Enable core:fmt tests.

Merge pull request #3506 from IllusionMan1212/windows-hid
core/sys/windows: added various procedures and constants for the hid library
2026-07-05 11:11:37 -07:00 · 2024-05-10 17:05:18 +02:00 · 2024-05-10 16:03:13 +01:00 · 2024-05-10 16:01:16 +01:00 · 2024-05-10 15:59:14 +01:00 · 2024-05-10 15:56:26 +01:00
957 changed files with 157775 additions and 73659 deletions
@@ -0,0 +1,10 @@
+root = true
+
+[*]
+indent_style = tab
+end_of_line = lf
+
+[*.y{,a}ml]
+indent_size = 2
+indent_style = space
+tab_width = 2
@@ -1 +1,6 @@
 *.odin linguist-language=Odin
+* text=auto
+
+# These files must always have *nix line-endings
+Makefile text eol=lf
+*.sh text eol=lf
@@ -3,11 +3,16 @@ on: [push, pull_request, workflow_dispatch]

 jobs:
  build_linux:
+    name: Ubuntu Build, Check, and Test
    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
@@ -46,6 +51,9 @@ jobs:
      - name: Odin check examples/all for Linux i386
        run: ./odin check examples/all -vet -strict-style -target:linux_i386
        timeout-minutes: 10
+      - name: Odin check examples/all for Linux arm64
+        run: ./odin check examples/all -vet -strict-style -target:linux_arm64
+        timeout-minutes: 10
      - name: Odin check examples/all for FreeBSD amd64
        run: ./odin check examples/all -vet -strict-style -target:freebsd_amd64
        timeout-minutes: 10
@@ -53,15 +61,14 @@ jobs:
        run: ./odin check examples/all -vet -strict-style -target:openbsd_amd64
        timeout-minutes: 10
  build_macOS:
-    runs-on: macos-latest
+    name: MacOS Build, Check, and Test
+    runs-on: macos-13
    steps:
      - uses: actions/checkout@v1
-      - name: Download LLVM, botan and setup PATH
+      - name: Download LLVM, and setup PATH
        run: |
-          brew install llvm@11 botan
-          echo "/usr/local/opt/llvm@11/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
@@ -87,9 +94,43 @@ jobs:
          cd tests/core
          make
        timeout-minutes: 10
-      - name: Vendor library tests
+      - name: Odin internals tests
        run: |
-          cd tests/vendor
+          cd tests/internal
+          make
+        timeout-minutes: 10
+  build_macOS_arm:
+    name: MacOS ARM Build, Check, and Test
+    runs-on: macos-14 # This is an arm/m1 runner.
+    steps:
+      - uses: actions/checkout@v1
+      - name: Download LLVM and setup PATH
+        run: |
+          brew install llvm@17
+          echo "/opt/homebrew/opt/llvm@17/bin" >> $GITHUB_PATH
+      - name: build odin
+        run: ./build_odin.sh release
+      - name: Odin version
+        run: ./odin version
+        timeout-minutes: 1
+      - name: Odin report
+        run: ./odin report
+        timeout-minutes: 1
+      - name: Odin check
+        run: ./odin check examples/demo -vet
+        timeout-minutes: 10
+      - name: Odin run
+        run: ./odin run examples/demo
+        timeout-minutes: 10
+      - name: Odin run -debug
+        run: ./odin run examples/demo -debug
+        timeout-minutes: 10
+      - name: Odin check examples/all
+        run: ./odin check examples/all -strict-style
+        timeout-minutes: 10
+      - name: Core library tests
+        run: |
+          cd tests/core
          make
        timeout-minutes: 10
      - name: Odin internals tests
@@ -97,20 +138,15 @@ jobs:
          cd tests/internal
          make
        timeout-minutes: 10
-      - name: Odin check examples/all for Darwin arm64
-        run: ./odin check examples/all -vet -strict-style -target:darwin_arm64
-        timeout-minutes: 10
-      - name: Odin check examples/all for Linux arm64
-        run: ./odin check examples/all -vet -strict-style -target:linux_arm64
-        timeout-minutes: 10
  build_windows:
-    runs-on: windows-2019
+    name: Windows Build, Check, and Test
+    runs-on: windows-2022
    steps:
      - uses: actions/checkout@v1
      - name: build Odin
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          ./build.bat 1
      - name: Odin version
        run: ./odin version
@@ -121,65 +157,65 @@ jobs:
      - name: Odin check
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          odin check examples/demo -vet
        timeout-minutes: 10
      - name: Odin run
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          odin run examples/demo
        timeout-minutes: 10
      - name: Odin run -debug
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          odin run examples/demo -debug
        timeout-minutes: 10
      - name: Odin check examples/all
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          odin check examples/all -strict-style
        timeout-minutes: 10
      - name: Core library tests
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          cd tests\core
          call build.bat
        timeout-minutes: 10
      - name: Vendor library tests
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          cd tests\vendor
          call build.bat
        timeout-minutes: 10
      - name: Odin internals tests
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          cd tests\internal
          call build.bat
        timeout-minutes: 10
      - name: Odin documentation tests
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          cd tests\documentation
-          call build.bat
+          rem call build.bat
        timeout-minutes: 10
      - name: core:math/big tests
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          cd tests\core\math\big
          call build.bat
        timeout-minutes: 10
      - name: Odin check examples/all for Windows 32bits
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          odin check examples/all -strict-style -target:windows_i386
        timeout-minutes: 10
@@ -7,26 +7,30 @@ on:

 jobs:
  build_windows:
-    runs-on: windows-2019
+    name: Windows Build
+    if: github.repository == 'odin-lang/Odin'
+    runs-on: windows-2022
    steps:
      - uses: actions/checkout@v1
      - name: build Odin
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          ./build.bat 1 1
      - name: Odin run
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
          odin run examples/demo
      - name: Copy artifacts
        run: |
          rm bin/llvm/windows/LLVM-C.lib
          mkdir dist
          cp odin.exe dist
+          cp LICENSE dist
          cp LLVM-C.dll dist
          cp -r shared dist
+          cp -r base dist
          cp -r core dist
          cp -r vendor dist
          cp -r bin dist
@@ -37,11 +41,17 @@ jobs:
          name: windows_artifacts
          path: dist
  build_ubuntu:
+    name: Ubuntu Build
+    if: github.repository == 'odin-lang/Odin'
    runs-on: ubuntu-latest
    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
@@ -50,46 +60,91 @@ jobs:
        run: |
          mkdir dist
          cp odin dist
+          cp LICENSE dist
          cp libLLVM* dist
          cp -r shared dist
+          cp -r base dist
          cp -r core dist
          cp -r vendor dist
          cp -r examples dist
+          # Zipping so executable permissions are retained, see https://github.com/actions/upload-artifact/issues/38
+          zip -r dist.zip dist
      - name: Upload artifact
        uses: actions/upload-artifact@v1
        with:
          name: ubuntu_artifacts
-          path: dist
+          path: dist.zip
  build_macos:
-    runs-on: macOS-latest
+    name: MacOS Build
+    if: github.repository == 'odin-lang/Odin'
+    runs-on: macos-13
    steps:
      - uses: actions/checkout@v1
      - name: Download LLVM and setup PATH
        run: |
-          brew install llvm@11
-          echo "/usr/local/opt/llvm@11/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
-        run: make nightly
-      - name: Odin run
-        run: ./odin run examples/demo
-      - name: Copy artifacts
+        # 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.
+        run: CXXFLAGS="-L/usr/lib/system -L/usr/lib" make nightly
+      - name: Bundle
        run: |
          mkdir dist
          cp odin dist
+          cp LICENSE dist
          cp -r shared dist
+          cp -r base dist
          cp -r core dist
          cp -r vendor dist
          cp -r examples dist
+          dylibbundler -b -x dist/odin -d dist/libs -od -p @executable_path/libs
+          # Zipping so executable permissions are retained, see https://github.com/actions/upload-artifact/issues/38
+          zip -r dist.zip dist
+      - name: Odin run
+        run: ./dist/odin run examples/demo
      - name: Upload artifact
        uses: actions/upload-artifact@v1
        with:
          name: macos_artifacts
-          path: dist
+          path: dist.zip
+  build_macos_arm:
+    name: MacOS ARM Build
+    if: github.repository == 'odin-lang/Odin'
+    runs-on: macos-14 # ARM machine
+    steps:
+      - uses: actions/checkout@v1
+      - name: Download LLVM and setup PATH
+        run: |
+          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.
+        run: CXXFLAGS="-L/usr/lib/system -L/usr/lib" make nightly
+      - name: Bundle
+        run: |
+          mkdir dist
+          cp odin dist
+          cp LICENSE dist
+          cp -r shared dist
+          cp -r base dist
+          cp -r core dist
+          cp -r vendor dist
+          cp -r examples dist
+          dylibbundler -b -x dist/odin -d dist/libs -od -p @executable_path/libs
+          # Zipping so executable permissions are retained, see https://github.com/actions/upload-artifact/issues/38
+          zip -r dist.zip dist
+      - name: Odin run
+        run: ./dist/odin run examples/demo
+      - name: Upload artifact
+        uses: actions/upload-artifact@v1
+        with:
+          name: macos_arm_artifacts
+          path: dist.zip
  upload_b2:
    runs-on: [ubuntu-latest]
-    needs: [build_windows, build_macos, build_ubuntu]
+    needs: [build_windows, build_macos, build_macos_arm, build_ubuntu]
    steps:
      - uses: actions/checkout@v1
      - uses: actions/setup-python@v2
@@ -120,6 +175,11 @@ jobs:
        with:
          name: macos_artifacts

+      - name: Download macOS arm artifacts
+        uses: actions/download-artifact@v1
+        with:
+          name: macos_arm_artifacts
+
      - name: Create archives and upload
        shell: bash
        env:
@@ -134,8 +194,9 @@ jobs:
          echo Uploading artifcates to B2
          chmod +x ./ci/upload_create_nightly.sh
          ./ci/upload_create_nightly.sh "$BUCKET" windows-amd64 windows_artifacts/
-          ./ci/upload_create_nightly.sh "$BUCKET" ubuntu-amd64 ubuntu_artifacts/
-          ./ci/upload_create_nightly.sh "$BUCKET" macos-amd64 macos_artifacts/
+          ./ci/upload_create_nightly.sh "$BUCKET" ubuntu-amd64 ubuntu_artifacts/dist.zip
+          ./ci/upload_create_nightly.sh "$BUCKET" macos-amd64 macos_artifacts/dist.zip
+          ./ci/upload_create_nightly.sh "$BUCKET" macos-arm64 macos_arm_artifacts/dist.zip

          echo Deleting old artifacts in B2
          python3 ci/delete_old_binaries.py "$BUCKET" "$DAYS_TO_KEEP"
@@ -22,6 +22,40 @@ bld/
 [Oo]bj/
 [Ll]og/
 ![Cc]ore/[Ll]og/
+tests/documentation/verify/
+tests/documentation/all.odin-doc
+tests/internal/test_map
+tests/internal/test_pow
+tests/internal/test_rtti
+tests/core/test_base64
+tests/core/test_cbor
+tests/core/test_core_compress
+tests/core/test_core_container
+tests/core/test_core_filepath
+tests/core/test_core_fmt
+tests/core/test_core_i18n
+tests/core/test_core_image
+tests/core/test_core_libc
+tests/core/test_core_match
+tests/core/test_core_math
+tests/core/test_core_net
+tests/core/test_core_os_exit
+tests/core/test_core_reflect
+tests/core/test_core_strings
+tests/core/test_core_time
+tests/core/test_crypto
+tests/core/test_hash
+tests/core/test_hex
+tests/core/test_hxa
+tests/core/test_json
+tests/core/test_linalg_glsl_math
+tests/core/test_noise
+tests/core/test_varint
+tests/core/test_xml
+tests/core/test_core_slice
+tests/core/test_core_thread
+tests/core/test_core_runtime
+tests/vendor/vendor_botan
 # Visual Studio 2015 cache/options directory
 .vs/
 # Visual Studio Code options directory
@@ -268,6 +302,7 @@ bin/

 # - Linux/MacOS
 odin
+!odin/
 odin.dSYM
 *.bin
 demo.bin
@@ -284,4 +319,9 @@ shared/
 *.sublime-workspace
 examples/bug/
 build.sh
-!core/debug/
+!core/debug/
+
+# RAD debugger project file
+*.raddbg
+
+misc/featuregen/featuregen
@@ -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:
@@ -11,7 +11,7 @@
        <img src="https://img.shields.io/badge/platforms-Windows%20|%20Linux%20|%20macOS-green.svg">
    </a>
    <br>
-    <a href="https://discord.gg/odinlang">
+    <a href="https://discord.com/invite/sVBPHEv">
        <img src="https://img.shields.io/discord/568138951836172421?logo=discord">
    </a>
    <a href="https://github.com/odin-lang/odin/actions">
@@ -110,7 +110,7 @@ typeid_of    :: proc($T: typeid) -> typeid ---
 swizzle :: proc(x: [N]T, indices: ..int) -> [len(indices)]T ---

 complex    :: proc(real, imag: Float) -> Complex_Type ---
-quaternion :: proc(real, imag, jmag, kmag: Float) -> Quaternion_Type ---
+quaternion :: proc(imag, jmag, kmag, real: Float) -> Quaternion_Type --- // fields must be named
 real       :: proc(value: Complex_Or_Quaternion) -> Float ---
 imag       :: proc(value: Complex_Or_Quaternion) -> Float ---
 jmag       :: proc(value: Quaternion) -> Float ---
@@ -5,6 +5,12 @@ package intrinsics
 // Package-Related
 is_package_imported :: proc(package_name: string) -> bool ---

+// Matrix Related Procedures
+transpose        :: proc(m: $T/matrix[$R, $C]$E)    -> matrix[C, R]E ---
+outer_product    :: proc(a: $A/[$X]$E, b: $B/[$Y]E) -> matrix[X, Y]E ---
+hadamard_product :: proc(a, b: $T/matrix[$R, $C]$E) -> T ---
+matrix_flatten   :: proc(m: $T/matrix[$R, $C]$E)    -> [R*C]E ---
+
 // Types
 soa_struct :: proc($N: int, $T: typeid) -> type/#soa[N]T

@@ -32,9 +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))) ---

@@ -162,9 +168,19 @@ type_is_matrix           :: proc($T: typeid) -> bool ---
 type_has_nil :: proc($T: typeid) -> bool ---

 type_is_specialization_of :: proc($T, $S: typeid) -> bool ---
-type_is_variant_of :: proc($U, $V: typeid) -> bool where type_is_union(U) ---

-type_has_field :: proc($T: typeid, $name: string) -> bool ---
+type_is_variant_of        :: proc($U, $V: typeid)          -> bool    where type_is_union(U) ---
+type_union_tag_type       :: proc($T: typeid)              -> typeid  where type_is_union(T) ---
+type_union_tag_offset     :: proc($T: typeid)              -> uintptr where type_is_union(T) ---
+type_union_base_tag_value :: proc($T: typeid)              -> int     where type_is_union(U) ---
+type_union_variant_count  :: proc($T: typeid)              -> int     where type_is_union(T) ---
+type_variant_type_of      :: proc($T: typeid, $index: int) -> typeid  where type_is_union(T) ---
+type_variant_index_of     :: proc($U, $V: typeid)          -> int     where type_is_union(U) ---
+
+type_bit_set_elem_type       :: proc($T: typeid) -> typeid where type_is_bit_set(T) ---
+type_bit_set_underlying_type :: proc($T: typeid) -> typeid where type_is_bit_set(T) ---
+
+type_has_field  :: proc($T: typeid, $name: string) -> bool ---
 type_field_type :: proc($T: typeid, $name: string) -> typeid ---

 type_proc_parameter_count :: proc($T: typeid) -> int where type_is_proc(T) ---
@@ -192,6 +208,7 @@ type_map_info      :: proc($T: typeid/map[$K]$V) -> ^runtime.Map_Info ---
 type_map_cell_info :: proc($T: typeid)           -> ^runtime.Map_Cell_Info ---

 type_convert_variants_to_pointers :: proc($T: typeid) -> typeid where type_is_union(T) ---
+type_merge :: proc($U, $V: typeid) -> typeid where type_is_union(U), type_is_union(V) ---

 constant_utf16_cstring :: proc($literal: string) -> [^]u16 ---

@@ -214,10 +231,10 @@ simd_shr_masked :: proc(a: #simd[N]T, b: #simd[N]Unsigned_Integer) -> #simd[N]T
 simd_add_sat :: proc(a, b: #simd[N]T) -> #simd[N]T ---
 simd_sub_sat :: proc(a, b: #simd[N]T) -> #simd[N]T ---

-simd_and     :: proc(a, b: #simd[N]T) -> #simd[N]T ---
-simd_or      :: proc(a, b: #simd[N]T) -> #simd[N]T ---
-simd_xor     :: proc(a, b: #simd[N]T) -> #simd[N]T ---
-simd_and_not :: proc(a, b: #simd[N]T) -> #simd[N]T ---
+simd_bit_and     :: proc(a, b: #simd[N]T) -> #simd[N]T ---
+simd_bit_or      :: proc(a, b: #simd[N]T) -> #simd[N]T ---
+simd_bit_xor     :: proc(a, b: #simd[N]T) -> #simd[N]T ---
+simd_bit_and_not :: proc(a, b: #simd[N]T) -> #simd[N]T ---

 simd_neg  :: proc(a: #simd[N]T) -> #simd[N]T ---

@@ -268,6 +285,12 @@ simd_reverse :: proc(a: #simd[N]T) -> #simd[N]T ---
 simd_rotate_left  :: proc(a: #simd[N]T, $offset: int) -> #simd[N]T ---
 simd_rotate_right :: proc(a: #simd[N]T, $offset: int) -> #simd[N]T ---

+// Checks if the current target supports the given target features.
+//
+// Takes a constant comma-seperated string (eg: "sha512,sse4.1"), or a procedure type which has either
+// `@(require_target_feature)` or `@(enable_target_feature)` as its input and returns a boolean indicating
+// if all listed features are supported.
+has_target_feature :: proc($test: $T) -> bool where type_is_string(T) || type_is_proc(T) ---

 // WASM targets only
 wasm_memory_grow :: proc(index, delta: uintptr) -> int ---
@@ -279,7 +302,9 @@ 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
+@(require_target_feature="atomics")
 wasm_memory_atomic_wait32   :: proc(ptr: ^u32, expected: u32, timeout_ns: i64) -> u32 ---
+@(require_target_feature="atomics")
 wasm_memory_atomic_notify32 :: proc(ptr: ^u32, waiters: u32) -> (waiters_woken_up: u32) ---

 // x86 Targets (i386, amd64)
@@ -18,9 +18,10 @@
 // This could change at a later date if the all these data structures are
 // implemented within the compiler rather than in this "preload" file
 //
+//+no-instrumentation
 package runtime

-import "core:intrinsics"
+import "base:intrinsics"

 // NOTE(bill): This must match the compiler's
 Calling_Convention :: enum u8 {
@@ -119,6 +120,7 @@ Type_Info_Struct :: struct {
 	tags:         []string,
 	is_packed:    bool,
 	is_raw_union: bool,
+	is_no_copy:   bool,
 	custom_align: bool,

 	equal: Equal_Proc, // set only when the struct has .Comparable set but does not have .Simple_Compare set
@@ -161,11 +163,11 @@ Type_Info_Simd_Vector :: struct {
 	count:      int,
 }
 Type_Info_Relative_Pointer :: struct {
-	pointer:      ^Type_Info,
+	pointer:      ^Type_Info, // ^T
 	base_integer: ^Type_Info,
 }
-Type_Info_Relative_Slice :: struct {
-	slice:        ^Type_Info,
+Type_Info_Relative_Multi_Pointer :: struct {
+	pointer:      ^Type_Info, // [^]T
 	base_integer: ^Type_Info,
 }
 Type_Info_Matrix :: struct {
@@ -175,10 +177,22 @@ 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,
 }
+Type_Info_Bit_Field :: struct {
+	backing_type: ^Type_Info,
+	names:        []string,
+	types:        []^Type_Info,
+	bit_sizes:    []uintptr,
+	bit_offsets:  []uintptr,
+	tags:         []string,
+}

 Type_Info_Flag :: enum u8 {
 	Comparable     = 0,
@@ -218,9 +232,10 @@ Type_Info :: struct {
 		Type_Info_Bit_Set,
 		Type_Info_Simd_Vector,
 		Type_Info_Relative_Pointer,
-		Type_Info_Relative_Slice,
+		Type_Info_Relative_Multi_Pointer,
 		Type_Info_Matrix,
 		Type_Info_Soa_Pointer,
+		Type_Info_Bit_Field,
 	},
 }

@@ -251,23 +266,25 @@ Typeid_Kind :: enum u8 {
 	Bit_Set,
 	Simd_Vector,
 	Relative_Pointer,
-	Relative_Slice,
+	Relative_Multi_Pointer,
 	Matrix,
+	Soa_Pointer,
+	Bit_Field,
 }
 #assert(len(Typeid_Kind) < 32)

-// Typeid_Bit_Field :: bit_field #align align_of(uintptr) {
-// 	index:    8*size_of(uintptr) - 8,
-// 	kind:     5, // Typeid_Kind
-// 	named:    1,
-// 	special:  1, // signed, cstring, etc
-// 	reserved: 1,
-// }
-// #assert(size_of(Typeid_Bit_Field) == size_of(uintptr));
+Typeid_Bit_Field :: bit_field uintptr {
+	index:    uintptr     | 8*size_of(uintptr) - 8,
+	kind:     Typeid_Kind | 5, // Typeid_Kind
+	named:    bool        | 1,
+	special:  bool        | 1, // signed, cstring, etc
+	reserved: bool        | 1,
+}
+#assert(size_of(Typeid_Bit_Field) == size_of(uintptr))

 // NOTE(bill): only the ones that are needed (not all types)
 // This will be set by the compiler
-type_table: []Type_Info
+type_table: []^Type_Info

 args__: []cstring

@@ -293,6 +310,14 @@ Source_Code_Location :: struct {
 	procedure:    string,
 }

+/*
+	Used by the built-in directory `#load_directory(path: string) -> []Load_Directory_File`
+*/
+Load_Directory_File :: struct {
+	name: string,
+	data: []byte, // immutable data
+}
+
 Assertion_Failure_Proc :: #type proc(prefix, message: string, loc: Source_Code_Location) -> !

 // Allocation Stuff
@@ -304,6 +329,7 @@ Allocator_Mode :: enum byte {
 	Query_Features,
 	Query_Info,
 	Alloc_Non_Zeroed,
+	Resize_Non_Zeroed,
 }

 Allocator_Mode_Set :: distinct bit_set[Allocator_Mode]
@@ -336,6 +362,8 @@ Kilobyte :: 1024 * Byte
 Megabyte :: 1024 * Kilobyte
 Gigabyte :: 1024 * Megabyte
 Terabyte :: 1024 * Gigabyte
+Petabyte :: 1024 * Terabyte
+Exabyte  :: 1024 * Petabyte

 // Logging stuff

@@ -424,7 +452,7 @@ Raw_Map :: struct {
 	// Map_Hash directly, though for consistency sake it's written as if it were
 	// an array of Map_Cell(Map_Hash).
 	data:      uintptr,   // 8-bytes on 64-bits, 4-bytes on 32-bits
-	len:       int,       // 8-bytes on 64-bits, 4-bytes on 32-bits
+	len:       uintptr,   // 8-bytes on 64-bits, 4-bytes on 32-bits
 	allocator: Allocator, // 16-bytes on 64-bits, 8-bytes on 32-bits
 }

@@ -470,7 +498,7 @@ Odin_OS_Type :: type_of(ODIN_OS)
 		arm32,
 		arm64,
 		wasm32,
-		wasm64,
+		wasm64p32,
 	}
 */
 Odin_Arch_Type :: type_of(ODIN_ARCH)
@@ -498,6 +526,29 @@ Odin_Build_Mode_Type :: type_of(ODIN_BUILD_MODE)
 Odin_Endian_Type :: type_of(ODIN_ENDIAN)


+/*
+	// Defined internally by the compiler
+	Odin_Platform_Subtarget_Type :: enum int {
+		Default,
+		iOS,
+	}
+*/
+Odin_Platform_Subtarget_Type :: type_of(ODIN_PLATFORM_SUBTARGET)
+
+/*
+	// Defined internally by the compiler
+	Odin_Sanitizer_Flag :: enum u32 {
+		Address = 0,
+		Memory  = 1,
+		Thread  = 2,
+	}
+	Odin_Sanitizer_Flags :: distinct bitset[Odin_Sanitizer_Flag; u32]
+
+	ODIN_SANITIZER_FLAGS // is a constant
+*/
+Odin_Sanitizer_Flags :: type_of(ODIN_SANITIZER_FLAGS)
+
+
 /////////////////////////////
 // Init Startup Procedures //
 /////////////////////////////
@@ -546,8 +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
 		}
 	}
@@ -562,10 +614,10 @@ __type_info_of :: proc "contextless" (id: typeid) -> ^Type_Info #no_bounds_check
 	if n < 0 || n >= len(type_table) {
 		n = 0
 	}
-	return &type_table[n]
+	return type_table[n]
 }

