Merge branch 'master' into syscall-fix

core/bufio/reader.odin

@@ -29,12 +29,12 @@ MIN_READ_BUFFER_SIZE :: 16
 @(private)
 DEFAULT_MAX_CONSECUTIVE_EMPTY_READS :: 128
 
-reader_init :: proc(b: ^Reader, rd: io.Reader, size: int = DEFAULT_BUF_SIZE, allocator := context.allocator) {
+reader_init :: proc(b: ^Reader, rd: io.Reader, size: int = DEFAULT_BUF_SIZE, allocator := context.allocator, loc := #caller_location) {
 	size := size
 	size = max(size, MIN_READ_BUFFER_SIZE)
 	reader_reset(b, rd)
 	b.buf_allocator = allocator
-	b.buf = make([]byte, size, allocator)
+	b.buf = make([]byte, size, allocator, loc)
 }
 
 reader_init_with_buf :: proc(b: ^Reader, rd: io.Reader, buf: []byte) {
@@ -81,7 +81,7 @@ _reader_read_new_chunk :: proc(b: ^Reader) -> io.Error {
 	for i := b.max_consecutive_empty_reads; i > 0; i -= 1 {
 		n, err := io.read(b.rd, b.buf[b.w:])
 		if n < 0 {
-			return .Negative_Read
+			return err if err != nil else .Negative_Read
 		}
 		b.w += n
 		if err != nil {
@@ -189,7 +189,7 @@ reader_read :: proc(b: ^Reader, p: []byte) -> (n: int, err: io.Error) {
 		if len(p) >= len(b.buf) {
 			n, b.err = io.read(b.rd, p)
 			if n < 0 {
-				return 0, .Negative_Read
+				return 0, b.err if b.err != nil else .Negative_Read
 			}
 
 			if n > 0 {
@@ -202,7 +202,7 @@ reader_read :: proc(b: ^Reader, p: []byte) -> (n: int, err: io.Error) {
 		b.r, b.w = 0, 0
 		n, b.err = io.read(b.rd, b.buf)
 		if n < 0 {
-			return 0, .Negative_Read
+			return 0, b.err if b.err != nil else .Negative_Read
 		}
 		if n == 0 {
 			return 0, _reader_consume_err(b)
@@ -290,7 +290,7 @@ reader_write_to :: proc(b: ^Reader, w: io.Writer) -> (n: i64, err: io.Error) {
 	write_buf :: proc(b: ^Reader, w: io.Writer) -> (i64, io.Error) {
 		n, err := io.write(w, b.buf[b.r:b.w])
 		if n < 0 {
-			return 0, .Negative_Write
+			return 0, err if err != nil else .Negative_Write
 		}
 		b.r += n
 		return i64(n), err

core/bufio/writer.odin

@@ -95,6 +95,10 @@ writer_write :: proc(b: ^Writer, p: []byte) -> (n: int, err: io.Error) {
 		m: int
 		if writer_buffered(b) == 0 {
 			m, b.err = io.write(b.wr, p)
+			if m < 0 && b.err == nil {
+				b.err = .Negative_Write
+				break
+			}
 		} else {
 			m = copy(b.buf[b.n:], p)
 			b.n += m

core/bytes/buffer.odin

@@ -27,19 +27,19 @@ Read_Op :: enum i8 {
 }
 
 
-buffer_init :: proc(b: ^Buffer, buf: []byte) {
-	resize(&b.buf, len(buf))
+buffer_init :: proc(b: ^Buffer, buf: []byte, loc := #caller_location) {
+	resize(&b.buf, len(buf), loc=loc)
 	copy(b.buf[:], buf)
 }
 
-buffer_init_string :: proc(b: ^Buffer, s: string) {
-	resize(&b.buf, len(s))
+buffer_init_string :: proc(b: ^Buffer, s: string, loc := #caller_location) {
+	resize(&b.buf, len(s), loc=loc)
 	copy(b.buf[:], s)
 }
 
-buffer_init_allocator :: proc(b: ^Buffer, len, cap: int, allocator := context.allocator) {
+buffer_init_allocator :: proc(b: ^Buffer, len, cap: int, allocator := context.allocator, loc := #caller_location) {
 	if b.buf == nil {
-		b.buf = make([dynamic]byte, len, cap, allocator)
+		b.buf = make([dynamic]byte, len, cap, allocator, loc)
 		return
 	}
 
@@ -96,28 +96,28 @@ buffer_truncate :: proc(b: ^Buffer, n: int) {
 }
 
 @(private)
-_buffer_try_grow :: proc(b: ^Buffer, n: int) -> (int, bool) {
+_buffer_try_grow :: proc(b: ^Buffer, n: int, loc := #caller_location) -> (int, bool) {
 	if l := len(b.buf); n <= cap(b.buf)-l {
-		resize(&b.buf, l+n)
+		resize(&b.buf, l+n, loc=loc)
 		return l, true
 	}
 	return 0, false
 }
 
 @(private)
-_buffer_grow :: proc(b: ^Buffer, n: int) -> int {
+_buffer_grow :: proc(b: ^Buffer, n: int, loc := #caller_location) -> int {
 	m := buffer_length(b)
 	if m == 0 && b.off != 0 {
 		buffer_reset(b)
 	}
-	if i, ok := _buffer_try_grow(b, n); ok {
+	if i, ok := _buffer_try_grow(b, n, loc=loc); ok {
 		return i
 	}
 
 	if b.buf == nil && n <= SMALL_BUFFER_SIZE {
 		// Fixes #2756 by preserving allocator if already set on Buffer via init_buffer_allocator
-		reserve(&b.buf, SMALL_BUFFER_SIZE)
-		resize(&b.buf, n)
+		reserve(&b.buf, SMALL_BUFFER_SIZE, loc=loc)
+		resize(&b.buf, n, loc=loc)
 		return 0
 	}
 
@@ -127,31 +127,31 @@ _buffer_grow :: proc(b: ^Buffer, n: int) -> int {
 	} else if c > max(int) - c - n {
 		panic("bytes.Buffer: too large")
 	} else {
-		resize(&b.buf, 2*c + n)
+		resize(&b.buf, 2*c + n, loc=loc)
 		copy(b.buf[:], b.buf[b.off:])
 	}
 	b.off = 0
-	resize(&b.buf, m+n)
+	resize(&b.buf, m+n, loc=loc)
 	return m
 }
 
-buffer_grow :: proc(b: ^Buffer, n: int) {
+buffer_grow :: proc(b: ^Buffer, n: int, loc := #caller_location) {
 	if n < 0 {
 		panic("bytes.buffer_grow: negative count")
 	}
-	m := _buffer_grow(b, n)
-	resize(&b.buf, m)
+	m := _buffer_grow(b, n, loc=loc)
+	resize(&b.buf, m, loc=loc)
 }
 
-buffer_write_at :: proc(b: ^Buffer, p: []byte, offset: int) -> (n: int, err: io.Error) {
+buffer_write_at :: proc(b: ^Buffer, p: []byte, offset: int, loc := #caller_location) -> (n: int, err: io.Error) {
 	b.last_read = .Invalid
 	if offset < 0 {
 		err = .Invalid_Offset
 		return
 	}
-	_, ok := _buffer_try_grow(b, offset+len(p))
+	_, ok := _buffer_try_grow(b, offset+len(p), loc=loc)
 	if !ok {
-		_ = _buffer_grow(b, offset+len(p))
+		_ = _buffer_grow(b, offset+len(p), loc=loc)
 	}
 	if len(b.buf) <= offset {
 		return 0, .Short_Write
@@ -160,47 +160,47 @@ buffer_write_at :: proc(b: ^Buffer, p: []byte, offset: int) -> (n: int, err: io.
 }
 
 
-buffer_write :: proc(b: ^Buffer, p: []byte) -> (n: int, err: io.Error) {
+buffer_write :: proc(b: ^Buffer, p: []byte, loc := #caller_location) -> (n: int, err: io.Error) {
 	b.last_read = .Invalid
-	m, ok := _buffer_try_grow(b, len(p))
+	m, ok := _buffer_try_grow(b, len(p), loc=loc)
 	if !ok {
-		m = _buffer_grow(b, len(p))
+		m = _buffer_grow(b, len(p), loc=loc)
 	}
 	return copy(b.buf[m:], p), nil
 }
 
-buffer_write_ptr :: proc(b: ^Buffer, ptr: rawptr, size: int) -> (n: int, err: io.Error) {
-	return buffer_write(b, ([^]byte)(ptr)[:size])
+buffer_write_ptr :: proc(b: ^Buffer, ptr: rawptr, size: int, loc := #caller_location) -> (n: int, err: io.Error) {
+	return buffer_write(b, ([^]byte)(ptr)[:size], loc=loc)
 }
 
-buffer_write_string :: proc(b: ^Buffer, s: string) -> (n: int, err: io.Error) {
+buffer_write_string :: proc(b: ^Buffer, s: string, loc := #caller_location) -> (n: int, err: io.Error) {
 	b.last_read = .Invalid
-	m, ok := _buffer_try_grow(b, len(s))
+	m, ok := _buffer_try_grow(b, len(s), loc=loc)
 	if !ok {
-		m = _buffer_grow(b, len(s))
+		m = _buffer_grow(b, len(s), loc=loc)
 	}
 	return copy(b.buf[m:], s), nil
 }
 
-buffer_write_byte :: proc(b: ^Buffer, c: byte) -> io.Error {
+buffer_write_byte :: proc(b: ^Buffer, c: byte, loc := #caller_location) -> io.Error {
 	b.last_read = .Invalid
-	m, ok := _buffer_try_grow(b, 1)
+	m, ok := _buffer_try_grow(b, 1, loc=loc)
 	if !ok {
-		m = _buffer_grow(b, 1)
+		m = _buffer_grow(b, 1, loc=loc)
 	}
 	b.buf[m] = c
 	return nil
 }
 
-buffer_write_rune :: proc(b: ^Buffer, r: rune) -> (n: int, err: io.Error) {
+buffer_write_rune :: proc(b: ^Buffer, r: rune, loc := #caller_location) -> (n: int, err: io.Error) {
 	if r < utf8.RUNE_SELF {
-		buffer_write_byte(b, byte(r))
+		buffer_write_byte(b, byte(r), loc=loc)
 		return 1, nil
 	}
 	b.last_read = .Invalid
-	m, ok := _buffer_try_grow(b, utf8.UTF_MAX)
+	m, ok := _buffer_try_grow(b, utf8.UTF_MAX, loc=loc)
 	if !ok {
-		m = _buffer_grow(b, utf8.UTF_MAX)
+		m = _buffer_grow(b, utf8.UTF_MAX, loc=loc)
 	}
 	res: [4]byte
 	res, n = utf8.encode_rune(r)
@@ -359,7 +359,7 @@ buffer_read_from :: proc(b: ^Buffer, r: io.Reader) -> (n: i64, err: io.Error) #n
 		resize(&b.buf, i)
 		m, e := io.read(r, b.buf[i:cap(b.buf)])
 		if m < 0 {
-			err = .Negative_Read
+			err = e if e != nil else .Negative_Read
 			return
 		}
 

core/bytes/bytes.odin

@@ -1167,3 +1167,28 @@ fields_proc :: proc(s: []byte, f: proc(rune) -> bool, allocator := context.alloc
 
 	return subslices[:]
 }
+
+// alias returns true iff a and b have a non-zero length, and any part of
+// a overlaps with b.
+alias :: proc "contextless" (a, b: []byte) -> bool {
+	a_len, b_len := len(a), len(b)
+	if a_len == 0 || b_len == 0 {
+		return false
+	}
+
+	a_start, b_start := uintptr(raw_data(a)), uintptr(raw_data(b))
+	a_end, b_end := a_start + uintptr(a_len-1), b_start + uintptr(b_len-1)
+
+	return a_start <= b_end && b_start <= a_end
+}
+
+// alias_inexactly returns true iff a and b have a non-zero length,
+// the base pointer of a and b are NOT equal, and any part of a overlaps
+// with b (ie: `alias(a, b)` with an exception that returns false for
+// `a == b`, `b = a[:len(a)-69]` and similar conditions).
+alias_inexactly :: proc "contextless" (a, b: []byte) -> bool {
+	if raw_data(a) == raw_data(b) {
+		return false
+	}
+	return alias(a, b)
+}

core/c/libc/errno.odin

@@ -40,7 +40,7 @@ when ODIN_OS == .FreeBSD {
 	ERANGE :: 34
 }
 
-when ODIN_OS == .OpenBSD {
+when ODIN_OS == .OpenBSD || ODIN_OS == .NetBSD {
 	@(private="file")
 	@(default_calling_convention="c")
 	foreign libc {

core/c/libc/signal.odin

@@ -34,20 +34,7 @@ when ODIN_OS == .Windows {
 	SIGTERM :: 15
 }
 
-when ODIN_OS == .Linux || ODIN_OS == .FreeBSD {
-	SIG_ERR  :: rawptr(~uintptr(0))
-	SIG_DFL  :: rawptr(uintptr(0))
-	SIG_IGN  :: rawptr(uintptr(1)) 
-
-	SIGABRT  :: 6
-	SIGFPE   :: 8
-	SIGILL   :: 4
-	SIGINT   :: 2
-	SIGSEGV  :: 11
-	SIGTERM  :: 15
-}
-
-when ODIN_OS == .Darwin {
+when ODIN_OS == .Linux || ODIN_OS == .FreeBSD || ODIN_OS == .Haiku || ODIN_OS == .OpenBSD || ODIN_OS == .NetBSD || ODIN_OS == .Darwin {
 	SIG_ERR  :: rawptr(~uintptr(0))
 	SIG_DFL  :: rawptr(uintptr(0))
 	SIG_IGN  :: rawptr(uintptr(1)) 

core/c/libc/stdio.odin

@@ -83,7 +83,7 @@ when ODIN_OS == .Linux {
 	}
 }
 
-when ODIN_OS == .OpenBSD {
+when ODIN_OS == .OpenBSD || ODIN_OS == .NetBSD {
 	fpos_t :: distinct i64
 
 	_IOFBF :: 0
@@ -102,10 +102,12 @@ when ODIN_OS == .OpenBSD {
 	SEEK_END :: 2
 
 	foreign libc {
-		stderr: ^FILE
-		stdin:  ^FILE
-		stdout: ^FILE
+		__sF: [3]FILE
 	}
+
+	stdin:  ^FILE = &__sF[0]
+	stdout: ^FILE = &__sF[1]
+	stderr: ^FILE = &__sF[2]
 }
 
 when ODIN_OS == .FreeBSD {
@@ -127,9 +129,9 @@ when ODIN_OS == .FreeBSD {
 	SEEK_END :: 2
 
 	foreign libc {
-		stderr: ^FILE
-		stdin:  ^FILE
-		stdout: ^FILE
+		@(link_name="__stderrp") stderr: ^FILE
+		@(link_name="__stdinp")  stdin:  ^FILE
+		@(link_name="__stdoutp") stdout: ^FILE
 	}
 }
 

core/c/libc/time.odin

@@ -45,7 +45,7 @@ when ODIN_OS == .Windows {
 	}
 }
 
-when ODIN_OS == .Linux || ODIN_OS == .FreeBSD || ODIN_OS == .Darwin || ODIN_OS == .OpenBSD || ODIN_OS == .Haiku {
+when ODIN_OS == .Linux || ODIN_OS == .FreeBSD || ODIN_OS == .Darwin || ODIN_OS == .OpenBSD || ODIN_OS == .NetBSD || ODIN_OS == .Haiku {
 	@(default_calling_convention="c")
 	foreign libc {
 		// 7.27.2 Time manipulation functions

core/c/libc/wctype.odin

@@ -22,7 +22,7 @@ when ODIN_OS == .Windows {
 	wctrans_t :: distinct int
 	wctype_t  :: distinct u32
 
-} else when ODIN_OS == .OpenBSD {
+} else when ODIN_OS == .OpenBSD || ODIN_OS == .NetBSD {
 	wctrans_t :: distinct rawptr
 	wctype_t  :: distinct rawptr
 

core/compress/common.odin

@@ -34,13 +34,13 @@ COMPRESS_OUTPUT_ALLOCATE_MIN :: int(#config(COMPRESS_OUTPUT_ALLOCATE_MIN, 1 << 2
 */
 when size_of(uintptr) == 8 {
 
-        // For 64-bit platforms, we set the default max buffer size to 4 GiB,
-        // which is GZIP and PKZIP's max payload size.
+	// For 64-bit platforms, we set the default max buffer size to 4 GiB,
+	// which is GZIP and PKZIP's max payload size.
 	COMPRESS_OUTPUT_ALLOCATE_MAX :: int(#config(COMPRESS_OUTPUT_ALLOCATE_MAX, 1 << 32))
 } else {
 	
 	// For 32-bit platforms, we set the default max buffer size to 512 MiB.
-        COMPRESS_OUTPUT_ALLOCATE_MAX :: int(#config(COMPRESS_OUTPUT_ALLOCATE_MAX, 1 << 29))
+	COMPRESS_OUTPUT_ALLOCATE_MAX :: int(#config(COMPRESS_OUTPUT_ALLOCATE_MAX, 1 << 29))
 }
 
 
@@ -186,7 +186,7 @@ input_size_from_stream :: proc(z: ^Context_Stream_Input) -> (res: i64, err: Erro
 
 input_size :: proc{input_size_from_memory, input_size_from_stream}
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_slice_from_memory :: #force_inline proc(z: ^Context_Memory_Input, size: int) -> (res: []u8, err: io.Error) {
 	#no_bounds_check {
 		if len(z.input_data) >= size {
@@ -203,7 +203,7 @@ read_slice_from_memory :: #force_inline proc(z: ^Context_Memory_Input, size: int
 	}
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_slice_from_stream :: #force_inline proc(z: ^Context_Stream_Input, size: int) -> (res: []u8, err: io.Error) {
 	// TODO: REMOVE ALL USE OF context.temp_allocator here
 	// there is literally no need for it
@@ -214,13 +214,13 @@ read_slice_from_stream :: #force_inline proc(z: ^Context_Stream_Input, size: int
 
 read_slice :: proc{read_slice_from_memory, read_slice_from_stream}
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_data :: #force_inline proc(z: ^$C, $T: typeid) -> (res: T, err: io.Error) {
 	b := read_slice(z, size_of(T)) or_return
 	return (^T)(&b[0])^, nil
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_u8_from_memory :: #force_inline proc(z: ^Context_Memory_Input) -> (res: u8, err: io.Error) {
 	#no_bounds_check {
 		if len(z.input_data) >= 1 {
@@ -232,7 +232,7 @@ read_u8_from_memory :: #force_inline proc(z: ^Context_Memory_Input) -> (res: u8,
 	return 0, .EOF
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_u8_from_stream :: #force_inline proc(z: ^Context_Stream_Input) -> (res: u8, err: io.Error) {
 	b := read_slice_from_stream(z, 1) or_return
 	return b[0], nil
@@ -242,7 +242,7 @@ read_u8 :: proc{read_u8_from_memory, read_u8_from_stream}
 
 // You would typically only use this at the end of Inflate, to drain bits from the code buffer
 // preferentially.
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_u8_prefer_code_buffer_lsb :: #force_inline proc(z: ^$C) -> (res: u8, err: io.Error) {
 	if z.num_bits >= 8 {
 		res = u8(read_bits_no_refill_lsb(z, 8))
@@ -257,7 +257,7 @@ read_u8_prefer_code_buffer_lsb :: #force_inline proc(z: ^$C) -> (res: u8, err: i
 	return
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_data_from_memory :: #force_inline proc(z: ^Context_Memory_Input, $T: typeid) -> (res: T, err: io.Error) {
 	size :: size_of(T)
 
@@ -275,7 +275,7 @@ peek_data_from_memory :: #force_inline proc(z: ^Context_Memory_Input, $T: typeid
 	}
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_data_at_offset_from_memory :: #force_inline proc(z: ^Context_Memory_Input, $T: typeid, #any_int offset: int) -> (res: T, err: io.Error) {
 	size :: size_of(T)
 
@@ -293,7 +293,7 @@ peek_data_at_offset_from_memory :: #force_inline proc(z: ^Context_Memory_Input,
 	}
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_data_from_stream :: #force_inline proc(z: ^Context_Stream_Input, $T: typeid) -> (res: T, err: io.Error) {
 	size :: size_of(T)
 
@@ -317,7 +317,7 @@ peek_data_from_stream :: #force_inline proc(z: ^Context_Stream_Input, $T: typeid
 	return res, .None
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_data_at_offset_from_stream :: #force_inline proc(z: ^Context_Stream_Input, $T: typeid, #any_int offset: int) -> (res: T, err: io.Error) {
 	size :: size_of(T)
 
@@ -352,14 +352,14 @@ peek_data :: proc{peek_data_from_memory, peek_data_from_stream, peek_data_at_off
 
 
 // Sliding window read back
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_back_byte :: #force_inline proc(z: ^$C, offset: i64) -> (res: u8, err: io.Error) {
 	// Look back into the sliding window.
 	return z.output.buf[z.bytes_written - offset], .None
 }
 
 // Generalized bit reader LSB
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 refill_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width := i8(48)) {
 	refill := u64(width)
 	b      := u64(0)
@@ -385,7 +385,7 @@ refill_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width :=
 }
 
 // Generalized bit reader LSB
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 refill_lsb_from_stream :: proc(z: ^Context_Stream_Input, width := i8(24)) {
 	refill := u64(width)
 
@@ -414,13 +414,13 @@ refill_lsb_from_stream :: proc(z: ^Context_Stream_Input, width := i8(24)) {
 refill_lsb :: proc{refill_lsb_from_memory, refill_lsb_from_stream}
 
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 consume_bits_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width: u8) {
 	z.code_buffer >>= width
 	z.num_bits -= u64(width)
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 consume_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width: u8) {
 	z.code_buffer >>= width
 	z.num_bits -= u64(width)
@@ -428,7 +428,7 @@ consume_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, wid
 
 consume_bits_lsb :: proc{consume_bits_lsb_from_memory, consume_bits_lsb_from_stream}
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_bits_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width: u8) -> u32 {
 	if z.num_bits < u64(width) {
 		refill_lsb(z)
@@ -436,7 +436,7 @@ peek_bits_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width:
 	return u32(z.code_buffer &~ (~u64(0) << width))
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width: u8) -> u32 {
 	if z.num_bits < u64(width) {
 		refill_lsb(z)
@@ -446,13 +446,13 @@ peek_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width:
 
 peek_bits_lsb :: proc{peek_bits_lsb_from_memory, peek_bits_lsb_from_stream}
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_bits_no_refill_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width: u8) -> u32 {
 	assert(z.num_bits >= u64(width))
 	return u32(z.code_buffer &~ (~u64(0) << width))
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_bits_no_refill_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width: u8) -> u32 {
 	assert(z.num_bits >= u64(width))
 	return u32(z.code_buffer &~ (~u64(0) << width))
@@ -460,14 +460,14 @@ peek_bits_no_refill_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Inp
 
 peek_bits_no_refill_lsb :: proc{peek_bits_no_refill_lsb_from_memory, peek_bits_no_refill_lsb_from_stream}
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_bits_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width: u8) -> u32 {
 	k := #force_inline peek_bits_lsb(z, width)
 	#force_inline consume_bits_lsb(z, width)
 	return k
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width: u8) -> u32 {
 	k := peek_bits_lsb(z, width)
 	consume_bits_lsb(z, width)
@@ -476,14 +476,14 @@ read_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width:
 
 read_bits_lsb :: proc{read_bits_lsb_from_memory, read_bits_lsb_from_stream}
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_bits_no_refill_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width: u8) -> u32 {
 	k := #force_inline peek_bits_no_refill_lsb(z, width)
 	#force_inline consume_bits_lsb(z, width)
 	return k
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_bits_no_refill_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width: u8) -> u32 {
 	k := peek_bits_no_refill_lsb(z, width)
 	consume_bits_lsb(z, width)
@@ -493,14 +493,14 @@ read_bits_no_refill_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Inp
 read_bits_no_refill_lsb :: proc{read_bits_no_refill_lsb_from_memory, read_bits_no_refill_lsb_from_stream}
 
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 discard_to_next_byte_lsb_from_memory :: proc(z: ^Context_Memory_Input) {
 	discard := u8(z.num_bits & 7)
 	#force_inline consume_bits_lsb(z, discard)
 }
 
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 discard_to_next_byte_lsb_from_stream :: proc(z: ^Context_Stream_Input) {
 	discard := u8(z.num_bits & 7)
 	consume_bits_lsb(z, discard)

core/compress/shoco/shoco.odin

@@ -98,7 +98,7 @@ decompress_slice_to_output_buffer :: proc(input: []u8, output: []u8, model := DE
 	validate_model(model) or_return
 
 	for inp < inp_end {
-		val  := transmute(i8)input[inp]
+		val  := i8(input[inp])
 		mark := int(-1)
 
 		for val < 0 {
@@ -274,12 +274,9 @@ compress_string_to_buffer :: proc(input: string, output: []u8, model := DEFAULT_
 				out_ptr := raw_data(output[out:])
 
 				switch pack.bytes_packed {
-				case 4:
-					intrinsics.unaligned_store(transmute(^u32)out_ptr, code)
-				case 2:
-					intrinsics.unaligned_store(transmute(^u16)out_ptr, u16(code))
-				case 1:
-					intrinsics.unaligned_store(transmute(^u8)out_ptr,  u8(code))
+				case 4: intrinsics.unaligned_store((^u32)(out_ptr), code)
+				case 2: intrinsics.unaligned_store((^u16)(out_ptr), u16(code))
+				case 1: intrinsics.unaligned_store( (^u8)(out_ptr),  u8(code))
 				case:
 					return out, .Unknown_Compression_Method
 				}

core/compress/zlib/zlib.odin

@@ -120,7 +120,7 @@ Huffman_Table :: struct {
 }
 
 // Implementation starts here
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 z_bit_reverse :: #force_inline proc(n: u16, bits: u8) -> (r: u16) {
 	assert(bits <= 16)
 	// NOTE: Can optimize with llvm.bitreverse.i64 or some bit twiddling
@@ -136,7 +136,7 @@ z_bit_reverse :: #force_inline proc(n: u16, bits: u8) -> (r: u16) {
 }
 
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 grow_buffer :: proc(buf: ^[dynamic]u8) -> (err: compress.Error) {
 	/*
 		That we get here at all means that we didn't pass an expected output size,
@@ -154,7 +154,7 @@ grow_buffer :: proc(buf: ^[dynamic]u8) -> (err: compress.Error) {
 	TODO: Make these return compress.Error.
 */
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 write_byte :: #force_inline proc(z: ^$C, c: u8) -> (err: io.Error) #no_bounds_check {
 	/*
 		Resize if needed.
@@ -173,7 +173,7 @@ write_byte :: #force_inline proc(z: ^$C, c: u8) -> (err: io.Error) #no_bounds_ch
 	return .None
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 repl_byte :: proc(z: ^$C, count: u16, c: u8) -> (err: io.Error) #no_bounds_check {
 	/*
 		TODO(Jeroen): Once we have a magic ring buffer, we can just peek/write into it
@@ -201,7 +201,7 @@ repl_byte :: proc(z: ^$C, count: u16, c: u8) -> (err: io.Error) #no_bounds_check
 	return .None
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 repl_bytes :: proc(z: ^$C, count: u16, distance: u16) -> (err: io.Error) {
 	/*
 		TODO(Jeroen): Once we have a magic ring buffer, we can just peek/write into it
@@ -234,8 +234,8 @@ allocate_huffman_table :: proc(allocator := context.allocator) -> (z: ^Huffman_T
 	return new(Huffman_Table, allocator), nil
 }
 
-@(optimization_mode="speed")
-build_huffman :: proc(z: ^Huffman_Table, code_lengths: []u8) -> (err: Error) {
+@(optimization_mode="favor_size")
+build_huffman :: #force_no_inline proc(z: ^Huffman_Table, code_lengths: []u8) -> (err: Error) {
 	sizes:     [HUFFMAN_MAX_BITS+1]int
 	next_code: [HUFFMAN_MAX_BITS+1]int
 
@@ -293,7 +293,7 @@ build_huffman :: proc(z: ^Huffman_Table, code_lengths: []u8) -> (err: Error) {
 	return nil
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 decode_huffman_slowpath :: proc(z: ^$C, t: ^Huffman_Table) -> (r: u16, err: Error) #no_bounds_check {
 	code := u16(compress.peek_bits_lsb(z,16))
 
@@ -324,7 +324,7 @@ decode_huffman_slowpath :: proc(z: ^$C, t: ^Huffman_Table) -> (r: u16, err: Erro
 	return r, nil
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 decode_huffman :: proc(z: ^$C, t: ^Huffman_Table) -> (r: u16, err: Error) #no_bounds_check {
 	if z.num_bits < 16 {
 		if z.num_bits > 63 {
@@ -344,7 +344,7 @@ decode_huffman :: proc(z: ^$C, t: ^Huffman_Table) -> (r: u16, err: Error) #no_bo
 	return decode_huffman_slowpath(z, t)
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 parse_huffman_block :: proc(z: ^$C, z_repeat, z_offset: ^Huffman_Table) -> (err: Error) #no_bounds_check {
 	#no_bounds_check for {
 		value, e := decode_huffman(z, z_repeat)
@@ -413,7 +413,7 @@ parse_huffman_block :: proc(z: ^$C, z_repeat, z_offset: ^Huffman_Table) -> (err:
 	}
 }
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 inflate_from_context :: proc(using ctx: ^compress.Context_Memory_Input, raw := false, expected_output_size := -1, allocator := context.allocator) -> (err: Error) #no_bounds_check {
 	/*
 		ctx.output must be a bytes.Buffer for now. We'll add a separate implementation that writes to a stream.
@@ -486,7 +486,7 @@ inflate_from_context :: proc(using ctx: ^compress.Context_Memory_Input, raw := f
 
 // TODO: Check alignment of reserve/resize.
 
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 inflate_raw :: proc(z: ^$C, expected_output_size := -1, allocator := context.allocator) -> (err: Error) #no_bounds_check {
 	context.allocator = allocator
 	expected_output_size := expected_output_size
@@ -670,4 +670,4 @@ inflate_from_byte_array_raw :: proc(input: []u8, buf: ^bytes.Buffer, raw := fals
 	return inflate_raw(&ctx, expected_output_size=expected_output_size)
 }
 
-inflate :: proc{inflate_from_context, inflate_from_byte_array}
+inflate :: proc{inflate_from_context, inflate_from_byte_array}

core/container/avl/avl.odin

@@ -5,13 +5,10 @@ The implementation is non-intrusive, and non-recursive.
 */
 package container_avl
 
-import "base:intrinsics"
-import "base:runtime"
+@(require) import "base:intrinsics"
+@(require) import "base:runtime"
 import "core:slice"
 
-_ :: intrinsics
-_ :: runtime
-
 // Originally based on the CC0 implementation by Eric Biggers
 // See: https://github.com/ebiggers/avl_tree/
 
@@ -90,7 +87,7 @@ init_cmp :: proc(
 init_ordered :: proc(
 	t: ^$T/Tree($Value),
 	node_allocator := context.allocator,
-) where intrinsics.type_is_ordered_numeric(Value) {
+) where intrinsics.type_is_ordered(Value) {
 	init_cmp(t, slice.cmp_proc(Value), node_allocator)
 }
 
@@ -675,4 +672,4 @@ iterator_first :: proc "contextless" (it: ^Iterator($Value)) {
 	if it._cur != nil {
 		it._next = node_next_or_prev_in_order(it._cur, it._direction)
 	}
-}
+}

core/container/bit_array/bit_array.odin

@@ -210,8 +210,11 @@ set :: proc(ba: ^Bit_Array, #any_int index: uint, set_to: bool = true, allocator
 
 	ba.max_index = max(idx, ba.max_index)
 
-	if set_to{ ba.bits[leg_index] |= 1 << uint(bit_index) }
-	else { ba.bits[leg_index] &= ~(1 << uint(bit_index)) }
+	if set_to {
+		ba.bits[leg_index] |=  1 << uint(bit_index)
+	} else {
+		ba.bits[leg_index] &~= 1 << uint(bit_index)
+	}
 
 	return true
 }
@@ -253,7 +256,7 @@ Inputs:
 - index: Which bit in the array
 */
 unsafe_unset :: proc(b: ^Bit_Array, bit: int) #no_bounds_check {
-	b.bits[bit >> INDEX_SHIFT] &= ~(1 << uint(bit & INDEX_MASK))
+	b.bits[bit >> INDEX_SHIFT] &~= 1 << uint(bit & INDEX_MASK)
 }
 /*
 A helper function to create a Bit Array with optional bias, in case your smallest index is non-zero (including negative).

core/container/intrusive/list/doc.odin

@@ -0,0 +1,46 @@
+/*
+Package list implements an intrusive doubly-linked list.
+
+An intrusive container requires a `Node` to be embedded in your own structure, like this:
+
+	My_String :: struct {
+		node:  list.Node,
+		value: string,
+	}
+
+Embedding the members of a `list.Node` in your structure with the `using` keyword is also allowed:
+
+	My_String :: struct {
+		using node: list.Node,
+		value: string,
+	}
+
+Here is a full example:
+
+	package test
+	
+	import "core:fmt"
+	import "core:container/intrusive/list"
+	
+	main :: proc() {
+	    l: list.List
+	
+	    one := My_String{value="Hello"}
+	    two := My_String{value="World"}
+	
+	    list.push_back(&l, &one.node)
+	    list.push_back(&l, &two.node)
+	
+	    iter := list.iterator_head(l, My_String, "node")
+	    for s in list.iterate_next(&iter) {
+	        fmt.println(s.value)
+	    }
+	}
+	
+	My_String :: struct {
+	    node:  list.Node,
+	    value: string,
+	}
+
+*/
+package container_intrusive_list

core/container/intrusive/list/intrusive_list.odin

@@ -18,11 +18,18 @@ List :: struct {
 	tail: ^Node,
 }
 
-
+// The list link you must include in your own structure.
 Node :: struct {
 	prev, next: ^Node,
 }
 
+/*
+Inserts a new element at the front of the list with O(1) time complexity.
+
+**Inputs**
+- list: The container list
+- node: The node member of the user-defined element structure
+*/
 push_front :: proc "contextless" (list: ^List, node: ^Node) {
 	if list.head != nil {
 		list.head.prev = node
@@ -33,7 +40,13 @@ push_front :: proc "contextless" (list: ^List, node: ^Node) {
 		node.prev, node.next = nil, nil
 	}
 }
+/*
+Inserts a new element at the back of the list with O(1) time complexity.
 
+**Inputs**
+- list: The container list
+- node: The node member of the user-defined element structure
+*/
 push_back :: proc "contextless" (list: ^List, node: ^Node) {
 	if list.tail != nil {
 		list.tail.next = node
@@ -45,6 +58,13 @@ push_back :: proc "contextless" (list: ^List, node: ^Node) {
 	}
 }
 
+/*
+Removes an element from a list with O(1) time complexity.
+
+**Inputs**
+- list: The container list
+- node: The node member of the user-defined element structure to be removed
+*/
 remove :: proc "contextless" (list: ^List, node: ^Node) {
 	if node != nil {
 		if node.next != nil {
@@ -61,7 +81,13 @@ remove :: proc "contextless" (list: ^List, node: ^Node) {
 		}
 	}
 }
+/*
+Removes from the given list all elements that satisfy a condition with O(N) time complexity.
 
+**Inputs**
+- list: The container list
+- to_erase: The condition procedure. It should return `true` if a node should be removed, `false` otherwise
+*/
 remove_by_proc :: proc(list: ^List, to_erase: proc(^Node) -> bool) {
 	for node := list.head; node != nil; {
 		next := node.next
@@ -82,7 +108,13 @@ remove_by_proc :: proc(list: ^List, to_erase: proc(^Node) -> bool) {
 		node = next
 	}
 }
+/*
+Removes from the given list all elements that satisfy a condition with O(N) time complexity.
 
+**Inputs**
+- list: The container list
+- to_erase: The _contextless_ condition procedure. It should return `true` if a node should be removed, `false` otherwise
+*/
 remove_by_proc_contextless :: proc(list: ^List, to_erase: proc "contextless" (^Node) -> bool) {
 	for node := list.head; node != nil; {
 		next := node.next
@@ -104,12 +136,26 @@ remove_by_proc_contextless :: proc(list: ^List, to_erase: proc "contextless" (^N
 	}
 }
 
+/*
+Checks whether the given list does not contain any element.
 
+**Inputs**
+- list: The container list
 
+**Returns** `true` if `list` is empty, `false` otherwise
+*/
 is_empty :: proc "contextless" (list: ^List) -> bool {
 	return list.head == nil
 }
 
+/*
+Removes and returns the element at the front of the list with O(1) time complexity.
+
+**Inputs**
+- list: The container list
+
+**Returns** The node member of the user-defined element structure, or `nil` if the list is empty
+*/
 pop_front :: proc "contextless" (list: ^List) -> ^Node {
 	link := list.head
 	if link == nil {
@@ -130,6 +176,14 @@ pop_front :: proc "contextless" (list: ^List) -> ^Node {
 	return link
 
 }
+/*
+Removes and returns the element at the back of the list with O(1) time complexity.
+
+**Inputs**
+- list: The container list
+
+**Returns** The node member of the user-defined element structure, or `nil` if the list is empty
+*/
 pop_back :: proc "contextless" (list: ^List) -> ^Node {
 	link := list.tail
 	if link == nil {
@@ -151,29 +205,102 @@ pop_back :: proc "contextless" (list: ^List) -> ^Node {
 }
 
 
+
 Iterator :: struct($T: typeid) {
 	curr:   ^Node,
 	offset: uintptr,
 }
 
+/*
+Creates an iterator pointing at the head of the given list. For an example, see `iterate_next`.
+
+**Inputs**
+- list: The container list
+- T: The type of the list's elements
+- field_name: The name of the node field in the `T` structure
+
+**Returns** An iterator pointing at the head of `list`
+
+*/
 iterator_head :: proc "contextless" (list: List, $T: typeid, $field_name: string) -> Iterator(T)
 	where intrinsics.type_has_field(T, field_name),
 	      intrinsics.type_field_type(T, field_name) == Node {
 	return {list.head, offset_of_by_string(T, field_name)}
 }
+/*
+Creates an iterator pointing at the tail of the given list. For an example, see `iterate_prev`.
 
+**Inputs**
+- list: The container list
+- T: The type of the list's elements
+- field_name: The name of the node field in the `T` structure
+
+**Returns** An iterator pointing at the tail of `list`
+
+*/
 iterator_tail :: proc "contextless" (list: List, $T: typeid, $field_name: string) -> Iterator(T)
 	where intrinsics.type_has_field(T, field_name),
 	      intrinsics.type_field_type(T, field_name) == Node {
 	return {list.tail, offset_of_by_string(T, field_name)}
 }
+/*
+Creates an iterator pointing at the specified node of a list.
 
+**Inputs**
+- node: a list node
+- T: The type of the list's elements
+- field_name: The name of the node field in the `T` structure
+
+**Returns** An iterator pointing at `node`
+
+*/
 iterator_from_node :: proc "contextless" (node: ^Node, $T: typeid, $field_name: string) -> Iterator(T)
 	where intrinsics.type_has_field(T, field_name),
 	      intrinsics.type_field_type(T, field_name) == Node {
 	return {node, offset_of_by_string(T, field_name)}
 }
 
+/*
+Retrieves the next element in a list and advances the iterator.
+
+**Inputs**  
+- it: The iterator
+
+**Returns**
+- ptr: The next list element
+- ok: `true` if the element is valid (the iterator could advance), `false` otherwise
+
+Example:
+
+	import "core:fmt"
+	import "core:container/intrusive/list"
+
+	iterate_next_example :: proc() {
+		l: list.List
+
+		one := My_Struct{value=1}
+		two := My_Struct{value=2}
+
+		list.push_back(&l, &one.node)
+		list.push_back(&l, &two.node)
+
+		it := list.iterator_head(l, My_Struct, "node")
+		for num in list.iterate_next(&it) {
+			fmt.println(num.value)
+		}
+	}
+
+	My_Struct :: struct {
+		node : list.Node,
+		value: int,
+	}
+
+Output:
+
+	1
+	2
+
+*/
 iterate_next :: proc "contextless" (it: ^Iterator($T)) -> (ptr: ^T, ok: bool) {
 	node := it.curr
 	if node == nil {
@@ -183,7 +310,47 @@ iterate_next :: proc "contextless" (it: ^Iterator($T)) -> (ptr: ^T, ok: bool) {
 
 	return (^T)(uintptr(node) - it.offset), true
 }
+/*
+Retrieves the previous element in a list and recede the iterator.
 
