|
@@ -0,0 +1,755 @@
|
|
|
+package main
|
|
|
+
|
|
|
+#assert(_BUFFER_SIZE > 0);
|
|
|
+
|
|
|
+import "core:fmt"
|
|
|
+import "core:strconv"
|
|
|
+import "core:mem"
|
|
|
+import "core:bits"
|
|
|
+import "core:hash"
|
|
|
+import "core:math"
|
|
|
+import "core:math/rand"
|
|
|
+import "core:os"
|
|
|
+import "core:raw"
|
|
|
+import "core:sort"
|
|
|
+import "core:strings"
|
|
|
+import "core:types"
|
|
|
+import "core:unicode/utf16"
|
|
|
+import "core:unicode/utf8"
|
|
|
+
|
|
|
+import "core:atomics"
|
|
|
+import "core:thread"
|
|
|
+import "core:sys/win32"
|
|
|
+
|
|
|
+@(link_name="general_stuff")
|
|
|
+general_stuff :: proc() {
|
|
|
+ fmt.println("# general_stuff");
|
|
|
+ { // `do` for inline statements rather than block
|
|
|
+ foo :: proc() do fmt.println("Foo!");
|
|
|
+ if false do foo();
|
|
|
+ for false do foo();
|
|
|
+ when false do foo();
|
|
|
+
|
|
|
+ if false do foo();
|
|
|
+ else do foo();
|
|
|
+ }
|
|
|
+
|
|
|
+ { // Removal of `++` and `--` (again)
|
|
|
+ x: int;
|
|
|
+ x += 1;
|
|
|
+ x -= 1;
|
|
|
+ }
|
|
|
+ { // Casting syntaxes
|
|
|
+ i := i32(137);
|
|
|
+ ptr := &i;
|
|
|
+
|
|
|
+ _ = (^f32)(ptr);
|
|
|
+ // ^f32(ptr) == ^(f32(ptr))
|
|
|
+ _ = cast(^f32)ptr;
|
|
|
+
|
|
|
+ _ = (^f32)(ptr)^;
|
|
|
+ _ = (cast(^f32)ptr)^;
|
|
|
+
|
|
|
+ // Questions: Should there be two ways to do it?
|
|
|
+ }
|
|
|
+
|
|
|
+ /*
|
|
|
+ * Remove *_val_of built-in procedures
|
|
|
+ * size_of, align_of, offset_of
|
|
|
+ * type_of, type_info_of
|
|
|
+ */
|
|
|
+
|
|
|
+ { // `expand_to_tuple` built-in procedure
|
|
|
+ Foo :: struct {
|
|
|
+ x: int,
|
|
|
+ b: bool,
|
|
|
+ }
|
|
|
+ f := Foo{137, true};
|
|
|
+ x, b := expand_to_tuple(f);
|
|
|
+ fmt.println(f);
|
|
|
+ fmt.println(x, b);
|
|
|
+ fmt.println(expand_to_tuple(f));
|
|
|
+ }
|
|
|
+
|
|
|
+ {
|
|
|
+ // .. half-closed range
|
|
|
+ // ... open range
|
|
|
+
|
|
|
+ for in 0..2 {} // 0, 1
|
|
|
+ for in 0...2 {} // 0, 1, 2
|
|
|
+ }
|
|
|
+
|
|
|
+ { // Multiple sized booleans
|
|
|
+
|
|
|
+ x0: bool; // default
|
|
|
+ x1: b8 = true;
|
|
|
+ x2: b16 = false;
|
|
|
+ x3: b32 = true;
|
|
|
+ x4: b64 = false;
|
|
|
+
|
|
|
+ fmt.printf("x1: %T = %v;\n", x1, x1);
|
|
|
+ fmt.printf("x2: %T = %v;\n", x2, x2);
|
|
|
+ fmt.printf("x3: %T = %v;\n", x3, x3);
|
|
|
+ fmt.printf("x4: %T = %v;\n", x4, x4);
|
|
|
+
|
|
|
+ // Having specific sized booleans is very useful when dealing with foreign code
|
|
|
+ // and to enforce specific alignment for a boolean, especially within a struct
|
|
|
+ }
|
|
|
+
|
|
|
+ { // `distinct` types
|
|
|
+ // Originally, all type declarations would create a distinct type unless #type_alias was present.