-when !ODIN_DISALLOW_RTTI {
+when !ODIN_NO_RTTI {
 	typeid_base :: proc "contextless" (id: typeid) -> typeid {
 		ti := type_info_of(id)
 		ti = type_info_base(ti)
@@ -637,8 +689,10 @@ default_assertion_failure_proc :: proc(prefix, message: string, loc: Source_Code
 	when ODIN_OS == .Freestanding {
 		// Do nothing
 	} else {
-		print_caller_location(loc)
-		print_string(" ")
+		when !ODIN_DISABLE_ASSERT {
+			print_caller_location(loc)
+			print_string(" ")
+		}
 		print_string(prefix)
 		if len(message) > 0 {
 			print_string(": ")
@@ -0,0 +1,902 @@
+package runtime
+
+import "base:intrinsics"
+
+@builtin
+Maybe :: union($T: typeid) {T}
+
+
+@(builtin, require_results)
+container_of :: #force_inline proc "contextless" (ptr: $P/^$Field_Type, $T: typeid, $field_name: string) -> ^T
+	where intrinsics.type_has_field(T, field_name),
+	      intrinsics.type_field_type(T, field_name) == Field_Type {
+	offset :: offset_of_by_string(T, field_name)
+	return (^T)(uintptr(ptr) - offset) if ptr != nil else nil
+}
+
+
+when !NO_DEFAULT_TEMP_ALLOCATOR {
+	@thread_local global_default_temp_allocator_data: Default_Temp_Allocator
+}
+
+@(builtin, disabled=NO_DEFAULT_TEMP_ALLOCATOR)
+init_global_temporary_allocator :: proc(size: int, backup_allocator := context.allocator) {
+	when !NO_DEFAULT_TEMP_ALLOCATOR {
+		default_temp_allocator_init(&global_default_temp_allocator_data, size, backup_allocator)
+	}
+}
+
+
+// `copy_slice` is a built-in procedure that copies elements from a source slice `src` to a destination slice `dst`.
+// The source and destination may overlap. Copy returns the number of elements copied, which will be the minimum
+// of len(src) and len(dst).
+//
+// Prefer the procedure group `copy`.
+@builtin
+copy_slice :: proc "contextless" (dst, src: $T/[]$E) -> int {
+	n := max(0, min(len(dst), len(src)))
+	if n > 0 {
+		intrinsics.mem_copy(raw_data(dst), raw_data(src), n*size_of(E))
+	}
+	return n
+}
+// `copy_from_string` is a built-in procedure that copies elements from a source string `src` to a destination slice `dst`.
+// The source and destination may overlap. Copy returns the number of elements copied, which will be the minimum
+// of len(src) and len(dst).
+//
+// Prefer the procedure group `copy`.
+@builtin
+copy_from_string :: proc "contextless" (dst: $T/[]$E/u8, src: $S/string) -> int {
+	n := max(0, min(len(dst), len(src)))
+	if n > 0 {
+		intrinsics.mem_copy(raw_data(dst), raw_data(src), n)
+	}
+	return n
+}
+// `copy` is a built-in procedure that copies elements from a source slice/string `src` to a destination slice `dst`.
+// The source and destination may overlap. Copy returns the number of elements copied, which will be the minimum
+// of len(src) and len(dst).
+@builtin
+copy :: proc{copy_slice, copy_from_string}
+
+
+
+// `unordered_remove` removed the element at the specified `index`. It does so by replacing the current end value
+// with the old value, and reducing the length of the dynamic array by 1.
+//
+// Note: This is an O(1) operation.
+// Note: If you the elements to remain in their order, use `ordered_remove`.
+// Note: If the index is out of bounds, this procedure will panic.
+@builtin
+unordered_remove :: proc(array: ^$D/[dynamic]$T, index: int, loc := #caller_location) #no_bounds_check {
+	bounds_check_error_loc(loc, index, len(array))
+	n := len(array)-1
+	if index != n {
+		array[index] = array[n]
+	}
+	(^Raw_Dynamic_Array)(array).len -= 1
+}
+// `ordered_remove` removed the element at the specified `index` whilst keeping the order of the other elements.
+//
+// Note: This is an O(N) operation.
+// Note: If you the elements do not have to remain in their order, prefer `unordered_remove`.
+// Note: If the index is out of bounds, this procedure will panic.
+@builtin
+ordered_remove :: proc(array: ^$D/[dynamic]$T, index: int, loc := #caller_location) #no_bounds_check {
+	bounds_check_error_loc(loc, index, len(array))
+	if index+1 < len(array) {
+		copy(array[index:], array[index+1:])
+	}
+	(^Raw_Dynamic_Array)(array).len -= 1
+}
+
+// `remove_range` removes a range of elements specified by the range `lo` and `hi`, whilst keeping the order of the other elements.
+//
+// Note: This is an O(N) operation.
+// Note: If the range is out of bounds, this procedure will panic.
+@builtin
+remove_range :: proc(array: ^$D/[dynamic]$T, lo, hi: int, loc := #caller_location) #no_bounds_check {
+	slice_expr_error_lo_hi_loc(loc, lo, hi, len(array))
+	n := max(hi-lo, 0)
+	if n > 0 {
+		if hi != len(array) {
+			copy(array[lo:], array[hi:])
+		}
+		(^Raw_Dynamic_Array)(array).len -= n
+	}
+}
+
+
+// `pop` will remove and return the end value of dynamic array `array` and reduces the length of `array` by 1.
+//
+// Note: If the dynamic array has no elements (`len(array) == 0`), this procedure will panic.
+@builtin
+pop :: proc(array: ^$T/[dynamic]$E, loc := #caller_location) -> (res: E) #no_bounds_check {
+	assert(len(array) > 0, loc=loc)
+	res = array[len(array)-1]
+	(^Raw_Dynamic_Array)(array).len -= 1
+	return res
+}
+
+
+// `pop_safe` trys to remove and return the end value of dynamic array `array` and reduces the length of `array` by 1.
+// If the operation is not possible, it will return false.
+@builtin
+pop_safe :: proc "contextless" (array: ^$T/[dynamic]$E) -> (res: E, ok: bool) #no_bounds_check {
+	if len(array) == 0 {
+		return
+	}
+	res, ok = array[len(array)-1], true
+	(^Raw_Dynamic_Array)(array).len -= 1
+	return
+}
+
+// `pop_front` will remove and return the first value of dynamic array `array` and reduces the length of `array` by 1.
+//
+// Note: If the dynamic array as no elements (`len(array) == 0`), this procedure will panic.
+@builtin
+pop_front :: proc(array: ^$T/[dynamic]$E, loc := #caller_location) -> (res: E) #no_bounds_check {
+	assert(len(array) > 0, loc=loc)
+	res = array[0]
+	if len(array) > 1 {
+		copy(array[0:], array[1:])
+	}
+	(^Raw_Dynamic_Array)(array).len -= 1
+	return res
+}
+
+// `pop_front_safe` trys to return and remove the first value of dynamic array `array` and reduces the length of `array` by 1.
+// If the operation is not possible, it will return false.
+@builtin
+pop_front_safe :: proc "contextless" (array: ^$T/[dynamic]$E) -> (res: E, ok: bool) #no_bounds_check {
+	if len(array) == 0 {
+		return
+	}
+	res, ok = array[0], true
+	if len(array) > 1 {
+		copy(array[0:], array[1:])
+	}
+	(^Raw_Dynamic_Array)(array).len -= 1
+	return
+}
+
+
+// `clear` will set the length of a passed dynamic array or map to `0`
+@builtin
+clear :: proc{clear_dynamic_array, clear_map}
+
+// `reserve` will try to reserve memory of a passed dynamic array or map to the requested element count (setting the `cap`).
+@builtin
+reserve :: proc{reserve_dynamic_array, reserve_map}
+
+@builtin
+non_zero_reserve :: proc{non_zero_reserve_dynamic_array}
+
+// `resize` will try to resize memory of a passed dynamic array to the requested element count (setting the `len`, and possibly `cap`).
+@builtin
+resize :: proc{resize_dynamic_array}
+
+@builtin
+non_zero_resize :: proc{non_zero_resize_dynamic_array}
+
+// Shrinks the capacity of a dynamic array or map down to the current length, or the given capacity.
+@builtin
+shrink :: proc{shrink_dynamic_array, shrink_map}
+
+// `free` will try to free the passed pointer, with the given `allocator` if the allocator supports this operation.
+@builtin
+free :: proc{mem_free}
+
+// `free_all` will try to free/reset all of the memory of the given `allocator` if the allocator supports this operation.
+@builtin
+free_all :: proc{mem_free_all}
+
+
+
+// `delete_string` will try to free the underlying data of the passed string, with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer the procedure group `delete`.
+@builtin
+delete_string :: proc(str: string, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
+	return mem_free_with_size(raw_data(str), len(str), allocator, loc)
+}
+// `delete_cstring` will try to free the underlying data of the passed string, with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer the procedure group `delete`.
+@builtin
+delete_cstring :: proc(str: cstring, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
+	return mem_free((^byte)(str), allocator, loc)
+}
+// `delete_dynamic_array` will try to free the underlying data of the passed dynamic array, with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer the procedure group `delete`.
+@builtin
+delete_dynamic_array :: proc(array: $T/[dynamic]$E, loc := #caller_location) -> Allocator_Error {
+	return mem_free_with_size(raw_data(array), cap(array)*size_of(E), array.allocator, loc)
+}
+// `delete_slice` will try to free the underlying data of the passed sliced, with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer the procedure group `delete`.
+@builtin
+delete_slice :: proc(array: $T/[]$E, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
+	return mem_free_with_size(raw_data(array), len(array)*size_of(E), allocator, loc)
+}
+// `delete_map` will try to free the underlying data of the passed map, with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer the procedure group `delete`.
+@builtin
+delete_map :: proc(m: $T/map[$K]$V, loc := #caller_location) -> Allocator_Error {
+	return map_free_dynamic(transmute(Raw_Map)m, map_info(T), loc)
+}
+
+
+// `delete` will try to free the underlying data of the passed built-in data structure (string, cstring, dynamic array, slice, or map), with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer `delete` over the specific `delete_*` procedures where possible.
+@builtin
+delete :: proc{
+	delete_string,
+	delete_cstring,
+	delete_dynamic_array,
+	delete_slice,
+	delete_map,
+	delete_soa_slice,
+	delete_soa_dynamic_array,
+}
+
+
+// The new built-in procedure allocates memory. The first argument is a type, not a value, and the value
+// return is a pointer to a newly allocated value of that type using the specified allocator, default is context.allocator
+@(builtin, require_results)
+new :: proc($T: typeid, allocator := context.allocator, loc := #caller_location) -> (^T, Allocator_Error) #optional_allocator_error {
+	return new_aligned(T, align_of(T), allocator, loc)
+}
+@(require_results)
+new_aligned :: proc($T: typeid, alignment: int, allocator := context.allocator, loc := #caller_location) -> (t: ^T, err: Allocator_Error) {
+	data := mem_alloc_bytes(size_of(T), alignment, allocator, loc) or_return
+	t = (^T)(raw_data(data))
+	return
+}
+
+@(builtin, require_results)
+new_clone :: proc(data: $T, allocator := context.allocator, loc := #caller_location) -> (t: ^T, err: Allocator_Error) #optional_allocator_error {
+	t_data := mem_alloc_bytes(size_of(T), align_of(T), allocator, loc) or_return
+	t = (^T)(raw_data(t_data))
+	if t != nil {
+		t^ = data
+	}
+	return
+}
+
+DEFAULT_RESERVE_CAPACITY :: 16
+
+@(require_results)
+make_aligned :: proc($T: typeid/[]$E, #any_int len: int, alignment: int, allocator := context.allocator, loc := #caller_location) -> (T, Allocator_Error) #optional_allocator_error {
+	make_slice_error_loc(loc, len)
+	data, err := mem_alloc_bytes(size_of(E)*len, alignment, allocator, loc)
+	if data == nil && size_of(E) != 0 {
+		return nil, err
+	}
+	s := Raw_Slice{raw_data(data), len}
+	return transmute(T)s, err
+}
+
+// `make_slice` allocates and initializes a slice. Like `new`, the first argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(builtin, require_results)
+make_slice :: proc($T: typeid/[]$E, #any_int len: int, allocator := context.allocator, loc := #caller_location) -> (T, Allocator_Error) #optional_allocator_error {
+	return make_aligned(T, len, align_of(E), allocator, loc)
+}
+// `make_dynamic_array` allocates and initializes a dynamic array. Like `new`, the first argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(builtin, require_results)
+make_dynamic_array :: proc($T: typeid/[dynamic]$E, allocator := context.allocator, loc := #caller_location) -> (T, Allocator_Error) #optional_allocator_error {
+	return make_dynamic_array_len_cap(T, 0, DEFAULT_RESERVE_CAPACITY, allocator, loc)
+}
+// `make_dynamic_array_len` allocates and initializes a dynamic array. Like `new`, the first argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(builtin, require_results)
+make_dynamic_array_len :: proc($T: typeid/[dynamic]$E, #any_int len: int, allocator := context.allocator, loc := #caller_location) -> (T, Allocator_Error) #optional_allocator_error {
+	return make_dynamic_array_len_cap(T, len, len, allocator, loc)
+}
+// `make_dynamic_array_len_cap` allocates and initializes a dynamic array. Like `new`, the first argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(builtin, require_results)
+make_dynamic_array_len_cap :: proc($T: typeid/[dynamic]$E, #any_int len: int, #any_int cap: int, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) #optional_allocator_error {
+	make_dynamic_array_error_loc(loc, len, cap)
+	array.allocator = allocator // initialize allocator before just in case it fails to allocate any memory
+	data := mem_alloc_bytes(size_of(E)*cap, align_of(E), allocator, loc) or_return
+	s := Raw_Dynamic_Array{raw_data(data), len, cap, allocator}
+	if data == nil && size_of(E) != 0 {
+		s.len, s.cap = 0, 0
+	}
+	array = transmute(T)s
+	return
+}
+// `make_map` allocates and initializes a dynamic array. Like `new`, the first argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(builtin, require_results)
+make_map :: proc($T: typeid/map[$K]$E, #any_int capacity: int = 1<<MAP_MIN_LOG2_CAPACITY, allocator := context.allocator, loc := #caller_location) -> (m: T, err: Allocator_Error) #optional_allocator_error {
+	make_map_expr_error_loc(loc, capacity)
+	context.allocator = allocator
+
+	err = reserve_map(&m, capacity, loc)
+	return
+}
+// `make_multi_pointer` allocates and initializes a dynamic array. Like `new`, the first 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`.
+@(builtin, require_results)
+make_multi_pointer :: proc($T: typeid/[^]$E, #any_int len: int, allocator := context.allocator, loc := #caller_location) -> (mp: T, err: Allocator_Error) #optional_allocator_error {
+	make_slice_error_loc(loc, len)
+	data := mem_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
+}
+
+
+// `make` built-in procedure allocates and initializes a value of type slice, dynamic array, map, or multi-pointer (only).
+//
+// Similar to `new`, the first argument is a type, not a value. Unlike new, make's return type is the same as the
+// type of its argument, not a pointer to it.
+// Make uses the specified allocator, default is context.allocator.
+@builtin
+make :: proc{
+	make_slice,
+	make_dynamic_array,
+	make_dynamic_array_len,
+	make_dynamic_array_len_cap,
+	make_map,
+	make_multi_pointer,
+}
+
+
+
+// `clear_map` will set the length of a passed map to `0`
+//
+// Note: Prefer the procedure group `clear`
+@builtin
+clear_map :: proc "contextless" (m: ^$T/map[$K]$V) {
+	if m == nil {
+		return
+	}
+	map_clear_dynamic((^Raw_Map)(m), map_info(T))
+}
+
+// `reserve_map` will try to reserve memory of a passed map to the requested element count (setting the `cap`).
+//
+// Note: Prefer the procedure group `reserve`
+@builtin
+reserve_map :: proc(m: ^$T/map[$K]$V, capacity: int, loc := #caller_location) -> Allocator_Error {
+	return __dynamic_map_reserve((^Raw_Map)(m), map_info(T), uint(capacity), loc) if m != nil else nil
+}
+
+// Shrinks the capacity of a map down to the current length.
+//
+// Note: Prefer the procedure group `shrink`
+@builtin
+shrink_map :: proc(m: ^$T/map[$K]$V, loc := #caller_location) -> (did_shrink: bool, err: Allocator_Error) {
+	if m != nil {
+		return map_shrink_dynamic((^Raw_Map)(m), map_info(T), loc)
+	}
+	return
+}
+
+// The delete_key built-in procedure deletes the element with the specified key (m[key]) from the map.
+// If m is nil, or there is no such element, this procedure is a no-op
+@builtin
+delete_key :: proc(m: ^$T/map[$K]$V, key: K) -> (deleted_key: K, deleted_value: V) {
+	if m != nil {
+		key := key
+		old_k, old_v, ok := map_erase_dynamic((^Raw_Map)(m), map_info(T), uintptr(&key))
+		if ok {
+			deleted_key   = (^K)(old_k)^
+			deleted_value = (^V)(old_v)^
+		}
+	}
+	return
+}
+
+_append_elem :: #force_inline proc(array: ^$T/[dynamic]$E, arg: E, should_zero: bool, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	if array == nil {
+		return 0, nil
+	}
+	when size_of(E) == 0 {
+		array := (^Raw_Dynamic_Array)(array)
+		array.len += 1
+		return 1, nil
+	} else {
+		if cap(array) < len(array)+1 {
+			cap := 2 * cap(array) + max(8, 1)
+
+			// do not 'or_return' here as it could be a partial success
+			if should_zero {
+				err = reserve(array, cap, loc)
+			} else {
+				err = non_zero_reserve(array, cap, loc) 
+			}
+		}
+		if cap(array)-len(array) > 0 {
+			a := (^Raw_Dynamic_Array)(array)
+			when size_of(E) != 0 {
+				data := ([^]E)(a.data)
+				assert(data != nil, loc=loc)
+				data[a.len] = arg
+			}
+			a.len += 1
+			return 1, err
+		}
+		return 0, err
+	}
+}
+
+@builtin
+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, #no_broadcast arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_elem(array, arg, false, loc=loc)
+}
+
+_append_elems :: #force_inline proc(array: ^$T/[dynamic]$E, should_zero: bool, loc := #caller_location, args: ..E) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	if array == nil {
+		return 0, nil
+	}
+
+	arg_len := len(args)
+	if arg_len <= 0 {
+		return 0, nil
+	}
+
+	when size_of(E) == 0 {
+		array := (^Raw_Dynamic_Array)(array)
+		array.len += arg_len
+		return arg_len, nil
+	} else {
+		if cap(array) < len(array)+arg_len {
+			cap := 2 * cap(array) + max(8, arg_len)
+
+			// do not 'or_return' here as it could be a partial success
+			if should_zero {
+				err = reserve(array, cap, loc)
+			} else {
+				err = non_zero_reserve(array, cap, loc)
+			}
+		}
+		arg_len = min(cap(array)-len(array), arg_len)
+		if arg_len > 0 {
+			a := (^Raw_Dynamic_Array)(array)
+			when size_of(E) != 0 {
+				data := ([^]E)(a.data)
+				assert(data != nil, loc=loc)
+				intrinsics.mem_copy(&data[a.len], raw_data(args), size_of(E) * arg_len)
+			}
+			a.len += arg_len
+		}
+		return arg_len, err
+	}
+}
+
+@builtin
+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, #no_broadcast args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_elems(array, false, loc, ..args)
+}
+
+// The append_string built-in procedure appends a string to the end of a [dynamic]u8 like type
+_append_elem_string :: proc(array: ^$T/[dynamic]$E/u8, arg: $A/string, should_zero: bool, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	args := transmute([]E)arg
+	if should_zero { 
+		return append_elems(array, ..args, loc=loc)
+	} else {
+		return non_zero_append_elems(array, ..args, loc=loc)
+	}
+}
+
+@builtin
+append_elem_string :: proc(array: ^$T/[dynamic]$E/u8, arg: $A/string, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_elem_string(array, arg, true, loc)
+}
+@builtin
+non_zero_append_elem_string :: proc(array: ^$T/[dynamic]$E/u8, arg: $A/string, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_elem_string(array, arg, false, loc)
+}
+
+
+// The append_string built-in procedure appends multiple strings to the end of a [dynamic]u8 like type
+@builtin
+append_string :: proc(array: ^$T/[dynamic]$E/u8, args: ..string, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	n_arg: int
+	for arg in args {
+		n_arg, err = append(array, ..transmute([]E)(arg), loc=loc)
+		n += n_arg
+		if err != nil {
+			return
+		}
+	}
+	return
+}
+
+// The append built-in procedure appends elements to the end of a dynamic array
+@builtin append :: proc{append_elem, append_elems, append_elem_string}
+@builtin non_zero_append :: proc{non_zero_append_elem, non_zero_append_elems, non_zero_append_elem_string}
+
+
+@builtin
+append_nothing :: proc(array: ^$T/[dynamic]$E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	if array == nil {
+		return 0, nil
+	}
+	prev_len := len(array)
+	resize(array, len(array)+1, loc) or_return
+	return len(array)-prev_len, nil
+}
+
+
+@builtin
+inject_at_elem :: proc(array: ^$T/[dynamic]$E, index: int, #no_broadcast arg: E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	if array == nil {
+		return
+	}
+	n := max(len(array), index)
+	m :: 1
+	new_size := n + m
+
+	resize(array, new_size, loc) or_return
+	when size_of(E) != 0 {
+		copy(array[index + m:], array[index:])
+		array[index] = arg
+	}
+	ok = true
+	return
+}
+
+@builtin
+inject_at_elems :: proc(array: ^$T/[dynamic]$E, index: int, #no_broadcast args: ..E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	if array == nil {
+		return
+	}
+	if len(args) == 0 {
+		ok = true
+		return
+	}
+
+	n := max(len(array), index)
+	m := len(args)
+	new_size := n + m
+
+	resize(array, new_size, loc) or_return
+	when size_of(E) != 0 {
+		copy(array[index + m:], array[index:])
+		copy(array[index:], args)
+	}
+	ok = true
+	return
+}
+
+@builtin
+inject_at_elem_string :: proc(array: ^$T/[dynamic]$E/u8, index: int, arg: string, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	if array == nil {
+		return
+	}
+	if len(arg) == 0 {
+		ok = true
+		return
+	}
+
+	n := max(len(array), index)
+	m := len(arg)
+	new_size := n + m
+
+	resize(array, new_size, loc) or_return
+	copy(array[index+m:], array[index:])
+	copy(array[index:], arg)
+	ok = true
+	return
+}
+
+@builtin inject_at :: proc{inject_at_elem, inject_at_elems, inject_at_elem_string}
+
+
+
+@builtin
+assign_at_elem :: proc(array: ^$T/[dynamic]$E, index: int, arg: E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	if index < len(array) {
+		array[index] = arg
+		ok = true
+	} else {
+		resize(array, index+1, loc) or_return
+		array[index] = arg
+		ok = true
+	}
+	return
+}
+
+
+@builtin
+assign_at_elems :: proc(array: ^$T/[dynamic]$E, index: int, args: ..E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	new_size := index + len(args)
+	if len(args) == 0 {
+		ok = true
+	} else if new_size < len(array) {
+		copy(array[index:], args)
+		ok = true
+	} else {
+		resize(array, new_size, loc) or_return
+		copy(array[index:], args)
+		ok = true
+	}
+	return
+}
+
+
+@builtin
+assign_at_elem_string :: proc(array: ^$T/[dynamic]$E/u8, index: int, arg: string, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	new_size := index + len(arg)
+	if len(arg) == 0 {
+		ok = true
+	} else if new_size < len(array) {
+		copy(array[index:], arg)
+		ok = true
+	} else {
+		resize(array, new_size, loc) or_return
+		copy(array[index:], arg)
+		ok = true
+	}
+	return
+}
+
+@builtin assign_at :: proc{assign_at_elem, assign_at_elems, assign_at_elem_string}
+
+
+
+
+// `clear_dynamic_array` will set the length of a passed dynamic array to `0`
+//
+// Note: Prefer the procedure group `clear`.
+@builtin
+clear_dynamic_array :: proc "contextless" (array: ^$T/[dynamic]$E) {
+	if array != nil {
+		(^Raw_Dynamic_Array)(array).len = 0
+	}
+}
+
+// `reserve_dynamic_array` will try to reserve memory of a passed dynamic array or map to the requested element count (setting the `cap`).
+//
+// Note: Prefer the procedure group `reserve`.
+_reserve_dynamic_array :: #force_inline proc(array: ^$T/[dynamic]$E, capacity: int, should_zero: bool, loc := #caller_location) -> Allocator_Error {
+	if array == nil {
+		return nil
+	}
+	a := (^Raw_Dynamic_Array)(array)
+
+	if capacity <= a.cap {
+		return nil
+	}
+
+	if a.allocator.procedure == nil {
+		a.allocator = context.allocator
+	}
+	assert(a.allocator.procedure != nil)
+
+	old_size  := a.cap * size_of(E)
+	new_size  := capacity * size_of(E)
+	allocator := a.allocator
+
+	new_data: []byte
+	if should_zero {
+		new_data = mem_resize(a.data, old_size, new_size, align_of(E), allocator, loc) or_return
+	} else {
+		new_data = non_zero_mem_resize(a.data, old_size, new_size, align_of(E), allocator, loc) or_return
+	}
+	if new_data == nil && new_size > 0 {
+		return .Out_Of_Memory
+	}
+
+	a.data = raw_data(new_data)
+	a.cap = capacity
+	return nil
+}
+
+@builtin
+reserve_dynamic_array :: proc(array: ^$T/[dynamic]$E, capacity: int, loc := #caller_location) -> Allocator_Error {
+	return _reserve_dynamic_array(array, capacity, true, loc)
+}
+
+@builtin
+non_zero_reserve_dynamic_array :: proc(array: ^$T/[dynamic]$E, capacity: int, loc := #caller_location) -> Allocator_Error {
+	return _reserve_dynamic_array(array, capacity, false, loc)
+}
+
+// `resize_dynamic_array` will try to resize memory of a passed dynamic array or map to the requested element count (setting the `len`, and possibly `cap`).
+//
+// Note: Prefer the procedure group `resize`
+_resize_dynamic_array :: #force_inline proc(array: ^$T/[dynamic]$E, length: int, should_zero: bool, loc := #caller_location) -> Allocator_Error {
+	if array == nil {
+		return nil
+	}
+	a := (^Raw_Dynamic_Array)(array)
+
+	if length <= a.cap {
+		if should_zero && a.len < length {
+			intrinsics.mem_zero(([^]E)(a.data)[a.len:], (length-a.len)*size_of(E))
+		}
+		a.len = max(length, 0)
+		return nil
+	}
+
+	if a.allocator.procedure == nil {
+		a.allocator = context.allocator
+	}
+	assert(a.allocator.procedure != nil)
+
+	old_size  := a.cap * size_of(E)
+	new_size  := length * size_of(E)
+	allocator := a.allocator
+
+	new_data : []byte
+	if should_zero {
+		new_data = mem_resize(a.data, old_size, new_size, align_of(E), allocator, loc) or_return
+	} else {
+		new_data = non_zero_mem_resize(a.data, old_size, new_size, align_of(E), allocator, loc) or_return
+	}
+	if new_data == nil && new_size > 0 {
+		return .Out_Of_Memory
+	}
+
+	a.data = raw_data(new_data)
+	a.len = length
+	a.cap = length
+	return nil
+}
+
+@builtin
+resize_dynamic_array :: proc(array: ^$T/[dynamic]$E, length: int, loc := #caller_location) -> Allocator_Error {
+	return _resize_dynamic_array(array, length, true, loc=loc)
+}
+
+@builtin
+non_zero_resize_dynamic_array :: proc(array: ^$T/[dynamic]$E, length: int, loc := #caller_location) -> Allocator_Error {
+	return _resize_dynamic_array(array, length, false, loc=loc)
+}
+
+/*
+	Shrinks the capacity of a dynamic array down to the current length, or the given capacity.
+
+	If `new_cap` is negative, then `len(array)` is used.
+
+	Returns false if `cap(array) < new_cap`, or the allocator report failure.
+
+	If `len(array) < new_cap`, then `len(array)` will be left unchanged.
+
+	Note: Prefer the procedure group `shrink`
+*/
+shrink_dynamic_array :: proc(array: ^$T/[dynamic]$E, new_cap := -1, loc := #caller_location) -> (did_shrink: bool, err: Allocator_Error) {
+	if array == nil {
+		return
+	}
+	a := (^Raw_Dynamic_Array)(array)
+
+	new_cap := new_cap if new_cap >= 0 else a.len
+
+	if new_cap > a.cap {
+		return
+	}
+
+	if a.allocator.procedure == nil {
+		a.allocator = context.allocator
+	}
+	assert(a.allocator.procedure != nil)
+
+	old_size := a.cap * size_of(E)
+	new_size := new_cap * size_of(E)
+
+	new_data := mem_resize(a.data, old_size, new_size, align_of(E), a.allocator, loc) or_return
+
+	a.data = raw_data(new_data)
+	a.len = min(new_cap, a.len)
+	a.cap = new_cap
+	return true, nil
+}
+
+@builtin
+map_insert :: proc(m: ^$T/map[$K]$V, key: K, value: V, loc := #caller_location) -> (ptr: ^V) {
+	key, value := key, value
+	return (^V)(__dynamic_map_set_without_hash((^Raw_Map)(m), map_info(T), rawptr(&key), rawptr(&value), loc))
+}
+
+// Explicitly inserts a key and value into a map `m`, the same as `map_insert`, but the return values differ.
+// - `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 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)
+	if kp != nil {
+		prev_key = (^K)(kp)^
+		found_previous = true
+	}
+	value_ptr = (^V)(vp)
+	return
+}
+
+
+@builtin
+card :: proc "contextless" (s: $S/bit_set[$E; $U]) -> int {
+	when size_of(S) == 1 {
+		return int(intrinsics.count_ones(transmute(u8)s))
+	} else when size_of(S) == 2 {
+		return int(intrinsics.count_ones(transmute(u16)s))
+	} else when size_of(S) == 4 {
+		return int(intrinsics.count_ones(transmute(u32)s))
+	} else when size_of(S) == 8 {
+		return int(intrinsics.count_ones(transmute(u64)s))
+	} else when size_of(S) == 16 {
+		return int(intrinsics.count_ones(transmute(u128)s))
+	} else {
+		#panic("Unhandled card bit_set size")
+	}
+}
+
+
+
+@builtin
+@(disabled=ODIN_DISABLE_ASSERT)
+assert :: proc(condition: bool, message := "", loc := #caller_location) {
+	if !condition {
+		// NOTE(bill): This is wrapped in a procedure call
+		// to improve performance to make the CPU not
+		// execute speculatively, making it about an order of
+		// magnitude faster
+		@(cold)
+		internal :: proc(message: string, loc: Source_Code_Location) {
+			p := context.assertion_failure_proc
+			if p == nil {
+				p = default_assertion_failure_proc
+			}
+			p("runtime assertion", message, loc)
+		}
+		internal(message, loc)
+	}
+}
+
+@builtin
+panic :: proc(message: string, loc := #caller_location) -> ! {
+	p := context.assertion_failure_proc
+	if p == nil {
+		p = default_assertion_failure_proc
+	}
+	p("panic", message, loc)
+}
+
+@builtin
+unimplemented :: proc(message := "", loc := #caller_location) -> ! {
+	p := context.assertion_failure_proc
+	if p == nil {
+		p = default_assertion_failure_proc
+	}
+	p("not yet implemented", message, loc)
+}
@@ -1,6 +1,6 @@
 package runtime

-import "core:intrinsics"
+import "base:intrinsics"
 _ :: intrinsics

 /*
@@ -50,6 +50,7 @@ Raw_SOA_Footer_Dynamic_Array :: struct {
 	allocator: Allocator,
 }

+@(builtin, require_results)
 raw_soa_footer_slice :: proc(array: ^$T/#soa[]$E) -> (footer: ^Raw_SOA_Footer_Slice) {
 	if array == nil {
 		return nil
@@ -58,6 +59,7 @@ raw_soa_footer_slice :: proc(array: ^$T/#soa[]$E) -> (footer: ^Raw_SOA_Footer_Sl
 	footer = (^Raw_SOA_Footer_Slice)(uintptr(array) + field_count*size_of(rawptr))
 	return
 }
+@(builtin, require_results)
 raw_soa_footer_dynamic_array :: proc(array: ^$T/#soa[dynamic]$E) -> (footer: ^Raw_SOA_Footer_Dynamic_Array) {
 	if array == nil {
 		return nil
@@ -78,7 +80,7 @@ raw_soa_footer :: proc{



-@builtin
+@(builtin, require_results)
 make_soa_aligned :: proc($T: typeid/#soa[]$E, length: int, alignment: int, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) #optional_allocator_error {
 	if length <= 0 {
 		return
@@ -137,32 +139,31 @@ make_soa_aligned :: proc($T: typeid/#soa[]$E, length: int, alignment: int, alloc
 	return
 }

-@builtin
+@(builtin, require_results)
 make_soa_slice :: proc($T: typeid/#soa[]$E, length: int, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) #optional_allocator_error {
 	return make_soa_aligned(T, length, align_of(E), allocator, loc)
 }

-@builtin
-make_soa_dynamic_array :: proc($T: typeid/#soa[dynamic]$E, allocator := context.allocator, loc := #caller_location) -> (array: T) {
+@(builtin, require_results)
+make_soa_dynamic_array :: proc($T: typeid/#soa[dynamic]$E, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) #optional_allocator_error {
 	context.allocator = allocator
-	reserve_soa(&array, DEFAULT_RESERVE_CAPACITY, loc)
-	return
+	reserve_soa(&array, DEFAULT_RESERVE_CAPACITY, loc) or_return
+	return array, nil
 }

-@builtin
-make_soa_dynamic_array_len :: proc($T: typeid/#soa[dynamic]$E, #any_int length: int, allocator := context.allocator, loc := #caller_location) -> (array: T) {
+@(builtin, require_results)
+make_soa_dynamic_array_len :: proc($T: typeid/#soa[dynamic]$E, #any_int length: int, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) #optional_allocator_error {
 	context.allocator = allocator
-	resize_soa(&array, length, loc)
-	return
+	resize_soa(&array, length, loc) or_return
+	return array, nil
 }

-@builtin
-make_soa_dynamic_array_len_cap :: proc($T: typeid/#soa[dynamic]$E, #any_int length, capacity: int, allocator := context.allocator, loc := #caller_location) -> (array: T) {
+@(builtin, require_results)
+make_soa_dynamic_array_len_cap :: proc($T: typeid/#soa[dynamic]$E, #any_int length, capacity: int, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) #optional_allocator_error {
 	context.allocator = allocator
-	if reserve_soa(&array, capacity, loc) {
-		resize_soa(&array, length, loc)
-	}
-	return
+	reserve_soa(&array, capacity, loc) or_return
+	resize_soa(&array, length, loc) or_return
+	return array, nil
 }


@@ -176,27 +177,25 @@ make_soa :: proc{


@builtin
-resize_soa :: proc(array: ^$T/#soa[dynamic]$E, length: int, loc := #caller_location) -> bool {
+resize_soa :: proc(array: ^$T/#soa[dynamic]$E, length: int, loc := #caller_location) -> Allocator_Error {
 	if array == nil {
-		return false
-	}
-	if !reserve_soa(array, length, loc) {
-		return false
+		return nil
 	}
+	reserve_soa(array, length, loc) or_return
 	footer := raw_soa_footer(array)
 	footer.len = length
-	return true
+	return nil
 }

@builtin
-reserve_soa :: proc(array: ^$T/#soa[dynamic]$E, capacity: int, loc := #caller_location) -> bool {
+reserve_soa :: proc(array: ^$T/#soa[dynamic]$E, capacity: int, loc := #caller_location) -> Allocator_Error {
 	if array == nil {
-		return false
+		return nil
 	}

 	old_cap := cap(array)
 	if capacity <= old_cap {
-		return true
+		return nil
 	}

 	if array.allocator.procedure == nil {
@@ -207,7 +206,7 @@ reserve_soa :: proc(array: ^$T/#soa[dynamic]$E, capacity: int, loc := #caller_lo
 	footer := raw_soa_footer(array)
 	if size_of(E) == 0 {
 		footer.cap = capacity
-		return true
+		return nil
 	}

 	ti := type_info_of(typeid_of(T))
@@ -238,13 +237,10 @@ reserve_soa :: proc(array: ^$T/#soa[dynamic]$E, capacity: int, loc := #caller_lo

 	old_data := (^rawptr)(array)^

-	new_bytes, err := array.allocator.procedure(
+	new_bytes := array.allocator.procedure(
 		array.allocator.data, .Alloc, new_size, max_align,
 		nil, old_size, loc,
-	)
-	if new_bytes == nil || err != nil {
-		return false
-	}
+	) or_return
 	new_data := raw_data(new_bytes)


@@ -269,32 +265,29 @@ reserve_soa :: proc(array: ^$T/#soa[dynamic]$E, capacity: int, loc := #caller_lo
 		new_offset += type.size * capacity
 	}

-	_, err = array.allocator.procedure(
+	array.allocator.procedure(
 		array.allocator.data, .Free, 0, max_align,
 		old_data, old_size, loc,
-	)
+	) or_return

-	return true
+	return nil
 }

@builtin
-append_soa_elem :: proc(array: ^$T/#soa[dynamic]$E, arg: E, loc := #caller_location) {
+append_soa_elem :: proc(array: ^$T/#soa[dynamic]$E, arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
 	if array == nil {
-		return
+		return 0, nil
 	}

-	arg_len := 1
-
-	if cap(array) <= len(array)+arg_len {
-		cap := 2 * cap(array) + max(8, arg_len)
-		_ = reserve_soa(array, cap, loc)
+	if cap(array) <= len(array) + 1 {
+		cap := 2 * cap(array) + 8
+		err = reserve_soa(array, cap, loc) // do not 'or_return' here as it could be a partial success
 	}
-	arg_len = min(cap(array)-len(array), arg_len)

 	footer := raw_soa_footer(array)

-	if size_of(E) > 0 && arg_len > 0 {
-		ti := type_info_of(typeid_of(T))
+	if size_of(E) > 0 && cap(array)-len(array) > 0 {
+		ti := type_info_of(T)
 		ti = type_info_base(ti)
 		si := &ti.variant.(Type_Info_Struct)
 		field_count: uintptr
@@ -326,12 +319,14 @@ append_soa_elem :: proc(array: ^$T/#soa[dynamic]$E, arg: E, loc := #caller_locat
 			soa_offset  += type.size * cap(array)
 			item_offset += type.size
 		}
+		footer.len += 1
+		return 1, err
 	}
-	footer.len += arg_len
+	return 0, err
 }

@builtin
-append_soa_elems :: proc(array: ^$T/#soa[dynamic]$E, args: ..E, loc := #caller_location) {
+append_soa_elems :: proc(array: ^$T/#soa[dynamic]$E, args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
 	if array == nil {
 		return
 	}
@@ -343,7 +338,7 @@ append_soa_elems :: proc(array: ^$T/#soa[dynamic]$E, args: ..E, loc := #caller_l

 	if cap(array) <= len(array)+arg_len {
 		cap := 2 * cap(array) + max(8, arg_len)
-		_ = reserve_soa(array, cap, loc)
+		err = reserve_soa(array, cap, loc) // do not 'or_return' here as it could be a partial success
 	}
 	arg_len = min(cap(array)-len(array), arg_len)

@@ -380,8 +375,8 @@ append_soa_elems :: proc(array: ^$T/#soa[dynamic]$E, args: ..E, loc := #caller_l
 			item_offset += type.size
 		}
 	}
-
 	footer.len += arg_len
+	return arg_len, err
 }


@@ -393,21 +388,23 @@ append_soa :: proc{
 }


-delete_soa_slice :: proc(array: $T/#soa[]$E, allocator := context.allocator, loc := #caller_location) {
+delete_soa_slice :: proc(array: $T/#soa[]$E, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
 	when intrinsics.type_struct_field_count(E) != 0 {
 		array := array
 		ptr := (^rawptr)(&array)^
-		free(ptr, allocator, loc)
+		free(ptr, allocator, loc) or_return
 	}
+	return nil
 }

-delete_soa_dynamic_array :: proc(array: $T/#soa[dynamic]$E, loc := #caller_location) {
+delete_soa_dynamic_array :: proc(array: $T/#soa[dynamic]$E, loc := #caller_location) -> Allocator_Error {
 	when intrinsics.type_struct_field_count(E) != 0 {
 		array := array
 		ptr := (^rawptr)(&array)^
 		footer := raw_soa_footer(&array)
-		free(ptr, footer.allocator, loc)
+		free(ptr, footer.allocator, loc) or_return
 	}
+	return nil
 }


@@ -416,3 +413,109 @@ delete_soa :: proc{
 	delete_soa_slice,
 	delete_soa_dynamic_array,
 }
+
+
+clear_soa_dynamic_array :: proc(array: ^$T/#soa[dynamic]$E) {
+	when intrinsics.type_struct_field_count(E) != 0 {
+		footer := raw_soa_footer(array)
+		footer.len = 0
+	}
+}
+
+@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
+}
@@ -1,6 +1,6 @@
 package runtime

-import "core:intrinsics"
+import "base:intrinsics"

 DEFAULT_ARENA_GROWING_MINIMUM_BLOCK_SIZE :: uint(DEFAULT_TEMP_ALLOCATOR_BACKING_SIZE)

@@ -28,11 +28,11 @@ safe_add :: #force_inline proc "contextless" (x, y: uint) -> (uint, bool) {
 }

@(require_results)
-memory_block_alloc :: proc(allocator: Allocator, capacity: uint, loc := #caller_location) -> (block: ^Memory_Block, err: Allocator_Error) {
-	total_size := uint(capacity + size_of(Memory_Block))
-	base_offset    := uintptr(size_of(Memory_Block))
+memory_block_alloc :: proc(allocator: Allocator, capacity: uint, alignment: uint, loc := #caller_location) -> (block: ^Memory_Block, err: Allocator_Error) {
+	total_size  := uint(capacity + max(alignment, size_of(Memory_Block)))
+	base_offset := uintptr(max(alignment, size_of(Memory_Block)))

-	min_alignment: int = max(16, align_of(Memory_Block))
+	min_alignment: int = max(16, align_of(Memory_Block), int(alignment))
 	data := mem_alloc(int(total_size), min_alignment, allocator, loc) or_return
 	block = (^Memory_Block)(raw_data(data))
 	end := uintptr(raw_data(data)[len(data):])
@@ -102,20 +102,20 @@ arena_alloc :: proc(arena: ^Arena, size, alignment: uint, loc := #caller_locatio
 	if size == 0 {
 		return
 	}
-
-	if arena.curr_block == nil || (safe_add(arena.curr_block.used, size) or_else 0) > arena.curr_block.capacity {
-		size = align_forward_uint(size, alignment)
+	
+	needed := align_forward_uint(size, alignment)
+	if arena.curr_block == nil || (safe_add(arena.curr_block.used, needed) or_else 0) > arena.curr_block.capacity {
 		if arena.minimum_block_size == 0 {
 			arena.minimum_block_size = DEFAULT_ARENA_GROWING_MINIMUM_BLOCK_SIZE
 		}

-		block_size := max(size, arena.minimum_block_size)
+		block_size := max(needed, arena.minimum_block_size)

 		if arena.backing_allocator.procedure == nil {
 			arena.backing_allocator = default_allocator()
 		}

-		new_block := memory_block_alloc(arena.backing_allocator, block_size, loc) or_return
+		new_block := memory_block_alloc(arena.backing_allocator, block_size, alignment, loc) or_return
 		new_block.prev = arena.curr_block
 		arena.curr_block = new_block
 		arena.total_capacity += new_block.capacity
@@ -134,7 +134,7 @@ arena_init :: proc(arena: ^Arena, size: uint, backing_allocator: Allocator, loc
 	arena^ = {}
 	arena.backing_allocator = backing_allocator
 	arena.minimum_block_size = max(size, 1<<12) // minimum block size of 4 KiB
-	new_block := memory_block_alloc(arena.backing_allocator, arena.minimum_block_size, loc) or_return
+	new_block := memory_block_alloc(arena.backing_allocator, arena.minimum_block_size, 0, loc) or_return
 	arena.curr_block = new_block
 	arena.total_capacity += new_block.capacity
 	return nil
@@ -195,7 +195,7 @@ arena_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
 		err = .Mode_Not_Implemented
 	case .Free_All:
 		arena_free_all(arena, location)
-	case .Resize:
+	case .Resize, .Resize_Non_Zeroed:
 		old_data := ([^]byte)(old_memory)

 		switch {
@@ -0,0 +1,12 @@
+package runtime
+
+when ODIN_DEFAULT_TO_NIL_ALLOCATOR {
+	default_allocator_proc :: nil_allocator_proc
+	default_allocator :: nil_allocator
+} else when ODIN_DEFAULT_TO_PANIC_ALLOCATOR {
+	default_allocator_proc :: panic_allocator_proc
+	default_allocator :: panic_allocator
+} else {
+	default_allocator :: heap_allocator
+	default_allocator_proc :: heap_allocator_proc
+}
@@ -0,0 +1,80 @@
+package runtime
+
+nil_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
+                               size, alignment: int,
+                               old_memory: rawptr, old_size: int, loc := #caller_location) -> ([]byte, Allocator_Error) {
+	switch mode {
+	case .Alloc, .Alloc_Non_Zeroed:
+		return nil, .Out_Of_Memory
+	case .Free:
+		return nil, .None
+	case .Free_All:
+		return nil, .Mode_Not_Implemented
+	case .Resize, .Resize_Non_Zeroed:
+		if size == 0 {
+			return nil, .None
+		}
+		return nil, .Out_Of_Memory
+	case .Query_Features:
+		return nil, .Mode_Not_Implemented
+	case .Query_Info:
+		return nil, .Mode_Not_Implemented
+	}
+	return nil, .None
+}
+
+nil_allocator :: proc() -> Allocator {
+	return Allocator{
+		procedure = nil_allocator_proc,
+		data = nil,
+	}
+}
+
+
+panic_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
+                             size, alignment: int,
+                             old_memory: rawptr, old_size: int, loc := #caller_location) -> ([]byte, Allocator_Error) {
+	switch mode {
+	case .Alloc:
+		if size > 0 {
+			panic("panic allocator, .Alloc called", loc=loc)
+		}
+	case .Alloc_Non_Zeroed:
+		if size > 0 {
+			panic("panic allocator, .Alloc_Non_Zeroed called", loc=loc)
+		}
+	case .Resize:
+		if size > 0 {
+			panic("panic allocator, .Resize called", loc=loc)
+		}
+	case .Resize_Non_Zeroed:
+		if size > 0 {
+			panic("panic allocator, .Alloc_Non_Zeroed called", loc=loc)
+		}
+	case .Free:
+		if old_memory != nil {
+			panic("panic allocator, .Free called", loc=loc)
+		}
+	case .Free_All:
+		panic("panic allocator, .Free_All called", loc=loc)
+
+	case .Query_Features:
+		set := (^Allocator_Mode_Set)(old_memory)
+		if set != nil {
+			set^ = {.Query_Features}
+		}
+		return nil, nil
+
+	case .Query_Info:
+		panic("panic allocator, .Query_Info called", loc=loc)
+	}
+
+	return nil, nil
+}
+
+panic_allocator :: proc() -> Allocator {
+	return Allocator{
+		procedure = panic_allocator_proc,
+		data = nil,
+	}
+}
@@ -0,0 +1,180 @@
+package runtime
+
+/*
+
+package runtime has numerous entities (declarations) which are required by the compiler to function.
+
+
+## Basic types and calls (and anything they rely on)
+
+Source_Code_Location
+Context
+Allocator
+Logger
+
+__init_context
+_cleanup_runtime
+
+
+## cstring calls
+
+cstring_to_string
+cstring_len
+
+
+
+## Required when RTTI is enabled (the vast majority of targets)
+
+Type_Info
+
+type_table
+__type_info_of
+
+
+## Hashing
+
+default_hasher
+default_hasher_cstring
+default_hasher_string
+
+
+## Pseudo-CRT required procedured due to LLVM but useful in general
+memset
+memcpy
+memove
+
+
+## Procedures required by the LLVM backend if u128/i128 is used
+umodti3
+udivti3
+modti3
+divti3
+fixdfti
+fixunsdfti
+fixunsdfdi
+floattidf
+floattidf_unsigned
+truncsfhf2
+truncdfhf2
+gnu_h2f_ieee
+gnu_f2h_ieee
+extendhfsf2
+
+## Procedures required by the LLVM backend if f16 is used
+__ashlti3 // wasm specific
+__multi3  // wasm specific
+
+
+## Required an entry point is defined (i.e. 'main')
+
+args__
+
+
+## When -no-crt is defined (and not a wasm target) (mostly due to LLVM)
+_tls_index
+_fltused
+
+
+## Bounds checking procedures (when not disabled with -no-bounds-check)
+
+bounds_check_error
+matrix_bounds_check_error
+slice_expr_error_hi
+slice_expr_error_lo_hi
+multi_pointer_slice_expr_error
+
+
+## Type assertion check
+
+type_assertion_check
+type_assertion_check2 // takes in typeid
+
+
+## Arithmetic
+
+quo_complex32
+quo_complex64
+quo_complex128
+
+mul_quaternion64
+mul_quaternion128
+mul_quaternion256
+
+quo_quaternion64
+quo_quaternion128
+quo_quaternion256
+
+abs_complex32
+abs_complex64
+abs_complex128
+
+abs_quaternion64
+abs_quaternion128
+abs_quaternion256
+
+
+## Comparison
+
+memory_equal
+memory_compare
+memory_compare_zero
+
+cstring_eq
+cstring_ne
+cstring_lt
+cstring_gt
+cstring_le
+cstring_gt
+
+string_eq
+string_ne
+string_lt
+string_gt
+string_le
+string_gt
+
+complex32_eq
+complex32_ne
+complex64_eq
+complex64_ne
+complex128_eq
+complex128_ne
+
+quaternion64_eq
+quaternion64_ne
+quaternion128_eq
+quaternion128_ne
+quaternion256_eq
+quaternion256_ne
+
+
+## Map specific calls
+
+map_seed_from_map_data
+__dynamic_map_check_grow // static map calls
+map_insert_hash_dynamic  // static map calls
+__dynamic_map_get // dynamic map calls
+__dynamic_map_set // dynamic map calls
+
+
+## Dynamic literals ([dymamic]T and map[K]V) (can be disabled with -no-dynamic-literals)
+
+__dynamic_array_reserve
+__dynamic_array_append
+
+__dynamic_map_reserve
+
+
+## Objective-C specific
+
+objc_lookUpClass
+sel_registerName
+objc_allocateClassPair
+
+
+## for-in `string` type
+
+string_decode_rune
+string_decode_last_rune // #reverse for
+
+*/
@@ -1,6 +1,6 @@
 package runtime

-import "core:intrinsics"
+import "base:intrinsics"
 _ :: intrinsics

 // High performance, cache-friendly, open-addressed Robin Hood hashing hash map
@@ -44,7 +44,7 @@ _ :: intrinsics
 MAP_LOAD_FACTOR :: 75

 // Minimum log2 capacity.
-MAP_MIN_LOG2_CAPACITY :: 6 // 64 elements
+MAP_MIN_LOG2_CAPACITY :: 3 // 8 elements

