Merge pull request #4449 from 0dminnimda/add-pie

Link as PIE for PIC by default
Fix previous commit for arrays
2026-07-05 03:01:38 -07:00 · 2024-11-04 14:25:31 +01:00 · 2024-11-04 12:51:54 +00:00 · 2024-11-04 15:50:43 +03:00 · 2024-11-04 12:49:43 +00:00 · 2024-11-04 15:44:07 +03:00
1526 changed files with 340707 additions and 112030 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
@@ -1,178 +1,282 @@
 name: CI
-on: [push, pull_request]
+on: [push, pull_request, workflow_dispatch]

 jobs:
-  build_linux:
+  build_netbsd:
+    name: NetBSD Build, Check, and Test
+    runs-on: ubuntu-latest
+    env:
+      PKGSRC_BRANCH: 2024Q2
+    steps:
+    - uses: actions/checkout@v4
+    - name: Build, Check, and Test
+      timeout-minutes: 15
+      uses: vmactions/netbsd-vm@v1
+      with:
+        release: "10.0"
+        envs: PKGSRC_BRANCH
+        usesh: true
+        copyback: false
+        prepare: |
+          PKG_PATH="https://cdn.NetBSD.org/pub/pkgsrc/packages/NetBSD/amd64/$(uname -r | cut -d_ -f1)_${PKGSRC_BRANCH}/All" /usr/sbin/pkg_add pkgin
+          pkgin -y in gmake git bash python311 llvm clang
+          ln -s /usr/pkg/bin/python3.11 /usr/bin/python3
+        run: |
+          set -e -x
+          git config --global --add safe.directory $(pwd)
+          gmake release
+          ./odin version
+          ./odin report
+          gmake -C vendor/stb/src
+          gmake -C vendor/cgltf/src
+          gmake -C vendor/miniaudio/src
+          ./odin check examples/all -vet -strict-style -disallow-do -target:netbsd_amd64
+          ./odin check examples/all -vet -strict-style -disallow-do -target:netbsd_arm64
+          ./odin test tests/core/normal.odin -file -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+          ./odin test tests/core/speed.odin -file -all-packages -vet -strict-style -disallow-do -o:speed -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+          ./odin test tests/vendor -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+          (cd tests/issues; ./run.sh)
+          ./odin check tests/benchmark -vet -strict-style -no-entry-point
+
+  build_freebsd:
+    name: FreeBSD 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: build odin
+    - uses: actions/checkout@v4
+    - name: Build, Check, and Test
+      timeout-minutes: 15
+      uses: vmactions/freebsd-vm@v1
+      with:
+        usesh: true
+        copyback: false
+        prepare: |
+          pkg install -y gmake git bash python3 libxml2 llvm17
+        run: |
+          # `set -e` is needed for test failures to register. https://github.com/vmactions/freebsd-vm/issues/72
+          set -e -x
+          git config --global --add safe.directory $(pwd)
+          gmake release
+          ./odin version
+          ./odin report
+          gmake -C vendor/stb/src
+          gmake -C vendor/cgltf/src
+          gmake -C vendor/miniaudio/src
+          ./odin check examples/all -vet -strict-style -disallow-do -target:freebsd_amd64
+          ./odin test tests/core/normal.odin -file -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+          ./odin test tests/core/speed.odin -file -all-packages -vet -strict-style -disallow-do -o:speed -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+          ./odin test tests/vendor -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+          (cd tests/issues; ./run.sh)
+          ./odin check tests/benchmark -vet -strict-style -no-entry-point
+  ci:
+    strategy:
+      fail-fast: false
+      matrix:
+        # MacOS 13 runs on Intel, 14 runs on ARM
+        os: [ubuntu-latest, macos-13, macos-14]
+    runs-on: ${{ matrix.os }}
+    name: ${{ matrix.os == 'macos-14' && 'MacOS ARM' || (matrix.os == 'macos-13' && 'MacOS Intel' || 'Ubuntu') }} Build, Check, and Test
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Download LLVM (Linux)
+        if: matrix.os == 'ubuntu-latest'
+        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: Download LLVM (MacOS Intel)
+        if: matrix.os == 'macos-13'
+        run: |
+          brew install llvm@17 lua@5.4
+          echo "/usr/local/opt/llvm@17/bin" >> $GITHUB_PATH
+
+      - name: Download LLVM (MacOS ARM)
+        if: matrix.os == 'macos-14'
+        run: |
+          brew install llvm@17 wasmtime lua@5.4
+          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: Compile needed Vendor
+        run: |
+          make -C vendor/stb/src
+          make -C vendor/cgltf/src
+          make -C vendor/miniaudio/src
      - 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: Normal Core library tests
+        run: ./odin test tests/core/normal.odin -file -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+      - name: Optimized Core library tests
+        run: ./odin test tests/core/speed.odin -o:speed -file -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
      - name: Vendor library tests
-        run: |
-          cd tests/vendor
-          make
-        timeout-minutes: 10
-      - name: Odin issues tests
+        run: ./odin test tests/vendor -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+      - name: Internals tests
+        run: ./odin test tests/internal -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+      - name: GitHub Issue tests
        run: |
          cd tests/issues
          ./run.sh
-        timeout-minutes: 10
+
+      - name: Check benchmarks
+        run: ./odin check tests/benchmark -vet -strict-style -no-entry-point
      - 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 FreeBSD amd64
-        run: ./odin check examples/all -vet -strict-style -target:freebsd_amd64
-        timeout-minutes: 10
-      - name: Odin check examples/all for OpenBSD amd64
-        run: ./odin check examples/all -vet -strict-style -target:openbsd_amd64
-        timeout-minutes: 10
-  build_macOS:
-    runs-on: macos-latest
-    steps:
-      - uses: actions/checkout@v1
-      - name: Download LLVM, botan 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
-      - 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: Vendor library tests
-        run: |
-          cd tests/vendor
-          make
-        timeout-minutes: 10
-      - name: Odin issues tests
-        run: |
-          cd tests/issues
-          ./run.sh
-        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
+        run: ./odin check examples/all -vet -strict-style -disallow-do -target:linux_i386
+        if: matrix.os == 'ubuntu-latest'
      - name: Odin check examples/all for Linux arm64
-        run: ./odin check examples/all -vet -strict-style -target:linux_arm64
-        timeout-minutes: 10
+        run: ./odin check examples/all -vet -strict-style -disallow-do -target:linux_arm64
+        if: matrix.os == 'ubuntu-latest'
+      - name: Odin check examples/all for FreeBSD amd64
+        run: ./odin check examples/all -vet -strict-style -disallow-do -target:freebsd_amd64
+        if: matrix.os == 'ubuntu-latest'
+      - name: Odin check examples/all for OpenBSD amd64
+        run: ./odin check examples/all -vet -strict-style -disallow-do -target:openbsd_amd64
+        if: matrix.os == 'ubuntu-latest'
+
+      - name: Run demo on WASI WASM32
+        run: |
+          ./odin build examples/demo -target:wasi_wasm32 -vet -strict-style -disallow-do -out:demo.wasm
+          wasmtime ./demo.wasm
+        if: matrix.os == 'macos-14'
+
  build_windows:
-    runs-on: windows-2019
+    name: Windows Build, Check, and Test
+    runs-on: windows-2022
+    timeout-minutes: 15
    steps:
-      - uses: actions/checkout@v1
+      - uses: actions/checkout@v4
      - 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
-        timeout-minutes: 1
      - name: Odin report
        run: ./odin report
-        timeout-minutes: 1
      - 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
-          odin run examples/demo -debug
-        timeout-minutes: 10
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          odin run examples/demo -debug -vet -strict-style -disallow-do
      - name: Odin check examples/all
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
-          odin check examples/all -strict-style
-        timeout-minutes: 10
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          odin check examples/all -vet -strict-style -disallow-do
      - name: Core library tests
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
-          cd tests\core
-          call build.bat
-        timeout-minutes: 10
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          odin test tests/core/normal.odin -file -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+      - name: Optimized core library tests
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          odin test tests/core/speed.odin -o:speed -file -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
      - name: Vendor library tests
        shell: cmd
        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
-          cd tests\vendor
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          copy vendor\lua\5.4\windows\*.dll .
+          odin test tests/vendor -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+      - name: Odin internals tests
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          odin test tests/internal -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true
+      - name: Check benchmarks
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          odin check tests/benchmark -vet -strict-style -no-entry-point
+      - name: Odin documentation tests
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          cd tests\documentation
          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 issues tests
-        shell: cmd
-        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
-          cd tests\issues
-          call run.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
+
+  build_linux_riscv64:
+    runs-on: ubuntu-latest
+    name: Linux riscv64 (emulated) Build, Check and Test
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Download LLVM (Linux)
+        run: |
+          wget https://apt.llvm.org/llvm.sh
+          chmod +x llvm.sh
+          sudo ./llvm.sh 18
+          echo "/usr/lib/llvm-18/bin" >> $GITHUB_PATH
+
+      - name: Build Odin
+        run: ./build_odin.sh release
+
+      - name: Odin version
+        run: ./odin version
+
+      - name: Odin report
+        run: ./odin report
+
+      - name: Compile needed Vendor
+        run: |
+          make -C vendor/stb/src
+          make -C vendor/cgltf/src
+          make -C vendor/miniaudio/src
+
+      - name: Odin check
+        run: ./odin check examples/all -target:linux_riscv64 -vet -strict-style -disallow-do
+
+      - name: Install riscv64 toolchain and qemu
+        run: sudo apt-get install -y qemu-user qemu-user-static gcc-12-riscv64-linux-gnu libc6-riscv64-cross
+
+      - name: Odin run
+        run: ./odin run examples/demo -vet -strict-style -disallow-do -target:linux_riscv64 -extra-linker-flags:"-fuse-ld=/usr/bin/riscv64-linux-gnu-gcc-12 -static -Wl,-static" -no-rpath
+
+      - name: Odin run -debug
+        run: ./odin run examples/demo -debug -vet -strict-style -disallow-do -target:linux_riscv64 -extra-linker-flags:"-fuse-ld=/usr/bin/riscv64-linux-gnu-gcc-12 -static -Wl,-static" -no-rpath
+
+      - name: Normal Core library tests
+        run: ./odin test tests/core/normal.odin -file -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true -target:linux_riscv64 -extra-linker-flags:"-fuse-ld=/usr/bin/riscv64-linux-gnu-gcc-12 -static -Wl,-static" -no-rpath
+
+      - name: Optimized Core library tests
+        run: ./odin test tests/core/speed.odin -o:speed -file -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true -target:linux_riscv64 -extra-linker-flags:"-fuse-ld=/usr/bin/riscv64-linux-gnu-gcc-12 -static -Wl,-static" -no-rpath
+
+      - name: Internals tests
+        run: ./odin test tests/internal -all-packages -vet -strict-style -disallow-do -define:ODIN_TEST_FANCY=false -define:ODIN_TEST_FAIL_ON_BAD_MEMORY=true -target:linux_riscv64 -extra-linker-flags:"-fuse-ld=/usr/bin/riscv64-linux-gnu-gcc-12 -static -Wl,-static" -no-rpath
@@ -7,117 +7,204 @@ 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
+      - uses: actions/checkout@v4
      - 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
          cp -r examples dist
      - name: Upload artifact
-        uses: actions/upload-artifact@v1
+        uses: actions/upload-artifact@v4
        with:
+          include-hidden-files: true
          name: windows_artifacts
          path: dist
-  build_ubuntu:
+  build_linux:
+    name: Linux Build
+    if: github.repository == 'odin-lang/Odin'
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/checkout@v1
+      - uses: actions/checkout@v4
+      - uses: jirutka/setup-alpine@v1
+        with:
+          branch: v3.20
      - name: (Linux) Download LLVM
-        run: sudo apt-get install llvm-11 clang-11
+        run: |
+          apk add --no-cache \
+          musl-dev llvm18-dev clang18 git mold lz4 \
+          libxml2-static llvm18-static zlib-static zstd-static \
+          make
+        shell: alpine.sh --root {0}
      - name: build odin
-        run: make nightly
+        # NOTE: this build does slow compile times because of musl
+        run: ci/build_linux_static.sh
+        shell: alpine.sh {0}
      - name: Odin run
        run: ./odin run examples/demo
      - name: Copy artifacts
        run: |
-          mkdir dist
-          cp odin dist
-          cp -r shared dist
-          cp -r core dist
-          cp -r vendor dist
-          cp -r examples dist
+          FILE="odin-linux-amd64-nightly+$(date -I)"
+          mkdir $FILE
+          cp odin $FILE
+          cp LICENSE $FILE
+          cp -r shared $FILE
+          cp -r base $FILE
+          cp -r core $FILE
+          cp -r vendor $FILE
+          cp -r examples $FILE
+          # Creating a tarball so executable permissions are retained, see https://github.com/actions/upload-artifact/issues/38
+          tar -czvf dist.tar.gz $FILE
+      - name: Odin run
+        run: |
+          FILE="odin-linux-amd64-nightly+$(date -I)"
+          $FILE/odin run examples/demo
      - name: Upload artifact
-        uses: actions/upload-artifact@v1
+        uses: actions/upload-artifact@v4
        with:
-          name: ubuntu_artifacts
-          path: dist
+          name: linux_artifacts
+          path: dist.tar.gz
  build_macos:
-    runs-on: macOS-latest
+    name: MacOS Build
+    if: github.repository == 'odin-lang/Odin'
+    runs-on: macos-13
    steps:
-      - uses: actions/checkout@v1
+      - uses: actions/checkout@v4
      - 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@18 dylibbundler
+          echo "/usr/local/opt/llvm@18/bin" >> $GITHUB_PATH
      - name: build odin
-        run: make nightly
-      - name: Odin run
-        run: ./odin run examples/demo
-      - name: Copy artifacts
+        # These -L makes the linker prioritize system libraries over LLVM libraries, this is mainly to
+        # not link with libunwind bundled with LLVM but link with libunwind on the system.
+        run: CXXFLAGS="-L/usr/lib/system -L/usr/lib" make nightly
+      - name: Bundle
        run: |
-          mkdir dist
-          cp odin dist
-          cp -r shared dist
-          cp -r core dist
-          cp -r vendor dist
-          cp -r examples dist
+          FILE="odin-macos-amd64-nightly+$(date -I)"
+          mkdir $FILE
+          cp odin $FILE
+          cp LICENSE $FILE
+          cp -r shared $FILE
+          cp -r base $FILE
+          cp -r core $FILE
+          cp -r vendor $FILE
+          cp -r examples $FILE
+          dylibbundler -b -x $FILE/odin -d $FILE/libs -od -p @executable_path/libs
+          # Creating a tarball so executable permissions are retained, see https://github.com/actions/upload-artifact/issues/38
+          tar -czvf dist.tar.gz $FILE
+      - name: Odin run
+        run: |
+          FILE="odin-macos-amd64-nightly+$(date -I)"
+          $FILE/odin run examples/demo
      - name: Upload artifact
-        uses: actions/upload-artifact@v1
+        uses: actions/upload-artifact@v4
        with:
          name: macos_artifacts
-          path: dist
+          path: dist.tar.gz
+  build_macos_arm:
+    name: MacOS ARM Build
+    if: github.repository == 'odin-lang/Odin'
+    runs-on: macos-14 # ARM machine
+    steps:
+      - uses: actions/checkout@v4
+      - name: Download LLVM and setup PATH
+        run: |
+          brew install llvm@18 dylibbundler
+          echo "/opt/homebrew/opt/llvm@18/bin" >> $GITHUB_PATH
+      - name: build odin
+        # These -L makes the linker prioritize system libraries over LLVM libraries, this is mainly to
+        # not link with libunwind bundled with LLVM but link with libunwind on the system.
+        run: CXXFLAGS="-L/usr/lib/system -L/usr/lib" make nightly
+      - name: Bundle
+        run: |
+          FILE="odin-macos-arm64-nightly+$(date -I)"
+          mkdir $FILE
+          cp odin $FILE
+          cp LICENSE $FILE
+          cp -r shared $FILE
+          cp -r base $FILE
+          cp -r core $FILE
+          cp -r vendor $FILE
+          cp -r examples $FILE
+          dylibbundler -b -x $FILE/odin -d $FILE/libs -od -p @executable_path/libs
+          # Creating a tarball so executable permissions are retained, see https://github.com/actions/upload-artifact/issues/38
+          tar -czvf dist.tar.gz $FILE
+      - name: Odin run
+        run: |
+          FILE="odin-macos-arm64-nightly+$(date -I)"
+          $FILE/odin run examples/demo
+      - name: Upload artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: macos_arm_artifacts
+          path: dist.tar.gz
  upload_b2:
    runs-on: [ubuntu-latest]
-    needs: [build_windows, build_macos, build_ubuntu]
+    needs: [build_windows, build_macos, build_macos_arm, build_linux]
    steps:
-      - uses: actions/checkout@v1
-      - uses: actions/setup-python@v2
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
        with:
          python-version: '3.8.x'

-      - name: Install B2 CLI
+      - name: Install B2 SDK
        shell: bash
        run: |
          python -m pip install --upgrade pip
-          pip install --upgrade b2
+          pip install --upgrade b2sdk

      - name: Display Python version
        run: python -c "import sys; print(sys.version)"

      - name: Download Windows artifacts
-        uses: actions/download-artifact@v1
+
+        uses: actions/download-artifact@v4.1.7
        with:
          name: windows_artifacts
+          path: windows_artifacts

      - name: Download Ubuntu artifacts
-        uses: actions/download-artifact@v1
+        uses: actions/download-artifact@v4.1.7
        with:
-          name: ubuntu_artifacts
+          name: linux_artifacts
+          path: linux_artifacts

      - name: Download macOS artifacts
-        uses: actions/download-artifact@v1
+        uses: actions/download-artifact@v4.1.7
        with:
          name: macos_artifacts
+          path: macos_artifacts
+
+      - name: Download macOS arm artifacts
+        uses: actions/download-artifact@v4.1.7
+        with:
+          name: macos_arm_artifacts
+          path: macos_arm_artifacts
+
+      - name: Debug
+        run: |
+          tree -L 2

      - name: Create archives and upload
        shell: bash
@@ -127,23 +214,10 @@ jobs:
          BUCKET: ${{ secrets.B2_BUCKET }}
          DAYS_TO_KEEP: ${{ secrets.B2_DAYS_TO_KEEP }}
        run: |
-          echo Authorizing B2 account
-          b2 authorize-account "$APPID" "$APPKEY"
-
-          echo Uploading artifcates to B2
-          chmod +x ./ci/upload_create_nightly.sh
-          ./ci/upload_create_nightly.sh "$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/
-
-          echo Deleting old artifacts in B2
-          python3 ci/delete_old_binaries.py "$BUCKET" "$DAYS_TO_KEEP"
-
-          echo Creating nightly.json
-          python3 ci/create_nightly_json.py "$BUCKET" > nightly.json
-
-          echo Uploading nightly.json
-          b2 upload-file "$BUCKET" nightly.json nightly.json
-
-          echo Clear B2 account info
-          b2 clear-account
+          file linux_artifacts/dist.tar.gz
+          python3 ci/nightly.py artifact windows-amd64 windows_artifacts/
+          python3 ci/nightly.py artifact linux-amd64 linux_artifacts/dist.tar.gz
+          python3 ci/nightly.py artifact macos-amd64 macos_artifacts/dist.tar.gz
+          python3 ci/nightly.py artifact macos-arm64 macos_arm_artifacts/dist.tar.gz
+          python3 ci/nightly.py prune
+          python3 ci/nightly.py json
@@ -0,0 +1,43 @@
+name: "Close Stale Issues & PRs"
+on:
+  workflow_dispatch:
+  schedule:
+    - cron: "0 21 * * *"
+    
+permissions:
+  issues: write
+  pull-requests: write
+
+jobs:
+  stale:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Close Stale Issues
+        uses: actions/stale@v7.0.0
+        with:
+#          stale-issue-message: |
+#            Hello!
+#            
+#            I am marking this issue as stale as it has not received any engagement from the community or maintainers 120 days. That does not imply that the issue has no merit! If you feel strongly about this issue
+#            - open a PR referencing and resolving the issue;
+#            - leave a comment on it and discuss ideas how you could contribute towards resolving it;
+#            - leave a comment and describe in detail why this issue is critical for your use case;
+#            - open a new issue with updated details and a plan on resolving the issue.
+#
+#            The motivation for this automation is to help prioritize issues in the backlog and not ignore, reject, or belittle anyone..
+#
+#          stale-pr-message: |
+#            Hello!
+#            
+#            I am marking this PR as stale as it has not received any engagement from the community or maintainers 120 days. That does not imply that the issue has no merit! If you feel strongly about this issue
+#            - leave a comment on it and discuss ideas how you could contribute towards resolving it;
+#            - leave a comment and describe in detail why this issue is critical for your use case;
+#
+#           The motivation for this automation is to help prioritize issues in the backlog and not ignore, reject, or belittle anyone..
+
+          days-before-stale: 120
+          days-before-close: -1
+          exempt-draft-pr: true
+          ascending: true
+          operations-per-run: 1000
+          exempt-issue-labels: "ignore"
@@ -17,6 +17,7 @@
 [Rr]eleases/
 x64/
 x86/
+!/core/simd/x86
 bld/
 [Bb]in/
 [Oo]bj/
@@ -265,12 +266,17 @@ bin/
 *.exe
 *.obj
 *.pdb
+*.res
+desktop.ini
+Thumbs.db

 # - Linux/MacOS
 odin
-odin.dSYM
+!odin/
+**/*.dSYM
 *.bin
 demo.bin
+libLLVM*.so*

 # shared collection
 shared/
@@ -283,3 +289,9 @@ shared/
 *.sublime-workspace
 examples/bug/
 build.sh
+!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:
@@ -1,4 +1,4 @@
-all: debug
+all: default

 demo:
 	./odin run examples/demo/demo.odin -file
@@ -6,12 +6,18 @@ demo:
 report:
 	./odin report

+default:
+	PROGRAM=make ./build_odin.sh # debug
+
 debug:
 	./build_odin.sh debug

 release:
 	./build_odin.sh release

+release-native:
+	./build_odin.sh release-native
+
 release_native:
 	./build_odin.sh release-native

@@ -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">
@@ -76,13 +76,13 @@ Answers to common questions about Odin.