+**Inputs**  
+- it: The iterator
+
+**Returns**
+- ptr: The previous list element
+- ok: `true` if the element is valid (the iterator could recede), `false` otherwise
+
+Example:
+
+	import "core:fmt"
+	import "core:container/intrusive/list"
+
+	iterate_next_example :: proc() {
+		l: list.List
+
+		one := My_Struct{value=1}
+		two := My_Struct{value=2}
+
+		list.push_back(&l, &one.node)
+		list.push_back(&l, &two.node)
+
+		it := list.iterator_tail(l, My_Struct, "node")
+		for num in list.iterate_prev(&it) {
+			fmt.println(num.value)
+		}
+	}
+
+	My_Struct :: struct {
+		node : list.Node,
+		value: int,
+	}
+
+Output:
+
+	2
+	1
+
+*/
 iterate_prev :: proc "contextless" (it: ^Iterator($T)) -> (ptr: ^T, ok: bool) {
 	node := it.curr
 	if node == nil {
@@ -192,4 +359,4 @@ iterate_prev :: proc "contextless" (it: ^Iterator($T)) -> (ptr: ^T, ok: bool) {
 	it.curr = node.prev
 
 	return (^T)(uintptr(node) - it.offset), true
-}
+}

core/container/lru/lru_cache.odin

@@ -70,8 +70,7 @@ set :: proc(c: ^$C/Cache($Key, $Value), key: Key, value: Value) -> runtime.Alloc
 	if c.count == c.capacity {
 		e = c.tail
 		_remove_node(c, e)
-	}
-	else {
+	} else {
 		c.count += 1
 		e = new(Node(Key, Value), c.node_allocator) or_return
 	}

core/container/queue/queue.odin

@@ -95,11 +95,11 @@ front_ptr :: proc(q: ^$Q/Queue($T)) -> ^T {
 }
 
 back :: proc(q: ^$Q/Queue($T)) -> T {
-	idx := (q.offset+uint(q.len))%builtin.len(q.data)
+	idx := (q.offset+uint(q.len - 1))%builtin.len(q.data)
 	return q.data[idx]
 }
 back_ptr :: proc(q: ^$Q/Queue($T)) -> ^T {
-	idx := (q.offset+uint(q.len))%builtin.len(q.data)
+	idx := (q.offset+uint(q.len - 1))%builtin.len(q.data)
 	return &q.data[idx]
 }
 
@@ -189,7 +189,7 @@ pop_front_safe :: proc(q: ^$Q/Queue($T)) -> (elem: T, ok: bool) {
 	return
 }
 
-// Push multiple elements to the front of the queue
+// Push multiple elements to the back of the queue
 push_back_elems :: proc(q: ^$Q/Queue($T), elems: ..T) -> (ok: bool, err: runtime.Allocator_Error)  {
 	n := uint(builtin.len(elems))
 	if space(q^) < int(n) {
@@ -241,7 +241,7 @@ clear :: proc(q: ^$Q/Queue($T)) {
 }
 
 
-// Internal growinh procedure
+// Internal growing procedure
 _grow :: proc(q: ^$Q/Queue($T), min_capacity: uint = 0) -> runtime.Allocator_Error {
 	new_capacity := max(min_capacity, uint(8), uint(builtin.len(q.data))*2)
 	n := uint(builtin.len(q.data))

core/container/rbtree/rbtree.odin

@@ -0,0 +1,568 @@
+// This package implements a red-black tree
+package container_rbtree
+
+@(require) import "base:intrinsics"
+@(require) import "base:runtime"
+import "core:slice"
+
+// Originally based on the CC0 implementation from literateprograms.org
+// But with API design mimicking `core:container/avl` for ease of use.
+
+// Direction specifies the traversal direction for a tree iterator.
+Direction :: enum i8 {
+	// Backward is the in-order backwards direction.
+	Backward = -1,
+	// Forward is the in-order forwards direction.
+	Forward  = 1,
+}
+
+Ordering :: slice.Ordering
+
+// Tree is a red-black tree
+Tree :: struct($Key: typeid, $Value: typeid) {
+	// user_data is a parameter that will be passed to the on_remove
+	// callback.
+	user_data: rawptr,
+	// on_remove is an optional callback that can be called immediately
+	// after a node is removed from the tree.
+	on_remove: proc(key: Key, value: Value, user_data: rawptr),
+
+	_root:           ^Node(Key, Value),
+	_node_allocator: runtime.Allocator,
+	_cmp_fn:         proc(Key, Key) -> Ordering,
+	_size:           int,
+}
+
+// Node is a red-black tree node.
+//
+// WARNING: It is unsafe to mutate value if the node is part of a tree
+// if doing so will alter the Node's sort position relative to other
+// elements in the tree.
+Node :: struct($Key: typeid, $Value: typeid) {
+	key:    Key,
+	value:  Value,
+
+	_parent: ^Node(Key, Value),
+	_left:   ^Node(Key, Value),
+	_right:  ^Node(Key, Value),
+	_color:  Color,
+}
+
+// Might store this in the node pointer in the future, but that'll require a decent amount of rework to pass ^^N instead of ^N
+Color :: enum uintptr {Black = 0, Red = 1}
+
+// Iterator is a tree iterator.
+//
+// WARNING: It is unsafe to modify the tree while iterating, except via
+// the iterator_remove method.
+Iterator :: struct($Key: typeid, $Value: typeid) {
+	_tree:        ^Tree(Key, Value),
+	_cur:         ^Node(Key, Value),
+	_next:        ^Node(Key, Value),
+	_direction:   Direction,
+	_called_next: bool,
+}
+
+// init initializes a tree.
+init :: proc {
+	init_ordered,
+	init_cmp,
+}
+
+// init_cmp initializes a tree.
+init_cmp :: proc(t: ^$T/Tree($Key, $Value), cmp_fn: proc(a, b: Key) -> Ordering, node_allocator := context.allocator) {
+	t._root   = nil
+	t._node_allocator = node_allocator
+	t._cmp_fn = cmp_fn
+	t._size = 0
+}
+
+// init_ordered initializes a tree containing ordered keys, with
+// a comparison function that results in an ascending order sort.
+init_ordered :: proc(t: ^$T/Tree($Key, $Value), node_allocator := context.allocator) where intrinsics.type_is_ordered(Key) {
+	init_cmp(t, slice.cmp_proc(Key), node_allocator)
+}
+
+// destroy de-initializes a tree.
+destroy :: proc(t: ^$T/Tree($Key, $Value), call_on_remove: bool = true) {
+	iter := iterator(t, .Forward)
+	for _ in iterator_next(&iter) {
+		iterator_remove(&iter, call_on_remove)
+	}
+}
+
+len :: proc "contextless" (t: ^$T/Tree($Key, $Value)) -> (node_count: int) {
+	return t._size
+}
+
+// first returns the first node in the tree (in-order) or nil iff
+// the tree is empty.
+first :: proc "contextless" (t: ^$T/Tree($Key, $Value)) -> ^Node(Key, Value) {
+	return tree_first_or_last_in_order(t, Direction.Backward)
+}
+
+// last returns the last element in the tree (in-order) or nil iff
+// the tree is empty.
+last :: proc "contextless" (t: ^$T/Tree($Key, $Value)) -> ^Node(Key, Value) {
+	return tree_first_or_last_in_order(t, Direction.Forward)
+}
+
+// find finds the key in the tree, and returns the corresponding node, or nil iff the value is not present.
+find :: proc(t: ^$T/Tree($Key, $Value), key: Key) -> (node: ^Node(Key, Value)) {
+	node = t._root
+	for node != nil {
+		switch t._cmp_fn(key, node.key) {
+		case .Equal:   return node
+		case .Less:    node = node._left
+		case .Greater: node = node._right
+		}
+	}
+	return node
+}
+
+// find_value finds the key in the tree, and returns the corresponding value, or nil iff the value is not present.
+find_value :: proc(t: ^$T/Tree($Key, $Value), key: Key) -> (value: Value, ok: bool) #optional_ok {
+	if n := find(t, key); n != nil {
+		return n.value, true
+	}
+	return
+}
+
+// find_or_insert attempts to insert the value into the tree, and returns
+// the node, a boolean indicating if the value was inserted, and the
+// node allocator error if relevant.  If the value is already present, the existing node is updated.
+find_or_insert :: proc(t: ^$T/Tree($Key, $Value), key: Key, value: Value) -> (n: ^Node(Key, Value), inserted: bool, err: runtime.Allocator_Error) {
+	n_ptr := &t._root
+	for n_ptr^ != nil {
+		n = n_ptr^
+		switch t._cmp_fn(key, n.key) {
+		case .Less:
+			n_ptr = &n._left
+		case .Greater:
+			n_ptr = &n._right
+		case .Equal:
+			return
+		}
+	}
+	_parent := n
+
+	n = new_clone(Node(Key, Value){key=key, value=value, _parent=_parent, _color=.Red}, t._node_allocator) or_return
+	n_ptr^ = n
+	insert_case1(t, n)
+	t._size += 1
+	return n, true, nil
+}
+
+// remove removes a node or value from the tree, and returns true iff the
+// removal was successful.  While the node's value will be left intact,
+// the node itself will be freed via the tree's node allocator.
+remove :: proc {
+	remove_key,
+	remove_node,
+}
+
+// remove_value removes a value from the tree, and returns true iff the
+// removal was successful.  While the node's key + value will be left intact,
+// the node itself will be freed via the tree's node allocator.
+remove_key :: proc(t: ^$T/Tree($Key, $Value), key: Key, call_on_remove := true) -> bool {
+	n := find(t, key)
+	if n == nil {
+		return false // Key not found, nothing to do
+	}
+	return remove_node(t, n, call_on_remove)
+}
+
+// remove_node removes a node from the tree, and returns true iff the
+// removal was successful.  While the node's key + value will be left intact,
+// the node itself will be freed via the tree's node allocator.
+remove_node :: proc(t: ^$T/Tree($Key, $Value), node: ^$N/Node(Key, Value), call_on_remove := true) -> (found: bool) {
+	if node._parent == node || (node._parent == nil && t._root != node) {
+		return false // Don't touch self-parented or dangling nodes.
+	}
+	node := node
+	if node._left != nil && node._right != nil {
+		// Copy key + value from predecessor and delete it instead
+		predecessor := maximum_node(node._left)
+		node.key   = predecessor.key
+		node.value = predecessor.value
+		node = predecessor
+	}
+
+	child := node._right == nil ? node._left : node._right
+	if node_color(node) == .Black {
+		node._color = node_color(child)
+		remove_case1(t, node)
+	}
+	replace_node(t, node, child)
+	if node._parent == nil && child != nil {
+		child._color = .Black // root should be black
+	}
+
+	if call_on_remove && t.on_remove != nil {
+		t.on_remove(node.key, node.value, t.user_data)
+	}
+	free(node, t._node_allocator)
+	t._size -= 1
+	return true
+}
+
+// iterator returns a tree iterator in the specified direction.
+iterator :: proc "contextless" (t: ^$T/Tree($Key, $Value), direction: Direction) -> Iterator(Key, Value) {
+	it: Iterator(Key, Value)
+	it._tree      = cast(^Tree(Key, Value))t
+	it._direction = direction
+
+	iterator_first(&it)
+
+	return it
+}
+
+// iterator_from_pos returns a tree iterator in the specified direction,
+// spanning the range [pos, last] (inclusive).
+iterator_from_pos :: proc "contextless" (t: ^$T/Tree($Key, $Value), pos: ^Node(Key, Value), direction: Direction) -> Iterator(Key, Value) {
+	it: Iterator(Key, Value)
+	it._tree        = transmute(^Tree(Key, Value))t
+	it._direction   = direction
+	it._next        = nil
+	it._called_next = false
+
+	if it._cur = pos; pos != nil {
+		it._next = node_next_or_prev_in_order(it._cur, it._direction)
+	}
+
+	return it
+}
+
+// iterator_get returns the node currently pointed to by the iterator,
+// or nil iff the node has been removed, the tree is empty, or the end
+// of the tree has been reached.
+iterator_get :: proc "contextless" (it: ^$I/Iterator($Key, $Value)) -> ^Node(Key, Value) {
+	return it._cur
+}
+
+// iterator_remove removes the node currently pointed to by the iterator,
+// and returns true iff the removal was successful.  Semantics are the
+// same as the Tree remove.
+iterator_remove :: proc(it: ^$I/Iterator($Key, $Value), call_on_remove: bool = true) -> bool {
+	if it._cur == nil {
+		return false
+	}
+
+	ok := remove_node(it._tree, it._cur , call_on_remove)
+	if ok {
+		it._cur = nil
+	}
+
+	return ok
+}
+
+// iterator_next advances the iterator and returns the (node, true) or
+// or (nil, false) iff the end of the tree has been reached.
+//
+// Note: The first call to iterator_next will return the first node instead
+// of advancing the iterator.
+iterator_next :: proc "contextless" (it: ^$I/Iterator($Key, $Value)) -> (^Node(Key, Value), bool) {
+	// This check is needed so that the first element gets returned from
+	// a brand-new iterator, and so that the somewhat contrived case where
+	// iterator_remove is called before the first call to iterator_next
+	// returns the correct value.
+	if !it._called_next {
+		it._called_next = true
+
+		// There can be the contrived case where iterator_remove is
+		// called before ever calling iterator_next, which needs to be
+		// handled as an actual call to next.
+		//
+		// If this happens it._cur will be nil, so only return the
+		// first value, if it._cur is valid.
+		if it._cur != nil {
+			return it._cur, true
+		}
+	}
+
+	if it._next == nil {
+		return nil, false
+	}
+
+	it._cur = it._next
+	it._next = node_next_or_prev_in_order(it._cur, it._direction)
+
+	return it._cur, true
+}
+
+@(private)
+tree_first_or_last_in_order :: proc "contextless" (t: ^$T/Tree($Key, $Value), direction: Direction) -> ^Node(Key, Value) {
+	first, sign := t._root, i8(direction)
+	if first != nil {
+		for {
+			tmp := node_get_child(first, sign)
+			if tmp == nil {
+				break
+			}
+			first = tmp
+		}
+	}
+	return first
+}
+
+@(private)
+node_get_child :: #force_inline proc "contextless" (n: ^Node($Key, $Value), sign: i8) -> ^Node(Key, Value) {
+	if sign < 0 {
+		return n._left
+	}
+	return n._right
+}
+
+@(private)
+node_next_or_prev_in_order :: proc "contextless" (n: ^Node($Key, $Value), direction: Direction) -> ^Node(Key, Value) {
+	next, tmp: ^Node(Key, Value)
+	sign := i8(direction)
+
+	if next = node_get_child(n, +sign); next != nil {
+		for {
+			tmp = node_get_child(next, -sign)
+			if tmp == nil {
+				break
+			}
+			next = tmp
+		}
+	} else {
+		tmp, next = n, n._parent
+		for next != nil && tmp == node_get_child(next, +sign) {
+			tmp, next = next, next._parent
+		}
+	}
+	return next
+}
+
+@(private)
+iterator_first :: proc "contextless" (it: ^Iterator($Key, $Value)) {
+	// This is private because behavior when the user manually calls
+	// iterator_first followed by iterator_next is unintuitive, since
+	// the first call to iterator_next MUST return the first node
+	// instead of advancing so that `for node in iterator_next(&next)`
+	// works as expected.
+
+	switch it._direction {
+	case .Forward:
+		it._cur = tree_first_or_last_in_order(it._tree, .Backward)
+	case .Backward:
+		it._cur = tree_first_or_last_in_order(it._tree, .Forward)
+	}
+
+	it._next = nil
+	it._called_next = false
+
+	if it._cur != nil {
+		it._next = node_next_or_prev_in_order(it._cur, it._direction)
+	}
+}
+
+@(private)
+grand_parent :: proc(n: ^$N/Node($Key, $Value)) -> (g: ^N) {
+	return n._parent._parent
+}
+
+@(private)
+sibling :: proc(n: ^$N/Node($Key, $Value)) -> (s: ^N) {
+	if n == n._parent._left {
+		return n._parent._right
+		} else {
+			return n._parent._left
+		}
+}
+
+@(private)
+uncle :: proc(n: ^$N/Node($Key, $Value)) -> (u: ^N) {
+	return sibling(n._parent)
+}
+
+@(private)
+rotate__left :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	r := n._right
+	replace_node(t, n, r)
+	n._right = r._left
+	if r._left != nil {
+		r._left._parent = n
+	}
+	r._left   = n
+	n._parent = r
+}
+
+@(private)
+rotate__right :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	l := n._left
+	replace_node(t, n, l)
+	n._left = l._right
+	if l._right != nil {
+		l._right._parent = n
+	}
+	l._right  = n
+	n._parent = l
+}
+
+@(private)
+replace_node :: proc(t: ^$T/Tree($Key, $Value), old_n: ^$N/Node(Key, Value), new_n: ^N) {
+	if old_n._parent == nil {
+		t._root = new_n
+	} else {
+		if (old_n == old_n._parent._left) {
+			old_n._parent._left  = new_n
+		} else {
+			old_n._parent._right = new_n
+		}
+	}
+	if new_n != nil {
+		new_n._parent = old_n._parent
+	}
+}
+
+@(private)
+insert_case1 :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	if n._parent == nil {
+		n._color = .Black
+	} else {
+		insert_case2(t, n)
+	}
+}
+
+@(private)
+insert_case2 :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	if node_color(n._parent) == .Black {
+		return // Tree is still valid
+	} else {
+		insert_case3(t, n)
+	}
+}
+
+@(private)
+insert_case3 :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	if node_color(uncle(n)) == .Red {
+		n._parent._color       = .Black
+		uncle(n)._color       = .Black
+		grand_parent(n)._color = .Red
+		insert_case1(t, grand_parent(n))
+	} else {
+		insert_case4(t, n)
+	}
+}
+
+@(private)
+insert_case4 :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	n := n
+	if n == n._parent._right && n._parent == grand_parent(n)._left {
+		rotate__left(t, n._parent)
+		n = n._left
+	} else if n == n._parent._left && n._parent == grand_parent(n)._right {
+		rotate__right(t, n._parent)
+		n = n._right
+	}
+	insert_case5(t, n)
+}
+
+@(private)
+insert_case5 :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	n._parent._color = .Black
+	grand_parent(n)._color = .Red
+	if n == n._parent._left && n._parent == grand_parent(n)._left {
+		rotate__right(t, grand_parent(n))
+	} else {
+		rotate__left(t, grand_parent(n))
+	}
+}
+
+// The maximum_node() helper function just walks _right until it reaches the last non-leaf:
+@(private)
+maximum_node :: proc(n: ^$N/Node($Key, $Value)) -> (max_node: ^N) {
+	n := n
+	for n._right != nil {
+		n = n._right
+	}
+	return n
+}
+
+@(private)
+remove_case1 :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	if n._parent == nil {
+		return
+	} else {
+		remove_case2(t, n)
+	}
+}
+
+@(private)
+remove_case2 :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	if node_color(sibling(n)) == .Red {
+		n._parent._color = .Red
+		sibling(n)._color = .Black
+		if n == n._parent._left {
+			rotate__left(t, n._parent)
+		} else {
+			rotate__right(t, n._parent)
+		}
+	}
+	remove_case3(t, n)
+}
+
+@(private)
+remove_case3 :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	if node_color(n._parent) == .Black &&
+		node_color(sibling(n)) == .Black &&
+		node_color(sibling(n)._left) == .Black &&
+		node_color(sibling(n)._right) == .Black {
+			sibling(n)._color = .Red
+			remove_case1(t, n._parent)
+	} else {
+		remove_case4(t, n)
+	}
+}
+
+@(private)
+remove_case4 :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	if node_color(n._parent) == .Red &&
+		node_color(sibling(n)) == .Black &&
+		node_color(sibling(n)._left) == .Black &&
+		node_color(sibling(n)._right) == .Black {
+			sibling(n)._color = .Red
+			n._parent._color = .Black
+	} else {
+		remove_case5(t, n)
+	}
+}
+
+@(private)
+remove_case5 :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	if n == n._parent._left &&
+		node_color(sibling(n)) == .Black &&
+		node_color(sibling(n)._left) == .Red &&
+		node_color(sibling(n)._right) == .Black {
+			sibling(n)._color = .Red
+			sibling(n)._left._color = .Black
+			rotate__right(t, sibling(n))
+	} else if n == n._parent._right &&
+		node_color(sibling(n)) == .Black &&
+		node_color(sibling(n)._right) == .Red &&
+		node_color(sibling(n)._left) == .Black {
+			sibling(n)._color = .Red
+			sibling(n)._right._color = .Black
+			rotate__left(t, sibling(n))
+	}
+	remove_case6(t, n)
+}
+
+@(private)
+remove_case6 :: proc(t: ^$T/Tree($Key, $Value), n: ^$N/Node(Key, Value)) {
+	sibling(n)._color = node_color(n._parent)
+	n._parent._color = .Black
+	if n == n._parent._left {
+		sibling(n)._right._color = .Black
+		rotate__left(t, n._parent)
+	} else {
+		sibling(n)._left._color = .Black
+		rotate__right(t, n._parent)
+	}
+}
+
+node_color :: proc(n: ^$N/Node($Key, $Value)) -> (c: Color) {
+	return n == nil ? .Black : n._color
+}

core/container/small_array/small_array.odin

@@ -119,20 +119,20 @@ consume :: proc "odin" (a: ^$A/Small_Array($N, $T), count: int, loc := #caller_l
 }
 
 ordered_remove :: proc "contextless" (a: ^$A/Small_Array($N, $T), index: int, loc := #caller_location) #no_bounds_check {
-    runtime.bounds_check_error_loc(loc, index, a.len)
-    if index+1 < a.len {
+	runtime.bounds_check_error_loc(loc, index, a.len)
+	if index+1 < a.len {
 		copy(a.data[index:], a.data[index+1:])
 	}
 	a.len -= 1
 }
 
 unordered_remove :: proc "contextless" (a: ^$A/Small_Array($N, $T), index: int, loc := #caller_location) #no_bounds_check {
-    runtime.bounds_check_error_loc(loc, index, a.len)
+	runtime.bounds_check_error_loc(loc, index, a.len)
 	n := a.len-1
 	if index != n {
 		a.data[index] = a.data[n]
 	}
-    a.len -= 1
+	a.len -= 1
 }
 
 clear :: proc "contextless" (a: ^$A/Small_Array($N, $T)) {

core/container/topological_sort/topological_sort.odin

@@ -61,7 +61,7 @@ add_dependency :: proc(sorter: ^$S/Sorter($K), key, dependency: K) -> bool {
 	}
 	find.dependents[key] = true
 
- 	find = &sorter.relations[key]
+	find = &sorter.relations[key]
 	if find == nil {
 		find = map_insert(&sorter.relations, key, make_relations(sorter))
 	}

core/crypto/_aes/aes.odin

@@ -0,0 +1,28 @@
+package _aes
+
+// KEY_SIZE_128 is the AES-128 key size in bytes.
+KEY_SIZE_128 :: 16
+// KEY_SIZE_192 is the AES-192 key size in bytes.
+KEY_SIZE_192 :: 24
+// KEY_SIZE_256 is the AES-256 key size in bytes.
+KEY_SIZE_256 :: 32
+
+// BLOCK_SIZE is the AES block size in bytes.
+BLOCK_SIZE :: 16
+
+// ROUNDS_128 is the number of rounds for AES-128.
+ROUNDS_128 :: 10
+// ROUNDS_192 is the number of rounds for AES-192.
+ROUNDS_192 :: 12
+// ROUNDS_256 is the number of rounds for AES-256.
+ROUNDS_256 :: 14
+
+// GHASH_KEY_SIZE is the GHASH key size in bytes.
+GHASH_KEY_SIZE :: 16
+// GHASH_BLOCK_SIZE is the GHASH block size in bytes.
+GHASH_BLOCK_SIZE :: 16
+// GHASH_TAG_SIZE is the GHASH tag size in bytes.
+GHASH_TAG_SIZE :: 16
+
+// RCON is the AES keyschedule round constants.
+RCON := [10]byte{0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1B, 0x36}

core/crypto/_aes/ct64/api.odin

@@ -0,0 +1,86 @@
+package aes_ct64
+
+import "base:intrinsics"
+import "core:mem"
+
+STRIDE :: 4
+
+// Context is a keyed AES (ECB) instance.
+Context :: struct {
+	_sk_exp:     [120]u64,
+	_num_rounds: int,
+}
+
+// init initializes a context for AES with the provided key.
+init :: proc(ctx: ^Context, key: []byte) {
+	skey: [30]u64 = ---
+
+	ctx._num_rounds = keysched(skey[:], key)
+	skey_expand(ctx._sk_exp[:], skey[:], ctx._num_rounds)
+}
+
+// encrypt_block sets `dst` to `AES-ECB-Encrypt(src)`.
+encrypt_block :: proc(ctx: ^Context, dst, src: []byte) {
+	q: [8]u64
+	load_blockx1(&q, src)
+	_encrypt(&q, ctx._sk_exp[:], ctx._num_rounds)
+	store_blockx1(dst, &q)
+}
+
+// encrypt_block sets `dst` to `AES-ECB-Decrypt(src)`.
+decrypt_block :: proc(ctx: ^Context, dst, src: []byte) {
+	q: [8]u64
+	load_blockx1(&q, src)
+	_decrypt(&q, ctx._sk_exp[:], ctx._num_rounds)
+	store_blockx1(dst, &q)
+}
+
+// encrypt_blocks sets `dst` to `AES-ECB-Encrypt(src[0], .. src[n])`.
+encrypt_blocks :: proc(ctx: ^Context, dst, src: [][]byte) {
+	q: [8]u64 = ---
+	src, dst := src, dst
+
+	n := len(src)
+	for n > 4 {
+		load_blocks(&q, src[0:4])
+		_encrypt(&q, ctx._sk_exp[:], ctx._num_rounds)
+		store_blocks(dst[0:4], &q)
+
+		src = src[4:]
+		dst = dst[4:]
+		n -= 4
+	}
+	if n > 0 {
+		load_blocks(&q, src)
+		_encrypt(&q, ctx._sk_exp[:], ctx._num_rounds)
+		store_blocks(dst, &q)
+	}
+}
+
+// decrypt_blocks sets dst to `AES-ECB-Decrypt(src[0], .. src[n])`.
+decrypt_blocks :: proc(ctx: ^Context, dst, src: [][]byte) {
+	q: [8]u64 = ---
+	src, dst := src, dst
+
+	n := len(src)
+	for n > 4 {
+		load_blocks(&q, src[0:4])
+		_decrypt(&q, ctx._sk_exp[:], ctx._num_rounds)
+		store_blocks(dst[0:4], &q)
+
+		src = src[4:]
+		dst = dst[4:]
+		n -= 4
+	}
+	if n > 0 {
+		load_blocks(&q, src)
+		_decrypt(&q, ctx._sk_exp[:], ctx._num_rounds)
+		store_blocks(dst, &q)
+	}
+}
+
+// reset sanitizes the Context.  The Context must be re-initialized to
+// be used again.
+reset :: proc(ctx: ^Context) {
+	mem.zero_explicit(ctx, size_of(ctx))
+}

core/crypto/_aes/ct64/ct64.odin

@@ -0,0 +1,265 @@
+// Copyright (c) 2016 Thomas Pornin <pornin@bolet.org>
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions
+// are met:
+//
+//   1. Redistributions of source code must retain the above copyright
+//      notice, this list of conditions and the following disclaimer.
+//
+// THIS SOFTWARE IS PROVIDED BY THE AUTHORS “AS IS” AND ANY EXPRESS OR
+// IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+// WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY
+// DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+// GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+// WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+// THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+package aes_ct64
+
+import "base:intrinsics"
+
+// Bitsliced AES for 64-bit general purpose (integer) registers.  Each
+// invocation will process up to 4 blocks at a time.  This implementation
+// is derived from the BearSSL ct64 code, and distributed under a 1-clause
+// BSD license with permission from the original author.
+//
+// WARNING: "hic sunt dracones"
+//
+// This package also deliberately exposes enough internals to be able to
+// function as a replacement for `AESENC` and `AESDEC` from AES-NI, to
+// allow the implementation of non-AES primitives that use the AES round
+// function such as AEGIS and Deoxys-II.  This should ONLY be done when
+// implementing something other than AES itself.
+
+sub_bytes :: proc "contextless" (q: ^[8]u64) {
+	// This S-box implementation is a straightforward translation of
+	// the circuit described by Boyar and Peralta in "A new
+	// combinational logic minimization technique with applications
+	// to cryptology" (https://eprint.iacr.org/2009/191.pdf).
+	//
+	// Note that variables x* (input) and s* (output) are numbered
+	// in "reverse" order (x0 is the high bit, x7 is the low bit).
+
+	x0 := q[7]
+	x1 := q[6]
+	x2 := q[5]
+	x3 := q[4]
+	x4 := q[3]
+	x5 := q[2]
+	x6 := q[1]
+	x7 := q[0]
+
+	// Top linear transformation.
+	y14 := x3 ~ x5
+	y13 := x0 ~ x6
+	y9 := x0 ~ x3
+	y8 := x0 ~ x5
+	t0 := x1 ~ x2
+	y1 := t0 ~ x7
+	y4 := y1 ~ x3
+	y12 := y13 ~ y14
+	y2 := y1 ~ x0
+	y5 := y1 ~ x6
+	y3 := y5 ~ y8
+	t1 := x4 ~ y12
+	y15 := t1 ~ x5
+	y20 := t1 ~ x1
+	y6 := y15 ~ x7
+	y10 := y15 ~ t0
+	y11 := y20 ~ y9
+	y7 := x7 ~ y11
+	y17 := y10 ~ y11
+	y19 := y10 ~ y8
+	y16 := t0 ~ y11
+	y21 := y13 ~ y16
+	y18 := x0 ~ y16
+
+	// Non-linear section.
+	t2 := y12 & y15
+	t3 := y3 & y6
+	t4 := t3 ~ t2
+	t5 := y4 & x7
+	t6 := t5 ~ t2
+	t7 := y13 & y16
+	t8 := y5 & y1
+	t9 := t8 ~ t7
+	t10 := y2 & y7
+	t11 := t10 ~ t7
+	t12 := y9 & y11
+	t13 := y14 & y17
+	t14 := t13 ~ t12
+	t15 := y8 & y10
+	t16 := t15 ~ t12
+	t17 := t4 ~ t14
+	t18 := t6 ~ t16
+	t19 := t9 ~ t14
+	t20 := t11 ~ t16
+	t21 := t17 ~ y20
+	t22 := t18 ~ y19
+	t23 := t19 ~ y21
+	t24 := t20 ~ y18
+
+	t25 := t21 ~ t22
+	t26 := t21 & t23
+	t27 := t24 ~ t26
+	t28 := t25 & t27
+	t29 := t28 ~ t22
+	t30 := t23 ~ t24
+	t31 := t22 ~ t26
+	t32 := t31 & t30
+	t33 := t32 ~ t24
+	t34 := t23 ~ t33
+	t35 := t27 ~ t33
+	t36 := t24 & t35
+	t37 := t36 ~ t34
+	t38 := t27 ~ t36
+	t39 := t29 & t38
+	t40 := t25 ~ t39
+
+	t41 := t40 ~ t37
+	t42 := t29 ~ t33
+	t43 := t29 ~ t40
+	t44 := t33 ~ t37
+	t45 := t42 ~ t41
+	z0 := t44 & y15
+	z1 := t37 & y6
+	z2 := t33 & x7
+	z3 := t43 & y16
+	z4 := t40 & y1
+	z5 := t29 & y7
+	z6 := t42 & y11
+	z7 := t45 & y17
+	z8 := t41 & y10
+	z9 := t44 & y12
+	z10 := t37 & y3
+	z11 := t33 & y4
+	z12 := t43 & y13
+	z13 := t40 & y5
+	z14 := t29 & y2
+	z15 := t42 & y9
+	z16 := t45 & y14
+	z17 := t41 & y8
+
+	// Bottom linear transformation.
+	t46 := z15 ~ z16
+	t47 := z10 ~ z11
+	t48 := z5 ~ z13
+	t49 := z9 ~ z10
+	t50 := z2 ~ z12
+	t51 := z2 ~ z5
+	t52 := z7 ~ z8
+	t53 := z0 ~ z3
+	t54 := z6 ~ z7
+	t55 := z16 ~ z17
+	t56 := z12 ~ t48
+	t57 := t50 ~ t53
+	t58 := z4 ~ t46
+	t59 := z3 ~ t54
+	t60 := t46 ~ t57
+	t61 := z14 ~ t57
+	t62 := t52 ~ t58
+	t63 := t49 ~ t58
+	t64 := z4 ~ t59
+	t65 := t61 ~ t62
+	t66 := z1 ~ t63
+	s0 := t59 ~ t63
+	s6 := t56 ~ ~t62
+	s7 := t48 ~ ~t60
+	t67 := t64 ~ t65
+	s3 := t53 ~ t66
+	s4 := t51 ~ t66
+	s5 := t47 ~ t65
+	s1 := t64 ~ ~s3
+	s2 := t55 ~ ~t67
+
+	q[7] = s0
+	q[6] = s1
+	q[5] = s2
+	q[4] = s3
+	q[3] = s4
+	q[2] = s5
+	q[1] = s6
+	q[0] = s7
+}
+
+orthogonalize :: proc "contextless" (q: ^[8]u64) {
+	CL2 :: 0x5555555555555555
+	CH2 :: 0xAAAAAAAAAAAAAAAA
+	q[0], q[1] = (q[0] & CL2) | ((q[1] & CL2) << 1), ((q[0] & CH2) >> 1) | (q[1] & CH2)
+	q[2], q[3] = (q[2] & CL2) | ((q[3] & CL2) << 1), ((q[2] & CH2) >> 1) | (q[3] & CH2)
+	q[4], q[5] = (q[4] & CL2) | ((q[5] & CL2) << 1), ((q[4] & CH2) >> 1) | (q[5] & CH2)
+	q[6], q[7] = (q[6] & CL2) | ((q[7] & CL2) << 1), ((q[6] & CH2) >> 1) | (q[7] & CH2)
+
+	CL4 :: 0x3333333333333333
+	CH4 :: 0xCCCCCCCCCCCCCCCC
+	q[0], q[2] = (q[0] & CL4) | ((q[2] & CL4) << 2), ((q[0] & CH4) >> 2) | (q[2] & CH4)
+	q[1], q[3] = (q[1] & CL4) | ((q[3] & CL4) << 2), ((q[1] & CH4) >> 2) | (q[3] & CH4)
+	q[4], q[6] = (q[4] & CL4) | ((q[6] & CL4) << 2), ((q[4] & CH4) >> 2) | (q[6] & CH4)
+	q[5], q[7] = (q[5] & CL4) | ((q[7] & CL4) << 2), ((q[5] & CH4) >> 2) | (q[7] & CH4)
+
+	CL8 :: 0x0F0F0F0F0F0F0F0F
+	CH8 :: 0xF0F0F0F0F0F0F0F0
+	q[0], q[4] = (q[0] & CL8) | ((q[4] & CL8) << 4), ((q[0] & CH8) >> 4) | (q[4] & CH8)
+	q[1], q[5] = (q[1] & CL8) | ((q[5] & CL8) << 4), ((q[1] & CH8) >> 4) | (q[5] & CH8)
+	q[2], q[6] = (q[2] & CL8) | ((q[6] & CL8) << 4), ((q[2] & CH8) >> 4) | (q[6] & CH8)
+	q[3], q[7] = (q[3] & CL8) | ((q[7] & CL8) << 4), ((q[3] & CH8) >> 4) | (q[7] & CH8)
+}
+
+@(require_results)
+interleave_in :: proc "contextless" (w: []u32) -> (q0, q1: u64) #no_bounds_check {
+	if len(w) < 4 {
+		intrinsics.trap()
+	}
+	x0, x1, x2, x3 := u64(w[0]), u64(w[1]), u64(w[2]), u64(w[3])
+	x0 |= (x0 << 16)
+	x1 |= (x1 << 16)
+	x2 |= (x2 << 16)
+	x3 |= (x3 << 16)
+	x0 &= 0x0000FFFF0000FFFF
+	x1 &= 0x0000FFFF0000FFFF
+	x2 &= 0x0000FFFF0000FFFF
+	x3 &= 0x0000FFFF0000FFFF
+	x0 |= (x0 << 8)
+	x1 |= (x1 << 8)
+	x2 |= (x2 << 8)
+	x3 |= (x3 << 8)
+	x0 &= 0x00FF00FF00FF00FF
+	x1 &= 0x00FF00FF00FF00FF
+	x2 &= 0x00FF00FF00FF00FF
+	x3 &= 0x00FF00FF00FF00FF
+	q0 = x0 | (x2 << 8)
+	q1 = x1 | (x3 << 8)
+	return
+}
+
+@(require_results)
+interleave_out :: proc "contextless" (q0, q1: u64) -> (w0, w1, w2, w3: u32) {
+	x0 := q0 & 0x00FF00FF00FF00FF
+	x1 := q1 & 0x00FF00FF00FF00FF
+	x2 := (q0 >> 8) & 0x00FF00FF00FF00FF
+	x3 := (q1 >> 8) & 0x00FF00FF00FF00FF
+	x0 |= (x0 >> 8)
+	x1 |= (x1 >> 8)
+	x2 |= (x2 >> 8)
+	x3 |= (x3 >> 8)
+	x0 &= 0x0000FFFF0000FFFF
+	x1 &= 0x0000FFFF0000FFFF
+	x2 &= 0x0000FFFF0000FFFF
+	x3 &= 0x0000FFFF0000FFFF
+	w0 = u32(x0) | u32(x0 >> 16)
+	w1 = u32(x1) | u32(x1 >> 16)
+	w2 = u32(x2) | u32(x2 >> 16)
+	w3 = u32(x3) | u32(x3 >> 16)
+	return
+}
+
+@(private)
+rotr32 :: #force_inline proc "contextless" (x: u64) -> u64 {
+	return (x << 32) | (x >> 32)
+}

core/crypto/_aes/ct64/ct64_dec.odin

@@ -0,0 +1,135 @@
+// Copyright (c) 2016 Thomas Pornin <pornin@bolet.org>
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions
+// are met:
+//
+//   1. Redistributions of source code must retain the above copyright
+//      notice, this list of conditions and the following disclaimer.
+//
+// THIS SOFTWARE IS PROVIDED BY THE AUTHORS “AS IS” AND ANY EXPRESS OR
+// IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+// WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY
+// DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+// GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+// WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+// THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+package aes_ct64
+
+import "base:intrinsics"
+
+inv_sub_bytes :: proc "contextless" (q: ^[8]u64) {
+	// AES S-box is:
+	//   S(x) = A(I(x)) ^ 0x63
+	// where I() is inversion in GF(256), and A() is a linear
+	// transform (0 is formally defined to be its own inverse).
+	// Since inversion is an involution, the inverse S-box can be
+	// computed from the S-box as:
+	//   iS(x) = B(S(B(x ^ 0x63)) ^ 0x63)
+	// where B() is the inverse of A(). Indeed, for any y in GF(256):
+	//   iS(S(y)) = B(A(I(B(A(I(y)) ^ 0x63 ^ 0x63))) ^ 0x63 ^ 0x63) = y
+	//
+	// Note: we reuse the implementation of the forward S-box,
+	// instead of duplicating it here, so that total code size is
+	// lower. By merging the B() transforms into the S-box circuit
+	// we could make faster CBC decryption, but CBC decryption is
+	// already quite faster than CBC encryption because we can
+	// process four blocks in parallel.
+
+	q0 := ~q[0]
+	q1 := ~q[1]
+	q2 := q[2]
+	q3 := q[3]
+	q4 := q[4]
+	q5 := ~q[5]
+	q6 := ~q[6]
+	q7 := q[7]
+	q[7] = q1 ~ q4 ~ q6
+	q[6] = q0 ~ q3 ~ q5
+	q[5] = q7 ~ q2 ~ q4
+	q[4] = q6 ~ q1 ~ q3
+	q[3] = q5 ~ q0 ~ q2
+	q[2] = q4 ~ q7 ~ q1
+	q[1] = q3 ~ q6 ~ q0
+	q[0] = q2 ~ q5 ~ q7
+
+	sub_bytes(q)
+
+	q0 = ~q[0]
+	q1 = ~q[1]
+	q2 = q[2]
+	q3 = q[3]
+	q4 = q[4]
+	q5 = ~q[5]
+	q6 = ~q[6]
+	q7 = q[7]
+	q[7] = q1 ~ q4 ~ q6
+	q[6] = q0 ~ q3 ~ q5
+	q[5] = q7 ~ q2 ~ q4
+	q[4] = q6 ~ q1 ~ q3
+	q[3] = q5 ~ q0 ~ q2
+	q[2] = q4 ~ q7 ~ q1
+	q[1] = q3 ~ q6 ~ q0
+	q[0] = q2 ~ q5 ~ q7
+}
+
+inv_shift_rows :: proc "contextless" (q: ^[8]u64) {
+	for x, i in q {
+		q[i] =
+			(x & 0x000000000000FFFF) |
+			((x & 0x000000000FFF0000) << 4) |
+			((x & 0x00000000F0000000) >> 12) |
+			((x & 0x000000FF00000000) << 8) |
+			((x & 0x0000FF0000000000) >> 8) |
+			((x & 0x000F000000000000) << 12) |
+			((x & 0xFFF0000000000000) >> 4)
+	}
+}
+
+inv_mix_columns :: proc "contextless" (q: ^[8]u64) {
+	q0 := q[0]
+	q1 := q[1]
+	q2 := q[2]
+	q3 := q[3]
+	q4 := q[4]
+	q5 := q[5]
+	q6 := q[6]
+	q7 := q[7]
+	r0 := (q0 >> 16) | (q0 << 48)
+	r1 := (q1 >> 16) | (q1 << 48)
+	r2 := (q2 >> 16) | (q2 << 48)
+	r3 := (q3 >> 16) | (q3 << 48)
+	r4 := (q4 >> 16) | (q4 << 48)
+	r5 := (q5 >> 16) | (q5 << 48)
+	r6 := (q6 >> 16) | (q6 << 48)
+	r7 := (q7 >> 16) | (q7 << 48)
+
+	q[0] = q5 ~ q6 ~ q7 ~ r0 ~ r5 ~ r7 ~ rotr32(q0 ~ q5 ~ q6 ~ r0 ~ r5)
+	q[1] = q0 ~ q5 ~ r0 ~ r1 ~ r5 ~ r6 ~ r7 ~ rotr32(q1 ~ q5 ~ q7 ~ r1 ~ r5 ~ r6)
+	q[2] = q0 ~ q1 ~ q6 ~ r1 ~ r2 ~ r6 ~ r7 ~ rotr32(q0 ~ q2 ~ q6 ~ r2 ~ r6 ~ r7)
+	q[3] = q0 ~ q1 ~ q2 ~ q5 ~ q6 ~ r0 ~ r2 ~ r3 ~ r5 ~ rotr32(q0 ~ q1 ~ q3 ~ q5 ~ q6 ~ q7 ~ r0 ~ r3 ~ r5 ~ r7)
+	q[4] = q1 ~ q2 ~ q3 ~ q5 ~ r1 ~ r3 ~ r4 ~ r5 ~ r6 ~ r7 ~ rotr32(q1 ~ q2 ~ q4 ~ q5 ~ q7 ~ r1 ~ r4 ~ r5 ~ r6)
+	q[5] = q2 ~ q3 ~ q4 ~ q6 ~ r2 ~ r4 ~ r5 ~ r6 ~ r7 ~ rotr32(q2 ~ q3 ~ q5 ~ q6 ~ r2 ~ r5 ~ r6 ~ r7)
+	q[6] = q3 ~ q4 ~ q5 ~ q7 ~ r3 ~ r5 ~ r6 ~ r7 ~ rotr32(q3 ~ q4 ~ q6 ~ q7 ~ r3 ~ r6 ~ r7)
+	q[7] = q4 ~ q5 ~ q6 ~ r4 ~ r6 ~ r7 ~ rotr32(q4 ~ q5 ~ q7 ~ r4 ~ r7)
+}
+
+@(private)
+_decrypt :: proc "contextless" (q: ^[8]u64, skey: []u64, num_rounds: int) {
+	add_round_key(q, skey[num_rounds << 3:])
+	for u := num_rounds - 1; u > 0; u -= 1 {
+		inv_shift_rows(q)
+		inv_sub_bytes(q)
+		add_round_key(q, skey[u << 3:])
+		inv_mix_columns(q)
+	}
+	inv_shift_rows(q)
+	inv_sub_bytes(q)
+	add_round_key(q, skey)
+}

core/crypto/_aes/ct64/ct64_enc.odin

@@ -0,0 +1,95 @@
+// Copyright (c) 2016 Thomas Pornin <pornin@bolet.org>
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions
+// are met:
+//
+//   1. Redistributions of source code must retain the above copyright
+//      notice, this list of conditions and the following disclaimer.
+//
+// THIS SOFTWARE IS PROVIDED BY THE AUTHORS “AS IS” AND ANY EXPRESS OR
+// IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+// WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY
+// DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+// GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+// WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+// THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+package aes_ct64
+
+import "base:intrinsics"
+
+add_round_key :: proc "contextless" (q: ^[8]u64, sk: []u64) #no_bounds_check {
+	if len(sk) < 8 {
+		intrinsics.trap()
+	}
+
+	q[0] ~= sk[0]
+	q[1] ~= sk[1]
+	q[2] ~= sk[2]
+	q[3] ~= sk[3]
+	q[4] ~= sk[4]
+	q[5] ~= sk[5]
+	q[6] ~= sk[6]
+	q[7] ~= sk[7]
+}
+
+shift_rows :: proc "contextless" (q: ^[8]u64) {
+	for x, i in q {
+		q[i] =
+			(x & 0x000000000000FFFF) |
+			((x & 0x00000000FFF00000) >> 4) |
+			((x & 0x00000000000F0000) << 12) |
+			((x & 0x0000FF0000000000) >> 8) |
+			((x & 0x000000FF00000000) << 8) |
+			((x & 0xF000000000000000) >> 12) |
+			((x & 0x0FFF000000000000) << 4)
+	}
+}
+
+mix_columns :: proc "contextless" (q: ^[8]u64) {
+	q0 := q[0]
+	q1 := q[1]
+	q2 := q[2]
+	q3 := q[3]
+	q4 := q[4]
+	q5 := q[5]
+	q6 := q[6]
+	q7 := q[7]
+	r0 := (q0 >> 16) | (q0 << 48)
+	r1 := (q1 >> 16) | (q1 << 48)
+	r2 := (q2 >> 16) | (q2 << 48)
+	r3 := (q3 >> 16) | (q3 << 48)
+	r4 := (q4 >> 16) | (q4 << 48)
+	r5 := (q5 >> 16) | (q5 << 48)
+	r6 := (q6 >> 16) | (q6 << 48)
+	r7 := (q7 >> 16) | (q7 << 48)
+
+	q[0] = q7 ~ r7 ~ r0 ~ rotr32(q0 ~ r0)
+	q[1] = q0 ~ r0 ~ q7 ~ r7 ~ r1 ~ rotr32(q1 ~ r1)
+	q[2] = q1 ~ r1 ~ r2 ~ rotr32(q2 ~ r2)
+	q[3] = q2 ~ r2 ~ q7 ~ r7 ~ r3 ~ rotr32(q3 ~ r3)
+	q[4] = q3 ~ r3 ~ q7 ~ r7 ~ r4 ~ rotr32(q4 ~ r4)
+	q[5] = q4 ~ r4 ~ r5 ~ rotr32(q5 ~ r5)
+	q[6] = q5 ~ r5 ~ r6 ~ rotr32(q6 ~ r6)
+	q[7] = q6 ~ r6 ~ r7 ~ rotr32(q7 ~ r7)
+}
+
+@(private)
+_encrypt :: proc "contextless" (q: ^[8]u64, skey: []u64, num_rounds: int) {
+	add_round_key(q, skey)
+	for u in 1 ..< num_rounds {
+		sub_bytes(q)
+		shift_rows(q)
+		mix_columns(q)
+		add_round_key(q, skey[u << 3:])
+	}
+	sub_bytes(q)
+	shift_rows(q)
+	add_round_key(q, skey[num_rounds << 3:])
+}

core/crypto/_aes/ct64/ct64_keysched.odin

@@ -0,0 +1,179 @@
+// Copyright (c) 2016 Thomas Pornin <pornin@bolet.org>
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions
+// are met:
+//
+//   1. Redistributions of source code must retain the above copyright
+//      notice, this list of conditions and the following disclaimer.
+//
+// THIS SOFTWARE IS PROVIDED BY THE AUTHORS “AS IS” AND ANY EXPRESS OR
+// IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+// WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY
+// DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+// GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+// WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+// THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+package aes_ct64
+
+import "base:intrinsics"
+import "core:crypto/_aes"
+import "core:encoding/endian"
+import "core:mem"
+
+@(private, require_results)
+sub_word :: proc "contextless" (x: u32) -> u32 {
+	q := [8]u64{u64(x), 0, 0, 0, 0, 0, 0, 0}
+
+	orthogonalize(&q)
+	sub_bytes(&q)
+	orthogonalize(&q)
+	ret := u32(q[0])
+
+	mem.zero_explicit(&q[0], size_of(u64))
+
+	return ret
+}
+
+@(private, require_results)
+keysched :: proc(comp_skey: []u64, key: []byte) -> int {
+	num_rounds, key_len := 0, len(key)
+	switch key_len {
+	case _aes.KEY_SIZE_128:
+		num_rounds = _aes.ROUNDS_128
+	case _aes.KEY_SIZE_192:
+		num_rounds = _aes.ROUNDS_192
+	case _aes.KEY_SIZE_256:
+		num_rounds = _aes.ROUNDS_256
+	case:
+		panic("crypto/aes: invalid AES key size")
+	}
+
+	skey: [60]u32 = ---
+	nk, nkf := key_len >> 2, (num_rounds + 1) << 2
+	for i in 0 ..< nk {
+		skey[i] = endian.unchecked_get_u32le(key[i << 2:])
+	}
+	tmp := skey[(key_len >> 2) - 1]
+	for i, j, k := nk, 0, 0; i < nkf; i += 1 {
+		if j == 0 {
+			tmp = (tmp << 24) | (tmp >> 8)
+			tmp = sub_word(tmp) ~ u32(_aes.RCON[k])
+		} else if nk > 6 && j == 4 {
+			tmp = sub_word(tmp)
+		}
+		tmp ~= skey[i - nk]
+		skey[i] = tmp
+		if j += 1; j == nk {
+			j = 0
+			k += 1
+		}
+	}
+
+	q: [8]u64 = ---
+	for i, j := 0, 0; i < nkf; i, j = i + 4, j + 2 {
+		q[0], q[4] = interleave_in(skey[i:])
+		q[1] = q[0]
+		q[2] = q[0]
+		q[3] = q[0]
+		q[5] = q[4]
+		q[6] = q[4]
+		q[7] = q[4]
+		orthogonalize(&q)
+		comp_skey[j + 0] =
+			(q[0] & 0x1111111111111111) |
+			(q[1] & 0x2222222222222222) |
+			(q[2] & 0x4444444444444444) |
+			(q[3] & 0x8888888888888888)
+		comp_skey[j + 1] =
+			(q[4] & 0x1111111111111111) |
+			(q[5] & 0x2222222222222222) |
+			(q[6] & 0x4444444444444444) |
+			(q[7] & 0x8888888888888888)
+	}
+
+	mem.zero_explicit(&skey, size_of(skey))
+	mem.zero_explicit(&q, size_of(q))
+
+	return num_rounds
+}
+
+@(private)
+skey_expand :: proc "contextless" (skey, comp_skey: []u64, num_rounds: int) {
+	n := (num_rounds + 1) << 1
+	for u, v := 0, 0; u < n; u, v = u + 1, v + 4 {
+		x0 := comp_skey[u]
+		x1, x2, x3 := x0, x0, x0
+		x0 &= 0x1111111111111111
+		x1 &= 0x2222222222222222
+		x2 &= 0x4444444444444444
+		x3 &= 0x8888888888888888
+		x1 >>= 1
+		x2 >>= 2
+		x3 >>= 3
+		skey[v + 0] = (x0 << 4) - x0
+		skey[v + 1] = (x1 << 4) - x1
+		skey[v + 2] = (x2 << 4) - x2
+		skey[v + 3] = (x3 << 4) - x3
+	}
+}
+
+orthogonalize_roundkey :: proc "contextless" (qq: []u64, key: []byte) {
+	if len(qq) < 8 || len(key) != 16 {
+		intrinsics.trap()
+	}
+
+	skey: [4]u32 = ---
+	skey[0] = endian.unchecked_get_u32le(key[0:])
+	skey[1] = endian.unchecked_get_u32le(key[4:])
+	skey[2] = endian.unchecked_get_u32le(key[8:])
+	skey[3] = endian.unchecked_get_u32le(key[12:])
+
+	q: [8]u64 = ---
+	q[0], q[4] = interleave_in(skey[:])
+	q[1] = q[0]
+	q[2] = q[0]
+	q[3] = q[0]
+	q[5] = q[4]
+	q[6] = q[4]
+	q[7] = q[4]
+	orthogonalize(&q)
+
+	comp_skey: [2]u64 = ---
+	comp_skey[0] =
+		(q[0] & 0x1111111111111111) |
+		(q[1] & 0x2222222222222222) |
+		(q[2] & 0x4444444444444444) |
+		(q[3] & 0x8888888888888888)
+	comp_skey[1] =
+		(q[4] & 0x1111111111111111) |
+		(q[5] & 0x2222222222222222) |
+		(q[6] & 0x4444444444444444) |
+		(q[7] & 0x8888888888888888)
+
+	for x, u in comp_skey {
+		x0 := x
+		x1, x2, x3 := x0, x0, x0
+		x0 &= 0x1111111111111111
+		x1 &= 0x2222222222222222
+		x2 &= 0x4444444444444444
+		x3 &= 0x8888888888888888
+		x1 >>= 1
+		x2 >>= 2
+		x3 >>= 3
+		qq[u * 4 + 0] = (x0 << 4) - x0
+		qq[u * 4 + 1] = (x1 << 4) - x1
+		qq[u * 4 + 2] = (x2 << 4) - x2
+		qq[u * 4 + 3] = (x3 << 4) - x3
+	}
+
+	mem.zero_explicit(&skey, size_of(skey))
+	mem.zero_explicit(&q, size_of(q))
+	mem.zero_explicit(&comp_skey, size_of(comp_skey))
+}