 // Has to be less than 100% though.
 #assert(MAP_LOAD_FACTOR < 100)
@@ -87,7 +87,7 @@ MAP_CACHE_LINE_SIZE :: 1 << MAP_CACHE_LINE_LOG2
 //
 // In the optimal case, len(Map_Cell(T){}.data) = 1 so the cell array can be treated
 // as a regular array of T, which is the case for hashes.
-Map_Cell :: struct($T: typeid) #align MAP_CACHE_LINE_SIZE {
+Map_Cell :: struct($T: typeid) #align(MAP_CACHE_LINE_SIZE) {
 	data: [MAP_CACHE_LINE_SIZE / size_of(T) when 0 < size_of(T) && size_of(T) < MAP_CACHE_LINE_SIZE else 1]T,
 }

@@ -184,7 +184,7 @@ map_cell_index_static :: #force_inline proc "contextless" (cells: [^]Map_Cell($T
 // len() for map
@(require_results)
 map_len :: #force_inline proc "contextless" (m: Raw_Map) -> int {
-	return m.len
+	return int(m.len)
 }

 // cap() for map
@@ -207,8 +207,8 @@ map_load_factor :: #force_inline proc "contextless" (log2_capacity: uintptr) ->
 }

@(require_results)
-map_resize_threshold :: #force_inline proc "contextless" (m: Raw_Map) -> int {
-	return int(map_load_factor(map_log2_cap(m)))
+map_resize_threshold :: #force_inline proc "contextless" (m: Raw_Map) -> uintptr {
+	return map_load_factor(map_log2_cap(m))
 }

 // The data stores the log2 capacity in the lower six bits. This is primarily
@@ -251,6 +251,26 @@ map_hash_is_valid :: #force_inline proc "contextless" (hash: Map_Hash) -> bool {
 	return (hash != 0) & (hash & TOMBSTONE_MASK == 0)
 }

+@(require_results)
+map_seed :: #force_inline proc "contextless" (m: Raw_Map) -> uintptr {
+	return map_seed_from_map_data(map_data(m))
+}
+
+// splitmix for uintptr
+@(require_results)
+map_seed_from_map_data :: #force_inline proc "contextless" (data: uintptr) -> uintptr {
+	when size_of(uintptr) == size_of(u64) {
+		mix := data + 0x9e3779b97f4a7c15
+		mix = (mix ~ (mix >> 30)) * 0xbf58476d1ce4e5b9
+		mix = (mix ~ (mix >> 27)) * 0x94d049bb133111eb
+		return mix ~ (mix >> 31)
+	} else {
+		mix := data + 0x9e3779b9
+		mix = (mix ~ (mix >> 16)) * 0x21f0aaad
+		mix = (mix ~ (mix >> 15)) * 0x735a2d97
+		return mix ~ (mix >> 15)
+	}
+}

 // Computes the desired position in the array. This is just index % capacity,
 // but a procedure as there's some math involved here to recover the capacity.
@@ -313,7 +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
@@ -330,6 +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) {
@@ -371,7 +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
@@ -394,32 +421,29 @@ map_insert_hash_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^
 	tk := map_cell_index_dynamic(sk, info.ks, 1)
 	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")
+		}

-	for {
-		hp := &hs[pos]
-		element_hash := hp^
+		element_hash := hs[pos]

 		if map_hash_is_empty(element_hash) {
 			k_dst := map_cell_index_dynamic(ks, info.ks, pos)
 			v_dst := map_cell_index_dynamic(vs, info.vs, pos)
 			intrinsics.mem_copy_non_overlapping(rawptr(k_dst), rawptr(k), size_of_k)
 			intrinsics.mem_copy_non_overlapping(rawptr(v_dst), rawptr(v), size_of_v)
-			hp^ = h
+			hs[pos] = h

 			return result if result != 0 else v_dst
 		}

+		if map_hash_is_deleted(element_hash) {
+			break swap_loop
+		}
+
 		if probe_distance := map_probe_distance(m^, element_hash, pos); distance > probe_distance {
-			if map_hash_is_deleted(element_hash) {
-				k_dst := map_cell_index_dynamic(ks, info.ks, pos)
-				v_dst := map_cell_index_dynamic(vs, info.vs, pos)
-				intrinsics.mem_copy_non_overlapping(rawptr(k_dst), rawptr(k), size_of_k)
-				intrinsics.mem_copy_non_overlapping(rawptr(v_dst), rawptr(v), size_of_v)
-				hp^ = h
-
-				return result if result != 0 else v_dst
-			}
-
 			if result == 0 {
 				result = map_cell_index_dynamic(vs, info.vs, pos)
 			}
@@ -436,8 +460,8 @@ map_insert_hash_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^
 			intrinsics.mem_copy_non_overlapping(rawptr(vp), rawptr(tv), size_of_v)

 			th := h
-			h = hp^
-			hp^ = th
+			h = hs[pos]
+			hs[pos] = th

 			distance = probe_distance
 		}
@@ -445,6 +469,103 @@ map_insert_hash_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^
 		pos = (pos + 1) & mask
 		distance += 1
 	}
+
+	// backward shift loop
+	hs[pos] = 0
+	look_ahead: uintptr = 1
+	for {
+		la_pos := (pos + look_ahead) & mask
+		element_hash := hs[la_pos]
+
+		if map_hash_is_deleted(element_hash) {
+			look_ahead += 1
+			hs[la_pos] = 0
+			continue
+		}
+
+		k_dst := map_cell_index_dynamic(ks, info.ks, pos)
+		v_dst := map_cell_index_dynamic(vs, info.vs, pos)
+
+		if map_hash_is_empty(element_hash) {
+			intrinsics.mem_copy_non_overlapping(rawptr(k_dst), rawptr(k), size_of_k)
+			intrinsics.mem_copy_non_overlapping(rawptr(v_dst), rawptr(v), size_of_v)
+			hs[pos] = h
+
+			return result if result != 0 else v_dst
+		}
+
+		k_src := map_cell_index_dynamic(ks, info.ks, la_pos)
+		v_src := map_cell_index_dynamic(vs, info.vs, la_pos)
+		probe_distance := map_probe_distance(m^, element_hash, la_pos)
+
+		if probe_distance < look_ahead {
+			// probed can be made ideal while placing saved (ending condition)
+			if result == 0 {
+				result = v_dst
+			}
+			intrinsics.mem_copy_non_overlapping(rawptr(k_dst), rawptr(k), size_of_k)
+			intrinsics.mem_copy_non_overlapping(rawptr(v_dst), rawptr(v), size_of_v)
+			hs[pos] = h
+
+			// This will be an ideal move
+			pos = (la_pos - probe_distance) & mask
+			look_ahead -= probe_distance
+
+			// shift until we hit ideal/empty
+			for probe_distance != 0 {
+				k_dst = map_cell_index_dynamic(ks, info.ks, pos)
+				v_dst = map_cell_index_dynamic(vs, info.vs, pos)
+
+				intrinsics.mem_copy_non_overlapping(rawptr(k_dst), rawptr(k_src), size_of_k)
+				intrinsics.mem_copy_non_overlapping(rawptr(v_dst), rawptr(v_src), size_of_v)
+				hs[pos] = element_hash
+				hs[la_pos] = 0
+
+				pos = (pos + 1) & mask
+				la_pos = (la_pos + 1) & mask
+				look_ahead = (la_pos - pos) & mask
+				element_hash = hs[la_pos]
+				if map_hash_is_empty(element_hash) {
+					return
+				}
+
+				probe_distance = map_probe_distance(m^, element_hash, la_pos)
+				if probe_distance == 0 {
+					return
+				}
+				// can be ideal?
+				if probe_distance < look_ahead {
+					pos = (la_pos - probe_distance) & mask
+				}
+				k_src = map_cell_index_dynamic(ks, info.ks, la_pos)
+				v_src = map_cell_index_dynamic(vs, info.vs, la_pos)
+			}
+			return
+		} else if distance < probe_distance - look_ahead {
+			// shift back probed
+			intrinsics.mem_copy_non_overlapping(rawptr(k_dst), rawptr(k_src), size_of_k)
+			intrinsics.mem_copy_non_overlapping(rawptr(v_dst), rawptr(v_src), size_of_v)
+			hs[pos] = element_hash
+			hs[la_pos] = 0
+		} else {
+			// place saved, save probed
+			if result == 0 {
+				result = v_dst
+			}
+			intrinsics.mem_copy_non_overlapping(rawptr(k_dst), rawptr(k), size_of_k)
+			intrinsics.mem_copy_non_overlapping(rawptr(v_dst), rawptr(v), size_of_v)
+			hs[pos] = h
+
+			intrinsics.mem_copy_non_overlapping(rawptr(k), rawptr(k_src), size_of_k)
+			intrinsics.mem_copy_non_overlapping(rawptr(v), rawptr(v_src), size_of_v)
+			h = hs[la_pos]
+			hs[la_pos] = 0
+			distance = probe_distance - look_ahead
+		}
+
+		pos = (pos + 1) & mask
+		distance += 1
+	}
 }

@(require_results)
@@ -503,6 +624,7 @@ map_reserve_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_
 		}
 		k := map_cell_index_dynamic(ks, info.ks, i)
 		v := map_cell_index_dynamic(vs, info.vs, i)
+		hash = info.key_hasher(rawptr(k), map_seed(resized))
 		_ = map_insert_hash_dynamic(&resized, info, hash, k, v)
 		// Only need to do this comparison on each actually added pair, so do not
 		// fold it into the for loop comparator as a micro-optimization.
@@ -519,7 +641,7 @@ map_reserve_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_


@(require_results)
-map_shrink_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, loc := #caller_location) -> Allocator_Error {
+map_shrink_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, loc := #caller_location) -> (did_shrink: bool, err: Allocator_Error) {
 	if m.allocator.procedure == nil {
 		m.allocator = context.allocator
 	}
@@ -529,7 +651,7 @@ map_shrink_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_I
 	// map needs to be within the max load factor.
 	log2_capacity := map_log2_cap(m^)
 	if uintptr(m.len) >= map_load_factor(log2_capacity - 1) {
-		return nil
+		return false, nil
 	}

 	shrunk := map_alloc_dynamic(info, log2_capacity - 1, m.allocator) or_return
@@ -550,6 +672,7 @@ map_shrink_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_I

 		k := map_cell_index_dynamic(ks, info.ks, i)
 		v := map_cell_index_dynamic(vs, info.vs, i)
+		hash = info.key_hasher(rawptr(k), map_seed(shrunk))
 		_ = map_insert_hash_dynamic(&shrunk, info, hash, k, v)
 		// Only need to do this comparison on each actually added pair, so do not
 		// fold it into the for loop comparator as a micro-optimization.
@@ -561,7 +684,7 @@ map_shrink_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_I

 	map_free_dynamic(m^, info, loc) or_return
 	m.data = shrunk.data
-	return nil
+	return true, nil
 }

@(require_results)
@@ -581,7 +704,7 @@ map_lookup_dynamic :: proc "contextless" (m: Raw_Map, #no_alias info: ^Map_Info,
 	if map_len(m) == 0 {
 		return 0, false
 	}
-	h := info.key_hasher(rawptr(k), 0)
+	h := info.key_hasher(rawptr(k), map_seed(m))
 	p := map_desired_position(m, h)
 	d := uintptr(0)
 	c := (uintptr(1) << map_log2_cap(m)) - 1
@@ -604,7 +727,7 @@ map_exists_dynamic :: proc "contextless" (m: Raw_Map, #no_alias info: ^Map_Info,
 	if map_len(m) == 0 {
 		return false
 	}
-	h := info.key_hasher(rawptr(k), 0)
+	h := info.key_hasher(rawptr(k), map_seed(m))
 	p := map_desired_position(m, h)
 	d := uintptr(0)
 	c := (uintptr(1) << map_log2_cap(m)) - 1
@@ -635,50 +758,19 @@ map_erase_dynamic :: #force_inline proc "contextless" (#no_alias m: ^Raw_Map, #n
 	m.len -= 1
 	ok = true

-	{ // coalesce tombstones
-		// HACK NOTE(bill): This is an ugly bodge but it is coalescing the tombstone slots
-		// TODO(bill): we should do backward shift deletion and not rely on tombstone slots
-		mask := (uintptr(1)<<map_log2_cap(m^)) - 1
-		curr_index := uintptr(index)
-
-		// TODO(bill): determine a good value for this empirically
-		// if we do not implement backward shift deletion
-		PROBE_COUNT :: 8
-		for _ in 0..<PROBE_COUNT {
-			next_index := (curr_index + 1) & mask
-			if next_index == index {
-				// looped around
-				break
-			}
-
-			// if the next element is empty or has zero probe distance, then any lookup
-			// will always fail on the next, so we can clear both of them
-			hash := hs[next_index]
-			if map_hash_is_empty(hash) || map_probe_distance(m^, hash, next_index) == 0 {
-				hs[curr_index] = 0
-				return
-			}
-
-			// now the next element will have a probe count of at least one,
-			// so it can use the delete slot instead
-			hs[curr_index] = hs[next_index]
-
-			mem_copy_non_overlapping(
-				rawptr(map_cell_index_dynamic(ks, info.ks, curr_index)),
-				rawptr(map_cell_index_dynamic(ks, info.ks, next_index)),
-				int(info.ks.size_of_type),
-			)
-			mem_copy_non_overlapping(
-				rawptr(map_cell_index_dynamic(vs, info.vs, curr_index)),
-				rawptr(map_cell_index_dynamic(vs, info.vs, next_index)),
-				int(info.vs.size_of_type),
-			)
-
-			curr_index = next_index
-		}
+	mask := (uintptr(1)<<map_log2_cap(m^)) - 1
+	curr_index := uintptr(index)
+	next_index := (curr_index + 1) & mask

+	// if the next element is empty or has zero probe distance, then any lookup
+	// will always fail on the next, so we can clear both of them
+	hash := hs[next_index]
+	if map_hash_is_empty(hash) || map_probe_distance(m^, hash, next_index) == 0 {
+		hs[curr_index] = 0
+	} else {
 		hs[curr_index] |= TOMBSTONE_MASK
 	}
+
 	return
 }

@@ -711,7 +803,7 @@ map_get :: proc "contextless" (m: $T/map[$K]$V, key: K) -> (stored_key: K, store
 	info := intrinsics.type_map_info(T)
 	key := key

-	h := info.key_hasher(&key, 0)
+	h := info.key_hasher(&key, map_seed(rm))
 	pos := map_desired_position(rm, h)
 	distance := uintptr(0)
 	mask := (uintptr(1) << map_log2_cap(rm)) - 1
@@ -761,16 +853,43 @@ __dynamic_map_get :: proc "contextless" (#no_alias m: ^Raw_Map, #no_alias info:
 	}
 }

-// IMPORTANT: USED WITHIN THE COMPILER
-__dynamic_map_check_grow :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, loc := #caller_location) -> Allocator_Error {
-	if m.len >= map_resize_threshold(m^) {
-		return map_grow_dynamic(m, info, loc)
+__dynamic_map_get_key_and_value :: proc "contextless" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, h: Map_Hash, key: rawptr) -> (key_ptr, value_ptr: rawptr) {
+	if m.len == 0 {
+		return nil, nil
 	}
-	return nil
+	pos := map_desired_position(m^, h)
+	distance := uintptr(0)
+	mask := (uintptr(1) << map_log2_cap(m^)) - 1
+	ks, vs, hs, _, _ := map_kvh_data_dynamic(m^, info)
+	for {
+		element_hash := hs[pos]
+		if map_hash_is_empty(element_hash) {
+			return nil, nil
+		} else if distance > map_probe_distance(m^, element_hash, pos) {
+			return nil, nil
+		} else if element_hash == h {
+			other_key := rawptr(map_cell_index_dynamic(ks, info.ks, pos))
+			if info.key_equal(key, other_key) {
+				key_ptr   = other_key
+				value_ptr = rawptr(map_cell_index_dynamic(vs, info.vs, pos))
+				return
+			}
+		}
+		pos = (pos + 1) & mask
+		distance += 1
+	}
+}
+
+// IMPORTANT: USED WITHIN THE COMPILER
+__dynamic_map_check_grow :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, loc := #caller_location) -> (err: Allocator_Error, has_grown: bool) {
+	if m.len >= map_resize_threshold(m^) {
+		return map_grow_dynamic(m, info, loc), true
+	}
+	return nil, false
 }

 __dynamic_map_set_without_hash :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, key, value: rawptr, loc := #caller_location) -> rawptr {
-	return __dynamic_map_set(m, info, info.key_hasher(key, 0), key, value, loc)
+	return __dynamic_map_set(m, info, info.key_hasher(key, map_seed(m^)), key, value, loc)
 }


@@ -781,14 +900,47 @@ __dynamic_map_set :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_In
 		return found
 	}

-	if __dynamic_map_check_grow(m, info, loc) != nil {
+	hash := hash
+	err, has_grown := __dynamic_map_check_grow(m, info, loc)
+	if err != nil {
 		return nil
 	}
+	if has_grown {
+		hash = info.key_hasher(key, map_seed(m^))
+	}

 	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) {
+	return __dynamic_map_set_extra(m, info, info.key_hasher(key, map_seed(m^)), key, value, loc)
+}
+
+__dynamic_map_set_extra :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, hash: Map_Hash, key, value: rawptr, loc := #caller_location) -> (prev_key_ptr, value_ptr: rawptr) {
+	if prev_key_ptr, value_ptr = __dynamic_map_get_key_and_value(m, info, hash, key); value_ptr != nil {
+		intrinsics.mem_copy_non_overlapping(value_ptr, value, info.vs.size_of_type)
+		return
+	}
+
+	hash := hash
+	err, has_grown := __dynamic_map_check_grow(m, info, loc)
+	if err != nil {
+		return nil, nil
+	}
+	if has_grown {
+		hash = info.key_hasher(key, map_seed(m^))
+	}
+
+	result := map_insert_hash_dynamic(m, info, hash, uintptr(key), uintptr(value))
+	if result != 0 {
+		m.len += 1
+	}
+	return nil, rawptr(result)
+}
+

 // IMPORTANT: USED WITHIN THE COMPILER
@(private)
@@ -0,0 +1,59 @@
+//+private
+//+build linux, darwin, freebsd, openbsd, haiku
+//+no-instrumentation
+package runtime
+
+import "base:intrinsics"
+
+when ODIN_BUILD_MODE == .Dynamic {
+	@(link_name="_odin_entry_point", linkage="strong", require/*, link_section=".init"*/)
+	_odin_entry_point :: proc "c" () {
+		context = default_context()
+		#force_no_inline _startup_runtime()
+		intrinsics.__entry_point()
+	}
+	@(link_name="_odin_exit_point", linkage="strong", require/*, link_section=".fini"*/)
+	_odin_exit_point :: proc "c" () {
+		context = default_context()
+		#force_no_inline _cleanup_runtime()
+	}
+	@(link_name="main", linkage="strong", require)
+	main :: proc "c" (argc: i32, argv: [^]cstring) -> i32 {
+		return 0
+	}
+} else when !ODIN_TEST && !ODIN_NO_ENTRY_POINT {
+	when ODIN_NO_CRT {
+		// NOTE(flysand): We need to start from assembly because we need
+		// to retrieve argc and argv from the stack
+		when ODIN_ARCH == .amd64 {
+			@require foreign import entry "entry_unix_no_crt_amd64.asm"
+			SYS_exit :: 60
+		} else when ODIN_ARCH == .i386 {
+			@require foreign import entry "entry_unix_no_crt_i386.asm"
+			SYS_exit :: 1
+		} else when ODIN_OS == .Darwin && ODIN_ARCH == .arm64 {
+			@require foreign import entry "entry_unix_no_crt_darwin_arm64.asm"
+			SYS_exit :: 1
+		}
+		@(link_name="_start_odin", linkage="strong", require)
+		_start_odin :: proc "c" (argc: i32, argv: [^]cstring) -> ! {
+			args__ = argv[:argc]
+			context = default_context()
+			#force_no_inline _startup_runtime()
+			intrinsics.__entry_point()
+			#force_no_inline _cleanup_runtime()
+			intrinsics.syscall(SYS_exit, 0)
+			unreachable()
+		}
+	} else {
+		@(link_name="main", linkage="strong", require)
+		main :: proc "c" (argc: i32, argv: [^]cstring) -> i32 {
+			args__ = argv[:argc]
+			context = default_context()
+			#force_no_inline _startup_runtime()
+			intrinsics.__entry_point()
+			#force_no_inline _cleanup_runtime()
+			return 0
+		}
+	}
+}
@@ -0,0 +1,43 @@
+bits 64
+
+extern _start_odin
+global _start
+
+section .text
+
+;; Entry point for programs that specify -no-crt option
+;; This entry point should be compatible with dynamic loaders on linux
+;; The parameters the dynamic loader passes to the _start function:
+;;    RDX = pointer to atexit function
+;; The stack layout is as follows:
+;;    +-------------------+
+;;            NULL
+;;    +-------------------+
+;;           envp[m]
+;;    +-------------------+
+;;            ...
+;;    +-------------------+
+;;           envp[0]
+;;    +-------------------+
+;;            NULL
+;;    +-------------------+
+;;           argv[n]
+;;    +-------------------+
+;;            ...
+;;    +-------------------+
+;;           argv[0]
+;;    +-------------------+
+;;            argc
+;;    +-------------------+ <------ RSP
+;;
+_start:
+    ;; Mark stack frame as the top of the stack
+    xor rbp, rbp
+    ;; Load argc into 1st param reg, argv into 2nd param reg
+    pop rdi
+    mov rdx, rsi
+    ;; Align stack pointer down to 16-bytes (sysv calling convention)
+    and rsp, -16
+    ;; Call into odin entry point
+    call _start_odin
+    jmp $$
@@ -0,0 +1,20 @@
+	.section __TEXT,__text
+
+	; NOTE(laytan): this should ideally be the -minimum-os-version flag but there is no nice way of preprocessing assembly in Odin.
+	; 10 seems to be the lowest it goes and I don't see it mess with any targeted os version so this seems fine.
+	.build_version macos, 10, 0
+
+	.extern __start_odin
+
+	.global _main
+	.align 2
+_main:
+	mov x5, sp       ; use x5 as the stack pointer
+
+	str x0, [x5]     ; get argc into x0 (kernel passes 32-bit int argc as 64-bits on stack to keep alignment)
+	str x1, [x5, #8] ; get argv into x1
+
+	and sp, x5, #~15 ; force 16-byte alignment of the stack
+	
+	bl __start_odin  ; call into Odin entry point
+	ret              ; should never get here
@@ -0,0 +1,18 @@
+bits 32
+
+extern _start_odin
+global _start
+
+section .text
+
+;; NOTE(flysand): For description see the corresponding *_amd64.asm file
+;; also I didn't test this on x86-32
+_start:
+    xor ebp, rbp
+    pop ecx
+    mov eax, esp
+    and esp, -16
+    push eax
+    push ecx
+    call _start_odin
+    jmp $$
@@ -1,8 +1,9 @@
 //+private
-//+build wasm32, wasm64
+//+build wasm32, wasm64p32
+//+no-instrumentation
 package runtime

-import "core:intrinsics"
+import "base:intrinsics"

 when !ODIN_TEST && !ODIN_NO_ENTRY_POINT {
 	@(link_name="_start", linkage="strong", require, export)
@@ -1,12 +1,13 @@
 //+private
 //+build windows
+//+no-instrumentation
 package runtime

-import "core:intrinsics"
+import "base:intrinsics"

 when ODIN_BUILD_MODE == .Dynamic {
 	@(link_name="DllMain", linkage="strong", require)
-	DllMain :: proc "stdcall" (hinstDLL: rawptr, fdwReason: u32, lpReserved: rawptr) -> b32 {
+	DllMain :: proc "system" (hinstDLL: rawptr, fdwReason: u32, lpReserved: rawptr) -> b32 {
 		context = default_context()

 		// Populate Windows DLL-specific global
@@ -28,7 +29,7 @@ when ODIN_BUILD_MODE == .Dynamic {
 } else when !ODIN_TEST && !ODIN_NO_ENTRY_POINT {
 	when ODIN_ARCH == .i386 || ODIN_NO_CRT {
 		@(link_name="mainCRTStartup", linkage="strong", require)
-		mainCRTStartup :: proc "stdcall" () -> i32 {
+		mainCRTStartup :: proc "system" () -> i32 {
 			context = default_context()
 			#force_no_inline _startup_runtime()
 			intrinsics.__entry_point()
@@ -1,5 +1,6 @@
 package runtime

+@(no_instrumentation)
 bounds_trap :: proc "contextless" () -> ! {
 	when ODIN_OS == .Windows {
 		windows_trap_array_bounds()
@@ -8,6 +9,7 @@ bounds_trap :: proc "contextless" () -> ! {
 	}
 }

+@(no_instrumentation)
 type_assertion_trap :: proc "contextless" () -> ! {
 	when ODIN_OS == .Windows {
 		windows_trap_type_assertion()
@@ -17,12 +19,13 @@ type_assertion_trap :: proc "contextless" () -> ! {
 }


+@(disabled=ODIN_NO_BOUNDS_CHECK)
 bounds_check_error :: proc "contextless" (file: string, line, column: i32, index, count: int) {
 	if uint(index) < uint(count) {
 		return
 	}
-	@(cold)
-	handle_error :: proc "contextless" (file: string, line, column: i32, index, count: int) {
+	@(cold, no_instrumentation)
+	handle_error :: proc "contextless" (file: string, line, column: i32, index, count: int) -> ! {
 		print_caller_location(Source_Code_Location{file, line, column, ""})
 		print_string(" Index ")
 		print_i64(i64(index))
@@ -34,6 +37,7 @@ bounds_check_error :: proc "contextless" (file: string, line, column: i32, index
 	handle_error(file, line, column, index, count)
 }

+@(no_instrumentation)
 slice_handle_error :: proc "contextless" (file: string, line, column: i32, lo, hi: int, len: int) -> ! {
 	print_caller_location(Source_Code_Location{file, line, column, ""})
 	print_string(" Invalid slice indices ")
@@ -46,6 +50,7 @@ slice_handle_error :: proc "contextless" (file: string, line, column: i32, lo, h
 	bounds_trap()
 }

+@(no_instrumentation)
 multi_pointer_slice_handle_error :: proc "contextless" (file: string, line, column: i32, lo, hi: int) -> ! {
 	print_caller_location(Source_Code_Location{file, line, column, ""})
 	print_string(" Invalid slice indices ")
@@ -57,6 +62,7 @@ multi_pointer_slice_handle_error :: proc "contextless" (file: string, line, colu
 }


+@(disabled=ODIN_NO_BOUNDS_CHECK)
 multi_pointer_slice_expr_error :: proc "contextless" (file: string, line, column: i32, lo, hi: int) {
 	if lo <= hi {
 		return
@@ -64,6 +70,7 @@ multi_pointer_slice_expr_error :: proc "contextless" (file: string, line, column
 	multi_pointer_slice_handle_error(file, line, column, lo, hi)
 }

+@(disabled=ODIN_NO_BOUNDS_CHECK)
 slice_expr_error_hi :: proc "contextless" (file: string, line, column: i32, hi: int, len: int) {
 	if 0 <= hi && hi <= len {
 		return
@@ -71,6 +78,7 @@ slice_expr_error_hi :: proc "contextless" (file: string, line, column: i32, hi:
 	slice_handle_error(file, line, column, 0, hi, len)
 }

+@(disabled=ODIN_NO_BOUNDS_CHECK)
 slice_expr_error_lo_hi :: proc "contextless" (file: string, line, column: i32, lo, hi: int, len: int) {
 	if 0 <= lo && lo <= len && lo <= hi && hi <= len {
 		return
@@ -78,12 +86,13 @@ slice_expr_error_lo_hi :: proc "contextless" (file: string, line, column: i32, l
 	slice_handle_error(file, line, column, lo, hi, len)
 }

+@(disabled=ODIN_NO_BOUNDS_CHECK)
 dynamic_array_expr_error :: proc "contextless" (file: string, line, column: i32, low, high, max: int) {
 	if 0 <= low && low <= high && high <= max {
 		return
 	}
-	@(cold)
-	handle_error :: proc "contextless" (file: string, line, column: i32, low, high, max: int) {
+	@(cold, no_instrumentation)
+	handle_error :: proc "contextless" (file: string, line, column: i32, low, high, max: int) -> ! {
 		print_caller_location(Source_Code_Location{file, line, column, ""})
 		print_string(" Invalid dynamic array indices ")
 		print_i64(i64(low))
@@ -98,13 +107,14 @@ dynamic_array_expr_error :: proc "contextless" (file: string, line, column: i32,
 }


+@(disabled=ODIN_NO_BOUNDS_CHECK)
 matrix_bounds_check_error :: proc "contextless" (file: string, line, column: i32, row_index, column_index, row_count, column_count: int) {
 	if uint(row_index) < uint(row_count) &&
 	   uint(column_index) < uint(column_count) {
 		return
 	}
-	@(cold)
-	handle_error :: proc "contextless" (file: string, line, column: i32, row_index, column_index, row_count, column_count: int) {
+	@(cold, no_instrumentation)
+	handle_error :: proc "contextless" (file: string, line, column: i32, row_index, column_index, row_count, column_count: int) -> ! {
 		print_caller_location(Source_Code_Location{file, line, column, ""})
 		print_string(" Matrix indices [")
 		print_i64(i64(row_index))
@@ -122,13 +132,13 @@ matrix_bounds_check_error :: proc "contextless" (file: string, line, column: i32
 }


-when ODIN_DISALLOW_RTTI {
+when ODIN_NO_RTTI {
 	type_assertion_check :: proc "contextless" (ok: bool, file: string, line, column: i32) {
 		if ok {
 			return
 		}
-		@(cold)
-		handle_error :: proc "contextless" (file: string, line, column: i32) {
+		@(cold, no_instrumentation)
+		handle_error :: proc "contextless" (file: string, line, column: i32) -> ! {
 			print_caller_location(Source_Code_Location{file, line, column, ""})
 			print_string(" Invalid type assertion\n")
 			type_assertion_trap()
@@ -140,8 +150,8 @@ when ODIN_DISALLOW_RTTI {
 		if ok {
 			return
 		}
-		@(cold)
-		handle_error :: proc "contextless" (file: string, line, column: i32) {
+		@(cold, no_instrumentation)
+		handle_error :: proc "contextless" (file: string, line, column: i32) -> ! {
 			print_caller_location(Source_Code_Location{file, line, column, ""})
 			print_string(" Invalid type assertion\n")
 			type_assertion_trap()
@@ -153,8 +163,8 @@ when ODIN_DISALLOW_RTTI {
 		if ok {
 			return
 		}
-		@(cold)
-		handle_error :: proc "contextless" (file: string, line, column: i32, from, to: typeid) {
+		@(cold, no_instrumentation)
+		handle_error :: proc "contextless" (file: string, line, column: i32, from, to: typeid) -> ! {
 			print_caller_location(Source_Code_Location{file, line, column, ""})
 			print_string(" Invalid type assertion from ")
 			print_typeid(from)
@@ -198,8 +208,8 @@ when ODIN_DISALLOW_RTTI {
 			return id
 		}

-		@(cold)
-		handle_error :: proc "contextless" (file: string, line, column: i32, from, to: typeid, from_data: rawptr) {
+		@(cold, no_instrumentation)
+		handle_error :: proc "contextless" (file: string, line, column: i32, from, to: typeid, from_data: rawptr) -> ! {

 			actual := variant_type(from, from_data)

@@ -220,12 +230,13 @@ when ODIN_DISALLOW_RTTI {
 }


+@(disabled=ODIN_NO_BOUNDS_CHECK)
 make_slice_error_loc :: #force_inline proc "contextless" (loc := #caller_location, len: int) {
 	if 0 <= len {
 		return
 	}
-	@(cold)
-	handle_error :: proc "contextless" (loc: Source_Code_Location, len: int) {
+	@(cold, no_instrumentation)
+	handle_error :: proc "contextless" (loc: Source_Code_Location, len: int) -> ! {
 		print_caller_location(loc)
 		print_string(" Invalid slice length for make: ")
 		print_i64(i64(len))
@@ -235,12 +246,13 @@ make_slice_error_loc :: #force_inline proc "contextless" (loc := #caller_locatio
 	handle_error(loc, len)
 }

-make_dynamic_array_error_loc :: #force_inline proc "contextless" (using loc := #caller_location, len, cap: int) {
+@(disabled=ODIN_NO_BOUNDS_CHECK)
+make_dynamic_array_error_loc :: #force_inline proc "contextless" (loc := #caller_location, len, cap: int) {
 	if 0 <= len && len <= cap {
 		return
 	}
-	@(cold)
-	handle_error :: proc "contextless" (loc: Source_Code_Location, len, cap: int) {
+	@(cold, no_instrumentation)
+	handle_error :: proc "contextless" (loc: Source_Code_Location, len, cap: int)  -> ! {
 		print_caller_location(loc)
 		print_string(" Invalid dynamic array parameters for make: ")
 		print_i64(i64(len))
@@ -252,12 +264,13 @@ make_dynamic_array_error_loc :: #force_inline proc "contextless" (using loc := #
 	handle_error(loc, len, cap)
 }

+@(disabled=ODIN_NO_BOUNDS_CHECK)
 make_map_expr_error_loc :: #force_inline proc "contextless" (loc := #caller_location, cap: int) {
 	if 0 <= cap {
 		return
 	}
-	@(cold)
-	handle_error :: proc "contextless" (loc: Source_Code_Location, cap: int) {
+	@(cold, no_instrumentation)
+	handle_error :: proc "contextless" (loc: Source_Code_Location, cap: int)  -> ! {
 		print_caller_location(loc)
 		print_string(" Invalid map capacity for make: ")
 		print_i64(i64(cap))
@@ -270,19 +283,22 @@ make_map_expr_error_loc :: #force_inline proc "contextless" (loc := #caller_loca



-
-bounds_check_error_loc :: #force_inline proc "contextless" (using loc := #caller_location, index, count: int) {
-	bounds_check_error(file_path, line, column, index, count)
+@(disabled=ODIN_NO_BOUNDS_CHECK)
+bounds_check_error_loc :: #force_inline proc "contextless" (loc := #caller_location, index, count: int) {
+	bounds_check_error(loc.file_path, loc.line, loc.column, index, count)
 }

-slice_expr_error_hi_loc :: #force_inline proc "contextless" (using loc := #caller_location, hi: int, len: int) {
-	slice_expr_error_hi(file_path, line, column, hi, len)
+@(disabled=ODIN_NO_BOUNDS_CHECK)
+slice_expr_error_hi_loc :: #force_inline proc "contextless" (loc := #caller_location, hi: int, len: int) {
+	slice_expr_error_hi(loc.file_path, loc.line, loc.column, hi, len)
 }

-slice_expr_error_lo_hi_loc :: #force_inline proc "contextless" (using loc := #caller_location, lo, hi: int, len: int) {
-	slice_expr_error_lo_hi(file_path, line, column, lo, hi, len)
+@(disabled=ODIN_NO_BOUNDS_CHECK)
+slice_expr_error_lo_hi_loc :: #force_inline proc "contextless" (loc := #caller_location, lo, hi: int, len: int) {
+	slice_expr_error_lo_hi(loc.file_path, loc.line, loc.column, lo, hi, len)
 }

