custom_deflate*/
+  unsigned (*custom_deflate)(unsigned char**, size_t*,
+                             const unsigned char*, size_t,
+                             const LodePNGCompressSettings*);
+
+  const void* custom_context; /*optional custom settings for custom functions*/
+};
+
+extern const LodePNGCompressSettings lodepng_default_compress_settings;
+void lodepng_compress_settings_init(LodePNGCompressSettings* settings);
+#endif /*LODEPNG_COMPILE_ENCODER*/
+
+#ifdef LODEPNG_COMPILE_PNG
+/*
+Color mode of an image. Contains all information required to decode the pixel
+bits to RGBA colors. This information is the same as used in the PNG file
+format, and is used both for PNG and raw image data in LodePNG.
+*/
+typedef struct LodePNGColorMode {
+  /*header (IHDR)*/
+  LodePNGColorType colortype; /*color type, see PNG standard or documentation further in this header file*/
+  unsigned bitdepth;  /*bits per sample, see PNG standard or documentation further in this header file*/
+
+  /*
+  palette (PLTE and tRNS)
+
+  Dynamically allocated with the colors of the palette, including alpha.
+  This field may not be allocated directly, use lodepng_color_mode_init first,
+  then lodepng_palette_add per color to correctly initialize it (to ensure size
+  of exactly 1024 bytes).
+
+  The alpha channels must be set as well, set them to 255 for opaque images.
+
+  When decoding, by default you can ignore this palette, since LodePNG already
+  fills the palette colors in the pixels of the raw RGBA output.
+
+  The palette is only supported for color type 3.
+  */
+  unsigned char* palette; /*palette in RGBARGBA... order. Must be either 0, or when allocated must have 1024 bytes*/
+  size_t palettesize; /*palette size in number of colors (amount of used bytes is 4 * palettesize)*/
+
+  /*
+  transparent color key (tRNS)
+
+  This color uses the same bit depth as the bitdepth value in this struct, which can be 1-bit to 16-bit.
+  For grayscale PNGs, r, g and b will all 3 be set to the same.
+
+  When decoding, by default you can ignore this information, since LodePNG sets
+  pixels with this key to transparent already in the raw RGBA output.
+
+  The color key is only supported for color types 0 and 2.
+  */
+  unsigned key_defined; /*is a transparent color key given? 0 = false, 1 = true*/
+  unsigned key_r;       /*red/grayscale component of color key*/
+  unsigned key_g;       /*green component of color key*/
+  unsigned key_b;       /*blue component of color key*/
+} LodePNGColorMode;
+
+/*init, cleanup and copy functions to use with this struct*/
+void lodepng_color_mode_init(LodePNGColorMode* info);
+void lodepng_color_mode_cleanup(LodePNGColorMode* info);
+/*return value is error code (0 means no error)*/
+unsigned lodepng_color_mode_copy(LodePNGColorMode* dest, const LodePNGColorMode* source);
+/* Makes a temporary LodePNGColorMode that does not need cleanup (no palette) */
+LodePNGColorMode lodepng_color_mode_make(LodePNGColorType colortype, unsigned bitdepth);
+
+void lodepng_palette_clear(LodePNGColorMode* info);
+/*add 1 color to the palette*/
+unsigned lodepng_palette_add(LodePNGColorMode* info,
+                             unsigned char r, unsigned char g, unsigned char b, unsigned char a);
+
+/*get the total amount of bits per pixel, based on colortype and bitdepth in the struct*/
+unsigned lodepng_get_bpp(const LodePNGColorMode* info);
+/*get the amount of color channels used, based on colortype in the struct.
+If a palette is used, it counts as 1 channel.*/
+unsigned lodepng_get_channels(const LodePNGColorMode* info);
+/*is it a grayscale type? (only colortype 0 or 4)*/
+unsigned lodepng_is_greyscale_type(const LodePNGColorMode* info);
+/*has it got an alpha channel? (only colortype 2 or 6)*/
+unsigned lodepng_is_alpha_type(const LodePNGColorMode* info);
+/*has it got a palette? (only colortype 3)*/
+unsigned lodepng_is_palette_type(const LodePNGColorMode* info);
+/*only returns true if there is a palette and there is a value in the palette with alpha < 255.
+Loops through the palette to check this.*/
+unsigned lodepng_has_palette_alpha(const LodePNGColorMode* info);
+/*
+Check if the given color info indicates the possibility of having non-opaque pixels in the PNG image.
+Returns true if the image can have translucent or invisible pixels (it still be opaque if it doesn't use such pixels).
+Returns false if the image can only have opaque pixels.
+In detail, it returns true only if it's a color type with alpha, or has a palette with non-opaque values,
+or if "key_defined" is true.
+*/
+unsigned lodepng_can_have_alpha(const LodePNGColorMode* info);
+/*Returns the byte size of a raw image buffer with given width, height and color mode*/
+size_t lodepng_get_raw_size(unsigned w, unsigned h, const LodePNGColorMode* color);
+
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+/*The information of a Time chunk in PNG.*/
+typedef struct LodePNGTime {
+  unsigned year;    /*2 bytes used (0-65535)*/
+  unsigned month;   /*1-12*/
+  unsigned day;     /*1-31*/
+  unsigned hour;    /*0-23*/
+  unsigned minute;  /*0-59*/
+  unsigned second;  /*0-60 (to allow for leap seconds)*/
+} LodePNGTime;
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+
+/*Information about the PNG image, except pixels, width and height.*/
+typedef struct LodePNGInfo {
+  /*header (IHDR), palette (PLTE) and transparency (tRNS) chunks*/
+  unsigned compression_method;/*compression method of the original file. Always 0.*/
+  unsigned filter_method;     /*filter method of the original file*/
+  unsigned interlace_method;  /*interlace method of the original file: 0=none, 1=Adam7*/
+  LodePNGColorMode color;     /*color type and bits, palette and transparency of the PNG file*/
+
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+  /*
+  Suggested background color chunk (bKGD)
+
+  This uses the same color mode and bit depth as the PNG (except no alpha channel),
+  with values truncated to the bit depth in the unsigned integer.
+
+  For grayscale and palette PNGs, the value is stored in background_r. The values
+  in background_g and background_b are then unused.
+
+  So when decoding, you may get these in a different color mode than the one you requested
+  for the raw pixels.
+
+  When encoding with auto_convert, you must use the color model defined in info_png.color for
+  these values. The encoder normally ignores info_png.color when auto_convert is on, but will
+  use it to interpret these values (and convert copies of them to its chosen color model).
+
+  When encoding, avoid setting this to an expensive color, such as a non-gray value
+  when the image is gray, or the compression will be worse since it will be forced to
+  write the PNG with a more expensive color mode (when auto_convert is on).
+
+  The decoder does not use this background color to edit the color of pixels. This is a
+  completely optional metadata feature.
+  */
+  unsigned background_defined; /*is a suggested background color given?*/
+  unsigned background_r;       /*red/gray/palette component of suggested background color*/
+  unsigned background_g;       /*green component of suggested background color*/
+  unsigned background_b;       /*blue component of suggested background color*/
+
+  /*
+  non-international text chunks (tEXt and zTXt)
+
+  The char** arrays each contain num strings. The actual messages are in
+  text_strings, while text_keys are keywords that give a short description what
+  the actual text represents, e.g. Title, Author, Description, or anything else.
+
+  All the string fields below including keys, names and language tags are null terminated.
+  The PNG specification uses null characters for the keys, names and tags, and forbids null
+  characters to appear in the main text which is why we can use null termination everywhere here.
+
+  A keyword is minimum 1 character and maximum 79 characters long. It's
+  discouraged to use a single line length longer than 79 characters for texts.
+
+  Don't allocate these text buffers yourself. Use the init/cleanup functions
+  correctly and use lodepng_add_text and lodepng_clear_text.
+  */
+  size_t text_num; /*the amount of texts in these char** buffers (there may be more texts in itext)*/
+  char** text_keys; /*the keyword of a text chunk (e.g. "Comment")*/
+  char** text_strings; /*the actual text*/
+
+  /*
+  international text chunks (iTXt)
+  Similar to the non-international text chunks, but with additional strings
+  "langtags" and "transkeys".
+  */
+  size_t itext_num; /*the amount of international texts in this PNG*/
+  char** itext_keys; /*the English keyword of the text chunk (e.g. "Comment")*/
+  char** itext_langtags; /*language tag for this text's language, ISO/IEC 646 string, e.g. ISO 639 language tag*/
+  char** itext_transkeys; /*keyword translated to the international language - UTF-8 string*/
+  char** itext_strings; /*the actual international text - UTF-8 string*/
+
+  /*time chunk (tIME)*/
+  unsigned time_defined; /*set to 1 to make the encoder generate a tIME chunk*/
+  LodePNGTime time;
+
+  /*phys chunk (pHYs)*/
+  unsigned phys_defined; /*if 0, there is no pHYs chunk and the values below are undefined, if 1 else there is one*/
+  unsigned phys_x; /*pixels per unit in x direction*/
+  unsigned phys_y; /*pixels per unit in y direction*/
+  unsigned phys_unit; /*may be 0 (unknown unit) or 1 (metre)*/
+
+  /*
+  Color profile related chunks: gAMA, cHRM, sRGB, iCPP
+
+  LodePNG does not apply any color conversions on pixels in the encoder or decoder and does not interpret these color
+  profile values. It merely passes on the information. If you wish to use color profiles and convert colors, please
+  use these values with a color management library.
+
+  See the PNG, ICC and sRGB specifications for more information about the meaning of these values.
+  */
+
+  /* gAMA chunk: optional, overridden by sRGB or iCCP if those are present. */
+  unsigned gama_defined; /* Whether a gAMA chunk is present (0 = not present, 1 = present). */
+  unsigned gama_gamma;   /* Gamma exponent times 100000 */
+
+  /* cHRM chunk: optional, overridden by sRGB or iCCP if those are present. */
+  unsigned chrm_defined; /* Whether a cHRM chunk is present (0 = not present, 1 = present). */
+  unsigned chrm_white_x; /* White Point x times 100000 */
+  unsigned chrm_white_y; /* White Point y times 100000 */
+  unsigned chrm_red_x;   /* Red x times 100000 */
+  unsigned chrm_red_y;   /* Red y times 100000 */
+  unsigned chrm_green_x; /* Green x times 100000 */
+  unsigned chrm_green_y; /* Green y times 100000 */
+  unsigned chrm_blue_x;  /* Blue x times 100000 */
+  unsigned chrm_blue_y;  /* Blue y times 100000 */
+
+  /*
+  sRGB chunk: optional. May not appear at the same time as iCCP.
+  If gAMA is also present gAMA must contain value 45455.
+  If cHRM is also present cHRM must contain respectively 31270,32900,64000,33000,30000,60000,15000,6000.
+  */
+  unsigned srgb_defined; /* Whether an sRGB chunk is present (0 = not present, 1 = present). */
+  unsigned srgb_intent;  /* Rendering intent: 0=perceptual, 1=rel. colorimetric, 2=saturation, 3=abs. colorimetric */
+
+  /*
+  iCCP chunk: optional. May not appear at the same time as sRGB.
+
+  LodePNG does not parse or use the ICC profile (except its color space header field for an edge case), a
+  separate library to handle the ICC data (not included in LodePNG) format is needed to use it for color
+  management and conversions.
+
+  For encoding, if iCCP is present, gAMA and cHRM are recommended to be added as well with values that match the ICC
+  profile as closely as possible, if you wish to do this you should provide the correct values for gAMA and cHRM and
+  enable their '_defined' flags since LodePNG will not automatically compute them from the ICC profile.
+
+  For encoding, the ICC profile is required by the PNG specification to be an "RGB" profile for non-gray
+  PNG color types and a "GRAY" profile for gray PNG color types. If you disable auto_convert, you must ensure
+  the ICC profile type matches your requested color type, else the encoder gives an error. If auto_convert is
+  enabled (the default), and the ICC profile is not a good match for the pixel data, this will result in an encoder
+  error if the pixel data has non-gray pixels for a GRAY profile, or a silent less-optimal compression of the pixel
+  data if the pixels could be encoded as grayscale but the ICC profile is RGB.
+
+  To avoid this do not set an ICC profile in the image unless there is a good reason for it, and when doing so
+  make sure you compute it carefully to avoid the above problems.
+  */
+  unsigned iccp_defined;      /* Whether an iCCP chunk is present (0 = not present, 1 = present). */
+  char* iccp_name;            /* Null terminated string with profile name, 1-79 bytes */
+  /*
+  The ICC profile in iccp_profile_size bytes.
+  Don't allocate this buffer yourself. Use the init/cleanup functions
+  correctly and use lodepng_set_icc and lodepng_clear_icc.
+  */
+  unsigned char* iccp_profile;
+  unsigned iccp_profile_size; /* The size of iccp_profile in bytes */
+
+  /* End of color profile related chunks */
+
+
+  /*
+  unknown chunks: chunks not known by LodePNG, passed on byte for byte.
+
+  There are 3 buffers, one for each position in the PNG where unknown chunks can appear.
+  Each buffer contains all unknown chunks for that position consecutively.
+  The 3 positions are:
+  0: between IHDR and PLTE, 1: between PLTE and IDAT, 2: between IDAT and IEND.
+
+  For encoding, do not store critical chunks or known chunks that are enabled with a "_defined" flag
+  above in here, since the encoder will blindly follow this and could then encode an invalid PNG file
+  (such as one with two IHDR chunks or the disallowed combination of sRGB with iCCP). But do use
+  this if you wish to store an ancillary chunk that is not supported by LodePNG (such as sPLT or hIST),
+  or any non-standard PNG chunk.
+
+  Do not allocate or traverse this data yourself. Use the chunk traversing functions declared
+  later, such as lodepng_chunk_next and lodepng_chunk_append, to read/write this struct.
+  */
+  unsigned char* unknown_chunks_data[3];
+  size_t unknown_chunks_size[3]; /*size in bytes of the unknown chunks, given for protection*/
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+} LodePNGInfo;
+
+/*init, cleanup and copy functions to use with this struct*/
+void lodepng_info_init(LodePNGInfo* info);
+void lodepng_info_cleanup(LodePNGInfo* info);
+/*return value is error code (0 means no error)*/
+unsigned lodepng_info_copy(LodePNGInfo* dest, const LodePNGInfo* source);
+
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+unsigned lodepng_add_text(LodePNGInfo* info, const char* key, const char* str); /*push back both texts at once*/
+void lodepng_clear_text(LodePNGInfo* info); /*use this to clear the texts again after you filled them in*/
+
+unsigned lodepng_add_itext(LodePNGInfo* info, const char* key, const char* langtag,
+                           const char* transkey, const char* str); /*push back the 4 texts of 1 chunk at once*/
+void lodepng_clear_itext(LodePNGInfo* info); /*use this to clear the itexts again after you filled them in*/
+
+/*replaces if exists*/
+unsigned lodepng_set_icc(LodePNGInfo* info, const char* name, const unsigned char* profile, unsigned profile_size);
+void lodepng_clear_icc(LodePNGInfo* info); /*use this to clear the texts again after you filled them in*/
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+
+/*
+Converts raw buffer from one color type to another color type, based on
+LodePNGColorMode structs to describe the input and output color type.
+See the reference manual at the end of this header file to see which color conversions are supported.
+return value = LodePNG error code (0 if all went ok, an error if the conversion isn't supported)
+The out buffer must have size (w * h * bpp + 7) / 8, where bpp is the bits per pixel
+of the output color type (lodepng_get_bpp).
+For < 8 bpp images, there should not be padding bits at the end of scanlines.
+For 16-bit per channel colors, uses big endian format like PNG does.
+Return value is LodePNG error code
+*/
+unsigned lodepng_convert(unsigned char* out, const unsigned char* in,
+                         const LodePNGColorMode* mode_out, const LodePNGColorMode* mode_in,
+                         unsigned w, unsigned h);
+
+#ifdef LODEPNG_COMPILE_DECODER
+/*
+Settings for the decoder. This contains settings for the PNG and the Zlib
+decoder, but not the Info settings from the Info structs.
+*/
+typedef struct LodePNGDecoderSettings {
+  LodePNGDecompressSettings zlibsettings; /*in here is the setting to ignore Adler32 checksums*/
+
+  /* Check LodePNGDecompressSettings for more ignorable errors such as ignore_adler32 */
+  unsigned ignore_crc; /*ignore CRC checksums*/
+  unsigned ignore_critical; /*ignore unknown critical chunks*/
+  unsigned ignore_end; /*ignore issues at end of file if possible (missing IEND chunk, too large chunk, ...)*/
+  /* TODO: make a system involving warnings with levels and a strict mode instead. Other potentially recoverable
+     errors: srgb rendering intent value, size of content of ancillary chunks, more than 79 characters for some
+     strings, placement/combination rules for ancillary chunks, crc of unknown chunks, allowed characters
+     in string keys, etc... */
+
+  unsigned color_convert; /*whether to convert the PNG to the color type you want. Default: yes*/
+
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+  unsigned read_text_chunks; /*if false but remember_unknown_chunks is true, they're stored in the unknown chunks*/
+  /*store all bytes from unknown chunks in the LodePNGInfo (off by default, useful for a png editor)*/
+  unsigned remember_unknown_chunks;
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+} LodePNGDecoderSettings;
+
+void lodepng_decoder_settings_init(LodePNGDecoderSettings* settings);
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+#ifdef LODEPNG_COMPILE_ENCODER
+/*automatically use color type with less bits per pixel if losslessly possible. Default: AUTO*/
+typedef enum LodePNGFilterStrategy {
+  /*every filter at zero*/
+  LFS_ZERO = 0,
+  /*every filter at 1, 2, 3 or 4 (paeth), unlike LFS_ZERO not a good choice, but for testing*/
+  LFS_ONE = 1,
+  LFS_TWO = 2,
+  LFS_THREE = 3,
+  LFS_FOUR = 4,
+  /*Use filter that gives minimum sum, as described in the official PNG filter heuristic.*/
+  LFS_MINSUM,
+  /*Use the filter type that gives smallest Shannon entropy for this scanline. Depending
+  on the image, this is better or worse than minsum.*/
+  LFS_ENTROPY,
+  /*
+  Brute-force-search PNG filters by compressing each filter for each scanline.
+  Experimental, very slow, and only rarely gives better compression than MINSUM.
+  */
+  LFS_BRUTE_FORCE,
+  /*use predefined_filters buffer: you specify the filter type for each scanline*/
+  LFS_PREDEFINED
+} LodePNGFilterStrategy;
+
+/*Gives characteristics about the integer RGBA colors of the image (count, alpha channel usage, bit depth, ...),
+which helps decide which color model to use for encoding.
+Used internally by default if "auto_convert" is enabled. Public because it's useful for custom algorithms.*/
+typedef struct LodePNGColorStats {
+  unsigned colored; /*not grayscale*/
+  unsigned key; /*image is not opaque and color key is possible instead of full alpha*/
+  unsigned short key_r; /*key values, always as 16-bit, in 8-bit case the byte is duplicated, e.g. 65535 means 255*/
+  unsigned short key_g;
+  unsigned short key_b;
+  unsigned alpha; /*image is not opaque and alpha channel or alpha palette required*/
+  unsigned numcolors; /*amount of colors, up to 257. Not valid if bits == 16 or allow_palette is disabled.*/
+  unsigned char palette[1024]; /*Remembers up to the first 256 RGBA colors, in no particular order, only valid when numcolors is valid*/
+  unsigned bits; /*bits per channel (not for palette). 1,2 or 4 for grayscale only. 16 if 16-bit per channel required.*/
+  size_t numpixels;
+
+  /*user settings for computing/using the stats*/
+  unsigned allow_palette; /*default 1. if 0, disallow choosing palette colortype in auto_choose_color, and don't count numcolors*/
+  unsigned allow_greyscale; /*default 1. if 0, choose RGB or RGBA even if the image only has gray colors*/
+} LodePNGColorStats;
+
+void lodepng_color_stats_init(LodePNGColorStats* stats);
+
+/*Get a LodePNGColorStats of the image. The stats must already have been inited.
+Returns error code (e.g. alloc fail) or 0 if ok.*/
+unsigned lodepng_compute_color_stats(LodePNGColorStats* stats,
+                                     const unsigned char* image, unsigned w, unsigned h,
+                                     const LodePNGColorMode* mode_in);
+
+/*Settings for the encoder.*/
+typedef struct LodePNGEncoderSettings {
+  LodePNGCompressSettings zlibsettings; /*settings for the zlib encoder, such as window size, ...*/
+
+  unsigned auto_convert; /*automatically choose output PNG color type. Default: true*/
+
+  /*If true, follows the official PNG heuristic: if the PNG uses a palette or lower than
+  8 bit depth, set all filters to zero. Otherwise use the filter_strategy. Note that to
+  completely follow the official PNG heuristic, filter_palette_zero must be true and
+  filter_strategy must be LFS_MINSUM*/
+  unsigned filter_palette_zero;
+  /*Which filter strategy to use when not using zeroes due to filter_palette_zero.
+  Set filter_palette_zero to 0 to ensure always using your chosen strategy. Default: LFS_MINSUM*/
+  LodePNGFilterStrategy filter_strategy;
+  /*used if filter_strategy is LFS_PREDEFINED. In that case, this must point to a buffer with
+  the same length as the amount of scanlines in the image, and each value must <= 5. You
+  have to cleanup this buffer, LodePNG will never free it. Don't forget that filter_palette_zero
+  must be set to 0 to ensure this is also used on palette or low bitdepth images.*/
+  const unsigned char* predefined_filters;
+
+  /*force creating a PLTE chunk if colortype is 2 or 6 (= a suggested palette).
+  If colortype is 3, PLTE is _always_ created.*/
+  unsigned force_palette;
+#ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
+  /*add LodePNG identifier and version as a text chunk, for debugging*/
+  unsigned add_id;
+  /*encode text chunks as zTXt chunks instead of tEXt chunks, and use compression in iTXt chunks*/
+  unsigned text_compression;
+#endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
+} LodePNGEncoderSettings;
+
+void lodepng_encoder_settings_init(LodePNGEncoderSettings* settings);
+#endif /*LODEPNG_COMPILE_ENCODER*/
+
+
+#if defined(LODEPNG_COMPILE_DECODER) || defined(LODEPNG_COMPILE_ENCODER)
+/*The settings, state and information for extended encoding and decoding.*/
+typedef struct LodePNGState {
+#ifdef LODEPNG_COMPILE_DECODER
+  LodePNGDecoderSettings decoder; /*the decoding settings*/
+#endif /*LODEPNG_COMPILE_DECODER*/
+#ifdef LODEPNG_COMPILE_ENCODER
+  LodePNGEncoderSettings encoder; /*the encoding settings*/
+#endif /*LODEPNG_COMPILE_ENCODER*/
+  LodePNGColorMode info_raw; /*specifies the format in which you would like to get the raw pixel buffer*/
+  LodePNGInfo info_png; /*info of the PNG image obtained after decoding*/
+  unsigned error;
+} LodePNGState;
+
+/*init, cleanup and copy functions to use with this struct*/
+void lodepng_state_init(LodePNGState* state);
+void lodepng_state_cleanup(LodePNGState* state);
+void lodepng_state_copy(LodePNGState* dest, const LodePNGState* source);
+#endif /* defined(LODEPNG_COMPILE_DECODER) || defined(LODEPNG_COMPILE_ENCODER) */
+
+#ifdef LODEPNG_COMPILE_DECODER
+/*
+Same as lodepng_decode_memory, but uses a LodePNGState to allow custom settings and
+getting much more information about the PNG image and color mode.
+*/
+unsigned lodepng_decode(unsigned char** out, unsigned* w, unsigned* h,
+                        LodePNGState* state,
+                        const unsigned char* in, size_t insize);
+
+/*
+Read the PNG header, but not the actual data. This returns only the information
+that is in the IHDR chunk of the PNG, such as width, height and color type. The
+information is placed in the info_png field of the LodePNGState.
+*/
+unsigned lodepng_inspect(unsigned* w, unsigned* h,
+                         LodePNGState* state,
+                         const unsigned char* in, size_t insize);
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+/*
+Reads one metadata chunk (other than IHDR) of the PNG file and outputs what it
+read in the state. Returns error code on failure.
+Use lodepng_inspect first with a new state, then e.g. lodepng_chunk_find_const
+to find the desired chunk type, and if non null use lodepng_inspect_chunk (with
+chunk_pointer - start_of_file as pos).
+Supports most metadata chunks from the PNG standard (gAMA, bKGD, tEXt, ...).
+Ignores unsupported, unknown, non-metadata or IHDR chunks (without error).
+Requirements: &in[pos] must point to start of a chunk, must use regular
+lodepng_inspect first since format of most other chunks depends on IHDR, and if
+there is a PLTE chunk, that one must be inspected before tRNS or bKGD.
+*/
+unsigned lodepng_inspect_chunk(LodePNGState* state, size_t pos,
+                               const unsigned char* in, size_t insize);
+
+#ifdef LODEPNG_COMPILE_ENCODER
+/*This function allocates the out buffer with standard malloc and stores the size in *outsize.*/
+unsigned lodepng_encode(unsigned char** out, size_t* outsize,
+                        const unsigned char* image, unsigned w, unsigned h,
+                        LodePNGState* state);
+#endif /*LODEPNG_COMPILE_ENCODER*/
+
+/*
+The lodepng_chunk functions are normally not needed, except to traverse the
+unknown chunks stored in the LodePNGInfo struct, or add new ones to it.
+It also allows traversing the chunks of an encoded PNG file yourself.
+
+The chunk pointer always points to the beginning of the chunk itself, that is
+the first byte of the 4 length bytes.
+
+In the PNG file format, chunks have the following format:
+-4 bytes length: length of the data of the chunk in bytes (chunk itself is 12 bytes longer)
+-4 bytes chunk type (ASCII a-z,A-Z only, see below)
+-length bytes of data (may be 0 bytes if length was 0)
+-4 bytes of CRC, computed on chunk name + data
+
+The first chunk starts at the 8th byte of the PNG file, the entire rest of the file
+exists out of concatenated chunks with the above format.
+
+PNG standard chunk ASCII naming conventions:
+-First byte: uppercase = critical, lowercase = ancillary
+-Second byte: uppercase = public, lowercase = private
+-Third byte: must be uppercase
+-Fourth byte: uppercase = unsafe to copy, lowercase = safe to copy
+*/
+
+/*
+Gets the length of the data of the chunk. Total chunk length has 12 bytes more.
+There must be at least 4 bytes to read from. If the result value is too large,
+it may be corrupt data.