|
|
|
+ // Now the behaviour has been reversed. All type declarations create a type alias unless `distinct` is present.
|
|
|
+ // If the type expression is `struct`, `union`, `enum`, `proc`, or `bit_field`, the types will always been distinct.
|
|
|
+
|
|
|
+ Int32 :: i32;
|
|
|
+ #assert(Int32 == i32);
|
|
|
+
|
|
|
+ My_Int32 :: distinct i32;
|
|
|
+ #assert(My_Int32 != i32);
|
|
|
+
|
|
|
+ My_Struct :: struct{x: int};
|
|
|
+ #assert(My_Struct != struct{x: int});
|
|
|
+ }
|
|
|
+}
|
|
|
+
|
|
|
+
|
|
|
+union_type :: proc() {
|
|
|
+ fmt.println("\n# union_type");
|
|
|
+ {
|
|
|
+ val: union{int, bool};
|
|
|
+ val = 137;
|
|
|
+ if i, ok := val.(int); ok {
|
|
|
+ fmt.println(i);
|
|
|
+ }
|
|
|
+ val = true;
|
|
|
+ fmt.println(val);
|
|
|
+
|
|
|
+ val = nil;
|
|
|
+
|
|
|
+ switch v in val {
|
|
|
+ case int: fmt.println("int", v);
|
|
|
+ case bool: fmt.println("bool", v);
|
|
|
+ case: fmt.println("nil");
|
|
|
+ }
|
|
|
+ }
|
|
|
+ {
|
|
|
+ // There is a duality between `any` and `union`
|
|
|
+ // An `any` has a pointer to the data and allows for any type (open)
|
|
|
+ // A `union` has as binary blob to store the data and allows only certain types (closed)
|
|
|
+ // The following code is with `any` but has the same syntax
|
|
|
+ val: any;
|
|
|
+ val = 137;
|
|
|
+ if i, ok := val.(int); ok {
|
|
|
+ fmt.println(i);
|
|
|
+ }
|
|
|
+ val = true;
|
|
|
+ fmt.println(val);
|
|
|
+
|
|
|
+ val = nil;
|
|
|
+
|
|
|
+ switch v in val {
|
|
|
+ case int: fmt.println("int", v);
|
|
|
+ case bool: fmt.println("bool", v);
|
|
|
+ case: fmt.println("nil");
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ Vector3 :: struct {x, y, z: f32};
|
|
|
+ Quaternion :: struct {x, y, z, w: f32};
|
|
|
+
|
|
|
+ // More realistic examples
|
|
|
+ {
|
|
|
+ // NOTE(bill): For the above basic examples, you may not have any
|
|
|
+ // particular use for it. However, my main use for them is not for these
|
|
|
+ // simple cases. My main use is for hierarchical types. Many prefer
|
|
|
+ // subtyping, embedding the base data into the derived types. Below is
|
|
|
+ // an example of this for a basic game Entity.
