anki-3d-engine/ThirdParty/Reproc/CHANGELOG.md

Changelog

14.2.4

• Bugfix: Fix a memory leak in reproc_start() on Windows (thanks @AokiYuune).
• Bugfix: Fix a memory leak in reproc++ array class move constructor.
• Allow passing zero-sized array's to reproc's input option (thanks @lightray22).

14.2.3

• Bugfix: Fix sign of EWOULDBLOCK error returned from reproc_read.

14.2.2

• Bugfix: Disallow using fork option when using reproc_run.

14.2.1

• Bugfix: reproc_run now handles forked child processes correctly.
• Bugfix: Sinks of different types can now be passed to reproc::drain.
• Bugfix: Processes on Windows returning negative exit codes don't cause asserts anymore.
• Bugfix: Dependency on librt on POSIX (except osx) systems is now explicit in CMake.
• Bugfix: Added missing stdout redirect option to reproc++.

14.2.0

• Added reproc_pid/process::pid to get the pid of the process
• Fixed compilation error when including reproc/drain.h in C++ code
• Added missing extern "C" block to reproc/run.h

14.1.0

• reproc_run/reproc::run now return the exit code of reproc_stop/process::stop even if the deadline is exceeded.
• A bug where deadlines wouldn't work was fixed.

14.0.0

• Renamed environment to env.
• Added configurable behavior to env option. Extra environment variables are now added via env.extra. Extra environment variables now extend the parent environment by default instead of replacing it.

13.0.1

• Bugfix: Reset the events parameter of every event source if a deadline expires when calling reproc_poll.
• Bugfix: Return 1 from reproc_poll if a deadline expires.
• Bugfix: Don't block in reproc_read on Windows if the nonblocking option was specified.

13.0.0

• Allow passing empty event sources to reproc_poll.

If the process member of a reproc_event_source object is NULL, reproc_poll ignores the event source.

• Return zero from reproc_poll if the given timeout expires instead of REPROC_ETIMEDOUT.

In reproc, we follow the general pattern that we don't modify output arguments if an error occurs. However, when reproc_poll errors, we still want to set all events of the given event sources to zero. To signal that we're modifying the output arguments if the timeout expires, we return zero instead of REPROC_ETIMEDOUT. This is also more consistent with poll and WSAPoll which have the same behaviour.

• If one or more events occur, return the number of processes with events from reproc_poll.

12.0.0

reproc

• Put pipes in blocking mode by default.

This allows using reproc_read and reproc_write directly without having to figure out reproc_poll.

• Add nonblocking option.

Allows putting pipes back in nonblocking mode if needed.

• Child process stderr streams are now redirected to the parent stderr stream by default.

Because pipes are blocking again by default, there's a (small) chance of deadlocks if we redirect both stdout and stderr to pipes. Redirecting stderr to the parent by default avoids that issue.

The other (bigger) issue is that if we redirect stderr to a pipe, there's a good chance users might forget to read from it and discard valuable error output from the child process.

By redirecting to the parent stderr by default, it's immediately noticeable when a child process is not behaving according to expectations as its error output will appear directly on the parent process stderr stream. Users can then still decide to explicitly discard the output from the stderr stream if needed.

• Turn timeout option into an argument for reproc_poll.

While deadlines can differ per process, the timeout is very likely to always be the same so we make it an argument to the only function that uses it, namely reproc_poll.

• In reproc_drain, call sink once with an empty buffer when a stream is closed.

This allows sinks to handle stream closure if needed.

• Change sinks to return int instead of bool. If a sink returns a non-zero value, reproc_drain exits immediately with the same value.

This change allows sinks to return their own errors without having to store extra state.

• reproc_sink_string now returns REPROC_ENOMEM from reproc_drain if a memory allocation fails and no longer frees any output that was read previously.

This allows the user to still do something with the remaining output even if a memory allocation failed. On the flipside, it is now required to always call reproc_free after calling reproc_drain with a sink string, even if it fails.

• Renamed sink.h to drain.h.

Reflect that sink.h contains reproc_drain by renaming it to drain.h.

• Add REPROC_REDIRECT_PATH and shorthand path options.

reproc++

• Equivalent changes as those done for reproc.