core/crypto/_aes/ct64/ghash.odin

@@ -0,0 +1,136 @@
+// Copyright (c) 2016 Thomas Pornin <pornin@bolet.org>
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions
+// are met:
+//
+//   1. Redistributions of source code must retain the above copyright
+//      notice, this list of conditions and the following disclaimer.
+//
+// THIS SOFTWARE IS PROVIDED BY THE AUTHORS “AS IS” AND ANY EXPRESS OR
+// IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+// WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY
+// DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+// GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+// WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+// THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+package aes_ct64
+
+import "base:intrinsics"
+import "core:crypto/_aes"
+import "core:encoding/endian"
+
+@(private = "file")
+bmul64 :: proc "contextless" (x, y: u64) -> u64 {
+	x0 := x & 0x1111111111111111
+	x1 := x & 0x2222222222222222
+	x2 := x & 0x4444444444444444
+	x3 := x & 0x8888888888888888
+	y0 := y & 0x1111111111111111
+	y1 := y & 0x2222222222222222
+	y2 := y & 0x4444444444444444
+	y3 := y & 0x8888888888888888
+	z0 := (x0 * y0) ~ (x1 * y3) ~ (x2 * y2) ~ (x3 * y1)
+	z1 := (x0 * y1) ~ (x1 * y0) ~ (x2 * y3) ~ (x3 * y2)
+	z2 := (x0 * y2) ~ (x1 * y1) ~ (x2 * y0) ~ (x3 * y3)
+	z3 := (x0 * y3) ~ (x1 * y2) ~ (x2 * y1) ~ (x3 * y0)
+	z0 &= 0x1111111111111111
+	z1 &= 0x2222222222222222
+	z2 &= 0x4444444444444444
+	z3 &= 0x8888888888888888
+	return z0 | z1 | z2 | z3
+}
+
+@(private = "file")
+rev64 :: proc "contextless" (x: u64) -> u64 {
+	x := x
+	x = ((x & 0x5555555555555555) << 1) | ((x >> 1) & 0x5555555555555555)
+	x = ((x & 0x3333333333333333) << 2) | ((x >> 2) & 0x3333333333333333)
+	x = ((x & 0x0F0F0F0F0F0F0F0F) << 4) | ((x >> 4) & 0x0F0F0F0F0F0F0F0F)
+	x = ((x & 0x00FF00FF00FF00FF) << 8) | ((x >> 8) & 0x00FF00FF00FF00FF)
+	x = ((x & 0x0000FFFF0000FFFF) << 16) | ((x >> 16) & 0x0000FFFF0000FFFF)
+	return (x << 32) | (x >> 32)
+}
+
+// ghash calculates the GHASH of data, with the key `key`, and input `dst`
+// and `data`, and stores the resulting digest in `dst`.
+//
+// Note: `dst` is both an input and an output, to support easy implementation
+// of GCM.
+ghash :: proc "contextless" (dst, key, data: []byte) {
+	if len(dst) != _aes.GHASH_BLOCK_SIZE || len(key) != _aes.GHASH_BLOCK_SIZE {
+		intrinsics.trap()
+	}
+
+	buf := data
+	l := len(buf)
+
+	y1 := endian.unchecked_get_u64be(dst[0:])
+	y0 := endian.unchecked_get_u64be(dst[8:])
+	h1 := endian.unchecked_get_u64be(key[0:])
+	h0 := endian.unchecked_get_u64be(key[8:])
+	h0r := rev64(h0)
+	h1r := rev64(h1)
+	h2 := h0 ~ h1
+	h2r := h0r ~ h1r
+
+	src: []byte
+	for l > 0 {
+		if l >= _aes.GHASH_BLOCK_SIZE {
+			src = buf
+			buf = buf[_aes.GHASH_BLOCK_SIZE:]
+			l -= _aes.GHASH_BLOCK_SIZE
+		} else {
+			tmp: [_aes.GHASH_BLOCK_SIZE]byte
+			copy(tmp[:], buf)
+			src = tmp[:]
+			l = 0
+		}
+		y1 ~= endian.unchecked_get_u64be(src)
+		y0 ~= endian.unchecked_get_u64be(src[8:])
+
+		y0r := rev64(y0)
+		y1r := rev64(y1)
+		y2 := y0 ~ y1
+		y2r := y0r ~ y1r
+
+		z0 := bmul64(y0, h0)
+		z1 := bmul64(y1, h1)
+		z2 := bmul64(y2, h2)
+		z0h := bmul64(y0r, h0r)
+		z1h := bmul64(y1r, h1r)
+		z2h := bmul64(y2r, h2r)
+		z2 ~= z0 ~ z1
+		z2h ~= z0h ~ z1h
+		z0h = rev64(z0h) >> 1
+		z1h = rev64(z1h) >> 1
+		z2h = rev64(z2h) >> 1
+
+		v0 := z0
+		v1 := z0h ~ z2
+		v2 := z1 ~ z2h
+		v3 := z1h
+
+		v3 = (v3 << 1) | (v2 >> 63)
+		v2 = (v2 << 1) | (v1 >> 63)
+		v1 = (v1 << 1) | (v0 >> 63)
+		v0 = (v0 << 1)
+
+		v2 ~= v0 ~ (v0 >> 1) ~ (v0 >> 2) ~ (v0 >> 7)
+		v1 ~= (v0 << 63) ~ (v0 << 62) ~ (v0 << 57)
+		v3 ~= v1 ~ (v1 >> 1) ~ (v1 >> 2) ~ (v1 >> 7)
+		v2 ~= (v1 << 63) ~ (v1 << 62) ~ (v1 << 57)
+
+		y0 = v2
+		y1 = v3
+	}
+
+	endian.unchecked_put_u64be(dst[0:], y1)
+	endian.unchecked_put_u64be(dst[8:], y0)
+}

core/crypto/_aes/ct64/helpers.odin

@@ -0,0 +1,75 @@
+package aes_ct64
+
+import "base:intrinsics"
+import "core:crypto/_aes"
+import "core:encoding/endian"
+
+load_blockx1 :: proc "contextless" (q: ^[8]u64, src: []byte) {
+	if len(src) != _aes.BLOCK_SIZE {
+		intrinsics.trap()
+	}
+
+	w: [4]u32 = ---
+	w[0] = endian.unchecked_get_u32le(src[0:])
+	w[1] = endian.unchecked_get_u32le(src[4:])
+	w[2] = endian.unchecked_get_u32le(src[8:])
+	w[3] = endian.unchecked_get_u32le(src[12:])
+	q[0], q[4] = interleave_in(w[:])
+	orthogonalize(q)
+}
+
+store_blockx1 :: proc "contextless" (dst: []byte, q: ^[8]u64) {
+	if len(dst) != _aes.BLOCK_SIZE {
+		intrinsics.trap()
+	}
+
+	orthogonalize(q)
+	w0, w1, w2, w3 := interleave_out(q[0], q[4])
+	endian.unchecked_put_u32le(dst[0:], w0)
+	endian.unchecked_put_u32le(dst[4:], w1)
+	endian.unchecked_put_u32le(dst[8:], w2)
+	endian.unchecked_put_u32le(dst[12:], w3)
+}
+
+load_blocks :: proc "contextless" (q: ^[8]u64, src: [][]byte) {
+	if n := len(src); n > STRIDE || n == 0 {
+		intrinsics.trap()
+	}
+
+	w: [4]u32 = ---
+	for s, i in src {
+		if len(s) != _aes.BLOCK_SIZE {
+			intrinsics.trap()
+		}
+
+		w[0] = endian.unchecked_get_u32le(s[0:])
+		w[1] = endian.unchecked_get_u32le(s[4:])
+		w[2] = endian.unchecked_get_u32le(s[8:])
+		w[3] = endian.unchecked_get_u32le(s[12:])
+		q[i], q[i + 4] = interleave_in(w[:])
+	}
+	orthogonalize(q)
+}
+
+store_blocks :: proc "contextless" (dst: [][]byte, q: ^[8]u64) {
+	if n := len(dst); n > STRIDE || n == 0 {
+		intrinsics.trap()
+	}
+
+	orthogonalize(q)
+	for d, i in dst {
+		// Allow storing [0,4] blocks.
+		if d == nil {
+			break
+		}
+		if len(d) != _aes.BLOCK_SIZE {
+			intrinsics.trap()
+		}
+
+		w0, w1, w2, w3 := interleave_out(q[i], q[i + 4])
+		endian.unchecked_put_u32le(d[0:], w0)
+		endian.unchecked_put_u32le(d[4:], w1)
+		endian.unchecked_put_u32le(d[8:], w2)
+		endian.unchecked_put_u32le(d[12:], w3)
+	}
+}

core/crypto/_aes/hw_intel/api.odin

@@ -0,0 +1,43 @@
+//+build amd64
+package aes_hw_intel
+
+import "core:sys/info"
+
+// is_supporte returns true iff hardware accelerated AES
+// is supported.
+is_supported :: proc "contextless" () -> bool {
+	features, ok := info.cpu_features.?
+	if !ok {
+		return false
+	}
+
+	// Note: Everything with AES-NI and PCLMULQDQ has support for
+	// the required SSE extxtensions.
+	req_features :: info.CPU_Features{
+		.sse2,
+		.ssse3,
+		.sse41,
+		.aes,
+		.pclmulqdq,
+	}
+	return features >= req_features
+}
+
+// Context is a keyed AES (ECB) instance.
+Context :: struct {
+	// Note: The ideal thing to do is for the expanded round keys to be
+	// arrays of `__m128i`, however that implies alignment (or using AVX).
+	//
+	// All the people using e-waste processors that don't support an
+	// insturction set that has been around for over 10 years are why
+	// we can't have nice things.
+	_sk_exp_enc: [15][16]byte,
+	_sk_exp_dec: [15][16]byte,
+	_num_rounds: int,
+}
+
+// init initializes a context for AES with the provided key.
+init :: proc(ctx: ^Context, key: []byte) {
+	keysched(ctx, key)
+}
+

core/crypto/_aes/hw_intel/ghash.odin

@@ -0,0 +1,281 @@
+// Copyright (c) 2017 Thomas Pornin <pornin@bolet.org>
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions
+// are met:
+//
+//   1. Redistributions of source code must retain the above copyright
+//      notice, this list of conditions and the following disclaimer.
+//
+// THIS SOFTWARE IS PROVIDED BY THE AUTHORS “AS IS” AND ANY EXPRESS OR
+// IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+// WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY
+// DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+// GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+// WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+// THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+//+build amd64
+package aes_hw_intel
+
+import "base:intrinsics"
+import "core:crypto/_aes"
+import "core:simd"
+import "core:simd/x86"
+
+@(private = "file")
+GHASH_STRIDE_HW :: 4
+@(private = "file")
+GHASH_STRIDE_BYTES_HW :: GHASH_STRIDE_HW * _aes.GHASH_BLOCK_SIZE
+
+// GHASH is defined over elements of GF(2^128) with "full little-endian"
+// representation: leftmost byte is least significant, and, within each
+// byte, leftmost _bit_ is least significant. The natural ordering in
+// x86 is "mixed little-endian": bytes are ordered from least to most
+// significant, but bits within a byte are in most-to-least significant
+// order. Going to full little-endian representation would require
+// reversing bits within each byte, which is doable but expensive.
+//
+// Instead, we go to full big-endian representation, by swapping bytes
+// around, which is done with a single _mm_shuffle_epi8() opcode (it
+// comes with SSSE3; all CPU that offer pclmulqdq also have SSSE3). We
+// can use a full big-endian representation because in a carryless
+// multiplication, we have a nice bit reversal property:
+//
+// rev_128(x) * rev_128(y) = rev_255(x * y)
+//
+// So by using full big-endian, we still get the right result, except
+// that it is right-shifted by 1 bit. The left-shift is relatively
+// inexpensive, and it can be mutualised.
+//
+// Since SSE2 opcodes do not have facilities for shitfting full 128-bit
+// values with bit precision, we have to break down values into 64-bit
+// chunks. We number chunks from 0 to 3 in left to right order.
+
+@(private = "file")
+byteswap_index := transmute(x86.__m128i)simd.i8x16{
+	// Note: simd.i8x16 is reverse order from x86._mm_set_epi8.
+	15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0,
+}
+
+@(private = "file", require_results, enable_target_feature = "sse2,ssse3")
+byteswap :: #force_inline proc "contextless" (x: x86.__m128i) -> x86.__m128i {
+	return x86._mm_shuffle_epi8(x, byteswap_index)
+}
+
+// From a 128-bit value kw, compute kx as the XOR of the two 64-bit
+// halves of kw (into the right half of kx; left half is unspecified),
+// and return kx.
+@(private = "file", require_results, enable_target_feature = "sse2")
+bk :: #force_inline proc "contextless" (kw: x86.__m128i) -> x86.__m128i {
+	return x86._mm_xor_si128(kw, x86._mm_shuffle_epi32(kw, 0x0e))
+}
+
+// Combine two 64-bit values (k0:k1) into a 128-bit (kw) value and
+// the XOR of the two values (kx), and return (kw, kx).
+@(private = "file", enable_target_feature = "sse2")
+pbk :: #force_inline proc "contextless" (k0, k1: x86.__m128i) -> (x86.__m128i, x86.__m128i) {
+	kw := x86._mm_unpacklo_epi64(k1, k0)
+	kx := x86._mm_xor_si128(k0, k1)
+	return kw, kx
+}
+
+// Left-shift by 1 bit a 256-bit value (in four 64-bit words).
+@(private = "file", require_results, enable_target_feature = "sse2")
+sl_256 :: #force_inline proc "contextless" (x0, x1, x2, x3: x86.__m128i) -> (x86.__m128i, x86.__m128i, x86.__m128i, x86.__m128i) {
+	x0, x1, x2, x3 := x0, x1, x2, x3
+
+	x0 = x86._mm_or_si128(x86._mm_slli_epi64(x0, 1), x86._mm_srli_epi64(x1, 63))
+	x1 = x86._mm_or_si128(x86._mm_slli_epi64(x1, 1), x86._mm_srli_epi64(x2, 63))
+	x2 = x86._mm_or_si128(x86._mm_slli_epi64(x2, 1), x86._mm_srli_epi64(x3, 63))
+	x3 = x86._mm_slli_epi64(x3, 1)
+
+	return x0, x1, x2, x3
+}
+
+// Perform reduction in GF(2^128).
+@(private = "file", require_results, enable_target_feature = "sse2")
+reduce_f128 :: #force_inline proc "contextless" (x0, x1, x2, x3: x86.__m128i) -> (x86.__m128i, x86.__m128i) {
+	x0, x1, x2 := x0, x1, x2
+
+	x1 = x86._mm_xor_si128(
+		x1,
+		x86._mm_xor_si128(
+			x86._mm_xor_si128(
+				x3,
+				x86._mm_srli_epi64(x3, 1)),
+			x86._mm_xor_si128(
+				x86._mm_srli_epi64(x3, 2),
+				x86._mm_srli_epi64(x3, 7))))
+	x2 = x86._mm_xor_si128(
+		x86._mm_xor_si128(
+			x2,
+			x86._mm_slli_epi64(x3, 63)),
+		x86._mm_xor_si128(
+			x86._mm_slli_epi64(x3, 62),
+			x86._mm_slli_epi64(x3, 57)))
+	x0 = x86._mm_xor_si128(
+		x0,
+		x86._mm_xor_si128(
+			x86._mm_xor_si128(
+				x2,
+				x86._mm_srli_epi64(x2, 1)),
+			x86._mm_xor_si128(
+				x86._mm_srli_epi64(x2, 2),
+				x86._mm_srli_epi64(x2, 7))))
+	x1 = x86._mm_xor_si128(
+		x86._mm_xor_si128(
+			x1,
+			x86._mm_slli_epi64(x2, 63)),
+		x86._mm_xor_si128(
+			x86._mm_slli_epi64(x2, 62),
+			x86._mm_slli_epi64(x2, 57)))
+
+	return x0, x1
+}
+
+// Square value kw in GF(2^128) into (dw,dx).
+@(private = "file", require_results, enable_target_feature = "sse2,pclmul")
+square_f128 :: #force_inline proc "contextless" (kw: x86.__m128i) -> (x86.__m128i, x86.__m128i) {
+	z1 := x86._mm_clmulepi64_si128(kw, kw, 0x11)
+	z3 := x86._mm_clmulepi64_si128(kw, kw, 0x00)
+	z0 := x86._mm_shuffle_epi32(z1, 0x0E)
+	z2 := x86._mm_shuffle_epi32(z3, 0x0E)
+	z0, z1, z2, z3 = sl_256(z0, z1, z2, z3)
+	z0, z1 = reduce_f128(z0, z1, z2, z3)
+	return pbk(z0, z1)
+}
+
+// ghash calculates the GHASH of data, with the key `key`, and input `dst`
+// and `data`, and stores the resulting digest in `dst`.
+//
+// Note: `dst` is both an input and an output, to support easy implementation
+// of GCM.
+@(enable_target_feature = "sse2,ssse3,pclmul")
+ghash :: proc "contextless" (dst, key, data: []byte) #no_bounds_check {
+	if len(dst) != _aes.GHASH_BLOCK_SIZE || len(key) != _aes.GHASH_BLOCK_SIZE {
+		intrinsics.trap()
+	}
+
+	// Note: BearSSL opts to copy the remainder into a zero-filled
+	// 64-byte buffer.  We do something slightly more simple.
+
+	// Load key and dst (h and y).
+	yw := intrinsics.unaligned_load((^x86.__m128i)(raw_data(dst)))
+	h1w := intrinsics.unaligned_load((^x86.__m128i)(raw_data(key)))
+	yw = byteswap(yw)
+	h1w = byteswap(h1w)
+	h1x := bk(h1w)
+
+	// Process 4 blocks at a time
+	buf := data
+	l := len(buf)
+	if l >= GHASH_STRIDE_BYTES_HW {
+		// Compute h2 = h^2
+		h2w, h2x := square_f128(h1w)
+
+		// Compute h3 = h^3 = h*(h^2)
+		t1 := x86._mm_clmulepi64_si128(h1w, h2w, 0x11)
+		t3 := x86._mm_clmulepi64_si128(h1w, h2w, 0x00)
+		t2 := x86._mm_xor_si128(
+			x86._mm_clmulepi64_si128(h1x, h2x, 0x00),
+			x86._mm_xor_si128(t1, t3))
+		t0 := x86._mm_shuffle_epi32(t1, 0x0E)
+		t1 = x86._mm_xor_si128(t1, x86._mm_shuffle_epi32(t2, 0x0E))
+		t2 = x86._mm_xor_si128(t2, x86._mm_shuffle_epi32(t3, 0x0E))
+		t0, t1, t2, t3 = sl_256(t0, t1, t2, t3)
+		t0, t1 = reduce_f128(t0, t1, t2, t3)
+		h3w, h3x := pbk(t0, t1)
+
+		// Compute h4 = h^4 = (h^2)^2
+		h4w, h4x := square_f128(h2w)
+
+		for l >= GHASH_STRIDE_BYTES_HW {
+			aw0 := intrinsics.unaligned_load((^x86.__m128i)(raw_data(buf)))
+			aw1 := intrinsics.unaligned_load((^x86.__m128i)(raw_data(buf[16:])))
+			aw2 := intrinsics.unaligned_load((^x86.__m128i)(raw_data(buf[32:])))
+			aw3 := intrinsics.unaligned_load((^x86.__m128i)(raw_data(buf[48:])))
+			aw0 = byteswap(aw0)
+			aw1 = byteswap(aw1)
+			aw2 = byteswap(aw2)
+			aw3 = byteswap(aw3)
+			buf, l = buf[GHASH_STRIDE_BYTES_HW:], l - GHASH_STRIDE_BYTES_HW
+
+			aw0 = x86._mm_xor_si128(aw0, yw)
+			ax1 := bk(aw1)
+			ax2 := bk(aw2)
+			ax3 := bk(aw3)
+			ax0 := bk(aw0)
+
+			t1 = x86._mm_xor_si128(
+				x86._mm_xor_si128(
+					x86._mm_clmulepi64_si128(aw0, h4w, 0x11),
+					x86._mm_clmulepi64_si128(aw1, h3w, 0x11)),
+				x86._mm_xor_si128(
+					x86._mm_clmulepi64_si128(aw2, h2w, 0x11),
+					x86._mm_clmulepi64_si128(aw3, h1w, 0x11)))
+			t3 = x86._mm_xor_si128(
+				x86._mm_xor_si128(
+					x86._mm_clmulepi64_si128(aw0, h4w, 0x00),
+					x86._mm_clmulepi64_si128(aw1, h3w, 0x00)),
+				x86._mm_xor_si128(
+					x86._mm_clmulepi64_si128(aw2, h2w, 0x00),
+					x86._mm_clmulepi64_si128(aw3, h1w, 0x00)))
+			t2 = x86._mm_xor_si128(
+				x86._mm_xor_si128(
+					x86._mm_clmulepi64_si128(ax0, h4x, 0x00),
+					x86._mm_clmulepi64_si128(ax1, h3x, 0x00)),
+				x86._mm_xor_si128(
+					x86._mm_clmulepi64_si128(ax2, h2x, 0x00),
+					x86._mm_clmulepi64_si128(ax3, h1x, 0x00)))
+			t2 = x86._mm_xor_si128(t2, x86._mm_xor_si128(t1, t3))
+			t0 = x86._mm_shuffle_epi32(t1, 0x0E)
+			t1 = x86._mm_xor_si128(t1, x86._mm_shuffle_epi32(t2, 0x0E))
+			t2 = x86._mm_xor_si128(t2, x86._mm_shuffle_epi32(t3, 0x0E))
+			t0, t1, t2, t3 = sl_256(t0, t1, t2, t3)
+			t0, t1 = reduce_f128(t0, t1, t2, t3)
+			yw = x86._mm_unpacklo_epi64(t1, t0)
+		}
+	}
+
+	// Process 1 block at a time
+	src: []byte
+	for l > 0 {
+		if l >= _aes.GHASH_BLOCK_SIZE {
+			src = buf
+			buf = buf[_aes.GHASH_BLOCK_SIZE:]
+			l -= _aes.GHASH_BLOCK_SIZE
+		} else {
+			tmp: [_aes.GHASH_BLOCK_SIZE]byte
+			copy(tmp[:], buf)
+			src = tmp[:]
+			l = 0
+		}
+
+		aw := intrinsics.unaligned_load((^x86.__m128i)(raw_data(src)))
+		aw = byteswap(aw)
+
+		aw = x86._mm_xor_si128(aw, yw)
+		ax := bk(aw)
+
+		t1 := x86._mm_clmulepi64_si128(aw, h1w, 0x11)
+		t3 := x86._mm_clmulepi64_si128(aw, h1w, 0x00)
+		t2 := x86._mm_clmulepi64_si128(ax, h1x, 0x00)
+		t2 = x86._mm_xor_si128(t2, x86._mm_xor_si128(t1, t3))
+		t0 := x86._mm_shuffle_epi32(t1, 0x0E)
+		t1 = x86._mm_xor_si128(t1, x86._mm_shuffle_epi32(t2, 0x0E))
+		t2 = x86._mm_xor_si128(t2, x86._mm_shuffle_epi32(t3, 0x0E))
+		t0, t1, t2, t3 = sl_256(t0, t1, t2, t3)
+		t0, t1 = reduce_f128(t0, t1, t2, t3)
+		yw = x86._mm_unpacklo_epi64(t1, t0)
+	}
+
+	// Write back the hash (dst, aka y)
+	yw = byteswap(yw)
+	intrinsics.unaligned_store((^x86.__m128i)(raw_data(dst)), yw)
+}

core/crypto/_aes/hw_intel/hw_intel_keysched.odin

@@ -0,0 +1,178 @@
+// Copyright (c) 2017 Thomas Pornin <pornin@bolet.org>
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions
+// are met:
+//
+//   1. Redistributions of source code must retain the above copyright
+//      notice, this list of conditions and the following disclaimer.
+//
+// THIS SOFTWARE IS PROVIDED BY THE AUTHORS “AS IS” AND ANY EXPRESS OR
+// IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+// WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY
+// DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
+// GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+// WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+// THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+//+build amd64
+package aes_hw_intel
+
+import "base:intrinsics"
+import "core:crypto/_aes"
+import "core:mem"
+import "core:simd/x86"
+
+// Intel AES-NI based implementation.  Inspiration taken from BearSSL.
+//
+// Note: This assumes that the SROA optimization pass is enabled to be
+// anything resembling performat otherwise, LLVM will not elide a massive
+// number of redundant loads/stores it generates for every intrinsic call.
+
+@(private = "file", require_results, enable_target_feature = "sse2")
+expand_step128 :: #force_inline proc(k1, k2: x86.__m128i) -> x86.__m128i {
+	k1, k2 := k1, k2
+
+	k2 = x86._mm_shuffle_epi32(k2, 0xff)
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	return x86._mm_xor_si128(k1, k2)
+}
+
+@(private = "file", require_results, enable_target_feature = "sse,sse2")
+expand_step192a :: #force_inline proc (k1_, k2_: ^x86.__m128i, k3: x86.__m128i) -> (x86.__m128i, x86.__m128i) {
+	k1, k2, k3 := k1_^, k2_^, k3
+
+	k3 = x86._mm_shuffle_epi32(k3, 0x55)
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	k1 = x86._mm_xor_si128(k1, k3)
+
+	tmp := k2
+	k2 = x86._mm_xor_si128(k2, x86._mm_slli_si128(k2, 0x04))
+	k2 = x86._mm_xor_si128(k2, x86._mm_shuffle_epi32(k1, 0xff))
+
+	k1_, k2_ := k1_, k2_
+	k1_^, k2_^ = k1, k2
+
+	r1 := transmute(x86.__m128i)(x86._mm_shuffle_ps(transmute(x86.__m128)(tmp), transmute(x86.__m128)(k1), 0x44))
+	r2 := transmute(x86.__m128i)(x86._mm_shuffle_ps(transmute(x86.__m128)(k1), transmute(x86.__m128)(k2), 0x4e))
+
+	return r1, r2
+}
+
+@(private = "file", require_results, enable_target_feature = "sse2")
+expand_step192b :: #force_inline proc (k1_, k2_: ^x86.__m128i, k3: x86.__m128i) -> x86.__m128i {
+	k1, k2, k3 := k1_^, k2_^, k3
+
+	k3 = x86._mm_shuffle_epi32(k3, 0x55)
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	k1 = x86._mm_xor_si128(k1, k3)
+
+	k2 = x86._mm_xor_si128(k2, x86._mm_slli_si128(k2, 0x04))
+	k2 = x86._mm_xor_si128(k2, x86._mm_shuffle_epi32(k1, 0xff))
+
+	k1_, k2_ := k1_, k2_
+	k1_^, k2_^ = k1, k2
+
+	return k1
+}
+
+@(private = "file", require_results, enable_target_feature = "sse2")
+expand_step256b :: #force_inline proc(k1, k2: x86.__m128i) -> x86.__m128i {
+	k1, k2 := k1, k2
+
+	k2 = x86._mm_shuffle_epi32(k2, 0xaa)
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	k1 = x86._mm_xor_si128(k1, x86._mm_slli_si128(k1, 0x04))
+	return x86._mm_xor_si128(k1, k2)
+}
+
+@(private = "file", enable_target_feature = "aes")
+derive_dec_keys :: proc(ctx: ^Context, sks: ^[15]x86.__m128i, num_rounds: int) {
+	intrinsics.unaligned_store((^x86.__m128i)(&ctx._sk_exp_dec[0]), sks[num_rounds])
+	for i in 1 ..< num_rounds {
+		tmp := x86._mm_aesimc_si128(sks[i])
+		intrinsics.unaligned_store((^x86.__m128i)(&ctx._sk_exp_dec[num_rounds - i]), tmp)
+	}
+	intrinsics.unaligned_store((^x86.__m128i)(&ctx._sk_exp_dec[num_rounds]), sks[0])
+}
+
+@(private, enable_target_feature = "sse,sse2,aes")
+keysched :: proc(ctx: ^Context, key: []byte) {
+	sks: [15]x86.__m128i = ---
+
+	// Compute the encryption keys.
+	num_rounds, key_len := 0, len(key)
+	switch key_len {
+	case _aes.KEY_SIZE_128:
+		sks[0] = intrinsics.unaligned_load((^x86.__m128i)(raw_data(key)))
+		sks[1] = expand_step128(sks[0], x86._mm_aeskeygenassist_si128(sks[0], 0x01))
+		sks[2] = expand_step128(sks[1], x86._mm_aeskeygenassist_si128(sks[1], 0x02))
+		sks[3] = expand_step128(sks[2], x86._mm_aeskeygenassist_si128(sks[2], 0x04))
+		sks[4] = expand_step128(sks[3], x86._mm_aeskeygenassist_si128(sks[3], 0x08))
+		sks[5] = expand_step128(sks[4], x86._mm_aeskeygenassist_si128(sks[4], 0x10))
+		sks[6] = expand_step128(sks[5], x86._mm_aeskeygenassist_si128(sks[5], 0x20))
+		sks[7] = expand_step128(sks[6], x86._mm_aeskeygenassist_si128(sks[6], 0x40))
+		sks[8] = expand_step128(sks[7], x86._mm_aeskeygenassist_si128(sks[7], 0x80))
+		sks[9] = expand_step128(sks[8], x86._mm_aeskeygenassist_si128(sks[8], 0x1b))
+		sks[10] = expand_step128(sks[9], x86._mm_aeskeygenassist_si128(sks[9], 0x36))
+		num_rounds = _aes.ROUNDS_128
+	case _aes.KEY_SIZE_192:
+		k0 := intrinsics.unaligned_load((^x86.__m128i)(raw_data(key)))
+		k1 := x86.__m128i{
+			intrinsics.unaligned_load((^i64)(raw_data(key[16:]))),
+			0,
+		}
+		sks[0] = k0
+		sks[1], sks[2] = expand_step192a(&k0, &k1, x86._mm_aeskeygenassist_si128(k1, 0x01))
+		sks[3] = expand_step192b(&k0, &k1, x86._mm_aeskeygenassist_si128(k1, 0x02))
+		sks[4], sks[5] = expand_step192a(&k0, &k1, x86._mm_aeskeygenassist_si128(k1, 0x04))
+		sks[6] = expand_step192b(&k0, &k1, x86._mm_aeskeygenassist_si128(k1, 0x08))
+		sks[7], sks[8] = expand_step192a(&k0, &k1, x86._mm_aeskeygenassist_si128(k1, 0x10))
+		sks[9] = expand_step192b(&k0, &k1, x86._mm_aeskeygenassist_si128(k1, 0x20))
+		sks[10], sks[11] = expand_step192a(&k0, &k1, x86._mm_aeskeygenassist_si128(k1, 0x40))
+		sks[12] = expand_step192b(&k0, &k1, x86._mm_aeskeygenassist_si128(k1, 0x80))
+		num_rounds = _aes.ROUNDS_192
+	case _aes.KEY_SIZE_256:
+		sks[0] = intrinsics.unaligned_load((^x86.__m128i)(raw_data(key)))
+		sks[1] = intrinsics.unaligned_load((^x86.__m128i)(raw_data(key[16:])))
+		sks[2] = expand_step128(sks[0], x86._mm_aeskeygenassist_si128(sks[1], 0x01))
+		sks[3] = expand_step256b(sks[1], x86._mm_aeskeygenassist_si128(sks[2], 0x01))
+		sks[4] = expand_step128(sks[2], x86._mm_aeskeygenassist_si128(sks[3], 0x02))
+		sks[5] = expand_step256b(sks[3], x86._mm_aeskeygenassist_si128(sks[4], 0x02))
+		sks[6] = expand_step128(sks[4], x86._mm_aeskeygenassist_si128(sks[5], 0x04))
+		sks[7] = expand_step256b(sks[5], x86._mm_aeskeygenassist_si128(sks[6], 0x04))
+		sks[8] = expand_step128(sks[6], x86._mm_aeskeygenassist_si128(sks[7], 0x08))
+		sks[9] = expand_step256b(sks[7], x86._mm_aeskeygenassist_si128(sks[8], 0x08))
+		sks[10] = expand_step128(sks[8], x86._mm_aeskeygenassist_si128(sks[9], 0x10))
+		sks[11] = expand_step256b(sks[9], x86._mm_aeskeygenassist_si128(sks[10], 0x10))
+		sks[12] = expand_step128(sks[10], x86._mm_aeskeygenassist_si128(sks[11], 0x20))
+		sks[13] = expand_step256b(sks[11], x86._mm_aeskeygenassist_si128(sks[12], 0x20))
+		sks[14] = expand_step128(sks[12], x86._mm_aeskeygenassist_si128(sks[13], 0x40))
+		num_rounds = _aes.ROUNDS_256
+	case:
+		panic("crypto/aes: invalid AES key size")
+	}
+	for i in 0 ..= num_rounds {
+		intrinsics.unaligned_store((^x86.__m128i)(&ctx._sk_exp_enc[i]), sks[i])
+	}
+
+	// Compute the decryption keys.  GCM and CTR do not need this, however
+	// ECB, CBC, OCB3, etc do.
+	derive_dec_keys(ctx, &sks, num_rounds)
+
+	ctx._num_rounds = num_rounds
+
+	mem.zero_explicit(&sks, size_of(sks))
+}

core/crypto/_edwards25519/edwards25519.odin

@@ -110,7 +110,7 @@ ge_set_bytes :: proc "contextless" (ge: ^Group_Element, b: []byte) -> bool {
 	if len(b) != 32 {
 		intrinsics.trap()
 	}
-	b_ := transmute(^[32]byte)(raw_data(b))
+	b_ := (^[32]byte)(raw_data(b))
 
 	// Do the work in a scratch element, so that ge is unchanged on
 	// failure.
@@ -169,7 +169,7 @@ ge_bytes :: proc "contextless" (ge: ^Group_Element, dst: []byte) {
 	if len(dst) != 32 {
 		intrinsics.trap()
 	}
-	dst_ := transmute(^[32]byte)(raw_data(dst))
+	dst_ := (^[32]byte)(raw_data(dst))
 
 	// Convert the element to affine (x, y) representation.
 	x, y, z_inv: field.Tight_Field_Element = ---, ---, ---

core/crypto/_edwards25519/edwards25519_scalar.odin

@@ -28,7 +28,7 @@ sc_set_bytes :: proc "contextless" (sc: ^Scalar, b: []byte) -> bool {
 	if len(b) != 32 {
 		intrinsics.trap()
 	}
-	b_ := transmute(^[32]byte)(raw_data(b))
+	b_ := (^[32]byte)(raw_data(b))
 	return field.fe_from_bytes(sc, b_)
 }
 
@@ -36,7 +36,7 @@ sc_set_bytes_rfc8032 :: proc "contextless" (sc: ^Scalar, b: []byte) {
 	if len(b) != 32 {
 		intrinsics.trap()
 	}
-	b_ := transmute(^[32]byte)(raw_data(b))
+	b_ := (^[32]byte)(raw_data(b))
 	field.fe_from_bytes_rfc8032(sc, b_)
 }
 

core/crypto/_fiat/field_curve25519/field.odin

@@ -6,13 +6,13 @@ import "core:mem"
 fe_relax_cast :: #force_inline proc "contextless" (
 	arg1: ^Tight_Field_Element,
 ) -> ^Loose_Field_Element {
-	return transmute(^Loose_Field_Element)(arg1)
+	return (^Loose_Field_Element)(arg1)
 }
 
 fe_tighten_cast :: #force_inline proc "contextless" (
 	arg1: ^Loose_Field_Element,
 ) -> ^Tight_Field_Element {
-	return transmute(^Tight_Field_Element)(arg1)
+	return (^Tight_Field_Element)(arg1)
 }
 
 fe_clear :: proc "contextless" (

core/crypto/_fiat/field_poly1305/field.odin

@@ -7,13 +7,13 @@ import "core:mem"
 fe_relax_cast :: #force_inline proc "contextless" (
 	arg1: ^Tight_Field_Element,
 ) -> ^Loose_Field_Element {
-	return transmute(^Loose_Field_Element)(arg1)
+	return (^Loose_Field_Element)(arg1)
 }
 
 fe_tighten_cast :: #force_inline proc "contextless" (
 	arg1: ^Loose_Field_Element,
 ) -> ^Tight_Field_Element {
-	return transmute(^Tight_Field_Element)(arg1)
+	return (^Tight_Field_Element)(arg1)
 }
 
 fe_from_bytes :: #force_inline proc "contextless" (

core/crypto/aes/aes.odin

@@ -0,0 +1,21 @@
+/*
+package aes implements the AES block cipher and some common modes.
+
+See:
+- https://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.197-upd1.pdf
+- https://nvlpubs.nist.gov/nistpubs/Legacy/SP/nistspecialpublication800-38a.pdf
+- https://nvlpubs.nist.gov/nistpubs/Legacy/SP/nistspecialpublication800-38d.pdf
+*/
+package aes
+
+import "core:crypto/_aes"
+
+// KEY_SIZE_128 is the AES-128 key size in bytes.
+KEY_SIZE_128 :: _aes.KEY_SIZE_128
+// KEY_SIZE_192 is the AES-192 key size in bytes.
+KEY_SIZE_192 :: _aes.KEY_SIZE_192
+// KEY_SIZE_256 is the AES-256 key size in bytes.
+KEY_SIZE_256 :: _aes.KEY_SIZE_256
+
+// BLOCK_SIZE is the AES block size in bytes.
+BLOCK_SIZE :: _aes.BLOCK_SIZE

core/crypto/aes/aes_ctr.odin

@@ -0,0 +1,201 @@
+package aes
+
+import "core:bytes"
+import "core:crypto/_aes/ct64"
+import "core:encoding/endian"
+import "core:math/bits"
+import "core:mem"
+
+// CTR_IV_SIZE is the size of the CTR mode IV in bytes.
+CTR_IV_SIZE :: 16
+
+// Context_CTR is a keyed AES-CTR instance.
+Context_CTR :: struct {
+	_impl:           Context_Impl,
+	_buffer:         [BLOCK_SIZE]byte,
+	_off:            int,
+	_ctr_hi:         u64,
+	_ctr_lo:         u64,
+	_is_initialized: bool,
+}
+
+// init_ctr initializes a Context_CTR with the provided key and IV.
+init_ctr :: proc(ctx: ^Context_CTR, key, iv: []byte, impl := Implementation.Hardware) {
+	if len(iv) != CTR_IV_SIZE {
+		panic("crypto/aes: invalid CTR IV size")
+	}
+
+	init_impl(&ctx._impl, key, impl)
+	ctx._off = BLOCK_SIZE
+	ctx._ctr_hi = endian.unchecked_get_u64be(iv[0:])
+	ctx._ctr_lo = endian.unchecked_get_u64be(iv[8:])
+	ctx._is_initialized = true
+}
+
+// xor_bytes_ctr XORs each byte in src with bytes taken from the AES-CTR
+// keystream, and writes the resulting output to dst.  dst and src MUST
+// alias exactly or not at all.
+xor_bytes_ctr :: proc(ctx: ^Context_CTR, dst, src: []byte) {
+	assert(ctx._is_initialized)
+
+	src, dst := src, dst
+	if dst_len := len(dst); dst_len < len(src) {
+		src = src[:dst_len]
+	}
+
+	if bytes.alias_inexactly(dst, src) {
+		panic("crypto/aes: dst and src alias inexactly")
+	}
+
+	for remaining := len(src); remaining > 0; {
+		// Process multiple blocks at once
+		if ctx._off == BLOCK_SIZE {
+			if nr_blocks := remaining / BLOCK_SIZE; nr_blocks > 0 {
+				direct_bytes := nr_blocks * BLOCK_SIZE
+				ctr_blocks(ctx, dst, src, nr_blocks)
+				remaining -= direct_bytes
+				if remaining == 0 {
+					return
+				}
+				dst = dst[direct_bytes:]
+				src = src[direct_bytes:]
+			}
+
+			// If there is a partial block, generate and buffer 1 block
+			// worth of keystream.
+			ctr_blocks(ctx, ctx._buffer[:], nil, 1)
+			ctx._off = 0
+		}
+
+		// Process partial blocks from the buffered keystream.
+		to_xor := min(BLOCK_SIZE - ctx._off, remaining)
+		buffered_keystream := ctx._buffer[ctx._off:]
+		for i := 0; i < to_xor; i = i + 1 {
+			dst[i] = buffered_keystream[i] ~ src[i]
+		}
+		ctx._off += to_xor
+		dst = dst[to_xor:]
+		src = src[to_xor:]
+		remaining -= to_xor
+	}
+}
+
+// keystream_bytes_ctr fills dst with the raw AES-CTR keystream output.
+keystream_bytes_ctr :: proc(ctx: ^Context_CTR, dst: []byte) {
+	assert(ctx._is_initialized)
+
+	dst := dst
+	for remaining := len(dst); remaining > 0; {
+		// Process multiple blocks at once
+		if ctx._off == BLOCK_SIZE {
+			if nr_blocks := remaining / BLOCK_SIZE; nr_blocks > 0 {
+				direct_bytes := nr_blocks * BLOCK_SIZE
+				ctr_blocks(ctx, dst, nil, nr_blocks)
+				remaining -= direct_bytes
+				if remaining == 0 {
+					return
+				}
+				dst = dst[direct_bytes:]
+			}
+
+			// If there is a partial block, generate and buffer 1 block
+			// worth of keystream.
+			ctr_blocks(ctx, ctx._buffer[:], nil, 1)
+			ctx._off = 0
+		}
+
+		// Process partial blocks from the buffered keystream.
+		to_copy := min(BLOCK_SIZE - ctx._off, remaining)
+		buffered_keystream := ctx._buffer[ctx._off:]
+		copy(dst[:to_copy], buffered_keystream[:to_copy])
+		ctx._off += to_copy
+		dst = dst[to_copy:]
+		remaining -= to_copy
+	}
+}
+
+// reset_ctr sanitizes the Context_CTR.  The Context_CTR must be
+// re-initialized to be used again.
+reset_ctr :: proc "contextless" (ctx: ^Context_CTR) {
+	reset_impl(&ctx._impl)
+	ctx._off = 0
+	ctx._ctr_hi = 0
+	ctx._ctr_lo = 0
+	mem.zero_explicit(&ctx._buffer, size_of(ctx._buffer))
+	ctx._is_initialized = false
+}
+
+@(private = "file")
+ctr_blocks :: proc(ctx: ^Context_CTR, dst, src: []byte, nr_blocks: int) #no_bounds_check {
+	// Use the optimized hardware implementation if available.
+	if _, is_hw := ctx._impl.(Context_Impl_Hardware); is_hw {
+		ctr_blocks_hw(ctx, dst, src, nr_blocks)
+		return
+	}
+
+	// Portable implementation.
+	ct64_inc_ctr := #force_inline proc "contextless" (dst: []byte, hi, lo: u64) -> (u64, u64) {
+		endian.unchecked_put_u64be(dst[0:], hi)
+		endian.unchecked_put_u64be(dst[8:], lo)
+
+		hi, lo := hi, lo
+		carry: u64
+		lo, carry = bits.add_u64(lo, 1, 0)
+		hi, _ = bits.add_u64(hi, 0, carry)
+		return hi, lo
+	}
+
+	impl := &ctx._impl.(ct64.Context)
+	src, dst := src, dst
+	nr_blocks := nr_blocks
+	ctr_hi, ctr_lo := ctx._ctr_hi, ctx._ctr_lo
+
+	tmp: [ct64.STRIDE][BLOCK_SIZE]byte = ---
+	ctrs: [ct64.STRIDE][]byte = ---
+	for i in 0 ..< ct64.STRIDE {
+		ctrs[i] = tmp[i][:]
+	}
+	for nr_blocks > 0 {
+		n := min(ct64.STRIDE, nr_blocks)
+		blocks := ctrs[:n]
+
+		for i in 0 ..< n {
+			ctr_hi, ctr_lo = ct64_inc_ctr(blocks[i], ctr_hi, ctr_lo)
+		}
+		ct64.encrypt_blocks(impl, blocks, blocks)
+
+		xor_blocks(dst, src, blocks)
+
+		if src != nil {
+			src = src[n * BLOCK_SIZE:]
+		}
+		dst = dst[n * BLOCK_SIZE:]
+		nr_blocks -= n
+	}
+
+	// Write back the counter.
+	ctx._ctr_hi, ctx._ctr_lo = ctr_hi, ctr_lo
+
+	mem.zero_explicit(&tmp, size_of(tmp))
+}
+
+@(private)
+xor_blocks :: #force_inline proc "contextless" (dst, src: []byte, blocks: [][]byte) {
+	// Note: This would be faster `core:simd` was used, however if
+	// performance of this implementation matters to where that
+	// optimization would be worth it, use chacha20poly1305, or a
+	// CPU that isn't e-waste.
+	#no_bounds_check {
+		if src != nil {
+				for i in 0 ..< len(blocks) {
+					off := i * BLOCK_SIZE
+					for j in 0 ..< BLOCK_SIZE {
+						blocks[i][j] ~= src[off + j]
+					}
+				}
+		}
+		for i in 0 ..< len(blocks) {
+			copy(dst[i * BLOCK_SIZE:], blocks[i])
+		}
+	}
+}