|
|
|
+
|
|
|
+ Entity :: struct {
|
|
|
+ id: u64,
|
|
|
+ name: string,
|
|
|
+ position: Vector3,
|
|
|
+ orientation: Quaternion,
|
|
|
+
|
|
|
+ derived: any,
|
|
|
+ }
|
|
|
+
|
|
|
+ Frog :: struct {
|
|
|
+ using entity: Entity,
|
|
|
+ jump_height: f32,
|
|
|
+ }
|
|
|
+
|
|
|
+ Monster :: struct {
|
|
|
+ using entity: Entity,
|
|
|
+ is_robot: bool,
|
|
|
+ is_zombie: bool,
|
|
|
+ }
|
|
|
+
|
|
|
+ // See `parametric_polymorphism` procedure for details
|
|
|
+ new_entity :: proc(T: type) -> ^Entity {
|
|
|
+ t := new(T);
|
|
|
+ t.derived = t^;
|
|
|
+ return t;
|
|
|
+ }
|
|
|
+
|
|
|
+ entity := new_entity(Monster);
|
|
|
+
|
|
|
+ switch e in entity.derived {
|
|
|
+ case Frog:
|
|
|
+ fmt.println("Ribbit");
|
|
|
+ case Monster:
|
|
|
+ if e.is_robot do fmt.println("Robotic");
|
|
|
+ if e.is_zombie do fmt.println("Grrrr!");
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ {
|
|
|
+ // NOTE(bill): A union can be used to achieve something similar. Instead
|
|
|
+ // of embedding the base data into the derived types, the derived data
|
|
|
+ // in embedded into the base type. Below is the same example of the
|
|
|
+ // basic game Entity but using an union.
|
|
|
+
|
|
|
+ Entity :: struct {
|
|
|
+ id: u64,
|
|
|
+ name: string,
|
|
|
+ position: Vector3,
|
|
|
+ orientation: Quaternion,
|
|
|
+
|
|
|
+ derived: union {Frog, Monster},
|
|
|
+ }
|
|
|
+
|
|
|
+ Frog :: struct {
|
|
|
+ using entity: ^Entity,
|
|
|
+ jump_height: f32,
|
|
|
+ }
|
|
|
+
|
|
|
+ Monster :: struct {
|
|
|
+ using entity: ^Entity,
|
|
|
+ is_robot: bool,
|
|
|
+ is_zombie: bool,
|
|
|
+ }
|
|
|
+
|
|
|
+ // See `parametric_polymorphism` procedure for details
|
|
|
+ new_entity :: proc(T: type) -> ^Entity {
|
|
|
+ t := new(Entity);
|
|
|
+ t.derived = T{entity = t};
|
|
|
+ return t;
|
|
|
+ }
|
|
|
+
|
|
|
+ entity := new_entity(Monster);
|
|
|
+
|
|
|
+ switch e in entity.derived {
|
|
|
+ case Frog:
|
|
|
+ fmt.println("Ribbit");
|
|
|
+ case Monster:
|
|
|
+ if e.is_robot do fmt.println("Robotic");
|
|
|
+ if e.is_zombie do fmt.println("Grrrr!");
|
|
|
+ }
|
|
|
+
|
|
|
+ // NOTE(bill): As you can see, the usage code has not changed, only its
|
|
|
+ // memory layout. Both approaches have their own advantages but they can
|
|
|
+ // be used together to achieve different results. The subtyping approach
|
|
|
+ // can allow for a greater control of the memory layout and memory
|
|
|
+ // allocation, e.g. storing the derivatives together. However, this is
|
|
|
+ // also its disadvantage. You must either preallocate arrays for each
|
|
|
+ // derivative separation (which can be easily missed) or preallocate a
|
|
|
+ // bunch of "raw" memory; determining the maximum size of the derived
|
|
|
+ // types would require the aid of metaprogramming. Unions solve this
|
|
|
+ // particular problem as the data is stored with the base data.
|
|
|
+ // Therefore, it is possible to preallocate, e.g. [100]Entity.
|
|
|
+
|
|
|
+ // It should be noted that the union approach can have the same memory
|
|
|
+ // layout as the any and with the same type restrictions by using a
|
|
|
+ // pointer type for the derivatives.
|
|
|
+
|
|
|
+ /*
|
|
|
+ Entity :: struct {
|
|
|
+ ...
|
|
|
+ derived: union{^Frog, ^Monster},
|
|
|
+ }
|
|
|
+
|
|
|
+ Frog :: struct {
|
|
|
+ using entity: Entity,
|
|
|
+ ...
|
|
|
+ }
|
|
|
+ Monster :: struct {
|
|
|
+ using entity: Entity,
|
|
|
+ ...