-dynamic_array_expr_error_loc :: #force_inline proc "contextless" (using loc := #caller_location, low, high, max: int) {
-	dynamic_array_expr_error(file_path, line, column, low, high, max)
+@(disabled=ODIN_NO_BOUNDS_CHECK)
+dynamic_array_expr_error_loc :: #force_inline proc "contextless" (loc := #caller_location, low, high, max: int) {
+	dynamic_array_expr_error(loc.file_path, loc.line, loc.column, low, high, max)
 }
@@ -0,0 +1,110 @@
+package runtime
+
+import "base:intrinsics"
+
+heap_allocator :: proc() -> Allocator {
+	return Allocator{
+		procedure = heap_allocator_proc,
+		data = nil,
+	}
+}
+
+heap_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
+                            size, alignment: int,
+                            old_memory: rawptr, old_size: int, loc := #caller_location) -> ([]byte, Allocator_Error) {
+	//
+	// NOTE(tetra, 2020-01-14): The heap doesn't respect alignment.
+	// Instead, we overallocate by `alignment + size_of(rawptr) - 1`, and insert
+	// padding. We also store the original pointer returned by heap_alloc right before
+	// the pointer we return to the user.
+	//
+
+	aligned_alloc :: proc(size, alignment: int, old_ptr: rawptr = nil, zero_memory := true) -> ([]byte, Allocator_Error) {
+		a := max(alignment, align_of(rawptr))
+		space := size + a - 1
+
+		allocated_mem: rawptr
+		if old_ptr != nil {
+			original_old_ptr := ([^]rawptr)(old_ptr)[-1]
+			allocated_mem = heap_resize(original_old_ptr, space+size_of(rawptr))
+		} else {
+			allocated_mem = heap_alloc(space+size_of(rawptr), zero_memory)
+		}
+		aligned_mem := rawptr(([^]u8)(allocated_mem)[size_of(rawptr):])
+
+		ptr := uintptr(aligned_mem)
+		aligned_ptr := (ptr - 1 + uintptr(a)) & -uintptr(a)
+		diff := int(aligned_ptr - ptr)
+		if (size + diff) > space || allocated_mem == nil {
+			return nil, .Out_Of_Memory
+		}
+
+		aligned_mem = rawptr(aligned_ptr)
+		([^]rawptr)(aligned_mem)[-1] = allocated_mem
+
+		return byte_slice(aligned_mem, size), nil
+	}
+
+	aligned_free :: proc(p: rawptr) {
+		if p != nil {
+			heap_free(([^]rawptr)(p)[-1])
+		}
+	}
+
+	aligned_resize :: proc(p: rawptr, old_size: int, new_size: int, new_alignment: int, zero_memory := true) -> (new_memory: []byte, err: Allocator_Error) {
+		if p == nil {
+			return nil, nil
+		}
+
+		new_memory = aligned_alloc(new_size, new_alignment, p, zero_memory) or_return
+
+		// NOTE: heap_resize does not zero the new memory, so we do it
+		if zero_memory && new_size > old_size {
+			new_region := raw_data(new_memory[old_size:])
+			intrinsics.mem_zero(new_region, new_size - old_size)
+		}
+		return
+	}
+
+	switch mode {
+	case .Alloc, .Alloc_Non_Zeroed:
+		return aligned_alloc(size, alignment, nil, mode == .Alloc)
+
+	case .Free:
+		aligned_free(old_memory)
+
+	case .Free_All:
+		return nil, .Mode_Not_Implemented
+
+	case .Resize, .Resize_Non_Zeroed:
+		if old_memory == nil {
+			return aligned_alloc(size, alignment, nil, mode == .Resize)
+		}
+		return aligned_resize(old_memory, old_size, size, alignment, mode == .Resize)
+
+	case .Query_Features:
+		set := (^Allocator_Mode_Set)(old_memory)
+		if set != nil {
+			set^ = {.Alloc, .Alloc_Non_Zeroed, .Free, .Resize, .Resize_Non_Zeroed, .Query_Features}
+		}
+		return nil, nil
+
+	case .Query_Info:
+		return nil, .Mode_Not_Implemented
+	}
+
+	return nil, nil
+}
+
+
+heap_alloc :: proc(size: int, zero_memory := true) -> rawptr {
+	return _heap_alloc(size, zero_memory)
+}
+
+heap_resize :: proc(ptr: rawptr, new_size: int) -> rawptr {
+	return _heap_resize(ptr, new_size)
+}
+
+heap_free :: proc(ptr: rawptr) {
+	_heap_free(ptr)
+}
@@ -0,0 +1,15 @@
+//+build js, wasi, freestanding, essence
+//+private
+package runtime
+
+_heap_alloc :: proc(size: int, zero_memory := true) -> rawptr {
+	unimplemented("base:runtime 'heap_alloc' procedure is not supported on this platform")
+}
+
+_heap_resize :: proc(ptr: rawptr, new_size: int) -> rawptr {
+	unimplemented("base:runtime 'heap_resize' procedure is not supported on this platform")
+}
+
+_heap_free :: proc(ptr: rawptr) {
+	unimplemented("base:runtime 'heap_free' procedure is not supported on this platform")
+}
@@ -0,0 +1,38 @@
+//+build linux, darwin, freebsd, openbsd, haiku
+//+private
+package runtime
+
+when ODIN_OS == .Darwin {
+	foreign import libc "system:System.framework"
+} else {
+	foreign import libc "system:c"
+}
+
+@(default_calling_convention="c")
+foreign libc {
+	@(link_name="malloc")   _unix_malloc   :: proc(size: int) -> rawptr ---
+	@(link_name="calloc")   _unix_calloc   :: proc(num, size: int) -> rawptr ---
+	@(link_name="free")     _unix_free     :: proc(ptr: rawptr) ---
+	@(link_name="realloc")  _unix_realloc  :: proc(ptr: rawptr, size: int) -> rawptr ---
+}
+
+_heap_alloc :: proc(size: int, zero_memory := true) -> rawptr {
+	if size <= 0 {
+		return nil
+	}
+	if zero_memory {
+		return _unix_calloc(1, size)
+	} else {
+		return _unix_malloc(size)
+	}
+}
+
+_heap_resize :: proc(ptr: rawptr, new_size: int) -> rawptr {
+	// NOTE: _unix_realloc doesn't guarantee new memory will be zeroed on
+	// POSIX platforms. Ensure your caller takes this into account.
+	return _unix_realloc(ptr, new_size)
+}
+
+_heap_free :: proc(ptr: rawptr) {
+	_unix_free(ptr)
+}
@@ -0,0 +1,39 @@
+package runtime
+
+foreign import kernel32 "system:Kernel32.lib"
+
+@(private="file")
+@(default_calling_convention="system")
+foreign kernel32 {
+	// NOTE(bill): The types are not using the standard names (e.g. DWORD and LPVOID) to just minimizing the dependency
+
+	// default_allocator
+	GetProcessHeap :: proc() -> rawptr ---
+	HeapAlloc      :: proc(hHeap: rawptr, dwFlags: u32, dwBytes: uint) -> rawptr ---
+	HeapReAlloc    :: proc(hHeap: rawptr, dwFlags: u32, lpMem: rawptr, dwBytes: uint) -> rawptr ---
+	HeapFree       :: proc(hHeap: rawptr, dwFlags: u32, lpMem: rawptr) -> b32 ---
+}
+
+_heap_alloc :: proc(size: int, zero_memory := true) -> rawptr {
+	HEAP_ZERO_MEMORY :: 0x00000008
+	return HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY if zero_memory else 0, uint(size))
+}
+_heap_resize :: proc(ptr: rawptr, new_size: int) -> rawptr {
+	if new_size == 0 {
+		_heap_free(ptr)
+		return nil
+	}
+	if ptr == nil {
+		return _heap_alloc(new_size)
+	}
+
+	HEAP_ZERO_MEMORY :: 0x00000008
+	return HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, ptr, uint(new_size))
+}
+_heap_free :: proc(ptr: rawptr) {
+	if ptr == nil {
+		return
+	}
+	HeapFree(GetProcessHeap(), 0, ptr)
+}
+
@@ -1,9 +1,9 @@
 package runtime

-import "core:intrinsics"
+import "base:intrinsics"

@(private="file")
-IS_WASM :: ODIN_ARCH == .wasm32 || ODIN_ARCH == .wasm64
+IS_WASM :: ODIN_ARCH == .wasm32 || ODIN_ARCH == .wasm64p32

@(private)
 RUNTIME_LINKAGE :: "strong" when (
@@ -11,7 +11,10 @@ RUNTIME_LINKAGE :: "strong" when (
 	ODIN_BUILD_MODE == .Dynamic ||
 	!ODIN_NO_CRT) &&
 	!IS_WASM) else "internal"
-RUNTIME_REQUIRE :: true
+RUNTIME_REQUIRE :: false // !ODIN_TILDE
+
+@(private)
+__float16 :: f16 when __ODIN_LLVM_F16_SUPPORTED else u16


@(private)
@@ -19,51 +22,7 @@ byte_slice :: #force_inline proc "contextless" (data: rawptr, len: int) -> []byt
 	return ([^]byte)(data)[:max(len, 0)]
 }

-bswap_16 :: proc "contextless" (x: u16) -> u16 {
-	return x>>8 | x<<8
-}
-
-bswap_32 :: proc "contextless" (x: u32) -> u32 {
-	return x>>24 | (x>>8)&0xff00 | (x<<8)&0xff0000 | x<<24
-}
-
-bswap_64 :: proc "contextless" (x: u64) -> u64 {
-	z := x
-	z = (z & 0x00000000ffffffff) << 32 | (z & 0xffffffff00000000) >> 32
-	z = (z & 0x0000ffff0000ffff) << 16 | (z & 0xffff0000ffff0000) >> 16
-	z = (z & 0x00ff00ff00ff00ff) << 8  | (z & 0xff00ff00ff00ff00) >> 8
-	return z
-}
-
-bswap_128 :: proc "contextless" (x: u128) -> u128 {
-	z := transmute([4]u32)x
-	z[0], z[3] = bswap_32(z[3]), bswap_32(z[0])
-	z[1], z[2] = bswap_32(z[2]), bswap_32(z[1])
-	return transmute(u128)z
-}
-
-bswap_f16 :: proc "contextless" (f: f16) -> f16 {
-	x := transmute(u16)f
-	z := bswap_16(x)
-	return transmute(f16)z
-
-}
-
-bswap_f32 :: proc "contextless" (f: f32) -> f32 {
-	x := transmute(u32)f
-	z := bswap_32(x)
-	return transmute(f32)z
-
-}
-
-bswap_f64 :: proc "contextless" (f: f64) -> f64 {
-	x := transmute(u64)f
-	z := bswap_64(x)
-	return transmute(f64)z
-}
-
-
-is_power_of_two_int :: #force_inline proc(x: int) -> bool {
+is_power_of_two_int :: #force_inline proc "contextless" (x: int) -> bool {
 	if x <= 0 {
 		return false
 	}
@@ -81,7 +40,7 @@ align_forward_int :: #force_inline proc(ptr, align: int) -> int {
 	return p
 }

-is_power_of_two_uintptr :: #force_inline proc(x: uintptr) -> bool {
+is_power_of_two_uintptr :: #force_inline proc "contextless" (x: uintptr) -> bool {
 	if x <= 0 {
 		return false
 	}
@@ -184,7 +143,7 @@ mem_free_all :: #force_inline proc(allocator := context.allocator, loc := #calle
 	return
 }

-mem_resize :: proc(ptr: rawptr, old_size, new_size: int, alignment: int = DEFAULT_ALIGNMENT, allocator := context.allocator, loc := #caller_location) -> (data: []byte, err: Allocator_Error) {
+_mem_resize :: #force_inline proc(ptr: rawptr, old_size, new_size: int, alignment: int = DEFAULT_ALIGNMENT, allocator := context.allocator, should_zero: bool, loc := #caller_location) -> (data: []byte, err: Allocator_Error) {
 	if allocator.procedure == nil {
 		return nil, nil
 	}
@@ -195,15 +154,27 @@ mem_resize :: proc(ptr: rawptr, old_size, new_size: int, alignment: int = DEFAUL
 		}
 		return
 	} else if ptr == nil {
-		return allocator.procedure(allocator.data, .Alloc, new_size, alignment, nil, 0, loc)
+		if should_zero {
+			return allocator.procedure(allocator.data, .Alloc, new_size, alignment, nil, 0, loc)
+		} else {
+			return allocator.procedure(allocator.data, .Alloc_Non_Zeroed, new_size, alignment, nil, 0, loc)
+		}
 	} else if old_size == new_size && uintptr(ptr) % uintptr(alignment) == 0 {
 		data = ([^]byte)(ptr)[:old_size]
 		return
 	}

-	data, err = allocator.procedure(allocator.data, .Resize, new_size, alignment, ptr, old_size, loc)
+	if should_zero {
+		data, err = allocator.procedure(allocator.data, .Resize, new_size, alignment, ptr, old_size, loc)
+	} else {
+		data, err = allocator.procedure(allocator.data, .Resize_Non_Zeroed, new_size, alignment, ptr, old_size, loc)
+	}
 	if err == .Mode_Not_Implemented {
-		data, err = allocator.procedure(allocator.data, .Alloc, new_size, alignment, nil, 0, loc)
+		if should_zero {
+			data, err = allocator.procedure(allocator.data, .Alloc, new_size, alignment, nil, 0, loc)
+		} else {
+			data, err = allocator.procedure(allocator.data, .Alloc_Non_Zeroed, new_size, alignment, nil, 0, loc)
+		}
 		if err != nil {
 			return
 		}
@@ -213,15 +184,30 @@ mem_resize :: proc(ptr: rawptr, old_size, new_size: int, alignment: int = DEFAUL
 	return
 }

+mem_resize :: proc(ptr: rawptr, old_size, new_size: int, alignment: int = DEFAULT_ALIGNMENT, allocator := context.allocator, loc := #caller_location) -> (data: []byte, err: Allocator_Error) {
+	return _mem_resize(ptr, old_size, new_size, alignment, allocator, true, loc)
+}
+non_zero_mem_resize :: proc(ptr: rawptr, old_size, new_size: int, alignment: int = DEFAULT_ALIGNMENT, allocator := context.allocator, loc := #caller_location) -> (data: []byte, err: Allocator_Error) {
+	return _mem_resize(ptr, old_size, new_size, alignment, allocator, false, loc)
+}
+
 memory_equal :: proc "contextless" (x, y: rawptr, n: int) -> bool {
 	switch {
 	case n == 0: return true
 	case x == y: return true
 	}
-	
 	a, b := ([^]byte)(x), ([^]byte)(y)
 	length := uint(n)
+
+	for i := uint(0); i < length; i += 1 {
+		if a[i] != b[i] {
+			return false
+		}
+	}
+	return true
 	
+/*
+
 	when size_of(uint) == 8 {
 		if word_length := length >> 3; word_length != 0 {
 			for _ in 0..<word_length {
@@ -276,6 +262,7 @@ memory_equal :: proc "contextless" (x, y: rawptr, n: int) -> bool {

 		return true
 	}
+*/

 }
 memory_compare :: proc "contextless" (a, b: rawptr, n: int) -> int #no_bounds_check {
@@ -401,6 +388,48 @@ cstring_to_string :: proc "contextless" (s: cstring) -> string {
 }


+cstring_eq :: proc "contextless" (lhs, rhs: cstring) -> bool {
+	x := ([^]byte)(lhs)
+	y := ([^]byte)(rhs)
+	if x == y {
+		return true
+	}
+	if (x == nil) ~ (y == nil) {
+		return false
+	}
+	xn := cstring_len(lhs)
+	yn := cstring_len(rhs)
+	if xn != yn {
+		return false
+	}
+	return #force_inline memory_equal(x, y, xn)
+}
+
+cstring_cmp :: proc "contextless" (lhs, rhs: cstring) -> int {
+	x := ([^]byte)(lhs)
+	y := ([^]byte)(rhs)
+	if x == y {
+		return 0
+	}
+	if (x == nil) ~ (y == nil) {
+		return -1 if x == nil else +1
+	}
+	xn := cstring_len(lhs)
+	yn := cstring_len(rhs)
+	ret := memory_compare(x, y, min(xn, yn))
+	if ret == 0 && xn != yn {
+		return -1 if xn < yn else +1
+	}
+	return ret
+}
+
+cstring_ne :: #force_inline proc "contextless" (a, b: cstring) -> bool { return !cstring_eq(a, b) }
+cstring_lt :: #force_inline proc "contextless" (a, b: cstring) -> bool { return cstring_cmp(a, b) < 0 }
+cstring_gt :: #force_inline proc "contextless" (a, b: cstring) -> bool { return cstring_cmp(a, b) > 0 }
+cstring_le :: #force_inline proc "contextless" (a, b: cstring) -> bool { return cstring_cmp(a, b) <= 0 }
+cstring_ge :: #force_inline proc "contextless" (a, b: cstring) -> bool { return cstring_cmp(a, b) >= 0 }
+
+
 complex32_eq :: #force_inline proc "contextless"  (a, b: complex32)  -> bool { return real(a) == real(b) && imag(a) == imag(b) }
 complex32_ne :: #force_inline proc "contextless"  (a, b: complex32)  -> bool { return real(a) != real(b) || imag(a) != imag(b) }

@@ -500,46 +529,73 @@ string_decode_rune :: #force_inline proc "contextless" (s: string) -> (rune, int
 	return rune(s0&MASK4)<<18 | rune(b1&MASKX)<<12 | rune(b2&MASKX)<<6 | rune(b3&MASKX), 4
 }

-abs_f16 :: #force_inline proc "contextless" (x: f16) -> f16 {
-	return -x if x < 0 else x
-}
-abs_f32 :: #force_inline proc "contextless" (x: f32) -> f32 {
-	return -x if x < 0 else x
-}
-abs_f64 :: #force_inline proc "contextless" (x: f64) -> f64 {
-	return -x if x < 0 else x
-}
+string_decode_last_rune :: proc "contextless" (s: string) -> (rune, int) {
+	RUNE_ERROR :: '\ufffd'
+	RUNE_SELF  :: 0x80
+	UTF_MAX    :: 4

-min_f16 :: #force_inline proc "contextless" (a, b: f16) -> f16 {
-	return a if a < b else b
-}
-min_f32 :: #force_inline proc "contextless" (a, b: f32) -> f32 {
-	return a if a < b else b
-}
-min_f64 :: #force_inline proc "contextless" (a, b: f64) -> f64 {
-	return a if a < b else b
-}
-max_f16 :: #force_inline proc "contextless" (a, b: f16) -> f16 {
-	return a if a > b else b
-}
-max_f32 :: #force_inline proc "contextless" (a, b: f32) -> f32 {
-	return a if a > b else b
-}
-max_f64 :: #force_inline proc "contextless" (a, b: f64) -> f64 {
-	return a if a > b else b
+	r: rune
+	size: int
+	start, end, limit: int
+
+	end = len(s)
+	if end == 0 {
+		return RUNE_ERROR, 0
+	}
+	start = end-1
+	r = rune(s[start])
+	if r < RUNE_SELF {
+		return r, 1
+	}
+
+	limit = max(end - UTF_MAX, 0)
+
+	for start-=1; start >= limit; start-=1 {
+		if (s[start] & 0xc0) != RUNE_SELF {
+			break
+		}
+	}
+
+	start = max(start, 0)
+	r, size = string_decode_rune(s[start:end])
+	if start+size != end {
+		return RUNE_ERROR, 1
+	}
+	return r, size
 }

 abs_complex32 :: #force_inline proc "contextless" (x: complex32) -> f16 {
-	r, i := real(x), imag(x)
-	return f16(intrinsics.sqrt(f32(r*r + i*i)))
+	p, q := abs(real(x)), abs(imag(x))
+	if p < q {
+		p, q = q, p
+	}
+	if p == 0 {
+		return 0
+	}
+	q = q / p
+	return p * f16(intrinsics.sqrt(f32(1 + q*q)))
 }
 abs_complex64 :: #force_inline proc "contextless" (x: complex64) -> f32 {
-	r, i := real(x), imag(x)
-	return intrinsics.sqrt(r*r + i*i)
+	p, q := abs(real(x)), abs(imag(x))
+	if p < q {
+		p, q = q, p
+	}
+	if p == 0 {
+		return 0
+	}
+	q = q / p
+	return p * intrinsics.sqrt(1 + q*q)
 }
 abs_complex128 :: #force_inline proc "contextless" (x: complex128) -> f64 {
-	r, i := real(x), imag(x)
-	return intrinsics.sqrt(r*r + i*i)
+	p, q := abs(real(x)), abs(imag(x))
+	if p < q {
+		p, q = q, p
+	}
+	if p == 0 {
+		return 0
+	}
+	q = q / p
+	return p * intrinsics.sqrt(1 + q*q)
 }
 abs_quaternion64 :: #force_inline proc "contextless" (x: quaternion64) -> f16 {
 	r, i, j, k := real(x), imag(x), jmag(x), kmag(x)
@@ -619,7 +675,7 @@ mul_quaternion64 :: proc "contextless" (q, r: quaternion64) -> quaternion64 {
 	t2 := r0*q2 + r1*q3 + r2*q0 - r3*q1
 	t3 := r0*q3 - r1*q2 + r2*q1 + r3*q0

-	return quaternion(t0, t1, t2, t3)
+	return quaternion(w=t0, x=t1, y=t2, z=t3)
 }

 mul_quaternion128 :: proc "contextless" (q, r: quaternion128) -> quaternion128 {
@@ -631,7 +687,7 @@ mul_quaternion128 :: proc "contextless" (q, r: quaternion128) -> quaternion128 {
 	t2 := r0*q2 + r1*q3 + r2*q0 - r3*q1
 	t3 := r0*q3 - r1*q2 + r2*q1 + r3*q0

-	return quaternion(t0, t1, t2, t3)
+	return quaternion(w=t0, x=t1, y=t2, z=t3)
 }

 mul_quaternion256 :: proc "contextless" (q, r: quaternion256) -> quaternion256 {
@@ -643,7 +699,7 @@ mul_quaternion256 :: proc "contextless" (q, r: quaternion256) -> quaternion256 {
 	t2 := r0*q2 + r1*q3 + r2*q0 - r3*q1
 	t3 := r0*q3 - r1*q2 + r2*q1 + r3*q0

-	return quaternion(t0, t1, t2, t3)
+	return quaternion(w=t0, x=t1, y=t2, z=t3)
 }

 quo_quaternion64 :: proc "contextless" (q, r: quaternion64) -> quaternion64 {
@@ -657,7 +713,7 @@ quo_quaternion64 :: proc "contextless" (q, r: quaternion64) -> quaternion64 {
 	t2 := (r0*q2 - r1*q3 - r2*q0 + r3*q1) * invmag2
 	t3 := (r0*q3 + r1*q2 + r2*q1 - r3*q0) * invmag2

-	return quaternion(t0, t1, t2, t3)
+	return quaternion(w=t0, x=t1, y=t2, z=t3)
 }

 quo_quaternion128 :: proc "contextless" (q, r: quaternion128) -> quaternion128 {
@@ -671,7 +727,7 @@ quo_quaternion128 :: proc "contextless" (q, r: quaternion128) -> quaternion128 {
 	t2 := (r0*q2 - r1*q3 - r2*q0 + r3*q1) * invmag2
 	t3 := (r0*q3 + r1*q2 + r2*q1 - r3*q0) * invmag2

-	return quaternion(t0, t1, t2, t3)
+	return quaternion(w=t0, x=t1, y=t2, z=t3)
 }

 quo_quaternion256 :: proc "contextless" (q, r: quaternion256) -> quaternion256 {
@@ -685,11 +741,11 @@ quo_quaternion256 :: proc "contextless" (q, r: quaternion256) -> quaternion256 {
 	t2 := (r0*q2 - r1*q3 - r2*q0 + r3*q1) * invmag2
 	t3 := (r0*q3 + r1*q2 + r2*q1 - r3*q0) * invmag2

-	return quaternion(t0, t1, t2, t3)
+	return quaternion(w=t0, x=t1, y=t2, z=t3)
 }

@(link_name="__truncsfhf2", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
-truncsfhf2 :: proc "c" (value: f32) -> u16 {
+truncsfhf2 :: proc "c" (value: f32) -> __float16 {
 	v: struct #raw_union { i: u32, f: f32 }
 	i, s, e, m: i32

@@ -703,7 +759,7 @@ truncsfhf2 :: proc "c" (value: f32) -> u16 {

 	if e <= 0 {
 		if e < -10 {
-			return u16(s)
+			return transmute(__float16)u16(s)
 		}
 		m = (m | 0x00800000) >> u32(1 - e)

@@ -711,14 +767,14 @@ truncsfhf2 :: proc "c" (value: f32) -> u16 {
 			m += 0x00002000
 		}

-		return u16(s | (m >> 13))
+		return transmute(__float16)u16(s | (m >> 13))
 	} else if e == 0xff - (127 - 15) {
 		if m == 0 {
-			return u16(s | 0x7c00) /* NOTE(bill): infinity */
+			return transmute(__float16)u16(s | 0x7c00) /* NOTE(bill): infinity */
 		} else {
 			/* NOTE(bill): NAN */
 			m >>= 13
-			return u16(s | 0x7c00 | m | i32(m == 0))
+			return transmute(__float16)u16(s | 0x7c00 | m | i32(m == 0))
 		}
 	} else {
 		if m & 0x00001000 != 0 {
@@ -738,23 +794,24 @@ truncsfhf2 :: proc "c" (value: f32) -> u16 {
 				intrinsics.volatile_store(&f, g)
 			}

-			return u16(s | 0x7c00)
+			return transmute(__float16)u16(s | 0x7c00)
 		}

-		return u16(s | (e << 10) | (m >> 13))
+		return transmute(__float16)u16(s | (e << 10) | (m >> 13))
 	}
 }


@(link_name="__truncdfhf2", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
-truncdfhf2 :: proc "c" (value: f64) -> u16 {
+truncdfhf2 :: proc "c" (value: f64) -> __float16 {
 	return truncsfhf2(f32(value))
 }

@(link_name="__gnu_h2f_ieee", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
-gnu_h2f_ieee :: proc "c" (value: u16) -> f32 {
+gnu_h2f_ieee :: proc "c" (value_: __float16) -> f32 {
 	fp32 :: struct #raw_union { u: u32, f: f32 }

+	value := transmute(u16)value_
 	v: fp32
 	magic, inf_or_nan: fp32
 	magic.u = u32((254 - 15) << 23)
@@ -771,12 +828,12 @@ gnu_h2f_ieee :: proc "c" (value: u16) -> f32 {


@(link_name="__gnu_f2h_ieee", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
-gnu_f2h_ieee :: proc "c" (value: f32) -> u16 {
+gnu_f2h_ieee :: proc "c" (value: f32) -> __float16 {
 	return truncsfhf2(value)
 }

@(link_name="__extendhfsf2", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
-extendhfsf2 :: proc "c" (value: u16) -> f32 {
+extendhfsf2 :: proc "c" (value: __float16) -> f32 {
 	return gnu_h2f_ieee(value)
 }

@@ -905,9 +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)
+	}
 }


@@ -977,3 +1036,23 @@ fixdfti :: proc(a: u64) -> i128 {
 	}

 }
+
+
+
+__write_bits :: proc "contextless" (dst, src: [^]byte, offset: uintptr, size: uintptr) {
+	for i in 0..<size {
+		j := offset+i
+		the_bit := byte((src[i>>3]) & (1<<(i&7)) != 0)
+		dst[j>>3] &~=       1<<(j&7)
+		dst[j>>3]  |= the_bit<<(j&7)
+	}
+}
+
+__read_bits :: proc "contextless" (dst, src: [^]byte, offset: uintptr, size: uintptr) {
+	for j in 0..<size {
+		i := offset+j
+		the_bit := byte((src[i>>3]) & (1<<(i&7)) != 0)
+		dst[j>>3] &~=       1<<(j&7)
+		dst[j>>3]  |= the_bit<<(j&7)
+	}
+}
@@ -0,0 +1,7 @@
+package runtime
+
+_OS_Errno :: distinct int
+
+stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	return _stderr_write(data)
+}
@@ -0,0 +1,22 @@
+//+build freebsd, openbsd
+//+private
+package runtime
+
+foreign import libc "system:c"
+
+@(default_calling_convention="c")
+foreign libc {
+	@(link_name="write")
+	_unix_write :: proc(fd: i32, buf: rawptr, size: int) -> int ---
+
+	__error :: 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 := __error()
+		return int(ret), _OS_Errno(err^ if err != nil else 0)
+	}
+	return int(ret), 0
+}
@@ -0,0 +1,15 @@
+//+build darwin
+//+private
+package runtime
+
+import "base:intrinsics"
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	WRITE  :: 0x2000004
+	STDERR :: 2
+	ret := intrinsics.syscall(WRITE, STDERR, uintptr(raw_data(data)), uintptr(len(data)))
+	if ret < 0 {
+		return 0, _OS_Errno(-ret)
+	}
+	return int(ret), 0
+}
@@ -1,7 +1,8 @@
 //+build freestanding
+//+private
 package runtime

 // TODO(bill): reimplement `os.write`
-_os_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
 	return 0, -1
 }
@@ -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
+}
@@ -0,0 +1,13 @@
+//+build js
+//+private
+package runtime
+
+foreign import "odin_env"
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	foreign odin_env {
+		write :: proc "contextless" (fd: u32, p: []byte) ---
+	}
+	write(1, data)
+	return len(data), 0
+}
@@ -0,0 +1,24 @@
+//+private
+package runtime
+
+import "base:intrinsics"
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	when ODIN_ARCH == .amd64 {
+		SYS_write :: uintptr(1)
+	} else when ODIN_ARCH == .arm64 {
+		SYS_write :: uintptr(64)
+	} else when ODIN_ARCH == .i386 {
+		SYS_write :: uintptr(4)
+	} else when ODIN_ARCH == .arm32 {
+		SYS_write :: uintptr(4)
+	}
+
+	stderr :: 2
+
+	ret := int(intrinsics.syscall(SYS_write, uintptr(stderr), uintptr(raw_data(data)), uintptr(len(data))))
+	if ret < 0 && ret > -4096 {
+		return 0, _OS_Errno(-ret)
+	}
+	return ret, 0
+}
@@ -1,9 +1,10 @@
 //+build wasi
+//+private
 package runtime

 import "core:sys/wasm/wasi"

-_os_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
 	data := (wasi.ciovec_t)(data)
 	n, err := wasi.fd_write(1, {data})
 	return int(n), _OS_Errno(err)
@@ -0,0 +1,51 @@
+//+build windows
+//+private
+package runtime
+
+foreign import kernel32 "system:Kernel32.lib"
+
+@(private="file")
+@(default_calling_convention="system")
+foreign kernel32 {
+	// NOTE(bill): The types are not using the standard names (e.g. DWORD and LPVOID) to just minimizing the dependency
+
+	// stderr_write
+	GetStdHandle         :: proc(which: u32) -> rawptr ---
+	SetHandleInformation :: proc(hObject: rawptr, dwMask: u32, dwFlags: u32) -> b32 ---
+	WriteFile            :: proc(hFile: rawptr, lpBuffer: rawptr, nNumberOfBytesToWrite: u32, lpNumberOfBytesWritten: ^u32, lpOverlapped: rawptr) -> b32 ---
+	GetLastError         :: proc() -> u32 ---
+}
+
+_stderr_write :: proc "contextless" (data: []byte) -> (n: int, err: _OS_Errno) #no_bounds_check {
+	if len(data) == 0 {
+		return 0, 0
+	}
+
+	STD_ERROR_HANDLE :: ~u32(0) -12 + 1
+	HANDLE_FLAG_INHERIT :: 0x00000001
+	MAX_RW :: 1<<30
+
+	h := GetStdHandle(STD_ERROR_HANDLE)
+	when size_of(uintptr) == 8 {
+		SetHandleInformation(h, HANDLE_FLAG_INHERIT, 0)
+	}
+
+	single_write_length: u32
+	total_write: i64
+	length := i64(len(data))
+
+	for total_write < length {
+		remaining := length - total_write
+		to_write := u32(min(i32(remaining), MAX_RW))
+
+		e := WriteFile(h, &data[total_write], to_write, &single_write_length, nil)
+		if single_write_length <= 0 || !e {
+			err = _OS_Errno(GetLastError())
+			n = int(total_write)
+			return
+		}
+		total_write += i64(single_write_length)
+	}
+	n = int(total_write)
+	return
+}
@@ -5,9 +5,14 @@ _INTEGER_DIGITS :: "0123456789abcdefghijklmnopqrstuvwxyz"
@(private="file")
 _INTEGER_DIGITS_VAR := _INTEGER_DIGITS

-when !ODIN_DISALLOW_RTTI {
-	print_any_single :: proc "contextless" (arg: any) {
+when !ODIN_NO_RTTI {
+	print_any_single :: #force_no_inline proc "contextless" (arg: any) {
 		x := arg
+		if x.data == nil {
+			print_string("nil")
+			return
+		}
+
 		if loc, ok := x.(Source_Code_Location); ok {
 			print_caller_location(loc)
 			return
@@ -48,6 +53,7 @@ when !ODIN_DISALLOW_RTTI {
 		case int:     print_int(v)
 		case uint:    print_uint(v)
 		case uintptr: print_uintptr(v)
+		case rawptr:  print_uintptr(uintptr(v))

 		case bool: print_string("true" if v else "false")
 		case b8:   print_string("true" if v else "false")
@@ -58,7 +64,7 @@ when !ODIN_DISALLOW_RTTI {
 		case:
 			ti := type_info_of(x.id)
 			#partial switch v in ti.variant {
-			case Type_Info_Pointer:
+			case Type_Info_Pointer, Type_Info_Multi_Pointer:
 				print_uintptr((^uintptr)(x.data)^)
 				return
 			}
@@ -66,8 +72,10 @@ when !ODIN_DISALLOW_RTTI {
 			print_string("<invalid-value>")
 		}
 	}
-	println_any :: proc "contextless" (args: ..any) {
+	println_any :: #force_no_inline proc "contextless" (args: ..any) {
+		context = default_context()
 		loop: for arg, i in args {
+			assert(arg.id != nil)
 			if i != 0 {
 				print_string(" ")
 			}
@@ -114,14 +122,14 @@ encode_rune :: proc "contextless" (c: rune) -> ([4]u8, int) {
 	return buf, 4
 }

-print_string :: proc "contextless" (str: string) -> (n: int) {
-	n, _ = os_write(transmute([]byte)str)
+print_string :: #force_no_inline proc "contextless" (str: string) -> (n: int) {
+	n, _ = stderr_write(transmute([]byte)str)
 	return
 }

-print_strings :: proc "contextless" (args: ..string) -> (n: int) {
+print_strings :: #force_no_inline proc "contextless" (args: ..string) -> (n: int) {
 	for str in args {
-		m, err := os_write(transmute([]byte)str)
+		m, err := stderr_write(transmute([]byte)str)
 		n += m
 		if err != 0 {
 			break
@@ -130,12 +138,12 @@ print_strings :: proc "contextless" (args: ..string) -> (n: int) {
 	return
 }

-print_byte :: proc "contextless" (b: byte) -> (n: int) {
-	n, _ = os_write([]byte{b})
+print_byte :: #force_no_inline proc "contextless" (b: byte) -> (n: int) {
+	n, _ = stderr_write([]byte{b})
 	return
 }

-print_encoded_rune :: proc "contextless" (r: rune) {
+print_encoded_rune :: #force_no_inline proc "contextless" (r: rune) {
 	print_byte('\'')

 	switch r {
@@ -162,7 +170,7 @@ print_encoded_rune :: proc "contextless" (r: rune) {
 	print_byte('\'')
 }

-print_rune :: proc "contextless" (r: rune) -> int #no_bounds_check {
+print_rune :: #force_no_inline proc "contextless" (r: rune) -> int #no_bounds_check {
 	RUNE_SELF :: 0x80

 	if r < RUNE_SELF {
@@ -170,12 +178,12 @@ print_rune :: proc "contextless" (r: rune) -> int #no_bounds_check {
 	}

 	b, n := encode_rune(r)
-	m, _ := os_write(b[:n])
+	m, _ := stderr_write(b[:n])
 	return m
 }


-print_u64 :: proc "contextless" (x: u64) #no_bounds_check {
+print_u64 :: #force_no_inline proc "contextless" (x: u64) #no_bounds_check {
 	a: [129]byte
 	i := len(a)
 	b := u64(10)
@@ -186,11 +194,11 @@ print_u64 :: proc "contextless" (x: u64) #no_bounds_check {
 	}
 	i -= 1; a[i] = _INTEGER_DIGITS_VAR[u % b]

-	os_write(a[i:])
+	stderr_write(a[i:])
 }


-print_i64 :: proc "contextless" (x: i64) #no_bounds_check {
+print_i64 :: #force_no_inline proc "contextless" (x: i64) #no_bounds_check {
 	b :: i64(10)

 	u := x
@@ -208,33 +216,37 @@ print_i64 :: proc "contextless" (x: i64) #no_bounds_check {
 		i -= 1; a[i] = '-'
 	}

-	os_write(a[i:])
+	stderr_write(a[i:])
 }

 print_uint    :: proc "contextless" (x: uint)    { print_u64(u64(x)) }
 print_uintptr :: proc "contextless" (x: uintptr) { print_u64(u64(x)) }
 print_int     :: proc "contextless" (x: int)     { print_i64(i64(x)) }

-print_caller_location :: proc "contextless" (using loc: Source_Code_Location) {
-	print_string(file_path)
+print_caller_location :: #force_no_inline proc "contextless" (loc: Source_Code_Location) {
+	print_string(loc.file_path)
 	when ODIN_ERROR_POS_STYLE == .Default {
 		print_byte('(')
-		print_u64(u64(line))
-		print_byte(':')
-		print_u64(u64(column))
+		print_u64(u64(loc.line))
+		if loc.column != 0 {
+			print_byte(':')
+			print_u64(u64(loc.column))
+		}
 		print_byte(')')
 	} else when ODIN_ERROR_POS_STYLE == .Unix {
 		print_byte(':')
-		print_u64(u64(line))
-		print_byte(':')
-		print_u64(u64(column))
+		print_u64(u64(loc.line))
+		if loc.column != 0 {
+			print_byte(':')
+			print_u64(u64(loc.column))
+		}
 		print_byte(':')
 	} else {
 		#panic("unhandled ODIN_ERROR_POS_STYLE")
 	}
 }
-print_typeid :: proc "contextless" (id: typeid) {
-	when ODIN_DISALLOW_RTTI {
+print_typeid :: #force_no_inline proc "contextless" (id: typeid) {
+	when ODIN_NO_RTTI {
 		if id == nil {
 			print_string("nil")
 		} else {
@@ -249,7 +261,9 @@ print_typeid :: proc "contextless" (id: typeid) {
 		}
 	}
 }
-print_type :: proc "contextless" (ti: ^Type_Info) {
+
+@(optimization_mode="size")
+print_type :: #force_no_inline proc "contextless" (ti: ^Type_Info) {
 	if ti == nil {
 		print_string("nil")
 		return
@@ -390,9 +404,9 @@ print_type :: proc "contextless" (ti: ^Type_Info) {
 		if info.is_packed    { print_string("#packed ") }
 		if info.is_raw_union { print_string("#raw_union ") }
 		if info.custom_align {
-			print_string("#align ")
+			print_string("#align(")
 			print_u64(u64(ti.align))
-			print_byte(' ')
+			print_string(") ")
 		}
 		print_byte('{')
 		for name, i in info.names {
@@ -406,8 +420,9 @@ print_type :: proc "contextless" (ti: ^Type_Info) {
 	case Type_Info_Union:
 		print_string("union ")
 		if info.custom_align {
-			print_string("#align ")
+			print_string("#align(")
 			print_u64(u64(ti.align))
+			print_string(") ")
 		}
 		if info.no_nil {
 			print_string("#no_nil ")
@@ -450,6 +465,20 @@ print_type :: proc "contextless" (ti: ^Type_Info) {
 		}
 		print_byte(']')

+	case Type_Info_Bit_Field:
+		print_string("bit_field ")
+		print_type(info.backing_type)
+		print_string(" {")
+		for name, i in info.names {
+			if i > 0 { print_string(", ") }
+			print_string(name)
+			print_string(": ")
+			print_type(info.types[i])
+			print_string(" | ")
+			print_u64(u64(info.bit_sizes[i]))
+		}
+		print_byte('}')
+

 	case Type_Info_Simd_Vector:
 		print_string("#simd[")
@@ -463,11 +492,11 @@ print_type :: proc "contextless" (ti: ^Type_Info) {
 		print_string(") ")
 		print_type(info.pointer)

-	case Type_Info_Relative_Slice:
+	case Type_Info_Relative_Multi_Pointer:
 		print_string("#relative(")
 		print_type(info.base_integer)
 		print_string(") ")
-		print_type(info.slice)
+		print_type(info.pointer)
 		
 	case Type_Info_Matrix:
 		print_string("matrix[")
@@ -4,7 +4,7 @@ when ODIN_NO_CRT && ODIN_OS == .Windows {
 	foreign import lib "system:NtDll.lib"
 	
