Remove ?

core:odin parsing fixes

.github/workflows/ci.yml

@@ -7,8 +7,12 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v1
-      - name: Download LLVM, botan
-        run: sudo apt-get install llvm-11 clang-11 libbotan-2-dev botan
+      - name: Download LLVM
+        run: |
+          wget https://apt.llvm.org/llvm.sh
+          chmod +x llvm.sh
+          sudo ./llvm.sh 17
+          echo "/usr/lib/llvm-17/bin" >> $GITHUB_PATH
       - name: build odin
         run: ./build_odin.sh release
       - name: Odin version
@@ -61,12 +65,10 @@ jobs:
     runs-on: macos-latest
     steps:
       - uses: actions/checkout@v1
-      - name: Download LLVM, botan and setup PATH
+      - name: Download LLVM, and setup PATH
         run: |
-          brew install llvm@13 botan
-          echo "/usr/local/opt/llvm@13/bin" >> $GITHUB_PATH
-          TMP_PATH=$(xcrun --show-sdk-path)/user/include
-          echo "CPATH=$TMP_PATH" >> $GITHUB_ENV
+          brew install llvm@17
+          echo "/usr/local/opt/llvm@17/bin" >> $GITHUB_PATH
       - name: build odin
         run: ./build_odin.sh release
       - name: Odin version
@@ -102,12 +104,10 @@ jobs:
     runs-on: macos-14 # This is an arm/m1 runner.
     steps:
       - uses: actions/checkout@v1
-      - name: Download LLVM, botan and setup PATH
+      - name: Download LLVM and setup PATH
         run: |
-          brew install llvm@13 botan
-          echo "/opt/homebrew/opt/llvm@13/bin" >> $GITHUB_PATH
-          TMP_PATH=$(xcrun --show-sdk-path)/user/include
-          echo "CPATH=$TMP_PATH" >> $GITHUB_ENV
+          brew install llvm@17
+          echo "/opt/homebrew/opt/llvm@17/bin" >> $GITHUB_PATH
       - name: build odin
         run: ./build_odin.sh release
       - name: Odin version

.github/workflows/nightly.yml

@@ -47,7 +47,11 @@ jobs:
     steps:
       - uses: actions/checkout@v1
       - name: (Linux) Download LLVM
-        run: sudo apt-get install llvm-11 clang-11
+        run: |
+          wget https://apt.llvm.org/llvm.sh
+          chmod +x llvm.sh
+          sudo ./llvm.sh 17
+          echo "/usr/lib/llvm-17/bin" >> $GITHUB_PATH
       - name: build odin
         run: make nightly
       - name: Odin run
@@ -78,10 +82,8 @@ jobs:
       - uses: actions/checkout@v1
       - name: Download LLVM and setup PATH
         run: |
-          brew install llvm@13 dylibbundler
-          echo "/usr/local/opt/llvm@13/bin" >> $GITHUB_PATH
-          TMP_PATH=$(xcrun --show-sdk-path)/user/include
-          echo "CPATH=$TMP_PATH" >> $GITHUB_ENV
+          brew install llvm@17 dylibbundler
+          echo "/usr/local/opt/llvm@17/bin" >> $GITHUB_PATH
       - name: build odin
         # These -L makes the linker prioritize system libraries over LLVM libraries, this is mainly to
         # not link with libunwind bundled with LLVM but link with libunwind on the system.
@@ -114,10 +116,8 @@ jobs:
       - uses: actions/checkout@v1
       - name: Download LLVM and setup PATH
         run: |
-          brew install llvm@13 dylibbundler
-          echo "/opt/homebrew/opt/llvm@13/bin" >> $GITHUB_PATH
-          TMP_PATH=$(xcrun --show-sdk-path)/user/include
-          echo "CPATH=$TMP_PATH" >> $GITHUB_ENV
+          brew install llvm@17 dylibbundler
+          echo "/opt/homebrew/opt/llvm@17/bin" >> $GITHUB_PATH
       - name: build odin
         # These -L makes the linker prioritize system libraries over LLVM libraries, this is mainly to
         # not link with libunwind bundled with LLVM but link with libunwind on the system.

LICENSE

@@ -1,4 +1,4 @@
-Copyright (c) 2016-2022 Ginger Bill. All rights reserved.
+Copyright (c) 2016-2024 Ginger Bill. All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:

base/intrinsics/intrinsics.odin

@@ -38,9 +38,9 @@ count_leading_zeros  :: proc(x: $T) -> T where type_is_integer(T) || type_is_sim
 reverse_bits         :: proc(x: $T) -> T where type_is_integer(T) || type_is_simd_vector(T) ---
 byte_swap            :: proc(x: $T) -> T where type_is_integer(T) || type_is_float(T) ---
 
-overflow_add :: proc(lhs, rhs: $T) -> (T, bool) #optional_ok ---
-overflow_sub :: proc(lhs, rhs: $T) -> (T, bool) #optional_ok ---
-overflow_mul :: proc(lhs, rhs: $T) -> (T, bool) #optional_ok ---
+overflow_add :: proc(lhs, rhs: $T) -> (T, bool) ---
+overflow_sub :: proc(lhs, rhs: $T) -> (T, bool) ---
+overflow_mul :: proc(lhs, rhs: $T) -> (T, bool) ---
 
 sqrt :: proc(x: $T) -> T where type_is_float(T) || (type_is_simd_vector(T) && type_is_float(type_elem_type(T))) ---
 
@@ -293,7 +293,9 @@ wasm_memory_size :: proc(index: uintptr)        -> int ---
 // 0 - indicates that the thread blocked and then was woken up
 // 1 - the loaded value from `ptr` did not match `expected`, the thread did not block
 // 2 - the thread blocked, but the timeout
+@(enable_target_feature="atomics")
 wasm_memory_atomic_wait32   :: proc(ptr: ^u32, expected: u32, timeout_ns: i64) -> u32 ---
+@(enable_target_feature="atomics")
 wasm_memory_atomic_notify32 :: proc(ptr: ^u32, waiters: u32) -> (waiters_woken_up: u32) ---
 
 // x86 Targets (i386, amd64)

base/runtime/core.odin

@@ -177,6 +177,10 @@ Type_Info_Matrix :: struct {
 	row_count:    int,
 	column_count: int,
 	// Total element count = column_count * elem_stride
+	layout: enum u8 {
+		Column_Major, // array of column vectors
+		Row_Major,    // array of row vectors
+	},
 }
 Type_Info_Soa_Pointer :: struct {
 	elem: ^Type_Info,
@@ -593,8 +597,9 @@ type_info_core :: proc "contextless" (info: ^Type_Info) -> ^Type_Info {
 	base := info
 	loop: for {
 		#partial switch i in base.variant {
-		case Type_Info_Named:  base = i.base
-		case Type_Info_Enum:   base = i.base
+		case Type_Info_Named:     base = i.base
+		case Type_Info_Enum:      base = i.base
+		case Type_Info_Bit_Field: base = i.backing_type
 		case: break loop
 		}
 	}

base/runtime/core_builtin.odin

@@ -447,12 +447,12 @@ _append_elem :: #force_inline proc(array: ^$T/[dynamic]$E, arg: E, should_zero:
 }
 
 @builtin
-append_elem :: proc(array: ^$T/[dynamic]$E, arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+append_elem :: proc(array: ^$T/[dynamic]$E, #no_broadcast arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
 	return _append_elem(array, arg, true, loc=loc)
 }
 
 @builtin
-non_zero_append_elem :: proc(array: ^$T/[dynamic]$E, arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+non_zero_append_elem :: proc(array: ^$T/[dynamic]$E, #no_broadcast arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
 	return _append_elem(array, arg, false, loc=loc)
 }
 
@@ -496,12 +496,12 @@ _append_elems :: #force_inline proc(array: ^$T/[dynamic]$E, should_zero: bool, l
 }
 
 @builtin
-append_elems :: proc(array: ^$T/[dynamic]$E, args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+append_elems :: proc(array: ^$T/[dynamic]$E, #no_broadcast args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
 	return _append_elems(array, true, loc, ..args)
 }
 
 @builtin
-non_zero_append_elems :: proc(array: ^$T/[dynamic]$E, args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+non_zero_append_elems :: proc(array: ^$T/[dynamic]$E, #no_broadcast args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
 	return _append_elems(array, false, loc, ..args)
 }
 
@@ -556,7 +556,7 @@ append_nothing :: proc(array: ^$T/[dynamic]$E, loc := #caller_location) -> (n: i
 
 
 @builtin
-inject_at_elem :: proc(array: ^$T/[dynamic]$E, index: int, arg: E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+inject_at_elem :: proc(array: ^$T/[dynamic]$E, index: int, #no_broadcast arg: E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
 	if array == nil {
 		return
 	}
@@ -574,7 +574,7 @@ inject_at_elem :: proc(array: ^$T/[dynamic]$E, index: int, arg: E, loc := #calle
 }
 
 @builtin
-inject_at_elems :: proc(array: ^$T/[dynamic]$E, index: int, args: ..E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+inject_at_elems :: proc(array: ^$T/[dynamic]$E, index: int, #no_broadcast args: ..E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
 	if array == nil {
 		return
 	}
@@ -740,6 +740,9 @@ _resize_dynamic_array :: #force_inline proc(array: ^$T/[dynamic]$E, length: int,
 	a := (^Raw_Dynamic_Array)(array)
 
 	if length <= a.cap {
+		if should_zero && a.len < length {
+			intrinsics.mem_zero(([^]E)(a.data)[a.len:], (length-a.len)*size_of(E))
+		}
 		a.len = max(length, 0)
 		return nil
 	}
@@ -825,16 +828,18 @@ map_insert :: proc(m: ^$T/map[$K]$V, key: K, value: V, loc := #caller_location)
 }
 
 // Explicitly inserts a key and value into a map `m`, the same as `map_insert`, but the return values differ.
-// - `prev_key_ptr` will return the previous pointer of a key if it exists, and `nil` otherwise.
+// - `prev_key` will return the previous pointer of a key if it exists, check `found_previous` if was previously found
 // - `value_ptr` will return the pointer of the memory where the insertion happens, and `nil` if the map failed to resize
-// - `found_previous` will be true if `prev_key_ptr != nil`
-@(require_results)
-map_insert_and_check_for_previous :: proc(m: ^$T/map[$K]$V, key: K, value: V, loc := #caller_location) -> (prev_key_ptr: ^K, value_ptr: ^V, found_previous: bool) {
+// - `found_previous` will be true a previous key was found
+@(builtin, require_results)
+map_upsert :: proc(m: ^$T/map[$K]$V, key: K, value: V, loc := #caller_location) -> (prev_key: K, value_ptr: ^V, found_previous: bool) {
 	key, value := key, value
 	kp, vp := __dynamic_map_set_extra_without_hash((^Raw_Map)(m), map_info(T), rawptr(&key), rawptr(&value), loc)
-	prev_key_ptr   = (^K)(kp)
-	value_ptr      = (^V)(vp)
-	found_previous = kp != nil
+	if kp != nil {
+		prev_key = (^K)(kp)^
+		found_previous = true
+	}
+	value_ptr = (^V)(vp)
 	return
 }
 

base/runtime/core_builtin_soa.odin

@@ -425,4 +425,97 @@ clear_soa_dynamic_array :: proc(array: ^$T/#soa[dynamic]$E) {
 @builtin
 clear_soa :: proc{
 	clear_soa_dynamic_array,
-}
+}
+
+// Converts soa slice into a soa dynamic array without cloning or allocating memory
+@(require_results)
+into_dynamic_soa :: proc(array: $T/#soa[]$E) -> #soa[dynamic]E {
+	d: #soa[dynamic]E
+	footer := raw_soa_footer_dynamic_array(&d)
+	footer^ = {
+		cap = len(array),
+		len = 0,
+		allocator = nil_allocator(),
+	}
+
+	field_count: uintptr
+	when intrinsics.type_is_array(E) {
+		field_count = len(E)
+	} else {
+		field_count = uintptr(intrinsics.type_struct_field_count(E))
+	}
+
+	array := array
+	dynamic_data := ([^]rawptr)(&d)[:field_count]
+	slice_data   := ([^]rawptr)(&array)[:field_count]
+	copy(dynamic_data, slice_data)
+
+	return d
+}
+
+// `unordered_remove_soa` removed the element at the specified `index`. It does so by replacing the current end value
+// with the old value, and reducing the length of the dynamic array by 1.
+//
+// Note: This is an O(1) operation.
+// Note: If you the elements to remain in their order, use `ordered_remove_soa`.
+// Note: If the index is out of bounds, this procedure will panic.
+@builtin
+unordered_remove_soa :: proc(array: ^$T/#soa[dynamic]$E, index: int, loc := #caller_location) #no_bounds_check {
+	bounds_check_error_loc(loc, index, len(array))
+	if index+1 < len(array) {
+		ti := type_info_of(typeid_of(T))
+		ti = type_info_base(ti)
+		si := &ti.variant.(Type_Info_Struct)
+
+		field_count: uintptr
+		when intrinsics.type_is_array(E) {
+			field_count = len(E)
+		} else {
+			field_count = uintptr(intrinsics.type_struct_field_count(E))
+		}
+
+		data := uintptr(array)
+		for i in 0..<field_count {
+			type := si.types[i].variant.(Type_Info_Pointer).elem
+
+			offset := rawptr((^uintptr)(data)^ + uintptr(index*type.size))
+			final := rawptr((^uintptr)(data)^ + uintptr((len(array)-1)*type.size))
+			mem_copy(offset, final, type.size)
+			data += size_of(rawptr)
+		}
+	}
+	raw_soa_footer_dynamic_array(array).len -= 1
+}
+
+// `ordered_remove_soa` removed the element at the specified `index` whilst keeping the order of the other elements.
+//
+// Note: This is an O(N) operation.
+// Note: If you the elements do not have to remain in their order, prefer `unordered_remove_soa`.
+// Note: If the index is out of bounds, this procedure will panic.
+@builtin
+ordered_remove_soa :: proc(array: ^$T/#soa[dynamic]$E, index: int, loc := #caller_location) #no_bounds_check {
+	bounds_check_error_loc(loc, index, len(array))
+	if index+1 < len(array) {
+		ti := type_info_of(typeid_of(T))
+		ti = type_info_base(ti)
+		si := &ti.variant.(Type_Info_Struct)
+
+		field_count: uintptr
+		when intrinsics.type_is_array(E) {
+			field_count = len(E)
+		} else {
+			field_count = uintptr(intrinsics.type_struct_field_count(E))
+		}
+
+		data := uintptr(array)
+		for i in 0..<field_count {
+			type := si.types[i].variant.(Type_Info_Pointer).elem
+
+			offset := (^uintptr)(data)^ + uintptr(index*type.size)
+			length := type.size*(len(array) - index - 1)
+			mem_copy(rawptr(offset), rawptr(offset + uintptr(type.size)), length)
+			data += size_of(rawptr)
+		}
+	}
+	raw_soa_footer_dynamic_array(array).len -= 1
+}

base/runtime/dynamic_map_internal.odin

@@ -333,7 +333,7 @@ map_kvh_data_values_dynamic :: proc "contextless" (m: Raw_Map, #no_alias info: ^
 }
 
 
-@(private, require_results)
+@(require_results)
 map_total_allocation_size :: #force_inline proc "contextless" (capacity: uintptr, info: ^Map_Info) -> uintptr {
 	round :: #force_inline proc "contextless" (value: uintptr) -> uintptr {
 		CACHE_MASK :: MAP_CACHE_LINE_SIZE - 1
@@ -350,6 +350,12 @@ map_total_allocation_size :: #force_inline proc "contextless" (capacity: uintptr
 	return size
 }
 
+@(require_results)
+map_total_allocation_size_from_value :: #force_inline proc "contextless" (m: $M/map[$K]$V) -> uintptr {
+	return map_total_allocation_size(uintptr(cap(m)), map_info(M))
+}
+
+
 // The only procedure which needs access to the context is the one which allocates the map.
 @(require_results)
 map_alloc_dynamic :: proc "odin" (info: ^Map_Info, log2_capacity: uintptr, allocator := context.allocator, loc := #caller_location) -> (result: Raw_Map, err: Allocator_Error) {
@@ -391,7 +397,8 @@ map_alloc_dynamic :: proc "odin" (info: ^Map_Info, log2_capacity: uintptr, alloc
 // arrays to reduce variance. This swapping can only be done with memcpy since
 // there is no type information.
 //
-// This procedure returns the address of the just inserted value.
+// This procedure returns the address of the just inserted value, and will
+// return 'nil' if there was no room to insert the entry
 @(require_results)
 map_insert_hash_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, h: Map_Hash, ik: uintptr, iv: uintptr) -> (result: uintptr) {
 	h        := h
@@ -415,6 +422,11 @@ map_insert_hash_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^
 	tv := map_cell_index_dynamic(sv, info.vs, 1)
 
 	swap_loop: for {
+		if distance > mask {
+			// Failed to find an empty slot and prevent infinite loop
+			panic("unable to insert into a map")
+		}
+
 		element_hash := hs[pos]
 
 		if map_hash_is_empty(element_hash) {
@@ -898,7 +910,9 @@ __dynamic_map_set :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_In
 	}
 
 	result := map_insert_hash_dynamic(m, info, hash, uintptr(key), uintptr(value))
-	m.len += 1
+	if result != 0 {
+		m.len += 1
+	}
 	return rawptr(result)
 }
 __dynamic_map_set_extra_without_hash :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, key, value: rawptr, loc := #caller_location) -> (prev_key_ptr, value_ptr: rawptr) {
@@ -921,7 +935,9 @@ __dynamic_map_set_extra :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^
 	}
 
 	result := map_insert_hash_dynamic(m, info, hash, uintptr(key), uintptr(value))
-	m.len += 1
+	if result != 0 {
+		m.len += 1
+	}
 	return nil, rawptr(result)
 }
 

base/runtime/entry_unix.odin

@@ -1,5 +1,5 @@
 //+private
-//+build linux, darwin, freebsd, openbsd
+//+build linux, darwin, freebsd, openbsd, haiku
 //+no-instrumentation
 package runtime
 

base/runtime/heap_allocator_unix.odin

@@ -1,4 +1,4 @@
-//+build linux, darwin, freebsd, openbsd
+//+build linux, darwin, freebsd, openbsd, haiku
 //+private
 package runtime
 
@@ -35,4 +35,4 @@ _heap_resize :: proc(ptr: rawptr, new_size: int) -> rawptr {
 
 _heap_free :: proc(ptr: rawptr) {
 	_unix_free(ptr)
-}
+}

base/runtime/internal.odin

@@ -962,9 +962,11 @@ udivmodti4 :: proc "c" (a, b: u128, rem: ^u128) -> u128 {
 	return udivmod128(a, b, rem)
 }
 
-@(link_name="__udivti3", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
-udivti3 :: proc "c" (a, b: u128) -> u128 {
-	return udivmodti4(a, b, nil)
+when !IS_WASM {
+	@(link_name="__udivti3", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
+	udivti3 :: proc "c" (a, b: u128) -> u128 {
+		return udivmodti4(a, b, nil)
+	}
 }
 
 

base/runtime/os_specific_darwin.odin

@@ -5,7 +5,7 @@ package runtime
 import "base:intrinsics"
 
 _stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
-	WRITE  :: 0x20000004
+	WRITE  :: 0x2000004
 	STDERR :: 2
 	ret := intrinsics.syscall(WRITE, STDERR, uintptr(raw_data(data)), uintptr(len(data)))
 	if ret < 0 {

base/runtime/os_specific_haiku.odin

@@ -0,0 +1,21 @@
+//+build haiku
+//+private
+package runtime
+
+foreign import libc "system:c"
+
+foreign libc {
+	@(link_name="write")
+	_unix_write :: proc(fd: i32, buf: rawptr, size: int) -> int ---
+
+	_errnop :: proc() -> ^i32 ---
+}
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	ret := _unix_write(2, raw_data(data), len(data))
+	if ret < len(data) {
+		err := _errnop()
+		return int(ret), _OS_Errno(err^ if err != nil else 0)
+	}
+	return int(ret), 0
+}

base/runtime/procs_wasm.odin

@@ -7,19 +7,25 @@ ti_int :: struct #raw_union {
 	all: i128,
 }
 
+@(private="file")
+ti_uint :: struct #raw_union {
+	using s: struct { lo, hi: u64 },
+	all: u128,
+}
+
 @(link_name="__ashlti3", linkage="strong")
-__ashlti3 :: proc "contextless" (a: i128, b_: u32) -> i128 {
+__ashlti3 :: proc "contextless" (la, ha: u64, b_: u32) -> i128 {
 	bits_in_dword :: size_of(u32)*8
 	b := u32(b_)
 	
 	input, result: ti_int
-	input.all = a
+	input.lo, input.hi = la, ha
 	if b & bits_in_dword != 0 {
 		result.lo = 0
 		result.hi = input.lo << (b-bits_in_dword)
 	} else {
 		if b == 0 {
-			return a
+			return input.all
 		}
 		result.lo = input.lo<<b
 		result.hi = (input.hi<<b) | (input.lo>>(bits_in_dword-b))
@@ -29,12 +35,20 @@ __ashlti3 :: proc "contextless" (a: i128, b_: u32) -> i128 {
 
 
 @(link_name="__multi3", linkage="strong")
-__multi3 :: proc "contextless" (a, b: i128) -> i128 {
+__multi3 :: proc "contextless" (la, ha, lb, hb: u64) -> i128 {
 	x, y, r: ti_int
-	
-	x.all = a
-	y.all = b
+
+	x.lo, x.hi = la, ha
+	y.lo, y.hi = lb, hb
 	r.all = i128(x.lo * y.lo) // TODO this is incorrect
 	r.hi += x.hi*y.lo + x.lo*y.hi
 	return r.all
-}
+}
+
+@(link_name="__udivti3", linkage="strong")
+udivti3 :: proc "c" (la, ha, lb, hb: u64) -> u128 {
+	a, b: ti_uint
+	a.lo, a.hi = la, ha
+	b.lo, b.hi = lb, hb
+	return udivmodti4(a.all, b.all, nil)
+}

build_odin.sh

@@ -56,7 +56,7 @@ fi
 
 case "$OS_NAME" in
 Darwin)
-	if [ "$OS_ARCH" == "arm64" ]; then
+	if [ "$OS_ARCH" = "arm64" ]; then
 		if [ $LLVM_VERSION_MAJOR -lt 13 ] || [ $LLVM_VERSION_MAJOR -gt 17 ]; then
 			error "Darwin Arm64 requires LLVM 13, 14 or 17"
 		fi
@@ -82,6 +82,11 @@ OpenBSD)
 	LDFLAGS="$LDFLAGS -liconv"
 	LDFLAGS="$LDFLAGS $($LLVM_CONFIG --libs core native --system-libs)"
 	;;
+Haiku)
+	CXXFLAGS="$CXXFLAGS $($LLVM_CONFIG --cxxflags --ldflags) -I/system/develop/headers/private/shared -I/system/develop/headers/private/kernel"
+	LDFLAGS="$LDFLAGS -liconv"
+	LDFLAGS="$LDFLAGS $($LLVM_CONFIG --libs core native --system-libs)"
+	;;
 *)
 	error "Platform \"$OS_NAME\" unsupported"
 	;;
@@ -96,7 +101,7 @@ build_odin() {
 		EXTRAFLAGS="-O3"
 		;;
 	release-native)
-		if [ "$OS_ARCH" == "arm64" ]; then
+		if [ "$OS_ARCH" = "arm64" ]; then
 			# Use preferred flag for Arm (ie arm64 / aarch64 / etc)
 			EXTRAFLAGS="-O3 -mcpu=native"
 		else

core/bufio/writer.odin

@@ -226,7 +226,6 @@ writer_to_writer :: proc(b: ^Writer) -> (s: io.Writer) {
 
 
 
-@(private)
 _writer_proc :: proc(stream_data: rawptr, mode: io.Stream_Mode, p: []byte, offset: i64, whence: io.Seek_From) -> (n: i64, err: io.Error) {
 	b := (^Writer)(stream_data)
 	#partial switch mode {

core/c/c.odin

@@ -104,3 +104,13 @@ NULL           :: rawptr(uintptr(0))
 NDEBUG         :: !ODIN_DEBUG
 
 CHAR_BIT :: 8
+
+// Since there are no types in C with an alignment larger than that of
+// max_align_t, which cannot be larger than sizeof(long double) as any other
+// exposed type wouldn't be valid C, the maximum alignment possible in a
+// strictly conformant C implementation is 16 on the platforms we care about.
+// The choice of 4096 bytes for storage of this type is more than enough on all
+// relevant platforms.
+va_list :: struct #align(16) {
+	_: [4096]u8,
+}

core/c/libc/errno.odin

@@ -80,6 +80,24 @@ when ODIN_OS == .Darwin {
 	ERANGE :: 34
 }
 
+when ODIN_OS == .Haiku {
+	@(private="file")
+	@(default_calling_convention="c")
+	foreign libc {
+		@(link_name="_errnop")
+		_get_errno :: proc() -> ^int ---
+	}
+
+	@(private="file")
+	B_GENERAL_ERROR_BASE :: min(i32)
+	@(private="file")
+	B_POSIX_ERROR_BASE   :: B_GENERAL_ERROR_BASE + 0x7000
+
+	EDOM   :: B_POSIX_ERROR_BASE + 16
+	EILSEQ :: B_POSIX_ERROR_BASE + 38
+	ERANGE :: B_POSIX_ERROR_BASE + 17
+}
+
 // Odin has no way to make an identifier "errno" behave as a function call to
 // read the value, or to produce an lvalue such that you can assign a different
 // error value to errno. To work around this, just expose it as a function like

core/c/libc/stdarg.odin

@@ -4,6 +4,8 @@ package libc
 
 import "base:intrinsics"
 
+import "core:c"
+
 @(private="file")
 @(default_calling_convention="none")
 foreign _ {
@@ -12,15 +14,7 @@ foreign _ {
 	@(link_name="llvm.va_copy")  _va_copy  :: proc(dst, src: ^i8) ---
 }
 
-// Since there are no types in C with an alignment larger than that of
-// max_align_t, which cannot be larger than sizeof(long double) as any other
-// exposed type wouldn't be valid C, the maximum alignment possible in a
-// strictly conformant C implementation is 16 on the platforms we care about.
-// The choice of 4096 bytes for storage of this type is more than enough on all
-// relevant platforms.
-va_list :: struct #align(16) {
-	_: [4096]u8,
-}
+va_list :: c.va_list
 
 va_start :: #force_inline proc(ap: ^va_list, _: any) {
 	_va_start(cast(^i8)ap)

core/c/libc/stdio.odin

@@ -163,6 +163,36 @@ when ODIN_OS == .Darwin {
 	}
 }
 
+when ODIN_OS == .Haiku {
+	fpos_t :: distinct i64
+	
+	_IOFBF        :: 0
+	_IOLBF        :: 1
+	_IONBF        :: 2
+
+	BUFSIZ        :: 8192
+
+	EOF           :: int(-1)
+
+	FOPEN_MAX     :: 128
+
+	FILENAME_MAX  :: 256
+
+	L_tmpnam      :: 512
+
+	SEEK_SET      :: 0
+	SEEK_CUR      :: 1
+	SEEK_END      :: 2
+
+	TMP_MAX       :: 32768
+
+	foreign libc {
+		stderr: ^FILE
+		stdin:  ^FILE
+		stdout: ^FILE
+	}
+}
+
 @(default_calling_convention="c")
 foreign libc {
 	// 7.21.4 Operations on files

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 {
+when ODIN_OS == .Linux || ODIN_OS == .FreeBSD || ODIN_OS == .Darwin || ODIN_OS == .OpenBSD || ODIN_OS == .Haiku {
 	@(default_calling_convention="c")
 	foreign libc {
 		// 7.27.2 Time manipulation functions

core/c/libc/wctype.odin

@@ -29,7 +29,11 @@ when ODIN_OS == .Windows {
 } else when ODIN_OS == .FreeBSD {
 	wctrans_t :: distinct int
 	wctype_t  :: distinct ulong
-	
+
+} else when ODIN_OS == .Haiku {
+	wctrans_t :: distinct i32
+	wctype_t  :: distinct i32
+
 }
 
 @(default_calling_convention="c")

core/crypto/README.md

@@ -14,6 +14,14 @@ constant-time byte comparison.
 - Best-effort is make to mitigate timing side-channels on reasonable
   architectures.  Architectures that are known to be unreasonable include
   but are not limited to i386, i486, and WebAssembly.
+- Implementations assume a 64-bit architecture (64-bit integer arithmetic
+  is fast, and includes add-with-carry, sub-with-borrow, and full-result
+  multiply).
+- Hardware sidechannels are explicitly out of scope for this package.
+  Notable examples include but are not limited to:
+  - Power/RF side-channels etc.
+  - Fault injection attacks etc.
+  - Hardware vulnerabilities ("apply mitigations or buy a new CPU").
 - The packages attempt to santize sensitive data, however this is, and
   will remain a "best-effort" implementation decision.  As Thomas Pornin
   puts it "In general, such memory cleansing is a fool's quest."

core/crypto/_edwards25519/edwards25519.odin

@@ -0,0 +1,428 @@
+package _edwards25519
+
+/*
+This implements the edwards25519 composite-order group, primarily for
+the purpose of implementing X25519, Ed25519, and ristretto255.  Use of
+this package for other purposes is NOT RECOMMENDED.
+
+See:
+- https://eprint.iacr.org/2011/368.pdf
+- https://datatracker.ietf.org/doc/html/rfc8032
+- https://www.hyperelliptic.org/EFD/g1p/auto-twisted-extended-1.html
+*/
+
+import "base:intrinsics"
+import "core:crypto"
+import field "core:crypto/_fiat/field_curve25519"
+import "core:mem"
+
+// Group_Element is an edwards25519 group element, as extended homogenous
+// coordinates, which represents the affine point `(x, y)` as `(X, Y, Z, T)`,
+// with the relations `x = X/Z`, `y = Y/Z`, and `x * y = T/Z`.
+//
+// d = -121665/121666 = 37095705934669439343138083508754565189542113879843219016388785533085940283555
+// a = -1
+//
+// Notes:
+// - There is considerable scope for optimization, however that
+//   will not change the external API, and this is simple and reasonably
+//   performant.
+// - The API delibarately makes it hard to create arbitrary group
+//   elements that are not on the curve.
+// - The group element decoding routine takes the opinionated stance of
+//   rejecting non-canonical encodings.
+
+FE_D := field.Tight_Field_Element {
+	929955233495203,
+	466365720129213,
+	1662059464998953,
+	2033849074728123,
+	1442794654840575,
+}
+@(private)
+FE_A := field.Tight_Field_Element {
+	2251799813685228,
+	2251799813685247,
+	2251799813685247,
+	2251799813685247,
+	2251799813685247,
+}
+@(private)
+FE_D2 := field.Tight_Field_Element {
+	1859910466990425,
+	932731440258426,
+	1072319116312658,
+	1815898335770999,
+	633789495995903,
+}
+@(private)
+GE_BASEPOINT := Group_Element {
+	field.Tight_Field_Element {
+		1738742601995546,
+		1146398526822698,
+		2070867633025821,
+		562264141797630,
+		587772402128613,
+	},
+	field.Tight_Field_Element {
+		1801439850948184,
+		1351079888211148,
+		450359962737049,
+		900719925474099,
+		1801439850948198,
+	},
+	field.Tight_Field_Element{1, 0, 0, 0, 0},
+	field.Tight_Field_Element {
+		1841354044333475,
+		16398895984059,
+		755974180946558,
+		900171276175154,
+		1821297809914039,
+	},
+}
+GE_IDENTITY := Group_Element {
+	field.Tight_Field_Element{0, 0, 0, 0, 0},
+	field.Tight_Field_Element{1, 0, 0, 0, 0},
+	field.Tight_Field_Element{1, 0, 0, 0, 0},
+	field.Tight_Field_Element{0, 0, 0, 0, 0},
+}
+
+Group_Element :: struct {
+	x: field.Tight_Field_Element,
+	y: field.Tight_Field_Element,
+	z: field.Tight_Field_Element,
+	t: field.Tight_Field_Element,
+}
+
+ge_clear :: proc "contextless" (ge: ^Group_Element) {
+	mem.zero_explicit(ge, size_of(Group_Element))
+}
+
+ge_set :: proc "contextless" (ge, a: ^Group_Element) {
+	field.fe_set(&ge.x, &a.x)
+	field.fe_set(&ge.y, &a.y)
+	field.fe_set(&ge.z, &a.z)
+	field.fe_set(&ge.t, &a.t)
+}
+
+@(require_results)
+ge_set_bytes :: proc "contextless" (ge: ^Group_Element, b: []byte) -> bool {
+	if len(b) != 32 {
+		intrinsics.trap()
+	}
+	b_ := transmute(^[32]byte)(raw_data(b))
+
+	// Do the work in a scratch element, so that ge is unchanged on
+	// failure.
+	tmp: Group_Element = ---
+	defer ge_clear(&tmp)
+	field.fe_one(&tmp.z) // Z = 1
+
+	// The encoding is the y-coordinate, with the x-coordinate polarity
+	// (odd/even) encoded in the MSB.
+	field.fe_from_bytes(&tmp.y, b_) // ignores high bit
+
+	// Recover the candidate x-coordinate via the curve equation:
+	// x^2 = (y^2 - 1) / (d * y^2 + 1) (mod p)
+
+	fe_tmp := &tmp.t // Use this to store intermediaries.
+	fe_one := &tmp.z
+
+	// x = num = y^2 - 1
+	field.fe_carry_square(fe_tmp, field.fe_relax_cast(&tmp.y)) // fe_tmp = y^2
+	field.fe_carry_sub(&tmp.x, fe_tmp, fe_one)
+
+	// den = d * y^2 + 1
+	field.fe_carry_mul(fe_tmp, field.fe_relax_cast(fe_tmp), field.fe_relax_cast(&FE_D))
+	field.fe_carry_add(fe_tmp, fe_tmp, fe_one)
+
+	// x = invsqrt(den/num)
+	is_square := field.fe_carry_sqrt_ratio_m1(
+		&tmp.x,
+		field.fe_relax_cast(&tmp.x),
+		field.fe_relax_cast(fe_tmp),
+	)
+	if is_square == 0 {
+		return false
+	}
+
+	// Pick the right x-coordinate.
+	field.fe_cond_negate(&tmp.x, &tmp.x, int(b[31] >> 7))
+
+	// t = x * y
+	field.fe_carry_mul(&tmp.t, field.fe_relax_cast(&tmp.x), field.fe_relax_cast(&tmp.y))
+
+	// Reject non-canonical encodings of ge.
+	buf: [32]byte = ---
+	field.fe_to_bytes(&buf, &tmp.y)
+	buf[31] |= byte(field.fe_is_negative(&tmp.x)) << 7
+	is_canonical := crypto.compare_constant_time(b, buf[:])
+
+	ge_cond_assign(ge, &tmp, is_canonical)
+
+	mem.zero_explicit(&buf, size_of(buf))
+
+	return is_canonical == 1
+}
+
+ge_bytes :: proc "contextless" (ge: ^Group_Element, dst: []byte) {
+	if len(dst) != 32 {
+		intrinsics.trap()
+	}
+	dst_ := transmute(^[32]byte)(raw_data(dst))
+
+	// Convert the element to affine (x, y) representation.
+	x, y, z_inv: field.Tight_Field_Element = ---, ---, ---
+	field.fe_carry_inv(&z_inv, field.fe_relax_cast(&ge.z))
+	field.fe_carry_mul(&x, field.fe_relax_cast(&ge.x), field.fe_relax_cast(&z_inv))
+	field.fe_carry_mul(&y, field.fe_relax_cast(&ge.y), field.fe_relax_cast(&z_inv))
+
+	// Encode the y-coordinate.
+	field.fe_to_bytes(dst_, &y)
+
+	// Copy the least significant bit of the x-coordinate to the most
+	// significant bit of the encoded y-coordinate.
+	dst_[31] |= byte((x[0] & 1) << 7)
+
+	field.fe_clear_vec([]^field.Tight_Field_Element{&x, &y, &z_inv})
+}
+
+ge_identity :: proc "contextless" (ge: ^Group_Element) {
+	field.fe_zero(&ge.x)
+	field.fe_one(&ge.y)
+	field.fe_one(&ge.z)
+	field.fe_zero(&ge.t)
+}
+
+ge_generator :: proc "contextless" (ge: ^Group_Element) {
+	ge_set(ge, &GE_BASEPOINT)
+}
+
+@(private)
+Addend_Group_Element :: struct {
+	y2_minus_x2:  field.Loose_Field_Element, // t1
+	y2_plus_x2:   field.Loose_Field_Element, // t3
+	k_times_t2:   field.Tight_Field_Element, // t4
+	two_times_z2: field.Loose_Field_Element, // t5
+}
+
+@(private)
+ge_addend_set :: proc "contextless" (ge_a: ^Addend_Group_Element, ge: ^Group_Element) {
+	field.fe_sub(&ge_a.y2_minus_x2, &ge.y, &ge.x)
+	field.fe_add(&ge_a.y2_plus_x2, &ge.y, &ge.x)
+	field.fe_carry_mul(&ge_a.k_times_t2, field.fe_relax_cast(&FE_D2), field.fe_relax_cast(&ge.t))
+	field.fe_add(&ge_a.two_times_z2, &ge.z, &ge.z)
+}
+
+@(private)
+ge_addend_conditional_assign :: proc "contextless" (ge_a, a: ^Addend_Group_Element, ctrl: int) {
+	field.fe_cond_select(&ge_a.y2_minus_x2, &ge_a.y2_minus_x2, &a.y2_minus_x2, ctrl)
+	field.fe_cond_select(&ge_a.y2_plus_x2, &ge_a.y2_plus_x2, &a.y2_plus_x2, ctrl)
+	field.fe_cond_select(&ge_a.k_times_t2, &ge_a.k_times_t2, &a.k_times_t2, ctrl)
+	field.fe_cond_select(&ge_a.two_times_z2, &ge_a.two_times_z2, &a.two_times_z2, ctrl)
+}
+
+@(private)
+Add_Scratch :: struct {
+	A, B, C, D: field.Tight_Field_Element,
+	E, F, G, H: field.Loose_Field_Element,
+	t0, t2:     field.Loose_Field_Element,
+}
+
+ge_add :: proc "contextless" (ge, a, b: ^Group_Element) {
+	b_: Addend_Group_Element = ---
+	ge_addend_set(&b_, b)
+
+	scratch: Add_Scratch = ---
+	ge_add_addend(ge, a, &b_, &scratch)
+
+	mem.zero_explicit(&b_, size_of(Addend_Group_Element))
+	mem.zero_explicit(&scratch, size_of(Add_Scratch))
+}
+
+@(private)
+ge_add_addend :: proc "contextless" (
+	ge, a: ^Group_Element,
+	b: ^Addend_Group_Element,
+	scratch: ^Add_Scratch,
+) {
+	// https://www.hyperelliptic.org/EFD/g1p/auto-twisted-extended-1.html#addition-add-2008-hwcd-3
+	// Assumptions: k=2*d.
+	//
+	// t0 = Y1-X1
+	// t1 = Y2-X2
+	// A = t0*t1
+	// t2 = Y1+X1
+	// t3 = Y2+X2
+	// B = t2*t3
+	// t4 = k*T2
+	// C = T1*t4
+	// t5 = 2*Z2
+	// D = Z1*t5
+	// E = B-A
+	// F = D-C
+	// G = D+C
+	// H = B+A
+	// X3 = E*F
+	// Y3 = G*H
+	// T3 = E*H
+	// Z3 = F*G
+	//
+	// In order to make the scalar multiply faster, the addend is provided
+	// as a `Addend_Group_Element` with t1, t3, t4, and t5 precomputed, as
+	// it is trivially obvious that those are the only values used by the
+	// formula that are directly dependent on `b`, and are only dependent
+	// on `b` and constants.  This saves 1 sub, 2 adds, and 1 multiply,
+	// each time the intermediate representation can be reused.
+
+	A, B, C, D := &scratch.A, &scratch.B, &scratch.C, &scratch.D
+	E, F, G, H := &scratch.E, &scratch.F, &scratch.G, &scratch.H
+	t0, t2 := &scratch.t0, &scratch.t2
+
+	field.fe_sub(t0, &a.y, &a.x)
+	t1 := &b.y2_minus_x2
+	field.fe_carry_mul(A, t0, t1)
+	field.fe_add(t2, &a.y, &a.x)
+	t3 := &b.y2_plus_x2
+	field.fe_carry_mul(B, t2, t3)
+	t4 := &b.k_times_t2
+	field.fe_carry_mul(C, field.fe_relax_cast(&a.t), field.fe_relax_cast(t4))
+	t5 := &b.two_times_z2
+	field.fe_carry_mul(D, field.fe_relax_cast(&a.z), t5)
+	field.fe_sub(E, B, A)
+	field.fe_sub(F, D, C)
+	field.fe_add(G, D, C)
+	field.fe_add(H, B, A)
+	field.fe_carry_mul(&ge.x, E, F)
+	field.fe_carry_mul(&ge.y, G, H)
+	field.fe_carry_mul(&ge.t, E, H)
+	field.fe_carry_mul(&ge.z, F, G)
+}
+
+@(private)
+Double_Scratch :: struct {
+	A, B, C, D, G: field.Tight_Field_Element,
+	t0, t2, t3:    field.Tight_Field_Element,
+	E, F, H:       field.Loose_Field_Element,
+	t1:            field.Loose_Field_Element,
+}
+
+ge_double :: proc "contextless" (ge, a: ^Group_Element, scratch: ^Double_Scratch = nil) {
+	// https://www.hyperelliptic.org/EFD/g1p/auto-twisted-extended-1.html#doubling-dbl-2008-hwcd
+	//
+	// A = X1^2
+	// B = Y1^2
+	// t0 = Z1^2
+	// C = 2*t0
+	// D = a*A
+	// t1 = X1+Y1
+	// t2 = t1^2
+	// t3 = t2-A
+	// E = t3-B
+	// G = D+B
+	// F = G-C
+	// H = D-B
+	// X3 = E*F
+	// Y3 = G*H
+	// T3 = E*H
+	// Z3 = F*G
+
+	sanitize, scratch := scratch == nil, scratch
+	if sanitize {
+		tmp: Double_Scratch = ---
+		scratch = &tmp
+	}
+
+	A, B, C, D, G := &scratch.A, &scratch.B, &scratch.C, &scratch.D, &scratch.G
+	t0, t2, t3 := &scratch.t0, &scratch.t2, &scratch.t3
+	E, F, H := &scratch.E, &scratch.F, &scratch.H
+	t1 := &scratch.t1
+
+	field.fe_carry_square(A, field.fe_relax_cast(&a.x))
+	field.fe_carry_square(B, field.fe_relax_cast(&a.y))
+	field.fe_carry_square(t0, field.fe_relax_cast(&a.z))
+	field.fe_carry_add(C, t0, t0)
+	field.fe_carry_mul(D, field.fe_relax_cast(&FE_A), field.fe_relax_cast(A))
+	field.fe_add(t1, &a.x, &a.y)
+	field.fe_carry_square(t2, t1)
+	field.fe_carry_sub(t3, t2, A)
+	field.fe_sub(E, t3, B)
+	field.fe_carry_add(G, D, B)
+	field.fe_sub(F, G, C)
+	field.fe_sub(H, D, B)
+	G_ := field.fe_relax_cast(G)
+	field.fe_carry_mul(&ge.x, E, F)
+	field.fe_carry_mul(&ge.y, G_, H)
+	field.fe_carry_mul(&ge.t, E, H)
+	field.fe_carry_mul(&ge.z, F, G_)
+
+	if sanitize {
+		mem.zero_explicit(scratch, size_of(Double_Scratch))
+	}
+}
+
+ge_negate :: proc "contextless" (ge, a: ^Group_Element) {
+	field.fe_carry_opp(&ge.x, &a.x)
+	field.fe_set(&ge.y, &a.y)
+	field.fe_set(&ge.z, &a.z)
+	field.fe_carry_opp(&ge.t, &a.t)
+}
+
+ge_cond_negate :: proc "contextless" (ge, a: ^Group_Element, ctrl: int) {
+	tmp: Group_Element = ---
+	ge_negate(&tmp, a)
+	ge_cond_assign(ge, &tmp, ctrl)
+
+	ge_clear(&tmp)
+}
+
+ge_cond_assign :: proc "contextless" (ge, a: ^Group_Element, ctrl: int) {
+	field.fe_cond_assign(&ge.x, &a.x, ctrl)
+	field.fe_cond_assign(&ge.y, &a.y, ctrl)
+	field.fe_cond_assign(&ge.z, &a.z, ctrl)
+	field.fe_cond_assign(&ge.t, &a.t, ctrl)
+}
+
+ge_cond_select :: proc "contextless" (ge, a, b: ^Group_Element, ctrl: int) {
+	field.fe_cond_select(&ge.x, &a.x, &b.x, ctrl)
+	field.fe_cond_select(&ge.y, &a.y, &b.y, ctrl)
+	field.fe_cond_select(&ge.z, &a.z, &b.z, ctrl)
+	field.fe_cond_select(&ge.t, &a.t, &b.t, ctrl)
+}
+
+@(require_results)
+ge_equal :: proc "contextless" (a, b: ^Group_Element) -> int {
+	// (x, y) ?= (x', y') -> (X/Z, Y/Z) ?= (X'/Z', Y'/Z')
+	// X/Z ?= X'/Z', Y/Z ?= Y'/Z' -> X*Z' ?= X'*Z, Y*Z' ?= Y'*Z
+	ax_bz, bx_az, ay_bz, by_az: field.Tight_Field_Element = ---, ---, ---, ---
+	field.fe_carry_mul(&ax_bz, field.fe_relax_cast(&a.x), field.fe_relax_cast(&b.z))
+	field.fe_carry_mul(&bx_az, field.fe_relax_cast(&b.x), field.fe_relax_cast(&a.z))
+	field.fe_carry_mul(&ay_bz, field.fe_relax_cast(&a.y), field.fe_relax_cast(&b.z))
+	field.fe_carry_mul(&by_az, field.fe_relax_cast(&b.y), field.fe_relax_cast(&a.z))
+
+	ret := field.fe_equal(&ax_bz, &bx_az) & field.fe_equal(&ay_bz, &by_az)
+
+	field.fe_clear_vec([]^field.Tight_Field_Element{&ax_bz, &ay_bz, &bx_az, &by_az})
+
+	return ret
+}
+
+@(require_results)
+ge_is_small_order :: proc "contextless" (ge: ^Group_Element) -> bool {
+	tmp: Group_Element = ---
+	ge_double(&tmp, ge)
+	ge_double(&tmp, &tmp)
+	ge_double(&tmp, &tmp)
+	return ge_equal(&tmp, &GE_IDENTITY) == 1
+}
+
+@(require_results)
+ge_in_prime_order_subgroup_vartime :: proc "contextless" (ge: ^Group_Element) -> bool {
+	// This is currently *very* expensive.  The faster method would be
+	// something like (https://eprint.iacr.org/2022/1164.pdf), however
+	// that is a ~50% speedup, and a lot of added complexity for something
+	// that is better solved by "just use ristretto255".
+	tmp: Group_Element = ---
+	_ge_scalarmult(&tmp, ge, &SC_ELL, true)
+	return ge_equal(&tmp, &GE_IDENTITY) == 1
+}