|
|
|
+
|
|
|
+ }
|
|
|
+ new_entity :: proc(T: type) -> ^Entity {
|
|
|
+ t := new(T);
|
|
|
+ t.derived = t;
|
|
|
+ return t;
|
|
|
+ }
|
|
|
+ */
|
|
|
+ }
|
|
|
+}
|
|
|
+
|
|
|
+parametric_polymorphism :: proc() {
|
|
|
+ fmt.println("# parametric_polymorphism");
|
|
|
+
|
|
|
+ print_value :: proc(value: $T) {
|
|
|
+ fmt.printf("print_value: %T %v\n", value, value);
|
|
|
+ }
|
|
|
+
|
|
|
+ v1: int = 1;
|
|
|
+ v2: f32 = 2.1;
|
|
|
+ v3: f64 = 3.14;
|
|
|
+ v4: string = "message";
|
|
|
+
|
|
|
+ print_value(v1);
|
|
|
+ print_value(v2);
|
|
|
+ print_value(v3);
|
|
|
+ print_value(v4);
|
|
|
+
|
|
|
+ fmt.println();
|
|
|
+
|
|
|
+ add :: proc(p, q: $T) -> T {
|
|
|
+ x: T = p + q;
|
|
|
+ return x;
|
|
|
+ }
|
|
|
+
|
|
|
+ a := add(3, 4);
|
|
|
+ fmt.printf("a: %T = %v\n", a, a);
|
|
|
+
|
|
|
+ b := add(3.2, 4.3);
|
|
|
+ fmt.printf("b: %T = %v\n", b, b);
|
|
|
+
|
|
|
+ // This is how `new` is implemented
|
|
|
+ alloc_type :: proc(T: type) -> ^T {
|
|
|
+ t := cast(^T)alloc(size_of(T), align_of(T));
|
|
|
+ t^ = T{}; // Use default initialization value
|
|
|
+ return t;
|
|
|
+ }
|
|
|
+
|
|
|
+ copy_slice :: proc(dst, src: []$T) -> int {
|
|
|
+ n := min(len(dst), len(src));
|
|
|
+ if n > 0 {
|
|
|
+ mem.copy(&dst[0], &src[0], n*size_of(T));
|
|
|
+ }
|
|
|
+ return n;
|
|
|
+ }
|
|
|
+
|
|
|
+ double_params :: proc(a: $A, b: $B) -> A {
|
|
|
+ return a + A(b);
|
|
|
+ }
|
|
|
+
|
|
|
+ fmt.println(double_params(12, 1.345));
|
|
|
+
|
|
|
+
|
|
|
+
|
|
|
+ { // Polymorphic Types and Type Specialization
|
|
|
+ Table_Slot :: struct(Key, Value: type) {
|
|
|
+ occupied: bool,
|
|
|
+ hash: u32,
|
|
|
+ key: Key,
|
|
|
+ value: Value,
|
|
|
+ }
|
|
|
+ TABLE_SIZE_MIN :: 32;
|
|
|
+ Table :: struct(Key, Value: type) {
|
|
|
+ count: int,
|
|
|
+ allocator: Allocator,
|
|
|
+ slots: []Table_Slot(Key, Value),
|
|
|
+ }
|
|
|
+
|
|
|
+ // Only allow types that are specializations of a (polymorphic) slice
|
|
|
+ make_slice :: proc(T: type/[]$E, len: int) -> T {
|
|
|
+ return make(T, len);
|
|
|
+ }
|
|
|
+
|
|
|
+
|
|
|
+ // Only allow types that are specializations of `Table`
|
|
|
+ allocate :: proc(table: ^$T/Table, capacity: int) {
|
|
|
+ c := context;
|
|
|
+ if table.allocator.procedure != nil do c.allocator = table.allocator;
|
|
|
+
|
|
|
+ context <- c {
|
|
|
+ table.slots = make_slice(type_of(table.slots), max(capacity, TABLE_SIZE_MIN));
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ expand :: proc(table: ^$T/Table) {
|
|
|
+ c := context;
|
|
|
+ if table.allocator.procedure != nil do c.allocator = table.allocator;