core/crypto/aes/aes_ctr_hw_intel.odin

@@ -0,0 +1,151 @@
+//+build amd64
+package aes
+
+import "base:intrinsics"
+import "core:crypto/_aes"
+import "core:math/bits"
+import "core:mem"
+import "core:simd/x86"
+
+@(private)
+CTR_STRIDE_HW :: 4
+@(private)
+CTR_STRIDE_BYTES_HW :: CTR_STRIDE_HW * BLOCK_SIZE
+
+@(private, enable_target_feature = "sse2,aes")
+ctr_blocks_hw :: proc(ctx: ^Context_CTR, dst, src: []byte, nr_blocks: int) #no_bounds_check {
+	hw_ctx := ctx._impl.(Context_Impl_Hardware)
+
+	sks: [15]x86.__m128i = ---
+	for i in 0 ..= hw_ctx._num_rounds {
+		sks[i] = intrinsics.unaligned_load((^x86.__m128i)(&hw_ctx._sk_exp_enc[i]))
+	}
+
+	hw_inc_ctr := #force_inline proc "contextless" (hi, lo: u64) -> (x86.__m128i, u64, u64) {
+		ret := x86.__m128i{
+			i64(intrinsics.byte_swap(hi)),
+			i64(intrinsics.byte_swap(lo)),
+		}
+
+		hi, lo := hi, lo
+		carry: u64
+
+		lo, carry = bits.add_u64(lo, 1, 0)
+		hi, _ = bits.add_u64(hi, 0, carry)
+		return ret, hi, lo
+	}
+
+	// The latency of AESENC depends on mfg and microarchitecture:
+	// - 7 -> up to Broadwell
+	// - 4 -> AMD and Skylake - Cascade Lake
+	// - 3 -> Ice Lake and newer
+	//
+	// This implementation does 4 blocks at once, since performance
+	// should be "adequate" across most CPUs.
+
+	src, dst := src, dst
+	nr_blocks := nr_blocks
+	ctr_hi, ctr_lo := ctx._ctr_hi, ctx._ctr_lo
+
+	blks: [CTR_STRIDE_HW]x86.__m128i = ---
+	for nr_blocks >= CTR_STRIDE_HW {
+		#unroll for i in 0..< CTR_STRIDE_HW {
+			blks[i], ctr_hi, ctr_lo = hw_inc_ctr(ctr_hi, ctr_lo)
+		}
+
+		#unroll for i in 0 ..< CTR_STRIDE_HW {
+			blks[i] = x86._mm_xor_si128(blks[i], sks[0])
+		}
+		#unroll for i in 1 ..= 9 {
+			#unroll for j in 0 ..< CTR_STRIDE_HW {
+				blks[j] = x86._mm_aesenc_si128(blks[j], sks[i])
+			}
+		}
+		switch hw_ctx._num_rounds {
+		case _aes.ROUNDS_128:
+			#unroll for i in 0 ..< CTR_STRIDE_HW {
+				blks[i] = x86._mm_aesenclast_si128(blks[i], sks[10])
+			}
+		case _aes.ROUNDS_192:
+			#unroll for i in 10 ..= 11 {
+				#unroll for j in 0 ..< CTR_STRIDE_HW {
+					blks[j] = x86._mm_aesenc_si128(blks[j], sks[i])
+				}
+			}
+			#unroll for i in 0 ..< CTR_STRIDE_HW {
+				blks[i] = x86._mm_aesenclast_si128(blks[i], sks[12])
+			}
+		case _aes.ROUNDS_256:
+			#unroll for i in 10 ..= 13 {
+				#unroll for j in 0 ..< CTR_STRIDE_HW {
+					blks[j] = x86._mm_aesenc_si128(blks[j], sks[i])
+				}
+			}
+			#unroll for i in 0 ..< CTR_STRIDE_HW {
+				blks[i] = x86._mm_aesenclast_si128(blks[i], sks[14])
+			}
+		}
+
+		xor_blocks_hw(dst, src, blks[:])
+
+		if src != nil {
+			src = src[CTR_STRIDE_BYTES_HW:]
+		}
+		dst = dst[CTR_STRIDE_BYTES_HW:]
+		nr_blocks -= CTR_STRIDE_HW
+	}
+
+	// Handle the remainder.
+	for nr_blocks > 0 {
+		blks[0], ctr_hi, ctr_lo = hw_inc_ctr(ctr_hi, ctr_lo)
+
+		blks[0] = x86._mm_xor_si128(blks[0], sks[0])
+		#unroll for i in 1 ..= 9 {
+			blks[0] = x86._mm_aesenc_si128(blks[0], sks[i])
+		}
+		switch hw_ctx._num_rounds {
+		case _aes.ROUNDS_128:
+			blks[0] = x86._mm_aesenclast_si128(blks[0], sks[10])
+		case _aes.ROUNDS_192:
+			#unroll for i in 10 ..= 11 {
+				blks[0] = x86._mm_aesenc_si128(blks[0], sks[i])
+			}
+			blks[0] = x86._mm_aesenclast_si128(blks[0], sks[12])
+		case _aes.ROUNDS_256:
+			#unroll for i in 10 ..= 13 {
+				blks[0] = x86._mm_aesenc_si128(blks[0], sks[i])
+			}
+			blks[0] = x86._mm_aesenclast_si128(blks[0], sks[14])
+		}
+
+		xor_blocks_hw(dst, src, blks[:1])
+
+		if src != nil {
+			src = src[BLOCK_SIZE:]
+		}
+		dst = dst[BLOCK_SIZE:]
+		nr_blocks -= 1
+	}
+
+	// Write back the counter.
+	ctx._ctr_hi, ctx._ctr_lo = ctr_hi, ctr_lo
+
+	mem.zero_explicit(&blks, size_of(blks))
+	mem.zero_explicit(&sks, size_of(sks))
+}
+
+@(private, enable_target_feature = "sse2")
+xor_blocks_hw :: proc(dst, src: []byte, blocks: []x86.__m128i) {
+	#no_bounds_check {
+		if src != nil {
+				for i in 0 ..< len(blocks) {
+					off := i * BLOCK_SIZE
+					tmp := intrinsics.unaligned_load((^x86.__m128i)(raw_data(src[off:])))
+					blocks[i] = x86._mm_xor_si128(blocks[i], tmp)
+				}
+		}
+		for i in 0 ..< len(blocks) {
+			intrinsics.unaligned_store((^x86.__m128i)(raw_data(dst[i * BLOCK_SIZE:])), blocks[i])
+		}
+	}
+}

core/crypto/aes/aes_ecb.odin

@@ -0,0 +1,57 @@
+package aes
+
+import "core:crypto/_aes/ct64"
+
+// Context_ECB is a keyed AES-ECB instance.
+//
+// WARNING: Using ECB mode is strongly discouraged unless it is being
+// used to implement higher level constructs.
+Context_ECB :: struct {
+	_impl:           Context_Impl,
+	_is_initialized: bool,
+}
+
+// init_ecb initializes a Context_ECB with the provided key.
+init_ecb :: proc(ctx: ^Context_ECB, key: []byte, impl := Implementation.Hardware) {
+	init_impl(&ctx._impl, key, impl)
+	ctx._is_initialized = true
+}
+
+// encrypt_ecb encrypts the BLOCK_SIZE buffer src, and writes the result to dst.
+encrypt_ecb :: proc(ctx: ^Context_ECB, dst, src: []byte) {
+	assert(ctx._is_initialized)
+
+	if len(dst) != BLOCK_SIZE || len(src) != BLOCK_SIZE {
+		panic("crypto/aes: invalid buffer size(s)")
+	}
+
+	switch &impl in ctx._impl {
+	case ct64.Context:
+		ct64.encrypt_block(&impl, dst, src)
+	case Context_Impl_Hardware:
+		encrypt_block_hw(&impl, dst, src)
+	}
+}
+
+// decrypt_ecb decrypts the BLOCK_SIZE buffer src, and writes the result to dst.
+decrypt_ecb :: proc(ctx: ^Context_ECB, dst, src: []byte) {
+	assert(ctx._is_initialized)
+
+	if len(dst) != BLOCK_SIZE || len(src) != BLOCK_SIZE {
+		panic("crypto/aes: invalid buffer size(s)")
+	}
+
+	switch &impl in ctx._impl {
+	case ct64.Context:
+		ct64.decrypt_block(&impl, dst, src)
+	case Context_Impl_Hardware:
+		decrypt_block_hw(&impl, dst, src)
+	}
+}
+
+// reset_ecb sanitizes the Context_ECB.  The Context_ECB must be
+// re-initialized to be used again.
+reset_ecb :: proc "contextless" (ctx: ^Context_ECB) {
+	reset_impl(&ctx._impl)
+	ctx._is_initialized = false
+}

core/crypto/aes/aes_ecb_hw_intel.odin

@@ -0,0 +1,58 @@
+//+build amd64
+package aes
+
+import "base:intrinsics"
+import "core:crypto/_aes"
+import "core:simd/x86"
+
+@(private, enable_target_feature = "sse2,aes")
+encrypt_block_hw :: proc(ctx: ^Context_Impl_Hardware, dst, src: []byte) {
+	blk := intrinsics.unaligned_load((^x86.__m128i)(raw_data(src)))
+
+	blk = x86._mm_xor_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_enc[0])))
+	#unroll for i in 1 ..= 9 {
+		blk = x86._mm_aesenc_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_enc[i])))
+	}
+	switch ctx._num_rounds {
+	case _aes.ROUNDS_128:
+		blk = x86._mm_aesenclast_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_enc[10])))
+	case _aes.ROUNDS_192:
+		#unroll for i in 10 ..= 11 {
+			blk = x86._mm_aesenc_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_enc[i])))
+		}
+		blk = x86._mm_aesenclast_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_enc[12])))
+	case _aes.ROUNDS_256:
+		#unroll for i in 10 ..= 13 {
+			blk = x86._mm_aesenc_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_enc[i])))
+		}
+		blk = x86._mm_aesenclast_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_enc[14])))
+	}
+
+	intrinsics.unaligned_store((^x86.__m128i)(raw_data(dst)), blk)
+}
+
+@(private, enable_target_feature = "sse2,aes")
+decrypt_block_hw :: proc(ctx: ^Context_Impl_Hardware, dst, src: []byte) {
+	blk := intrinsics.unaligned_load((^x86.__m128i)(raw_data(src)))
+
+	blk = x86._mm_xor_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_dec[0])))
+	#unroll for i in 1 ..= 9 {
+		blk = x86._mm_aesdec_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_dec[i])))
+	}
+	switch ctx._num_rounds {
+	case _aes.ROUNDS_128:
+		blk = x86._mm_aesdeclast_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_dec[10])))
+	case _aes.ROUNDS_192:
+		#unroll for i in 10 ..= 11 {
+			blk = x86._mm_aesdec_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_dec[i])))
+		}
+		blk = x86._mm_aesdeclast_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_dec[12])))
+	case _aes.ROUNDS_256:
+		#unroll for i in 10 ..= 13 {
+			blk = x86._mm_aesdec_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_dec[i])))
+		}
+		blk = x86._mm_aesdeclast_si128(blk, intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_dec[14])))
+	}
+
+	intrinsics.unaligned_store((^x86.__m128i)(raw_data(dst)), blk)
+}

core/crypto/aes/aes_gcm.odin

@@ -0,0 +1,269 @@
+package aes
+
+import "core:bytes"
+import "core:crypto"
+import "core:crypto/_aes"
+import "core:crypto/_aes/ct64"
+import "core:encoding/endian"
+import "core:mem"
+
+// GCM_NONCE_SIZE is the default size of the GCM nonce in bytes.
+GCM_NONCE_SIZE :: 12
+// GCM_NONCE_SIZE_MAX is the maximum size of the GCM nonce in bytes.
+GCM_NONCE_SIZE_MAX :: 0x2000000000000000 // floor((2^64 - 1) / 8) bits
+// GCM_TAG_SIZE is the size of a GCM tag in bytes.
+GCM_TAG_SIZE :: _aes.GHASH_TAG_SIZE
+
+@(private)
+GCM_A_MAX :: max(u64) / 8 // 2^64 - 1 bits -> bytes
+@(private)
+GCM_P_MAX :: 0xfffffffe0 // 2^39 - 256 bits -> bytes
+
+// Context_GCM is a keyed AES-GCM instance.
+Context_GCM :: struct {
+	_impl:           Context_Impl,
+	_is_initialized: bool,
+}
+
+// init_gcm initializes a Context_GCM with the provided key.
+init_gcm :: proc(ctx: ^Context_GCM, key: []byte, impl := Implementation.Hardware) {
+	init_impl(&ctx._impl, key, impl)
+	ctx._is_initialized = true
+}
+
+// seal_gcm encrypts the plaintext and authenticates the aad and ciphertext,
+// with the provided Context_GCM and nonce, stores the output in dst and tag.
+//
+// dst and plaintext MUST alias exactly or not at all.
+seal_gcm :: proc(ctx: ^Context_GCM, dst, tag, nonce, aad, plaintext: []byte) {
+	assert(ctx._is_initialized)
+
+	gcm_validate_common_slice_sizes(tag, nonce, aad, plaintext)
+	if len(dst) != len(plaintext) {
+		panic("crypto/aes: invalid destination ciphertext size")
+	}
+	if bytes.alias_inexactly(dst, plaintext) {
+		panic("crypto/aes: dst and plaintext alias inexactly")
+	}
+
+	if impl, is_hw := ctx._impl.(Context_Impl_Hardware); is_hw {
+		gcm_seal_hw(&impl, dst, tag, nonce, aad, plaintext)
+		return
+	}
+
+	h: [_aes.GHASH_KEY_SIZE]byte
+	j0: [_aes.GHASH_BLOCK_SIZE]byte
+	j0_enc: [_aes.GHASH_BLOCK_SIZE]byte
+	s: [_aes.GHASH_TAG_SIZE]byte
+	init_ghash_ct64(ctx, &h, &j0, &j0_enc, nonce)
+
+	// Note: Our GHASH implementation handles appending padding.
+	ct64.ghash(s[:], h[:], aad)
+	gctr_ct64(ctx, dst, &s, plaintext, &h, &j0, true)
+	final_ghash_ct64(&s, &h, &j0_enc, len(aad), len(plaintext))
+	copy(tag, s[:])
+
+	mem.zero_explicit(&h, len(h))
+	mem.zero_explicit(&j0, len(j0))
+	mem.zero_explicit(&j0_enc, len(j0_enc))
+}
+
+// open_gcm authenticates the aad and ciphertext, and decrypts the ciphertext,
+// with the provided Context_GCM, nonce, and tag, and stores the output in dst,
+// returning true iff the authentication was successful.  If authentication
+// fails, the destination buffer will be zeroed.
+//
+// dst and plaintext MUST alias exactly or not at all.
+open_gcm :: proc(ctx: ^Context_GCM, dst, nonce, aad, ciphertext, tag: []byte) -> bool {
+	assert(ctx._is_initialized)
+
+	gcm_validate_common_slice_sizes(tag, nonce, aad, ciphertext)
+	if len(dst) != len(ciphertext) {
+		panic("crypto/aes: invalid destination plaintext size")
+	}
+	if bytes.alias_inexactly(dst, ciphertext) {
+		panic("crypto/aes: dst and ciphertext alias inexactly")
+	}
+
+	if impl, is_hw := ctx._impl.(Context_Impl_Hardware); is_hw {
+		return gcm_open_hw(&impl, dst, nonce, aad, ciphertext, tag)
+	}
+
+	h: [_aes.GHASH_KEY_SIZE]byte
+	j0: [_aes.GHASH_BLOCK_SIZE]byte
+	j0_enc: [_aes.GHASH_BLOCK_SIZE]byte
+	s: [_aes.GHASH_TAG_SIZE]byte
+	init_ghash_ct64(ctx, &h, &j0, &j0_enc, nonce)
+
+	ct64.ghash(s[:], h[:], aad)
+	gctr_ct64(ctx, dst, &s, ciphertext, &h, &j0, false)
+	final_ghash_ct64(&s, &h, &j0_enc, len(aad), len(ciphertext))
+
+	ok := crypto.compare_constant_time(s[:], tag) == 1
+	if !ok {
+		mem.zero_explicit(raw_data(dst), len(dst))
+	}
+
+	mem.zero_explicit(&h, len(h))
+	mem.zero_explicit(&j0, len(j0))
+	mem.zero_explicit(&j0_enc, len(j0_enc))
+	mem.zero_explicit(&s, len(s))
+
+	return ok
+}
+
+// reset_ctr sanitizes the Context_GCM.  The Context_GCM must be
+// re-initialized to be used again.
+reset_gcm :: proc "contextless" (ctx: ^Context_GCM) {
+	reset_impl(&ctx._impl)
+	ctx._is_initialized = false
+}
+
+@(private = "file")
+gcm_validate_common_slice_sizes :: proc(tag, nonce, aad, text: []byte) {
+	if len(tag) != GCM_TAG_SIZE {
+		panic("crypto/aes: invalid GCM tag size")
+	}
+
+	// The specification supports nonces in the range [1, 2^64) bits.
+	if l := len(nonce); l == 0 || u64(l) >= GCM_NONCE_SIZE_MAX {
+		panic("crypto/aes: invalid GCM nonce size")
+	}
+
+	if aad_len := u64(len(aad)); aad_len > GCM_A_MAX {
+		panic("crypto/aes: oversized GCM aad")
+	}
+	if text_len := u64(len(text)); text_len > GCM_P_MAX {
+		panic("crypto/aes: oversized GCM src data")
+	}
+}
+
+@(private = "file")
+init_ghash_ct64 :: proc(
+	ctx: ^Context_GCM,
+	h: ^[_aes.GHASH_KEY_SIZE]byte,
+	j0: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	j0_enc: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	nonce: []byte,
+) {
+	impl := &ctx._impl.(ct64.Context)
+
+	// 1. Let H = CIPH(k, 0^128)
+	ct64.encrypt_block(impl, h[:], h[:])
+
+	// Define a block, J0, as follows:
+	if l := len(nonce); l == GCM_NONCE_SIZE {
+		// if len(IV) = 96, then let J0 = IV || 0^31 || 1
+		copy(j0[:], nonce)
+		j0[_aes.GHASH_BLOCK_SIZE - 1] = 1
+	} else {
+		// If len(IV) != 96, then let s = 128 ceil(len(IV)/128) - len(IV),
+		// and let J0 = GHASHH(IV || 0^(s+64) || ceil(len(IV))^64).
+		ct64.ghash(j0[:], h[:], nonce)
+
+		tmp: [_aes.GHASH_BLOCK_SIZE]byte
+		endian.unchecked_put_u64be(tmp[8:], u64(l) * 8)
+		ct64.ghash(j0[:], h[:], tmp[:])
+	}
+
+	// ECB encrypt j0, so that we can just XOR with the tag.  In theory
+	// this could be processed along with the final GCTR block, to
+	// potentially save a call to AES-ECB, but... just use AES-NI.
+	ct64.encrypt_block(impl, j0_enc[:], j0[:])
+}
+
+@(private = "file")
+final_ghash_ct64 :: proc(
+	s: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	h: ^[_aes.GHASH_KEY_SIZE]byte,
+	j0: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	a_len: int,
+	t_len: int,
+) {
+	blk: [_aes.GHASH_BLOCK_SIZE]byte
+	endian.unchecked_put_u64be(blk[0:], u64(a_len) * 8)
+	endian.unchecked_put_u64be(blk[8:], u64(t_len) * 8)
+
+	ct64.ghash(s[:], h[:], blk[:])
+	for i in 0 ..< len(s) {
+		s[i] ~= j0[i]
+	}
+}
+
+@(private = "file")
+gctr_ct64 :: proc(
+	ctx: ^Context_GCM,
+	dst: []byte,
+	s: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	src: []byte,
+	h: ^[_aes.GHASH_KEY_SIZE]byte,
+	nonce: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	is_seal: bool,
+) #no_bounds_check {
+	ct64_inc_ctr32 := #force_inline proc "contextless" (dst: []byte, ctr: u32) -> u32 {
+		endian.unchecked_put_u32be(dst[12:], ctr)
+		return ctr + 1
+	}
+
+	// Setup the counter blocks.
+	tmp, tmp2: [ct64.STRIDE][BLOCK_SIZE]byte = ---, ---
+	ctrs, blks: [ct64.STRIDE][]byte = ---, ---
+	ctr := endian.unchecked_get_u32be(nonce[GCM_NONCE_SIZE:]) + 1
+	for i in 0 ..< ct64.STRIDE {
+		// Setup scratch space for the keystream.
+		blks[i] = tmp2[i][:]
+
+		// Pre-copy the IV to all the counter blocks.
+		ctrs[i] = tmp[i][:]
+		copy(ctrs[i], nonce[:GCM_NONCE_SIZE])
+	}
+
+	impl := &ctx._impl.(ct64.Context)
+	src, dst := src, dst
+
+	nr_blocks := len(src) / BLOCK_SIZE
+	for nr_blocks > 0 {
+		n := min(ct64.STRIDE, nr_blocks)
+		l := n * BLOCK_SIZE
+
+		if !is_seal {
+			ct64.ghash(s[:], h[:], src[:l])
+		}
+
+		// The keystream is written to a separate buffer, as we will
+		// reuse the first 96-bits of each counter.
+		for i in 0 ..< n {
+			ctr = ct64_inc_ctr32(ctrs[i], ctr)
+		}
+		ct64.encrypt_blocks(impl, blks[:n], ctrs[:n])
+
+		xor_blocks(dst, src, blks[:n])
+
+		if is_seal {
+			ct64.ghash(s[:], h[:], dst[:l])
+		}
+
+		src = src[l:]
+		dst = dst[l:]
+		nr_blocks -= n
+	}
+	if l := len(src); l > 0 {
+		if !is_seal {
+			ct64.ghash(s[:], h[:], src[:l])
+		}
+
+		ct64_inc_ctr32(ctrs[0], ctr)
+		ct64.encrypt_block(impl, ctrs[0], ctrs[0])
+
+		for i in 0 ..< l {
+			dst[i] = src[i] ~ ctrs[0][i]
+		}
+
+		if is_seal {
+			ct64.ghash(s[:], h[:], dst[:l])
+		}
+	}
+
+	mem.zero_explicit(&tmp, size_of(tmp))
+	mem.zero_explicit(&tmp2, size_of(tmp2))
+}

core/crypto/aes/aes_gcm_hw_intel.odin

@@ -0,0 +1,243 @@
+//+build amd64
+package aes
+
+import "base:intrinsics"
+import "core:crypto"
+import "core:crypto/_aes"
+import "core:crypto/_aes/hw_intel"
+import "core:encoding/endian"
+import "core:mem"
+import "core:simd/x86"
+
+@(private)
+gcm_seal_hw :: proc(ctx: ^Context_Impl_Hardware, dst, tag, nonce, aad, plaintext: []byte) {
+	h: [_aes.GHASH_KEY_SIZE]byte
+	j0: [_aes.GHASH_BLOCK_SIZE]byte
+	j0_enc: [_aes.GHASH_BLOCK_SIZE]byte
+	s: [_aes.GHASH_TAG_SIZE]byte
+	init_ghash_hw(ctx, &h, &j0, &j0_enc, nonce)
+
+	// Note: Our GHASH implementation handles appending padding.
+	hw_intel.ghash(s[:], h[:], aad)
+	gctr_hw(ctx, dst, &s, plaintext, &h, &j0, true)
+	final_ghash_hw(&s, &h, &j0_enc, len(aad), len(plaintext))
+	copy(tag, s[:])
+
+	mem.zero_explicit(&h, len(h))
+	mem.zero_explicit(&j0, len(j0))
+	mem.zero_explicit(&j0_enc, len(j0_enc))
+}
+
+@(private)
+gcm_open_hw :: proc(ctx: ^Context_Impl_Hardware, dst, nonce, aad, ciphertext, tag: []byte) -> bool {
+	h: [_aes.GHASH_KEY_SIZE]byte
+	j0: [_aes.GHASH_BLOCK_SIZE]byte
+	j0_enc: [_aes.GHASH_BLOCK_SIZE]byte
+	s: [_aes.GHASH_TAG_SIZE]byte
+	init_ghash_hw(ctx, &h, &j0, &j0_enc, nonce)
+
+	hw_intel.ghash(s[:], h[:], aad)
+	gctr_hw(ctx, dst, &s, ciphertext, &h, &j0, false)
+	final_ghash_hw(&s, &h, &j0_enc, len(aad), len(ciphertext))
+
+	ok := crypto.compare_constant_time(s[:], tag) == 1
+	if !ok {
+		mem.zero_explicit(raw_data(dst), len(dst))
+	}
+
+	mem.zero_explicit(&h, len(h))
+	mem.zero_explicit(&j0, len(j0))
+	mem.zero_explicit(&j0_enc, len(j0_enc))
+	mem.zero_explicit(&s, len(s))
+
+	return ok
+}
+
+@(private = "file")
+init_ghash_hw :: proc(
+	ctx: ^Context_Impl_Hardware,
+	h: ^[_aes.GHASH_KEY_SIZE]byte,
+	j0: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	j0_enc: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	nonce: []byte,
+) {
+	// 1. Let H = CIPH(k, 0^128)
+	encrypt_block_hw(ctx, h[:], h[:])
+
+	// Define a block, J0, as follows:
+	if l := len(nonce); l == GCM_NONCE_SIZE {
+		// if len(IV) = 96, then let J0 = IV || 0^31 || 1
+		copy(j0[:], nonce)
+		j0[_aes.GHASH_BLOCK_SIZE - 1] = 1
+	} else {
+		// If len(IV) != 96, then let s = 128 ceil(len(IV)/128) - len(IV),
+		// and let J0 = GHASHH(IV || 0^(s+64) || ceil(len(IV))^64).
+		hw_intel.ghash(j0[:], h[:], nonce)
+
+		tmp: [_aes.GHASH_BLOCK_SIZE]byte
+		endian.unchecked_put_u64be(tmp[8:], u64(l) * 8)
+		hw_intel.ghash(j0[:], h[:], tmp[:])
+	}
+
+	// ECB encrypt j0, so that we can just XOR with the tag.
+	encrypt_block_hw(ctx, j0_enc[:], j0[:])
+}
+
+@(private = "file", enable_target_feature = "sse2")
+final_ghash_hw :: proc(
+	s: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	h: ^[_aes.GHASH_KEY_SIZE]byte,
+	j0: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	a_len: int,
+	t_len: int,
+) {
+	blk: [_aes.GHASH_BLOCK_SIZE]byte
+	endian.unchecked_put_u64be(blk[0:], u64(a_len) * 8)
+	endian.unchecked_put_u64be(blk[8:], u64(t_len) * 8)
+
+	hw_intel.ghash(s[:], h[:], blk[:])
+	j0_vec := intrinsics.unaligned_load((^x86.__m128i)(j0))
+	s_vec := intrinsics.unaligned_load((^x86.__m128i)(s))
+	s_vec = x86._mm_xor_si128(s_vec, j0_vec)
+	intrinsics.unaligned_store((^x86.__m128i)(s), s_vec)
+}
+
+@(private = "file", enable_target_feature = "sse2,sse4.1,aes")
+gctr_hw :: proc(
+	ctx: ^Context_Impl_Hardware,
+	dst: []byte,
+	s: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	src: []byte,
+	h: ^[_aes.GHASH_KEY_SIZE]byte,
+	nonce: ^[_aes.GHASH_BLOCK_SIZE]byte,
+	is_seal: bool,
+) #no_bounds_check {
+	sks: [15]x86.__m128i = ---
+	for i in 0 ..= ctx._num_rounds {
+		sks[i] = intrinsics.unaligned_load((^x86.__m128i)(&ctx._sk_exp_enc[i]))
+	}
+
+	// Setup the counter block
+	ctr_blk := intrinsics.unaligned_load((^x86.__m128i)(nonce))
+	ctr := endian.unchecked_get_u32be(nonce[GCM_NONCE_SIZE:]) + 1
+
+	src, dst := src, dst
+
+	// Note: Instead of doing GHASH and CTR separately, it is more
+	// performant to interleave (stitch) the two operations together.
+	// This results in an unreadable mess, so we opt for simplicity
+	// as performance is adequate.
+
+	blks: [CTR_STRIDE_HW]x86.__m128i = ---
+	nr_blocks := len(src) / BLOCK_SIZE
+	for nr_blocks >= CTR_STRIDE_HW {
+		if !is_seal {
+			hw_intel.ghash(s[:], h[:], src[:CTR_STRIDE_BYTES_HW])
+		}
+
+		#unroll for i in 0 ..< CTR_STRIDE_HW {
+			blks[i], ctr = hw_inc_ctr32(&ctr_blk, ctr)
+		}
+
+		#unroll for i in 0 ..< CTR_STRIDE_HW {
+			blks[i] = x86._mm_xor_si128(blks[i], sks[0])
+		}
+		#unroll for i in 1 ..= 9 {
+			#unroll for j in 0 ..< CTR_STRIDE_HW {
+				blks[j] = x86._mm_aesenc_si128(blks[j], sks[i])
+			}
+		}
+		switch ctx._num_rounds {
+		case _aes.ROUNDS_128:
+			#unroll for i in 0 ..< CTR_STRIDE_HW {
+				blks[i] = x86._mm_aesenclast_si128(blks[i], sks[10])
+			}
+		case _aes.ROUNDS_192:
+			#unroll for i in 10 ..= 11 {
+				#unroll for j in 0 ..< CTR_STRIDE_HW {
+					blks[j] = x86._mm_aesenc_si128(blks[j], sks[i])
+				}
+			}
+			#unroll for i in 0 ..< CTR_STRIDE_HW {
+				blks[i] = x86._mm_aesenclast_si128(blks[i], sks[12])
+			}
+		case _aes.ROUNDS_256:
+			#unroll for i in 10 ..= 13 {
+				#unroll for j in 0 ..< CTR_STRIDE_HW {
+					blks[j] = x86._mm_aesenc_si128(blks[j], sks[i])
+				}
+			}
+			#unroll for i in 0 ..< CTR_STRIDE_HW {
+				blks[i] = x86._mm_aesenclast_si128(blks[i], sks[14])
+			}
+		}
+
+		xor_blocks_hw(dst, src, blks[:])
+
+		if is_seal {
+			hw_intel.ghash(s[:], h[:], dst[:CTR_STRIDE_BYTES_HW])
+		}
+
+		src = src[CTR_STRIDE_BYTES_HW:]
+		dst = dst[CTR_STRIDE_BYTES_HW:]
+		nr_blocks -= CTR_STRIDE_HW
+	}
+
+	// Handle the remainder.
+	for n := len(src); n > 0; {
+		l := min(n, BLOCK_SIZE)
+		if !is_seal {
+			hw_intel.ghash(s[:], h[:], src[:l])
+		}
+
+		blks[0], ctr = hw_inc_ctr32(&ctr_blk, ctr)
+
+		blks[0] = x86._mm_xor_si128(blks[0], sks[0])
+		#unroll for i in 1 ..= 9 {
+			blks[0] = x86._mm_aesenc_si128(blks[0], sks[i])
+		}
+		switch ctx._num_rounds {
+		case _aes.ROUNDS_128:
+			blks[0] = x86._mm_aesenclast_si128(blks[0], sks[10])
+		case _aes.ROUNDS_192:
+			#unroll for i in 10 ..= 11 {
+				blks[0] = x86._mm_aesenc_si128(blks[0], sks[i])
+			}
+			blks[0] = x86._mm_aesenclast_si128(blks[0], sks[12])
+		case _aes.ROUNDS_256:
+			#unroll for i in 10 ..= 13 {
+				blks[0] = x86._mm_aesenc_si128(blks[0], sks[i])
+			}
+			blks[0] = x86._mm_aesenclast_si128(blks[0], sks[14])
+		}
+
+		if l == BLOCK_SIZE {
+			xor_blocks_hw(dst, src, blks[:1])
+		} else {
+			blk: [BLOCK_SIZE]byte
+			copy(blk[:], src)
+			xor_blocks_hw(blk[:], blk[:], blks[:1])
+			copy(dst, blk[:l])
+		}
+		if is_seal {
+			hw_intel.ghash(s[:], h[:], dst[:l])
+		}
+
+		dst = dst[l:]
+		src = src[l:]
+		n -= l
+	}
+
+	mem.zero_explicit(&blks, size_of(blks))
+	mem.zero_explicit(&sks, size_of(sks))
+}
+
+// BUG: Sticking this in gctr_hw (like the other implementations) crashes
+// the compiler.
+//
+// src/check_expr.cpp(7892): Assertion Failure: `c->curr_proc_decl->entity`
+@(private = "file", enable_target_feature = "sse4.1")
+hw_inc_ctr32 :: #force_inline proc "contextless" (src: ^x86.__m128i, ctr: u32) -> (x86.__m128i, u32) {
+	ret := x86._mm_insert_epi32(src^, i32(intrinsics.byte_swap(ctr)), 3)
+	return ret, ctr + 1
+}

core/crypto/aes/aes_impl.odin

@@ -0,0 +1,41 @@
+package aes
+
+import "core:crypto/_aes/ct64"
+import "core:mem"
+import "core:reflect"
+
+@(private)
+Context_Impl :: union {
+	ct64.Context,
+	Context_Impl_Hardware,
+}
+
+// Implementation is an AES implementation.  Most callers will not need
+// to use this as the package will automatically select the most performant
+// implementation available (See `is_hardware_accelerated()`).
+Implementation :: enum {
+	Portable,
+	Hardware,
+}
+
+@(private)
+init_impl :: proc(ctx: ^Context_Impl, key: []byte, impl: Implementation) {
+	impl := impl
+	if !is_hardware_accelerated() {
+		impl = .Portable
+	}
+
+	switch impl {
+	case .Portable:
+		reflect.set_union_variant_typeid(ctx^, typeid_of(ct64.Context))
+		ct64.init(&ctx.(ct64.Context), key)
+	case .Hardware:
+		reflect.set_union_variant_typeid(ctx^, typeid_of(Context_Impl_Hardware))
+		init_impl_hw(&ctx.(Context_Impl_Hardware), key)
+	}
+}
+
+@(private)
+reset_impl :: proc "contextless" (ctx: ^Context_Impl) {
+	mem.zero_explicit(ctx, size_of(Context_Impl))
+}

core/crypto/aes/aes_impl_hw_gen.odin

@@ -0,0 +1,44 @@
+//+build !amd64
+package aes
+
+@(private = "file")
+ERR_HW_NOT_SUPPORTED :: "crypto/aes: hardware implementation unsupported"
+
+// is_hardware_accelerated returns true iff hardware accelerated AES
+// is supported.
+is_hardware_accelerated :: proc "contextless" () -> bool {
+	return false
+}
+
+@(private)
+Context_Impl_Hardware :: struct {}
+
+@(private)
+init_impl_hw :: proc(ctx: ^Context_Impl_Hardware, key: []byte) {
+	panic(ERR_HW_NOT_SUPPORTED)
+}
+
+@(private)
+encrypt_block_hw :: proc(ctx: ^Context_Impl_Hardware, dst, src: []byte) {
+	panic(ERR_HW_NOT_SUPPORTED)
+}
+
+@(private)
+decrypt_block_hw :: proc(ctx: ^Context_Impl_Hardware, dst, src: []byte) {
+	panic(ERR_HW_NOT_SUPPORTED)
+}
+
+@(private)
+ctr_blocks_hw :: proc(ctx: ^Context_CTR, dst, src: []byte, nr_blocks: int) {
+	panic(ERR_HW_NOT_SUPPORTED)
+}
+
+@(private)
+gcm_seal_hw :: proc(ctx: ^Context_Impl_Hardware, dst, tag, nonce, aad, plaintext: []byte) {
+	panic(ERR_HW_NOT_SUPPORTED)
+}
+
+@(private)
+gcm_open_hw :: proc(ctx: ^Context_Impl_Hardware, dst, nonce, aad, ciphertext, tag: []byte) -> bool {
+	panic(ERR_HW_NOT_SUPPORTED)
+}

core/crypto/aes/aes_impl_hw_intel.odin

@@ -0,0 +1,18 @@
+//+build amd64
+package aes
+
+import "core:crypto/_aes/hw_intel"
+
+// is_hardware_accelerated returns true iff hardware accelerated AES
+// is supported.
+is_hardware_accelerated :: proc "contextless" () -> bool {
+	return hw_intel.is_supported()
+}
+
+@(private)
+Context_Impl_Hardware :: hw_intel.Context
+
+@(private, enable_target_feature = "sse2,aes")
+init_impl_hw :: proc(ctx: ^Context_Impl_Hardware, key: []byte) {
+	hw_intel.init(ctx, key)
+}

core/crypto/chacha20/chacha20.odin

@@ -7,6 +7,7 @@ See:
 */
 package chacha20
 
+import "core:bytes"
 import "core:encoding/endian"
 import "core:math/bits"
 import "core:mem"
@@ -121,14 +122,15 @@ seek :: proc(ctx: ^Context, block_nr: u64) {
 xor_bytes :: proc(ctx: ^Context, dst, src: []byte) {
 	assert(ctx._is_initialized)
 
-	// TODO: Enforcing that dst and src alias exactly or not at all
-	// is a good idea, though odd aliasing should be extremely uncommon.
-
 	src, dst := src, dst
 	if dst_len := len(dst); dst_len < len(src) {
 		src = src[:dst_len]
 	}
 
+	if bytes.alias_inexactly(dst, src) {
+		panic("crypto/chacha20: dst and src alias inexactly")
+	}
+
 	for remaining := len(src); remaining > 0; {
 		// Process multiple blocks at once
 		if ctx._off == _BLOCK_SIZE {

core/crypto/crypto.odin

@@ -4,6 +4,7 @@ helper routines.
 */
 package crypto
 
+import "base:runtime"
 import "core:mem"
 
 // compare_constant_time returns 1 iff a and b are equal, 0 otherwise.
@@ -49,6 +50,9 @@ compare_byte_ptrs_constant_time :: proc "contextless" (a, b: ^byte, n: int) -> i
 // the system entropy source.  This routine will block if the system entropy
 // source is not ready yet.  All system entropy source failures are treated
 // as catastrophic, resulting in a panic.
+//
+// Support for the system entropy source can be checked with the
+// `HAS_RAND_BYTES` boolean constant.
 rand_bytes :: proc (dst: []byte) {
 	// zero-fill the buffer first
 	mem.zero_explicit(raw_data(dst), len(dst))
@@ -56,8 +60,27 @@ rand_bytes :: proc (dst: []byte) {
 	_rand_bytes(dst)
 }
 
-// has_rand_bytes returns true iff the target has support for accessing the
-// system entropty source.
-has_rand_bytes :: proc () -> bool {
-	return _has_rand_bytes()
+// random_generator returns a `runtime.Random_Generator` backed by the
+// system entropy source.
+//
+// Support for the system entropy source can be checked with the
+// `HAS_RAND_BYTES` boolean constant.
+random_generator :: proc() -> runtime.Random_Generator {
+	return {
+		procedure = proc(data: rawptr, mode: runtime.Random_Generator_Mode, p: []byte) {
+			switch mode {
+			case .Read:
+				rand_bytes(p)
+			case .Reset:
+				// do nothing
+			case .Query_Info:
+				if len(p) != size_of(runtime.Random_Generator_Query_Info) {
+					return
+				}
+				info := (^runtime.Random_Generator_Query_Info)(raw_data(p))
+				info^ += {.Uniform, .Cryptographic, .External_Entropy}
+			}
+		},
+		data = nil,
+	}
 }

core/crypto/kmac/kmac.odin

@@ -61,7 +61,7 @@ init_256 :: proc(ctx: ^Context, key, domain_sep: []byte) {
 update :: proc(ctx: ^Context, data: []byte) {
 	assert(ctx.is_initialized)
 
-	shake.write(transmute(^shake.Context)(ctx), data)
+	shake.write((^shake.Context)(ctx), data)
 }
 
 // final finalizes the Context, writes the tag to dst, and calls reset
@@ -75,7 +75,7 @@ final :: proc(ctx: ^Context, dst: []byte) {
 		panic("crypto/kmac: invalid KMAC tag_size, too short")
 	}
 
-	_sha3.final_cshake(transmute(^_sha3.Context)(ctx), dst)
+	_sha3.final_cshake((^_sha3.Context)(ctx), dst)
 }
 
 // clone clones the Context other into ctx.
@@ -84,7 +84,7 @@ clone :: proc(ctx, other: ^Context) {
 		return
 	}
 
-	shake.clone(transmute(^shake.Context)(ctx), transmute(^shake.Context)(other))
+	shake.clone((^shake.Context)(ctx), (^shake.Context)(other))
 }
 
 // reset sanitizes the Context.  The Context must be re-initialized to
@@ -94,7 +94,7 @@ reset :: proc(ctx: ^Context) {
 		return
 	}
 
-	shake.reset(transmute(^shake.Context)(ctx))
+	shake.reset((^shake.Context)(ctx))
 }
 
 @(private)
@@ -107,7 +107,7 @@ _init_kmac :: proc(ctx: ^Context, key, s: []byte, sec_strength: int) {
 		panic("crypto/kmac: invalid KMAC key, too short")
 	}
 
-	ctx_ := transmute(^_sha3.Context)(ctx)
+	ctx_ := (^_sha3.Context)(ctx)
 	_sha3.init_cshake(ctx_, N_KMAC, s, sec_strength)
 	_sha3.bytepad(ctx_, [][]byte{key}, _sha3.rate_cshake(sec_strength))
 }

core/crypto/legacy/keccak/keccak.odin

@@ -66,12 +66,12 @@ init_512 :: proc(ctx: ^Context) {
 @(private)
 _init :: proc(ctx: ^Context) {
 	ctx.dsbyte = _sha3.DS_KECCAK
-	_sha3.init(transmute(^_sha3.Context)(ctx))
+	_sha3.init((^_sha3.Context)(ctx))
 }
 
 // update adds more data to the Context.
 update :: proc(ctx: ^Context, data: []byte) {
-	_sha3.update(transmute(^_sha3.Context)(ctx), data)
+	_sha3.update((^_sha3.Context)(ctx), data)
 }
 
 // final finalizes the Context, writes the digest to hash, and calls
@@ -80,16 +80,16 @@ update :: proc(ctx: ^Context, data: []byte) {
 // Iff finalize_clone is set, final will work on a copy of the Context,
 // which is useful for for calculating rolling digests.
 final :: proc(ctx: ^Context, hash: []byte, finalize_clone: bool = false) {
-	_sha3.final(transmute(^_sha3.Context)(ctx), hash, finalize_clone)
+	_sha3.final((^_sha3.Context)(ctx), hash, finalize_clone)
 }
 
 // clone clones the Context other into ctx.
 clone :: proc(ctx, other: ^Context) {
-	_sha3.clone(transmute(^_sha3.Context)(ctx), transmute(^_sha3.Context)(other))
+	_sha3.clone((^_sha3.Context)(ctx), (^_sha3.Context)(other))
 }
 
 // reset sanitizes the Context.  The Context must be re-initialized to
 // be used again.
 reset :: proc(ctx: ^Context) {
-	_sha3.reset(transmute(^_sha3.Context)(ctx))
+	_sha3.reset((^_sha3.Context)(ctx))
 }

core/crypto/rand_bsd.odin

@@ -1,16 +1,15 @@
-//+build freebsd, openbsd
+//+build freebsd, openbsd, netbsd
 package crypto
 
 foreign import libc "system:c"
 
+HAS_RAND_BYTES :: true
+
 foreign libc {
 	arc4random_buf :: proc(buf: [^]byte, nbytes: uint) ---
 }
 
+@(private)
 _rand_bytes :: proc(dst: []byte) {
 	arc4random_buf(raw_data(dst), len(dst))
 }
-
-_has_rand_bytes :: proc() -> bool {
-	return true
-}

core/crypto/rand_darwin.odin

@@ -5,14 +5,13 @@ import "core:fmt"
 import CF "core:sys/darwin/CoreFoundation"
 import Sec "core:sys/darwin/Security"
 
+HAS_RAND_BYTES :: true
+
+@(private)
 _rand_bytes :: proc(dst: []byte) {
 	err := Sec.RandomCopyBytes(count=len(dst), bytes=raw_data(dst))
 	if err != .Success {
-        msg := CF.StringCopyToOdinString(Sec.CopyErrorMessageString(err))
-        panic(fmt.tprintf("crypto/rand_bytes: SecRandomCopyBytes returned non-zero result: %v %s", err, msg))
+		msg := CF.StringCopyToOdinString(Sec.CopyErrorMessageString(err))
+		fmt.panicf("crypto/rand_bytes: SecRandomCopyBytes returned non-zero result: %v %s", err, msg)
 	}
 }
-
-_has_rand_bytes :: proc() -> bool {
-	return true
-}

core/crypto/rand_generic.odin

@@ -2,14 +2,14 @@
 //+build !windows
 //+build !openbsd
 //+build !freebsd
+//+build !netbsd
 //+build !darwin
 //+build !js
 package crypto
 
+HAS_RAND_BYTES :: false
+
+@(private)
 _rand_bytes :: proc(dst: []byte) {
 	unimplemented("crypto: rand_bytes not supported on this OS")
 }
-
-_has_rand_bytes :: proc() -> bool {
-	return false
-}

core/crypto/rand_js.odin

@@ -6,8 +6,12 @@ foreign odin_env {
 	env_rand_bytes :: proc "contextless" (buf: []byte) ---
 }
 