core/crypto/_edwards25519/edwards25519_scalar.odin

@@ -0,0 +1,61 @@
+package _edwards25519
+
+import "base:intrinsics"
+import field "core:crypto/_fiat/field_scalar25519"
+import "core:mem"
+
+Scalar :: field.Montgomery_Domain_Field_Element
+
+// WARNING: This is non-canonical and only to be used when checking if
+// a group element is on the prime-order subgroup.
+@(private)
+SC_ELL := field.Non_Montgomery_Domain_Field_Element {
+	field.ELL[0],
+	field.ELL[1],
+	field.ELL[2],
+	field.ELL[3],
+}
+
+sc_set_u64 :: proc "contextless" (sc: ^Scalar, i: u64) {
+	tmp := field.Non_Montgomery_Domain_Field_Element{i, 0, 0, 0}
+	field.fe_to_montgomery(sc, &tmp)
+
+	mem.zero_explicit(&tmp, size_of(tmp))
+}
+
+@(require_results)
+sc_set_bytes :: proc "contextless" (sc: ^Scalar, b: []byte) -> bool {
+	if len(b) != 32 {
+		intrinsics.trap()
+	}
+	b_ := transmute(^[32]byte)(raw_data(b))
+	return field.fe_from_bytes(sc, b_)
+}
+
+sc_set_bytes_rfc8032 :: proc "contextless" (sc: ^Scalar, b: []byte) {
+	if len(b) != 32 {
+		intrinsics.trap()
+	}
+	b_ := transmute(^[32]byte)(raw_data(b))
+	field.fe_from_bytes_rfc8032(sc, b_)
+}
+
+sc_clear :: proc "contextless" (sc: ^Scalar) {
+	mem.zero_explicit(sc, size_of(Scalar))
+}
+
+sc_set :: field.fe_set
+sc_set_bytes_wide :: field.fe_from_bytes_wide
+sc_bytes :: field.fe_to_bytes
+
+sc_zero :: field.fe_zero
+sc_one :: field.fe_one
+
+sc_add :: field.fe_add
+sc_sub :: field.fe_sub
+sc_negate :: field.fe_opp
+sc_mul :: field.fe_mul
+sc_square :: field.fe_square
+
+sc_cond_assign :: field.fe_cond_assign
+sc_equal :: field.fe_equal

core/crypto/_edwards25519/edwards25519_scalar_mul.odin

@@ -0,0 +1,288 @@
+package _edwards25519
+
+import field "core:crypto/_fiat/field_scalar25519"
+import "core:math/bits"
+import "core:mem"
+
+// GE_BASEPOINT_TABLE is 1 * G, ... 15 * G, in precomputed format.
+//
+// Note: When generating, the values were reduced to Tight_Field_Element
+// ranges, even though that is not required.
+@(private)
+GE_BASEPOINT_TABLE := Multiply_Table {
+	{
+		{62697248952638, 204681361388450, 631292143396476, 338455783676468, 1213667448819585},
+		{1288382639258501, 245678601348599, 269427782077623, 1462984067271730, 137412439391563},
+		{301289933810280, 1259582250014073, 1422107436869536, 796239922652654, 1953934009299142},
+		{2, 0, 0, 0, 0},
+	},
+	{
+		{1519297034332653, 1098796920435767, 1823476547744119, 808144629470969, 2110930855619772},
+		{338005982828284, 1667856962156925, 100399270107451, 1604566703601691, 1950338038771369},
+		{1920505767731247, 1443759578976892, 1659852098357048, 1484431291070208, 275018744912646},
+		{763163817085987, 2195095074806923, 2167883174351839, 1868059999999762, 911071066608705},
+	},
+	{
+		{960627541894068, 1314966688943942, 1126875971034044, 2059608312958945, 605975666152586},
+		{1714478358025626, 2209607666607510, 1600912834284834, 496072478982142, 481970031861896},
+		{851735079403194, 1088965826757164, 141569479297499, 602804610059257, 2004026468601520},
+		{197585529552380, 324719066578543, 564481854250498, 1173818332764578, 35452976395676},
+	},
+	{
+		{1152980410747203, 2196804280851952, 25745194962557, 1915167295473129, 1266299690309224},
+		{809905889679060, 979732230071345, 1509972345538142, 188492426534402, 818965583123815},
+		{997685409185036, 1451818320876327, 2126681166774509, 2000509606057528, 235432372486854},
+		{887734189279642, 1460338685162044, 877378220074262, 102436391401299, 153369156847490},
+	},
+	{
+		{2056621900836770, 1821657694132497, 1627986892909426, 1163363868678833, 1108873376459226},
+		{1187697490593623, 1066539945237335, 885654531892000, 1357534489491782, 359370291392448},
+		{1509033452137525, 1305318174298508, 613642471748944, 1987256352550234, 1044283663101541},
+		{220105720697037, 387661783287620, 328296827867762, 360035589590664, 795213236824054},
+	},
+	{
+		{1820794733038396, 1612235121681074, 757405923441402, 1094031020892801, 231025333128907},
+		{1639067873254194, 1484176557946322, 300800382144789, 1329915446659183, 1211704578730455},
+		{641900794791527, 1711751746971612, 179044712319955, 576455585963824, 1852617592509865},
+		{743549047192397, 685091042550147, 1952415336873496, 1965124675654685, 513364998442917},
+	},
+	{
+		{1004557076870448, 1762911374844520, 1330807633622723, 384072910939787, 953849032243810},
+		{2178275058221458, 257933183722891, 376684351537894, 2010189102001786, 1981824297484148},
+		{1332915663881114, 1286540505502549, 1741691283561518, 977214932156314, 1764059494778091},
+		{429702949064027, 1368332611650677, 2019867176450999, 2212258376161746, 526160996742554},
+	},
+	{
+		{2098932988258576, 2203688382075948, 2120400160059479, 1748488020948146, 1203264167282624},
+		{677131386735829, 1850249298025188, 672782146532031, 2144145693078904, 2088656272813787},
+		{1065622343976192, 1573853211848116, 223560413590068, 333846833073379, 27832122205830},
+		{1781008836504573, 917619542051793, 544322748939913, 882577394308384, 1720521246471195},
+	},
+	{
+		{660120928379860, 2081944024858618, 1878411111349191, 424587356517195, 2111317439894005},
+		{1834193977811532, 1864164086863319, 797334633289424, 150410812403062, 2085177078466389},
+		{1438117271371866, 783915531014482, 388731514584658, 292113935417795, 1945855002546714},
+		{1678140823166658, 679103239148744, 614102761596238, 1052962498997885, 1863983323810390},
+	},
+	{
+		{1690309392496233, 1116333140326275, 1377242323631039, 717196888780674, 82724646713353},
+		{1722370213432106, 74265192976253, 264239578448472, 1714909985012994, 2216984958602173},
+		{2010482366920922, 1294036471886319, 566466395005815, 1631955803657320, 1751698647538458},
+		{1073230604155753, 1159087041338551, 1664057985455483, 127472702826203, 1339591128522371},
+	},
+	{
+		{478053307175577, 2179515791720985, 21146535423512, 1831683844029536, 462805561553981},
+		{1945267486565588, 1298536818409655, 2214511796262989, 1904981051429012, 252904800782086},
+		{268945954671210, 222740425595395, 1208025911856230, 1080418823003555, 75929831922483},
+		{1884784014268948, 643868448202966, 978736549726821, 46385971089796, 1296884812292320},
+	},
+	{
+		{1861159462859103, 7077532564710, 963010365896826, 1938780006785270, 766241051941647},
+		{1778966986051906, 1713995999765361, 1394565822271816, 1366699246468722, 1213407027149475},
+		{1978989286560907, 2135084162045594, 1951565508865477, 671788336314416, 293123929458176},
+		{902608944504080, 2167765718046481, 1285718473078022, 1222562171329269, 492109027844479},
+	},
+	{
+		{1820807832746213, 1029220580458586, 1101997555432203, 1039081975563572, 202477981158221},
+		{1866134980680205, 2222325502763386, 1830284629571201, 1046966214478970, 418381946936795},
+		{1783460633291322, 1719505443254998, 1810489639976220, 877049370713018, 2187801198742619},
+		{197118243000763, 305493867565736, 518814410156522, 1656246186645170, 901894734874934},
+	},
+	{
+		{225454942125915, 478410476654509, 600524586037746, 643450007230715, 1018615928259319},
+		{1733330584845708, 881092297970296, 507039890129464, 496397090721598, 2230888519577628},
+		{690155664737246, 1010454785646677, 753170144375012, 1651277613844874, 1622648796364156},
+		{1321310321891618, 1089655277873603, 235891750867089, 815878279563688, 1709264240047556},
+	},
+	{
+		{805027036551342, 1387174275567452, 1156538511461704, 1465897486692171, 1208567094120903},
+		{2228417017817483, 202885584970535, 2182114782271881, 2077405042592934, 1029684358182774},
+		{460447547653983, 627817697755692, 524899434670834, 1228019344939427, 740684787777653},
+		{849757462467675, 447476306919899, 422618957298818, 302134659227815, 675831828440895},
+	},
+}
+
+ge_scalarmult :: proc "contextless" (ge, p: ^Group_Element, sc: ^Scalar) {
+	tmp: field.Non_Montgomery_Domain_Field_Element
+	field.fe_from_montgomery(&tmp, sc)
+
+	_ge_scalarmult(ge, p, &tmp)
+
+	mem.zero_explicit(&tmp, size_of(tmp))
+}
+
+ge_scalarmult_basepoint :: proc "contextless" (ge: ^Group_Element, sc: ^Scalar) {
+	// Something like the comb method from "Fast and compact elliptic-curve
+	// cryptography" Section 3.3, would be more performant, but more
+	// complex.
+	//
+	// - https://eprint.iacr.org/2012/309
+	ge_scalarmult(ge, &GE_BASEPOINT, sc)
+}
+
+ge_scalarmult_vartime :: proc "contextless" (ge, p: ^Group_Element, sc: ^Scalar) {
+	tmp: field.Non_Montgomery_Domain_Field_Element
+	field.fe_from_montgomery(&tmp, sc)
+
+	_ge_scalarmult(ge, p, &tmp, true)
+}
+
+ge_double_scalarmult_basepoint_vartime :: proc "contextless" (
+	ge: ^Group_Element,
+	a: ^Scalar,
+	A: ^Group_Element,
+	b: ^Scalar,
+) {
+	// Strauss-Shamir, commonly referred to as the "Shamir trick",
+	// saves half the doublings, relative to doing this the naive way.
+	//
+	// ABGLSV-Pornin (https://eprint.iacr.org/2020/454) is faster,
+	// but significantly more complex, and has incompatibilities with
+	// mixed-order group elements.
+
+	tmp_add: Add_Scratch = ---
+	tmp_addend: Addend_Group_Element = ---
+	tmp_dbl: Double_Scratch = ---
+	tmp: Group_Element = ---
+
+	A_tbl: Multiply_Table = ---
+	mul_tbl_set(&A_tbl, A, &tmp_add)
+
+	sc_a, sc_b: field.Non_Montgomery_Domain_Field_Element
+	field.fe_from_montgomery(&sc_a, a)
+	field.fe_from_montgomery(&sc_b, b)
+
+	ge_identity(&tmp)
+	for i := 31; i >= 0; i = i - 1 {
+		limb := i / 8
+		shift := uint(i & 7) * 8
+
+		limb_byte_a := sc_a[limb] >> shift
+		limb_byte_b := sc_b[limb] >> shift
+
+		hi_a, lo_a := (limb_byte_a >> 4) & 0x0f, limb_byte_a & 0x0f
+		hi_b, lo_b := (limb_byte_b >> 4) & 0x0f, limb_byte_b & 0x0f
+
+		if i != 31 {
+			ge_double(&tmp, &tmp, &tmp_dbl)
+			ge_double(&tmp, &tmp, &tmp_dbl)
+			ge_double(&tmp, &tmp, &tmp_dbl)
+			ge_double(&tmp, &tmp, &tmp_dbl)
+		}
+		mul_tbl_add(&tmp, &A_tbl, hi_a, &tmp_add, &tmp_addend, true)
+		mul_tbl_add(&tmp, &GE_BASEPOINT_TABLE, hi_b, &tmp_add, &tmp_addend, true)
+
+		ge_double(&tmp, &tmp, &tmp_dbl)
+		ge_double(&tmp, &tmp, &tmp_dbl)
+		ge_double(&tmp, &tmp, &tmp_dbl)
+		ge_double(&tmp, &tmp, &tmp_dbl)
+		mul_tbl_add(&tmp, &A_tbl, lo_a, &tmp_add, &tmp_addend, true)
+		mul_tbl_add(&tmp, &GE_BASEPOINT_TABLE, lo_b, &tmp_add, &tmp_addend, true)
+	}
+
+	ge_set(ge, &tmp)
+}
+
+@(private)
+_ge_scalarmult :: proc "contextless" (
+	ge, p: ^Group_Element,
+	sc: ^field.Non_Montgomery_Domain_Field_Element,
+	unsafe_is_vartime := false,
+) {
+	// Do the simplest possible thing that works and provides adequate,
+	// performance, which is windowed add-then-multiply.
+
+	tmp_add: Add_Scratch = ---
+	tmp_addend: Addend_Group_Element = ---
+	tmp_dbl: Double_Scratch = ---
+	tmp: Group_Element = ---
+
+	p_tbl: Multiply_Table = ---
+	mul_tbl_set(&p_tbl, p, &tmp_add)
+
+	ge_identity(&tmp)
+	for i := 31; i >= 0; i = i - 1 {
+		limb := i / 8
+		shift := uint(i & 7) * 8
+		limb_byte := sc[limb] >> shift
+
+		hi, lo := (limb_byte >> 4) & 0x0f, limb_byte & 0x0f
+
+		if i != 31 {
+			ge_double(&tmp, &tmp, &tmp_dbl)
+			ge_double(&tmp, &tmp, &tmp_dbl)
+			ge_double(&tmp, &tmp, &tmp_dbl)
+			ge_double(&tmp, &tmp, &tmp_dbl)
+		}
+		mul_tbl_add(&tmp, &p_tbl, hi, &tmp_add, &tmp_addend, unsafe_is_vartime)
+
+		ge_double(&tmp, &tmp, &tmp_dbl)
+		ge_double(&tmp, &tmp, &tmp_dbl)
+		ge_double(&tmp, &tmp, &tmp_dbl)
+		ge_double(&tmp, &tmp, &tmp_dbl)
+		mul_tbl_add(&tmp, &p_tbl, lo, &tmp_add, &tmp_addend, unsafe_is_vartime)
+	}
+
+	ge_set(ge, &tmp)
+
+	if !unsafe_is_vartime {
+		ge_clear(&tmp)
+		mem.zero_explicit(&tmp_add, size_of(Add_Scratch))
+		mem.zero_explicit(&tmp_addend, size_of(Addend_Group_Element))
+		mem.zero_explicit(&tmp_dbl, size_of(Double_Scratch))
+	}
+}
+
+@(private)
+Multiply_Table :: [15]Addend_Group_Element // 0 = inf, which is implicit.
+
+@(private)
+mul_tbl_set :: proc "contextless" (
+	tbl: ^Multiply_Table,
+	ge: ^Group_Element,
+	tmp_add: ^Add_Scratch,
+) {
+	tmp: Group_Element = ---
+	ge_set(&tmp, ge)
+
+	ge_addend_set(&tbl[0], ge)
+	for i := 1; i < 15; i = i + 1 {
+		ge_add_addend(&tmp, &tmp, &tbl[0], tmp_add)
+		ge_addend_set(&tbl[i], &tmp)
+	}
+
+	ge_clear(&tmp)
+}
+
+@(private)
+mul_tbl_add :: proc "contextless" (
+	ge: ^Group_Element,
+	tbl: ^Multiply_Table,
+	idx: u64,
+	tmp_add: ^Add_Scratch,
+	tmp_addend: ^Addend_Group_Element,
+	unsafe_is_vartime: bool,
+) {
+	// Variable time lookup, with the addition omitted entirely if idx == 0.
+	if unsafe_is_vartime {
+		// Skip adding the point at infinity.
+		if idx != 0 {
+			ge_add_addend(ge, ge, &tbl[idx - 1], tmp_add)
+		}
+		return
+	}
+
+	// Constant time lookup.
+	tmp_addend^ = {
+		// Point at infinity (0, 1, 1, 0) in precomputed form
+		{1, 0, 0, 0, 0}, // y - x
+		{1, 0, 0, 0, 0}, // y + x
+		{0, 0, 0, 0, 0}, // t * 2d
+		{2, 0, 0, 0, 0}, // z * 2
+	}
+	for i := u64(1); i < 16; i = i + 1 {
+		_, ctrl := bits.sub_u64(0, (i ~ idx), 0)
+		ge_addend_conditional_assign(tmp_addend, &tbl[i - 1], int(~ctrl) & 1)
+	}
+	ge_add_addend(ge, ge, tmp_addend, tmp_add)
+}

core/crypto/_fiat/fiat.odin

@@ -9,7 +9,7 @@ package fiat
 u1 :: distinct u8
 i1 :: distinct i8
 
-@(optimization_mode="none")
+@(optimization_mode = "none")
 cmovznz_u64 :: proc "contextless" (arg1: u1, arg2, arg3: u64) -> (out1: u64) {
 	x1 := (u64(arg1) * 0xffffffffffffffff)
 	x2 := ((x1 & arg3) | ((~x1) & arg2))
@@ -17,7 +17,7 @@ cmovznz_u64 :: proc "contextless" (arg1: u1, arg2, arg3: u64) -> (out1: u64) {
 	return
 }
 
-@(optimization_mode="none")
+@(optimization_mode = "none")
 cmovznz_u32 :: proc "contextless" (arg1: u1, arg2, arg3: u32) -> (out1: u32) {
 	x1 := (u32(arg1) * 0xffffffff)
 	x2 := ((x1 & arg3) | ((~x1) & arg2))

core/crypto/_fiat/field_curve25519/field.odin

@@ -3,14 +3,32 @@ package field_curve25519
 import "core:crypto"
 import "core:mem"
 
-fe_relax_cast :: #force_inline proc "contextless" (arg1: ^Tight_Field_Element) -> ^Loose_Field_Element {
+fe_relax_cast :: #force_inline proc "contextless" (
+	arg1: ^Tight_Field_Element,
+) -> ^Loose_Field_Element {
 	return transmute(^Loose_Field_Element)(arg1)
 }
 
-fe_tighten_cast :: #force_inline proc "contextless" (arg1: ^Loose_Field_Element) -> ^Tight_Field_Element {
+fe_tighten_cast :: #force_inline proc "contextless" (
+	arg1: ^Loose_Field_Element,
+) -> ^Tight_Field_Element {
 	return transmute(^Tight_Field_Element)(arg1)
 }
 
+fe_clear :: proc "contextless" (
+	arg1: $T,
+) where T == ^Tight_Field_Element || T == ^Loose_Field_Element {
+	mem.zero_explicit(arg1, size_of(arg1^))
+}
+
+fe_clear_vec :: proc "contextless" (
+	arg1: $T,
+) where T == []^Tight_Field_Element || T == []^Loose_Field_Element {
+	for fe in arg1 {
+		fe_clear(fe)
+	}
+}
+
 fe_from_bytes :: proc "contextless" (out1: ^Tight_Field_Element, arg1: ^[32]byte) {
 	// Ignore the unused bit by copying the input and masking the bit off
 	// prior to deserialization.
@@ -23,12 +41,25 @@ fe_from_bytes :: proc "contextless" (out1: ^Tight_Field_Element, arg1: ^[32]byte
 	mem.zero_explicit(&tmp1, size_of(tmp1))
 }
 
+fe_is_negative :: proc "contextless" (arg1: ^Tight_Field_Element) -> int {
+	tmp1: [32]byte = ---
+
+	fe_to_bytes(&tmp1, arg1)
+	ret := tmp1[0] & 1
+
+	mem.zero_explicit(&tmp1, size_of(tmp1))
+
+	return int(ret)
+}
+
 fe_equal :: proc "contextless" (arg1, arg2: ^Tight_Field_Element) -> int {
-	tmp2: [32]byte = ---
+	tmp1, tmp2: [32]byte = ---, ---
 
+	fe_to_bytes(&tmp1, arg1)
 	fe_to_bytes(&tmp2, arg2)
-	ret := fe_equal_bytes(arg1, &tmp2)
+	ret := crypto.compare_constant_time(tmp1[:], tmp2[:])
 
+	mem.zero_explicit(&tmp1, size_of(tmp1))
 	mem.zero_explicit(&tmp2, size_of(tmp2))
 
 	return ret
@@ -46,7 +77,11 @@ fe_equal_bytes :: proc "contextless" (arg1: ^Tight_Field_Element, arg2: ^[32]byt
 	return ret
 }
 
-fe_carry_pow2k :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element, arg2: uint) {
+fe_carry_pow2k :: proc "contextless" (
+	out1: ^Tight_Field_Element,
+	arg1: ^Loose_Field_Element,
+	arg2: uint,
+) {
 	// Special case: `arg1^(2 * 0) = 1`, though this should never happen.
 	if arg2 == 0 {
 		fe_one(out1)
@@ -54,27 +89,46 @@ fe_carry_pow2k :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element,
 	}
 
 	fe_carry_square(out1, arg1)
-	for _ in 1..<arg2 {
+	for _ in 1 ..< arg2 {
 		fe_carry_square(out1, fe_relax_cast(out1))
 	}
 }
 
+fe_carry_add :: #force_inline proc "contextless" (out1, arg1, arg2: ^Tight_Field_Element) {
+	fe_add(fe_relax_cast(out1), arg1, arg2)
+	fe_carry(out1, fe_relax_cast(out1))
+}
+
+fe_carry_sub :: #force_inline proc "contextless" (out1, arg1, arg2: ^Tight_Field_Element) {
+	fe_sub(fe_relax_cast(out1), arg1, arg2)
+	fe_carry(out1, fe_relax_cast(out1))
+}
+
 fe_carry_opp :: #force_inline proc "contextless" (out1, arg1: ^Tight_Field_Element) {
 	fe_opp(fe_relax_cast(out1), arg1)
 	fe_carry(out1, fe_relax_cast(out1))
 }
 
-fe_carry_invsqrt :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) -> int {
-	// Inverse square root taken from Monocypher.
+fe_carry_abs :: #force_inline proc "contextless" (out1, arg1: ^Tight_Field_Element) {
+	fe_cond_negate(out1, arg1, fe_is_negative(arg1))
+}
 
+fe_carry_sqrt_ratio_m1 :: proc "contextless" (
+	out1: ^Tight_Field_Element,
+	arg1: ^Loose_Field_Element, // u
+	arg2: ^Loose_Field_Element, // v
+) -> int {
+	// SQRT_RATIO_M1(u, v) from RFC 9496 - 4.2, based on the inverse
+	// square root from Monocypher.
+
+	w: Tight_Field_Element = ---
+	fe_carry_mul(&w, arg1, arg2) // u * v
+
+	// r = tmp1 = u * w^((p-5)/8)
 	tmp1, tmp2, tmp3: Tight_Field_Element = ---, ---, ---
-
-	// t0 = x^((p-5)/8)
-	// Can be achieved with a simple double & add ladder,
-	// but it would be slower.
-	fe_carry_pow2k(&tmp1, arg1, 1)
+	fe_carry_pow2k(&tmp1, fe_relax_cast(&w), 1)
 	fe_carry_pow2k(&tmp2, fe_relax_cast(&tmp1), 2)
-	fe_carry_mul(&tmp2, arg1, fe_relax_cast(&tmp2))
+	fe_carry_mul(&tmp2, fe_relax_cast(&w), fe_relax_cast(&tmp2))
 	fe_carry_mul(&tmp1, fe_relax_cast(&tmp1), fe_relax_cast(&tmp2))
 	fe_carry_pow2k(&tmp1, fe_relax_cast(&tmp1), 1)
 	fe_carry_mul(&tmp1, fe_relax_cast(&tmp2), fe_relax_cast(&tmp1))
@@ -93,46 +147,121 @@ fe_carry_invsqrt :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element
 	fe_carry_pow2k(&tmp2, fe_relax_cast(&tmp2), 50)
 	fe_carry_mul(&tmp1, fe_relax_cast(&tmp2), fe_relax_cast(&tmp1))
 	fe_carry_pow2k(&tmp1, fe_relax_cast(&tmp1), 2)
-	fe_carry_mul(&tmp1, fe_relax_cast(&tmp1), arg1)
+	fe_carry_mul(&tmp1, fe_relax_cast(&tmp1), fe_relax_cast(&w)) // w^((p-5)/8)
 
-	// quartic = x^((p-1)/4)
-	quartic := &tmp2
-	fe_carry_square(quartic, fe_relax_cast(&tmp1))
-	fe_carry_mul(quartic, fe_relax_cast(quartic), arg1)
+	fe_carry_mul(&tmp1, fe_relax_cast(&tmp1), arg1) // u * w^((p-5)/8)
 
-	// Serialize quartic once to save on repeated serialization/sanitization.
-	quartic_buf: [32]byte = ---
-	fe_to_bytes(&quartic_buf, quartic)
-	check := &tmp3
+	// Serialize `check` once to save on repeated serialization.
+	r, check := &tmp1, &tmp2
+	b: [32]byte = ---
+	fe_carry_square(check, fe_relax_cast(r))
+	fe_carry_mul(check, fe_relax_cast(check), arg2) // check * v
+	fe_to_bytes(&b, check)
 
-	fe_one(check)
-	p1 := fe_equal_bytes(check, &quartic_buf)
-	fe_carry_opp(check, check)
-	m1 := fe_equal_bytes(check, &quartic_buf)
-	fe_carry_opp(check, &SQRT_M1)
-	ms := fe_equal_bytes(check, &quartic_buf)
+	u, neg_u, neg_u_i := &tmp3, &w, check
+	fe_carry(u, arg1)
+	fe_carry_opp(neg_u, u)
+	fe_carry_mul(neg_u_i, fe_relax_cast(neg_u), fe_relax_cast(&FE_SQRT_M1))
 
-	// if quartic == -1 or sqrt(-1)
-	// then  isr = x^((p-1)/4) * sqrt(-1)
-	// else  isr = x^((p-1)/4)
-	fe_carry_mul(out1, fe_relax_cast(&tmp1), fe_relax_cast(&SQRT_M1))
-	fe_cond_assign(out1, &tmp1, (m1|ms) ~ 1)
+	correct_sign_sqrt := fe_equal_bytes(u, &b)
+	flipped_sign_sqrt := fe_equal_bytes(neg_u, &b)
+	flipped_sign_sqrt_i := fe_equal_bytes(neg_u_i, &b)
 
-	mem.zero_explicit(&tmp1, size_of(tmp1))
-	mem.zero_explicit(&tmp2, size_of(tmp2))
-	mem.zero_explicit(&tmp3, size_of(tmp3))
-	mem.zero_explicit(&quartic_buf, size_of(quartic_buf))
+	r_prime := check
+	fe_carry_mul(r_prime, fe_relax_cast(r), fe_relax_cast(&FE_SQRT_M1))
+	fe_cond_assign(r, r_prime, flipped_sign_sqrt | flipped_sign_sqrt_i)
 
-	return p1 | m1
+	// Pick the non-negative square root.
+	fe_carry_abs(out1, r)
+
+	fe_clear_vec([]^Tight_Field_Element{&w, &tmp1, &tmp2, &tmp3})
+	mem.zero_explicit(&b, size_of(b))
+
+	return correct_sign_sqrt | flipped_sign_sqrt
 }
 
-fe_carry_inv :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
+fe_carry_inv :: proc "contextless" (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
 	tmp1: Tight_Field_Element
 
 	fe_carry_square(&tmp1, arg1)
-	_ = fe_carry_invsqrt(&tmp1, fe_relax_cast(&tmp1))
+	_ = fe_carry_sqrt_ratio_m1(&tmp1, fe_relax_cast(&FE_ONE), fe_relax_cast(&tmp1))
 	fe_carry_square(&tmp1, fe_relax_cast(&tmp1))
 	fe_carry_mul(out1, fe_relax_cast(&tmp1), arg1)
 
-	mem.zero_explicit(&tmp1, size_of(tmp1))
+	fe_clear(&tmp1)
+}
+
+fe_zero :: proc "contextless" (out1: ^Tight_Field_Element) {
+	out1[0] = 0
+	out1[1] = 0
+	out1[2] = 0
+	out1[3] = 0
+	out1[4] = 0
+}
+
+fe_one :: proc "contextless" (out1: ^Tight_Field_Element) {
+	out1[0] = 1
+	out1[1] = 0
+	out1[2] = 0
+	out1[3] = 0
+	out1[4] = 0
+}
+
+fe_set :: proc "contextless" (out1, arg1: ^Tight_Field_Element) {
+	x1 := arg1[0]
+	x2 := arg1[1]
+	x3 := arg1[2]
+	x4 := arg1[3]
+	x5 := arg1[4]
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+	out1[3] = x4
+	out1[4] = x5
+}
+
+@(optimization_mode = "none")
+fe_cond_swap :: #force_no_inline proc "contextless" (out1, out2: ^Tight_Field_Element, arg1: int) {
+	mask := (u64(arg1) * 0xffffffffffffffff)
+	x := (out1[0] ~ out2[0]) & mask
+	x1, y1 := out1[0] ~ x, out2[0] ~ x
+	x = (out1[1] ~ out2[1]) & mask
+	x2, y2 := out1[1] ~ x, out2[1] ~ x
+	x = (out1[2] ~ out2[2]) & mask
+	x3, y3 := out1[2] ~ x, out2[2] ~ x
+	x = (out1[3] ~ out2[3]) & mask
+	x4, y4 := out1[3] ~ x, out2[3] ~ x
+	x = (out1[4] ~ out2[4]) & mask
+	x5, y5 := out1[4] ~ x, out2[4] ~ x
+	out1[0], out2[0] = x1, y1
+	out1[1], out2[1] = x2, y2
+	out1[2], out2[2] = x3, y3
+	out1[3], out2[3] = x4, y4
+	out1[4], out2[4] = x5, y5
+}
+
+@(optimization_mode = "none")
+fe_cond_select :: #force_no_inline proc "contextless" (
+	out1, arg1, arg2: $T,
+	arg3: int,
+) where T == ^Tight_Field_Element || T == ^Loose_Field_Element {
+	mask := (u64(arg3) * 0xffffffffffffffff)
+	x1 := ((mask & arg2[0]) | ((~mask) & arg1[0]))
+	x2 := ((mask & arg2[1]) | ((~mask) & arg1[1]))
+	x3 := ((mask & arg2[2]) | ((~mask) & arg1[2]))
+	x4 := ((mask & arg2[3]) | ((~mask) & arg1[3]))
+	x5 := ((mask & arg2[4]) | ((~mask) & arg1[4]))
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+	out1[3] = x4
+	out1[4] = x5
+}
+
+fe_cond_negate :: proc "contextless" (out1, arg1: ^Tight_Field_Element, ctrl: int) {
+	tmp1: Tight_Field_Element = ---
+	fe_carry_opp(&tmp1, arg1)
+	fe_cond_select(out1, arg1, &tmp1, ctrl)
+
+	fe_clear(&tmp1)
 }

core/crypto/_fiat/field_curve25519/field51.odin

@@ -30,8 +30,6 @@ package field_curve25519
 //
 // While the base implementation is provably correct, this implementation
 // makes no such claims as the port and optimizations were done by hand.
-// At some point, it may be worth adding support to fiat-crypto for
-// generating Odin output.
 //
 // TODO:
 //  * When fiat-crypto supports it, using a saturated 64-bit limbs
@@ -44,7 +42,10 @@ import "core:math/bits"
 Loose_Field_Element :: distinct [5]u64
 Tight_Field_Element :: distinct [5]u64
 
-SQRT_M1 := Tight_Field_Element{
+FE_ZERO := Tight_Field_Element{0, 0, 0, 0, 0}
+FE_ONE := Tight_Field_Element{1, 0, 0, 0, 0}
+
+FE_SQRT_M1 := Tight_Field_Element {
 	1718705420411056,
 	234908883556509,
 	2233514472574048,
@@ -52,7 +53,13 @@ SQRT_M1 := Tight_Field_Element{
 	765476049583133,
 }
 
-_addcarryx_u51 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+_addcarryx_u51 :: #force_inline proc "contextless" (
+	arg1: fiat.u1,
+	arg2, arg3: u64,
+) -> (
+	out1: u64,
+	out2: fiat.u1,
+) {
 	x1 := ((u64(arg1) + arg2) + arg3)
 	x2 := (x1 & 0x7ffffffffffff)
 	x3 := fiat.u1((x1 >> 51))
@@ -61,7 +68,13 @@ _addcarryx_u51 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u
 	return
 }
 
-_subborrowx_u51 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+_subborrowx_u51 :: #force_inline proc "contextless" (
+	arg1: fiat.u1,
+	arg2, arg3: u64,
+) -> (
+	out1: u64,
+	out2: fiat.u1,
+) {
 	x1 := ((i64(arg2) - i64(arg1)) - i64(arg3))
 	x2 := fiat.i1((x1 >> 51))
 	x3 := (u64(x1) & 0x7ffffffffffff)
@@ -70,7 +83,7 @@ _subborrowx_u51 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3:
 	return
 }
 
-fe_carry_mul :: proc (out1: ^Tight_Field_Element, arg1, arg2: ^Loose_Field_Element) {
+fe_carry_mul :: proc "contextless" (out1: ^Tight_Field_Element, arg1, arg2: ^Loose_Field_Element) {
 	x2, x1 := bits.mul_u64(arg1[4], (arg2[4] * 0x13))
 	x4, x3 := bits.mul_u64(arg1[4], (arg2[3] * 0x13))
 	x6, x5 := bits.mul_u64(arg1[4], (arg2[2] * 0x13))
@@ -169,7 +182,7 @@ fe_carry_mul :: proc (out1: ^Tight_Field_Element, arg1, arg2: ^Loose_Field_Eleme
 	out1[4] = x152
 }
 
-fe_carry_square :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
+fe_carry_square :: proc "contextless" (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
 	x1 := (arg1[4] * 0x13)
 	x2 := (x1 * 0x2)
 	x3 := (arg1[4] * 0x2)
@@ -305,8 +318,11 @@ fe_opp :: proc "contextless" (out1: ^Loose_Field_Element, arg1: ^Tight_Field_Ele
 	out1[4] = x5
 }
 
-@(optimization_mode="none")
-fe_cond_assign :: #force_no_inline proc "contextless" (out1, arg1: ^Tight_Field_Element, arg2: int) {
+@(optimization_mode = "none")
+fe_cond_assign :: #force_no_inline proc "contextless" (
+	out1, arg1: ^Tight_Field_Element,
+	arg2: int,
+) {
 	x1 := fiat.cmovznz_u64(fiat.u1(arg2), out1[0], arg1[0])
 	x2 := fiat.cmovznz_u64(fiat.u1(arg2), out1[1], arg1[1])
 	x3 := fiat.cmovznz_u64(fiat.u1(arg2), out1[2], arg1[2])
@@ -527,7 +543,10 @@ fe_relax :: proc "contextless" (out1: ^Loose_Field_Element, arg1: ^Tight_Field_E
 	out1[4] = x5
 }
 
-fe_carry_scmul_121666 :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
+fe_carry_scmul_121666 :: proc "contextless" (
+	out1: ^Tight_Field_Element,
+	arg1: ^Loose_Field_Element,
+) {
 	x2, x1 := bits.mul_u64(0x1db42, arg1[4])
 	x4, x3 := bits.mul_u64(0x1db42, arg1[3])
 	x6, x5 := bits.mul_u64(0x1db42, arg1[2])
@@ -565,54 +584,3 @@ fe_carry_scmul_121666 :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_El
 	out1[3] = x27
 	out1[4] = x32
 }
-
-// The following routines were added by hand, and do not come from fiat-crypto.
-
-fe_zero :: proc "contextless" (out1: ^Tight_Field_Element) {
-	out1[0] = 0
-	out1[1] = 0
-	out1[2] = 0
-	out1[3] = 0
-	out1[4] = 0
-}
-
-fe_one :: proc "contextless" (out1: ^Tight_Field_Element) {
-	out1[0] = 1
-	out1[1] = 0
-	out1[2] = 0
-	out1[3] = 0
-	out1[4] = 0
-}
-
-fe_set :: proc "contextless" (out1, arg1: ^Tight_Field_Element) {
-	x1 := arg1[0]
-	x2 := arg1[1]
-	x3 := arg1[2]
-	x4 := arg1[3]
-	x5 := arg1[4]
-	out1[0] = x1
-	out1[1] = x2
-	out1[2] = x3
-	out1[3] = x4
-	out1[4] = x5
-}
-
-@(optimization_mode="none")
-fe_cond_swap :: #force_no_inline proc "contextless" (out1, out2: ^Tight_Field_Element, arg1: int) {
-	mask := -u64(arg1)
-	x := (out1[0] ~ out2[0]) & mask
-	x1, y1 := out1[0] ~ x, out2[0] ~ x
-	x = (out1[1] ~ out2[1]) & mask
-	x2, y2 := out1[1] ~ x, out2[1] ~ x
-	x = (out1[2] ~ out2[2]) & mask
-	x3, y3 := out1[2] ~ x, out2[2] ~ x
-	x = (out1[3] ~ out2[3]) & mask
-	x4, y4 := out1[3] ~ x, out2[3] ~ x
-	x = (out1[4] ~ out2[4]) & mask
-	x5, y5 := out1[4] ~ x, out2[4] ~ x
-	out1[0], out2[0] = x1, y1
-	out1[1], out2[1] = x2, y2
-	out1[2], out2[2] = x3, y3
-	out1[3], out2[3] = x4, y4
-	out1[4], out2[4] = x5, y5
-}

core/crypto/_fiat/field_poly1305/field.odin

@@ -1,17 +1,26 @@
 package field_poly1305
 
+import "base:intrinsics"
 import "core:encoding/endian"
 import "core:mem"
 
-fe_relax_cast :: #force_inline proc "contextless" (arg1: ^Tight_Field_Element) -> ^Loose_Field_Element {
+fe_relax_cast :: #force_inline proc "contextless" (
+	arg1: ^Tight_Field_Element,
+) -> ^Loose_Field_Element {
 	return transmute(^Loose_Field_Element)(arg1)
 }
 
-fe_tighten_cast :: #force_inline proc "contextless" (arg1: ^Loose_Field_Element) -> ^Tight_Field_Element {
+fe_tighten_cast :: #force_inline proc "contextless" (
+	arg1: ^Loose_Field_Element,
+) -> ^Tight_Field_Element {
 	return transmute(^Tight_Field_Element)(arg1)
 }
 
-fe_from_bytes :: #force_inline proc (out1: ^Tight_Field_Element, arg1: []byte, arg2: byte) {
+fe_from_bytes :: #force_inline proc "contextless" (
+	out1: ^Tight_Field_Element,
+	arg1: []byte,
+	arg2: byte,
+) {
 	// fiat-crypto's deserialization routine effectively processes a
 	// single byte at a time, and wants 256-bits of input for a value
 	// that will be 128-bits or 129-bits.
@@ -20,7 +29,9 @@ fe_from_bytes :: #force_inline proc (out1: ^Tight_Field_Element, arg1: []byte, a
 	// makes implementing the actual MAC block processing considerably
 	// neater.
 