 Documentation for all the official packages part of the [core](https://pkg.odin-lang.org/core/) and [vendor](https://pkg.odin-lang.org/vendor/) library collections.

-#### [The Odin Wiki](https://github.com/odin-lang/Odin/wiki)
+#### [Odin Documentation](https://odin-lang.org/docs/)

-A wiki maintained by the Odin community.
+Documentation for the Odin language itself.

 #### [Odin Discord](https://discord.gg/sVBPHEv)

-Get live support and talk with other odiners on the Odin Discord.
+Get live support and talk with other Odin programmers on the Odin Discord.

 ### Articles

@@ -94,7 +94,15 @@ cap :: proc(array: Array_Type) -> int ---

 size_of      :: proc($T: typeid) -> int ---
 align_of     :: proc($T: typeid) -> int ---
-offset_of    :: proc($T: typeid) -> uintptr ---
+
+// e.g. offset_of(t.f), where t is an instance of the type T
+offset_of_selector :: proc(selector: $T) -> uintptr ---
+// e.g. offset_of(T, f), where T can be the type instead of a variable
+offset_of_member   :: proc($T: typeid, member: $M) -> uintptr ---
+offset_of :: proc{offset_of_selector, offset_of_member}
+// e.g. offset_of(T, "f"), where T can be the type instead of a variable
+offset_of_by_string :: proc($T: typeid, member: string) -> uintptr ---
+
 type_of      :: proc(x: expr) -> type ---
 type_info_of :: proc($T: typeid) -> ^runtime.Type_Info ---
 typeid_of    :: proc($T: typeid) -> typeid ---
@@ -102,14 +110,14 @@ 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 ---
 kmag       :: proc(value: Quaternion) -> Float ---
 conj       :: proc(value: Complex_Or_Quaternion) -> Complex_Or_Quaternion ---

-expand_to_tuple :: proc(value: Struct_Or_Array) -> (A, B, C, ...) ---
+expand_values :: proc(value: Struct_Or_Array) -> (A, B, C, ...) ---

 min   :: proc(values: ..T) -> T ---
 max   :: proc(values: ..T) -> T ---
@@ -118,3 +126,5 @@ clamp :: proc(value, minimum, maximum: T) -> T ---

 soa_zip :: proc(slices: ...) -> #soa[]Struct ---
 soa_unzip :: proc(value: $S/#soa[]$E) -> (slices: ...) ---
+
+unreachable :: proc() -> ! ---
@@ -1,10 +1,16 @@
 // This is purely for documentation
-//+ignore
+#+build ignore
 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 +38,12 @@ 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) where type_is_integer(T) #optional_ok ---
+overflow_sub :: proc(lhs, rhs: $T) -> (T, bool) where type_is_integer(T) #optional_ok ---
+overflow_mul :: proc(lhs, rhs: $T) -> (T, bool) where type_is_integer(T) #optional_ok ---
+
+saturating_add :: proc(lhs, rhs: $T) -> T where type_is_integer(T) ---
+saturating_sub :: proc(lhs, rhs: $T) -> T where type_is_integer(T) ---

 sqrt :: proc(x: $T) -> T where type_is_float(T) || (type_is_simd_vector(T) && type_is_float(type_elem_type(T))) ---

@@ -67,6 +76,8 @@ expect :: proc(val, expected_val: T) -> T ---

 // Linux and Darwin Only
 syscall :: proc(id: uintptr, args: ..uintptr) -> uintptr ---
+// FreeBSD, NetBSD, et cetera
+syscall_bsd :: proc(id: uintptr, args: ..uintptr) -> (uintptr, bool) ---


 // Atomics
@@ -161,10 +172,23 @@ 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, $V: typeid) -> bool where type_is_union(U) ---
+type_is_matrix_row_major    :: proc($T: typeid) -> bool where type_is_matrix(T) ---
+type_is_matrix_column_major :: proc($T: typeid) -> bool where type_is_matrix(T) ---

-type_has_field :: proc($T: typeid, $name: string) -> bool ---
+type_is_specialization_of :: proc($T, $S: typeid) -> bool ---
+
+type_is_variant_of        :: proc($U, $V: typeid)          -> bool    where type_is_union(U) ---
+type_union_tag_type       :: proc($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($T: typeid) -> int where type_is_proc(T) ---
@@ -173,7 +197,8 @@ type_proc_return_count    :: proc($T: typeid) -> int where type_is_proc(T) ---
 type_proc_parameter_type  :: proc($T: typeid, index: int) -> typeid where type_is_proc(T) ---
 type_proc_return_type     :: proc($T: typeid, index: int) -> typeid where type_is_proc(T) ---

-type_struct_field_count :: proc($T: typeid) -> int where type_is_struct(T) ---
+type_struct_field_count          :: proc($T: typeid) -> int  where type_is_struct(T) ---
+type_struct_has_implicit_padding :: proc($T: typeid) -> bool where type_is_struct(T) ---

 type_polymorphic_record_parameter_count :: proc($T: typeid) -> typeid ---
 type_polymorphic_record_parameter_value :: proc($T: typeid, index: int) -> $V ---
@@ -188,14 +213,27 @@ type_field_index_of :: proc($T: typeid, $name: string) -> uintptr ---
 type_equal_proc  :: proc($T: typeid) -> (equal:  proc "contextless" (rawptr, rawptr) -> bool)                 where type_is_comparable(T) ---
 type_hasher_proc :: proc($T: typeid) -> (hasher: proc "contextless" (data: rawptr, seed: uintptr) -> uintptr) where type_is_comparable(T) ---

+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, $V: typeid) -> typeid where type_is_union(U), type_is_union(V) ---
+
+type_has_shared_fields :: proc($U, $V: typeid) -> bool typeid where type_is_struct(U), type_is_struct(V) ---
+
 constant_utf16_cstring :: proc($literal: string) -> [^]u16 ---

+constant_log2 :: proc($v: $T) -> T where type_is_integer(T) ---
+
 // SIMD related
 simd_add  :: proc(a, b: #simd[N]T) -> #simd[N]T ---
 simd_sub  :: proc(a, b: #simd[N]T) -> #simd[N]T ---
 simd_mul  :: proc(a, b: #simd[N]T) -> #simd[N]T ---
 simd_div  :: proc(a, b: #simd[N]T) -> #simd[N]T where type_is_float(T) ---

+simd_saturating_add  :: proc(a, b: #simd[N]T) -> #simd[N]T where type_is_integer(T) ---
+simd_saturating_sub  :: proc(a, b: #simd[N]T) -> #simd[N]T where type_is_integer(T) ---
+
 // Keeps Odin's Behaviour
 // (x << y) if y <= mask else 0
 simd_shl :: proc(a: #simd[N]T, b: #simd[N]Unsigned_Integer) -> #simd[N]T ---
@@ -206,13 +244,10 @@ simd_shr :: proc(a: #simd[N]T, b: #simd[N]Unsigned_Integer) -> #simd[N]T ---
 simd_shl_masked :: proc(a: #simd[N]T, b: #simd[N]Unsigned_Integer) -> #simd[N]T ---
 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 ---

@@ -237,13 +272,28 @@ simd_lanes_ge :: proc(a, b: #simd[N]T) -> #simd[N]Integer ---
 simd_extract :: proc(a: #simd[N]T, idx: uint) -> T ---
 simd_replace :: proc(a: #simd[N]T, idx: uint, elem: T) -> #simd[N]T ---

-simd_reduce_add_ordered :: proc(a: #simd[N]T) -> T ---
-simd_reduce_mul_ordered :: proc(a: #simd[N]T) -> T ---
-simd_reduce_min         :: proc(a: #simd[N]T) -> T ---
-simd_reduce_max         :: proc(a: #simd[N]T) -> T ---
-simd_reduce_and         :: proc(a: #simd[N]T) -> T ---
-simd_reduce_or          :: proc(a: #simd[N]T) -> T ---
-simd_reduce_xor         :: proc(a: #simd[N]T) -> T ---
+simd_reduce_add_ordered :: proc(a: #simd[N]T) -> T where type_is_integer(T) || type_is_float(T)---
+simd_reduce_mul_ordered :: proc(a: #simd[N]T) -> T where type_is_integer(T) || type_is_float(T)---
+simd_reduce_min         :: proc(a: #simd[N]T) -> T where type_is_integer(T) || type_is_float(T)---
+simd_reduce_max         :: proc(a: #simd[N]T) -> T where type_is_integer(T) || type_is_float(T)---
+simd_reduce_and         :: proc(a: #simd[N]T) -> T where type_is_integer(T) || type_is_float(T)---
+simd_reduce_or          :: proc(a: #simd[N]T) -> T where type_is_integer(T) || type_is_float(T)---
+simd_reduce_xor         :: proc(a: #simd[N]T) -> T where type_is_integer(T) || type_is_float(T)---
+
+simd_reduce_any         :: proc(a: #simd[N]T) -> T where type_is_boolean(T) ---
+simd_reduce_all         :: proc(a: #simd[N]T) -> T where type_is_boolean(T) ---
+
+
+simd_gather       :: proc(ptr: #simd[N]rawptr, val: #simd[N]T, mask: #simd[N]U) -> #simd[N]T where type_is_integer(U) || type_is_boolean(U) ---
+simd_scatter      :: proc(ptr: #simd[N]rawptr, val: #simd[N]T, mask: #simd[N]U)              where type_is_integer(U) || type_is_boolean(U) ---
+
+simd_masked_load  :: proc(ptr: rawptr, val: #simd[N]T, mask: #simd[N]U) -> #simd[N]T where type_is_integer(U) || type_is_boolean(U) ---
+simd_masked_store :: proc(ptr: rawptr, val: #simd[N]T, mask: #simd[N]U)              where type_is_integer(U) || type_is_boolean(U) ---
+
+simd_masked_expand_load    :: proc(ptr: rawptr, val: #simd[N]T, mask: #simd[N]U) -> #simd[N]T where type_is_integer(U) || type_is_boolean(U) ---
+simd_masked_compress_store :: proc(ptr: rawptr, val: #simd[N]T, mask: #simd[N]U)              where type_is_integer(U) || type_is_boolean(U) ---
+
+

 simd_shuffle :: proc(a, b: #simd[N]T, indices: ..int) -> #simd[len(indices)]T ---
 simd_select  :: proc(cond: #simd[N]boolean_or_integer, true, false: #simd[N]T) -> #simd[N]T ---
@@ -257,12 +307,22 @@ simd_nearest :: proc(a: #simd[N]any_float) -> #simd[N]any_float ---

 simd_to_bits :: proc(v: #simd[N]T) -> #simd[N]Integer where size_of(T) == size_of(Integer), type_is_unsigned(Integer) ---

-// equivalent a swizzle with descending indices, e.g. reserve(a, 3, 2, 1, 0)
-simd_reverse :: proc(a: #simd[N]T) -> #simd[N]T ---
+// equivalent to a swizzle with descending indices, e.g. reserve(a, 3, 2, 1, 0)
+simd_lanes_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 ---
+simd_lanes_rotate_left  :: proc(a: #simd[N]T, $offset: int) -> #simd[N]T ---
+simd_lanes_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) ---
+
+
+// Returns the value of the procedure where `x` must be a call expression
+procedure_of :: proc(x: $T) -> T where type_is_proc(T) ---

 // WASM targets only
 wasm_memory_grow :: proc(index, delta: uintptr) -> int ---
@@ -274,11 +334,13 @@ 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)
-x86_cpuid  :: proc(ax, cx: u32) -> (eax, ebc, ecx, edx: u32) ---
+x86_cpuid  :: proc(ax, cx: u32) -> (eax, ebx, ecx, edx: u32) ---
 x86_xgetbv :: proc(cx: u32) -> (eax, edx: u32) ---


@@ -295,6 +357,9 @@ objc_register_selector :: proc($name: string) -> objc_SEL   ---
 objc_find_class        :: proc($name: string) -> objc_Class ---
 objc_register_class    :: proc($name: string) -> objc_Class ---

+
+valgrind_client_request :: proc(default: uintptr, request: uintptr, a0, a1, a2, a3, a4: uintptr) -> uintptr ---
+
 // Internal compiler use only

-__entry_point :: proc() ---
+__entry_point :: proc() ---
@@ -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 {
@@ -65,7 +66,7 @@ Type_Info_Named :: struct {
 	name: string,
 	base: ^Type_Info,
 	pkg:  string,
-	loc:  Source_Code_Location,
+	loc:  ^Source_Code_Location,
 }
 Type_Info_Integer    :: struct {signed: bool, endianness: Platform_Endianness}
 Type_Info_Rune       :: struct {}
@@ -83,8 +84,8 @@ Type_Info_Multi_Pointer :: struct {
 	elem: ^Type_Info,
 }
 Type_Info_Procedure :: struct {
-	params:     ^Type_Info, // Type_Info_Tuple
-	results:    ^Type_Info, // Type_Info_Tuple
+	params:     ^Type_Info, // Type_Info_Parameters
+	results:    ^Type_Info, // Type_Info_Parameters
 	variadic:   bool,
 	convention: Calling_Convention,
 }
@@ -104,27 +105,39 @@ Type_Info_Enumerated_Array :: struct {
 }
 Type_Info_Dynamic_Array :: struct {elem: ^Type_Info, elem_size: int}
 Type_Info_Slice         :: struct {elem: ^Type_Info, elem_size: int}
-Type_Info_Tuple :: struct { // Only used for procedures parameters and results
+
+Type_Info_Parameters :: struct { // Only used for procedures parameters and results
 	types:        []^Type_Info,
 	names:        []string,
 }
+Type_Info_Tuple :: Type_Info_Parameters // Will be removed eventually
+
+Type_Info_Struct_Flags :: distinct bit_set[Type_Info_Struct_Flag; u8]
+Type_Info_Struct_Flag :: enum u8 {
+	packed    = 0,
+	raw_union = 1,
+	no_copy   = 2,
+	align     = 3,
+}

 Type_Info_Struct :: struct {
-	types:        []^Type_Info,
-	names:        []string,
-	offsets:      []uintptr,
-	usings:       []bool,
-	tags:         []string,
-	is_packed:    bool,
-	is_raw_union: bool,
-	custom_align: bool,
+	// Slice these with `field_count`
+	types:   [^]^Type_Info `fmt:"v,field_count"`,
+	names:   [^]string     `fmt:"v,field_count"`,
+	offsets: [^]uintptr    `fmt:"v,field_count"`,
+	usings:  [^]bool       `fmt:"v,field_count"`,
+	tags:    [^]string     `fmt:"v,field_count"`,

-	equal: Equal_Proc, // set only when the struct has .Comparable set but does not have .Simple_Compare set
+	field_count: i32,
+
+	flags: Type_Info_Struct_Flags,

 	// These are only set iff this structure is an SOA structure
 	soa_kind:      Type_Info_Struct_Soa_Kind,
+	soa_len:       i32,
 	soa_base_type: ^Type_Info,
-	soa_len:       int,
+
+	equal: Equal_Proc, // set only when the struct has .Comparable set but does not have .Simple_Compare set
 }
 Type_Info_Union :: struct {
 	variants:     []^Type_Info,
@@ -138,16 +151,14 @@ Type_Info_Union :: struct {
 	shared_nil:   bool,
 }
 Type_Info_Enum :: struct {
-	base:      ^Type_Info,
-	names:     []string,
-	values:    []Type_Info_Enum_Value,
+	base:   ^Type_Info,
+	names:  []string,
+	values: []Type_Info_Enum_Value,
 }
 Type_Info_Map :: struct {
-	key:              ^Type_Info,
-	value:            ^Type_Info,
-	generated_struct: ^Type_Info,
-	key_equal:        Equal_Proc,
-	key_hasher:       Hasher_Proc,
+	key:      ^Type_Info,
+	value:    ^Type_Info,
+	map_info: ^Map_Info,
 }
 Type_Info_Bit_Set :: struct {
 	elem:       ^Type_Info,
@@ -161,11 +172,11 @@ 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 {
@@ -175,6 +186,22 @@ 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     `fmt:"v,field_count"`,
+	types:        [^]^Type_Info `fmt:"v,field_count"`,
+	bit_sizes:    [^]uintptr    `fmt:"v,field_count"`,
+	bit_offsets:  [^]uintptr    `fmt:"v,field_count"`,
+	tags:         [^]string     `fmt:"v,field_count"`,
+	field_count:  int,
 }

 Type_Info_Flag :: enum u8 {
@@ -207,7 +234,7 @@ Type_Info :: struct {
 		Type_Info_Enumerated_Array,
 		Type_Info_Dynamic_Array,
 		Type_Info_Slice,
-		Type_Info_Tuple,
+		Type_Info_Parameters,
 		Type_Info_Struct,
 		Type_Info_Union,
 		Type_Info_Enum,
@@ -215,8 +242,10 @@ 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,
 	},
 }

@@ -247,23 +276,25 @@ 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

@@ -278,6 +309,8 @@ when ODIN_OS == .Windows {
 		Thread_Detach  = 3,
 	}
 	dll_forward_reason: DLL_Forward_Reason
+
+	dll_instance: rawptr
 }

 // IMPORTANT NOTE(bill): Must be in this order (as the compiler relies upon it)
@@ -289,6 +322,14 @@ 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
@@ -299,6 +340,8 @@ Allocator_Mode :: enum byte {
 	Resize,
 	Query_Features,
 	Query_Info,
+	Alloc_Non_Zeroed,
+	Resize_Non_Zeroed,
 }

 Allocator_Mode_Set :: distinct bit_set[Allocator_Mode]
@@ -326,6 +369,14 @@ Allocator :: struct {
 	data:      rawptr,
 }

+Byte     :: 1
+Kilobyte :: 1024 * Byte
+Megabyte :: 1024 * Kilobyte
+Gigabyte :: 1024 * Megabyte
+Terabyte :: 1024 * Gigabyte
+Petabyte :: 1024 * Terabyte
+Exabyte  :: 1024 * Petabyte
+
 // Logging stuff

 Logger_Level :: enum uint {
@@ -358,11 +409,34 @@ Logger :: struct {
 	options:      Logger_Options,
 }

+
+Random_Generator_Mode :: enum {
+	Read,
+	Reset,
+	Query_Info,
+}
+
+Random_Generator_Query_Info_Flag :: enum u32 {
+	Cryptographic,
+	Uniform,
+	External_Entropy,
+	Resettable,
+}
+Random_Generator_Query_Info :: distinct bit_set[Random_Generator_Query_Info_Flag; u32]
+
+Random_Generator_Proc :: #type proc(data: rawptr, mode: Random_Generator_Mode, p: []byte)
+
+Random_Generator :: struct {
+	procedure: Random_Generator_Proc,
+	data:      rawptr,
+}
+
 Context :: struct {
 	allocator:              Allocator,
 	temp_allocator:         Allocator,
 	assertion_failure_proc: Assertion_Failure_Proc,
 	logger:                 Logger,
+	random_generator:       Random_Generator,

 	user_ptr:   rawptr,
 	user_index: int,
@@ -389,9 +463,32 @@ Raw_Dynamic_Array :: struct {
 	allocator: Allocator,
 }

+// The raw, type-erased representation of a map.
+//
+// 32-bytes on 64-bit
+// 16-bytes on 32-bit
 Raw_Map :: struct {
-	hashes:  []int,
-	entries: Raw_Dynamic_Array,
+	// A single allocation spanning all keys, values, and hashes.
+	// {
+	//   k: Map_Cell(K) * (capacity / ks_per_cell)
+	//   v: Map_Cell(V) * (capacity / vs_per_cell)
+	//   h: Map_Cell(H) * (capacity / hs_per_cell)
+	// }
+	//
+	// The data is allocated assuming 64-byte alignment, meaning the address is
+	// always a multiple of 64. This means we have 6 bits of zeros in the pointer
+	// to store the capacity. We can store a value as large as 2^6-1 or 63 in
+	// there. This conveniently is the maximum log2 capacity we can have for a map
+	// as Odin uses signed integers to represent capacity.
+	//
+	// Since the hashes are backed by Map_Hash, which is just a 64-bit unsigned
+	// integer, the cell structure for hashes is unnecessary because 64/8 is 8 and
+	// requires no padding, meaning it can be indexed as a regular array of
+	// 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:       uintptr,   // 8-bytes on 64-bits, 4-bytes on 32-bits
+	allocator: Allocator, // 16-bytes on 64-bits, 8-bytes on 32-bits
 }

 Raw_Any :: struct {
@@ -403,6 +500,21 @@ Raw_Cstring :: struct {
 	data: [^]byte,
 }

+Raw_Soa_Pointer :: struct {
+	data:  rawptr,
+	index: int,
+}
+
+Raw_Complex32     :: struct {real, imag: f16}
+Raw_Complex64     :: struct {real, imag: f32}
+Raw_Complex128    :: struct {real, imag: f64}
+Raw_Quaternion64  :: struct {imag, jmag, kmag: f16, real: f16}
+Raw_Quaternion128 :: struct {imag, jmag, kmag: f32, real: f32}
+Raw_Quaternion256 :: struct {imag, jmag, kmag: f64, real: f64}
+Raw_Quaternion64_Vector_Scalar  :: struct {vector: [3]f16, scalar: f16}
+Raw_Quaternion128_Vector_Scalar :: struct {vector: [3]f32, scalar: f32}
+Raw_Quaternion256_Vector_Scalar :: struct {vector: [3]f64, scalar: f64}
+

 /*
 	// Defined internally by the compiler
@@ -414,8 +526,11 @@ Raw_Cstring :: struct {
 		Essence,
 		FreeBSD,
 		OpenBSD,
+		NetBSD,
+		Haiku,
 		WASI,
 		JS,
+		Orca,
 		Freestanding,
 	}
 */
@@ -430,16 +545,30 @@ Odin_OS_Type :: type_of(ODIN_OS)
 		arm32,
 		arm64,
 		wasm32,
-		wasm64,
+		wasm64p32,
+		riscv64,
 	}
 */
 Odin_Arch_Type :: type_of(ODIN_ARCH)