|
|
|
+
|
|
|
+ context <- c {
|
|
|
+ old_slots := table.slots;
|
|
|
+
|
|
|
+ cap := max(2*len(table.slots), TABLE_SIZE_MIN);
|
|
|
+ allocate(table, cap);
|
|
|
+
|
|
|
+ for s in old_slots do if s.occupied {
|
|
|
+ put(table, s.key, s.value);
|
|
|
+ }
|
|
|
+
|
|
|
+ free(old_slots);
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ // Polymorphic determination of a polymorphic struct
|
|
|
+ // put :: proc(table: ^$T/Table, key: T.Key, value: T.Value) {
|
|
|
+ put :: proc(table: ^Table($Key, $Value), key: Key, value: Value) {
|
|
|
+ hash := get_hash(key); // Ad-hoc method which would fail in a different scope
|
|
|
+ index := find_index(table, key, hash);
|
|
|
+ if index < 0 {
|
|
|
+ if f64(table.count) >= 0.75*f64(len(table.slots)) {
|
|
|
+ expand(table);
|
|
|
+ }
|
|
|
+ assert(table.count <= len(table.slots));
|
|
|
+
|
|
|
+ hash := get_hash(key);
|
|
|
+ index = int(hash % u32(len(table.slots)));
|
|
|
+
|
|
|
+ for table.slots[index].occupied {
|
|
|
+ if index += 1; index >= len(table.slots) {
|
|
|
+ index = 0;
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ table.count += 1;
|
|
|
+ }
|
|
|
+
|
|
|
+ slot := &table.slots[index];
|
|
|
+ slot.occupied = true;
|
|
|
+ slot.hash = hash;
|
|
|
+ slot.key = key;
|
|
|
+ slot.value = value;
|
|
|
+ }
|
|
|
+
|
|
|
+
|
|
|
+ // find :: proc(table: ^$T/Table, key: T.Key) -> (T.Value, bool) {
|
|
|
+ find :: proc(table: ^Table($Key, $Value), key: Key) -> (Value, bool) {
|
|
|
+ hash := get_hash(key);
|
|
|
+ index := find_index(table, key, hash);
|
|
|
+ if index < 0 {
|
|
|
+ return Value{}, false;
|
|
|
+ }
|
|
|
+ return table.slots[index].value, true;
|
|
|
+ }
|
|
|
+
|
|
|
+ find_index :: proc(table: ^Table($Key, $Value), key: Key, hash: u32) -> int {
|
|
|
+ if len(table.slots) <= 0 do return -1;
|
|
|
+
|
|
|
+ index := int(hash % u32(len(table.slots)));
|
|
|
+ for table.slots[index].occupied {
|
|
|
+ if table.slots[index].hash == hash {
|
|
|
+ if table.slots[index].key == key {
|
|
|
+ return index;
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ if index += 1; index >= len(table.slots) {
|
|
|
+ index = 0;
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ return -1;
|
|
|
+ }
|
|
|
+
|
|
|
+ get_hash :: proc(s: string) -> u32 { // fnv32a
|
|
|
+ h: u32 = 0x811c9dc5;
|
|
|
+ for i in 0..len(s) {
|
|
|
+ h = (h ~ u32(s[i])) * 0x01000193;
|
|
|
+ }
|
|
|
+ return h;
|
|
|
+ }
|
|
|
+
|
|
|
+
|
|
|
+ table: Table(string, int);
|
|
|
+
|
|
|
+ for i in 0..36 do put(&table, "Hellope", i);
|
|
|
+ for i in 0..42 do put(&table, "World!", i);