+*/
+unsigned lodepng_chunk_length(const unsigned char* chunk);
+
+/*puts the 4-byte type in null terminated string*/
+void lodepng_chunk_type(char type[5], const unsigned char* chunk);
+
+/*check if the type is the given type*/
+unsigned char lodepng_chunk_type_equals(const unsigned char* chunk, const char* type);
+
+/*0: it's one of the critical chunk types, 1: it's an ancillary chunk (see PNG standard)*/
+unsigned char lodepng_chunk_ancillary(const unsigned char* chunk);
+
+/*0: public, 1: private (see PNG standard)*/
+unsigned char lodepng_chunk_private(const unsigned char* chunk);
+
+/*0: the chunk is unsafe to copy, 1: the chunk is safe to copy (see PNG standard)*/
+unsigned char lodepng_chunk_safetocopy(const unsigned char* chunk);
+
+/*get pointer to the data of the chunk, where the input points to the header of the chunk*/
+unsigned char* lodepng_chunk_data(unsigned char* chunk);
+const unsigned char* lodepng_chunk_data_const(const unsigned char* chunk);
+
+/*returns 0 if the crc is correct, 1 if it's incorrect (0 for OK as usual!)*/
+unsigned lodepng_chunk_check_crc(const unsigned char* chunk);
+
+/*generates the correct CRC from the data and puts it in the last 4 bytes of the chunk*/
+void lodepng_chunk_generate_crc(unsigned char* chunk);
+
+/*
+Iterate to next chunks, allows iterating through all chunks of the PNG file.
+Input must be at the beginning of a chunk (result of a previous lodepng_chunk_next call,
+or the 8th byte of a PNG file which always has the first chunk), or alternatively may
+point to the first byte of the PNG file (which is not a chunk but the magic header, the
+function will then skip over it and return the first real chunk).
+Will output pointer to the start of the next chunk, or at or beyond end of the file if there
+is no more chunk after this or possibly if the chunk is corrupt.
+Start this process at the 8th byte of the PNG file.
+In a non-corrupt PNG file, the last chunk should have name "IEND".
+*/
+unsigned char* lodepng_chunk_next(unsigned char* chunk, unsigned char* end);
+const unsigned char* lodepng_chunk_next_const(const unsigned char* chunk, const unsigned char* end);
+
+/*Finds the first chunk with the given type in the range [chunk, end), or returns NULL if not found.*/
+unsigned char* lodepng_chunk_find(unsigned char* chunk, unsigned char* end, const char type[5]);
+const unsigned char* lodepng_chunk_find_const(const unsigned char* chunk, const unsigned char* end, const char type[5]);
+
+/*
+Appends chunk to the data in out. The given chunk should already have its chunk header.
+The out variable and outsize are updated to reflect the new reallocated buffer.
+Returns error code (0 if it went ok)
+*/
+unsigned lodepng_chunk_append(unsigned char** out, size_t* outsize, const unsigned char* chunk);
+
+/*
+Appends new chunk to out. The chunk to append is given by giving its length, type
+and data separately. The type is a 4-letter string.
+The out variable and outsize are updated to reflect the new reallocated buffer.
+Returne error code (0 if it went ok)
+*/
+unsigned lodepng_chunk_create(unsigned char** out, size_t* outsize, unsigned length,
+                              const char* type, const unsigned char* data);
+
+
+/*Calculate CRC32 of buffer*/
+unsigned lodepng_crc32(const unsigned char* buf, size_t len);
+#endif /*LODEPNG_COMPILE_PNG*/
+
+
+#ifdef LODEPNG_COMPILE_ZLIB
+/*
+This zlib part can be used independently to zlib compress and decompress a
+buffer. It cannot be used to create gzip files however, and it only supports the
+part of zlib that is required for PNG, it does not support dictionaries.
+*/
+
+#ifdef LODEPNG_COMPILE_DECODER
+/*Inflate a buffer. Inflate is the decompression step of deflate. Out buffer must be freed after use.*/
+unsigned lodepng_inflate(unsigned char** out, size_t* outsize,
+                         const unsigned char* in, size_t insize,
+                         const LodePNGDecompressSettings* settings);
+
+/*
+Decompresses Zlib data. Reallocates the out buffer and appends the data. The
+data must be according to the zlib specification.
+Either, *out must be NULL and *outsize must be 0, or, *out must be a valid
+buffer and *outsize its size in bytes. out must be freed by user after usage.
+*/
+unsigned lodepng_zlib_decompress(unsigned char** out, size_t* outsize,
+                                 const unsigned char* in, size_t insize,
+                                 const LodePNGDecompressSettings* settings);
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+#ifdef LODEPNG_COMPILE_ENCODER
+/*
+Compresses data with Zlib. Reallocates the out buffer and appends the data.
+Zlib adds a small header and trailer around the deflate data.
+The data is output in the format of the zlib specification.
+Either, *out must be NULL and *outsize must be 0, or, *out must be a valid
+buffer and *outsize its size in bytes. out must be freed by user after usage.
+*/
+unsigned lodepng_zlib_compress(unsigned char** out, size_t* outsize,
+                               const unsigned char* in, size_t insize,
+                               const LodePNGCompressSettings* settings);
+
+/*
+Find length-limited Huffman code for given frequencies. This function is in the
+public interface only for tests, it's used internally by lodepng_deflate.
+*/
+unsigned lodepng_huffman_code_lengths(unsigned* lengths, const unsigned* frequencies,
+                                      size_t numcodes, unsigned maxbitlen);
+
+/*Compress a buffer with deflate. See RFC 1951. Out buffer must be freed after use.*/
+unsigned lodepng_deflate(unsigned char** out, size_t* outsize,
+                         const unsigned char* in, size_t insize,
+                         const LodePNGCompressSettings* settings);
+
+#endif /*LODEPNG_COMPILE_ENCODER*/
+#endif /*LODEPNG_COMPILE_ZLIB*/
+
+#ifdef LODEPNG_COMPILE_DISK
+/*
+Load a file from disk into buffer. The function allocates the out buffer, and
+after usage you should free it.
+out: output parameter, contains pointer to loaded buffer.
+outsize: output parameter, size of the allocated out buffer
+filename: the path to the file to load
+return value: error code (0 means ok)
+*/
+unsigned lodepng_load_file(unsigned char** out, size_t* outsize, const char* filename);
+
+/*
+Save a file from buffer to disk. Warning, if it exists, this function overwrites
+the file without warning!
+buffer: the buffer to write
+buffersize: size of the buffer to write
+filename: the path to the file to save to
+return value: error code (0 means ok)
+*/
+unsigned lodepng_save_file(const unsigned char* buffer, size_t buffersize, const char* filename);
+#endif /*LODEPNG_COMPILE_DISK*/
+
+#ifdef LODEPNG_COMPILE_CPP
+/* The LodePNG C++ wrapper uses std::vectors instead of manually allocated memory buffers. */
+namespace lodepng {
+#ifdef LODEPNG_COMPILE_PNG
+class State : public LodePNGState {
+  public:
+    State();
+    State(const State& other);
+    ~State();
+    State& operator=(const State& other);
+};
+
+#ifdef LODEPNG_COMPILE_DECODER
+/* Same as other lodepng::decode, but using a State for more settings and information. */
+unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
+                State& state,
+                const unsigned char* in, size_t insize);
+unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
+                State& state,
+                const std::vector<unsigned char>& in);
+#endif /*LODEPNG_COMPILE_DECODER*/
+
+#ifdef LODEPNG_COMPILE_ENCODER
+/* Same as other lodepng::encode, but using a State for more settings and information. */
+unsigned encode(std::vector<unsigned char>& out,
+                const unsigned char* in, unsigned w, unsigned h,
+                State& state);
+unsigned encode(std::vector<unsigned char>& out,
+                const std::vector<unsigned char>& in, unsigned w, unsigned h,
+                State& state);
+#endif /*LODEPNG_COMPILE_ENCODER*/
+
+#ifdef LODEPNG_COMPILE_DISK
+/*
+Load a file from disk into an std::vector.
+return value: error code (0 means ok)
+*/
+unsigned load_file(std::vector<unsigned char>& buffer, const std::string& filename);
+
+/*
+Save the binary data in an std::vector to a file on disk. The file is overwritten
+without warning.
+*/
+unsigned save_file(const std::vector<unsigned char>& buffer, const std::string& filename);
+#endif /* LODEPNG_COMPILE_DISK */
+#endif /* LODEPNG_COMPILE_PNG */
+
+#ifdef LODEPNG_COMPILE_ZLIB
+#ifdef LODEPNG_COMPILE_DECODER
+/* Zlib-decompress an unsigned char buffer */
+unsigned decompress(std::vector<unsigned char>& out, const unsigned char* in, size_t insize,
+                    const LodePNGDecompressSettings& settings = lodepng_default_decompress_settings);
+
+/* Zlib-decompress an std::vector */
+unsigned decompress(std::vector<unsigned char>& out, const std::vector<unsigned char>& in,
+                    const LodePNGDecompressSettings& settings = lodepng_default_decompress_settings);
+#endif /* LODEPNG_COMPILE_DECODER */
+
+#ifdef LODEPNG_COMPILE_ENCODER
+/* Zlib-compress an unsigned char buffer */
+unsigned compress(std::vector<unsigned char>& out, const unsigned char* in, size_t insize,
+                  const LodePNGCompressSettings& settings = lodepng_default_compress_settings);
+
+/* Zlib-compress an std::vector */
+unsigned compress(std::vector<unsigned char>& out, const std::vector<unsigned char>& in,
+                  const LodePNGCompressSettings& settings = lodepng_default_compress_settings);
+#endif /* LODEPNG_COMPILE_ENCODER */
+#endif /* LODEPNG_COMPILE_ZLIB */
+} /* namespace lodepng */
+#endif /*LODEPNG_COMPILE_CPP*/
+
+/*
+TODO:
+[.] test if there are no memory leaks or security exploits - done a lot but needs to be checked often
+[.] check compatibility with various compilers  - done but needs to be redone for every newer version
+[X] converting color to 16-bit per channel types
+[X] support color profile chunk types (but never let them touch RGB values by default)
+[ ] support all public PNG chunk types (almost done except sBIT, sPLT and hIST)
+[ ] make sure encoder generates no chunks with size > (2^31)-1
+[ ] partial decoding (stream processing)
+[X] let the "isFullyOpaque" function check color keys and transparent palettes too
+[X] better name for the variables "codes", "codesD", "codelengthcodes", "clcl" and "lldl"
+[ ] allow treating some errors like warnings, when image is recoverable (e.g. 69, 57, 58)
+[ ] make warnings like: oob palette, checksum fail, data after iend, wrong/unknown crit chunk, no null terminator in text, ...
+[ ] error messages with line numbers (and version)
+[ ] errors in state instead of as return code?
+[ ] new errors/warnings like suspiciously big decompressed ztxt or iccp chunk
+[ ] let the C++ wrapper catch exceptions coming from the standard library and return LodePNG error codes
+[ ] allow user to provide custom color conversion functions, e.g. for premultiplied alpha, padding bits or not, ...
+[ ] allow user to give data (void*) to custom allocator
+[X] provide alternatives for C library functions not present on some platforms (memcpy, ...)
+*/
+
+#endif /*LODEPNG_H inclusion guard*/
+
+/*
+LodePNG Documentation
+---------------------
+
+0. table of contents
+--------------------
+
+  1. about
+   1.1. supported features
+   1.2. features not supported
+  2. C and C++ version
+  3. security
+  4. decoding
+  5. encoding
+  6. color conversions
+    6.1. PNG color types
+    6.2. color conversions
+    6.3. padding bits
+    6.4. A note about 16-bits per channel and endianness
+  7. error values
+  8. chunks and PNG editing
+  9. compiler support
+  10. examples
+   10.1. decoder C++ example
+   10.2. decoder C example
+  11. state settings reference
+  12. changes
+  13. contact information
+
+
+1. about
+--------
+
+PNG is a file format to store raster images losslessly with good compression,
+supporting different color types and alpha channel.
+
+LodePNG is a PNG codec according to the Portable Network Graphics (PNG)
+Specification (Second Edition) - W3C Recommendation 10 November 2003.
+
+The specifications used are:
+
+*) Portable Network Graphics (PNG) Specification (Second Edition):
+     http://www.w3.org/TR/2003/REC-PNG-20031110
+*) RFC 1950 ZLIB Compressed Data Format version 3.3:
+     http://www.gzip.org/zlib/rfc-zlib.html
+*) RFC 1951 DEFLATE Compressed Data Format Specification ver 1.3:
+     http://www.gzip.org/zlib/rfc-deflate.html
+
+The most recent version of LodePNG can currently be found at
+http://lodev.org/lodepng/
+
+LodePNG works both in C (ISO C90) and C++, with a C++ wrapper that adds
+extra functionality.
+
+LodePNG exists out of two files:
+-lodepng.h: the header file for both C and C++
+-lodepng.c(pp): give it the name lodepng.c or lodepng.cpp (or .cc) depending on your usage
+
+If you want to start using LodePNG right away without reading this doc, get the
+examples from the LodePNG website to see how to use it in code, or check the
+smaller examples in chapter 13 here.
+
+LodePNG is simple but only supports the basic requirements. To achieve
+simplicity, the following design choices were made: There are no dependencies
+on any external library. There are functions to decode and encode a PNG with
+a single function call, and extended versions of these functions taking a
+LodePNGState struct allowing to specify or get more information. By default
+the colors of the raw image are always RGB or RGBA, no matter what color type
+the PNG file uses. To read and write files, there are simple functions to
+convert the files to/from buffers in memory.
+
+This all makes LodePNG suitable for loading textures in games, demos and small
+programs, ... It's less suitable for full fledged image editors, loading PNGs
+over network (it requires all the image data to be available before decoding can
+begin), life-critical systems, ...
+
+1.1. supported features
+-----------------------
+
+The following features are supported by the decoder:
+
+*) decoding of PNGs with any color type, bit depth and interlace mode, to a 24- or 32-bit color raw image,
+   or the same color type as the PNG
+*) encoding of PNGs, from any raw image to 24- or 32-bit color, or the same color type as the raw image
+*) Adam7 interlace and deinterlace for any color type
+*) loading the image from harddisk or decoding it from a buffer from other sources than harddisk
+*) support for alpha channels, including RGBA color model, translucent palettes and color keying
+*) zlib decompression (inflate)
+*) zlib compression (deflate)
+*) CRC32 and ADLER32 checksums
+*) colorimetric color profile conversions: currently experimentally available in lodepng_util.cpp only,
+   plus alternatively ability to pass on chroma/gamma/ICC profile information to other color management system.
+*) handling of unknown chunks, allowing making a PNG editor that stores custom and unknown chunks.
+*) the following chunks are supported by both encoder and decoder:
+    IHDR: header information
+    PLTE: color palette
+    IDAT: pixel data
+    IEND: the final chunk
+    tRNS: transparency for palettized images
+    tEXt: textual information
+    zTXt: compressed textual information
+    iTXt: international textual information
+    bKGD: suggested background color
+    pHYs: physical dimensions
+    tIME: modification time
+    cHRM: RGB chromaticities
+    gAMA: RGB gamma correction
+    iCCP: ICC color profile
+    sRGB: rendering intent
+
+1.2. features not supported
+---------------------------
+
+The following features are _not_ supported:
+
+*) some features needed to make a conformant PNG-Editor might be still missing.
+*) partial loading/stream processing. All data must be available and is processed in one call.
+*) The following public chunks are not (yet) supported but treated as unknown chunks by LodePNG:
+    sBIT
+    hIST
+    sPLT
+
+
+2. C and C++ version
+--------------------
+
+The C version uses buffers allocated with alloc that you need to free()
+yourself. You need to use init and cleanup functions for each struct whenever
+using a struct from the C version to avoid exploits and memory leaks.
+
+The C++ version has extra functions with std::vectors in the interface and the
+lodepng::State class which is a LodePNGState with constructor and destructor.
+
+These files work without modification for both C and C++ compilers because all
+the additional C++ code is in "#ifdef __cplusplus" blocks that make C-compilers
+ignore it, and the C code is made to compile both with strict ISO C90 and C++.
+
+To use the C++ version, you need to rename the source file to lodepng.cpp
+(instead of lodepng.c), and compile it with a C++ compiler.
+
+To use the C version, you need to rename the source file to lodepng.c (instead
+of lodepng.cpp), and compile it with a C compiler.
+
+
+3. Security
+-----------
+
+Even if carefully designed, it's always possible that LodePNG contains possible
+exploits. If you discover one, please let me know, and it will be fixed.
+
+When using LodePNG, care has to be taken with the C version of LodePNG, as well
+as the C-style structs when working with C++. The following conventions are used
+for all C-style structs:
+
+-if a struct has a corresponding init function, always call the init function when making a new one
+-if a struct has a corresponding cleanup function, call it before the struct disappears to avoid memory leaks
+-if a struct has a corresponding copy function, use the copy function instead of "=".
+ The destination must also be inited already.
+
+
+4. Decoding
+-----------
+
+Decoding converts a PNG compressed image to a raw pixel buffer.
+
+Most documentation on using the decoder is at its declarations in the header
+above. For C, simple decoding can be done with functions such as
+lodepng_decode32, and more advanced decoding can be done with the struct
+LodePNGState and lodepng_decode. For C++, all decoding can be done with the
+various lodepng::decode functions, and lodepng::State can be used for advanced
+features.
+
+When using the LodePNGState, it uses the following fields for decoding:
+*) LodePNGInfo info_png: it stores extra information about the PNG (the input) in here
+*) LodePNGColorMode info_raw: here you can say what color mode of the raw image (the output) you want to get
+*) LodePNGDecoderSettings decoder: you can specify a few extra settings for the decoder to use
+
+LodePNGInfo info_png
+--------------------
+
+After decoding, this contains extra information of the PNG image, except the actual
+pixels, width and height because these are already gotten directly from the decoder
+functions.
+
+It contains for example the original color type of the PNG image, text comments,
+suggested background color, etc... More details about the LodePNGInfo struct are
+at its declaration documentation.
+
+LodePNGColorMode info_raw
+-------------------------
+
+When decoding, here you can specify which color type you want
+the resulting raw image to be. If this is different from the colortype of the
+PNG, then the decoder will automatically convert the result. This conversion
+always works, except if you want it to convert a color PNG to grayscale or to
+a palette with missing colors.
+
+By default, 32-bit color is used for the result.
+
+LodePNGDecoderSettings decoder
+------------------------------
+
+The settings can be used to ignore the errors created by invalid CRC and Adler32
+chunks, and to disable the decoding of tEXt chunks.
+
+There's also a setting color_convert, true by default. If false, no conversion
+is done, the resulting data will be as it was in the PNG (after decompression)
+and you'll have to puzzle the colors of the pixels together yourself using the
+color type information in the LodePNGInfo.
+
+
+5. Encoding
+-----------
+
+Encoding converts a raw pixel buffer to a PNG compressed image.
+
+Most documentation on using the encoder is at its declarations in the header
+above. For C, simple encoding can be done with functions such as
+lodepng_encode32, and more advanced decoding can be done with the struct
+LodePNGState and lodepng_encode. For C++, all encoding can be done with the
+various lodepng::encode functions, and lodepng::State can be used for advanced
+features.
+
+Like the decoder, the encoder can also give errors. However it gives less errors
+since the encoder input is trusted, the decoder input (a PNG image that could
+be forged by anyone) is not trusted.
+
+When using the LodePNGState, it uses the following fields for encoding:
+*) LodePNGInfo info_png: here you specify how you want the PNG (the output) to be.
+*) LodePNGColorMode info_raw: here you say what color type of the raw image (the input) has
+*) LodePNGEncoderSettings encoder: you can specify a few settings for the encoder to use
+
+LodePNGInfo info_png
+--------------------
+
+When encoding, you use this the opposite way as when decoding: for encoding,
+you fill in the values you want the PNG to have before encoding. By default it's
+not needed to specify a color type for the PNG since it's automatically chosen,
+but it's possible to choose it yourself given the right settings.
+
+The encoder will not always exactly match the LodePNGInfo struct you give,
+it tries as close as possible. Some things are ignored by the encoder. The
+encoder uses, for example, the following settings from it when applicable:
+colortype and bitdepth, text chunks, time chunk, the color key, the palette, the
+background color, the interlace method, unknown chunks, ...
+
+When encoding to a PNG with colortype 3, the encoder will generate a PLTE chunk.
+If the palette contains any colors for which the alpha channel is not 255 (so
+there are translucent colors in the palette), it'll add a tRNS chunk.
+
+LodePNGColorMode info_raw
+-------------------------
+
+You specify the color type of the raw image that you give to the input here,
+including a possible transparent color key and palette you happen to be using in
+your raw image data.
+
+By default, 32-bit color is assumed, meaning your input has to be in RGBA
+format with 4 bytes (unsigned chars) per pixel.
+
+LodePNGEncoderSettings encoder
+------------------------------
+
+The following settings are supported (some are in sub-structs):
+*) auto_convert: when this option is enabled, the encoder will
+automatically choose the smallest possible color mode (including color key) that
+can encode the colors of all pixels without information loss.
+*) btype: the block type for LZ77. 0 = uncompressed, 1 = fixed huffman tree,
+   2 = dynamic huffman tree (best compression). Should be 2 for proper
+   compression.
+*) use_lz77: whether or not to use LZ77 for compressed block types. Should be
+   true for proper compression.
+*) windowsize: the window size used by the LZ77 encoder (1 - 32768). Has value
+   2048 by default, but can be set to 32768 for better, but slow, compression.
+*) force_palette: if colortype is 2 or 6, you can make the encoder write a PLTE
+   chunk if force_palette is true. This can used as suggested palette to convert
+   to by viewers that don't support more than 256 colors (if those still exist)
+*) add_id: add text chunk "Encoder: LodePNG <version>" to the image.
+*) text_compression: default 1. If 1, it'll store texts as zTXt instead of tEXt chunks.
+  zTXt chunks use zlib compression on the text. This gives a smaller result on
+  large texts but a larger result on small texts (such as a single program name).
+  It's all tEXt or all zTXt though, there's no separate setting per text yet.
+
+
+6. color conversions
+--------------------
+
+An important thing to note about LodePNG, is that the color type of the PNG, and
+the color type of the raw image, are completely independent. By default, when
+you decode a PNG, you get the result as a raw image in the color type you want,
+no matter whether the PNG was encoded with a palette, grayscale or RGBA color.
+And if you encode an image, by default LodePNG will automatically choose the PNG
+color type that gives good compression based on the values of colors and amount
+of colors in the image. It can be configured to let you control it instead as
+well, though.
+
+To be able to do this, LodePNG does conversions from one color mode to another.
+It can convert from almost any color type to any other color type, except the
+following conversions: RGB to grayscale is not supported, and converting to a
+palette when the palette doesn't have a required color is not supported. This is
+not supported on purpose: this is information loss which requires a color
+reduction algorithm that is beyond the scope of a PNG encoder (yes, RGB to gray
+is easy, but there are multiple ways if you want to give some channels more
+weight).
+
+By default, when decoding, you get the raw image in 32-bit RGBA or 24-bit RGB
+color, no matter what color type the PNG has. And by default when encoding,
+LodePNG automatically picks the best color model for the output PNG, and expects
+the input image to be 32-bit RGBA or 24-bit RGB. So, unless you want to control
+the color format of the images yourself, you can skip this chapter.
+
+6.1. PNG color types
+--------------------
+
+A PNG image can have many color types, ranging from 1-bit color to 64-bit color,
+as well as palettized color modes. After the zlib decompression and unfiltering
+in the PNG image is done, the raw pixel data will have that color type and thus
+a certain amount of bits per pixel. If you want the output raw image after
+decoding to have another color type, a conversion is done by LodePNG.
+
+The PNG specification gives the following color types:
+
+0: grayscale, bit depths 1, 2, 4, 8, 16
+2: RGB, bit depths 8 and 16
+3: palette, bit depths 1, 2, 4 and 8
+4: grayscale with alpha, bit depths 8 and 16
+6: RGBA, bit depths 8 and 16
+
+Bit depth is the amount of bits per pixel per color channel. So the total amount
+of bits per pixel is: amount of channels * bitdepth.
+
+6.2. color conversions
+----------------------
+
+As explained in the sections about the encoder and decoder, you can specify
+color types and bit depths in info_png and info_raw to change the default
+behaviour.
+
+If, when decoding, you want the raw image to be something else than the default,
+you need to set the color type and bit depth you want in the LodePNGColorMode,
+or the parameters colortype and bitdepth of the simple decoding function.
+
+If, when encoding, you use another color type than the default in the raw input
+image, you need to specify its color type and bit depth in the LodePNGColorMode
+of the raw image, or use the parameters colortype and bitdepth of the simple
+encoding function.
+
+If, when encoding, you don't want LodePNG to choose the output PNG color type
+but control it yourself, you need to set auto_convert in the encoder settings
+to false, and specify the color type you want in the LodePNGInfo of the
+encoder (including palette: it can generate a palette if auto_convert is true,
+otherwise not).
+
+If the input and output color type differ (whether user chosen or auto chosen),
+LodePNG will do a color conversion, which follows the rules below, and may
+sometimes result in an error.
+
+To avoid some confusion:
+-the decoder converts from PNG to raw image
+-the encoder converts from raw image to PNG
+-the colortype and bitdepth in LodePNGColorMode info_raw, are those of the raw image
+-the colortype and bitdepth in the color field of LodePNGInfo info_png, are those of the PNG
+-when encoding, the color type in LodePNGInfo is ignored if auto_convert
+ is enabled, it is automatically generated instead
+-when decoding, the color type in LodePNGInfo is set by the decoder to that of the original
+ PNG image, but it can be ignored since the raw image has the color type you requested instead
+-if the color type of the LodePNGColorMode and PNG image aren't the same, a conversion
+ between the color types is done if the color types are supported. If it is not
+ supported, an error is returned. If the types are the same, no conversion is done.
+-even though some conversions aren't supported, LodePNG supports loading PNGs from any
+ colortype and saving PNGs to any colortype, sometimes it just requires preparing
+ the raw image correctly before encoding.
+-both encoder and decoder use the same color converter.
+
+The function lodepng_convert does the color conversion. It is available in the
+interface but normally isn't needed since the encoder and decoder already call
+it.
+
+Non supported color conversions:
+-color to grayscale when non-gray pixels are present: no error is thrown, but
+the result will look ugly because only the red channel is taken (it assumes all
+three channels are the same in this case so ignores green and blue). The reason
+no error is given is to allow converting from three-channel grayscale images to
+one-channel even if there are numerical imprecisions.
+-anything to palette when the palette does not have an exact match for a from-color
+in it: in this case an error is thrown
+
+Supported color conversions:
+-anything to 8-bit RGB, 8-bit RGBA, 16-bit RGB, 16-bit RGBA
+-any gray or gray+alpha, to gray or gray+alpha
+-anything to a palette, as long as the palette has the requested colors in it
+-removing alpha channel
+-higher to smaller bitdepth, and vice versa
+
+If you want no color conversion to be done (e.g. for speed or control):
+-In the encoder, you can make it save a PNG with any color type by giving the
+raw color mode and LodePNGInfo the same color mode, and setting auto_convert to
+false.