+Odin_Arch_Types :: bit_set[Odin_Arch_Type]
+
+ALL_ODIN_ARCH_TYPES :: Odin_Arch_Types{
+	.amd64,
+	.i386,
+	.arm32,
+	.arm64,
+	.wasm32,
+	.wasm64p32,
+	.riscv64,
+}
+
 /*
 	// Defined internally by the compiler
 	Odin_Build_Mode_Type :: enum int {
 		Executable,
 		Dynamic,
+		Static,
 		Object,
 		Assembly,
 		LLVM_IR,
@@ -457,6 +586,58 @@ Odin_Build_Mode_Type :: type_of(ODIN_BUILD_MODE)
 */
 Odin_Endian_Type :: type_of(ODIN_ENDIAN)

+Odin_OS_Types :: bit_set[Odin_OS_Type]
+
+ALL_ODIN_OS_TYPES :: Odin_OS_Types{
+	.Windows,
+	.Darwin,
+	.Linux,
+	.Essence,
+	.FreeBSD,
+	.OpenBSD,
+	.NetBSD,
+	.Haiku,
+	.WASI,
+	.JS,
+	.Orca,
+	.Freestanding,
+}
+
+/*
+	// 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 bit_set[Odin_Sanitizer_Flag; u32]
+
+	ODIN_SANITIZER_FLAGS // is a constant
+*/
+Odin_Sanitizer_Flags :: type_of(ODIN_SANITIZER_FLAGS)
+
+/*
+	// Defined internally by the compiler
+	Odin_Optimization_Mode :: enum int {
+		None       = -1,
+		Minimal    =  0,
+		Size       =  1,
+		Speed      =  2,
+		Aggressive =  3,
+	}
+
+	ODIN_OPTIMIZATION_MODE // is a constant
+*/
+Odin_Optimization_Mode :: type_of(ODIN_OPTIMIZATION_MODE)

 /////////////////////////////
 // Init Startup Procedures //
@@ -467,11 +648,8 @@ Odin_Endian_Type :: type_of(ODIN_ENDIAN)
 foreign {
 	@(link_name="__$startup_runtime")
 	_startup_runtime :: proc "odin" () ---
-}
-
-@(link_name="__$cleanup_runtime")
-_cleanup_runtime :: proc() {
-	default_temp_allocator_destroy(&global_default_temp_allocator_data)
+	@(link_name="__$cleanup_runtime")
+	_cleanup_runtime :: proc "odin" () ---
 }

 _cleanup_runtime_contextless :: proc "contextless" () {
@@ -509,8 +687,9 @@ 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
 		}
 	}
@@ -525,10 +704,10 @@ __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)
@@ -584,7 +763,9 @@ __init_context :: proc "contextless" (c: ^Context) {
 	c.allocator.data = nil

 	c.temp_allocator.procedure = default_temp_allocator_proc
-	c.temp_allocator.data = &global_default_temp_allocator_data
+	when !NO_DEFAULT_TEMP_ALLOCATOR {
+		c.temp_allocator.data = &global_default_temp_allocator_data
+	}
 	
 	when !ODIN_DISABLE_ASSERT {
 		c.assertion_failure_proc = default_assertion_failure_proc
@@ -592,20 +773,40 @@ __init_context :: proc "contextless" (c: ^Context) {

 	c.logger.procedure = default_logger_proc
 	c.logger.data = nil
+
+	c.random_generator.procedure = default_random_generator_proc
+	c.random_generator.data = nil
 }

 default_assertion_failure_proc :: proc(prefix, message: string, loc: Source_Code_Location) -> ! {
+	default_assertion_contextless_failure_proc(prefix, message, loc)
+}
+
+default_assertion_contextless_failure_proc :: proc "contextless" (prefix, message: string, loc: Source_Code_Location) -> ! {
 	when ODIN_OS == .Freestanding {
 		// Do nothing
 	} else {
-		print_caller_location(loc)
-		print_string(" ")
+		when ODIN_OS != .Orca && !ODIN_DISABLE_ASSERT {
+			print_caller_location(loc)
+			print_string(" ")
+		}
 		print_string(prefix)
 		if len(message) > 0 {
 			print_string(": ")
 			print_string(message)
 		}
-		print_byte('\n')
+
+		when ODIN_OS == .Orca {
+			assert_fail(
+				cstring(raw_data(loc.file_path)),
+				cstring(raw_data(loc.procedure)),
+				loc.line,
+				"",
+				cstring(raw_data(orca_stderr_buffer[:orca_stderr_buffer_idx])),
+			)
+		} else {
+			print_byte('\n')
+		}
 	}
 	trap()
 }
@@ -0,0 +1,535 @@
+package runtime
+
+import "base:intrinsics"
+_ :: intrinsics
+
+/*
+
+	SOA types are implemented with this sort of layout:
+
+	SOA Fixed Array
+	struct {
+		f0: [N]T0,
+		f1: [N]T1,
+		f2: [N]T2,
+	}
+
+	SOA Slice
+	struct {
+		f0: ^T0,
+		f1: ^T1,
+		f2: ^T2,
+
+		len: int,
+	}
+
+	SOA Dynamic Array
+	struct {
+		f0: ^T0,
+		f1: ^T1,
+		f2: ^T2,
+
+		len: int,
+		cap: int,
+		allocator: Allocator,
+	}
+
+	A footer is used rather than a header purely to simplify access to the fields internally
+	i.e. field index of the AOS == SOA
+
+*/
+
+
+Raw_SOA_Footer_Slice :: struct {
+	len: int,
+}
+
+Raw_SOA_Footer_Dynamic_Array :: struct {
+	len: int,
+	cap: int,
+	allocator: Allocator,
+}
+
+@(builtin, require_results)
+raw_soa_footer_slice :: proc(array: ^$T/#soa[]$E) -> (footer: ^Raw_SOA_Footer_Slice) {
+	if array == nil {
+		return nil
+	}
+	field_count := uintptr(len(E) when intrinsics.type_is_array(E) else intrinsics.type_struct_field_count(E))
+	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
+	}
+	field_count := uintptr(len(E) when intrinsics.type_is_array(E) else intrinsics.type_struct_field_count(E))
+	footer = (^Raw_SOA_Footer_Dynamic_Array)(uintptr(array) + field_count*size_of(rawptr))
+	return
+}
+raw_soa_footer :: proc{
+	raw_soa_footer_slice,
+	raw_soa_footer_dynamic_array,
+}
+
+
+
+@(builtin, require_results)
+make_soa_aligned :: proc($T: typeid/#soa[]$E, #any_int length, alignment: int, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) #optional_allocator_error {
+	if length <= 0 {
+		return
+	}
+
+	footer := raw_soa_footer(&array)
+	if size_of(E) == 0 {
+		footer.len = length
+		return
+	}
+
+	max_align := max(alignment, align_of(E))
+
+	ti := type_info_of(typeid_of(T))
+	ti = type_info_base(ti)
+	si := &ti.variant.(Type_Info_Struct)
+
+	field_count := uintptr(len(E) when intrinsics.type_is_array(E) else intrinsics.type_struct_field_count(E))
+
+	total_size := 0
+	for i in 0..<field_count {
+		type := si.types[i].variant.(Type_Info_Multi_Pointer).elem
+		total_size += type.size * length
+		total_size = align_forward_int(total_size, max_align)
+	}
+
+	allocator := allocator
+	if allocator.procedure == nil {
+		allocator = context.allocator
+	}
+	assert(allocator.procedure != nil)
+
+	new_bytes: []byte
+	new_bytes, err = allocator.procedure(
+		allocator.data, .Alloc, total_size, max_align,
+		nil, 0, loc,
+	)
+	if new_bytes == nil || err != nil {
+		return
+	}
+	new_data := raw_data(new_bytes)
+
+	data := uintptr(&array)
+	offset := 0
+	for i in 0..<field_count {
+		type := si.types[i].variant.(Type_Info_Multi_Pointer).elem
+
+		offset = align_forward_int(offset, max_align)
+
+		(^uintptr)(data)^ = uintptr(new_data) + uintptr(offset)
+		data += size_of(rawptr)
+		offset += type.size * length
+	}
+	footer.len = length
+
+	return
+}
+
+@(builtin, require_results)
+make_soa_slice :: proc($T: typeid/#soa[]$E, #any_int 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, 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, 0, loc) or_return
+	return array, nil
+}
+
+@(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) or_return
+	return array, nil
+}
+
+@(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
+	reserve_soa(&array, capacity, loc) or_return
+	resize_soa(&array, length, loc) or_return
+	return array, nil
+}
+
+
+@builtin
+make_soa :: proc{
+	make_soa_slice,
+	make_soa_dynamic_array,
+	make_soa_dynamic_array_len,
+	make_soa_dynamic_array_len_cap,
+}
+
+
+@builtin
+resize_soa :: proc(array: ^$T/#soa[dynamic]$E, #any_int length: int, loc := #caller_location) -> Allocator_Error {
+	if array == nil {
+		return nil
+	}
+	reserve_soa(array, length, loc) or_return
+	footer := raw_soa_footer(array)
+	footer.len = length
+	return nil
+}
+
+@builtin
+non_zero_resize_soa :: proc(array: ^$T/#soa[dynamic]$E, #any_int length: int, loc := #caller_location) -> Allocator_Error {
+	if array == nil {
+		return nil
+	}
+	non_zero_reserve_soa(array, length, loc) or_return
+	footer := raw_soa_footer(array)
+	footer.len = length
+	return nil
+}
+
+@builtin
+reserve_soa :: proc(array: ^$T/#soa[dynamic]$E, #any_int capacity: int, loc := #caller_location) -> Allocator_Error {
+	return _reserve_soa(array, capacity, true, loc)
+}
+
+@builtin
+non_zero_reserve_soa :: proc(array: ^$T/#soa[dynamic]$E, #any_int capacity: int, loc := #caller_location) -> Allocator_Error {
+	return _reserve_soa(array, capacity, false, loc)
+}
+
+_reserve_soa :: proc(array: ^$T/#soa[dynamic]$E, capacity: int, zero_memory: bool, loc := #caller_location) -> Allocator_Error {
+	if array == nil {
+		return nil
+	}
+
+	old_cap := cap(array)
+	if capacity <= old_cap {
+		return nil
+	}
+
+	if array.allocator.procedure == nil {
+		array.allocator = context.allocator
+	}
+	assert(array.allocator.procedure != nil)
+
+	footer := raw_soa_footer(array)
+	if size_of(E) == 0 {
+		footer.cap = capacity
+		return nil
+	}
+
+	ti := type_info_of(typeid_of(T))
+	ti = type_info_base(ti)
+	si := &ti.variant.(Type_Info_Struct)
+
+	field_count := uintptr(len(E) when intrinsics.type_is_array(E) else intrinsics.type_struct_field_count(E))
+	assert(footer.cap == old_cap)
+
+	old_size := 0
+	new_size := 0
+
+	max_align :: align_of(E)
+	for i in 0..<field_count {
+		type := si.types[i].variant.(Type_Info_Multi_Pointer).elem
+
+		old_size += type.size * old_cap
+		new_size += type.size * capacity
+
+		old_size = align_forward_int(old_size, max_align)
+		new_size = align_forward_int(new_size, max_align)
+	}
+
+	old_data := (^rawptr)(array)^
+
+	new_bytes := array.allocator.procedure(
+		array.allocator.data, .Alloc if zero_memory else .Alloc_Non_Zeroed, new_size, max_align,
+		nil, old_size, loc,
+	) or_return
+	new_data := raw_data(new_bytes)
+
+
+	footer.cap = capacity
+
+	old_offset := 0
+	new_offset := 0
+	for i in 0..<field_count {
+		type := si.types[i].variant.(Type_Info_Multi_Pointer).elem
+
+		old_offset = align_forward_int(old_offset, max_align)
+		new_offset = align_forward_int(new_offset, max_align)
+
+		new_data_elem := rawptr(uintptr(new_data) + uintptr(new_offset))
+		old_data_elem := rawptr(uintptr(old_data) + uintptr(old_offset))
+
+		mem_copy(new_data_elem, old_data_elem, type.size * old_cap)
+
+		(^rawptr)(uintptr(array) + i*size_of(rawptr))^ = new_data_elem
+
+		old_offset += type.size * old_cap
+		new_offset += type.size * capacity
+	}
+
+	array.allocator.procedure(
+		array.allocator.data, .Free, 0, max_align,
+		old_data, old_size, loc,
+	) or_return
+
+	return nil
+}
+
+
+@builtin
+append_soa_elem :: proc(array: ^$T/#soa[dynamic]$E, #no_broadcast arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_soa_elem(array, true, arg, loc)
+}
+
+@builtin
+non_zero_append_soa_elem :: proc(array: ^$T/#soa[dynamic]$E, #no_broadcast arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_soa_elem(array, false, arg, loc)
+}
+
+_append_soa_elem :: proc(array: ^$T/#soa[dynamic]$E, zero_memory: bool, #no_broadcast arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	if array == nil {
+		return 0, nil
+	}
+
+	if cap(array) <= len(array) + 1 {
+		// Same behavior as append_soa_elems but there's only one arg, so we always just add DEFAULT_DYNAMIC_ARRAY_CAPACITY.
+		cap := 2 * cap(array) + DEFAULT_DYNAMIC_ARRAY_CAPACITY
+		err = _reserve_soa(array, cap, zero_memory, loc) // do not 'or_return' here as it could be a partial success
+	}
+
+	footer := raw_soa_footer(array)
+
+	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(len(E) when intrinsics.type_is_array(E) else intrinsics.type_struct_field_count(E))
+
+		data := (^rawptr)(array)^
+
+		soa_offset := 0
+		item_offset := 0
+
+		arg_copy := arg
+		arg_ptr := &arg_copy
+
+		max_align :: align_of(E)
+		for i in 0..<field_count {
+			type := si.types[i].variant.(Type_Info_Multi_Pointer).elem
+
+			soa_offset  = align_forward_int(soa_offset, max_align)
+			item_offset = align_forward_int(item_offset, type.align)
+
+			dst := rawptr(uintptr(data) + uintptr(soa_offset) + uintptr(type.size * footer.len))
+			src := rawptr(uintptr(arg_ptr) + uintptr(item_offset))
+			mem_copy(dst, src, type.size)
+
+			soa_offset  += type.size * cap(array)
+			item_offset += type.size
+		}
+		footer.len += 1
+		return 1, err
+	}
+	return 0, err
+}
+
+@builtin
+append_soa_elems :: proc(array: ^$T/#soa[dynamic]$E, #no_broadcast args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_soa_elems(array, true, args=args, loc=loc)
+}
+
+@builtin
+non_zero_append_soa_elems :: proc(array: ^$T/#soa[dynamic]$E, #no_broadcast args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_soa_elems(array, false, args=args, loc=loc)
+}
+
+
+_append_soa_elems :: proc(array: ^$T/#soa[dynamic]$E, zero_memory: bool, #no_broadcast args: []E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	if array == nil {
+		return
+	}
+
+	arg_len := len(args)
+	if arg_len == 0 {
+		return
+	}
+
+	if cap(array) <= len(array)+arg_len {
+		cap := 2 * cap(array) + max(DEFAULT_DYNAMIC_ARRAY_CAPACITY, arg_len)
+		err = _reserve_soa(array, cap, zero_memory, 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))
+		ti = type_info_base(ti)
+		si := &ti.variant.(Type_Info_Struct)
+		field_count := uintptr(len(E) when intrinsics.type_is_array(E) else intrinsics.type_struct_field_count(E))
+
+		data := (^rawptr)(array)^
+
+		soa_offset := 0
+		item_offset := 0
+
+		args_ptr := &args[0]
+
+		max_align :: align_of(E)
+		for i in 0..<field_count {
+			type := si.types[i].variant.(Type_Info_Multi_Pointer).elem
+
+			soa_offset  = align_forward_int(soa_offset, max_align)
+			item_offset = align_forward_int(item_offset, type.align)
+
+			dst := uintptr(data) + uintptr(soa_offset) + uintptr(type.size * footer.len)
+			src := uintptr(args_ptr) + uintptr(item_offset)
+			for j in 0..<arg_len {
+				d := rawptr(dst + uintptr(j*type.size))
+				s := rawptr(src + uintptr(j*size_of(E)))
+				mem_copy(d, s, type.size)
+			}
+
+			soa_offset  += type.size * cap(array)
+			item_offset += type.size
+		}
+	}
+	footer.len += arg_len
+	return arg_len, err
+}
+
+
+// The append_soa built-in procedure appends elements to the end of an #soa dynamic array
+@builtin
+append_soa :: proc{
+	append_soa_elem,
+	append_soa_elems,
+}
+
+
+delete_soa_slice :: proc(array: $T/#soa[]$E, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
+	field_count :: len(E) when intrinsics.type_is_array(E) else intrinsics.type_struct_field_count(E)
+	when field_count != 0 {
+		array := array
+		ptr := (^rawptr)(&array)^
+		free(ptr, allocator, loc) or_return
+	}
+	return nil
+}
+
+delete_soa_dynamic_array :: proc(array: $T/#soa[dynamic]$E, loc := #caller_location) -> Allocator_Error {
+	field_count :: len(E) when intrinsics.type_is_array(E) else intrinsics.type_struct_field_count(E)
+	when field_count != 0 {
+		array := array
+		ptr := (^rawptr)(&array)^
+		footer := raw_soa_footer(&array)
+		free(ptr, footer.allocator, loc) or_return
+	}
+	return nil
+}
+
+
+@builtin
+delete_soa :: proc{
+	delete_soa_slice,
+	delete_soa_dynamic_array,
+}
+
+
+clear_soa_dynamic_array :: proc(array: ^$T/#soa[dynamic]$E) {
+	field_count :: len(E) when intrinsics.type_is_array(E) else intrinsics.type_struct_field_count(E)
+	when field_count != 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(len(E) when intrinsics.type_is_array(E) else 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, #any_int 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(len(E) when intrinsics.type_is_array(E) else intrinsics.type_struct_field_count(E))
+
+		data := uintptr(array)
+		for i in 0..<field_count {
+			type := si.types[i].variant.(Type_Info_Multi_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, #any_int 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(len(E) when intrinsics.type_is_array(E) else intrinsics.type_struct_field_count(E))
+
+		data := uintptr(array)
+		for i in 0..<field_count {
+			type := si.types[i].variant.(Type_Info_Multi_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
+}
@@ -0,0 +1,15 @@
+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 when ODIN_OS != .Orca && (ODIN_ARCH == .wasm32 || ODIN_ARCH == .wasm64p32) {
+	default_allocator :: default_wasm_allocator
+	default_allocator_proc :: wasm_allocator_proc
+} 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 +1,306 @@
+package runtime
+
+import "base:intrinsics"
+
+DEFAULT_ARENA_GROWING_MINIMUM_BLOCK_SIZE :: uint(DEFAULT_TEMP_ALLOCATOR_BACKING_SIZE)
+
+Memory_Block :: struct {
+	prev:      ^Memory_Block,
+	allocator: Allocator,
+	base:      [^]byte,
+	used:      uint,
+	capacity:  uint,
+}
+
+// NOTE: This is a growing arena that is only used for the default temp allocator.
+// For your own growing arena needs, prefer `Arena` from `core:mem/virtual`.
+Arena :: struct {
+	backing_allocator:  Allocator,
+	curr_block:         ^Memory_Block,
+	total_used:         uint,
+	total_capacity:     uint,
+	minimum_block_size: uint,
+	temp_count:         uint,
+}
+
+@(private, require_results)
+safe_add :: #force_inline proc "contextless" (x, y: uint) -> (uint, bool) {
+	z, did_overflow := intrinsics.overflow_add(x, y)
+	return z, !did_overflow
+}
+
+@(require_results)
+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), 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):])
+
+	block.allocator = allocator
+	block.base = ([^]byte)(uintptr(block) + base_offset)
+	block.capacity = uint(end - uintptr(block.base))
+
+	// Should be zeroed
+	assert(block.used == 0)
+	assert(block.prev == nil)
+	return
+}
+
+memory_block_dealloc :: proc(block_to_free: ^Memory_Block, loc := #caller_location) {
+	if block_to_free != nil {
+		allocator := block_to_free.allocator
+		mem_free(block_to_free, allocator, loc)
+	}
+}
+
+@(require_results)
+alloc_from_memory_block :: proc(block: ^Memory_Block, min_size, alignment: uint) -> (data: []byte, err: Allocator_Error) {
+	calc_alignment_offset :: proc "contextless" (block: ^Memory_Block, alignment: uintptr) -> uint {
+		alignment_offset := uint(0)
+		ptr := uintptr(block.base[block.used:])
+		mask := alignment-1
+		if ptr & mask != 0 {
+			alignment_offset = uint(alignment - (ptr & mask))
+		}
+		return alignment_offset
+
+	}
+	if block == nil {
+		return nil, .Out_Of_Memory
+	}
+	alignment_offset := calc_alignment_offset(block, uintptr(alignment))
+	size, size_ok := safe_add(min_size, alignment_offset)
+	if !size_ok {
+		err = .Out_Of_Memory
+		return
+	}
+
+	if to_be_used, ok := safe_add(block.used, size); !ok || to_be_used > block.capacity {
+		err = .Out_Of_Memory
+		return
+	}
+	data = block.base[block.used+alignment_offset:][:min_size]
+	block.used += size
+	return
+}
+
+@(require_results)
+arena_alloc :: proc(arena: ^Arena, size, alignment: uint, loc := #caller_location) -> (data: []byte, err: Allocator_Error) {
+	align_forward_uint :: proc "contextless" (ptr, align: uint) -> uint {
+		p := ptr
+		modulo := p & (align-1)
+		if modulo != 0 {
+			p += align - modulo
+		}
+		return p
+	}
+
+	assert(alignment & (alignment-1) == 0, "non-power of two alignment", loc)
+
+	size := size
+	if size == 0 {
+		return
+	}
+	
+	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(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, alignment, loc) or_return
+		new_block.prev = arena.curr_block
+		arena.curr_block = new_block
+		arena.total_capacity += new_block.capacity
+	}
+
+	prev_used := arena.curr_block.used
+	data, err = alloc_from_memory_block(arena.curr_block, size, alignment)
+	arena.total_used += arena.curr_block.used - prev_used
+	return
+}
+
+// `arena_init` will initialize the arena with a usable block.
+// This procedure is not necessary to use the Arena as the default zero as `arena_alloc` will set things up if necessary
+@(require_results)
+arena_init :: proc(arena: ^Arena, size: uint, backing_allocator: Allocator, loc := #caller_location) -> Allocator_Error {
+	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, 0, loc) or_return
+	arena.curr_block = new_block
+	arena.total_capacity += new_block.capacity
+	return nil
+}
+
+
+arena_free_last_memory_block :: proc(arena: ^Arena, loc := #caller_location) {
+	if free_block := arena.curr_block; free_block != nil {
+		arena.curr_block = free_block.prev
+
+		arena.total_capacity -= free_block.capacity
+		memory_block_dealloc(free_block, loc)
+	}
+}
+
+// `arena_free_all` will free all but the first memory block, and then reset the memory block
+arena_free_all :: proc(arena: ^Arena, loc := #caller_location) {
+	for arena.curr_block != nil && arena.curr_block.prev != nil {
+		arena_free_last_memory_block(arena, loc)
+	}
+
+	if arena.curr_block != nil {
+		intrinsics.mem_zero(arena.curr_block.base, arena.curr_block.used)
+		arena.curr_block.used = 0
+	}
+	arena.total_used = 0
+}
+
+arena_destroy :: proc(arena: ^Arena, loc := #caller_location) {
+	for arena.curr_block != nil {
+		free_block := arena.curr_block
+		arena.curr_block = free_block.prev
+
+		arena.total_capacity -= free_block.capacity
+		memory_block_dealloc(free_block, loc)
+	}
+	arena.total_used = 0
+	arena.total_capacity = 0
+}
+
+arena_allocator :: proc(arena: ^Arena) -> Allocator {
+	return Allocator{arena_allocator_proc, arena}
+}
+
+arena_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
+                             size, alignment: int,
+                             old_memory: rawptr, old_size: int,
+                             location := #caller_location) -> (data: []byte, err: Allocator_Error) {
+	arena := (^Arena)(allocator_data)
+
+	size, alignment := uint(size), uint(alignment)
+	old_size := uint(old_size)
+
+	switch mode {
+	case .Alloc, .Alloc_Non_Zeroed:
+		return arena_alloc(arena, size, alignment, location)
+	case .Free:
+		err = .Mode_Not_Implemented
+	case .Free_All:
+		arena_free_all(arena, location)
+	case .Resize, .Resize_Non_Zeroed:
+		old_data := ([^]byte)(old_memory)
+
+		switch {
+		case old_data == nil:
+			return arena_alloc(arena, size, alignment, location)
+		case size == old_size:
+			// return old memory
+			data = old_data[:size]
+			return
+		case size == 0:
+			err = .Mode_Not_Implemented
+			return
+		case (uintptr(old_data) & uintptr(alignment-1) == 0) && size < old_size:
+			// shrink data in-place
+			data = old_data[:size]
+			return
+		}
+
+		new_memory := arena_alloc(arena, size, alignment, location) or_return
+		if new_memory == nil {
+			return
+		}
+		copy(new_memory, old_data[:old_size])
+		return new_memory, nil
+	case .Query_Features:
+		set := (^Allocator_Mode_Set)(old_memory)
+		if set != nil {
+			set^ = {.Alloc, .Alloc_Non_Zeroed, .Free_All, .Resize, .Query_Features}
+		}
+	case .Query_Info:
+		err = .Mode_Not_Implemented
+	}
+
+	return
+}
+
+
+
+
+Arena_Temp :: struct {
+	arena: ^Arena,
+	block: ^Memory_Block,
+	used:  uint,
+}
+
+@(require_results)
+arena_temp_begin :: proc(arena: ^Arena, loc := #caller_location) -> (temp: Arena_Temp) {
+	assert(arena != nil, "nil arena", loc)
+
+	temp.arena = arena
+	temp.block = arena.curr_block
+	if arena.curr_block != nil {
+		temp.used = arena.curr_block.used
+	}
+	arena.temp_count += 1
+	return
+}
+
+arena_temp_end :: proc(temp: Arena_Temp, loc := #caller_location) {
+	if temp.arena == nil {
+		assert(temp.block == nil)
+		assert(temp.used == 0)
+		return
+	}
+	arena := temp.arena
+
+	if temp.block != nil {
+		memory_block_found := false
+		for block := arena.curr_block; block != nil; block = block.prev {
+			if block == temp.block {
+				memory_block_found = true
+				break
+			}
+		}
+		if !memory_block_found {
+			assert(arena.curr_block == temp.block, "memory block stored within Arena_Temp not owned by Arena", loc)
+		}
+
+		for arena.curr_block != temp.block {
+			arena_free_last_memory_block(arena)
+		}
+
+		if block := arena.curr_block; block != nil {
+			assert(block.used >= temp.used, "out of order use of arena_temp_end", loc)
+			amount_to_zero := min(block.used-temp.used, block.capacity-block.used)
+			intrinsics.mem_zero(block.base[temp.used:], amount_to_zero)
+			block.used = temp.used
+		}
+	}
+
+	assert(arena.temp_count > 0, "double-use of arena_temp_end", loc)
+	arena.temp_count -= 1
+}
+
+// Ignore the use of a `arena_temp_begin` entirely
+arena_temp_ignore :: proc(temp: Arena_Temp, loc := #caller_location) {
+	assert(temp.arena != nil, "nil arena", loc)
+	arena := temp.arena
+
+	assert(arena.temp_count > 0, "double-use of arena_temp_end", loc)
+	arena.temp_count -= 1
+}
+
+arena_check_temp :: proc(arena: ^Arena, loc := #caller_location) {
+	assert(arena.temp_count == 0, "Arena_Temp not been ended", loc)
+}
@@ -0,0 +1,79 @@
+package runtime
+
+DEFAULT_TEMP_ALLOCATOR_BACKING_SIZE: int : #config(DEFAULT_TEMP_ALLOCATOR_BACKING_SIZE, 4 * Megabyte)
+NO_DEFAULT_TEMP_ALLOCATOR: bool : ODIN_OS == .Freestanding || ODIN_DEFAULT_TO_NIL_ALLOCATOR
+
+when NO_DEFAULT_TEMP_ALLOCATOR {
+	Default_Temp_Allocator :: struct {}
+	
+	default_temp_allocator_init :: proc(s: ^Default_Temp_Allocator, size: int, backing_allocator := context.allocator) {}
+	
+	default_temp_allocator_destroy :: proc(s: ^Default_Temp_Allocator) {}
+	
+	default_temp_allocator_proc :: nil_allocator_proc
+
+	@(require_results)
+	default_temp_allocator_temp_begin :: proc(loc := #caller_location) -> (temp: Arena_Temp) {
+		return
+	}
+
+	default_temp_allocator_temp_end :: proc(temp: Arena_Temp, loc := #caller_location) {
+	}
+} else {
+	Default_Temp_Allocator :: struct {
+		arena: Arena,
+	}
+	
+	default_temp_allocator_init :: proc(s: ^Default_Temp_Allocator, size: int, backing_allocator := context.allocator) {
+		_ = arena_init(&s.arena, uint(size), backing_allocator)
+	}
+
+	default_temp_allocator_destroy :: proc(s: ^Default_Temp_Allocator) {
+		if s != nil {
+			arena_destroy(&s.arena)
+			s^ = {}
+		}
+	}
+
+	default_temp_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
+	                                    size, alignment: int,
+	                                    old_memory: rawptr, old_size: int, loc := #caller_location) -> (data: []byte, err: Allocator_Error) {
+
+		s := (^Default_Temp_Allocator)(allocator_data)
+		return arena_allocator_proc(&s.arena, mode, size, alignment, old_memory, old_size, loc)
+	}
+
+	@(require_results)
+	default_temp_allocator_temp_begin :: proc(loc := #caller_location) -> (temp: Arena_Temp) {
+		if context.temp_allocator.data == &global_default_temp_allocator_data {
+			temp = arena_temp_begin(&global_default_temp_allocator_data.arena, loc)
+		}
+		return
+	}
+
+	default_temp_allocator_temp_end :: proc(temp: Arena_Temp, loc := #caller_location) {
+		arena_temp_end(temp, loc)
+	}
+
+	@(fini, private)
+	_destroy_temp_allocator_fini :: proc() {
+		default_temp_allocator_destroy(&global_default_temp_allocator_data)
+	}
+}
+
+@(deferred_out=default_temp_allocator_temp_end)
+DEFAULT_TEMP_ALLOCATOR_TEMP_GUARD :: #force_inline proc(ignore := false, loc := #caller_location) -> (Arena_Temp, Source_Code_Location) {
+	if ignore {
+		return {}, loc
+	} else {
+		return default_temp_allocator_temp_begin(loc), loc
+	}
+}
+
+
+default_temp_allocator :: proc(allocator: ^Default_Temp_Allocator) -> Allocator {
+	return Allocator{
+		procedure = default_temp_allocator_proc,
+		data      = allocator,
+	}
+}
@@ -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 ([dynamic]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
+
+*/
@@ -29,11 +29,15 @@ __dynamic_array_reserve :: proc(array_: rawptr, elem_size, elem_align: int, cap:
 	new_size  := cap * elem_size
 	allocator := array.allocator