-	assert(len(arg1) == 16)
+	if len(arg1) != 16 {
+		intrinsics.trap()
+	}
 
 	// While it may be unwise to do deserialization here on our
 	// own when fiat-crypto provides equivalent functionality,
@@ -51,3 +62,35 @@ fe_from_u64s :: proc "contextless" (out1: ^Tight_Field_Element, lo, hi: u64) {
 	// This routine is only used to deserialize `r` which is confidential.
 	mem.zero_explicit(&tmp, size_of(tmp))
 }
+
+fe_zero :: proc "contextless" (out1: ^Tight_Field_Element) {
+	out1[0] = 0
+	out1[1] = 0
+	out1[2] = 0
+}
+
+fe_set :: #force_inline proc "contextless" (out1, arg1: ^Tight_Field_Element) {
+	x1 := arg1[0]
+	x2 := arg1[1]
+	x3 := arg1[2]
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+}
+
+@(optimization_mode = "none")
+fe_cond_swap :: #force_no_inline proc "contextless" (
+	out1, out2: ^Tight_Field_Element,
+	arg1: bool,
+) {
+	mask := (u64(arg1) * 0xffffffffffffffff)
+	x := (out1[0] ~ out2[0]) & mask
+	x1, y1 := out1[0] ~ x, out2[0] ~ x
+	x = (out1[1] ~ out2[1]) & mask
+	x2, y2 := out1[1] ~ x, out2[1] ~ x
+	x = (out1[2] ~ out2[2]) & mask
+	x3, y3 := out1[2] ~ x, out2[2] ~ x
+	out1[0], out2[0] = x1, y1
+	out1[1], out2[1] = x2, y2
+	out1[2], out2[2] = x3, y3
+}

core/crypto/_fiat/field_poly1305/field4344.odin

@@ -39,7 +39,13 @@ import "core:math/bits"
 Loose_Field_Element :: distinct [3]u64
 Tight_Field_Element :: distinct [3]u64
 
-_addcarryx_u44 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+_addcarryx_u44 :: #force_inline proc "contextless" (
+	arg1: fiat.u1,
+	arg2, arg3: u64,
+) -> (
+	out1: u64,
+	out2: fiat.u1,
+) {
 	x1 := ((u64(arg1) + arg2) + arg3)
 	x2 := (x1 & 0xfffffffffff)
 	x3 := fiat.u1((x1 >> 44))
@@ -48,7 +54,13 @@ _addcarryx_u44 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u
 	return
 }
 
-_subborrowx_u44 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+_subborrowx_u44 :: #force_inline proc "contextless" (
+	arg1: fiat.u1,
+	arg2, arg3: u64,
+) -> (
+	out1: u64,
+	out2: fiat.u1,
+) {
 	x1 := ((i64(arg2) - i64(arg1)) - i64(arg3))
 	x2 := fiat.i1((x1 >> 44))
 	x3 := (u64(x1) & 0xfffffffffff)
@@ -57,7 +69,13 @@ _subborrowx_u44 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3:
 	return
 }
 
-_addcarryx_u43 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+_addcarryx_u43 :: #force_inline proc "contextless" (
+	arg1: fiat.u1,
+	arg2, arg3: u64,
+) -> (
+	out1: u64,
+	out2: fiat.u1,
+) {
 	x1 := ((u64(arg1) + arg2) + arg3)
 	x2 := (x1 & 0x7ffffffffff)
 	x3 := fiat.u1((x1 >> 43))
@@ -66,7 +84,13 @@ _addcarryx_u43 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u
 	return
 }
 
-_subborrowx_u43 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+_subborrowx_u43 :: #force_inline proc "contextless" (
+	arg1: fiat.u1,
+	arg2, arg3: u64,
+) -> (
+	out1: u64,
+	out2: fiat.u1,
+) {
 	x1 := ((i64(arg2) - i64(arg1)) - i64(arg3))
 	x2 := fiat.i1((x1 >> 43))
 	x3 := (u64(x1) & 0x7ffffffffff)
@@ -75,7 +99,7 @@ _subborrowx_u43 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3:
 	return
 }
 
-fe_carry_mul :: proc (out1: ^Tight_Field_Element, arg1, arg2: ^Loose_Field_Element) {
+fe_carry_mul :: proc "contextless" (out1: ^Tight_Field_Element, arg1, arg2: ^Loose_Field_Element) {
 	x2, x1 := bits.mul_u64(arg1[2], (arg2[2] * 0x5))
 	x4, x3 := bits.mul_u64(arg1[2], (arg2[1] * 0xa))
 	x6, x5 := bits.mul_u64(arg1[1], (arg2[2] * 0xa))
@@ -120,7 +144,7 @@ fe_carry_mul :: proc (out1: ^Tight_Field_Element, arg1, arg2: ^Loose_Field_Eleme
 	out1[2] = x62
 }
 
-fe_carry_square :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
+fe_carry_square :: proc "contextless" (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
 	x1 := (arg1[2] * 0x5)
 	x2 := (x1 * 0x2)
 	x3 := (arg1[2] * 0x2)
@@ -201,8 +225,11 @@ fe_opp :: proc "contextless" (out1: ^Loose_Field_Element, arg1: ^Tight_Field_Ele
 	out1[2] = x3
 }
 
-@(optimization_mode="none")
-fe_cond_assign :: #force_no_inline proc "contextless" (out1, arg1: ^Tight_Field_Element, arg2: bool) {
+@(optimization_mode = "none")
+fe_cond_assign :: #force_no_inline proc "contextless" (
+	out1, arg1: ^Tight_Field_Element,
+	arg2: bool,
+) {
 	x1 := fiat.cmovznz_u64(fiat.u1(arg2), out1[0], arg1[0])
 	x2 := fiat.cmovznz_u64(fiat.u1(arg2), out1[1], arg1[1])
 	x3 := fiat.cmovznz_u64(fiat.u1(arg2), out1[2], arg1[2])
@@ -325,34 +352,3 @@ fe_relax :: proc "contextless" (out1: ^Loose_Field_Element, arg1: ^Tight_Field_E
 	out1[1] = x2
 	out1[2] = x3
 }
-
-// The following routines were added by hand, and do not come from fiat-crypto.
-
-fe_zero :: proc "contextless" (out1: ^Tight_Field_Element) {
-	out1[0] = 0
-	out1[1] = 0
-	out1[2] = 0
-}
-
-fe_set :: #force_inline proc "contextless" (out1, arg1: ^Tight_Field_Element) {
-	x1 := arg1[0]
-	x2 := arg1[1]
-	x3 := arg1[2]
-	out1[0] = x1
-	out1[1] = x2
-	out1[2] = x3
-}
-
-@(optimization_mode="none")
-fe_cond_swap :: #force_no_inline proc "contextless" (out1, out2: ^Tight_Field_Element, arg1: bool) {
-	mask := -u64(arg1)
-	x := (out1[0] ~ out2[0]) & mask
-	x1, y1 := out1[0] ~ x, out2[0] ~ x
-	x = (out1[1] ~ out2[1]) & mask
-	x2, y2 := out1[1] ~ x, out2[1] ~ x
-	x = (out1[2] ~ out2[2]) & mask
-	x3, y3 := out1[2] ~ x, out2[2] ~ x
-	out1[0], out2[0] = x1, y1
-	out1[1], out2[1] = x2, y2
-	out1[2], out2[2] = x3, y3
-}

core/crypto/_fiat/field_scalar25519/field.odin

@@ -0,0 +1,153 @@
+package field_scalar25519
+
+import "base:intrinsics"
+import "core:encoding/endian"
+import "core:math/bits"
+import "core:mem"
+
+@(private)
+_TWO_168 := Montgomery_Domain_Field_Element {
+	0x5b8ab432eac74798,
+	0x38afddd6de59d5d7,
+	0xa2c131b399411b7c,
+	0x6329a7ed9ce5a30,
+}
+@(private)
+_TWO_336 := Montgomery_Domain_Field_Element {
+	0xbd3d108e2b35ecc5,
+	0x5c3a3718bdf9c90b,
+	0x63aa97a331b4f2ee,
+	0x3d217f5be65cb5c,
+}
+
+fe_clear :: proc "contextless" (arg1: ^Montgomery_Domain_Field_Element) {
+	mem.zero_explicit(arg1, size_of(Montgomery_Domain_Field_Element))
+}
+
+fe_from_bytes :: proc "contextless" (
+	out1: ^Montgomery_Domain_Field_Element,
+	arg1: ^[32]byte,
+	unsafe_assume_canonical := false,
+) -> bool {
+	tmp := Non_Montgomery_Domain_Field_Element {
+		endian.unchecked_get_u64le(arg1[0:]),
+		endian.unchecked_get_u64le(arg1[8:]),
+		endian.unchecked_get_u64le(arg1[16:]),
+		endian.unchecked_get_u64le(arg1[24:]),
+	}
+	defer mem.zero_explicit(&tmp, size_of(tmp))
+
+	// Check that tmp is in the the range [0, ELL).
+	if !unsafe_assume_canonical {
+		_, borrow := bits.sub_u64(ELL[0] - 1, tmp[0], 0)
+		_, borrow = bits.sub_u64(ELL[1], tmp[1], borrow)
+		_, borrow = bits.sub_u64(ELL[2], tmp[2], borrow)
+		_, borrow = bits.sub_u64(ELL[3], tmp[3], borrow)
+		if borrow != 0 {
+			return false
+		}
+	}
+
+	fe_to_montgomery(out1, &tmp)
+
+	return true
+}
+
+fe_from_bytes_rfc8032 :: proc "contextless" (
+	out1: ^Montgomery_Domain_Field_Element,
+	arg1: ^[32]byte,
+) {
+	tmp: [64]byte
+	copy(tmp[:], arg1[:])
+
+	// Apply "clamping" as in RFC 8032.
+	tmp[0] &= 248
+	tmp[31] &= 127
+	tmp[31] |= 64 // Sets the 254th bit, so the encoding is non-canonical.
+
+	fe_from_bytes_wide(out1, &tmp)
+
+	mem.zero_explicit(&tmp, size_of(tmp))
+}
+
+fe_from_bytes_wide :: proc "contextless" (
+	out1: ^Montgomery_Domain_Field_Element,
+	arg1: ^[64]byte,
+) {
+	tmp: Montgomery_Domain_Field_Element
+	// Use Frank Denis' trick, as documented by Filippo Valsorda
+	// at https://words.filippo.io/dispatches/wide-reduction/
+	//
+	// x = c * 2^336 + b * 2^168 + a  mod l
+	_fe_from_bytes_short(out1, arg1[:21]) // a
+
+	_fe_from_bytes_short(&tmp, arg1[21:42]) // b
+	fe_mul(&tmp, &tmp, &_TWO_168) // b * 2^168
+	fe_add(out1, out1, &tmp) // a + b * 2^168
+
+	_fe_from_bytes_short(&tmp, arg1[42:]) // c
+	fe_mul(&tmp, &tmp, &_TWO_336) // c * 2^336
+	fe_add(out1, out1, &tmp) // a + b * 2^168 + c * 2^336
+
+	fe_clear(&tmp)
+}
+
+@(private)
+_fe_from_bytes_short :: proc "contextless" (out1: ^Montgomery_Domain_Field_Element, arg1: []byte) {
+	// INVARIANT: len(arg1) < 32.
+	if len(arg1) >= 32 {
+		intrinsics.trap()
+	}
+	tmp: [32]byte
+	copy(tmp[:], arg1)
+
+	_ = fe_from_bytes(out1, &tmp, true)
+	mem.zero_explicit(&tmp, size_of(tmp))
+}
+
+fe_to_bytes :: proc "contextless" (out1: []byte, arg1: ^Montgomery_Domain_Field_Element) {
+	if len(out1) != 32 {
+		intrinsics.trap()
+	}
+
+	tmp: Non_Montgomery_Domain_Field_Element
+	fe_from_montgomery(&tmp, arg1)
+
+	endian.unchecked_put_u64le(out1[0:], tmp[0])
+	endian.unchecked_put_u64le(out1[8:], tmp[1])
+	endian.unchecked_put_u64le(out1[16:], tmp[2])
+	endian.unchecked_put_u64le(out1[24:], tmp[3])
+
+	mem.zero_explicit(&tmp, size_of(tmp))
+}
+
+fe_equal :: proc "contextless" (arg1, arg2: ^Montgomery_Domain_Field_Element) -> int {
+	tmp: Montgomery_Domain_Field_Element
+	fe_sub(&tmp, arg1, arg2)
+
+	// This will only underflow iff arg1 == arg2, and we return the borrow,
+	// which will be 1.
+	_, borrow := bits.sub_u64(fe_non_zero(&tmp), 1, 0)
+
+	fe_clear(&tmp)
+
+	return int(borrow)
+}
+
+fe_zero :: proc "contextless" (out1: ^Montgomery_Domain_Field_Element) {
+	out1[0] = 0
+	out1[1] = 0
+	out1[2] = 0
+	out1[3] = 0
+}
+
+fe_set :: proc "contextless" (out1, arg1: ^Montgomery_Domain_Field_Element) {
+	x1 := arg1[0]
+	x2 := arg1[1]
+	x3 := arg1[2]
+	x4 := arg1[3]
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+	out1[3] = x4
+}

core/crypto/_fiat/field_scalar25519/field64.odin

@@ -0,0 +1,535 @@
+// The BSD 1-Clause License (BSD-1-Clause)
+//
+// Copyright (c) 2015-2020 the fiat-crypto authors (see the AUTHORS file)
+// 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 fiat-crypto 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 Berkeley Software Design,
+// Inc. 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 field_scalar25519
+
+// The file provides arithmetic on the field Z/(2^252+27742317777372353535851937790883648493)
+// using a 64-bit Montgomery form internal representation.  It is derived
+// primarily from the machine generated Golang output from the fiat-crypto
+// project.
+//
+// While the base implementation is provably correct, this implementation
+// makes no such claims as the port and optimizations were done by hand.
+
+import fiat "core:crypto/_fiat"
+import "core:math/bits"
+
+// ELL is the saturated representation of the field order, least-significant
+// limb first.
+ELL :: [4]u64{0x5812631a5cf5d3ed, 0x14def9dea2f79cd6, 0x0, 0x1000000000000000}
+
+Montgomery_Domain_Field_Element :: distinct [4]u64
+Non_Montgomery_Domain_Field_Element :: distinct [4]u64
+
+fe_mul :: proc "contextless" (out1, arg1, arg2: ^Montgomery_Domain_Field_Element) {
+	x1 := arg1[1]
+	x2 := arg1[2]
+	x3 := arg1[3]
+	x4 := arg1[0]
+	x6, x5 := bits.mul_u64(x4, arg2[3])
+	x8, x7 := bits.mul_u64(x4, arg2[2])
+	x10, x9 := bits.mul_u64(x4, arg2[1])
+	x12, x11 := bits.mul_u64(x4, arg2[0])
+	x13, x14 := bits.add_u64(x12, x9, u64(0x0))
+	x15, x16 := bits.add_u64(x10, x7, u64(fiat.u1(x14)))
+	x17, x18 := bits.add_u64(x8, x5, u64(fiat.u1(x16)))
+	x19 := (u64(fiat.u1(x18)) + x6)
+	_, x20 := bits.mul_u64(x11, 0xd2b51da312547e1b)
+	x23, x22 := bits.mul_u64(x20, 0x1000000000000000)
+	x25, x24 := bits.mul_u64(x20, 0x14def9dea2f79cd6)
+	x27, x26 := bits.mul_u64(x20, 0x5812631a5cf5d3ed)
+	x28, x29 := bits.add_u64(x27, x24, u64(0x0))
+	x30 := (u64(fiat.u1(x29)) + x25)
+	_, x32 := bits.add_u64(x11, x26, u64(0x0))
+	x33, x34 := bits.add_u64(x13, x28, u64(fiat.u1(x32)))
+	x35, x36 := bits.add_u64(x15, x30, u64(fiat.u1(x34)))
+	x37, x38 := bits.add_u64(x17, x22, u64(fiat.u1(x36)))
+	x39, x40 := bits.add_u64(x19, x23, u64(fiat.u1(x38)))
+	x42, x41 := bits.mul_u64(x1, arg2[3])
+	x44, x43 := bits.mul_u64(x1, arg2[2])
+	x46, x45 := bits.mul_u64(x1, arg2[1])
+	x48, x47 := bits.mul_u64(x1, arg2[0])
+	x49, x50 := bits.add_u64(x48, x45, u64(0x0))
+	x51, x52 := bits.add_u64(x46, x43, u64(fiat.u1(x50)))
+	x53, x54 := bits.add_u64(x44, x41, u64(fiat.u1(x52)))
+	x55 := (u64(fiat.u1(x54)) + x42)
+	x56, x57 := bits.add_u64(x33, x47, u64(0x0))
+	x58, x59 := bits.add_u64(x35, x49, u64(fiat.u1(x57)))
+	x60, x61 := bits.add_u64(x37, x51, u64(fiat.u1(x59)))
+	x62, x63 := bits.add_u64(x39, x53, u64(fiat.u1(x61)))
+	x64, x65 := bits.add_u64(u64(fiat.u1(x40)), x55, u64(fiat.u1(x63)))
+	_, x66 := bits.mul_u64(x56, 0xd2b51da312547e1b)
+	x69, x68 := bits.mul_u64(x66, 0x1000000000000000)
+	x71, x70 := bits.mul_u64(x66, 0x14def9dea2f79cd6)
+	x73, x72 := bits.mul_u64(x66, 0x5812631a5cf5d3ed)
+	x74, x75 := bits.add_u64(x73, x70, u64(0x0))
+	x76 := (u64(fiat.u1(x75)) + x71)
+	_, x78 := bits.add_u64(x56, x72, u64(0x0))
+	x79, x80 := bits.add_u64(x58, x74, u64(fiat.u1(x78)))
+	x81, x82 := bits.add_u64(x60, x76, u64(fiat.u1(x80)))
+	x83, x84 := bits.add_u64(x62, x68, u64(fiat.u1(x82)))
+	x85, x86 := bits.add_u64(x64, x69, u64(fiat.u1(x84)))
+	x87 := (u64(fiat.u1(x86)) + u64(fiat.u1(x65)))
+	x89, x88 := bits.mul_u64(x2, arg2[3])
+	x91, x90 := bits.mul_u64(x2, arg2[2])
+	x93, x92 := bits.mul_u64(x2, arg2[1])
+	x95, x94 := bits.mul_u64(x2, arg2[0])
+	x96, x97 := bits.add_u64(x95, x92, u64(0x0))
+	x98, x99 := bits.add_u64(x93, x90, u64(fiat.u1(x97)))
+	x100, x101 := bits.add_u64(x91, x88, u64(fiat.u1(x99)))
+	x102 := (u64(fiat.u1(x101)) + x89)
+	x103, x104 := bits.add_u64(x79, x94, u64(0x0))
+	x105, x106 := bits.add_u64(x81, x96, u64(fiat.u1(x104)))
+	x107, x108 := bits.add_u64(x83, x98, u64(fiat.u1(x106)))
+	x109, x110 := bits.add_u64(x85, x100, u64(fiat.u1(x108)))
+	x111, x112 := bits.add_u64(x87, x102, u64(fiat.u1(x110)))
+	_, x113 := bits.mul_u64(x103, 0xd2b51da312547e1b)
+	x116, x115 := bits.mul_u64(x113, 0x1000000000000000)
+	x118, x117 := bits.mul_u64(x113, 0x14def9dea2f79cd6)
+	x120, x119 := bits.mul_u64(x113, 0x5812631a5cf5d3ed)
+	x121, x122 := bits.add_u64(x120, x117, u64(0x0))
+	x123 := (u64(fiat.u1(x122)) + x118)
+	_, x125 := bits.add_u64(x103, x119, u64(0x0))
+	x126, x127 := bits.add_u64(x105, x121, u64(fiat.u1(x125)))
+	x128, x129 := bits.add_u64(x107, x123, u64(fiat.u1(x127)))
+	x130, x131 := bits.add_u64(x109, x115, u64(fiat.u1(x129)))
+	x132, x133 := bits.add_u64(x111, x116, u64(fiat.u1(x131)))
+	x134 := (u64(fiat.u1(x133)) + u64(fiat.u1(x112)))
+	x136, x135 := bits.mul_u64(x3, arg2[3])
+	x138, x137 := bits.mul_u64(x3, arg2[2])
+	x140, x139 := bits.mul_u64(x3, arg2[1])
+	x142, x141 := bits.mul_u64(x3, arg2[0])
+	x143, x144 := bits.add_u64(x142, x139, u64(0x0))
+	x145, x146 := bits.add_u64(x140, x137, u64(fiat.u1(x144)))
+	x147, x148 := bits.add_u64(x138, x135, u64(fiat.u1(x146)))
+	x149 := (u64(fiat.u1(x148)) + x136)
+	x150, x151 := bits.add_u64(x126, x141, u64(0x0))
+	x152, x153 := bits.add_u64(x128, x143, u64(fiat.u1(x151)))
+	x154, x155 := bits.add_u64(x130, x145, u64(fiat.u1(x153)))
+	x156, x157 := bits.add_u64(x132, x147, u64(fiat.u1(x155)))
+	x158, x159 := bits.add_u64(x134, x149, u64(fiat.u1(x157)))
+	_, x160 := bits.mul_u64(x150, 0xd2b51da312547e1b)
+	x163, x162 := bits.mul_u64(x160, 0x1000000000000000)
+	x165, x164 := bits.mul_u64(x160, 0x14def9dea2f79cd6)
+	x167, x166 := bits.mul_u64(x160, 0x5812631a5cf5d3ed)
+	x168, x169 := bits.add_u64(x167, x164, u64(0x0))
+	x170 := (u64(fiat.u1(x169)) + x165)
+	_, x172 := bits.add_u64(x150, x166, u64(0x0))
+	x173, x174 := bits.add_u64(x152, x168, u64(fiat.u1(x172)))
+	x175, x176 := bits.add_u64(x154, x170, u64(fiat.u1(x174)))
+	x177, x178 := bits.add_u64(x156, x162, u64(fiat.u1(x176)))
+	x179, x180 := bits.add_u64(x158, x163, u64(fiat.u1(x178)))
+	x181 := (u64(fiat.u1(x180)) + u64(fiat.u1(x159)))
+	x182, x183 := bits.sub_u64(x173, 0x5812631a5cf5d3ed, u64(0x0))
+	x184, x185 := bits.sub_u64(x175, 0x14def9dea2f79cd6, u64(fiat.u1(x183)))
+	x186, x187 := bits.sub_u64(x177, u64(0x0), u64(fiat.u1(x185)))
+	x188, x189 := bits.sub_u64(x179, 0x1000000000000000, u64(fiat.u1(x187)))
+	_, x191 := bits.sub_u64(x181, u64(0x0), u64(fiat.u1(x189)))
+	x192 := fiat.cmovznz_u64(fiat.u1(x191), x182, x173)
+	x193 := fiat.cmovznz_u64(fiat.u1(x191), x184, x175)
+	x194 := fiat.cmovznz_u64(fiat.u1(x191), x186, x177)
+	x195 := fiat.cmovznz_u64(fiat.u1(x191), x188, x179)
+	out1[0] = x192
+	out1[1] = x193
+	out1[2] = x194
+	out1[3] = x195
+}
+
+fe_square :: proc "contextless" (out1, arg1: ^Montgomery_Domain_Field_Element) {
+	x1 := arg1[1]
+	x2 := arg1[2]
+	x3 := arg1[3]
+	x4 := arg1[0]
+	x6, x5 := bits.mul_u64(x4, arg1[3])
+	x8, x7 := bits.mul_u64(x4, arg1[2])
+	x10, x9 := bits.mul_u64(x4, arg1[1])
+	x12, x11 := bits.mul_u64(x4, arg1[0])
+	x13, x14 := bits.add_u64(x12, x9, u64(0x0))
+	x15, x16 := bits.add_u64(x10, x7, u64(fiat.u1(x14)))
+	x17, x18 := bits.add_u64(x8, x5, u64(fiat.u1(x16)))
+	x19 := (u64(fiat.u1(x18)) + x6)
+	_, x20 := bits.mul_u64(x11, 0xd2b51da312547e1b)
+	x23, x22 := bits.mul_u64(x20, 0x1000000000000000)
+	x25, x24 := bits.mul_u64(x20, 0x14def9dea2f79cd6)
+	x27, x26 := bits.mul_u64(x20, 0x5812631a5cf5d3ed)
+	x28, x29 := bits.add_u64(x27, x24, u64(0x0))
+	x30 := (u64(fiat.u1(x29)) + x25)
+	_, x32 := bits.add_u64(x11, x26, u64(0x0))
+	x33, x34 := bits.add_u64(x13, x28, u64(fiat.u1(x32)))
+	x35, x36 := bits.add_u64(x15, x30, u64(fiat.u1(x34)))
+	x37, x38 := bits.add_u64(x17, x22, u64(fiat.u1(x36)))
+	x39, x40 := bits.add_u64(x19, x23, u64(fiat.u1(x38)))
+	x42, x41 := bits.mul_u64(x1, arg1[3])
+	x44, x43 := bits.mul_u64(x1, arg1[2])
+	x46, x45 := bits.mul_u64(x1, arg1[1])
+	x48, x47 := bits.mul_u64(x1, arg1[0])
+	x49, x50 := bits.add_u64(x48, x45, u64(0x0))
+	x51, x52 := bits.add_u64(x46, x43, u64(fiat.u1(x50)))
+	x53, x54 := bits.add_u64(x44, x41, u64(fiat.u1(x52)))
+	x55 := (u64(fiat.u1(x54)) + x42)
+	x56, x57 := bits.add_u64(x33, x47, u64(0x0))
+	x58, x59 := bits.add_u64(x35, x49, u64(fiat.u1(x57)))
+	x60, x61 := bits.add_u64(x37, x51, u64(fiat.u1(x59)))
+	x62, x63 := bits.add_u64(x39, x53, u64(fiat.u1(x61)))
+	x64, x65 := bits.add_u64(u64(fiat.u1(x40)), x55, u64(fiat.u1(x63)))
+	_, x66 := bits.mul_u64(x56, 0xd2b51da312547e1b)
+	x69, x68 := bits.mul_u64(x66, 0x1000000000000000)
+	x71, x70 := bits.mul_u64(x66, 0x14def9dea2f79cd6)
+	x73, x72 := bits.mul_u64(x66, 0x5812631a5cf5d3ed)
+	x74, x75 := bits.add_u64(x73, x70, u64(0x0))
+	x76 := (u64(fiat.u1(x75)) + x71)
+	_, x78 := bits.add_u64(x56, x72, u64(0x0))
+	x79, x80 := bits.add_u64(x58, x74, u64(fiat.u1(x78)))
+	x81, x82 := bits.add_u64(x60, x76, u64(fiat.u1(x80)))
+	x83, x84 := bits.add_u64(x62, x68, u64(fiat.u1(x82)))
+	x85, x86 := bits.add_u64(x64, x69, u64(fiat.u1(x84)))
+	x87 := (u64(fiat.u1(x86)) + u64(fiat.u1(x65)))
+	x89, x88 := bits.mul_u64(x2, arg1[3])
+	x91, x90 := bits.mul_u64(x2, arg1[2])
+	x93, x92 := bits.mul_u64(x2, arg1[1])
+	x95, x94 := bits.mul_u64(x2, arg1[0])
+	x96, x97 := bits.add_u64(x95, x92, u64(0x0))
+	x98, x99 := bits.add_u64(x93, x90, u64(fiat.u1(x97)))
+	x100, x101 := bits.add_u64(x91, x88, u64(fiat.u1(x99)))
+	x102 := (u64(fiat.u1(x101)) + x89)
+	x103, x104 := bits.add_u64(x79, x94, u64(0x0))
+	x105, x106 := bits.add_u64(x81, x96, u64(fiat.u1(x104)))
+	x107, x108 := bits.add_u64(x83, x98, u64(fiat.u1(x106)))
+	x109, x110 := bits.add_u64(x85, x100, u64(fiat.u1(x108)))
+	x111, x112 := bits.add_u64(x87, x102, u64(fiat.u1(x110)))
+	_, x113 := bits.mul_u64(x103, 0xd2b51da312547e1b)
+	x116, x115 := bits.mul_u64(x113, 0x1000000000000000)
+	x118, x117 := bits.mul_u64(x113, 0x14def9dea2f79cd6)
+	x120, x119 := bits.mul_u64(x113, 0x5812631a5cf5d3ed)
+	x121, x122 := bits.add_u64(x120, x117, u64(0x0))
+	x123 := (u64(fiat.u1(x122)) + x118)
+	_, x125 := bits.add_u64(x103, x119, u64(0x0))
+	x126, x127 := bits.add_u64(x105, x121, u64(fiat.u1(x125)))
+	x128, x129 := bits.add_u64(x107, x123, u64(fiat.u1(x127)))
+	x130, x131 := bits.add_u64(x109, x115, u64(fiat.u1(x129)))
+	x132, x133 := bits.add_u64(x111, x116, u64(fiat.u1(x131)))
+	x134 := (u64(fiat.u1(x133)) + u64(fiat.u1(x112)))
+	x136, x135 := bits.mul_u64(x3, arg1[3])
+	x138, x137 := bits.mul_u64(x3, arg1[2])
+	x140, x139 := bits.mul_u64(x3, arg1[1])
+	x142, x141 := bits.mul_u64(x3, arg1[0])
+	x143, x144 := bits.add_u64(x142, x139, u64(0x0))
+	x145, x146 := bits.add_u64(x140, x137, u64(fiat.u1(x144)))
+	x147, x148 := bits.add_u64(x138, x135, u64(fiat.u1(x146)))
+	x149 := (u64(fiat.u1(x148)) + x136)
+	x150, x151 := bits.add_u64(x126, x141, u64(0x0))
+	x152, x153 := bits.add_u64(x128, x143, u64(fiat.u1(x151)))
+	x154, x155 := bits.add_u64(x130, x145, u64(fiat.u1(x153)))
+	x156, x157 := bits.add_u64(x132, x147, u64(fiat.u1(x155)))
+	x158, x159 := bits.add_u64(x134, x149, u64(fiat.u1(x157)))
+	_, x160 := bits.mul_u64(x150, 0xd2b51da312547e1b)
+	x163, x162 := bits.mul_u64(x160, 0x1000000000000000)
+	x165, x164 := bits.mul_u64(x160, 0x14def9dea2f79cd6)
+	x167, x166 := bits.mul_u64(x160, 0x5812631a5cf5d3ed)
+	x168, x169 := bits.add_u64(x167, x164, u64(0x0))
+	x170 := (u64(fiat.u1(x169)) + x165)
+	_, x172 := bits.add_u64(x150, x166, u64(0x0))
+	x173, x174 := bits.add_u64(x152, x168, u64(fiat.u1(x172)))
+	x175, x176 := bits.add_u64(x154, x170, u64(fiat.u1(x174)))
+	x177, x178 := bits.add_u64(x156, x162, u64(fiat.u1(x176)))
+	x179, x180 := bits.add_u64(x158, x163, u64(fiat.u1(x178)))
+	x181 := (u64(fiat.u1(x180)) + u64(fiat.u1(x159)))
+	x182, x183 := bits.sub_u64(x173, 0x5812631a5cf5d3ed, u64(0x0))
+	x184, x185 := bits.sub_u64(x175, 0x14def9dea2f79cd6, u64(fiat.u1(x183)))
+	x186, x187 := bits.sub_u64(x177, u64(0x0), u64(fiat.u1(x185)))
+	x188, x189 := bits.sub_u64(x179, 0x1000000000000000, u64(fiat.u1(x187)))
+	_, x191 := bits.sub_u64(x181, u64(0x0), u64(fiat.u1(x189)))
+	x192 := fiat.cmovznz_u64(fiat.u1(x191), x182, x173)
+	x193 := fiat.cmovznz_u64(fiat.u1(x191), x184, x175)
+	x194 := fiat.cmovznz_u64(fiat.u1(x191), x186, x177)
+	x195 := fiat.cmovznz_u64(fiat.u1(x191), x188, x179)
+	out1[0] = x192
+	out1[1] = x193
+	out1[2] = x194
+	out1[3] = x195
+}
+
+fe_add :: proc "contextless" (out1, arg1, arg2: ^Montgomery_Domain_Field_Element) {
+	x1, x2 := bits.add_u64(arg1[0], arg2[0], u64(0x0))
+	x3, x4 := bits.add_u64(arg1[1], arg2[1], u64(fiat.u1(x2)))
+	x5, x6 := bits.add_u64(arg1[2], arg2[2], u64(fiat.u1(x4)))
+	x7, x8 := bits.add_u64(arg1[3], arg2[3], u64(fiat.u1(x6)))
+	x9, x10 := bits.sub_u64(x1, 0x5812631a5cf5d3ed, u64(0x0))
+	x11, x12 := bits.sub_u64(x3, 0x14def9dea2f79cd6, u64(fiat.u1(x10)))
+	x13, x14 := bits.sub_u64(x5, u64(0x0), u64(fiat.u1(x12)))
+	x15, x16 := bits.sub_u64(x7, 0x1000000000000000, u64(fiat.u1(x14)))
+	_, x18 := bits.sub_u64(u64(fiat.u1(x8)), u64(0x0), u64(fiat.u1(x16)))
+	x19 := fiat.cmovznz_u64(fiat.u1(x18), x9, x1)
+	x20 := fiat.cmovznz_u64(fiat.u1(x18), x11, x3)
+	x21 := fiat.cmovznz_u64(fiat.u1(x18), x13, x5)
+	x22 := fiat.cmovznz_u64(fiat.u1(x18), x15, x7)
+	out1[0] = x19
+	out1[1] = x20
+	out1[2] = x21
+	out1[3] = x22
+}
+
+fe_sub :: proc "contextless" (out1, arg1, arg2: ^Montgomery_Domain_Field_Element) {
+	x1, x2 := bits.sub_u64(arg1[0], arg2[0], u64(0x0))
+	x3, x4 := bits.sub_u64(arg1[1], arg2[1], u64(fiat.u1(x2)))
+	x5, x6 := bits.sub_u64(arg1[2], arg2[2], u64(fiat.u1(x4)))
+	x7, x8 := bits.sub_u64(arg1[3], arg2[3], u64(fiat.u1(x6)))
+	x9 := fiat.cmovznz_u64(fiat.u1(x8), u64(0x0), 0xffffffffffffffff)
+	x10, x11 := bits.add_u64(x1, (x9 & 0x5812631a5cf5d3ed), u64(0x0))
+	x12, x13 := bits.add_u64(x3, (x9 & 0x14def9dea2f79cd6), u64(fiat.u1(x11)))
+	x14, x15 := bits.add_u64(x5, u64(0x0), u64(fiat.u1(x13)))
+	x16, _ := bits.add_u64(x7, (x9 & 0x1000000000000000), u64(fiat.u1(x15)))
+	out1[0] = x10
+	out1[1] = x12
+	out1[2] = x14
+	out1[3] = x16
+}
+
+fe_opp :: proc "contextless" (out1, arg1: ^Montgomery_Domain_Field_Element) {
+	x1, x2 := bits.sub_u64(u64(0x0), arg1[0], u64(0x0))
+	x3, x4 := bits.sub_u64(u64(0x0), arg1[1], u64(fiat.u1(x2)))
+	x5, x6 := bits.sub_u64(u64(0x0), arg1[2], u64(fiat.u1(x4)))
+	x7, x8 := bits.sub_u64(u64(0x0), arg1[3], u64(fiat.u1(x6)))
+	x9 := fiat.cmovznz_u64(fiat.u1(x8), u64(0x0), 0xffffffffffffffff)
+	x10, x11 := bits.add_u64(x1, (x9 & 0x5812631a5cf5d3ed), u64(0x0))
+	x12, x13 := bits.add_u64(x3, (x9 & 0x14def9dea2f79cd6), u64(fiat.u1(x11)))
+	x14, x15 := bits.add_u64(x5, u64(0x0), u64(fiat.u1(x13)))
+	x16, _ := bits.add_u64(x7, (x9 & 0x1000000000000000), u64(fiat.u1(x15)))
+	out1[0] = x10
+	out1[1] = x12
+	out1[2] = x14
+	out1[3] = x16
+}
+
+fe_one :: proc "contextless" (out1: ^Montgomery_Domain_Field_Element) {
+	out1[0] = 0xd6ec31748d98951d
+	out1[1] = 0xc6ef5bf4737dcf70
+	out1[2] = 0xfffffffffffffffe
+	out1[3] = 0xfffffffffffffff
+}
+
+fe_non_zero :: proc "contextless" (arg1: ^Montgomery_Domain_Field_Element) -> u64 {
+	return arg1[0] | (arg1[1] | (arg1[2] | arg1[3]))
+}
+
+@(optimization_mode = "none")
+fe_cond_assign :: #force_no_inline proc "contextless" (
+	out1, arg1: ^Montgomery_Domain_Field_Element,
+	arg2: int,
+) {
+	x1 := fiat.cmovznz_u64(fiat.u1(arg2), out1[0], arg1[0])
+	x2 := fiat.cmovznz_u64(fiat.u1(arg2), out1[1], arg1[1])
+	x3 := fiat.cmovznz_u64(fiat.u1(arg2), out1[2], arg1[2])
+	x4 := fiat.cmovznz_u64(fiat.u1(arg2), out1[3], arg1[3])
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+	out1[3] = x4
+}
+
+fe_from_montgomery :: proc "contextless" (
+	out1: ^Non_Montgomery_Domain_Field_Element,
+	arg1: ^Montgomery_Domain_Field_Element,
+) {
+	x1 := arg1[0]
+	_, x2 := bits.mul_u64(x1, 0xd2b51da312547e1b)
+	x5, x4 := bits.mul_u64(x2, 0x1000000000000000)
+	x7, x6 := bits.mul_u64(x2, 0x14def9dea2f79cd6)
+	x9, x8 := bits.mul_u64(x2, 0x5812631a5cf5d3ed)
+	x10, x11 := bits.add_u64(x9, x6, u64(0x0))
+	_, x13 := bits.add_u64(x1, x8, u64(0x0))
+	x14, x15 := bits.add_u64(u64(0x0), x10, u64(fiat.u1(x13)))
+	x16, x17 := bits.add_u64(x14, arg1[1], u64(0x0))
+	_, x18 := bits.mul_u64(x16, 0xd2b51da312547e1b)
+	x21, x20 := bits.mul_u64(x18, 0x1000000000000000)
+	x23, x22 := bits.mul_u64(x18, 0x14def9dea2f79cd6)
+	x25, x24 := bits.mul_u64(x18, 0x5812631a5cf5d3ed)
+	x26, x27 := bits.add_u64(x25, x22, u64(0x0))
+	_, x29 := bits.add_u64(x16, x24, u64(0x0))
+	x30, x31 := bits.add_u64(
+		(u64(fiat.u1(x17)) + (u64(fiat.u1(x15)) + (u64(fiat.u1(x11)) + x7))),
+		x26,
+		u64(fiat.u1(x29)),
+	)
+	x32, x33 := bits.add_u64(x4, (u64(fiat.u1(x27)) + x23), u64(fiat.u1(x31)))
+	x34, x35 := bits.add_u64(x5, x20, u64(fiat.u1(x33)))
+	x36, x37 := bits.add_u64(x30, arg1[2], u64(0x0))
+	x38, x39 := bits.add_u64(x32, u64(0x0), u64(fiat.u1(x37)))
+	x40, x41 := bits.add_u64(x34, u64(0x0), u64(fiat.u1(x39)))
+	_, x42 := bits.mul_u64(x36, 0xd2b51da312547e1b)
+	x45, x44 := bits.mul_u64(x42, 0x1000000000000000)
+	x47, x46 := bits.mul_u64(x42, 0x14def9dea2f79cd6)
+	x49, x48 := bits.mul_u64(x42, 0x5812631a5cf5d3ed)
+	x50, x51 := bits.add_u64(x49, x46, u64(0x0))
+	_, x53 := bits.add_u64(x36, x48, u64(0x0))
+	x54, x55 := bits.add_u64(x38, x50, u64(fiat.u1(x53)))
+	x56, x57 := bits.add_u64(x40, (u64(fiat.u1(x51)) + x47), u64(fiat.u1(x55)))
+	x58, x59 := bits.add_u64(
+		(u64(fiat.u1(x41)) + (u64(fiat.u1(x35)) + x21)),
+		x44,
+		u64(fiat.u1(x57)),
+	)
+	x60, x61 := bits.add_u64(x54, arg1[3], u64(0x0))
+	x62, x63 := bits.add_u64(x56, u64(0x0), u64(fiat.u1(x61)))
+	x64, x65 := bits.add_u64(x58, u64(0x0), u64(fiat.u1(x63)))
+	_, x66 := bits.mul_u64(x60, 0xd2b51da312547e1b)
+	x69, x68 := bits.mul_u64(x66, 0x1000000000000000)
+	x71, x70 := bits.mul_u64(x66, 0x14def9dea2f79cd6)
+	x73, x72 := bits.mul_u64(x66, 0x5812631a5cf5d3ed)
+	x74, x75 := bits.add_u64(x73, x70, u64(0x0))
+	_, x77 := bits.add_u64(x60, x72, u64(0x0))
+	x78, x79 := bits.add_u64(x62, x74, u64(fiat.u1(x77)))
+	x80, x81 := bits.add_u64(x64, (u64(fiat.u1(x75)) + x71), u64(fiat.u1(x79)))
+	x82, x83 := bits.add_u64(
+		(u64(fiat.u1(x65)) + (u64(fiat.u1(x59)) + x45)),
+		x68,
+		u64(fiat.u1(x81)),
+	)
+	x84 := (u64(fiat.u1(x83)) + x69)
+	x85, x86 := bits.sub_u64(x78, 0x5812631a5cf5d3ed, u64(0x0))
+	x87, x88 := bits.sub_u64(x80, 0x14def9dea2f79cd6, u64(fiat.u1(x86)))
+	x89, x90 := bits.sub_u64(x82, u64(0x0), u64(fiat.u1(x88)))
+	x91, x92 := bits.sub_u64(x84, 0x1000000000000000, u64(fiat.u1(x90)))
+	_, x94 := bits.sub_u64(u64(0x0), u64(0x0), u64(fiat.u1(x92)))
+	x95 := fiat.cmovznz_u64(fiat.u1(x94), x85, x78)
+	x96 := fiat.cmovznz_u64(fiat.u1(x94), x87, x80)
+	x97 := fiat.cmovznz_u64(fiat.u1(x94), x89, x82)
+	x98 := fiat.cmovznz_u64(fiat.u1(x94), x91, x84)
+	out1[0] = x95
+	out1[1] = x96
+	out1[2] = x97
+	out1[3] = x98
+}
+
+fe_to_montgomery :: proc "contextless" (
+	out1: ^Montgomery_Domain_Field_Element,
+	arg1: ^Non_Montgomery_Domain_Field_Element,
+) {
+	x1 := arg1[1]
+	x2 := arg1[2]
+	x3 := arg1[3]
+	x4 := arg1[0]
+	x6, x5 := bits.mul_u64(x4, 0x399411b7c309a3d)
+	x8, x7 := bits.mul_u64(x4, 0xceec73d217f5be65)
+	x10, x9 := bits.mul_u64(x4, 0xd00e1ba768859347)
+	x12, x11 := bits.mul_u64(x4, 0xa40611e3449c0f01)
+	x13, x14 := bits.add_u64(x12, x9, u64(0x0))
+	x15, x16 := bits.add_u64(x10, x7, u64(fiat.u1(x14)))
+	x17, x18 := bits.add_u64(x8, x5, u64(fiat.u1(x16)))
+	_, x19 := bits.mul_u64(x11, 0xd2b51da312547e1b)
+	x22, x21 := bits.mul_u64(x19, 0x1000000000000000)
+	x24, x23 := bits.mul_u64(x19, 0x14def9dea2f79cd6)
+	x26, x25 := bits.mul_u64(x19, 0x5812631a5cf5d3ed)
+	x27, x28 := bits.add_u64(x26, x23, u64(0x0))
+	_, x30 := bits.add_u64(x11, x25, u64(0x0))
+	x31, x32 := bits.add_u64(x13, x27, u64(fiat.u1(x30)))
+	x33, x34 := bits.add_u64(x15, (u64(fiat.u1(x28)) + x24), u64(fiat.u1(x32)))
+	x35, x36 := bits.add_u64(x17, x21, u64(fiat.u1(x34)))
+	x38, x37 := bits.mul_u64(x1, 0x399411b7c309a3d)
+	x40, x39 := bits.mul_u64(x1, 0xceec73d217f5be65)
+	x42, x41 := bits.mul_u64(x1, 0xd00e1ba768859347)
+	x44, x43 := bits.mul_u64(x1, 0xa40611e3449c0f01)
+	x45, x46 := bits.add_u64(x44, x41, u64(0x0))
+	x47, x48 := bits.add_u64(x42, x39, u64(fiat.u1(x46)))
+	x49, x50 := bits.add_u64(x40, x37, u64(fiat.u1(x48)))
+	x51, x52 := bits.add_u64(x31, x43, u64(0x0))
+	x53, x54 := bits.add_u64(x33, x45, u64(fiat.u1(x52)))
+	x55, x56 := bits.add_u64(x35, x47, u64(fiat.u1(x54)))
+	x57, x58 := bits.add_u64(
+		((u64(fiat.u1(x36)) + (u64(fiat.u1(x18)) + x6)) + x22),
+		x49,
+		u64(fiat.u1(x56)),
+	)
+	_, x59 := bits.mul_u64(x51, 0xd2b51da312547e1b)
+	x62, x61 := bits.mul_u64(x59, 0x1000000000000000)
+	x64, x63 := bits.mul_u64(x59, 0x14def9dea2f79cd6)
+	x66, x65 := bits.mul_u64(x59, 0x5812631a5cf5d3ed)
+	x67, x68 := bits.add_u64(x66, x63, u64(0x0))
+	_, x70 := bits.add_u64(x51, x65, u64(0x0))
+	x71, x72 := bits.add_u64(x53, x67, u64(fiat.u1(x70)))
+	x73, x74 := bits.add_u64(x55, (u64(fiat.u1(x68)) + x64), u64(fiat.u1(x72)))
+	x75, x76 := bits.add_u64(x57, x61, u64(fiat.u1(x74)))
+	x78, x77 := bits.mul_u64(x2, 0x399411b7c309a3d)
+	x80, x79 := bits.mul_u64(x2, 0xceec73d217f5be65)
+	x82, x81 := bits.mul_u64(x2, 0xd00e1ba768859347)
+	x84, x83 := bits.mul_u64(x2, 0xa40611e3449c0f01)
+	x85, x86 := bits.add_u64(x84, x81, u64(0x0))
+	x87, x88 := bits.add_u64(x82, x79, u64(fiat.u1(x86)))
+	x89, x90 := bits.add_u64(x80, x77, u64(fiat.u1(x88)))
+	x91, x92 := bits.add_u64(x71, x83, u64(0x0))
+	x93, x94 := bits.add_u64(x73, x85, u64(fiat.u1(x92)))
+	x95, x96 := bits.add_u64(x75, x87, u64(fiat.u1(x94)))
+	x97, x98 := bits.add_u64(
+		((u64(fiat.u1(x76)) + (u64(fiat.u1(x58)) + (u64(fiat.u1(x50)) + x38))) + x62),
+		x89,
+		u64(fiat.u1(x96)),
+	)
+	_, x99 := bits.mul_u64(x91, 0xd2b51da312547e1b)
+	x102, x101 := bits.mul_u64(x99, 0x1000000000000000)
+	x104, x103 := bits.mul_u64(x99, 0x14def9dea2f79cd6)
+	x106, x105 := bits.mul_u64(x99, 0x5812631a5cf5d3ed)
+	x107, x108 := bits.add_u64(x106, x103, u64(0x0))
+	_, x110 := bits.add_u64(x91, x105, u64(0x0))
+	x111, x112 := bits.add_u64(x93, x107, u64(fiat.u1(x110)))
+	x113, x114 := bits.add_u64(x95, (u64(fiat.u1(x108)) + x104), u64(fiat.u1(x112)))
+	x115, x116 := bits.add_u64(x97, x101, u64(fiat.u1(x114)))
+	x118, x117 := bits.mul_u64(x3, 0x399411b7c309a3d)
+	x120, x119 := bits.mul_u64(x3, 0xceec73d217f5be65)
+	x122, x121 := bits.mul_u64(x3, 0xd00e1ba768859347)
+	x124, x123 := bits.mul_u64(x3, 0xa40611e3449c0f01)
+	x125, x126 := bits.add_u64(x124, x121, u64(0x0))
+	x127, x128 := bits.add_u64(x122, x119, u64(fiat.u1(x126)))
+	x129, x130 := bits.add_u64(x120, x117, u64(fiat.u1(x128)))
+	x131, x132 := bits.add_u64(x111, x123, u64(0x0))
+	x133, x134 := bits.add_u64(x113, x125, u64(fiat.u1(x132)))
+	x135, x136 := bits.add_u64(x115, x127, u64(fiat.u1(x134)))
+	x137, x138 := bits.add_u64(
+		((u64(fiat.u1(x116)) + (u64(fiat.u1(x98)) + (u64(fiat.u1(x90)) + x78))) + x102),
+		x129,
+		u64(fiat.u1(x136)),
+	)
+	_, x139 := bits.mul_u64(x131, 0xd2b51da312547e1b)
+	x142, x141 := bits.mul_u64(x139, 0x1000000000000000)
+	x144, x143 := bits.mul_u64(x139, 0x14def9dea2f79cd6)
+	x146, x145 := bits.mul_u64(x139, 0x5812631a5cf5d3ed)
+	x147, x148 := bits.add_u64(x146, x143, u64(0x0))
+	_, x150 := bits.add_u64(x131, x145, u64(0x0))
+	x151, x152 := bits.add_u64(x133, x147, u64(fiat.u1(x150)))
+	x153, x154 := bits.add_u64(x135, (u64(fiat.u1(x148)) + x144), u64(fiat.u1(x152)))
+	x155, x156 := bits.add_u64(x137, x141, u64(fiat.u1(x154)))
+	x157 := ((u64(fiat.u1(x156)) + (u64(fiat.u1(x138)) + (u64(fiat.u1(x130)) + x118))) + x142)
+	x158, x159 := bits.sub_u64(x151, 0x5812631a5cf5d3ed, u64(0x0))
+	x160, x161 := bits.sub_u64(x153, 0x14def9dea2f79cd6, u64(fiat.u1(x159)))
+	x162, x163 := bits.sub_u64(x155, u64(0x0), u64(fiat.u1(x161)))
+	x164, x165 := bits.sub_u64(x157, 0x1000000000000000, u64(fiat.u1(x163)))
+	_, x167 := bits.sub_u64(u64(0x0), u64(0x0), u64(fiat.u1(x165)))
+	x168 := fiat.cmovznz_u64(fiat.u1(x167), x158, x151)
+	x169 := fiat.cmovznz_u64(fiat.u1(x167), x160, x153)
+	x170 := fiat.cmovznz_u64(fiat.u1(x167), x162, x155)
+	x171 := fiat.cmovznz_u64(fiat.u1(x167), x164, x157)
+	out1[0] = x168
+	out1[1] = x169
+	out1[2] = x170
+	out1[3] = x171
+}