+HAS_RAND_BYTES :: true
+
+@(private)
 _MAX_PER_CALL_BYTES :: 65536 // 64kiB
 
+@(private)
 _rand_bytes :: proc(dst: []byte) {
 	dst := dst
 
@@ -18,7 +22,3 @@ _rand_bytes :: proc(dst: []byte) {
 		dst = dst[to_read:]
 	}
 }
-
-_has_rand_bytes :: proc() -> bool {
-	return true
-}

core/crypto/rand_linux.odin

@@ -4,8 +4,12 @@ import "core:fmt"
 
 import "core:sys/linux"
 
+HAS_RAND_BYTES :: true
+
+@(private)
 _MAX_PER_CALL_BYTES :: 33554431 // 2^25 - 1
 
+@(private)
 _rand_bytes :: proc (dst: []byte) {
 	dst := dst
 	l := len(dst)
@@ -28,13 +32,9 @@ _rand_bytes :: proc (dst: []byte) {
 			// All other failures are things that should NEVER happen
 			// unless the kernel interface changes (ie: the Linux
 			// developers break userland).
-			panic(fmt.tprintf("crypto: getrandom failed: %v", errno))
+			fmt.panicf("crypto: getrandom failed: %v", errno)
 		}
 		l -= n_read
 		dst = dst[n_read:]
 	}
 }
-
-_has_rand_bytes :: proc() -> bool {
-	return true
-}

core/crypto/rand_windows.odin

@@ -4,24 +4,23 @@ import win32 "core:sys/windows"
 import "core:os"
 import "core:fmt"
 
+HAS_RAND_BYTES :: true
+
+@(private)
 _rand_bytes :: proc(dst: []byte) {
 	ret := (os.Errno)(win32.BCryptGenRandom(nil, raw_data(dst), u32(len(dst)), win32.BCRYPT_USE_SYSTEM_PREFERRED_RNG))
 	if ret != os.ERROR_NONE {
 		switch ret {
-			case os.ERROR_INVALID_HANDLE:
-				// The handle to the first parameter is invalid.
-				// This should not happen here, since we explicitly pass nil to it
-				panic("crypto: BCryptGenRandom Invalid handle for hAlgorithm")
-			case os.ERROR_INVALID_PARAMETER:
-				// One of the parameters was invalid
-				panic("crypto: BCryptGenRandom Invalid parameter")
-			case:
-				// Unknown error
-				panic(fmt.tprintf("crypto: BCryptGenRandom failed: %d\n", ret))
+		case os.ERROR_INVALID_HANDLE:
+			// The handle to the first parameter is invalid.
+			// This should not happen here, since we explicitly pass nil to it
+			panic("crypto: BCryptGenRandom Invalid handle for hAlgorithm")
+		case os.ERROR_INVALID_PARAMETER:
+			// One of the parameters was invalid
+			panic("crypto: BCryptGenRandom Invalid parameter")
+		case:
+			// Unknown error
+			fmt.panicf("crypto: BCryptGenRandom failed: %d\n", ret)
 		}
 	}
 }
-
-_has_rand_bytes :: proc() -> bool {
-	return true
-}

core/crypto/ristretto255/ristretto255.odin

@@ -112,7 +112,7 @@ ge_set_bytes :: proc "contextless" (ge: ^Group_Element, b: []byte) -> bool {
 		return false
 	}
 
-	b_ := transmute(^[32]byte)(raw_data(b))
+	b_ := (^[32]byte)(raw_data(b))
 
 	s: field.Tight_Field_Element = ---
 	defer field.fe_clear(&s)
@@ -297,7 +297,7 @@ ge_bytes :: proc(ge: ^Group_Element, dst: []byte) {
 	// 2.  Return the 32-byte little-endian encoding of s.  More
 	// specifically, this is the encoding of the canonical
 	// representation of s as an integer between 0 and p-1, inclusive.
-	dst_ := transmute(^[32]byte)(raw_data(dst))
+	dst_ := (^[32]byte)(raw_data(dst))
 	field.fe_to_bytes(dst_, &tmp)
 
 	field.fe_clear_vec([]^field.Tight_Field_Element{&u1, &u2, &tmp, &z_inv, &ix0, &iy0, &x, &y})
@@ -417,7 +417,7 @@ ge_is_identity :: proc(ge: ^Group_Element) -> int {
 
 @(private)
 ge_map :: proc "contextless" (ge: ^Group_Element, b: []byte) {
-	b_ := transmute(^[32]byte)(raw_data(b))
+	b_ := (^[32]byte)(raw_data(b))
 
 	// The MAP function is defined on 32-byte strings as:
 	//

core/crypto/ristretto255/ristretto255_scalar.odin

@@ -46,7 +46,7 @@ sc_set_bytes_wide :: proc(sc: ^Scalar, b: []byte) {
 		panic("crypto/ristretto255: invalid wide input size")
 	}
 
-	b_ := transmute(^[WIDE_SCALAR_SIZE]byte)(raw_data(b))
+	b_ := (^[WIDE_SCALAR_SIZE]byte)(raw_data(b))
 	grp.sc_set_bytes_wide(sc, b_)
 }
 

core/crypto/sha3/sha3.odin

@@ -68,12 +68,12 @@ init_512 :: proc(ctx: ^Context) {
 @(private)
 _init :: proc(ctx: ^Context) {
 	ctx.dsbyte = _sha3.DS_SHA3
-	_sha3.init(transmute(^_sha3.Context)(ctx))
+	_sha3.init((^_sha3.Context)(ctx))
 }
 
 // update adds more data to the Context.
 update :: proc(ctx: ^Context, data: []byte) {
-	_sha3.update(transmute(^_sha3.Context)(ctx), data)
+	_sha3.update((^_sha3.Context)(ctx), data)
 }
 
 // final finalizes the Context, writes the digest to hash, and calls
@@ -82,16 +82,16 @@ update :: proc(ctx: ^Context, data: []byte) {
 // Iff finalize_clone is set, final will work on a copy of the Context,
 // which is useful for for calculating rolling digests.
 final :: proc(ctx: ^Context, hash: []byte, finalize_clone: bool = false) {
-	_sha3.final(transmute(^_sha3.Context)(ctx), hash, finalize_clone)
+	_sha3.final((^_sha3.Context)(ctx), hash, finalize_clone)
 }
 
 // clone clones the Context other into ctx.
 clone :: proc(ctx, other: ^Context) {
-	_sha3.clone(transmute(^_sha3.Context)(ctx), transmute(^_sha3.Context)(other))
+	_sha3.clone((^_sha3.Context)(ctx), (^_sha3.Context)(other))
 }
 
 // reset sanitizes the Context.  The Context must be re-initialized to
 // be used again.
 reset :: proc(ctx: ^Context) {
-	_sha3.reset(transmute(^_sha3.Context)(ctx))
+	_sha3.reset((^_sha3.Context)(ctx))
 }

core/crypto/shake/shake.odin

@@ -24,35 +24,35 @@ Context :: distinct _sha3.Context
 
 // init_128 initializes a Context for SHAKE128.
 init_128 :: proc(ctx: ^Context) {
-	_sha3.init_cshake(transmute(^_sha3.Context)(ctx), nil, nil, 128)
+	_sha3.init_cshake((^_sha3.Context)(ctx), nil, nil, 128)
 }
 
 // init_256 initializes a Context for SHAKE256.
 init_256 :: proc(ctx: ^Context) {
-	_sha3.init_cshake(transmute(^_sha3.Context)(ctx), nil, nil, 256)
+	_sha3.init_cshake((^_sha3.Context)(ctx), nil, nil, 256)
 }
 
 // init_cshake_128 initializes a Context for cSHAKE128.
 init_cshake_128 :: proc(ctx: ^Context, domain_sep: []byte) {
-	_sha3.init_cshake(transmute(^_sha3.Context)(ctx), nil, domain_sep, 128)
+	_sha3.init_cshake((^_sha3.Context)(ctx), nil, domain_sep, 128)
 }
 
 // init_cshake_256 initializes a Context for cSHAKE256.
 init_cshake_256 :: proc(ctx: ^Context, domain_sep: []byte) {
-	_sha3.init_cshake(transmute(^_sha3.Context)(ctx), nil, domain_sep, 256)
+	_sha3.init_cshake((^_sha3.Context)(ctx), nil, domain_sep, 256)
 }
 
 // write writes more data into the SHAKE instance.  This MUST not be called
 // after any reads have been done, and attempts to do so will panic.
 write :: proc(ctx: ^Context, data: []byte) {
-	_sha3.update(transmute(^_sha3.Context)(ctx), data)
+	_sha3.update((^_sha3.Context)(ctx), data)
 }
 
 // read reads output from the SHAKE instance.  There is no practical upper
 // limit to the amount of data that can be read from SHAKE.  After read has
 // been called one or more times, further calls to write will panic.
 read :: proc(ctx: ^Context, dst: []byte) {
-	ctx_ := transmute(^_sha3.Context)(ctx)
+	ctx_ := (^_sha3.Context)(ctx)
 	if !ctx.is_finalized {
 		_sha3.shake_xof(ctx_)
 	}
@@ -62,11 +62,11 @@ read :: proc(ctx: ^Context, dst: []byte) {
 
 // clone clones the Context other into ctx.
 clone :: proc(ctx, other: ^Context) {
-	_sha3.clone(transmute(^_sha3.Context)(ctx), transmute(^_sha3.Context)(other))
+	_sha3.clone((^_sha3.Context)(ctx), (^_sha3.Context)(other))
 }
 
 // reset sanitizes the Context.  The Context must be re-initialized to
 // be used again.
 reset :: proc(ctx: ^Context) {
-	_sha3.reset(transmute(^_sha3.Context)(ctx))
+	_sha3.reset((^_sha3.Context)(ctx))
 }

core/crypto/tuplehash/tuplehash.odin

@@ -13,19 +13,19 @@ Context :: distinct _sha3.Context
 
 // init_128 initializes a Context for TupleHash128 or TupleHashXOF128.
 init_128 :: proc(ctx: ^Context, domain_sep: []byte) {
-	_sha3.init_cshake(transmute(^_sha3.Context)(ctx), N_TUPLEHASH, domain_sep, 128)
+	_sha3.init_cshake((^_sha3.Context)(ctx), N_TUPLEHASH, domain_sep, 128)
 }
 
 // init_256 initializes a Context for TupleHash256 or TupleHashXOF256.
 init_256 :: proc(ctx: ^Context, domain_sep: []byte) {
-	_sha3.init_cshake(transmute(^_sha3.Context)(ctx), N_TUPLEHASH, domain_sep, 256)
+	_sha3.init_cshake((^_sha3.Context)(ctx), N_TUPLEHASH, domain_sep, 256)
 }
 
 // write_element writes a tuple element into the TupleHash or TupleHashXOF
 // instance.  This MUST not be called after any reads have been done, and
 // any attempts to do so will panic.
 write_element :: proc(ctx: ^Context, data: []byte) {
-	_, _ = _sha3.encode_string(transmute(^_sha3.Context)(ctx), data)
+	_, _ = _sha3.encode_string((^_sha3.Context)(ctx), data)
 }
 
 // final finalizes the Context, writes the digest to hash, and calls
@@ -34,7 +34,7 @@ write_element :: proc(ctx: ^Context, data: []byte) {
 // Iff finalize_clone is set, final will work on a copy of the Context,
 // which is useful for for calculating rolling digests.
 final :: proc(ctx: ^Context, hash: []byte, finalize_clone: bool = false) {
-	_sha3.final_cshake(transmute(^_sha3.Context)(ctx), hash, finalize_clone)
+	_sha3.final_cshake((^_sha3.Context)(ctx), hash, finalize_clone)
 }
 
 // read reads output from the TupleHashXOF instance.  There is no practical
@@ -42,7 +42,7 @@ final :: proc(ctx: ^Context, hash: []byte, finalize_clone: bool = false) {
 // After read has been called one or more times, further calls to
 // write_element will panic.
 read :: proc(ctx: ^Context, dst: []byte) {
-	ctx_ := transmute(^_sha3.Context)(ctx)
+	ctx_ := (^_sha3.Context)(ctx)
 	if !ctx.is_finalized {
 		_sha3.encode_byte_len(ctx_, 0, false) // right_encode
 		_sha3.shake_xof(ctx_)
@@ -53,13 +53,13 @@ read :: proc(ctx: ^Context, dst: []byte) {
 
 // clone clones the Context other into ctx.
 clone :: proc(ctx, other: ^Context) {
-	_sha3.clone(transmute(^_sha3.Context)(ctx), transmute(^_sha3.Context)(other))
+	_sha3.clone((^_sha3.Context)(ctx), (^_sha3.Context)(other))
 }
 
 // reset sanitizes the Context.  The Context must be re-initialized to
 // be used again.
 reset :: proc(ctx: ^Context) {
-	_sha3.reset(transmute(^_sha3.Context)(ctx))
+	_sha3.reset((^_sha3.Context)(ctx))
 }
 
 @(private)

core/debug/trace/doc.odin

@@ -0,0 +1,51 @@
+/*
+A debug stack trace library. Only works when debug symbols are enabled `-debug`.
+
+Example:
+	import "base:runtime"
+	import "core:debug/trace"
+
+	import "core:fmt"
+
+	global_trace_ctx: trace.Context
+
+	debug_trace_assertion_failure_proc :: proc(prefix, message: string, loc := #caller_location) -> ! {
+		runtime.print_caller_location(loc)
+		runtime.print_string(" ")
+		runtime.print_string(prefix)
+		if len(message) > 0 {
+			runtime.print_string(": ")
+			runtime.print_string(message)
+		}
+		runtime.print_byte('\n')
+
+		ctx := &trace_ctx
+		if !trace.in_resolve(ctx) {
+			buf: [64]trace.Frame
+			runtime.print_string("Debug Trace:\n")
+			frames := trace.frames(ctx, 1, buf[:])
+			for f, i in frames {
+				fl := trace.resolve(ctx, f, context.temp_allocator)
+				if fl.loc.file_path == "" && fl.loc.line == 0 {
+					continue
+				}
+				runtime.print_caller_location(fl.loc)
+				runtime.print_string(" - frame ")
+				runtime.print_int(i)
+				runtime.print_byte('\n')
+			}
+		}
+		runtime.trap()
+	}
+
+	main :: proc() {
+		trace.init(&global_trace_ctx)
+		defer trace.destroy(&global_trace_ctx)
+
+		context.assertion_failure_proc = debug_trace_assertion_failure_proc
+
+		...
+	}
+
+*/
+package debug_trace

core/debug/trace/trace.odin

@@ -0,0 +1,47 @@
+package debug_trace
+
+import "base:intrinsics"
+import "base:runtime"
+
+Frame :: distinct uintptr
+
+Frame_Location :: struct {
+	using loc: runtime.Source_Code_Location,
+	allocator: runtime.Allocator,
+}
+
+delete_frame_location :: proc(fl: Frame_Location) -> runtime.Allocator_Error {
+	allocator := fl.allocator
+	delete(fl.loc.procedure, allocator) or_return
+	delete(fl.loc.file_path, allocator) or_return
+	return nil
+}
+
+Context :: struct {
+	in_resolve: bool, // atomic
+	impl: _Context,
+}
+
+init :: proc(ctx: ^Context) -> bool {
+	return _init(ctx)
+}
+
+destroy :: proc(ctx: ^Context) -> bool {
+	return _destroy(ctx)
+}
+
+@(require_results)
+frames :: proc(ctx: ^Context, skip: uint, frames_buffer: []Frame) -> []Frame {
+	return _frames(ctx, skip, frames_buffer)
+}
+
+@(require_results)
+resolve :: proc(ctx: ^Context, frame: Frame, allocator: runtime.Allocator) -> (result: Frame_Location) {
+	return _resolve(ctx, frame, allocator)
+}
+
+
+@(require_results)
+in_resolve :: proc "contextless" (ctx: ^Context) -> bool {
+	return intrinsics.atomic_load(&ctx.in_resolve)
+}

core/debug/trace/trace_cpp.odin

@@ -0,0 +1,195 @@
+//+private file
+//+build linux, darwin
+package debug_trace
+
+import "base:intrinsics"
+import "base:runtime"
+import "core:strings"
+import "core:fmt"
+import "core:c"
+
+// NOTE: Relies on C++23 which adds <stacktrace> and becomes ABI and that can be used
+foreign import stdcpplibbacktrace "system:stdc++_libbacktrace"
+
+foreign import libdl "system:dl"
+
+backtrace_state :: struct {}
+backtrace_error_callback   :: proc "c" (data: rawptr, msg: cstring, errnum: c.int)
+backtrace_simple_callback  :: proc "c" (data: rawptr, pc: uintptr) -> c.int
+backtrace_full_callback    :: proc "c" (data: rawptr, pc: uintptr, filename: cstring, lineno: c.int, function: cstring) -> c.int
+backtrace_syminfo_callback :: proc "c" (data: rawptr, pc: uintptr, symname: cstring, symval: uintptr, symsize: uintptr)
+
+@(default_calling_convention="c", link_prefix="__glibcxx_")
+foreign stdcpplibbacktrace {
+	backtrace_create_state :: proc(
+		filename:       cstring,
+		threaded:       c.int,
+		error_callback: backtrace_error_callback,
+		data:           rawptr,
+	) -> ^backtrace_state ---
+	backtrace_simple  :: proc(
+		state:          ^backtrace_state,
+		skip:           c.int,
+		callback:       backtrace_simple_callback,
+		error_callback: backtrace_error_callback,
+		data:           rawptr,
+	) -> c.int ---
+	backtrace_pcinfo  :: proc(
+		state:          ^backtrace_state,
+		pc:             uintptr,
+		callback:       backtrace_full_callback,
+		error_callback: backtrace_error_callback,
+		data:           rawptr,
+	) -> c.int ---
+	backtrace_syminfo :: proc(
+		state:          ^backtrace_state,
+		addr:           uintptr,
+		callback:       backtrace_syminfo_callback,
+		error_callback: backtrace_error_callback,
+		data:           rawptr,
+	) -> c.int ---
+
+	// NOTE(bill): this is technically an internal procedure, but it is exposed
+	backtrace_free    :: proc(
+		state: ^backtrace_state,
+		p:              rawptr,
+		size:           c.size_t,                 // unused
+		error_callback: backtrace_error_callback, // unused
+		data:           rawptr,                   // unused
+		) ---
+}
+
+Dl_info :: struct {
+	dli_fname: cstring,
+	dli_fbase: rawptr,
+	dli_sname: cstring,
+	dli_saddr: rawptr,
+}
+
+@(default_calling_convention="c")
+foreign libdl {
+	dladdr :: proc(addr: rawptr, info: ^Dl_info) -> c.int ---
+}
+
+@(private="package")
+_Context :: struct {
+	state: ^backtrace_state,
+}
+
+@(private="package")
+_init :: proc(ctx: ^Context) -> (ok: bool) {
+	defer if !ok { destroy(ctx) }
+
+	ctx.impl.state = backtrace_create_state("odin-debug-trace", 1, nil, ctx)
+	return ctx.impl.state != nil
+}
+
+@(private="package")
+_destroy :: proc(ctx: ^Context) -> bool {
+	if ctx != nil {
+		backtrace_free(ctx.impl.state, nil, 0, nil, nil)
+	}
+	return true
+}
+
+@(private="package")
+_frames :: proc "contextless" (ctx: ^Context, skip: uint, frames_buffer: []Frame) -> (frames: []Frame) {
+	Backtrace_Context :: struct {
+		ctx:         ^Context,
+		frames:      []Frame,
+		frame_count: int,
+	}
+
+	btc := &Backtrace_Context{
+		ctx = ctx,
+		frames = frames_buffer,
+	}
+	backtrace_simple(
+		ctx.impl.state,
+		c.int(skip + 2),
+		proc "c" (user: rawptr, address: uintptr) -> c.int {
+			btc := (^Backtrace_Context)(user)
+			address := Frame(address)
+			if address == 0 {
+				return 1
+			}
+			if btc.frame_count == len(btc.frames) {
+				return 1
+			}
+			btc.frames[btc.frame_count] = address
+			btc.frame_count += 1
+			return 0
+		},
+		nil,
+		btc,
+	)
+
+	if btc.frame_count > 0 {
+		frames = btc.frames[:btc.frame_count]
+	}
+	return
+}
+
+@(private="package")
+_resolve :: proc(ctx: ^Context, frame: Frame, allocator: runtime.Allocator) -> Frame_Location {
+	intrinsics.atomic_store(&ctx.in_resolve, true)
+	defer intrinsics.atomic_store(&ctx.in_resolve, false)
+
+	Backtrace_Context :: struct {
+		rt_ctx:    runtime.Context,
+		allocator: runtime.Allocator,
+		frame:     Frame_Location,
+	}
+
+	btc := &Backtrace_Context{
+		rt_ctx = context,
+		allocator = allocator,
+	}
+	done := backtrace_pcinfo(
+		ctx.impl.state,
+		uintptr(frame),
+		proc "c" (data: rawptr, address: uintptr, file: cstring, line: c.int, symbol: cstring) -> c.int {
+			btc := (^Backtrace_Context)(data)
+			context = btc.rt_ctx
+
+			frame := &btc.frame
+
+			if file != nil {
+				frame.file_path = strings.clone_from_cstring(file, btc.allocator)
+			} else if info: Dl_info; dladdr(rawptr(address), &info) != 0 && info.dli_fname != "" {
+				frame.file_path = strings.clone_from_cstring(info.dli_fname, btc.allocator)
+			}
+			if symbol != nil {
+				frame.procedure = strings.clone_from_cstring(symbol, btc.allocator)
+			} else if info: Dl_info; dladdr(rawptr(address), &info) != 0 && info.dli_sname != "" {
+				frame.procedure = strings.clone_from_cstring(info.dli_sname, btc.allocator)
+			} else {
+				frame.procedure = fmt.aprintf("(procedure: 0x%x)", allocator=btc.allocator)
+			}
+			frame.line = i32(line)
+			return 0
+		},
+		nil,
+		btc,
+	)
+	if done != 0 {
+		return btc.frame
+	}
+
+	// NOTE(bill): pcinfo cannot resolve, but it might be possible to get the procedure name at least
+	backtrace_syminfo(
+		ctx.impl.state,
+		uintptr(frame),
+		proc "c" (data: rawptr, address: uintptr, symbol: cstring, _ignore0, _ignore1: uintptr) {
+			if symbol != nil {
+				btc := (^Backtrace_Context)(data)
+				context = btc.rt_ctx
+				btc.frame.procedure = strings.clone_from_cstring(symbol, btc.allocator)
+			}
+		},
+		nil,
+		btc,
+	)
+
+	return btc.frame
+}

core/debug/trace/trace_nil.odin

@@ -0,0 +1,20 @@
+//+build !windows !linux !darwin
+package debug_trace
+
+import "base:runtime"
+
+_Context :: struct {
+}
+
+_init :: proc(ctx: ^Context) -> (ok: bool) {
+	return true
+}
+_destroy :: proc(ctx: ^Context) -> bool {
+	return true
+}
+_frames :: proc(ctx: ^Context, skip: uint, frames_buffer: []Frame) -> []Frame {
+	return nil
+}
+_resolve :: proc(ctx: ^Context, frame: Frame, allocator: runtime.Allocator) -> (result: Frame_Location) {
+	return
+}

core/debug/trace/trace_windows.odin

@@ -0,0 +1,68 @@
+//+private
+//+build windows
+package debug_trace
+
+import "base:intrinsics"
+import "base:runtime"
+
+import win32 "core:sys/windows"
+import "core:fmt"
+
+_Context :: struct {
+	hProcess: win32.HANDLE,
+	lock:     win32.SRWLOCK,
+}
+
+_init :: proc "contextless" (ctx: ^Context) -> (ok: bool) {
+	defer if !ok { _destroy(ctx) }
+	ctx.impl.hProcess = win32.GetCurrentProcess()
+	win32.SymInitialize(ctx.impl.hProcess, nil, true) or_return
+	win32.SymSetOptions(win32.SYMOPT_LOAD_LINES)
+	return true
+}
+
+_destroy :: proc "contextless" (ctx: ^Context) -> bool {
+	if ctx != nil {
+		win32.SymCleanup(ctx.impl.hProcess)
+	}
+	return true
+}
+
+_frames :: proc "contextless" (ctx: ^Context, skip: uint, frames_buffer: []Frame) -> []Frame {
+	frame_count := win32.RtlCaptureStackBackTrace(u32(skip) + 2, u32(len(frames_buffer)), ([^]rawptr)(&frames_buffer[0]), nil)
+	for i in 0..<frame_count {
+		// NOTE: Return address is one after the call instruction so subtract a byte to
+		// end up back inside the call instruction which is needed for SymFromAddr.
+		frames_buffer[i] -= 1
+	}
+	return frames_buffer[:frame_count]
+}
+
+
+_resolve :: proc(ctx: ^Context, frame: Frame, allocator: runtime.Allocator) -> (fl: Frame_Location) {
+	intrinsics.atomic_store(&ctx.in_resolve, true)
+	defer intrinsics.atomic_store(&ctx.in_resolve, false)
+
+	// NOTE(bill): Dbghelp is not thread-safe
+	win32.AcquireSRWLockExclusive(&ctx.impl.lock)
+	defer win32.ReleaseSRWLockExclusive(&ctx.impl.lock)
+
+	data: [size_of(win32.SYMBOL_INFOW) + size_of([256]win32.WCHAR)]byte
+	symbol := (^win32.SYMBOL_INFOW)(&data[0])
+	symbol.SizeOfStruct = size_of(symbol)
+	symbol.MaxNameLen = 255
+	if win32.SymFromAddrW(ctx.impl.hProcess, win32.DWORD64(frame), &{}, symbol) {
+		fl.procedure, _ = win32.wstring_to_utf8(&symbol.Name[0], -1, allocator)
+	} else {
+		fl.procedure = fmt.aprintf("(procedure: 0x%x)", frame, allocator=allocator)
+	}
+
+	line: win32.IMAGEHLP_LINE64
+	line.SizeOfStruct = size_of(line)
+	if win32.SymGetLineFromAddrW64(ctx.impl.hProcess, win32.DWORD64(frame), &{}, &line) {
+		fl.file_path, _ = win32.wstring_to_utf8(line.FileName, -1, allocator)
+		fl.line = i32(line.LineNumber)
+	}
+
+	return
+}

core/dynlib/lib.odin

@@ -16,15 +16,12 @@ Library :: distinct rawptr
 Loads a dynamic library from the filesystem. The paramater `global_symbols` makes the symbols in the loaded
 library available to resolve references in subsequently loaded libraries.
 
-The paramater `global_symbols` is only used for the platforms `linux`, `darwin`, `freebsd` and `openbsd`.
+The parameter `global_symbols` is only used for the platforms `linux`, `darwin`, `freebsd` and `openbsd`.
 On `windows` this paramater is ignored.
 
 The underlying behaviour is platform specific.
 On `linux`, `darwin`, `freebsd` and `openbsd` refer to `dlopen`.
-On `windows` refer to `LoadLibraryW`.
-
-**Implicit Allocators**
-`context.temp_allocator`
+On `windows` refer to `LoadLibraryW`. Also temporarily needs an allocator to convert a string.
 
 Example:
 	import "core:dynlib"
@@ -79,10 +76,7 @@ Loads the address of a procedure/variable from a dynamic library.
 
 The underlying behaviour is platform specific.
 On `linux`, `darwin`, `freebsd` and `openbsd` refer to `dlsym`.
-On `windows` refer to `GetProcAddress`.
-
-**Implicit Allocators**
-`context.temp_allocator`
+On `windows` refer to `GetProcAddress`. Also temporarily needs an allocator to convert a string.
 
 Example:
 	import "core:dynlib"
@@ -177,9 +171,7 @@ initialize_symbols :: proc(
 	return count, count > 0
 }
 
-/*
-Returns an error message for the last failed procedure call.
-*/
+// Returns an error message for the last failed procedure call.
 last_error :: proc() -> string {
 	return _last_error()
-}
+}

core/dynlib/lib_js.odin

@@ -16,4 +16,4 @@ _symbol_address :: proc(library: Library, symbol: string) -> (ptr: rawptr, found
 
 _last_error :: proc() -> string {
 	return ""
-}
+}

core/dynlib/lib_unix.odin

@@ -1,4 +1,4 @@
-//+build linux, darwin, freebsd, openbsd
+//+build linux, darwin, freebsd, openbsd, netbsd
 //+private
 package dynlib
 
@@ -26,4 +26,4 @@ _symbol_address :: proc(library: Library, symbol: string) -> (ptr: rawptr, found
 _last_error :: proc() -> string {
 	err := os.dlerror()
 	return "unknown" if err == "" else err
-}
+}

core/dynlib/lib_windows.odin

@@ -4,14 +4,12 @@ package dynlib
 
 import win32 "core:sys/windows"
 import "core:strings"
-import "base:runtime"
 import "core:reflect"
 
-_load_library :: proc(path: string, global_symbols := false) -> (Library, bool) {
+_load_library :: proc(path: string, global_symbols := false, allocator := context.temp_allocator) -> (Library, bool) {
 	// NOTE(bill): 'global_symbols' is here only for consistency with POSIX which has RTLD_GLOBAL
-
-	runtime.DEFAULT_TEMP_ALLOCATOR_TEMP_GUARD()
-	wide_path := win32.utf8_to_wstring(path, context.temp_allocator)
+	wide_path := win32.utf8_to_wstring(path, allocator)
+	defer free(wide_path, allocator)
 	handle := cast(Library)win32.LoadLibraryW(wide_path)
 	return handle, handle != nil
 }
@@ -21,9 +19,9 @@ _unload_library :: proc(library: Library) -> bool {
 	return bool(ok)
 }
 
-_symbol_address :: proc(library: Library, symbol: string) -> (ptr: rawptr, found: bool) {
-	runtime.DEFAULT_TEMP_ALLOCATOR_TEMP_GUARD()
-	c_str := strings.clone_to_cstring(symbol, context.temp_allocator)
+_symbol_address :: proc(library: Library, symbol: string, allocator := context.temp_allocator) -> (ptr: rawptr, found: bool) {
+	c_str := strings.clone_to_cstring(symbol, allocator)
+	defer delete(c_str, allocator)
 	ptr = win32.GetProcAddress(cast(win32.HMODULE)library, c_str)
 	found = ptr != nil
 	return
@@ -33,4 +31,4 @@ _last_error :: proc() -> string {
 	err := win32.System_Error(win32.GetLastError())
 	err_msg := reflect.enum_string(err)
 	return "unknown" if err_msg == "" else err_msg
-}
+}

core/encoding/ansi/ansi.odin

@@ -0,0 +1,137 @@
+package ansi
+
+BEL     :: "\a" // Bell
+BS      :: "\b" // Backspace
+ESC     :: "\e" // Escape
+
+// Fe Escape sequences
+
+CSI     :: ESC + "["  // Control Sequence Introducer
+OSC     :: ESC + "]"  // Operating System Command
+ST      :: ESC + "\\" // String Terminator
+
+// CSI sequences
+
+CUU     :: "A"  // Cursor Up
+CUD     :: "B"  // Cursor Down
+CUF     :: "C"  // Cursor Forward
+CUB     :: "D"  // Cursor Back
+CNL     :: "E"  // Cursor Next Line
+CPL     :: "F"  // Cursor Previous Line
+CHA     :: "G"  // Cursor Horizontal Absolute
+CUP     :: "H"  // Cursor Position
+ED      :: "J"  // Erase in Display
+EL      :: "K"  // Erase in Line
+SU      :: "S"  // Scroll Up
+SD      :: "T"  // Scroll Down
+HVP     :: "f"  // Horizontal Vertical Position
+SGR     :: "m"  // Select Graphic Rendition
+AUX_ON  :: "5i" // AUX Port On
+AUX_OFF :: "4i" // AUX Port Off
+DSR     :: "6n" // Device Status Report
+
+// CSI: private sequences
+
+SCP          :: "s"    // Save Current Cursor Position
+RCP          :: "u"    // Restore Saved Cursor Position
+DECAWM_ON    :: "?7h"  // Auto Wrap Mode (Enabled)
+DECAWM_OFF   :: "?7l"  // Auto Wrap Mode (Disabled)
+DECTCEM_SHOW :: "?25h" // Text Cursor Enable Mode (Visible)
+DECTCEM_HIDE :: "?25l" // Text Cursor Enable Mode (Invisible)
+
+// SGR sequences
+
+RESET                   :: "0"
+BOLD                    :: "1"
+FAINT                   :: "2"
+ITALIC                  :: "3" // Not widely supported.
+UNDERLINE               :: "4"
+BLINK_SLOW              :: "5"
+BLINK_RAPID             :: "6" // Not widely supported.
+INVERT                  :: "7" // Also known as reverse video.
+HIDE                    :: "8" // Not widely supported.
+STRIKE                  :: "9"
+FONT_PRIMARY            :: "10"
+FONT_ALT1               :: "11"
+FONT_ALT2               :: "12"
+FONT_ALT3               :: "13"
+FONT_ALT4               :: "14"
+FONT_ALT5               :: "15"
+FONT_ALT6               :: "16"
+FONT_ALT7               :: "17"
+FONT_ALT8               :: "18"
+FONT_ALT9               :: "19"
+FONT_FRAKTUR            :: "20" // Rarely supported.
+UNDERLINE_DOUBLE        :: "21" // May be interpreted as "disable bold."
+NO_BOLD_FAINT           :: "22"
+NO_ITALIC_BLACKLETTER   :: "23"
+NO_UNDERLINE            :: "24"
+NO_BLINK                :: "25"
+PROPORTIONAL_SPACING    :: "26"
+NO_REVERSE              :: "27"
+NO_HIDE                 :: "28"
+NO_STRIKE               :: "29"
+
+FG_BLACK                :: "30"
+FG_RED                  :: "31"
+FG_GREEN                :: "32"
+FG_YELLOW               :: "33"
+FG_BLUE                 :: "34"
+FG_MAGENTA              :: "35"
+FG_CYAN                 :: "36"
+FG_WHITE                :: "37"
+FG_COLOR                :: "38"
+FG_COLOR_8_BIT          :: "38;5" // Followed by ";n" where n is in 0..=255
+FG_COLOR_24_BIT         :: "38;2" // Followed by ";r;g;b" where r,g,b are in 0..=255
+FG_DEFAULT              :: "39"
+
+BG_BLACK                :: "40"
+BG_RED                  :: "41"
+BG_GREEN                :: "42"
+BG_YELLOW               :: "43"
+BG_BLUE                 :: "44"
+BG_MAGENTA              :: "45"
+BG_CYAN                 :: "46"
+BG_WHITE                :: "47"
+BG_COLOR                :: "48"
+BG_COLOR_8_BIT          :: "48;5" // Followed by ";n" where n is in 0..=255
+BG_COLOR_24_BIT         :: "48;2" // Followed by ";r;g;b" where r,g,b are in 0..=255
+BG_DEFAULT              :: "49"
+
+NO_PROPORTIONAL_SPACING :: "50"
+FRAMED                  :: "51"
+ENCIRCLED               :: "52"
+OVERLINED               :: "53"
+NO_FRAME_ENCIRCLE       :: "54"
+NO_OVERLINE             :: "55"
+
+// SGR: non-standard bright colors
+
+FG_BRIGHT_BLACK         :: "90" // Also known as grey.
+FG_BRIGHT_RED           :: "91"
+FG_BRIGHT_GREEN         :: "92"
+FG_BRIGHT_YELLOW        :: "93"
+FG_BRIGHT_BLUE          :: "94"
+FG_BRIGHT_MAGENTA       :: "95"
+FG_BRIGHT_CYAN          :: "96"
+FG_BRIGHT_WHITE         :: "97"
+
+BG_BRIGHT_BLACK         :: "100" // Also known as grey.
+BG_BRIGHT_RED           :: "101"
+BG_BRIGHT_GREEN         :: "102"
+BG_BRIGHT_YELLOW        :: "103"
+BG_BRIGHT_BLUE          :: "104"
+BG_BRIGHT_MAGENTA       :: "105"
+BG_BRIGHT_CYAN          :: "106"
+BG_BRIGHT_WHITE         :: "107"
+
+// Fp Escape sequences
+
+DECSC :: ESC + "7" // DEC Save Cursor
+DECRC :: ESC + "8" // DEC Restore Cursor
+
+// OSC sequences
+
+WINDOW_TITLE :: "2"  // Followed by ";<text>" ST.
+HYPERLINK    :: "8"  // Followed by ";[params];<URI>" ST. Closed by OSC HYPERLINK ";;" ST.
+CLIPBOARD    :: "52" // Followed by ";c;<Base64-encoded string>" ST.

core/encoding/ansi/doc.odin

@@ -0,0 +1,20 @@
+/*
+package ansi implements constant references to many widely-supported ANSI
+escape codes, primarily used in terminal emulators for enhanced graphics, such
+as colors, text styling, and animated displays.
+
+For example, you can print out a line of cyan text like this:
+	fmt.println(ansi.CSI + ansi.FG_CYAN + ansi.SGR + "Hellope!" + ansi.CSI + ansi.RESET + ansi.SGR)
+
+Multiple SGR (Select Graphic Rendition) codes can be joined by semicolons:
+	fmt.println(ansi.CSI + ansi.BOLD + ";" + ansi.FG_BLUE + ansi.SGR + "Hellope!" + ansi.CSI + ansi.RESET + ansi.SGR)
+
+If your terminal supports 24-bit true color mode, you can also do this:
+	fmt.println(ansi.CSI + ansi.FG_COLOR_24_BIT + ";0;255;255" + ansi.SGR + "Hellope!" + ansi.CSI + ansi.RESET + ansi.SGR)
+
+For more information, see:
+	1. https://en.wikipedia.org/wiki/ANSI_escape_code
+	2. https://www.vt100.net/docs/vt102-ug/chapter5.html
+	3. https://invisible-island.net/xterm/ctlseqs/ctlseqs.html
+*/
+package ansi

core/encoding/base32/base32.odin

@@ -8,141 +8,141 @@ package encoding_base32
 // truncate it from the encoded output.
 
 ENC_TABLE := [32]byte {
-    'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H',
-    'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P',
-    'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X',
-    'Y', 'Z', '2', '3', '4', '5', '6', '7',
+	'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H',
+	'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P',
+	'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X',
+	'Y', 'Z', '2', '3', '4', '5', '6', '7',
 }
 
 PADDING :: '='
 
 DEC_TABLE := [?]u8 {
-     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
-     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
-     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
-     0,  0, 26, 27, 28, 29, 30, 31,  0,  0,  0,  0,  0,  0,  0,  0,
-     0,  0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14,
-    15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25,  0,  0,  0,  0,  0,
-     0,  0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14,
-    15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25,  0,  0,  0,  0,  0,
-     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
-     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
-     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
-     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
-     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
-     0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
+	 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
+	 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
+	 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
+	 0,  0, 26, 27, 28, 29, 30, 31,  0,  0,  0,  0,  0,  0,  0,  0,
+	 0,  0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14,
+	15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25,  0,  0,  0,  0,  0,
+	 0,  0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14,
+	15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25,  0,  0,  0,  0,  0,
+	 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
+	 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
+	 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
+	 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
+	 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
+	 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,
 }
 
 encode :: proc(data: []byte, ENC_TBL := ENC_TABLE, allocator := context.allocator) -> string {
-    out_length := (len(data) + 4) / 5 * 8
-    out := make([]byte, out_length)
-    _encode(out, data)
-    return string(out)
+	out_length := (len(data) + 4) / 5 * 8
+	out := make([]byte, out_length)
+	_encode(out, data)
+	return string(out)
 }
 
 @private
 _encode :: proc(out, data: []byte, ENC_TBL := ENC_TABLE, allocator := context.allocator) {
-    out := out
-    data := data
+	out := out
+	data := data
 
-    for len(data) > 0 {
-        carry: byte
-        switch len(data) {
-            case:
-                out[7] = ENC_TABLE[data[4] & 0x1f]
-                carry = data[4] >> 5
-                fallthrough
-            case 4:
-                out[6] = ENC_TABLE[carry | (data[3] << 3) & 0x1f]
-                out[5] = ENC_TABLE[(data[3] >> 2) & 0x1f]
-                carry = data[3] >> 7
-                fallthrough
-            case 3:
-                out[4] = ENC_TABLE[carry | (data[2] << 1) & 0x1f]
-                carry = (data[2] >> 4) & 0x1f
-                fallthrough
-            case 2:
-                out[3] = ENC_TABLE[carry | (data[1] << 4) & 0x1f]
-                out[2] = ENC_TABLE[(data[1] >> 1) & 0x1f]
-                carry = (data[1] >> 6) & 0x1f
-                fallthrough
-            case 1:
-                out[1] = ENC_TABLE[carry | (data[0] << 2) & 0x1f]
-                out[0] = ENC_TABLE[data[0] >> 3]
-        }
+	for len(data) > 0 {
+		carry: byte
+		switch len(data) {
+		case:
+			out[7] = ENC_TABLE[data[4] & 0x1f]
+			carry = data[4] >> 5
+			fallthrough
+		case 4:
+			out[6] = ENC_TABLE[carry | (data[3] << 3) & 0x1f]
+			out[5] = ENC_TABLE[(data[3] >> 2) & 0x1f]
+			carry = data[3] >> 7
+			fallthrough
+		case 3:
+			out[4] = ENC_TABLE[carry | (data[2] << 1) & 0x1f]
+			carry = (data[2] >> 4) & 0x1f
+			fallthrough
+		case 2:
+			out[3] = ENC_TABLE[carry | (data[1] << 4) & 0x1f]
+			out[2] = ENC_TABLE[(data[1] >> 1) & 0x1f]
+			carry = (data[1] >> 6) & 0x1f
+			fallthrough
+		case 1:
+			out[1] = ENC_TABLE[carry | (data[0] << 2) & 0x1f]
+			out[0] = ENC_TABLE[data[0] >> 3]
+		}
 
-        if len(data) < 5 {
-            out[7] = byte(PADDING)
-            if len(data) < 4 {
-                out[6] = byte(PADDING)
-                out[5] = byte(PADDING)
-                if len(data) < 3 {
-                    out[4] = byte(PADDING)
-                    if len(data) < 2 {
-                        out[3] = byte(PADDING)
-                        out[2] = byte(PADDING)
-                    }
-                }
-            }
-            break
-        }
-        data = data[5:]
-        out = out[8:]
-    }
+		if len(data) < 5 {
+			out[7] = byte(PADDING)
+			if len(data) < 4 {
+				out[6] = byte(PADDING)
+				out[5] = byte(PADDING)
+				if len(data) < 3 {
+					out[4] = byte(PADDING)
+					if len(data) < 2 {
+						out[3] = byte(PADDING)
+						out[2] = byte(PADDING)
+					}
+				}
+			}
+			break
+		}
+		data = data[5:]
+		out = out[8:]
+	}
 }
 
 decode :: proc(data: string, DEC_TBL := DEC_TABLE, allocator := context.allocator) -> []byte #no_bounds_check{
-    if len(data) == 0 {
-        return nil
-    }
+	if len(data) == 0 {
+		return nil
+	}
 
-    outi := 0
-    data := data
+	outi := 0
+	data := data
 
-    out := make([]byte, len(data) / 8 * 5, allocator)
-    end := false
-    for len(data) > 0 && !end {
-        dbuf : [8]byte
-        dlen := 8
+	out := make([]byte, len(data) / 8 * 5, allocator)
+	end := false
+	for len(data) > 0 && !end {
+		dbuf : [8]byte
+		dlen := 8
 
-        for j := 0; j < 8; {
-            if len(data) == 0 {
-                dlen, end = j, true
-                break
-            }
-            input := data[0]
-            data = data[1:]
-            if input == byte(PADDING) && j >= 2 && len(data) < 8 {
-                assert(!(len(data) + j < 8 - 1), "Corrupted input")
-                for k := 0; k < 8-1-j; k +=1 {
-                    assert(len(data) < k || data[k] == byte(PADDING), "Corrupted input")
-                }
-                dlen, end = j, true
-                assert(dlen != 1 && dlen != 3 && dlen != 6, "Corrupted input")
-                break
-            }
-            dbuf[j] = DEC_TABLE[input]
-            assert(dbuf[j] != 0xff, "Corrupted input")
-            j += 1
-        }
+		for j := 0; j < 8; {
+			if len(data) == 0 {
+				dlen, end = j, true
+				break
+			}
+			input := data[0]
+			data = data[1:]
+			if input == byte(PADDING) && j >= 2 && len(data) < 8 {
+				assert(!(len(data) + j < 8 - 1), "Corrupted input")
+				for k := 0; k < 8-1-j; k +=1 {
+					assert(len(data) < k || data[k] == byte(PADDING), "Corrupted input")
+				}
+				dlen, end = j, true
+				assert(dlen != 1 && dlen != 3 && dlen != 6, "Corrupted input")
+				break
+			}
+			dbuf[j] = DEC_TABLE[input]
+			assert(dbuf[j] != 0xff, "Corrupted input")
+			j += 1
+		}
 
-        switch dlen {
-            case 8:
-                out[outi + 4] = dbuf[6] << 5 | dbuf[7]
-                fallthrough
-            case 7:
-                out[outi + 3] = dbuf[4] << 7 | dbuf[5] << 2 | dbuf[6] >> 3
-                fallthrough
-            case 5:
-                out[outi + 2] = dbuf[3] << 4 | dbuf[4] >> 1
-                fallthrough
-            case 4:
-                out[outi + 1] = dbuf[1] << 6 | dbuf[2] << 1 | dbuf[3] >> 4
-                fallthrough
-            case 2:
-                out[outi + 0] = dbuf[0] << 3 | dbuf[1] >> 2
-        }
-        outi += 5
-    }
-    return out
+		switch dlen {
+		case 8:
+			out[outi + 4] = dbuf[6] << 5 | dbuf[7]
+			fallthrough
+		case 7:
+			out[outi + 3] = dbuf[4] << 7 | dbuf[5] << 2 | dbuf[6] >> 3
+			fallthrough
+		case 5:
+			out[outi + 2] = dbuf[3] << 4 | dbuf[4] >> 1
+			fallthrough
+		case 4:
+			out[outi + 1] = dbuf[1] << 6 | dbuf[2] << 1 | dbuf[3] >> 4
+			fallthrough
+		case 2:
+			out[outi + 0] = dbuf[0] << 3 | dbuf[1] >> 2
+		}
+		outi += 5
+	}
+	return out
 }

core/encoding/cbor/cbor.odin

@@ -320,8 +320,8 @@ to_diagnostic_format :: proc {
 
 // Turns the given CBOR value into a human-readable string.
 // See docs on the proc group `diagnose` for more info.
-to_diagnostic_format_string :: proc(val: Value, padding := 0, allocator := context.allocator) -> (string, mem.Allocator_Error) #optional_allocator_error {
-	b := strings.builder_make(allocator)
+to_diagnostic_format_string :: proc(val: Value, padding := 0, allocator := context.allocator, loc := #caller_location) -> (string, mem.Allocator_Error) #optional_allocator_error {
+	b := strings.builder_make(allocator, loc)
 	w := strings.to_stream(&b)
 	err := to_diagnostic_format_writer(w, val, padding)
 	if err == .EOF {

core/encoding/cbor/coding.odin

@@ -95,24 +95,25 @@ decode :: decode_from
 
 // Decodes the given string as CBOR.
 // See docs on the proc group `decode` for more information.
-decode_from_string :: proc(s: string, flags: Decoder_Flags = {}, allocator := context.allocator) -> (v: Value, err: Decode_Error) {
+decode_from_string :: proc(s: string, flags: Decoder_Flags = {}, allocator := context.allocator, loc := #caller_location) -> (v: Value, err: Decode_Error) {
 	r: strings.Reader
 	strings.reader_init(&r, s)
-	return decode_from_reader(strings.reader_to_stream(&r), flags, allocator)
+	return decode_from_reader(strings.reader_to_stream(&r), flags, allocator, loc)
 }
 