• Remove use of reprocxx.

Meson gained support for CMake subprojects containing targets with special characters so we rename directories and CMake targets back to reproc++.

11.0.0

General

• Compilation now happens with compiler extensions disabled (-std=c99 and -std=c++11).

reproc

• Add inherit and discard options as shorthands to set all members of the redirect options to REPROC_REDIRECT_INHERIT and REPROC_REDIRECT_DISCARD respectively.

• Add reproc_run and reproc_run_ex which allows running a process using only a single function.

Running a simple process with reproc required calling reproc_new, reproc_start, reproc_wait and optionally reproc_drain with all the associated error handling. reproc_run encapsulates all this boilerplate.

• Add input option that writes the given to the child process stdin pipe before starting the process.

This allows passing input to the process when using reproc_run.

• Add deadline option that specifies a point in time beyond which reproc_poll will return REPROC_ETIMEDOUT.

• Add REPROC_DEADLINE that makes reproc_wait wait until the deadline specified in the deadline option has expired.

By default, if the deadline option is set, reproc_destroy waits until the deadline expires before sending a SIGTERM signal to the child process.

• Add (POSIX only) fork option that makes reproc_start a safe alternative to fork with support for all of reproc's other features.

• Return the amount of bytes written from reproc_write and stop handling partial writes.

Now that reproc uses nonblocking pipes, it doesn't make sense to handle partial writes anymore. The input option can be used as an alternative. reproc_start keeps writing until all data from input is written to the child process stdin pipe or until a call to reproc_write returns an error.

• Add reproc_poll to query child process events of one or more child processes.

We now support polling multiple child processes for events. reproc_poll mimicks the POSIX poll function but instead of pollfds, it takes a list of event sources which consist out of a process, the events we're interested in and an output field which is filled in by reproc_poll that contains the events that occurred for that child process.

• Stop reading from both stdout and stderr in reproc_read.

Because we now have reproc_poll, reproc_read was simplified to again read from the given stream which is passed again as an argument. To avoid deadlocks, call reproc_poll to figure out the first stream that has data available to read.

• Support polling for process exit events.

By adding REPROC_EVENT_EXIT to the list of interested events, we can poll for child process exit events.

• Add a dependency on Winsock2 on Windows.

To implement reproc_poll, we redirect to sockets on Windows which allows us to use WSAPoll which is required to implement reproc_poll. Using sockets on Windows requires linking with the ws2_32 library. This dependency is automatically handled by CMake and pkg-config.

• Move in, out and err options out of stdio directly into redirect.

This reduces the amount of boilerplate when redirecting streams.

• Rename REPROC_REDIRECT_INHERIT to REPROC_REDIRECT_PARENT.

REPROC_REDIRECT_PARENT more clearly indicates that we're redirecting to the parent's corresponding standard stream.

• Add support for redirecting to operating system handles.

This allows redirecting to operating system-specific handles. On POSIX systems, this feature expects file descriptors. On Windows, it expects HANDLEs or SOCKETs.

• Add support for redirecting to FILE *s.

This gives users a cross-platform way to redirect standard streams to files.

• Add support for redirecting stderr to stdout.

For high-traffic scenarios, it doesn't make sense to allocate a separate pipe for stderr if its output is only going to be combined with the output of stdout. By using REPROC_REDIRECT_STDOUT for stderr, its output is written directly to stdout by the child process.

• Turn redirect's in, out and err options into instances of the new reproc_redirecŧ struct.

An enum didn't cut it anymore for the new file and handle redirect options since those require extra fields to allow specifying which file or handle to redirect to.

• Add redirect.file option as a shorthand for redirecting stdout and stderr to the same file.

reproc++

• reproc++ includes mostly the same changes done to reproc so we only document the differences.

• Add fork method instead of fork option.

Adding a fork option would have required changing start to return std::pair<bool, std::error_code> to allow determining whether we're in the parent or the child process after a fork. However, the bool return value would only be valid when the fork option was enabled. Thus, this approach would unnecessarily complicate all other use cases of start that don't require fork. To solve the issue, we made fork a separate method instead.

10.0.3

reproc

• Fixed issue where reproc_wait would assert when invoked with a timeout of zero on POSIX.