-	new_data, err := allocator.procedure(allocator.data, .Resize, new_size, elem_align, array.data, old_size, loc)
+	new_data, err := mem_resize(array.data, old_size, new_size, elem_align, allocator, loc)
 	if err != nil {
 		return false
 	}
-	if new_data != nil || elem_size == 0 {
+	if elem_size == 0 {
+		array.data = raw_data(new_data)
+		array.cap = cap
+		return true
+	} else if new_data != nil {
 		array.data = raw_data(new_data)
 		array.cap = min(cap, len(new_data)/elem_size)
 		return true
@@ -55,11 +59,13 @@ __dynamic_array_shrink :: proc(array_: rawptr, elem_size, elem_align: int, new_c
 		return
 	}

+	new_cap := new_cap
+	new_cap = max(new_cap, 0)
 	old_size  := array.cap * elem_size
 	new_size  := new_cap * elem_size
 	allocator := array.allocator

-	new_data, err := allocator.procedure(allocator.data, .Resize, new_size, elem_align, array.data, old_size, loc)
+	new_data, err := mem_resize(array.data, old_size, new_size, elem_align, allocator, loc)
 	if err != nil {
 		return
 	}
@@ -0,0 +1,991 @@
+package runtime
+
+import "base:intrinsics"
+_ :: intrinsics
+
+// High performance, cache-friendly, open-addressed Robin Hood hashing hash map
+// data structure with various optimizations for Odin.
+//
+// Copyright 2022 (c) Dale Weiler
+//
+// The core of the hash map data structure is the Raw_Map struct which is a
+// type-erased representation of the map. This type-erased representation is
+// used in two ways: static and dynamic. When static type information is known,
+// the procedures suffixed with _static should be used instead of _dynamic. The
+// static procedures are optimized since they have type information. Hashing of
+// keys, comparison of keys, and data lookup are all optimized. When type
+// information is not known, the procedures suffixed with _dynamic should be
+// used. The representation of the map is the same for both static and dynamic,
+// and procedures of each can be mixed and matched. The purpose of the dynamic
+// representation is to enable reflection and runtime manipulation of the map.
+// The dynamic procedures all take an additional Map_Info structure parameter
+// which carries runtime values describing the size, alignment, and offset of
+// various traits of a given key and value type pair. The Map_Info value can
+// be created by calling map_info(K, V) with the key and value typeids.
+//
+// This map implementation makes extensive use of uintptr for representing
+// sizes, lengths, capacities, masks, pointers, offsets, and addresses to avoid
+// expensive sign extension and masking that would be generated if types were
+// casted all over. The only place regular ints show up is in the cap() and
+// len() implementations.
+//
+// To make this map cache-friendly it uses a novel strategy to ensure keys and
+// values of the map are always cache-line aligned and that no single key or
+// value of any type ever straddles a cache-line. This cache efficiency makes
+// for quick lookups because the linear-probe always addresses data in a cache
+// friendly way. This is enabled through the use of a special meta-type called
+// a Map_Cell which packs as many values of a given type into a local array adding
+// internal padding to round to MAP_CACHE_LINE_SIZE. One other benefit to storing
+// the internal data in this manner is false sharing no longer occurs when using
+// a map, enabling efficient concurrent access of the map data structure with
+// minimal locking if desired.
+
+// With Robin Hood hashing a maximum load factor of 75% is ideal.
+MAP_LOAD_FACTOR :: 75
+
+// Minimum log2 capacity.
+MAP_MIN_LOG2_CAPACITY :: 3 // 8 elements
+
+// Has to be less than 100% though.
+#assert(MAP_LOAD_FACTOR < 100)
+
+// This is safe to change. The log2 size of a cache-line. At minimum it has to
+// be six though. Higher cache line sizes are permitted.
+MAP_CACHE_LINE_LOG2 :: 6
+
+// The size of a cache-line.
+MAP_CACHE_LINE_SIZE :: 1 << MAP_CACHE_LINE_LOG2
+
+// The minimum cache-line size allowed by this implementation is 64 bytes since
+// we need 6 bits in the base pointer to store the integer log2 capacity, which
+// at maximum is 63. Odin uses signed integers to represent length and capacity,
+// so only 63 bits are needed in the maximum case.
+#assert(MAP_CACHE_LINE_SIZE >= 64)
+
+// Map_Cell type that packs multiple T in such a way to ensure that each T stays
+// aligned by align_of(T) and such that align_of(Map_Cell(T)) % MAP_CACHE_LINE_SIZE == 0
+//
+// This means a value of type T will never straddle a cache-line.
+//
+// When multiple Ts can fit in a single cache-line the data array will have more
+// than one element. When it cannot, the data array will have one element and
+// an array of Map_Cell(T) will be padded to stay a multiple of MAP_CACHE_LINE_SIZE.
+//
+// We rely on the type system to do all the arithmetic and padding for us here.
+//
+// The usual array[index] indexing for []T backed by a []Map_Cell(T) becomes a bit
+// more involved as there now may be internal padding. The indexing now becomes
+//
+//  N :: len(Map_Cell(T){}.data)
+//  i := index / N
+//  j := index % N
+//  cell[i].data[j]
+//
+// However, since len(Map_Cell(T){}.data) is a compile-time constant, there are some
+// optimizations we can do to eliminate the need for any divisions as N will
+// be bounded by [1, 64).
+//
+// 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($T: typeid) #align(MAP_CACHE_LINE_SIZE) {
+	data: [MAP_CACHE_LINE_SIZE / size_of(T) when 0 < size_of(T) && size_of(T) < MAP_CACHE_LINE_SIZE else 1]T,
+}
+
+// So we can operate on a cell data structure at runtime without any type
+// information, we have a simple table that stores some traits about the cell.
+//
+// 32-bytes on 64-bit
+// 16-bytes on 32-bit
+Map_Cell_Info :: struct {
+	size_of_type:      uintptr, // 8-bytes on 64-bit, 4-bytes on 32-bits
+	align_of_type:     uintptr, // 8-bytes on 64-bit, 4-bytes on 32-bits
+	size_of_cell:      uintptr, // 8-bytes on 64-bit, 4-bytes on 32-bits
+	elements_per_cell: uintptr, // 8-bytes on 64-bit, 4-bytes on 32-bits
+}
+
+// map_cell_info :: proc "contextless" ($T: typeid) -> ^Map_Cell_Info {...}
+map_cell_info :: intrinsics.type_map_cell_info
+
+// Same as the above procedure but at runtime with the cell Map_Cell_Info value.
+@(require_results)
+map_cell_index_dynamic :: #force_inline proc "contextless" (base: uintptr, #no_alias info: ^Map_Cell_Info, index: uintptr) -> uintptr {
+	// Micro-optimize the common cases to save on integer division.
+	elements_per_cell := uintptr(info.elements_per_cell)
+	size_of_cell      := uintptr(info.size_of_cell)
+	switch elements_per_cell {
+	case 1:
+		return base + (index * size_of_cell)
+	case 2:
+		cell_index   := index >> 1
+		data_index   := index & 1
+		size_of_type := uintptr(info.size_of_type)
+		return base + (cell_index * size_of_cell) + (data_index * size_of_type)
+	case:
+		cell_index   := index / elements_per_cell
+		data_index   := index % elements_per_cell
+		size_of_type := uintptr(info.size_of_type)
+		return base + (cell_index * size_of_cell) + (data_index * size_of_type)
+	}
+}
+
+// Same as above procedure but with compile-time constant index.
+@(require_results)
+map_cell_index_dynamic_const :: proc "contextless" (base: uintptr, #no_alias info: ^Map_Cell_Info, $INDEX: uintptr) -> uintptr {
+	elements_per_cell := uintptr(info.elements_per_cell)
+	size_of_cell      := uintptr(info.size_of_cell)
+	size_of_type      := uintptr(info.size_of_type)
+	cell_index        := INDEX / elements_per_cell
+	data_index        := INDEX % elements_per_cell
+	return base + (cell_index * size_of_cell) + (data_index * size_of_type)
+}
+
+// We always round the capacity to a power of two so this becomes [16]Foo, which
+// works out to [4]Cell(Foo).
+//
+// The following compile-time procedure indexes such a [N]Cell(T) structure as
+// if it were a flat array accounting for the internal padding introduced by the
+// Cell structure.
+@(require_results)
+map_cell_index_static :: #force_inline proc "contextless" (cells: [^]Map_Cell($T), index: uintptr) -> ^T #no_bounds_check {
+	N :: size_of(Map_Cell(T){}.data) / size_of(T) when size_of(T) > 0 else 1
+
+	#assert(N <= MAP_CACHE_LINE_SIZE)
+
+	when size_of(Map_Cell(T)) == size_of([N]T) {
+		// No padding case, can treat as a regular array of []T.
+
+		return &([^]T)(cells)[index]
+	} else when (N & (N - 1)) == 0 && N <= 8*size_of(uintptr) {
+		// Likely case, N is a power of two because T is a power of two.
+
+		// Compute the integer log 2 of N, this is the shift amount to index the
+		// correct cell. Odin's intrinsics.count_leading_zeros does not produce a
+		// constant, hence this approach. We only need to check up to N = 64.
+		SHIFT :: 1 when N < 2  else
+		         2 when N < 4  else
+		         3 when N < 8  else
+		         4 when N < 16 else
+		         5 when N < 32 else 6
+		#assert(SHIFT <= MAP_CACHE_LINE_LOG2)
+		// Unique case, no need to index data here since only one element.
+		when N == 1 {
+			return &cells[index >> SHIFT].data[0]
+		} else {
+			return &cells[index >> SHIFT].data[index & (N - 1)]
+		}
+	} else {
+		// Least likely (and worst case), we pay for a division operation but we
+		// assume the compiler does not actually generate a division. N will be in the
+		// range [1, CACHE_LINE_SIZE) and not a power of two.
+		return &cells[index / N].data[index % N]
+	}
+}
+
+// len() for map
+@(require_results)
+map_len :: #force_inline proc "contextless" (m: Raw_Map) -> int {
+	return int(m.len)
+}
+
+// cap() for map
+@(require_results)
+map_cap :: #force_inline proc "contextless" (m: Raw_Map) -> int {
+	// The data uintptr stores the capacity in the lower six bits which gives the
+	// a maximum value of 2^6-1, or 63. We store the integer log2 of capacity
+	// since our capacity is always a power of two. We only need 63 bits as Odin
+	// represents length and capacity as a signed integer.
+	return 0 if m.data == 0 else 1 << map_log2_cap(m)
+}
+
+// Query the load factor of the map. This is not actually configurable, but
+// some math is needed to compute it. Compute it as a fixed point percentage to
+// avoid floating point operations. This division can be optimized out by
+// multiplying by the multiplicative inverse of 100.
+@(require_results)
+map_load_factor :: #force_inline proc "contextless" (log2_capacity: uintptr) -> uintptr {
+	return ((uintptr(1) << log2_capacity) * MAP_LOAD_FACTOR) / 100
+}
+
+@(require_results)
+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
+// used in the implementation rather than map_cap since the check for data = 0
+// isn't necessary in the implementation. cap() on the otherhand needs to work
+// when called on an empty map.
+@(require_results)
+map_log2_cap :: #force_inline proc "contextless" (m: Raw_Map) -> uintptr {
+	return m.data & (64 - 1)
+}
+
+// Canonicalize the data by removing the tagged capacity stored in the lower six
+// bits of the data uintptr.
+@(require_results)
+map_data :: #force_inline proc "contextless" (m: Raw_Map) -> uintptr {
+	return m.data &~ uintptr(64 - 1)
+}
+
+
+Map_Hash :: uintptr
+
+TOMBSTONE_MASK :: 1<<(size_of(Map_Hash)*8 - 1)
+
+// Procedure to check if a slot is empty for a given hash. This is represented
+// by the zero value to make the zero value useful. This is a procedure just
+// for prose reasons.
+@(require_results)
+map_hash_is_empty :: #force_inline proc "contextless" (hash: Map_Hash) -> bool {
+	return hash == 0
+}
+
+@(require_results)
+map_hash_is_deleted :: #force_no_inline proc "contextless" (hash: Map_Hash) -> bool {
+	// The MSB indicates a tombstone
+	return hash & TOMBSTONE_MASK != 0
+}
+@(require_results)
+map_hash_is_valid :: #force_inline proc "contextless" (hash: Map_Hash) -> bool {
+	// The MSB indicates a tombstone
+	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.
+@(require_results)
+map_desired_position :: #force_inline proc "contextless" (m: Raw_Map, hash: Map_Hash) -> uintptr {
+	// We do not use map_cap since we know the capacity will not be zero here.
+	capacity := uintptr(1) << map_log2_cap(m)
+	return uintptr(hash & Map_Hash(capacity - 1))
+}
+
+@(require_results)
+map_probe_distance :: #force_inline proc "contextless" (m: Raw_Map, hash: Map_Hash, slot: uintptr) -> uintptr {
+	// We do not use map_cap since we know the capacity will not be zero here.
+	capacity := uintptr(1) << map_log2_cap(m)
+	return (slot + capacity - map_desired_position(m, hash)) & (capacity - 1)
+}
+
+// When working with the type-erased structure at runtime we need information
+// about the map to make working with it possible. This info structure stores
+// that.
+//
+// `Map_Info` and `Map_Cell_Info` are read only data structures and cannot be
+// modified after creation
+//
+// 32-bytes on 64-bit
+// 16-bytes on 32-bit
+Map_Info :: struct {
+	ks: ^Map_Cell_Info, // 8-bytes on 64-bit, 4-bytes on 32-bit
+	vs: ^Map_Cell_Info, // 8-bytes on 64-bit, 4-bytes on 32-bit
+	key_hasher: proc "contextless" (key: rawptr, seed: Map_Hash) -> Map_Hash, // 8-bytes on 64-bit, 4-bytes on 32-bit
+	key_equal:  proc "contextless" (lhs, rhs: rawptr) -> bool,                // 8-bytes on 64-bit, 4-bytes on 32-bit
+}
+
+
+// The Map_Info structure is basically a pseudo-table of information for a given K and V pair.
+// map_info :: proc "contextless" ($T: typeid/map[$K]$V) -> ^Map_Info {...}
+map_info :: intrinsics.type_map_info
+
+@(require_results)
+map_kvh_data_dynamic :: proc "contextless" (m: Raw_Map, #no_alias info: ^Map_Info) -> (ks: uintptr, vs: uintptr, hs: [^]Map_Hash, sk: uintptr, sv: uintptr) {
+	INFO_HS := intrinsics.type_map_cell_info(Map_Hash)
+
+	capacity := uintptr(1) << map_log2_cap(m)
+	ks   = map_data(m)
+	vs   = map_cell_index_dynamic(ks,  info.ks, capacity) // Skip past ks to get start of vs
+	hs_ := map_cell_index_dynamic(vs,  info.vs, capacity) // Skip past vs to get start of hs
+	sk   = map_cell_index_dynamic(hs_, INFO_HS, capacity) // Skip past hs to get start of sk
+	// Need to skip past two elements in the scratch key space to get to the start
+	// of the scratch value space, of which there's only two elements as well.
+	sv = map_cell_index_dynamic_const(sk, info.ks, 2)
+
+	hs = ([^]Map_Hash)(hs_)
+	return
+}
+
+@(require_results)
+map_kvh_data_values_dynamic :: proc "contextless" (m: Raw_Map, #no_alias info: ^Map_Info) -> (vs: uintptr) {
+	capacity := uintptr(1) << map_log2_cap(m)
+	return map_cell_index_dynamic(map_data(m), info.ks, capacity) // Skip past ks to get start of vs
+}
+
+
+@(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
+		return (value + CACHE_MASK) &~ CACHE_MASK
+	}
+	INFO_HS := intrinsics.type_map_cell_info(Map_Hash)
+
+	size := uintptr(0)
+	size = round(map_cell_index_dynamic(size, info.ks, capacity))
+	size = round(map_cell_index_dynamic(size, info.vs, capacity))
+	size = round(map_cell_index_dynamic(size, INFO_HS, capacity))
+	size = round(map_cell_index_dynamic(size, info.ks, 2)) // Two additional ks for scratch storage
+	size = round(map_cell_index_dynamic(size, info.vs, 2)) // Two additional vs for scratch storage