 	@(private="file")
-	@(default_calling_convention="stdcall")
+	@(default_calling_convention="system")
 	foreign lib {
 		RtlMoveMemory :: proc(dst, s: rawptr, length: int) ---
 		RtlFillMemory :: proc(dst: rawptr, length: int, fill: i32) ---
@@ -25,19 +25,30 @@ when ODIN_NO_CRT && ODIN_OS == .Windows {
 		RtlMoveMemory(dst, src, len)
 		return dst
 	}
-} else when ODIN_NO_CRT || (ODIN_ARCH == .wasm32 || ODIN_ARCH == .wasm64) {
+} else when ODIN_NO_CRT || (ODIN_ARCH == .wasm32 || ODIN_ARCH == .wasm64p32) {
 	@(link_name="memset", linkage="strong", require)
 	memset :: proc "c" (ptr: rawptr, val: i32, len: int) -> rawptr {
 		if ptr != nil && len != 0 {
 			b := byte(val)
 			p := ([^]byte)(ptr)
-			for i in 0..<len {
+			for i := 0; i < len; i += 1 {
 				p[i] = b
 			}
 		}
 		return ptr
 	}
-	
+
+	@(link_name="bzero", linkage="strong", require)
+	bzero :: proc "c" (ptr: rawptr, len: int) -> rawptr {
+		if ptr != nil && len != 0 {
+			p := ([^]byte)(ptr)
+			for i := 0; i < len; i += 1 {
+				p[i] = 0
+			}
+		}
+		return ptr
+	}
+
 	@(link_name="memmove", linkage="strong", require)
 	memmove :: proc "c" (dst, src: rawptr, len: int) -> rawptr {
 		d, s := ([^]byte)(dst), ([^]byte)(src)
@@ -75,7 +86,7 @@ when ODIN_NO_CRT && ODIN_OS == .Windows {
 		if ptr != nil && len != 0 {
 			b := byte(val)
 			p := ([^]byte)(ptr)
-			for i in 0..<len {
+			for i := 0; i < len; i += 1 {
 				p[i] = b
 			}
 		}
@@ -3,7 +3,7 @@ package runtime

 foreign import "system:Foundation.framework"

-import "core:intrinsics"
+import "base:intrinsics"

 objc_id :: ^intrinsics.objc_object
 objc_Class :: ^intrinsics.objc_class
@@ -1,4 +1,4 @@
-//+build js wasm32
+//+build js
 package runtime

 init_default_context_for_js: Context
@@ -1,4 +1,4 @@
-//+build wasm32
+//+build wasm32, wasm64p32
 package runtime

@(private="file")
@@ -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 "c" (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 "c" (a: i128, b_: u32) -> i128 {


@(link_name="__multi3", linkage="strong")
-__multi3 :: proc "c" (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)
+}
@@ -0,0 +1,79 @@
+bits 64
+
+global __chkstk
+global _tls_index
+global _fltused
+
+section .data
+	_tls_index: dd 0
+	_fltused:   dd 0x9875
+
+section .text
+; NOTE(flysand): The function call to __chkstk is called
+; by the compiler, when we're allocating arrays larger than
+; a page size. The reason is because the OS doesn't map the
+; whole stack into memory all at once, but does so page-by-page.
+; When the next page is touched, the CPU generates a page fault,
+; which *the OS* is handling by allocating the next page in the
+; stack until we reach the limit of stack size.
+;
+; This page is called the guard page, touching it will extend
+; the size of the stack and overwrite the stack limit in the TEB.
+;
+; If we allocate a large enough array and start writing from the
+; bottom of it, it's possible that we may start touching
+; non-contiguous pages which are unmapped. OS only maps the stack
+; page into the memory if the page above it was also mapped.
+;
+; Therefore the compilers insert this routine, the sole purpose
+; of which is to step through the stack starting from the RSP
+; down to the new RSP after allocation, and touch every page
+; of the new allocation so that the stack is fully mapped for
+; the new allocation
+;
+; I've gotten this code by disassembling the output of MSVC long
+; time ago. I don't remember if I've cleaned it up, but it definately
+; stinks.
+;
+; Additional notes:
+;   RAX (passed as parameter) holds the allocation's size
+;   GS:[0x10] references the current stack limit
+;     (i.e. bottom of the stack (i.e. lowest address accessible))
+;
+; Also this stuff is windows-only kind of thing, because linux people
+; didn't think stack that grows is cool enough for them, but the kernel
+; totally supports this kind of stack.
+__chkstk:
+	;; Allocate 16 bytes to store values of r10 and r11
+	sub   rsp, 0x10
+	mov   [rsp], r10
+	mov   [rsp+0x8], r11
+	;; Set r10 to point to the stack as of the moment of the function call
+	lea   r10, [rsp+0x18]
+	;; Subtract r10 til the bottom of the stack allocation, if we overflow
+	;; reset r10 to 0, we'll crash with segfault anyway
+	xor   r11, r11
+	sub   r10, rax
+	cmovb r10, r11
+	;; Load r11 with the bottom of the stack (lowest allocated address)
+	mov   r11, gs:[0x10] ; NOTE(flysand): gs:[0x10] is stack limit
+	;; If the bottom of the allocation is above the bottom of the stack,
+	;; we don't need to probe
+	cmp   r10, r11
+	jnb   .end
+	;; Align the bottom of the allocation down to page size
+	and   r10w, 0xf000
+.loop:
+	;; Move the pointer to the next guard page, and touch it by loading 0
+	;; into that page
+	lea   r11, [r11-0x1000]
+	mov   byte [r11], 0x0
+	;; Did we reach the bottom of the allocation?
+	cmp   r10, r11
+	jnz   .loop
+.end:
+	;; Restore previous r10 and r11 and return
+	mov   r10, [rsp]
+	mov   r11, [rsp+0x8]
+	add   rsp, 0x10
+	ret
@@ -1,11 +1,12 @@
 //+private
+//+no-instrumentation
 package runtime

 foreign import kernel32 "system:Kernel32.lib"

@(private)
 foreign kernel32 {
-	RaiseException :: proc "stdcall" (dwExceptionCode, dwExceptionFlags, nNumberOfArguments: u32, lpArguments: ^uint) -> ! ---
+	RaiseException :: proc "system" (dwExceptionCode, dwExceptionFlags, nNumberOfArguments: u32, lpArguments: ^uint) -> ! ---
 }

 windows_trap_array_bounds :: proc "contextless" () -> ! {
@@ -1,4 +1,5 @@
 //+private
+//+no-instrumentation
 package runtime

@require foreign import "system:int64.lib"
@@ -12,7 +13,7 @@ windows_trap_array_bounds :: proc "contextless" () -> ! {
 	EXCEPTION_ARRAY_BOUNDS_EXCEEDED :: 0xC000008C

 	foreign kernel32 {
-		RaiseException :: proc "stdcall" (dwExceptionCode, dwExceptionFlags, nNumberOfArguments: DWORD, lpArguments: ^ULONG_PTR) -> ! ---
+		RaiseException :: proc "system" (dwExceptionCode, dwExceptionFlags, nNumberOfArguments: DWORD, lpArguments: ^ULONG_PTR) -> ! ---
 	}

 	RaiseException(EXCEPTION_ARRAY_BOUNDS_EXCEEDED, 0, 0, nil)
@@ -1,6 +1,6 @@
 package runtime

-import "core:intrinsics"
+import "base:intrinsics"

 udivmod128 :: proc "c" (a, b: u128, rem: ^u128) -> u128 {
 	_ctz :: intrinsics.count_trailing_zeros
@@ -3,18 +3,20 @@
 setlocal EnableDelayedExpansion

 where /Q cl.exe || (
-  set __VSCMD_ARG_NO_LOGO=1
-  for /f "tokens=*" %%i in ('"C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe" -latest -requires Microsoft.VisualStudio.Workload.NativeDesktop -property installationPath') do set VS=%%i
-  if "!VS!" equ "" (
-    echo ERROR: Visual Studio installation not found
-    exit /b 1
-  )  
-  call "!VS!\VC\Auxiliary\Build\vcvarsall.bat" amd64 || exit /b 1
+	set __VSCMD_ARG_NO_LOGO=1
+	for /f "tokens=*" %%i in ('"C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe" -latest -requires Microsoft.VisualStudio.Workload.NativeDesktop -property installationPath') do set VS=%%i
+	if "!VS!" equ "" (
+		echo ERROR: Visual Studio installation not found
+		exit /b 1
+	)
+	call "!VS!\VC\Auxiliary\Build\vcvarsall.bat" amd64 || exit /b 1
 )

 if "%VSCMD_ARG_TGT_ARCH%" neq "x64" (
-  echo ERROR: please run this from MSVC x64 native tools command prompt, 32-bit target is not supported!
-  exit /b 1
+	if "%ODIN_IGNORE_MSVC_CHECK%" == "" (
+		echo ERROR: please run this from MSVC x64 native tools command prompt, 32-bit target is not supported!
+		exit /b 1
+	)
 )

 for /f "usebackq tokens=1,2 delims=,=- " %%i in (`wmic os get LocalDateTime /value`) do @if %%i==LocalDateTime (
@@ -48,8 +50,14 @@ set odin_version_raw="dev-%curr_year%-%curr_month%"
 set compiler_flags= -nologo -Oi -TP -fp:precise -Gm- -MP -FC -EHsc- -GR- -GF
 set compiler_defines= -DODIN_VERSION_RAW=\"%odin_version_raw%\"

-for /f %%i in ('git rev-parse --short HEAD') do set GIT_SHA=%%i
+if not exist .git\ goto skip_git_hash
+for /f "tokens=1,2" %%i IN ('git show "--pretty=%%cd %%h" "--date=format:%%Y-%%m" --no-patch --no-notes HEAD') do (
+	set odin_version_raw=dev-%%i
+	set GIT_SHA=%%j
+)
 if %ERRORLEVEL% equ 0 set compiler_defines=%compiler_defines% -DGIT_SHA=\"%GIT_SHA%\"
+:skip_git_hash
+
 if %nightly% equ 1 set compiler_defines=%compiler_defines% -DNIGHTLY

 if %release_mode% EQU 0 ( rem Debug
@@ -72,6 +80,16 @@ set libs= ^
 	Synchronization.lib ^
 	bin\llvm\windows\LLVM-C.lib

+rem DO NOT TOUCH!
+rem THIS TILDE STUFF IS FOR DEVELOPMENT ONLY!
+set tilde_backend=0
+if %tilde_backend% EQU 1 (
+	set libs=%libs% src\tilde\tb.lib
+	set compiler_defines=%compiler_defines% -DODIN_TILDE_BACKEND
+)
+rem DO NOT TOUCH!
+
+
 set linker_flags= -incremental:no -opt:ref -subsystem:console

 if %release_mode% EQU 0 ( rem Debug
@@ -92,8 +110,9 @@ if %errorlevel% neq 0 goto end_of_build
 call build_vendor.bat
 if %errorlevel% neq 0 goto end_of_build

-if %release_mode% EQU 0 odin run examples/demo
+rem If the demo doesn't run for you and your CPU is more than a decade old, try -microarch:native
+if %release_mode% EQU 0 odin run examples/demo -- Hellope World

 del *.obj > NUL 2> NUL

-:end_of_build
+:end_of_build
@@ -1,123 +1,98 @@
-#!/usr/bin/env bash
+#!/usr/bin/env sh
 set -eu

-: ${CXX=clang++}
 : ${CPPFLAGS=}
 : ${CXXFLAGS=}
 : ${LDFLAGS=}
-: ${ODIN_VERSION=dev-$(date +"%Y-%m")}
+: ${LLVM_CONFIG=}

-CPPFLAGS="$CPPFLAGS -DODIN_VERSION_RAW=\"$ODIN_VERSION\""
+CPPFLAGS="$CPPFLAGS -DODIN_VERSION_RAW=\"dev-$(date +"%Y-%m")\""
 CXXFLAGS="$CXXFLAGS -std=c++14"
-LDFLAGS="$LDFLAGS -pthread -lm -lstdc++"
-
-GIT_SHA=$(git rev-parse --short HEAD || :)
-if [ "$GIT_SHA" ]; then CPPFLAGS="$CPPFLAGS -DGIT_SHA=\"$GIT_SHA\""; fi
-
 DISABLED_WARNINGS="-Wno-switch -Wno-macro-redefined -Wno-unused-value"
-OS=$(uname)
+LDFLAGS="$LDFLAGS -pthread -lm -lstdc++"
+OS_ARCH="$(uname -m)"
+OS_NAME="$(uname -s)"

-panic() {
-	printf "%s\n" "$1"
+if [ -d ".git" ] && [ -n "$(command -v git)" ]; then
+	GIT_SHA=$(git show --pretty='%h' --no-patch --no-notes HEAD)
+	CPPFLAGS="$CPPFLAGS -DGIT_SHA=\"$GIT_SHA\""
+fi
+
+error() {
+	printf "ERROR: %s\n" "$1"
 	exit 1
 }

-version() { echo "$@" | awk -F. '{ printf("%d%03d%03d%03d\n", $1,$2,$3,$4); }'; }
-
-config_darwin() {
-	ARCH=$(uname -m)
-	: ${LLVM_CONFIG=llvm-config}
-
-	# allow for arm only llvm's with version 13
-	if [ ARCH == arm64 ]; then
-		MIN_LLVM_VERSION=("13.0.0")
+if [ -z "$LLVM_CONFIG" ]; then
+	# darwin, linux, openbsd
+	if   [ -n "$(command -v llvm-config-18)" ]; then LLVM_CONFIG="llvm-config-18"
+	elif [ -n "$(command -v llvm-config-17)" ]; then LLVM_CONFIG="llvm-config-17"
+	elif [ -n "$(command -v llvm-config-14)" ]; then LLVM_CONFIG="llvm-config-14"
+	elif [ -n "$(command -v llvm-config-13)" ]; then LLVM_CONFIG="llvm-config-13"
+	elif [ -n "$(command -v llvm-config-12)" ]; then LLVM_CONFIG="llvm-config-12"
+	elif [ -n "$(command -v llvm-config-11)" ]; then LLVM_CONFIG="llvm-config-11"
+	# freebsd
+	elif [ -n "$(command -v llvm-config18)" ]; then  LLVM_CONFIG="llvm-config18"
+	elif [ -n "$(command -v llvm-config17)" ]; then  LLVM_CONFIG="llvm-config17"
+	elif [ -n "$(command -v llvm-config14)" ]; then  LLVM_CONFIG="llvm-config14"
+	elif [ -n "$(command -v llvm-config13)" ]; then  LLVM_CONFIG="llvm-config13"
+	elif [ -n "$(command -v llvm-config12)" ]; then  LLVM_CONFIG="llvm-config12"
+	elif [ -n "$(command -v llvm-config11)" ]; then  LLVM_CONFIG="llvm-config11"
+	# fallback
+	elif [ -n "$(command -v llvm-config)" ]; then LLVM_CONFIG="llvm-config"
 	else
-		# allow for x86 / amd64 all llvm versions beginning from 11
-		MIN_LLVM_VERSION=("11.1.0")
+		error "No llvm-config command found. Set LLVM_CONFIG to proceed."
 	fi
+fi

-	if [ $(version $($LLVM_CONFIG --version)) -lt $(version $MIN_LLVM_VERSION) ]; then
-		if [ ARCH == arm64 ]; then
-			panic "Requirement: llvm-config must be base version 13 for arm64"
-		else
-			panic "Requirement: llvm-config must be base version greater than 11 for amd64/x86"
+: ${CXX=$($LLVM_CONFIG --bindir)/clang++}
+
+LLVM_VERSION="$($LLVM_CONFIG --version)"
+LLVM_VERSION_MAJOR="$(echo $LLVM_VERSION | awk -F. '{print $1}')"
+LLVM_VERSION_MINOR="$(echo $LLVM_VERSION | awk -F. '{print $2}')"
+LLVM_VERSION_PATCH="$(echo $LLVM_VERSION | awk -F. '{print $3}')"
+
+if [ $LLVM_VERSION_MAJOR -lt 11 ] || ([ $LLVM_VERSION_MAJOR -gt 14 ] && [ $LLVM_VERSION_MAJOR -lt 17 ]) || [ $LLVM_VERSION_MAJOR -gt 18 ]; then
+	error "Invalid LLVM version $LLVM_VERSION: must be 11, 12, 13, 14, 17 or 18"
+fi
+
+case "$OS_NAME" in
+Darwin)
+	if [ "$OS_ARCH" = "arm64" ]; then
+		if [ $LLVM_VERSION_MAJOR -lt 13 ]; then
+			error "Invalid LLVM version $LLVM_VERSION: Darwin Arm64 requires LLVM 13, 14, 17 or 18"
 		fi
 	fi

-	MAX_LLVM_VERSION=("14.999.999")
-	if [ $(version $($LLVM_CONFIG --version)) -gt $(version $MAX_LLVM_VERSION) ]; then
-		echo "Tried to use " $(which $LLVM_CONFIG) "version" $($LLVM_CONFIG --version)
-		panic "Requirement: llvm-config must be base version smaller than 15"
-	fi
-
-	LDFLAGS="$LDFLAGS -liconv -ldl -framework System"
 	CXXFLAGS="$CXXFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
-	LDFLAGS="$LDFLAGS -lLLVM-C"
-}
-
-config_freebsd() {
-	: ${LLVM_CONFIG=}
-
-	if [ ! "$LLVM_CONFIG" ]; then
-		if which llvm-config11 > /dev/null 2>&1; then
-			LLVM_CONFIG=llvm-config11
-		elif which llvm-config12 > /dev/null 2>&1; then
-			LLVM_CONFIG=llvm-config12
-		elif which llvm-config13 > /dev/null 2>&1; then
-			LLVM_CONFIG=llvm-config13
-		else
-			panic "Unable to find LLVM-config"
-		fi
-	fi
-
+	LDFLAGS="$LDFLAGS -liconv -ldl -framework System -lLLVM"
+	;;
+FreeBSD)
 	CXXFLAGS="$CXXFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
 	LDFLAGS="$LDFLAGS $($LLVM_CONFIG --libs core native --system-libs)"
-}
-
-config_openbsd() {
-	: ${LLVM_CONFIG=/usr/local/bin/llvm-config}
-
-	LDFLAGS="$LDFLAGS -liconv"
+	;;
+Linux)
 	CXXFLAGS="$CXXFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
-	LDFLAGS="$LDFLAGS $($LLVM_CONFIG --libs core native --system-libs)"
-}
-
-config_linux() {
-	: ${LLVM_CONFIG=}
-
-	if [ ! "$LLVM_CONFIG" ]; then
-		if which llvm-config > /dev/null 2>&1; then
-			LLVM_CONFIG=llvm-config
-		elif which llvm-config-11 > /dev/null 2>&1; then
-			LLVM_CONFIG=llvm-config-11
-		elif which llvm-config-11-64 > /dev/null 2>&1; then
-			LLVM_CONFIG=llvm-config-11-64
-		else
-			panic "Unable to find LLVM-config"
-		fi
-	fi
-
-	MIN_LLVM_VERSION=("11.0.0")
-	if [ $(version $($LLVM_CONFIG --version)) -lt $(version $MIN_LLVM_VERSION) ]; then
-		echo "Tried to use " $(which $LLVM_CONFIG) "version" $($LLVM_CONFIG --version)
-		panic "Requirement: llvm-config must be base version greater than 11"
-	fi
-
-	MAX_LLVM_VERSION=("14.999.999")
-	if [ $(version $($LLVM_CONFIG --version)) -gt $(version $MAX_LLVM_VERSION) ]; then
-		echo "Tried to use " $(which $LLVM_CONFIG) "version" $($LLVM_CONFIG --version)
-		panic "Requirement: llvm-config must be base version smaller than 15"
-	fi
-
-	LDFLAGS="$LDFLAGS -ldl"
-	CXXFLAGS="$CXXFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
-	LDFLAGS="$LDFLAGS $($LLVM_CONFIG  --libs core native --system-libs --libfiles) -Wl,-rpath=\$ORIGIN"
-
-	# Creates a copy of the llvm library in the build dir, this is meant to support compiler explorer.
-	# The annoyance is that this copy can be cluttering the development folder. TODO: split staging folders
-	# for development and compiler explorer builds
+	LDFLAGS="$LDFLAGS -ldl $($LLVM_CONFIG --libs core native --system-libs --libfiles)"
+	# Copy libLLVM*.so into current directory for linking
+	# NOTE: This is needed by the Linux release pipeline!
 	cp $(readlink -f $($LLVM_CONFIG --libfiles)) ./
-}
+	LDFLAGS="$LDFLAGS -Wl,-rpath=\$ORIGIN"
+	;;
+OpenBSD)
+	CXXFLAGS="$CXXFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
+	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"
+	;;
+esac

 build_odin() {
 	case $1 in
@@ -128,13 +103,20 @@ build_odin() {
 		EXTRAFLAGS="-O3"
 		;;
 	release-native)
-		EXTRAFLAGS="-O3 -march=native"
+		if [ "$OS_ARCH" = "arm64" ]; then
+			# Use preferred flag for Arm (ie arm64 / aarch64 / etc)
+			EXTRAFLAGS="-O3 -mcpu=native"
+		else
+			# Use preferred flag for x86 / amd64
+			EXTRAFLAGS="-O3 -march=native"
+		fi
 		;;
 	nightly)
 		EXTRAFLAGS="-DNIGHTLY -O3"
 		;;
 	*)
-		panic "Build mode unsupported!"
+		error "Build mode \"$1\" unsupported!"
+		;;
 	esac

 	set -x
@@ -143,57 +125,23 @@ build_odin() {
 }

 run_demo() {
-	./odin run examples/demo/demo.odin -file
+	./odin run examples/demo/demo.odin -file -- Hellope World
 }

-have_which() {
-	if ! which which > /dev/null 2>&1; then
-		panic "Could not find \`which\`"
-	fi
-}
-
-have_which
-
-case $OS in
-Linux)
-	config_linux
-	;;
-Darwin)
-	config_darwin
-	;;
-OpenBSD)
-	config_openbsd
-	;;
-FreeBSD)
-	config_freebsd
-	;;
-*)
-	panic "Platform unsupported!"
-esac
-
-if [[ $# -eq 0 ]]; then
+if [ $# -eq 0 ]; then
 	build_odin debug
 	run_demo
-	exit 0
-fi
-
-if [[ $# -eq 1 ]]; then
+elif [ $# -eq 1 ]; then
 	case $1 in
 	report)
-		if [[ ! -f "./odin" ]]; then
-			build_odin debug
-		fi
-
+		[ ! -f "./odin" ] && build_odin debug
 		./odin report
-		exit 0
 		;;
 	*)
 		build_odin $1
 		;;
 	esac
-
 	run_demo
-	exit 0
 else
-	panic "Too many arguments!"
+	error "Too many arguments!"
 fi
@@ -1,5 +1,7 @@
 #!/bin/bash

+set -e
+
 bucket=$1
 platform=$2
 artifact=$3
@@ -9,5 +11,15 @@ filename="odin-$platform-nightly+$now.zip"

 echo "Creating archive $filename from $artifact and uploading to $bucket"

-7z a -bd "output/$filename" -r "$artifact"
-b2 upload-file --noProgress "$bucket" "output/$filename" "nightly/$filename"
+# If this is already zipped up (done before artifact upload to keep permissions in tact), just move it.
+if [ "${artifact: -4}" == ".zip" ]
+then
+	echo "Artifact already a zip"
+	mkdir -p "output"
+	mv "$artifact" "output/$filename"
+else
+	echo "Artifact needs to be zipped"
+	7z a -bd "output/$filename" -r "$artifact"
+fi
+
+b2 upload-file --noProgress "$bucket" "output/$filename" "nightly/$filename"
@@ -2,25 +2,25 @@ package bufio

 import "core:io"

-// Loadahead_Reader provides io lookahead.
+// Lookahead_Reader provides io lookahead.
 // This is useful for tokenizers/parsers.
-// Loadahead_Reader is similar to bufio.Reader, but unlike bufio.Reader, Loadahead_Reader's buffer size
+// Lookahead_Reader is similar to bufio.Reader, but unlike bufio.Reader, Lookahead_Reader's buffer size
 // will EXACTLY match the specified size, whereas bufio.Reader's buffer size may differ from the specified size.
 // This makes sure that the buffer will not be accidentally read beyond the expected size.
-Loadahead_Reader :: struct {
+Lookahead_Reader :: struct {
 	r:   io.Reader,
 	buf: []byte,
 	n:   int,
 }

-lookahead_reader_init :: proc(lr: ^Loadahead_Reader, r: io.Reader, buf: []byte) -> ^Loadahead_Reader {
+lookahead_reader_init :: proc(lr: ^Lookahead_Reader, r: io.Reader, buf: []byte) -> ^Lookahead_Reader {
 	lr.r = r
 	lr.buf = buf
 	lr.n = 0
 	return lr
 }

-lookahead_reader_buffer :: proc(lr: ^Loadahead_Reader) -> []byte {
+lookahead_reader_buffer :: proc(lr: ^Lookahead_Reader) -> []byte {
 	return lr.buf[:lr.n]
 }

@@ -28,7 +28,7 @@ lookahead_reader_buffer :: proc(lr: ^Loadahead_Reader) -> []byte {
 // lookahead_reader_peek returns a slice of the Lookahead_Reader which holds n bytes
 // If the Lookahead_Reader cannot hold enough bytes, it will read from the underlying reader to populate the rest.
 // NOTE: The returned buffer is not a copy of the underlying buffer
-lookahead_reader_peek :: proc(lr: ^Loadahead_Reader, n: int) -> ([]byte, io.Error) {
+lookahead_reader_peek :: proc(lr: ^Lookahead_Reader, n: int) -> ([]byte, io.Error) {
 	switch {
 	case n < 0:
 		return nil, .Negative_Read
@@ -58,13 +58,13 @@ lookahead_reader_peek :: proc(lr: ^Loadahead_Reader, n: int) -> ([]byte, io.Erro
 // lookahead_reader_peek_all returns a slice of the Lookahead_Reader populating the full buffer
 // If the Lookahead_Reader cannot hold enough bytes, it will read from the underlying reader to populate the rest.
 // NOTE: The returned buffer is not a copy of the underlying buffer
-lookahead_reader_peek_all :: proc(lr: ^Loadahead_Reader) -> ([]byte, io.Error) {
+lookahead_reader_peek_all :: proc(lr: ^Lookahead_Reader) -> ([]byte, io.Error) {
 	return lookahead_reader_peek(lr, len(lr.buf))
 }


 // lookahead_reader_consume drops the first n populated bytes from the Lookahead_Reader.
-lookahead_reader_consume :: proc(lr: ^Loadahead_Reader, n: int) -> io.Error {
+lookahead_reader_consume :: proc(lr: ^Lookahead_Reader, n: int) -> io.Error {
 	switch {
 	case n == 0:
 		return nil
@@ -78,6 +78,6 @@ lookahead_reader_consume :: proc(lr: ^Loadahead_Reader, n: int) -> io.Error {
 	return nil
 }

-lookahead_reader_consume_all :: proc(lr: ^Loadahead_Reader) -> io.Error {
+lookahead_reader_consume_all :: proc(lr: ^Lookahead_Reader) -> io.Error {
 	return lookahead_reader_consume(lr, lr.n)
 }
@@ -14,51 +14,29 @@ read_writer_init :: proc(rw: ^Read_Writer, r: ^Reader, w: ^Writer) {
 }

 read_writer_to_stream :: proc(rw: ^Read_Writer) -> (s: io.Stream) {
-	s.stream_data = rw
-	s.stream_vtable = &_read_writer_vtable
+	s.procedure = _read_writer_procedure
+	s.data = rw
 	return
 }

@(private)
-_read_writer_vtable := io.Stream_VTable{
-	impl_read = proc(s: io.Stream, p: []byte) -> (n: int, err: io.Error) {
-		b := (^Read_Writer)(s.stream_data).r
-		return reader_read(b, p)
-	},
-	impl_unread_byte = proc(s: io.Stream) -> io.Error {
-		b := (^Read_Writer)(s.stream_data).r
-		return reader_unread_byte(b)
-	},
-	impl_read_rune = proc(s: io.Stream) -> (r: rune, size: int, err: io.Error) {
-		b := (^Read_Writer)(s.stream_data).r
-		return reader_read_rune(b)
-	},
-	impl_unread_rune = proc(s: io.Stream) -> io.Error {
-		b := (^Read_Writer)(s.stream_data).r
-		return reader_unread_rune(b)
-	},
-	impl_write_to = proc(s: io.Stream, w: io.Writer) -> (n: i64, err: io.Error) {
-		b := (^Read_Writer)(s.stream_data).r
-		return reader_write_to(b, w)
-	},
-	impl_flush = proc(s: io.Stream)  -> io.Error {
-		b := (^Read_Writer)(s.stream_data).w
-		return writer_flush(b)
-	},
-	impl_write = proc(s: io.Stream, p: []byte) -> (n: int, err: io.Error) {
-		b := (^Read_Writer)(s.stream_data).w
-		return writer_write(b, p)
-	},
-	impl_write_byte = proc(s: io.Stream, c: byte) -> io.Error {
-		b := (^Read_Writer)(s.stream_data).w
-		return writer_write_byte(b, c)
-	},
-	impl_write_rune = proc(s: io.Stream, r: rune) -> (int, io.Error) {
-		b := (^Read_Writer)(s.stream_data).w
-		return writer_write_rune(b, r)
-	},
-	impl_read_from = proc(s: io.Stream, r: io.Reader) -> (n: i64, err: io.Error) {
-		b := (^Read_Writer)(s.stream_data).w
-		return writer_read_from(b, r)
-	},
-}
+_read_writer_procedure := proc(stream_data: rawptr, mode: io.Stream_Mode, p: []byte, offset: i64, whence: io.Seek_From) -> (n: i64, err: io.Error) {
+	rw := (^Read_Writer)(stream_data)
+	n_int: int
+	#partial switch mode {
+	case .Flush:
+		err = writer_flush(rw.w)
+		return
+	case .Read:
+		n_int, err = reader_read(rw.r, p)
+		n = i64(n_int)
+		return
+	case .Write:
+		n_int, err = writer_write(rw.w, p)
+		n = i64(n_int)
+		return
+	case .Query:
+		return io.query_utility({.Flush, .Read, .Write, .Query})
+	}
+	return 0, .Empty
+}
@@ -81,7 +81,7 @@ _reader_read_new_chunk :: proc(b: ^Reader) -> io.Error {
 	for i := b.max_consecutive_empty_reads; i > 0; i -= 1 {
 		n, err := io.read(b.rd, b.buf[b.w:])
 		if n < 0 {
-			return .Negative_Read
+			return err if err != nil else .Negative_Read
 		}
 		b.w += n
 		if err != nil {
@@ -118,9 +118,7 @@ reader_peek :: proc(b: ^Reader, n: int) -> (data: []byte, err: io.Error) {
 	b.last_rune_size = -1

 	for b.w-b.r < n && b.w-b.r < len(b.buf) && b.err == nil {
-		if fill_err := _reader_read_new_chunk(b); fill_err != nil {
-			return nil, fill_err
-		}
+		_reader_read_new_chunk(b) or_return
 	}

 	if n > len(b.buf) {
@@ -156,9 +154,7 @@ reader_discard :: proc(b: ^Reader, n: int) -> (discarded: int, err: io.Error) {
 	for {
 		skip := reader_buffered(b)
 		if skip == 0 {
-			if fill_err := _reader_read_new_chunk(b); fill_err != nil {
-				return 0, fill_err
-			}
+			_reader_read_new_chunk(b) or_return
 			skip = reader_buffered(b)
 		}
 		skip = min(skip, remaining)
@@ -193,7 +189,7 @@ reader_read :: proc(b: ^Reader, p: []byte) -> (n: int, err: io.Error) {
 		if len(p) >= len(b.buf) {
 			n, b.err = io.read(b.rd, p)
 			if n < 0 {
-				return 0, .Negative_Read
+				return 0, b.err if b.err != nil else .Negative_Read
 			}

 			if n > 0 {
@@ -206,7 +202,7 @@ reader_read :: proc(b: ^Reader, p: []byte) -> (n: int, err: io.Error) {
 		b.r, b.w = 0, 0
 		n, b.err = io.read(b.rd, b.buf)
 		if n < 0 {
-			return 0, .Negative_Read
+			return 0, b.err if b.err != nil else .Negative_Read
 		}
 		if n == 0 {
 			return 0, _reader_consume_err(b)
@@ -223,20 +219,18 @@ reader_read :: proc(b: ^Reader, p: []byte) -> (n: int, err: io.Error) {

 // reader_read_byte reads and returns a single byte
 // If no byte is available, it return an error
-reader_read_byte :: proc(b: ^Reader) -> (byte, io.Error) {
+reader_read_byte :: proc(b: ^Reader) -> (c: byte, err: io.Error) {
 	b.last_rune_size = -1
 	for b.r == b.w {
 		if b.err != nil {
 			return 0, _reader_consume_err(b)
 		}
-		if err := _reader_read_new_chunk(b); err != nil {
-			return 0, err
-		}
+		_reader_read_new_chunk(b) or_return
 	}
-	c := b.buf[b.r]
+	c = b.buf[b.r]
 	b.r += 1
 	b.last_byte = int(c)
-	return c, nil
+	return
 }

 // reader_unread_byte unreads the last byte. Only the most recently read byte can be unread
@@ -264,15 +258,12 @@ reader_read_rune :: proc(b: ^Reader) -> (r: rune, size: int, err: io.Error) {
 	    !utf8.full_rune(b.buf[b.r:b.w]) &&
 	    b.err == nil &&
 	    b.w-b.w < len(b.buf) {
-		if err = _reader_read_new_chunk(b); err != nil {
-			return
-		}
+		_reader_read_new_chunk(b) or_return
 	}

 	b.last_rune_size = -1
 	if b.r == b.w {
-		err = _reader_consume_err(b)
-		return
+		return 0, 0, _reader_consume_err(b)
 	}
 	r, size = rune(b.buf[b.r]), 1
 	if r >= utf8.RUNE_SELF {
@@ -299,45 +290,26 @@ reader_write_to :: proc(b: ^Reader, w: io.Writer) -> (n: i64, err: io.Error) {
 	write_buf :: proc(b: ^Reader, w: io.Writer) -> (i64, io.Error) {
 		n, err := io.write(w, b.buf[b.r:b.w])
 		if n < 0 {
-			return 0, .Negative_Write
+			return 0, err if err != nil else .Negative_Write
 		}
 		b.r += n
 		return i64(n), err
 	}

-	n, err = write_buf(b, w)
-	if err != nil {
-		return
-	}
+	n = write_buf(b, w) or_return

 	m: i64
-	if nr, ok := io.to_writer_to(b.rd); ok {
-		m, err = io.write_to(nr, w)
-		n += m
-		return n, err
-	}
-
-	if nw, ok := io.to_reader_from(w); ok {
-		m, err = io.read_from(nw, b.rd)
-		n += m
-		return n, err
-	}
-
 	if b.w-b.r < len(b.buf) {
-		if err = _reader_read_new_chunk(b); err != nil {
-			return
-		}
+		_reader_read_new_chunk(b) or_return
 	}

 	for b.r < b.w {
 		m, err = write_buf(b, w)
-		n += m
+		n += m // this needs to be done before returning
 		if err != nil {
 			return
 		}
-		if err = _reader_read_new_chunk(b); err != nil {
-			return
-		}
+		_reader_read_new_chunk(b) or_return
 	}

 	if b.err == .EOF {
@@ -352,48 +324,28 @@ reader_write_to :: proc(b: ^Reader, w: io.Writer) -> (n: i64, err: io.Error) {

 // reader_to_stream converts a Reader into an io.Stream
 reader_to_stream :: proc(b: ^Reader) -> (s: io.Stream) {
-	s.stream_data = b
-	s.stream_vtable = &_reader_vtable
+	s.data = b
+	s.procedure = _reader_proc
 	return
 }



@(private)
-_reader_vtable := io.Stream_VTable{
-	impl_destroy = proc(s: io.Stream) -> io.Error {
-		b := (^Reader)(s.stream_data)
+_reader_proc :: proc(stream_data: rawptr, mode: io.Stream_Mode, p: []byte, offset: i64, whence: io.Seek_From) -> (n: i64, err: io.Error) {
+	b := (^Reader)(stream_data)
+	#partial switch mode {
+	case .Read:
+		return io._i64_err(reader_read(b, p))
+	case .Destroy:
 		reader_destroy(b)
-		return nil
-	},
-	impl_read = proc(s: io.Stream, p: []byte) -> (n: int, err: io.Error) {
-		b := (^Reader)(s.stream_data)
-		return reader_read(b, p)
-	},
-	impl_read_byte = proc(s: io.Stream) -> (c: byte, err: io.Error) {
-		b := (^Reader)(s.stream_data)
-		return reader_read_byte(b)
-	},
-	impl_unread_byte = proc(s: io.Stream) -> io.Error {
-		b := (^Reader)(s.stream_data)
-		return reader_unread_byte(b)
-	},
-	impl_read_rune = proc(s: io.Stream) -> (r: rune, size: int, err: io.Error) {
-		b := (^Reader)(s.stream_data)
-		return reader_read_rune(b)
-	},
-	impl_unread_rune = proc(s: io.Stream) -> io.Error {
-		b := (^Reader)(s.stream_data)
-		return reader_unread_rune(b)
-	},
-	impl_write_to = proc(s: io.Stream, w: io.Writer) -> (n: i64, err: io.Error) {
-		b := (^Reader)(s.stream_data)
-		return reader_write_to(b, w)
-	},
+		return
+	case .Query:
+		return io.query_utility({.Read, .Destroy, .Query})
+	}
+	return 0, .Empty
 }

-
-
 //
 // Utility procedures
 //
@@ -435,9 +387,7 @@ reader_read_slice :: proc(b: ^Reader, delim: byte) -> (line: []byte, err: io.Err

 		s = b.w - b.r

-		if err = _reader_read_new_chunk(b); err != nil {
-			break
-		}
+		_reader_read_new_chunk(b) or_break
 	}

 	if i := len(line)-1; i >= 0 {
@@ -4,7 +4,7 @@ import "core:bytes"
 import "core:io"
 import "core:mem"
 import "core:unicode/utf8"
-import "core:intrinsics"
+import "base:intrinsics"

 // Extra errors returns by scanning procedures
 Scanner_Extra_Error :: enum i32 {
@@ -95,6 +95,10 @@ writer_write :: proc(b: ^Writer, p: []byte) -> (n: int, err: io.Error) {
 		m: int
 		if writer_buffered(b) == 0 {
 			m, b.err = io.write(b.wr, p)
+			if m < 0 && b.err == nil {
+				b.err = .Negative_Write
+				break
+			}
 		} else {
 			m = copy(b.buf[b.n:], p)
 			b.n += m
@@ -159,7 +163,7 @@ writer_write_rune :: proc(b: ^Writer, r: rune) -> (size: int, err: io.Error) {
 	return
 }