|
|
|
+
|
|
|
+ found, _ := find(&table, "Hellope");
|
|
|
+ fmt.printf("`found` is %v\n", found);
|
|
|
+
|
|
|
+ found, _ = find(&table, "World!");
|
|
|
+ fmt.printf("`found` is %v\n", found);
|
|
|
+
|
|
|
+ // I would not personally design a hash table like this in production
|
|
|
+ // but this is a nice basic example
|
|
|
+ // A better approach would either use a `u64` or equivalent for the key
|
|
|
+ // and let the user specify the hashing function or make the user store
|
|
|
+ // the hashing procedure with the table
|
|
|
+ }
|
|
|
+}
|
|
|
+
|
|
|
+
|
|
|
+
|
|
|
+
|
|
|
+prefix_table := [?]string{
|
|
|
+ "White",
|
|
|
+ "Red",
|
|
|
+ "Green",
|
|
|
+ "Blue",
|
|
|
+ "Octarine",
|
|
|
+ "Black",
|
|
|
+};
|
|
|
+
|
|
|
+threading_example :: proc() {
|
|
|
+ when ODIN_OS == "windows" {
|
|
|
+ fmt.println("# threading_example");
|
|
|
+
|
|
|
+ unordered_remove :: proc(array: ^[dynamic]$T, index: int, loc := #caller_location) {
|
|
|
+ __bounds_check_error_loc(loc, index, len(array));
|
|
|
+ array[index] = array[len(array)-1];
|
|
|
+ pop(array);
|
|
|
+ }
|
|
|
+ ordered_remove :: proc(array: ^[dynamic]$T, index: int, loc := #caller_location) {
|
|
|
+ __bounds_check_error_loc(loc, index, len(array));
|
|
|
+ copy(array[index..], array[index+1..]);
|
|
|
+ pop(array);
|
|
|
+ }
|
|
|
+
|
|
|
+ worker_proc :: proc(t: ^thread.Thread) -> int {
|
|
|
+ for iteration in 1...5 {
|
|
|
+ fmt.printf("Thread %d is on iteration %d\n", t.user_index, iteration);
|
|
|
+ fmt.printf("`%s`: iteration %d\n", prefix_table[t.user_index], iteration);
|
|
|
+ // win32.sleep(1);
|
|
|
+ }
|
|
|
+ return 0;
|
|
|
+ }
|
|
|
+
|
|
|
+ threads := make([dynamic]^thread.Thread, 0, len(prefix_table));
|
|
|
+ defer free(threads);
|
|
|
+
|
|
|
+ for in prefix_table {
|
|
|
+ if t := thread.create(worker_proc); t != nil {
|
|
|
+ t.init_context = context;
|
|
|
+ t.use_init_context = true;
|
|
|
+ t.user_index = len(threads);
|
|
|
+ append(&threads, t);
|
|
|
+ thread.start(t);
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ for len(threads) > 0 {
|
|
|
+ for i := 0; i < len(threads); /**/ {
|
|
|
+ if t := threads[i]; thread.is_done(t) {
|
|
|
+ fmt.printf("Thread %d is done\n", t.user_index);
|
|
|
+ thread.destroy(t);
|
|
|
+
|
|
|
+ ordered_remove(&threads, i);
|
|
|
+ } else {
|
|
|
+ i += 1;
|
|
|
+ }
|
|
|
+ }
|
|
|
+ }
|
|
|
+ }
|
|
|
+}
|
|
|
+
|
|
|
+array_programming :: proc() {
|
|
|
+ fmt.println("# array_programming");
|
|
|
+ {
|
|
|
+ a := [3]f32{1, 2, 3};
|
|
|
+ b := [3]f32{5, 6, 7};
|
|
|
+ c := a * b;
|
|