+-In the decoder, you can make it store the pixel data in the same color type
+as the PNG has, by setting the color_convert setting to false. Settings in
+info_raw are then ignored.
+
+6.3. padding bits
+-----------------
+
+In the PNG file format, if a less than 8-bit per pixel color type is used and the scanlines
+have a bit amount that isn't a multiple of 8, then padding bits are used so that each
+scanline starts at a fresh byte. But that is NOT true for the LodePNG raw input and output.
+The raw input image you give to the encoder, and the raw output image you get from the decoder
+will NOT have these padding bits, e.g. in the case of a 1-bit image with a width
+of 7 pixels, the first pixel of the second scanline will the 8th bit of the first byte,
+not the first bit of a new byte.
+
+6.4. A note about 16-bits per channel and endianness
+----------------------------------------------------
+
+LodePNG uses unsigned char arrays for 16-bit per channel colors too, just like
+for any other color format. The 16-bit values are stored in big endian (most
+significant byte first) in these arrays. This is the opposite order of the
+little endian used by x86 CPU's.
+
+LodePNG always uses big endian because the PNG file format does so internally.
+Conversions to other formats than PNG uses internally are not supported by
+LodePNG on purpose, there are myriads of formats, including endianness of 16-bit
+colors, the order in which you store R, G, B and A, and so on. Supporting and
+converting to/from all that is outside the scope of LodePNG.
+
+This may mean that, depending on your use case, you may want to convert the big
+endian output of LodePNG to little endian with a for loop. This is certainly not
+always needed, many applications and libraries support big endian 16-bit colors
+anyway, but it means you cannot simply cast the unsigned char* buffer to an
+unsigned short* buffer on x86 CPUs.
+
+
+7. error values
+---------------
+
+All functions in LodePNG that return an error code, return 0 if everything went
+OK, or a non-zero code if there was an error.
+
+The meaning of the LodePNG error values can be retrieved with the function
+lodepng_error_text: given the numerical error code, it returns a description
+of the error in English as a string.
+
+Check the implementation of lodepng_error_text to see the meaning of each code.
+
+
+8. chunks and PNG editing
+-------------------------
+
+If you want to add extra chunks to a PNG you encode, or use LodePNG for a PNG
+editor that should follow the rules about handling of unknown chunks, or if your
+program is able to read other types of chunks than the ones handled by LodePNG,
+then that's possible with the chunk functions of LodePNG.
+
+A PNG chunk has the following layout:
+
+4 bytes length
+4 bytes type name
+length bytes data
+4 bytes CRC
+
+8.1. iterating through chunks
+-----------------------------
+
+If you have a buffer containing the PNG image data, then the first chunk (the
+IHDR chunk) starts at byte number 8 of that buffer. The first 8 bytes are the
+signature of the PNG and are not part of a chunk. But if you start at byte 8
+then you have a chunk, and can check the following things of it.
+
+NOTE: none of these functions check for memory buffer boundaries. To avoid
+exploits, always make sure the buffer contains all the data of the chunks.
+When using lodepng_chunk_next, make sure the returned value is within the
+allocated memory.
+
+unsigned lodepng_chunk_length(const unsigned char* chunk):
+
+Get the length of the chunk's data. The total chunk length is this length + 12.
+
+void lodepng_chunk_type(char type[5], const unsigned char* chunk):
+unsigned char lodepng_chunk_type_equals(const unsigned char* chunk, const char* type):
+
+Get the type of the chunk or compare if it's a certain type
+
+unsigned char lodepng_chunk_critical(const unsigned char* chunk):
+unsigned char lodepng_chunk_private(const unsigned char* chunk):
+unsigned char lodepng_chunk_safetocopy(const unsigned char* chunk):
+
+Check if the chunk is critical in the PNG standard (only IHDR, PLTE, IDAT and IEND are).
+Check if the chunk is private (public chunks are part of the standard, private ones not).
+Check if the chunk is safe to copy. If it's not, then, when modifying data in a critical
+chunk, unsafe to copy chunks of the old image may NOT be saved in the new one if your
+program doesn't handle that type of unknown chunk.
+
+unsigned char* lodepng_chunk_data(unsigned char* chunk):
+const unsigned char* lodepng_chunk_data_const(const unsigned char* chunk):
+
+Get a pointer to the start of the data of the chunk.
+
+unsigned lodepng_chunk_check_crc(const unsigned char* chunk):
+void lodepng_chunk_generate_crc(unsigned char* chunk):
+
+Check if the crc is correct or generate a correct one.
+
+unsigned char* lodepng_chunk_next(unsigned char* chunk):
+const unsigned char* lodepng_chunk_next_const(const unsigned char* chunk):
+
+Iterate to the next chunk. This works if you have a buffer with consecutive chunks. Note that these
+functions do no boundary checking of the allocated data whatsoever, so make sure there is enough
+data available in the buffer to be able to go to the next chunk.
+
+unsigned lodepng_chunk_append(unsigned char** out, size_t* outsize, const unsigned char* chunk):
+unsigned lodepng_chunk_create(unsigned char** out, size_t* outsize, unsigned length,
+                              const char* type, const unsigned char* data):
+
+These functions are used to create new chunks that are appended to the data in *out that has
+length *outsize. The append function appends an existing chunk to the new data. The create
+function creates a new chunk with the given parameters and appends it. Type is the 4-letter
+name of the chunk.
+
+8.2. chunks in info_png
+-----------------------
+
+The LodePNGInfo struct contains fields with the unknown chunk in it. It has 3
+buffers (each with size) to contain 3 types of unknown chunks:
+the ones that come before the PLTE chunk, the ones that come between the PLTE
+and the IDAT chunks, and the ones that come after the IDAT chunks.
+It's necessary to make the distinction between these 3 cases because the PNG
+standard forces to keep the ordering of unknown chunks compared to the critical
+chunks, but does not force any other ordering rules.
+
+info_png.unknown_chunks_data[0] is the chunks before PLTE
+info_png.unknown_chunks_data[1] is the chunks after PLTE, before IDAT
+info_png.unknown_chunks_data[2] is the chunks after IDAT
+
+The chunks in these 3 buffers can be iterated through and read by using the same
+way described in the previous subchapter.
+
+When using the decoder to decode a PNG, you can make it store all unknown chunks
+if you set the option settings.remember_unknown_chunks to 1. By default, this
+option is off (0).
+
+The encoder will always encode unknown chunks that are stored in the info_png.
+If you need it to add a particular chunk that isn't known by LodePNG, you can
+use lodepng_chunk_append or lodepng_chunk_create to the chunk data in
+info_png.unknown_chunks_data[x].
+
+Chunks that are known by LodePNG should not be added in that way. E.g. to make
+LodePNG add a bKGD chunk, set background_defined to true and add the correct
+parameters there instead.
+
+
+9. compiler support
+-------------------
+
+No libraries other than the current standard C library are needed to compile
+LodePNG. For the C++ version, only the standard C++ library is needed on top.
+Add the files lodepng.c(pp) and lodepng.h to your project, include
+lodepng.h where needed, and your program can read/write PNG files.
+
+It is compatible with C90 and up, and C++03 and up.
+
+If performance is important, use optimization when compiling! For both the
+encoder and decoder, this makes a large difference.
+
+Make sure that LodePNG is compiled with the same compiler of the same version
+and with the same settings as the rest of the program, or the interfaces with
+std::vectors and std::strings in C++ can be incompatible.
+
+CHAR_BITS must be 8 or higher, because LodePNG uses unsigned chars for octets.
+
+*) gcc and g++
+
+LodePNG is developed in gcc so this compiler is natively supported. It gives no
+warnings with compiler options "-Wall -Wextra -pedantic -ansi", with gcc and g++
+version 4.7.1 on Linux, 32-bit and 64-bit.
+
+*) Clang
+
+Fully supported and warning-free.
+
+*) Mingw
+
+The Mingw compiler (a port of gcc for Windows) should be fully supported by
+LodePNG.
+
+*) Visual Studio and Visual C++ Express Edition
+
+LodePNG should be warning-free with warning level W4. Two warnings were disabled
+with pragmas though: warning 4244 about implicit conversions, and warning 4996
+where it wants to use a non-standard function fopen_s instead of the standard C
+fopen.
+
+Visual Studio may want "stdafx.h" files to be included in each source file and
+give an error "unexpected end of file while looking for precompiled header".
+This is not standard C++ and will not be added to the stock LodePNG. You can
+disable it for lodepng.cpp only by right clicking it, Properties, C/C++,
+Precompiled Headers, and set it to Not Using Precompiled Headers there.
+
+NOTE: Modern versions of VS should be fully supported, but old versions, e.g.
+VS6, are not guaranteed to work.
+
+*) Compilers on Macintosh
+
+LodePNG has been reported to work both with gcc and LLVM for Macintosh, both for
+C and C++.
+
+*) Other Compilers
+
+If you encounter problems on any compilers, feel free to let me know and I may
+try to fix it if the compiler is modern and standards compliant.
+
+
+10. examples
+------------
+
+This decoder example shows the most basic usage of LodePNG. More complex
+examples can be found on the LodePNG website.
+
+10.1. decoder C++ example
+-------------------------
+
+#include "lodepng.h"
+#include <iostream>
+
+int main(int argc, char *argv[]) {
+  const char* filename = argc > 1 ? argv[1] : "test.png";
+
+  //load and decode
+  std::vector<unsigned char> image;
+  unsigned width, height;
+  unsigned error = lodepng::decode(image, width, height, filename);
+
+  //if there's an error, display it
+  if(error) std::cout << "decoder error " << error << ": " << lodepng_error_text(error) << std::endl;
+
+  //the pixels are now in the vector "image", 4 bytes per pixel, ordered RGBARGBA..., use it as texture, draw it, ...
+}
+
+10.2. decoder C example
+-----------------------
+
+#include "lodepng.h"
+
+int main(int argc, char *argv[]) {
+  unsigned error;
+  unsigned char* image;
+  size_t width, height;
+  const char* filename = argc > 1 ? argv[1] : "test.png";
+
+  error = lodepng_decode32_file(&image, &width, &height, filename);
+
+  if(error) printf("decoder error %u: %s\n", error, lodepng_error_text(error));
+
+  / * use image here * /
+
+  free(image);
+  return 0;
+}
+
+11. state settings reference
+----------------------------
+
+A quick reference of some settings to set on the LodePNGState
+
+For decoding:
+
+state.decoder.zlibsettings.ignore_adler32: ignore ADLER32 checksums
+state.decoder.zlibsettings.custom_...: use custom inflate function
+state.decoder.ignore_crc: ignore CRC checksums
+state.decoder.ignore_critical: ignore unknown critical chunks
+state.decoder.ignore_end: ignore missing IEND chunk. May fail if this corruption causes other errors
+state.decoder.color_convert: convert internal PNG color to chosen one
+state.decoder.read_text_chunks: whether to read in text metadata chunks
+state.decoder.remember_unknown_chunks: whether to read in unknown chunks
+state.info_raw.colortype: desired color type for decoded image
+state.info_raw.bitdepth: desired bit depth for decoded image
+state.info_raw....: more color settings, see struct LodePNGColorMode
+state.info_png....: no settings for decoder but ouput, see struct LodePNGInfo
+
+For encoding:
+
+state.encoder.zlibsettings.btype: disable compression by setting it to 0
+state.encoder.zlibsettings.use_lz77: use LZ77 in compression
+state.encoder.zlibsettings.windowsize: tweak LZ77 windowsize
+state.encoder.zlibsettings.minmatch: tweak min LZ77 length to match
+state.encoder.zlibsettings.nicematch: tweak LZ77 match where to stop searching
+state.encoder.zlibsettings.lazymatching: try one more LZ77 matching
+state.encoder.zlibsettings.custom_...: use custom deflate function
+state.encoder.auto_convert: choose optimal PNG color type, if 0 uses info_png
+state.encoder.filter_palette_zero: PNG filter strategy for palette
+state.encoder.filter_strategy: PNG filter strategy to encode with
+state.encoder.force_palette: add palette even if not encoding to one
+state.encoder.add_id: add LodePNG identifier and version as a text chunk
+state.encoder.text_compression: use compressed text chunks for metadata
+state.info_raw.colortype: color type of raw input image you provide
+state.info_raw.bitdepth: bit depth of raw input image you provide
+state.info_raw: more color settings, see struct LodePNGColorMode
+state.info_png.color.colortype: desired color type if auto_convert is false
+state.info_png.color.bitdepth: desired bit depth if auto_convert is false
+state.info_png.color....: more color settings, see struct LodePNGColorMode
+state.info_png....: more PNG related settings, see struct LodePNGInfo
+
+
+12. changes
+-----------
+
+The version number of LodePNG is the date of the change given in the format
+yyyymmdd.
+
+Some changes aren't backwards compatible. Those are indicated with a (!)
+symbol.
+
+Not all changes are listed here, the commit history in github lists more:
+https://github.com/lvandeve/lodepng
+
+*) 06 mar 2020: simplified some of the dynamic memory allocations.
+*) 12 jan 2020: (!) added 'end' argument to lodepng_chunk_next to allow correct
+   overflow checks.
+*) 14 aug 2019: around 25% faster decoding thanks to huffman lookup tables.
+*) 15 jun 2019: (!) auto_choose_color API changed (for bugfix: don't use palette
+   if gray ICC profile) and non-ICC LodePNGColorProfile renamed to
+   LodePNGColorStats.
+*) 30 dec 2018: code style changes only: removed newlines before opening braces.
+*) 10 sep 2018: added way to inspect metadata chunks without full decoding.
+*) 19 aug 2018: (!) fixed color mode bKGD is encoded with and made it use
+   palette index in case of palette.
+*) 10 aug 2018: (!) added support for gAMA, cHRM, sRGB and iCCP chunks. This
+   change is backwards compatible unless you relied on unknown_chunks for those.
+*) 11 jun 2018: less restrictive check for pixel size integer overflow
+*) 14 jan 2018: allow optionally ignoring a few more recoverable errors
+*) 17 sep 2017: fix memory leak for some encoder input error cases
+*) 27 nov 2016: grey+alpha auto color model detection bugfix
+*) 18 apr 2016: Changed qsort to custom stable sort (for platforms w/o qsort).
+*) 09 apr 2016: Fixed colorkey usage detection, and better file loading (within
+   the limits of pure C90).
+*) 08 dec 2015: Made load_file function return error if file can't be opened.
+*) 24 okt 2015: Bugfix with decoding to palette output.
+*) 18 apr 2015: Boundary PM instead of just package-merge for faster encoding.
+*) 24 aug 2014: Moved to github
+*) 23 aug 2014: Reduced needless memory usage of decoder.
+*) 28 jun 2014: Removed fix_png setting, always support palette OOB for
+    simplicity. Made ColorProfile public.
+*) 09 jun 2014: Faster encoder by fixing hash bug and more zeros optimization.
+*) 22 dec 2013: Power of two windowsize required for optimization.
+*) 15 apr 2013: Fixed bug with LAC_ALPHA and color key.
+*) 25 mar 2013: Added an optional feature to ignore some PNG errors (fix_png).
+*) 11 mar 2013: (!) Bugfix with custom free. Changed from "my" to "lodepng_"
+    prefix for the custom allocators and made it possible with a new #define to
+    use custom ones in your project without needing to change lodepng's code.
+*) 28 jan 2013: Bugfix with color key.
+*) 27 okt 2012: Tweaks in text chunk keyword length error handling.
+*) 8 okt 2012: (!) Added new filter strategy (entropy) and new auto color mode.
+    (no palette). Better deflate tree encoding. New compression tweak settings.
+    Faster color conversions while decoding. Some internal cleanups.
+*) 23 sep 2012: Reduced warnings in Visual Studio a little bit.
+*) 1 sep 2012: (!) Removed #define's for giving custom (de)compression functions
+    and made it work with function pointers instead.
+*) 23 jun 2012: Added more filter strategies. Made it easier to use custom alloc
+    and free functions and toggle #defines from compiler flags. Small fixes.
+*) 6 may 2012: (!) Made plugging in custom zlib/deflate functions more flexible.
+*) 22 apr 2012: (!) Made interface more consistent, renaming a lot. Removed
+    redundant C++ codec classes. Reduced amount of structs. Everything changed,
+    but it is cleaner now imho and functionality remains the same. Also fixed
+    several bugs and shrunk the implementation code. Made new samples.
+*) 6 nov 2011: (!) By default, the encoder now automatically chooses the best
+    PNG color model and bit depth, based on the amount and type of colors of the
+    raw image. For this, autoLeaveOutAlphaChannel replaced by auto_choose_color.
+*) 9 okt 2011: simpler hash chain implementation for the encoder.
+*) 8 sep 2011: lz77 encoder lazy matching instead of greedy matching.
+*) 23 aug 2011: tweaked the zlib compression parameters after benchmarking.
+    A bug with the PNG filtertype heuristic was fixed, so that it chooses much
+    better ones (it's quite significant). A setting to do an experimental, slow,
+    brute force search for PNG filter types is added.
+*) 17 aug 2011: (!) changed some C zlib related function names.
+*) 16 aug 2011: made the code less wide (max 120 characters per line).
+*) 17 apr 2011: code cleanup. Bugfixes. Convert low to 16-bit per sample colors.
+*) 21 feb 2011: fixed compiling for C90. Fixed compiling with sections disabled.
+*) 11 dec 2010: encoding is made faster, based on suggestion by Peter Eastman
+    to optimize long sequences of zeros.
+*) 13 nov 2010: added LodePNG_InfoColor_hasPaletteAlpha and
+    LodePNG_InfoColor_canHaveAlpha functions for convenience.
+*) 7 nov 2010: added LodePNG_error_text function to get error code description.
+*) 30 okt 2010: made decoding slightly faster
+*) 26 okt 2010: (!) changed some C function and struct names (more consistent).
+     Reorganized the documentation and the declaration order in the header.
+*) 08 aug 2010: only changed some comments and external samples.
+*) 05 jul 2010: fixed bug thanks to warnings in the new gcc version.
+*) 14 mar 2010: fixed bug where too much memory was allocated for char buffers.
+*) 02 sep 2008: fixed bug where it could create empty tree that linux apps could
+    read by ignoring the problem but windows apps couldn't.
+*) 06 jun 2008: added more error checks for out of memory cases.
+*) 26 apr 2008: added a few more checks here and there to ensure more safety.
+*) 06 mar 2008: crash with encoding of strings fixed
+*) 02 feb 2008: support for international text chunks added (iTXt)
+*) 23 jan 2008: small cleanups, and #defines to divide code in sections
+*) 20 jan 2008: support for unknown chunks allowing using LodePNG for an editor.
+*) 18 jan 2008: support for tIME and pHYs chunks added to encoder and decoder.
+*) 17 jan 2008: ability to encode and decode compressed zTXt chunks added
+    Also various fixes, such as in the deflate and the padding bits code.
+*) 13 jan 2008: Added ability to encode Adam7-interlaced images. Improved
+    filtering code of encoder.
+*) 07 jan 2008: (!) changed LodePNG to use ISO C90 instead of C++. A
+    C++ wrapper around this provides an interface almost identical to before.
+    Having LodePNG be pure ISO C90 makes it more portable. The C and C++ code
+    are together in these files but it works both for C and C++ compilers.
+*) 29 dec 2007: (!) changed most integer types to unsigned int + other tweaks
+*) 30 aug 2007: bug fixed which makes this Borland C++ compatible
+*) 09 aug 2007: some VS2005 warnings removed again
+*) 21 jul 2007: deflate code placed in new namespace separate from zlib code
+*) 08 jun 2007: fixed bug with 2- and 4-bit color, and small interlaced images
+*) 04 jun 2007: improved support for Visual Studio 2005: crash with accessing
+    invalid std::vector element [0] fixed, and level 3 and 4 warnings removed
+*) 02 jun 2007: made the encoder add a tag with version by default
+*) 27 may 2007: zlib and png code separated (but still in the same file),
+    simple encoder/decoder functions added for more simple usage cases
+*) 19 may 2007: minor fixes, some code cleaning, new error added (error 69),
+    moved some examples from here to lodepng_examples.cpp
+*) 12 may 2007: palette decoding bug fixed
+*) 24 apr 2007: changed the license from BSD to the zlib license
+*) 11 mar 2007: very simple addition: ability to encode bKGD chunks.
+*) 04 mar 2007: (!) tEXt chunk related fixes, and support for encoding
+    palettized PNG images. Plus little interface change with palette and texts.
+*) 03 mar 2007: Made it encode dynamic Huffman shorter with repeat codes.
+    Fixed a bug where the end code of a block had length 0 in the Huffman tree.
+*) 26 feb 2007: Huffman compression with dynamic trees (BTYPE 2) now implemented
+    and supported by the encoder, resulting in smaller PNGs at the output.
+*) 27 jan 2007: Made the Adler-32 test faster so that a timewaste is gone.
+*) 24 jan 2007: gave encoder an error interface. Added color conversion from any
+    greyscale type to 8-bit greyscale with or without alpha.
+*) 21 jan 2007: (!) Totally changed the interface. It allows more color types
+    to convert to and is more uniform. See the manual for how it works now.
+*) 07 jan 2007: Some cleanup & fixes, and a few changes over the last days:
+    encode/decode custom tEXt chunks, separate classes for zlib & deflate, and
+    at last made the decoder give errors for incorrect Adler32 or Crc.
+*) 01 jan 2007: Fixed bug with encoding PNGs with less than 8 bits per channel.
+*) 29 dec 2006: Added support for encoding images without alpha channel, and
+    cleaned out code as well as making certain parts faster.
+*) 28 dec 2006: Added "Settings" to the encoder.
+*) 26 dec 2006: The encoder now does LZ77 encoding and produces much smaller files now.
+    Removed some code duplication in the decoder. Fixed little bug in an example.
+*) 09 dec 2006: (!) Placed output parameters of public functions as first parameter.
+    Fixed a bug of the decoder with 16-bit per color.
+*) 15 okt 2006: Changed documentation structure
+*) 09 okt 2006: Encoder class added. It encodes a valid PNG image from the
+    given image buffer, however for now it's not compressed.
+*) 08 sep 2006: (!) Changed to interface with a Decoder class
+*) 30 jul 2006: (!) LodePNG_InfoPng , width and height are now retrieved in different
+    way. Renamed decodePNG to decodePNGGeneric.
+*) 29 jul 2006: (!) Changed the interface: image info is now returned as a
+    struct of type LodePNG::LodePNG_Info, instead of a vector, which was a bit clumsy.
+*) 28 jul 2006: Cleaned the code and added new error checks.
+    Corrected terminology "deflate" into "inflate".
+*) 23 jun 2006: Added SDL example in the documentation in the header, this
+    example allows easy debugging by displaying the PNG and its transparency.
+*) 22 jun 2006: (!) Changed way to obtain error value. Added
+    loadFile function for convenience. Made decodePNG32 faster.
+*) 21 jun 2006: (!) Changed type of info vector to unsigned.
+    Changed position of palette in info vector. Fixed an important bug that
+    happened on PNGs with an uncompressed block.
+*) 16 jun 2006: Internally changed unsigned into unsigned where
+    needed, and performed some optimizations.
+*) 07 jun 2006: (!) Renamed functions to decodePNG and placed them
+    in LodePNG namespace. Changed the order of the parameters. Rewrote the
+    documentation in the header. Renamed files to lodepng.cpp and lodepng.h
+*) 22 apr 2006: Optimized and improved some code
+*) 07 sep 2005: (!) Changed to std::vector interface
+*) 12 aug 2005: Initial release (C++, decoder only)
+
+
+13. contact information
+-----------------------
+
+Feel free to contact me with suggestions, problems, comments, ... concerning
+LodePNG. If you encounter a PNG image that doesn't work properly with this
+decoder, feel free to send it and I'll use it to find and fix the problem.
+
+My email address is (puzzle the account and domain together with an @ symbol):
+Domain: gmail dot com.
+Account: lode dot vandevenne.
+
+
+Copyright (c) 2005-2020 Lode Vandevenne
+*/

+ 3225 - 0
Tests/Dependencies/nanobench/nanobench.h

@@ -0,0 +1,3225 @@
+//  __   _ _______ __   _  _____  ______  _______ __   _ _______ _     _
+//  | \  | |_____| | \  | |     | |_____] |______ | \  | |       |_____|
+//  |  \_| |     | |  \_| |_____| |_____] |______ |  \_| |_____  |     |
+//
+// Microbenchmark framework for C++11/14/17/20
+// https://github.com/martinus/nanobench
+//
+// Licensed under the MIT License <http://opensource.org/licenses/MIT>.
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2019-2020 Martin Ankerl <[email protected]>
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all
+// copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+
+#ifndef ANKERL_NANOBENCH_H_INCLUDED
+#define ANKERL_NANOBENCH_H_INCLUDED
+
+// see https://semver.org/
+#define ANKERL_NANOBENCH_VERSION_MAJOR 4 // incompatible API changes
+#define ANKERL_NANOBENCH_VERSION_MINOR 0 // backwards-compatible changes
+#define ANKERL_NANOBENCH_VERSION_PATCH 0 // backwards-compatible bug fixes
+
+///////////////////////////////////////////////////////////////////////////////////////////////////
+// public facing api - as minimal as possible
+///////////////////////////////////////////////////////////////////////////////////////////////////
+
+#include <chrono>  // high_resolution_clock
+#include <cstring> // memcpy
+#include <iosfwd>  // for std::ostream* custom output target in Config
+#include <string>  // all names
+#include <vector>  // holds all results
+
+#define ANKERL_NANOBENCH(x) ANKERL_NANOBENCH_PRIVATE_##x()
+
+#define ANKERL_NANOBENCH_PRIVATE_CXX() __cplusplus
+#define ANKERL_NANOBENCH_PRIVATE_CXX98() 199711L
+#define ANKERL_NANOBENCH_PRIVATE_CXX11() 201103L
+#define ANKERL_NANOBENCH_PRIVATE_CXX14() 201402L
+#define ANKERL_NANOBENCH_PRIVATE_CXX17() 201703L
+
+#if ANKERL_NANOBENCH(CXX) >= ANKERL_NANOBENCH(CXX17)
+#    define ANKERL_NANOBENCH_PRIVATE_NODISCARD() [[nodiscard]]
+#else
+#    define ANKERL_NANOBENCH_PRIVATE_NODISCARD()
+#endif
+
+#if defined(__clang__)
+#    define ANKERL_NANOBENCH_PRIVATE_IGNORE_PADDED_PUSH() \
+        _Pragma("clang diagnostic push") _Pragma("clang diagnostic ignored \"-Wpadded\"")
+#    define ANKERL_NANOBENCH_PRIVATE_IGNORE_PADDED_POP() _Pragma("clang diagnostic pop")
+#else
+#    define ANKERL_NANOBENCH_PRIVATE_IGNORE_PADDED_PUSH()
+#    define ANKERL_NANOBENCH_PRIVATE_IGNORE_PADDED_POP()
+#endif
+
+#if defined(__GNUC__)
+#    define ANKERL_NANOBENCH_PRIVATE_IGNORE_EFFCPP_PUSH() _Pragma("GCC diagnostic push") _Pragma("GCC diagnostic ignored \"-Weffc++\"")
+#    define ANKERL_NANOBENCH_PRIVATE_IGNORE_EFFCPP_POP() _Pragma("GCC diagnostic pop")
+#else
+#    define ANKERL_NANOBENCH_PRIVATE_IGNORE_EFFCPP_PUSH()
+#    define ANKERL_NANOBENCH_PRIVATE_IGNORE_EFFCPP_POP()
+#endif
+
+#if defined(ANKERL_NANOBENCH_LOG_ENABLED)
+#    include <iostream>
+#    define ANKERL_NANOBENCH_LOG(x) std::cout << __FUNCTION__ << "@" << __LINE__ << ": " << x << std::endl
+#else
+#    define ANKERL_NANOBENCH_LOG(x)
+#endif
+
+#if defined(__linux__) && !defined(ANKERL_NANOBENCH_DISABLE_PERF_COUNTERS)
+#    define ANKERL_NANOBENCH_PRIVATE_PERF_COUNTERS() 1
+#else
+#    define ANKERL_NANOBENCH_PRIVATE_PERF_COUNTERS() 0
+#endif
+
+#if defined(__clang__)
+#    define ANKERL_NANOBENCH_NO_SANITIZE(...) __attribute__((no_sanitize(__VA_ARGS__)))
+#else
+#    define ANKERL_NANOBENCH_NO_SANITIZE(...)