core/crypto/_sha3/sha3.odin

@@ -7,8 +7,12 @@ package _sha3
     List of contributors:
         zhibog, dotbmp:  Initial implementation.
 
-    Implementation of the Keccak hashing algorithm, standardized as SHA3 in <https://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.202.pdf>
-    To use the original Keccak padding, set the is_keccak bool to true, otherwise it will use SHA3 padding.
+    Implementation of the Keccak hashing algorithm, standardized as SHA3
+    in <https://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.202.pdf>.
+
+    As the only difference between the legacy Keccak and SHA3 is the domain
+    separation byte, set dsbyte to the appropriate value to pick the desired
+    algorithm.
 */
 
 import "core:math/bits"
@@ -16,47 +20,56 @@ import "core:mem"
 
 ROUNDS :: 24
 
+RATE_128 :: 1344 / 8 // ONLY for SHAKE128.
 RATE_224 :: 1152 / 8
 RATE_256 :: 1088 / 8
 RATE_384 :: 832 / 8
 RATE_512 :: 576 / 8
 
+DS_KECCAK :: 0x01
+DS_SHA3 :: 0x06
+DS_SHAKE :: 0x1f
+DS_CSHAKE :: 0x04
+
 Context :: struct {
-	st:        struct #raw_union {
+	st:             struct #raw_union {
 		b: [200]u8,
 		q: [25]u64,
 	},
-	pt:        int,
-	rsiz:      int,
-	mdlen:     int,
-	is_keccak: bool,
-
+	pt:             int,
+	rsiz:           int,
+	mdlen:          int,
+	dsbyte:         byte,
 	is_initialized: bool,
 	is_finalized:   bool, // For SHAKE (unlimited squeeze is allowed)
 }
 
+@(private)
+keccakf_rndc := [?]u64 {
+	0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
+	0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
+	0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
+	0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
+	0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
+	0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
+	0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
+	0x8000000000008080, 0x0000000080000001, 0x8000000080008008,
+}
+
+@(private)
+keccakf_rotc := [?]int {
+	1, 3, 6, 10, 15, 21, 28, 36, 45, 55, 2, 14,
+	27, 41, 56, 8, 25, 43, 62, 18, 39, 61, 20, 44,
+}
+
+@(private)
+keccakf_piln := [?]i32 {
+	10, 7, 11, 17, 18, 3, 5, 16, 8, 21, 24, 4,
+	15, 23, 19, 13, 12, 2, 20, 14, 22, 9, 6, 1,
+}
+
+@(private)
 keccakf :: proc "contextless" (st: ^[25]u64) {
-	keccakf_rndc := [?]u64 {
-		0x0000000000000001, 0x0000000000008082, 0x800000000000808a,
-		0x8000000080008000, 0x000000000000808b, 0x0000000080000001,
-		0x8000000080008081, 0x8000000000008009, 0x000000000000008a,
-		0x0000000000000088, 0x0000000080008009, 0x000000008000000a,
-		0x000000008000808b, 0x800000000000008b, 0x8000000000008089,
-		0x8000000000008003, 0x8000000000008002, 0x8000000000000080,
-		0x000000000000800a, 0x800000008000000a, 0x8000000080008081,
-		0x8000000000008080, 0x0000000080000001, 0x8000000080008008,
-	}
-
-	keccakf_rotc := [?]int {
-		1, 3, 6, 10, 15, 21, 28, 36, 45, 55, 2, 14,
-		27, 41, 56, 8, 25, 43, 62, 18, 39, 61, 20, 44,
-	}
-
-	keccakf_piln := [?]i32 {
-		10, 7, 11, 17, 18, 3, 5, 16, 8, 21, 24, 4,
-		15, 23, 19, 13, 12, 2, 20, 14, 22, 9, 6, 1,
-	}
-
 	i, j, r: i32 = ---, ---, ---
 	t: u64 = ---
 	bc: [5]u64 = ---
@@ -140,9 +153,6 @@ final :: proc(ctx: ^Context, hash: []byte, finalize_clone: bool = false) {
 	assert(ctx.is_initialized)
 
 	if len(hash) < ctx.mdlen {
-		if ctx.is_keccak {
-			panic("crypto/keccac: invalid destination digest size")
-		}
 		panic("crypto/sha3: invalid destination digest size")
 	}
 
@@ -152,13 +162,9 @@ final :: proc(ctx: ^Context, hash: []byte, finalize_clone: bool = false) {
 		clone(&tmp_ctx, ctx)
 		ctx = &tmp_ctx
 	}
-	defer(reset(ctx))
+	defer (reset(ctx))
 
-	if ctx.is_keccak {
-		ctx.st.b[ctx.pt] ~= 0x01
-	} else {
-		ctx.st.b[ctx.pt] ~= 0x06
-	}
+	ctx.st.b[ctx.pt] ~= ctx.dsbyte
 
 	ctx.st.b[ctx.rsiz - 1] ~= 0x80
 	keccakf(&ctx.st.q)
@@ -183,7 +189,7 @@ shake_xof :: proc(ctx: ^Context) {
 	assert(ctx.is_initialized)
 	assert(!ctx.is_finalized)
 
-	ctx.st.b[ctx.pt] ~= 0x1F
+	ctx.st.b[ctx.pt] ~= ctx.dsbyte
 	ctx.st.b[ctx.rsiz - 1] ~= 0x80
 	keccakf(&ctx.st.q)
 	ctx.pt = 0

core/crypto/_sha3/sp800_185.odin

@@ -0,0 +1,145 @@
+package _sha3
+
+import "core:encoding/endian"
+import "core:math/bits"
+
+init_cshake :: proc(ctx: ^Context, n, s: []byte, sec_strength: int) {
+	ctx.mdlen = sec_strength / 8
+
+	// No domain separator is equivalent to vanilla SHAKE.
+	if len(n) == 0 && len(s) == 0 {
+		ctx.dsbyte = DS_SHAKE
+		init(ctx)
+		return
+	}
+
+	ctx.dsbyte = DS_CSHAKE
+	init(ctx)
+	bytepad(ctx, [][]byte{n, s}, rate_cshake(sec_strength))
+}
+
+final_cshake :: proc(ctx: ^Context, dst: []byte, finalize_clone: bool = false) {
+	ctx := ctx
+	if finalize_clone {
+		tmp_ctx: Context
+		clone(&tmp_ctx, ctx)
+		ctx = &tmp_ctx
+	}
+	defer reset(ctx)
+
+	encode_byte_len(ctx, len(dst), false) // right_encode
+	shake_xof(ctx)
+	shake_out(ctx, dst)
+}
+
+rate_cshake :: #force_inline proc(sec_strength: int) -> int {
+	switch sec_strength {
+	case 128:
+		return RATE_128
+	case 256:
+		return RATE_256
+	}
+
+	panic("crypto/sha3: invalid security strength")
+}
+
+// right_encode and left_encode are defined to support 0 <= x < 2^2040
+// however, the largest value we will ever need to encode is `max(int) * 8`.
+//
+// This is unfortunate as the extreme upper edge is larger than
+// `max(u64)`.  While such values are impractical at present,
+// they are possible (ie: https://arxiv.org/pdf/quant-ph/9908043.pdf).
+//
+// Thus we support 0 <= x < 2^128.
+
+@(private)
+_PAD: [RATE_128]byte // Biggest possible value of w per spec.
+
+bytepad :: proc(ctx: ^Context, x_strings: [][]byte, w: int) {
+	// 1. z = left_encode(w) || X.
+	z_hi: u64
+	z_lo := left_right_encode(ctx, 0, u64(w), true)
+	for x in x_strings {
+		// All uses of bytepad in SP 800-185 use the output from
+		// one or more encode_string values for `X`.
+		hi, lo := encode_string(ctx, x)
+
+		carry: u64
+		z_lo, carry = bits.add_u64(z_lo, lo, 0)
+		z_hi, carry = bits.add_u64(z_hi, hi, carry)
+
+		// This isn't actually possible, at least with the currently
+		// defined SP 800-185 routines.
+		if carry != 0 {
+			panic("crypto/sha3: bytepad input length overflow")
+		}
+	}
+
+	// We skip this step as we are doing a byte-oriented implementation
+	// rather than a bit oriented one.
+	//
+	// 2. while len(z) mod 8 ≠ 0:
+	//    z = z || 0
+
+	// 3. while (len(z)/8) mod w ≠ 0:
+	//    z = z || 00000000
+	z_len := u128(z_hi) << 64 | u128(z_lo)
+	z_rem := int(z_len % u128(w))
+	pad := _PAD[:w - z_rem]
+
+	// We just add the padding to the state, instead of returning z.
+	//
+	// 4. return z.
+	update(ctx, pad)
+}
+
+encode_string :: #force_inline proc(ctx: ^Context, s: []byte) -> (u64, u64) {
+	l := encode_byte_len(ctx, len(s), true) // left_encode
+	update(ctx, s)
+
+	lo, hi := bits.add_u64(l, u64(len(s)), 0)
+
+	return hi, lo
+}
+
+encode_byte_len :: #force_inline proc(ctx: ^Context, l: int, is_left: bool) -> u64 {
+	hi, lo := bits.mul_u64(u64(l), 8)
+	return left_right_encode(ctx, hi, lo, is_left)
+}
+
+@(private)
+left_right_encode :: proc(ctx: ^Context, hi, lo: u64, is_left: bool) -> u64 {
+	HI_OFFSET :: 1
+	LO_OFFSET :: HI_OFFSET + 8
+	RIGHT_OFFSET :: LO_OFFSET + 8
+	BUF_LEN :: RIGHT_OFFSET + 1
+
+	buf: [BUF_LEN]byte // prefix + largest uint + postfix
+
+	endian.unchecked_put_u64be(buf[HI_OFFSET:], hi)
+	endian.unchecked_put_u64be(buf[LO_OFFSET:], lo)
+
+	// 2. Strip leading `0x00` bytes.
+	off: int
+	for off = HI_OFFSET; off < RIGHT_OFFSET - 1; off = off + 1 {// Note: Minimum size is 1, not 0.
+		if buf[off] != 0 {
+			break
+		}
+	}
+	n := byte(RIGHT_OFFSET - off)
+
+	// 3. Prefix (left_encode) or postfix (right_encode) the length in bytes.
+	b: []byte
+	switch is_left {
+	case true:
+		buf[off - 1] = n // n | x
+		b = buf[off - 1:RIGHT_OFFSET]
+	case false:
+		buf[RIGHT_OFFSET] = n // x | n
+		b = buf[off:]
+	}
+
+	update(ctx, b)
+
+	return u64(len(b))
+}

core/crypto/chacha20/chacha20.odin

@@ -1,11 +1,21 @@
+/*
+package chacha20 implements the ChaCha20 and XChaCha20 stream ciphers.
+
+See:
+- https://datatracker.ietf.org/doc/html/rfc8439
+- https://datatracker.ietf.org/doc/draft-irtf-cfrg-xchacha/03/
+*/
 package chacha20
 
 import "core:encoding/endian"
 import "core:math/bits"
 import "core:mem"
 
+// KEY_SIZE is the (X)ChaCha20 key size in bytes.
 KEY_SIZE :: 32
+// NONCE_SIZE is the ChaCha20 nonce size in bytes.
 NONCE_SIZE :: 12
+// XNONCE_SIZE is the XChaCha20 nonce size in bytes.
 XNONCE_SIZE :: 24
 
 @(private)
@@ -19,25 +29,26 @@ _STATE_SIZE_U32 :: 16
 _ROUNDS :: 20
 
 @(private)
-_SIGMA_0 : u32 : 0x61707865
+_SIGMA_0: u32 : 0x61707865
 @(private)
-_SIGMA_1 : u32 : 0x3320646e
+_SIGMA_1: u32 : 0x3320646e
 @(private)
-_SIGMA_2 : u32 : 0x79622d32
+_SIGMA_2: u32 : 0x79622d32
 @(private)
-_SIGMA_3 : u32 : 0x6b206574
+_SIGMA_3: u32 : 0x6b206574
 
+// Context is a ChaCha20 or XChaCha20 instance.
 Context :: struct {
-	_s: [_STATE_SIZE_U32]u32,
-
-	_buffer: [_BLOCK_SIZE]byte,
-	_off: int,
-
+	_s:              [_STATE_SIZE_U32]u32,
+	_buffer:         [_BLOCK_SIZE]byte,
+	_off:            int,
 	_is_ietf_flavor: bool,
 	_is_initialized: bool,
 }
 
-init :: proc (ctx: ^Context, key, nonce: []byte) {
+// init inititializes a Context for ChaCha20 or XChaCha20 with the provided
+// key and nonce.
+init :: proc(ctx: ^Context, key, nonce: []byte) {
 	if len(key) != KEY_SIZE {
 		panic("crypto/chacha20: invalid ChaCha20 key size")
 	}
@@ -89,7 +100,8 @@ init :: proc (ctx: ^Context, key, nonce: []byte) {
 	ctx._is_initialized = true
 }
 
-seek :: proc (ctx: ^Context, block_nr: u64) {
+// seek seeks the (X)ChaCha20 stream counter to the specified block.
+seek :: proc(ctx: ^Context, block_nr: u64) {
 	assert(ctx._is_initialized)
 
 	if ctx._is_ietf_flavor {
@@ -103,7 +115,10 @@ seek :: proc (ctx: ^Context, block_nr: u64) {
 	ctx._off = _BLOCK_SIZE
 }
 
-xor_bytes :: proc (ctx: ^Context, dst, src: []byte) {
+// xor_bytes XORs each byte in src with bytes taken from the (X)ChaCha20
+// keystream, and writes the resulting output to dst.  Dst and src MUST
+// alias exactly or not at all.
+xor_bytes :: proc(ctx: ^Context, dst, src: []byte) {
 	assert(ctx._is_initialized)
 
 	// TODO: Enforcing that dst and src alias exactly or not at all
@@ -147,7 +162,8 @@ xor_bytes :: proc (ctx: ^Context, dst, src: []byte) {
 	}
 }
 
-keystream_bytes :: proc (ctx: ^Context, dst: []byte) {
+// keystream_bytes fills dst with the raw (X)ChaCha20 keystream output.
+keystream_bytes :: proc(ctx: ^Context, dst: []byte) {
 	assert(ctx._is_initialized)
 
 	dst := dst
@@ -180,7 +196,9 @@ keystream_bytes :: proc (ctx: ^Context, dst: []byte) {
 	}
 }
 
-reset :: proc (ctx: ^Context) {
+// reset sanitizes the Context.  The Context must be re-initialized to
+// be used again.
+reset :: proc(ctx: ^Context) {
 	mem.zero_explicit(&ctx._s, size_of(ctx._s))
 	mem.zero_explicit(&ctx._buffer, size_of(ctx._buffer))
 
@@ -188,7 +206,7 @@ reset :: proc (ctx: ^Context) {
 }
 
 @(private)
-_do_blocks :: proc (ctx: ^Context, dst, src: []byte, nr_blocks: int) {
+_do_blocks :: proc(ctx: ^Context, dst, src: []byte, nr_blocks: int) {
 	// Enforce the maximum consumed keystream per nonce.
 	//
 	// While all modern "standard" definitions of ChaCha20 use

core/crypto/chacha20poly1305/chacha20poly1305.odin

@@ -1,3 +1,10 @@
+/*
+package chacha20poly1305 implements the AEAD_CHACHA20_POLY1305 Authenticated
+Encryption with Additional Data algorithm.
+
+See:
+- https://www.rfc-editor.org/rfc/rfc8439
+*/
 package chacha20poly1305
 
 import "core:crypto"
@@ -6,8 +13,11 @@ import "core:crypto/poly1305"
 import "core:encoding/endian"
 import "core:mem"
 
+// KEY_SIZE is the chacha20poly1305 key size in bytes.
 KEY_SIZE :: chacha20.KEY_SIZE
+// NONCE_SIZE is the chacha20poly1305 nonce size in bytes.
 NONCE_SIZE :: chacha20.NONCE_SIZE
+// TAG_SIZE is the chacha20poly1305 tag size in bytes.
 TAG_SIZE :: poly1305.TAG_SIZE
 
 @(private)
@@ -49,6 +59,8 @@ _update_mac_pad16 :: #force_inline proc (ctx: ^poly1305.Context, x_len: int) {
 	}
 }
 
+// encrypt encrypts the plaintext and authenticates the aad and ciphertext,
+// with the provided key and nonce, stores the output in ciphertext and tag.
 encrypt :: proc (ciphertext, tag, key, nonce, aad, plaintext: []byte) {
 	_validate_common_slice_sizes(tag, key, nonce, aad, plaintext)
 	if len(ciphertext) != len(plaintext) {
@@ -95,6 +107,11 @@ encrypt :: proc (ciphertext, tag, key, nonce, aad, plaintext: []byte) {
 	poly1305.final(&mac_ctx, tag) // Implicitly sanitizes context.
 }
 
+// decrypt authenticates the aad and ciphertext, and decrypts the ciphertext,
+// with the provided key, nonce, and tag, and stores the output in plaintext,
+// returning true iff the authentication was successful.
+//
+// If authentication fails, the destination plaintext buffer will be zeroed.
 decrypt :: proc (plaintext, tag, key, nonce, aad, ciphertext: []byte) -> bool {
 	_validate_common_slice_sizes(tag, key, nonce, aad, ciphertext)
 	if len(ciphertext) != len(plaintext) {

core/crypto/crypto.odin

@@ -1,3 +1,7 @@
+/*
+package crypto implements a selection of cryptography algorithms and useful
+helper routines.
+*/
 package crypto
 
 import "core:mem"
@@ -51,3 +55,9 @@ 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()
+}

core/crypto/ed25519/ed25519.odin

@@ -0,0 +1,314 @@
+/*
+package ed25519 implements the Ed25519 EdDSA signature algorithm.
+
+See:
+- https://datatracker.ietf.org/doc/html/rfc8032
+- https://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.186-5.pdf
+- https://eprint.iacr.org/2020/1244.pdf
+*/
+package ed25519
+
+import "core:crypto"
+import grp "core:crypto/_edwards25519"
+import "core:crypto/sha2"
+import "core:mem"
+
+// PRIVATE_KEY_SIZE is the byte-encoded private key size.
+PRIVATE_KEY_SIZE :: 32
+// PUBLIC_KEY_SIZE is the byte-encoded public key size.
+PUBLIC_KEY_SIZE :: 32
+// SIGNATURE_SIZE is the byte-encoded signature size.
+SIGNATURE_SIZE :: 64
+
+@(private)
+NONCE_SIZE :: 32
+
+// Private_Key is an Ed25519 private key.
+Private_Key :: struct {
+	// WARNING: All of the members are to be treated as internal (ie:
+	// the Private_Key structure is intended to be opaque).  There are
+	// subtle vulnerabilities that can be introduced if the internal
+	// values are allowed to be altered.
+	//
+	// See: https://github.com/MystenLabs/ed25519-unsafe-libs
+	_b:              [PRIVATE_KEY_SIZE]byte,
+	_s:              grp.Scalar,
+	_nonce:          [NONCE_SIZE]byte,
+	_pub_key:        Public_Key,
+	_is_initialized: bool,
+}
+
+// Public_Key is an Ed25519 public key.
+Public_Key :: struct {
+	// WARNING: All of the members are to be treated as internal (ie:
+	// the Public_Key structure is intended to be opaque).
+	_b:              [PUBLIC_KEY_SIZE]byte,
+	_neg_A:          grp.Group_Element,
+	_is_valid:       bool,
+	_is_initialized: bool,
+}
+
+// private_key_set_bytes decodes a byte-encoded private key, and returns
+// true iff the operation was successful.
+private_key_set_bytes :: proc(priv_key: ^Private_Key, b: []byte) -> bool {
+	if len(b) != PRIVATE_KEY_SIZE {
+		return false
+	}
+
+	// Derive the private key.
+	ctx: sha2.Context_512 = ---
+	h_bytes: [sha2.DIGEST_SIZE_512]byte = ---
+	sha2.init_512(&ctx)
+	sha2.update(&ctx, b)
+	sha2.final(&ctx, h_bytes[:])
+
+	copy(priv_key._b[:], b)
+	copy(priv_key._nonce[:], h_bytes[32:])
+	grp.sc_set_bytes_rfc8032(&priv_key._s, h_bytes[:32])
+
+	// Derive the corresponding public key.
+	A: grp.Group_Element = ---
+	grp.ge_scalarmult_basepoint(&A, &priv_key._s)
+	grp.ge_bytes(&A, priv_key._pub_key._b[:])
+	grp.ge_negate(&priv_key._pub_key._neg_A, &A)
+	priv_key._pub_key._is_valid = !grp.ge_is_small_order(&A)
+	priv_key._pub_key._is_initialized = true
+
+	priv_key._is_initialized = true
+
+	return true
+}
+
+// private_key_bytes sets dst to byte-encoding of priv_key.
+private_key_bytes :: proc(priv_key: ^Private_Key, dst: []byte) {
+	if !priv_key._is_initialized {
+		panic("crypto/ed25519: uninitialized private key")
+	}
+	if len(dst) != PRIVATE_KEY_SIZE {
+		panic("crypto/ed25519: invalid destination size")
+	}
+
+	copy(dst, priv_key._b[:])
+}
+
+// private_key_clear clears priv_key to the uninitialized state.
+private_key_clear :: proc "contextless" (priv_key: ^Private_Key) {
+	mem.zero_explicit(priv_key, size_of(Private_Key))
+}
+
+// sign writes the signature by priv_key over msg to sig.
+sign :: proc(priv_key: ^Private_Key, msg, sig: []byte) {
+	if !priv_key._is_initialized {
+		panic("crypto/ed25519: uninitialized private key")
+	}
+	if len(sig) != SIGNATURE_SIZE {
+		panic("crypto/ed25519: invalid destination size")
+	}
+
+	// 1. Compute the hash of the private key d, H(d) = (h_0, h_1, ..., h_2b-1)
+	// using SHA-512 for Ed25519.  H(d) may be precomputed.
+	//
+	// 2. Using the second half of the digest hdigest2 = hb || ... || h2b-1,
+	// define:
+	//
+	// 2.1 For Ed25519, r = SHA-512(hdigest2 || M); Interpret r as a
+	// 64-octet little-endian integer.
+	ctx: sha2.Context_512 = ---
+	digest_bytes: [sha2.DIGEST_SIZE_512]byte = ---
+	sha2.init_512(&ctx)
+	sha2.update(&ctx, priv_key._nonce[:])
+	sha2.update(&ctx, msg)
+	sha2.final(&ctx, digest_bytes[:])
+
+	r: grp.Scalar = ---
+	grp.sc_set_bytes_wide(&r, &digest_bytes)
+
+	// 3. Compute the point [r]G. The octet string R is the encoding of
+	// the point [r]G.
+	R: grp.Group_Element = ---
+	R_bytes := sig[:32]
+	grp.ge_scalarmult_basepoint(&R, &r)
+	grp.ge_bytes(&R, R_bytes)
+
+	// 4. Derive s from H(d) as in the key pair generation algorithm.
+	// Use octet strings R, Q, and M to define:
+	//
+	// 4.1 For Ed25519, digest = SHA-512(R || Q || M).
+	// Interpret digest as a little-endian integer.
+	sha2.init_512(&ctx)
+	sha2.update(&ctx, R_bytes)
+	sha2.update(&ctx, priv_key._pub_key._b[:]) // Q in NIST terminology.
+	sha2.update(&ctx, msg)
+	sha2.final(&ctx, digest_bytes[:])
+
+	sc: grp.Scalar = --- // `digest` in NIST terminology.
+	grp.sc_set_bytes_wide(&sc, &digest_bytes)
+
+	// 5. Compute S = (r + digest × s) mod n. The octet string S is the
+	// encoding of the resultant integer.
+	grp.sc_mul(&sc, &sc, &priv_key._s)
+	grp.sc_add(&sc, &sc, &r)
+
+	// 6. Form the signature as the concatenation of the octet strings
+	// R and S.
+	grp.sc_bytes(sig[32:], &sc)
+
+	grp.sc_clear(&r)
+}
+
+// public_key_set_bytes decodes a byte-encoded public key, and returns
+// true iff the operation was successful.
+public_key_set_bytes :: proc "contextless" (pub_key: ^Public_Key, b: []byte) -> bool {
+	if len(b) != PUBLIC_KEY_SIZE {
+		return false
+	}
+
+	A: grp.Group_Element = ---
+	if !grp.ge_set_bytes(&A, b) {
+		return false
+	}
+
+	copy(pub_key._b[:], b)
+	grp.ge_negate(&pub_key._neg_A, &A)
+	pub_key._is_valid = !grp.ge_is_small_order(&A)
+	pub_key._is_initialized = true
+
+	return true
+}
+
+// public_key_set_priv sets pub_key to the public component of priv_key.
+public_key_set_priv :: proc(pub_key: ^Public_Key, priv_key: ^Private_Key) {
+	if !priv_key._is_initialized {
+		panic("crypto/ed25519: uninitialized public key")
+	}
+
+	src := &priv_key._pub_key
+	copy(pub_key._b[:], src._b[:])
+	grp.ge_set(&pub_key._neg_A, &src._neg_A)
+	pub_key._is_valid = src._is_valid
+	pub_key._is_initialized = src._is_initialized
+}
+
+// public_key_bytes sets dst to byte-encoding of pub_key.
+public_key_bytes :: proc(pub_key: ^Public_Key, dst: []byte) {
+	if !pub_key._is_initialized {
+		panic("crypto/ed25519: uninitialized public key")
+	}
+	if len(dst) != PUBLIC_KEY_SIZE {
+		panic("crypto/ed25519: invalid destination size")
+	}
+
+	copy(dst, pub_key._b[:])
+}
+
+// public_key_equal returns true iff pub_key is equal to other.
+public_key_equal :: proc(pub_key, other: ^Public_Key) -> bool {
+	if !pub_key._is_initialized || !other._is_initialized {
+		panic("crypto/ed25519: uninitialized public key")
+	}
+
+	return crypto.compare_constant_time(pub_key._b[:], other._b[:]) == 1
+}
+
+// verify returns true iff sig is a valid signature by pub_key over msg.
+//
+// The optional `allow_small_order_A` parameter will make this
+// implementation strictly compatible with FIPS 186-5, at the expense of
+// SBS-security.  Doing so is NOT recommended, and the disallowed
+// public keys all have a known discrete-log.
+verify :: proc(pub_key: ^Public_Key, msg, sig: []byte, allow_small_order_A := false) -> bool {
+	switch {
+	case !pub_key._is_initialized:
+		return false
+	case len(sig) != SIGNATURE_SIZE:
+		return false
+	}
+
+	// TLDR: Just use ristretto255.
+	//
+	// While there are two "standards" for EdDSA, existing implementations
+	// diverge (sometimes dramatically).  This implementation opts for
+	// "Algorithm 2" from "Taming the Many EdDSAs", which provides the
+	// strongest notion of security (SUF-CMA + SBS).
+	//
+	// The relevant properties are:
+	// - Reject non-canonical S.
+	// - Reject non-canonical A/R.
+	// - Reject small-order A (Extra non-standard check).
+	// - Cofactored verification equation.
+	//
+	// There are 19 possible non-canonical group element encodings of
+	// which:
+	// - 2 are small order
+	// - 10 are mixed order
+	// - 7 are not on the curve
+	//
+	// While historical implementations have been lax about enforcing
+	// that A/R are canonically encoded, that behavior is mandated by
+	// both the RFC and FIPS specification.  No valid key generation
+	// or sign implementation will ever produce non-canonically encoded
+	// public keys or signatures.
+	//
+	// There are 8 small-order group elements, 1 which is in the
+	// prime-order sub-group, and thus the probability that a properly
+	// generated A is small-order is cryptographically insignificant.
+	//
+	// While both the RFC and FIPS standard allow for either the
+	// cofactored or non-cofactored equation.  It is possible to
+	// artificially produce signatures that are valid for the former
+	// but not the latter.  This will NEVER occur with a valid sign
+	// implementation.  The choice of the latter is to be compatible
+	// with ABGLSV-Pornin, batch verification, and FROST (among other
+	// things).
+
+	s_bytes, r_bytes := sig[32:], sig[:32]
+
+	// 1. Reject the signature if S is not in the range [0, L).
+	s: grp.Scalar = ---
+	if !grp.sc_set_bytes(&s, s_bytes) {
+		return false
+	}
+
+	// 2. Reject the signature if the public key A is one of 8 small
+	// order points.
+	//
+	// As this check is optional and not part of the standard, we allow
+	// the caller to bypass it if desired.  Disabling the check makes
+	// the scheme NOT SBS-secure.
+	if !pub_key._is_valid && !allow_small_order_A {
+		return false
+	}
+
+	// 3. Reject the signature if A or R are non-canonical.
+	//
+	// Note: All initialized public keys are guaranteed to be canonical.
+	neg_R: grp.Group_Element = ---
+	if !grp.ge_set_bytes(&neg_R, r_bytes) {
+		return false
+	}
+	grp.ge_negate(&neg_R, &neg_R)
+
+	// 4. Compute the hash SHA512(R||A||M) and reduce it mod L to get a
+	// scalar h.
+	ctx: sha2.Context_512 = ---
+	h_bytes: [sha2.DIGEST_SIZE_512]byte = ---
+	sha2.init_512(&ctx)
+	sha2.update(&ctx, r_bytes)
+	sha2.update(&ctx, pub_key._b[:])
+	sha2.update(&ctx, msg)
+	sha2.final(&ctx, h_bytes[:])
+
+	h: grp.Scalar = ---
+	grp.sc_set_bytes_wide(&h, &h_bytes)
+
+	// 5. Accept if 8(s * G) - 8R - 8(h * A) = 0
+	//
+	// > first compute V = SB − R − hA and then accept if V is one of
+	// > 8 small order points (or alternatively compute 8V with 3
+	// > doublings and check against the neutral element)
+	V: grp.Group_Element = ---
+	grp.ge_double_scalarmult_basepoint_vartime(&V, &h, &pub_key._neg_A, &s)
+	grp.ge_add(&V, &V, &neg_R)
+
+	return grp.ge_is_small_order(&V)
+}

core/crypto/hkdf/hkdf.odin

@@ -0,0 +1,103 @@
+/*
+package hkdf implements the HKDF HMAC-based Extract-and-Expand Key
+Derivation Function.
+
+See: https://www.rfc-editor.org/rfc/rfc5869
+*/
+package hkdf
+
+import "core:crypto/hash"
+import "core:crypto/hmac"
+import "core:mem"
+
+// extract_and_expand derives output keying material (OKM) via the
+// HKDF-Extract and HKDF-Expand algorithms, with the specified has
+// function, salt, input keying material (IKM), and optional info.
+// The dst buffer must be less-than-or-equal to 255 HMAC tags.
+extract_and_expand :: proc(algorithm: hash.Algorithm, salt, ikm, info, dst: []byte) {
+	h_len := hash.DIGEST_SIZES[algorithm]
+
+	tmp: [hash.MAX_DIGEST_SIZE]byte
+	prk := tmp[:h_len]
+	defer mem.zero_explicit(raw_data(prk), h_len)
+
+	extract(algorithm, salt, ikm, prk)
+	expand(algorithm, prk, info, dst)
+}
+
+// extract derives a pseudorandom key (PRK) via the HKDF-Extract algorithm,
+// with the specified hash function, salt, and input keying material (IKM).
+// It requires that the dst buffer be the HMAC tag size for the specified
+// hash function.
+extract :: proc(algorithm: hash.Algorithm, salt, ikm, dst: []byte) {
+	// PRK = HMAC-Hash(salt, IKM)
+	hmac.sum(algorithm, dst, ikm, salt)
+}
+
+// expand derives output keying material (OKM) via the HKDF-Expand algorithm,
+// with the specified hash function, pseudorandom key (PRK), and optional
+// info.  The dst buffer must be less-than-or-equal to 255 HMAC tags.
+expand :: proc(algorithm: hash.Algorithm, prk, info, dst: []byte) {
+	h_len := hash.DIGEST_SIZES[algorithm]
+
+	// (<= 255*HashLen)
+	dk_len := len(dst)
+	switch {
+	case dk_len == 0:
+		return
+	case dk_len > h_len * 255:
+		panic("crypto/hkdf: derived key too long")
+	case:
+	}
+
+	// The output OKM is calculated as follows:
+	//
+	// N = ceil(L/HashLen)
+	// T = T(1) | T(2) | T(3) | ... | T(N)
+	// OKM = first L octets of T
+	//
+	// where:
+	// T(0) = empty string (zero length)
+	// T(1) = HMAC-Hash(PRK, T(0) | info | 0x01)
+	// T(2) = HMAC-Hash(PRK, T(1) | info | 0x02)
+	// T(3) = HMAC-Hash(PRK, T(2) | info | 0x03)
+	// ...
+
+	n := dk_len / h_len
+	r := dk_len % h_len
+
+	base: hmac.Context
+	defer hmac.reset(&base)
+
+	hmac.init(&base, algorithm, prk)
+
+	dst_blk := dst
+	prev: []byte
+
+	for i in 1 ..= n {
+		_F(&base, prev, info, i, dst_blk[:h_len])
+
+		prev = dst_blk[:h_len]
+		dst_blk = dst_blk[h_len:]
+	}
+
+	if r > 0 {
+		tmp: [hash.MAX_DIGEST_SIZE]byte
+		blk := tmp[:h_len]
+		defer mem.zero_explicit(raw_data(blk), h_len)
+
+		_F(&base, prev, info, n + 1, blk)
+		copy(dst_blk, blk)
+	}
+}
+
+@(private)
+_F :: proc(base: ^hmac.Context, prev, info: []byte, i: int, dst_blk: []byte) {
+	prf: hmac.Context
+
+	hmac.clone(&prf, base)
+	hmac.update(&prf, prev)
+	hmac.update(&prf, info)
+	hmac.update(&prf, []byte{u8(i)})
+	hmac.final(&prf, dst_blk)
+}

core/crypto/hmac/hmac.odin

@@ -11,7 +11,7 @@ import "core:crypto/hash"
 import "core:mem"
 
 // sum will compute the HMAC with the specified algorithm and key
-// over msg, and write the computed digest to dst.  It requires that
+// over msg, and write the computed tag to dst.  It requires that
 // the dst buffer is the tag size.
 sum :: proc(algorithm: hash.Algorithm, dst, msg, key: []byte) {
 	ctx: Context
@@ -78,6 +78,18 @@ final :: proc(ctx: ^Context, dst: []byte) {
 	hash.final(&ctx._o_hash, dst)
 }
 