|
+ d := a + b;
|
|
|
+ e := 1 + (c - d) / 2;
|
|
|
+ fmt.printf("%.1f\n", e); // [0.5, 3.0, 6.5]
|
|
|
+ }
|
|
|
+
|
|
|
+ {
|
|
|
+ a := [3]f32{1, 2, 3};
|
|
|
+ b := swizzle(a, 2, 1, 0);
|
|
|
+ assert(b == [3]f32{3, 2, 1});
|
|
|
+
|
|
|
+ c := swizzle(a, 0, 0);
|
|
|
+ assert(c == [2]f32{1, 1});
|
|
|
+ assert(c == 1);
|
|
|
+ }
|
|
|
+
|
|
|
+ {
|
|
|
+ Vector3 :: distinct [3]f32;
|
|
|
+ a := Vector3{1, 2, 3};
|
|
|
+ b := Vector3{5, 6, 7};
|
|
|
+ c := (a * b)/2 + 1;
|
|
|
+ d := c.x + c.y + c.z;
|
|
|
+ fmt.printf("%.1f\n", d); // 22.0
|
|
|
+
|
|
|
+ cross :: proc(a, b: Vector3) -> Vector3 {
|
|
|
+ i := swizzle(a, 1, 2, 0) * swizzle(b, 2, 0, 1);
|
|
|
+ j := swizzle(a, 2, 0, 1) * swizzle(b, 1, 2, 0);
|
|
|
+ return i - j;
|
|
|
+ }
|
|
|
+
|
|
|
+ blah :: proc(a: Vector3) -> f32 {
|
|
|
+ return a.x + a.y + a.z;
|
|
|
+ }
|
|
|
+
|
|
|
+ x := cross(a, b);
|
|
|
+ fmt.println(x);
|
|
|
+ fmt.println(blah(x));
|
|
|
+ }
|
|
|
+}
|
|
|
+
|
|
|
+
|
|
|
+using println in import "core:fmt"
|
|
|
+
|
|
|
+using_in :: proc() {
|
|
|
+ fmt.println("# using in");
|
|
|
+ using print in fmt;
|
|
|
+
|
|
|
+ println("Hellope1");
|
|
|
+ print("Hellope2\n");
|
|
|
+
|
|
|
+ Foo :: struct {
|
|
|
+ x, y: int,
|
|
|
+ b: bool,
|
|
|
+ }
|
|
|
+ f: Foo;
|
|
|
+ f.x, f.y = 123, 321;
|
|
|
+ println(f);
|
|
|
+ using x, y in f;
|
|
|
+ x, y = 456, 654;
|
|
|
+ println(f);
|
|
|
+}
|
|
|
+
|
|
|
+named_proc_return_parameters :: proc() {
|
|
|
+ fmt.println("# named proc return parameters");
|
|
|
+
|
|
|
+ foo0 :: proc() -> int {
|
|
|
+ return 123;
|
|
|
+ }
|
|
|
+ foo1 :: proc() -> (a: int) {
|
|
|
+ a = 123;
|
|
|
+ return;
|
|
|
+ }
|
|
|
+ foo2 :: proc() -> (a, b: int) {
|
|
|
+ // Named return values act like variables within the scope
|
|
|
+ a = 321;
|
|
|
+ b = 567;
|
|
|
+ return b, a;
|
|
|
+ }
|
|
|
+ fmt.println("foo0 =", foo0()); // 123
|
|
|
+ fmt.println("foo1 =", foo1()); // 123
|
|
|
+ fmt.println("foo2 =", foo2()); // 567 321
|
|
|
+}
|
|
|
+
|
|
|
+
|
|
|
+enum_export :: proc() {
|
|
|
+ fmt.println("# enum #export");
|
|
|
+
|
|
|
+ Foo :: enum #export {A, B, C};
|
|
|
+
|
|
|
+ f0 := A;
|
|
|
+ f1 := B;
|
|
|
+ f2 := C;
|
|
|
+ fmt.println(f0, f1, f2);
|
|
|
+}
|
|
|
+
|
|
|
+explicit_procedure_overloading :: proc() {
|
|
|
+ fmt.println("# explicit procedure overloading");
|
|
|
+
|
|
|
+ add_ints :: proc(a, b: int) -> int {
|
|
|
+ x := a + b;
|
|
|
+ fmt.println("add_ints", x);
|
|
|
+ return x;
|
|
|
+ }
|
|
|
+ add_floats :: proc(a, b: f32) -> f32 {
|
|
|
+ x := a + b;
|
|
|
+ fmt.println("add_floats", x);
|
|
|