+#endif
+
+#if defined(_MSC_VER)
+#    define ANKERL_NANOBENCH_PRIVATE_NOINLINE() __declspec(noinline)
+#else
+#    define ANKERL_NANOBENCH_PRIVATE_NOINLINE() __attribute__((noinline))
+#endif
+
+// workaround missing "is_trivially_copyable" in g++ < 5.0
+// See https://stackoverflow.com/a/31798726/48181
+#if defined(__GNUC__) && __GNUC__ < 5
+#    define ANKERL_NANOBENCH_IS_TRIVIALLY_COPYABLE(...) __has_trivial_copy(__VA_ARGS__)
+#else
+#    define ANKERL_NANOBENCH_IS_TRIVIALLY_COPYABLE(...) std::is_trivially_copyable<__VA_ARGS__>::value
+#endif
+
+// declarations ///////////////////////////////////////////////////////////////////////////////////
+
+namespace ankerl {
+namespace nanobench {
+
+using Clock = std::conditional<std::chrono::high_resolution_clock::is_steady, std::chrono::high_resolution_clock,
+                               std::chrono::steady_clock>::type;
+class Bench;
+struct Config;
+class Result;
+class Rng;
+class BigO;
+
+/**
+ * @brief Renders output from a mustache-like template and benchmark results.
+ *
+ * The templating facility here is heavily inspired by [mustache - logic-less templates](https://mustache.github.io/).
+ * It adds a few more features that are necessary to get all of the captured data out of nanobench. Please read the
+ * excellent [mustache manual](https://mustache.github.io/mustache.5.html) to see what this is all about.
+ *
+ * nanobench output has two nested layers, *result* and *measurement*.  Here is a hierarchy of the allowed tags:
+ *
+ * * `{{#result}}` Marks the begin of the result layer. Whatever comes after this will be instantiated as often as
+ *   a benchmark result is available. Within it, you can use these tags:
+ *
+ *    * `{{title}}` See Bench::title().
+ *
+ *    * `{{name}}` Benchmark name, usually directly provided with Bench::run(), but can also be set with Bench::name().
+ *
+ *    * `{{unit}}` Unit, e.g. `byte`. Defaults to `op`, see Bench::title().
+ *
+ *    * `{{batch}}` Batch size, see Bench::batch().
+ *
+ *    * `{{complexityN}}` Value used for asymptotic complexity calculation. See Bench::complexityN().
+ *
+ *    * `{{epochs}}` Number of epochs, see Bench::epochs().
+ *
+ *    * `{{clockResolution}}` Accuracy of the clock, i.e. what's the smallest time possible to measure with the clock.
+ *      For modern systems, this can be around 20 ns. This value is automatically determined by nanobench at the first
+ *      benchmark that is run, and used as a static variable throughout the application's runtime.
+ *
+ *    * `{{clockResolutionMultiple}}` Configuration multiplier for `clockResolution`. See Bench::clockResolutionMultiple().
+ *      This is the target runtime for each measurement (epoch). That means the more accurate your clock is, the faster
+ *      will be the benchmark. Basing the measurement's runtime on the clock resolution is the main reason why nanobench is so fast.
+ *
+ *    * `{{maxEpochTime}}` Configuration for a maximum time each measurement (epoch) is allowed to take. Note that at least
+ *      a single iteration will be performed, even when that takes longer than maxEpochTime. See Bench::maxEpochTime().
+ *
+ *    * `{{minEpochTime}}` Minimum epoch time, usually not set. See Bench::minEpochTime().
+ *
+ *    * `{{minEpochIterations}}` See Bench::minEpochIterations().
+ *
+ *    * `{{epochIterations}}` See Bench::epochIterations().
+ *
+ *    * `{{warmup}}` Number of iterations used before measuring starts. See Bench::warmup().
+ *
+ *    * `{{relative}}` True or false, depending on the setting you have used. See Bench::relative().
+ *
+ *    Apart from these tags, it is also possible to use some mathematical operations on the measurement data. The operations
+ *    are of the form `{{command(name)}}`.  Currently `name` can be one of `elapsed`, `iterations`. If performance counters
+ *    are available (currently only on current Linux systems), you also have `pagefaults`, `cpucycles`,
+ *    `contextswitches`, `instructions`, `branchinstructions`, and `branchmisses`. All the measuers (except `iterations`) are
+ *    provided for a single iteration (so `elapsed` is the time a single iteration took). The following tags are available:
+ *
+ *    * `{{median(<name>>)}}` Calculate median of a measurement data set, e.g. `{{median(elapsed)}}`.
+ *
+ *    * `{{average(<name>)}}` Average (mean) calculation.
+ *
+ *    * `{{medianAbsolutePercentError(<name>)}}` Calculates MdAPE, the Median Absolute Percentage Error. The MdAPE is an excellent
+ *      metric for the variation of measurements. It is more robust to outliers than the
+ *      [Mean absolute percentage error (M-APE)](https://en.wikipedia.org/wiki/Mean_absolute_percentage_error).
+ *      @f[
+ *       \mathrm{medianAbsolutePercentError}(e) = \mathrm{median}\{| \frac{e_i - \mathrm{median}\{e\}}{e_i}| \}
+ *      @f]
+ *      E.g. for *elapsed*: First, @f$ \mathrm{median}\{elapsed\} @f$ is calculated. This is used to calculate the absolute percentage
+ *      error to this median for each measurement, as in  @f$ | \frac{e_i - \mathrm{median}\{e\}}{e_i}| @f$. All these results
+ *      are sorted, and the middle value is chosen as the median absolute percent error.
+ *
+ *      This measurement is a bit hard to interpret, but it is very robust against outliers. E.g. a value of 5% means that half of the
+ *      measurements deviate less than 5% from the median, and the other deviate more than 5% from the median.
+ *
+ *    * `{{sum(<name>)}}` Sums of all the measurements. E.g. `{{sum(iterations)}}` will give you the total number of iterations
+*        measured in this benchmark.
+ *
+ *    * `{{minimum(<name>)}}` Minimum of all measurements.
+ *
+ *    * `{{maximum(<name>)}}` Maximum of all measurements.
+ *
+ *    * `{{sumProduct(<first>, <second>)}}` Calculates the sum of the products of corresponding measures:
+ *      @f[
+ *          \mathrm{sumProduct}(a,b) = \sum_{i=1}^{n}a_i\cdot b_i
+ *      @f]
+ *      E.g. to calculate total runtime of the benchmark, you multiply iterations with elapsed time for each measurement, and
+ *      sum these results up:
+ *      `{{sumProduct(iterations, elapsed)}}`.
+ *
+ *    * `{{#measurement}}` To access individual measurement results, open the begin tag for measurements.
+ *
+ *       * `{{elapsed}}` Average elapsed time per iteration, in seconds.
+ *
+ *       * `{{iterations}}` Number of iterations in the measurement. The number of iterations will fluctuate due
+ *         to some applied randomness, to enhance accuracy.
+ *
+ *       * `{{pagefaults}}` Average number of pagefaults per iteration.
+ *
+ *       * `{{cpucycles}}` Average number of CPU cycles processed per iteration.
+ *
+ *       * `{{contextswitches}}` Average number of context switches per iteration.
+ *
+ *       * `{{instructions}}` Average number of retired instructions per iteration.
+ *
+ *       * `{{branchinstructions}}` Average number of branches executed per iteration.
+ *
+ *       * `{{branchmisses}}` Average number of branches that were missed per iteration.
+ *
+ *    * `{{/measurement}}` Ends the measurement tag.
+ *
+ * * `{{/result}}` Marks the end of the result layer. This is the end marker for the template part that will be instantiated
+ *   for each benchmark result.
+ *
+ *
+ *  For the layer tags *result* and *measurement* you additionally can use these special markers:
+ *
+ *  * ``{{#-first}}`` - Begin marker of a template that will be instantiated *only for the first* entry in the layer. Use is only
+ *    allowed between the begin and end marker of the layer allowed. So between ``{{#result}}`` and ``{{/result}}``, or between
+ *    ``{{#measurement}}`` and ``{{/measurement}}``. Finish the template with ``{{/-first}}``.
+ *
+ *  * ``{{^-first}}`` - Begin marker of a template that will be instantiated *for each except the first* entry in the layer. This,
+ *    this is basically the inversion of ``{{#-first}}``. Use is only allowed between the begin and end marker of the layer allowed.
+ *    So between ``{{#result}}`` and ``{{/result}}``, or between ``{{#measurement}}`` and ``{{/measurement}}``.
+ *
+ *  * ``{{/-first}}`` - End marker for either ``{{#-first}}`` or ``{{^-first}}``.
+ *
+ *  * ``{{#-last}}`` - Begin marker of a template that will be instantiated *only for the last* entry in the layer. Use is only
+ *    allowed between the begin and end marker of the layer allowed. So between ``{{#result}}`` and ``{{/result}}``, or between
+ *    ``{{#measurement}}`` and ``{{/measurement}}``. Finish the template with ``{{/-last}}``.
+ *
+ *  * ``{{^-last}}`` - Begin marker of a template that will be instantiated *for each except the last* entry in the layer. This,
+ *    this is basically the inversion of ``{{#-last}}``. Use is only allowed between the begin and end marker of the layer allowed.
+ *    So between ``{{#result}}`` and ``{{/result}}``, or between ``{{#measurement}}`` and ``{{/measurement}}``.
+ *
+ *  * ``{{/-last}}`` - End marker for either ``{{#-last}}`` or ``{{^-last}}``.
+ *
+   @verbatim embed:rst
+
+   For an overview of all the possible data you can get out of nanobench, please see the tutorial at :ref:`tutorial-template-json`.
+
+   The templates that ship with nanobench are:
+
+   * :cpp:func:`templates::csv() <ankerl::nanobench::templates::csv()>`
+   * :cpp:func:`templates::json() <ankerl::nanobench::templates::json()>`
+   * :cpp:func:`templates::htmlBoxplot() <ankerl::nanobench::templates::htmlBoxplot()>`
+
+   @endverbatim
+ *
+ * @param mustacheTemplate The template.
+ * @param bench Benchmark, containing all the results.
+ * @param out Output for the generated output.
+ */
+void render(char const* mustacheTemplate, Bench const& bench, std::ostream& out);
+
+/**
+ * Same as render(char const* mustacheTemplate, Bench const& bench, std::ostream& out), but for when
+ * you only have results available.
+ *
+ * @param mustacheTemplate The template.
+ * @param results All the results to be used for rendering.
+ * @param out Output for the generated output.
+ */
+void render(char const* mustacheTemplate, std::vector<Result> const& results, std::ostream& out);
+
+// Contains mustache-like templates
+namespace templates {
+
+/*!
+  @brief CSV data for the benchmark results.
+
+  Generates a comma-separated values dataset. First line is the header, each following line is a summary of each benchmark run.
+
+  @verbatim embed:rst
+  See the tutorial at :ref:`tutorial-template-csv` for an example.
+  @endverbatim
+ */
+char const* csv() noexcept;
+
+/*!
+  @brief HTML output that uses plotly to generate an interactive boxplot chart. See the tutorial for an example output.
+
+  The output uses only the elapsed time, and displays each epoch as a single dot.
+  @verbatim embed:rst
+  See the tutorial at :ref:`tutorial-template-html` for an example.
+  @endverbatim
+
+  @see ankerl::nanobench::render()
+ */
+char const* htmlBoxplot() noexcept;
+
+/*!
+  @brief Template to generate JSON data.
+
+  The generated JSON data contains *all* data that has been generated. All times are as double values, in seconds. The output can get
+  quite large.
+  @verbatim embed:rst
+  See the tutorial at :ref:`tutorial-template-json` for an example.
+  @endverbatim
+ */
+char const* json() noexcept;
+
+} // namespace templates
+
+namespace detail {
+
+template <typename T>
+struct PerfCountSet;
+
+class IterationLogic;
+class PerformanceCounters;
+
+#if ANKERL_NANOBENCH(PERF_COUNTERS)
+class LinuxPerformanceCounters;
+#endif
+
+} // namespace detail
+} // namespace nanobench
+} // namespace ankerl
+
+// definitions ////////////////////////////////////////////////////////////////////////////////////
+
+namespace ankerl {
+namespace nanobench {
+namespace detail {
+
+template <typename T>
+struct PerfCountSet {
+    T pageFaults{};
+    T cpuCycles{};
+    T contextSwitches{};
+    T instructions{};
+    T branchInstructions{};
+    T branchMisses{};
+};
+
+} // namespace detail
+
+ANKERL_NANOBENCH(IGNORE_PADDED_PUSH)
+struct Config {
+    // actual benchmark config
+    std::string mBenchmarkTitle = "benchmark";
+    std::string mBenchmarkName = "noname";
+    std::string mUnit = "op";
+    double mBatch = 1.0;
+    double mComplexityN = -1.0;
+    size_t mNumEpochs = 11;
+    size_t mClockResolutionMultiple = static_cast<size_t>(1000);
+    std::chrono::nanoseconds mMaxEpochTime = std::chrono::milliseconds(100);
+    std::chrono::nanoseconds mMinEpochTime{};
+    uint64_t mMinEpochIterations{1};
+    uint64_t mEpochIterations{0}; // If not 0, run *exactly* these number of iterations per epoch.
+    uint64_t mWarmup = 0;
+    std::ostream* mOut = nullptr;
+    bool mShowPerformanceCounters = true;
+    bool mIsRelative = false;
+
+    Config();
+    ~Config();
+    Config& operator=(Config const&);
+    Config& operator=(Config&&);
+    Config(Config const&);
+    Config(Config&&) noexcept;
+};
+ANKERL_NANOBENCH(IGNORE_PADDED_POP)
+
+// Result returned after a benchmark has finished. Can be used as a baseline for relative().
+ANKERL_NANOBENCH(IGNORE_PADDED_PUSH)
+class Result {
+public:
+    enum class Measure : size_t {
+        elapsed,
+        iterations,
+        pagefaults,
+        cpucycles,
+        contextswitches,
+        instructions,
+        branchinstructions,
+        branchmisses,
+        _size
+    };
+
+    explicit Result(Config const& benchmarkConfig);
+
+    ~Result();
+    Result& operator=(Result const&);
+    Result& operator=(Result&&);
+    Result(Result const&);
+    Result(Result&&) noexcept;
+
+    // adds new measurement results
+    // all values are scaled by iters (except iters...)
+    void add(Clock::duration totalElapsed, uint64_t iters, detail::PerformanceCounters const& pc);
+
+    ANKERL_NANOBENCH(NODISCARD) Config const& config() const noexcept;
+
+    ANKERL_NANOBENCH(NODISCARD) double median(Measure m) const;
+    ANKERL_NANOBENCH(NODISCARD) double medianAbsolutePercentError(Measure m) const;
+    ANKERL_NANOBENCH(NODISCARD) double average(Measure m) const;
+    ANKERL_NANOBENCH(NODISCARD) double sum(Measure m) const noexcept;
+    ANKERL_NANOBENCH(NODISCARD) double sumProduct(Measure m1, Measure m2) const noexcept;
+    ANKERL_NANOBENCH(NODISCARD) double minimum(Measure m) const noexcept;
+    ANKERL_NANOBENCH(NODISCARD) double maximum(Measure m) const noexcept;
+
+    ANKERL_NANOBENCH(NODISCARD) bool has(Measure m) const noexcept;
+    ANKERL_NANOBENCH(NODISCARD) double get(size_t idx, Measure m) const;
+    ANKERL_NANOBENCH(NODISCARD) bool empty() const noexcept;
+    ANKERL_NANOBENCH(NODISCARD) size_t size() const noexcept;
+
+    // Finds string, if not found, returns _size.
+    static Measure fromString(std::string const& str);
+
+private:
+    Config mConfig{};
+    std::vector<std::vector<double>> mNameToMeasurements{};
+};
+ANKERL_NANOBENCH(IGNORE_PADDED_POP)
+
+/**
+ * An extremely fast random generator. Currently, this implements *RomuDuoJr*, developed by Mark Overton. Source:
+ * http://www.romu-random.org/
+ *
+ * RomuDuoJr is extremely fast and provides reasonable good randomness. Not enough for large jobs, but definitely
+ * good enough for a benchmarking framework.
+ *
+ *  * Estimated capacity: @f$ 2^{51} @f$ bytes
+ *  * Register pressure: 4
+ *  * State size: 128 bits
+ *
+ * This random generator is a drop-in replacement for the generators supplied by ``<random>``. It is not
+ * cryptographically secure. It's intended purpose is to be very fast so that benchmarks that make use
+ * of randomness are not distorted too much by the random generator.
+ *
+ * Rng also provides a few non-standard helpers, optimized for speed.
+ */
+class Rng final {
+public:
+    /**
+     * @brief This RNG provides 64bit randomness.
+     */
+    using result_type = uint64_t;
+
+    static constexpr uint64_t(min)();
+    static constexpr uint64_t(max)();
+
+    /**
+     * As a safety precausion, we don't allow copying. Copying a PRNG would mean you would have two random generators that produce the
+     * same sequence, which is generally not what one wants. Instead create a new rng with the default constructor Rng(), which is
+     * automatically seeded from `std::random_device`. If you really need a copy, use copy().
+     */
+    Rng(Rng const&) = delete;
+
+    /**
+     * Same as Rng(Rng const&), we don't allow assignment. If you need a new Rng create one with the default constructor Rng().
+     */
+    Rng& operator=(Rng const&) = delete;
+
+    // moving is ok
+    Rng(Rng&&) noexcept = default;
+    Rng& operator=(Rng&&) noexcept = default;
+    ~Rng() noexcept = default;
+
+    /**
+     * @brief Creates a new Random generator with random seed.
+     *
+     * Instead of a default seed (as the random generators from the STD), this properly seeds the random generator from
+     * `std::random_device`. It guarantees correct seeding. Note that seeding can be relatively slow, depending on the source of
+     * randomness used. So it is best to create a Rng once and use it for all your randomness purposes.
+     */
+    Rng();
+
+    /*!
+      Creates a new Rng that is seeded with a specific seed. Each Rng created from the same seed will produce the same randomness
+      sequence. This can be useful for deterministic behavior.
+
+      @verbatim embed:rst
+      .. note::
+
+         The random algorithm might change between nanobench releases. Whenever a faster and/or better random
+         generator becomes available, I will switch the implementation.
+      @endverbatim
+
+      As per the Romu paper, this seeds the Rng with splitMix64 algorithm and performs 10 initial rounds for further mixing up of the
+      internal state.
+
+      @param seed  The 64bit seed. All values are allowed, even 0.
+     */
+    explicit Rng(uint64_t seed) noexcept;
+    Rng(uint64_t x, uint64_t y) noexcept;
+
+    /**
+     * Creates a copy of the Rng, thus the copy provides exactly the same random sequence as the original.
+     */
+    ANKERL_NANOBENCH(NODISCARD) Rng copy() const noexcept;
+
+    /**
+     * @brief Produces a 64bit random value. This should be very fast, thus it is marked as inline. In my benchmark, this is ~46 times
+     * faster than `std::default_random_engine` for producing 64bit random values. It seems that the fastest std contender is
+     * `std::mt19937_64`. Still, this RNG is 2-3 times as fast.
+     *
+     * @return uint64_t The next 64 bit random value.
+     */
+    inline uint64_t operator()() noexcept;
+
+    // This is slightly biased. See
+
+    /**
+     * Generates a random number between 0 and range (excluding range).
+     *
+     * The algorithm only produces 32bit numbers, and is slightly biased. The effect is quite small unless your range is close to the
+     * maximum value of an integer. It is possible to correct the bias with rejection sampling (see
+     * [here](https://lemire.me/blog/2016/06/30/fast-random-shuffling/), but this is most likely irrelevant in practices for the
+     * purposes of this Rng.
+     *
+     * See Daniel Lemire's blog post [A fast alternative to the modulo
+     * reduction](https://lemire.me/blog/2016/06/27/a-fast-alternative-to-the-modulo-reduction/)
+     *
+     * @param range Upper exclusive range. E.g a value of 3 will generate random numbers 0, 1, 2.
+     * @return uint32_t Generated random values in range [0, range(.
+     */
+    inline uint32_t bounded(uint32_t range) noexcept;
+
+    // random double in range [0, 1(
+    // see http://prng.di.unimi.it/
+
+    /**
+     * Provides a random uniform double value between 0 and 1. This uses the method described in [Generating uniform doubles in the
+     * unit interval](http://prng.di.unimi.it/), and is extremely fast.
+     *
+     * @return double Uniformly distributed double value in range [0,1(, excluding 1.
+     */
+    inline double uniform01() noexcept;
+
+    /**
+     * Shuffles all entries in the given container. Although this has a slight bias due to the implementation of bounded(), this is
+     * preferable to `std::shuffle` because it is over 5 times faster. See Daniel Lemire's blog post [Fast random
+     * shuffling](https://lemire.me/blog/2016/06/30/fast-random-shuffling/).
+     *
+     * @param container The whole container will be shuffled.
+     */
+    template <typename Container>
+    void shuffle(Container& container) noexcept;
+
+private:
+    static constexpr uint64_t rotl(uint64_t x, unsigned k) noexcept;
+
+    uint64_t mX;
+    uint64_t mY;
+};
+
+/**
+ * @brief Main entry point to nanobench's benchmarking facility.
+ *
+ * It holds configuration and results from one or more benchmark runs. Usually it is used in a single line, where the object is
+ * constructed, configured, and then a benchmark is run. E.g. like this:
+ *
+ *     ankerl::nanobench::Bench().unit("byte").batch(1000).run("random fluctuations", [&] {
+ *         // here be the benchmark code
+ *     });
+ *
+ * In that example Bench() constructs the benchmark, it is then configured with unit() and batch(), and after configuration a
+ * benchmark is executed with run(). Once run() has finished, it prints the result to `std::cout`. It would also store the results
+ * in the Bench instance, but in this case the object is immediately destroyed so it's not available any more.
+ */
+ANKERL_NANOBENCH(IGNORE_PADDED_PUSH)
+class Bench {
+public:
+    /**
+     * @brief Creates a new benchmark for configuration and running of benchmarks.
+     */
+    Bench();
+
+    Bench(Bench&& other);
+    Bench& operator=(Bench&& other);
+    Bench(Bench const& other);
+    Bench& operator=(Bench const& other);
+    ~Bench() noexcept;
+
+    /*!
+      @brief Repeatedly calls `op()` based on the configuration, and performs measurements.
+
+      This call is marked with `noinline` to prevent the compiler to optimize beyond different benchmarks. This can have quite a big
+      effect on benchmark accuracy.
+
+      @verbatim embed:rst
+      .. note::
+
+        Each call to your lambda must have a side effect that the compiler can't possibly optimize it away. E.g. add a result to an
+        externally defined number (like `x` in the above example), and finally call `doNotOptimizeAway` on the variables the compiler
+        must not remove. You can also use :cpp:func:`ankerl::nanobench::doNotOptimizeAway` directly in the lambda, but be aware that
+        this has a small overhead.
+
+      @endverbatim
+
+      @tparam Op The code to benchmark.
+     */
+    template <typename Op>
+    ANKERL_NANOBENCH(NOINLINE)
+    Bench& run(char const* benchmarkName, Op&& op);
+
+    template <typename Op>
+    ANKERL_NANOBENCH(NOINLINE)
+    Bench& run(std::string const& benchmarkName, Op&& op);
+
+    /**
+     * @brief Same as run(char const* benchmarkName, Op op), but instead uses the previously set name.
+     * @tparam Op The code to benchmark.
+     */
+    template <typename Op>
+    ANKERL_NANOBENCH(NOINLINE)
+    Bench& run(Op&& op);
+
+    /**
+     * @brief Title of the benchmark, will be shown in the table header. Changing the title will start a new markdown table.
+     *
+     * @param benchmarkTitle The title of the benchmark.
+     */
+    Bench& title(char const* benchmarkTitle);
+    Bench& title(std::string const& benchmarkTitle);
+    ANKERL_NANOBENCH(NODISCARD) std::string const& title() const noexcept;
+
+    /// Name of the benchmark, will be shown in the table row.
+    Bench& name(char const* benchmarkName);
+    Bench& name(std::string const& benchmarkName);
+    ANKERL_NANOBENCH(NODISCARD) std::string const& name() const noexcept;
+
+    /**
+     * @brief Sets the batch size.
+     *
+     * E.g. number of processed byte, or some other metric for the size of the processed data in each iteration. If you benchmark
+     * hashing of a 1000 byte long string and want byte/sec as a result, you can specify 1000 as the batch size.
+     *
+     * @tparam T Any input type is internally cast to `double`.
+     * @param b batch size
+     */
+    template <typename T>
+    Bench& batch(T b) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) double batch() const noexcept;
+
+    /**
+     * @brief Sets the operation unit.