+// clone clones the Context other into ctx.
+clone :: proc(ctx, other: ^Context) {
+	if ctx == other {
+		return
+	}
+
+	hash.clone(&ctx._o_hash, &other._o_hash)
+	hash.clone(&ctx._i_hash, &other._i_hash)
+	ctx._tag_sz = other._tag_sz
+	ctx._is_initialized = other._is_initialized
+}
+
 // reset sanitizes the Context.  The Context must be re-initialized to
 // be used again.
 reset :: proc(ctx: ^Context) {

core/crypto/kmac/kmac.odin

@@ -0,0 +1,116 @@
+/*
+package kmac implements the KMAC MAC algorithm.
+
+See:
+- https://nvlpubs.nist.gov/nistpubs/specialpublications/nist.sp.800-185.pdf
+*/
+package kmac
+
+import "../_sha3"
+import "core:crypto"
+import "core:crypto/shake"
+
+// MIN_KEY_SIZE_128 is the minimum key size for KMAC128 in bytes.
+MIN_KEY_SIZE_128 :: 128 / 8
+// MIN_KEY_SIZE_256 is the minimum key size for KMAC256 in bytes.
+MIN_KEY_SIZE_256 :: 256 / 8
+
+// MIN_TAG_SIZE is the absolute minimum tag size for KMAC in bytes (8.4.2).
+// Most callers SHOULD use at least 128-bits if not 256-bits for the tag
+// size.
+MIN_TAG_SIZE :: 32 / 8
+
+// sum will compute the KMAC with the specified security strength,
+// key, and domain separator over msg, and write the computed digest to
+// dst.
+sum :: proc(sec_strength: int, dst, msg, key, domain_sep: []byte) {
+	ctx: Context
+
+	_init_kmac(&ctx, key, domain_sep, sec_strength)
+	update(&ctx, msg)
+	final(&ctx, dst)
+}
+
+// verify will verify the KMAC tag computed with the specified security
+// strength, key and domain separator over msg and return true iff the
+// tag is valid.
+verify :: proc(sec_strength: int, tag, msg, key, domain_sep: []byte, allocator := context.temp_allocator) -> bool {
+	derived_tag := make([]byte, len(tag), allocator)
+
+	sum(sec_strength, derived_tag, msg, key, domain_sep)
+
+	return crypto.compare_constant_time(derived_tag, tag) == 1
+}
+
+// Context is a KMAC instance.
+Context :: distinct shake.Context
+
+// init_128 initializes a Context for KMAC28.  This routine will panic if
+// the key length is less than MIN_KEY_SIZE_128.
+init_128 :: proc(ctx: ^Context, key, domain_sep: []byte) {
+	_init_kmac(ctx, key, domain_sep, 128)
+}
+
+// init_256 initializes a Context for KMAC256.  This routine will panic if
+// the key length is less than MIN_KEY_SIZE_256.
+init_256 :: proc(ctx: ^Context, key, domain_sep: []byte) {
+	_init_kmac(ctx, key, domain_sep, 256)
+}
+
+// update adds more data to the Context.
+update :: proc(ctx: ^Context, data: []byte) {
+	assert(ctx.is_initialized)
+
+	shake.write(transmute(^shake.Context)(ctx), data)
+}
+
+// final finalizes the Context, writes the tag to dst, and calls reset
+// on the Context.  This routine will panic if the dst length is less than
+// MIN_TAG_SIZE.
+final :: proc(ctx: ^Context, dst: []byte) {
+	assert(ctx.is_initialized)
+	defer reset(ctx)
+
+	if len(dst) < MIN_TAG_SIZE {
+		panic("crypto/kmac: invalid KMAC tag_size, too short")
+	}
+
+	_sha3.final_cshake(transmute(^_sha3.Context)(ctx), dst)
+}
+
+// clone clones the Context other into ctx.
+clone :: proc(ctx, other: ^Context) {
+	if ctx == other {
+		return
+	}
+
+	shake.clone(transmute(^shake.Context)(ctx), transmute(^shake.Context)(other))
+}
+
+// reset sanitizes the Context.  The Context must be re-initialized to
+// be used again.
+reset :: proc(ctx: ^Context) {
+	if !ctx.is_initialized {
+		return
+	}
+
+	shake.reset(transmute(^shake.Context)(ctx))
+}
+
+@(private)
+_init_kmac :: proc(ctx: ^Context, key, s: []byte, sec_strength: int) {
+	if ctx.is_initialized {
+		reset(ctx)
+	}
+
+	if len(key) < sec_strength / 8 {
+		panic("crypto/kmac: invalid KMAC key, too short")
+	}
+
+	ctx_ := transmute(^_sha3.Context)(ctx)
+	_sha3.init_cshake(ctx_, N_KMAC, s, sec_strength)
+	_sha3.bytepad(ctx_, [][]byte{key}, _sha3.rate_cshake(sec_strength))
+}
+
+@(private)
+N_KMAC := []byte{'K', 'M', 'A', 'C'}

core/crypto/legacy/keccak/keccak.odin

@@ -65,7 +65,7 @@ init_512 :: proc(ctx: ^Context) {
 
 @(private)
 _init :: proc(ctx: ^Context) {
-	ctx.is_keccak = true
+	ctx.dsbyte = _sha3.DS_KECCAK
 	_sha3.init(transmute(^_sha3.Context)(ctx))
 }
 

core/crypto/pbkdf2/pbkdf2.odin

@@ -0,0 +1,122 @@
+/*
+package pbkdf2 implements the PBKDF2 password-based key derivation function.
+
+See: https://www.rfc-editor.org/rfc/rfc2898
+*/
+package pbkdf2
+
+import "core:crypto/hash"
+import "core:crypto/hmac"
+import "core:encoding/endian"
+import "core:mem"
+
+// derive invokes PBKDF2-HMAC with the specified hash algorithm, password,
+// salt, iteration count, and outputs the derived key to dst.
+derive :: proc(
+	hmac_hash: hash.Algorithm,
+	password: []byte,
+	salt: []byte,
+	iterations: u32,
+	dst: []byte,
+) {
+	h_len := hash.DIGEST_SIZES[hmac_hash]
+
+	// 1. If dkLen > (2^32 - 1) * hLen, output "derived key too long"
+	// and stop.
+
+	dk_len := len(dst)
+	switch {
+	case dk_len == 0:
+		return
+	case u64(dk_len) > u64(max(u32)) * u64(h_len):
+		// This is so beyond anything that is practical or reasonable,
+		// so just panic instead of returning an error.
+		panic("crypto/pbkdf2: derived key too long")
+	case:
+	}
+
+	// 2. Let l be the number of hLen-octet blocks in the derived key,
+	// rounding up, and let r be the number of octets in the last block.
+
+	l := dk_len / h_len // Don't need to round up.
+	r := dk_len % h_len
+
+	// 3. For each block of the derived key apply the function F defined
+	// below to the password P, the salt S, the iteration count c, and
+	// the block index to compute the block.
+	//
+	// 4. Concatenate the blocks and extract the first dkLen octets to
+	// produce a derived key DK.
+	//
+	// 5. Output the derived key DK.
+
+	// Each iteration of F is always `PRF (P, ...)`, so instantiate the
+	// PRF, and clone since memcpy is faster than having to re-initialize
+	// HMAC repeatedly.
+
+	base: hmac.Context
+	defer hmac.reset(&base)
+
+	hmac.init(&base, hmac_hash, password)
+
+	// Process all of the blocks that will be written directly to dst.
+	dst_blk := dst
+	for i in 1 ..= l { 	// F expects i starting at 1.
+		_F(&base, salt, iterations, u32(i), dst_blk[:h_len])
+		dst_blk = dst_blk[h_len:]
+	}
+
+	// Instead of rounding l up, just proceass the one extra block iff
+	// r != 0.
+	if r > 0 {
+		tmp: [hash.MAX_DIGEST_SIZE]byte
+		blk := tmp[:h_len]
+		defer mem.zero_explicit(raw_data(blk), h_len)
+
+		_F(&base, salt, iterations, u32(l + 1), blk)
+		copy(dst_blk, blk)
+	}
+}
+
+@(private)
+_F :: proc(base: ^hmac.Context, salt: []byte, c: u32, i: u32, dst_blk: []byte) {
+	h_len := len(dst_blk)
+
+	tmp: [hash.MAX_DIGEST_SIZE]byte
+	u := tmp[:h_len]
+	defer mem.zero_explicit(raw_data(u), h_len)
+
+	// F (P, S, c, i) = U_1 \xor U_2 \xor ... \xor U_c
+	//
+	// where
+	//
+	// U_1 = PRF (P, S || INT (i)) ,
+	// U_2 = PRF (P, U_1) ,
+	// ...
+	// U_c = PRF (P, U_{c-1}) .
+	//
+	// Here, INT (i) is a four-octet encoding of the integer i, most
+	// significant octet first.
+
+	prf: hmac.Context
+
+	// U_1: PRF (P, S || INT (i))
+	hmac.clone(&prf, base)
+	hmac.update(&prf, salt)
+	endian.unchecked_put_u32be(u, i) // Use u as scratch space.
+	hmac.update(&prf, u[:4])
+	hmac.final(&prf, u)
+	copy(dst_blk, u)
+
+	// U_2 ... U_c: U_n = PRF (P, U_(n-1))
+	for _ in 1 ..< c {
+		hmac.clone(&prf, base)
+		hmac.update(&prf, u)
+		hmac.final(&prf, u)
+
+		// XOR dst_blk and u.
+		for v, i in u {
+			dst_blk[i] ~= v
+		}
+	}
+}

core/crypto/poly1305/poly1305.odin

@@ -1,17 +1,31 @@
+/*
+package poly1305 implements the Poly1305 one-time MAC algorithm.
+
+See:
+- https://datatracker.ietf.org/doc/html/rfc8439
+*/
 package poly1305
 
 import "core:crypto"
 import field "core:crypto/_fiat/field_poly1305"
 import "core:encoding/endian"
+import "core:math/bits"
 import "core:mem"
 
+// KEY_SIZE is the Poly1305 key size in bytes.
 KEY_SIZE :: 32
+// TAG_SIZE is the Poly1305 tag size in bytes.
 TAG_SIZE :: 16
 
 @(private)
 _BLOCK_SIZE :: 16
 
-sum :: proc (dst, msg, key: []byte) {
+// sum will compute the Poly1305 MAC with the key over msg, and write
+// the computed tag to dst.  It requires that the dst buffer is the tag
+// size.
+//
+// The key SHOULD be unique and MUST be unpredictable for each invocation.
+sum :: proc(dst, msg, key: []byte) {
 	ctx: Context = ---
 
 	init(&ctx, key)
@@ -19,9 +33,12 @@ sum :: proc (dst, msg, key: []byte) {
 	final(&ctx, dst)
 }
 
-verify :: proc (tag, msg, key: []byte) -> bool {
+// verify will verify the Poly1305 tag computed with the key over msg and
+// return true iff the tag is valid.  It requires that the tag is correctly
+// sized.
+verify :: proc(tag, msg, key: []byte) -> bool {
 	ctx: Context = ---
-	derived_tag: [16]byte = ---
+	derived_tag: [TAG_SIZE]byte = ---
 
 	init(&ctx, key)
 	update(&ctx, msg)
@@ -30,18 +47,19 @@ verify :: proc (tag, msg, key: []byte) -> bool {
 	return crypto.compare_constant_time(derived_tag[:], tag) == 1
 }
 
+// Context is a Poly1305 instance.
 Context :: struct {
-	_r: field.Tight_Field_Element,
-	_a: field.Tight_Field_Element,
-	_s: field.Tight_Field_Element,
-
-	_buffer: [_BLOCK_SIZE]byte,
-	_leftover: int,
-
+	_r:              field.Tight_Field_Element,
+	_a:              field.Tight_Field_Element,
+	_s:              [2]u64,
+	_buffer:         [_BLOCK_SIZE]byte,
+	_leftover:       int,
 	_is_initialized: bool,
 }
 
-init :: proc (ctx: ^Context, key: []byte) {
+// init initializes a Context with the specified key.  The key SHOULD be
+// unique and MUST be unpredictable for each invocation.
+init :: proc(ctx: ^Context, key: []byte) {
 	if len(key) != KEY_SIZE {
 		panic("crypto/poly1305: invalid key size")
 	}
@@ -49,11 +67,12 @@ init :: proc (ctx: ^Context, key: []byte) {
 	// r = le_bytes_to_num(key[0..15])
 	// r = clamp(r) (r &= 0xffffffc0ffffffc0ffffffc0fffffff)
 	tmp_lo := endian.unchecked_get_u64le(key[0:]) & 0x0ffffffc0fffffff
-	tmp_hi := endian.unchecked_get_u64le(key[8:]) & 0xffffffc0ffffffc
+	tmp_hi := endian.unchecked_get_u64le(key[8:]) & 0x0ffffffc0ffffffc
 	field.fe_from_u64s(&ctx._r, tmp_lo, tmp_hi)
 
 	// s = le_bytes_to_num(key[16..31])
-	field.fe_from_bytes(&ctx._s, key[16:32], 0)
+	ctx._s[0] = endian.unchecked_get_u64le(key[16:])
+	ctx._s[1] = endian.unchecked_get_u64le(key[24:])
 
 	// a = 0
 	field.fe_zero(&ctx._a)
@@ -64,7 +83,8 @@ init :: proc (ctx: ^Context, key: []byte) {
 	ctx._is_initialized = true
 }
 
-update :: proc (ctx: ^Context, data: []byte) {
+// update adds more data to the Context.
+update :: proc(ctx: ^Context, data: []byte) {
 	assert(ctx._is_initialized)
 
 	msg := data
@@ -101,8 +121,11 @@ update :: proc (ctx: ^Context, data: []byte) {
 	}
 }
 
-final :: proc (ctx: ^Context, dst: []byte) {
+// final finalizes the Context, writes the tag to dst, and calls
+// reset on the Context.
+final :: proc(ctx: ^Context, dst: []byte) {
 	assert(ctx._is_initialized)
+	defer reset(ctx)
 
 	if len(dst) != TAG_SIZE {
 		panic("poly1305: invalid destination tag size")
@@ -117,19 +140,25 @@ final :: proc (ctx: ^Context, dst: []byte) {
 		_blocks(ctx, ctx._buffer[:], true)
 	}
 
-	// a += s
-	field.fe_add(field.fe_relax_cast(&ctx._a), &ctx._a, &ctx._s) // _a unreduced
-	field.fe_carry(&ctx._a, field.fe_relax_cast(&ctx._a)) // _a reduced
-
-	// return num_to_16_le_bytes(a)
+	// a += s (NOT mod p)
 	tmp: [32]byte = ---
 	field.fe_to_bytes(&tmp, &ctx._a)
-	copy_slice(dst, tmp[0:16])
 
-	reset(ctx)
+	c: u64
+	lo := endian.unchecked_get_u64le(tmp[0:])
+	hi := endian.unchecked_get_u64le(tmp[8:])
+
+	lo, c = bits.add_u64(lo, ctx._s[0], 0)
+	hi, _ = bits.add_u64(hi, ctx._s[1], c)
+
+	// return num_to_16_le_bytes(a)
+	endian.unchecked_put_u64le(dst[0:], lo)
+	endian.unchecked_put_u64le(dst[8:], hi)
 }
 
-reset :: proc (ctx: ^Context) {
+// reset sanitizes the Context.  The Context must be re-initialized to
+// be used again.
+reset :: proc(ctx: ^Context) {
 	mem.zero_explicit(&ctx._r, size_of(ctx._r))
 	mem.zero_explicit(&ctx._a, size_of(ctx._a))
 	mem.zero_explicit(&ctx._s, size_of(ctx._s))
@@ -139,7 +168,7 @@ reset :: proc (ctx: ^Context) {
 }
 
 @(private)
-_blocks :: proc (ctx: ^Context, msg: []byte, final := false) {
+_blocks :: proc "contextless" (ctx: ^Context, msg: []byte, final := false) {
 	n: field.Tight_Field_Element = ---
 	final_byte := byte(!final)
 

core/crypto/rand_bsd.odin

@@ -0,0 +1,16 @@
+//+build freebsd, openbsd
+package crypto
+
+foreign import libc "system:c"
+
+foreign libc {
+	arc4random_buf :: proc(buf: [^]byte, nbytes: uint) ---
+}
+
+_rand_bytes :: proc(dst: []byte) {
+	arc4random_buf(raw_data(dst), len(dst))
+}
+
+_has_rand_bytes :: proc () -> bool {
+	return true
+}

core/crypto/rand_darwin.odin

@@ -0,0 +1,16 @@
+package crypto
+
+import "core:fmt"
+import "core:sys/darwin"
+
+_rand_bytes :: proc(dst: []byte) {
+	res := darwin.SecRandomCopyBytes(count=len(dst), bytes=raw_data(dst))
+	if res != .Success {
+		msg := darwin.CFStringCopyToOdinString(darwin.SecCopyErrorMessageString(res))
+		panic(fmt.tprintf("crypto/rand_bytes: SecRandomCopyBytes returned non-zero result: %v %s", res, msg))
+	}
+}
+
+_has_rand_bytes :: proc () -> bool {
+	return true
+}

core/crypto/rand_generic.odin

@@ -1,7 +1,15 @@
+//+build !linux
+//+build !windows
+//+build !openbsd
+//+build !freebsd
+//+build !darwin
+//+build !js
 package crypto
 
-when ODIN_OS != .Linux && ODIN_OS != .OpenBSD && ODIN_OS != .Windows && ODIN_OS != .JS {
-	_rand_bytes :: proc(dst: []byte) {
-		unimplemented("crypto: rand_bytes not supported on this OS")
-	}
+_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

@@ -18,3 +18,7 @@ _rand_bytes :: proc(dst: []byte) {
 		dst = dst[to_read:]
 	}
 }
+
+_has_rand_bytes :: proc () -> bool {
+	return true
+}

core/crypto/rand_linux.odin

@@ -34,3 +34,7 @@ _rand_bytes :: proc (dst: []byte) {
 		dst = dst[n_read:]
 	}
 }
+
+_has_rand_bytes :: proc () -> bool {
+	return true
+}

core/crypto/rand_openbsd.odin

@@ -1,12 +0,0 @@
-package crypto
-
-import "core:c"
-
-foreign import libc "system:c"
-foreign libc {
-	arc4random_buf :: proc "c" (buf: rawptr, nbytes: c.size_t) ---
-}
-
-_rand_bytes :: proc (dst: []byte) {
-	arc4random_buf(raw_data(dst), len(dst))
-}

core/crypto/rand_windows.odin

@@ -21,3 +21,7 @@ _rand_bytes :: proc(dst: []byte) {
 		}
 	}
 }
+
+_has_rand_bytes :: proc () -> bool {
+	return true
+}

core/crypto/ristretto255/ristretto255.odin

@@ -0,0 +1,510 @@
+/*
+package ristretto255 implement the ristretto255 prime-order group.
+
+See:
+- https://www.rfc-editor.org/rfc/rfc9496
+*/
+package ristretto255
+
+import grp "core:crypto/_edwards25519"
+import field "core:crypto/_fiat/field_curve25519"
+import "core:mem"
+
+// ELEMENT_SIZE is the size of a byte-encoded ristretto255 group element.
+ELEMENT_SIZE :: 32
+// WIDE_ELEMENT_SIZE is the side of a wide byte-encoded ristretto255
+// group element.
+WIDE_ELEMENT_SIZE :: 64
+
+@(private)
+FE_NEG_ONE := field.Tight_Field_Element {
+	2251799813685228,
+	2251799813685247,
+	2251799813685247,
+	2251799813685247,
+	2251799813685247,
+}
+@(private)
+FE_INVSQRT_A_MINUS_D := field.Tight_Field_Element {
+	278908739862762,
+	821645201101625,
+	8113234426968,
+	1777959178193151,
+	2118520810568447,
+}
+@(private)
+FE_ONE_MINUS_D_SQ := field.Tight_Field_Element {
+	1136626929484150,
+	1998550399581263,
+	496427632559748,
+	118527312129759,
+	45110755273534,
+}
+@(private)
+FE_D_MINUS_ONE_SQUARED := field.Tight_Field_Element {
+	1507062230895904,
+	1572317787530805,
+	683053064812840,
+	317374165784489,
+	1572899562415810,
+}
+@(private)
+FE_SQRT_AD_MINUS_ONE := field.Tight_Field_Element {
+	2241493124984347,
+	425987919032274,
+	2207028919301688,
+	1220490630685848,
+	974799131293748,
+}
+@(private)
+GE_IDENTITY := Group_Element{grp.GE_IDENTITY, true}
+
+// Group_Element is a ristretto255 group element.  The zero-initialized
+// value is invalid.
+Group_Element :: struct {
+	// WARNING: While the internal representation is an Edwards25519
+	// group element, this is not guaranteed to always be the case,
+	// and your code *WILL* break if you mess with `_p`.
+	_p:              grp.Group_Element,
+	_is_initialized: bool,
+}
+
+// ge_clear clears ge to the uninitialized state.
+ge_clear :: proc "contextless" (ge: ^Group_Element) {
+	mem.zero_explicit(ge, size_of(Group_Element))
+}
+
+// ge_set sets `ge = a`.
+ge_set :: proc(ge, a: ^Group_Element) {
+	_ge_assert_initialized([]^Group_Element{a})
+
+	grp.ge_set(&ge._p, &a._p)
+	ge._is_initialized = true
+}
+
+// ge_identity sets ge to the identity (neutral) element.
+ge_identity :: proc "contextless" (ge: ^Group_Element) {
+	grp.ge_identity(&ge._p)
+	ge._is_initialized = true
+}
+
+// ge_generator sets ge to the group generator.
+ge_generator :: proc "contextless" (ge: ^Group_Element) {
+	grp.ge_generator(&ge._p)
+	ge._is_initialized = true
+}
+
+// ge_set_bytes sets ge to the result of decoding b as a ristretto255
+// group element, and returns true on success.
+@(require_results)
+ge_set_bytes :: proc "contextless" (ge: ^Group_Element, b: []byte) -> bool {
+	// 1.  Interpret the string as an unsigned integer s in little-endian
+	//     representation.  If the length of the string is not 32 bytes or
+	//     if the resulting value is >= p, decoding fails.
+	//
+	// 2.  If IS_NEGATIVE(s) returns TRUE, decoding fails.
+
+	if len(b) != ELEMENT_SIZE {
+		return false
+	}
+	if b[31] & 128 != 0 || b[0] & 1 != 0 {
+		// Fail early if b is clearly > p, or negative.
+		return false
+	}
+
+	b_ := transmute(^[32]byte)(raw_data(b))
+
+	s: field.Tight_Field_Element = ---
+	defer field.fe_clear(&s)
+
+	field.fe_from_bytes(&s, b_)
+	if field.fe_equal_bytes(&s, b_) != 1 {
+		// Reject non-canonical encodings of s.
+		return false
+	}
+
+	// 3.  Process s as follows:
+	v, u1, u2: field.Loose_Field_Element = ---, ---, ---
+	tmp, u2_sqr: field.Tight_Field_Element = ---, ---
+
+	// ss = s^2
+	// u1 = 1 - ss
+	// u2 = 1 + ss
+	// u2_sqr = u2^2
+	field.fe_carry_square(&tmp, field.fe_relax_cast(&s))
+	field.fe_sub(&u1, &field.FE_ONE, &tmp)
+	field.fe_add(&u2, &field.FE_ONE, &tmp)
+	field.fe_carry_square(&u2_sqr, &u2)
+
+	// v = -(D * u1^2) - u2_sqr
+	field.fe_carry_square(&tmp, &u1)
+	field.fe_carry_mul(&tmp, field.fe_relax_cast(&grp.FE_D), field.fe_relax_cast(&tmp))
+	field.fe_carry_add(&tmp, &tmp, &u2_sqr)
+	field.fe_opp(&v, &tmp)
+
+	// (was_square, invsqrt) = SQRT_RATIO_M1(1, v * u2_sqr)
+	field.fe_carry_mul(&tmp, &v, field.fe_relax_cast(&u2_sqr))
+	was_square := field.fe_carry_sqrt_ratio_m1(
+		&tmp,
+		field.fe_relax_cast(&field.FE_ONE),
+		field.fe_relax_cast(&tmp),
+	)
+
+	// den_x = invsqrt * u2
+	// den_y = invsqrt * den_x * v
+	x, y, t: field.Tight_Field_Element = ---, ---, ---
+	field.fe_carry_mul(&x, field.fe_relax_cast(&tmp), &u2)
+	field.fe_carry_mul(&y, field.fe_relax_cast(&tmp), field.fe_relax_cast(&x))
+	field.fe_carry_mul(&y, field.fe_relax_cast(&y), &v)
+
+	// x = CT_ABS(2 * s * den_x)
+	field.fe_carry_mul(&x, field.fe_relax_cast(&s), field.fe_relax_cast(&x))
+	field.fe_carry_add(&x, &x, &x)
+	field.fe_carry_abs(&x, &x)
+
+	// y = u1 * den_y
+	field.fe_carry_mul(&y, &u1, field.fe_relax_cast(&y))
+
+	// t = x * y
+	field.fe_carry_mul(&t, field.fe_relax_cast(&x), field.fe_relax_cast(&y))
+
+	field.fe_clear_vec([]^field.Loose_Field_Element{&v, &u1, &u2})
+	field.fe_clear_vec([]^field.Tight_Field_Element{&tmp, &u2_sqr})
+	defer field.fe_clear_vec([]^field.Tight_Field_Element{&x, &y, &t})
+
+	// 4.  If was_square is FALSE, IS_NEGATIVE(t) returns TRUE, or y = 0,
+	// decoding fails.  Otherwise, return the group element represented
+	// by the internal representation (x, y, 1, t) as the result of
+	// decoding.
+
+	switch {
+	case was_square == 0:
+		// Not sure why the RFC doesn't have this just fail early.
+		return false
+	case field.fe_is_negative(&t) != 0:
+		return false
+	case field.fe_equal(&y, &field.FE_ZERO) != 0:
+		return false
+	}
+
+	field.fe_set(&ge._p.x, &x)
+	field.fe_set(&ge._p.y, &y)
+	field.fe_one(&ge._p.z)
+	field.fe_set(&ge._p.t, &t)
+	ge._is_initialized = true
+
+	return true
+}
+
+// ge_set_wide_bytes sets ge to the result of deriving a ristretto255
+// group element, from a wide (512-bit) byte string.
+ge_set_wide_bytes :: proc(ge: ^Group_Element, b: []byte) {
+	if len(b) != WIDE_ELEMENT_SIZE {
+		panic("crypto/ristretto255: invalid wide input size")
+	}
+
+	// The element derivation function on an input string b proceeds as
+	// follows:
+	//
+	// 1.  Compute P1 as MAP(b[0:32]).
+	// 2.  Compute P2 as MAP(b[32:64]).
+	// 3.  Return P1 + P2.
+
+	p1, p2: Group_Element = ---, ---
+	ge_map(&p1, b[0:32])
+	ge_map(&p2, b[32:64])
+
+	ge_add(ge, &p1, &p2)
+
+	ge_clear(&p1)
+	ge_clear(&p2)
+}
+
+// ge_bytes sets dst to the canonical encoding of ge.
+ge_bytes :: proc(ge: ^Group_Element, dst: []byte) {
+	_ge_assert_initialized([]^Group_Element{ge})
+	if len(dst) != ELEMENT_SIZE {
+		panic("crypto/ristretto255: invalid destination size")
+	}
+
+	x0, y0, z0, t0 := &ge._p.x, &ge._p.y, &ge._p.z, &ge._p.t
+
+	// 1.  Process the internal representation into a field element s as
+	// follows:
+
+	// u1 = (z0 + y0) * (z0 - y0)
+	// u2 = x0 * y0
+	u1, u2: field.Tight_Field_Element = ---, ---
+	tmp1, tmp2: field.Loose_Field_Element = ---, ---
+	field.fe_add(&tmp1, z0, y0)
+	field.fe_sub(&tmp2, z0, y0)
+	field.fe_carry_mul(&u1, &tmp1, &tmp2)
+	field.fe_carry_mul(&u2, field.fe_relax_cast(x0), field.fe_relax_cast(y0))
+
+	// Ignore was_square since this is always square.
+	// (_, invsqrt) = SQRT_RATIO_M1(1, u1 * u2^2)
+	tmp: field.Tight_Field_Element = ---
+	field.fe_carry_square(&tmp, field.fe_relax_cast(&u2))
+	field.fe_carry_mul(&tmp, field.fe_relax_cast(&u1), field.fe_relax_cast(&tmp))
+	_ = field.fe_carry_sqrt_ratio_m1(
+		&tmp,
+		field.fe_relax_cast(&field.FE_ONE),
+		field.fe_relax_cast(&tmp),
+	)
+
+	// den1 = invsqrt * u1
+	// den2 = invsqrt * u2
+	// z_inv = den1 * den2 * t0
+	den1, den2 := &u1, &u2
+	z_inv: field.Tight_Field_Element = ---
+	field.fe_carry_mul(den1, field.fe_relax_cast(&tmp), field.fe_relax_cast(&u1))
+	field.fe_carry_mul(den2, field.fe_relax_cast(&tmp), field.fe_relax_cast(&u2))
+	field.fe_carry_mul(&z_inv, field.fe_relax_cast(den1), field.fe_relax_cast(den2))
+	field.fe_carry_mul(&z_inv, field.fe_relax_cast(&z_inv), field.fe_relax_cast(t0))
+
+	// rotate = IS_NEGATIVE(t0 * z_inv)
+	// Note: Reordered from the RFC because invsqrt is no longer needed.
+	field.fe_carry_mul(&tmp, field.fe_relax_cast(t0), field.fe_relax_cast(&z_inv))
+	rotate := field.fe_is_negative(&tmp)
+
+	// ix0 = x0 * SQRT_M1
+	// iy0 = y0 * SQRT_M1
+	// enchanted_denominator = den1 * INVSQRT_A_MINUS_D
+	ix0, iy0: field.Tight_Field_Element = ---, ---
+	field.fe_carry_mul(&ix0, field.fe_relax_cast(x0), field.fe_relax_cast(&field.FE_SQRT_M1))
+	field.fe_carry_mul(&iy0, field.fe_relax_cast(y0), field.fe_relax_cast(&field.FE_SQRT_M1))
+	field.fe_carry_mul(&tmp, field.fe_relax_cast(den1), field.fe_relax_cast(&FE_INVSQRT_A_MINUS_D))
+
+	// Conditionally rotate x and y.
+	// x = CT_SELECT(iy0 IF rotate ELSE x0)
+	// y = CT_SELECT(ix0 IF rotate ELSE y0)
+	// z = z0
+	// den_inv = CT_SELECT(enchanted_denominator IF rotate ELSE den2)
+	x, y: field.Tight_Field_Element = ---, ---
+	field.fe_cond_select(&x, x0, &iy0, rotate)
+	field.fe_cond_select(&y, y0, &ix0, rotate)
+	field.fe_cond_select(&tmp, den2, &tmp, rotate)
+
+	// y = CT_SELECT(-y IF IS_NEGATIVE(x * z_inv) ELSE y)
+	field.fe_carry_mul(&x, field.fe_relax_cast(&x), field.fe_relax_cast(&z_inv))
+	field.fe_cond_negate(&y, &y, field.fe_is_negative(&x))
+
+	// s = CT_ABS(den_inv * (z - y))
+	field.fe_sub(&tmp1, z0, &y)
+	field.fe_carry_mul(&tmp, field.fe_relax_cast(&tmp), &tmp1)
+	field.fe_carry_abs(&tmp, &tmp)
+
+	// 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))
+	field.fe_to_bytes(dst_, &tmp)
+
+	field.fe_clear_vec([]^field.Tight_Field_Element{&u1, &u2, &tmp, &z_inv, &ix0, &iy0, &x, &y})
+	field.fe_clear_vec([]^field.Loose_Field_Element{&tmp1, &tmp2})
+}
+
+// ge_add sets `ge = a + b`.
+ge_add :: proc(ge, a, b: ^Group_Element) {
+	_ge_assert_initialized([]^Group_Element{a, b})
+
+	grp.ge_add(&ge._p, &a._p, &b._p)
+	ge._is_initialized = true
+}
+
+// ge_double sets `ge = a + a`.
+ge_double :: proc(ge, a: ^Group_Element) {
+	_ge_assert_initialized([]^Group_Element{a})
+
+	grp.ge_double(&ge._p, &a._p)
+	ge._is_initialized = true
+}
+
+// ge_negate sets `ge = -a`.
+ge_negate :: proc(ge, a: ^Group_Element) {
+	_ge_assert_initialized([]^Group_Element{a})
+
+	grp.ge_negate(&ge._p, &a._p)
+	ge._is_initialized = true
+}
+
+// ge_scalarmult sets `ge = A * sc`.
+ge_scalarmult :: proc(ge, A: ^Group_Element, sc: ^Scalar) {
+	_ge_assert_initialized([]^Group_Element{A})
+
+	grp.ge_scalarmult(&ge._p, &A._p, sc)
+	ge._is_initialized = true
+}
+
+// ge_scalarmult_generator sets `ge = G * sc`
+ge_scalarmult_generator :: proc "contextless" (ge: ^Group_Element, sc: ^Scalar) {
+	grp.ge_scalarmult_basepoint(&ge._p, sc)
+	ge._is_initialized = true
+}
+
+// ge_scalarmult_vartime sets `ge = A * sc` in variable time.
+ge_scalarmult_vartime :: proc(ge, A: ^Group_Element, sc: ^Scalar) {
+	_ge_assert_initialized([]^Group_Element{A})
+
+	grp.ge_scalarmult_vartime(&ge._p, &A._p, sc)
+	ge._is_initialized = true
+}
+
+// ge_double_scalarmult_generator_vartime sets `ge = A * a + G * b` in variable
+// time.
+ge_double_scalarmult_generator_vartime :: proc(
+	ge: ^Group_Element,
+	a: ^Scalar,
+	A: ^Group_Element,
+	b: ^Scalar,
+) {
+	_ge_assert_initialized([]^Group_Element{A})
+
+	grp.ge_double_scalarmult_basepoint_vartime(&ge._p, a, &A._p, b)
+	ge._is_initialized = true
+}
+
+// ge_cond_negate sets `ge = a` iff `ctrl == 0` and `ge = -a` iff `ctrl == 1`.
+// Behavior for all other values of ctrl are undefined,
+ge_cond_negate :: proc(ge, a: ^Group_Element, ctrl: int) {
+	_ge_assert_initialized([]^Group_Element{a})
+
+	grp.ge_cond_negate(&ge._p, &a._p, ctrl)
+	ge._is_initialized = true
+}
+
+// ge_cond_assign sets `ge = ge` iff `ctrl == 0` and `ge = a` iff `ctrl == 1`.
+// Behavior for all other values of ctrl are undefined,
+ge_cond_assign :: proc(ge, a: ^Group_Element, ctrl: int) {
+	_ge_assert_initialized([]^Group_Element{ge, a})
+
+	grp.ge_cond_assign(&ge._p, &a._p, ctrl)
+}
+
+// ge_cond_select sets `ge = a` iff `ctrl == 0` and `ge = b` iff `ctrl == 1`.
+// Behavior for all other values of ctrl are undefined,
+ge_cond_select :: proc(ge, a, b: ^Group_Element, ctrl: int) {
+	_ge_assert_initialized([]^Group_Element{a, b})
+
+	grp.ge_cond_select(&ge._p, &a._p, &b._p, ctrl)
+	ge._is_initialized = true
+}
+
+// ge_equal returns 1 iff `a == b`, and 0 otherwise.
+@(require_results)
+ge_equal :: proc(a, b: ^Group_Element) -> int {
+	_ge_assert_initialized([]^Group_Element{a, b})
+
+	// CT_EQ(x1 * y2, y1 * x2) | CT_EQ(y1 * y2, x1 * x2)
+	ax_by, ay_bx, ay_by, ax_bx: field.Tight_Field_Element = ---, ---, ---, ---
+	field.fe_carry_mul(&ax_by, field.fe_relax_cast(&a._p.x), field.fe_relax_cast(&b._p.y))
+	field.fe_carry_mul(&ay_bx, field.fe_relax_cast(&a._p.y), field.fe_relax_cast(&b._p.x))
+	field.fe_carry_mul(&ay_by, field.fe_relax_cast(&a._p.y), field.fe_relax_cast(&b._p.y))
+	field.fe_carry_mul(&ax_bx, field.fe_relax_cast(&a._p.x), field.fe_relax_cast(&b._p.x))
+
+	ret := field.fe_equal(&ax_by, &ay_bx) | field.fe_equal(&ay_by, &ax_bx)
+
+	field.fe_clear_vec([]^field.Tight_Field_Element{&ax_by, &ay_bx, &ay_by, &ax_bx})
+
+	return ret
+}
+
+// ge_is_identity returns 1 iff `ge` is the identity element, and 0 otherwise.
+@(require_results)
+ge_is_identity :: proc(ge: ^Group_Element) -> int {
+	return ge_equal(ge, &GE_IDENTITY)
+}
+
+@(private)
+ge_map :: proc "contextless" (ge: ^Group_Element, b: []byte) {
+	b_ := transmute(^[32]byte)(raw_data(b))
+
+	// The MAP function is defined on 32-byte strings as:
+	//
+	// 1.  Mask the most significant bit in the final byte of the string,
+	// and interpret the string as an unsigned integer r in little-
+	// endian representation.  Reduce r modulo p to obtain a field
+	// element t.
+	// *  Masking the most significant bit is equivalent to interpreting
+	// the whole string as an unsigned integer in little-endian
+	// representation and then reducing it modulo 2^255.
+	t: field.Tight_Field_Element = ---
+	field.fe_from_bytes(&t, b_)
+
+	// 2.  Process t as follows:
+	//
+	// r = SQRT_M1 * t^2
+	// u = (r + 1) * ONE_MINUS_D_SQ
+	// v = (-1 - r*D) * (r + D)
+	tmp1: field.Loose_Field_Element = ---
+	r, u, v: field.Tight_Field_Element = ---, ---, ---
+
+	field.fe_carry_square(&r, field.fe_relax_cast(&t))
+	field.fe_carry_mul(&r, field.fe_relax_cast(&field.FE_SQRT_M1), field.fe_relax_cast(&r))
+
+	field.fe_add(&tmp1, &field.FE_ONE, &r)
+	field.fe_carry_mul(&u, &tmp1, field.fe_relax_cast(&FE_ONE_MINUS_D_SQ))
+
+	field.fe_carry_mul(&v, field.fe_relax_cast(&r), field.fe_relax_cast(&grp.FE_D))
+	field.fe_carry_add(&v, &field.FE_ONE, &v)
+	field.fe_carry_opp(&v, &v)
+	field.fe_add(&tmp1, &r, &grp.FE_D)
+	field.fe_carry_mul(&v, field.fe_relax_cast(&v), &tmp1)
+
+	// (was_square, s) = SQRT_RATIO_M1(u, v)
+	// s_prime = -CT_ABS(s*t)
+	// s = CT_SELECT(s IF was_square ELSE s_prime)
+	// c = CT_SELECT(-1 IF was_square ELSE r)
+	s, s_prime, c: field.Tight_Field_Element = ---, ---, ---
+	was_square := field.fe_carry_sqrt_ratio_m1(
+		&s,
+		field.fe_relax_cast(&u),
+		field.fe_relax_cast(&v),
+	)
+	field.fe_carry_mul(&s_prime, field.fe_relax_cast(&s), field.fe_relax_cast(&t))
+	field.fe_carry_abs(&s_prime, &s_prime)
+	field.fe_carry_opp(&s_prime, &s_prime)
+	field.fe_cond_select(&s, &s_prime, &s, was_square)
+	field.fe_cond_select(&c, &r, &FE_NEG_ONE, was_square)
+
+	// N = c * (r - 1) * D_MINUS_ONE_SQ - v
+	N: field.Tight_Field_Element = ---
+	field.fe_sub(&tmp1, &r, &field.FE_ONE)
+	field.fe_carry_mul(&N, field.fe_relax_cast(&c), &tmp1)
+	field.fe_carry_mul(&N, field.fe_relax_cast(&N), field.fe_relax_cast(&FE_D_MINUS_ONE_SQUARED))
+	field.fe_carry_sub(&N, &N, &v)
+
+	// w0 = 2 * s * v
+	// w1 = N * SQRT_AD_MINUS_ONE
+	// w2 = 1 - s^2
+	// w3 = 1 + s^2
+	w0, w1: field.Tight_Field_Element = ---, ---
+	w2, w3: field.Loose_Field_Element = ---, ---
+	field.fe_carry_mul(&w0, field.fe_relax_cast(&s), field.fe_relax_cast(&v))
+	field.fe_carry_add(&w0, &w0, &w0)
+	field.fe_carry_mul(&w1, field.fe_relax_cast(&N), field.fe_relax_cast(&FE_SQRT_AD_MINUS_ONE))
+	field.fe_carry_square(&s, field.fe_relax_cast(&s))
+	field.fe_sub(&w2, &field.FE_ONE, &s)
+	field.fe_add(&w3, &field.FE_ONE, &s)
+
+	// 3.  Return the group element represented by the internal
+	// representation (w0*w3, w2*w1, w1*w3, w0*w2).
+
+	field.fe_carry_mul(&ge._p.x, field.fe_relax_cast(&w0), &w3)
+	field.fe_carry_mul(&ge._p.y, &w2, field.fe_relax_cast(&w1))
+	field.fe_carry_mul(&ge._p.z, field.fe_relax_cast(&w1), &w3)
+	field.fe_carry_mul(&ge._p.t, field.fe_relax_cast(&w0), &w2)
+	ge._is_initialized = true
+
+	field.fe_clear_vec([]^field.Tight_Field_Element{&r, &u, &v, &s, &s_prime, &c, &N, &w0, &w1})
+	field.fe_clear_vec([]^field.Loose_Field_Element{&tmp1, &w2, &w3})
+}
+
+@(private)
+_ge_assert_initialized :: proc(ges: []^Group_Element) {
+	for ge in ges {
+		if !ge._is_initialized {
+			panic("crypto/ristretto255: uninitialized group element")
+		}
+	}
+}

core/crypto/ristretto255/ristretto255_scalar.odin

@@ -0,0 +1,97 @@
+package ristretto255
+
+import grp "core:crypto/_edwards25519"
+
+// SCALAR_SIZE is the size of a byte-encoded ristretto255 scalar.
+SCALAR_SIZE :: 32
+// WIDE_SCALAR_SIZE is the size of a wide byte-encoded ristretto255
+// scalar.
+WIDE_SCALAR_SIZE :: 64
+
+// Scalar is a ristretto255 scalar.  The zero-initialized value is valid,
+// and represents `0`.
+Scalar :: grp.Scalar
+
+// sc_clear clears sc to the uninitialized state.
+sc_clear :: proc "contextless" (sc: ^Scalar) {
+	grp.sc_clear(sc)
+}
+
+// sc_set sets `sc = a`.
+sc_set :: proc "contextless" (sc, a: ^Scalar) {
+	grp.sc_set(sc, a)
+}
+
+// sc_set_u64 sets `sc = i`.
+sc_set_u64 :: proc "contextless" (sc: ^Scalar, i: u64) {
+	grp.sc_set_u64(sc, i)
+}
+
+// sc_set_bytes sets sc to the result of decoding b as a ristretto255
+// scalar, and returns true on success.
+@(require_results)
+sc_set_bytes :: proc(sc: ^Scalar, b: []byte) -> bool {
+	if len(b) != SCALAR_SIZE {
+		return false
+	}
+
+	return grp.sc_set_bytes(sc, b)
+}
+
+// sc_set_wide_bytes sets sc to the result of deriving a ristretto255
+// scalar, from a wide (512-bit) byte string by interpreting b as a
+// little-endian value, and reducing it mod the group order.
+sc_set_bytes_wide :: proc(sc: ^Scalar, b: []byte) {
+	if len(b) != WIDE_SCALAR_SIZE {
+		panic("crypto/ristretto255: invalid wide input size")
+	}
+
+	b_ := transmute(^[WIDE_SCALAR_SIZE]byte)(raw_data(b))
+	grp.sc_set_bytes_wide(sc, b_)
+}
+
+// sc_bytes sets dst to the canonical encoding of sc.
+sc_bytes :: proc(sc: ^Scalar, dst: []byte) {
+	if len(dst) != SCALAR_SIZE {
+		panic("crypto/ristretto255: invalid destination size")
+	}
+
+	grp.sc_bytes(dst, sc)
+}
+
+// sc_add sets `sc = a + b`.
+sc_add :: proc "contextless" (sc, a, b: ^Scalar) {
+	grp.sc_add(sc, a, b)
+}
+
+// sc_sub sets `sc = a - b`.
+sc_sub :: proc "contextless" (sc, a, b: ^Scalar) {
+	grp.sc_sub(sc, a, b)
+}
+
+// sc_negate sets `sc = -a`.
+sc_negate :: proc "contextless" (sc, a: ^Scalar) {
+	grp.sc_negate(sc, a)
+}
+
+// sc_mul sets `sc = a * b`.
+sc_mul :: proc "contextless" (sc, a, b: ^Scalar) {
+	grp.sc_mul(sc, a, b)
+}
+
+// sc_square sets `sc = a^2`.
+sc_square :: proc "contextless" (sc, a: ^Scalar) {
+	grp.sc_square(sc, a)
+}
+
+// sc_cond_assign sets `sc = sc` iff `ctrl == 0` and `sc = a` iff `ctrl == 1`.
+// Behavior for all other values of ctrl are undefined,
+sc_cond_assign :: proc(sc, a: ^Scalar, ctrl: int) {
+	grp.sc_cond_assign(sc, a, ctrl)
+}
+
+// sc_equal returns 1 iff `a == b`, and 0 otherwise.
+@(require_results)
+sc_equal :: proc(a, b: ^Scalar) -> int {
+	return grp.sc_equal(a, b)
+}

core/crypto/sha3/sha3.odin

@@ -67,6 +67,7 @@ init_512 :: proc(ctx: ^Context) {
 
 @(private)
 _init :: proc(ctx: ^Context) {
+	ctx.dsbyte = _sha3.DS_SHA3
 	_sha3.init(transmute(^_sha3.Context)(ctx))
 }
 

core/crypto/shake/shake.odin

@@ -1,10 +1,11 @@
 /*
-package shake implements the SHAKE XOF algorithm family.
+package shake implements the SHAKE and cSHAKE XOF algorithm families.
 