-// writer_write writes a string into the buffer
+// writer_write_string writes a string into the buffer
 // It returns the number of bytes written
 // If n < len(p), it will return an error explaining why the write is short
 writer_write_string :: proc(b: ^Writer, s: string) -> (int, io.Error) {
@@ -173,14 +177,6 @@ writer_read_from :: proc(b: ^Writer, r: io.Reader) -> (n: i64, err: io.Error) {
 	if b.err != nil {
 		return 0, b.err
 	}
-	if writer_buffered(b) == 0 {
-		if w, ok := io.to_reader_from(b.wr); !ok {
-			n, err = io.read_from(w, r)
-			b.err = err
-			return
-		}
-	}
-
 	for {
 		if writer_available(b) == 0 {
 			writer_flush(b) or_return
@@ -222,38 +218,34 @@ writer_read_from :: proc(b: ^Writer, r: io.Reader) -> (n: i64, err: io.Error) {

 // writer_to_stream converts a Writer into an io.Stream
 writer_to_stream :: proc(b: ^Writer) -> (s: io.Stream) {
-	s.stream_data = b
-	s.stream_vtable = &_writer_vtable
+	s.data = b
+	s.procedure = _writer_proc
 	return
 }

-
-
-@(private)
-_writer_vtable := io.Stream_VTable{
-	impl_destroy = proc(s: io.Stream) -> io.Error {
-		b := (^Writer)(s.stream_data)
-		writer_destroy(b)
-		return nil
-	},
-	impl_flush = proc(s: io.Stream)  -> io.Error {
-		b := (^Writer)(s.stream_data)
-		return writer_flush(b)
-	},
-	impl_write = proc(s: io.Stream, p: []byte) -> (n: int, err: io.Error) {
-		b := (^Writer)(s.stream_data)
-		return writer_write(b, p)
-	},
-	impl_write_byte = proc(s: io.Stream, c: byte) -> io.Error {
-		b := (^Writer)(s.stream_data)
-		return writer_write_byte(b, c)
-	},
-	impl_write_rune = proc(s: io.Stream, r: rune) -> (int, io.Error) {
-		b := (^Writer)(s.stream_data)
-		return writer_write_rune(b, r)
-	},
-	impl_read_from = proc(s: io.Stream, r: io.Reader) -> (n: i64, err: io.Error) {
-		b := (^Writer)(s.stream_data)
-		return writer_read_from(b, r)
-	},
+// writer_to_stream converts a Writer into an io.Stream
+writer_to_writer :: proc(b: ^Writer) -> (s: io.Writer) {
+	return writer_to_stream(b)
+}
+
+
+
+_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 {
+	case .Flush:
+		err = writer_flush(b)
+		return
+	case .Write:
+		n_int: int
+		n_int, err = writer_write(b, p)
+		n = i64(n_int)
+		return
+	case .Destroy:
+		writer_destroy(b)
+		return
+	case .Query:
+		return io.query_utility({.Flush, .Write, .Destroy, .Query})
+	}
+	return 0, .Empty
 }
@@ -38,6 +38,11 @@ buffer_init_string :: proc(b: ^Buffer, s: string) {
 }

 buffer_init_allocator :: proc(b: ^Buffer, len, cap: int, allocator := context.allocator) {
+	if b.buf == nil {
+		b.buf = make([dynamic]byte, len, cap, allocator)
+		return
+	}
+
 	b.buf.allocator = allocator
 	reserve(&b.buf, cap)
 	resize(&b.buf, len)
@@ -108,8 +113,11 @@ _buffer_grow :: proc(b: ^Buffer, n: int) -> int {
 	if i, ok := _buffer_try_grow(b, n); ok {
 		return i
 	}
+
 	if b.buf == nil && n <= SMALL_BUFFER_SIZE {
-		b.buf = make([dynamic]byte, n, SMALL_BUFFER_SIZE)
+		// Fixes #2756 by preserving allocator if already set on Buffer via init_buffer_allocator
+		reserve(&b.buf, SMALL_BUFFER_SIZE)
+		resize(&b.buf, n)
 		return 0
 	}

@@ -351,7 +359,7 @@ buffer_read_from :: proc(b: ^Buffer, r: io.Reader) -> (n: i64, err: io.Error) #n
 		resize(&b.buf, i)
 		m, e := io.read(r, b.buf[i:cap(b.buf)])
 		if m < 0 {
-			err = .Negative_Read
+			err = e if e != nil else .Negative_Read
 			return
 		}

@@ -370,69 +378,31 @@ buffer_read_from :: proc(b: ^Buffer, r: io.Reader) -> (n: i64, err: io.Error) #n


 buffer_to_stream :: proc(b: ^Buffer) -> (s: io.Stream) {
-	s.stream_data = b
-	s.stream_vtable = &_buffer_vtable
+	s.data = b
+	s.procedure = _buffer_proc
 	return
 }

@(private)
-_buffer_vtable := io.Stream_VTable{
-	impl_size = proc(s: io.Stream) -> i64 {
-		b := (^Buffer)(s.stream_data)
-		return i64(buffer_capacity(b))
-	},
-	impl_read = proc(s: io.Stream, p: []byte) -> (n: int, err: io.Error) {
-		b := (^Buffer)(s.stream_data)
-		return buffer_read(b, p)
-	},
-	impl_read_at = proc(s: io.Stream, p: []byte, offset: i64) -> (n: int, err: io.Error) {
-		b := (^Buffer)(s.stream_data)
-		return buffer_read_at(b, p, int(offset))
-	},
-	impl_read_byte = proc(s: io.Stream) -> (byte, io.Error) {
-		b := (^Buffer)(s.stream_data)
-		return buffer_read_byte(b)
-	},
-	impl_read_rune = proc(s: io.Stream) -> (r: rune, size: int, err: io.Error) {
-		b := (^Buffer)(s.stream_data)
-		return buffer_read_rune(b)
-	},
-	impl_write = proc(s: io.Stream, p: []byte) -> (n: int, err: io.Error) {
-		b := (^Buffer)(s.stream_data)
-		return buffer_write(b, p)
-	},
-	impl_write_at = proc(s: io.Stream, p: []byte, offset: i64) -> (n: int, err: io.Error) {
-		b := (^Buffer)(s.stream_data)
-		return buffer_write_at(b, p, int(offset))
-	},
-	impl_write_byte = proc(s: io.Stream, c: byte) -> io.Error {
-		b := (^Buffer)(s.stream_data)
-		return buffer_write_byte(b, c)
-	},
-	impl_write_rune = proc(s: io.Stream, r: rune) -> (int, io.Error) {
-		b := (^Buffer)(s.stream_data)
-		return buffer_write_rune(b, r)
-	},
-	impl_unread_byte = proc(s: io.Stream) -> io.Error {
-		b := (^Buffer)(s.stream_data)
-		return buffer_unread_byte(b)
-	},
-	impl_unread_rune = proc(s: io.Stream) -> io.Error {
-		b := (^Buffer)(s.stream_data)
-		return buffer_unread_rune(b)
-	},
-	impl_destroy = proc(s: io.Stream) -> io.Error {
-		b := (^Buffer)(s.stream_data)
+_buffer_proc :: proc(stream_data: rawptr, mode: io.Stream_Mode, p: []byte, offset: i64, whence: io.Seek_From) -> (n: i64, err: io.Error) {
+	b := (^Buffer)(stream_data)
+	#partial switch mode {
+	case .Read:
+		return io._i64_err(buffer_read(b, p))
+	case .Read_At:
+		return io._i64_err(buffer_read_at(b, p, int(offset)))
+	case .Write:
+		return io._i64_err(buffer_write(b, p))
+	case .Write_At:
+		return io._i64_err(buffer_write_at(b, p, int(offset)))
+	case .Size:
+		n = i64(buffer_capacity(b))
+		return
+	case .Destroy:
 		buffer_destroy(b)
-		return nil
-	},
-	impl_write_to = proc(s: io.Stream, w: io.Writer) -> (n: i64, err: io.Error) {
-		b := (^Buffer)(s.stream_data)
-		return buffer_write_to(b, w)
-	},
-	impl_read_from = proc(s: io.Stream, r: io.Reader) -> (n: i64, err: io.Error) {
-		b := (^Buffer)(s.stream_data)
-		return buffer_read_from(b, r)
-	},
+		return
+	case .Query:
+		return io.query_utility({.Read, .Read_At, .Write, .Write_At, .Size, .Destroy})
+	}
+	return 0, .Empty
 }
-
@@ -895,7 +895,7 @@ split_multi_iterator :: proc(s: ^[]byte, substrs: [][]byte, skip_empty := false)



-// scrub scruvs invalid utf-8 characters and replaces them with the replacement string
+// Scrubs invalid utf-8 characters and replaces them with the replacement string
 // Adjacent invalid bytes are only replaced once
 scrub :: proc(s: []byte, replacement: []byte, allocator := context.allocator) -> []byte {
 	str := s
@@ -16,8 +16,8 @@ reader_init :: proc(r: ^Reader, s: []byte) {
 }

 reader_to_stream :: proc(r: ^Reader) -> (s: io.Stream) {
-	s.stream_data = r
-	s.stream_vtable = &_reader_vtable
+	s.data = r
+	s.procedure = _reader_proc
 	return
 }

@@ -137,41 +137,22 @@ reader_write_to :: proc(r: ^Reader, w: io.Writer) -> (n: i64, err: io.Error) {


@(private)
-_reader_vtable := io.Stream_VTable{
-	impl_size = proc(s: io.Stream) -> i64 {
-		r := (^Reader)(s.stream_data)
-		return reader_size(r)
-	},
-	impl_read = proc(s: io.Stream, p: []byte) -> (n: int, err: io.Error) {
-		r := (^Reader)(s.stream_data)
-		return reader_read(r, p)
-	},
-	impl_read_at = proc(s: io.Stream, p: []byte, off: i64) -> (n: int, err: io.Error) {
-		r := (^Reader)(s.stream_data)
-		return reader_read_at(r, p, off)
-	},
-	impl_read_byte = proc(s: io.Stream) -> (byte, io.Error) {
-		r := (^Reader)(s.stream_data)
-		return reader_read_byte(r)
-	},
-	impl_unread_byte = proc(s: io.Stream) -> io.Error {
-		r := (^Reader)(s.stream_data)
-		return reader_unread_byte(r)
-	},
-	impl_read_rune = proc(s: io.Stream) -> (ch: rune, size: int, err: io.Error) {
-		r := (^Reader)(s.stream_data)
-		return reader_read_rune(r)
-	},
-	impl_unread_rune = proc(s: io.Stream) -> io.Error {
-		r := (^Reader)(s.stream_data)
-		return reader_unread_rune(r)
-	},
-	impl_seek = proc(s: io.Stream, offset: i64, whence: io.Seek_From) -> (i64, io.Error) {
-		r := (^Reader)(s.stream_data)
-		return reader_seek(r, offset, whence)
-	},
-	impl_write_to = proc(s: io.Stream, w: io.Writer) -> (n: i64, err: io.Error) {
-		r := (^Reader)(s.stream_data)
-		return reader_write_to(r, w)
-	},
+_reader_proc :: proc(stream_data: rawptr, mode: io.Stream_Mode, p: []byte, offset: i64, whence: io.Seek_From) -> (n: i64, err: io.Error) {
+	r := (^Reader)(stream_data)
+	#partial switch mode {
+	case .Read:
+		return io._i64_err(reader_read(r, p))
+	case .Read_At:
+		return io._i64_err(reader_read_at(r, p, offset))
+	case .Seek:
+		n, err = reader_seek(r, offset, whence)
+		return
+	case .Size:
+		n = reader_size(r)
+		return
+	case .Query:
+		return io.query_utility({.Read, .Read_At, .Seek, .Size, .Query})
+	}
+	return 0, .Empty
 }
+
@@ -1,6 +1,6 @@
 package c

-import builtin "core:builtin"
+import builtin "base:builtin"

 char           :: builtin.u8  // assuming -funsigned-char

@@ -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,
+}
@@ -1118,7 +1118,7 @@ expand_macro :: proc(cpp: ^Preprocessor, rest: ^^Token, tok: ^Token) -> bool {

 search_include_next :: proc(cpp: ^Preprocessor, filename: string) -> (path: string, ok: bool) {
 	for ; cpp.include_next_index < len(cpp.include_paths); cpp.include_next_index += 1 {
-		tpath := filepath.join(elems={cpp.include_paths[cpp.include_next_index], filename}, allocator=context.temp_allocator)
+		tpath := filepath.join({cpp.include_paths[cpp.include_next_index], filename}, allocator=context.temp_allocator)
 		if os.exists(tpath) {
 			return strings.clone(tpath), true
 		}
@@ -1136,7 +1136,7 @@ search_include_paths :: proc(cpp: ^Preprocessor, filename: string) -> (path: str
 	}

 	for include_path in cpp.include_paths {
-		tpath := filepath.join(elems={include_path, filename}, allocator=context.temp_allocator)
+		tpath := filepath.join({include_path, filename}, allocator=context.temp_allocator)
 		if os.exists(tpath) {
 			path, ok = strings.clone(tpath), true
 			cpp.filepath_cache[filename] = path
@@ -1,34 +1,31 @@
 /*
-package demo
+Example:
+	package demo

-import tokenizer "core:c/frontend/tokenizer"
-import preprocessor "core:c/frontend/preprocessor"
-import "core:fmt"
+	import tokenizer "core:c/frontend/tokenizer"
+	import preprocessor "core:c/frontend/preprocessor"
+	import "core:fmt"

-main :: proc() {
-	t := &tokenizer.Tokenizer{};
-	tokenizer.init_defaults(t);
+	main :: proc() {
+		t := &tokenizer.Tokenizer{};
+		tokenizer.init_defaults(t);

-	cpp := &preprocessor.Preprocessor{};
-	cpp.warn, cpp.err = t.warn, t.err;
-	preprocessor.init_lookup_tables(cpp);
-	preprocessor.init_default_macros(cpp);
-	cpp.include_paths = {"my/path/to/include"};
+		cpp := &preprocessor.Preprocessor{};
+		cpp.warn, cpp.err = t.warn, t.err;
+		preprocessor.init_lookup_tables(cpp);
+		preprocessor.init_default_macros(cpp);
+		cpp.include_paths = {"my/path/to/include"};

-	tok := tokenizer.tokenize_file(t, "the/source/file.c", 1);
+		tok := tokenizer.tokenize_file(t, "the/source/file.c", 1);

-	tok = preprocessor.preprocess(cpp, tok);
-	if tok != nil {
-		for t := tok; t.kind != .EOF; t = t.next {
-			fmt.println(t.lit);
+		tok = preprocessor.preprocess(cpp, tok);
+		if tok != nil {
+			for t := tok; t.kind != .EOF; t = t.next {
+				fmt.println(t.lit);
+			}
 		}
+
+		fmt.println("[Done]");
 	}
-
-	fmt.println("[Done]");
-}
 */
-
-
 package c_frontend_tokenizer
-
-
@@ -127,7 +127,7 @@ advance_rune :: proc(t: ^Tokenizer) {
 }

 advance_rune_n :: proc(t: ^Tokenizer, n: int) {
-	for in 0..<n {
+	for _ in 0..<n {
 		advance_rune(t)
 	}
 }
@@ -67,7 +67,7 @@ foreign libc {
 	crealf  :: proc(z: complex_float) -> float ---
 }

-import builtin "core:builtin"
+import builtin "base:builtin"

 complex_float  :: distinct builtin.complex64
 complex_double :: distinct builtin.complex128
@@ -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
@@ -2,7 +2,7 @@ package libc

 // 7.12 Mathematics

-import "core:intrinsics"
+import "base:intrinsics"

 when ODIN_OS == .Windows {
 	foreign import libc "system:libucrt.lib"
@@ -63,4 +63,4 @@ foreign libc {
 // 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.
-jmp_buf :: struct #align 16 { _: [4096]char, }
+jmp_buf :: struct #align(16) { _: [4096]char, }
@@ -2,7 +2,9 @@ package libc

 // 7.16 Variable arguments

-import "core:intrinsics"
+import "base:intrinsics"
+
+import "core:c"

@(private="file")
@(default_calling_convention="none")
@@ -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)
@@ -2,7 +2,7 @@ package libc

 // 7.17 Atomics

-import "core:intrinsics"
+import "base:intrinsics"

 ATOMIC_BOOL_LOCK_FREE     :: true
 ATOMIC_CHAR_LOCK_FREE     :: true
@@ -1,5 +1,7 @@
 package libc

+import "core:io"
+
 when ODIN_OS == .Windows {
 	foreign import libc {
 		"system:libucrt.lib",
@@ -161,6 +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
@@ -182,7 +214,7 @@ foreign libc {
 	fscanf    :: proc(stream: ^FILE, format: cstring, #c_vararg args: ..any) -> int ---
 	printf    :: proc(format: cstring, #c_vararg args: ..any) -> int ---
 	scanf     :: proc(format: cstring, #c_vararg args: ..any) -> int ---
-	snprintf  :: proc(s: [^]char, format: cstring, #c_vararg args: ..any) -> int ---
+	snprintf  :: proc(s: [^]char, n: size_t, format: cstring, #c_vararg args: ..any) -> int ---
 	sscanf    :: proc(s, format: cstring, #c_vararg args: ..any) -> int ---
 	vfprintf  :: proc(stream: ^FILE, format: cstring, arg: ^va_list) -> int ---
 	vfscanf   :: proc(stream: ^FILE, format: cstring, arg: ^va_list) -> int ---
@@ -218,3 +250,102 @@ foreign libc {
 	ferror    :: proc(stream: ^FILE) -> int ---
 	perror    :: proc(s: cstring) ---
 }
+
+to_stream :: proc(file: ^FILE) -> io.Stream {
+	stream_proc :: proc(stream_data: rawptr, mode: io.Stream_Mode, p: []byte, offset: i64, whence: io.Seek_From) -> (n: i64, err: io.Error) {
+		unknown_or_eof :: proc(f: ^FILE) -> io.Error {
+			switch {
+			case ferror(f) != 0:
+				return .Unknown
+			case feof(f) != 0:
+				return .EOF
+			case:
+				return nil
+			}
+		}
+
+		file := (^FILE)(stream_data)
+		switch mode {
+		case .Close:
+			if fclose(file) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+		case .Flush:
+			if fflush(file) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+		case .Read:
+			n = i64(fread(raw_data(p), size_of(byte), len(p), file))
+			if n == 0 { err = unknown_or_eof(file) }
+
+		case .Read_At:
+			curr := ftell(file)
+			if curr == -1 {
+				return 0, unknown_or_eof(file)
+			}
+
+			if fseek(file, long(offset), SEEK_SET) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+			defer fseek(file, long(curr), SEEK_SET)
+
+			n = i64(fread(raw_data(p), size_of(byte), len(p), file))
+			if n == 0 { err = unknown_or_eof(file) }
+		
+		case .Write:
+			n = i64(fwrite(raw_data(p), size_of(byte), len(p), file))
+			if n == 0 { err = unknown_or_eof(file) }
+
+		case .Write_At:
+			curr := ftell(file)
+			if curr == -1 {
+				return 0, unknown_or_eof(file)
+			}
+
+			if fseek(file, long(offset), SEEK_SET) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+			defer fseek(file, long(curr), SEEK_SET)
+
+			n = i64(fwrite(raw_data(p), size_of(byte), len(p), file))
+			if n == 0 { err = unknown_or_eof(file) }
+
+		case .Seek:
+			if fseek(file, long(offset), int(whence)) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+		
+		case .Size:
+			curr := ftell(file)
+			if curr == -1 {
+				return 0, unknown_or_eof(file)
+			}
+			defer fseek(file, curr, SEEK_SET)
+
+			if fseek(file, 0, SEEK_END) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+			n = i64(ftell(file))
+			if n == -1 {
+				return 0, unknown_or_eof(file)
+			}
+
+		case .Destroy:
+			return 0, .Empty
+		
+		case .Query:
+			return io.query_utility({ .Close, .Flush, .Read, .Read_At, .Write, .Write_At, .Seek, .Size })
+		}
+		return
+	}
+
+	return {
+		data      = file,
+		procedure = stream_proc,
+	}
+}
@@ -1,6 +1,6 @@
 package libc

-import "core:runtime"
+import "base:runtime"

 // 7.24 String handling

@@ -35,17 +35,17 @@ when ODIN_OS == .Windows {
 	clock_t        :: distinct long
 	time_t         :: distinct i64

-	timespec :: struct #align 8 {
+	timespec :: struct #align(8) {
 		tv_sec:  time_t,
 		tv_nsec: long,
 	}

-	tm :: struct #align 8 {
+	tm :: struct #align(8) {
 		tm_sec, tm_min, tm_hour, tm_mday, tm_mon, tm_year, tm_wday, tm_yday, tm_isdst: int,
 	}
 }

-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
@@ -2,6 +2,8 @@ package libc

 import "core:c"

+#assert(!ODIN_NO_CRT, `"core:c/libc" cannot be imported when '-no-crt' is used`)
+
 char           :: c.char // assuming -funsigned-char

 schar          :: c.schar
@@ -98,7 +98,7 @@ foreign libc {
 }

 // Large enough and aligned enough for any wide-spread in-use libc.
-mbstate_t :: struct #align 16 { _: [32]char, }
+mbstate_t :: struct #align(16) { _: [32]char, }

 // Odin does not have default argument promotion so the need for a separate type
 // here isn't necessary, though make it distinct just to be safe.
@@ -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")
@@ -12,7 +12,7 @@ package compress

 import "core:io"
 import "core:bytes"
-import "core:runtime"
+import "base:runtime"

 /*
 	These settings bound how much compression algorithms will allocate for their output buffer.
@@ -20,10 +20,9 @@ import "core:runtime"

 */

-/*
-	When a decompression routine doesn't stream its output, but writes to a buffer,
-	we pre-allocate an output buffer to speed up decompression. The default is 1 MiB.
-*/
+
+// When a decompression routine doesn't stream its output, but writes to a buffer,
+// we pre-allocate an output buffer to speed up decompression. The default is 1 MiB.
 COMPRESS_OUTPUT_ALLOCATE_MIN :: int(#config(COMPRESS_OUTPUT_ALLOCATE_MIN, 1 << 20))

 /*
@@ -34,16 +33,14 @@ COMPRESS_OUTPUT_ALLOCATE_MIN :: int(#config(COMPRESS_OUTPUT_ALLOCATE_MIN, 1 << 2

 */
 when size_of(uintptr) == 8 {
-	/*
-		For 64-bit platforms, we set the default max buffer size to 4 GiB,
-		which is GZIP and PKZIP's max payload size.
-	*/	
+
+        // For 64-bit platforms, we set the default max buffer size to 4 GiB,
+        // which is GZIP and PKZIP's max payload size.
 	COMPRESS_OUTPUT_ALLOCATE_MAX :: int(#config(COMPRESS_OUTPUT_ALLOCATE_MAX, 1 << 32))
 } else {
-	/*
-		For 32-bit platforms, we set the default max buffer size to 512 MiB.
-	*/
-	COMPRESS_OUTPUT_ALLOCATE_MAX :: int(#config(COMPRESS_OUTPUT_ALLOCATE_MAX, 1 << 29))
+	
+	// For 32-bit platforms, we set the default max buffer size to 512 MiB.
+        COMPRESS_OUTPUT_ALLOCATE_MAX :: int(#config(COMPRESS_OUTPUT_ALLOCATE_MAX, 1 << 29))
 }


@@ -69,9 +66,8 @@ General_Error :: enum {
 	Incompatible_Options,
 	Unimplemented,

-	/*
-		Memory errors
-	*/
+	// Memory errors
+
 	Allocation_Failed,
 	Resize_Failed,
 }
@@ -86,17 +82,16 @@ GZIP_Error :: enum {
 	Payload_Length_Invalid,
 	Payload_CRC_Invalid,

-	/*
-		GZIP's payload can be a maximum of max(u32le), or 4 GiB.
-		If you tell it you expect it to contain more, that's obviously an error.
-	*/
-	Payload_Size_Exceeds_Max_Payload,
-	/*
-		For buffered instead of streamed output, the payload size can't exceed
-		the max set by the `COMPRESS_OUTPUT_ALLOCATE_MAX` switch in compress/common.odin.
+	// GZIP's payload can be a maximum of max(u32le), or 4 GiB.
+	// If you tell it you expect it to contain more, that's obviously an error.
+
+	Payload_Size_Exceeds_Max_Payload,
+
+	// For buffered instead of streamed output, the payload size can't exceed
+	// the max set by the `COMPRESS_OUTPUT_ALLOCATE_MAX` switch in compress/common.odin.
+	//
+	// You can tweak this setting using `-define:COMPRESS_OUTPUT_ALLOCATE_MAX=size_in_bytes`

-		You can tweak this setting using `-define:COMPRESS_OUTPUT_ALLOCATE_MAX=size_in_bytes`
-	*/
 	Output_Exceeds_COMPRESS_OUTPUT_ALLOCATE_MAX,

 }
@@ -137,9 +132,8 @@ Context_Memory_Input :: struct #packed {
 	code_buffer:       u64,
 	num_bits:          u64,

-	/*
-		If we know the data size, we can optimize the reads and writes.
-	*/
+	// If we know the data size, we can optimize the reads and writes.
+
 	size_packed:       i64,
 	size_unpacked:     i64,
 }
@@ -159,18 +153,16 @@ Context_Stream_Input :: struct #packed {
 	code_buffer:       u64,
 	num_bits:          u64,

-	/*
-		If we know the data size, we can optimize the reads and writes.
-	*/
+	// If we know the data size, we can optimize the reads and writes.
+
 	size_packed:       i64,
 	size_unpacked:     i64,

-	/*
-		Flags:
-			`input_fully_in_memory`
-				true  = This tells us we read input from `input_data` exclusively. [] = EOF.
-				false = Try to refill `input_data` from the `input` stream.
-	*/
+	// Flags:
+	// `input_fully_in_memory`
+	//   true  = This tells us we read input from `input_data` exclusively. [] = EOF.
+	//   false = Try to refill `input_data` from the `input` stream.
+
 	input_fully_in_memory: b8,

 	padding: [1]u8,
@@ -188,7 +180,8 @@ input_size_from_memory :: proc(z: ^Context_Memory_Input) -> (res: i64, err: Erro
 }

 input_size_from_stream :: proc(z: ^Context_Stream_Input) -> (res: i64, err: Error) {
-	return io.size(z.input), nil
+	res, _ = io.size(z.input)
+	return
 }

 input_size :: proc{input_size_from_memory, input_size_from_stream}
@@ -213,26 +206,18 @@ read_slice_from_memory :: #force_inline proc(z: ^Context_Memory_Input, size: int
@(optimization_mode="speed")
 read_slice_from_stream :: #force_inline proc(z: ^Context_Stream_Input, size: int) -> (res: []u8, err: io.Error) {
 	// TODO: REMOVE ALL USE OF context.temp_allocator here
-	// the is literally no need for it
+	// there is literally no need for it
 	b := make([]u8, size, context.temp_allocator)
-	_, e := z.input->impl_read(b[:])
-	if e == .None {
-		return b, .None
-	}
-
-	return []u8{}, e
+	_ = io.read(z.input, b[:]) or_return
+	return b, nil
 }

 read_slice :: proc{read_slice_from_memory, read_slice_from_stream}

@(optimization_mode="speed")
 read_data :: #force_inline proc(z: ^$C, $T: typeid) -> (res: T, err: io.Error) {
-	b, e := read_slice(z, size_of(T))
-	if e == .None {
-		return (^T)(&b[0])^, .None
-	}
-
-	return T{}, e
+	b := read_slice(z, size_of(T)) or_return
+	return (^T)(&b[0])^, nil
 }

@(optimization_mode="speed")
@@ -249,20 +234,14 @@ read_u8_from_memory :: #force_inline proc(z: ^Context_Memory_Input) -> (res: u8,

@(optimization_mode="speed")
 read_u8_from_stream :: #force_inline proc(z: ^Context_Stream_Input) -> (res: u8, err: io.Error) {
-	b, e := read_slice_from_stream(z, 1)
-	if e == .None {
-		return b[0], .None
-	}
-
-	return 0, e
+	b := read_slice_from_stream(z, 1) or_return
+	return b[0], nil
 }

 read_u8 :: proc{read_u8_from_memory, read_u8_from_stream}

-/*
-	You would typically only use this at the end of Inflate, to drain bits from the code buffer
-	preferentially.
-*/
+// You would typically only use this at the end of Inflate, to drain bits from the code buffer
+// preferentially.
@(optimization_mode="speed")
 read_u8_prefer_code_buffer_lsb :: #force_inline proc(z: ^$C) -> (res: u8, err: io.Error) {
 	if z.num_bits >= 8 {
@@ -319,12 +298,9 @@ peek_data_from_stream :: #force_inline proc(z: ^Context_Stream_Input, $T: typeid
 	size :: size_of(T)

 	// Get current position to read from.
-	curr, e1 := z.input->impl_seek(0, .Current)
-	if e1 != .None {
-		return T{}, e1
-	}
-	r, e2 := io.to_reader_at(z.input)
-	if !e2 {
+	curr := z.input->impl_seek(0, .Current) or_return
+	r, e1 := io.to_reader_at(z.input)
+	if !e1 {
 		return T{}, .Empty
 	}
 	when size <= 128 {
@@ -332,8 +308,8 @@ peek_data_from_stream :: #force_inline proc(z: ^Context_Stream_Input, $T: typeid
 	} else {
 		b := make([]u8, size, context.temp_allocator)
 	}
-	_, e3 := io.read_at(r, b[:], curr)
-	if e3 != .None {
+	_, e2 := io.read_at(r, b[:], curr)
+	if e2 != .None {
 		return T{}, .Empty
 	}

@@ -346,16 +322,9 @@ peek_data_at_offset_from_stream :: #force_inline proc(z: ^Context_Stream_Input,
 	size :: size_of(T)

 	// Get current position to return to.
-	cur_pos, e1 := z.input->impl_seek(0, .Current)
-	if e1 != .None {
-		return T{}, e1
-	}
-
+	cur_pos := z.input->impl_seek(0, .Current) or_return
 	// Seek to offset.
-	pos, e2 := z.input->impl_seek(offset, .Start)
-	if e2 != .None {
-		return T{}, e2
-	}
+	pos := z.input->impl_seek(offset, .Start) or_return

 	r, e3 := io.to_reader_at(z.input)
 	if !e3 {
@@ -464,7 +433,7 @@ peek_bits_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width:
 	if z.num_bits < u64(width) {
 		refill_lsb(z)
 	}
-	return u32(z.code_buffer & ~(~u64(0) << width))
+	return u32(z.code_buffer &~ (~u64(0) << width))
 }

@(optimization_mode="speed")
@@ -472,7 +441,7 @@ peek_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width:
 	if z.num_bits < u64(width) {
 		refill_lsb(z)
 	}
-	return u32(z.code_buffer & ~(~u64(0) << width))
+	return u32(z.code_buffer &~ (~u64(0) << width))
 }

 peek_bits_lsb :: proc{peek_bits_lsb_from_memory, peek_bits_lsb_from_stream}
@@ -480,13 +449,13 @@ peek_bits_lsb :: proc{peek_bits_lsb_from_memory, peek_bits_lsb_from_stream}
@(optimization_mode="speed")
 peek_bits_no_refill_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width: u8) -> u32 {
 	assert(z.num_bits >= u64(width))
-	return u32(z.code_buffer & ~(~u64(0) << width))
+	return u32(z.code_buffer &~ (~u64(0) << width))
 }

@(optimization_mode="speed")
 peek_bits_no_refill_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width: u8) -> u32 {
 	assert(z.num_bits >= u64(width))
-	return u32(z.code_buffer & ~(~u64(0) << width))
+	return u32(z.code_buffer &~ (~u64(0) << width))
 }

 peek_bits_no_refill_lsb :: proc{peek_bits_no_refill_lsb_from_memory, peek_bits_no_refill_lsb_from_stream}
@@ -1,5 +1,5 @@
 //+build ignore
-package gzip
+package compress_gzip

 /*
 	Copyright 2021 Jeroen van Rijn <nom@duclavier.com>.
@@ -1,4 +1,4 @@
-package gzip
+package compress_gzip

 /*
 	Copyright 2021 Jeroen van Rijn <nom@duclavier.com>.
@@ -335,10 +335,8 @@ load_from_context :: proc(z: ^$C, buf: ^bytes.Buffer, known_gzip_size := -1, exp

 	// fmt.printf("GZIP: Expected Payload Size: %v\n", expected_output_size);

-	zlib_error := zlib.inflate_raw(z=z, expected_output_size=expected_output_size)
-	if zlib_error != nil {
-		return zlib_error
-	}
+	zlib.inflate_raw(z, expected_output_size=expected_output_size) or_return
+
 	/*
 		Read CRC32 using the ctx bit reader because zlib may leave bytes in there.
 	*/
@@ -5,7 +5,7 @@
 */

 // package shoco is an implementation of the shoco short string compressor
-package shoco
+package compress_shoco

 DEFAULT_MODEL :: Shoco_Model {
 	min_char = 39,
@@ -9,9 +9,9 @@
 */

 // package shoco is an implementation of the shoco short string compressor
-package shoco
+package compress_shoco

-import "core:intrinsics"
+import "base:intrinsics"
 import "core:compress"

 Shoco_Pack :: struct {
@@ -177,12 +177,10 @@ decompress_slice_to_string :: proc(input: []u8, model := DEFAULT_MODEL, allocato
 	max_output_size := decompress_bound(len(input), model)

 	buf: [dynamic]u8
-	if !resize(&buf, max_output_size) {
-		return "", .Out_Of_Memory
-	}
+	resize(&buf, max_output_size) or_return

 	length, result := decompress_slice_to_output_buffer(input, buf[:])
-	resize(&buf, length)
+	resize(&buf, length) or_return
 	return string(buf[:]), result
 }
 decompress :: proc{decompress_slice_to_output_buffer, decompress_slice_to_string}
@@ -307,12 +305,10 @@ compress_string :: proc(input: string, model := DEFAULT_MODEL, allocator := cont
 	max_output_size := compress_bound(len(input))

 	buf: [dynamic]u8
-	if !resize(&buf, max_output_size) {
-		return {}, .Out_Of_Memory
-	}
+	resize(&buf, max_output_size) or_return

 	length, result := compress_string_to_buffer(input, buf[:])
-	resize(&buf, length)
+	resize(&buf, length) or_return
 	return buf[:length], result
 }
 compress :: proc{compress_string_to_buffer, compress_string}
@@ -1,5 +1,5 @@
 //+build ignore
-package zlib
+package compress_zlib

 /*
 	Copyright 2021 Jeroen van Rijn <nom@duclavier.com>.
@@ -1,4 +1,5 @@
-package zlib
+//+vet !using-param
+package compress_zlib

 /*
 	Copyright 2021 Jeroen van Rijn <nom@duclavier.com>.
@@ -146,15 +147,7 @@ grow_buffer :: proc(buf: ^[dynamic]u8) -> (err: compress.Error) {
 		Double until we reach the maximum allowed.
 	*/
 	new_size := min(len(buf) << 1, compress.COMPRESS_OUTPUT_ALLOCATE_MAX)
-	resize(buf, new_size)
-	if len(buf) != new_size {
-		/*
-			Resize failed.
-		*/
-		return .Resize_Failed
-	}
-
-	return nil
+	return resize(buf, new_size)
 }

 /*
@@ -181,7 +174,7 @@ write_byte :: #force_inline proc(z: ^$C, c: u8) -> (err: io.Error) #no_bounds_ch
 }

@(optimization_mode="speed")
-repl_byte :: proc(z: ^$C, count: u16, c: u8) -> (err: io.Error) 	#no_bounds_check {
+repl_byte :: proc(z: ^$C, count: u16, c: u8) -> (err: io.Error) #no_bounds_check {
 	/*
 		TODO(Jeroen): Once we have a magic ring buffer, we can just peek/write into it
 		without having to worry about wrapping, so no need for a temp allocation to give to
@@ -305,10 +298,10 @@ decode_huffman_slowpath :: proc(z: ^$C, t: ^Huffman_Table) -> (r: u16, err: Erro
 	code := u16(compress.peek_bits_lsb(z,16))

 	k := int(z_bit_reverse(code, 16))
-	s: u8

-	#no_bounds_check for s = HUFFMAN_FAST_BITS+1; ; {
-		if k < t.maxcode[s] {
+	s: u8 = HUFFMAN_FAST_BITS+1
+	for {
+		#no_bounds_check if k < t.maxcode[s] {
 			break
 		}
 		s += 1
@@ -471,7 +464,7 @@ inflate_from_context :: proc(using ctx: ^compress.Context_Memory_Input, raw := f
 	}

 	// Parse ZLIB stream without header.
-	inflate_raw(z=ctx, expected_output_size=expected_output_size) or_return
+	inflate_raw(ctx, expected_output_size=expected_output_size) or_return

 	if !raw {
 		compress.discard_to_next_byte_lsb(ctx)
@@ -509,8 +502,8 @@ inflate_raw :: proc(z: ^$C, expected_output_size := -1, allocator := context.all
 		/*
 			Try to pre-allocate the output buffer.
 		*/
-		reserve(&z.output.buf, expected_output_size)
-		resize (&z.output.buf, expected_output_size)
+		reserve(&z.output.buf, expected_output_size) or_return
+		resize (&z.output.buf, expected_output_size) or_return
 	}

 	if len(z.output.buf) != expected_output_size {
@@ -653,7 +646,7 @@ inflate_raw :: proc(z: ^$C, expected_output_size := -1, allocator := context.all
 	}

 	if int(z.bytes_written) != len(z.output.buf) {
-		resize(&z.output.buf, int(z.bytes_written))
+		resize(&z.output.buf, int(z.bytes_written)) or_return
 	}

 	return nil
@@ -665,7 +658,7 @@ inflate_from_byte_array :: proc(input: []u8, buf: ^bytes.Buffer, raw := false, e
 	ctx.input_data = input
 	ctx.output = buf