 // Reads a CBOR value from the given reader.
 // See docs on the proc group `decode` for more information.
-decode_from_reader :: proc(r: io.Reader, flags: Decoder_Flags = {}, allocator := context.allocator) -> (v: Value, err: Decode_Error) {
+decode_from_reader :: proc(r: io.Reader, flags: Decoder_Flags = {}, allocator := context.allocator, loc := #caller_location) -> (v: Value, err: Decode_Error) {
 	return decode_from_decoder(
 		Decoder{ DEFAULT_MAX_PRE_ALLOC, flags, r },
 		allocator=allocator,
+		loc = loc,
 	)
 }
 
 // Reads a CBOR value from the given decoder.
 // See docs on the proc group `decode` for more information.
-decode_from_decoder :: proc(d: Decoder, allocator := context.allocator) -> (v: Value, err: Decode_Error) {
+decode_from_decoder :: proc(d: Decoder, allocator := context.allocator, loc := #caller_location) -> (v: Value, err: Decode_Error) {
 	context.allocator = allocator
 	
 	d := d
@@ -121,13 +122,13 @@ decode_from_decoder :: proc(d: Decoder, allocator := context.allocator) -> (v: V
 		d.max_pre_alloc = DEFAULT_MAX_PRE_ALLOC
 	}
 
-	v, err = _decode_from_decoder(d)
+	v, err = _decode_from_decoder(d, {}, allocator, loc)
 	// Normal EOF does not exist here, we try to read the exact amount that is said to be provided.
 	if err == .EOF { err = .Unexpected_EOF }
 	return
 }
 
-_decode_from_decoder :: proc(d: Decoder, hdr: Header = Header(0)) -> (v: Value, err: Decode_Error) {
+_decode_from_decoder :: proc(d: Decoder, hdr: Header = Header(0), allocator := context.allocator, loc := #caller_location) -> (v: Value, err: Decode_Error) {
 	hdr := hdr
 	r := d.reader
 	if hdr == Header(0) { hdr = _decode_header(r) or_return }
@@ -161,11 +162,11 @@ _decode_from_decoder :: proc(d: Decoder, hdr: Header = Header(0)) -> (v: Value,
 	switch maj {
 	case .Unsigned: return _decode_tiny_u8(add)
 	case .Negative: return Negative_U8(_decode_tiny_u8(add) or_return), nil
-	case .Bytes:    return _decode_bytes_ptr(d, add)
-	case .Text:     return _decode_text_ptr(d, add)
-	case .Array:    return _decode_array_ptr(d, add)
-	case .Map:      return _decode_map_ptr(d, add)
-	case .Tag:      return _decode_tag_ptr(d, add)
+	case .Bytes:    return _decode_bytes_ptr(d, add, .Bytes, allocator, loc)
+	case .Text:     return _decode_text_ptr(d, add, allocator, loc)
+	case .Array:    return _decode_array_ptr(d, add, allocator, loc)
+	case .Map:      return _decode_map_ptr(d, add, allocator, loc)
+	case .Tag:      return _decode_tag_ptr(d, add, allocator, loc)
 	case .Other:    return _decode_tiny_simple(add)
 	case:           return nil, .Bad_Major
 	}
@@ -203,27 +204,27 @@ encode :: encode_into
 
 // Encodes the CBOR value into binary CBOR allocated on the given allocator.
 // See the docs on the proc group `encode_into` for more info.
-encode_into_bytes :: proc(v: Value, flags := ENCODE_SMALL, allocator := context.allocator, temp_allocator := context.temp_allocator) -> (data: []byte, err: Encode_Error) {
-	b := strings.builder_make(allocator) or_return
+encode_into_bytes :: proc(v: Value, flags := ENCODE_SMALL, allocator := context.allocator, temp_allocator := context.temp_allocator, loc := #caller_location) -> (data: []byte, err: Encode_Error) {
+	b := strings.builder_make(allocator, loc) or_return
 	encode_into_builder(&b, v, flags, temp_allocator) or_return
 	return b.buf[:], nil
 }
 
 // Encodes the CBOR value into binary CBOR written to the given builder.
 // See the docs on the proc group `encode_into` for more info.
-encode_into_builder :: proc(b: ^strings.Builder, v: Value, flags := ENCODE_SMALL, temp_allocator := context.temp_allocator) -> Encode_Error {
-	return encode_into_writer(strings.to_stream(b), v, flags, temp_allocator)
+encode_into_builder :: proc(b: ^strings.Builder, v: Value, flags := ENCODE_SMALL, temp_allocator := context.temp_allocator, loc := #caller_location) -> Encode_Error {
+	return encode_into_writer(strings.to_stream(b), v, flags, temp_allocator, loc=loc)
 }
 
 // Encodes the CBOR value into binary CBOR written to the given writer.
 // See the docs on the proc group `encode_into` for more info.
-encode_into_writer :: proc(w: io.Writer, v: Value, flags := ENCODE_SMALL, temp_allocator := context.temp_allocator) -> Encode_Error {
-	return encode_into_encoder(Encoder{flags, w, temp_allocator}, v)
+encode_into_writer :: proc(w: io.Writer, v: Value, flags := ENCODE_SMALL, temp_allocator := context.temp_allocator, loc := #caller_location) -> Encode_Error {
+	return encode_into_encoder(Encoder{flags, w, temp_allocator}, v, loc=loc)
 }
 
 // Encodes the CBOR value into binary CBOR written to the given encoder.
 // See the docs on the proc group `encode_into` for more info.
-encode_into_encoder :: proc(e: Encoder, v: Value) -> Encode_Error {
+encode_into_encoder :: proc(e: Encoder, v: Value, loc := #caller_location) -> Encode_Error {
 	e := e
 
 	if e.temp_allocator.procedure == nil {
@@ -232,7 +233,7 @@ encode_into_encoder :: proc(e: Encoder, v: Value) -> Encode_Error {
 
 	if .Self_Described_CBOR in e.flags {
 		_encode_u64(e, TAG_SELF_DESCRIBED_CBOR, .Tag) or_return
-		e.flags &~= { .Self_Described_CBOR }
+		e.flags -= { .Self_Described_CBOR }
 	}
 
 	switch v_spec in v {
@@ -366,21 +367,21 @@ _encode_u64_exact :: proc(w: io.Writer, v: u64, major: Major = .Unsigned) -> (er
 	return
 }
 
-_decode_bytes_ptr :: proc(d: Decoder, add: Add, type: Major = .Bytes) -> (v: ^Bytes, err: Decode_Error) {
-	v = new(Bytes) or_return
-	defer if err != nil { free(v) }
+_decode_bytes_ptr :: proc(d: Decoder, add: Add, type: Major = .Bytes, allocator := context.allocator, loc := #caller_location) -> (v: ^Bytes, err: Decode_Error) {
+	v = new(Bytes, allocator, loc) or_return
+	defer if err != nil { free(v, allocator, loc) }
 
-	v^ = _decode_bytes(d, add, type) or_return
+	v^ = _decode_bytes(d, add, type, allocator, loc) or_return
 	return
 }
 
-_decode_bytes :: proc(d: Decoder, add: Add, type: Major = .Bytes, allocator := context.allocator) -> (v: Bytes, err: Decode_Error) {
+_decode_bytes :: proc(d: Decoder, add: Add, type: Major = .Bytes, allocator := context.allocator, loc := #caller_location) -> (v: Bytes, err: Decode_Error) {
 	context.allocator = allocator
 
 	add := add
 	n, scap := _decode_len_str(d, add) or_return
 	
-	buf := strings.builder_make(0, scap) or_return
+	buf := strings.builder_make(0, scap, allocator, loc) or_return
 	defer if err != nil { strings.builder_destroy(&buf) }
 	buf_stream := strings.to_stream(&buf)
 
@@ -422,44 +423,44 @@ _decode_bytes :: proc(d: Decoder, add: Add, type: Major = .Bytes, allocator := c
 _encode_bytes :: proc(e: Encoder, val: Bytes, major: Major = .Bytes) -> (err: Encode_Error) {
 	assert(len(val) >= 0)
 	_encode_u64(e, u64(len(val)), major) or_return
-    _, err = io.write_full(e.writer, val[:])
+	_, err = io.write_full(e.writer, val[:])
 	return
 }
 
-_decode_text_ptr :: proc(d: Decoder, add: Add) -> (v: ^Text, err: Decode_Error) {
-	v = new(Text) or_return
+_decode_text_ptr :: proc(d: Decoder, add: Add, allocator := context.allocator, loc := #caller_location) -> (v: ^Text, err: Decode_Error) {
+	v = new(Text, allocator, loc) or_return
 	defer if err != nil { free(v) }
 
-	v^ = _decode_text(d, add) or_return
+	v^ = _decode_text(d, add, allocator, loc) or_return
 	return
 }
 
-_decode_text :: proc(d: Decoder, add: Add, allocator := context.allocator) -> (v: Text, err: Decode_Error) {
-	return (Text)(_decode_bytes(d, add, .Text, allocator) or_return), nil
+_decode_text :: proc(d: Decoder, add: Add, allocator := context.allocator, loc := #caller_location) -> (v: Text, err: Decode_Error) {
+	return (Text)(_decode_bytes(d, add, .Text, allocator, loc) or_return), nil
 }
 
 _encode_text :: proc(e: Encoder, val: Text) -> Encode_Error {
-    return _encode_bytes(e, transmute([]byte)val, .Text)
+	return _encode_bytes(e, transmute([]byte)val, .Text)
 }
 
-_decode_array_ptr :: proc(d: Decoder, add: Add) -> (v: ^Array, err: Decode_Error) {
-	v = new(Array) or_return
+_decode_array_ptr :: proc(d: Decoder, add: Add, allocator := context.allocator, loc := #caller_location) -> (v: ^Array, err: Decode_Error) {
+	v = new(Array, allocator, loc) or_return
 	defer if err != nil { free(v) }
 
-	v^ = _decode_array(d, add) or_return
+	v^ = _decode_array(d, add, allocator, loc) or_return
 	return
 }
 
-_decode_array :: proc(d: Decoder, add: Add) -> (v: Array, err: Decode_Error) {
+_decode_array :: proc(d: Decoder, add: Add, allocator := context.allocator, loc := #caller_location) -> (v: Array, err: Decode_Error) {
 	n, scap := _decode_len_container(d, add) or_return
-	array := make([dynamic]Value, 0, scap) or_return
+	array := make([dynamic]Value, 0, scap, allocator, loc) or_return
 	defer if err != nil {
-		for entry in array { destroy(entry) }
-		delete(array)
+		for entry in array { destroy(entry, allocator) }
+		delete(array, loc)
 	}
 	
 	for i := 0; n == -1 || i < n; i += 1 {
-		val, verr := _decode_from_decoder(d)
+		val, verr := _decode_from_decoder(d, {}, allocator, loc)
 		if n == -1 && verr == .Break {
 			break
 		} else if verr != nil {
@@ -479,45 +480,45 @@ _decode_array :: proc(d: Decoder, add: Add) -> (v: Array, err: Decode_Error) {
 _encode_array :: proc(e: Encoder, arr: Array) -> Encode_Error {
 	assert(len(arr) >= 0)
 	_encode_u64(e, u64(len(arr)), .Array)
-    for val in arr {
-        encode(e, val) or_return
-    }
-    return nil
+	for val in arr {
+		encode(e, val) or_return
+	}
+	return nil
 }
 
-_decode_map_ptr :: proc(d: Decoder, add: Add) -> (v: ^Map, err: Decode_Error) {
-	v = new(Map) or_return
+_decode_map_ptr :: proc(d: Decoder, add: Add, allocator := context.allocator, loc := #caller_location) -> (v: ^Map, err: Decode_Error) {
+	v = new(Map, allocator, loc) or_return
 	defer if err != nil { free(v) }
 
-	v^ = _decode_map(d, add) or_return
+	v^ = _decode_map(d, add, allocator, loc) or_return
 	return
 }
 
-_decode_map :: proc(d: Decoder, add: Add) -> (v: Map, err: Decode_Error) {
+_decode_map :: proc(d: Decoder, add: Add, allocator := context.allocator, loc := #caller_location) -> (v: Map, err: Decode_Error) {
 	n, scap := _decode_len_container(d, add) or_return
-	items := make([dynamic]Map_Entry, 0, scap) or_return
+	items := make([dynamic]Map_Entry, 0, scap, allocator, loc) or_return
 	defer if err != nil { 
 		for entry in items {
 			destroy(entry.key)
 			destroy(entry.value)
 		}
-		delete(items)
+		delete(items, loc)
 	}
 
 	for i := 0; n == -1 || i < n; i += 1 {
-		key, kerr := _decode_from_decoder(d)
+		key, kerr := _decode_from_decoder(d, {}, allocator, loc)
 		if n == -1 && kerr == .Break {
 			break
 		} else if kerr != nil {
 			return nil, kerr
 		} 
 
-		value := _decode_from_decoder(d) or_return
+		value := _decode_from_decoder(d, {}, allocator, loc) or_return
 
 		append(&items, Map_Entry{
 			key   = key,
 			value = value,
-		}) or_return
+		}, loc) or_return
 	}
 
 	if .Shrink_Excess in d.flags { shrink(&items) }
@@ -575,23 +576,23 @@ _encode_map :: proc(e: Encoder, m: Map) -> (err: Encode_Error) {
 		encode(e, entry.entry.value) or_return
 	}
 
-    return nil
+	return nil
 }
 
-_decode_tag_ptr :: proc(d: Decoder, add: Add) -> (v: Value, err: Decode_Error) {
-	tag := _decode_tag(d, add) or_return
+_decode_tag_ptr :: proc(d: Decoder, add: Add, allocator := context.allocator, loc := #caller_location) -> (v: Value, err: Decode_Error) {
+	tag := _decode_tag(d, add, allocator, loc) or_return
 	if t, ok := tag.?; ok {
 		defer if err != nil { destroy(t.value) }
-		tp := new(Tag) or_return
+		tp := new(Tag, allocator, loc) or_return
 		tp^ = t
 		return tp, nil
 	}
 
 	// no error, no tag, this was the self described CBOR tag, skip it.
-	return _decode_from_decoder(d)
+	return _decode_from_decoder(d, {}, allocator, loc)
 }
 
-_decode_tag :: proc(d: Decoder, add: Add) -> (v: Maybe(Tag), err: Decode_Error) {
+_decode_tag :: proc(d: Decoder, add: Add, allocator := context.allocator, loc := #caller_location) -> (v: Maybe(Tag), err: Decode_Error) {
 	num := _decode_uint_as_u64(d.reader, add) or_return
 
 	// CBOR can be wrapped in a tag that decoders can use to see/check if the binary data is CBOR.
@@ -602,7 +603,7 @@ _decode_tag :: proc(d: Decoder, add: Add) -> (v: Maybe(Tag), err: Decode_Error)
 
 	t := Tag{
 		number = num,
-		value = _decode_from_decoder(d) or_return,
+		value = _decode_from_decoder(d, {}, allocator, loc) or_return,
 	}
 
 	if nested, ok := t.value.(^Tag); ok {
@@ -625,7 +626,7 @@ _decode_uint_as_u64 :: proc(r: io.Reader, add: Add) -> (nr: u64, err: Decode_Err
 
 _encode_tag :: proc(e: Encoder, val: Tag) -> Encode_Error {
 	_encode_u64(e, val.number, .Tag) or_return
-    return encode(e, val.value)
+	return encode(e, val.value)
 }
 
 _decode_simple :: proc(r: io.Reader) -> (v: Simple, err: io.Error) {
@@ -738,16 +739,16 @@ _encode_nil :: proc(w: io.Writer) -> io.Error {
 // Streaming
 
 encode_stream_begin :: proc(w: io.Writer, major: Major) -> (err: io.Error) {
-    assert(major >= Major(.Bytes) && major <= Major(.Map), "illegal stream type")
+	assert(major >= Major(.Bytes) && major <= Major(.Map), "illegal stream type")
 
-    header := (u8(major) << 5) | u8(Add.Length_Unknown)
-    _, err = io.write_full(w, {header})
+	header := (u8(major) << 5) | u8(Add.Length_Unknown)
+	_, err = io.write_full(w, {header})
 	return
 }
 
 encode_stream_end :: proc(w: io.Writer) -> io.Error {
-    header := (u8(Major.Other) << 5) | u8(Add.Break)
-    _, err := io.write_full(w, {header})
+	header := (u8(Major.Other) << 5) | u8(Add.Break)
+	_, err := io.write_full(w, {header})
 	return err
 }
 
@@ -756,8 +757,8 @@ encode_stream_text       :: _encode_text
 encode_stream_array_item :: encode
 
 encode_stream_map_entry :: proc(e: Encoder, key: Value, val: Value) -> Encode_Error {
-    encode(e, key) or_return
-    return encode(e, val)
+	encode(e, key) or_return
+	return encode(e, val)
 }
 
 // For `Bytes` and `Text` strings: Decodes the number of items the header says follows.
@@ -883,4 +884,4 @@ _encode_deterministic_f64 :: proc(w: io.Writer, v: f64) -> io.Error {
 	}
 
 	return _encode_f64_exact(w, v)
-}
+}

core/encoding/cbor/doc.odin

@@ -77,8 +77,11 @@ You can look at the default tags provided for pointers on how these implementati
 Example:
 	package main
 
+	import "base:intrinsics"
+
 	import "core:encoding/cbor"
 	import "core:fmt"
+	import "core:reflect"
 	import "core:time"
 
 	Possibilities :: union {
@@ -93,9 +96,32 @@ Example:
 		ignore_this: ^Data `cbor:"-"`,     // Ignored by implementation.
 		renamed: f32 `cbor:"renamed :)"`,  // Renamed when encoded.
 		my_union: Possibilities,           // Union support.
+
+		my_raw: [8]u32 `cbor_tag:"raw"`, // Custom tag that just writes the value as bytes.
 	}
 
 	main :: proc() {
+		// Example custom tag implementation that instead of breaking down all parts,
+		// just writes the value as a big byte blob. This is an advanced feature but very powerful.
+		RAW_TAG_NR :: 200
+		cbor.tag_register_number({
+			marshal = proc(_: ^cbor.Tag_Implementation, e: cbor.Encoder, v: any) -> cbor.Marshal_Error {
+				cbor._encode_u8(e.writer, RAW_TAG_NR, .Tag) or_return
+				return cbor.err_conv(cbor._encode_bytes(e, reflect.as_bytes(v)))
+			},
+			unmarshal = proc(_: ^cbor.Tag_Implementation, d: cbor.Decoder, _: cbor.Tag_Number, v: any) -> (cbor.Unmarshal_Error) {
+				hdr := cbor._decode_header(d.reader) or_return
+				maj, add := cbor._header_split(hdr)
+				if maj != .Bytes {
+					return .Bad_Tag_Value
+				}
+
+				bytes := cbor.err_conv(cbor._decode_bytes(d, add, maj)) or_return
+				intrinsics.mem_copy_non_overlapping(v.data, raw_data(bytes), len(bytes))
+				return nil
+			},
+		}, RAW_TAG_NR, "raw")
+
 		now := time.Time{_nsec = 1701117968 * 1e9}
 
 		data := Data{
@@ -105,21 +131,22 @@ Example:
 			ignore_this = &Data{},
 			renamed     = 123123.125,
 			my_union    = 3,
+			my_raw      = {1=1, 2=2, 3=3},
 		}
-		
+
 		// Marshal the struct into binary CBOR.
 		binary, err := cbor.marshal(data, cbor.ENCODE_FULLY_DETERMINISTIC)
-		assert(err == nil)
+		fmt.assertf(err == nil, "marshal error: %v", err)
 		defer delete(binary)
-		
+
 		// Decode the binary data into a `cbor.Value`.
 		decoded, derr := cbor.decode(string(binary))
-		assert(derr == nil)
+		fmt.assertf(derr == nil, "decode error: %v", derr)
 		defer cbor.destroy(decoded)
 
 		// Turn the CBOR into a human readable representation defined as the diagnostic format in [[RFC 8949 Section 8;https://www.rfc-editor.org/rfc/rfc8949.html#name-diagnostic-notation]].
 		diagnosis, eerr := cbor.to_diagnostic_format(decoded)
-		assert(eerr == nil)
+		fmt.assertf(eerr == nil, "to diagnostic error: %v", eerr)
 		defer delete(diagnosis)
 
 		fmt.println(diagnosis)
@@ -127,6 +154,7 @@ Example:
 
 Output:
 	{
+		"my_raw": 200(h'00001000200030000000000000000000'),
 		"my_union": 1010([
 			"int",
 			3

core/encoding/cbor/marshal.odin

@@ -45,8 +45,8 @@ marshal :: marshal_into
 
 // Marshals the given value into a CBOR byte stream (allocated using the given allocator).
 // See docs on the `marshal_into` proc group for more info.
-marshal_into_bytes :: proc(v: any, flags := ENCODE_SMALL, allocator := context.allocator, temp_allocator := context.temp_allocator) -> (bytes: []byte, err: Marshal_Error) {
-	b, alloc_err := strings.builder_make(allocator)
+marshal_into_bytes :: proc(v: any, flags := ENCODE_SMALL, allocator := context.allocator, temp_allocator := context.temp_allocator, loc := #caller_location) -> (bytes: []byte, err: Marshal_Error) {
+	b, alloc_err := strings.builder_make(allocator, loc=loc)
  	// The builder as a stream also returns .EOF if it ran out of memory so this is consistent.
 	if alloc_err != nil {
 		return nil, .EOF
@@ -85,7 +85,7 @@ marshal_into_encoder :: proc(e: Encoder, v: any) -> (err: Marshal_Error) {
 
 	if .Self_Described_CBOR in e.flags {
 		err_conv(_encode_u64(e, TAG_SELF_DESCRIBED_CBOR, .Tag)) or_return
-		e.flags &~= { .Self_Described_CBOR }
+		e.flags -= { .Self_Described_CBOR }
 	}
 
 	if v == nil {
@@ -97,11 +97,14 @@ marshal_into_encoder :: proc(e: Encoder, v: any) -> (err: Marshal_Error) {
 		return impl->marshal(e, v)
 	}
 
-	ti := runtime.type_info_base(type_info_of(v.id))
-	a := any{v.data, ti.id}
+	ti := runtime.type_info_core(type_info_of(v.id))
+	return _marshal_into_encoder(e, v, ti)
+}
 
+_marshal_into_encoder :: proc(e: Encoder, v: any, ti: ^runtime.Type_Info) -> (err: Marshal_Error) {
+	a := any{v.data, ti.id}
 	#partial switch info in ti.variant {
-	case runtime.Type_Info_Named:
+	case runtime.Type_Info_Named, runtime.Type_Info_Enum, runtime.Type_Info_Bit_Field:
 		unreachable()
 
 	case runtime.Type_Info_Pointer:
@@ -223,18 +226,38 @@ marshal_into_encoder :: proc(e: Encoder, v: any) -> (err: Marshal_Error) {
 		}
 
 		err_conv(_encode_u64(e, u64(info.count), .Array)) or_return
+
+		if impl, ok := _tag_implementations_type[info.elem.id]; ok {
+			for i in 0..<info.count {
+				data := uintptr(v.data) + uintptr(i*info.elem_size)
+				impl->marshal(e, any{rawptr(data), info.elem.id}) or_return
+			}
+			return
+		}
+
+		elem_ti := runtime.type_info_core(type_info_of(info.elem.id))
 		for i in 0..<info.count {
 			data := uintptr(v.data) + uintptr(i*info.elem_size)
-			marshal_into(e, any{rawptr(data), info.elem.id}) or_return
+			_marshal_into_encoder(e, any{rawptr(data), info.elem.id}, elem_ti) or_return
 		}
 		return
 
 	case runtime.Type_Info_Enumerated_Array:
 		// index := runtime.type_info_base(info.index).variant.(runtime.Type_Info_Enum)
 		err_conv(_encode_u64(e, u64(info.count), .Array)) or_return
+
+		if impl, ok := _tag_implementations_type[info.elem.id]; ok {
+			for i in 0..<info.count {
+				data := uintptr(v.data) + uintptr(i*info.elem_size)
+				impl->marshal(e, any{rawptr(data), info.elem.id}) or_return
+			}
+			return
+		}
+
+		elem_ti := runtime.type_info_core(type_info_of(info.elem.id))
 		for i in 0..<info.count {
 			data := uintptr(v.data) + uintptr(i*info.elem_size)
-			marshal_into(e, any{rawptr(data), info.elem.id}) or_return
+			_marshal_into_encoder(e, any{rawptr(data), info.elem.id}, elem_ti) or_return
 		}
 		return
 		
@@ -246,9 +269,19 @@ marshal_into_encoder :: proc(e: Encoder, v: any) -> (err: Marshal_Error) {
 
 		array := (^mem.Raw_Dynamic_Array)(v.data)
 		err_conv(_encode_u64(e, u64(array.len), .Array)) or_return
+
+		if impl, ok := _tag_implementations_type[info.elem.id]; ok {
+			for i in 0..<array.len {
+				data := uintptr(array.data) + uintptr(i*info.elem_size)
+				impl->marshal(e, any{rawptr(data), info.elem.id}) or_return
+			}
+			return
+		}
+
+		elem_ti := runtime.type_info_core(type_info_of(info.elem.id))
 		for i in 0..<array.len {
 			data := uintptr(array.data) + uintptr(i*info.elem_size)
-			marshal_into(e, any{rawptr(data), info.elem.id}) or_return
+			_marshal_into_encoder(e, any{rawptr(data), info.elem.id}, elem_ti) or_return
 		}
 		return
 
@@ -260,9 +293,19 @@ marshal_into_encoder :: proc(e: Encoder, v: any) -> (err: Marshal_Error) {
 
 		array := (^mem.Raw_Slice)(v.data)
 		err_conv(_encode_u64(e, u64(array.len), .Array)) or_return
+
+		if impl, ok := _tag_implementations_type[info.elem.id]; ok {
+			for i in 0..<array.len {
+				data := uintptr(array.data) + uintptr(i*info.elem_size)
+				impl->marshal(e, any{rawptr(data), info.elem.id}) or_return
+			}
+			return
+		}
+
+		elem_ti := runtime.type_info_core(type_info_of(info.elem.id))
 		for i in 0..<array.len {
 			data := uintptr(array.data) + uintptr(i*info.elem_size)
-			marshal_into(e, any{rawptr(data), info.elem.id}) or_return
+			_marshal_into_encoder(e, any{rawptr(data), info.elem.id}, elem_ti) or_return
 		}
 		return
 
@@ -308,7 +351,8 @@ marshal_into_encoder :: proc(e: Encoder, v: any) -> (err: Marshal_Error) {
 				builder := strings.builder_from_slice(res[:])
 				e.writer = strings.to_stream(&builder)
 
-				assert(_encode_u64(e, u64(len(str)), .Text) == nil)
+				err := _encode_u64(e, u64(len(str)), .Text)
+				assert(err == nil)
 				res[9] = u8(len(builder.buf))
 				assert(res[9] < 10)
 				return
@@ -463,7 +507,7 @@ marshal_into_encoder :: proc(e: Encoder, v: any) -> (err: Marshal_Error) {
 		}
 		
 		n: u64; {
-			for _, i in info.names {
+			for _, i in info.names[:info.field_count] {
 				if field_name(info, i) != "-" {
 					n += 1
 				}
@@ -479,7 +523,7 @@ marshal_into_encoder :: proc(e: Encoder, v: any) -> (err: Marshal_Error) {
 			entries := make([dynamic]Name, 0, n, e.temp_allocator) or_return
 			defer delete(entries)
 
-			for _, i in info.names {
+			for _, i in info.names[:info.field_count] {
 				fname := field_name(info, i)
 				if fname == "-" {
 					continue
@@ -497,7 +541,7 @@ marshal_into_encoder :: proc(e: Encoder, v: any) -> (err: Marshal_Error) {
 				marshal_entry(e, info, v, entry.name, entry.field) or_return
 			}
 		} else {
-			for _, i in info.names {
+			for _, i in info.names[:info.field_count] {
 				fname := field_name(info, i)
 				if fname == "-" {
 					continue
@@ -542,9 +586,6 @@ marshal_into_encoder :: proc(e: Encoder, v: any) -> (err: Marshal_Error) {
 
 		return marshal_into(e, any{v.data, vti.id})
 
-	case runtime.Type_Info_Enum:
-		return marshal_into(e, any{v.data, info.base.id})
-
 	case runtime.Type_Info_Bit_Set:
 		// Store bit_set as big endian just like the protocol.
 		do_byte_swap := !reflect.bit_set_is_big_endian(v)

core/encoding/cbor/tags.odin

@@ -95,7 +95,6 @@ tag_register_number :: proc(impl: Tag_Implementation, nr: Tag_Number, id: string
 }
 
 // Controls initialization of default tag implementations.
-// JS and WASI default to a panic allocator so we don't want to do it on those.
 INITIALIZE_DEFAULT_TAGS :: #config(CBOR_INITIALIZE_DEFAULT_TAGS, !ODIN_DEFAULT_TO_PANIC_ALLOCATOR && !ODIN_DEFAULT_TO_NIL_ALLOCATOR)
 
 @(private, init, disabled=!INITIALIZE_DEFAULT_TAGS)

core/encoding/cbor/unmarshal.odin

@@ -31,8 +31,8 @@ unmarshal :: proc {
 	unmarshal_from_string,
 }
 
-unmarshal_from_reader :: proc(r: io.Reader, ptr: ^$T, flags := Decoder_Flags{}, allocator := context.allocator, temp_allocator := context.temp_allocator) -> (err: Unmarshal_Error) {
-	err = unmarshal_from_decoder(Decoder{ DEFAULT_MAX_PRE_ALLOC, flags, r }, ptr, allocator, temp_allocator)
+unmarshal_from_reader :: proc(r: io.Reader, ptr: ^$T, flags := Decoder_Flags{}, allocator := context.allocator, temp_allocator := context.temp_allocator, loc := #caller_location) -> (err: Unmarshal_Error) {
+	err = unmarshal_from_decoder(Decoder{ DEFAULT_MAX_PRE_ALLOC, flags, r }, ptr, allocator, temp_allocator, loc)
 
 	// Normal EOF does not exist here, we try to read the exact amount that is said to be provided.
 	if err == .EOF { err = .Unexpected_EOF }
@@ -40,21 +40,21 @@ unmarshal_from_reader :: proc(r: io.Reader, ptr: ^$T, flags := Decoder_Flags{},
 }
 
 // Unmarshals from a string, see docs on the proc group `Unmarshal` for more info.
-unmarshal_from_string :: proc(s: string, ptr: ^$T, flags := Decoder_Flags{}, allocator := context.allocator, temp_allocator := context.temp_allocator) -> (err: Unmarshal_Error) {
+unmarshal_from_string :: proc(s: string, ptr: ^$T, flags := Decoder_Flags{}, allocator := context.allocator, temp_allocator := context.temp_allocator, loc := #caller_location) -> (err: Unmarshal_Error) {
 	sr: strings.Reader
 	r := strings.to_reader(&sr, s)
 
-	err = unmarshal_from_reader(r, ptr, flags, allocator, temp_allocator)
+	err = unmarshal_from_reader(r, ptr, flags, allocator, temp_allocator, loc)
 
 	// Normal EOF does not exist here, we try to read the exact amount that is said to be provided.
 	if err == .EOF { err = .Unexpected_EOF }
 	return
 }
 
-unmarshal_from_decoder :: proc(d: Decoder, ptr: ^$T, allocator := context.allocator, temp_allocator := context.temp_allocator) -> (err: Unmarshal_Error) {
+unmarshal_from_decoder :: proc(d: Decoder, ptr: ^$T, allocator := context.allocator, temp_allocator := context.temp_allocator, loc := #caller_location) -> (err: Unmarshal_Error) {
 	d := d
 
-	err = _unmarshal_any_ptr(d, ptr, nil, allocator, temp_allocator)
+	err = _unmarshal_any_ptr(d, ptr, nil, allocator, temp_allocator, loc)
 
 	// Normal EOF does not exist here, we try to read the exact amount that is said to be provided.
 	if err == .EOF { err = .Unexpected_EOF }
@@ -62,7 +62,7 @@ unmarshal_from_decoder :: proc(d: Decoder, ptr: ^$T, allocator := context.alloca
 
 }
 
-_unmarshal_any_ptr :: proc(d: Decoder, v: any, hdr: Maybe(Header) = nil, allocator := context.allocator, temp_allocator := context.temp_allocator) -> Unmarshal_Error {
+_unmarshal_any_ptr :: proc(d: Decoder, v: any, hdr: Maybe(Header) = nil, allocator := context.allocator, temp_allocator := context.temp_allocator, loc := #caller_location) -> Unmarshal_Error {
 	context.allocator = allocator
 	context.temp_allocator = temp_allocator
 	v := v
@@ -78,10 +78,10 @@ _unmarshal_any_ptr :: proc(d: Decoder, v: any, hdr: Maybe(Header) = nil, allocat
 	}
 	
 	data := any{(^rawptr)(v.data)^, ti.variant.(reflect.Type_Info_Pointer).elem.id}	
-	return _unmarshal_value(d, data, hdr.? or_else (_decode_header(d.reader) or_return))
+	return _unmarshal_value(d, data, hdr.? or_else (_decode_header(d.reader) or_return), allocator, temp_allocator, loc)
 }
 
-_unmarshal_value :: proc(d: Decoder, v: any, hdr: Header) -> (err: Unmarshal_Error) {
+_unmarshal_value :: proc(d: Decoder, v: any, hdr: Header, allocator := context.allocator, temp_allocator := context.temp_allocator, loc := #caller_location) -> (err: Unmarshal_Error) {
 	v := v
 	ti := reflect.type_info_base(type_info_of(v.id))
 	r := d.reader
@@ -96,7 +96,8 @@ _unmarshal_value :: proc(d: Decoder, v: any, hdr: Header) -> (err: Unmarshal_Err
 			ti = reflect.type_info_base(variant)
 			if !reflect.is_pointer_internally(variant) {
 				tag := any{rawptr(uintptr(v.data) + u.tag_offset), u.tag_type.id}
-				assert(_assign_int(tag, 1))
+				assigned := _assign_int(tag, 1)
+				assert(assigned)
 			}
 		}
 	}
@@ -104,7 +105,7 @@ _unmarshal_value :: proc(d: Decoder, v: any, hdr: Header) -> (err: Unmarshal_Err
 	// Allow generic unmarshal by doing it into a `Value`.
 	switch &dst in v {
 	case Value:
-		dst = err_conv(_decode_from_decoder(d, hdr)) or_return
+		dst = err_conv(_decode_from_decoder(d, hdr, allocator, loc)) or_return
 		return
 	}
 
@@ -273,13 +274,13 @@ _unmarshal_value :: proc(d: Decoder, v: any, hdr: Header) -> (err: Unmarshal_Err
 
 		 // NOTE: Because this is a special type and not to be treated as a general integer,
 		 // We only put the value of it in fields that are explicitly of type `Simple`.
-		 switch &dst in v {
-		 case Simple:
-			 dst = decoded
-			 return
-		 case:
-		 	return _unsupported(v, hdr, add)
-		 }
+		switch &dst in v {
+		case Simple:
+			dst = decoded
+			return
+		case:
+			return _unsupported(v, hdr, add)
+		}
 
 	case .Tag:
 		switch &dst in v {
@@ -308,7 +309,7 @@ _unmarshal_value :: proc(d: Decoder, v: any, hdr: Header) -> (err: Unmarshal_Err
 		if impl, ok := _tag_implementations_nr[nr]; ok {
 			return impl->unmarshal(d, nr, v)
 		} else if nr == TAG_OBJECT_TYPE {
-			return _unmarshal_union(d, v, ti, hdr)
+			return _unmarshal_union(d, v, ti, hdr, loc=loc)
 		} else {
 			// Discard the tag info and unmarshal as its value.
 			return _unmarshal_value(d, v, _decode_header(r) or_return)
@@ -316,19 +317,19 @@ _unmarshal_value :: proc(d: Decoder, v: any, hdr: Header) -> (err: Unmarshal_Err
 
 		return _unsupported(v, hdr, add)
 
-	case .Bytes: return _unmarshal_bytes(d, v, ti, hdr, add)
-	case .Text:  return _unmarshal_string(d, v, ti, hdr, add)
-	case .Array: return _unmarshal_array(d, v, ti, hdr, add)
-	case .Map:   return _unmarshal_map(d, v, ti, hdr, add)
+	case .Bytes: return _unmarshal_bytes(d, v, ti, hdr, add, allocator=allocator, loc=loc)
+	case .Text:  return _unmarshal_string(d, v, ti, hdr, add, allocator=allocator, loc=loc)
+	case .Array: return _unmarshal_array(d, v, ti, hdr, add, allocator=allocator, loc=loc)
+	case .Map:   return _unmarshal_map(d, v, ti, hdr, add, allocator=allocator, loc=loc)
 
 	case:        return .Bad_Major
 	}
 }
 
-_unmarshal_bytes :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header, add: Add) -> (err: Unmarshal_Error) {
+_unmarshal_bytes :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header, add: Add, allocator := context.allocator, loc := #caller_location) -> (err: Unmarshal_Error) {
 	#partial switch t in ti.variant {
 	case reflect.Type_Info_String:
-		bytes := err_conv(_decode_bytes(d, add)) or_return
+		bytes := err_conv(_decode_bytes(d, add, allocator=allocator, loc=loc)) or_return
 
 		if t.is_cstring {
 			raw  := (^cstring)(v.data)
@@ -347,7 +348,7 @@ _unmarshal_bytes :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 
 		if elem_base.id != byte { return _unsupported(v, hdr) }
 
-		bytes := err_conv(_decode_bytes(d, add)) or_return
+		bytes := err_conv(_decode_bytes(d, add, allocator=allocator, loc=loc)) or_return
 		raw   := (^mem.Raw_Slice)(v.data)
 		raw^   = transmute(mem.Raw_Slice)bytes
 		return
@@ -357,12 +358,12 @@ _unmarshal_bytes :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 
 		if elem_base.id != byte { return _unsupported(v, hdr) }
 		
-		bytes         := err_conv(_decode_bytes(d, add)) or_return
+		bytes         := err_conv(_decode_bytes(d, add, allocator=allocator, loc=loc)) or_return
 		raw           := (^mem.Raw_Dynamic_Array)(v.data)
 		raw.data       = raw_data(bytes)
 		raw.len        = len(bytes)
 		raw.cap        = len(bytes)
-		raw.allocator  = context.allocator
+		raw.allocator  = allocator
 		return
 
 	case reflect.Type_Info_Array:
@@ -385,10 +386,10 @@ _unmarshal_bytes :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 	return _unsupported(v, hdr)
 }
 
-_unmarshal_string :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header, add: Add) -> (err: Unmarshal_Error) {
+_unmarshal_string :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header, add: Add, allocator := context.allocator, temp_allocator := context.temp_allocator, loc := #caller_location) -> (err: Unmarshal_Error) {
 	#partial switch t in ti.variant {
 	case reflect.Type_Info_String:
-		text := err_conv(_decode_text(d, add)) or_return
+		text := err_conv(_decode_text(d, add, allocator, loc)) or_return
 
 		if t.is_cstring {
 			raw := (^cstring)(v.data)
@@ -403,8 +404,8 @@ _unmarshal_string :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Heade
 
 	// Enum by its variant name.
 	case reflect.Type_Info_Enum:
-		text := err_conv(_decode_text(d, add, allocator=context.temp_allocator)) or_return
-		defer delete(text, context.temp_allocator)
+		text := err_conv(_decode_text(d, add, allocator=temp_allocator, loc=loc)) or_return
+		defer delete(text, temp_allocator, loc)
 
 		for name, i in t.names {
 			if name == text {
@@ -414,8 +415,8 @@ _unmarshal_string :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Heade
 		}
 	
 	case reflect.Type_Info_Rune:
-		text := err_conv(_decode_text(d, add, allocator=context.temp_allocator)) or_return
-		defer delete(text, context.temp_allocator)
+		text := err_conv(_decode_text(d, add, allocator=temp_allocator, loc=loc)) or_return
+		defer delete(text, temp_allocator, loc)
 
 		r := (^rune)(v.data)
 		dr, n := utf8.decode_rune(text)
@@ -430,13 +431,15 @@ _unmarshal_string :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Heade
 	return _unsupported(v, hdr)
 }
 
-_unmarshal_array :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header, add: Add) -> (err: Unmarshal_Error) {
+_unmarshal_array :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header, add: Add, allocator := context.allocator, loc := #caller_location) -> (err: Unmarshal_Error) {
 	assign_array :: proc(
 		d: Decoder,
 		da: ^mem.Raw_Dynamic_Array,
 		elemt: ^reflect.Type_Info,
 		length: int,
 		growable := true,
+		allocator := context.allocator,
+		loc := #caller_location,
 	) -> (out_of_space: bool, err: Unmarshal_Error) {
 		for idx: uintptr = 0; length == -1 || idx < uintptr(length); idx += 1 {
 			elem_ptr := rawptr(uintptr(da.data) + idx*uintptr(elemt.size))
@@ -450,13 +453,13 @@ _unmarshal_array :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 				if !growable { return true, .Out_Of_Memory }
 
 				cap := 2 * da.cap
-				ok := runtime.__dynamic_array_reserve(da, elemt.size, elemt.align, cap)
+				ok := runtime.__dynamic_array_reserve(da, elemt.size, elemt.align, cap, loc)
  				
 				// NOTE: Might be lying here, but it is at least an allocator error.
 				if !ok { return false, .Out_Of_Memory }
 			}
 			
-			err = _unmarshal_value(d, elem, hdr)
+			err = _unmarshal_value(d, elem, hdr, allocator=allocator, loc=loc)
 			if length == -1 && err == .Break { break }
 			if err != nil { return }
 
@@ -469,10 +472,10 @@ _unmarshal_array :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 	// Allow generically storing the values array.
 	switch &dst in v {
 	case ^Array:
-		dst = err_conv(_decode_array_ptr(d, add)) or_return
+		dst = err_conv(_decode_array_ptr(d, add, allocator=allocator, loc=loc)) or_return
 		return
 	case Array:
-		dst = err_conv(_decode_array(d, add)) or_return
+		dst = err_conv(_decode_array(d, add, allocator=allocator, loc=loc)) or_return
 		return
 	}
 
@@ -480,8 +483,8 @@ _unmarshal_array :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 	case reflect.Type_Info_Slice:
 		length, scap := err_conv(_decode_len_container(d, add)) or_return
 
-		data := mem.alloc_bytes_non_zeroed(t.elem.size * scap, t.elem.align) or_return
-		defer if err != nil { mem.free_bytes(data) }
+		data := mem.alloc_bytes_non_zeroed(t.elem.size * scap, t.elem.align, allocator=allocator, loc=loc) or_return
+		defer if err != nil { mem.free_bytes(data, allocator=allocator, loc=loc) }
 
 		da := mem.Raw_Dynamic_Array{raw_data(data), 0, length, context.allocator }
 
@@ -489,7 +492,7 @@ _unmarshal_array :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 
 		if .Shrink_Excess in d.flags {
 			// Ignoring an error here, but this is not critical to succeed.
-			_ = runtime.__dynamic_array_shrink(&da, t.elem.size, t.elem.align, da.len)
+			_ = runtime.__dynamic_array_shrink(&da, t.elem.size, t.elem.align, da.len, loc=loc)
 		}
 
 		raw      := (^mem.Raw_Slice)(v.data)
@@ -500,8 +503,8 @@ _unmarshal_array :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 	case reflect.Type_Info_Dynamic_Array:
 		length, scap := err_conv(_decode_len_container(d, add)) or_return
 
-		data := mem.alloc_bytes_non_zeroed(t.elem.size * scap, t.elem.align) or_return
-		defer if err != nil { mem.free_bytes(data) }
+		data := mem.alloc_bytes_non_zeroed(t.elem.size * scap, t.elem.align, loc=loc) or_return
+		defer if err != nil { mem.free_bytes(data, allocator=allocator, loc=loc) }
 
 		raw           := (^mem.Raw_Dynamic_Array)(v.data)
 		raw.data       = raw_data(data) 
@@ -513,47 +516,41 @@ _unmarshal_array :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 
 		if .Shrink_Excess in d.flags {
 			// Ignoring an error here, but this is not critical to succeed.
-			_ = runtime.__dynamic_array_shrink(raw, t.elem.size, t.elem.align, raw.len)
+			_ = runtime.__dynamic_array_shrink(raw, t.elem.size, t.elem.align, raw.len, loc=loc)
 		}
 		return
 
 	case reflect.Type_Info_Array:
-		_, scap := err_conv(_decode_len_container(d, add)) or_return
-		length := min(scap, t.count)
-	
+		length, _ := err_conv(_decode_len_container(d, add)) or_return
 		if length > t.count {
 			return _unsupported(v, hdr)
 		}
 
-		da := mem.Raw_Dynamic_Array{rawptr(v.data), 0, length, context.allocator }
+		da := mem.Raw_Dynamic_Array{rawptr(v.data), 0, length, allocator }
 
 		out_of_space := assign_array(d, &da, t.elem, length, growable=false) or_return
 		if out_of_space { return _unsupported(v, hdr) }
 		return
 
 	case reflect.Type_Info_Enumerated_Array:
-		_, scap := err_conv(_decode_len_container(d, add)) or_return
-		length := min(scap, t.count)
-	
+		length, _ := err_conv(_decode_len_container(d, add)) or_return
 		if length > t.count {
 			return _unsupported(v, hdr)
 		}
 