 The SHA3 hash algorithm can be found in the crypto/sha3.
 
 See:
 - https://nvlpubs.nist.gov/nistpubs/fips/nist.fips.202.pdf
+- https://nvlpubs.nist.gov/nistpubs/specialpublications/nist.sp.800-185.pdf
 */
 package shake
 
@@ -18,24 +19,27 @@ package shake
 
 import "../_sha3"
 
-// Context is a SHAKE128 or SHAKE256 instance.
+// Context is a SHAKE128, SHAKE256, cSHAKE128, or cSHAKE256 instance.
 Context :: distinct _sha3.Context
 
 // init_128 initializes a Context for SHAKE128.
 init_128 :: proc(ctx: ^Context) {
-	ctx.mdlen = 128 / 8
-	_init(ctx)
+	_sha3.init_cshake(transmute(^_sha3.Context)(ctx), nil, nil, 128)
 }
 
 // init_256 initializes a Context for SHAKE256.
 init_256 :: proc(ctx: ^Context) {
-	ctx.mdlen = 256 / 8
-	_init(ctx)
+	_sha3.init_cshake(transmute(^_sha3.Context)(ctx), nil, nil, 256)
 }
 
-@(private)
-_init :: proc(ctx: ^Context) {
-	_sha3.init(transmute(^_sha3.Context)(ctx))
+// 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)
+}
+
+// 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)
 }
 
 // write writes more data into the SHAKE instance.  This MUST not be called

core/crypto/tuplehash/tuplehash.odin

@@ -0,0 +1,66 @@
+/*
+package tuplehash implements the TupleHash and TupleHashXOF algorithms.
+
+See:
+- https://nvlpubs.nist.gov/nistpubs/specialpublications/nist.sp.800-185.pdf
+*/
+package tuplehash
+
+import "../_sha3"
+
+// Context is a TupleHash or TupleHashXOF instance.
+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)
+}
+
+// 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)
+}
+
+// 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)
+}
+
+// final finalizes the Context, writes the digest to hash, and calls
+// reset on the Context.
+//
+// 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)
+}
+
+// read reads output from the TupleHashXOF instance.  There is no practical
+// upper limit to the amount of data that can be read from TupleHashXOF.
+// 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)
+	if !ctx.is_finalized {
+		_sha3.encode_byte_len(ctx_, 0, false) // right_encode
+		_sha3.shake_xof(ctx_)
+	}
+
+	_sha3.shake_out(ctx_, dst)
+}
+
+// clone clones the Context other into ctx.
+clone :: proc(ctx, other: ^Context) {
+	_sha3.clone(transmute(^_sha3.Context)(ctx), transmute(^_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))
+}
+
+@(private)
+N_TUPLEHASH := []byte{'T', 'u', 'p', 'l', 'e', 'H', 'a', 's', 'h'}

core/crypto/x25519/x25519.odin

@@ -1,9 +1,18 @@
+/*
+package x25519 implements the X25519 (aka curve25519) Elliptic-Curve
+Diffie-Hellman key exchange protocol.
+
+See:
+- https://www.rfc-editor.org/rfc/rfc7748
+*/
 package x25519
 
 import field "core:crypto/_fiat/field_curve25519"
 import "core:mem"
 
+// SCALAR_SIZE is the size of a X25519 scalar (private key) in bytes.
 SCALAR_SIZE :: 32
+// POINT_SIZE is the size of a X25519 point (public key/shared secret) in bytes.
 POINT_SIZE :: 32
 
 @(private)
@@ -14,11 +23,11 @@ _scalar_bit :: #force_inline proc "contextless" (s: ^[32]byte, i: int) -> u8 {
 	if i < 0 {
 		return 0
 	}
-	return (s[i>>3] >> uint(i&7)) & 1
+	return (s[i >> 3] >> uint(i & 7)) & 1
 }
 
 @(private)
-_scalarmult :: proc (out, scalar, point: ^[32]byte) {
+_scalarmult :: proc "contextless" (out, scalar, point: ^[32]byte) {
 	// Montgomery pseduo-multiplication taken from Monocypher.
 
 	// computes the scalar product
@@ -26,7 +35,7 @@ _scalarmult :: proc (out, scalar, point: ^[32]byte) {
 	field.fe_from_bytes(&x1, point)
 
 	// computes the actual scalar product (the result is in x2 and z2)
-	x2, x3, z2, z3: field.Tight_Field_Element =  ---, ---, ---, ---
+	x2, x3, z2, z3: field.Tight_Field_Element = ---, ---, ---, ---
 	t0, t1: field.Loose_Field_Element = ---, ---
 
 	// Montgomery ladder
@@ -38,7 +47,7 @@ _scalarmult :: proc (out, scalar, point: ^[32]byte) {
 	field.fe_one(&z3)
 
 	swap: int
-	for pos := 255-1; pos >= 0; pos = pos - 1 	{
+	for pos := 255 - 1; pos >= 0; pos = pos - 1 {
 		// constant time conditional swap before ladder step
 		b := int(_scalar_bit(scalar, pos))
 		swap ~= b // xor trick avoids swapping at the end of the loop
@@ -85,16 +94,13 @@ _scalarmult :: proc (out, scalar, point: ^[32]byte) {
 	field.fe_carry_mul(&x2, field.fe_relax_cast(&x2), field.fe_relax_cast(&z2))
 	field.fe_to_bytes(out, &x2)
 
-	mem.zero_explicit(&x1, size_of(x1))
-	mem.zero_explicit(&x2, size_of(x2))
-	mem.zero_explicit(&x3, size_of(x3))
-	mem.zero_explicit(&z2, size_of(z2))
-	mem.zero_explicit(&z3, size_of(z3))
-	mem.zero_explicit(&t0, size_of(t0))
-	mem.zero_explicit(&t1, size_of(t1))
+	field.fe_clear_vec([]^field.Tight_Field_Element{&x1, &x2, &x3, &z2, &z3})
+	field.fe_clear_vec([]^field.Loose_Field_Element{&t0, &t1})
 }
 
-scalarmult :: proc (dst, scalar, point: []byte) {
+// scalarmult "multiplies" the provided scalar and point, and writes the
+// resulting point to dst.
+scalarmult :: proc(dst, scalar, point: []byte) {
 	if len(scalar) != SCALAR_SIZE {
 		panic("crypto/x25519: invalid scalar size")
 	}
@@ -123,7 +129,8 @@ scalarmult :: proc (dst, scalar, point: []byte) {
 	mem.zero_explicit(&d, size_of(d))
 }
 
-scalarmult_basepoint :: proc (dst, scalar: []byte) {
-	// TODO/perf: Switch to using a precomputed table.
+// scalarmult_basepoint "multiplies" the provided scalar with the X25519
+// base point and writes the resulting point to dst.
+scalarmult_basepoint :: proc(dst, scalar: []byte) {
 	scalarmult(dst, scalar, _BASE_POINT[:])
 }

core/encoding/hxa/read.odin

@@ -177,7 +177,7 @@ read :: proc(data: []byte, filename := "<input>", print_error := false, allocato
 	}
 	defer file.nodes = file.nodes[:node_count]
 
-	for node_idx in 0..<header.internal_node_count {
+	for _ in 0..<header.internal_node_count {
 		node := &file.nodes[node_count]
 		type := read_value(r, Node_Type) or_return
 		if type > max(Node_Type) {

core/encoding/json/marshal.odin

@@ -51,6 +51,11 @@ Marshal_Options :: struct {
 	// NOTE: This will temp allocate and sort a list for each map.
 	sort_maps_by_key: bool,
 
+	// Output enum value's name instead of its underlying value.
+	//
+	// NOTE: If a name isn't found it'll use the underlying value.
+	use_enum_names: bool,
+
 	// Internal state
 	indentation: int,
 	mjson_skipped_first_braces_start: bool,
@@ -241,7 +246,6 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 		opt_write_end(w, opt, ']') or_return
 		
 	case runtime.Type_Info_Enumerated_Array:
-		index := runtime.type_info_base(info.index).variant.(runtime.Type_Info_Enum)
 		opt_write_start(w, opt, '[') or_return
 		for i in 0..<info.count {
 			opt_write_iteration(w, opt, i) or_return
@@ -294,14 +298,14 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 
 					// check for string type
 					{
-						v := any{key, info.key.id}
-						ti := runtime.type_info_base(type_info_of(v.id))
-						a := any{v.data, ti.id}
+						kv  := any{key, info.key.id}
+						kti := runtime.type_info_base(type_info_of(kv.id))
+						ka  := any{kv.data, kti.id}
 						name: string
 
-						#partial switch info in ti.variant {
+						#partial switch info in kti.variant {
 						case runtime.Type_Info_String:
-							switch s in a {
+							switch s in ka {
 							case string: name = s
 							case cstring: name = string(s)
 							}
@@ -331,13 +335,13 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 
 					// check for string type
 					{
-						v := any{key, info.key.id}
-						ti := runtime.type_info_base(type_info_of(v.id))
-						a := any{v.data, ti.id}
+						kv  := any{key, info.key.id}
+						kti := runtime.type_info_base(type_info_of(kv.id))
+						ka  := any{kv.data, kti.id}
 
-						#partial switch info in ti.variant {
+						#partial switch info in kti.variant {
 						case runtime.Type_Info_String:
-							switch s in a {
+							switch s in ka {
 							case string: name = s
 							case cstring: name = string(s)
 							}
@@ -362,24 +366,93 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 		opt_write_end(w, opt, '}') or_return
 
 	case runtime.Type_Info_Struct:
-		opt_write_start(w, opt, '{') or_return
-		
-		for name, i in info.names {
-			opt_write_iteration(w, opt, i) or_return
-			if json_name := string(reflect.struct_tag_get(auto_cast info.tags[i], "json")); json_name != "" {
-				opt_write_key(w, opt, json_name) or_return
-			} else {
-				opt_write_key(w, opt, name) or_return
+		is_omitempty :: proc(v: any) -> bool {
+			v := v
+			if v == nil {
+				return true
 			}
-
-			id := info.types[i].id
-			data := rawptr(uintptr(v.data) + info.offsets[i])
-			marshal_to_writer(w, any{data, id}, opt) or_return
+			ti := runtime.type_info_core(type_info_of(v.id))
+			#partial switch info in ti.variant {
+			case runtime.Type_Info_String:
+				switch x in v {
+				case string:
+					return x == ""
+				case cstring:
+					return x == nil || x == ""
+				}
+			case runtime.Type_Info_Any:
+				return v.(any) == nil
+			case runtime.Type_Info_Type_Id:
+				return v.(typeid) == nil
+			case runtime.Type_Info_Pointer,
+			     runtime.Type_Info_Multi_Pointer,
+			     runtime.Type_Info_Procedure:
+			     	return (^rawptr)(v.data)^ == nil
+			case runtime.Type_Info_Dynamic_Array:
+			     	return (^runtime.Raw_Dynamic_Array)(v.data).len == 0
+			case runtime.Type_Info_Slice:
+			     	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 false
 		}
 
+		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 {
+				omitempty := false
+
+				json_name, extra := json_name_from_tag_value(reflect.struct_tag_get(reflect.Struct_Tag(info.tags[i]), "json"))
+				for flag in strings.split_iterator(&extra, ",") {
+					switch flag {
+					case "omitempty":
+						omitempty = true
+					}
+				}
+
+				id := info.types[i].id
+				data := rawptr(uintptr(v.data) + info.offsets[i])
+				the_value := any{data, id}
+
+				if is_omitempty(the_value) {
+					continue
+				}
+
+				opt_write_iteration(w, opt, i) or_return
+				if json_name != "" {
+					opt_write_key(w, opt, json_name) or_return
+				} else {
+					// Marshal the fields of 'using _: T' fields directly into the parent struct
+					if info.usings[i] && name == "_" {
+						marshal_struct_fields(w, the_value, opt) or_return
+						continue
+					} else {
+						opt_write_key(w, opt, name) or_return
+					}
+				}
+
+
+				marshal_to_writer(w, the_value, opt) or_return
+			}
+			return
+		}
+		
+		opt_write_start(w, opt, '{') or_return
+		marshal_struct_fields(w, v, opt) or_return
 		opt_write_end(w, opt, '}') or_return
 
 	case runtime.Type_Info_Union:
+		if len(info.variants) == 0 || v.data == nil {
+			io.write_string(w, "null") or_return
+			return nil
+		}
+
 		tag_ptr := uintptr(v.data) + info.tag_offset
 		tag_any := any{rawptr(tag_ptr), info.tag_type.id}
 
@@ -404,7 +477,16 @@ marshal_to_writer :: proc(w: io.Writer, v: any, opt: ^Marshal_Options) -> (err:
 		}
 
 	case runtime.Type_Info_Enum:
-		return marshal_to_writer(w, any{v.data, info.base.id}, opt)
+		if !opt.use_enum_names || len(info.names) == 0 {
+			return marshal_to_writer(w, any{v.data, info.base.id}, opt)
+		} else {
+			name, found := reflect.enum_name_from_value_any(v)
+			if found {
+				return marshal_to_writer(w, name, opt)
+			} else {
+				return marshal_to_writer(w, any{v.data, info.base.id}, opt)
+			}
+		}
 
 	case runtime.Type_Info_Bit_Set:
 		is_bit_set_different_endian_to_platform :: proc(ti: ^runtime.Type_Info) -> bool {

core/encoding/json/parser.odin

@@ -204,8 +204,8 @@ parse_array :: proc(p: ^Parser) -> (value: Value, err: Error) {
 }
 
 @(private)
-bytes_make :: proc(size, alignment: int, allocator: mem.Allocator) -> (bytes: []byte, err: Error) {
-	b, berr := mem.alloc_bytes(size, alignment, allocator)
+bytes_make :: proc(size, alignment: int, allocator: mem.Allocator, loc := #caller_location) -> (bytes: []byte, err: Error) {
+	b, berr := mem.alloc_bytes(size, alignment, allocator, loc)
 	if berr != nil {
 		if berr == .Out_Of_Memory {
 			err = .Out_Of_Memory
@@ -217,9 +217,9 @@ bytes_make :: proc(size, alignment: int, allocator: mem.Allocator) -> (bytes: []
 	return
 }
 
-clone_string :: proc(s: string, allocator: mem.Allocator) -> (str: string, err: Error) {
+clone_string :: proc(s: string, allocator: mem.Allocator, loc := #caller_location) -> (str: string, err: Error) {
 	n := len(s)
-	b := bytes_make(n+1, 1, allocator) or_return
+	b := bytes_make(n+1, 1, allocator, loc) or_return
 	copy(b, s)
 	if len(b) > n {
 		b[n] = 0
@@ -290,7 +290,7 @@ parse_object :: proc(p: ^Parser) -> (value: Value, err: Error) {
 
 
 // IMPORTANT NOTE(bill): unquote_string assumes a mostly valid string
-unquote_string :: proc(token: Token, spec: Specification, allocator := context.allocator) -> (value: string, err: Error) {
+unquote_string :: proc(token: Token, spec: Specification, allocator := context.allocator, loc := #caller_location) -> (value: string, err: Error) {
 	get_u2_rune :: proc(s: string) -> rune {
 		if len(s) < 4 || s[0] != '\\' || s[1] != 'x' {
 			return -1
@@ -359,7 +359,7 @@ unquote_string :: proc(token: Token, spec: Specification, allocator := context.a
 		i += w
 	}
 	if i == len(s) {
-		return clone_string(s, allocator)
+		return clone_string(s, allocator, loc)
 	}
 
 	b := bytes_make(len(s) + 2*utf8.UTF_MAX, 1, allocator) or_return

core/encoding/json/unmarshal.odin

@@ -6,6 +6,7 @@ import "core:reflect"
 import "core:strconv"
 import "core:strings"
 import "base:runtime"
+import "base:intrinsics"
 
 Unmarshal_Data_Error :: enum {
 	Invalid_Data,
@@ -342,6 +343,16 @@ unmarshal_expect_token :: proc(p: ^Parser, kind: Token_Kind, loc := #caller_loca
 	return prev
 }
 
+@(private)
+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:]
+	}
+	return
+}
+
 
 @(private)
 unmarshal_object :: proc(p: ^Parser, v: any, end_token: Token_Kind) -> (err: Unmarshal_Error) {
@@ -368,16 +379,23 @@ unmarshal_object :: proc(p: ^Parser, v: any, end_token: Token_Kind) -> (err: Unm
 			unmarshal_expect_token(p, .Colon)						
 			
 			fields := reflect.struct_fields_zipped(ti.id)
-			
-			runtime.DEFAULT_TEMP_ALLOCATOR_TEMP_GUARD(ignore = context.temp_allocator == context.allocator)
 
-			field_used := make([]bool, len(fields), context.temp_allocator)
-			
+			field_test :: #force_inline proc "contextless" (field_used: [^]byte, offset: uintptr) -> bool {
+				prev_set := field_used[offset/8] & byte(offset&7) != 0
+				field_used[offset/8] |= byte(offset&7)
+				return prev_set
+			}
+
+			field_used_bytes := (reflect.size_of_typeid(ti.id)+7)/8
+			field_used := intrinsics.alloca(field_used_bytes, 1)
+			intrinsics.mem_zero(field_used, field_used_bytes)
+
 			use_field_idx := -1
 			
 			for field, field_idx in fields {
 				tag_value := string(reflect.struct_tag_get(field.tag, "json"))
-				if key == tag_value {
+				json_name, _ := json_name_from_tag_value(tag_value)
+				if key == json_name {
 					use_field_idx = field_idx
 					break
 				}
@@ -392,14 +410,45 @@ unmarshal_object :: proc(p: ^Parser, v: any, end_token: Token_Kind) -> (err: Unm
 				}
 			}
 			
-			if use_field_idx >= 0 {
-				if field_used[use_field_idx] {
+			check_children_using_fields :: proc(key: string, parent: typeid) -> (
+				offset: uintptr,
+				type: ^reflect.Type_Info,
+				found: bool,
+			) {
+				for field in reflect.struct_fields_zipped(parent) {
+					if field.is_using && field.name == "_" {
+						offset, type, found = check_children_using_fields(key, field.type.id)
+						if found {
+							offset += field.offset
+							return
+						}
+					}
+
+					if field.name == key {
+						offset = field.offset
+						type = field.type
+						found = true
+						return
+					}
+				}
+				return
+			}
+
+			offset: uintptr
+			type: ^reflect.Type_Info
+			field_found: bool = use_field_idx >= 0
+
+			if field_found {
+				offset = fields[use_field_idx].offset
+				type = fields[use_field_idx].type
+			} else {
+				offset, type, field_found = check_children_using_fields(key, ti.id)
+			}
+
+			if field_found {
+				if field_test(field_used, offset) {
 					return .Multiple_Use_Field
 				}
-				field_used[use_field_idx] = true
-				offset := fields[use_field_idx].offset
-				type := fields[use_field_idx].type
-				name := fields[use_field_idx].name
 				
 				field_ptr := rawptr(uintptr(v.data) + offset)
 				field := any{field_ptr, type.id}
@@ -411,6 +460,12 @@ unmarshal_object :: proc(p: ^Parser, v: any, end_token: Token_Kind) -> (err: Unm
 				continue struct_loop
 			} else {
 				// allows skipping unused struct fields
+
+				// NOTE(bill): prevent possible memory leak if a string is unquoted
+				allocator := p.allocator
+				defer p.allocator = allocator
+				p.allocator = mem.nil_allocator()
+
 				parse_value(p) or_return
 				if parse_comma(p) {
 					break struct_loop
@@ -439,9 +494,9 @@ unmarshal_object :: proc(p: ^Parser, v: any, end_token: Token_Kind) -> (err: Unm
 			
 			
 			mem.zero_slice(elem_backing)
-			if err := unmarshal_value(p, map_backing_value); err != nil {
+			if uerr := unmarshal_value(p, map_backing_value); uerr != nil {
 				delete(key, p.allocator)
-				return err
+				return uerr
 			}
 
 			key_ptr := rawptr(&key)

core/fmt/fmt.odin

@@ -972,7 +972,7 @@ fmt_bad_verb :: proc(fi: ^Info, verb: rune) {
 //
 fmt_bool :: proc(fi: ^Info, b: bool, verb: rune) {
 	switch verb {
-	case 't', 'v':
+	case 't', 'v', 'w':
 		fmt_string(fi, b ? "true" : "false", 's')
 	case:
 		fmt_bad_verb(fi, verb)
@@ -1209,7 +1209,7 @@ fmt_rune :: proc(fi: ^Info, r: rune, verb: rune) {
 	switch verb {
 	case 'c', 'r', 'v':
 		io.write_rune(fi.writer, r, &fi.n)
-	case 'q':
+	case 'q', 'w':
 		fi.n += io.write_quoted_rune(fi.writer, r)
 	case:
 		fmt_int(fi, u64(r), false, 32, verb)
@@ -1226,7 +1226,8 @@ fmt_rune :: proc(fi: ^Info, r: rune, verb: rune) {
 //
 fmt_int :: proc(fi: ^Info, u: u64, is_signed: bool, bit_size: int, verb: rune) {
 	switch verb {
-	case 'v': _fmt_int(fi, u, 10, is_signed, bit_size, __DIGITS_LOWER)
+	case 'v', 'w':
+		_fmt_int(fi, u, 10, is_signed, bit_size, __DIGITS_LOWER)
 	case 'b': _fmt_int(fi, u,  2, is_signed, bit_size, __DIGITS_LOWER)
 	case 'o': _fmt_int(fi, u,  8, is_signed, bit_size, __DIGITS_LOWER)
 	case 'i', 'd': _fmt_int(fi, u, 10, is_signed, bit_size, __DIGITS_LOWER)
@@ -1261,7 +1262,8 @@ fmt_int :: proc(fi: ^Info, u: u64, is_signed: bool, bit_size: int, verb: rune) {
 //
 fmt_int_128 :: proc(fi: ^Info, u: u128, is_signed: bool, bit_size: int, verb: rune) {
 	switch verb {
-	case 'v': _fmt_int_128(fi, u, 10, is_signed, bit_size, __DIGITS_LOWER)
+	case 'v', 'w':
+		_fmt_int_128(fi, u, 10, is_signed, bit_size, __DIGITS_LOWER)
 	case 'b': _fmt_int_128(fi, u,  2, is_signed, bit_size, __DIGITS_LOWER)
 	case 'o': _fmt_int_128(fi, u,  8, is_signed, bit_size, __DIGITS_LOWER)
 	case 'i', 'd': _fmt_int_128(fi, u, 10, is_signed, bit_size, __DIGITS_LOWER)
@@ -1351,7 +1353,7 @@ _fmt_float_as :: proc(fi: ^Info, v: f64, bit_size: int, verb: rune, float_fmt: b
 //
 fmt_float :: proc(fi: ^Info, v: f64, bit_size: int, verb: rune) {
 	switch verb {
-	case 'g', 'G', 'v':
+	case 'g', 'G', 'v', 'w':
 		_fmt_float_as(fi, v, bit_size, verb, 'g', -1)
 	case 'f', 'F':
 		_fmt_float_as(fi, v, bit_size, verb, 'f', 3)
@@ -1424,7 +1426,7 @@ fmt_string :: proc(fi: ^Info, s: string, verb: rune) {
 			io.write_string(fi.writer, s, &fi.n)
 		}
 
-	case 'q': // quoted string
+	case 'q', 'w': // quoted string
 		io.write_quoted_string(fi.writer, s, '"', &fi.n)
 
 	case 'x', 'X':
@@ -1467,7 +1469,7 @@ fmt_cstring :: proc(fi: ^Info, s: cstring, verb: rune) {
 fmt_pointer :: proc(fi: ^Info, p: rawptr, verb: rune) {
 	u := u64(uintptr(p))
 	switch verb {
-	case 'p', 'v':
+	case 'p', 'v', 'w':
 		if !fi.hash && verb == 'v' {
 			io.write_string(fi.writer, "0x", &fi.n)
 		}
@@ -1535,7 +1537,7 @@ string_to_enum_value :: proc($T: typeid, s: string) -> (T, bool) {
 // Inputs:
 // - fi: Pointer to the Info struct containing format settings.
 // - v: The enum value to format.
-// - verb: The format specifier character (e.g. 'i','d','f','s','v','q').
+// - verb: The format specifier character (e.g. 'i','d','f','s','v','q','w').
 //
 fmt_enum :: proc(fi: ^Info, v: any, verb: rune) {
 	if v.id == nil || v.data == nil {
@@ -1559,6 +1561,15 @@ fmt_enum :: proc(fi: ^Info, v: any, verb: rune) {
 				fmt_arg(fi, any{v.data, runtime.type_info_base(e.base).id}, 'i')
 				io.write_string(fi.writer, ")", &fi.n)
 			}
+		case 'w':
+			if str, ok := enum_value_to_string(v); ok {
+				io.write_byte(fi.writer, '.', &fi.n)
+				io.write_string(fi.writer, str, &fi.n)
+			} else {
+				io.write_string(fi.writer, "%!(BAD ENUM VALUE=", &fi.n)
+				fmt_arg(fi, any{v.data, runtime.type_info_base(e.base).id}, 'i')
+				io.write_string(fi.writer, ")", &fi.n)
+			}
 		}
 	}
 }
@@ -1747,8 +1758,8 @@ fmt_write_indent :: proc(fi: ^Info) {
 // - verb: The formatting verb to be used for the array elements.
 //
 fmt_write_array :: proc(fi: ^Info, array_data: rawptr, count: int, elem_size: int, elem_id: typeid, verb: rune) {
-	io.write_byte(fi.writer, '[', &fi.n)
-	defer io.write_byte(fi.writer, ']', &fi.n)
+	io.write_byte(fi.writer, '[' if verb != 'w' else '{', &fi.n)
+	defer io.write_byte(fi.writer, ']' if verb != 'w' else '}', &fi.n)
 
 	if count <= 0 {
 		return
@@ -1821,6 +1832,13 @@ handle_tag :: proc(data: rawptr, info: reflect.Type_Info_Struct, idx: int, verb:
 		r, w := utf8.decode_rune_in_string(value)
 		value = value[w:]
 		if value == "" || value[0] == ',' {
+			if verb^ == 'w' {
+				// TODO(bill): is this a good idea overriding that field tags if 'w' is used?
+				switch r {
+				case 's': r = 'q'
+				case:     r = 'w'
+				}
+			}
 			verb^ = r
 			if len(value) > 0 && value[0] == ',' {
 				field_name := value[1:]
@@ -1832,7 +1850,7 @@ handle_tag :: proc(data: rawptr, info: reflect.Type_Info_Struct, idx: int, verb:
 					switch r {
 					case 's', 'q':
 						handle_optional_len(data, info, field_name, optional_len)
-					case 'v':
+					case 'v', 'w':
 						#partial switch reflect.type_kind(info.types[idx].id) {
 						case .String, .Multi_Pointer, .Array, .Slice, .Dynamic_Array:
 							handle_optional_len(data, info, field_name, optional_len)
@@ -1854,7 +1872,7 @@ handle_tag :: proc(data: rawptr, info: reflect.Type_Info_Struct, idx: int, verb:
 // - type_name: The name of the type being formatted
 //
 fmt_struct :: proc(fi: ^Info, v: any, the_verb: rune, info: runtime.Type_Info_Struct, type_name: string) {
-	if the_verb != 'v' {
+	if the_verb != 'v' && the_verb != 'w' {
 		fmt_bad_verb(fi, the_verb)
 		return
 	}
@@ -1871,7 +1889,7 @@ fmt_struct :: proc(fi: ^Info, v: any, the_verb: rune, info: runtime.Type_Info_St
 	is_soa := info.soa_kind != .None
 
 	io.write_string(fi.writer, type_name, &fi.n)
-	io.write_byte(fi.writer, '[' if is_soa else '{', &fi.n)
+	io.write_byte(fi.writer, '[' if is_soa && the_verb == 'v' else '{', &fi.n)
 	fi.record_level += 1
 	defer fi.record_level -= 1
 
@@ -1889,7 +1907,7 @@ fmt_struct :: proc(fi: ^Info, v: any, the_verb: rune, info: runtime.Type_Info_St
 		if hash {
 			for _ in 0..<indent { io.write_byte(fi.writer, '\t', &fi.n) }
 		}
-		io.write_byte(fi.writer, ']' if is_soa else '}', &fi.n)
+		io.write_byte(fi.writer, ']' if is_soa && the_verb == 'v' else '}', &fi.n)
 	}
 
 	if is_soa {
@@ -1972,7 +1990,7 @@ fmt_struct :: proc(fi: ^Info, v: any, the_verb: rune, info: runtime.Type_Info_St
 		for name, i in info.names {
 			optional_len: int = -1
 			use_nul_termination: bool = false
-			verb := 'v'
+			verb := the_verb if the_verb == 'w' else 'v'
 			if handle_tag(v.data, info, i, &verb, &optional_len, &use_nul_termination) {
 				continue
 			}
@@ -2051,7 +2069,7 @@ fmt_array_nul_terminated :: proc(fi: ^Info, data: rawptr, max_n: int, elem_size:
 // - n: The number of elements in the array.
 // - elem_size: The size of each element in the array.
 // - elem: Pointer to the type information of the array element.
-// - verb: The formatting verb (e.g. 's','q','p').
+// - verb: The formatting verb (e.g. 's','q','p','w').
 //
 fmt_array :: proc(fi: ^Info, data: rawptr, n: int, elem_size: int, elem: ^reflect.Type_Info, verb: rune) {
 	if data == nil && n > 0 {
@@ -2130,141 +2148,143 @@ fmt_named :: proc(fi: ^Info, v: any, verb: rune, info: runtime.Type_Info_Named)
 	}
 
 	// Built-in Custom Formatters for core library types
-	switch a in v {
-	case runtime.Source_Code_Location:
-		io.write_string(fi.writer, a.file_path, &fi.n)
+	if verb != 'w' {
+		switch a in v {
+		case runtime.Source_Code_Location:
+			io.write_string(fi.writer, a.file_path, &fi.n)
 
-		when ODIN_ERROR_POS_STYLE == .Default {
-			io.write_byte(fi.writer, '(', &fi.n)
-			io.write_int(fi.writer, int(a.line), 10, &fi.n)
-			io.write_byte(fi.writer, ':', &fi.n)
-			io.write_int(fi.writer, int(a.column), 10, &fi.n)
-			io.write_byte(fi.writer, ')', &fi.n)
-		} else when ODIN_ERROR_POS_STYLE == .Unix {
-			io.write_byte(fi.writer, ':', &fi.n)
-			io.write_int(fi.writer, int(a.line), 10, &fi.n)
-			io.write_byte(fi.writer, ':', &fi.n)
-			io.write_int(fi.writer, int(a.column), 10, &fi.n)
-			io.write_byte(fi.writer, ':', &fi.n)
-		} else {
-			#panic("Unhandled ODIN_ERROR_POS_STYLE")
-		}
-		return
-
-	case time.Duration:
-		ffrac :: proc(buf: []byte, v: u64, prec: int) -> (nw: int, nv: u64) {
-			v := v
-			w := len(buf)
-			print := false
-			for _ in 0..<prec {
-				digit := v % 10
-				print = print || digit != 0
-				if print {
-					w -= 1
-					buf[w] = byte(digit) + '0'
-				}
-				v /= 10
-			}
-			if print {
-				w -= 1
-				buf[w] = '.'
-			}
-			return w, v
-		}
-		fint :: proc(buf: []byte, v: u64) -> int {
-			v := v
-			w := len(buf)
-			if v == 0 {
-				w -= 1
-				buf[w] = '0'
+			when ODIN_ERROR_POS_STYLE == .Default {
+				io.write_byte(fi.writer, '(', &fi.n)
+				io.write_int(fi.writer, int(a.line), 10, &fi.n)
+				io.write_byte(fi.writer, ':', &fi.n)
+				io.write_int(fi.writer, int(a.column), 10, &fi.n)
+				io.write_byte(fi.writer, ')', &fi.n)
+			} else when ODIN_ERROR_POS_STYLE == .Unix {
+				io.write_byte(fi.writer, ':', &fi.n)
+				io.write_int(fi.writer, int(a.line), 10, &fi.n)
+				io.write_byte(fi.writer, ':', &fi.n)
+				io.write_int(fi.writer, int(a.column), 10, &fi.n)
+				io.write_byte(fi.writer, ':', &fi.n)
 			} else {
-				for v > 0 {
-					w -= 1
-					buf[w] = byte(v%10) + '0'
+				#panic("Unhandled ODIN_ERROR_POS_STYLE")
+			}
+			return
+
+		case time.Duration:
+			ffrac :: proc(buf: []byte, v: u64, prec: int) -> (nw: int, nv: u64) {
+				v := v
+				w := len(buf)
+				print := false
+				for _ in 0..<prec {
+					digit := v % 10
+					print = print || digit != 0
+					if print {
+						w -= 1
+						buf[w] = byte(digit) + '0'
+					}
 					v /= 10
 				}
+				if print {
+					w -= 1
+					buf[w] = '.'
+				}
+				return w, v
 			}
-			return w
-		}
-
-		buf: [32]byte
-		w := len(buf)
-		u := u64(a)
-		neg := a < 0
-		if neg {
-			u = -u
-		}
-
-		if u < u64(time.Second) {
-			prec: int
-			w -= 1
-			buf[w] = 's'
-			w -= 1
-			switch {
-			case u == 0:
-				io.write_string(fi.writer, "0s", &fi.n)
-				return
-			case u < u64(time.Microsecond):
-				prec = 0
-				buf[w] = 'n'
-			case u < u64(time.Millisecond):
-				prec = 3
-				// U+00B5 'µ' micro sign == 0xC2 0xB5
-				w -= 1 // Need room for two bytes
-				copy(buf[w:], "µ")
-			case:
-				prec = 6
-				buf[w] = 'm'
+			fint :: proc(buf: []byte, v: u64) -> int {
+				v := v
+				w := len(buf)
+				if v == 0 {
+					w -= 1
+					buf[w] = '0'
+				} else {
+					for v > 0 {
+						w -= 1
+						buf[w] = byte(v%10) + '0'
+						v /= 10
+					}
+				}
+				return w
 			}
-			w, u = ffrac(buf[:w], u, prec)
-			w = fint(buf[:w], u)
-		} else {
-			w -= 1
-			buf[w] = 's'
-			w, u = ffrac(buf[:w], u, 9)
-			w = fint(buf[:w], u%60)
-			u /= 60
-			if u > 0 {
+
+			buf: [32]byte
+			w := len(buf)
+			u := u64(a)
+			neg := a < 0
+			if neg {
+				u = -u
+			}
+
+			if u < u64(time.Second) {
+				prec: int
 				w -= 1
-				buf[w] = 'm'
+				buf[w] = 's'
+				w -= 1
+				switch {
+				case u == 0:
+					io.write_string(fi.writer, "0s", &fi.n)
+					return
+				case u < u64(time.Microsecond):
+					prec = 0
+					buf[w] = 'n'
+				case u < u64(time.Millisecond):
+					prec = 3
+					// U+00B5 'µ' micro sign == 0xC2 0xB5
+					w -= 1 // Need room for two bytes
+					copy(buf[w:], "µ")
+				case:
+					prec = 6
+					buf[w] = 'm'
+				}
+				w, u = ffrac(buf[:w], u, prec)
+				w = fint(buf[:w], u)
+			} else {
+				w -= 1
+				buf[w] = 's'
+				w, u = ffrac(buf[:w], u, 9)
 				w = fint(buf[:w], u%60)
 				u /= 60
 				if u > 0 {
 					w -= 1
-					buf[w] = 'h'
-					w = fint(buf[:w], u)
+					buf[w] = 'm'
+					w = fint(buf[:w], u%60)
+					u /= 60
+					if u > 0 {
+						w -= 1
+						buf[w] = 'h'
+						w = fint(buf[:w], u)
+					}
 				}
 			}
+
+			if neg {
+				w -= 1
+				buf[w] = '-'
+			}
+			io.write_string(fi.writer, string(buf[w:]), &fi.n)
+			return
+
+		case time.Time:
+			t := a
+			y, mon, d := time.date(t)
+			h, min, s := time.clock(t)
+			ns := (t._nsec - (t._nsec/1e9 + time.UNIX_TO_ABSOLUTE)*1e9) % 1e9
+			write_padded_number(fi, i64(y), 4)
+			io.write_byte(fi.writer, '-', &fi.n)
+			write_padded_number(fi, i64(mon), 2)
+			io.write_byte(fi.writer, '-', &fi.n)
+			write_padded_number(fi, i64(d), 2)
+			io.write_byte(fi.writer, ' ', &fi.n)
+
+			write_padded_number(fi, i64(h), 2)
+			io.write_byte(fi.writer, ':', &fi.n)
+			write_padded_number(fi, i64(min), 2)
+			io.write_byte(fi.writer, ':', &fi.n)
+			write_padded_number(fi, i64(s), 2)
+			io.write_byte(fi.writer, '.', &fi.n)
+			write_padded_number(fi, (ns), 9)
+			io.write_string(fi.writer, " +0000 UTC", &fi.n)
+			return
 		}
-
-		if neg {
-			w -= 1
-			buf[w] = '-'
-		}
-		io.write_string(fi.writer, string(buf[w:]), &fi.n)
-		return
-
-	case time.Time:
-		t := a
-		y, mon, d := time.date(t)
-		h, min, s := time.clock(t)
-		ns := (t._nsec - (t._nsec/1e9 + time.UNIX_TO_ABSOLUTE)*1e9) % 1e9
-		write_padded_number(fi, i64(y), 4)
-		io.write_byte(fi.writer, '-', &fi.n)
-		write_padded_number(fi, i64(mon), 2)
-		io.write_byte(fi.writer, '-', &fi.n)
-		write_padded_number(fi, i64(d), 2)
-		io.write_byte(fi.writer, ' ', &fi.n)
-
-		write_padded_number(fi, i64(h), 2)
-		io.write_byte(fi.writer, ':', &fi.n)
-		write_padded_number(fi, i64(min), 2)
-		io.write_byte(fi.writer, ':', &fi.n)
-		write_padded_number(fi, i64(s), 2)
-		io.write_byte(fi.writer, '.', &fi.n)
-		write_padded_number(fi, (ns), 9)
-		io.write_string(fi.writer, " +0000 UTC", &fi.n)
-		return
 	}
 
 	#partial switch b in info.base.variant {
@@ -2275,6 +2295,23 @@ fmt_named :: proc(fi: ^Info, v: any, verb: rune, info: runtime.Type_Info_Named)
 	case runtime.Type_Info_Bit_Set:
 		fmt_bit_set(fi, v, verb = verb)
 	case:
+		if verb == 'w' {
+			#partial switch _ in info.base.variant {
+			case runtime.Type_Info_Array,
+			     runtime.Type_Info_Enumerated_Array,
+			     runtime.Type_Info_Dynamic_Array,
+			     runtime.Type_Info_Slice,
+			     runtime.Type_Info_Struct,
+			     runtime.Type_Info_Union,
+			     runtime.Type_Info_Enum,
+			     runtime.Type_Info_Map,
+			     runtime.Type_Info_Bit_Set,
+			     runtime.Type_Info_Simd_Vector,
+			     runtime.Type_Info_Matrix,
+			     runtime.Type_Info_Bit_Field:
+				io.write_string(fi.writer, info.name, &fi.n)
+			}
+		}
 		fmt_value(fi, any{v.data, info.base.id}, verb)
 	}
 }
@@ -2341,8 +2378,13 @@ fmt_union :: proc(fi: ^Info, v: any, verb: rune, info: runtime.Type_Info_Union,
 // - info: A runtime.Type_Info_Matrix struct containing matrix type information.
 //
 fmt_matrix :: proc(fi: ^Info, v: any, verb: rune, info: runtime.Type_Info_Matrix) {
-	io.write_string(fi.writer, "matrix[", &fi.n)
-	defer io.write_byte(fi.writer, ']', &fi.n)
+	if verb == 'w' {
+		io.write_byte(fi.writer, '{', &fi.n)
+	} else {
+		io.write_string(fi.writer, "matrix", &fi.n)
+		io.write_byte(fi.writer, '[', &fi.n)
+	}
+	defer io.write_byte(fi.writer, ']' if verb != 'w' else '}', &fi.n)
 
 	fi.indent += 1
 
@@ -2354,7 +2396,11 @@ fmt_matrix :: proc(fi: ^Info, v: any, verb: rune, info: runtime.Type_Info_Matrix
 			for col in 0..<info.column_count {
 				if col > 0 { io.write_string(fi.writer, ", ", &fi.n) }
 
-				offset := (row + col*info.elem_stride)*info.elem_size
+				offset: int
+				switch info.layout {
+				case .Column_Major: offset = (row + col*info.elem_stride)*info.elem_size
+				case .Row_Major:    offset = (col + row*info.elem_stride)*info.elem_size
+				}
 
 				data := uintptr(v.data) + uintptr(offset)
 				fmt_arg(fi, any{rawptr(data), info.elem.id}, verb)
@@ -2368,7 +2414,11 @@ fmt_matrix :: proc(fi: ^Info, v: any, verb: rune, info: runtime.Type_Info_Matrix
 			for col in 0..<info.column_count {
 				if col > 0 { io.write_string(fi.writer, ", ", &fi.n) }
 
-				offset := (row + col*info.elem_stride)*info.elem_size
+				offset: int
+				switch info.layout {
+				case .Column_Major: offset = (row + col*info.elem_stride)*info.elem_size
+				case .Row_Major:    offset = (col + row*info.elem_stride)*info.elem_size
+				}
 
 				data := uintptr(v.data) + uintptr(offset)
 				fmt_arg(fi, any{rawptr(data), info.elem.id}, verb)
@@ -2413,8 +2463,8 @@ fmt_bit_field :: proc(fi: ^Info, v: any, verb: rune, info: runtime.Type_Info_Bit
 		return false
 	}
 
-	io.write_string(fi.writer, type_name if len(type_name) != 0 else "bit_field", &fi.n)
-	io.write_string(fi.writer, "{", &fi.n)
+	io.write_string(fi.writer, type_name if len(type_name) != 0 || verb == 'w' else "bit_field", &fi.n)
+	io.write_byte(fi.writer, '{', &fi.n)
 
 	hash   := fi.hash;   defer fi.hash = hash
 	indent := fi.indent; defer fi.indent -= 1
@@ -2631,11 +2681,12 @@ fmt_value :: proc(fi: ^Info, v: any, verb: rune) {
 		defer fi.record_level -= 1
 
 		if fi.hash {
-			io.write_string(fi.writer, "[\n", &fi.n)
+			io.write_byte(fi.writer, '[' if verb != 'w' else '{', &fi.n)
+			io.write_byte(fi.writer, '\n', &fi.n)
 			defer {
 				io.write_byte(fi.writer, '\n', &fi.n)
 				fmt_write_indent(fi)
-				io.write_byte(fi.writer, ']', &fi.n)
+				io.write_byte(fi.writer, ']' if verb != 'w' else '}', &fi.n)
 			}
 			indent := fi.indent
 			fi.indent += 1
@@ -2659,8 +2710,8 @@ fmt_value :: proc(fi: ^Info, v: any, verb: rune) {
 				io.write_string(fi.writer, ",\n", &fi.n)
 			}
 		} else {
-			io.write_byte(fi.writer, '[', &fi.n)
-			defer io.write_byte(fi.writer, ']', &fi.n)
+			io.write_byte(fi.writer, '[' if verb != 'w' else '{', &fi.n)
+			defer io.write_byte(fi.writer, ']' if verb != 'w' else '}', &fi.n)
 			for i in 0..<info.count {
 				if i > 0 { io.write_string(fi.writer, ", ", &fi.n) }
 