+     *
+     * Defaults to "op". Could be e.g. "byte" for string processing. This is used for the table header, e.g. to show `ns/byte`. Use
+     * singular (*byte*, not *bytes*). A change clears the currently collected results.
+     *
+     * @param unit The unit name.
+     */
+    Bench& unit(char const* unit);
+    Bench& unit(std::string const& unit);
+    ANKERL_NANOBENCH(NODISCARD) std::string const& unit() const noexcept;
+
+    /**
+     * @brief Set the output stream where the resulting markdown table will be printed to.
+     *
+     * The default is `&std::cout`. You can disable all output by setting `nullptr`.
+     *
+     * @param outstream Pointer to output stream, can be `nullptr`.
+     */
+    Bench& output(std::ostream* outstream) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) std::ostream* output() const noexcept;
+
+    /**
+     * Modern processors have a very accurate clock, being able to measure as low as 20 nanoseconds. This is the main trick nanobech to
+     * be so fast: we find out how accurate the clock is, then run the benchmark only so often that the clock's accuracy is good enough
+     * for accurate measurements.
+     *
+     * The default is to run one epoch for 1000 times the clock resolution. So for 20ns resolution and 11 epochs, this gives a total
+     * runtime of
+     *
+     * @f[
+     * 20ns * 1000 * 11 \approx 0.2ms
+     * @f]
+     *
+     * To be precise, nanobench adds a 0-20% random noise to each evaluation. This is to prevent any aliasing effects, and further
+     * improves accuracy.
+     *
+     * Total runtime will be higher though: Some initial time is needed to find out the target number of iterations for each epoch, and
+     * there is some overhead involved to start & stop timers and calculate resulting statistics and writing the output.
+     *
+     * @param multiple Target number of times of clock resolution. Usually 1000 is a good compromise between runtime and accuracy.
+     */
+    Bench& clockResolutionMultiple(size_t multiple) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) size_t clockResolutionMultiple() const noexcept;
+
+    /**
+     * @brief Controls number of epochs, the number of measurements to perform.
+     *
+     * The reported result will be the median of evaluation of each epoch. The higher you choose this, the more
+     * deterministic the result be and outliers will be more easily removed. Also the `err%` will be more accurate the higher this
+     * number is. Note that the `err%` will not necessarily decrease when number of epochs is increased. But it will be a more accurate
+     * representation of the benchmarked code's runtime stability.
+     *
+     * Choose the value wisely. In practice, 11 has been shown to be a reasonable choice between runtime performance and accuracy.
+     * This setting goes hand in hand with minEpocIterations() (or minEpochTime()). If you are more interested in *median* runtime, you
+     * might want to increase epochs(). If you are more interested in *mean* runtime, you might want to increase minEpochIterations()
+     * instead.
+     *
+     * @param numEpochs Number of epochs.
+     */
+    Bench& epochs(size_t numEpochs) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) size_t epochs() const noexcept;
+
+    /**
+     * @brief Upper limit for the runtime of each epoch.
+     *
+     * As a safety precausion if the clock is not very accurate, we can set an upper limit for the maximum evaluation time per
+     * epoch. Default is 100ms. At least a single evaluation of the benchmark is performed.
+     *
+     * @see minEpochTime(), minEpochIterations()
+     *
+     * @param t Maximum target runtime for a single epoch.
+     */
+    Bench& maxEpochTime(std::chrono::nanoseconds t) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) std::chrono::nanoseconds maxEpochTime() const noexcept;
+
+    /**
+     * @brief Minimum time each epoch should take.
+     *
+     * Default is zero, so we are fully relying on clockResolutionMultiple(). In most cases this is exactly what you want. If you see
+     * that the evaluation is unreliable with a high `err%`, you can increase either minEpochTime() or minEpochIterations().
+     *
+     * @see maxEpochTime(), minEpochIterations()
+     *
+     * @param t Minimum time each epoch should take.
+     */
+    Bench& minEpochTime(std::chrono::nanoseconds t) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) std::chrono::nanoseconds minEpochTime() const noexcept;
+
+    /**
+     * @brief Sets the minimum number of iterations each epoch should take.
+     *
+     * Default is 1, and we rely on clockResolutionMultiple(). If the `err%` is high and you want a more smooth result, you might want
+     * to increase the minimum number or iterations, or increase the minEpochTime().
+     *
+     * @see minEpochTime(), maxEpochTime(), minEpochIterations()
+     *
+     * @param numIters Minimum number of iterations per epoch.
+     */
+    Bench& minEpochIterations(uint64_t numIters) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) uint64_t minEpochIterations() const noexcept;
+
+    /**
+     * Sets exactly the number of iterations for each epoch. Ignores all other epoch limits. This forces nanobench to use exactly
+     * the given number of iterations for each epoch, not more and not less. Default is 0 (disabled).
+     *
+     * @param numIters Exact number of iterations to use. Set to 0 to disable.
+     */
+    Bench& epochIterations(uint64_t numIters) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) uint64_t epochIterations() const noexcept;
+
+    /**
+     * @brief Sets a number of iterations that are initially performed without any measurements.
+     *
+     * Some benchmarks need a few evaluations to warm up caches / database / whatever access. Normally this should not be needed, since
+     * we show the median result so initial outliers will be filtered away automatically. If the warmup effect is large though, you
+     * might want to set it. Default is 0.
+     *
+     * @param numWarmupIters Number of warmup iterations.
+     */
+    Bench& warmup(uint64_t numWarmupIters) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) uint64_t warmup() const noexcept;
+
+    /**
+     * @brief Marks the next run as the baseline.
+     *
+     * Call `relative(true)` to mark the run as the baseline. Successive runs will be compared to this run. It is calculated by
+     *
+     * @f[
+     * 100\% * \frac{baseline}{runtime}
+     * @f]
+     *
+     *  * 100% means it is exactly as fast as the baseline
+     *  * >100% means it is faster than the baseline. E.g. 200% means the current run is twice as fast as the baseline.
+     *  * <100% means it is slower than the baseline. E.g. 50% means it is twice as slow as the baseline.
+     *
+     * See the tutorial section "Comparing Results" for example usage.
+     *
+     * @param isRelativeEnabled True to enable processing
+     */
+    Bench& relative(bool isRelativeEnabled) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) bool relative() const noexcept;
+
+    /**
+     * @brief Enables/disables performance counters.
+     *
+     * On Linux nanobench has a powerful feature to use performance counters. This enables counting of retired instructions, count
+     * number of branches, missed branches, etc. On default this is enabled, but you can disable it if you don't need that feature.
+     *
+     * @param showPerformanceCounters True to enable, false to disable.
+     */
+    Bench& performanceCounters(bool showPerformanceCounters) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) bool performanceCounters() const noexcept;
+
+    /**
+     * @brief Retrieves all benchmark results collected by the bench object so far.
+     *
+     * Each call to run() generates a Result that is stored within the Bench instance. This is mostly for advanced users who want to
+     * see all the nitty gritty detials.
+     *
+     * @return All results collected so far.
+     */
+    ANKERL_NANOBENCH(NODISCARD) std::vector<Result> const& results() const noexcept;
+
+    /*!
+      @verbatim embed:rst
+
+      Convenience shortcut to :cpp:func:`ankerl::nanobench::doNotOptimizeAway`.
+
+      @endverbatim
+     */
+    template <typename Arg>
+    Bench& doNotOptimizeAway(Arg&& arg);
+
+    /*!
+      @verbatim embed:rst
+
+      Sets N for asymptotic complexity calculation, so it becomes possible to calculate `Big O
+      <https://en.wikipedia.org/wiki/Big_O_notation>`_ from multiple benchmark evaluations.
+
+      Use :cpp:func:`ankerl::nanobench::Bench::complexityBigO` when the evaluation has finished. See the tutorial
+      :ref:`asymptotic-complexity` for details.
+
+      @endverbatim
+
+      @tparam T Any type is cast to `double`.
+      @param b Length of N for the next benchmark run, so it is possible to calculate `bigO`.
+     */
+    template <typename T>
+    Bench& complexityN(T b) noexcept;
+    ANKERL_NANOBENCH(NODISCARD) double complexityN() const noexcept;
+
+    /*!
+      Calculates [Big O](https://en.wikipedia.org/wiki/Big_O_notation>) of the results with all preconfigured complexity functions.
+      Currently these complexity functions are fitted into the benchmark results:
+
+       @f$ \mathcal{O}(1) @f$,
+       @f$ \mathcal{O}(n) @f$,
+       @f$ \mathcal{O}(\log{}n) @f$,
+       @f$ \mathcal{O}(n\log{}n) @f$,
+       @f$ \mathcal{O}(n^2) @f$,
+       @f$ \mathcal{O}(n^3) @f$.
+
+      If we e.g. evaluate the complexity of `std::sort`, this is the result of `std::cout << bench.complexityBigO()`:
+
+      ```
+      |   coefficient |   err% | complexity
+      |--------------:|-------:|------------
+      |   5.08935e-09 |   2.6% | O(n log n)
+      |   6.10608e-08 |   8.0% | O(n)
+      |   1.29307e-11 |  47.2% | O(n^2)
+      |   2.48677e-15 |  69.6% | O(n^3)
+      |   9.88133e-06 | 132.3% | O(log n)
+      |   5.98793e-05 | 162.5% | O(1)
+      ```
+
+      So in this case @f$ \mathcal{O}(n\log{}n) @f$ provides the best approximation.
+
+      @verbatim embed:rst
+      See the tutorial :ref:`asymptotic-complexity` for details.
+      @endverbatim
+      @return Evaluation results, which can be printed or otherwise inspected.
+     */
+    std::vector<BigO> complexityBigO() const;
+
+    /**
+     * @brief Calculates bigO for a custom function.
+     *
+     * E.g. to calculate the mean squared error for @f$ \mathcal{O}(\log{}\log{}n) @f$, which is not part of the default set of
+     * complexityBigO(), you can do this:
+     *
+     * ```
+     * auto logLogN = bench.complexityBigO("O(log log n)", [](double n) {
+     *     return std::log2(std::log2(n));
+     * });
+     * ```
+     *
+     * The resulting mean squared error can be printed with `std::cout << logLogN`. E.g. it prints something like this:
+     *
+     * ```text
+     * 2.46985e-05 * O(log log n), rms=1.48121
+     * ```
+     *
+     * @tparam Op Type of mapping operation.
+     * @param name Name for the function, e.g. "O(log log n)"
+     * @param op Op's operator() maps a `double` with the desired complexity function, e.g. `log2(log2(n))`.
+     * @return BigO Error calculation, which is streamable to std::cout.
+     */
+    template <typename Op>
+    BigO complexityBigO(char const* name, Op op) const;
+
+    template <typename Op>
+    BigO complexityBigO(std::string const& name, Op op) const;
+
+    /*!
+      @verbatim embed:rst
+
+      Convenience shortcut to :cpp:func:`ankerl::nanobench::render`.
+
+      @endverbatim
+     */
+    Bench& render(char const* templateContent, std::ostream& os);
+
+    Bench& config(Config const& benchmarkConfig);
+    ANKERL_NANOBENCH(NODISCARD) Config const& config() const noexcept;
+
+private:
+    Config mConfig{};
+    std::vector<Result> mResults{};
+};
+ANKERL_NANOBENCH(IGNORE_PADDED_POP)
+
+/**
+ * @brief Makes sure none of the given arguments are optimized away by the compiler.
+ *
+ * @tparam Arg Type of the argument that shouldn't be optimized away.
+ * @param arg The input that we mark as being used, even though we don't do anything with it.
+ */
+template <typename Arg>
+void doNotOptimizeAway(Arg&& arg);
+
+namespace detail {
+
+#if defined(_MSC_VER)
+void doNotOptimizeAwaySink(void const*);
+
+template <typename T>
+void doNotOptimizeAway(T const& val);
+
+#else
+
+// see folly's Benchmark.h
+template <typename T>
+constexpr bool doNotOptimizeNeedsIndirect() {
+    using Decayed = typename std::decay<T>::type;
+    return !ANKERL_NANOBENCH_IS_TRIVIALLY_COPYABLE(Decayed) || sizeof(Decayed) > sizeof(long) || std::is_pointer<Decayed>::value;
+}
+
+template <typename T>
+typename std::enable_if<!doNotOptimizeNeedsIndirect<T>()>::type doNotOptimizeAway(T const& val) {
+    // NOLINTNEXTLINE(hicpp-no-assembler)
+    asm volatile("" ::"r"(val));
+}
+
+template <typename T>
+typename std::enable_if<doNotOptimizeNeedsIndirect<T>()>::type doNotOptimizeAway(T const& val) {
+    // NOLINTNEXTLINE(hicpp-no-assembler)
+    asm volatile("" ::"m"(val) : "memory");
+}
+#endif
+
+// internally used, but visible because run() is templated.
+// Not movable/copy-able, so we simply use a pointer instead of unique_ptr. This saves us from
+// having to include <memory>, and the template instantiation overhead of unique_ptr which is unfortunately quite significant.
+ANKERL_NANOBENCH(IGNORE_EFFCPP_PUSH)
+class IterationLogic {
+public:
+    explicit IterationLogic(Bench const& config) noexcept;
+    ~IterationLogic();
+
+    ANKERL_NANOBENCH(NODISCARD) uint64_t numIters() const noexcept;
+    void add(std::chrono::nanoseconds elapsed, PerformanceCounters const& pc) noexcept;
+    void moveResultTo(std::vector<Result>& results) noexcept;
+
+private:
+    struct Impl;
+    Impl* mPimpl;
+};
+ANKERL_NANOBENCH(IGNORE_EFFCPP_POP)
+
+ANKERL_NANOBENCH(IGNORE_PADDED_PUSH)
+class PerformanceCounters {
+public:
+    PerformanceCounters(PerformanceCounters const&) = delete;
+    PerformanceCounters& operator=(PerformanceCounters const&) = delete;
+
+    PerformanceCounters();
+    ~PerformanceCounters();
+
+    void beginMeasure();
+    void endMeasure();
+    void updateResults(uint64_t numIters);
+
+    ANKERL_NANOBENCH(NODISCARD) PerfCountSet<uint64_t> const& val() const noexcept;
+    ANKERL_NANOBENCH(NODISCARD) PerfCountSet<bool> const& has() const noexcept;
+
+private:
+#if ANKERL_NANOBENCH(PERF_COUNTERS)
+    LinuxPerformanceCounters* mPc = nullptr;
+#endif
+    PerfCountSet<uint64_t> mVal{};
+    PerfCountSet<bool> mHas{};
+};
+ANKERL_NANOBENCH(IGNORE_PADDED_POP)
+
+// Gets the singleton
+PerformanceCounters& performanceCounters();
+
+} // namespace detail
+
+class BigO {
+public:
+    using RangeMeasure = std::vector<std::pair<double, double>>;
+
+    template <typename Op>
+    static RangeMeasure mapRangeMeasure(RangeMeasure data, Op op) {
+        for (auto& rangeMeasure : data) {
+            rangeMeasure.first = op(rangeMeasure.first);
+        }
+        return data;
+    }
+
+    static RangeMeasure collectRangeMeasure(std::vector<Result> const& results);
+
+    template <typename Op>
+    BigO(char const* bigOName, RangeMeasure const& rangeMeasure, Op rangeToN)
+        : BigO(bigOName, mapRangeMeasure(rangeMeasure, rangeToN)) {}
+
+    template <typename Op>
+    BigO(std::string const& bigOName, RangeMeasure const& rangeMeasure, Op rangeToN)
+        : BigO(bigOName, mapRangeMeasure(rangeMeasure, rangeToN)) {}
+
+    BigO(char const* bigOName, RangeMeasure const& scaledRangeMeasure);
+    BigO(std::string const& bigOName, RangeMeasure const& scaledRangeMeasure);
+    ANKERL_NANOBENCH(NODISCARD) std::string const& name() const noexcept;
+    ANKERL_NANOBENCH(NODISCARD) double constant() const noexcept;
+    ANKERL_NANOBENCH(NODISCARD) double normalizedRootMeanSquare() const noexcept;
+    ANKERL_NANOBENCH(NODISCARD) bool operator<(BigO const& other) const noexcept;
+
+private:
+    std::string mName{};
+    double mConstant{};
+    double mNormalizedRootMeanSquare{};
+};
+std::ostream& operator<<(std::ostream& os, BigO const& bigO);
+std::ostream& operator<<(std::ostream& os, std::vector<ankerl::nanobench::BigO> const& bigOs);
+
+} // namespace nanobench
+} // namespace ankerl
+
+// implementation /////////////////////////////////////////////////////////////////////////////////
+
+namespace ankerl {
+namespace nanobench {
+
+constexpr uint64_t(Rng::min)() {
+    return 0;
+}
+
+constexpr uint64_t(Rng::max)() {
+    return (std::numeric_limits<uint64_t>::max)();
+}
+
+ANKERL_NANOBENCH_NO_SANITIZE("integer")
+uint64_t Rng::operator()() noexcept {
+    auto x = mX;
+
+    mX = UINT64_C(15241094284759029579) * mY;
+    mY = rotl(mY - x, 27);
+
+    return x;
+}
+
+ANKERL_NANOBENCH_NO_SANITIZE("integer")
+uint32_t Rng::bounded(uint32_t range) noexcept {
+    uint64_t r32 = static_cast<uint32_t>(operator()());
+    auto multiresult = r32 * range;
+    return static_cast<uint32_t>(multiresult >> 32U);
+}
+
+double Rng::uniform01() noexcept {
+    auto i = (UINT64_C(0x3ff) << 52U) | (operator()() >> 12U);
+    // can't use union in c++ here for type puning, it's undefined behavior.
+    // std::memcpy is optimized anyways.
+    double d;
+    std::memcpy(&d, &i, sizeof(double));
+    return d - 1.0;
+}
+
+template <typename Container>
+void Rng::shuffle(Container& container) noexcept {
+    auto size = static_cast<uint32_t>(container.size());
+    for (auto i = size; i > 1U; --i) {
+        using std::swap;
+        auto p = bounded(i); // number in [0, i)
+        swap(container[i - 1], container[p]);
+    }
+}
+
+constexpr uint64_t Rng::rotl(uint64_t x, unsigned k) noexcept {
+    return (x << k) | (x >> (64U - k));
+}
+
+template <typename Op>
+ANKERL_NANOBENCH_NO_SANITIZE("integer")
+Bench& Bench::run(Op&& op) {
+    // It is important that this method is kept short so the compiler can do better optimizations/ inlining of op()
+    detail::IterationLogic iterationLogic(*this);
+    auto& pc = detail::performanceCounters();
+
+    while (auto n = iterationLogic.numIters()) {
+        pc.beginMeasure();
+        Clock::time_point before = Clock::now();
+        while (n-- > 0) {
+            op();
+        }
+        Clock::time_point after = Clock::now();
+        pc.endMeasure();
+        pc.updateResults(iterationLogic.numIters());
+        iterationLogic.add(after - before, pc);
+    }
+    iterationLogic.moveResultTo(mResults);
+    return *this;
+}
+
+// Performs all evaluations.
+template <typename Op>
+Bench& Bench::run(char const* benchmarkName, Op&& op) {
+    name(benchmarkName);
+    return run(std::forward<Op>(op));
+}
+
+template <typename Op>
+Bench& Bench::run(std::string const& benchmarkName, Op&& op) {
+    name(benchmarkName);
+    return run(std::forward<Op>(op));
+}
+
+template <typename Op>
+BigO Bench::complexityBigO(char const* benchmarkName, Op op) const {
+    return BigO(benchmarkName, BigO::collectRangeMeasure(mResults), op);
+}
+
+template <typename Op>
+BigO Bench::complexityBigO(std::string const& benchmarkName, Op op) const {
+    return BigO(benchmarkName, BigO::collectRangeMeasure(mResults), op);
+}
+
+// Set the batch size, e.g. number of processed bytes, or some other metric for the size of the processed data in each iteration.
+// Any argument is cast to double.
+template <typename T>
+Bench& Bench::batch(T b) noexcept {
+    mConfig.mBatch = static_cast<double>(b);
+    return *this;
+}
+
+// Sets the computation complexity of the next run. Any argument is cast to double.
+template <typename T>
+Bench& Bench::complexityN(T n) noexcept {
+    mConfig.mComplexityN = static_cast<double>(n);
+    return *this;
+}
+
+// Convenience: makes sure none of the given arguments are optimized away by the compiler.
+template <typename Arg>
+Bench& Bench::doNotOptimizeAway(Arg&& arg) {
+    detail::doNotOptimizeAway(std::forward<Arg>(arg));
+    return *this;
+}
+
+// Makes sure none of the given arguments are optimized away by the compiler.