-		da := mem.Raw_Dynamic_Array{rawptr(v.data), 0, length, context.allocator }
+		da := mem.Raw_Dynamic_Array{rawptr(v.data), 0, length, allocator }
 
 		out_of_space := assign_array(d, &da, t.elem, length, growable=false) or_return
 		if out_of_space { return _unsupported(v, hdr) }
 		return
 
 	case reflect.Type_Info_Complex:
-		_, scap := err_conv(_decode_len_container(d, add)) or_return
-		length := min(scap, 2)
-	
+		length, _ := err_conv(_decode_len_container(d, add)) or_return
 		if length > 2 {
 			return _unsupported(v, hdr)
 		}
 
-		da := mem.Raw_Dynamic_Array{rawptr(v.data), 0, 2, context.allocator }
+		da := mem.Raw_Dynamic_Array{rawptr(v.data), 0, 2, allocator }
 
 		info: ^runtime.Type_Info
 		switch ti.id {
@@ -568,14 +565,12 @@ _unmarshal_array :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 		return
 	
 	case reflect.Type_Info_Quaternion:
-		_, scap := err_conv(_decode_len_container(d, add)) or_return
-		length := min(scap, 4)
-	
+		length, _ := err_conv(_decode_len_container(d, add)) or_return
 		if length > 4 {
 			return _unsupported(v, hdr)
 		}
 
-		da := mem.Raw_Dynamic_Array{rawptr(v.data), 0, 4, context.allocator }
+		da := mem.Raw_Dynamic_Array{rawptr(v.data), 0, 4, allocator }
 
 		info: ^runtime.Type_Info
 		switch ti.id {
@@ -593,17 +588,17 @@ _unmarshal_array :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 	}
 }
 
-_unmarshal_map :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header, add: Add) -> (err: Unmarshal_Error) {
+_unmarshal_map :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header, add: Add, allocator := context.allocator, loc := #caller_location) -> (err: Unmarshal_Error) {
 	r := d.reader
-	decode_key :: proc(d: Decoder, v: any, allocator := context.allocator) -> (k: string, err: Unmarshal_Error) {
+	decode_key :: proc(d: Decoder, v: any, allocator := context.allocator, loc := #caller_location) -> (k: string, err: Unmarshal_Error) {
 		entry_hdr := _decode_header(d.reader) or_return
 		entry_maj, entry_add := _header_split(entry_hdr)
 		#partial switch entry_maj {
 		case .Text:
-			k = err_conv(_decode_text(d, entry_add, allocator)) or_return
+			k = err_conv(_decode_text(d, entry_add, allocator=allocator, loc=loc)) or_return
 			return
 		case .Bytes:
-			bytes := err_conv(_decode_bytes(d, entry_add, allocator=allocator)) or_return
+			bytes := err_conv(_decode_bytes(d, entry_add, allocator=allocator, loc=loc)) or_return
 			k = string(bytes)
 			return
 		case:
@@ -615,23 +610,23 @@ _unmarshal_map :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header,
 	// Allow generically storing the map array.
 	switch &dst in v {
 	case ^Map:
-		dst = err_conv(_decode_map_ptr(d, add)) or_return
+		dst = err_conv(_decode_map_ptr(d, add, allocator=allocator, loc=loc)) or_return
 		return
 	case Map:
-		dst = err_conv(_decode_map(d, add)) or_return
+		dst = err_conv(_decode_map(d, add, allocator=allocator, loc=loc)) or_return
 		return
 	}
 
 	#partial switch t in ti.variant {
 	case reflect.Type_Info_Struct:
-		if t.is_raw_union {
+		if .raw_union in t.flags {
 			return _unsupported(v, hdr)
 		}
 
 		length, _ := err_conv(_decode_len_container(d, add)) or_return
 		unknown := length == -1
 		fields := reflect.struct_fields_zipped(ti.id)
-	
+
 		for idx := 0; idx < len(fields) && (unknown || idx < length); idx += 1 {
 			// Decode key, keys can only be strings.
 			key: string
@@ -644,7 +639,7 @@ _unmarshal_map :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header,
 				key = keyv
 			}
 			defer delete(key, context.temp_allocator)
-			
+
 			// Find matching field.
 			use_field_idx := -1
 			{
@@ -754,7 +749,7 @@ _unmarshal_map :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header,
 // Unmarshal into a union, based on the `TAG_OBJECT_TYPE` tag of the spec, it denotes a tag which
 // contains an array of exactly two elements, the first is a textual representation of the following
 // CBOR value's type.
-_unmarshal_union :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header) -> (err: Unmarshal_Error) {
+_unmarshal_union :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header, loc := #caller_location) -> (err: Unmarshal_Error) {
 	r := d.reader
 	#partial switch t in ti.variant {
 	case reflect.Type_Info_Union:
@@ -792,7 +787,7 @@ _unmarshal_union :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 			case reflect.Type_Info_Named:
 				if vti.name == target_name {
 					reflect.set_union_variant_raw_tag(v, tag)
-					return _unmarshal_value(d, any{v.data, variant.id}, _decode_header(r) or_return)
+					return _unmarshal_value(d, any{v.data, variant.id}, _decode_header(r) or_return, loc=loc)
 				}
 
 			case:
@@ -804,7 +799,7 @@ _unmarshal_union :: proc(d: Decoder, v: any, ti: ^reflect.Type_Info, hdr: Header
 				
 				if variant_name == target_name {
 					reflect.set_union_variant_raw_tag(v, tag)
-					return _unmarshal_value(d, any{v.data, variant.id}, _decode_header(r) or_return)
+					return _unmarshal_value(d, any{v.data, variant.id}, _decode_header(r) or_return, loc=loc)
 				}
 			}
 		}

core/encoding/csv/example.odin

@@ -0,0 +1,88 @@
+//+build ignore
+package encoding_csv
+
+import "core:fmt"
+import "core:encoding/csv"
+import "core:os"
+
+// Requires keeping the entire CSV file in memory at once
+iterate_csv_from_string :: proc(filename: string) {
+	r: csv.Reader
+	r.trim_leading_space  = true
+	r.reuse_record        = true // Without it you have to delete(record)
+	r.reuse_record_buffer = true // Without it you have to each of the fields within it
+	defer csv.reader_destroy(&r)
+
+	if csv_data, ok := os.read_entire_file(filename); ok {
+		csv.reader_init_with_string(&r, string(csv_data))
+		defer delete(csv_data)
+	} else {
+		fmt.printfln("Unable to open file: %v", filename)
+		return
+	}
+
+	for r, i, err in csv.iterator_next(&r) {
+		if err != nil { /* Do something with error */ }
+		for f, j in r {
+			fmt.printfln("Record %v, field %v: %q", i, j, f)
+		}
+	}
+}
+
+// Reads the CSV as it's processed (with a small buffer)
+iterate_csv_from_stream :: proc(filename: string) {
+	fmt.printfln("Hellope from %v", filename)
+	r: csv.Reader
+	r.trim_leading_space  = true
+	r.reuse_record        = true // Without it you have to delete(record)
+	r.reuse_record_buffer = true // Without it you have to each of the fields within it
+	defer csv.reader_destroy(&r)
+
+	handle, errno := os.open(filename)
+	if errno != os.ERROR_NONE {
+		fmt.printfln("Error opening file: %v", filename)
+		return
+	}
+	defer os.close(handle)
+	csv.reader_init(&r, os.stream_from_handle(handle))
+
+	for r, i in csv.iterator_next(&r) {
+		for f, j in r {
+			fmt.printfln("Record %v, field %v: %q", i, j, f)
+		}
+	}
+	fmt.printfln("Error: %v", csv.iterator_last_error(r))
+}
+
+// Read all records at once
+read_csv_from_string :: proc(filename: string) {
+	r: csv.Reader
+	r.trim_leading_space  = true
+	r.reuse_record        = true // Without it you have to delete(record)
+	r.reuse_record_buffer = true // Without it you have to each of the fields within it
+	defer csv.reader_destroy(&r)
+
+	if csv_data, ok := os.read_entire_file(filename); ok {
+		csv.reader_init_with_string(&r, string(csv_data))
+		defer delete(csv_data)
+	} else {
+		fmt.printfln("Unable to open file: %v", filename)
+		return
+	}
+
+	records, err := csv.read_all(&r)
+	if err != nil { /* Do something with CSV parse error */ }
+
+	defer {
+		for rec in records {
+			delete(rec)
+		}
+		delete(records)
+	}
+
+	for r, i in records {
+		for f, j in r {
+			fmt.printfln("Record %v, field %v: %q", i, j, f)
+		}
+	}
+}

core/encoding/csv/reader.odin

@@ -57,6 +57,9 @@ Reader :: struct {
 	field_indices: [dynamic]int,
 	last_record:   [dynamic]string,
 	sr: strings.Reader, // used by reader_init_with_string
+
+	// Set and used by the iterator. Query using `iterator_last_error`
+	last_iterator_error: Error,
 }
 
 
@@ -121,6 +124,27 @@ reader_destroy :: proc(r: ^Reader) {
 	bufio.reader_destroy(&r.r)
 }
 
+/*
+	Returns a record at a time.
+
+	for record, row_idx in csv.iterator_next(&r) { ... }
+
+	TIP: If you process the results within the loop and don't need to own the results,
+	you can set the Reader's `reuse_record` and `reuse_record_reuse_record_buffer` to true;
+	you won't need to delete the record or its fields.
+*/
+iterator_next :: proc(r: ^Reader) -> (record: []string, idx: int, err: Error, more: bool) {
+	record, r.last_iterator_error = read(r)
+	return record, r.line_count - 1, r.last_iterator_error, r.last_iterator_error == nil
+}
+
+// Get last CSV parse error if we ignored it in the iterator loop
+//
+// for record, row_idx in csv.iterator_next(&r) { ... }
+iterator_last_error :: proc(r: Reader) -> (err: Error) {
+	return r.last_iterator_error
+}
+
 // read reads a single record (a slice of fields) from r
 //
 // All \r\n sequences are normalized to \n, including multi-line field
@@ -147,7 +171,7 @@ is_io_error :: proc(err: Error, io_err: io.Error) -> bool {
 
 // read_all reads all the remaining records from r.
 // Each record is a slice of fields.
-// read_all is defined to read until an EOF, and does not treat, and does not treat EOF as an error
+// read_all is defined to read until an EOF, and does not treat EOF as an error
 @(require_results)
 read_all :: proc(r: ^Reader, allocator := context.allocator) -> ([][]string, Error) {
 	context.allocator = allocator
@@ -460,5 +484,4 @@ _read_record :: proc(r: ^Reader, dst: ^[dynamic]string, allocator := context.all
 		r.fields_per_record = len(dst)
 	}
 	return dst[:], err
-
-}
+}

core/encoding/entity/entity.odin

@@ -56,38 +56,27 @@ CDATA_END     :: "]]>"
 COMMENT_START :: "<!--"
 COMMENT_END   :: "-->"
 
-/*
-	Default: CDATA and comments are passed through unchanged.
-*/
+// Default: CDATA and comments are passed through unchanged.
 XML_Decode_Option :: enum u8 {
-	/*
-		Do not decode & entities. It decodes by default.
-		If given, overrides `Decode_CDATA`.
-	*/
+	// Do not decode & entities. It decodes by default. If given, overrides `Decode_CDATA`.
 	No_Entity_Decode,
 
-	/*
-		CDATA is unboxed.
-	*/
+	// CDATA is unboxed.
 	Unbox_CDATA,
 
-	/*
-		Unboxed CDATA is decoded as well.
-		Ignored if `.Unbox_CDATA` is not given.
-	*/
+	// Unboxed CDATA is decoded as well. Ignored if `.Unbox_CDATA` is not given.
 	Decode_CDATA,
 
-	/*
-		Comments are stripped.
-	*/
+	// Comments are stripped.
 	Comment_Strip,
+
+	// Normalize whitespace
+	Normalize_Whitespace,
 }
 XML_Decode_Options :: bit_set[XML_Decode_Option; u8]
 
-/*
-	Decode a string that may include SGML/XML/HTML entities.
-	The caller has to free the result.
-*/
+// Decode a string that may include SGML/XML/HTML entities.
+// The caller has to free the result.
 decode_xml :: proc(input: string, options := XML_Decode_Options{}, allocator := context.allocator) -> (decoded: string, err: Error) {
 	context.allocator = allocator
 
@@ -100,14 +89,14 @@ decode_xml :: proc(input: string, options := XML_Decode_Options{}, allocator :=
 	t := Tokenizer{src=input}
 	in_data := false
 
+	prev: rune = ' '
+
 	loop: for {
 		advance(&t) or_return
 		if t.r < 0 { break loop }
 
-		/*
-			Below here we're never inside a CDATA tag.
-			At most we'll see the start of one, but that doesn't affect the logic.
-		*/
+		// Below here we're never inside a CDATA tag. At most we'll see the start of one,
+		// but that doesn't affect the logic.
 		switch t.r {
 		case '<':
 			/*
@@ -126,9 +115,7 @@ decode_xml :: proc(input: string, options := XML_Decode_Options{}, allocator :=
 			in_data = _handle_xml_special(&t, &builder, options) or_return
 
 		case ']':
-			/*
-				If we're unboxing _and_ decoding CDATA, we'll have to check for the end tag.
-			*/
+			// If we're unboxing _and_ decoding CDATA, we'll have to check for the end tag.
 			if in_data {
 				if t.read_offset + len(CDATA_END) < len(t.src) {
 					if string(t.src[t.offset:][:len(CDATA_END)]) == CDATA_END {
@@ -143,22 +130,16 @@ decode_xml :: proc(input: string, options := XML_Decode_Options{}, allocator :=
 
 		case:
 			if in_data && .Decode_CDATA not_in options {
-				/*
-					Unboxed, but undecoded.
-				*/
+				// Unboxed, but undecoded.
 				write_rune(&builder, t.r)
 				continue
 			}
 
 			if t.r == '&' {
 				if entity, entity_err := _extract_xml_entity(&t); entity_err != .None {
-					/*
-						We read to the end of the string without closing the entity.
-						Pass through as-is.
-					*/
+					// We read to the end of the string without closing the entity. Pass through as-is.
 					write_string(&builder, entity)
 				} else {
-
 					if .No_Entity_Decode not_in options {
 						if decoded, ok := xml_decode_entity(entity); ok {
 							write_rune(&builder, decoded)
@@ -166,19 +147,41 @@ decode_xml :: proc(input: string, options := XML_Decode_Options{}, allocator :=
 						}
 					}
 
-					/*
-						Literal passthrough because the decode failed or we want entities not decoded.
-					*/
+					// Literal passthrough because the decode failed or we want entities not decoded.
 					write_string(&builder, "&")
 					write_string(&builder, entity)
 					write_string(&builder, ";")
 				}
 			} else {
-				write_rune(&builder, t.r)
+				// Handle AV Normalization: https://www.w3.org/TR/2006/REC-xml11-20060816/#AVNormalize
+				if .Normalize_Whitespace in options {
+					switch t.r {
+					case ' ', '\r', '\n', '\t':
+						if prev != ' ' {
+							write_rune(&builder, ' ')
+							prev = ' '
+						}
+					case:
+						write_rune(&builder, t.r)
+						prev = t.r
+					}
+				} else {
+					// https://www.w3.org/TR/2006/REC-xml11-20060816/#sec-line-ends
+					switch t.r {
+					case '\n', 0x85, 0x2028:
+						write_rune(&builder, '\n')
+					case '\r': // Do nothing until next character
+					case:
+						if prev == '\r' { // Turn a single carriage return into a \n
+							write_rune(&builder, '\n')
+						}
+						write_rune(&builder, t.r)
+					}
+					prev = t.r
+				}
 			}
 		}
 	}
-
 	return strings.clone(strings.to_string(builder), allocator), err
 }
 
@@ -253,24 +256,18 @@ xml_decode_entity :: proc(entity: string) -> (decoded: rune, ok: bool) {
 		return rune(val), true
 
 	case:
-		/*
-			Named entity.
-		*/
+		// Named entity.
 		return named_xml_entity_to_rune(entity)
 	}
 }
 
-/*
-	Private XML helper to extract `&<stuff>;` entity.
-*/
+// Private XML helper to extract `&<stuff>;` entity.
 @(private="file")
 _extract_xml_entity :: proc(t: ^Tokenizer) -> (entity: string, err: Error) {
 	assert(t != nil && t.r == '&')
 
-	/*
-		All of these would be in the ASCII range.
-		Even if one is not, it doesn't matter. All characters we need to compare to extract are.
-	*/
+	// All of these would be in the ASCII range.
+	// Even if one is not, it doesn't matter. All characters we need to compare to extract are.
 
 	length := len(t.src)
 	found  := false
@@ -292,9 +289,7 @@ _extract_xml_entity :: proc(t: ^Tokenizer) -> (entity: string, err: Error) {
 	return string(t.src[t.offset : t.read_offset]), .Invalid_Entity_Encoding
 }
 
-/*
-	Private XML helper for CDATA and comments.
-*/
+// Private XML helper for CDATA and comments.
 @(private="file")
 _handle_xml_special :: proc(t: ^Tokenizer, builder: ^strings.Builder, options: XML_Decode_Options) -> (in_data: bool, err: Error) {
 	assert(t != nil && t.r == '<')
@@ -304,20 +299,14 @@ _handle_xml_special :: proc(t: ^Tokenizer, builder: ^strings.Builder, options: X
 		t.read_offset += len(CDATA_START) - 1
 
 		if .Unbox_CDATA in options && .Decode_CDATA in options {
-			/*
-				We're unboxing _and_ decoding CDATA
-			*/
+			// We're unboxing _and_ decoding CDATA
 			return true, .None
 		}
 
-		/*
-			CDATA is passed through.
-		*/
+		// CDATA is passed through.
 		offset := t.offset
 
-		/*
-			Scan until end of CDATA.
-		*/
+		// Scan until end of CDATA.
 		for {
 			advance(t) or_return
 			if t.r < 0 { return true, .CDATA_Not_Terminated }
@@ -341,14 +330,10 @@ _handle_xml_special :: proc(t: ^Tokenizer, builder: ^strings.Builder, options: X
 
 	} else if string(t.src[t.offset:][:len(COMMENT_START)]) == COMMENT_START {
 		t.read_offset += len(COMMENT_START)
-		/*
-			Comment is passed through by default.
-		*/
+		// Comment is passed through by default.
 		offset := t.offset
 
-		/*
-			Scan until end of Comment.
-		*/
+		// Scan until end of Comment.
 		for {
 			advance(t) or_return
 			if t.r < 0 { return true, .Comment_Not_Terminated }

core/encoding/hex/hex.odin

@@ -2,8 +2,8 @@ package encoding_hex
 
 import "core:strings"
 
-encode :: proc(src: []byte, allocator := context.allocator) -> []byte #no_bounds_check {
-	dst := make([]byte, len(src) * 2, allocator)
+encode :: proc(src: []byte, allocator := context.allocator, loc := #caller_location) -> []byte #no_bounds_check {
+	dst := make([]byte, len(src) * 2, allocator, loc)
 	for i, j := 0, 0; i < len(src); i += 1 {
 		v := src[i]
 		dst[j]   = HEXTABLE[v>>4]
@@ -15,12 +15,12 @@ encode :: proc(src: []byte, allocator := context.allocator) -> []byte #no_bounds
 }
 
 
-decode :: proc(src: []byte, allocator := context.allocator) -> (dst: []byte, ok: bool) #no_bounds_check {
+decode :: proc(src: []byte, allocator := context.allocator, loc := #caller_location) -> (dst: []byte, ok: bool) #no_bounds_check {
 	if len(src) % 2 == 1 {
 		return
 	}
 
-	dst = make([]byte, len(src) / 2, allocator)
+	dst = make([]byte, len(src) / 2, allocator, loc)
 	for i, j := 0, 1; j < len(src); j += 2 {
 		p := src[j-1]
 		q := src[j]
@@ -69,5 +69,4 @@ hex_digit :: proc(char: byte) -> (u8, bool) {
 	case 'A' ..= 'F': return char - 'A' + 10, true
 	case:             return 0, false
 	}
-}
-
+}

core/encoding/hxa/hxa.odin

@@ -160,34 +160,35 @@ CONVENTION_SOFT_TRANSFORM :: "transform"
 
 /* destroy procedures */
 
-meta_destroy :: proc(meta: Meta, allocator := context.allocator) {
+meta_destroy :: proc(meta: Meta, allocator := context.allocator, loc := #caller_location) {
 	if nested, ok := meta.value.([]Meta); ok {
 		for m in nested {
-			meta_destroy(m)
+			meta_destroy(m, loc=loc)
 		}
-		delete(nested, allocator)
+		delete(nested, allocator, loc=loc)
 	}
 }
-nodes_destroy :: proc(nodes: []Node, allocator := context.allocator) {
+nodes_destroy :: proc(nodes: []Node, allocator := context.allocator, loc := #caller_location) {
 	for node in nodes {
 		for meta in node.meta_data {
-			meta_destroy(meta)
+			meta_destroy(meta, loc=loc)
 		}
-		delete(node.meta_data, allocator)
+		delete(node.meta_data, allocator, loc=loc)
 
 		switch n in node.content {
 		case Node_Geometry:
-			delete(n.corner_stack, allocator)
-			delete(n.edge_stack, allocator)
-			delete(n.face_stack, allocator)
+			delete(n.corner_stack, allocator, loc=loc)
+			delete(n.vertex_stack, allocator, loc=loc)
+			delete(n.edge_stack,   allocator, loc=loc)
+			delete(n.face_stack,   allocator, loc=loc)
 		case Node_Image:
-			delete(n.image_stack, allocator)
+			delete(n.image_stack,  allocator, loc=loc)
 		}
 	}
-	delete(nodes, allocator)
+	delete(nodes, allocator, loc=loc)
 }
 
-file_destroy :: proc(file: File) {
-	nodes_destroy(file.nodes, file.allocator)
-	delete(file.backing, file.allocator)
-}
+file_destroy :: proc(file: File, loc := #caller_location) {
+	nodes_destroy(file.nodes, file.allocator, loc=loc)
+	delete(file.backing, file.allocator, loc=loc)
+}

core/encoding/hxa/read.odin

@@ -11,24 +11,21 @@ Read_Error :: enum {
 	Unable_To_Read_File,
 }
 
-read_from_file :: proc(filename: string, print_error := false, allocator := context.allocator) -> (file: File, err: Read_Error) {
+read_from_file :: proc(filename: string, print_error := false, allocator := context.allocator, loc := #caller_location) -> (file: File, err: Read_Error) {
 	context.allocator = allocator
 
-	data, ok := os.read_entire_file(filename)
+	data, ok := os.read_entire_file(filename, allocator, loc)
 	if !ok {
 		err = .Unable_To_Read_File
+		delete(data, allocator, loc)
 		return
 	}
-	defer if !ok {
-		delete(data)
-	} else {
-		file.backing = data
-	}
-	file, err = read(data, filename, print_error, allocator)
+	file, err = read(data, filename, print_error, allocator, loc)
+	file.backing   = data
 	return
 }
 
-read :: proc(data: []byte, filename := "<input>", print_error := false, allocator := context.allocator) -> (file: File, err: Read_Error) {
+read :: proc(data: []byte, filename := "<input>", print_error := false, allocator := context.allocator, loc := #caller_location) -> (file: File, err: Read_Error) {
 	Reader :: struct {
 		filename:    string,
 		data:        []byte,
@@ -79,8 +76,8 @@ read :: proc(data: []byte, filename := "<input>", print_error := false, allocato
 		return string(data[:len]), nil
 	}
 
-	read_meta :: proc(r: ^Reader, capacity: u32le) -> (meta_data: []Meta, err: Read_Error) {
-		meta_data = make([]Meta, int(capacity))
+	read_meta :: proc(r: ^Reader, capacity: u32le, allocator := context.allocator, loc := #caller_location) -> (meta_data: []Meta, err: Read_Error) {
+		meta_data = make([]Meta, int(capacity), allocator=allocator)
 		count := 0
 		defer meta_data = meta_data[:count]
 		for &m in meta_data {
@@ -111,10 +108,10 @@ read :: proc(data: []byte, filename := "<input>", print_error := false, allocato
 		return
 	}
 
-	read_layer_stack :: proc(r: ^Reader, capacity: u32le) -> (layers: Layer_Stack, err: Read_Error) {
+	read_layer_stack :: proc(r: ^Reader, capacity: u32le, allocator := context.allocator, loc := #caller_location) -> (layers: Layer_Stack, err: Read_Error) {
 		stack_count := read_value(r, u32le) or_return
 		layer_count := 0
-		layers = make(Layer_Stack, stack_count)
+		layers = make(Layer_Stack, stack_count, allocator=allocator, loc=loc)
 		defer layers = layers[:layer_count]
 		for &layer in layers {
 			layer.name = read_name(r) or_return
@@ -170,7 +167,8 @@ read :: proc(data: []byte, filename := "<input>", print_error := false, allocato
 
 	node_count := 0
 	file.header = header^
-	file.nodes = make([]Node, header.internal_node_count)
+	file.nodes = make([]Node, header.internal_node_count, allocator=allocator, loc=loc)
+	file.allocator = allocator
 	defer if err != nil {
 		nodes_destroy(file.nodes)
 		file.nodes = nil
@@ -198,15 +196,15 @@ read :: proc(data: []byte, filename := "<input>", print_error := false, allocato
 		case .Geometry:
 			g: Node_Geometry
 
-			g.vertex_count      = read_value(r, u32le)                     or_return
-			g.vertex_stack      = read_layer_stack(r, g.vertex_count)      or_return
-			g.edge_corner_count = read_value(r, u32le)                     or_return
-			g.corner_stack      = read_layer_stack(r, g.edge_corner_count) or_return
+			g.vertex_count      = read_value(r, u32le)                               or_return
+			g.vertex_stack      = read_layer_stack(r, g.vertex_count, loc=loc)       or_return
+			g.edge_corner_count = read_value(r, u32le)                               or_return
+			g.corner_stack      = read_layer_stack(r, g.edge_corner_count, loc=loc)  or_return
 			if header.version > 2 {
-				g.edge_stack = read_layer_stack(r, g.edge_corner_count) or_return
+				g.edge_stack = read_layer_stack(r, g.edge_corner_count, loc=loc) or_return
 			}
-			g.face_count = read_value(r, u32le)              or_return
-			g.face_stack = read_layer_stack(r, g.face_count) or_return
+			g.face_count = read_value(r, u32le)                       or_return
+			g.face_stack = read_layer_stack(r, g.face_count, loc=loc) or_return
 
 			node.content = g
 
@@ -233,4 +231,4 @@ read :: proc(data: []byte, filename := "<input>", print_error := false, allocato
 	}
 
 	return
-}
+}

core/encoding/ini/ini.odin

@@ -0,0 +1,191 @@
+package encoding_ini
+
+import "base:runtime"
+import "base:intrinsics"
+import "core:strings"
+import "core:strconv"
+import "core:io"
+import "core:os"
+import "core:fmt"
+_ :: fmt
+
+Options :: struct {
+	comment: string,
+	key_lower_case: bool,
+}
+
+DEFAULT_OPTIONS :: Options {
+	comment = ";",
+	key_lower_case = false,
+}
+
+Iterator :: struct {
+	section: string,
+	_src:    string,
+	options: Options,
+}
+
+iterator_from_string :: proc(src: string, options := DEFAULT_OPTIONS) -> Iterator {
+	return {
+		section = "",
+		options = options,
+		_src = src,
+	}
+}
+
+
+// Returns the raw `key` and `value`. `ok` will be false if no more key=value pairs cannot be found.
+// They key and value may be quoted, which may require the use of `strconv.unquote_string`.
+iterate :: proc(it: ^Iterator) -> (key, value: string, ok: bool) {
+	for line_ in strings.split_lines_iterator(&it._src) {
+		line := strings.trim_space(line_)
+
+		if len(line) == 0 {
+			continue
+		}
+
+		if line[0] == '[' {
+			end_idx := strings.index_byte(line, ']')
+			if end_idx < 0 {
+				end_idx = len(line)
+			}
+			it.section = line[1:end_idx]
+			continue
+		}
+
+		if it.options.comment != "" && strings.has_prefix(line, it.options.comment) {
+			continue
+		}
+
+		equal := strings.index(line, " =") // check for things keys that `ctrl+= = zoom_in`
+		quote := strings.index_byte(line, '"')
+		if equal < 0 || quote > 0 && quote < equal {
+			equal = strings.index_byte(line, '=')
+			if equal < 0 {
+				continue
+			}
+		} else {
+			equal += 1
+		}
+
+		key = strings.trim_space(line[:equal])
+		value = strings.trim_space(line[equal+1:])
+		ok = true
+		return
+	}
+
+	it.section = ""
+	return
+}
+
+Map :: distinct map[string]map[string]string
+
+load_map_from_string :: proc(src: string, allocator: runtime.Allocator, options := DEFAULT_OPTIONS) -> (m: Map, err: runtime.Allocator_Error) {
+	unquote :: proc(val: string) -> (string, runtime.Allocator_Error) {
+		if len(val) > 0 && (val[0] == '"' || val[0] == '\'') {
+			v, allocated, ok := strconv.unquote_string(val)
+			if !ok {
+				return strings.clone(val)
+			}
+			if allocated {
+				return v, nil
+			}
+		}
+		return strings.clone(val)
+	}
+
+	context.allocator = allocator
+
+	it := iterator_from_string(src, options)
+
+	for key, value in iterate(&it) {
+		section := it.section
+		if section not_in m {
+			section = strings.clone(section) or_return
+			m[section] = {}
+		}
+
+		// store key-value pair
+		pairs := &m[section]
+		new_key := unquote(key) or_return
+		if options.key_lower_case {
+			old_key := new_key
+			new_key = strings.to_lower(key) or_return
+			delete(old_key) or_return
+		}
+		pairs[new_key], err = unquote(value) or_return
+	}
+	return
+}
+
+load_map_from_path :: proc(path: string, allocator: runtime.Allocator, options := DEFAULT_OPTIONS) -> (m: Map, err: runtime.Allocator_Error, ok: bool) {
+	data := os.read_entire_file(path, allocator) or_return
+	defer delete(data, allocator)
+	m, err = load_map_from_string(string(data), allocator, options)
+	ok = err == nil
+	defer if !ok {
+		delete_map(m)
+	}
+	return
+}
+
+save_map_to_string :: proc(m: Map, allocator: runtime.Allocator) -> (data: string) {
+	b := strings.builder_make(allocator)
+	_, _ = write_map(strings.to_writer(&b), m)
+	return strings.to_string(b)
+}
+
+delete_map :: proc(m: Map) {
+	allocator := m.allocator
+	for section, pairs in m {
+		for key, value in pairs {
+			delete(key, allocator)
+			delete(value, allocator)
+		}
+		delete(section)
+		delete(pairs)
+	}
+	delete(m)
+}
+
+write_section :: proc(w: io.Writer, name: string, n_written: ^int = nil) -> (n: int, err: io.Error) {
+	defer if n_written != nil { n_written^ += n }
+	io.write_byte  (w, '[',  &n) or_return
+	io.write_string(w, name, &n) or_return
+	io.write_byte  (w, ']',  &n) or_return
+	return
+}
+
+write_pair :: proc(w: io.Writer, key: string, value: $T, n_written: ^int = nil) -> (n: int, err: io.Error) {
+	defer if n_written != nil { n_written^ += n }
+	io.write_string(w, key,   &n) or_return
+	io.write_string(w, " = ", &n) or_return
+	when intrinsics.type_is_string(T) {
+		val := string(value)
+		if len(val) > 0 && (val[0] == ' ' || val[len(val)-1] == ' ') {
+			io.write_quoted_string(w, val, n_written=&n) or_return
+		} else {
+			io.write_string(w, val, &n) or_return
+		}
+	} else {
+		n += fmt.wprint(w, value)
+	}
+	io.write_byte(w, '\n', &n) or_return
+	return
+}
+
+write_map :: proc(w: io.Writer, m: Map) -> (n: int, err: io.Error) {
+	section_index := 0
+	for section, pairs in m {
+		if section_index == 0 && section == "" {
+			// ignore section
+		} else {
+			write_section(w, section, &n) or_return
+		}
+		for key, value in pairs {
+			write_pair(w, key, value, &n) or_return
+		}
+		section_index += 1
+	}
+	return
+}

core/encoding/json/marshal.odin

@@ -25,7 +25,7 @@ Marshal_Options :: struct {
 	spec: Specification,
 
 	// Use line breaks & tabs/spaces
-	pretty: bool, 
+	pretty: bool,
 
 	// Use spaces for indentation instead of tabs
 	use_spaces: bool,
@@ -34,7 +34,7 @@ Marshal_Options :: struct {
 	spaces: int,
 
 	// Output uint as hex in JSON5 & MJSON
-	write_uint_as_hex: bool, 
+	write_uint_as_hex: bool,
 
 	// If spec is MJSON and this is true, then keys will be quoted.
 	//
@@ -62,8 +62,8 @@ Marshal_Options :: struct {
 	mjson_skipped_first_braces_end: bool,
 }
 
-marshal :: proc(v: any, opt: Marshal_Options = {}, allocator := context.allocator) -> (data: []byte, err: Marshal_Error) {
-	b := strings.builder_make(allocator)
+marshal :: proc(v: any, opt: Marshal_Options = {}, allocator := context.allocator, loc := #caller_location) -> (data: []byte, err: Marshal_Error) {
+	b := strings.builder_make(allocator, loc)
 	defer if err != nil {
 		strings.builder_destroy(&b)
 	}
@@ -100,45 +100,14 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 
 	case runtime.Type_Info_Integer:
 		buf: [40]byte
-		u: u128
-		switch i in a {
-		case i8:      u = u128(i)
-		case i16:     u = u128(i)
-		case i32:     u = u128(i)
-		case i64:     u = u128(i)
-		case i128:    u = u128(i)
-		case int:     u = u128(i)
-		case u8:      u = u128(i)
-		case u16:     u = u128(i)
-		case u32:     u = u128(i)
-		case u64:     u = u128(i)
-		case u128:    u = u128(i)
-		case uint:    u = u128(i)
-		case uintptr: u = u128(i)
-
-		case i16le:  u = u128(i)
-		case i32le:  u = u128(i)
-		case i64le:  u = u128(i)
-		case u16le:  u = u128(i)
-		case u32le:  u = u128(i)
-		case u64le:  u = u128(i)
-		case u128le: u = u128(i)
-
-		case i16be:  u = u128(i)
-		case i32be:  u = u128(i)
-		case i64be:  u = u128(i)
-		case u16be:  u = u128(i)
-		case u32be:  u = u128(i)
-		case u64be:  u = u128(i)
-		case u128be: u = u128(i)
-		}
+		u := cast_any_int_to_u128(a)
 
 		s: string
 
 		// allow uints to be printed as hex
 		if opt.write_uint_as_hex && (opt.spec == .JSON5 || opt.spec == .MJSON) {
 			switch i in a {
-			case u8, u16, u32, u64, u128: 
+			case u8, u16, u32, u64, u128:
 				s = strconv.append_bits_128(buf[:], u, 16, info.signed, 8*ti.size, "0123456789abcdef", { .Prefix })
 
 			case:
@@ -239,7 +208,7 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 	case runtime.Type_Info_Array:
 		opt_write_start(w, opt, '[') or_return
 		for i in 0..<info.count {
-			opt_write_iteration(w, opt, i) or_return
+			opt_write_iteration(w, opt, i == 0) or_return
 			data := uintptr(v.data) + uintptr(i*info.elem_size)
 			marshal_to_writer(w, any{rawptr(data), info.elem.id}, opt) or_return
 		}
@@ -248,7 +217,7 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 	case runtime.Type_Info_Enumerated_Array:
 		opt_write_start(w, opt, '[') or_return
 		for i in 0..<info.count {
-			opt_write_iteration(w, opt, i) or_return
+			opt_write_iteration(w, opt, i == 0) or_return
 			data := uintptr(v.data) + uintptr(i*info.elem_size)
 			marshal_to_writer(w, any{rawptr(data), info.elem.id}, opt) or_return
 		}
@@ -258,7 +227,7 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 		opt_write_start(w, opt, '[') or_return
 		array := cast(^mem.Raw_Dynamic_Array)v.data
 		for i in 0..<array.len {
-			opt_write_iteration(w, opt, i) or_return
+			opt_write_iteration(w, opt, i == 0) or_return
 			data := uintptr(array.data) + uintptr(i*info.elem_size)
 			marshal_to_writer(w, any{rawptr(data), info.elem.id}, opt) or_return
 		}
@@ -268,7 +237,7 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 		opt_write_start(w, opt, '[') or_return
 		slice := cast(^mem.Raw_Slice)v.data
 		for i in 0..<slice.len {
-			opt_write_iteration(w, opt, i) or_return
+			opt_write_iteration(w, opt, i == 0) or_return
 			data := uintptr(slice.data) + uintptr(i*info.elem_size)
 			marshal_to_writer(w, any{rawptr(data), info.elem.id}, opt) or_return
 		}
@@ -290,7 +259,7 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 				for bucket_index in 0..<map_cap {
 					runtime.map_hash_is_valid(hs[bucket_index]) or_continue
 
-					opt_write_iteration(w, opt, i) or_return
+					opt_write_iteration(w, opt, i == 0) or_return
 					i += 1
 
 					key   := rawptr(runtime.map_cell_index_dynamic(ks, info.map_info.ks, bucket_index))
@@ -310,7 +279,12 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 							case cstring: name = string(s)
 							}
 							opt_write_key(w, opt, name) or_return
-
+						case runtime.Type_Info_Integer:
+							buf: [40]byte
+							u := cast_any_int_to_u128(ka)
+							name = strconv.append_bits_128(buf[:], u, 10, info.signed, 8*kti.size, "0123456789", nil)
+							
+							opt_write_key(w, opt, name) or_return
 						case: return .Unsupported_Type
 						}
 					}
@@ -356,7 +330,7 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 				slice.sort_by(sorted[:], proc(i, j: Entry) -> bool { return i.key < j.key })
 
 				for s, i in sorted {
-					opt_write_iteration(w, opt, i) or_return
+					opt_write_iteration(w, opt, i == 0) or_return
 					opt_write_key(w, opt, s.key) or_return
 					marshal_to_writer(w, s.value, opt) or_return
 				}
@@ -387,17 +361,17 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 			case runtime.Type_Info_Pointer,
 			     runtime.Type_Info_Multi_Pointer,
 			     runtime.Type_Info_Procedure:
-			     	return (^rawptr)(v.data)^ == nil
+				return (^rawptr)(v.data)^ == nil
 			case runtime.Type_Info_Dynamic_Array:
-			     	return (^runtime.Raw_Dynamic_Array)(v.data).len == 0
+				return (^runtime.Raw_Dynamic_Array)(v.data).len == 0
 			case runtime.Type_Info_Slice:
-			     	return (^runtime.Raw_Slice)(v.data).len == 0
+				return (^runtime.Raw_Slice)(v.data).len == 0
 			case runtime.Type_Info_Union,
 			     runtime.Type_Info_Bit_Set,
 			     runtime.Type_Info_Soa_Pointer:
 				return reflect.is_nil(v)
 			case runtime.Type_Info_Map:
-			     	return (^runtime.Raw_Map)(v.data).len == 0
+				return (^runtime.Raw_Map)(v.data).len == 0
 			}
 			return false
 		}
@@ -405,10 +379,16 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 		marshal_struct_fields :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err: Marshal_Error) {
 			ti := runtime.type_info_base(type_info_of(v.id))
 			info := ti.variant.(runtime.Type_Info_Struct)
-			for name, i in info.names {
+			first_iteration := true
+			for name, i in info.names[:info.field_count] {
 				omitempty := false
 
 				json_name, extra := json_name_from_tag_value(reflect.struct_tag_get(reflect.Struct_Tag(info.tags[i]), "json"))
+
+				if json_name == "-" {
+					continue
+				}
+
 				for flag in strings.split_iterator(&extra, ",") {
 					switch flag {
 					case "omitempty":
@@ -420,11 +400,12 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 				data := rawptr(uintptr(v.data) + info.offsets[i])
 				the_value := any{data, id}
 
-				if is_omitempty(the_value) {
+				if omitempty && is_omitempty(the_value) {
 					continue
 				}
 
-				opt_write_iteration(w, opt, i) or_return
+				opt_write_iteration(w, opt, first_iteration) or_return
+				first_iteration = false
 				if json_name != "" {
 					opt_write_key(w, opt, json_name) or_return
 				} else {
@@ -469,12 +450,15 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 		case: panic("Invalid union tag type")
 		}
 
-		if v.data == nil || tag == 0 {
-			io.write_string(w, "null") or_return
-		} else {
-			id := info.variants[tag-1].id
-			return marshal_to_writer(w, any{v.data, id}, opt)
+		if !info.no_nil {
+			if tag == 0 {
+				io.write_string(w, "null") or_return
+				return nil
+			}
+			tag -= 1
 		}
+		id := info.variants[tag].id
+		return marshal_to_writer(w, any{v.data, id}, opt)
 
 	case runtime.Type_Info_Enum:
 		if !opt.use_enum_names || len(info.names) == 0 {
@@ -536,8 +520,6 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 		case: panic("unknown bit_size size")
 		}
 		io.write_u64(w, bit_data) or_return
-
-		return .Unsupported_Type
 	}
 
 	return
@@ -587,10 +569,10 @@ opt_write_start :: proc(w: io.Writer, opt: ^Marshal_Options, c: byte) -> (err: i
 }
 
 // insert comma separation and write indentations
-opt_write_iteration :: proc(w: io.Writer, opt: ^Marshal_Options, iteration: int) -> (err: io.Error) {
+opt_write_iteration :: proc(w: io.Writer, opt: ^Marshal_Options, first_iteration: bool) -> (err: io.Error) {
 	switch opt.spec {
-	case .JSON, .JSON5: 
-		if iteration > 0 {
+	case .JSON, .JSON5:
+		if !first_iteration {
 			io.write_byte(w, ',') or_return
 
 			if opt.pretty {
@@ -600,8 +582,8 @@ opt_write_iteration :: proc(w: io.Writer, opt: ^Marshal_Options, iteration: int)
 
 		opt_write_indentation(w, opt) or_return
 
-	case .MJSON: 
-		if iteration > 0 {
+	case .MJSON:
+		if !first_iteration {
 			// on pretty no commas necessary
 			if opt.pretty {
 				io.write_byte(w, '\n') or_return
@@ -654,3 +636,41 @@ opt_write_indentation :: proc(w: io.Writer, opt: ^Marshal_Options) -> (err: io.E
 
 	return
 }
+
+@(private)
+cast_any_int_to_u128 :: proc(any_int_value: any) -> u128 {
+	u: u128 = 0
+	switch i in any_int_value {
+	case i8:      u = u128(i)
+	case i16:     u = u128(i)
+	case i32:     u = u128(i)
+	case i64:     u = u128(i)
+	case i128:    u = u128(i)
+	case int:     u = u128(i)
+	case u8:      u = u128(i)
+	case u16:     u = u128(i)
+	case u32:     u = u128(i)
+	case u64:     u = u128(i)
+	case u128:    u = u128(i)
+	case uint:    u = u128(i)
+	case uintptr: u = u128(i)
+
+	case i16le:  u = u128(i)
+	case i32le:  u = u128(i)
+	case i64le:  u = u128(i)
+	case u16le:  u = u128(i)
+	case u32le:  u = u128(i)
+	case u64le:  u = u128(i)
+	case u128le: u = u128(i)
+
+	case i16be:  u = u128(i)
+	case i32be:  u = u128(i)
+	case i64be:  u = u128(i)
+	case u16be:  u = u128(i)
+	case u32be:  u = u128(i)
+	case u64be:  u = u128(i)
+	case u128be: u = u128(i)
+	}
+
+	return u
+}

core/encoding/json/parser.odin

@@ -28,27 +28,27 @@ make_parser_from_string :: proc(data: string, spec := DEFAULT_SPECIFICATION, par
 }
 
 
-parse :: proc(data: []byte, spec := DEFAULT_SPECIFICATION, parse_integers := false, allocator := context.allocator) -> (Value, Error) {
-	return parse_string(string(data), spec, parse_integers, allocator)
+parse :: proc(data: []byte, spec := DEFAULT_SPECIFICATION, parse_integers := false, allocator := context.allocator, loc := #caller_location) -> (Value, Error) {
+	return parse_string(string(data), spec, parse_integers, allocator, loc)
 }
 
-parse_string :: proc(data: string, spec := DEFAULT_SPECIFICATION, parse_integers := false, allocator := context.allocator) -> (Value, Error) {
+parse_string :: proc(data: string, spec := DEFAULT_SPECIFICATION, parse_integers := false, allocator := context.allocator, loc := #caller_location) -> (Value, Error) {
 	context.allocator = allocator
 	p := make_parser_from_string(data, spec, parse_integers, allocator)
 
 	switch p.spec {
 	case .JSON:
-		return parse_object(&p)
+		return parse_object(&p, loc)
 	case .JSON5:
-		return parse_value(&p)
+		return parse_value(&p, loc)
 	case .SJSON:
 		#partial switch p.curr_token.kind {
 		case .Ident, .String:
-			return parse_object_body(&p, .EOF)
+			return parse_object_body(&p, .EOF, loc)
 		}
-		return parse_value(&p)
+		return parse_value(&p, loc)
 	}
-	return parse_object(&p)
+	return parse_object(&p, loc)
 }
 
 token_end_pos :: proc(tok: Token) -> Pos {
@@ -106,7 +106,7 @@ parse_comma :: proc(p: ^Parser) -> (do_break: bool) {
 	return false
 }
 
-parse_value :: proc(p: ^Parser) -> (value: Value, err: Error) {
+parse_value :: proc(p: ^Parser, loc := #caller_location) -> (value: Value, err: Error) {
 	err = .None
 	token := p.curr_token
 	#partial switch token.kind {
@@ -142,13 +142,13 @@ parse_value :: proc(p: ^Parser) -> (value: Value, err: Error) {
 		
 	case .String:
 		advance_token(p)
-		return unquote_string(token, p.spec, p.allocator)
+		return unquote_string(token, p.spec, p.allocator, loc)
 
 	case .Open_Brace:
-		return parse_object(p)
+		return parse_object(p, loc)
 
 	case .Open_Bracket:
-		return parse_array(p)
+		return parse_array(p, loc)
 
 	case:
 		if p.spec != .JSON {
@@ -176,7 +176,7 @@ parse_value :: proc(p: ^Parser) -> (value: Value, err: Error) {
 	return
 }
 