@@ -2725,38 +2776,61 @@ fmt_value :: proc(fi: ^Info, v: any, verb: rune) {
 
 
 	case runtime.Type_Info_Map:
-		if verb != 'v' {
+		switch verb {
+		case:
 			fmt_bad_verb(fi, verb)
-			return
-		}
-
-		io.write_string(fi.writer, "map[", &fi.n)
-		defer io.write_byte(fi.writer, ']', &fi.n)
-		fi.record_level += 1
-		defer fi.record_level -= 1
-
-		m := (^mem.Raw_Map)(v.data)
-		if m != nil {
-			if info.map_info == nil {
-				return
+		case 'v', 'w':
+			if verb == 'v' {
+				io.write_string(fi.writer, "map", &fi.n)
 			}
-			map_cap := uintptr(runtime.map_cap(m^))
-			ks, vs, hs, _, _ := runtime.map_kvh_data_dynamic(m^, info.map_info)
-			j := 0
-			for bucket_index in 0..<map_cap {
-				runtime.map_hash_is_valid(hs[bucket_index]) or_continue
+			io.write_byte(fi.writer, '[' if verb != 'w' else '{', &fi.n)
+			defer io.write_byte(fi.writer, ']' if verb != 'w' else '}', &fi.n)
 
-				if j > 0 {
-					io.write_string(fi.writer, ", ", &fi.n)
+
+			hash   := fi.hash;   defer fi.hash = hash
+			indent := fi.indent; defer fi.indent -= 1
+			do_trailing_comma := hash
+
+			fi.indent += 1
+			if hash	{
+				io.write_byte(fi.writer, '\n', &fi.n)
+			}
+			defer {
+				if hash {
+					for _ in 0..<indent { io.write_byte(fi.writer, '\t', &fi.n) }
 				}
-				j += 1
+			}
 
-				key   := runtime.map_cell_index_dynamic(ks, info.map_info.ks, bucket_index)
-				value := runtime.map_cell_index_dynamic(vs, info.map_info.vs, bucket_index)
+			m := (^mem.Raw_Map)(v.data)
+			if m != nil {
+				if info.map_info == nil {
+					return
+				}
+				map_cap := uintptr(runtime.map_cap(m^))
+				ks, vs, hs, _, _ := runtime.map_kvh_data_dynamic(m^, info.map_info)
+				j := 0
+				for bucket_index in 0..<map_cap {
+					runtime.map_hash_is_valid(hs[bucket_index]) or_continue
 
-				fmt_arg(&Info{writer = fi.writer}, any{rawptr(key), info.key.id}, 'v')
-				io.write_string(fi.writer, "=", &fi.n)
-				fmt_arg(fi, any{rawptr(value), info.value.id}, 'v')
+					if !do_trailing_comma && j > 0 { io.write_string(fi.writer, ", ") }
+					if hash {
+						fmt_write_indent(fi)
+					}
+					j += 1
+
+					key   := runtime.map_cell_index_dynamic(ks, info.map_info.ks, bucket_index)
+					value := runtime.map_cell_index_dynamic(vs, info.map_info.vs, bucket_index)
+
+					fmt_arg(&Info{writer = fi.writer}, any{rawptr(key), info.key.id}, verb)
+					if hash {
+						io.write_string(fi.writer, " = ", &fi.n)
+					} else {
+						io.write_string(fi.writer, "=", &fi.n)
+					}
+					fmt_arg(fi, any{rawptr(value), info.value.id}, verb)
+
+					if do_trailing_comma { io.write_string(fi.writer, ",\n", &fi.n) }
+				}
 			}
 		}
 
@@ -2811,11 +2885,11 @@ fmt_value :: proc(fi: ^Info, v: any, verb: rune) {
 // - fi: A pointer to an Info struct containing formatting information.
 // - c: The complex128 value to be formatted.
 // - bits: The number of bits in the complex number (32 or 64).
-// - verb: The formatting verb rune ('f', 'F', 'v', 'h', 'H').
+// - verb: The formatting verb rune ('f', 'F', 'v', 'h', 'H', 'w').
 //
 fmt_complex :: proc(fi: ^Info, c: complex128, bits: int, verb: rune) {
 	switch verb {
-	case 'f', 'F', 'v', 'h', 'H':
+	case 'f', 'F', 'v', 'h', 'H', 'w':
 		r, i := real(c), imag(c)
 		fmt_float(fi, r, bits/2, verb)
 		if !fi.plus && i >= 0 {
@@ -2835,11 +2909,11 @@ fmt_complex :: proc(fi: ^Info, c: complex128, bits: int, verb: rune) {
 // - fi: A pointer to an Info struct containing formatting information.
 // - q: The quaternion256 value to be formatted.
 // - bits: The number of bits in the quaternion number (64, 128, or 256).
-// - verb: The formatting verb rune ('f', 'F', 'v', 'h', 'H').
+// - verb: The formatting verb rune ('f', 'F', 'v', 'h', 'H', 'w').
 //
 fmt_quaternion  :: proc(fi: ^Info, q: quaternion256, bits: int, verb: rune) {
 	switch verb {
-	case 'f', 'F', 'v', 'h', 'H':
+	case 'f', 'F', 'v', 'h', 'H', 'w':
 		r, i, j, k := real(q), imag(q), jmag(q), kmag(q)
 
 		fmt_float(fi, r, bits/4, verb)

core/hash/crc.odin

@@ -1,7 +1,7 @@
 package hash
 
 @(optimization_mode="speed")
-crc64_ecma_182 :: proc(data: []byte, seed := u64(0)) -> (result: u64) #no_bounds_check {
+crc64_ecma_182 :: proc "contextless" (data: []byte, seed := u64(0)) -> (result: u64) #no_bounds_check {
 	result = seed
 	#no_bounds_check for b in data {
 		result = result<<8 ~ _crc64_table_ecma_182[((result>>56) ~ u64(b)) & 0xff]
@@ -15,7 +15,7 @@ crc64_ecma_182 :: proc(data: []byte, seed := u64(0)) -> (result: u64) #no_bounds
 	Based on Mark Adler's v1.4 implementation in C under the ZLIB license.
 */
 @(optimization_mode="speed")
-crc64_xz :: proc(data: []byte, seed := u64(0)) -> u64 #no_bounds_check {
+crc64_xz :: proc "contextless" (data: []byte, seed := u64(0)) -> u64 #no_bounds_check {
 	data := data
 	result := ~u64le(seed)
 
@@ -53,7 +53,7 @@ crc64_xz :: proc(data: []byte, seed := u64(0)) -> u64 #no_bounds_check {
 	Generator polynomial: x^64 + x^4 + x^3 + x + 1
 */
 @(optimization_mode="speed")
-crc64_iso_3306 :: proc(data: []byte, seed := u64(0)) -> u64 #no_bounds_check {
+crc64_iso_3306 :: proc "contextless" (data: []byte, seed := u64(0)) -> u64 #no_bounds_check {
 
 	result := seed
 
@@ -70,7 +70,7 @@ crc64_iso_3306 :: proc(data: []byte, seed := u64(0)) -> u64 #no_bounds_check {
 	return result
 }
 
-crc64_iso_3306_inverse :: proc(data: []byte, seed := u64(0)) -> u64 {
+crc64_iso_3306_inverse :: proc "contextless" (data: []byte, seed := u64(0)) -> u64 {
 	result := #force_inline crc64_iso_3306(data, ~seed)
 	return ~result
 }

core/hash/crc32.odin

@@ -3,7 +3,7 @@ package hash
 import "base:intrinsics"
 
 @(optimization_mode="speed")
-crc32 :: proc(data: []byte, seed := u32(0)) -> u32 #no_bounds_check {
+crc32 :: proc "contextless" (data: []byte, seed := u32(0)) -> u32 #no_bounds_check {
 	crc := ~seed
 	buffer := raw_data(data)
 	length := len(data)
@@ -323,7 +323,7 @@ crc32_table := [8][256]u32{
 
 /*
 @(optimization_mode="speed")
-crc32 :: proc(data: []byte, seed := u32(0)) -> u32 {
+crc32 :: proc "contextless" (data: []byte, seed := u32(0)) -> u32 {
 	result := ~u32(seed);
 	 #no_bounds_check for b in data {
 		result = result>>8 ~ _crc32_table[(result ~ u32(b)) & 0xff];

core/hash/hash.odin

@@ -4,7 +4,7 @@ import "core:mem"
 import "base:intrinsics"
 
 @(optimization_mode="speed")
-adler32 :: proc(data: []byte, seed := u32(1)) -> u32 #no_bounds_check {
+adler32 :: proc "contextless" (data: []byte, seed := u32(1)) -> u32 #no_bounds_check {
 
 	ADLER_CONST :: 65521
 
@@ -47,7 +47,7 @@ adler32 :: proc(data: []byte, seed := u32(1)) -> u32 #no_bounds_check {
 }
 
 @(optimization_mode="speed")
-djb2 :: proc(data: []byte, seed := u32(5381)) -> u32 {
+djb2 :: proc "contextless" (data: []byte, seed := u32(5381)) -> u32 {
 	hash: u32 = seed
 	for b in data {
 		hash = (hash << 5) + hash + u32(b) // hash * 33 + u32(b)
@@ -55,7 +55,7 @@ djb2 :: proc(data: []byte, seed := u32(5381)) -> u32 {
 	return hash
 }
 
-djbx33a :: proc(data: []byte, seed := u32(5381)) -> (result: [16]byte) #no_bounds_check {
+djbx33a :: proc "contextless" (data: []byte, seed := u32(5381)) -> (result: [16]byte) #no_bounds_check {
 	state := [4]u32{seed, seed, seed, seed}
 	
 	s: u32 = 0
@@ -74,7 +74,7 @@ djbx33a :: proc(data: []byte, seed := u32(5381)) -> (result: [16]byte) #no_bound
 
 // If you have a choice, prefer fnv32a
 @(optimization_mode="speed")
-fnv32_no_a :: proc(data: []byte, seed := u32(0x811c9dc5)) -> u32 {
+fnv32_no_a :: proc "contextless" (data: []byte, seed := u32(0x811c9dc5)) -> u32 {
 	h: u32 = seed
 	for b in data {
 		h = (h * 0x01000193) ~ u32(b)
@@ -87,7 +87,7 @@ fnv64 :: fnv64_no_a // NOTE(bill): Not a fan of these aliases but seems necessar
 
 // If you have a choice, prefer fnv64a
 @(optimization_mode="speed")
-fnv64_no_a :: proc(data: []byte, seed := u64(0xcbf29ce484222325)) -> u64 {
+fnv64_no_a :: proc "contextless" (data: []byte, seed := u64(0xcbf29ce484222325)) -> u64 {
 	h: u64 = seed
 	for b in data {
 		h = (h * 0x100000001b3) ~ u64(b)
@@ -95,7 +95,7 @@ fnv64_no_a :: proc(data: []byte, seed := u64(0xcbf29ce484222325)) -> u64 {
 	return h
 }
 @(optimization_mode="speed")
-fnv32a :: proc(data: []byte, seed := u32(0x811c9dc5)) -> u32 {
+fnv32a :: proc "contextless" (data: []byte, seed := u32(0x811c9dc5)) -> u32 {
 	h: u32 = seed
 	for b in data {
 		h = (h ~ u32(b)) * 0x01000193
@@ -104,7 +104,7 @@ fnv32a :: proc(data: []byte, seed := u32(0x811c9dc5)) -> u32 {
 }
 
 @(optimization_mode="speed")
-fnv64a :: proc(data: []byte, seed := u64(0xcbf29ce484222325)) -> u64 {
+fnv64a :: proc "contextless" (data: []byte, seed := u64(0xcbf29ce484222325)) -> u64 {
 	h: u64 = seed
 	for b in data {
 		h = (h ~ u64(b)) * 0x100000001b3
@@ -113,7 +113,7 @@ fnv64a :: proc(data: []byte, seed := u64(0xcbf29ce484222325)) -> u64 {
 }
 
 @(optimization_mode="speed")
-jenkins :: proc(data: []byte, seed := u32(0)) -> u32 {
+jenkins :: proc "contextless" (data: []byte, seed := u32(0)) -> u32 {
 	hash: u32 = seed
 	for b in data {
 		hash += u32(b)
@@ -127,7 +127,7 @@ jenkins :: proc(data: []byte, seed := u32(0)) -> u32 {
 }
 
 @(optimization_mode="speed")
-murmur32 :: proc(data: []byte, seed := u32(0)) -> u32 {
+murmur32 :: proc "contextless" (data: []byte, seed := u32(0)) -> u32 {
 	c1_32: u32 : 0xcc9e2d51
 	c2_32: u32 : 0x1b873593
 
@@ -178,7 +178,7 @@ murmur32 :: proc(data: []byte, seed := u32(0)) -> u32 {
 
 // See https://github.com/aappleby/smhasher/blob/master/src/MurmurHash2.cpp#L96
 @(optimization_mode="speed")
-murmur64a :: proc(data: []byte, seed := u64(0x9747b28c)) -> u64 {
+murmur64a :: proc "contextless" (data: []byte, seed := u64(0x9747b28c)) -> u64 {
 	m :: 0xc6a4a7935bd1e995
 	r :: 47
 
@@ -219,7 +219,7 @@ murmur64a :: proc(data: []byte, seed := u64(0x9747b28c)) -> u64 {
 
 // See https://github.com/aappleby/smhasher/blob/master/src/MurmurHash2.cpp#L140
 @(optimization_mode="speed")
-murmur64b :: proc(data: []byte, seed := u64(0x9747b28c)) -> u64 {
+murmur64b :: proc "contextless" (data: []byte, seed := u64(0x9747b28c)) -> u64 {
 	m :: 0x5bd1e995
 	r :: 24
 
@@ -287,7 +287,7 @@ murmur64b :: proc(data: []byte, seed := u64(0x9747b28c)) -> u64 {
 }
 
 @(optimization_mode="speed")
-sdbm :: proc(data: []byte, seed := u32(0)) -> u32 {
+sdbm :: proc "contextless" (data: []byte, seed := u32(0)) -> u32 {
 	hash: u32 = seed
 	for b in data {
 		hash = u32(b) + (hash<<6) + (hash<<16) - hash

core/hash/mini.odin

@@ -1,6 +1,6 @@
 package hash
 
-ginger_hash8 :: proc(x: u8) -> u8 {
+ginger_hash8 :: proc "contextless" (x: u8) -> u8 {
 	h := x * 251
 	h += ~(x << 3)
 	h ~=  (x >> 1)
@@ -11,7 +11,7 @@ ginger_hash8 :: proc(x: u8) -> u8 {
 }
 
 
-ginger_hash16 :: proc(x: u16) -> u16 {
+ginger_hash16 :: proc "contextless" (x: u16) -> u16 {
 	z := (x << 8) | (x >> 8)
 	h := z
 	h += ~(z << 5)
@@ -24,14 +24,14 @@ ginger_hash16 :: proc(x: u16) -> u16 {
 }
 
 
-ginger8 :: proc(data: []byte) -> u8 {
+ginger8 :: proc "contextless" (data: []byte) -> u8 {
 	h := ginger_hash8(0)
 	for b in data {
 		h ~= ginger_hash8(b)
 	}
 	return h
 }
-ginger16 :: proc(data: []byte) -> u16 {
+ginger16 :: proc "contextless" (data: []byte) -> u16 {
 	h := ginger_hash16(0)
 	for b in data {
 		h ~= ginger_hash16(u16(b))

core/hash/xxhash/streaming.odin

@@ -129,7 +129,7 @@ XXH3_create_state :: proc(allocator := context.allocator) -> (res: ^XXH3_state,
 }
 
 XXH3_destroy_state :: proc(state: ^XXH3_state, allocator := context.allocator) -> (err: Error) {
-	free(state)
+	free(state, allocator)
 	return .None
 }
 

core/hash/xxhash/xxhash_32.odin

@@ -19,15 +19,15 @@ xxh_u32    :: u32
 XXH32_DEFAULT_SEED :: XXH32_hash(0)
 
 XXH32_state :: struct {
-   total_len_32: XXH32_hash,    /*!< Total length hashed, modulo 2^32 */
-   large_len:    XXH32_hash,    /*!< Whether the hash is >= 16 (handles @ref total_len_32 overflow) */
-   v1:           XXH32_hash,    /*!< First accumulator lane */
-   v2:           XXH32_hash,    /*!< Second accumulator lane */
-   v3:           XXH32_hash,    /*!< Third accumulator lane */
-   v4:           XXH32_hash,    /*!< Fourth accumulator lane */
-   mem32:        [4]XXH32_hash, /*!< Internal buffer for partial reads. Treated as unsigned char[16]. */
-   memsize:      XXH32_hash,    /*!< Amount of data in @ref mem32 */
-   reserved:     XXH32_hash,    /*!< Reserved field. Do not read or write to it, it may be removed. */
+	total_len_32: XXH32_hash,    /*!< Total length hashed, modulo 2^32 */
+	large_len:    XXH32_hash,    /*!< Whether the hash is >= 16 (handles @ref total_len_32 overflow) */
+	v1:           XXH32_hash,    /*!< First accumulator lane */
+	v2:           XXH32_hash,    /*!< Second accumulator lane */
+	v3:           XXH32_hash,    /*!< Third accumulator lane */
+	v4:           XXH32_hash,    /*!< Fourth accumulator lane */
+	mem32:        [4]XXH32_hash, /*!< Internal buffer for partial reads. Treated as unsigned char[16]. */
+	memsize:      XXH32_hash,    /*!< Amount of data in @ref mem32 */
+	reserved:     XXH32_hash,    /*!< Reserved field. Do not read or write to it, it may be removed. */
 }
 
 XXH32_canonical :: struct {

core/image/common.odin

@@ -651,7 +651,7 @@ alpha_add_if_missing :: proc(img: ^Image, alpha_key := Alpha_Key{}, allocator :=
 				// We have keyed alpha.
 				o: GA_Pixel
 				for p in inp {
-					if p == key.r {
+					if p.r == key.r {
 						o = GA_Pixel{0, key.g}
 					} else {
 						o = GA_Pixel{p.r, 255}
@@ -710,7 +710,7 @@ alpha_add_if_missing :: proc(img: ^Image, alpha_key := Alpha_Key{}, allocator :=
 				// We have keyed alpha.
 				o: GA_Pixel_16
 				for p in inp {
-					if p == key.r {
+					if p.r == key.r {
 						o = GA_Pixel_16{0, key.g}
 					} else {
 						o = GA_Pixel_16{p.r, 65535}
@@ -842,11 +842,11 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al
 			bg  := G_Pixel{}
 			if temp_bg, temp_bg_ok := img.background.(RGB_Pixel_16); temp_bg_ok {
 				// Background is RGB 16-bit, take just the red channel's topmost byte.
-				bg = u8(temp_bg.r >> 8)
+				bg.r = u8(temp_bg.r >> 8)
 			}
 
 			for p in inp {
-				out[0] = bg if p == key else p
+				out[0] = bg if p.r == key else p
 				out    = out[1:]
 			}
 
@@ -865,8 +865,8 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al
 				for p in inp {
 					a := f32(p.g) / 255.0
 					c := ((1.0 - a) * bg + a * f32(p.r))
-					out[0] = u8(c)
-					out    = out[1:]
+					out[0].r = u8(c)
+					out      = out[1:]
 				}
 
 			} else if .alpha_premultiply in options {
@@ -874,14 +874,14 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al
 				for p in inp {
 					a := f32(p.g) / 255.0
 					c := f32(p.r) * a
-					out[0] = u8(c)
-					out    = out[1:]
+					out[0].r = u8(c)
+					out      = out[1:]
 				}
 			} else {
 				// Just drop alpha on the floor.
 				for p in inp {
-					out[0] = p.r
-					out    = out[1:]
+					out[0].r = p.r
+					out      = out[1:]
 				}
 			}
 
@@ -951,11 +951,11 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al
 			bg  := G_Pixel_16{}
 			if temp_bg, temp_bg_ok := img.background.(RGB_Pixel_16); temp_bg_ok {
 				// Background is RGB 16-bit, take just the red channel.
-				bg = temp_bg.r
+				bg.r = temp_bg.r
 			}
 
 			for p in inp {
-				out[0] = bg if p == key else p
+				out[0] = bg if p.r == key else p
 				out    = out[1:]
 			}
 
@@ -974,8 +974,8 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al
 				for p in inp {
 					a := f32(p.g) / 65535.0
 					c := ((1.0 - a) * bg + a * f32(p.r))
-					out[0] = u16(c)
-					out    = out[1:]
+					out[0].r = u16(c)
+					out      = out[1:]
 				}
 
 			} else if .alpha_premultiply in options {
@@ -983,14 +983,14 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al
 				for p in inp {
 					a := f32(p.g) / 65535.0
 					c := f32(p.r) * a
-					out[0] = u16(c)
-					out    = out[1:]
+					out[0].r = u16(c)
+					out      = out[1:]
 				}
 			} else {
 				// Just drop alpha on the floor.
 				for p in inp {
-					out[0] = p.r
-					out    = out[1:]
+					out[0].r = p.r
+					out      = out[1:]
 				}
 			}
 

core/image/netpbm/netpbm.odin

@@ -199,8 +199,8 @@ save_to_buffer :: proc(img: ^Image, custom_info: Info = {}, allocator := context
 				for x in 0 ..< img.width {
 					i := y * img.width + x
 					for c in 0 ..< img.channels {
-						i := i * img.channels + c
-						fmt.sbprintf(&data, "%i ", pixels[i])
+						j := i * img.channels + c
+						fmt.sbprintf(&data, "%i ", pixels[j])
 					}
 					fmt.sbprint(&data, "\n")
 				}
@@ -213,8 +213,8 @@ save_to_buffer :: proc(img: ^Image, custom_info: Info = {}, allocator := context
 				for x in 0 ..< img.width {
 					i := y * img.width + x
 					for c in 0 ..< img.channels {
-						i := i * img.channels + c
-						fmt.sbprintf(&data, "%i ", pixels[i])
+						j := i * img.channels + c
+						fmt.sbprintf(&data, "%i ", pixels[j])
 					}
 					fmt.sbprint(&data, "\n")
 				}
@@ -283,7 +283,7 @@ _parse_header_pnm :: proc(data: []byte) -> (header: Header, length: int, err: Er
 	current_field := 0
 	current_value := header_fields[0]
 
-	parse_loop: for d, i in data[SIG_LENGTH:] {
+	parse_loop: for d in data[SIG_LENGTH:] {
 		length += 1
 
 		// handle comments
@@ -728,4 +728,4 @@ _register :: proc() {
 		_ = destroy(img)
 	}
 	image.register(.NetPBM, loader, destroyer)
-}
+}

core/io/multi.odin

@@ -4,7 +4,6 @@ Multi_Reader :: struct {
 	readers: [dynamic]Reader,
 }
 
-@(private)
 _multi_reader_proc :: proc(stream_data: rawptr, mode: Stream_Mode, p: []byte, offset: i64, whence: Seek_From) -> (n: i64, err: Error) {
 	if mode == .Query {
 		return query_utility({.Read, .Query})
@@ -58,7 +57,6 @@ Multi_Writer :: struct {
 	writers: [dynamic]Writer,
 }
 
-@(private)
 _multi_writer_proc :: proc(stream_data: rawptr, mode: Stream_Mode, p: []byte, offset: i64, whence: Seek_From) -> (n: i64, err: Error) {
 	if mode == .Query {
 		return query_utility({.Write, .Query})

core/math/big/internal.odin

@@ -1181,28 +1181,18 @@ internal_cmp_digit :: internal_compare_digit
 */
 internal_int_compare_magnitude :: #force_inline proc(a, b: ^Int) -> (comparison: int) {
 	assert_if_nil(a, b)
-	/*
-		Compare based on used digits.
-	*/
+
+	// Compare based on used digits.
 	if a.used != b.used {
-		if a.used > b.used {
-			return +1
-		}
-		return -1
+		return +1 if a.used > b.used else -1
 	}
 
-	/*
-		Same number of used digits, compare based on their value.
-	*/
+	// Same number of used digits, compare based on their value.
 	#no_bounds_check for n := a.used - 1; n >= 0; n -= 1 {
 		if a.digit[n] != b.digit[n] {
-			if a.digit[n] > b.digit[n] {
-				return +1
-			}
-			return -1
+			return +1 if a.digit[n] > b.digit[n] else -1
 		}
 	}
-
 	return 0
 }
 internal_compare_magnitude :: proc { internal_int_compare_magnitude, }
@@ -2046,9 +2036,9 @@ internal_int_inverse_modulo :: proc(dest, a, b: ^Int, allocator := context.alloc
 	if internal_is_positive(a) && internal_eq(b, 1) { return internal_zero(dest)	}
 
 	/*
-		`b` cannot be negative and has to be > 1
+		`b` cannot be negative and b has to be > 1
 	*/
-	if internal_is_negative(b) || internal_gt(b, 1) { return .Invalid_Argument }
+	if internal_is_negative(b) || !internal_gt(b, 1) { return .Invalid_Argument }
 
 	/*
 		If the modulus is odd we can use a faster routine instead.
@@ -2057,6 +2047,7 @@ internal_int_inverse_modulo :: proc(dest, a, b: ^Int, allocator := context.alloc
 
 	return _private_inverse_modulo(dest, a, b)
 }
+internal_int_invmod :: internal_int_inverse_modulo
 internal_invmod :: proc{ internal_int_inverse_modulo, }
 
 /*
@@ -2954,4 +2945,4 @@ internal_zero_unused :: proc { internal_int_zero_unused, }
 
 /*
 	==========================    End of low-level routines   ==========================
-*/
+*/

core/math/big/prime.odin

@@ -44,7 +44,7 @@ internal_int_prime_is_divisible :: proc(a: ^Int, allocator := context.allocator)
 	Computes res == G**X mod P.
 	Assumes `res`, `G`, `X` and `P` to not be `nil` and for `G`, `X` and `P` to have been initialized.
 */
-internal_int_exponent_mod :: proc(res, G, X, P: ^Int, allocator := context.allocator) -> (err: Error) {
+internal_int_power_modulo :: proc(res, G, X, P: ^Int, allocator := context.allocator) -> (err: Error) {
 	context.allocator = allocator
 
 	dr: int
@@ -112,6 +112,9 @@ internal_int_exponent_mod :: proc(res, G, X, P: ^Int, allocator := context.alloc
 	*/
 	return _private_int_exponent_mod(res, G, X, P, 0)
 }
+internal_int_exponent_mod :: internal_int_power_modulo
+internal_int_powmod :: internal_int_power_modulo
+internal_powmod :: proc { internal_int_power_modulo, }
 
 /*
 	Kronecker/Legendre symbol (a|p)
@@ -1109,7 +1112,7 @@ internal_int_prime_next_prime :: proc(a: ^Int, trials: int, bbs_style: bool, all
 		Generate the restable.
 	*/
 	for x := 1; x < _PRIME_TAB_SIZE; x += 1 {
-		res_tab = internal_mod(a, _private_prime_table[x]) or_return
+		res_tab = cast(type_of(res_tab))(internal_mod(a, _private_prime_table[x]) or_return)
 	}
 
 	for {
@@ -1411,4 +1414,4 @@ number_of_rabin_miller_trials :: proc(bit_size: int) -> (number_of_trials: int)
 	case:
 		return  2		/* For keysizes bigger than 10_240 use always at least 2 Rounds */
 	}
-}
+}

core/math/fixed/fixed.odin

@@ -39,7 +39,6 @@ init_from_f64 :: proc(x: ^$T/Fixed($Backing, $Fraction_Width), val: f64) {
 }
 
 init_from_parts :: proc(x: ^$T/Fixed($Backing, $Fraction_Width), integer, fraction: Backing) {
-	i, f := math.modf(val)
 	x.i  = fraction
 	x.i &= 1<<Fraction_Width - 1
 	x.i |= integer
@@ -103,37 +102,51 @@ round :: proc(x: $T/Fixed($Backing, $Fraction_Width)) -> Backing {
 	return (x.i + (1 << (Fraction_Width - 1))) >> Fraction_Width
 }
 
-
-
 @(require_results)
 append :: proc(dst: []byte, x: $T/Fixed($Backing, $Fraction_Width)) -> string {
+	Integer_Width :: 8*size_of(Backing) - Fraction_Width
+
 	x := x
 	buf: [48]byte
 	i := 0
-	if x.i < 0 {
+
+	if !intrinsics.type_is_unsigned(Backing) && x.i == min(Backing) {
+		// edge case handling for signed numbers
 		buf[i] = '-'
 		i += 1
-		x.i = -x.i
-	}
-
-	integer := x.i >> Fraction_Width
-	fraction := x.i & (1<<Fraction_Width - 1)
-
-	s := strconv.append_uint(buf[i:], u64(integer), 10)
-	i += len(s)
-	if fraction != 0 {
-		buf[i] = '.'
-		i += 1
-		for fraction > 0 {
-			fraction *= 10
-			buf[i] = byte('0' + (fraction>>Fraction_Width))
+		i += copy(buf[i:], _power_of_two_table[Integer_Width])
+	} else {
+		if x.i < 0 {
+			buf[i] = '-'
 			i += 1
-			fraction &= 1<<Fraction_Width - 1
+			x.i = -x.i
+		}
+
+		when size_of(Backing) < 16 {
+			T :: u64
+			append_uint :: strconv.append_uint
+		} else {
+			T :: u128
+			append_uint :: strconv.append_u128
+		}
+
+		integer := T(x.i) >> Fraction_Width
+		fraction := T(x.i) & (1<<Fraction_Width - 1)
+
+		s := append_uint(buf[i:], integer, 10)
+		i += len(s)
+		if fraction != 0 {
+			buf[i] = '.'
+			i += 1
+			for fraction > 0 {
+				fraction *= 10
+				buf[i] = byte('0' + (fraction>>Fraction_Width) % 10)
+				i += 1
+				fraction &= 1<<Fraction_Width - 1
+			}
 		}
 	}
 
-
-
 	n := copy(dst, buf[:i])
 	return string(dst[:i])
 }
@@ -147,3 +160,137 @@ to_string :: proc(x: $T/Fixed($Backing, $Fraction_Width), allocator := context.a
 	copy(str, s)
 	return string(str)
 }
+
+
+@(private)
+_power_of_two_table := [129]string{
+	"0.5",
+	"1",
+	"2",
+	"4",
+	"8",
+	"16",
+	"32",
+	"64",
+	"128",
+	"256",
+	"512",
+	"1024",
+	"2048",
+	"4096",
+	"8192",
+	"16384",
+	"32768",
+	"65536",
+	"131072",
+	"262144",
+	"524288",
+	"1048576",
+	"2097152",
+	"4194304",
+	"8388608",
+	"16777216",
+	"33554432",
+	"67108864",
+	"134217728",
+	"268435456",
+	"536870912",
+	"1073741824",
+	"2147483648",
+	"4294967296",
+	"8589934592",
+	"17179869184",
+	"34359738368",
+	"68719476736",
+	"137438953472",
+	"274877906944",
+	"549755813888",
+	"1099511627776",
+	"2199023255552",
+	"4398046511104",
+	"8796093022208",
+	"17592186044416",
+	"35184372088832",
+	"70368744177664",
+	"140737488355328",
+	"281474976710656",
+	"562949953421312",
+	"1125899906842624",
+	"2251799813685248",
+	"4503599627370496",
+	"9007199254740992",
+	"18014398509481984",
+	"36028797018963968",
+	"72057594037927936",
+	"144115188075855872",
+	"288230376151711744",
+	"576460752303423488",
+	"1152921504606846976",
+	"2305843009213693952",
+	"4611686018427387904",
+	"9223372036854775808",
+	"18446744073709551616",
+	"36893488147419103232",
+	"73786976294838206464",
+	"147573952589676412928",
+	"295147905179352825856",
+	"590295810358705651712",
+	"1180591620717411303424",
+	"2361183241434822606848",
+	"4722366482869645213696",
+	"9444732965739290427392",
+	"18889465931478580854784",
+	"37778931862957161709568",
+	"75557863725914323419136",
+	"151115727451828646838272",
+	"302231454903657293676544",
+	"604462909807314587353088",
+	"1208925819614629174706176",
+	"2417851639229258349412352",
+	"4835703278458516698824704",
+	"9671406556917033397649408",
+	"19342813113834066795298816",
+	"38685626227668133590597632",
+	"77371252455336267181195264",
+	"154742504910672534362390528",
+	"309485009821345068724781056",
+	"618970019642690137449562112",
+	"1237940039285380274899124224",
+	"2475880078570760549798248448",
+	"4951760157141521099596496896",
+	"9903520314283042199192993792",
+	"19807040628566084398385987584",
+	"39614081257132168796771975168",
+	"79228162514264337593543950336",
+	"158456325028528675187087900672",
+	"316912650057057350374175801344",
+	"633825300114114700748351602688",
+	"1267650600228229401496703205376",
+	"2535301200456458802993406410752",
+	"5070602400912917605986812821504",
+	"10141204801825835211973625643008",
+	"20282409603651670423947251286016",
+	"40564819207303340847894502572032",
+	"81129638414606681695789005144064",
+	"162259276829213363391578010288128",
+	"324518553658426726783156020576256",
+	"649037107316853453566312041152512",
+	"1298074214633706907132624082305024",
+	"2596148429267413814265248164610048",
+	"5192296858534827628530496329220096",
+	"10384593717069655257060992658440192",
+	"20769187434139310514121985316880384",
+	"41538374868278621028243970633760768",
+	"83076749736557242056487941267521536",
+	"166153499473114484112975882535043072",
+	"332306998946228968225951765070086144",
+	"664613997892457936451903530140172288",
+	"1329227995784915872903807060280344576",
+	"2658455991569831745807614120560689152",
+	"5316911983139663491615228241121378304",
+	"10633823966279326983230456482242756608",
+	"21267647932558653966460912964485513216",
+	"42535295865117307932921825928971026432",
+	"85070591730234615865843651857942052864",
+	"170141183460469231731687303715884105728",
+}

core/math/linalg/extended.odin

@@ -326,6 +326,7 @@ ln :: proc "contextless" (x: $T) -> (out: T) where IS_FLOAT(ELEM_TYPE(T)) {
 
 @(require_results)
 log2 :: proc "contextless" (x: $T) -> (out: T) where IS_FLOAT(ELEM_TYPE(T)) {
+	INVLN2 :: 1.4426950408889634073599246810018921374266459541529859341354494069
 	when IS_ARRAY(T) {
 		for i in 0..<len(T) {
 			out[i] = INVLN2 * math.ln(x[i])
@@ -338,6 +339,7 @@ log2 :: proc "contextless" (x: $T) -> (out: T) where IS_FLOAT(ELEM_TYPE(T)) {
 
 @(require_results)
 log10 :: proc "contextless" (x: $T) -> (out: T) where IS_FLOAT(ELEM_TYPE(T)) {
+	INVLN10 :: 0.4342944819032518276511289189166050822943970058036665661144537831
 	when IS_ARRAY(T) {
 		for i in 0..<len(T) {
 			out[i] = INVLN10 * math.ln(x[i])
@@ -355,7 +357,7 @@ log :: proc "contextless" (x, b: $T) -> (out: T) where IS_FLOAT(ELEM_TYPE(T)) {
 			out[i] = math.ln(x[i]) / math.ln(cast(ELEM_TYPE(T))b[i])
 		}
 	} else {
-		out = INVLN10 * math.ln(x) / math.ln(cast(ELEM_TYPE(T))b)
+		out = math.ln(x) / math.ln(cast(ELEM_TYPE(T))b)
 	}
 	return
 }

core/math/linalg/specific.odin

@@ -2711,6 +2711,7 @@ to_quaternion :: proc{
 
 @(require_results)
 matrix2_orthonormalize_f16 :: proc "contextless" (m: Matrix2f16) -> (r: Matrix2f16) #no_bounds_check {
+	r = m
 	r[0] = normalize(m[0])
 
 	d0 := dot(r[0], r[1])
@@ -2721,6 +2722,7 @@ matrix2_orthonormalize_f16 :: proc "contextless" (m: Matrix2f16) -> (r: Matrix2f
 }
 @(require_results)
 matrix2_orthonormalize_f32 :: proc "contextless" (m: Matrix2f32) -> (r: Matrix2f32) #no_bounds_check {
+	r = m
 	r[0] = normalize(m[0])
 
 	d0 := dot(r[0], r[1])
@@ -2731,6 +2733,7 @@ matrix2_orthonormalize_f32 :: proc "contextless" (m: Matrix2f32) -> (r: Matrix2f
 }
 @(require_results)
 matrix2_orthonormalize_f64 :: proc "contextless" (m: Matrix2f64) -> (r: Matrix2f64) #no_bounds_check {
+	r = m
 	r[0] = normalize(m[0])
 
 	d0 := dot(r[0], r[1])
@@ -2748,6 +2751,7 @@ matrix2_orthonormalize :: proc{
 
 @(require_results)
 matrix3_orthonormalize_f16 :: proc "contextless" (m: Matrix3f16) -> (r: Matrix3f16) #no_bounds_check {
+	r = m
 	r[0] = normalize(m[0])
 
 	d0 := dot(r[0], r[1])
@@ -2763,6 +2767,7 @@ matrix3_orthonormalize_f16 :: proc "contextless" (m: Matrix3f16) -> (r: Matrix3f
 }
 @(require_results)
 matrix3_orthonormalize_f32 :: proc "contextless" (m: Matrix3f32) -> (r: Matrix3f32) #no_bounds_check {
+	r = m
 	r[0] = normalize(m[0])
 
 	d0 := dot(r[0], r[1])
@@ -2778,6 +2783,7 @@ matrix3_orthonormalize_f32 :: proc "contextless" (m: Matrix3f32) -> (r: Matrix3f
 }
 @(require_results)
 matrix3_orthonormalize_f64 :: proc "contextless" (m: Matrix3f64) -> (r: Matrix3f64) #no_bounds_check {
+	r = m
 	r[0] = normalize(m[0])
 
 	d0 := dot(r[0], r[1])

core/mem/alloc.odin

@@ -83,11 +83,7 @@ free :: proc(ptr: rawptr, allocator := context.allocator, loc := #caller_locatio
 }
 
 free_with_size :: proc(ptr: rawptr, byte_count: int, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
-	if ptr == nil || allocator.procedure == nil {
-		return nil
-	}
-	_, err := allocator.procedure(allocator.data, .Free, 0, 0, ptr, byte_count, loc)
-	return err
+	return runtime.mem_free_with_size(ptr, byte_count, allocator, loc)
 }
 
 free_bytes :: proc(bytes: []byte, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
@@ -130,19 +126,19 @@ query_info :: proc(pointer: rawptr, allocator: Allocator, loc := #caller_locatio
 
 
 delete_string :: proc(str: string, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
-	return free_with_size(raw_data(str), len(str), allocator, loc)
+	return runtime.delete_string(str, allocator, loc)
 }
 delete_cstring :: proc(str: cstring, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
-	return free((^byte)(str), allocator, loc)
+	return runtime.delete_cstring(str, allocator, loc)
 }
 delete_dynamic_array :: proc(array: $T/[dynamic]$E, loc := #caller_location) -> Allocator_Error {
-	return free_with_size(raw_data(array), cap(array)*size_of(E), array.allocator, loc)
+	return runtime.delete_dynamic_array(array, loc)
 }
 delete_slice :: proc(array: $T/[]$E, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
-	return free_with_size(raw_data(array), len(array)*size_of(E), allocator, loc)
+	return runtime.delete_slice(array, allocator, loc)
 }
 delete_map :: proc(m: $T/map[$K]$V, loc := #caller_location) -> Allocator_Error {
-	return runtime.map_free_dynamic(transmute(Raw_Map)m, runtime.map_info(T), loc)
+	return runtime.delete_map(m, loc)
 }
 
 
@@ -161,71 +157,40 @@ new :: proc($T: typeid, allocator := context.allocator, loc := #caller_location)
 }
 @(require_results)
 new_aligned :: proc($T: typeid, alignment: int, allocator := context.allocator, loc := #caller_location) -> (t: ^T, err: Allocator_Error) {
-	data := alloc_bytes(size_of(T), alignment, allocator, loc) or_return
-	t = (^T)(raw_data(data))
-	return
+	return runtime.new_aligned(T, alignment, allocator, loc)
 }
 @(require_results)
 new_clone :: proc(data: $T, allocator := context.allocator, loc := #caller_location) -> (t: ^T, err: Allocator_Error) {
-	backing := alloc_bytes(size_of(T), align_of(T), allocator, loc) or_return
-	t = (^T)(raw_data(backing))
-	if t != nil {
-		t^ = data
-		return t, nil
-	}
-	return nil, .Out_Of_Memory
+	return runtime.new_clone(data, allocator, loc)
 }
 
 @(require_results)
 make_aligned :: proc($T: typeid/[]$E, #any_int len: int, alignment: int, allocator := context.allocator, loc := #caller_location) -> (slice: T, err: Allocator_Error) {
-	runtime.make_slice_error_loc(loc, len)
-	data := alloc_bytes(size_of(E)*len, alignment, allocator, loc) or_return
-	if data == nil && size_of(E) != 0 {
-		return
-	}
-	slice = transmute(T)Raw_Slice{raw_data(data), len}
-	return
+	return runtime.make_aligned(T, len, alignment, allocator, loc)
 }
 @(require_results)
 make_slice :: proc($T: typeid/[]$E, #any_int len: int, allocator := context.allocator, loc := #caller_location) -> (T, Allocator_Error) {
-	return make_aligned(T, len, align_of(E), allocator, loc)
+	return runtime.make_slice(T, len, allocator, loc)
 }
 @(require_results)
 make_dynamic_array :: proc($T: typeid/[dynamic]$E, allocator := context.allocator, loc := #caller_location) -> (T, Allocator_Error) {
-	return make_dynamic_array_len_cap(T, 0, 16, allocator, loc)
+	return runtime.make_dynamic_array(T, allocator, loc)
 }
 @(require_results)
 make_dynamic_array_len :: proc($T: typeid/[dynamic]$E, #any_int len: int, allocator := context.allocator, loc := #caller_location) -> (T, Allocator_Error) {
-	return make_dynamic_array_len_cap(T, len, len, allocator, loc)
+	return runtime.make_dynamic_array(T, len, allocator, loc)
 }
 @(require_results)
 make_dynamic_array_len_cap :: proc($T: typeid/[dynamic]$E, #any_int len: int, #any_int cap: int, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) {
-	runtime.make_dynamic_array_error_loc(loc, len, cap)
-	data := alloc_bytes(size_of(E)*cap, align_of(E), allocator, loc) or_return
-	s := Raw_Dynamic_Array{raw_data(data), len, cap, allocator}
-	if data == nil && size_of(E) != 0 {
-		s.len, s.cap = 0, 0
-	}
-	array = transmute(T)s
-	return
+	return runtime.make_dynamic_array(T, len, cap, allocator, loc)
 }
 @(require_results)
 make_map :: proc($T: typeid/map[$K]$E, #any_int cap: int = 1<<runtime.MAP_MIN_LOG2_CAPACITY, allocator := context.allocator, loc := #caller_location) -> (m: T, err: Allocator_Error) {
-	runtime.make_map_expr_error_loc(loc, cap)
-	context.allocator = allocator
-
-	err = reserve_map(&m, cap, loc)
-	return
+	return runtime.make_map(T, cap, allocator, loc)
 }
 @(require_results)
 make_multi_pointer :: proc($T: typeid/[^]$E, #any_int len: int, allocator := context.allocator, loc := #caller_location) -> (mp: T, err: Allocator_Error) {
-	runtime.make_slice_error_loc(loc, len)
-	data := alloc_bytes(size_of(E)*len, align_of(E), allocator, loc) or_return
-	if data == nil && size_of(E) != 0 {
-		return
-	}
-	mp = cast(T)raw_data(data)
-	return
+	return runtime.make_multi_pointer(T, len, allocator, loc)
 }
 
 make :: proc{

core/mem/allocators.odin

@@ -84,11 +84,11 @@ arena_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
 	case .Free_All:
 		arena.offset = 0
 