• Fixed issue where reproc_wait would not return REPROC_ETIMEDOUT when invoked with a timeout of zero on POSIX.

10.0.2

• Update CMake project version.

10.0.1

reproc

• Pass timeout once via reproc_options instead of passing it via reproc_read, reproc_write and reproc_drain.

reproc++

• Pass timeout once via reproc::options instead of passing it via process::read, process::write and reproc::drain.

10.0.0

reproc

• Remove reproc_parse.

Instead of checking for REPROC_EPIPE (previously REPROC_ERROR_STREAM_CLOSED), simply check if the given parser has a full message available. If it doesn't, the output streams closed unexpectedly.

• Remove reproc_running and reproc_exit_status.

When calling reproc_running, it would wait with a zero timeout if the process was still running and check if the wait timed out. However, a call to wait can fail for other reasons as well which were all ignored by reproc_running. Instead of reproc_running, A call to reproc_wait with a timeout of zero should be used to check if a process is still running. reproc_wait now also returns the exit status if the process exits or has already exited which removes the need for reproc_exit_status.

• Read from both stdout and stderr in reproc_read to avoid deadlocks and indicate which stream reproc_read was read from.

Previously, users would indicate the stream they wanted to read from when calling reproc_read. However, this lead to issues with programs that write to both stdout and stderr as a user wouldn't know whether stdout or stderr would have output available to read. Reading from only the stdout stream didn't work as the parent could be blocked on reading from stdout while the child was simultaneously blocked on writing to stderr leading to a deadlock. To get around this, users had to start up a separate thread to read from both stdout and stderr at the same time which was a lot of extra work just to get the output of external programs that write to both stdout and stderr. Now, reproc takes care of avoiding the deadlock by checking which of stdout/stderr can be read from, doing the actual read and indicating to the user which stream was read from.

Practically, instead of passing REPROC_STREAM_OUT or REPROC_STREAM_ERR to reproc_read, you now pass a pointer to a REPROC_STREAM variable instead which reproc_read will set to REPROC_STREAM_OUT or REPROC_STREAM_ERR depending on which stream it read from.

If both streams have been closed by the child process, reproc_read returns REPROC_EPIPE.

Because of the changes to reproc_read, reproc_drain now also reads from both stdout and stderr and indicates the stream that was read from to the given sink function via an extra argument passed to the sink.

• Read the output of both stdout and stderr into a single contiguous null-terminated string in reproc_sink_string.

• Remove the bytes_written parameter of reproc_write.

reproc_write now always writes size bytes to the standard input of the child process. Partial writes do not have to be handled by users anymore and are instead handled by reproc internally.

• Define _GNU_SOURCE and _WIN32_WINNT only in the implementation files that need them.

This helps keep track of where we're using functionality that requires extra definitions and makes building reproc in all kinds of build systems simpler as the compiler invocations to build reproc get smaller as a result.

• Change the error handling in the public API to return negative errno (POSIX) or GetLastError (Windows) style values. REPROC_ERROR is replaced by extern constants that are assigned the correct error value based on the platform reproc is built for. Instead of returning REPROC_ERROR, most functions in reproc's API now return int when they can fail. Because system errors are now returned directly, there's no need anymore for REPROC_ERROR and reproc_error_system and they has been removed.

Error handling before 10.0.0:

  REPROC_ERROR error = reproc_start(...);
  if (error) {
    goto finish;
  }

  finish:
  if (error) {
    fprintf(stderr, "%s", reproc_strerror(error));
  }

Error handling from 10.0.0 onwards:

  int r = reproc_start(...);
  if (r < 0) {
    goto finish;
  }

  finish:
  if (r < 0) {
    fprintf(stderr, "%s", reproc_strerror(r));
  }

• Hide the internals of reproc_t.

Instances of reproc_t are now allocated on the heap by calling reproc_new. reproc_destroy releases the memory allocated by reproc_new.

• Take optional arguments via the reproc_options struct in reproc_start.

When using designated initializers, calls to reproc_start are now much more readable than before. Using a struct also makes it much easier to set all options to their default values (reproc_options options = { 0 };). Finally, we can add more options in further releases without requiring existing users to change their code.