-parse_array :: proc(p: ^Parser) -> (value: Value, err: Error) {
+parse_array :: proc(p: ^Parser, loc := #caller_location) -> (value: Value, err: Error) {
 	err = .None
 	expect_token(p, .Open_Bracket) or_return
 
@@ -184,14 +184,14 @@ parse_array :: proc(p: ^Parser) -> (value: Value, err: Error) {
 	array.allocator = p.allocator
 	defer if err != nil {
 		for elem in array {
-			destroy_value(elem)
+			destroy_value(elem, loc=loc)
 		}
-		delete(array)
+		delete(array, loc)
 	}
 
 	for p.curr_token.kind != .Close_Bracket {
-		elem := parse_value(p) or_return
-		append(&array, elem)
+		elem := parse_value(p, loc) or_return
+		append(&array, elem, loc)
 		
 		if parse_comma(p) {
 			break
@@ -228,38 +228,39 @@ clone_string :: proc(s: string, allocator: mem.Allocator, loc := #caller_locatio
 	return
 }
 
-parse_object_key :: proc(p: ^Parser, key_allocator: mem.Allocator) -> (key: string, err: Error) {
+parse_object_key :: proc(p: ^Parser, key_allocator: mem.Allocator, loc := #caller_location) -> (key: string, err: Error) {
 	tok := p.curr_token
 	if p.spec != .JSON {
 		if allow_token(p, .Ident) {
-			return clone_string(tok.text, key_allocator)
+			return clone_string(tok.text, key_allocator, loc)
 		}
 	}
 	if tok_err := expect_token(p, .String); tok_err != nil {
 		err = .Expected_String_For_Object_Key
 		return
 	}
-	return unquote_string(tok, p.spec, key_allocator)
+	return unquote_string(tok, p.spec, key_allocator, loc)
 }
 
-parse_object_body :: proc(p: ^Parser, end_token: Token_Kind) -> (obj: Object, err: Error) {
-	obj.allocator = p.allocator
+parse_object_body :: proc(p: ^Parser, end_token: Token_Kind, loc := #caller_location) -> (obj: Object, err: Error) {
+	obj = make(Object, allocator=p.allocator, loc=loc)
+
 	defer if err != nil {
 		for key, elem in obj {
-			delete(key, p.allocator)
-			destroy_value(elem)
+			delete(key, p.allocator, loc)
+			destroy_value(elem, loc=loc)
 		}
-		delete(obj)
+		delete(obj, loc)
 	}
 
 	for p.curr_token.kind != end_token {
-		key := parse_object_key(p, p.allocator) or_return
+		key := parse_object_key(p, p.allocator, loc) or_return
 		parse_colon(p) or_return
-		elem := parse_value(p) or_return
+		elem := parse_value(p, loc) or_return
 
 		if key in obj {
 			err = .Duplicate_Object_Key
-			delete(key, p.allocator)
+			delete(key, p.allocator, loc)
 			return
 		}
 
@@ -267,7 +268,7 @@ parse_object_body :: proc(p: ^Parser, end_token: Token_Kind) -> (obj: Object, er
 		// inserting empty key/values into the object and for those we do not
 		// want to allocate anything
 		if key != "" {
-			reserve_error := reserve(&obj, len(obj) + 1)
+			reserve_error := reserve(&obj, len(obj) + 1, loc)
 			if reserve_error == mem.Allocator_Error.Out_Of_Memory {
 				return nil, .Out_Of_Memory
 			}
@@ -281,9 +282,9 @@ parse_object_body :: proc(p: ^Parser, end_token: Token_Kind) -> (obj: Object, er
 	return obj, .None
 }
 
-parse_object :: proc(p: ^Parser) -> (value: Value, err: Error) {
+parse_object :: proc(p: ^Parser, loc := #caller_location) -> (value: Value, err: Error) {
 	expect_token(p, .Open_Brace) or_return
-	obj := parse_object_body(p, .Close_Brace) or_return
+	obj := parse_object_body(p, .Close_Brace, loc) or_return
 	expect_token(p, .Close_Brace) or_return
 	return obj, .None
 }
@@ -480,4 +481,4 @@ unquote_string :: proc(token: Token, spec: Specification, allocator := context.a
 	}
 
 	return string(b[:w]), nil
-}
+}

core/encoding/json/types.odin

@@ -89,22 +89,22 @@ Error :: enum {
 
 
 
-destroy_value :: proc(value: Value, allocator := context.allocator) {
+destroy_value :: proc(value: Value, allocator := context.allocator, loc := #caller_location) {
 	context.allocator = allocator
 	#partial switch v in value {
 	case Object:
 		for key, elem in v {
-			delete(key)
-			destroy_value(elem)
+			delete(key, loc=loc)
+			destroy_value(elem, loc=loc)
 		}
-		delete(v)
+		delete(v, loc=loc)
 	case Array:
 		for elem in v {
-			destroy_value(elem)
+			destroy_value(elem, loc=loc)
 		}
-		delete(v)
+		delete(v, loc=loc)
 	case String:
-		delete(v)
+		delete(v, loc=loc)
 	}
 }
 

core/encoding/json/unmarshal.odin

@@ -348,7 +348,7 @@ json_name_from_tag_value :: proc(value: string) -> (json_name, extra: string) {
 	json_name = value
 	if comma_index := strings.index_byte(json_name, ','); comma_index >= 0 {
 		json_name = json_name[:comma_index]
-		extra = json_name[comma_index:]
+		extra = value[1 + comma_index:]
 	}
 	return
 }
@@ -368,7 +368,7 @@ unmarshal_object :: proc(p: ^Parser, v: any, end_token: Token_Kind) -> (err: Unm
 	
 	#partial switch t in ti.variant {
 	case reflect.Type_Info_Struct:
-		if t.is_raw_union {
+		if .raw_union in t.flags {
 			return UNSUPPORTED_TYPE
 		}
 	
@@ -475,7 +475,7 @@ unmarshal_object :: proc(p: ^Parser, v: any, end_token: Token_Kind) -> (err: Unm
 		}
 		
 	case reflect.Type_Info_Map:
-		if !reflect.is_string(t.key) {
+		if !reflect.is_string(t.key) && !reflect.is_integer(t.key) {
 			return UNSUPPORTED_TYPE
 		}
 		raw_map := (^mem.Raw_Map)(v.data)
@@ -492,25 +492,39 @@ unmarshal_object :: proc(p: ^Parser, v: any, end_token: Token_Kind) -> (err: Unm
 			key, _ := parse_object_key(p, p.allocator)
 			unmarshal_expect_token(p, .Colon)
 			
-			
+
 			mem.zero_slice(elem_backing)
 			if uerr := unmarshal_value(p, map_backing_value); uerr != nil {
 				delete(key, p.allocator)
 				return uerr
 			}
 
-			key_ptr := rawptr(&key)
+			key_ptr: rawptr
 
-			key_cstr: cstring
-			if reflect.is_cstring(t.key) {
-				key_cstr = cstring(raw_data(key))
-				key_ptr = &key_cstr
+			#partial switch tk in t.key.variant {
+				case runtime.Type_Info_String:			
+					key_ptr = rawptr(&key)
+					key_cstr: cstring
+					if reflect.is_cstring(t.key) {
+						key_cstr = cstring(raw_data(key))
+						key_ptr = &key_cstr
+					}
+				case runtime.Type_Info_Integer:
+					i, ok := strconv.parse_i128(key)
+					if !ok	{ return UNSUPPORTED_TYPE }
+					key_ptr = rawptr(&i)
+				case: return UNSUPPORTED_TYPE
 			}
-			
+
 			set_ptr := runtime.__dynamic_map_set_without_hash(raw_map, t.map_info, key_ptr, map_backing_value.data)
 			if set_ptr == nil {
 				delete(key, p.allocator)
 			} 
+
+			// there's no need to keep string value on the heap, since it was copied into map 
+			if reflect.is_integer(t.key) {
+				delete(key, p.allocator)
+			}
 			
 			if parse_comma(p) {
 				break map_loop

core/encoding/uuid/LICENSE

@@ -0,0 +1,28 @@
+BSD 3-Clause License
+
+Copyright (c) 2024, Feoramund
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+3. Neither the name of the copyright holder nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

core/encoding/uuid/definitions.odin

@@ -0,0 +1,67 @@
+package uuid
+
+// A RFC 4122 Universally Unique Identifier
+Identifier :: distinct [16]u8
+
+EXPECTED_LENGTH :: 8 + 4 + 4 + 4 + 12 + 4
+
+VERSION_BYTE_INDEX :: 6
+VARIANT_BYTE_INDEX :: 8
+
+// The number of 100-nanosecond intervals between 1582-10-15 and 1970-01-01.
+HNS_INTERVALS_BETWEEN_GREG_AND_UNIX :: 141427 * 24 * 60 * 60 * 1000 * 1000 * 10
+
+VERSION_7_TIME_MASK     :: 0xffffffff_ffff0000_00000000_00000000
+VERSION_7_TIME_SHIFT    :: 80
+VERSION_7_COUNTER_MASK  :: 0x00000000_00000fff_00000000_00000000
+VERSION_7_COUNTER_SHIFT :: 64
+
+@(private)
+NO_CSPRNG_ERROR :: "The context random generator is not cryptographic. See the documentation for an example of how to set one up."
+@(private)
+BIG_CLOCK_ERROR :: "The clock sequence can only hold 14 bits of data, therefore no number greater than 16,383 (0x3FFF)."
+@(private)
+VERSION_7_BIG_COUNTER_ERROR :: "This implementation of the version 7 UUID counter can only hold 12 bits of data, therefore no number greater than 4,095 (0xFFF)."
+
+Read_Error :: enum {
+	None,
+	Invalid_Length,
+	Invalid_Hexadecimal,
+	Invalid_Separator,
+}
+
+Variant_Type :: enum {
+	Unknown,
+	Reserved_Apollo_NCS,    // 0b0xx
+	RFC_4122,               // 0b10x
+	Reserved_Microsoft_COM, // 0b110
+	Reserved_Future,        // 0b111
+}
+
+// Name string is a fully-qualified domain name.
+@(rodata)
+Namespace_DNS := Identifier {
+	0x6b, 0xa7, 0xb8, 0x10, 0x9d, 0xad, 0x11, 0xd1,
+	0x80, 0xb4, 0x00, 0xc0, 0x4f, 0xd4, 0x30, 0xc8,
+}
+
+// Name string is a URL.
+@(rodata)
+Namespace_URL := Identifier {
+	0x6b, 0xa7, 0xb8, 0x11, 0x9d, 0xad, 0x11, 0xd1,
+	0x80, 0xb4, 0x00, 0xc0, 0x4f, 0xd4, 0x30, 0xc8,
+}
+
+// Name string is an ISO OID.
+@(rodata)
+Namespace_OID := Identifier {
+	0x6b, 0xa7, 0xb8, 0x12, 0x9d, 0xad, 0x11, 0xd1,
+	0x80, 0xb4, 0x00, 0xc0, 0x4f, 0xd4, 0x30, 0xc8,
+}
+
+// Name string is an X.500 DN (in DER or a text output format).
+@(rodata)
+Namespace_X500 := Identifier {
+	0x6b, 0xa7, 0xb8, 0x14, 0x9d, 0xad, 0x11, 0xd1,
+	0x80, 0xb4, 0x00, 0xc0, 0x4f, 0xd4, 0x30, 0xc8,
+}

core/encoding/uuid/doc.odin

@@ -0,0 +1,46 @@
+/*
+package uuid implements Universally Unique Identifiers according to the
+standard originally outlined in RFC 4122 with additions from RFC 9562.
+
+The UUIDs are textually represented and read in the following string format:
+`00000000-0000-v000-V000-000000000000`
+
+`v` is where the version bits reside, and `V` is where the variant bits reside.
+The meaning of the other bits is version-dependent.
+
+Outside of string representations, UUIDs are represented in memory by a 128-bit
+structure organized as an array of 16 bytes.
+
+
+Of the UUID versions which may make use of random number generation, a
+requirement is placed upon them that the underlying generator be
+cryptographically-secure, per RFC 9562's suggestion.
+
+- Version 1 without a node argument.
+- Version 4 in all cases.
+- Version 6 without either a clock or node argument.
+- Version 7 in all cases.
+
+Here's an example of how to set up one:
+	
+	import "core:crypto"
+	import "core:encoding/uuid"
+
+	main :: proc() {
+		my_uuid: uuid.Identifier
+
+		{
+			// This scope will have a CSPRNG.
+			context.random_generator = crypto.random_generator()
+			my_uuid = uuid.generate_v7()
+		}
+
+		// Back to the default random number generator.
+	}
+
+
+For more information on the specifications, see here:
+- https://www.rfc-editor.org/rfc/rfc4122.html
+- https://www.rfc-editor.org/rfc/rfc9562.html
+*/
+package uuid

core/encoding/uuid/generation.odin

@@ -0,0 +1,333 @@
+package uuid
+
+import "base:runtime"
+import "core:crypto/hash"
+import "core:math/rand"
+import "core:time"
+
+/*
+Generate a version 1 UUID.
+
+Inputs:
+- clock_seq: The clock sequence, a number which must be initialized to a random number once in the lifetime of a system.
+- node: An optional 48-bit spatially unique identifier, specified to be the IEEE 802 address of the system.
+  If one is not provided or available, 48 bits of random state will take its place.
+- timestamp: A timestamp from the `core:time` package, or `nil` to use the current time.
+
+Returns:
+- result: The generated UUID.
+*/
+generate_v1 :: proc(clock_seq: u16, node: Maybe([6]u8) = nil, timestamp: Maybe(time.Time) = nil) -> (result: Identifier) {
+	assert(clock_seq <= 0x3FFF, BIG_CLOCK_ERROR)
+	unix_time_in_hns_intervals := time.to_unix_nanoseconds(timestamp.? or_else time.now()) / 100
+
+	uuid_timestamp := cast(u64le)(HNS_INTERVALS_BETWEEN_GREG_AND_UNIX + unix_time_in_hns_intervals)
+	uuid_timestamp_octets := transmute([8]u8)uuid_timestamp
+
+	result[0] = uuid_timestamp_octets[0]
+	result[1] = uuid_timestamp_octets[1]
+	result[2] = uuid_timestamp_octets[2]
+	result[3] = uuid_timestamp_octets[3]
+	result[4] = uuid_timestamp_octets[4]
+	result[5] = uuid_timestamp_octets[5]
+
+	result[6] = uuid_timestamp_octets[6] >> 4
+	result[7] = uuid_timestamp_octets[6] << 4 | uuid_timestamp_octets[7]
+
+	if realized_node, ok := node.?; ok {
+		mutable_node := realized_node
+		runtime.mem_copy_non_overlapping(&result[10], &mutable_node[0], 6)
+	} else {
+		assert(.Cryptographic in runtime.random_generator_query_info(context.random_generator), NO_CSPRNG_ERROR)
+		bytes_generated := rand.read(result[10:])
+		assert(bytes_generated == 6, "RNG failed to generate 6 bytes for UUID v1.")
+	}
+
+	result[VERSION_BYTE_INDEX] |= 0x10
+	result[VARIANT_BYTE_INDEX] |= 0x80
+
+	result[8] |= cast(u8)(clock_seq & 0x3F00 >> 8)
+	result[9]  = cast(u8)clock_seq
+
+	return
+}
+
+/*
+Generate a version 4 UUID.
+
+This UUID will be pseudorandom, save for 6 pre-determined version and variant bits.
+
+Returns:
+- result: The generated UUID.
+*/
+generate_v4 :: proc() -> (result: Identifier) {
+	assert(.Cryptographic in runtime.random_generator_query_info(context.random_generator), NO_CSPRNG_ERROR)
+	bytes_generated := rand.read(result[:])
+	assert(bytes_generated == 16, "RNG failed to generate 16 bytes for UUID v4.")
+
+	result[VERSION_BYTE_INDEX] &= 0x0F
+	result[VERSION_BYTE_INDEX] |= 0x40
+
+	result[VARIANT_BYTE_INDEX] &= 0x3F
+	result[VARIANT_BYTE_INDEX] |= 0x80
+
+	return
+}
+
+/*
+Generate a version 6 UUID.
+
+Inputs:
+- clock_seq: The clock sequence from version 1, now made optional.
+  If unspecified, it will be replaced with random bits.
+- node: An optional 48-bit spatially unique identifier, specified to be the IEEE 802 address of the system.
+  If one is not provided or available, 48 bits of random state will take its place.
+- timestamp: A timestamp from the `core:time` package, or `nil` to use the current time.
+
+Returns:
+- result: The generated UUID.
+*/
+generate_v6 :: proc(clock_seq: Maybe(u16) = nil, node: Maybe([6]u8) = nil, timestamp: Maybe(time.Time) = nil) -> (result: Identifier) {
+	unix_time_in_hns_intervals := time.to_unix_nanoseconds(timestamp.? or_else time.now()) / 100
+
+	uuid_timestamp := cast(u128be)(HNS_INTERVALS_BETWEEN_GREG_AND_UNIX + unix_time_in_hns_intervals)
+
+	result = transmute(Identifier)(
+		uuid_timestamp & 0x0FFFFFFF_FFFFF000 << 68 |
+		uuid_timestamp & 0x00000000_00000FFF << 64
+	)
+
+	if realized_clock_seq, ok := clock_seq.?; ok {
+		assert(realized_clock_seq <= 0x3FFF, BIG_CLOCK_ERROR)
+		result[8] |= cast(u8)(realized_clock_seq & 0x3F00 >> 8)
+		result[9]  = cast(u8)realized_clock_seq
+	} else {
+		assert(.Cryptographic in runtime.random_generator_query_info(context.random_generator), NO_CSPRNG_ERROR)
+		temporary: [2]u8
+		bytes_generated := rand.read(temporary[:])
+		assert(bytes_generated == 2, "RNG failed to generate 2 bytes for UUID v1.")
+		result[8] |= temporary[0] & 0x3F
+		result[9]  = temporary[1]
+	}
+
+	if realized_node, ok := node.?; ok {
+		mutable_node := realized_node
+		runtime.mem_copy_non_overlapping(&result[10], &mutable_node[0], 6)
+	} else {
+		assert(.Cryptographic in runtime.random_generator_query_info(context.random_generator), NO_CSPRNG_ERROR)
+		bytes_generated := rand.read(result[10:])
+		assert(bytes_generated == 6, "RNG failed to generate 6 bytes for UUID v1.")
+	}
+
+	result[VERSION_BYTE_INDEX] |= 0x60
+	result[VARIANT_BYTE_INDEX] |= 0x80
+
+	return
+}
+
+/*
+Generate a version 7 UUID.
+
+This UUID will be pseudorandom, save for 6 pre-determined version and variant
+bits and a 48-bit timestamp.
+
+It is designed with time-based sorting in mind, such as for database usage, as
+the highest bits are allocated from the timestamp of when it is created.
+
+Inputs:
+- timestamp: A timestamp from the `core:time` package, or `nil` to use the current time.
+
+Returns:
+- result: The generated UUID.
+*/
+generate_v7_basic :: proc(timestamp: Maybe(time.Time) = nil) -> (result: Identifier) {
+	assert(.Cryptographic in runtime.random_generator_query_info(context.random_generator), NO_CSPRNG_ERROR)
+	unix_time_in_milliseconds := time.to_unix_nanoseconds(timestamp.? or_else time.now()) / 1e6
+
+	result = transmute(Identifier)(cast(u128be)unix_time_in_milliseconds << VERSION_7_TIME_SHIFT)
+
+	bytes_generated := rand.read(result[6:])
+	assert(bytes_generated == 10, "RNG failed to generate 10 bytes for UUID v7.")
+
+	result[VERSION_BYTE_INDEX] &= 0x0F
+	result[VERSION_BYTE_INDEX] |= 0x70
+
+	result[VARIANT_BYTE_INDEX] &= 0x3F
+	result[VARIANT_BYTE_INDEX] |= 0x80
+
+	return
+}
+
+/*
+Generate a version 7 UUID that has an incremented counter.
+
+This UUID will be pseudorandom, save for 6 pre-determined version and variant
+bits, a 48-bit timestamp, and 12 bits of counter state.
+
+It is designed with time-based sorting in mind, such as for database usage, as
+the highest bits are allocated from the timestamp of when it is created.
+
+This procedure is preferable if you are generating hundreds or thousands of
+UUIDs as a batch within the span of a millisecond. Do note that the counter
+only has 12 bits of state, thus `counter` cannot exceed the number 4,095.
+
+Example:
+
+	import "core:uuid"
+
+	// Create a batch of UUIDs all at once.
+	batch: [dynamic]uuid.Identifier
+
+	for i: u16 = 0; i < 1000; i += 1 {
+		my_uuid := uuid.generate_v7_counter(i)
+		append(&batch, my_uuid)
+	}
+
+Inputs:
+- counter: A 12-bit value which should be incremented each time a UUID is generated in a batch.
+- timestamp: A timestamp from the `core:time` package, or `nil` to use the current time.
+
+Returns:
+- result: The generated UUID.
+*/
+generate_v7_with_counter :: proc(counter: u16, timestamp: Maybe(time.Time) = nil) -> (result: Identifier) {
+	assert(.Cryptographic in runtime.random_generator_query_info(context.random_generator), NO_CSPRNG_ERROR)
+	assert(counter <= 0x0fff, VERSION_7_BIG_COUNTER_ERROR)
+	unix_time_in_milliseconds := time.to_unix_nanoseconds(timestamp.? or_else time.now()) / 1e6
+
+	result = transmute(Identifier)(
+		cast(u128be)unix_time_in_milliseconds << VERSION_7_TIME_SHIFT |
+		cast(u128be)counter << VERSION_7_COUNTER_SHIFT
+	)
+
+	bytes_generated := rand.read(result[8:])
+	assert(bytes_generated == 8, "RNG failed to generate 8 bytes for UUID v7.")
+
+	result[VERSION_BYTE_INDEX] &= 0x0F
+	result[VERSION_BYTE_INDEX] |= 0x70
+
+	result[VARIANT_BYTE_INDEX] &= 0x3F
+	result[VARIANT_BYTE_INDEX] |= 0x80
+
+	return
+}
+
+generate_v7 :: proc {
+	generate_v7_basic,
+	generate_v7_with_counter,
+}
+
+/*
+Generate a version 8 UUID using a specific hashing algorithm.
+
+This UUID is generated by hashing a name with a namespace.
+
+Note that all version 8 UUIDs are for experimental or vendor-specific use
+cases, per the specification. This use case in particular is for offering a
+non-legacy alternative to UUID versions 3 and 5.
+
+Inputs:
+- namespace: An `Identifier` that is used to represent the underlying namespace.
+  This can be any one of the `Namespace_*` values provided in this package.
+- name: The byte slice which will be hashed with the namespace.
+- algorithm: A hashing algorithm from `core:crypto/hash`.
+
+Returns:
+- result: The generated UUID.
+
+Example:
+	import "core:crypto/hash"
+	import "core:encoding/uuid"
+	import "core:fmt"
+
+	main :: proc() {
+		my_uuid := uuid.generate_v8_hash(uuid.Namespace_DNS, "www.odin-lang.org", .SHA256)
+		my_uuid_string := uuid.to_string(my_uuid, context.temp_allocator)
+		fmt.println(my_uuid_string)
+	}
+
+Output:
+
+	3730f688-4bff-8dce-9cbf-74a3960c5703
+
+*/
+generate_v8_hash_bytes :: proc(
+	namespace: Identifier,
+	name: []byte,
+	algorithm: hash.Algorithm,
+) -> (
+	result: Identifier,
+) {
+	// 128 bytes should be enough for the foreseeable future.
+	digest: [128]byte
+
+	assert(hash.DIGEST_SIZES[algorithm] >= 16, "Per RFC 9562, the hashing algorithm used must generate a digest of 128 bits or larger.")
+	assert(hash.DIGEST_SIZES[algorithm] < len(digest), "Digest size is too small for this algorithm. The buffer must be increased.")
+
+	hash_context: hash.Context
+	hash.init(&hash_context, algorithm)
+
+	mutable_namespace := namespace
+	hash.update(&hash_context, mutable_namespace[:])
+	hash.update(&hash_context, name[:])
+	hash.final(&hash_context, digest[:])
+
+	runtime.mem_copy_non_overlapping(&result, &digest, 16)
+
+	result[VERSION_BYTE_INDEX] &= 0x0F
+	result[VERSION_BYTE_INDEX] |= 0x80
+
+	result[VARIANT_BYTE_INDEX] &= 0x3F
+	result[VARIANT_BYTE_INDEX] |= 0x80
+
+	return
+}
+
+/*
+Generate a version 8 UUID using a specific hashing algorithm.
+
+This UUID is generated by hashing a name with a namespace.
+
+Note that all version 8 UUIDs are for experimental or vendor-specific use
+cases, per the specification. This use case in particular is for offering a
+non-legacy alternative to UUID versions 3 and 5.
+
+Inputs:
+- namespace: An `Identifier` that is used to represent the underlying namespace.
+  This can be any one of the `Namespace_*` values provided in this package.
+- name: The string which will be hashed with the namespace.
+- algorithm: A hashing algorithm from `core:crypto/hash`.
+
+Returns:
+- result: The generated UUID.
+
+Example:
+	import "core:crypto/hash"
+	import "core:encoding/uuid"
+	import "core:fmt"
+
+	main :: proc() {
+		my_uuid := uuid.generate_v8_hash(uuid.Namespace_DNS, "www.odin-lang.org", .SHA256)
+		my_uuid_string := uuid.to_string(my_uuid, context.temp_allocator)
+		fmt.println(my_uuid_string)
+	}
+
+Output:
+
+	3730f688-4bff-8dce-9cbf-74a3960c5703
+
+*/
+generate_v8_hash_string :: proc(
+	namespace: Identifier,
+	name: string,
+	algorithm: hash.Algorithm,
+) -> (
+	result: Identifier,
+) {
+	return generate_v8_hash_bytes(namespace, transmute([]byte)name, algorithm)
+}
+
+generate_v8_hash :: proc {
+	generate_v8_hash_bytes,
+	generate_v8_hash_string,
+}

core/encoding/uuid/legacy/legacy.odin

@@ -0,0 +1,146 @@
+/*
+package uuid/legacy implements versions 3 and 5 of UUID generation, both of
+which are using hashing algorithms (MD5 and SHA1, respectively) that are known
+these days to no longer be secure.
+*/
+package uuid_legacy
+
+import "base:runtime"
+import "core:crypto/legacy/md5"
+import "core:crypto/legacy/sha1"
+import "core:encoding/uuid"
+
+Identifier :: uuid.Identifier
+VERSION_BYTE_INDEX :: uuid.VERSION_BYTE_INDEX
+VARIANT_BYTE_INDEX :: uuid.VARIANT_BYTE_INDEX
+
+
+/*
+Generate a version 3 UUID.
+
+This UUID is generated with a MD5 hash of a name and a namespace.
+
+Inputs:
+- namespace: An `Identifier` that is used to represent the underlying namespace.
+  This can be any one of the `Namespace_*` values provided in the `uuid` package.
+- name: The byte slice which will be hashed with the namespace.
+
+Returns:
+- result: The generated UUID.
+*/
+generate_v3_bytes :: proc(
+	namespace: Identifier,
+	name: []byte,
+) -> (
+	result: Identifier,
+) {
+	namespace := namespace
+
+	ctx: md5.Context
+	md5.init(&ctx)
+	md5.update(&ctx, namespace[:])
+	md5.update(&ctx, name)
+	md5.final(&ctx, result[:])
+
+	result[VERSION_BYTE_INDEX] &= 0x0F
+	result[VERSION_BYTE_INDEX] |= 0x30
+
+	result[VARIANT_BYTE_INDEX] &= 0x3F
+	result[VARIANT_BYTE_INDEX] |= 0x80
+
+	return
+}
+
+/*
+Generate a version 3 UUID.
+
+This UUID is generated with a MD5 hash of a name and a namespace.
+
+Inputs:
+- namespace: An `Identifier` that is used to represent the underlying namespace.
+  This can be any one of the `Namespace_*` values provided in the `uuid` package.
+- name: The string which will be hashed with the namespace.
+
+Returns:
+- result: The generated UUID.
+*/
+generate_v3_string :: proc(
+	namespace: Identifier,
+	name: string,
+) -> (
+	result: Identifier,
+) {
+	return generate_v3_bytes(namespace, transmute([]byte)name)
+}
+
+generate_v3 :: proc {
+	generate_v3_bytes,
+	generate_v3_string,
+}
+
+/*
+Generate a version 5 UUID.
+
+This UUID is generated with a SHA1 hash of a name and a namespace.
+
+Inputs:
+- namespace: An `Identifier` that is used to represent the underlying namespace.
+  This can be any one of the `Namespace_*` values provided in the `uuid` package.
+- name: The byte slice which will be hashed with the namespace.
+
+Returns:
+- result: The generated UUID.
+*/
+generate_v5_bytes :: proc(
+	namespace: Identifier,
+	name: []byte,
+) -> (
+	result: Identifier,
+) {
+	namespace := namespace
+	digest: [sha1.DIGEST_SIZE]byte
+
+	ctx: sha1.Context
+	sha1.init(&ctx)
+	sha1.update(&ctx, namespace[:])
+	sha1.update(&ctx, name)
+	sha1.final(&ctx, digest[:])
+
+	runtime.mem_copy_non_overlapping(&result, &digest, 16)
+
+	result[VERSION_BYTE_INDEX] &= 0x0F
+	result[VERSION_BYTE_INDEX] |= 0x50
+
+	result[VARIANT_BYTE_INDEX] &= 0x3F
+	result[VARIANT_BYTE_INDEX] |= 0x80
+
+	return
+}
+
+/*
+Generate a version 5 UUID.
+
+This UUID is generated with a SHA1 hash of a name and a namespace.
+
+Inputs:
+- namespace: An `Identifier` that is used to represent the underlying namespace.
+  This can be any one of the `Namespace_*` values provided in the `uuid` package.
+- name: The string which will be hashed with the namespace.
+
+Returns:
+- result: The generated UUID.
+*/
+generate_v5_string :: proc(
+	namespace: Identifier,
+	name: string,
+) -> (
+	result: Identifier,
+) {
+	return generate_v5_bytes(namespace, transmute([]byte)name)
+}
+
+generate_v5 :: proc {
+	generate_v5_bytes,
+	generate_v5_string,
+}
+

core/encoding/uuid/reading.odin

@@ -0,0 +1,242 @@
+package uuid
+
+import "base:runtime"
+import "core:time"
+
+/*
+Convert a string to a UUID.
+
+Inputs:
+- str: A string in the 8-4-4-4-12 format.
+
+Returns:
+- id: The converted identifier, or `nil` if there is an error.
+- error: A description of the error, or `nil` if successful.
+*/
+read :: proc "contextless" (str: string) -> (id: Identifier, error: Read_Error) #no_bounds_check {
+	// Only exact-length strings are acceptable.
+	if len(str) != EXPECTED_LENGTH {
+		return {}, .Invalid_Length
+	}
+
+	// Check ahead to see if the separators are in the right places.
+	if str[8] != '-' || str[13] != '-' || str[18] != '-' || str[23] != '-' {
+		return {}, .Invalid_Separator
+	}
+
+	read_nibble :: proc "contextless" (nibble: u8) -> u8 {
+		switch nibble {
+		case '0' ..= '9':
+			return nibble - '0'
+		case 'A' ..= 'F':
+			return nibble - 'A' + 10
+		case 'a' ..= 'f':
+			return nibble - 'a' + 10
+		case:
+			// Return an error value.
+			return 0xFF
+		}
+	}
+
+	index := 0
+	octet_index := 0
+
+	CHUNKS :: [5]int{8, 4, 4, 4, 12}
+
+	for chunk in CHUNKS {
+		for i := index; i < index + chunk; i += 2 {
+			high := read_nibble(str[i])
+			low := read_nibble(str[i + 1])
+
+			if high | low > 0xF {
+				return {}, .Invalid_Hexadecimal
+			}
+
+			id[octet_index] = low | high << 4
+			octet_index += 1
+		}
+
+		index += chunk + 1
+	}
+
+	return
+}
+
+/*
+Get the version of a UUID.
+
+Inputs:
+- id: The identifier.
+
+Returns:
+- number: The version number.
+*/
+version :: proc "contextless" (id: Identifier) -> (number: int) #no_bounds_check {
+	return cast(int)(id[VERSION_BYTE_INDEX] & 0xF0 >> 4)
+}
+
+/*
+Get the variant of a UUID.
+
+Inputs:
+- id: The identifier.
+
+Returns:
+- variant: The variant type.
+*/
+variant :: proc "contextless" (id: Identifier) -> (variant: Variant_Type) #no_bounds_check {
+	switch {
+	case id[VARIANT_BYTE_INDEX] & 0x80 == 0:
+		return .Reserved_Apollo_NCS
+	case id[VARIANT_BYTE_INDEX] & 0xC0 == 0x80:
+		return .RFC_4122
+	case id[VARIANT_BYTE_INDEX] & 0xE0 == 0xC0:
+		return .Reserved_Microsoft_COM
+	case id[VARIANT_BYTE_INDEX] & 0xF0 == 0xE0:
+		return .Reserved_Future
+	case:
+		return .Unknown
+	}
+}
+
+/*
+Get the clock sequence of a version 1 or version 6 UUID.
+
+Inputs:
+- id: The identifier.
+
+Returns:
+- clock_seq: The 14-bit clock sequence field.
+*/
+clock_seq :: proc "contextless" (id: Identifier) -> (clock_seq: u16) {
+	return cast(u16)id[9] | cast(u16)id[8] & 0x3F << 8
+}
+
+/*
+Get the node of a version 1 or version 6 UUID.
+
+Inputs:
+- id: The identifier.
+
+Returns:
+- node: The 48-bit spatially unique identifier.
+*/
+node :: proc "contextless" (id: Identifier) -> (node: [6]u8) {
+	mutable_id := id
+	runtime.mem_copy_non_overlapping(&node, &mutable_id[10], 6)
+	return
+}
+
+/*
+Get the raw timestamp of a version 1 UUID.
+
+Inputs:
+- id: The identifier.
+
+Returns:
+- timestamp: The timestamp, in 100-nanosecond intervals since 1582-10-15.
+*/
+raw_time_v1 :: proc "contextless" (id: Identifier) -> (timestamp: u64) {
+	timestamp_octets: [8]u8
+
+	timestamp_octets[0] = id[0]
+	timestamp_octets[1] = id[1]
+	timestamp_octets[2] = id[2]
+	timestamp_octets[3] = id[3]
+	timestamp_octets[4] = id[4]
+	timestamp_octets[5] = id[5]
+
+	timestamp_octets[6] = id[6] << 4 | id[7] >> 4
+	timestamp_octets[7] = id[7] & 0xF
+
+	return cast(u64)transmute(u64le)timestamp_octets
+}
+
+
+/*
+Get the timestamp of a version 1 UUID.
+
+Inputs:
+- id: The identifier.
+
+Returns:
+- timestamp: The timestamp of the UUID.
+*/
+time_v1 :: proc "contextless" (id: Identifier) -> (timestamp: time.Time) {
+	return time.from_nanoseconds(cast(i64)(raw_time_v1(id) - HNS_INTERVALS_BETWEEN_GREG_AND_UNIX) * 100)
+}
+
+/*
+Get the raw timestamp of a version 6 UUID.
+
+Inputs:
+- id: The identifier.
+
+Returns:
+- timestamp: The timestamp, in 100-nanosecond intervals since 1582-10-15.
+*/
+raw_time_v6 :: proc "contextless" (id: Identifier) -> (timestamp: u64) {
+	temporary := transmute(u128be)id
+
+	timestamp |= cast(u64)(temporary & 0xFFFFFFFF_FFFF0000_00000000_00000000 >> 68)
+	timestamp |= cast(u64)(temporary & 0x00000000_00000FFF_00000000_00000000 >> 64)
+
+	return timestamp
+}
+
+/*
+Get the timestamp of a version 6 UUID.
+
+Inputs:
+- id: The identifier.
+
+Returns:
+- timestamp: The timestamp, in 100-nanosecond intervals since 1582-10-15.
+*/
+time_v6 :: proc "contextless" (id: Identifier) -> (timestamp: time.Time) {
+	return time.from_nanoseconds(cast(i64)(raw_time_v6(id) - HNS_INTERVALS_BETWEEN_GREG_AND_UNIX) * 100)
+}
+
+/*
+Get the raw timestamp of a version 7 UUID.
+
+Inputs:
+- id: The identifier.
+
+Returns:
+- timestamp: The timestamp, in milliseconds since the UNIX epoch.
+*/
+raw_time_v7 :: proc "contextless" (id: Identifier) -> (timestamp: u64) {
+	time_bits := transmute(u128be)id & VERSION_7_TIME_MASK
+	return cast(u64)(time_bits >> VERSION_7_TIME_SHIFT)
+}
+
+/*
+Get the timestamp of a version 7 UUID.
+
+Inputs:
+- id: The identifier.
+
+Returns:
+- timestamp: The timestamp, in milliseconds since the UNIX epoch.
+*/
+time_v7 :: proc "contextless" (id: Identifier) -> (timestamp: time.Time) {
+	return time.from_nanoseconds(cast(i64)raw_time_v7(id) * 1e6)
+}
+
+/*
+Get the 12-bit counter value of a version 7 UUID.
+
+The UUID must have been generated with a counter, otherwise this procedure will
+return random bits.
+
+Inputs:
+- id: The identifier.
+
+Returns:
+- counter: The 12-bit counter value.
+*/
+counter_v7 :: proc "contextless" (id: Identifier) -> (counter: u16) {
+	counter_bits := transmute(u128be)id & VERSION_7_COUNTER_MASK
+	return cast(u16)(counter_bits >> VERSION_7_COUNTER_SHIFT)
+}

core/encoding/uuid/stamping.odin

@@ -0,0 +1,89 @@
+package uuid
+
+import "base:runtime"
+
+/*
+Stamp a 128-bit integer as being a valid version 8 UUID.
+
+Per the specification, all version 8 UUIDs are either for experimental or
+vendor-specific purposes. This procedure allows for converting arbitrary data
+into custom UUIDs.
+
+Inputs:
+- integer: Any integer type.
+
+Returns:
+- result: A valid version 8 UUID.
+*/
+stamp_v8_int :: proc(#any_int integer: u128) -> (result: Identifier) {
+	result = transmute(Identifier)cast(u128be)integer
+
+	result[VERSION_BYTE_INDEX] &= 0x0F
+	result[VERSION_BYTE_INDEX] |= 0x80
+
+	result[VARIANT_BYTE_INDEX] &= 0x3F
+	result[VARIANT_BYTE_INDEX] |= 0x80
+
+	return
+}
+
+/*
+Stamp an array of 16 bytes as being a valid version 8 UUID.
+
+Per the specification, all version 8 UUIDs are either for experimental or
+vendor-specific purposes. This procedure allows for converting arbitrary data
+into custom UUIDs.
+
+Inputs:
+- array: An array of 16 bytes.
+
+Returns:
+- result: A valid version 8 UUID.
+*/
+stamp_v8_array :: proc(array: [16]u8) -> (result: Identifier) {
+	result = Identifier(array)
+
+	result[VERSION_BYTE_INDEX] &= 0x0F
+	result[VERSION_BYTE_INDEX] |= 0x80
+
+	result[VARIANT_BYTE_INDEX] &= 0x3F
+	result[VARIANT_BYTE_INDEX] |= 0x80
+
+	return
+}
+
+/*
+Stamp a slice of bytes as being a valid version 8 UUID.
+
+If the slice is less than 16 bytes long, the data available will be used.
+If it is longer than 16 bytes, only the first 16 will be used.
+
+This procedure does not modify the underlying slice.
+
+Per the specification, all version 8 UUIDs are either for experimental or
+vendor-specific purposes. This procedure allows for converting arbitrary data
+into custom UUIDs.
+
+Inputs:
+- slice: A slice of bytes.
+
+Returns:
+- result: A valid version 8 UUID.
+*/
+stamp_v8_slice :: proc(slice: []u8) -> (result: Identifier) {
+	runtime.mem_copy_non_overlapping(&result, &slice[0], min(16, len(slice)))
+
+	result[VERSION_BYTE_INDEX] &= 0x0F
+	result[VERSION_BYTE_INDEX] |= 0x80
+
+	result[VARIANT_BYTE_INDEX] &= 0x3F
+	result[VARIANT_BYTE_INDEX] |= 0x80
+
+	return
+}
+
+stamp_v8 :: proc {
+	stamp_v8_int,
+	stamp_v8_array,
+	stamp_v8_slice,
+}

core/encoding/uuid/writing.odin

@@ -0,0 +1,131 @@
+package uuid
+
+import "base:runtime"
+import "core:io"
+import "core:strconv"
+import "core:strings"
+
+/*
+Write a UUID in the 8-4-4-4-12 format.
+
+This procedure performs error checking with every byte written.
+
+If you can guarantee beforehand that your stream has enough space to hold the
+UUID (32 bytes), then it is better to use `unsafe_write` instead as that will
+be faster.
+
+Inputs:
+- w: A writable stream.
+- id: The identifier to convert.
+
+Returns:
+- error: An `io` error, if one occurred, otherwise `nil`.
+*/
+write :: proc(w: io.Writer, id: Identifier) -> (error: io.Error) #no_bounds_check {
+	write_octet :: proc (w: io.Writer, octet: u8) -> io.Error #no_bounds_check {
+		high_nibble := octet >> 4
+		low_nibble := octet & 0xF
+
+		io.write_byte(w, strconv.digits[high_nibble]) or_return
+		io.write_byte(w, strconv.digits[low_nibble]) or_return
+		return nil
+	}
+
+	for index in  0 ..<  4 { write_octet(w, id[index]) or_return }
+	io.write_byte(w, '-') or_return
+	for index in  4 ..<  6 { write_octet(w, id[index]) or_return }
+	io.write_byte(w, '-') or_return
+	for index in  6 ..<  8 { write_octet(w, id[index]) or_return }
+	io.write_byte(w, '-') or_return
+	for index in  8 ..< 10 { write_octet(w, id[index]) or_return }
+	io.write_byte(w, '-') or_return
+	for index in 10 ..< 16 { write_octet(w, id[index]) or_return }
+
+	return nil
+}
+
+/*
+Write a UUID in the 8-4-4-4-12 format.
+
+This procedure performs no error checking on the underlying stream.
+
+Inputs:
+- w: A writable stream.
+- id: The identifier to convert.
+*/
+unsafe_write :: proc(w: io.Writer, id: Identifier) #no_bounds_check {
+	write_octet :: proc (w: io.Writer, octet: u8) #no_bounds_check {
+		high_nibble := octet >> 4
+		low_nibble := octet & 0xF
+
+		io.write_byte(w, strconv.digits[high_nibble])
+		io.write_byte(w, strconv.digits[low_nibble])
+	}
+
+	for index in  0 ..<  4 { write_octet(w, id[index]) }
+	io.write_byte(w, '-')
+	for index in  4 ..<  6 { write_octet(w, id[index]) }
+	io.write_byte(w, '-')
+	for index in  6 ..<  8 { write_octet(w, id[index]) }
+	io.write_byte(w, '-')
+	for index in  8 ..< 10 { write_octet(w, id[index]) }
+	io.write_byte(w, '-')
+	for index in 10 ..< 16 { write_octet(w, id[index]) }
+}
+
+/*
+Convert a UUID to a string in the 8-4-4-4-12 format.
+
+*Allocates Using Provided Allocator*
+
+Inputs:
+- id: The identifier to convert.
+- allocator: (default: context.allocator)
+- loc: The caller location for debugging purposes (default: #caller_location)
+
+Returns:
+- str: The allocated and converted string.
+- error: An optional allocator error if one occured, `nil` otherwise.
+*/
+to_string_allocated :: proc(
+	id: Identifier,
+	allocator := context.allocator,
+	loc := #caller_location,
+) -> (
+	str: string,
+	error: runtime.Allocator_Error,
+) #optional_allocator_error {
+	buf := make([]byte, EXPECTED_LENGTH, allocator, loc) or_return
+	builder := strings.builder_from_bytes(buf[:])
+	unsafe_write(strings.to_writer(&builder), id)
+	return strings.to_string(builder), nil
+}
+
+/*
+Convert a UUID to a string in the 8-4-4-4-12 format.
+
+Inputs:
+- id: The identifier to convert.
+- buffer: A byte buffer to store the result. Must be at least 32 bytes large.
+- loc: The caller location for debugging purposes (default: #caller_location)
+
+Returns:
+- str: The converted string which will be stored in `buffer`.
+*/
+to_string_buffer :: proc(
+	id: Identifier,
+	buffer: []byte,
+	loc := #caller_location,
+) -> (
+	str: string,
+) {
+	assert(len(buffer) >= EXPECTED_LENGTH, "The buffer provided is not at least 32 bytes large.", loc)
+	builder := strings.builder_from_bytes(buffer)
+	unsafe_write(strings.to_writer(&builder), id)
+	return strings.to_string(builder)
+}
+
+to_string :: proc {
+	to_string_allocated,
+	to_string_buffer,
+}

core/encoding/xml/debug_print.odin

@@ -33,7 +33,7 @@ print :: proc(writer: io.Writer, doc: ^Document) -> (written: int, err: io.Error
 		written += fmt.wprintf(writer, "[DOCTYPE]  %v\n", doc.doctype.ident)
 
 		if len(doc.doctype.rest) > 0 {
-		 	fmt.wprintf(writer, "\t%v\n", doc.doctype.rest)
+			fmt.wprintf(writer, "\t%v\n", doc.doctype.rest)
 		}
 	}
 
@@ -42,10 +42,10 @@ print :: proc(writer: io.Writer, doc: ^Document) -> (written: int, err: io.Error
 	}
 
 	if len(doc.elements) > 0 {
-	 	fmt.wprintln(writer, " --- ")
-	 	print_element(writer, doc, 0)
-	 	fmt.wprintln(writer, " --- ")
-	 }
+		fmt.wprintln(writer, " --- ")
+		print_element(writer, doc, 0)
+		fmt.wprintln(writer, " --- ")
+	}
 
 	return written, .None
 }