+ return x;
|
|
|
+ }
|
|
|
+ add_numbers :: proc(a: int, b: f32, c: u8) -> int {
|
|
|
+ x := int(a) + int(b) + int(c);
|
|
|
+ fmt.println("add_numbers", x);
|
|
|
+ return x;
|
|
|
+ }
|
|
|
+
|
|
|
+ add :: proc[add_ints, add_floats, add_numbers];
|
|
|
+
|
|
|
+ add(int(1), int(2));
|
|
|
+ add(f32(1), f32(2));
|
|
|
+ add(int(1), f32(2), u8(3));
|
|
|
+
|
|
|
+ add(1, 2); // untyped ints coerce to int tighter than f32
|
|
|
+ add(1.0, 2.0); // untyped floats coerce to f32 tighter than int
|
|
|
+ add(1, 2, 3); // three parameters
|
|
|
+
|
|
|
+ // Ambiguous answers
|
|
|
+ // add(1.0, 2);
|
|
|
+ // add(1, 2.0);
|
|
|
+}
|
|
|
+
|
|
|
+complete_switch :: proc() {
|
|
|
+ fmt.println("# complete_switch");
|
|
|
+ { // enum
|
|
|
+ Foo :: enum #export {
|
|
|
+ A,
|
|
|
+ B,
|
|
|
+ C,
|
|
|
+ D,
|
|
|
+ }
|
|
|
+
|
|
|
+ b := Foo.B;
|
|
|
+ f := Foo.A;
|
|
|
+ #complete switch f {
|
|
|
+ case A: fmt.println("A");
|
|
|
+ case B: fmt.println("B");
|
|
|
+ case C: fmt.println("C");
|
|
|
+ case D: fmt.println("D");
|
|
|
+ case: fmt.println("?");
|
|
|
+ }
|
|
|
+ }
|
|
|
+ { // union
|
|
|
+ Foo :: union {int, bool};
|
|
|
+ f: Foo = 123;
|
|
|
+ #complete switch in f {
|
|
|
+ case int: fmt.println("int");
|
|
|
+ case bool: fmt.println("bool");
|
|
|
+ case:
|
|
|
+ }
|
|
|
+ }
|
|
|
+}
|
|
|
+
|
|
|
+
|
|
|
+cstring_example :: proc() {
|
|
|
+ W :: "Hellope";
|
|
|
+ X :: cstring(W);
|
|
|
+ Y :: string(X);
|
|
|
+
|
|
|
+ w := W;
|
|
|
+ x: cstring = X;
|
|
|
+ y: string = Y;
|
|
|
+ z := string(x);
|
|
|
+ fmt.println(x, y, z);
|
|
|
+ fmt.println(len(x), len(y), len(z));
|
|
|
+ fmt.println(len(W), len(X), len(Y));
|
|
|
+ // IMPORTANT NOTE for cstring variables
|
|
|
+ // len(cstring) is O(N)
|
|
|
+ // cast(cstring)string is O(N)
|
|
|
+}
|
|
|
+
|
|
|
+deprecated_attribute :: proc() {
|
|
|
+ @(deprecated="Use foo_v2 instead")
|
|
|
+ foo_v1 :: proc(x: int) {
|
|
|
+ fmt.println("foo_v1");
|
|
|
+ }
|
|
|
+ foo_v2 :: proc(x: int) {
|
|
|
+ fmt.println("foo_v2");
|
|
|
+ }
|
|
|
+
|
|
|
+ // NOTE: Uncomment to see the warning messages
|
|
|
+ // foo_v1(1);
|
|
|
+}
|
|
|
+
|
|
|
+
|
|
|
+main :: proc() {
|
|
|
+ fmt.println("HERE\n");
|
|
|
+ when true {
|
|
|
+ general_stuff();
|
|
|
+ union_type();
|
|
|
+ parametric_polymorphism();
|
|
|
+ threading_example();
|
|
|
+ array_programming();
|
|
|
+ using_in();
|
|
|
+ named_proc_return_parameters();
|
|
|
+ enum_export();
|
|
|
+ explicit_procedure_overloading();
|
|
|
+ complete_switch();
|
|
|
+ cstring_example();
|
|
|
+ deprecated_attribute();
|
|
|
+ }
|
|
|
+}
|