+template <typename Arg>
+void doNotOptimizeAway(Arg&& arg) {
+    detail::doNotOptimizeAway(std::forward<Arg>(arg));
+}
+
+namespace detail {
+
+#if defined(_MSC_VER)
+template <typename T>
+void doNotOptimizeAway(T const& val) {
+    doNotOptimizeAwaySink(&val);
+}
+
+#endif
+
+} // namespace detail
+} // namespace nanobench
+} // namespace ankerl
+
+#if defined(ANKERL_NANOBENCH_IMPLEMENT)
+
+///////////////////////////////////////////////////////////////////////////////////////////////////
+// implementation part - only visible in .cpp
+///////////////////////////////////////////////////////////////////////////////////////////////////
+
+#    include <algorithm> // sort, reverse
+#    include <atomic>    // compare_exchange_strong in loop overhead
+#    include <cstdlib>   // getenv
+#    include <cstring>   // strstr, strncmp
+#    include <fstream>   // ifstream to parse proc files
+#    include <iomanip>   // setw, setprecision
+#    include <iostream>  // cout
+#    include <numeric>   // accumulate
+#    include <random>    // random_device
+#    include <sstream>   // to_s in Number
+#    include <stdexcept> // throw for rendering templates
+#    include <tuple>     // std::tie
+#    if defined(__linux__)
+#        include <unistd.h> //sysconf
+#    endif
+#    if ANKERL_NANOBENCH(PERF_COUNTERS)
+#        include <map> // map
+
+#        include <linux/perf_event.h>
+#        include <sys/ioctl.h>
+#        include <sys/syscall.h>
+#        include <unistd.h>
+#    endif
+
+// declarations ///////////////////////////////////////////////////////////////////////////////////
+
+namespace ankerl {
+namespace nanobench {
+
+// helper stuff that is only intended to be used internally
+namespace detail {
+
+struct TableInfo;
+
+// formatting utilities
+namespace fmt {
+
+class NumSep;
+class StreamStateRestorer;
+class Number;
+class MarkDownColumn;
+class MarkDownCode;
+
+} // namespace fmt
+} // namespace detail
+} // namespace nanobench
+} // namespace ankerl
+
+// definitions ////////////////////////////////////////////////////////////////////////////////////
+
+namespace ankerl {
+namespace nanobench {
+
+uint64_t splitMix64(uint64_t& state) noexcept;
+
+namespace detail {
+
+// helpers to get double values
+template <typename T>
+inline double d(T t) noexcept {
+    return static_cast<double>(t);
+}
+inline double d(Clock::duration duration) noexcept {
+    return std::chrono::duration_cast<std::chrono::duration<double>>(duration).count();
+}
+
+// Calculates clock resolution once, and remembers the result
+inline Clock::duration clockResolution() noexcept;
+
+} // namespace detail
+
+namespace templates {
+
+char const* csv() noexcept {
+    return R"DELIM("title";"name";"unit";"batch";"elapsed";"error %";"instructions";"branches";"branch misses";"total"
+{{#result}}"{{title}}";"{{name}}";"{{unit}}";{{batch}};{{median(elapsed)}};{{medianAbsolutePercentError(elapsed)}};{{median(instructions)}};{{median(branchinstructions)}};{{median(branchmisses)}};{{sumProduct(iterations, elapsed)}}
+{{/result}})DELIM";
+}
+
+char const* htmlBoxplot() noexcept {
+    return R"DELIM(<html>
+
+<head>
+    <script src="https://cdn.plot.ly/plotly-latest.min.js"></script>
+</head>
+
+<body>
+    <div id="myDiv"></div>
+    <script>
+        var data = [
+            {{#result}}{
+                name: '{{name}}',
+                y: [{{#measurement}}{{elapsed}}{{^-last}}, {{/last}}{{/measurement}}],
+            },
+            {{/result}}
+        ];
+        var title = '{{title}}';
+
+        data = data.map(a => Object.assign(a, { boxpoints: 'all', pointpos: 0, type: 'box' }));
+        var layout = { title: { text: title }, showlegend: false, yaxis: { title: 'time per unit', rangemode: 'tozero', autorange: true } }; Plotly.newPlot('myDiv', data, layout, {responsive: true});
+    </script>
+</body>
+
+</html>)DELIM";
+}
+
+char const* json() noexcept {
+    return R"DELIM({
+    "results": [
+{{#result}}        {
+            "title": "{{title}}",
+            "name": "{{name}}",
+            "unit": "{{unit}}",
+            "batch": {{batch}},
+            "complexityN": {{complexityN}},
+            "epochs": {{epochs}},
+            "clockResolution": {{clockResolution}},
+            "clockResolutionMultiple": {{clockResolutionMultiple}},
+            "maxEpochTime": {{maxEpochTime}},
+            "minEpochTime": {{minEpochTime}},
+            "minEpochIterations": {{minEpochIterations}},
+            "epochIterations": {{epochIterations}},
+            "warmup": {{warmup}},
+            "relative": {{relative}},
+            "median(elapsed)": {{median(elapsed)}},
+            "medianAbsolutePercentError(elapsed)": {{medianAbsolutePercentError(elapsed)}},
+            "median(instructions)": {{median(instructions)}},
+            "medianAbsolutePercentError(instructions)": {{medianAbsolutePercentError(instructions)}},
+            "median(cpucycles)": {{median(cpucycles)}},
+            "median(contextswitches)": {{median(contextswitches)}},
+            "median(pagefaults)": {{median(pagefaults)}},
+            "median(branchinstructions)": {{median(branchinstructions)}},
+            "median(branchmisses)": {{median(branchmisses)}},
+            "totalTime": {{sumProduct(iterations, elapsed)}},
+            "measurements": [
+{{#measurement}}                {
+                    "iterations": {{iterations}},
+                    "elapsed": {{elapsed}},
+                    "pagefaults": {{pagefaults}},
+                    "cpucycles": {{cpucycles}},
+                    "contextswitches": {{contextswitches}},
+                    "instructions": {{instructions}},
+                    "branchinstructions": {{branchinstructions}},
+                    "branchmisses": {{branchmisses}}
+                }{{^-last}},{{/-last}}
+{{/measurement}}            ]
+        }{{^-last}},{{/-last}}
+{{/result}}    ]
+})DELIM";
+}
+
+ANKERL_NANOBENCH(IGNORE_PADDED_PUSH)
+struct Node {
+    enum class Type { tag, content, section, inverted_section };
+
+    char const* begin;
+    char const* end;
+    std::vector<Node> children;
+    Type type;
+
+    template <size_t N>
+    // NOLINTNEXTLINE(hicpp-avoid-c-arrays,modernize-avoid-c-arrays,cppcoreguidelines-avoid-c-arrays)
+    bool operator==(char const (&str)[N]) const noexcept {
+        return static_cast<size_t>(std::distance(begin, end) + 1) == N && 0 == strncmp(str, begin, N - 1);
+    }
+};
+ANKERL_NANOBENCH(IGNORE_PADDED_POP)
+
+static std::vector<Node> parseMustacheTemplate(char const** tpl) {
+    std::vector<Node> nodes;
+
+    while (true) {
+        auto begin = std::strstr(*tpl, "{{");
+        auto end = begin;
+        if (begin != nullptr) {
+            begin += 2;
+            end = std::strstr(begin, "}}");
+        }
+
+        if (begin == nullptr || end == nullptr) {
+            // nothing found, finish node
+            nodes.emplace_back(Node{*tpl, *tpl + std::strlen(*tpl), std::vector<Node>{}, Node::Type::content});
+            return nodes;
+        }
+
+        nodes.emplace_back(Node{*tpl, begin - 2, std::vector<Node>{}, Node::Type::content});
+
+        // we found a tag
+        *tpl = end + 2;
+        switch (*begin) {
+        case '/':
+            // finished! bail out
+            return nodes;
+
+        case '#':
+            nodes.emplace_back(Node{begin + 1, end, parseMustacheTemplate(tpl), Node::Type::section});
+            break;
+
+        case '^':
+            nodes.emplace_back(Node{begin + 1, end, parseMustacheTemplate(tpl), Node::Type::inverted_section});
+            break;
+
+        default:
+            nodes.emplace_back(Node{begin, end, std::vector<Node>{}, Node::Type::tag});
+            break;
+        }
+    }
+}
+
+static bool generateFirstLast(Node const& n, size_t idx, size_t size, std::ostream& out) {
+    bool matchFirst = n == "-first";
+    bool matchLast = n == "-last";
+    if (!matchFirst && !matchLast) {
+        return false;
+    }
+
+    bool doWrite = false;
+    if (n.type == Node::Type::section) {
+        doWrite = (matchFirst && idx == 0) || (matchLast && idx == size - 1);
+    } else if (n.type == Node::Type::inverted_section) {
+        doWrite = (matchFirst && idx != 0) || (matchLast && idx != size - 1);
+    }
+
+    if (doWrite) {
+        for (auto const& child : n.children) {
+            if (child.type == Node::Type::content) {
+                out.write(child.begin, std::distance(child.begin, child.end));
+            }
+        }
+    }
+    return true;
+}
+
+static bool matchCmdArgs(std::string const& str, std::vector<std::string>& matchResult) {
+    matchResult.clear();
+    auto idxOpen = str.find('(');
+    auto idxClose = str.find(')', idxOpen);
+    if (idxClose == std::string::npos) {
+        return false;
+    }
+
+    matchResult.emplace_back(str.substr(0, idxOpen));
+
+    // split by comma
+    matchResult.emplace_back(std::string{});
+    for (size_t i = idxOpen + 1; i != idxClose; ++i) {
+        if (str[i] == ' ' || str[i] == '\t') {
+            // skip whitespace
+            continue;
+        }
+        if (str[i] == ',') {
+            // got a comma => new string
+            matchResult.emplace_back(std::string{});
+            continue;
+        }
+        // no whitespace no comma, append
+        matchResult.back() += str[i];
+    }
+    return true;
+}
+
+static bool generateConfigTag(Node const& n, Config const& config, std::ostream& out) {
+    using detail::d;
+
+    if (n == "title") {
+        out << config.mBenchmarkTitle;
+        return true;
+    } else if (n == "name") {
+        out << config.mBenchmarkName;
+        return true;
+    } else if (n == "unit") {
+        out << config.mUnit;
+        return true;
+    } else if (n == "batch") {
+        out << config.mBatch;
+        return true;
+    } else if (n == "complexityN") {
+        out << config.mComplexityN;
+        return true;
+    } else if (n == "epochs") {
+        out << config.mNumEpochs;
+        return true;
+    } else if (n == "clockResolution") {
+        out << d(detail::clockResolution());
+        return true;
+    } else if (n == "clockResolutionMultiple") {
+        out << config.mClockResolutionMultiple;
+        return true;
+    } else if (n == "maxEpochTime") {
+        out << d(config.mMaxEpochTime);
+        return true;
+    } else if (n == "minEpochTime") {
+        out << d(config.mMinEpochTime);
+        return true;
+    } else if (n == "minEpochIterations") {
+        out << config.mMinEpochIterations;
+        return true;
+    } else if (n == "epochIterations") {
+        out << config.mEpochIterations;
+        return true;
+    } else if (n == "warmup") {
+        out << config.mWarmup;
+        return true;
+    } else if (n == "relative") {
+        out << config.mIsRelative;
+        return true;
+    }
+    return false;
+}
+
+static std::ostream& generateResultTag(Node const& n, Result const& r, std::ostream& out) {
+    if (generateConfigTag(n, r.config(), out)) {
+        return out;
+    }
+    // match e.g. "median(elapsed)"
+    // g++ 4.8 doesn't implement std::regex :(
+    // static std::regex const regOpArg1("^([a-zA-Z]+)\\(([a-zA-Z]*)\\)$");
+    // std::cmatch matchResult;
+    // if (std::regex_match(n.begin, n.end, matchResult, regOpArg1)) {
+    std::vector<std::string> matchResult;
+    if (matchCmdArgs(std::string(n.begin, n.end), matchResult)) {
+        if (matchResult.size() == 2) {
+            auto m = Result::fromString(matchResult[1]);
+            if (m == Result::Measure::_size) {
+                return out << 0.0;
+            }
+
+            if (matchResult[0] == "median") {
+                return out << r.median(m);
+            }
+            if (matchResult[0] == "average") {
+                return out << r.average(m);
+            }
+            if (matchResult[0] == "medianAbsolutePercentError") {
+                return out << r.medianAbsolutePercentError(m);
+            }
+            if (matchResult[0] == "sum") {
+                return out << r.sum(m);
+            }
+            if (matchResult[0] == "minimum") {
+                return out << r.minimum(m);
+            }
+            if (matchResult[0] == "maximum") {
+                return out << r.maximum(m);
+            }
+        } else if (matchResult.size() == 3) {
+            auto m1 = Result::fromString(matchResult[1]);
+            auto m2 = Result::fromString(matchResult[2]);
+            if (m1 == Result::Measure::_size || m2 == Result::Measure::_size) {
+                return out << 0.0;
+            }
+
+            if (matchResult[0] == "sumProduct") {
+                return out << r.sumProduct(m1, m2);
+            }
+        }
+    }
+
+    // match e.g. "sumProduct(elapsed, iterations)"
+    // static std::regex const regOpArg2("^([a-zA-Z]+)\\(([a-zA-Z]*)\\s*,\\s+([a-zA-Z]*)\\)$");
+
+    // nothing matches :(
+    throw std::runtime_error("command '" + std::string(n.begin, n.end) + "' not understood");
+}
+
+static void generateResultMeasurement(std::vector<Node> const& nodes, size_t idx, Result const& r, std::ostream& out) {
+    for (auto const& n : nodes) {
+        if (!generateFirstLast(n, idx, r.size(), out)) {
+            ANKERL_NANOBENCH_LOG("n.type=" << static_cast<int>(n.type));
+            switch (n.type) {
+            case Node::Type::content:
+                out.write(n.begin, std::distance(n.begin, n.end));
+                break;
+
+            case Node::Type::inverted_section:
+                throw std::runtime_error("got a inverted section inside measurement");
+
+            case Node::Type::section:
+                throw std::runtime_error("got a section inside measurement");
+
+            case Node::Type::tag: {
+                auto m = Result::fromString(std::string(n.begin, n.end));
+                if (m == Result::Measure::_size || !r.has(m)) {
+                    out << 0.0;
+                } else {
+                    out << r.get(idx, m);
+                }
+                break;
+            }
+            }
+        }
+    }
+}
+
+static void generateResult(std::vector<Node> const& nodes, size_t idx, std::vector<Result> const& results, std::ostream& out) {
+    auto const& r = results[idx];
+    for (auto const& n : nodes) {
+        if (!generateFirstLast(n, idx, results.size(), out)) {
+            ANKERL_NANOBENCH_LOG("n.type=" << static_cast<int>(n.type));
+            switch (n.type) {
+            case Node::Type::content:
+                out.write(n.begin, std::distance(n.begin, n.end));
+                break;
+
+            case Node::Type::inverted_section:
+                throw std::runtime_error("got a inverted section inside result");
+
+            case Node::Type::section:
+                if (n == "measurement") {
+                    for (size_t i = 0; i < r.size(); ++i) {
+                        generateResultMeasurement(n.children, i, r, out);
+                    }
+                } else {
+                    throw std::runtime_error("got a section inside result");
+                }
+                break;
+
+            case Node::Type::tag:
+                generateResultTag(n, r, out);
+                break;
+            }
+        }
+    }
+}
+
+} // namespace templates
+
+// helper stuff that only intended to be used internally
+namespace detail {
+
+char const* getEnv(char const* name);
+bool isEndlessRunning(std::string const& name);
+
+template <typename T>
+T parseFile(std::string const& filename);
+
+void gatherStabilityInformation(std::vector<std::string>& warnings, std::vector<std::string>& recommendations);
+void printStabilityInformationOnce(std::ostream* os);
+
+// remembers the last table settings used. When it changes, a new table header is automatically written for the new entry.
+uint64_t& singletonHeaderHash() noexcept;
+
+// determines resolution of the given clock. This is done by measuring multiple times and returning the minimum time difference.
+Clock::duration calcClockResolution(size_t numEvaluations) noexcept;
+
+// formatting utilities
+namespace fmt {
+
+// adds thousands separator to numbers
+ANKERL_NANOBENCH(IGNORE_PADDED_PUSH)
+class NumSep : public std::numpunct<char> {
+public:
+    explicit NumSep(char sep);
+    char do_thousands_sep() const override;
+    std::string do_grouping() const override;
+
+private:
+    char mSep;
+};
+ANKERL_NANOBENCH(IGNORE_PADDED_POP)
+
+// RAII to save & restore a stream's state
+ANKERL_NANOBENCH(IGNORE_PADDED_PUSH)
+class StreamStateRestorer {
+public:
+    explicit StreamStateRestorer(std::ostream& s);
+    ~StreamStateRestorer();
+
+    // sets back all stream info that we remembered at construction
+    void restore();
+
+    // don't allow copying / moving
+    StreamStateRestorer(StreamStateRestorer const&) = delete;
+    StreamStateRestorer& operator=(StreamStateRestorer const&) = delete;
+    StreamStateRestorer(StreamStateRestorer&&) = delete;
+    StreamStateRestorer& operator=(StreamStateRestorer&&) = delete;
+
+private:
+    std::ostream& mStream;
+    std::locale mLocale;
+    std::streamsize const mPrecision;
+    std::streamsize const mWidth;
+    std::ostream::char_type const mFill;
+    std::ostream::fmtflags const mFmtFlags;
+};
+ANKERL_NANOBENCH(IGNORE_PADDED_POP)
+
+// Number formatter
+class Number {
+public:
+    Number(int width, int precision, double value);
+    Number(int width, int precision, int64_t value);
+    std::string to_s() const;
+
+private:
+    friend std::ostream& operator<<(std::ostream& os, Number const& n);
+    std::ostream& write(std::ostream& os) const;
+
+    int mWidth;
+    int mPrecision;
+    double mValue;
+};
+
+// helper replacement for std::to_string of signed/unsigned numbers so we are locale independent
+std::string to_s(uint64_t s);
+
+std::ostream& operator<<(std::ostream& os, Number const& n);
+
+class MarkDownColumn {
+public:
+    MarkDownColumn(int w, int prec, std::string const& tit, std::string const& suff, double val);
+    std::string title() const;
+    std::string separator() const;
+    std::string invalid() const;
+    std::string value() const;
+
+private:
+    int mWidth;
+    int mPrecision;
+    std::string mTitle;
+    std::string mSuffix;
+    double mValue;
+};
+
+// Formats any text as markdown code, escaping backticks.
+class MarkDownCode {
+public:
+    explicit MarkDownCode(std::string const& what);
+
+private:
+    friend std::ostream& operator<<(std::ostream& os, MarkDownCode const& mdCode);
+    std::ostream& write(std::ostream& os) const;
+
+    std::string mWhat{};
+};
+
+std::ostream& operator<<(std::ostream& os, MarkDownCode const& mdCode);
+
+} // namespace fmt
+} // namespace detail
+} // namespace nanobench
+} // namespace ankerl
+
+// implementation /////////////////////////////////////////////////////////////////////////////////
+
+namespace ankerl {
+namespace nanobench {
+
+void render(char const* mustacheTemplate, std::vector<Result> const& results, std::ostream& out) {
+    detail::fmt::StreamStateRestorer restorer(out);
+
+    out.precision(std::numeric_limits<double>::digits10);
+    auto nodes = templates::parseMustacheTemplate(&mustacheTemplate);
+
+    for (auto const& n : nodes) {
+        ANKERL_NANOBENCH_LOG("n.type=" << static_cast<int>(n.type));
+        switch (n.type) {
+        case templates::Node::Type::content:
+            out.write(n.begin, std::distance(n.begin, n.end));
+            break;
+
+        case templates::Node::Type::inverted_section:
+            throw std::runtime_error("unknown list '" + std::string(n.begin, n.end) + "'");
+
+        case templates::Node::Type::section:
+            if (n == "result") {
+                const size_t nbResults = results.size();
+                for (size_t i = 0; i < nbResults; ++i) {
+                    generateResult(n.children, i, results, out);
+                }
+            } else {
+                throw std::runtime_error("unknown section '" + std::string(n.begin, n.end) + "'");
+            }
+            break;
+
+        case templates::Node::Type::tag:
+            // This just uses the last result's config.
+            if (!generateConfigTag(n, results.back().config(), out)) {
+                throw std::runtime_error("unknown tag '" + std::string(n.begin, n.end) + "'");
+            }
+            break;
+        }
+    }
+}
+
+void render(char const* mustacheTemplate, const Bench& bench, std::ostream& out) {
+    render(mustacheTemplate, bench.results(), out);
+}
+
+namespace detail {
+
+PerformanceCounters& performanceCounters() {
+#    if defined(__clang__)
+#        pragma clang diagnostic push
+#        pragma clang diagnostic ignored "-Wexit-time-destructors"
+#    endif
+    static PerformanceCounters pc;
+#    if defined(__clang__)
+#        pragma clang diagnostic pop
+#    endif
+    return pc;
+}
+
+// Windows version of doNotOptimizeAway
+// see https://github.com/google/benchmark/blob/master/include/benchmark/benchmark.h#L307
+// see https://github.com/facebook/folly/blob/master/folly/Benchmark.h#L280
+// see https://docs.microsoft.com/en-us/cpp/preprocessor/optimize
+#    if defined(_MSC_VER)
+#        pragma optimize("", off)
+void doNotOptimizeAwaySink(void const*) {}
+#        pragma optimize("", on)
+#    endif
+
+template <typename T>
+T parseFile(std::string const& filename) {
+    std::ifstream fin(filename);
+    T num{};
+    fin >> num;
+    return num;
+}
+
+char const* getEnv(char const* name) {
+#    if defined(_MSC_VER)
+#        pragma warning(push)
+#        pragma warning(disable : 4996) // getenv': This function or variable may be unsafe.
+#    endif
+    return std::getenv(name);
+#    if defined(_MSC_VER)
+#        pragma warning(pop)
+#    endif
+}
+
+bool isEndlessRunning(std::string const& name) {
+    auto endless = getEnv("NANOBENCH_ENDLESS");
+    return nullptr != endless && endless == name;
+}
+
+void gatherStabilityInformation(std::vector<std::string>& warnings, std::vector<std::string>& recommendations) {
+    warnings.clear();
+    recommendations.clear();
+
+    bool recommendCheckFlags = false;
+
+#    if defined(DEBUG)
+    warnings.emplace_back("DEBUG defined");
+    recommendCheckFlags = true;
+#    endif
+
+    bool recommendPyPerf = false;
+#    if defined(__linux__)
+    auto nprocs = sysconf(_SC_NPROCESSORS_CONF);
+    if (nprocs <= 0) {
+        warnings.emplace_back("couldn't figure out number of processors - no governor, turbo check possible");
+    } else {
+
+        // check frequency scaling
+        for (long id = 0; id < nprocs; ++id) {
+            auto idStr = detail::fmt::to_s(static_cast<uint64_t>(id));
+            auto sysCpu = "/sys/devices/system/cpu/cpu" + idStr;
+            auto minFreq = parseFile<int64_t>(sysCpu + "/cpufreq/scaling_min_freq");
+            auto maxFreq = parseFile<int64_t>(sysCpu + "/cpufreq/scaling_max_freq");
+            if (minFreq != maxFreq) {
+                auto minMHz = static_cast<double>(minFreq) / 1000.0;
+                auto maxMHz = static_cast<double>(maxFreq) / 1000.0;
+                warnings.emplace_back("CPU frequency scaling enabled: CPU " + idStr + " between " +
+                                      detail::fmt::Number(1, 1, minMHz).to_s() + " and " + detail::fmt::Number(1, 1, maxMHz).to_s() +
+                                      " MHz");
+                recommendPyPerf = true;
+                break;
+            }
+        }
+
+        auto currentGovernor = parseFile<std::string>("/sys/devices/system/cpu/cpu0/cpufreq/scaling_governor");
+        if ("performance" != currentGovernor) {
+            warnings.emplace_back("CPU governor is '" + currentGovernor + "' but should be 'performance'");
+            recommendPyPerf = true;
+        }
+
+        if (0 == parseFile<int>("/sys/devices/system/cpu/intel_pstate/no_turbo")) {
+            warnings.emplace_back("Turbo is enabled, CPU frequency will fluctuate");
+            recommendPyPerf = true;
+        }
+    }
+#    endif
+
+    if (recommendCheckFlags) {
+        recommendations.emplace_back("Make sure you compile for Release");
+    }
+    if (recommendPyPerf) {
+        recommendations.emplace_back("Use 'pyperf system tune' before benchmarking. See https://github.com/vstinner/pyperf");
+    }
+}
+
+void printStabilityInformationOnce(std::ostream* outStream) {
+    static bool shouldPrint = true;
+    if (shouldPrint && outStream) {
+        auto& os = *outStream;
+        shouldPrint = false;
+        std::vector<std::string> warnings;
+        std::vector<std::string> recommendations;
+        gatherStabilityInformation(warnings, recommendations);
+        if (warnings.empty()) {
+            return;
+        }
+
+        os << "Warning, results might be unstable:" << std::endl;
+        for (auto const& w : warnings) {
+            os << "* " << w << std::endl;
+        }
+
+        os << std::endl << "Recommendations" << std::endl;
+        for (auto const& r : recommendations) {
+            os << "* " << r << std::endl;
+        }
+    }
+}
+
+// remembers the last table settings used. When it changes, a new table header is automatically written for the new entry.
+uint64_t& singletonHeaderHash() noexcept {
+    static uint64_t sHeaderHash{};
+    return sHeaderHash;
+}
+
+ANKERL_NANOBENCH_NO_SANITIZE("integer")
+inline uint64_t fnv1a(std::string const& str) noexcept {
+    auto val = UINT64_C(14695981039346656037);
+    for (auto c : str) {
+        val = (val ^ static_cast<uint8_t>(c)) * UINT64_C(1099511628211);
+    }
+    return val;
+}
+
+ANKERL_NANOBENCH_NO_SANITIZE("integer")
+inline uint64_t hash_combine(uint64_t seed, uint64_t val) {
+    return seed ^ (val + UINT64_C(0x9e3779b9) + (seed << 6U) + (seed >> 2U));
+}
+
+// determines resolution of the given clock. This is done by measuring multiple times and returning the minimum time difference.
+Clock::duration calcClockResolution(size_t numEvaluations) noexcept {
+    auto bestDuration = Clock::duration::max();
+    Clock::time_point tBegin;
+    Clock::time_point tEnd;
+    for (size_t i = 0; i < numEvaluations; ++i) {
+        tBegin = Clock::now();
+        do {
+            tEnd = Clock::now();
+        } while (tBegin == tEnd);
+        bestDuration = (std::min)(bestDuration, tEnd - tBegin);
+    }
+    return bestDuration;
+}
+
+// Calculates clock resolution once, and remembers the result
+Clock::duration clockResolution() noexcept {
+    static Clock::duration sResolution = calcClockResolution(20);
+    return sResolution;
+}
+
+ANKERL_NANOBENCH(IGNORE_PADDED_PUSH)
+struct IterationLogic::Impl {
+    enum class State { warmup, upscaling_runtime, measuring, endless };
+
+    explicit Impl(Bench const& bench)
+        : mBench(bench)
+        , mResult(bench.config()) {
+        printStabilityInformationOnce(mBench.output());
+
+        // determine target runtime per epoch
+        mTargetRuntimePerEpoch = detail::clockResolution() * mBench.clockResolutionMultiple();
+        if (mTargetRuntimePerEpoch > mBench.maxEpochTime()) {
+            mTargetRuntimePerEpoch = mBench.maxEpochTime();
+        }
+        if (mTargetRuntimePerEpoch < mBench.minEpochTime()) {
+            mTargetRuntimePerEpoch = mBench.minEpochTime();
+        }
+
+        if (isEndlessRunning(mBench.name())) {
+            std::cerr << "NANOBENCH_ENDLESS set: running '" << mBench.name() << "' endlessly" << std::endl;
+            mNumIters = (std::numeric_limits<uint64_t>::max)();
+            mState = State::endless;
+        } else if (0 != mBench.warmup()) {
+            mNumIters = mBench.warmup();
+            mState = State::warmup;
+        } else if (0 != mBench.epochIterations()) {
+            // exact number of iterations
+            mNumIters = mBench.epochIterations();
+            mState = State::measuring;
+        } else {
+            mNumIters = mBench.minEpochIterations();
+            mState = State::upscaling_runtime;
+        }
+    }
+
+    // directly calculates new iters based on elapsed&iters, and adds a 10% noise. Makes sure we don't underflow.
+    ANKERL_NANOBENCH(NODISCARD) uint64_t calcBestNumIters(std::chrono::nanoseconds elapsed, uint64_t iters) noexcept {
+        auto doubleElapsed = d(elapsed);
+        auto doubleTargetRuntimePerEpoch = d(mTargetRuntimePerEpoch);
+        auto doubleNewIters = doubleTargetRuntimePerEpoch / doubleElapsed * d(iters);
+
+        auto doubleMinEpochIters = d(mBench.minEpochIterations());
+        if (doubleNewIters < doubleMinEpochIters) {
+            doubleNewIters = doubleMinEpochIters;
+        }
+        doubleNewIters *= 1.0 + 0.2 * mRng.uniform01();
+
+        // +0.5 for correct rounding when casting
+        // NOLINTNEXTLINE(bugprone-incorrect-roundings)
+        return static_cast<uint64_t>(doubleNewIters + 0.5);
+    }
+
+    ANKERL_NANOBENCH_NO_SANITIZE("integer") void upscale(std::chrono::nanoseconds elapsed) {
+        if (elapsed * 10 < mTargetRuntimePerEpoch) {
+            // we are far below the target runtime. Multiply iterations by 10 (with overflow check)
+            if (mNumIters * 10 < mNumIters) {
+                // overflow :-(
+                showResult("iterations overflow. Maybe your code got optimized away?");
+                mNumIters = 0;
+                return;
+            }
+            mNumIters *= 10;
+        } else {
+            mNumIters = calcBestNumIters(elapsed, mNumIters);
+        }
+    }
+
+    void add(std::chrono::nanoseconds elapsed, PerformanceCounters const& pc) noexcept {
+#    if defined(ANKERL_NANOBENCH_LOG_ENABLED)
+        auto oldIters = mNumIters;
+#    endif
+
+        switch (mState) {
+        case State::warmup:
+            if (isCloseEnoughForMeasurements(elapsed)) {
+                // if elapsed is close enough, we can skip upscaling and go right to measurements
+                // still, we don't add the result to the measurements.