• Support redirecting the child process standard streams to /dev/null (POSIX) or NUL (Windows) in reproc_start via the redirect field in reproc_options.

This is especially useful when you're not interested in the output of a child process as redirecting to /dev/null doesn't require regularly flushing the output pipes of the process to prevent deadlocks as is the case when redirecting to pipes.

• Support redirecting the child process standard streams to the parent process standard streams in reproc_starŧ via the redirect field in reproc_options.

This is useful when you want to interleave child process output with the parent process output.

• Modify reproc_start and reproc_destroy to work like the reproc++ process class constructor and destructor.

The stop_actions field in reproc_options can be used to define up to three stop actions that are executed when reproc_destroy is called if the child process is still running. If no explicit stop actions are given, reproc_destroy defaults to waiting indefinitely for the child process to exit.

• Return the amount of bytes read from reproc_read if it succeeds.

This is made possible by the new error handling scheme. Because errors are all negative values, we can use the positive range of an int as the normal return value if no errors occur.

• Return the exit status from reproc_wait and reproc_stop if they succeed.

Same reasoning as above. If the child process has already exited, reproc_wait and reproc_stop simply returns the exit status again.

• Do nothing when NULL is passed to reproc_destroy and always return NULL from reproc_destroy.

This allows reproc_destroy to be safely called on the same instance multiple times when assigning the result of reproc_destroy to the same instance (process = reproc_destroy(process)).

• Take stop actions via the reproc_stop_actions struct in reproc_stop.

This makes it easier to store stop action configurations both in and outside of reproc.

• Add 256 to signal exit codes returned by reproc_wait and reproc_stop.

This prevents conflicts with normal exit codes.

• Add REPROC_SIGTERM and REPROC_SIGKILL constants to match against signal exit codes.

These also work on Windows and correspond to the exit codes returned by sending the CTRL-BREAK signal and calling TerminateProcess respectively.

• Rename REPROC_CLEANUP to REPROC_STOP.

Naming the enum after the function it is passed to (reproc_stop) is simpler than using a different name.

• Rewrite tests in C using CTest and assert and remove doctest.

Doctest is a great library but we don't really lose anything major by replacing it with CTest and asserts. On the other hand, we lose a dependency, don't need to download stuff from CMake anymore and tests compile significantly faster.

Tests are now executed by running cmake --build build --target test.

• Return REPROC_EINVAL from public API functions when passed invalid arguments.

• Make reproc_strerror thread-safe.

• Move reproc_drain to sink.h.

• Make reproc_drain take a separate sink for each output stream. Sinks are now passed via the reproc_sink type.

Using separate sinks for both output streams allows for a lot more flexibility. To use a single sink for both output streams, simply pass the same sink to both the out and err arguments of reproc_drain.

• Turn reproc_sink_string and reproc_sink_discard into functions that return sinks and hide the actual functions in sink.c.

• Add reproc_free to sink.h which must be used to free memory allocated by reproc_sink_string.

This avoids issues with allocating across module (DLL) boundaries on Windows.

• Support passing timeouts to reproc_read, reproc_write and reproc_drain.

Pass REPROC_INFINITE as the timeout to retain the old behaviour.

• Use int to represent timeout values.

• Renamed stop_actions field of reproc_options to stop.

reproc++

• Remove process::parse, process::exit_status and process::running.

Consequence of the equivalents in reproc being removed.

• Take separate out and err arguments in the sink::string and sink::ostream constructors that receive output from the stdout and stderr streams of the child process respectively.

To combine the output from the stdout and stderr streams, simply pass the same string or ostream to both the out and err arguments.

• Modify process::read to return a tuple of the stream read from, the amount of bytes read and an error code. The stream read from and amount of bytes read are only valid if process::read succeeds.

std::tie can be used pre-C++17 to assign the tuple's contents to separate variables.

• Modify process::wait and process::stop to return a pair of exit status and error code. The exit status is only valid if process::wait or process::stop succeeds.

• Alias reproc::error to std::errc.

As OS errors are now used everywhere, we can simply use std::errc for all error handling instead of defining our own error code.

• Add signal::terminate and signal::kill constants.