+	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) {
+	result.allocator = allocator // set the allocator always
+	if log2_capacity == 0 {
+		return
+	}
+
+	if log2_capacity >= 64 {
+		// Overflowed, would be caused by log2_capacity > 64
+		return {}, .Out_Of_Memory
+	}
+
+	capacity := uintptr(1) << max(log2_capacity, MAP_MIN_LOG2_CAPACITY)
+
+	CACHE_MASK :: MAP_CACHE_LINE_SIZE - 1
+
+	size := map_total_allocation_size(capacity, info)
+
+	data := mem_alloc_non_zeroed(int(size), MAP_CACHE_LINE_SIZE, allocator, loc) or_return
+	data_ptr := uintptr(raw_data(data))
+	if data_ptr == 0 {
+		err = .Out_Of_Memory
+		return
+	}
+	if intrinsics.expect(data_ptr & CACHE_MASK != 0, false) {
+		panic("allocation not aligned to a cache line", loc)
+	} else {
+		result.data = data_ptr | log2_capacity // Tagged pointer representation for capacity.
+		result.len = 0
+
+		map_clear_dynamic(&result, info)
+	}
+	return
+}
+
+// This procedure has to stack allocate storage to store local keys during the
+// Robin Hood hashing technique where elements are swapped in the backing
+// 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, 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
+	pos      := map_desired_position(m^, h)
+	distance := uintptr(0)
+	mask     := (uintptr(1) << map_log2_cap(m^)) - 1
+
+	ks, vs, hs, sk, sv := map_kvh_data_dynamic(m^, info)
+
+	// Avoid redundant loads of these values
+	size_of_k := info.ks.size_of_type
+	size_of_v := info.vs.size_of_type
+
+	k := map_cell_index_dynamic(sk, info.ks, 0)
+	v := map_cell_index_dynamic(sv, info.vs, 0)
+	intrinsics.mem_copy_non_overlapping(rawptr(k), rawptr(ik), size_of_k)
+	intrinsics.mem_copy_non_overlapping(rawptr(v), rawptr(iv), size_of_v)
+
+	// Temporary k and v dynamic storage for swap below
+	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")
+		}
+
+		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)
+			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 result == 0 {
+				result = map_cell_index_dynamic(vs, info.vs, pos)
+			}
+
+			kp := map_cell_index_dynamic(ks, info.ks, pos)
+			vp := map_cell_index_dynamic(vs, info.vs, pos)
+
+			intrinsics.mem_copy_non_overlapping(rawptr(tk), rawptr(k), size_of_k)
+			intrinsics.mem_copy_non_overlapping(rawptr(k),  rawptr(kp), size_of_k)
+			intrinsics.mem_copy_non_overlapping(rawptr(kp), rawptr(tk), size_of_k)
+
+			intrinsics.mem_copy_non_overlapping(rawptr(tv), rawptr(v), size_of_v)
+			intrinsics.mem_copy_non_overlapping(rawptr(v),  rawptr(vp), size_of_v)
+			intrinsics.mem_copy_non_overlapping(rawptr(vp), rawptr(tv), size_of_v)
+
+			th := h
+			h = hs[pos]
+			hs[pos] = th
+
+			distance = probe_distance
+		}
+
+		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)
+map_grow_dynamic :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, loc := #caller_location) -> Allocator_Error {
+	log2_capacity := map_log2_cap(m^)
+	new_capacity := uintptr(1) << max(log2_capacity + 1, MAP_MIN_LOG2_CAPACITY)
+	return map_reserve_dynamic(m, info, new_capacity, loc)
+}
+
+
+@(require_results)
+map_reserve_dynamic :: #force_no_inline proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, new_capacity: uintptr, loc := #caller_location) -> Allocator_Error {
+	@(require_results)
+	ceil_log2 :: #force_inline proc "contextless" (x: uintptr) -> uintptr {
+		z := intrinsics.count_leading_zeros(x)
+		if z > 0 && x & (x-1) != 0 {
+			z -= 1
+		}
+		return size_of(uintptr)*8 - 1 - z
+	}
+
+	if m.allocator.procedure == nil {
+		m.allocator = context.allocator
+	}
+
+	new_capacity := new_capacity
+	old_capacity := uintptr(map_cap(m^))
+
+	if old_capacity >= new_capacity {
+		return nil
+	}
+
+	// ceiling nearest power of two
+	log2_new_capacity := ceil_log2(new_capacity)
+
+	log2_min_cap := max(MAP_MIN_LOG2_CAPACITY, log2_new_capacity)
+
+	if m.data == 0 {
+		m^ = map_alloc_dynamic(info, log2_min_cap, m.allocator, loc) or_return
+		return nil
+	}
+
+	resized := map_alloc_dynamic(info, log2_min_cap, m.allocator, loc) or_return
+
+	ks, vs, hs, _, _ := map_kvh_data_dynamic(m^, info)
+
+	// Cache these loads to avoid hitting them in the for loop.
+	n := m.len
+	for i in 0..<old_capacity {
+		hash := hs[i]
+		if map_hash_is_empty(hash) {
+			continue
+		}
+		if map_hash_is_deleted(hash) {
+			continue
+		}
+		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.
+		n -= 1
+		if n == 0 {
+			break
+		}
+	}
+
+	map_free_dynamic(m^, info, loc) or_return
+	m.data = resized.data
+	return nil
+}
+
+
+@(require_results)
+map_shrink_dynamic :: #force_no_inline 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
+	}
+
+	// Cannot shrink the capacity if the number of items in the map would exceed
+	// one minus the current log2 capacity's resize threshold. That is the shrunk
+	// 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 false, nil
+	}
+
+	shrunk := map_alloc_dynamic(info, log2_capacity - 1, m.allocator) or_return
+
+	capacity := uintptr(1) << log2_capacity
+
+	ks, vs, hs, _, _ := map_kvh_data_dynamic(m^, info)
+
+	n := m.len
+	for i in 0..<capacity {
+		hash := hs[i]
+		if map_hash_is_empty(hash) {
+			continue
+		}
+		if map_hash_is_deleted(hash) {
+			continue
+		}
+
+		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.
+		n -= 1
+		if n == 0 {
+			break
+		}
+	}
+
+	map_free_dynamic(m^, info, loc) or_return
+	m.data = shrunk.data
+	return true, nil
+}
+
+@(require_results)
+map_free_dynamic :: #force_no_inline proc "odin" (m: Raw_Map, info: ^Map_Info, loc := #caller_location) -> Allocator_Error {
+	ptr := rawptr(map_data(m))
+	size := int(map_total_allocation_size(uintptr(map_cap(m)), info))
+	err := mem_free_with_size(ptr, size, m.allocator, loc)
+	#partial switch err {
+	case .None, .Mode_Not_Implemented:
+		return nil
+	}
+	return err
+}
+
+@(require_results)
+map_lookup_dynamic :: #force_no_inline proc "contextless" (m: Raw_Map, #no_alias info: ^Map_Info, k: uintptr) -> (index: uintptr, ok: bool) {
+	if map_len(m) == 0 {
+		return 0, false
+	}
+	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
+	ks, _, hs, _, _ := map_kvh_data_dynamic(m, info)
+	for {
+		element_hash := hs[p]
+		if map_hash_is_empty(element_hash) {
+			return 0, false
+		} else if d > map_probe_distance(m, element_hash, p) {
+			return 0, false
+		} else if element_hash == h && info.key_equal(rawptr(k), rawptr(map_cell_index_dynamic(ks, info.ks, p))) {
+			return p, true
+		}
+		p = (p + 1) & c
+		d += 1
+	}
+}
+@(require_results)
+map_exists_dynamic :: #force_no_inline proc "contextless" (m: Raw_Map, #no_alias info: ^Map_Info, k: uintptr) -> (ok: bool) {
+	if map_len(m) == 0 {
+		return false
+	}
+	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
+	ks, _, hs, _, _ := map_kvh_data_dynamic(m, info)
+	for {
+		element_hash := hs[p]
+		if map_hash_is_empty(element_hash) {
+			return false
+		} else if d > map_probe_distance(m, element_hash, p) {
+			return false
+		} else if element_hash == h && info.key_equal(rawptr(k), rawptr(map_cell_index_dynamic(ks, info.ks, p))) {
+			return true
+		}
+		p = (p + 1) & c
+		d += 1
+	}
+}
+
+
+
+@(require_results)
+map_erase_dynamic :: #force_no_inline proc "contextless" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, k: uintptr) -> (old_k, old_v: uintptr, ok: bool) {
+	index := map_lookup_dynamic(m^, info, k) or_return
+	ks, vs, hs, _, _ := map_kvh_data_dynamic(m^, info)
+	hs[index] |= TOMBSTONE_MASK
+	old_k = map_cell_index_dynamic(ks, info.ks, index)
+	old_v = map_cell_index_dynamic(vs, info.vs, index)
+	m.len -= 1
+	ok = true
+
+	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
+}
+
+map_clear_dynamic :: #force_inline proc "contextless" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info) {
+	if m.data == 0 {
+		return
+	}
+	_, _, hs, _, _ := map_kvh_data_dynamic(m^, info)
+	intrinsics.mem_zero(rawptr(hs), map_cap(m^) * size_of(Map_Hash))
+	m.len = 0
+}
+
+
+@(require_results)
+map_kvh_data_static :: #force_inline proc "contextless" (m: $T/map[$K]$V) -> (ks: [^]Map_Cell(K), vs: [^]Map_Cell(V), hs: [^]Map_Hash) {
+	capacity := uintptr(cap(m))
+	ks = ([^]Map_Cell(K))(map_data(transmute(Raw_Map)m))
+	vs = ([^]Map_Cell(V))(map_cell_index_static(ks, capacity))
+	hs = ([^]Map_Hash)(map_cell_index_static(vs, capacity))
+	return
+}
+
+
+@(require_results)
+map_get :: proc "contextless" (m: $T/map[$K]$V, key: K) -> (stored_key: K, stored_value: V, ok: bool) {
+	rm := transmute(Raw_Map)m
+	if rm.len == 0 {
+		return
+	}
+	info := intrinsics.type_map_info(T)
+	key := key
+
+	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
+	ks, vs, hs := map_kvh_data_static(m)
+	for {
+		element_hash := hs[pos]
+		if map_hash_is_empty(element_hash) {
+			return
+		} else if distance > map_probe_distance(rm, element_hash, pos) {
+			return
+		} else if element_hash == h {
+			element_key := map_cell_index_static(ks, pos)
+			if info.key_equal(&key, rawptr(element_key)) {
+				element_value := map_cell_index_static(vs, pos)
+				stored_key   = (^K)(element_key)^
+				stored_value = (^V)(element_value)^
+				ok = true
+				return
+			}
+
+		}
+		pos = (pos + 1) & mask
+		distance += 1
+	}
+}
+
+// IMPORTANT: USED WITHIN THE COMPILER
+__dynamic_map_get :: proc "contextless" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, h: Map_Hash, key: rawptr) -> (ptr: rawptr) {
+	if m.len == 0 {
+		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
+		} else if distance > map_probe_distance(m^, element_hash, pos) {
+			return nil
+		} else if element_hash == h && info.key_equal(key, rawptr(map_cell_index_dynamic(ks, info.ks, pos))) {
+			return rawptr(map_cell_index_dynamic(vs, info.vs, pos))
+		}
+		pos = (pos + 1) & mask
+		distance += 1
+	}
+}
+
+__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
+	}
+	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, map_seed(m^)), key, value, loc)
+}
+
+
+// IMPORTANT: USED WITHIN THE COMPILER
+__dynamic_map_set :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, hash: Map_Hash, key, value: rawptr, loc := #caller_location) -> rawptr {
+	if found := __dynamic_map_get(m, info, hash, key); found != nil {
+		intrinsics.mem_copy_non_overlapping(found, value, info.vs.size_of_type)
+		return found
+	}
+
+	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))
+	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)
+__dynamic_map_reserve :: proc "odin" (#no_alias m: ^Raw_Map, #no_alias info: ^Map_Info, new_capacity: uint, loc := #caller_location) -> Allocator_Error {
+	return map_reserve_dynamic(m, info, uintptr(new_capacity), loc)
+}
+
+
+
+// NOTE: the default hashing algorithm derives from fnv64a, with some minor modifications to work for `map` type:
+//
+//     * Convert a `0` result to `1`
+//         * "empty entry"
+//     * Prevent the top bit from being set
+//         * "deleted entry"
+//
+// Both of these modification are necessary for the implementation of the `map`
+
+INITIAL_HASH_SEED :: 0xcbf29ce484222325
+
+HASH_MASK :: 1 << (8*size_of(uintptr) - 1) -1
+
+default_hasher :: #force_inline proc "contextless" (data: rawptr, seed: uintptr, N: int) -> uintptr {
+	h := u64(seed) + INITIAL_HASH_SEED
+	p := ([^]byte)(data)
+	for _ in 0..<N {
+		h = (h ~ u64(p[0])) * 0x100000001b3
+		p = p[1:]
+	}
+	h &= HASH_MASK
+	return uintptr(h) | uintptr(uintptr(h) == 0)
+}
+
+default_hasher_string :: proc "contextless" (data: rawptr, seed: uintptr) -> uintptr {
+	str := (^[]byte)(data)
+	return default_hasher(raw_data(str^), seed, len(str))
+}
+default_hasher_cstring :: proc "contextless" (data: rawptr, seed: uintptr) -> uintptr {
+	h := u64(seed) + INITIAL_HASH_SEED
+	if ptr := (^[^]byte)(data)^; ptr != nil {
+		for ptr[0] != 0 {
+			h = (h ~ u64(ptr[0])) * 0x100000001b3
+			ptr = ptr[1:]
+		}
+	}
+	h &= HASH_MASK
+	return uintptr(h) | uintptr(uintptr(h) == 0)
+}
@@ -0,0 +1,62 @@
+#+private
+#+build linux, darwin, freebsd, openbsd, netbsd, 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
+		} else when ODIN_ARCH == .riscv64 {
+			@require foreign import entry "entry_unix_no_crt_riscv64.asm"
+			SYS_exit :: 93
+		}
+		@(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 $$
@@ -0,0 +1,20 @@
+	.section __TEXT,__text
+
+	; NOTE(laytan): this should ideally be the -minimum-os-version flag but there is no nice way of preprocessing assembly in Odin.
+	; 10 seems to be the lowest it goes and I don't see it mess with any targeted os version so this seems fine.
+	.build_version macos, 10, 0
+
+	.extern __start_odin
+
+	.global _main
+	.align 2
+_main:
+	mov x5, sp       ; use x5 as the stack pointer
+
+	str x0, [x5]     ; get argc into x0 (kernel passes 32-bit int argc as 64-bits on stack to keep alignment)
+	str x1, [x5, #8] ; get argv into x1
+
+	and sp, x5, #~15 ; force 16-byte alignment of the stack
+	
+	bl __start_odin  ; call into Odin entry point
+	ret              ; should never get here
@@ -0,0 +1,18 @@
+bits 32
+
+extern _start_odin
+global _start
+
+section .text
+
+;; NOTE(flysand): For description see the corresponding *_amd64.asm file
+;; also I didn't test this on x86-32
+_start:
+    xor ebp, rbp
+    pop ecx
+    mov eax, esp
+    and esp, -16
+    push eax
+    push ecx
+    call _start_odin
+    jmp $$
@@ -0,0 +1,10 @@
+.text
+
+.globl _start
+
+_start:
+	ld a0, 0(sp)
+	addi a1, sp, 8
+	addi sp, sp, ~15
+	call _start_odin
+	ebreak
@@ -0,0 +1,39 @@
+#+private
+#+build wasm32, wasm64p32
+#+no-instrumentation
+package runtime
+
+import "base:intrinsics"
+
+when !ODIN_TEST && !ODIN_NO_ENTRY_POINT {
+	when ODIN_OS == .Orca {
+		@(linkage="strong", require, export)
+		oc_on_init :: proc "c" () {
+			context = default_context()
+			#force_no_inline _startup_runtime()
+			intrinsics.__entry_point()
+		}
+		@(linkage="strong", require, export)
+		oc_on_terminate :: proc "c" () {
+			context = default_context()
+			#force_no_inline _cleanup_runtime()
+		}
+	} else {
+		@(link_name="_start", linkage="strong", require, export)
+		_start :: proc "c" () {
+			context = default_context()
+
+			when ODIN_OS == .WASI {
+				_wasi_setup_args()
+			}
+
+			#force_no_inline _startup_runtime()
+			intrinsics.__entry_point()
+		}
+		@(link_name="_end", linkage="strong", require, export)
+		_end :: proc "c" () {
+			context = default_context()
+			#force_no_inline _cleanup_runtime()
+		}
+	}
+}
@@ -1,16 +1,18 @@
-//+private
-//+build windows
+#+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
+		// Populate Windows DLL-specific globals
 		dll_forward_reason = DLL_Forward_Reason(fdwReason)
+		dll_instance       = hinstDLL

 		switch dll_forward_reason {
 		case .Process_Attach:
@@ -28,7 +30,7 @@ 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()
@@ -1,28 +1,35 @@
 package runtime

+@(no_instrumentation)
 bounds_trap :: proc "contextless" () -> ! {
 	when ODIN_OS == .Windows {
 		windows_trap_array_bounds()
+	} else when ODIN_OS == .Orca {
+		abort_ext("", "", 0, "bounds trap")
 	} else {
 		trap()
 	}
 }

+@(no_instrumentation)
 type_assertion_trap :: proc "contextless" () -> ! {
 	when ODIN_OS == .Windows {
 		windows_trap_type_assertion()
+	} else when ODIN_OS == .Orca {
+		abort_ext("", "", 0, "type assertion trap")
 	} else {
 		trap()
 	}
 }