+                mState = State::measuring;
+                mNumIters = calcBestNumIters(elapsed, mNumIters);
+            } else {
+                // not close enough: switch to upscaling
+                mState = State::upscaling_runtime;
+                upscale(elapsed);
+            }
+            break;
+
+        case State::upscaling_runtime:
+            if (isCloseEnoughForMeasurements(elapsed)) {
+                // if we are close enough, add measurement and switch to always measuring
+                mState = State::measuring;
+                mTotalElapsed += elapsed;
+                mTotalNumIters += mNumIters;
+                mResult.add(elapsed, mNumIters, pc);
+                mNumIters = calcBestNumIters(mTotalElapsed, mTotalNumIters);
+            } else {
+                upscale(elapsed);
+            }
+            break;
+
+        case State::measuring:
+            // just add measurements - no questions asked. Even when runtime is low. But we can't ignore
+            // that fluctuation, or else we would bias the result
+            mTotalElapsed += elapsed;
+            mTotalNumIters += mNumIters;
+            mResult.add(elapsed, mNumIters, pc);
+            if (0 != mBench.epochIterations()) {
+                mNumIters = mBench.epochIterations();
+            } else {
+                mNumIters = calcBestNumIters(mTotalElapsed, mTotalNumIters);
+            }
+            break;
+
+        case State::endless:
+            mNumIters = (std::numeric_limits<uint64_t>::max)();
+            break;
+        }
+
+        if (static_cast<uint64_t>(mResult.size()) == mBench.epochs()) {
+            // we got all the results that we need, finish it
+            showResult("");
+            mNumIters = 0;
+        }
+
+        ANKERL_NANOBENCH_LOG(mBench.name() << ": " << detail::fmt::Number(20, 3, static_cast<double>(elapsed.count())) << " elapsed, "
+                                           << detail::fmt::Number(20, 3, static_cast<double>(mTargetRuntimePerEpoch.count()))
+                                           << " target. oldIters=" << oldIters << ", mNumIters=" << mNumIters
+                                           << ", mState=" << static_cast<int>(mState));
+    }
+
+    void showResult(std::string const& errorMessage) const {
+        ANKERL_NANOBENCH_LOG(errorMessage);
+
+        if (mBench.output() != nullptr) {
+            // prepare column data ///////
+            std::vector<fmt::MarkDownColumn> columns;
+
+            auto rMedian = mResult.median(Result::Measure::elapsed);
+
+            if (mBench.relative()) {
+                double d = 100.0;
+                if (!mBench.results().empty()) {
+                    d = rMedian <= 0.0 ? 0.0 : mBench.results().front().median(Result::Measure::elapsed) / rMedian * 100.0;
+                }
+                columns.emplace_back(11, 1, "relative", "%", d);
+            }
+
+            if (mBench.complexityN() > 0) {
+                columns.emplace_back(14, 0, "complexityN", "", mBench.complexityN());
+            }
+
+            columns.emplace_back(22, 2, "ns/" + mBench.unit(), "", 1e9 * rMedian / mBench.batch());
+            columns.emplace_back(22, 2, mBench.unit() + "/s", "", rMedian <= 0.0 ? 0.0 : mBench.batch() / rMedian);
+
+            double rErrorMedian = mResult.medianAbsolutePercentError(Result::Measure::elapsed);
+            columns.emplace_back(10, 1, "err%", "%", rErrorMedian * 100.0);
+
+            double rInsMedian = -1.0;
+            if (mResult.has(Result::Measure::instructions)) {
+                rInsMedian = mResult.median(Result::Measure::instructions);
+                columns.emplace_back(18, 2, "ins/" + mBench.unit(), "", rInsMedian / mBench.batch());
+            }
+
+            double rCycMedian = -1.0;
+            if (mResult.has(Result::Measure::cpucycles)) {
+                rCycMedian = mResult.median(Result::Measure::cpucycles);
+                columns.emplace_back(18, 2, "cyc/" + mBench.unit(), "", rCycMedian / mBench.batch());
+            }
+            if (rInsMedian > 0.0 && rCycMedian > 0.0) {
+                columns.emplace_back(9, 3, "IPC", "", rCycMedian <= 0.0 ? 0.0 : rInsMedian / rCycMedian);
+            }
+            if (mResult.has(Result::Measure::branchinstructions)) {
+                double rBraMedian = mResult.median(Result::Measure::branchinstructions);
+                columns.emplace_back(17, 2, "bra/" + mBench.unit(), "", rBraMedian / mBench.batch());
+                if (mResult.has(Result::Measure::branchmisses)) {
+                    double p = 0.0;
+                    if (rBraMedian >= 1e-9) {
+                        p = 100.0 * mResult.median(Result::Measure::branchmisses) / rBraMedian;
+                    }
+                    columns.emplace_back(10, 1, "miss%", "%", p);
+                }
+            }
+
+            columns.emplace_back(12, 2, "total", "", mResult.sum(Result::Measure::elapsed));
+
+            // write everything
+            auto& os = *mBench.output();
+
+            uint64_t hash = 0;
+            hash = hash_combine(fnv1a(mBench.unit()), hash);
+            hash = hash_combine(fnv1a(mBench.title()), hash);
+            hash = hash_combine(mBench.relative(), hash);
+            hash = hash_combine(mBench.performanceCounters(), hash);
+
+            if (hash != singletonHeaderHash()) {
+                singletonHeaderHash() = hash;
+
+                // no result yet, print header
+                os << std::endl;
+                for (auto const& col : columns) {
+                    os << col.title();
+                }
+                os << "| " << mBench.title() << std::endl;
+
+                for (auto const& col : columns) {
+                    os << col.separator();
+                }
+                os << "|:" << std::string(mBench.title().size() + 1U, '-') << std::endl;
+            }
+
+            if (!errorMessage.empty()) {
+                for (auto const& col : columns) {
+                    os << col.invalid();
+                }
+                os << "| :boom: " << fmt::MarkDownCode(mBench.name()) << " (" << errorMessage << ')' << std::endl;
+            } else {
+                for (auto const& col : columns) {
+                    os << col.value();
+                }
+                os << "| ";
+                auto showUnstable = rErrorMedian >= 0.05;
+                if (showUnstable) {
+                    os << ":wavy_dash: ";
+                }
+                os << fmt::MarkDownCode(mBench.name());
+                if (showUnstable) {
+                    auto avgIters = static_cast<double>(mTotalNumIters) / static_cast<double>(mBench.epochs());
+                    // NOLINTNEXTLINE(bugprone-incorrect-roundings)
+                    auto suggestedIters = static_cast<uint64_t>(avgIters * 10 + 0.5);
+
+                    os << " (Unstable with ~" << detail::fmt::Number(1, 1, avgIters)
+                       << " iters. Increase `minEpochIterations` to e.g. " << suggestedIters << ")";
+                }
+                os << std::endl;
+            }
+        }
+    }
+
+    ANKERL_NANOBENCH(NODISCARD) bool isCloseEnoughForMeasurements(std::chrono::nanoseconds elapsed) const noexcept {
+        return elapsed * 3 >= mTargetRuntimePerEpoch * 2;
+    }
+
+    uint64_t mNumIters = 1;
+    Bench const& mBench;
+    std::chrono::nanoseconds mTargetRuntimePerEpoch{};
+    Result mResult;
+    Rng mRng{123};
+    std::chrono::nanoseconds mTotalElapsed{};
+    uint64_t mTotalNumIters = 0;
+
+    State mState = State::upscaling_runtime;
+};
+ANKERL_NANOBENCH(IGNORE_PADDED_POP)
+
+IterationLogic::IterationLogic(Bench const& bench) noexcept
+    : mPimpl(new Impl(bench)) {}
+
+IterationLogic::~IterationLogic() {
+    if (mPimpl) {
+        delete mPimpl;
+    }
+}
+
+uint64_t IterationLogic::numIters() const noexcept {
+    ANKERL_NANOBENCH_LOG(mPimpl->mBench.name() << ": mNumIters=" << mPimpl->mNumIters);
+    return mPimpl->mNumIters;
+}
+
+void IterationLogic::add(std::chrono::nanoseconds elapsed, PerformanceCounters const& pc) noexcept {
+    mPimpl->add(elapsed, pc);
+}
+
+void IterationLogic::moveResultTo(std::vector<Result>& results) noexcept {
+    results.emplace_back(std::move(mPimpl->mResult));
+}
+
+#    if ANKERL_NANOBENCH(PERF_COUNTERS)
+
+ANKERL_NANOBENCH(IGNORE_PADDED_PUSH)
+class LinuxPerformanceCounters {
+public:
+    struct Target {
+        Target(uint64_t* targetValue_, bool correctMeasuringOverhead_, bool correctLoopOverhead_)
+            : targetValue(targetValue_)
+            , correctMeasuringOverhead(correctMeasuringOverhead_)
+            , correctLoopOverhead(correctLoopOverhead_) {}
+
+        uint64_t* targetValue{};
+        bool correctMeasuringOverhead{};
+        bool correctLoopOverhead{};
+    };
+
+    ~LinuxPerformanceCounters();
+
+    // quick operation
+    inline void start() {}
+
+    inline void stop() {}
+
+    bool monitor(perf_sw_ids swId, Target target);
+    bool monitor(perf_hw_id hwId, Target target);
+
+    bool hasError() const noexcept {
+        return mHasError;
+    }
+
+    // Just reading data is faster than enable & disabling.
+    // we subtract data ourselves.
+    inline void beginMeasure() {
+        if (mHasError) {
+            return;
+        }
+
+        // NOLINTNEXTLINE(hicpp-signed-bitwise)
+        mHasError = -1 == ioctl(mFd, PERF_EVENT_IOC_RESET, PERF_IOC_FLAG_GROUP);
+        if (mHasError) {
+            return;
+        }
+
+        // NOLINTNEXTLINE(hicpp-signed-bitwise)
+        mHasError = -1 == ioctl(mFd, PERF_EVENT_IOC_ENABLE, PERF_IOC_FLAG_GROUP);
+    }
+
+    inline void endMeasure() {
+        if (mHasError) {
+            return;
+        }
+
+        // NOLINTNEXTLINE(hicpp-signed-bitwise)
+        mHasError = (-1 == ioctl(mFd, PERF_EVENT_IOC_DISABLE, PERF_IOC_FLAG_GROUP));
+        if (mHasError) {
+            return;
+        }
+
+        auto const numBytes = sizeof(uint64_t) * mCounters.size();
+        auto ret = read(mFd, mCounters.data(), numBytes);
+        mHasError = ret != static_cast<ssize_t>(numBytes);
+    }
+
+    void updateResults(uint64_t numIters);
+
+    // rounded integer division
+    template <typename T>
+    static inline T divRounded(T a, T divisor) {
+        return (a + divisor / 2) / divisor;
+    }
+
+    template <typename Op>
+    ANKERL_NANOBENCH_NO_SANITIZE("integer")
+    void calibrate(Op&& op) {
+        // clear current calibration data,
+        for (auto& v : mCalibratedOverhead) {
+            v = UINT64_C(0);
+        }
+
+        // create new calibration data
+        auto newCalibration = mCalibratedOverhead;
+        for (auto& v : newCalibration) {
+            v = (std::numeric_limits<uint64_t>::max)();
+        }
+        for (size_t iter = 0; iter < 100; ++iter) {
+            beginMeasure();
+            op();
+            endMeasure();
+            if (mHasError) {
+                return;
+            }
+
+            for (size_t i = 0; i < newCalibration.size(); ++i) {
+                auto diff = mCounters[i];
+                if (newCalibration[i] > diff) {
+                    newCalibration[i] = diff;
+                }
+            }
+        }
+
+        mCalibratedOverhead = std::move(newCalibration);
+
+        {
+            // calibrate loop overhead. For branches & instructions this makes sense, not so much for everything else like cycles.
+            // marsaglia's xorshift: mov, sal/shr, xor. Times 3.
+            // This has the nice property that the compiler doesn't seem to be able to optimize multiple calls any further.
+            // see https://godbolt.org/z/49RVQ5
+            uint64_t const numIters = 100000U + (std::random_device{}() & 3);
+            uint64_t n = numIters;
+            uint32_t x = 1234567;
+            auto fn = [&]() {
+                x ^= x << 13;
+                x ^= x >> 17;
+                x ^= x << 5;
+            };
+
+            beginMeasure();
+            while (n-- > 0) {
+                fn();
+            }
+            endMeasure();
+            detail::doNotOptimizeAway(x);
+            auto measure1 = mCounters;
+
+            n = numIters;
+            beginMeasure();
+            while (n-- > 0) {
+                // we now run *twice* so we can easily calculate the overhead
+                fn();
+                fn();
+            }
+            endMeasure();
+            detail::doNotOptimizeAway(x);
+            auto measure2 = mCounters;
+
+            for (size_t i = 0; i < mCounters.size(); ++i) {
+                // factor 2 because we have two instructions per loop
+                auto m1 = measure1[i] > mCalibratedOverhead[i] ? measure1[i] - mCalibratedOverhead[i] : 0;
+                auto m2 = measure2[i] > mCalibratedOverhead[i] ? measure2[i] - mCalibratedOverhead[i] : 0;
+                auto overhead = m1 * 2 > m2 ? m1 * 2 - m2 : 0;
+
+                mLoopOverhead[i] = divRounded(overhead, numIters);
+            }
+        }
+    }
+
+private:
+    bool monitor(uint32_t type, uint64_t eventid, Target target);
+
+    std::map<uint64_t, Target> mIdToTarget{};
+
+    // start with minimum size of 3 for read_format
+    std::vector<uint64_t> mCounters{3};
+    std::vector<uint64_t> mCalibratedOverhead{3};
+    std::vector<uint64_t> mLoopOverhead{3};
+
+    uint64_t mTimeEnabledNanos = 0;
+    uint64_t mTimeRunningNanos = 0;
+    int mFd = -1;
+    bool mHasError = false;
+};
+ANKERL_NANOBENCH(IGNORE_PADDED_POP)
+
+LinuxPerformanceCounters::~LinuxPerformanceCounters() {
+    if (-1 != mFd) {
+        close(mFd);
+    }
+}
+
+bool LinuxPerformanceCounters::monitor(perf_sw_ids swId, LinuxPerformanceCounters::Target target) {
+    return monitor(PERF_TYPE_SOFTWARE, swId, target);
+}
+
+bool LinuxPerformanceCounters::monitor(perf_hw_id hwId, LinuxPerformanceCounters::Target target) {
+    return monitor(PERF_TYPE_HARDWARE, hwId, target);
+}
+
+// overflow is ok, it's checked
+ANKERL_NANOBENCH_NO_SANITIZE("integer")
+void LinuxPerformanceCounters::updateResults(uint64_t numIters) {
+    // clear old data
+    for (auto& id_value : mIdToTarget) {
+        *id_value.second.targetValue = UINT64_C(0);
+    }
+
+    if (mHasError) {
+        return;
+    }
+
+    mTimeEnabledNanos = mCounters[1] - mCalibratedOverhead[1];
+    mTimeRunningNanos = mCounters[2] - mCalibratedOverhead[2];
+
+    for (uint64_t i = 0; i < mCounters[0]; ++i) {
+        auto idx = static_cast<size_t>(3 + i * 2 + 0);
+        auto id = mCounters[idx + 1U];
+
+        auto it = mIdToTarget.find(id);
+        if (it != mIdToTarget.end()) {
+
+            auto& tgt = it->second;
+            *tgt.targetValue = mCounters[idx];
+            if (tgt.correctMeasuringOverhead) {
+                if (*tgt.targetValue >= mCalibratedOverhead[idx]) {
+                    *tgt.targetValue -= mCalibratedOverhead[idx];
+                } else {
+                    *tgt.targetValue = 0U;
+                }
+            }
+            if (tgt.correctLoopOverhead) {
+                auto correctionVal = mLoopOverhead[idx] * numIters;
+                if (*tgt.targetValue >= correctionVal) {
+                    *tgt.targetValue -= correctionVal;
+                } else {
+                    *tgt.targetValue = 0U;
+                }
+            }
+        }
+    }
+}
+
+bool LinuxPerformanceCounters::monitor(uint32_t type, uint64_t eventid, Target target) {
+    *target.targetValue = (std::numeric_limits<uint64_t>::max)();
+    if (mHasError) {
+        return false;
+    }
+
+    auto pea = perf_event_attr();
+    std::memset(&pea, 0, sizeof(perf_event_attr));
+    pea.type = type;
+    pea.size = sizeof(perf_event_attr);
+    pea.config = eventid;
+    pea.disabled = 1; // start counter as disabled
+    pea.exclude_kernel = 1;
+    pea.exclude_hv = 1;
+
+    // NOLINTNEXTLINE(hicpp-signed-bitwise)
+    pea.read_format = PERF_FORMAT_GROUP | PERF_FORMAT_ID | PERF_FORMAT_TOTAL_TIME_ENABLED | PERF_FORMAT_TOTAL_TIME_RUNNING;
+
+    const int pid = 0;                    // the current process
+    const int cpu = -1;                   // all CPUs
+#        if defined(PERF_FLAG_FD_CLOEXEC) // since Linux 3.14
+    const unsigned long flags = PERF_FLAG_FD_CLOEXEC;
+#        else
+    const unsigned long flags = 0;
+#        endif
+
+    auto fd = static_cast<int>(syscall(__NR_perf_event_open, &pea, pid, cpu, mFd, flags));
+    if (-1 == fd) {
+        return false;
+    }
+    if (-1 == mFd) {
+        // first call: set to fd, and use this from now on
+        mFd = fd;
+    }
+    uint64_t id = 0;
+    // NOLINTNEXTLINE(hicpp-signed-bitwise)
+    if (-1 == ioctl(fd, PERF_EVENT_IOC_ID, &id)) {
+        // couldn't get id
+        return false;
+    }
+
+    // insert into map, rely on the fact that map's references are constant.
+    mIdToTarget.emplace(id, target);
+
+    // prepare readformat with the correct size (after the insert)
+    auto size = 3 + 2 * mIdToTarget.size();
+    mCounters.resize(size);
+    mCalibratedOverhead.resize(size);
+    mLoopOverhead.resize(size);
+
+    return true;
+}
+
+PerformanceCounters::PerformanceCounters()
+    : mPc(new LinuxPerformanceCounters())
+    , mVal()
+    , mHas() {
+
+    mHas.pageFaults = mPc->monitor(PERF_COUNT_SW_PAGE_FAULTS, LinuxPerformanceCounters::Target(&mVal.pageFaults, true, false));
+    mHas.cpuCycles = mPc->monitor(PERF_COUNT_HW_REF_CPU_CYCLES, LinuxPerformanceCounters::Target(&mVal.cpuCycles, true, false));
+    mHas.contextSwitches =
+        mPc->monitor(PERF_COUNT_SW_CONTEXT_SWITCHES, LinuxPerformanceCounters::Target(&mVal.contextSwitches, true, false));
+    mHas.instructions = mPc->monitor(PERF_COUNT_HW_INSTRUCTIONS, LinuxPerformanceCounters::Target(&mVal.instructions, true, true));
+    mHas.branchInstructions =
+        mPc->monitor(PERF_COUNT_HW_BRANCH_INSTRUCTIONS, LinuxPerformanceCounters::Target(&mVal.branchInstructions, true, false));
+    mHas.branchMisses = mPc->monitor(PERF_COUNT_HW_BRANCH_MISSES, LinuxPerformanceCounters::Target(&mVal.branchMisses, true, false));
+    // mHas.branchMisses = false;
+
+    mPc->start();
+    mPc->calibrate([] {
+        auto before = ankerl::nanobench::Clock::now();
+        auto after = ankerl::nanobench::Clock::now();
+        (void)before;
+        (void)after;
+    });
+
+    if (mPc->hasError()) {
+        // something failed, don't monitor anything.
+        mHas = PerfCountSet<bool>{};
+    }
+}
+
+PerformanceCounters::~PerformanceCounters() {
+    if (nullptr != mPc) {
+        delete mPc;
+    }
+}
+
+void PerformanceCounters::beginMeasure() {
+    mPc->beginMeasure();
+}
+
+void PerformanceCounters::endMeasure() {
+    mPc->endMeasure();
+}
+
+void PerformanceCounters::updateResults(uint64_t numIters) {
+    mPc->updateResults(numIters);
+}
+
+#    else
+
+PerformanceCounters::PerformanceCounters() = default;
+PerformanceCounters::~PerformanceCounters() = default;
+void PerformanceCounters::beginMeasure() {}
+void PerformanceCounters::endMeasure() {}
+void PerformanceCounters::updateResults(uint64_t) {}
+
+#    endif
+
+ANKERL_NANOBENCH(NODISCARD) PerfCountSet<uint64_t> const& PerformanceCounters::val() const noexcept {
+    return mVal;
+}
+ANKERL_NANOBENCH(NODISCARD) PerfCountSet<bool> const& PerformanceCounters::has() const noexcept {
+    return mHas;
+}
+
+// formatting utilities
+namespace fmt {
+
+// adds thousands separator to numbers
+NumSep::NumSep(char sep)
+    : mSep(sep) {}
+
+char NumSep::do_thousands_sep() const {
+    return mSep;
+}
+
+std::string NumSep::do_grouping() const {
+    return "\003";
+}
+
+// RAII to save & restore a stream's state
+StreamStateRestorer::StreamStateRestorer(std::ostream& s)
+    : mStream(s)
+    , mLocale(s.getloc())
+    , mPrecision(s.precision())
+    , mWidth(s.width())
+    , mFill(s.fill())
+    , mFmtFlags(s.flags()) {}
+
+StreamStateRestorer::~StreamStateRestorer() {
+    restore();
+}
+
+// sets back all stream info that we remembered at construction
+void StreamStateRestorer::restore() {
+    mStream.imbue(mLocale);
+    mStream.precision(mPrecision);
+    mStream.width(mWidth);
+    mStream.fill(mFill);
+    mStream.flags(mFmtFlags);
+}
+
+Number::Number(int width, int precision, int64_t value)
+    : mWidth(width)
+    , mPrecision(precision)
+    , mValue(static_cast<double>(value)) {}
+
+Number::Number(int width, int precision, double value)
+    : mWidth(width)
+    , mPrecision(precision)
+    , mValue(value) {}
+
+std::ostream& Number::write(std::ostream& os) const {
+    StreamStateRestorer restorer(os);
+    os.imbue(std::locale(os.getloc(), new NumSep(',')));
+    os << std::setw(mWidth) << std::setprecision(mPrecision) << std::fixed << mValue;
+    return os;
+}
+
+std::string Number::to_s() const {
+    std::stringstream ss;
+    write(ss);
+    return ss.str();
+}
+
+std::string to_s(uint64_t n) {
+    std::string str;
+    do {
+        str += static_cast<char>('0' + static_cast<char>(n % 10));
+        n /= 10;
+    } while (n != 0);
+    std::reverse(str.begin(), str.end());
+    return str;
+}
+
+std::ostream& operator<<(std::ostream& os, Number const& n) {
+    return n.write(os);
+}
+
+MarkDownColumn::MarkDownColumn(int w, int prec, std::string const& tit, std::string const& suff, double val)
+    : mWidth(w)
+    , mPrecision(prec)
+    , mTitle(tit)
+    , mSuffix(suff)
+    , mValue(val) {}
+
+std::string MarkDownColumn::title() const {
+    std::stringstream ss;
+    ss << '|' << std::setw(mWidth - 2) << std::right << mTitle << ' ';
+    return ss.str();
+}
+
+std::string MarkDownColumn::separator() const {
+    std::string sep(static_cast<size_t>(mWidth), '-');
+    sep.front() = '|';
+    sep.back() = ':';
+    return sep;
+}
+
+std::string MarkDownColumn::invalid() const {
+    std::string sep(static_cast<size_t>(mWidth), ' ');
+    sep.front() = '|';
+    sep[sep.size() - 2] = '-';
+    return sep;
+}
+
+std::string MarkDownColumn::value() const {
+    std::stringstream ss;
+    auto width = mWidth - 2 - static_cast<int>(mSuffix.size());
+    ss << '|' << Number(width, mPrecision, mValue) << mSuffix << ' ';
+    return ss.str();
+}
+
+// Formats any text as markdown code, escaping backticks.
+MarkDownCode::MarkDownCode(std::string const& what) {
+    mWhat.reserve(what.size() + 2);
+    mWhat.push_back('`');
+    for (char c : what) {
+        mWhat.push_back(c);
+        if ('`' == c) {
+            mWhat.push_back('`');
+        }
+    }
+    mWhat.push_back('`');
+}
+
+std::ostream& MarkDownCode::write(std::ostream& os) const {
+    return os << mWhat;
+}
+
+std::ostream& operator<<(std::ostream& os, MarkDownCode const& mdCode) {
+    return mdCode.write(os);
+}
+} // namespace fmt
+} // namespace detail
+
+// provide implementation here so it's only generated once
+Config::Config() = default;
+Config::~Config() = default;
+Config& Config::operator=(Config const&) = default;
+Config& Config::operator=(Config&&) = default;
+Config::Config(Config const&) = default;
+Config::Config(Config&&) noexcept = default;
+
+// provide implementation here so it's only generated once
+Result::~Result() = default;
+Result& Result::operator=(Result const&) = default;
+Result& Result::operator=(Result&&) = default;
+Result::Result(Result const&) = default;
+Result::Result(Result&&) noexcept = default;
+
+namespace detail {
+template <typename T>
+inline constexpr typename std::underlying_type<T>::type u(T val) noexcept {
+    return static_cast<typename std::underlying_type<T>::type>(val);
+}
+} // namespace detail
+
+// Result returned after a benchmark has finished. Can be used as a baseline for relative().
+Result::Result(Config const& benchmarkConfig)
+    : mConfig(benchmarkConfig)
+    , mNameToMeasurements{detail::u(Result::Measure::_size)} {}
+
+void Result::add(Clock::duration totalElapsed, uint64_t iters, detail::PerformanceCounters const& pc) {
+    using detail::d;
+    using detail::u;
+
+    double dIters = d(iters);
+    mNameToMeasurements[u(Result::Measure::iterations)].push_back(dIters);
+
+    mNameToMeasurements[u(Result::Measure::elapsed)].push_back(d(totalElapsed) / dIters);
+    if (pc.has().pageFaults) {
+        mNameToMeasurements[u(Result::Measure::pagefaults)].push_back(d(pc.val().pageFaults) / dIters);
+    }
+    if (pc.has().cpuCycles) {
+        mNameToMeasurements[u(Result::Measure::cpucycles)].push_back(d(pc.val().cpuCycles) / dIters);
+    }
+    if (pc.has().contextSwitches) {
+        mNameToMeasurements[u(Result::Measure::contextswitches)].push_back(d(pc.val().contextSwitches) / dIters);
+    }
+    if (pc.has().instructions) {
+        mNameToMeasurements[u(Result::Measure::instructions)].push_back(d(pc.val().instructions) / dIters);
+    }
+    if (pc.has().branchInstructions) {
+        double branchInstructions = 0.0;
+        // correcting branches: remove branch introduced by the while (...) loop for each iteration.
+        if (pc.val().branchInstructions > iters + 1U) {
+            branchInstructions = d(pc.val().branchInstructions - (iters + 1U));
+        }
+        mNameToMeasurements[u(Result::Measure::branchinstructions)].push_back(branchInstructions / dIters);
+
+        if (pc.has().branchMisses) {
+            // correcting branch misses
+            double branchMisses = d(pc.val().branchMisses);
+            if (branchMisses > branchInstructions) {
+                // can't have branch misses when there were branches...