These are aliases for REPROC_SIGTERM and REPROC_SIGKILL respectively.

• Inline all sink implementations in sink.hpp.

• Add sink::thread_safe::string which is a thread-safe version of sink::string.

• Move process::drain out of the process class and move it to sink.hpp.

process.drain(...) becomes reproc::drain(process, ...).

• Make reproc::drain take a separate sink for each output stream.

Same reasoning as reproc_drain.

• Modify all included sinks to support the new reproc::drain behaviour.

• Support passing timeouts to process::read, process::write and reproc::drain.

They still default to waiting indefinitely which matches their old behaviour.

• Renamed stop_actions field of reproc::options to stop.

CMake

• Drop required CMake version to CMake 3.12.
• Add CMake 3.16 as a supported CMake version.
• Build reproc++ with -pthread when REPROC_MULTITHREADED is enabled.

See https://github.com/DaanDeMeyer/reproc/issues/24 for more information.

• Add REPROC_WARNINGS option (default: OFF) to build with compiler warnings.
• Add REPROC_DEVELOP option (default: OFF) which enables a lot of options to simplify developing reproc.

By default, most of reproc's CMake options are disabled to make including reproc in other projects as simple as possible. However, when working on reproc, we usually wants most options enabled instead. To make enabling all options simpler, REPROC_DEVELOP was added from which most other options take their default value. As a result, enabling REPROC_DEVELOP enables all options related to developing reproc. Additionally, REPROC_DEVELOP takes its initial value from an environment variable of the same name so it can be set once and always take effect whenever running CMake on reproc's source tree.

• Add REPROC_OBJECT_LIBRARIES option to build CMake object libraries.

In CMake, linking a library against a static library doesn't actually copy the object files from the static library into the library. Instead, both static libraries have to be installed and depended on by the final executable. By using CMake object libraries, the object files are copied into the depending static library and no extra artifacts are produced.

• Enable REPROC_INSTALL by default unless REPROC_OBJECT_LIBRARIES is enabled.

As REPROC_OBJECT_LIBRARIES can now be used to depend on reproc without generating extra artifacts, we assume that users not using REPROC_OBJECT_LIBRARIES will want to install the produced artifacts.

• Rename reproc++ to reprocxx inside the CMake build files.

This was done to allow using reproc as a Meson subproject. Meson doesn't accept the '+' character in target names so we use 'x' instead.

• Modify the export headers so that the only extra define necessary is REPROC_SHARED when using reproc as a shared library on Windows.

Naturally, this define is added as a CMake usage requirement and doesn't have to be worried about when using reproc via add_subdirectory or find_package.

9.0.0

General

• Drop support for Windows XP.

• Add support for custom environments.

reproc_start and process::start now take an extra environment parameter that allows specifying custom environments.

IMPORTANT: The environment parameter was inserted before the working_directory parameter so make sure to update existing usages of reproc_start and process::start so that the environment and working_directory arguments are specified in the correct order.

To keep the previous behaviour, pass nullptr as the environment to reproc_start/process::start or use the process::start overload without the environment parameter.

• Remove argc parameter from reproc_start and process::start.

We can trivially calculate argc internally in reproc since argv is required to end with a NULL value.

• Improve implementation of reproc_wait with a timeout on POSIX systems.

Instead of spawning a new process to implement the timeout, we now use sigtimedwait on Linux and kqueue on macOS to wait on SIGCHLD signals and check if the process we're waiting on has exited after each received SIGCHLD signal.

• Remove vfork usage.

Clang analyzer was indicating a host of errors in our usage of vfork. We also discovered tests were behaving differently on macOS depending on whether vfork was enabled or disabled. As we do not have the expertise to verify if vfork is working correctly, we opt to remove it.

• Ensure passing a custom working directory and a relative executable path behaves consistently on all supported platforms.

Previously, calling reproc_start with a relative executable path combined with a custom working directory would behave differently depending on which platform the code was executed on. On POSIX systems, the relative executable path would be resolved relative to the custom working directory. On Windows, the relative executable path would be resolved relative to the parent process working directory. Now, relative executable paths are always resolved relative to the parent process working directory.