+@(disabled=ODIN_NO_BOUNDS_CHECK)
 bounds_check_error :: proc "contextless" (file: string, line, column: i32, index, count: int) {
-	if 0 <= index && index < count {
+	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 +41,7 @@ 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 +54,7 @@ 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 +66,7 @@ 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 +74,7 @@ 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 +82,7 @@ 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 +90,13 @@ 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 +111,14 @@ 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 0 <= row_index && row_index < row_count &&
-	   0 <= column_index && column_index < column_count {
+	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 +136,13 @@ 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 +154,8 @@ 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 +167,8 @@ 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 +212,8 @@ 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 +234,13 @@ 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 +250,13 @@ 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 +268,13 @@ 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 +287,22 @@ 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 +1,119 @@
+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, old_size: int, zero_memory := true) -> ([]byte, Allocator_Error) {
+		// Not(flysand): We need to reserve enough space for alignment, which
+		// includes the user data itself, the space to store the pointer to
+		// allocation start, as well as the padding required to align both
+		// the user data and the pointer.
+		a := max(alignment, align_of(rawptr))
+		space := a-1 + size_of(rawptr) + size
+		allocated_mem: rawptr
+
+		force_copy := old_ptr != nil && alignment > align_of(rawptr)
+
+		if old_ptr != nil && !force_copy {
+			original_old_ptr := ([^]rawptr)(old_ptr)[-1]
+			allocated_mem = heap_resize(original_old_ptr, space)
+		} else {
+			allocated_mem = heap_alloc(space, zero_memory)
+		}
+		aligned_mem := rawptr(([^]u8)(allocated_mem)[size_of(rawptr):])
+
+		ptr := uintptr(aligned_mem)
+		aligned_ptr := (ptr + uintptr(a)-1) & ~(uintptr(a)-1)
+		if allocated_mem == nil {
+			aligned_free(old_ptr)
+			aligned_free(allocated_mem)
+			return nil, .Out_Of_Memory
+		}
+
+		aligned_mem = rawptr(aligned_ptr)
+		([^]rawptr)(aligned_mem)[-1] = allocated_mem
+
+		if force_copy {
+			mem_copy_non_overlapping(aligned_mem, old_ptr, min(old_size, size))
+			aligned_free(old_ptr)
+		}
+
+		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 aligned_alloc(new_size, new_alignment, nil, old_size, zero_memory)
+		}
+
+		new_memory = aligned_alloc(new_size, new_alignment, p, old_size, 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, 0, mode == .Alloc)
+
+	case .Free:
+		aligned_free(old_memory)
+
+	case .Free_All:
+		return nil, .Mode_Not_Implemented
+
+	case .Resize, .Resize_Non_Zeroed:
+		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 "contextless" (size: int, zero_memory := true) -> rawptr {
+	return _heap_alloc(size, zero_memory)
+}
+
+heap_resize :: proc "contextless" (ptr: rawptr, new_size: int) -> rawptr {
+	return _heap_resize(ptr, new_size)
+}
+
+heap_free :: proc "contextless" (ptr: rawptr) {
+	_heap_free(ptr)
+}
@@ -0,0 +1,29 @@
+#+build orca
+#+private
+package runtime
+
+foreign {
+	@(link_name="malloc")   _orca_malloc   :: proc "c" (size: int) -> rawptr ---
+	@(link_name="calloc")   _orca_calloc   :: proc "c" (num, size: int) -> rawptr ---
+	@(link_name="free")     _orca_free     :: proc "c" (ptr: rawptr) ---
+	@(link_name="realloc")  _orca_realloc  :: proc "c" (ptr: rawptr, size: int) -> rawptr ---
+}
+
+_heap_alloc :: proc "contextless" (size: int, zero_memory := true) -> rawptr {
+	if size <= 0 {
+		return nil
+	}
+	if zero_memory {
+		return _orca_calloc(1, size)
+	} else {
+		return _orca_malloc(size)
+	}
+}
+
+_heap_resize :: proc "contextless" (ptr: rawptr, new_size: int) -> rawptr {
+	return _orca_realloc(ptr, new_size)
+}
+
+_heap_free :: proc "contextless" (ptr: rawptr) {
+	_orca_free(ptr)
+}
@@ -0,0 +1,18 @@
+#+build js, wasi, freestanding, essence
+#+private
+package runtime
+
+_heap_alloc :: proc "contextless" (size: int, zero_memory := true) -> rawptr {
+	context = default_context()
+	unimplemented("base:runtime 'heap_alloc' procedure is not supported on this platform")
+}
+
+_heap_resize :: proc "contextless" (ptr: rawptr, new_size: int) -> rawptr {
+	context = default_context()
+	unimplemented("base:runtime 'heap_resize' procedure is not supported on this platform")
+}
+
+_heap_free :: proc "contextless" (ptr: rawptr) {
+	context = default_context()
+	unimplemented("base:runtime 'heap_free' procedure is not supported on this platform")
+}
@@ -0,0 +1,38 @@
+#+build linux, darwin, freebsd, openbsd, netbsd, 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 "contextless" (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 "contextless" (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 "contextless" (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 "contextless" (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 "contextless" (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 "contextless" (ptr: rawptr) {
+	if ptr == nil {
+		return
+	}
+	HeapFree(GetProcessHeap(), 0, ptr)
+}
+
@@ -1,17 +1,20 @@
+#+vet !cast
 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 (
-	(ODIN_USE_SEPARATE_MODULES || 
+	ODIN_USE_SEPARATE_MODULES || 
 	ODIN_BUILD_MODE == .Dynamic ||
-	!ODIN_NO_CRT) &&
-	!IS_WASM) else "internal"
-RUNTIME_REQUIRE :: true
+	!ODIN_NO_CRT) else "internal"
+RUNTIME_REQUIRE :: false // !ODIN_TILDE
+
+@(private)
+__float16 :: f16 when __ODIN_LLVM_F16_SUPPORTED else u16


@(private)
@@ -19,58 +22,14 @@ 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
 	}
 	return (x & (x-1)) == 0
 }

-align_forward_int :: #force_inline proc(ptr, align: int) -> int {
+align_forward_int :: #force_inline proc "odin" (ptr, align: int) -> int {
 	assert(is_power_of_two_int(align))

 	p := ptr
@@ -81,14 +40,32 @@ 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_uint :: #force_inline proc "contextless" (x: uint) -> bool {
 	if x <= 0 {
 		return false
 	}
 	return (x & (x-1)) == 0
 }

-align_forward_uintptr :: #force_inline proc(ptr, align: uintptr) -> uintptr {
+align_forward_uint :: #force_inline proc "odin" (ptr, align: uint) -> uint {
+	assert(is_power_of_two_uint(align))
+
+	p := ptr
+	modulo := p & (align-1)
+	if modulo != 0 {
+		p += align - modulo
+	}
+	return p
+}
+
+is_power_of_two_uintptr :: #force_inline proc "contextless" (x: uintptr) -> bool {
+	if x <= 0 {
+		return false
+	}
+	return (x & (x-1)) == 0
+}
+
+align_forward_uintptr :: #force_inline proc "odin" (ptr, align: uintptr) -> uintptr {
 	assert(is_power_of_two_uintptr(align))

 	p := ptr
@@ -99,11 +76,23 @@ align_forward_uintptr :: #force_inline proc(ptr, align: uintptr) -> uintptr {
 	return p
 }

+is_power_of_two :: proc {
+	is_power_of_two_int,
+	is_power_of_two_uint,
+	is_power_of_two_uintptr,
+}
+
+align_forward :: proc {
+	align_forward_int,
+	align_forward_uint,
+	align_forward_uintptr,
+}
+
 mem_zero :: proc "contextless" (data: rawptr, len: int) -> rawptr {
 	if data == nil {
 		return nil
 	}
-	if len < 0 {
+	if len <= 0 {
 		return data
 	}
 	intrinsics.mem_zero(data, len)
@@ -111,59 +100,72 @@ mem_zero :: proc "contextless" (data: rawptr, len: int) -> rawptr {
 }

 mem_copy :: proc "contextless" (dst, src: rawptr, len: int) -> rawptr {
-	if src == nil {
-		return dst
+	if src != nil && dst != src && len > 0 {
+		// NOTE(bill): This _must_ be implemented like C's memmove
+		intrinsics.mem_copy(dst, src, len)
 	}
-
-	// NOTE(bill): This _must_ be implemented like C's memmove
-	intrinsics.mem_copy(dst, src, len)
 	return dst
 }

 mem_copy_non_overlapping :: proc "contextless" (dst, src: rawptr, len: int) -> rawptr {
-	if src == nil {
-		return dst
+	if src != nil && dst != src && len > 0 {
+		// NOTE(bill): This _must_ be implemented like C's memcpy
+		intrinsics.mem_copy_non_overlapping(dst, src, len)
 	}
-
-	// NOTE(bill): This _must_ be implemented like C's memcpy
-	intrinsics.mem_copy_non_overlapping(dst, src, len)
 	return dst
 }

 DEFAULT_ALIGNMENT :: 2*align_of(rawptr)

 mem_alloc_bytes :: #force_inline proc(size: int, alignment: int = DEFAULT_ALIGNMENT, allocator := context.allocator, loc := #caller_location) -> ([]byte, Allocator_Error) {
-	if size == 0 {
-		return nil, nil
-	}
-	if allocator.procedure == nil {
+	assert(is_power_of_two_int(alignment), "Alignment must be a power of two", loc)
+	if size == 0 || allocator.procedure == nil{
 		return nil, nil
 	}
 	return allocator.procedure(allocator.data, .Alloc, size, alignment, nil, 0, loc)
 }

-mem_alloc :: #force_inline proc(size: int, alignment: int = DEFAULT_ALIGNMENT, allocator := context.allocator, loc := #caller_location) -> (rawptr, Allocator_Error) {
-	if size == 0 {
+mem_alloc :: #force_inline proc(size: int, alignment: int = DEFAULT_ALIGNMENT, allocator := context.allocator, loc := #caller_location) -> ([]byte, Allocator_Error) {
+	assert(is_power_of_two_int(alignment), "Alignment must be a power of two", loc)
+	if size == 0 || allocator.procedure == nil {
 		return nil, nil
 	}
-	if allocator.procedure == nil {
+	return allocator.procedure(allocator.data, .Alloc, size, alignment, nil, 0, loc)
+}
+
+mem_alloc_non_zeroed :: #force_inline proc(size: int, alignment: int = DEFAULT_ALIGNMENT, allocator := context.allocator, loc := #caller_location) -> ([]byte, Allocator_Error) {
+	assert(is_power_of_two_int(alignment), "Alignment must be a power of two", loc)
+	if size == 0 || allocator.procedure == nil {
 		return nil, nil
 	}
-	data, err := allocator.procedure(allocator.data, .Alloc, size, alignment, nil, 0, loc)
-	return raw_data(data), err
+	return allocator.procedure(allocator.data, .Alloc_Non_Zeroed, size, alignment, nil, 0, loc)
 }

 mem_free :: #force_inline proc(ptr: rawptr, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
-	if ptr == nil {
-		return .None
-	}
-	if allocator.procedure == nil {
-		return .None
+	if ptr == nil || allocator.procedure == nil {
+		return nil
 	}
 	_, err := allocator.procedure(allocator.data, .Free, 0, 0, ptr, 0, loc)
 	return err
 }

+mem_free_with_size :: #force_inline proc(ptr: rawptr, byte_count: int, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
+	if ptr == nil || allocator.procedure == nil {
+		return nil
+	}
+	_, err := allocator.procedure(allocator.data, .Free, 0, 0, ptr, byte_count, loc)
+	return err
+}
+
+mem_free_bytes :: #force_inline proc(bytes: []byte, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
+	if bytes == nil || allocator.procedure == nil {
+		return nil
+	}
+	_, err := allocator.procedure(allocator.data, .Free, 0, 0, raw_data(bytes), len(bytes), loc)
+	return err
+}
+
+
 mem_free_all :: #force_inline proc(allocator := context.allocator, loc := #caller_location) -> (err: Allocator_Error) {
 	if allocator.procedure != nil {
 		_, err = allocator.procedure(allocator.data, .Free_All, 0, 0, nil, 0, loc)
@@ -171,33 +173,77 @@ mem_free_all :: #force_inline proc(allocator := context.allocator, loc := #calle
 	return
 }

-mem_resize :: #force_inline proc(ptr: rawptr, old_size, new_size: int, alignment: int = DEFAULT_ALIGNMENT, allocator := context.allocator, loc := #caller_location) -> (new_ptr: rawptr, err: Allocator_Error) {
-	new_data: []byte
-	switch {
-	case allocator.procedure == nil:
-		return
-	case new_size == 0:
-		new_data, err = allocator.procedure(allocator.data, .Free, 0, 0, ptr, 0, loc)
-	case ptr == nil:
-		new_data, err = allocator.procedure(allocator.data, .Alloc, new_size, alignment, nil, 0, loc)
-	case:
-		new_data, err = allocator.procedure(allocator.data, .Resize, new_size, alignment, ptr, old_size, loc)
+_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) {
+	assert(is_power_of_two_int(alignment), "Alignment must be a power of two", loc)
+	if allocator.procedure == nil {
+		return nil, nil
+	}
+	if new_size == 0 {
+		if ptr != nil {
+			_, err = allocator.procedure(allocator.data, .Free, 0, 0, ptr, old_size, loc)
+			return
+		}
+		return
+	} else if ptr == nil {
+		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
+	}
+
+	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 {
+		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
+		}
+		copy(data, ([^]byte)(ptr)[:old_size])
+		_, err = allocator.procedure(allocator.data, .Free, 0, 0, ptr, old_size, loc)
 	}
-	new_ptr = raw_data(new_data)
 	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) {
+	assert(is_power_of_two_int(alignment), "Alignment must be a power of two", loc)
+	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) {
+	assert(is_power_of_two_int(alignment), "Alignment must be a power of two", loc)
+	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 i in 0..<word_length {
+			for _ in 0..<word_length {
 				if intrinsics.unaligned_load((^u64)(a)) != intrinsics.unaligned_load((^u64)(b)) {
 					return false
 				}
@@ -228,7 +274,7 @@ memory_equal :: proc "contextless" (x, y: rawptr, n: int) -> bool {
 		return true
 	} else {
 		if word_length := length >> 2; word_length != 0 {
-			for i in 0..<word_length {
+			for _ in 0..<word_length {
 				if intrinsics.unaligned_load((^u32)(a)) != intrinsics.unaligned_load((^u32)(b)) {
 					return false
 				}
@@ -249,6 +295,7 @@ 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 {
@@ -341,7 +388,12 @@ string_eq :: proc "contextless" (lhs, rhs: string) -> bool {
 string_cmp :: proc "contextless" (a, b: string) -> int {
 	x := transmute(Raw_String)a
 	y := transmute(Raw_String)b
-	return memory_compare(x.data, y.data, min(x.len, y.len))
+
+	ret := memory_compare(x.data, y.data, min(x.len, y.len))
+	if ret == 0 && x.len != y.len {
+		return -1 if x.len < y.len else +1
+	}
+	return ret
 }

 string_ne :: #force_inline proc "contextless" (a, b: string) -> bool { return !string_eq(a, b) }
@@ -369,6 +421,48 @@ 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) }

@@ -392,7 +486,7 @@ quaternion256_ne :: #force_inline proc "contextless" (a, b: quaternion256) -> bo
 string_decode_rune :: #force_inline proc "contextless" (s: string) -> (rune, int) {
 	// NOTE(bill): Duplicated here to remove dependency on package unicode/utf8

-	@static accept_sizes := [256]u8{
+	@(static, rodata) accept_sizes := [256]u8{
 		0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, // 0x00-0x0f
 		0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, // 0x10-0x1f
 		0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, // 0x20-0x2f
@@ -413,7 +507,7 @@ string_decode_rune :: #force_inline proc "contextless" (s: string) -> (rune, int
 	}
 	Accept_Range :: struct {lo, hi: u8}

-	@static accept_ranges := [5]Accept_Range{
+	@(static, rodata) accept_ranges := [5]Accept_Range{
 		{0x80, 0xbf},
 		{0xa0, 0xbf},
 		{0x80, 0x9f},
@@ -468,46 +562,73 @@ 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)
@@ -524,21 +645,24 @@ abs_quaternion256 :: #force_inline proc "contextless" (x: quaternion256) -> f64


 quo_complex32 :: proc "contextless" (n, m: complex32) -> complex32 {
-	e, f: f16
+	nr, ni := f32(real(n)), f32(imag(n))
+	mr, mi := f32(real(m)), f32(imag(m))

-	if abs(real(m)) >= abs(imag(m)) {
-		ratio := imag(m) / real(m)
-		denom := real(m) + ratio*imag(m)
-		e = (real(n) + imag(n)*ratio) / denom
-		f = (imag(n) - real(n)*ratio) / denom
+	e, f: f32
+
+	if abs(mr) >= abs(mi) {
+		ratio := mi / mr
+		denom := mr + ratio*mi
+		e = (nr + ni*ratio) / denom
+		f = (ni - nr*ratio) / denom
 	} else {
-		ratio := real(m) / imag(m)
-		denom := imag(m) + ratio*real(m)
-		e = (real(n)*ratio + imag(n)) / denom
-		f = (imag(n)*ratio - real(n)) / denom
+		ratio := mr / mi
+		denom := mi + ratio*mr
+		e = (nr*ratio + ni) / denom
+		f = (ni*ratio - nr) / denom
 	}

-	return complex(e, f)
+	return complex(f16(e), f16(f))
 }


@@ -579,15 +703,15 @@ quo_complex128 :: proc "contextless" (n, m: complex128) -> complex128 {
 }

 mul_quaternion64 :: proc "contextless" (q, r: quaternion64) -> quaternion64 {
-	q0, q1, q2, q3 := real(q), imag(q), jmag(q), kmag(q)
-	r0, r1, r2, r3 := real(r), imag(r), jmag(r), kmag(r)
+	q0, q1, q2, q3 := f32(real(q)), f32(imag(q)), f32(jmag(q)), f32(kmag(q))
+	r0, r1, r2, r3 := f32(real(r)), f32(imag(r)), f32(jmag(r)), f32(kmag(r))

 	t0 := r0*q0 - r1*q1 - r2*q2 - r3*q3
 	t1 := r0*q1 + r1*q0 - r2*q3 + r3*q2
 	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=f16(t0), x=f16(t1), y=f16(t2), z=f16(t3))
 }

 mul_quaternion128 :: proc "contextless" (q, r: quaternion128) -> quaternion128 {
@@ -599,7 +723,7 @@ 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 {
@@ -611,12 +735,12 @@ 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 {
-	q0, q1, q2, q3 := real(q), imag(q), jmag(q), kmag(q)
-	r0, r1, r2, r3 := real(r), imag(r), jmag(r), kmag(r)
+	q0, q1, q2, q3 := f32(real(q)), f32(imag(q)), f32(jmag(q)), f32(kmag(q))
+	r0, r1, r2, r3 := f32(real(r)), f32(imag(r)), f32(jmag(r)), f32(kmag(r))

 	invmag2 := 1.0 / (r0*r0 + r1*r1 + r2*r2 + r3*r3)

@@ -625,7 +749,7 @@ 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=f16(t0), x=f16(t1), y=f16(t2), z=f16(t3))
 }

 quo_quaternion128 :: proc "contextless" (q, r: quaternion128) -> quaternion128 {
@@ -639,7 +763,7 @@ 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 {
@@ -653,11 +777,11 @@ 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

@@ -671,7 +795,7 @@ 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)

@@ -679,14 +803,14 @@ 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 {
@@ -706,23 +830,28 @@ 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="__aeabi_d2h", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
+aeabi_d2h :: proc "c" (value: f64) -> __float16 {
+	return truncsfhf2(f32(value))
+}

@(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)
@@ -739,12 +868,12 @@ 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)
 }

@@ -752,9 +881,6 @@ extendhfsf2 :: proc "c" (value: u16) -> f32 {

@(link_name="__floattidf", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
 floattidf :: proc "c" (a: i128) -> f64 {
-when IS_WASM {
-	return 0
-} else {
 	DBL_MANT_DIG :: 53
 	if a == 0 {
 		return 0.0
@@ -794,14 +920,10 @@ when IS_WASM {
 	fb[0] = u32(a)                           // mantissa-low
 	return transmute(f64)fb
 }
-}


@(link_name="__floattidf_unsigned", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
 floattidf_unsigned :: proc "c" (a: u128) -> f64 {
-when IS_WASM {
-	return 0
-} else {
 	DBL_MANT_DIG :: 53
 	if a == 0 {
 		return 0.0
@@ -839,7 +961,6 @@ when IS_WASM {
 	fb[0] = u32(a)                           // mantissa-low
 	return transmute(f64)fb
 }
-}



@@ -873,9 +994,11 @@ 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)
+	}
 }


@@ -887,26 +1010,44 @@ modti3 :: proc "c" (a, b: i128) -> i128 {
 	bn := (b ~ s_b) - s_b

 	r: u128 = ---
-	_ = udivmod128(transmute(u128)an, transmute(u128)bn, &r)
-	return (transmute(i128)r ~ s_a) - s_a
+	_ = udivmod128(u128(an), u128(bn), &r)
+	return (i128(r) ~ s_a) - s_a
 }


@(link_name="__divmodti4", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
 divmodti4 :: proc "c" (a, b: i128, rem: ^i128) -> i128 {
-	u := udivmod128(transmute(u128)a, transmute(u128)b, cast(^u128)rem)
-	return transmute(i128)u
+	s_a := a >> (128 - 1) // -1 if negative or 0
+	s_b := b >> (128 - 1)
+	an := (a ~ s_a) - s_a // absolute
+	bn := (b ~ s_b) - s_b
+
+	s_b   ~= s_a // quotient sign
+	u_s_b := u128(s_b)
+	u_s_a := u128(s_a)
+
+	r: u128 = ---
+	u := i128((udivmodti4(u128(an), u128(bn), &r) ~ u_s_b) - u_s_b) // negate if negative
+	rem^ = i128((r ~ u_s_a) - u_s_a)
+	return u
 }

@(link_name="__divti3", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
 divti3 :: proc "c" (a, b: i128) -> i128 {
-	u := udivmodti4(transmute(u128)a, transmute(u128)b, nil)
-	return transmute(i128)u
+	s_a := a >> (128 - 1) // -1 if negative or 0
+	s_b := b >> (128 - 1)
+	an := (a ~ s_a) - s_a // absolute
+	bn := (b ~ s_b) - s_b
+
+	s_a   ~= s_b // quotient sign
+	u_s_a := u128(s_a)
+
+	return i128((udivmodti4(u128(an), u128(bn), nil) ~ u_s_a) - u_s_a) // negate if negative
 }