-	return inflate_from_context(ctx=&ctx, raw=raw, expected_output_size=expected_output_size)
+	return inflate_from_context(&ctx, raw=raw, expected_output_size=expected_output_size)
 }

 inflate_from_byte_array_raw :: proc(input: []u8, buf: ^bytes.Buffer, raw := false, expected_output_size := -1) -> (err: Error) {
@@ -674,7 +667,7 @@ inflate_from_byte_array_raw :: proc(input: []u8, buf: ^bytes.Buffer, raw := fals
 	ctx.input_data = input
 	ctx.output = buf

-	return inflate_raw(z=&ctx, expected_output_size=expected_output_size)
+	return inflate_raw(&ctx, expected_output_size=expected_output_size)
 }

 inflate :: proc{inflate_from_context, inflate_from_byte_array}
@@ -0,0 +1,678 @@
+/*
+package avl implements an AVL tree.
+
+The implementation is non-intrusive, and non-recursive.
+*/
+package container_avl
+
+import "base:intrinsics"
+import "base:runtime"
+import "core:slice"
+
+_ :: intrinsics
+_ :: runtime
+
+// Originally based on the CC0 implementation by Eric Biggers
+// See: https://github.com/ebiggers/avl_tree/
+
+// Direction specifies the traversal direction for a tree iterator.
+Direction :: enum i8 {
+	// Backward is the in-order backwards direction.
+	Backward = -1,
+	// Forward is the in-order forwards direction.
+	Forward  = 1,
+}
+
+// Ordering specifies order when inserting/finding values into the tree.
+Ordering :: slice.Ordering
+
+// Tree is an AVL tree.
+Tree :: struct($Value: typeid) {
+	// user_data is a parameter that will be passed to the on_remove
+	// callback.
+	user_data: rawptr,
+	// on_remove is an optional callback that can be called immediately
+	// after a node is removed from the tree.
+	on_remove: proc(value: Value, user_data: rawptr),
+
+	_root:           ^Node(Value),
+	_node_allocator: runtime.Allocator,
+	_cmp_fn:         proc(a, b: Value) -> Ordering,
+	_size:           int,
+}
+
+// Node is an AVL tree node.
+//
+// WARNING: It is unsafe to mutate value if the node is part of a tree
+// if doing so will alter the Node's sort position relative to other
+// elements in the tree.
+Node :: struct($Value: typeid) {
+	value: Value,
+
+	_parent:  ^Node(Value),
+	_left:    ^Node(Value),
+	_right:   ^Node(Value),
+	_balance: i8,
+}
+
+// Iterator is a tree iterator.
+//
+// WARNING: It is unsafe to modify the tree while iterating, except via
+// the iterator_remove method.
+Iterator :: struct($Value: typeid) {
+	_tree:        ^Tree(Value),
+	_cur:         ^Node(Value),
+	_next:        ^Node(Value),
+	_direction:   Direction,
+	_called_next: bool,
+}
+
+// init initializes a tree.
+init :: proc {
+	init_ordered,
+	init_cmp,
+}
+
+// init_cmp initializes a tree.
+init_cmp :: proc(
+	t: ^$T/Tree($Value),
+	cmp_fn: proc(a, b: Value) -> Ordering,
+	node_allocator := context.allocator,
+) {
+	t._root = nil
+	t._node_allocator = node_allocator
+	t._cmp_fn = cmp_fn
+	t._size = 0
+}
+
+// init_ordered initializes a tree containing ordered items, with
+// a comparison function that results in an ascending order sort.
+init_ordered :: proc(
+	t: ^$T/Tree($Value),
+	node_allocator := context.allocator,
+) where intrinsics.type_is_ordered_numeric(Value) {
+	init_cmp(t, slice.cmp_proc(Value), node_allocator)
+}
+
+// destroy de-initializes a tree.
+destroy :: proc(t: ^$T/Tree($Value), call_on_remove: bool = true) {
+	iter := iterator(t, Direction.Forward)
+	for _ in iterator_next(&iter) {
+		iterator_remove(&iter, call_on_remove)
+	}
+}
+
+// len returns the number of elements in the tree.
+len :: proc "contextless" (t: ^$T/Tree($Value)) -> int {
+	return t._size
+}
+
+// first returns the first node in the tree (in-order) or nil iff
+// the tree is empty.
+first :: proc "contextless" (t: ^$T/Tree($Value)) -> ^Node(Value) {
+	return tree_first_or_last_in_order(t, Direction.Backward)
+}
+
+// last returns the last element in the tree (in-order) or nil iff
+// the tree is empty.
+last :: proc "contextless" (t: ^$T/Tree($Value)) -> ^Node(Value) {
+	return tree_first_or_last_in_order(t, Direction.Forward)
+}
+
+// find finds the value in the tree, and returns the corresponding
+// node or nil iff the value is not present.
+find :: proc(t: ^$T/Tree($Value), value: Value) -> ^Node(Value) {
+	cur := t._root
+	descend_loop: for cur != nil {
+		switch t._cmp_fn(value, cur.value) {
+		case .Less:
+			cur = cur._left
+		case .Greater:
+			cur = cur._right
+		case .Equal:
+			break descend_loop
+		}
+	}
+
+	return cur
+}
+
+// find_or_insert attempts to insert the value into the tree, and returns
+// the node, a boolean indicating if the value was inserted, and the
+// node allocator error if relevant.  If the value is already
+// present, the existing node is returned un-altered.
+find_or_insert :: proc(
+	t: ^$T/Tree($Value),
+	value: Value,
+) -> (
+	n: ^Node(Value),
+	inserted: bool,
+	err: runtime.Allocator_Error,
+) {
+	n_ptr := &t._root
+	for n_ptr^ != nil {
+		n = n_ptr^
+		switch t._cmp_fn(value, n.value) {
+		case .Less:
+			n_ptr = &n._left
+		case .Greater:
+			n_ptr = &n._right
+		case .Equal:
+			return
+		}
+	}
+
+	parent := n
+	n = new(Node(Value), t._node_allocator) or_return
+	n.value = value
+	n._parent = parent
+	n_ptr^ = n
+	tree_rebalance_after_insert(t, n)
+
+	t._size += 1
+	inserted = true
+
+	return
+}
+
+// remove removes a node or value from the tree, and returns true iff the
+// removal was successful.  While the node's value will be left intact,
+// the node itself will be freed via the tree's node allocator.
+remove :: proc {
+	remove_value,
+	remove_node,
+}
+
+// remove_value removes a value from the tree, and returns true iff the
+// removal was successful.  While the node's value will be left intact,
+// the node itself will be freed via the tree's node allocator.
+remove_value :: proc(t: ^$T/Tree($Value), value: Value, call_on_remove: bool = true) -> bool {
+	n := find(t, value)
+	if n == nil {
+		return false
+	}
+	return remove_node(t, n, call_on_remove)
+}
+
+// remove_node removes a node from the tree, and returns true iff the
+// removal was successful.  While the node's value will be left intact,
+// the node itself will be freed via the tree's node allocator.
+remove_node :: proc(t: ^$T/Tree($Value), node: ^Node(Value), call_on_remove: bool = true) -> bool {
+	if node._parent == node || (node._parent == nil && t._root != node) {
+		return false
+	}
+	defer {
+		if call_on_remove && t.on_remove != nil {
+			t.on_remove(node.value, t.user_data)
+		}
+		free(node, t._node_allocator)
+	}
+
+	parent: ^Node(Value)
+	left_deleted: bool
+
+	t._size -= 1
+	if node._left != nil && node._right != nil {
+		parent, left_deleted = tree_swap_with_successor(t, node)
+	} else {
+		child := node._left
+		if child == nil {
+			child = node._right
+		}
+		parent = node._parent
+		if parent != nil {
+			if node == parent._left {
+				parent._left = child
+				left_deleted = true
+			} else {
+				parent._right = child
+				left_deleted = false
+			}
+			if child != nil {
+				child._parent = parent
+			}
+		} else {
+			if child != nil {
+				child._parent = parent
+			}
+			t._root = child
+			node_reset(node)
+			return true
+		}
+	}
+
+	for {
+		if left_deleted {
+			parent = tree_handle_subtree_shrink(t, parent, +1, &left_deleted)
+		} else {
+			parent = tree_handle_subtree_shrink(t, parent, -1, &left_deleted)
+		}
+		if parent == nil {
+			break
+		}
+	}
+	node_reset(node)
+
+	return true
+}
+
+// iterator returns a tree iterator in the specified direction.
+iterator :: proc "contextless" (t: ^$T/Tree($Value), direction: Direction) -> Iterator(Value) {
+	it: Iterator(Value)
+	it._tree = transmute(^Tree(Value))t
+	it._direction = direction
+
+	iterator_first(&it)
+
+	return it
+}
+
+// iterator_from_pos returns a tree iterator in the specified direction,
+// spanning the range [pos, last] (inclusive).
+iterator_from_pos :: proc "contextless" (
+	t: ^$T/Tree($Value),
+	pos: ^Node(Value),
+	direction: Direction,
+) -> Iterator(Value) {
+	it: Iterator(Value)
+	it._tree = transmute(^Tree(Value))t
+	it._direction = direction
+	it._next = nil
+	it._called_next = false
+
+	if it._cur = pos; pos != nil {
+		it._next = node_next_or_prev_in_order(it._cur, it._direction)
+	}
+
+	return it
+}
+
+// iterator_get returns the node currently pointed to by the iterator,
+// or nil iff the node has been removed, the tree is empty, or the end
+// of the tree has been reached.
+iterator_get :: proc "contextless" (it: ^$I/Iterator($Value)) -> ^Node(Value) {
+	return it._cur
+}
+
+// iterator_remove removes the node currently pointed to by the iterator,
+// and returns true iff the removal was successful.  Semantics are the
+// same as the Tree remove.
+iterator_remove :: proc(it: ^$I/Iterator($Value), call_on_remove: bool = true) -> bool {
+	if it._cur == nil {
+		return false
+	}
+
+	ok := remove_node(it._tree, it._cur, call_on_remove)
+	if ok {
+		it._cur = nil
+	}
+
+	return ok
+}
+
+// iterator_next advances the iterator and returns the (node, true) or
+// or (nil, false) iff the end of the tree has been reached.
+//
+// Note: The first call to iterator_next will return the first node instead
+// of advancing the iterator.
+iterator_next :: proc "contextless" (it: ^$I/Iterator($Value)) -> (^Node(Value), bool) {
+	// This check is needed so that the first element gets returned from
+	// a brand-new iterator, and so that the somewhat contrived case where
+	// iterator_remove is called before the first call to iterator_next
+	// returns the correct value.
+	if !it._called_next {
+		it._called_next = true
+
+		// There can be the contrived case where iterator_remove is
+		// called before ever calling iterator_next, which needs to be
+		// handled as an actual call to next.
+		//
+		// If this happens it._cur will be nil, so only return the
+		// first value, if it._cur is valid.
+		if it._cur != nil {
+			return it._cur, true
+		}
+	}
+
+	if it._next == nil {
+		return nil, false
+	}
+
+	it._cur = it._next
+	it._next = node_next_or_prev_in_order(it._cur, it._direction)
+
+	return it._cur, true
+}
+
+@(private)
+tree_first_or_last_in_order :: proc "contextless" (
+	t: ^$T/Tree($Value),
+	direction: Direction,
+) -> ^Node(Value) {
+	first, sign := t._root, i8(direction)
+	if first != nil {
+		for {
+			tmp := node_get_child(first, +sign)
+			if tmp == nil {
+				break
+			}
+			first = tmp
+		}
+	}
+
+	return first
+}
+
+@(private)
+tree_replace_child :: proc "contextless" (
+	t: ^$T/Tree($Value),
+	parent, old_child, new_child: ^Node(Value),
+) {
+	if parent != nil {
+		if old_child == parent._left {
+			parent._left = new_child
+		} else {
+			parent._right = new_child
+		}
+	} else {
+		t._root = new_child
+	}
+}
+
+@(private)
+tree_rotate :: proc "contextless" (t: ^$T/Tree($Value), a: ^Node(Value), sign: i8) {
+	b := node_get_child(a, -sign)
+	e := node_get_child(b, +sign)
+	p := a._parent
+
+	node_set_child(a, -sign, e)
+	a._parent = b
+
+	node_set_child(b, +sign, a)
+	b._parent = p
+
+	if e != nil {
+		e._parent = a
+	}
+
+	tree_replace_child(t, p, a, b)
+}
+
+@(private)
+tree_double_rotate :: proc "contextless" (
+	t: ^$T/Tree($Value),
+	b, a: ^Node(Value),
+	sign: i8,
+) -> ^Node(Value) {
+	e := node_get_child(b, +sign)
+	f := node_get_child(e, -sign)
+	g := node_get_child(e, +sign)
+	p := a._parent
+	e_bal := e._balance
+
+	node_set_child(a, -sign, g)
+	a_bal := -e_bal
+	if sign * e_bal >= 0 {
+		a_bal = 0
+	}
+	node_set_parent_balance(a, e, a_bal)
+
+	node_set_child(b, +sign, f)
+	b_bal := -e_bal
+	if sign * e_bal <= 0 {
+		b_bal = 0
+	}
+	node_set_parent_balance(b, e, b_bal)
+
+	node_set_child(e, +sign, a)
+	node_set_child(e, -sign, b)
+	node_set_parent_balance(e, p, 0)
+
+	if g != nil {
+		g._parent = a
+	}
+
+	if f != nil {
+		f._parent = b
+	}
+
+	tree_replace_child(t, p, a, e)
+
+	return e
+}
+
+@(private)
+tree_handle_subtree_growth :: proc "contextless" (
+	t: ^$T/Tree($Value),
+	node, parent: ^Node(Value),
+	sign: i8,
+) -> bool {
+	old_balance_factor := parent._balance
+	if old_balance_factor == 0 {
+		node_adjust_balance_factor(parent, sign)
+		return false
+	}
+
+	new_balance_factor := old_balance_factor + sign
+	if new_balance_factor == 0 {
+		node_adjust_balance_factor(parent, sign)
+		return true
+	}
+
+	if sign * node._balance > 0 {
+		tree_rotate(t, parent, -sign)
+		node_adjust_balance_factor(parent, -sign)
+		node_adjust_balance_factor(node, -sign)
+	} else {
+		tree_double_rotate(t, node, parent, -sign)
+	}
+
+	return true
+}
+
+@(private)
+tree_rebalance_after_insert :: proc "contextless" (t: ^$T/Tree($Value), inserted: ^Node(Value)) {
+	node, parent := inserted, inserted._parent
+	switch {
+	case parent == nil:
+		return
+	case node == parent._left:
+		node_adjust_balance_factor(parent, -1)
+	case:
+		node_adjust_balance_factor(parent, +1)
+	}
+
+	if parent._balance == 0 {
+		return
+	}
+
+	for done := false; !done; {
+		node = parent
+		if parent = node._parent; parent == nil {
+			return
+		}
+
+		if node == parent._left {
+			done = tree_handle_subtree_growth(t, node, parent, -1)
+		} else {
+			done = tree_handle_subtree_growth(t, node, parent, +1)
+		}
+	}
+}
+
+@(private)
+tree_swap_with_successor :: proc "contextless" (
+	t: ^$T/Tree($Value),
+	x: ^Node(Value),
+) -> (
+	^Node(Value),
+	bool,
+) {
+	ret: ^Node(Value)
+	left_deleted: bool
+
+	y := x._right
+	if y._left == nil {
+		ret = y
+	} else {
+		q: ^Node(Value)
+
+		for {
+			q = y
+			if y = y._left; y._left == nil {
+				break
+			}
+		}
+
+		if q._left = y._right; q._left != nil {
+			q._left._parent = q
+		}
+		y._right = x._right
+		x._right._parent = y
+		ret = q
+		left_deleted = true
+	}
+
+	y._left = x._left
+	x._left._parent = y
+
+	y._parent = x._parent
+	y._balance = x._balance
+
+	tree_replace_child(t, x._parent, x, y)
+
+	return ret, left_deleted
+}
+
+@(private)
+tree_handle_subtree_shrink :: proc "contextless" (
+	t: ^$T/Tree($Value),
+	parent: ^Node(Value),
+	sign: i8,
+	left_deleted: ^bool,
+) -> ^Node(Value) {
+	old_balance_factor := parent._balance
+	if old_balance_factor == 0 {
+		node_adjust_balance_factor(parent, sign)
+		return nil
+	}
+
+	node: ^Node(Value)
+	new_balance_factor := old_balance_factor + sign
+	if new_balance_factor == 0 {
+		node_adjust_balance_factor(parent, sign)
+		node = parent
+	} else {
+		node = node_get_child(parent, sign)
+		if sign * node._balance >= 0 {
+			tree_rotate(t, parent, -sign)
+			if node._balance == 0 {
+				node_adjust_balance_factor(node, -sign)
+				return nil
+			}
+			node_adjust_balance_factor(parent, -sign)
+			node_adjust_balance_factor(node, -sign)
+		} else {
+			node = tree_double_rotate(t, node, parent, -sign)
+		}
+	}
+
+	parent := parent
+	if parent = node._parent; parent != nil {
+		left_deleted^ = node == parent._left
+	}
+	return parent
+}
+
+@(private)
+node_reset :: proc "contextless" (n: ^Node($Value)) {
+	// Mostly pointless as n will be deleted after this is called, but
+	// attempt to be able to catch cases of n not being in the tree.
+	n._parent = n
+	n._left = nil
+	n._right = nil
+	n._balance = 0
+}
+
+@(private)
+node_set_parent_balance :: #force_inline proc "contextless" (
+	n, parent: ^Node($Value),
+	balance: i8,
+) {
+	n._parent = parent
+	n._balance = balance
+}
+
+@(private)
+node_get_child :: #force_inline proc "contextless" (n: ^Node($Value), sign: i8) -> ^Node(Value) {
+	if sign < 0 {
+		return n._left
+	}
+	return n._right
+}
+
+@(private)
+node_next_or_prev_in_order :: proc "contextless" (
+	n: ^Node($Value),
+	direction: Direction,
+) -> ^Node(Value) {
+	next, tmp: ^Node(Value)
+	sign := i8(direction)
+
+	if next = node_get_child(n, +sign); next != nil {
+		for {
+			tmp = node_get_child(next, -sign)
+			if tmp == nil {
+				break
+			}
+			next = tmp
+		}
+	} else {
+		tmp, next = n, n._parent
+		for next != nil && tmp == node_get_child(next, +sign) {
+			tmp, next = next, next._parent
+		}
+	}
+	return next
+}
+
+@(private)
+node_set_child :: #force_inline proc "contextless" (
+	n: ^Node($Value),
+	sign: i8,
+	child: ^Node(Value),
+) {
+	if sign < 0 {
+		n._left = child
+	} else {
+		n._right = child
+	}
+}
+
+@(private)
+node_adjust_balance_factor :: #force_inline proc "contextless" (n: ^Node($Value), amount: i8) {
+	n._balance += amount
+}
+
+@(private)
+iterator_first :: proc "contextless" (it: ^Iterator($Value)) {
+	// This is private because behavior when the user manually calls
+	// iterator_first followed by iterator_next is unintuitive, since
+	// the first call to iterator_next MUST return the first node
+	// instead of advancing so that `for node in iterator_next(&next)`
+	// works as expected.
+
+	switch it._direction {
+	case .Forward:
+		it._cur = tree_first_or_last_in_order(it._tree, .Backward)
+	case .Backward:
+		it._cur = tree_first_or_last_in_order(it._tree, .Forward)
+	}
+
+	it._next = nil
+	it._called_next = false
+
+	if it._cur != nil {
+		it._next = node_next_or_prev_in_order(it._cur, it._direction)
+	}
+}
@@ -1,6 +1,6 @@
-package dynamic_bit_array
+package container_dynamic_bit_array

-import "core:intrinsics"
+import "base:intrinsics"
 import "core:mem"

 /*
@@ -27,27 +27,28 @@ Bit_Array_Iterator :: struct {
 	word_idx: int,
 	bit_idx:  uint,
 }
-
 /*
-	In:
-		- ba:   ^Bit_Array - the array to iterate over
+Wraps a `Bit_Array` into an Iterator

-	Out:
-		- it:   ^Bit_Array_Iterator - the iterator that holds iteration state
+Inputs:
+- ba: Pointer to the Bit_Array
+
+Returns:
+- it: Iterator struct
 */
 make_iterator :: proc (ba: ^Bit_Array) -> (it: Bit_Array_Iterator) {
 	return Bit_Array_Iterator { array = ba }
 }
-
 /*
-	In:
-		- it:    ^Bit_Array_Iterator - the iterator struct that holds the state.
+Returns the next bit, including its set-state. ok=false once exhausted

-	Out:
-		- set:    bool - the state of the bit at `index`
-		- index:  int - the next bit of the Bit_Array referenced by `it`.
-		- ok:	  bool - `true` if the iterator returned a valid index,
-			  `false` if there were no more bits
+Inputs:
+- it: The iterator that holds the state.
+
+Returns:
+- set: `true` if the bit at `index` is set.
+- index: The next bit of the Bit_Array referenced by `it`.
+- ok: `true` if the iterator can continue, `false` if the iterator is done
 */
 iterate_by_all :: proc (it: ^Bit_Array_Iterator) -> (set: bool, index: int, ok: bool) {
 	index = it.word_idx * NUM_BITS + int(it.bit_idx) + it.array.bias
@@ -64,39 +65,51 @@ iterate_by_all :: proc (it: ^Bit_Array_Iterator) -> (set: bool, index: int, ok:

 	return set, index, true
 }
-
 /*
-	In:
-		- it:     ^Bit_Array_Iterator - the iterator struct that holds the state.
+Returns the next Set Bit, for example if `0b1010`, then the iterator will return index={1, 3} over two calls.

-	Out:
-		- index:  int - the next set bit of the Bit_Array referenced by `it`.
-		- ok:	  bool - `true` if the iterator returned a valid index,
-			  `false` if there were no more bits set
+Inputs:
+- it: The iterator that holds the state.
+
+Returns:
+- index: The next *set* bit of the Bit_Array referenced by `it`.
+- ok: `true` if the iterator can continue, `false` if the iterator is done
 */
 iterate_by_set :: proc (it: ^Bit_Array_Iterator) -> (index: int, ok: bool) {
 	return iterate_internal_(it, true)
 }
-
 /*
-	In:
-		- it:	  ^Bit_Array_Iterator - the iterator struct that holds the state.
+Returns the next Unset Bit, for example if `0b1010`, then the iterator will return index={0, 2} over two calls.

-	Out:
-		- index:  int - the next unset bit of the Bit_Array referenced by `it`.
-		- ok:	  bool - `true` if the iterator returned a valid index,
-			  `false` if there were no more unset bits
+Inputs:
+- it: The iterator that holds the state.
+
+Returns:
+- index: The next *unset* bit of the Bit_Array referenced by `it`.
+- ok: `true` if the iterator can continue, `false` if the iterator is done
 */
 iterate_by_unset:: proc (it: ^Bit_Array_Iterator) -> (index: int, ok: bool) {
 	return iterate_internal_(it, false)
 }
+/*
+Iterates through set/unset bits

+*Private*
+
+Inputs:
+- it: The iterator that holds the state.
+- ITERATE_SET_BITS: `true` for returning only set bits, false for returning only unset bits
+
+Returns:
+- index: The next *unset* bit of the Bit_Array referenced by `it`.
+- ok: `true` if the iterator can continue, `false` if the iterator is done
+*/
@(private="file")
 iterate_internal_ :: proc (it: ^Bit_Array_Iterator, $ITERATE_SET_BITS: bool) -> (index: int, ok: bool) {
 	word := it.array.bits[it.word_idx] if len(it.array.bits) > it.word_idx else 0
 	when ! ITERATE_SET_BITS { word = ~word }

-	// if the word is empty or we have already gone over all the bits in it,
+	// If the word is empty or we have already gone over all the bits in it,
 	// b.bit_idx is greater than the index of any set bit in the word,
 	// meaning that word >> b.bit_idx == 0.
 	for it.word_idx < len(it.array.bits) && word >> it.bit_idx == 0 {
@@ -106,14 +119,14 @@ iterate_internal_ :: proc (it: ^Bit_Array_Iterator, $ITERATE_SET_BITS: bool) ->
 		when ! ITERATE_SET_BITS { word = ~word }
 	}

-	// if we are iterating the set bits, reaching the end of the array means we have no more bits to check
+	// If we are iterating the set bits, reaching the end of the array means we have no more bits to check
 	when ITERATE_SET_BITS {
 		if it.word_idx >= len(it.array.bits) {
 			return 0, false
 		}
 	}

-	// reaching here means that the word has some set bits
+	// Reaching here means that the word has some set bits
 	it.bit_idx += uint(intrinsics.count_trailing_zeros(word >> it.bit_idx))
 	index = it.word_idx * NUM_BITS + int(it.bit_idx) + it.array.bias

@@ -124,24 +137,21 @@ iterate_internal_ :: proc (it: ^Bit_Array_Iterator, $ITERATE_SET_BITS: bool) ->
 	}
 	return index, index <= it.array.max_index
 }
-
-
 /*
-	In:
-		- ba:    ^Bit_Array - a pointer to the Bit Array
-		- index: The bit index. Can be an enum member.
+Gets the state of a bit in the bit-array

-	Out:
-		- res:   The bit you're interested in.
-		- ok:    Whether the index was valid. Returns `false` if the index is smaller than the bias.
+Inputs:
+- ba: Pointer to the Bit_Array
+- index: Which bit in the array

-	The `ok` return value may be ignored.
+Returns:
+- res: `true` if the bit at `index` is set.
+- ok: Whether the index was valid. Returns `false` if the index is smaller than the bias.
 */
-get :: proc(ba: ^Bit_Array, #any_int index: uint, allocator := context.allocator) -> (res: bool, ok: bool) {
+get :: proc(ba: ^Bit_Array, #any_int index: uint) -> (res: bool, ok: bool) #optional_ok {
 	idx := int(index) - ba.bias

 	if ba == nil || int(index) < ba.bias { return false, false }
-	context.allocator = allocator

 	leg_index := idx >> INDEX_SHIFT
 	bit_index := idx &  INDEX_MASK
@@ -157,18 +167,36 @@ get :: proc(ba: ^Bit_Array, #any_int index: uint, allocator := context.allocator

 	return res, true
 }
-
 /*
-	In:
-		- ba:    ^Bit_Array - a pointer to the Bit Array
-		- index: The bit index. Can be an enum member.
+Gets the state of a bit in the bit-array

-	Out:
-		- ok:    Whether or not we managed to set requested bit.
+*Bypasses all Checks*

-	`set` automatically resizes the Bit Array to accommodate the requested index if needed.
+Inputs:
+- ba: Pointer to the Bit_Array
+- index: Which bit in the array
+
+Returns:
+- `true` if bit is set
 */
-set :: proc(ba: ^Bit_Array, #any_int index: uint, allocator := context.allocator) -> (ok: bool) {
+unsafe_get :: #force_inline proc(ba: ^Bit_Array, #any_int index: uint) -> bool #no_bounds_check {
+	return bool((ba.bits[index >> INDEX_SHIFT] >> uint(index & INDEX_MASK)) & 1)
+}
+/*
+Sets the state of a bit in the bit-array
+
+*Conditionally Allocates (Resizes backing data when `index > len(ba.bits)`)*
+
+Inputs:
+- ba: Pointer to the Bit_Array
+- index: Which bit in the array
+- set_to: `true` sets the bit on, `false` to turn it off
+- allocator: (default is context.allocator)
+
+Returns:
+- ok: Whether the set was successful, `false` on allocation failure or bad index
+*/
+set :: proc(ba: ^Bit_Array, #any_int index: uint, set_to: bool = true, allocator := context.allocator) -> (ok: bool) {

 	idx := int(index) - ba.bias

@@ -181,65 +209,97 @@ set :: proc(ba: ^Bit_Array, #any_int index: uint, allocator := context.allocator
 	resize_if_needed(ba, leg_index) or_return

 	ba.max_index = max(idx, ba.max_index)
-	ba.bits[leg_index] |= 1 << uint(bit_index)
+
+	if set_to{ ba.bits[leg_index] |= 1 << uint(bit_index) }
+	else { ba.bits[leg_index] &= ~(1 << uint(bit_index)) }
+
 	return true
 }
-
 /*
-	In:
-		- ba:    ^Bit_Array - a pointer to the Bit Array
-		- index: The bit index. Can be an enum member.
+Sets the state of a bit in the bit-array

-	Out:
-		- ok:    Whether or not we managed to unset requested bit.
+*Bypasses all checks*

-	`unset` automatically resizes the Bit Array to accommodate the requested index if needed.
+Inputs:
+- ba: Pointer to the Bit_Array
+- index: Which bit in the array
 */
-unset :: proc(ba: ^Bit_Array, #any_int index: uint, allocator := context.allocator) -> (ok: bool) {
-
-	idx := int(index) - ba.bias
-
-	if ba == nil || int(index) < ba.bias { return false }
-	context.allocator = allocator
-
-	leg_index := idx >> INDEX_SHIFT
-	bit_index := idx &  INDEX_MASK
-
-	resize_if_needed(ba, leg_index) or_return
-
-	ba.max_index = max(idx, ba.max_index)
-	ba.bits[leg_index] &= ~(1 << uint(bit_index))
-	return true
+unsafe_set :: proc(ba: ^Bit_Array, bit: int) #no_bounds_check {
+	ba.bits[bit >> INDEX_SHIFT] |= 1 << uint(bit & INDEX_MASK)
 }
-
 /*
-	A helper function to create a Bit Array with optional bias, in case your smallest index is non-zero (including negative).
+Unsets the state of a bit in the bit-array. (Convienence wrapper for `set`)
+
+*Conditionally Allocates (Resizes backing data when `index > len(ba.bits)`)*
+
+Inputs:
+- ba: Pointer to the Bit_Array
+- index: Which bit in the array
+- allocator: (default is context.allocator)
+
+Returns:
+- ok: Whether the unset was successful, `false` on allocation failure or bad index
 */
-create :: proc(max_index: int, min_index := 0, allocator := context.allocator) -> (res: ^Bit_Array, ok: bool) #optional_ok {
+unset :: #force_inline proc(ba: ^Bit_Array, #any_int index: uint, allocator := context.allocator) -> (ok: bool) {
+	return set(ba, index, false, allocator)
+}
+/*
+Unsets the state of a bit in the bit-array
+
+*Bypasses all Checks*
+
+Inputs:
+- ba: Pointer to the Bit_Array
+- index: Which bit in the array
+*/
+unsafe_unset :: proc(b: ^Bit_Array, bit: int) #no_bounds_check {
+	b.bits[bit >> INDEX_SHIFT] &= ~(1 << uint(bit & INDEX_MASK))
+}
+/*
+A helper function to create a Bit Array with optional bias, in case your smallest index is non-zero (including negative).
+
+*Allocates (`new(Bit_Array) & make(ba.bits)`)*
+
+Inputs:
+- max_index: maximum starting index
+- min_index: minimum starting index (used as a bias)
+- allocator: (default is context.allocator)
+
+Returns:
+- ba: Allocates a bit_Array, backing data is set to `max-min / 64` indices, rounded up (eg 65 - 0 allocates for [2]u64).
+*/
+create :: proc(max_index: int, min_index: int = 0, allocator := context.allocator) -> (res: ^Bit_Array, ok: bool) #optional_ok {
 	context.allocator = allocator
 	size_in_bits := max_index - min_index

 	if size_in_bits < 1 { return {}, false }

 	legs := size_in_bits >> INDEX_SHIFT
-
+	if size_in_bits & INDEX_MASK > 0 {legs+=1}
+	bits, err := make([dynamic]u64, legs)
+	ok = err == mem.Allocator_Error.None
 	res = new(Bit_Array)
+	res.bits         = bits
 	res.bias         = min_index
 	res.max_index    = max_index
 	res.free_pointer = true
-	return res, resize_if_needed(res, legs)
+	return
 }
-
 /*
-	Sets all bits to `false`.
+Sets all values in the Bit_Array to zero.
+
+Inputs:
+- ba: The target Bit_Array
 */
 clear :: proc(ba: ^Bit_Array) {
 	if ba == nil { return }
 	mem.zero_slice(ba.bits[:])
 }
-
 /*
-	Releases the memory used by the Bit Array.
+Deallocates the Bit_Array and its backing storage
+
+Inputs:
+- ba: The target Bit_Array
 */
 destroy :: proc(ba: ^Bit_Array) {
 	if ba == nil { return }
@@ -248,9 +308,8 @@ destroy :: proc(ba: ^Bit_Array) {
 		free(ba)
 	}
 }
-
 /*
-	Resizes the Bit Array. For internal use.
+	Resizes the Bit Array. For internal use. Provisions needed capacity+1
 	If you want to reserve the memory for a given-sized Bit Array up front, you can use `create`.
 */
@(private="file")
@@ -1,53 +1,52 @@
-package dynamic_bit_array
-
 /*
-	The Bit Array can be used in several ways:
+The Bit Array can be used in several ways:

-	-- By default you don't need to instantiate a Bit Array:
+- By default you don't need to instantiate a Bit Array:

-		package test
+	package test

-		import "core:fmt"
-		import "core:container/bit_array"
+	import "core:fmt"
+	import "core:container/bit_array"

-		main :: proc() {
-			using bit_array
+	main :: proc() {
+		using bit_array

-			bits: Bit_Array
+		bits: Bit_Array

-			// returns `true`
-			fmt.println(set(&bits, 42))
+		// returns `true`
+		fmt.println(set(&bits, 42))

-			// returns `false`, `false`, because this Bit Array wasn't created to allow negative indices.
-			was_set, was_retrieved := get(&bits, -1)
-			fmt.println(was_set, was_retrieved) 
-			destroy(&bits)
+		// returns `false`, `false`, because this Bit Array wasn't created to allow negative indices.
+		was_set, was_retrieved := get(&bits, -1)
+		fmt.println(was_set, was_retrieved) 
+		destroy(&bits)
+	}
+
+- A Bit Array can optionally allow for negative indices, if the minimum value was given during creation:
+
+	package test
+
+	import "core:fmt"
+	import "core:container/bit_array"
+
+	main :: proc() {
+		Foo :: enum int {
+			Negative_Test = -42,
+			Bar           = 420,
+			Leaves        = 69105,
 		}

-	-- A Bit Array can optionally allow for negative indices, if the mininum value was given during creation:
+		using bit_array

-		package test
+		bits := create(int(max(Foo)), int(min(Foo)))
+		defer destroy(bits)

-		import "core:fmt"
-		import "core:container/bit_array"
-
-		main :: proc() {
-			Foo :: enum int {
-				Negative_Test = -42,
-				Bar           = 420,
-				Leaves        = 69105,
-			}
-
-			using bit_array
-
-			bits := create(int(max(Foo)), int(min(Foo)))
-			defer destroy(bits)
-
-			fmt.printf("Set(Bar):           %v\n",     set(bits, Foo.Bar))
-			fmt.printf("Get(Bar):           %v, %v\n", get(bits, Foo.Bar))
-			fmt.printf("Set(Negative_Test): %v\n",     set(bits, Foo.Negative_Test))
-			fmt.printf("Get(Leaves):        %v, %v\n", get(bits, Foo.Leaves))
-			fmt.printf("Get(Negative_Test): %v, %v\n", get(bits, Foo.Negative_Test))
-			fmt.printf("Freed.\n")
-		}
-*/
+		fmt.printf("Set(Bar):           %v\n",     set(bits, Foo.Bar))
+		fmt.printf("Get(Bar):           %v, %v\n", get(bits, Foo.Bar))
+		fmt.printf("Set(Negative_Test): %v\n",     set(bits, Foo.Negative_Test))
+		fmt.printf("Get(Leaves):        %v, %v\n", get(bits, Foo.Leaves))
+		fmt.printf("Get(Negative_Test): %v, %v\n", get(bits, Foo.Negative_Test))
+		fmt.printf("Freed.\n")
+	}
+*/
+package container_dynamic_bit_array
@@ -1,6 +1,6 @@
 package container_intrusive_list

-import "core:intrinsics"
+import "base:intrinsics"

 // An intrusive doubly-linked list
 //
@@ -20,10 +20,10 @@ List :: struct {


 Node :: struct {
-	next, prev: ^Node,
+	prev, next: ^Node,
 }

-push_front :: proc(list: ^List, node: ^Node) {
+push_front :: proc "contextless" (list: ^List, node: ^Node) {
 	if list.head != nil {
 		list.head.prev = node
 		node.prev, node.next = nil, list.head
@@ -34,7 +34,7 @@ push_front :: proc(list: ^List, node: ^Node) {
 	}
 }

-push_back :: proc(list: ^List, node: ^Node) {
+push_back :: proc "contextless" (list: ^List, node: ^Node) {
 	if list.tail != nil {
 		list.tail.next = node
 		node.prev, node.next = list.tail, nil
@@ -45,7 +45,7 @@ push_back :: proc(list: ^List, node: ^Node) {
 	}
 }

-remove :: proc(list: ^List, node: ^Node) {
+remove :: proc "contextless" (list: ^List, node: ^Node) {
 	if node != nil {
 		if node.next != nil {
 			node.next.prev = node.prev
@@ -83,12 +83,34 @@ remove_by_proc :: proc(list: ^List, to_erase: proc(^Node) -> bool) {
 	}
 }

+remove_by_proc_contextless :: proc(list: ^List, to_erase: proc "contextless" (^Node) -> bool) {
+	for node := list.head; node != nil; {
+		next := node.next
+		if to_erase(node) {
+			if node.next != nil {
+				node.next.prev = node.prev
+			}
+			if node.prev != nil {
+				node.prev.next = node.next
+			}
+			if list.head == node {
+				list.head = node.next
+			}
+			if list.tail == node {
+				list.tail = node.prev
+			}
+		}
+		node = next
+	}
+}

-is_empty :: proc(list: ^List) -> bool {
+
+
+is_empty :: proc "contextless" (list: ^List) -> bool {
 	return list.head == nil
 }

-pop_front :: proc(list: ^List) -> ^Node {
+pop_front :: proc "contextless" (list: ^List) -> ^Node {
 	link := list.head
 	if link == nil {
 		return nil
@@ -108,7 +130,7 @@ pop_front :: proc(list: ^List) -> ^Node {
 	return link

 }
-pop_back :: proc(list: ^List) -> ^Node {
+pop_back :: proc "contextless" (list: ^List) -> ^Node {
 	link := list.tail
 	if link == nil {
 		return nil
@@ -134,25 +156,25 @@ Iterator :: struct($T: typeid) {
 	offset: uintptr,
 }