+                branchMisses = branchInstructions;
+            }
+
+            // assuming at least one missed branch for the loop
+            branchMisses -= 1.0;
+            if (branchMisses < 1.0) {
+                branchMisses = 1.0;
+            }
+            mNameToMeasurements[u(Result::Measure::branchmisses)].push_back(branchMisses / dIters);
+        }
+    }
+}
+
+Config const& Result::config() const noexcept {
+    return mConfig;
+}
+
+inline double calcMedian(std::vector<double>& data) {
+    if (data.empty()) {
+        return 0.0;
+    }
+    std::sort(data.begin(), data.end());
+
+    auto midIdx = data.size() / 2U;
+    if (1U == (data.size() & 1U)) {
+        return data[midIdx];
+    }
+    return (data[midIdx - 1U] + data[midIdx]) / 2U;
+}
+
+double Result::median(Measure m) const {
+    // create a copy so we can sort
+    auto data = mNameToMeasurements[detail::u(m)];
+    return calcMedian(data);
+}
+
+double Result::average(Measure m) const {
+    using detail::d;
+    auto const& data = mNameToMeasurements[detail::u(m)];
+    if (data.empty()) {
+        return 0.0;
+    }
+
+    // create a copy so we can sort
+    return sum(m) / d(data.size());
+}
+
+double Result::medianAbsolutePercentError(Measure m) const {
+    // create copy
+    auto data = mNameToMeasurements[detail::u(m)];
+
+    // calculates MdAPE which is the median of percentage error
+    // see https://www.spiderfinancial.com/support/documentation/numxl/reference-manual/forecasting-performance/mdape
+    auto med = calcMedian(data);
+
+    // transform the data to absolute error
+    for (auto& x : data) {
+        x = (x - med) / x;
+        if (x < 0) {
+            x = -x;
+        }
+    }
+    return calcMedian(data);
+}
+
+double Result::sum(Measure m) const noexcept {
+    auto const& data = mNameToMeasurements[detail::u(m)];
+    return std::accumulate(data.begin(), data.end(), 0.0);
+}
+
+double Result::sumProduct(Measure m1, Measure m2) const noexcept {
+    auto const& data1 = mNameToMeasurements[detail::u(m1)];
+    auto const& data2 = mNameToMeasurements[detail::u(m2)];
+
+    if (data1.size() != data2.size()) {
+        return 0.0;
+    }
+
+    double result = 0.0;
+    for (size_t i = 0, s = data1.size(); i != s; ++i) {
+        result += data1[i] * data2[i];
+    }
+    return result;
+}
+
+bool Result::has(Measure m) const noexcept {
+    return !mNameToMeasurements[detail::u(m)].empty();
+}
+
+double Result::get(size_t idx, Measure m) const {
+    auto const& data = mNameToMeasurements[detail::u(m)];
+    return data.at(idx);
+}
+
+bool Result::empty() const noexcept {
+    return 0U == size();
+}
+
+size_t Result::size() const noexcept {
+    auto const& data = mNameToMeasurements[detail::u(Measure::elapsed)];
+    return data.size();
+}
+
+double Result::minimum(Measure m) const noexcept {
+    auto const& data = mNameToMeasurements[detail::u(m)];
+    if (data.empty()) {
+        return 0.0;
+    }
+
+    // here its save to assume that at least one element is there
+    return *std::min_element(data.begin(), data.end());
+}
+
+double Result::maximum(Measure m) const noexcept {
+    auto const& data = mNameToMeasurements[detail::u(m)];
+    if (data.empty()) {
+        return 0.0;
+    }
+
+    // here its save to assume that at least one element is there
+    return *std::max_element(data.begin(), data.end());
+}
+
+Result::Measure Result::fromString(std::string const& str) {
+    if (str == "elapsed") {
+        return Measure::elapsed;
+    } else if (str == "iterations") {
+        return Measure::iterations;
+    } else if (str == "pagefaults") {
+        return Measure::pagefaults;
+    } else if (str == "cpucycles") {
+        return Measure::cpucycles;
+    } else if (str == "contextswitches") {
+        return Measure::contextswitches;
+    } else if (str == "instructions") {
+        return Measure::instructions;
+    } else if (str == "branchinstructions") {
+        return Measure::branchinstructions;
+    } else if (str == "branchmisses") {
+        return Measure::branchmisses;
+    } else {
+        // not found, return _size
+        return Measure::_size;
+    }
+}
+
+// Configuration of a microbenchmark.
+Bench::Bench() {
+    mConfig.mOut = &std::cout;
+}
+
+Bench::Bench(Bench&&) = default;
+Bench& Bench::operator=(Bench&&) = default;
+Bench::Bench(Bench const&) = default;
+Bench& Bench::operator=(Bench const&) = default;
+Bench::~Bench() noexcept = default;
+
+double Bench::batch() const noexcept {
+    return mConfig.mBatch;
+}
+
+double Bench::complexityN() const noexcept {
+    return mConfig.mComplexityN;
+}
+
+// Set a baseline to compare it to. 100% it is exactly as fast as the baseline, >100% means it is faster than the baseline, <100%
+// means it is slower than the baseline.
+Bench& Bench::relative(bool isRelativeEnabled) noexcept {
+    mConfig.mIsRelative = isRelativeEnabled;
+    return *this;
+}
+bool Bench::relative() const noexcept {
+    return mConfig.mIsRelative;
+}
+
+Bench& Bench::performanceCounters(bool showPerformanceCounters) noexcept {
+    mConfig.mShowPerformanceCounters = showPerformanceCounters;
+    return *this;
+}
+bool Bench::performanceCounters() const noexcept {
+    return mConfig.mShowPerformanceCounters;
+}
+
+// Operation unit. Defaults to "op", could be e.g. "byte" for string processing.
+// If u differs from currently set unit, the stored results will be cleared.
+// Use singular (byte, not bytes).
+Bench& Bench::unit(char const* u) {
+    if (u != mConfig.mUnit) {
+        mResults.clear();
+    }
+    mConfig.mUnit = u;
+    return *this;
+}
+
+Bench& Bench::unit(std::string const& u) {
+    return unit(u.c_str());
+}
+
+std::string const& Bench::unit() const noexcept {
+    return mConfig.mUnit;
+}
+
+// If benchmarkTitle differs from currently set title, the stored results will be cleared.
+Bench& Bench::title(const char* benchmarkTitle) {
+    if (benchmarkTitle != mConfig.mBenchmarkTitle) {
+        mResults.clear();
+    }
+    mConfig.mBenchmarkTitle = benchmarkTitle;
+    return *this;
+}
+Bench& Bench::title(std::string const& benchmarkTitle) {
+    if (benchmarkTitle != mConfig.mBenchmarkTitle) {
+        mResults.clear();
+    }
+    mConfig.mBenchmarkTitle = benchmarkTitle;
+    return *this;
+}
+
+std::string const& Bench::title() const noexcept {
+    return mConfig.mBenchmarkTitle;
+}
+
+Bench& Bench::name(const char* benchmarkName) {
+    mConfig.mBenchmarkName = benchmarkName;
+    return *this;
+}
+
+Bench& Bench::name(std::string const& benchmarkName) {
+    mConfig.mBenchmarkName = benchmarkName;
+    return *this;
+}
+
+std::string const& Bench::name() const noexcept {
+    return mConfig.mBenchmarkName;
+}
+
+// Number of epochs to evaluate. The reported result will be the median of evaluation of each epoch.
+Bench& Bench::epochs(size_t numEpochs) noexcept {
+    mConfig.mNumEpochs = numEpochs;
+    return *this;
+}
+size_t Bench::epochs() const noexcept {
+    return mConfig.mNumEpochs;
+}
+
+// Desired evaluation time is a multiple of clock resolution. Default is to be 1000 times above this measurement precision.
+Bench& Bench::clockResolutionMultiple(size_t multiple) noexcept {
+    mConfig.mClockResolutionMultiple = multiple;
+    return *this;
+}
+size_t Bench::clockResolutionMultiple() const noexcept {
+    return mConfig.mClockResolutionMultiple;
+}
+
+// Sets the maximum time each epoch should take. Default is 100ms.
+Bench& Bench::maxEpochTime(std::chrono::nanoseconds t) noexcept {
+    mConfig.mMaxEpochTime = t;
+    return *this;
+}
+std::chrono::nanoseconds Bench::maxEpochTime() const noexcept {
+    return mConfig.mMaxEpochTime;
+}
+
+// Sets the maximum time each epoch should take. Default is 100ms.
+Bench& Bench::minEpochTime(std::chrono::nanoseconds t) noexcept {
+    mConfig.mMinEpochTime = t;
+    return *this;
+}
+std::chrono::nanoseconds Bench::minEpochTime() const noexcept {
+    return mConfig.mMinEpochTime;
+}
+
+Bench& Bench::minEpochIterations(uint64_t numIters) noexcept {
+    mConfig.mMinEpochIterations = (numIters == 0) ? 1 : numIters;
+    return *this;
+}
+uint64_t Bench::minEpochIterations() const noexcept {
+    return mConfig.mMinEpochIterations;
+}
+
+Bench& Bench::epochIterations(uint64_t numIters) noexcept {
+    mConfig.mEpochIterations = numIters;
+    return *this;
+}
+uint64_t Bench::epochIterations() const noexcept {
+    return mConfig.mEpochIterations;
+}
+
+Bench& Bench::warmup(uint64_t numWarmupIters) noexcept {
+    mConfig.mWarmup = numWarmupIters;
+    return *this;
+}
+uint64_t Bench::warmup() const noexcept {
+    return mConfig.mWarmup;
+}
+
+Bench& Bench::config(Config const& benchmarkConfig) {
+    mConfig = benchmarkConfig;
+    return *this;
+}
+Config const& Bench::config() const noexcept {
+    return mConfig;
+}
+
+Bench& Bench::output(std::ostream* outstream) noexcept {
+    mConfig.mOut = outstream;
+    return *this;
+}
+
+ANKERL_NANOBENCH(NODISCARD) std::ostream* Bench::output() const noexcept {
+    return mConfig.mOut;
+}
+
+std::vector<Result> const& Bench::results() const noexcept {
+    return mResults;
+}
+
+Bench& Bench::render(char const* templateContent, std::ostream& os) {
+    ::ankerl::nanobench::render(templateContent, *this, os);
+    return *this;
+}
+
+std::vector<BigO> Bench::complexityBigO() const {
+    std::vector<BigO> bigOs;
+    auto rangeMeasure = BigO::collectRangeMeasure(mResults);
+    bigOs.emplace_back("O(1)", rangeMeasure, [](double) {
+        return 1.0;
+    });
+    bigOs.emplace_back("O(n)", rangeMeasure, [](double n) {
+        return n;
+    });
+    bigOs.emplace_back("O(log n)", rangeMeasure, [](double n) {
+        return std::log2(n);
+    });
+    bigOs.emplace_back("O(n log n)", rangeMeasure, [](double n) {
+        return n * std::log2(n);
+    });
+    bigOs.emplace_back("O(n^2)", rangeMeasure, [](double n) {
+        return n * n;
+    });
+    bigOs.emplace_back("O(n^3)", rangeMeasure, [](double n) {
+        return n * n * n;
+    });
+    std::sort(bigOs.begin(), bigOs.end());
+    return bigOs;
+}
+
+Rng::Rng()
+    : mX(0)
+    , mY(0) {
+    std::random_device rd;
+    std::uniform_int_distribution<uint64_t> dist;
+    do {
+        mX = dist(rd);
+        mY = dist(rd);
+    } while (mX == 0 && mY == 0);
+}
+
+ANKERL_NANOBENCH_NO_SANITIZE("integer")
+uint64_t splitMix64(uint64_t& state) noexcept {
+    uint64_t z = (state += UINT64_C(0x9e3779b97f4a7c15));
+    z = (z ^ (z >> 30U)) * UINT64_C(0xbf58476d1ce4e5b9);
+    z = (z ^ (z >> 27U)) * UINT64_C(0x94d049bb133111eb);
+    return z ^ (z >> 31U);
+}
+
+// Seeded as described in romu paper (update april 2020)
+Rng::Rng(uint64_t seed) noexcept
+    : mX(splitMix64(seed))
+    , mY(splitMix64(seed)) {
+    for (size_t i = 0; i < 10; ++i) {
+        operator()();
+    }
+}
+
+// only internally used to copy the RNG.
+Rng::Rng(uint64_t x, uint64_t y) noexcept
+    : mX(x)
+    , mY(y) {}
+
+Rng Rng::copy() const noexcept {
+    return Rng{mX, mY};
+}
+
+BigO::RangeMeasure BigO::collectRangeMeasure(std::vector<Result> const& results) {
+    BigO::RangeMeasure rangeMeasure;
+    for (auto const& result : results) {
+        if (result.config().mComplexityN > 0.0) {
+            rangeMeasure.emplace_back(result.config().mComplexityN, result.median(Result::Measure::elapsed));
+        }
+    }
+    return rangeMeasure;
+}
+
+BigO::BigO(std::string const& bigOName, RangeMeasure const& rangeMeasure)
+    : mName(bigOName) {
+
+    // estimate the constant factor
+    double sumRangeMeasure = 0.0;
+    double sumRangeRange = 0.0;
+
+    for (size_t i = 0; i < rangeMeasure.size(); ++i) {
+        sumRangeMeasure += rangeMeasure[i].first * rangeMeasure[i].second;
+        sumRangeRange += rangeMeasure[i].first * rangeMeasure[i].first;
+    }
+    mConstant = sumRangeMeasure / sumRangeRange;
+
+    // calculate root mean square
+    double err = 0.0;
+    double sumMeasure = 0.0;
+    for (size_t i = 0; i < rangeMeasure.size(); ++i) {
+        auto diff = mConstant * rangeMeasure[i].first - rangeMeasure[i].second;
+        err += diff * diff;
+
+        sumMeasure += rangeMeasure[i].second;
+    }
+
+    auto n = static_cast<double>(rangeMeasure.size());
+    auto mean = sumMeasure / n;
+    mNormalizedRootMeanSquare = std::sqrt(err / n) / mean;
+}
+
+BigO::BigO(const char* bigOName, RangeMeasure const& rangeMeasure)
+    : BigO(std::string(bigOName), rangeMeasure) {}
+
+std::string const& BigO::name() const noexcept {
+    return mName;
+}
+
+double BigO::constant() const noexcept {
+    return mConstant;
+}
+
+double BigO::normalizedRootMeanSquare() const noexcept {
+    return mNormalizedRootMeanSquare;
+}
+
+bool BigO::operator<(BigO const& other) const noexcept {
+    return std::tie(mNormalizedRootMeanSquare, mName) < std::tie(other.mNormalizedRootMeanSquare, other.mName);
+}
+
+std::ostream& operator<<(std::ostream& os, BigO const& bigO) {
+    return os << bigO.constant() << " * " << bigO.name() << ", rms=" << bigO.normalizedRootMeanSquare();
+}
+
+std::ostream& operator<<(std::ostream& os, std::vector<ankerl::nanobench::BigO> const& bigOs) {
+    detail::fmt::StreamStateRestorer restorer(os);
+    os << std::endl << "|   coefficient |   err% | complexity" << std::endl << "|--------------:|-------:|------------" << std::endl;
+    for (auto const& bigO : bigOs) {
+        os << "|" << std::setw(14) << std::setprecision(7) << std::scientific << bigO.constant() << " ";
+        os << "|" << detail::fmt::Number(6, 1, bigO.normalizedRootMeanSquare() * 100.0) << "% ";
+        os << "| " << bigO.name();
+        os << std::endl;
+    }
+    return os;
+}
+
+} // namespace nanobench
+} // namespace ankerl
+
+#endif // ANKERL_NANOBENCH_IMPLEMENT
+#endif // ANKERL_NANOBENCH_H_INCLUDED

+ 2 - 0
Tests/Output/.gitignore

@@ -0,0 +1,2 @@
+*.png
+*.log

+ 3 - 0
Tests/Output/Readme.txt

@@ -0,0 +1,3 @@
+By default, the VisualTests suite will output screenshots and diff images into this directory, and read previous screenshots from this directory.
+
+Use the CMake options VISUAL_TESTS_OUTPUT_DIRECTORY and VISUAL_TESTS_INPUT_DIRECTORY to specify other directories.

+ 129 - 0
Tests/Source/Benchmarks/DataExpression.cpp

@@ -0,0 +1,129 @@
+/*
+ * This source file is part of RmlUi, the HTML/CSS Interface Middleware
+ *
+ * For the latest information, see http://github.com/mikke89/RmlUi
+ *
+ * Copyright (c) 2008-2010 CodePoint Ltd, Shift Technology Ltd
+ * Copyright (c) 2019 The RmlUi Team, and contributors
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ */
+
+
+#include "../../../Source/Core/DataExpression.cpp"
+
+#include <RmlUi/Core/DataModel.h>
+#include <doctest.h>
+#include <nanobench.h>
+
+using namespace Rml;
+using namespace ankerl;
+
+static DataTypeRegister type_register;
+static DataModel model(type_register.GetTransformFuncRegister());
+static DataExpressionInterface interface(&model, nullptr);
+
+
+TEST_CASE("Data expressions")
+{
+	float radius = 6.0f;
+	String color_name = "color";
+	Colourb color_value = Colourb(180, 100, 255);
+
+
+	DataModelConstructor constructor(&model, &type_register);
+	constructor.Bind("radius", &radius);
+	constructor.Bind("color_name", &color_name);
+	constructor.BindFunc("color_value", [&](Variant& variant) {
+		variant = ToString(color_value);
+	});
+
+
+	nanobench::Bench bench;
+	bench.title("Data expression");
+	bench.relative(true);
+
+	auto bench_expression = [&](const String& expression, const char* parse_name, const char* execute_name) {
+		DataParser parser(expression, interface);
+
+		bool result = true;
+		bench.run(parse_name, [&] {
+			result &= parser.Parse(false);
+			});
+
+		REQUIRE(result);
+
+		Program program = parser.ReleaseProgram();
+		AddressList addresses = parser.ReleaseAddresses();
+		DataInterpreter interpreter(program, addresses, interface);
+
+		bench.run(execute_name, [&] {
+			result &= interpreter.Run();
+		});
+
+		REQUIRE(result);
+	};
+
+
+	bench_expression(
+		"2 * 2",
+		"Simple (parse)",
+		"Simple (execute)"
+	);
+
+	bench_expression(
+		"true || false ? true && radius==1+2 ? 'Absolutely!' : color_value : 'no'",
+		"Complex (parse)",
+		"Complex (execute)"
+	);
+
+	auto bench_assignment = [&](const String& expression, const char* parse_name, const char* execute_name) {
+		DataParser parser(expression, interface); 
+		
+		bool result = true;
+		bench.run(parse_name, [&] {
+			result &= parser.Parse(true);
+			});
+
+		REQUIRE(result);
+
+		Program program = parser.ReleaseProgram();
+		AddressList addresses = parser.ReleaseAddresses();
+		DataInterpreter interpreter(program, addresses, interface);
+
+		bench.run(execute_name, [&] {
+			result &= interpreter.Run();
+		});
+
+		REQUIRE(result);
+	};
+
+	bench_assignment(
+		"radius = 15",
+		"Simple assign (parse)",
+		"Simple assign (execute)"
+	);
+
+	bench_assignment(
+		"radius = radius*radius*3.14; color_name = 'image-color'",
+		"Complex assign (parse)",
+		"Complex assign (execute)"
+	);
+}

+ 293 - 0
Tests/Source/Benchmarks/Element.cpp

@@ -0,0 +1,293 @@
+/*
+ * This source file is part of RmlUi, the HTML/CSS Interface Middleware
+ *
+ * For the latest information, see http://github.com/mikke89/RmlUi
+ *
+ * Copyright (c) 2008-2010 CodePoint Ltd, Shift Technology Ltd
+ * Copyright (c) 2019 The RmlUi Team, and contributors
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ */
+
+#include "../Common/TestsShell.h"
+#include "../Common/TestsInterface.h"
+#include <RmlUi/Core/Context.h>
+#include <RmlUi/Core/Element.h>
+#include <RmlUi/Core/ElementDocument.h>
+#include <RmlUi/Core/Types.h>
+
+#include <doctest.h>
+#include <nanobench.h>
+
+using namespace ankerl;
+using namespace Rml;
+
+static String GenerateRml(const int num_rows)
+{
+	static nanobench::Rng rng;
+
+	Rml::String rml;
+	rml.reserve(1000 * num_rows);
+
+	for (int i = 0; i < num_rows; i++)
+	{
+		int index = rng() % 1000;
+		int route = rng() % 50;
+		int max = (rng() % 40) + 10;
+		int value = rng() % max;
+		Rml::String rml_row = Rml::CreateString(1000, R"(
+			<div class="row">
+				<div class="col col1"><button class="expand" index="%d">+</button>&nbsp;<a>Route %d</a></div>
+				<div class="col col23"><input type="range" class="assign_range" min="0" max="%d" value="%d"/></div>
+				<div class="col col4">Assigned</div>
+				<select>
+					<option>Red</option><option>Blue</option><option selected>Green</option><option style="background-color: yellow;">Yellow</option>
+				</select>
+				<div class="inrow unmark_collapse">
+					<div class="col col123 assign_text">Assign to route</div>
+					<div class="col col4">
+						<input type="submit" class="vehicle_depot_assign_confirm" quantity="0">Confirm</input>
+					</div>
+				</div>
+			</div>)",
+			index,
+			route,
+			max,
+			value
+		);
+		rml += rml_row;
+	}
+
+	return rml;
+}
+
+
+TEST_CASE("Elements (shell)")
+{
+	Context* context = TestsShell::GetMainContext();
+	REQUIRE(context);
+
+	ElementDocument* document = context->LoadDocument("basic/benchmark/data/benchmark.rml");
+	REQUIRE(document);
+	document->Show();
+
+	Element* el = document->GetElementById("performance");
+	REQUIRE(el);
+
+	nanobench::Bench bench;
+	bench.title("Elements (shell)");
+	bench.relative(true);
+
+	constexpr int num_rows = 50;
+	const String rml = GenerateRml(num_rows);
+
+	el->SetInnerRML(rml);
+	context->Update();
+	context->Render();
+
+	bench.run("Update (unmodified)", [&] {
+		context->Update();
+	});
+
+	bench.run("Render", [&] {
+		TestsShell::PrepareRenderBuffer();
+		context->Render();
+		TestsShell::PresentRenderBuffer();
+	});
+
+	bench.run("SetInnerRML", [&] {
+		el->SetInnerRML(rml);
+	});
+
+	bench.run("SetInnerRML + Update", [&] {
+		el->SetInnerRML(rml);
+		context->Update();
+	});
+
+	bench.run("SetInnerRML + Update + Render", [&] {
+		el->SetInnerRML(rml);
+		context->Update();
+		TestsShell::PrepareRenderBuffer();
+		context->Render();
+		TestsShell::PresentRenderBuffer();
+	});
+
+	document->Close();
+}
+
+
+TEST_CASE("Elements (dummy interface)")
+{
+	TestsRenderInterface render_interface;
+	Context* context = TestsShell::CreateContext("element_dummy", &render_interface);
+	REQUIRE(context);
+
+	ElementDocument* document = context->LoadDocument("basic/benchmark/data/benchmark.rml");
+	REQUIRE(document);
+	document->Show();
+
+	Element* el = document->GetElementById("performance");
+	REQUIRE(el);
+
+	nanobench::Bench bench;
+	bench.title("Elements (dummy interface)");
+	bench.relative(true);
+
+	constexpr int num_rows = 50;
+	const String rml = GenerateRml(num_rows);
+
+	el->SetInnerRML(rml);
+	context->Update();
+	context->Render();
+
+	bench.run("Update (unmodified)", [&] {
+		context->Update();
+	});
+
+	bench.run("Render", [&] {
+		context->Render();
+	});
+
+	bench.run("SetInnerRML", [&] {
+		el->SetInnerRML(rml);
+	});
+
+	bench.run("SetInnerRML + Update", [&] {
+		el->SetInnerRML(rml);
+		context->Update();
+	});
+
+	bench.run("SetInnerRML + Update + Render", [&] {
+		el->SetInnerRML(rml);
+		context->Update();
+		context->Render();
+	});
+
+	render_interface.ResetCounters();
+	context->Render();
+	auto& counters = render_interface.GetCounters();
+
+	const String msg = CreateString(256,
+		"Stats for single Context::Render() with n=%d rows: \n"
+		"Render calls: %zu\n"
+		"Scissor enable: %zu\n"
+		"Scissor set: %zu\n"
+		"Texture load: %zu\n"
+		"Texture generate: %zu\n"
+		"Texture release: %zu\n"
+		"Transform set: %zu\n",
+		num_rows,
+		counters.render_calls,
+		counters.enable_scissor,
+		counters.set_scissor,
+		counters.load_texture,
+		counters.generate_texture,
+		counters.release_texture,
+		counters.set_transform
+	);
+	MESSAGE(msg);
+
+	document->Close();
+	TestsShell::RemoveContext(context);
+}
+
+
+TEST_CASE("Elements asymptotic complexity (dummy interface)")
+{
+	TestsRenderInterface render_interface;
+	Context* context = TestsShell::CreateContext("element_complexity", &render_interface);
+	REQUIRE(context);
+
+	ElementDocument* document = context->LoadDocument("basic/benchmark/data/benchmark.rml");
+	REQUIRE(document);
+	document->Show();
+
+	Element* el = document->GetElementById("performance");
+	REQUIRE(el);
+
+
+	struct BenchDef {
+		const char* title;
+		Function<void(const String& rml)> run;
+	};
+
+	Vector<BenchDef> bench_list = {
+		{
+			"SetInnerRML",
+			[&](const String& rml) {
+				el->SetInnerRML(rml);
+			}
+		},
+		{
+			"Update (unmodified)",
+			[&](const String& /*rml*/) {
+				context->Update();
+			}
+		},
+		{
+			"Render",
+			[&](const String& /*rml*/) {
+				context->Render();
+			}
+		},
+		{
+			"SetInnerRML + Update",
+			[&](const String& rml) {
+				el->SetInnerRML(rml);
+				context->Update();
+			}
+		},
+		{
+			"SetInnerRML + Update + Render",
+			[&](const String& rml) {
+				el->SetInnerRML(rml);
+				context->Update();
+				context->Render();
+			}
+		},
+	};
+
+	for (auto& bench_def : bench_list)
+	{
+		nanobench::Bench bench;
+		bench.title(bench_def.title);
+		bench.relative(true);
+
+		// Running the benchmark multiple times, with different number of rows.
+		for (const int num_rows : { 1, 2, 5, 10, 20, 50, 100, 200, 500 })
+		{
+			const String rml = GenerateRml(num_rows);
+
+			el->SetInnerRML(rml);
+			context->Update();
+			context->Render();
+
+			bench.complexityN(num_rows).run(bench_def.title, [&]() {
+				bench_def.run(rml);
+			});
+		}
+
+#ifdef RMLUI_BENCHMARKS_SHOW_COMPLEXITY
+		MESSAGE(bench.complexityBigO());
+#endif
+	}
+
+	TestsShell::RemoveContext(context);
+}

+ 55 - 0
Tests/Source/Benchmarks/main.cpp

@@ -0,0 +1,55 @@
+/*
+ * This source file is part of RmlUi, the HTML/CSS Interface Middleware
+ *
+ * For the latest information, see http://github.com/mikke89/RmlUi
+ *
+ * Copyright (c) 2008-2010 CodePoint Ltd, Shift Technology Ltd
+ * Copyright (c) 2019 The RmlUi Team, and contributors
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ */
+
+#include "../Common/TestsShell.h"
+
+#define ANKERL_NANOBENCH_IMPLEMENT
+#include <nanobench.h>
+
+#define DOCTEST_CONFIG_IMPLEMENT
+#include <doctest.h>
+
+
+int main(int argc, char** argv) {
+
+    // Initialize and run doctest
+    doctest::Context doctest_context;
+
+    doctest_context.applyCommandLine(argc, argv);
+
+    int doctest_result = doctest_context.run();
+
+    if (doctest_context.shouldExit())
+        return doctest_result;
+
+    // RmlUi is initialized during doctest run above as necessary.
+    // Clean everything up here.
+    TestsShell::ShutdownShell();
+
+    return doctest_result;
+}

Some files were not shown because too many files changed in this diff