@(link_name="__fixdfti", linkage=RUNTIME_LINKAGE, require=RUNTIME_REQUIRE)
-fixdfti :: proc(a: u64) -> i128 {
+fixdfti :: proc "c" (a: u64) -> i128 {
 	significandBits :: 52
 	typeWidth       :: (size_of(u64)*8)
 	exponentBits    :: (typeWidth - significandBits - 1)
@@ -945,3 +1086,23 @@ 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 +1,26 @@
+#+build freebsd, openbsd, netbsd
+#+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 ---
+
+	when ODIN_OS == .NetBSD {
+		@(link_name="__errno") __error :: proc() -> ^i32 ---
+	} else {
+		__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 +1,28 @@
+#+build darwin
+#+private
+package runtime
+
+import "base:intrinsics"
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	STDERR :: 2
+	when ODIN_NO_CRT {
+		WRITE  :: 0x2000004
+		ret := intrinsics.syscall(WRITE, STDERR, uintptr(raw_data(data)), uintptr(len(data)))
+		if ret < 0 {
+			return 0, _OS_Errno(-ret)
+		}
+		return int(ret), 0
+	} else {
+		foreign {
+			write   :: proc(handle: i32, buffer: [^]byte, count: uint) -> int ---
+			__error :: proc() -> ^i32 ---
+		}
+
+		if ret := write(STDERR, raw_data(data), len(data)); ret >= 0 {
+			return int(ret), 0
+		}
+
+		return 0, _OS_Errno(__error()^)
+	}
+}
@@ -0,0 +1,8 @@
+#+build freestanding
+#+private
+package runtime
+
+// TODO(bill): reimplement `os.write`
+_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 +1,26 @@
+#+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)
+	} else when ODIN_ARCH == .riscv64 {
+		SYS_write :: uintptr(64)
+	}
+
+	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
+}
@@ -0,0 +1,43 @@
+#+build orca
+#+private
+package runtime
+
+import "base:intrinsics"
+
+// Constants allowing to specify the level of logging verbosity.
+log_level :: enum u32 {
+	// Only errors are logged.
+	ERROR = 0,
+	// Only warnings and errors are logged.
+	WARNING = 1,
+	// All messages are logged.
+	INFO = 2,
+	COUNT = 3,
+}
+
+@(default_calling_convention="c", link_prefix="oc_")
+foreign {
+	abort_ext   :: proc(file: cstring, function: cstring, line: i32, fmt: cstring, #c_vararg args: ..any) -> ! ---
+	assert_fail :: proc(file: cstring, function: cstring, line: i32, src: cstring, fmt: cstring, #c_vararg args: ..any) -> ! ---
+	log_ext     :: proc(level: log_level, function: cstring, file: cstring, line: i32, fmt: cstring, #c_vararg args: ..any) ---
+}
+
+// NOTE: This is all pretty gross, don't look.
+
+// WASM is single threaded so this should be fine.
+orca_stderr_buffer:     [4096]byte
+orca_stderr_buffer_idx: int
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	for b in data {
+		orca_stderr_buffer[orca_stderr_buffer_idx] = b
+		orca_stderr_buffer_idx += 1
+
+		if b == '\n' || orca_stderr_buffer_idx == len(orca_stderr_buffer)-1 {
+			log_ext(.ERROR, "", "", 0, cstring(raw_data(orca_stderr_buffer[:orca_stderr_buffer_idx])))
+			orca_stderr_buffer_idx = 0
+		}
+	}
+
+	return len(data), 0
+}
@@ -0,0 +1,55 @@
+#+build wasi
+#+private
+package runtime
+
+foreign import wasi "wasi_snapshot_preview1"
+
+@(default_calling_convention="contextless")
+foreign wasi {
+	fd_write :: proc(
+		fd: i32,
+		iovs: [][]byte,
+		n: ^uint,
+	) -> u16 ---
+
+	@(private="file")
+	args_sizes_get :: proc(
+		num_of_args:  ^uint,
+		size_of_args: ^uint,
+	) -> u16 ---
+
+	@(private="file")
+	args_get :: proc(
+		argv:     [^]cstring,
+		argv_buf: [^]byte,
+	) -> u16 ---
+}
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	n: uint
+	err := fd_write(1, {data}, &n)
+	return int(n), _OS_Errno(err)
+}
+
+_wasi_setup_args :: proc() {
+	num_of_args, size_of_args: uint
+	if errno := args_sizes_get(&num_of_args, &size_of_args); errno != 0 {
+		return
+	}
+
+	err: Allocator_Error
+	if args__, err = make([]cstring, num_of_args); err != nil {
+		return
+	}
+
+	args_buf: []byte
+	if args_buf, err = make([]byte, size_of_args); err != nil {
+		delete(args__)
+		return
+	}
+
+	if errno := args_get(raw_data(args__), raw_data(args_buf)); errno != 0 {
+		delete(args__)
+		delete(args_buf)
+	}
+}
@@ -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
+}
@@ -2,6 +2,90 @@ package runtime

 _INTEGER_DIGITS :: "0123456789abcdefghijklmnopqrstuvwxyz"

+@(private="file")
+_INTEGER_DIGITS_VAR := _INTEGER_DIGITS
+
+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
+		}
+		x.id = typeid_base(x.id)
+		switch v in x {
+		case typeid:     print_typeid(v)
+		case ^Type_Info: print_type(v)
+
+		case string:  print_string(v)
+		case cstring: print_string(string(v))
+		case []byte:  print_string(string(v))
+
+		case rune:  print_rune(v)
+
+		case u8:    print_u64(u64(v))
+		case u16:   print_u64(u64(v))
+		case u16le: print_u64(u64(v))
+		case u16be: print_u64(u64(v))
+		case u32:   print_u64(u64(v))
+		case u32le: print_u64(u64(v))
+		case u32be: print_u64(u64(v))
+		case u64:   print_u64(u64(v))
+		case u64le: print_u64(u64(v))
+		case u64be: print_u64(u64(v))
+
+		case i8:    print_i64(i64(v))
+		case i16:   print_i64(i64(v))
+		case i16le: print_i64(i64(v))
+		case i16be: print_i64(i64(v))
+		case i32:   print_i64(i64(v))
+		case i32le: print_i64(i64(v))
+		case i32be: print_i64(i64(v))
+		case i64:   print_i64(i64(v))
+		case i64le: print_i64(i64(v))
+		case i64be: print_i64(i64(v))
+
+		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")
+		case b16:  print_string("true" if v else "false")
+		case b32:  print_string("true" if v else "false")
+		case b64:  print_string("true" if v else "false")
+
+		case:
+			ti := type_info_of(x.id)
+			#partial switch v in ti.variant {
+			case Type_Info_Pointer, Type_Info_Multi_Pointer:
+				print_uintptr((^uintptr)(x.data)^)
+				return
+			}
+
+			print_string("<invalid-value>")
+		}
+	}
+	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(" ")
+			}
+			print_any_single(arg)
+		}
+		print_string("\n")
+	}
+}
+
+
 encode_rune :: proc "contextless" (c: rune) -> ([4]u8, int) {
 	r := c

@@ -38,14 +122,14 @@ encode_rune :: proc "contextless" (c: rune) -> ([4]u8, int) {
 	return buf, 4
 }

-print_string :: proc "contextless" (str: string) -> (int, _OS_Errno) {
-	return 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, err: _OS_Errno) {
+print_strings :: #force_no_inline proc "contextless" (args: ..string) -> (n: int) {
 	for str in args {
-		m: int
-		m, err = os_write(transmute([]byte)str)
+		m, err := stderr_write(transmute([]byte)str)
 		n += m
 		if err != 0 {
 			break
@@ -54,11 +138,12 @@ print_strings :: proc "contextless" (args: ..string) -> (n: int, err: _OS_Errno)
 	return
 }

-print_byte :: proc "contextless" (b: byte) -> (int, _OS_Errno) {
-	return 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 {
@@ -74,11 +159,10 @@ print_encoded_rune :: proc "contextless" (r: rune) {
 		if r <= 0 {
 			print_string("\\x00")
 		} else if r < 32 {
-			digits := _INTEGER_DIGITS
 			n0, n1 := u8(r) >> 4, u8(r) & 0xf
 			print_string("\\x")
-			print_byte(digits[n0])
-			print_byte(digits[n1])
+			print_byte(_INTEGER_DIGITS_VAR[n0])
+			print_byte(_INTEGER_DIGITS_VAR[n1])
 		} else {
 			print_rune(r)
 		}
@@ -86,7 +170,7 @@ print_encoded_rune :: proc "contextless" (r: rune) {
 	print_byte('\'')
 }

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

 	if r < RUNE_SELF {
@@ -94,29 +178,27 @@ print_rune :: proc "contextless" (r: rune) -> (int, _OS_Errno) #no_bounds_check
 	}

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


-print_u64 :: proc "contextless" (x: u64) #no_bounds_check {
-	digits := _INTEGER_DIGITS
-
+print_u64 :: #force_no_inline proc "contextless" (x: u64) #no_bounds_check {
 	a: [129]byte
 	i := len(a)
 	b := u64(10)
 	u := x
 	for u >= b {
-		i -= 1; a[i] = digits[u % b]
+		i -= 1; a[i] = _INTEGER_DIGITS_VAR[u % b]
 		u /= b
 	}
-	i -= 1; a[i] = digits[u % b]
+	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 {
-	digits := _INTEGER_DIGITS
+print_i64 :: #force_no_inline proc "contextless" (x: i64) #no_bounds_check {
 	b :: i64(10)

 	u := x
@@ -126,41 +208,45 @@ print_i64 :: proc "contextless" (x: i64) #no_bounds_check {
 	a: [129]byte
 	i := len(a)
 	for u >= b {
-		i -= 1; a[i] = digits[u % b]
+		i -= 1; a[i] = _INTEGER_DIGITS_VAR[u % b]
 		u /= b
 	}
-	i -= 1; a[i] = digits[u % b]
+	i -= 1; a[i] = _INTEGER_DIGITS_VAR[u % b]
 	if neg {
 		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 {
@@ -175,7 +261,9 @@ print_typeid :: proc "contextless" (id: typeid) {
 		}
 	}
 }
-print_type :: proc "contextless" (ti: ^Type_Info) {
+
+@(optimization_mode="favor_size")
+print_type :: #force_no_inline proc "contextless" (ti: ^Type_Info) {
 	if ti == nil {
 		print_string("nil")
 		return
@@ -228,12 +316,15 @@ print_type :: proc "contextless" (ti: ^Type_Info) {
 	case Type_Info_Multi_Pointer:
 		print_string("[^]")
 		print_type(info.elem)
+	case Type_Info_Soa_Pointer:
+		print_string("#soa ^")
+		print_type(info.elem)
 	case Type_Info_Procedure:
 		print_string("proc")
 		if info.params == nil {
 			print_string("()")
 		} else {
-			t := info.params.variant.(Type_Info_Tuple)
+			t := info.params.variant.(Type_Info_Parameters)
 			print_byte('(')
 			for t, i in t.types {
 				if i > 0 { print_string(", ") }
@@ -245,7 +336,7 @@ print_type :: proc "contextless" (ti: ^Type_Info) {
 			print_string(" -> ")
 			print_type(info.results)
 		}
-	case Type_Info_Tuple:
+	case Type_Info_Parameters:
 		count := len(info.names)
 		if count != 1 { print_byte('(') }
 		for name, i in info.names {
@@ -310,15 +401,16 @@ print_type :: proc "contextless" (ti: ^Type_Info) {
 		}

 		print_string("struct ")
-		if info.is_packed    { print_string("#packed ") }
-		if info.is_raw_union { print_string("#raw_union ") }
-		if info.custom_align {
-			print_string("#align ")
+		if .packed    in info.flags { print_string("#packed ") }
+		if .raw_union in info.flags { print_string("#raw_union ") }
+		if .no_copy   in info.flags { print_string("#no_copy ") }
+		if .align in info.flags {
+			print_string("#align(")
 			print_u64(u64(ti.align))
-			print_byte(' ')
+			print_string(") ")
 		}
 		print_byte('{')
-		for name, i in info.names {
+		for name, i in info.names[:info.field_count] {
 			if i > 0 { print_string(", ") }
 			print_string(name)
 			print_string(": ")
@@ -329,8 +421,9 @@ 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 ")
@@ -373,6 +466,20 @@ 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[:info.field_count] {
+			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[")
@@ -386,11 +493,11 @@ 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[")
@@ -0,0 +1,101 @@
+package runtime
+
+when ODIN_NO_CRT && ODIN_OS == .Windows {
+	foreign import lib "system:NtDll.lib"
+	
+	@(private="file")
+	@(default_calling_convention="system")
+	foreign lib {
+		RtlMoveMemory :: proc(dst, s: rawptr, length: int) ---
+		RtlFillMemory :: proc(dst: rawptr, length: int, fill: i32) ---
+	}
+	
+	@(link_name="memset", linkage="strong", require)
+	memset :: proc "c" (ptr: rawptr, val: i32, len: int) -> rawptr {
+		RtlFillMemory(ptr, len, val)
+		return ptr
+	}
+	@(link_name="memmove", linkage="strong", require)
+	memmove :: proc "c" (dst, src: rawptr, len: int) -> rawptr {
+		RtlMoveMemory(dst, src, len)
+		return dst
+	}
+	@(link_name="memcpy", linkage="strong", require)
+	memcpy :: proc "c" (dst, src: rawptr, len: int) -> rawptr {
+		RtlMoveMemory(dst, src, len)
+		return dst
+	}
+} else when ODIN_NO_CRT || (ODIN_OS != .Orca && (ODIN_ARCH == .wasm32 || ODIN_ARCH == .wasm64p32)) {
+	// NOTE: on wasm, calls to these procs are generated (by LLVM) with type `i32` instead of `int`.
+	//
+	// NOTE: `#any_int` is also needed, because calls that we generate (and package code)
+	//       will be using `int` and need to be converted.
+	int_t :: i32 when ODIN_ARCH == .wasm64p32 else int
+
+	@(link_name="memset", linkage="strong", require)
+	memset :: proc "c" (ptr: rawptr, val: i32, #any_int len: int_t) -> rawptr {
+		if ptr != nil && len != 0 {
+			b := byte(val)
+			p := ([^]byte)(ptr)
+			for i := int_t(0); i < len; i += 1 {
+				p[i] = b
+			}
+		}
+		return ptr
+	}
+
+	@(link_name="bzero", linkage="strong", require)
+	bzero :: proc "c" (ptr: rawptr, #any_int len: int_t) -> rawptr {
+		if ptr != nil && len != 0 {
+			p := ([^]byte)(ptr)
+			for i := int_t(0); i < len; i += 1 {
+				p[i] = 0
+			}
+		}
+		return ptr
+	}
+
+	@(link_name="memmove", linkage="strong", require)
+	memmove :: proc "c" (dst, src: rawptr, #any_int len: int_t) -> rawptr {
+		d, s := ([^]byte)(dst), ([^]byte)(src)
+		if d == s || len == 0 {
+			return dst
+		}
+		if d > s && uintptr(d)-uintptr(s) < uintptr(len) {
+			for i := len-1; i >= 0; i -= 1 {
+				d[i] = s[i]
+			}
+			return dst
+		}
+
+		if s > d && uintptr(s)-uintptr(d) < uintptr(len) {
+			for i := int_t(0); i < len; i += 1 {
+				d[i] = s[i]
+			}
+			return dst
+		}
+		return memcpy(dst, src, len)
+	}
+	@(link_name="memcpy", linkage="strong", require)
+	memcpy :: proc "c" (dst, src: rawptr, #any_int len: int_t) -> rawptr {
+		d, s := ([^]byte)(dst), ([^]byte)(src)
+		if d != s {
+			for i := int_t(0); i < len; i += 1 {
+				d[i] = s[i]
+			}
+		}
+		return d
+		
+	}
+} else {
+	memset :: proc "c" (ptr: rawptr, val: i32, len: int) -> rawptr {
+		if ptr != nil && len != 0 {
+			b := byte(val)
+			p := ([^]byte)(ptr)
+			for i := 0; i < len; i += 1 {
+				p[i] = b
+			}
+		}
+		return ptr
+	}
+}
@@ -1,9 +1,9 @@
-//+private
+#+private
 package runtime

 foreign import "system:Foundation.framework"

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

 objc_id :: ^intrinsics.objc_object
 objc_Class :: ^intrinsics.objc_class
@@ -12,7 +12,7 @@ objc_SEL :: ^intrinsics.objc_selector
 foreign Foundation {
 	objc_lookUpClass :: proc "c" (name: cstring) -> objc_Class ---
 	sel_registerName :: proc "c" (name: cstring) -> objc_SEL ---
-	objc_allocateClassPair :: proc "c" (superclass: objc_Class, name: cstring, extraBytes: uint) ---
+	objc_allocateClassPair :: proc "c" (superclass: objc_Class, name: cstring, extraBytes: uint) -> objc_Class ---

 	objc_msgSend        :: proc "c" (self: objc_id, op: objc_SEL, #c_vararg args: ..any) ---
 	objc_msgSend_fpret  :: proc "c" (self: objc_id, op: objc_SEL, #c_vararg args: ..any) -> f64 ---
@@ -1,4 +1,4 @@
-//+build js wasm32
+#+build js
 package runtime

 init_default_context_for_js: Context
@@ -0,0 +1,97 @@
+#+build wasm32, wasm64p32
+package runtime
+
+@(private="file")
+ti_int :: struct #raw_union {
+	using s: struct { lo, hi: u64 },
+	all: i128,
+}
+
+@(private="file")
+ti_uint :: struct #raw_union {
+	using s: struct { lo, hi: u64 },
+	all: u128,
+}
+
+@(link_name="__ashlti3", linkage="strong")
+__ashlti3 :: proc "contextless" (a: i128, b: u32) -> i128 {
+	bits :: 64
+	
+	input: ti_int = ---
+	result: ti_int = ---
+	input.all = a
+	if b & bits != 0 {
+		result.lo = 0
+		result.hi = input.lo << (b-bits)
+	} else {
+		if b == 0 {
+			return a
+		}
+		result.lo = input.lo<<b
+		result.hi = (input.hi<<b) | (input.lo>>(bits-b))
+	}
+	return result.all
+}
+
+__ashlti3_unsigned :: proc "contextless" (a: u128, b: u32) -> u128 {
+	return cast(u128)__ashlti3(cast(i128)a, b)
+}
+
+@(link_name="__mulddi3", linkage="strong")
+__mulddi3 :: proc "contextless" (a, b: u64) -> i128 {
+	r: ti_int
+	bits :: 32
+
+	mask :: ~u64(0) >> bits
+	r.lo = (a & mask) * (b & mask)
+	t := r.lo >> bits
+	r.lo &= mask
+	t += (a >> bits) * (b & mask)
+	r.lo += (t & mask) << bits
+	r.hi = t >> bits
+	t = r.lo >> bits
+	r.lo &= mask
+	t += (b >> bits) * (a & mask)
+	r.lo += (t & mask) << bits
+	r.hi += t >> bits
+	r.hi += (a >> bits) * (b >> bits)
+	return r.all
+}
+
+@(link_name="__multi3", linkage="strong")
+__multi3 :: proc "contextless" (a, b: i128) -> i128 {
+	x, y, r: ti_int
+
+	x.all = a
+	y.all = b
+	r.all = __mulddi3(x.lo, y.lo)
+	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)
+}
+
+@(link_name="__lshrti3", linkage="strong")
+__lshrti3 :: proc "c" (a: i128, b: u32) -> i128 {
+	bits :: 64
+
+	input, result: ti_int
+	input.all = a
+	if b & bits != 0 {
+		result.hi = 0
+		result.lo = input.hi >> (b - bits)
+	} else if b == 0 {
+		return a
+	} else {
+		result.hi = input.hi >> b
+		result.lo = (input.hi << (bits - b)) | (input.lo >> b)
+	}
+
+	return result.all
+}
@@ -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
+#+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
+#+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)
@@ -0,0 +1,127 @@
+package runtime
+
+import "base:intrinsics"
+
+@(require_results)
+random_generator_read_bytes :: proc(rg: Random_Generator, p: []byte) -> bool {
+	if rg.procedure != nil {
+		rg.procedure(rg.data, .Read, p)
+		return true
+	}
+	return false
+}
+
+@(require_results)
+random_generator_read_ptr :: proc(rg: Random_Generator, p: rawptr, len: uint) -> bool {
+	if rg.procedure != nil {
+		rg.procedure(rg.data, .Read, ([^]byte)(p)[:len])
+		return true
+	}
+	return false
+}
+
+@(require_results)
+random_generator_query_info :: proc(rg: Random_Generator) -> (info: Random_Generator_Query_Info) {
+	if rg.procedure != nil {
+		rg.procedure(rg.data, .Query_Info, ([^]byte)(&info)[:size_of(info)])
+	}
+	return
+}
+
+
+random_generator_reset_bytes :: proc(rg: Random_Generator, p: []byte) {
+	if rg.procedure != nil {
+		rg.procedure(rg.data, .Reset, p)
+	}
+}
+
+random_generator_reset_u64 :: proc(rg: Random_Generator, p: u64) {
+	if rg.procedure != nil {
+		p := p
+		rg.procedure(rg.data, .Reset, ([^]byte)(&p)[:size_of(p)])
+	}
+}
+
+
+Default_Random_State :: struct {
+	state: u64,
+	inc:   u64,
+}
+
+default_random_generator_proc :: proc(data: rawptr, mode: Random_Generator_Mode, p: []byte) {
+	@(require_results)
+	read_u64 :: proc "contextless" (r: ^Default_Random_State) -> u64 {
+		old_state := r.state
+		r.state = old_state * 6364136223846793005 + (r.inc|1)
+		xor_shifted := (((old_state >> 59) + 5) ~ old_state) * 12605985483714917081
+		rot := (old_state >> 59)
+		return (xor_shifted >> rot) | (xor_shifted << ((-rot) & 63))
+	}
+
+	@(thread_local)
+	global_rand_seed: Default_Random_State
+
+	init :: proc "contextless" (r: ^Default_Random_State, seed: u64) {
+		seed := seed
+		if seed == 0 {
+			seed = u64(intrinsics.read_cycle_counter())
+		}
+		r.state = 0
+		r.inc = (seed << 1) | 1
+		_ = read_u64(r)
+		r.state += seed
+		_ = read_u64(r)
+	}
+
+	r: ^Default_Random_State = ---
+	if data == nil {
+		r = &global_rand_seed
+	} else {
+		r = cast(^Default_Random_State)data
+	}
+
+	switch mode {
+	case .Read:
+		if r.state == 0 && r.inc == 0 {
+			init(r, 0)
+		}
+
+		switch len(p) {
+		case size_of(u64):
+			// Fast path for a 64-bit destination.
+			intrinsics.unaligned_store((^u64)(raw_data(p)), read_u64(r))
+		case:
+			// All other cases.
+			pos := i8(0)
+			val := u64(0)
+			for &v in p {
+				if pos == 0 {
+					val = read_u64(r)
+					pos = 7
+				}
+				v = byte(val)
+				val >>= 8
+				pos -= 1
+			}
+		}
+
+	case .Reset:
+		seed: u64
+		mem_copy_non_overlapping(&seed, raw_data(p), min(size_of(seed), len(p)))
+		init(r, seed)
+
+	case .Query_Info:
+		if len(p) != size_of(Random_Generator_Query_Info) {
+			return
+		}
+		info := (^Random_Generator_Query_Info)(raw_data(p))
+		info^ += {.Uniform, .Resettable}
+	}
+}
+
+default_random_generator :: proc "contextless" (state: ^Default_Random_State = nil) -> Random_Generator {
+	return {
+		procedure = default_random_generator_proc,
+		data = state,
+	}
+}
@@ -0,0 +1,34 @@
+package runtime
+
+Thread_Local_Cleaner :: #type proc "odin" ()
+
+@(private="file")
+thread_local_cleaners: [8]Thread_Local_Cleaner
+
+// Add a procedure that will be run at the end of a thread for the purpose of
+// deallocating state marked as `thread_local`.
+//
+// Intended to be called in an `init` procedure of a package with
+// dynamically-allocated memory that is stored in `thread_local` variables.
+add_thread_local_cleaner :: proc "contextless" (p: Thread_Local_Cleaner) {
+	for &v in thread_local_cleaners {
+		if v == nil {
+			v = p
+			return
+		}
+	}
+	panic_contextless("There are no more thread-local cleaner slots available.")
+}
+
+// Run all of the thread-local cleaner procedures.
+//
+// Intended to be called by the internals of a threading API at the end of a
+// thread's lifetime.
+run_thread_local_cleaners :: proc "odin" () {
+	for p in thread_local_cleaners {
+		if p == nil {
+			break
+		}
+		p()
+	}
+}
@@ -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
@@ -58,7 +58,7 @@ udivmod128 :: proc "c" (a, b: u128, rem: ^u128) -> u128 {
 			return u128(n[high] >> _ctz(d[high]))
 		}