-iterator_head :: proc(list: List, $T: typeid, $field_name: string) -> Iterator(T)
+iterator_head :: proc "contextless" (list: List, $T: typeid, $field_name: string) -> Iterator(T)
 	where intrinsics.type_has_field(T, field_name),
 	      intrinsics.type_field_type(T, field_name) == Node {
 	return {list.head, offset_of_by_string(T, field_name)}
 }

-iterator_tail :: proc(list: List, $T: typeid, $field_name: string) -> Iterator(T)
+iterator_tail :: proc "contextless" (list: List, $T: typeid, $field_name: string) -> Iterator(T)
 	where intrinsics.type_has_field(T, field_name),
 	      intrinsics.type_field_type(T, field_name) == Node {
 	return {list.tail, offset_of_by_string(T, field_name)}
 }

-iterator_from_node :: proc(node: ^Node, $T: typeid, $field_name: string) -> Iterator(T)
+iterator_from_node :: proc "contextless" (node: ^Node, $T: typeid, $field_name: string) -> Iterator(T)
 	where intrinsics.type_has_field(T, field_name),
 	      intrinsics.type_field_type(T, field_name) == Node {
 	return {node, offset_of_by_string(T, field_name)}
 }

-iterate_next :: proc(it: ^Iterator($T)) -> (ptr: ^T, ok: bool) {
+iterate_next :: proc "contextless" (it: ^Iterator($T)) -> (ptr: ^T, ok: bool) {
 	node := it.curr
 	if node == nil {
 		return nil, false
@@ -162,7 +184,7 @@ iterate_next :: proc(it: ^Iterator($T)) -> (ptr: ^T, ok: bool) {
 	return (^T)(uintptr(node) - it.offset), true
 }

-iterate_prev :: proc(it: ^Iterator($T)) -> (ptr: ^T, ok: bool) {
+iterate_prev :: proc "contextless" (it: ^Iterator($T)) -> (ptr: ^T, ok: bool) {
 	node := it.curr
 	if node == nil {
 		return nil, false
@@ -1,7 +1,7 @@
 package container_lru

-import "core:runtime"
-import "core:intrinsics"
+import "base:runtime"
+import "base:intrinsics"
 _ :: runtime
 _ :: intrinsics

@@ -1,6 +1,6 @@
 package container_priority_queue

-import "core:builtin"
+import "base:builtin"

 Priority_Queue :: struct($T: typeid) {
 	queue: [dynamic]T,
@@ -140,3 +140,18 @@ remove :: proc(pq: ^$Q/Priority_Queue($T), i: int) -> (value: T, ok: bool) {
 	return
 }

+peek_safe :: proc(pq: $Q/Priority_Queue($T), loc := #caller_location) -> (res: T, ok: bool) {
+	if builtin.len(pq.queue) > 0 {
+		return pq.queue[0], true
+	}
+	return
+}
+
+peek :: proc(pq: $Q/Priority_Queue($T), loc := #caller_location) -> (res: T) {
+	assert(condition=builtin.len(pq.queue)>0, loc=loc)
+
+	if builtin.len(pq.queue) > 0 {
+		return pq.queue[0]
+	}
+	return
+}
@@ -1,7 +1,7 @@
 package container_queue

-import "core:builtin"
-import "core:runtime"
+import "base:builtin"
+import "base:runtime"
 _ :: runtime

 // Dynamically resizable double-ended queue/ring-buffer
@@ -14,7 +14,7 @@ Queue :: struct($T: typeid) {
 DEFAULT_CAPACITY :: 16

 // Procedure to initialize a queue
-init :: proc(q: ^$Q/Queue($T), capacity := DEFAULT_CAPACITY, allocator := context.allocator) -> bool {
+init :: proc(q: ^$Q/Queue($T), capacity := DEFAULT_CAPACITY, allocator := context.allocator) -> runtime.Allocator_Error {
 	if q.data.allocator.procedure == nil {
 		q.data.allocator = allocator
 	}
@@ -22,7 +22,9 @@ init :: proc(q: ^$Q/Queue($T), capacity := DEFAULT_CAPACITY, allocator := contex
 	return reserve(q, capacity)
 }

-// Procedure to initialize a queue from a fixed backing slice
+// Procedure to initialize a queue from a fixed backing slice.
+// The contents of the `backing` will be overwritten as items are pushed onto the `Queue`.
+// Any previous contents are not available.
 init_from_slice :: proc(q: ^$Q/Queue($T), backing: []T) -> bool {
 	clear(q)
 	q.data = transmute([dynamic]T)runtime.Raw_Dynamic_Array{
@@ -34,6 +36,21 @@ init_from_slice :: proc(q: ^$Q/Queue($T), backing: []T) -> bool {
 	return true
 }

+// Procedure to initialize a queue from a fixed backing slice.
+// Existing contents are preserved and available on the queue.
+init_with_contents :: proc(q: ^$Q/Queue($T), backing: []T) -> bool {
+	clear(q)
+	q.data = transmute([dynamic]T)runtime.Raw_Dynamic_Array{
+		data = raw_data(backing),
+		len = builtin.len(backing),
+		cap = builtin.len(backing),
+		allocator = {procedure=runtime.nil_allocator_proc, data=nil},
+	}
+	q.len    = len(backing)
+	q.offset = len(backing)
+	return true
+}
+
 // Procedure to destroy a queue
 destroy :: proc(q: ^$Q/Queue($T)) {
 	delete(q.data)
@@ -55,11 +72,11 @@ space :: proc(q: $Q/Queue($T)) -> int {
 }

 // Reserve enough space for at least the specified capacity
-reserve :: proc(q: ^$Q/Queue($T), capacity: int) -> bool {
-	if uint(capacity) > q.len {
+reserve :: proc(q: ^$Q/Queue($T), capacity: int) -> runtime.Allocator_Error {
+	if capacity > space(q^) {
 		return _grow(q, uint(capacity)) 
 	}
-	return true
+	return nil
 }


@@ -112,25 +129,25 @@ peek_back :: proc(q: ^$Q/Queue($T), loc := #caller_location) -> ^T {
 }

 // Push an element to the back of the queue
-push_back :: proc(q: ^$Q/Queue($T), elem: T) -> bool {
+push_back :: proc(q: ^$Q/Queue($T), elem: T) -> (ok: bool, err: runtime.Allocator_Error) {
 	if space(q^) == 0 {
 		_grow(q) or_return
 	}
 	idx := (q.offset+uint(q.len))%builtin.len(q.data)
 	q.data[idx] = elem
 	q.len += 1
-	return true
+	return true, nil
 }

 // Push an element to the front of the queue
-push_front :: proc(q: ^$Q/Queue($T), elem: T) -> bool {
+push_front :: proc(q: ^$Q/Queue($T), elem: T) -> (ok: bool, err: runtime.Allocator_Error)  {
 	if space(q^) == 0 {
 		_grow(q) or_return
 	}	
 	q.offset = uint(q.offset - 1 + builtin.len(q.data)) % builtin.len(q.data)
 	q.len += 1
 	q.data[q.offset] = elem
-	return true
+	return true, nil
 }


@@ -173,7 +190,7 @@ pop_front_safe :: proc(q: ^$Q/Queue($T)) -> (elem: T, ok: bool) {
 }

 // Push multiple elements to the front of the queue
-push_back_elems :: proc(q: ^$Q/Queue($T), elems: ..T) -> bool {
+push_back_elems :: proc(q: ^$Q/Queue($T), elems: ..T) -> (ok: bool, err: runtime.Allocator_Error)  {
 	n := uint(builtin.len(elems))
 	if space(q^) < int(n) {
 		_grow(q, q.len + n) or_return
@@ -188,7 +205,7 @@ push_back_elems :: proc(q: ^$Q/Queue($T), elems: ..T) -> bool {
 	copy(q.data[insert_from:], elems[:insert_to])
 	copy(q.data[:insert_from], elems[insert_to:])
 	q.len += n
-	return true
+	return true, nil
 }

 // Consume `n` elements from the front of the queue
@@ -225,7 +242,7 @@ clear :: proc(q: ^$Q/Queue($T)) {


 // Internal growinh procedure
-_grow :: proc(q: ^$Q/Queue($T), min_capacity: uint = 0) -> bool {
+_grow :: proc(q: ^$Q/Queue($T), min_capacity: uint = 0) -> runtime.Allocator_Error {
 	new_capacity := max(min_capacity, uint(8), uint(builtin.len(q.data))*2)
 	n := uint(builtin.len(q.data))
 	builtin.resize(&q.data, int(new_capacity)) or_return
@@ -234,5 +251,5 @@ _grow :: proc(q: ^$Q/Queue($T), min_capacity: uint = 0) -> bool {
 		copy(q.data[new_capacity-diff:], q.data[q.offset:][:diff])
 		q.offset += new_capacity - n
 	}
-	return true
+	return nil
 }
@@ -1,7 +1,7 @@
 package container_small_array

-import "core:builtin"
-import "core:runtime"
+import "base:builtin"
+import "base:runtime"
 _ :: runtime

 Small_Array :: struct($N: int, $T: typeid) where N >= 0 {
@@ -3,8 +3,8 @@
 // map type is being used to accelerate lookups.
 package container_topological_sort

-import "core:intrinsics"
-import "core:runtime"
+import "base:intrinsics"
+import "base:runtime"
 _ :: intrinsics
 _ :: runtime

@@ -32,7 +32,7 @@ init :: proc(sorter: ^$S/Sorter($K)) {
 }

 destroy :: proc(sorter: ^$S/Sorter($K)) {
-	for _, v in &sorter.relations {
+	for _, v in sorter.relations {
 		delete(v.dependents)
 	}
 	delete(sorter.relations)
@@ -80,11 +80,13 @@ sort :: proc(sorter: ^$S/Sorter($K)) -> (sorted, cycled: [dynamic]K) {
 		}
 	}

-	for root in &sorted do for k, _ in relations[root].dependents {
-		relation := &relations[k]
-		relation.dependencies -= 1
-		if relation.dependencies == 0 {
-			append(&sorted, k)
+	for root in sorted {
+		for k, _ in relations[root].dependents {
+			relation := &relations[k]
+			relation.dependencies -= 1
+			if relation.dependencies == 0 {
+				append(&sorted, k)
+			}
 		}
 	}

@@ -1,95 +1,32 @@
 # crypto
-A crypto library for the Odin language
+
+A cryptography library for the Odin language.

 ## Supported
-This library offers various algorithms implemented in Odin.
-Please see the chart below for the options.  

-## Hashing algorithms
-| Algorithm                                                                                                    |                  |
-|:-------------------------------------------------------------------------------------------------------------|:-----------------|
-| [BLAKE](https://web.archive.org/web/20190915215948/https://131002.net/blake)                                 | &#10004;&#65039; |
-| [BLAKE2B](https://datatracker.ietf.org/doc/html/rfc7693)                                                     | &#10004;&#65039; |
-| [BLAKE2S](https://datatracker.ietf.org/doc/html/rfc7693)                                                     | &#10004;&#65039; |
-| [GOST](https://datatracker.ietf.org/doc/html/rfc5831)                                                        | &#10004;&#65039; |
-| [Grøstl](http://www.groestl.info/Groestl.zip)                                                                | &#10004;&#65039; |
-| [HAVAL](https://web.archive.org/web/20150111210116/http://labs.calyptix.com/haval.php)                       | &#10004;&#65039; |
-| [JH](https://www3.ntu.edu.sg/home/wuhj/research/jh/index.html)                                               | &#10004;&#65039; |
-| [Keccak](https://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.202.pdf)                                           | &#10004;&#65039; |
-| [MD2](https://datatracker.ietf.org/doc/html/rfc1319)                                                         | &#10004;&#65039; |
-| [MD4](https://datatracker.ietf.org/doc/html/rfc1320)                                                         | &#10004;&#65039; |
-| [MD5](https://datatracker.ietf.org/doc/html/rfc1321)                                                         | &#10004;&#65039; |
-| [RIPEMD](https://homes.esat.kuleuven.be/~bosselae/ripemd160.html)                                            | &#10004;&#65039; |
-| [SHA-1](https://datatracker.ietf.org/doc/html/rfc3174)                                                       | &#10004;&#65039; |
-| [SHA-2](https://csrc.nist.gov/csrc/media/publications/fips/180/2/archive/2002-08-01/documents/fips180-2.pdf) | &#10004;&#65039; |
-| [SHA-3](https://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.202.pdf)                                            | &#10004;&#65039; |
-| [SHAKE](https://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.202.pdf)                                            | &#10004;&#65039; |
-| [SM3](https://datatracker.ietf.org/doc/html/draft-sca-cfrg-sm3-02)                                           | &#10004;&#65039; |
-| [Streebog](https://datatracker.ietf.org/doc/html/rfc6986)                                                    | &#10004;&#65039; |
-| [Tiger](https://www.cs.technion.ac.il/~biham/Reports/Tiger/)                                                 | &#10004;&#65039; |
-| [Tiger2](https://www.cs.technion.ac.il/~biham/Reports/Tiger/)                                                | &#10004;&#65039; |
-| [Whirlpool](https://web.archive.org/web/20171129084214/http://www.larc.usp.br/~pbarreto/WhirlpoolPage.html)  | &#10004;&#65039; |
+This package offers various algorithms implemented in Odin, along with
+useful helpers such as access to the system entropy source, and a
+constant-time byte comparison.

-#### High level API
-Each hash algorithm contains a procedure group named `hash`, or if the algorithm provides more than one digest size `hash_<size>`\*.  
-Included in these groups are six procedures.
-* `hash_string` - Hash a given string and return the computed hash. Just calls `hash_bytes` internally
-* `hash_bytes` - Hash a given byte slice and return the computed hash
-* `hash_string_to_buffer` - Hash a given string and put the computed hash in the second proc parameter. Just calls `hash_bytes_to_buffer` internally
-* `hash_bytes_to_buffer` - Hash a given string and put the computed hash in the second proc parameter. The destination buffer has to be at least as big as the digest size of the hash
-* `hash_stream` - Takes a stream from io.Stream and returns the computed hash from it
-* `hash_file` - Takes a file handle and returns the computed hash from it. A second optional boolean parameter controls if the file is streamed (this is the default) or read at once (set to true)
+## Implementation considerations

-\* On some algorithms there is another part to the name, since they might offer control about additional parameters.  
-For instance, `HAVAL` offers different sizes as well as three different round amounts.  
-Computing a 256-bit hash with 3 rounds is therefore achieved by calling `haval.hash_256_3(...)`.
+- The crypto packages are not thread-safe.
+- 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."
+- All of these packages have not received independent third party review.

-#### Low level API
-The above mentioned procedures internally call three procedures: `init`, `update` and `final`.
-You may also directly call them, if you wish.
+## License

-#### Example
-```odin
-package crypto_example
-
-// Import the desired package
-import "core:crypto/md4"
-
-main :: proc() {
-    input := "foo"
-
-    // Compute the hash, using the high level API
-    computed_hash := md4.hash(input)
-
-    // Variant that takes a destination buffer, instead of returning the computed hash
-    hash := make([]byte, md4.DIGEST_SIZE) // @note: Destination buffer has to be at least as big as the digest size of the hash
-    md4.hash(input, hash[:])
-
-    // Compute the hash, using the low level API
-    ctx: md4.Md4_Context
-    computed_hash_low: [16]byte
-    md4.init(&ctx)
-    md4.update(&ctx, transmute([]byte)input)
-    md4.final(&ctx, computed_hash_low[:])
-}
-```
-For example uses of all available algorithms, please see the tests within `tests/core/crypto`.
-
-#### Thread safety
-The crypto package is not thread-safe at the moment. This may change in the future.
-
-### Disclaimer
-The algorithms were ported out of curiosity and due to interest in the field.
-We have not had any of the code verified by a third party or tested/fuzzed by any automatic means.
-Wherever we were able to find official test vectors, those were used to verify the implementation.
-We do not recommend using them in a production environment, without any additional testing and/or verification.
-
-### ToDo
-* Ciphers (Symmetric, Asymmetric)
-* MACs (Message Authentication Code)
-* CSPRNGs (Cryptographically Secure PseudoRandom Number Generator)
-* KDFs (Key Derivation Function)
-* KEAs (Key Exchange Algorithm)
-
-### License
 This library is made available under the BSD-3 license.
@@ -10,12 +10,13 @@ package _blake2
    Implementation of the BLAKE2 hashing algorithm, as defined in <https://datatracker.ietf.org/doc/html/rfc7693> and <https://www.blake2.net/>
 */

-import "../util"
+import "core:encoding/endian"
+import "core:mem"

-BLAKE2S_BLOCK_SIZE  :: 64
-BLAKE2S_SIZE        :: 32
-BLAKE2B_BLOCK_SIZE  :: 128
-BLAKE2B_SIZE        :: 64
+BLAKE2S_BLOCK_SIZE :: 64
+BLAKE2S_SIZE :: 32
+BLAKE2B_BLOCK_SIZE :: 128
+BLAKE2B_SIZE :: 64

 Blake2s_Context :: struct {
 	h:            [8]u32,
@@ -28,7 +29,8 @@ Blake2s_Context :: struct {
 	is_keyed:     bool,
 	size:         byte,
 	is_last_node: bool,
-	cfg:		  Blake2_Config,
+
+	is_initialized: bool,
 }

 Blake2b_Context :: struct {
@@ -42,15 +44,18 @@ Blake2b_Context :: struct {
 	is_keyed:     bool,
 	size:         byte,
 	is_last_node: bool,
-	cfg:		  Blake2_Config,
+
+	is_initialized: bool,
 }

 Blake2_Config :: struct {
-    size:   byte,
-	key:    []byte, 
-	salt:   []byte, 
+	size:   byte,
+	key:    []byte,
+	salt:   []byte,
 	person: []byte,
-	tree:   union{Blake2_Tree},
+	tree:   union {
+		Blake2_Tree,
+	},
 }

 Blake2_Tree :: struct {
@@ -63,11 +68,13 @@ Blake2_Tree :: struct {
 	is_last_node:    bool,
 }

+@(private)
 BLAKE2S_IV := [8]u32 {
 	0x6a09e667, 0xbb67ae85, 0x3c6ef372, 0xa54ff53a,
 	0x510e527f, 0x9b05688c, 0x1f83d9ab, 0x5be0cd19,
 }

+@(private)
 BLAKE2B_IV := [8]u64 {
 	0x6a09e667f3bcc908, 0xbb67ae8584caa73b,
 	0x3c6ef372fe94f82b, 0xa54ff53a5f1d36f1,
@@ -75,80 +82,94 @@ BLAKE2B_IV := [8]u64 {
 	0x1f83d9abfb41bd6b, 0x5be0cd19137e2179,
 }

-init :: proc(ctx: ^$T) {
+init :: proc(ctx: ^$T, cfg: ^Blake2_Config) {
 	when T == Blake2s_Context {
-		block_size :: BLAKE2S_BLOCK_SIZE
+		max_size :: BLAKE2S_SIZE
 	} else when T == Blake2b_Context {
-		block_size :: BLAKE2B_BLOCK_SIZE
+		max_size :: BLAKE2B_SIZE
 	}

-	p := make([]byte, block_size)
-	defer delete(p)
+	if cfg.size > max_size {
+		panic("blake2: requested output size exceeeds algorithm max")
+	}

-	p[0] = ctx.cfg.size
-	p[1] = byte(len(ctx.cfg.key))
+	// To save having to allocate a scratch buffer, use the internal
+	// data buffer (`ctx.x`), as it is exactly the correct size.
+	p := ctx.x[:]

-	if ctx.cfg.salt != nil {
+	p[0] = cfg.size
+	p[1] = byte(len(cfg.key))
+
+	if cfg.salt != nil {
 		when T == Blake2s_Context {
-			copy(p[16:], ctx.cfg.salt)
+			copy(p[16:], cfg.salt)
 		} else when T == Blake2b_Context {
-			copy(p[32:], ctx.cfg.salt)
+			copy(p[32:], cfg.salt)
 		}
 	}
-	if ctx.cfg.person != nil {
+	if cfg.person != nil {
 		when T == Blake2s_Context {
-			copy(p[24:], ctx.cfg.person)
+			copy(p[24:], cfg.person)
 		} else when T == Blake2b_Context {
-			copy(p[48:], ctx.cfg.person)
+			copy(p[48:], cfg.person)
 		}
 	}

-	if ctx.cfg.tree != nil {
-		p[2] = ctx.cfg.tree.(Blake2_Tree).fanout
-		p[3] = ctx.cfg.tree.(Blake2_Tree).max_depth
-		util.PUT_U32_LE(p[4:], ctx.cfg.tree.(Blake2_Tree).leaf_size)
+	if cfg.tree != nil {
+		p[2] = cfg.tree.(Blake2_Tree).fanout
+		p[3] = cfg.tree.(Blake2_Tree).max_depth
+		endian.unchecked_put_u32le(p[4:], cfg.tree.(Blake2_Tree).leaf_size)
 		when T == Blake2s_Context {
-			p[8]  = byte(ctx.cfg.tree.(Blake2_Tree).node_offset)
-			p[9]  = byte(ctx.cfg.tree.(Blake2_Tree).node_offset >> 8)
-			p[10] = byte(ctx.cfg.tree.(Blake2_Tree).node_offset >> 16)
-			p[11] = byte(ctx.cfg.tree.(Blake2_Tree).node_offset >> 24)
-			p[12] = byte(ctx.cfg.tree.(Blake2_Tree).node_offset >> 32)
-			p[13] = byte(ctx.cfg.tree.(Blake2_Tree).node_offset >> 40)
-			p[14] = ctx.cfg.tree.(Blake2_Tree).node_depth
-			p[15] = ctx.cfg.tree.(Blake2_Tree).inner_hash_size
+			p[8] = byte(cfg.tree.(Blake2_Tree).node_offset)
+			p[9] = byte(cfg.tree.(Blake2_Tree).node_offset >> 8)
+			p[10] = byte(cfg.tree.(Blake2_Tree).node_offset >> 16)
+			p[11] = byte(cfg.tree.(Blake2_Tree).node_offset >> 24)
+			p[12] = byte(cfg.tree.(Blake2_Tree).node_offset >> 32)
+			p[13] = byte(cfg.tree.(Blake2_Tree).node_offset >> 40)
+			p[14] = cfg.tree.(Blake2_Tree).node_depth
+			p[15] = cfg.tree.(Blake2_Tree).inner_hash_size
 		} else when T == Blake2b_Context {
-			util.PUT_U64_LE(p[8:], ctx.cfg.tree.(Blake2_Tree).node_offset)
-			p[16] = ctx.cfg.tree.(Blake2_Tree).node_depth
-			p[17] = ctx.cfg.tree.(Blake2_Tree).inner_hash_size
+			endian.unchecked_put_u64le(p[8:], cfg.tree.(Blake2_Tree).node_offset)
+			p[16] = cfg.tree.(Blake2_Tree).node_depth
+			p[17] = cfg.tree.(Blake2_Tree).inner_hash_size
 		}
 	} else {
 		p[2], p[3] = 1, 1
 	}
-	ctx.size = ctx.cfg.size
+	ctx.size = cfg.size
 	for i := 0; i < 8; i += 1 {
 		when T == Blake2s_Context {
-			ctx.h[i] = BLAKE2S_IV[i] ~ util.U32_LE(p[i * 4:])
+			ctx.h[i] = BLAKE2S_IV[i] ~ endian.unchecked_get_u32le(p[i * 4:])
 		}
 		when T == Blake2b_Context {
-			ctx.h[i] = BLAKE2B_IV[i] ~ util.U64_LE(p[i * 8:])
+			ctx.h[i] = BLAKE2B_IV[i] ~ endian.unchecked_get_u64le(p[i * 8:])
 		}
 	}
-	if ctx.cfg.tree != nil && ctx.cfg.tree.(Blake2_Tree).is_last_node {
+
+	mem.zero(&ctx.x, size_of(ctx.x)) // Done with the scratch space, no barrier.
+
+	if cfg.tree != nil && cfg.tree.(Blake2_Tree).is_last_node {
 		ctx.is_last_node = true
 	}
-	if len(ctx.cfg.key) > 0 {
-		copy(ctx.padded_key[:], ctx.cfg.key)
+	if len(cfg.key) > 0 {
+		copy(ctx.padded_key[:], cfg.key)
 		update(ctx, ctx.padded_key[:])
 		ctx.is_keyed = true
 	}
 	copy(ctx.ih[:], ctx.h[:])
-	copy(ctx.h[:],  ctx.ih[:])
+	copy(ctx.h[:], ctx.ih[:])
 	if ctx.is_keyed {
 		update(ctx, ctx.padded_key[:])
 	}
+
+	ctx.nx = 0
+
+	ctx.is_initialized = true
 }

-update :: proc "contextless" (ctx: ^$T, p: []byte) {
+update :: proc(ctx: ^$T, p: []byte) {
+	assert(ctx.is_initialized)
+
 	p := p
 	when T == Blake2s_Context {
 		block_size :: BLAKE2S_BLOCK_SIZE
@@ -174,15 +195,43 @@ update :: proc "contextless" (ctx: ^$T, p: []byte) {
 	ctx.nx += copy(ctx.x[ctx.nx:], p)
 }

-final :: proc "contextless" (ctx: ^$T, hash: []byte) {
-	when T == Blake2s_Context {
-		blake2s_final(ctx, hash)
+final :: proc(ctx: ^$T, hash: []byte, finalize_clone: bool = false) {
+	assert(ctx.is_initialized)
+
+	ctx := ctx
+	if finalize_clone {
+		tmp_ctx: T
+		clone(&tmp_ctx, ctx)
+		ctx = &tmp_ctx
 	}
-	when T == Blake2b_Context {
+	defer(reset(ctx))
+
+	when T == Blake2s_Context {
+		if len(hash) < int(ctx.size) {
+			panic("crypto/blake2s: invalid destination digest size")
+		}
+		blake2s_final(ctx, hash)
+	} else when T == Blake2b_Context {
+		if len(hash) < int(ctx.size) {
+			panic("crypto/blake2b: invalid destination digest size")
+		}
 		blake2b_final(ctx, hash)
 	}
 }

+clone :: proc(ctx, other: ^$T) {
+	ctx^ = other^
+}
+
+reset :: proc(ctx: ^$T) {
+	if !ctx.is_initialized {
+		return
+	}
+
+	mem.zero_explicit(ctx, size_of(ctx^))
+}
+
+@(private)
 blake2s_final :: proc "contextless" (ctx: ^Blake2s_Context, hash: []byte) {
 	if ctx.is_keyed {
 		for i := 0; i < len(ctx.padded_key); i += 1 {
@@ -203,16 +252,14 @@ blake2s_final :: proc "contextless" (ctx: ^Blake2s_Context, hash: []byte) {

 	blocks(ctx, ctx.x[:])

-	j := 0
-	for s, _ in ctx.h[:(ctx.size - 1) / 4 + 1] {
-		hash[j + 0] = byte(s >> 0)
-		hash[j + 1] = byte(s >> 8)
-		hash[j + 2] = byte(s >> 16)
-		hash[j + 3] = byte(s >> 24)
-		j += 4
+	dst: [BLAKE2S_SIZE]byte
+	for i := 0; i < BLAKE2S_SIZE / 4; i += 1 {
+		endian.unchecked_put_u32le(dst[i * 4:], ctx.h[i])
 	}
+	copy(hash, dst[:])
 }

+@(private)
 blake2b_final :: proc "contextless" (ctx: ^Blake2b_Context, hash: []byte) {
 	if ctx.is_keyed {
 		for i := 0; i < len(ctx.padded_key); i += 1 {
@@ -229,56 +276,52 @@ blake2b_final :: proc "contextless" (ctx: ^Blake2b_Context, hash: []byte) {
 	ctx.f[0] = 0xffffffffffffffff
 	if ctx.is_last_node {
 		ctx.f[1] = 0xffffffffffffffff
-	} 
+	}

 	blocks(ctx, ctx.x[:])

-	j := 0
-	for s, _ in ctx.h[:(ctx.size - 1) / 8 + 1] {
-		hash[j + 0] = byte(s >> 0)
-		hash[j + 1] = byte(s >> 8)
-		hash[j + 2] = byte(s >> 16)
-		hash[j + 3] = byte(s >> 24)
-		hash[j + 4] = byte(s >> 32)
-		hash[j + 5] = byte(s >> 40)
-		hash[j + 6] = byte(s >> 48)
-		hash[j + 7] = byte(s >> 56)
-		j += 8
+	dst: [BLAKE2B_SIZE]byte
+	for i := 0; i < BLAKE2B_SIZE / 8; i += 1 {
+		endian.unchecked_put_u64le(dst[i * 8:], ctx.h[i])
 	}
+	copy(hash, dst[:])
 }

+@(private)
 blocks :: proc "contextless" (ctx: ^$T, p: []byte) {
 	when T == Blake2s_Context {
 		blake2s_blocks(ctx, p)
-	}
-	when T == Blake2b_Context {
+	} else when T == Blake2b_Context {
 		blake2b_blocks(ctx, p)
 	}
 }

+@(private)
 blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []byte) {
-	h0, h1, h2, h3, h4, h5, h6, h7 := ctx.h[0], ctx.h[1], ctx.h[2], ctx.h[3], ctx.h[4], ctx.h[5], ctx.h[6], ctx.h[7]
+	h0, h1, h2, h3, h4, h5, h6, h7 :=
+		ctx.h[0], ctx.h[1], ctx.h[2], ctx.h[3], ctx.h[4], ctx.h[5], ctx.h[6], ctx.h[7]
 	p := p
 	for len(p) >= BLAKE2S_BLOCK_SIZE {
 		ctx.t[0] += BLAKE2S_BLOCK_SIZE
 		if ctx.t[0] < BLAKE2S_BLOCK_SIZE {
 			ctx.t[1] += 1
-		} 
+		}
 		v0, v1, v2, v3, v4, v5, v6, v7 := h0, h1, h2, h3, h4, h5, h6, h7
-		v8  := BLAKE2S_IV[0]
-		v9  := BLAKE2S_IV[1]
+		v8 := BLAKE2S_IV[0]
+		v9 := BLAKE2S_IV[1]
 		v10 := BLAKE2S_IV[2]
 		v11 := BLAKE2S_IV[3]
 		v12 := BLAKE2S_IV[4] ~ ctx.t[0]
 		v13 := BLAKE2S_IV[5] ~ ctx.t[1]
 		v14 := BLAKE2S_IV[6] ~ ctx.f[0]
 		v15 := BLAKE2S_IV[7] ~ ctx.f[1]
-		m: [16]u32
-		j := 0
+
+		m: [16]u32 = ---
 		for i := 0; i < 16; i += 1 {
-			m[i] = u32(p[j]) | u32(p[j + 1]) << 8 | u32(p[j + 2]) << 16 | u32(p[j + 3]) << 24
-			j += 4
+			m[i] = endian.unchecked_get_u32le(p[i * 4:])
 		}
+
+		// Round 1
 		v0 += m[0]
 		v0 += v4
 		v12 ~= v0
@@ -391,6 +434,8 @@ blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (32 - 7) | v5 >> 7
+
+		// Round 2
 		v0 += m[14]
 		v0 += v4
 		v12 ~= v0
@@ -503,6 +548,8 @@ blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (32 - 7) | v5 >> 7
+
+		// Round 3
 		v0 += m[11]
 		v0 += v4
 		v12 ~= v0
@@ -615,6 +662,8 @@ blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (32 - 7) | v5 >> 7
+
+		// Round 4
 		v0 += m[7]
 		v0 += v4
 		v12 ~= v0
@@ -727,6 +776,8 @@ blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (32 - 7) | v5 >> 7
+
+		// Round 5
 		v0 += m[9]
 		v0 += v4
 		v12 ~= v0
@@ -839,6 +890,8 @@ blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (32 - 7) | v5 >> 7
+
+		// Round 6
 		v0 += m[2]
 		v0 += v4
 		v12 ~= v0
@@ -951,6 +1004,8 @@ blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (32 - 7) | v5 >> 7
+
+		// Round 7
 		v0 += m[12]
 		v0 += v4
 		v12 ~= v0
@@ -1063,6 +1118,8 @@ blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (32 - 7) | v5 >> 7
+
+		// Round 8
 		v0 += m[13]
 		v0 += v4
 		v12 ~= v0
@@ -1175,6 +1232,8 @@ blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (32 - 7) | v5 >> 7
+
+		// Round 9
 		v0 += m[6]
 		v0 += v4
 		v12 ~= v0
@@ -1287,6 +1346,8 @@ blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (32 - 7) | v5 >> 7
+
+		// Round 10
 		v0 += m[10]
 		v0 += v4
 		v12 ~= v0
@@ -1399,6 +1460,7 @@ blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (32 - 7) | v5 >> 7
+
 		h0 ~= v0 ~ v8
 		h1 ~= v1 ~ v9
 		h2 ~= v2 ~ v10
@@ -1407,19 +1469,23 @@ blake2s_blocks :: #force_inline proc "contextless" (ctx: ^Blake2s_Context, p: []
 		h5 ~= v5 ~ v13
 		h6 ~= v6 ~ v14
 		h7 ~= v7 ~ v15
+
 		p = p[BLAKE2S_BLOCK_SIZE:]
 	}
-	ctx.h[0], ctx.h[1], ctx.h[2], ctx.h[3], ctx.h[4], ctx.h[5], ctx.h[6], ctx.h[7] = h0, h1, h2, h3, h4, h5, h6, h7
+	ctx.h[0], ctx.h[1], ctx.h[2], ctx.h[3], ctx.h[4], ctx.h[5], ctx.h[6], ctx.h[7] =
+		h0, h1, h2, h3, h4, h5, h6, h7
 }

+@(private)
 blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []byte) {
-	h0, h1, h2, h3, h4, h5, h6, h7 := ctx.h[0], ctx.h[1], ctx.h[2], ctx.h[3], ctx.h[4], ctx.h[5], ctx.h[6], ctx.h[7]
+	h0, h1, h2, h3, h4, h5, h6, h7 :=
+		ctx.h[0], ctx.h[1], ctx.h[2], ctx.h[3], ctx.h[4], ctx.h[5], ctx.h[6], ctx.h[7]
 	p := p
 	for len(p) >= BLAKE2B_BLOCK_SIZE {
 		ctx.t[0] += BLAKE2B_BLOCK_SIZE
 		if ctx.t[0] < BLAKE2B_BLOCK_SIZE {
-			ctx.t[1]+=1
-		} 
+			ctx.t[1] += 1
+		}
 		v0, v1, v2, v3, v4, v5, v6, v7 := h0, h1, h2, h3, h4, h5, h6, h7
 		v8 := BLAKE2B_IV[0]
 		v9 := BLAKE2B_IV[1]
@@ -1429,13 +1495,13 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v13 := BLAKE2B_IV[5] ~ ctx.t[1]
 		v14 := BLAKE2B_IV[6] ~ ctx.f[0]
 		v15 := BLAKE2B_IV[7] ~ ctx.f[1]
+
 		m: [16]u64 = ---
-		j := 0
-		for i := 0; i < 16; i+=1 {
-			m[i] = u64(p[j]) 		   | u64(p[j + 1]) << 8  | u64(p[j + 2]) << 16 | u64(p[j + 3]) << 24 |
-				   u64(p[j + 4]) << 32 | u64(p[j + 5]) << 40 | u64(p[j + 6]) << 48 | u64(p[j + 7]) << 56
-			j += 8
+		for i := 0; i < 16; i += 1 {
+			m[i] = endian.unchecked_get_u64le(p[i * 8:])
 		}
+
+		// Round 1
 		v0 += m[0]
 		v0 += v4
 		v12 ~= v0
@@ -1548,6 +1614,8 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
+		// Round 2
 		v0 += m[14]
 		v0 += v4
 		v12 ~= v0
@@ -1660,6 +1728,8 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
+		// Round 3
 		v0 += m[11]
 		v0 += v4
 		v12 ~= v0
@@ -1772,6 +1842,8 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
+		// Round 4
 		v0 += m[7]
 		v0 += v4
 		v12 ~= v0
@@ -1884,6 +1956,8 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
+		// Round 5
 		v0 += m[9]
 		v0 += v4
 		v12 ~= v0
@@ -1996,6 +2070,8 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
+		// Round 6
 		v0 += m[2]
 		v0 += v4
 		v12 ~= v0
@@ -2108,6 +2184,8 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
+		// Round 7
 		v0 += m[12]
 		v0 += v4
 		v12 ~= v0
@@ -2220,6 +2298,8 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
+		// Round 8
 		v0 += m[13]
 		v0 += v4
 		v12 ~= v0
@@ -2332,6 +2412,8 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
+		// Round 9
 		v0 += m[6]
 		v0 += v4
 		v12 ~= v0
@@ -2444,6 +2526,8 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
+		// Round 10
 		v0 += m[10]
 		v0 += v4
 		v12 ~= v0
@@ -2556,6 +2640,8 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
+		// Round 11
 		v0 += m[0]
 		v0 += v4
 		v12 ~= v0
@@ -2668,6 +2754,8 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
+		// Round 12
 		v0 += m[14]
 		v0 += v4
 		v12 ~= v0
@@ -2780,6 +2868,7 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		v10 += v15
 		v5 ~= v10
 		v5 = v5 << (64 - 63) | v5 >> 63
+
 		h0 ~= v0 ~ v8
 		h1 ~= v1 ~ v9
 		h2 ~= v2 ~ v10
@@ -2788,7 +2877,9 @@ blake2b_blocks :: #force_inline proc "contextless" (ctx: ^Blake2b_Context, p: []
 		h5 ~= v5 ~ v13
 		h6 ~= v6 ~ v14
 		h7 ~= v7 ~ v15
+
 		p = p[BLAKE2B_BLOCK_SIZE:]
 	}
-	ctx.h[0], ctx.h[1], ctx.h[2], ctx.h[3], ctx.h[4], ctx.h[5], ctx.h[6], ctx.h[7] = h0, h1, h2, h3, h4, h5, h6, h7
-}
+	ctx.h[0], ctx.h[1], ctx.h[2], ctx.h[3], ctx.h[4], ctx.h[5], ctx.h[6], ctx.h[7] =
+		h0, h1, h2, h3, h4, h5, h6, h7
+}
@@ -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
+}
--- a/Show More
+++ b/Show More