• Reimplement reproc_drain/process::drain in terms of reproc_parse/process::parse.

Like reproc_parse and process::parse, reproc_drain and process::drain are now guaranteed to always be called once with an empty buffer before reading any actual data.

We now also guarantee that the initial empty buffer is not NULL or nullptr so the received data and size can always be safely passed to memcpy.

• Add MinGW support.

MinGW CI builds were also added to prevent regressions in MinGW support.

reproc

• Update reproc_strerror to return the actual system error string of the error code returned by reproc_system_error instead of "system error" when passed REPROC_ERROR_SYSTEM as argument.

This should make debugging reproc errors a lot easier.

• Add reproc_sink_string in sink.h, a sink that stores all process output in a single null-terminated C string.

• Add reproc_sink_discard in sink.h, a sink that discards all process output.

reproc++

• Move sinks into sink namespace and remove _sink suffix from all sinks.

• Add discard sink that discards all output read from a stream.

This is useful when a child process produces a lot of output that we're not interested in and cannot handle the output stream being closed or full. When this is the case, simply start a thread that drains the stream with a discard sink.

• Update process::start to work with any kind of string type.

Every string type that implements a size method and the index operator can now be passed in a container to process::start. working_directory now takes a const char * instead of a std::string *.

• Fix compilation error when using process::parse.

8.0.1

• Correctly escape arguments on Windows.

See #18 for more information.

8.0.0

• Change reproc_parse and reproc_drain argument order.

context is now the last argument instead of the first.

• Use uint8_t * as buffer type instead of char * or void *

uint8_t * more clearly indicates reproc is working with buffers of bytes than char * and void *. We choose uint8_t * over char * to avoid errors caused by passing data read by reproc directly to functions that expect null-terminated strings (data read by reproc is not null-terminated).

7.0.0

General

• Rework error handling.

Trying to abstract platform-specific errors in REPROC_ERROR and reproc::errc turned out to be harder than expected. On POSIX it remains very hard to figure out which errors actually have a chance of happening and matching reproc::errc values to std::errc values is also ambiguous and prone to errors. On Windows, there's hardly any documentation on which system errors functions can return so 90% of the time we were just returning REPROC_UNKNOWN_ERROR. Furthermore, many operating system errors will be fatal for most users and we suspect they'll all be handled similarly (stopping the application or retrying).

As a result, in this release we stop trying to abstract system errors in reproc. All system errors in REPROC_ERROR were replaced by a single value (REPROC_ERROR_SYSTEM). reproc::errc was renamed to reproc::error and turned into an error code instead of an error condition and only contains the reproc-specific errors.

reproc users can still retrieve the specific system error using reproc_system_error.

reproc++ users can still match against specific system errors using the std::errc error condition enum (https://en.cppreference.com/w/cpp/error/errc) or print a string presentation of the error using the message method of std::error_code.

All values from REPROC_ERROR are now prefixed with REPROC_ERROR instead of REPROC which helps reduce clutter in code completion.

• Azure Pipelines CI now includes Visual Studio 2019.

• Various smaller improvements and fixes.

CMake

• Introduce REPROC_MULTITHREADED to configure whether reproc should link against pthreads.

By default, REPROC_MULTITHREADED is enabled to prevent accidental undefined behaviour caused by forgetting to enable REPROC_MULTITHREADED. Advanced users might want to disable REPROC_MULTITHREADED when they know for certain their code won't use more than a single thread.

• doctest is now downloaded at configure time instead of being vendored inside the reproc repository.

doctest is only downloaded if REPROC_TEST is enabled.

6.0.0

General

• Added Azure Pipelines CI.

Azure Pipelines provides 10 parallel jobs which is more than Travis and Appveyor combined. If it turns out to be reliable Appveyor and Travis will likely be dropped in the future. For now, all three are enabled.

• Code cleanup and refactoring.

CMake

• Renamed REPROC_TESTS to REPROC_TEST.
• Renamed test executable from tests to test.

reproc

• Renamed reproc_type to reproc_t.

We chose reproc_type initially because _t belongs to POSIX but we switch to using _t because reproc is a sufficiently unique name that we don't have to worry about naming conflicts.

• reproc now keeps track of whether a process has exited and its exit status.