-    case .Resize:
-        return default_resize_bytes_align(byte_slice(old_memory, old_size), size, alignment, arena_allocator(arena))
+	case .Resize:
+		return default_resize_bytes_align(byte_slice(old_memory, old_size), size, alignment, arena_allocator(arena))
 
-    case .Resize_Non_Zeroed:
-        return default_resize_bytes_align_non_zeroed(byte_slice(old_memory, old_size), size, alignment, arena_allocator(arena))
+	case .Resize_Non_Zeroed:
+		return default_resize_bytes_align_non_zeroed(byte_slice(old_memory, old_size), size, alignment, arena_allocator(arena))
 
 	case .Query_Features:
 		set := (^Allocator_Mode_Set)(old_memory)
@@ -860,3 +860,282 @@ panic_allocator :: proc() -> Allocator {
 		data = nil,
 	}
 }
+
+
+
+
+
+
+Buddy_Block :: struct #align(align_of(uint)) {
+	size:    uint,
+	is_free: bool,
+}
+
+@(require_results)
+buddy_block_next :: proc(block: ^Buddy_Block) -> ^Buddy_Block {
+	return (^Buddy_Block)(([^]byte)(block)[block.size:])
+}
+
+@(require_results)
+buddy_block_split :: proc(block: ^Buddy_Block, size: uint) -> ^Buddy_Block {
+	block := block
+	if block != nil && size != 0 {
+		// Recursive Split
+		for size < block.size {
+			sz := block.size >> 1
+			block.size = sz
+			block = buddy_block_next(block)
+			block.size = sz
+			block.is_free = true
+		}
+		if size <= block.size {
+			return block
+		}
+	}
+	// Block cannot fit the requested allocation size
+	return nil
+}
+
+buddy_block_coalescence :: proc(head, tail: ^Buddy_Block) {
+	for {
+		// Keep looping until there are no more buddies to coalesce
+		block := head
+		buddy := buddy_block_next(block)
+
+		no_coalescence := true
+		for block < tail && buddy < tail { // make sure the buddies are within the range
+			if block.is_free && buddy.is_free && block.size == buddy.size {
+				// Coalesce buddies into one
+				block.size <<= 1
+				block = buddy_block_next(block)
+				if block < tail {
+					buddy = buddy_block_next(block)
+					no_coalescence = false
+				}
+			} else if block.size < buddy.size {
+				// The buddy block is split into smaller blocks
+				block = buddy
+				buddy = buddy_block_next(buddy)
+			} else {
+				block = buddy_block_next(buddy)
+				if block < tail {
+					// Leave the buddy block for the next iteration
+					buddy = buddy_block_next(block)
+				}
+			}
+		}
+
+		if no_coalescence {
+			return
+		}
+	}
+}
+
+
+@(require_results)
+buddy_block_find_best :: proc(head, tail: ^Buddy_Block, size: uint) -> ^Buddy_Block {
+	assert(size != 0)
+
+	best_block: ^Buddy_Block
+	block := head                    // left
+	buddy := buddy_block_next(block) // right
+
+	// The entire memory section between head and tail is free,
+	// just call 'buddy_block_split' to get the allocation
+	if buddy == tail && block.is_free {
+		return buddy_block_split(block, size)
+	}
+
+	// Find the block which is the 'best_block' to requested allocation sized
+	for block < tail && buddy < tail { // make sure the buddies are within the range
+		// If both buddies are free, coalesce them together
+		// NOTE: this is an optimization to reduce fragmentation
+		//       this could be completely ignored
+		if block.is_free && buddy.is_free && block.size == buddy.size {
+			block.size <<= 1
+			if size <= block.size && (best_block == nil || block.size <= best_block.size) {
+				best_block = block
+			}
+
+			block = buddy_block_next(buddy)
+			if block < tail {
+				// Delay the buddy block for the next iteration
+				buddy = buddy_block_next(block)
+			}
+			continue
+		}
+
+
+		if block.is_free && size <= block.size &&
+		   (best_block == nil || block.size <= best_block.size) {
+			best_block = block
+		}
+
+		if buddy.is_free && size <= buddy.size &&
+		   (best_block == nil || buddy.size < best_block.size) {
+			// If each buddy are the same size, then it makes more sense
+			// to pick the buddy as it "bounces around" less
+			best_block = buddy
+		}
+
+		if (block.size <= buddy.size) {
+			block = buddy_block_next(buddy)
+			if (block < tail) {
+				// Delay the buddy block for the next iteration
+				buddy = buddy_block_next(block)
+			}
+		} else {
+			// Buddy was split into smaller blocks
+			block = buddy
+			buddy = buddy_block_next(buddy)
+		}
+	}
+
+	if best_block != nil {
+		// This will handle the case if the 'best_block' is also the perfect fit
+		return buddy_block_split(best_block, size)
+	}
+
+	// Maybe out of memory
+	return nil
+}
+
+
+Buddy_Allocator :: struct {
+	head: ^Buddy_Block,
+	tail: ^Buddy_Block,
+	alignment: uint,
+}
+
+@(require_results)
+buddy_allocator :: proc(b: ^Buddy_Allocator) -> Allocator {
+	return Allocator{
+		procedure = buddy_allocator_proc,
+		data      = b,
+	}
+}
+
+buddy_allocator_init :: proc(b: ^Buddy_Allocator, data: []byte, alignment: uint) {
+	assert(data != nil)
+	assert(is_power_of_two(uintptr(len(data))))
+	assert(is_power_of_two(uintptr(alignment)))
+
+	alignment := alignment
+	if alignment < size_of(Buddy_Block) {
+		alignment = size_of(Buddy_Block)
+	}
+
+	ptr := raw_data(data)
+	assert(uintptr(ptr) % uintptr(alignment) == 0, "data is not aligned to minimum alignment")
+
+	b.head = (^Buddy_Block)(ptr)
+
+	b.head.size = len(data)
+	b.head.is_free = true
+
+	b.tail = buddy_block_next(b.head)
+
+	b.alignment = alignment
+}
+
+@(require_results)
+buddy_block_size_required :: proc(b: ^Buddy_Allocator, size: uint) -> uint {
+	size := size
+	actual_size := b.alignment
+	size += size_of(Buddy_Block)
+	size = align_forward_uint(size, b.alignment)
+
+	for size > actual_size {
+		actual_size <<= 1
+	}
+
+	return actual_size
+}
+
+@(require_results)
+buddy_allocator_alloc :: proc(b: ^Buddy_Allocator, size: uint, zeroed: bool) -> ([]byte, Allocator_Error) {
+	if size != 0 {
+		actual_size := buddy_block_size_required(b, size)
+
+		found := buddy_block_find_best(b.head, b.tail, actual_size)
+		if found != nil {
+			// Try to coalesce all the free buddy blocks and then search again
+			buddy_block_coalescence(b.head, b.tail)
+			found = buddy_block_find_best(b.head, b.tail, actual_size)
+		}
+		if found == nil {
+			return nil, .Out_Of_Memory
+		}
+		found.is_free = false
+
+		data := ([^]byte)(found)[b.alignment:][:size]
+		if zeroed {
+			zero_slice(data)
+		}
+		return data, nil
+	}
+	return nil, nil
+}
+
+buddy_allocator_free :: proc(b: ^Buddy_Allocator, ptr: rawptr) -> Allocator_Error {
+	if ptr != nil {
+		if !(b.head <= ptr && ptr <= b.tail) {
+			return .Invalid_Pointer
+		}
+
+		block := (^Buddy_Block)(([^]byte)(ptr)[-b.alignment:])
+		block.is_free = true
+
+		buddy_block_coalescence(b.head, b.tail)
+	}
+	return nil
+}
+
+buddy_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
+                             size, alignment: int,
+                             old_memory: rawptr, old_size: int,loc := #caller_location) -> ([]byte, Allocator_Error) {
+
+	b := (^Buddy_Allocator)(allocator_data)
+
+	switch mode {
+	case .Alloc, .Alloc_Non_Zeroed:
+		return buddy_allocator_alloc(b, uint(size), mode == .Alloc)
+	case .Resize:
+		return default_resize_bytes_align(byte_slice(old_memory, old_size), size, alignment, buddy_allocator(b))
+	case .Resize_Non_Zeroed:
+		return default_resize_bytes_align_non_zeroed(byte_slice(old_memory, old_size), size, alignment, buddy_allocator(b))
+	case .Free:
+		return nil, buddy_allocator_free(b, old_memory)
+	case .Free_All:
+
+		alignment := b.alignment
+		head := ([^]byte)(b.head)
+		tail := ([^]byte)(b.tail)
+		data := head[:ptr_sub(tail, head)]
+		buddy_allocator_init(b, data, alignment)
+
+	case .Query_Features:
+		set := (^Allocator_Mode_Set)(old_memory)
+		if set != nil {
+			set^ = {.Query_Features, .Alloc, .Alloc_Non_Zeroed, .Resize, .Resize_Non_Zeroed, .Free, .Free_All, .Query_Info}
+		}
+		return nil, nil
+
+	case .Query_Info:
+		info := (^Allocator_Query_Info)(old_memory)
+		if info != nil && info.pointer != nil {
+			ptr := old_memory
+			if !(b.head <= ptr && ptr <= b.tail) {
+				return nil, .Invalid_Pointer
+			}
+
+			block := (^Buddy_Block)(([^]byte)(ptr)[-b.alignment:])
+			info.size = int(block.size)
+			info.alignment = int(b.alignment)
+			return byte_slice(info, size_of(info^)), nil
+		}
+		return nil, nil
+	}
+
+	return nil, nil
+}

core/mem/mem.odin

@@ -45,6 +45,8 @@ copy_non_overlapping :: proc "contextless" (dst, src: rawptr, len: int) -> rawpt
 	intrinsics.mem_copy_non_overlapping(dst, src, len)
 	return dst
 }
+
+@(require_results)
 compare :: proc "contextless" (a, b: []byte) -> int {
 	res := compare_byte_ptrs(raw_data(a), raw_data(b), min(len(a), len(b)))
 	if res == 0 && len(a) != len(b) {

core/mem/mutex_allocator.odin

@@ -0,0 +1,33 @@
+//+build !freestanding
+package mem
+
+import "core:sync"
+
+Mutex_Allocator :: struct {
+	backing: Allocator,
+	mutex:   sync.Mutex,
+}
+
+mutex_allocator_init :: proc(m: ^Mutex_Allocator, backing_allocator: Allocator) {
+	m.backing = backing_allocator
+	m.mutex = {}
+}
+
+
+@(require_results)
+mutex_allocator :: proc(m: ^Mutex_Allocator) -> Allocator {
+	return Allocator{
+		procedure = mutex_allocator_proc,
+		data = m,
+	}
+}
+
+mutex_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
+                             size, alignment: int,
+                             old_memory: rawptr, old_size: int, loc := #caller_location) -> (result: []byte, err: Allocator_Error) {
+	m := (^Mutex_Allocator)(allocator_data)
+
+	sync.mutex_guard(&m.mutex)
+	return m.backing.procedure(m.backing.data, mode, size, alignment, old_memory, old_size, loc)
+}
+

core/mem/virtual/arena.odin

@@ -17,18 +17,23 @@ Arena_Kind :: enum uint {
 	Buffer: A single `Memory_Block` created from a user provided []byte.
 */
 Arena :: struct {
-	kind:               Arena_Kind,
-	curr_block:         ^Memory_Block,
-	total_used:         uint,
-	total_reserved:     uint,
-	minimum_block_size: uint,
-	temp_count:         uint,
-	mutex:              sync.Mutex,
+	kind:                Arena_Kind,
+	curr_block:          ^Memory_Block,
+
+	total_used:          uint,
+	total_reserved:      uint,
+
+	default_commit_size: uint, // commit size <= reservation size
+	minimum_block_size:  uint, // block size == total reservation
+
+	temp_count:          uint,
+	mutex:               sync.Mutex,
 }
 
 
 // 1 MiB should be enough to start with
 DEFAULT_ARENA_STATIC_COMMIT_SIZE         :: mem.Megabyte
+DEFAULT_ARENA_GROWING_COMMIT_SIZE        :: 8*mem.Megabyte
 DEFAULT_ARENA_GROWING_MINIMUM_BLOCK_SIZE :: DEFAULT_ARENA_STATIC_COMMIT_SIZE
 
 // 1 GiB on 64-bit systems, 128 MiB on 32-bit systems by default
@@ -102,8 +107,20 @@ arena_alloc :: proc(arena: ^Arena, size: uint, alignment: uint, loc := #caller_l
 		if arena.curr_block == nil || (safe_add(arena.curr_block.used, needed) or_else 0) > arena.curr_block.reserved {
 			if arena.minimum_block_size == 0 {
 				arena.minimum_block_size = DEFAULT_ARENA_GROWING_MINIMUM_BLOCK_SIZE
+				arena.minimum_block_size = mem.align_forward_uint(arena.minimum_block_size, DEFAULT_PAGE_SIZE)
+			}
+			if arena.default_commit_size == 0 {
+				arena.default_commit_size = min(DEFAULT_ARENA_GROWING_COMMIT_SIZE, arena.minimum_block_size)
+				arena.default_commit_size = mem.align_forward_uint(arena.default_commit_size, DEFAULT_PAGE_SIZE)
 			}
 
+			if arena.default_commit_size != 0 {
+				arena.default_commit_size, arena.minimum_block_size =
+					min(arena.default_commit_size, arena.minimum_block_size),
+					max(arena.default_commit_size, arena.minimum_block_size)
+			}
+
+			needed = max(needed, arena.default_commit_size)
 			block_size := max(needed, arena.minimum_block_size)
 
 			new_block := memory_block_alloc(needed, block_size, alignment, {}) or_return
@@ -113,7 +130,7 @@ arena_alloc :: proc(arena: ^Arena, size: uint, alignment: uint, loc := #caller_l
 		}
 
 		prev_used := arena.curr_block.used
-		data, err = alloc_from_memory_block(arena.curr_block, size, alignment)
+		data, err = alloc_from_memory_block(arena.curr_block, size, alignment, default_commit_size=arena.default_commit_size)
 		arena.total_used += arena.curr_block.used - prev_used
 	case .Static:
 		if arena.curr_block == nil {
@@ -122,12 +139,17 @@ arena_alloc :: proc(arena: ^Arena, size: uint, alignment: uint, loc := #caller_l
 			}
 			arena_init_static(arena, reserved=arena.minimum_block_size, commit_size=DEFAULT_ARENA_STATIC_COMMIT_SIZE) or_return
 		}
-		fallthrough
+		if arena.curr_block == nil {
+			return nil, .Out_Of_Memory
+		}
+		data, err = alloc_from_memory_block(arena.curr_block, size, alignment, default_commit_size=arena.default_commit_size)
+		arena.total_used = arena.curr_block.used
+
 	case .Buffer:
 		if arena.curr_block == nil {
 			return nil, .Out_Of_Memory
 		}
-		data, err = alloc_from_memory_block(arena.curr_block, size, alignment)
+		data, err = alloc_from_memory_block(arena.curr_block, size, alignment, default_commit_size=0)
 		arena.total_used = arena.curr_block.used
 	}
 	return

core/mem/virtual/arena_util.odin

@@ -0,0 +1,67 @@
+package mem_virtual
+
+import "base:runtime"
+_ :: runtime
+
+// The `new` procedure allocates memory for a type `T` from a `virtual.Arena`. The second argument is a type,
+// not a value, and the value return is a pointer to a newly allocated value of that type using the specified allocator.
+@(require_results)
+new :: proc(arena: ^Arena, $T: typeid, loc := #caller_location) -> (ptr: ^T, err: Allocator_Error) {
+	return new_aligned(arena, T, align_of(T), loc)
+}
+
+// The `new_aligned` procedure allocates memory for a type `T` from a `virtual.Arena` with a specified `alignment`.
+// The second argument is a type, not a value, and the value return is a pointer to a newly allocated value of
+// that type using the specified allocator.
+@(require_results)
+new_aligned :: proc(arena: ^Arena, $T: typeid, alignment: uint, loc := #caller_location) -> (ptr: ^T, err: Allocator_Error) {
+	data := arena_alloc(arena, size_of(T), alignment, loc) or_return
+	ptr = (^T)(raw_data(data))
+	return
+}
+
+// `make_slice` allocates and initializes a slice. Like `new`, the second argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(require_results)
+make_slice :: proc(arena: ^Arena, $T: typeid/[]$E, #any_int len: int, loc := #caller_location) -> (T, Allocator_Error) {
+	return make_aligned(arena, T, len, align_of(E), loc)
+}
+
+// `make_aligned` allocates and initializes a slice. Like `new`, the second argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(require_results)
+make_aligned :: proc(arena: ^Arena, $T: typeid/[]$E, #any_int len: int, alignment: uint, loc := #caller_location) -> (T, Allocator_Error) {
+	runtime.make_slice_error_loc(loc, len)
+	data, err := arena_alloc(arena, size_of(E)*uint(len), alignment, loc)
+	if data == nil && size_of(E) != 0 {
+		return nil, err
+	}
+	s := ([^]E)(raw_data(data))[:len]
+	return T(s), err
+}
+
+
+// `make_multi_pointer` allocates and initializes a dynamic array. Like `new`, the second argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// This is "similar" to doing `raw_data(make([]E, len, allocator))`.
+//
+// Note: Prefer using the procedure group `make`.
+@(require_results)
+make_multi_pointer :: proc(arena: ^Arena, $T: typeid/[^]$E, #any_int len: int, loc := #caller_location) -> (T, Allocator_Error) {
+	runtime.make_slice_error_loc(loc, len)
+	data, err := arena_alloc(arena, size_of(E)*uint(len), align_of(E), loc)
+	if data == nil && size_of(E) != 0 {
+		return nil, err
+	}
+	return (T)(raw_data(data)), err
+}
+
+make :: proc{
+	make_slice,
+	make_multi_pointer,
+}

core/mem/virtual/virtual.odin

@@ -112,7 +112,7 @@ memory_block_alloc :: proc(committed, reserved: uint, alignment: uint = 0, flags
 }
 
 @(require_results)
-alloc_from_memory_block :: proc(block: ^Memory_Block, min_size, alignment: uint) -> (data: []byte, err: Allocator_Error) {
+alloc_from_memory_block :: proc(block: ^Memory_Block, min_size, alignment: uint, default_commit_size: uint = 0) -> (data: []byte, err: Allocator_Error) {
 	calc_alignment_offset :: proc "contextless" (block: ^Memory_Block, alignment: uintptr) -> uint {
 		alignment_offset := uint(0)
 		ptr := uintptr(block.base[block.used:])
@@ -123,7 +123,7 @@ alloc_from_memory_block :: proc(block: ^Memory_Block, min_size, alignment: uint)
 		return alignment_offset
 		
 	}
-	do_commit_if_necessary :: proc(block: ^Memory_Block, size: uint) -> (err: Allocator_Error) {
+	do_commit_if_necessary :: proc(block: ^Memory_Block, size: uint, default_commit_size: uint) -> (err: Allocator_Error) {
 		if block.committed - block.used < size {
 			pmblock := (^Platform_Memory_Block)(block)
 			base_offset := uint(uintptr(pmblock.block.base) - uintptr(pmblock))
@@ -133,7 +133,7 @@ alloc_from_memory_block :: proc(block: ^Memory_Block, min_size, alignment: uint)
 			extra_size := max(size, block.committed>>1)
 			platform_total_commit := base_offset + block.used + extra_size
 			platform_total_commit = align_formula(platform_total_commit, DEFAULT_PAGE_SIZE)
-			platform_total_commit = min(platform_total_commit, pmblock.reserved)
+			platform_total_commit = min(max(platform_total_commit, default_commit_size), pmblock.reserved)
 
 			assert(pmblock.committed <= pmblock.reserved)
 			assert(pmblock.committed < platform_total_commit)
@@ -163,7 +163,7 @@ alloc_from_memory_block :: proc(block: ^Memory_Block, min_size, alignment: uint)
 		return
 	}
 	assert(block.committed <= block.reserved)
-	do_commit_if_necessary(block, size) or_return
+	do_commit_if_necessary(block, size, default_commit_size) or_return
 
 	data = block.base[block.used+alignment_offset:][:min_size]
 	block.used += size

core/net/addr.odin

@@ -119,9 +119,9 @@ aton :: proc(address_and_maybe_port: string, family: Address_Family, allow_decim
 		}
 
 		a: [4]u8 = ---
-		for v, i in buf {
+		for v, j in buf {
 			if v > 255 { return {}, false }
-			a[i] = u8(v)
+			a[j] = u8(v)
 		}
 		return IP4_Address(a), true
 

core/net/dns.odin

@@ -816,7 +816,6 @@ parse_response :: proc(response: []u8, filter: DNS_Record_Type = nil, allocator
 
 		dq_sz :: 4
 		hn_sz := skip_hostname(response, cur_idx) or_return
-		dns_query := mem.slice_data_cast([]u16be, response[cur_idx+hn_sz:cur_idx+hn_sz+dq_sz])
 
 		cur_idx += hn_sz + dq_sz
 	}

core/net/dns_windows.odin

@@ -85,11 +85,9 @@ _get_dns_records_os :: proc(hostname: string, type: DNS_Record_Type, allocator :
 			append(&recs, record)
 
 		case .CNAME:
-
-			hostname := strings.clone(string(r.Data.CNAME))
 			record := DNS_Record_CNAME{
 				base      = base_record,
-				host_name = hostname,
+				host_name = strings.clone(string(r.Data.CNAME)),
 			}
 			append(&recs, record)
 
@@ -107,10 +105,9 @@ _get_dns_records_os :: proc(hostname: string, type: DNS_Record_Type, allocator :
 			}
 
 		case .NS:
-			hostname := strings.clone(string(r.Data.NS))
 			record := DNS_Record_NS{
 				base      = base_record,
-				host_name = hostname,
+				host_name = strings.clone(string(r.Data.NS)),
 			}
 			append(&recs, record)
 

core/net/socket.odin

@@ -161,11 +161,10 @@ recv_any :: proc(socket: Any_Socket, buf: []byte) -> (
 ) {
 	switch socktype in socket {
 	case TCP_Socket:
-		bytes_read, err := recv_tcp(socktype, buf)
-		return bytes_read, nil, err
+		bytes_read, err = recv_tcp(socktype, buf)
+		return
 	case UDP_Socket:
-		bytes_read, endpoint, err := recv_udp(socktype, buf)
-		return bytes_read, endpoint, err
+		return recv_udp(socktype, buf)
 	case: panic("Not supported")
 	}
 }

core/odin/ast/ast.odin

@@ -597,6 +597,7 @@ Field_Flag :: enum {
 	Any_Int,
 	Subtype,
 	By_Ptr,
+	No_Broadcast,
 
 	Results,
 	Tags,
@@ -616,6 +617,7 @@ field_flag_strings := [Field_Flag]string{
 	.Any_Int            = "#any_int",
 	.Subtype            = "#subtype",
 	.By_Ptr             = "#by_ptr",
+	.No_Broadcast       = "#no_broadcast",
 
 	.Results            = "results",
 	.Tags               = "field tag",
@@ -624,12 +626,13 @@ field_flag_strings := [Field_Flag]string{
 }
 
 field_hash_flag_strings := []struct{key: string, flag: Field_Flag}{
-	{"no_alias", .No_Alias},
-	{"c_vararg", .C_Vararg},
-	{"const",    .Const},
-	{"any_int",  .Any_Int},
-	{"subtype",  .Subtype},
-	{"by_ptr",   .By_Ptr},
+	{"no_alias",     .No_Alias},
+	{"c_vararg",     .C_Vararg},
+	{"const",        .Const},
+	{"any_int",      .Any_Int},
+	{"subtype",      .Subtype},
+	{"by_ptr",       .By_Ptr},
+	{"no_broadcast", .No_Broadcast},
 }
 
 
@@ -650,6 +653,7 @@ Field_Flags_Signature :: Field_Flags{
 	.Const,
 	.Any_Int,
 	.By_Ptr,
+	.No_Broadcast,
 	.Default_Parameters,
 }
 
@@ -838,6 +842,23 @@ Matrix_Type :: struct {
 	elem:         ^Expr,
 }
 
+Bit_Field_Type :: struct {
+	using node:   Expr,
+	tok_pos:      tokenizer.Pos,
+	backing_type: ^Expr,
+	open:         tokenizer.Pos,
+	fields:       []^Bit_Field_Field,
+	close:        tokenizer.Pos,
+}
+
+Bit_Field_Field :: struct {
+	using node: Node,
+	docs:       ^Comment_Group,
+	name:       ^Expr,
+	type:       ^Expr,
+	bit_size:   ^Expr,
+	comments:   ^Comment_Group,
+}
 
 Any_Node :: union {
 	^Package,
@@ -894,6 +915,7 @@ Any_Node :: union {
 	^Map_Type,
 	^Relative_Type,
 	^Matrix_Type,
+	^Bit_Field_Type,
 
 	^Bad_Stmt,
 	^Empty_Stmt,
@@ -924,6 +946,7 @@ Any_Node :: union {
 	^Attribute,
 	^Field,
 	^Field_List,
+	^Bit_Field_Field,
 }
 
 
@@ -978,6 +1001,7 @@ Any_Expr :: union {
 	^Map_Type,
 	^Relative_Type,
 	^Matrix_Type,
+	^Bit_Field_Type,
 }
 
 

core/odin/ast/clone.odin

@@ -336,6 +336,13 @@ clone_node :: proc(node: ^Node) -> ^Node {
 		case ^Relative_Type:
 			r.tag = clone(r.tag)
 			r.type = clone(r.type)
+		case ^Bit_Field_Type:
+			r.backing_type = clone(r.backing_type)
+			r.fields       = auto_cast clone(r.fields)
+		case ^Bit_Field_Field:
+			r.name     = clone(r.name)
+			r.type     = clone(r.type)
+			r.bit_size = clone(r.bit_size)
 		case:
 			fmt.panicf("Unhandled node kind: %v", r)
 		}

core/odin/ast/walk.odin

@@ -414,7 +414,15 @@ walk :: proc(v: ^Visitor, node: ^Node) {
 		walk(v, n.row_count)
 		walk(v, n.column_count)
 		walk(v, n.elem)
-
+	case ^Bit_Field_Type:
+		walk(v, n.backing_type)
+		for f in n.fields {
+			walk(v, f)
+		}
+	case ^Bit_Field_Field:
+		walk(v, n.name)
+		walk(v, n.type)
+		walk(v, n.bit_size)
 	case:
 		fmt.panicf("ast.walk: unexpected node type %T", n)
 	}

core/odin/doc-format/doc_format.odin

@@ -11,7 +11,7 @@ String :: distinct Array(byte)
 
 Version_Type_Major :: 0
 Version_Type_Minor :: 3
-Version_Type_Patch :: 0
+Version_Type_Patch :: 1
 
 Version_Type :: struct {
 	major, minor, patch: u8,
@@ -102,13 +102,15 @@ Entity_Flag :: enum u32le {
 	Foreign = 0,
 	Export  = 1,
 
-	Param_Using     = 2, // using
-	Param_Const     = 3, // #const
-	Param_Auto_Cast = 4, // auto_cast
-	Param_Ellipsis  = 5, // Variadic parameter
-	Param_CVararg   = 6, // #c_vararg
-	Param_No_Alias  = 7, // #no_alias
-	Param_Any_Int   = 8, // #any_int
+	Param_Using        = 2, // using
+	Param_Const        = 3, // #const
+	Param_Auto_Cast    = 4, // auto_cast
+	Param_Ellipsis     = 5, // Variadic parameter
+	Param_CVararg      = 6, // #c_vararg
+	Param_No_Alias     = 7, // #no_alias
+	Param_Any_Int      = 8, // #any_int
+	Param_By_Ptr       = 9, // #by_ptr
+	Param_No_Broadcast = 10, // #no_broadcast
 
 	Bit_Field_Field = 19,
 
@@ -247,7 +249,6 @@ Type :: struct {
 	// .Bit_Set            - <=2 types: 0=element type, 1=underlying type (Underlying_Type flag will be set)
 	// .Simd_Vector        - 1 type:    0=element
 	// .Relative_Pointer   - 2 types:   0=pointer type, 1=base integer
-	// .Relative_Slice     - 2 types:   0=slice type, 1=base integer
 	// .Multi_Pointer      - 1 type:    0=element
 	// .Matrix             - 1 type:    0=element
 	// .Soa_Pointer        - 1 type:    0=element

core/odin/parser/parser.odin

@@ -416,24 +416,28 @@ end_of_line_pos :: proc(p: ^Parser, tok: tokenizer.Token) -> tokenizer.Pos {
 }
 
 expect_closing_brace_of_field_list :: proc(p: ^Parser) -> tokenizer.Token {
+	return expect_closing_token_of_field_list(p, .Close_Brace, "field list")
+}
+
+expect_closing_token_of_field_list :: proc(p: ^Parser, closing_kind: tokenizer.Token_Kind, msg: string) -> tokenizer.Token {
 	token := p.curr_tok
-	if allow_token(p, .Close_Brace) {
+	if allow_token(p, closing_kind) {
 		return token
 	}
 	if allow_token(p, .Semicolon) && !tokenizer.is_newline(token) {
 		str := tokenizer.token_to_string(token)
 		error(p, end_of_line_pos(p, p.prev_tok), "expected a comma, got %s", str)
 	}
-	expect_brace := expect_token(p, .Close_Brace)
+	expect_closing := expect_token_after(p, closing_kind, msg)
 
-	if expect_brace.kind != .Close_Brace {
-		for p.curr_tok.kind != .Close_Brace && p.curr_tok.kind != .EOF && !is_non_inserted_semicolon(p.curr_tok) {
+	if expect_closing.kind != closing_kind {
+		for p.curr_tok.kind != closing_kind && p.curr_tok.kind != .EOF && !is_non_inserted_semicolon(p.curr_tok) {
 			advance_token(p)
 		}
 		return p.curr_tok
 	} 
 
-	return expect_brace
+	return expect_closing
 }
 
 expect_closing_parentheses_of_field_list :: proc(p: ^Parser) -> tokenizer.Token {
@@ -531,7 +535,7 @@ is_semicolon_optional_for_node :: proc(p: ^Parser, node: ^ast.Node) -> bool {
 		return is_semicolon_optional_for_node(p, n.type)
 	case ^ast.Pointer_Type:
 		return is_semicolon_optional_for_node(p, n.elem)
-	case ^ast.Struct_Type, ^ast.Union_Type, ^ast.Enum_Type:
+	case ^ast.Struct_Type, ^ast.Union_Type, ^ast.Enum_Type, ^ast.Bit_Set_Type, ^ast.Bit_Field_Type:
 		// Require semicolon within a procedure body
 		return p.curr_proc == nil
 	case ^ast.Proc_Lit:
@@ -1354,6 +1358,7 @@ parse_stmt :: proc(p: ^Parser) -> ^ast.Stmt {
 
 		rs := ast.new(ast.Return_Stmt, tok.pos, end)
 		rs.results = results[:]
+		expect_semicolon(p, rs)
 		return rs
 
 	case .Break, .Continue, .Fallthrough:
@@ -2790,6 +2795,48 @@ parse_operand :: proc(p: ^Parser, lhs: bool) -> ^ast.Expr {
 		mt.column_count = column_count
 		mt.elem = elem
 		return mt
+	
+	case .Bit_Field:
+		tok := expect_token(p, .Bit_Field)
+
+		backing_type := parse_type_or_ident(p)
+		if backing_type == nil {
+			token := advance_token(p)
+			error(p, token.pos, "Expected a backing type for a 'bit_field'")
+		}
+
+		skip_possible_newline_for_literal(p)
+		open := expect_token_after(p, .Open_Brace, "bit_field")
+
+		fields: [dynamic]^ast.Bit_Field_Field
+		for p.curr_tok.kind != .Close_Brace && p.curr_tok.kind != .EOF {
+			name := parse_ident(p)
+			expect_token(p, .Colon)
+			type := parse_type(p)
+			expect_token(p, .Or)
+			bit_size := parse_expr(p, true)
+
+			field := ast.new(ast.Bit_Field_Field, name.pos, bit_size)
+
+			field.name     = name
+			field.type     = type
+			field.bit_size = bit_size
+
+			append(&fields, field)
+
+			allow_token(p, .Comma) or_break
+		}
+
+		close := expect_closing_brace_of_field_list(p)
+
+		bf := ast.new(ast.Bit_Field_Type, tok.pos, close.pos)
+
+		bf.tok_pos      = tok.pos
+		bf.backing_type = backing_type
+		bf.open         = open.pos
+		bf.fields       = fields[:]
+		bf.close        = close.pos
+		return bf
 
 	case .Asm:
 		tok := expect_token(p, .Asm)
@@ -2897,7 +2944,8 @@ is_literal_type :: proc(expr: ^ast.Expr) -> bool {
 		^ast.Map_Type,
 		^ast.Bit_Set_Type,
 		^ast.Matrix_Type,
-		^ast.Call_Expr:
+		^ast.Call_Expr,
+		^ast.Bit_Field_Type:
 		return true
 	}
 	return false
@@ -2947,7 +2995,8 @@ parse_literal_value :: proc(p: ^Parser, type: ^ast.Expr) -> ^ast.Comp_Lit {
 	}
 	p.expr_level -= 1
 
-	close := expect_token_after(p, .Close_Brace, "compound literal")
+  	skip_possible_newline(p)
+	close := expect_closing_brace_of_field_list(p)
 
 	pos := type.pos if type != nil else open.pos
 	lit := ast.new(ast.Comp_Lit, pos, end_pos(close))
@@ -3010,7 +3059,7 @@ parse_call_expr :: proc(p: ^Parser, operand: ^ast.Expr) -> ^ast.Expr {
 		allow_token(p, .Comma) or_break
 	}
 
-	close := expect_token_after(p, .Close_Paren, "argument list")
+	close := expect_closing_token_of_field_list(p, .Close_Paren, "argument list")
 	p.expr_level -= 1
 
 	ce := ast.new(ast.Call_Expr, operand.pos, end_pos(close))

core/odin/printer/printer.odin

@@ -643,7 +643,7 @@ align_switch_stmt :: proc(p: ^Printer, index: int) {
 	format_tokens := make([dynamic]TokenAndLength, 0, brace_token.parameter_count, context.temp_allocator)
 
 	//find all the switch cases that are one lined
-	for line, line_index in p.lines[brace_line + 1:] {
+	for line in p.lines[brace_line + 1:] {
 
 		case_found  := false
 		colon_found := false
@@ -716,7 +716,7 @@ align_enum :: proc(p: ^Printer, index: int) {
 
 	format_tokens := make([dynamic]TokenAndLength, 0, brace_token.parameter_count, context.temp_allocator)
 
-	for line, line_index in p.lines[brace_line + 1:] {
+	for line in p.lines[brace_line + 1:] {
 		length := 0
 
 		for format_token, i in line.format_tokens {
@@ -880,7 +880,7 @@ align_comments :: proc(p: ^Printer) {
 
 			length := 0
 
-			for format_token, i in line.format_tokens {
+			for format_token in line.format_tokens {
 				if format_token.kind == .Comment {
 					current_info.length = max(current_info.length, length)
 					current_info.end = line_index

core/odin/printer/visit.odin

@@ -445,7 +445,7 @@ visit_decl :: proc(p: ^Printer, decl: ^ast.Decl, called_in_stmt := false) {
 
 		for value in v.values {
 			#partial switch a in value.derived {
-			case ^ast.Union_Type, ^ast.Enum_Type, ^ast.Struct_Type:
+			case ^ast.Union_Type, ^ast.Enum_Type, ^ast.Struct_Type, ^ast.Bit_Field_Type:
 				add_semicolon = false || called_in_stmt
 			case ^ast.Proc_Lit:
 				add_semicolon = false
@@ -488,6 +488,37 @@ visit_exprs :: proc(p: ^Printer, list: []^ast.Expr, options := List_Options{}) {
 	}
 }
 
+@(private)
+visit_bit_field_fields :: proc(p: ^Printer, list: []^ast.Bit_Field_Field, options := List_Options{}) {
+	if len(list) == 0 {
+		return
+	}
+
+	// we have to newline the expressions to respect the source
+	for v, i in list {
+		// Don't move the first expression, it looks bad
+		if i != 0 && .Enforce_Newline in options {
+			newline_position(p, 1)
+		} else if i != 0 {
+			move_line_limit(p, v.pos, 1)
+		}
+
+		visit_expr(p, v.name, options)
+		push_generic_token(p, .Colon, 0)
+		visit_expr(p, v.type, options)
+		push_generic_token(p, .Or, 1)
+		visit_expr(p, v.bit_size, options)
+
+		if (i != len(list) - 1 || .Trailing in options) && .Add_Comma in options {
+			push_generic_token(p, .Comma, 0)
+		}
+	}
+
+	if len(list) > 1 && .Enforce_Newline in options {
+		newline_position(p, 1)
+	}
+}
+
 @(private)
 visit_attributes :: proc(p: ^Printer, attributes: [dynamic]^ast.Attribute) {
 	if len(attributes) == 0 {
@@ -1293,6 +1324,25 @@ visit_expr :: proc(p: ^Printer, expr: ^ast.Expr, options := List_Options{}) {
 		visit_expr(p, v.column_count)
 		push_generic_token(p, .Close_Bracket, 0)
 		visit_expr(p, v.elem)
+	case ^ast.Bit_Field_Type:
+		push_generic_token(p, .Bit_Field, 1)
+
+		visit_expr(p, v.backing_type)
+
+		if len(v.fields) == 0 || v.pos.line == v.close.line {
+			push_generic_token(p, .Open_Brace, 1)
+			visit_bit_field_fields(p, v.fields, {.Add_Comma})
+			push_generic_token(p, .Close_Brace, 0)
+		} else {
+			visit_begin_brace(p, v.pos, .Generic, len(v.fields))
+			newline_position(p, 1)
+			set_source_position(p, v.fields[0].pos)
+			visit_bit_field_fields(p, v.fields, {.Add_Comma, .Trailing, .Enforce_Newline})
+			set_source_position(p, v.close)
+			visit_end_brace(p, v.close)
+		}
+
+		set_source_position(p, v.close)
 	case:
 		panic(fmt.aprint(expr.derived))
 	}
@@ -1462,9 +1512,9 @@ visit_binary_expr :: proc(p: ^Printer, binary: ^ast.Binary_Expr) {
 	}
 
 	either_implicit_selector := false
-	if _, ok := binary.left.derived.(^ast.Implicit_Selector_Expr); ok {
+	if _, lok := binary.left.derived.(^ast.Implicit_Selector_Expr); lok {
 		either_implicit_selector = true
-	} else if _, ok := binary.right.derived.(^ast.Implicit_Selector_Expr); ok {
+	} else if _, rok := binary.right.derived.(^ast.Implicit_Selector_Expr); rok {
 		either_implicit_selector = true
 	}
 

core/odin/tokenizer/tokenizer.odin

@@ -39,6 +39,7 @@ init :: proc(t: ^Tokenizer, src: string, path: string, err: Error_Handler = defa
 	t.read_offset = 0
 	t.line_offset = 0
 	t.line_count = len(src) > 0 ? 1 : 0
+	t.insert_semicolon = false
 	t.error_count = 0
 	t.path = path
 

core/os/os2/errors.odin

@@ -13,6 +13,12 @@ General_Error :: enum u32 {
 
 	Timeout,
 
+	Broken_Pipe,
+
+	// Indicates that an attempt to retrieve a file's size was made, but the
+	// file doesn't have a size.
+	No_Size,
+
 	Invalid_File,
 	Invalid_Dir,
 	Invalid_Path,
@@ -51,6 +57,8 @@ error_string :: proc(ferr: Error) -> string {
 		case .Not_Exist:         return "file does not exist"
 		case .Closed:            return "file already closed"
 		case .Timeout:           return "i/o timeout"
+		case .Broken_Pipe:       return "Broken pipe"
+		case .No_Size:           return "file has no definite size"
 		case .Invalid_File:      return "invalid file"
 		case .Invalid_Dir:       return "invalid directory"
 		case .Invalid_Path:      return "invalid path"

core/os/os2/file.odin

@@ -6,6 +6,7 @@ import "base:runtime"
 
 File :: struct {
 	impl: _File,
+	stream: io.Stream,
 }
 
 File_Mode :: distinct u32
@@ -72,56 +73,56 @@ name :: proc(f: ^File) -> string {
 
 close :: proc(f: ^File) -> Error {
 	if f != nil {
-		return io.close(f.impl.stream)
+		return io.close(f.stream)
 	}
 	return nil
 }
 
 seek :: proc(f: ^File, offset: i64, whence: io.Seek_From) -> (ret: i64, err: Error) {
 	if f != nil {
-		return io.seek(f.impl.stream, offset, whence)
+		return io.seek(f.stream, offset, whence)
 	}
 	return 0, .Invalid_File
 }
 
 read :: proc(f: ^File, p: []byte) -> (n: int, err: Error) {
 	if f != nil {
-		return io.read(f.impl.stream, p)
+		return io.read(f.stream, p)
 	}
 	return 0, .Invalid_File
 }
 
 read_at :: proc(f: ^File, p: []byte, offset: i64) -> (n: int, err: Error) {
 	if f != nil {
-		return io.read_at(f.impl.stream, p, offset)
+		return io.read_at(f.stream, p, offset)
 	}
 	return 0, .Invalid_File
 }
 
 write :: proc(f: ^File, p: []byte) -> (n: int, err: Error) {
 	if f != nil {
-		return io.write(f.impl.stream, p)
+		return io.write(f.stream, p)
 	}
 	return 0, .Invalid_File
 }
 
 write_at :: proc(f: ^File, p: []byte, offset: i64) -> (n: int, err: Error) {
 	if f != nil {
-		return io.write_at(f.impl.stream, p, offset)
+		return io.write_at(f.stream, p, offset)
 	}
 	return 0, .Invalid_File
 }
 
 file_size :: proc(f: ^File) -> (n: i64, err: Error) {
 	if f != nil {
-		return io.size(f.impl.stream)
+		return io.size(f.stream)
 	}
 	return 0, .Invalid_File
 }
 
 flush :: proc(f: ^File) -> Error {
 	if f != nil {
-		return io.flush(f.impl.stream)
+		return io.flush(f.stream)
 	}
 	return nil
 }

core/os/os2/file_linux.odin

@@ -33,8 +33,6 @@ _File :: struct {
 	name: string,
 	fd: int,
 	allocator: runtime.Allocator,
-
-	stream: io.Stream,
 }
 
 _file_allocator :: proc() -> runtime.Allocator {
@@ -75,7 +73,7 @@ _new_file :: proc(fd: uintptr, _: string) -> ^File {
 	file.impl.fd = int(fd)
 	file.impl.allocator = _file_allocator()
 	file.impl.name = _get_full_path(file.impl.fd, file.impl.allocator)
-	file.impl.stream = {
+	file.stream = {
 		data = file,
 		procedure = _file_stream_proc,
 	}