Keeping track of whether the child process has exited allows us to remove the restriction that reproc_wait, reproc_terminate, reproc_kill and reproc_stop cannot be called again on the same process after completing successfully once. Now, if the process has already exited, these methods don't do anything and return REPROC_SUCCESS.

• Added reproc_running to allow checking whether a child process is still running.

• Added reproc_exit_status to allow querying the exit status of a process after it has exited.

• reproc_wait and reproc_stop lost their exit_status output parameter.

Use reproc_exit_status instead to retrieve the exit status.

reproc++

• Added process::running and process::exit_status.

These delegate to reproc_running and reproc_exit_status respectively.

• process::wait and process::stop lost their exit_status output parameter.

Use process::exit_status instead.

5.0.1

reproc++

• Fixed compilation error caused by defining reproc::process's move assignment operator as default in the header which is not allowed when a std::unique_ptr member of an incomplete type is present.

5.0.0

General

• Added and rewrote implementation documentation.
• General refactoring and simplification of the source code.

CMake

• Raised minimum CMake version to 3.13.

Tests are now added to a single target reproc-tests in each subdirectory included with add_subdirectory. Dependencies required to run the added tests are added to reproc-tests with target_link_libraries. Before CMake 3.13, target_link_libraries could not modify targets created outside of the current directory which is why CMake 3.13 is needed.

• REPROC_CI was renamed to REPROC_WARNINGS_AS_ERRORS.

This is a side effect of upgrading cddm. The variable was renamed in cddm to more clearly indicate its purpose.

• Removed namespace from reproc's targets.

To link against reproc or reproc++, you now have to link against the target without a namespace prefix:

  find_package(reproc) # or add_subdirectory(external/reproc)
  target_link_libraries(myapp PRIVATE reproc)

  find_package(reproc++) # or add_subdirectory(external/reproc++)
  target_link_libraries(myapp PRIVATE reproc++)

This change was made because of a change in cddm (a collection of CMake functions to make setting up new projects easier) that removed namespacing and aliases of library targets in favor of namespacing within the target name itself. This change was made because the original target can still conflict with other targets even after adding an alias. This can cause problems when using generic names for targets inside the library itself. An example clarifies the problem:

Imagine reproc added a target for working with processes asynchronously. In the previous naming scheme, we'd do the following in reproc's CMake build files:

  add_library(async "")
  add_library(reproc::async ALIAS async)

However, there's a non-negligible chance that someone using reproc might also have a target named async which would result in a conflict when using reproc with add_subdirectory since there'd be two targets with the same name. With the new naming scheme, we'd do the following instead:

  add_library(reproc-async "")

This has almost zero chance of conflicting with user's target names. The advantage is that with this scheme we can use common target names without conflicting with user's target names which was not the case with the previous naming scheme.

reproc

• Removed undefined behaviour in Windows implementation caused by casting an int to an unsigned int.

• Added a note to reproc_start docs about the behaviour of using a executable path relative to the working directory combined with a custom working directory for the child process on different platforms.

• We now retrieve the file descriptor limit in the parent process (using sysconf) instead of in the child process because sysconf is not guaranteed to be async-signal-safe which all functions called in a child process after forking should be.

• Fixed compilation issue when ATTRIBUTE_LIST_FOUND was undefined (#15).

reproc++

• Generified process::start so it works with any container of std::string satisfying the SequenceContainer interface.

4.0.0

General

• Internal improvements and documentation fixes.

reproc

• Added reproc_parse which mimics reproc++'s process::parse.
• Added reproc_drain which mimics reproc++'s process::drain along with an example that explains how to use it.

Because C doesn't support lambda's, both of these functions take a function pointer and an extra context argument which is passed to the function pointer each time it is called. The context argument can be used to store any data needed by the given function pointer.

reproc++

• Renamed the process::read overload which takes a parser to process::parse.

This breaking change was done to keep consistency with reproc where we added reproc_parse. We couldn't add another reproc_read since C doesn't support overloading so we made the decision to rename process::read to process::parse instead.

• Changed process::drain sinks to return a boolean instead of void.

Before this change, the only way to stop draining a process was to throw an exception from the sink. By changing sinks to return bool, a sink can tell drain to stop if an error occurs by returning false. The error itself can be stored in the sink if needed.

3.1.3

CMake

• Update project version in CMakeLists.txt from 3.0.0 to the actual latest version (3.1.3).

3.1.2

pkg-config

• Fix pkg-config install prefix.

3.1.0

CMake

• Added REPROC_INSTALL_PKGCONFIG to control whether pkg-config files are installed or not (default: ON).

The vcpkg package manager has no need for the pkg-config files so we added an option to disable installing them.

• Added REPROC_INSTALL_CMAKECONFIGDIR and REPROC_INSTALL_PKGCONFIGDIR to control where cmake config files and pkg-config files are installed respectively (default: ${CMAKE_INSTALL_LIBDIR}/cmake and ${CMAKE_INSTALL_LIBDIR}/pkgconfig).

reproc already uses the values from GNUInstallDirs when generating its install rules which are cache variables that be overridden by users. However, GNUInstallDirs does not include variables for the installation directories of CMake config files and pkg-config files. vcpkg requires cmake config files to be installed to a different directory than the directory reproc used until now. These options were added to allow vcpkg to control where the config files are installed to.

3.0.0

General

• Removed support for Doxygen (and as a result REPROC_DOCS).

All the Doxygen directives made the header docstrings rather hard to read directly. Doxygen's output was also too complicated for a simple library such as reproc. Finally, Doxygen doesn't really provide any intuitive support for documenting a set of libraries. I have an idea for a Doxygen alternative using libclang and cmark but I'm not sure when I'll be able to implement it.

CMake

• Renamed REPROCXX option to REPROC++.

REPROCXX was initially chosen because CMake didn't recommend using anything other than letters and underscores for variable names. However, REPROC++ turns out to work without any problems so we use it since it's the expected name for an option to build reproc++.

• Stopped modifying the default CMAKE_INSTALL_PREFIX on Windows.

In 2.0.0, when installing to the default CMAKE_INSTALL_PREFIX, you would end up with C:\Program Files (x86)\reproc and C:\Program Files (x86)\reproc++ when installing reproc. In 3.0.0, the default CMAKE_INSTALL_PREFIX isn't modified anymore and all libraries are installed to CMAKE_INSTALL_PREFIX in exactly the same way as they are on UNIX systems (include and lib subdirectories directly beneath the installation directory). Sticking to the defaults makes it easy to include reproc in various package managers such as vcpkg.

reproc

• reproc_terminate and reproc_kill don't call reproc_wait internally anymore. reproc_stop has been changed to call reproc_wait after calling reproc_terminate or reproc_kill so it still behaves the same.

Previously, calling reproc_terminate on a list of processes would only call reproc_terminate on the next process after the previous process had exited or the timeout had expired. This made terminating multiple processes take longer than required. By removing the reproc_wait call from reproc_terminate, users can first call reproc_terminate on all processes before waiting for each of them with reproc_wait which makes terminating multiple processes much faster.

• Default to using vfork instead of fork on POSIX systems.

This change was made to increase reproc_start's performance when the parent process is using a large amount of memory. In these scenario's, vfork can be a lot faster than fork. Care is taken to make sure signal handlers in the child don't corrupt the state of the parent process. This change induces an extra constraint in that set*id functions cannot be called while a call to reproc_start is in process, but this situation is rare enough that the tradeoff for better performance seems worth it.

A dependency on pthreads had to be added in order to safely use vfork (we needed access to pthread_sigmask). The CMake and pkg-config files have been updated to automatically find pthreads so users don't have to find it themselves.

• Renamed reproc_error_to_string to reproc_strerror.

The C standard library has strerror for retrieving a string representation of an error. By using the same function name (prefixed with reproc) for a function that does the same for reproc's errors, new users will immediately know what the function does.

reproc++

• reproc++ now takes timeouts as std::chrono::duration values (more specific reproc::milliseconds) instead of unsigned ints.

Taking the reproc::milliseconds type explains a lot more about the expected argument than taking an unsigned int. C++14 also added chrono literals which make constructing reproc::milliseconds values a lot more concise (reproc::milliseconds(2000) => 2000ms).