-		sr = transmute(u32)(i32(_clz(d[high])) - i32(_clz(n[high])))
+		sr = u32(i32(_clz(d[high])) - i32(_clz(n[high])))
 		if sr > U64_BITS - 2 {
 			if rem != nil {
 				rem^ = a
@@ -107,7 +107,7 @@ udivmod128 :: proc "c" (a, b: u128, rem: ^u128) -> u128 {
 				r[low]  = n[high] >> (sr - U64_BITS)
 			}
 		} else {
-			sr = transmute(u32)(i32(_clz(d[high])) - i32(_clz(n[high])))
+			sr = u32(i32(_clz(d[high])) - i32(_clz(n[high])))

 			if sr > U64_BITS - 1 {
 				if rem != nil {
@@ -143,7 +143,7 @@ udivmod128 :: proc "c" (a, b: u128, rem: ^u128) -> u128 {
 		r_all = transmute(u128)r
 		s := i128(b - r_all - 1) >> (U128_BITS - 1)
 		carry = u32(s & 1)
-		r_all -= b & transmute(u128)s
+		r_all -= b & u128(s)
 		r = transmute([2]u64)r_all
 	}

@@ -0,0 +1,871 @@
+#+build wasm32, wasm64p32
+package runtime
+
+import "base:intrinsics"
+
+/*
+Port of emmalloc, modified for use in Odin.
+
+Invariants:
+	- Per-allocation header overhead is 8 bytes, smallest allocated payload
+	  amount is 8 bytes, and a multiple of 4 bytes.
+	- Acquired memory blocks are subdivided into disjoint regions that lie
+	  next to each other.
+	- A region is either in used or free.
+	  Used regions may be adjacent, and a used and unused region
+	  may be adjacent, but not two unused ones - they would be
+	  merged.
+	- Memory allocation takes constant time, unless the alloc needs to wasm_memory_grow()
+	  or memory is very close to being exhausted.
+	- Free and used regions are managed inside "root regions", which are slabs
+	  of memory acquired via wasm_memory_grow().
+	- Memory retrieved using wasm_memory_grow() can not be given back to the OS.
+	  Therefore, frees are internal to the allocator.
+
+Copyright (c) 2010-2014 Emscripten authors, see AUTHORS file.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.
+*/
+
+WASM_Allocator :: struct #no_copy {
+	// The minimum alignment of allocations.
+	alignment: uint,
+	// A region that contains as payload a single forward linked list of pointers to
+	// root regions of each disjoint region blocks.
+	list_of_all_regions: ^Root_Region,
+	// For each of the buckets, maintain a linked list head node. The head node for each
+	// free region is a sentinel node that does not actually represent any free space, but
+	// the sentinel is used to avoid awkward testing against (if node == freeRegionHeadNode)
+	// when adding and removing elements from the linked list, i.e. we are guaranteed that
+	// the sentinel node is always fixed and there, and the actual free region list elements
+	// start at free_region_buckets[i].next each.
+	free_region_buckets: [NUM_FREE_BUCKETS]Region,
+	// A bitmask that tracks the population status for each of the 64 distinct memory regions:
+	// a zero at bit position i means that the free list bucket i is empty. This bitmask is
+	// used to avoid redundant scanning of the 64 different free region buckets: instead by
+	// looking at the bitmask we can find in constant time an index to a free region bucket
+	// that contains free memory of desired size.
+	free_region_buckets_used: BUCKET_BITMASK_T,
+	// Because wasm memory can only be allocated in pages of 64k at a time, we keep any
+	// spilled/unused bytes that are left from the allocated pages here, first using this
+	// when bytes are needed.
+	spill: []byte,
+	// Mutex for thread safety, only used if the target feature "atomics" is enabled.
+	mu: Mutex_State,
+}
+
+// Not required to be called, called on first allocation otherwise.
+wasm_allocator_init :: proc(a: ^WASM_Allocator, alignment: uint = 8) {
+	assert(is_power_of_two(alignment), "alignment must be a power of two")
+	assert(alignment > 4, "alignment must be more than 4")
+
+	a.alignment = alignment
+
+	for i in 0..<NUM_FREE_BUCKETS {
+		a.free_region_buckets[i].next = &a.free_region_buckets[i]
+		a.free_region_buckets[i].prev = a.free_region_buckets[i].next
+	}
+
+	if !claim_more_memory(a, 3*size_of(Region)) {
+		panic("wasm_allocator: initial memory could not be allocated")
+	}
+}
+
+global_default_wasm_allocator_data: WASM_Allocator
+
+default_wasm_allocator :: proc() -> Allocator {
+	return wasm_allocator(&global_default_wasm_allocator_data)
+}
+
+wasm_allocator :: proc(a: ^WASM_Allocator) -> Allocator {
+	return {
+		data      = a,
+		procedure = wasm_allocator_proc,
+	}
+}
+
+wasm_allocator_proc :: proc(a: rawptr, mode: Allocator_Mode, size, alignment: int, old_memory: rawptr, old_size: int, loc := #caller_location) -> ([]byte, Allocator_Error) {
+	a := (^WASM_Allocator)(a)
+	if a == nil {
+		a = &global_default_wasm_allocator_data
+	}
+
+	if a.alignment == 0 {
+		wasm_allocator_init(a)
+	}
+
+	switch mode {
+	case .Alloc:
+		ptr := aligned_alloc(a, uint(alignment), uint(size), loc)
+		if ptr == nil {
+			return nil, .Out_Of_Memory
+		}
+		intrinsics.mem_zero(ptr, size)
+		return ([^]byte)(ptr)[:size], nil
+
+	case .Alloc_Non_Zeroed:
+		ptr := aligned_alloc(a, uint(alignment), uint(size), loc)
+		if ptr == nil {
+			return nil, .Out_Of_Memory
+		}
+		return ([^]byte)(ptr)[:size], nil
+
+	case .Resize:
+		ptr := aligned_realloc(a, old_memory, uint(alignment), uint(size), loc)
+		if ptr == nil {
+			return nil, .Out_Of_Memory
+		}
+
+		bytes := ([^]byte)(ptr)[:size]
+
+		if size > old_size {
+			new_region := raw_data(bytes[old_size:])
+			intrinsics.mem_zero(new_region, size - old_size)
+		}
+
+		return bytes, nil
+
+	case .Resize_Non_Zeroed:
+		ptr := aligned_realloc(a, old_memory, uint(alignment), uint(size), loc)
+		if ptr == nil {
+			return nil, .Out_Of_Memory
+		}
+		return ([^]byte)(ptr)[:size], nil
+
+	case .Free:
+		free(a, old_memory, loc)
+		return nil, nil
+
+	case .Free_All, .Query_Info:
+		return nil, .Mode_Not_Implemented
+
+	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
+	}
+
+	unreachable()
+}
+
+// Returns the allocated size of the allocator (both free and used).
+// If `nil` is given, the global allocator is used.
+wasm_allocator_size :: proc(a: ^WASM_Allocator = nil) -> (size: uint) {
+	a := a
+	if a == nil {
+		a = &global_default_wasm_allocator_data
+	}
+
+	lock(a)
+	defer unlock(a)
+
+	root := a.list_of_all_regions
+	for root != nil {
+		size += uint(uintptr(root.end_ptr) - uintptr(root))
+		root = root.next
+	}
+
+	size += len(a.spill)
+
+	return
+}
+
+// Returns the amount of free memory on the allocator.
+// If `nil` is given, the global allocator is used.
+wasm_allocator_free_space :: proc(a: ^WASM_Allocator = nil) -> (free: uint) {
+	a := a
+	if a == nil {
+		a = &global_default_wasm_allocator_data
+	}
+
+	lock(a)
+	defer unlock(a)
+
+	bucket_index: u64 = 0
+	bucket_mask := a.free_region_buckets_used
+
+	for bucket_mask != 0 {
+		index_add := intrinsics.count_trailing_zeros(bucket_mask)
+		bucket_index += index_add
+		bucket_mask >>= index_add
+		for free_region := a.free_region_buckets[bucket_index].next; free_region != &a.free_region_buckets[bucket_index]; free_region = free_region.next {
+			free += free_region.size - REGION_HEADER_SIZE
+		}
+		bucket_index += 1
+		bucket_mask >>= 1
+	}
+
+	free += len(a.spill)
+
+	return
+}
+
+@(private="file")
+NUM_FREE_BUCKETS :: 64
+@(private="file")
+BUCKET_BITMASK_T :: u64
+
+// Dynamic memory is subdivided into regions, in the format
+
+// <size:u32> ..... <size:u32> | <size:u32> ..... <size:u32> | <size:u32> ..... <size:u32> | .....
+
+// That is, at the bottom and top end of each memory region, the size of that region is stored. That allows traversing the
+// memory regions backwards and forwards. Because each allocation must be at least a multiple of 4 bytes, the lowest two bits of
+// each size field is unused. Free regions are distinguished by used regions by having the FREE_REGION_FLAG bit present
+// in the size field. I.e. for free regions, the size field is odd, and for used regions, the size field reads even.
+@(private="file")
+FREE_REGION_FLAG :: 0x1
+
+// Attempts to alloc more than this many bytes would cause an overflow when calculating the size of a region,
+// therefore allocations larger than this are short-circuited immediately on entry.
+@(private="file")
+MAX_ALLOC_SIZE :: 0xFFFFFFC7
+
+// A free region has the following structure:
+// <size:uint> <prevptr> <nextptr> ... <size:uint>
+
+@(private="file")
+Region :: struct {
+	size: uint,
+	prev, next: ^Region,
+	_at_the_end_of_this_struct_size: uint,
+}
+
+// Each memory block starts with a Root_Region at the beginning.
+// The Root_Region specifies the size of the region block, and forms a linked
+// list of all Root_Regions in the program, starting with `list_of_all_regions`
+// below.
+@(private="file")
+Root_Region :: struct {
+	size:    u32,
+	next:    ^Root_Region,
+	end_ptr: ^byte,
+}
+
+@(private="file")
+Mutex_State :: enum u32 {
+	Unlocked = 0,
+	Locked   = 1,
+	Waiting  = 2,
+}
+
+@(private="file")
+lock :: proc(a: ^WASM_Allocator) {
+	when intrinsics.has_target_feature("atomics") {
+		@(cold)
+		lock_slow :: proc(a: ^WASM_Allocator, curr_state: Mutex_State) {
+			new_state := curr_state // Make a copy of it
+
+			spin_lock: for spin in 0..<i32(100) {
+				state, ok := intrinsics.atomic_compare_exchange_weak_explicit(&a.mu, .Unlocked, new_state, .Acquire, .Consume)
+				if ok {
+					return
+				}
+
+				if state == .Waiting {
+					break spin_lock
+				}
+
+				for i := min(spin+1, 32); i > 0; i -= 1 {
+					intrinsics.cpu_relax()
+				}
+			}
+
+			// Set just in case 100 iterations did not do it
+			new_state = .Waiting
+
+			for {
+				if intrinsics.atomic_exchange_explicit(&a.mu, .Waiting, .Acquire) == .Unlocked {
+					return
+				}
+
+				ret := intrinsics.wasm_memory_atomic_wait32((^u32)(&a.mu), u32(new_state), -1)
+				assert(ret != 0)
+				intrinsics.cpu_relax()
+			}
+		}
+
+
+		if v := intrinsics.atomic_exchange_explicit(&a.mu, .Locked, .Acquire); v != .Unlocked {
+			lock_slow(a, v)
+		}
+	}
+}
+
+@(private="file")
+unlock :: proc(a: ^WASM_Allocator) {
+	when intrinsics.has_target_feature("atomics") {
+		@(cold)
+		unlock_slow :: proc(a: ^WASM_Allocator) {
+			for {
+				s := intrinsics.wasm_memory_atomic_notify32((^u32)(&a.mu), 1)
+				if s >= 1 {
+					return
+				}
+			}
+		}
+
+		switch intrinsics.atomic_exchange_explicit(&a.mu, .Unlocked, .Release) {
+		case .Unlocked:
+			unreachable()
+		case .Locked:
+		// Okay
+		case .Waiting:
+			unlock_slow(a)
+		}
+	}
+}
+
+@(private="file")
+assert_locked :: proc(a: ^WASM_Allocator) {
+	when intrinsics.has_target_feature("atomics") {
+		assert(intrinsics.atomic_load(&a.mu) != .Unlocked)
+	}
+}
+
+@(private="file")
+has_alignment_uintptr :: proc(ptr: uintptr, #any_int alignment: uintptr) -> bool {
+	return ptr & (alignment-1) == 0
+}
+
+@(private="file")
+has_alignment_uint :: proc(ptr: uint, alignment: uint) -> bool {
+	return ptr & (alignment-1) == 0
+}
+
+@(private="file")
+has_alignment :: proc {
+	has_alignment_uintptr,
+	has_alignment_uint,
+}
+
+@(private="file")
+REGION_HEADER_SIZE :: 2*size_of(uint)
+
+@(private="file")
+SMALLEST_ALLOCATION_SIZE :: 2*size_of(rawptr)
+
+// Subdivide regions of free space into distinct circular doubly linked lists, where each linked list
+// represents a range of free space blocks. The following function compute_free_list_bucket() converts
+// an allocation size to the bucket index that should be looked at.
+#assert(NUM_FREE_BUCKETS == 64, "Following function is tailored specifically for the NUM_FREE_BUCKETS == 64 case")
+@(private="file")
+compute_free_list_bucket :: proc(size: uint) -> uint {
+	if size < 128 { return (size >> 3) - 1 }
+
+	clz := intrinsics.count_leading_zeros(i32(size))
+	bucket_index: i32 = ((clz > 19) \
+		?     110 - (clz<<2) + ((i32)(size >> (u32)(29-clz)) ~ 4) \
+		: min( 71 - (clz<<1) + ((i32)(size >> (u32)(30-clz)) ~ 2), NUM_FREE_BUCKETS-1))
+
+	assert(bucket_index >= 0)
+	assert(bucket_index < NUM_FREE_BUCKETS)
+	return uint(bucket_index)
+}
+
+@(private="file")
+prev_region :: proc(region: ^Region) -> ^Region {
+	prev_region_size := ([^]uint)(region)[-1]
+	prev_region_size  = prev_region_size & ~uint(FREE_REGION_FLAG)
+	return (^Region)(uintptr(region)-uintptr(prev_region_size))
+}
+
+@(private="file")
+next_region :: proc(region: ^Region) -> ^Region {
+	return (^Region)(uintptr(region)+uintptr(region.size))
+}
+
+@(private="file")
+region_ceiling_size :: proc(region: ^Region) -> uint {
+	return ([^]uint)(uintptr(region)+uintptr(region.size))[-1]
+}
+
+@(private="file")
+region_is_free :: proc(r: ^Region) -> bool {
+	return region_ceiling_size(r) & FREE_REGION_FLAG >= 1
+}
+
+@(private="file")
+region_is_in_use :: proc(r: ^Region) -> bool {
+	return r.size == region_ceiling_size(r)
+}
+
+@(private="file")
+region_payload_start_ptr :: proc(r: ^Region) -> [^]byte {
+	return ([^]byte)(r)[size_of(uint):]
+}
+
+@(private="file")
+region_payload_end_ptr :: proc(r: ^Region) -> [^]byte {
+	return ([^]byte)(r)[r.size-size_of(uint):]
+}
+
+@(private="file")
+create_used_region :: proc(ptr: rawptr, size: uint) {
+	assert(has_alignment(uintptr(ptr), size_of(uint)))
+	assert(has_alignment(size, size_of(uint)))
+	assert(size >= size_of(Region))
+
+	uptr := ([^]uint)(ptr)
+	uptr[0] = size
+	uptr[size/size_of(uint)-1] = size
+}
+
+@(private="file")
+create_free_region :: proc(ptr: rawptr, size: uint) {
+	assert(has_alignment(uintptr(ptr), size_of(uint)))
+	assert(has_alignment(size, size_of(uint)))
+	assert(size >= size_of(Region))
+
+	free_region := (^Region)(ptr)
+	free_region.size = size
+	([^]uint)(ptr)[size/size_of(uint)-1] = size | FREE_REGION_FLAG
+}
+
+@(private="file")
+prepend_to_free_list :: proc(region: ^Region, prepend_to: ^Region) {
+	assert(region_is_free(region))
+	region.next = prepend_to
+	region.prev = prepend_to.prev
+	prepend_to.prev = region
+	region.prev.next = region
+}
+
+@(private="file")
+unlink_from_free_list :: proc(region: ^Region) {
+	assert(region_is_free(region))
+	region.prev.next = region.next
+	region.next.prev = region.prev
+}
+
+@(private="file")
+link_to_free_list :: proc(a: ^WASM_Allocator, free_region: ^Region) {
+	assert(free_region.size >= size_of(Region))
+	bucket_index := compute_free_list_bucket(free_region.size-REGION_HEADER_SIZE)
+	free_list_head := &a.free_region_buckets[bucket_index]
+	free_region.prev = free_list_head
+	free_region.next = free_list_head.next
+	free_list_head.next = free_region
+	free_region.next.prev = free_region
+	a.free_region_buckets_used |= BUCKET_BITMASK_T(1) << bucket_index
+}
+
+@(private="file")
+claim_more_memory :: proc(a: ^WASM_Allocator, num_bytes: uint) -> bool {
+
+	PAGE_SIZE :: 64 * 1024
+
+	page_alloc :: proc(page_count: int) -> []byte {
+		prev_page_count := intrinsics.wasm_memory_grow(0, uintptr(page_count))
+		if prev_page_count < 0 { return nil }
+
+		ptr := ([^]byte)(uintptr(prev_page_count) * PAGE_SIZE)
+		return ptr[:page_count * PAGE_SIZE]
+	}
+
+	alloc :: proc(a: ^WASM_Allocator, num_bytes: uint) -> (bytes: [^]byte) #no_bounds_check {
+		if uint(len(a.spill)) >= num_bytes {
+			bytes = raw_data(a.spill[:num_bytes])
+			a.spill = a.spill[num_bytes:]
+			return
+		}
+
+		pages := int((num_bytes / PAGE_SIZE) + 1)
+		allocated := page_alloc(pages)
+		if allocated == nil { return nil }
+
+		// If the allocated memory is a direct continuation of the spill from before,
+		// we can just extend the spill.
+		spill_end := uintptr(raw_data(a.spill)) + uintptr(len(a.spill))
+		if spill_end == uintptr(raw_data(allocated)) {
+			raw_spill := (^Raw_Slice)(&a.spill)
+			raw_spill.len += len(allocated)
+		} else {
+			// Otherwise, we have to "waste" the previous spill.
+			// Now this is probably uncommon, and will only happen if another code path
+			// is also requesting pages.
+			a.spill = allocated
+		}
+
+		bytes = raw_data(a.spill)
+		a.spill = a.spill[num_bytes:]
+		return
+	}
+
+	num_bytes := num_bytes
+	num_bytes  = align_forward(num_bytes, a.alignment)
+
+	start_ptr := alloc(a, uint(num_bytes))
+	if start_ptr == nil { return false }
+
+	assert(has_alignment(uintptr(start_ptr), align_of(uint)))
+	end_ptr := start_ptr[num_bytes:]
+
+	end_sentinel_region := (^Region)(end_ptr[-size_of(Region):])
+	create_used_region(end_sentinel_region, size_of(Region))
+
+	// If we are the sole user of wasm_memory_grow(), it will feed us continuous/consecutive memory addresses - take advantage
+	// of that if so: instead of creating two disjoint memory regions blocks, expand the previous one to a larger size.
+	prev_alloc_end_address := a.list_of_all_regions != nil ? a.list_of_all_regions.end_ptr : nil
+	if start_ptr == prev_alloc_end_address {
+		prev_end_sentinel := prev_region((^Region)(start_ptr))
+		assert(region_is_in_use(prev_end_sentinel))
+		prev_region := prev_region(prev_end_sentinel)
+
+		a.list_of_all_regions.end_ptr = end_ptr
+
+		// Two scenarios, either the last region of the previous block was in use, in which case we need to create
+		// a new free region in the newly allocated space; or it was free, in which case we can extend that region
+		// to cover a larger size.
+		if region_is_free(prev_region) {
+			new_free_region_size := uint(uintptr(end_sentinel_region) - uintptr(prev_region))
+			unlink_from_free_list(prev_region)
+			create_free_region(prev_region, new_free_region_size)
+			link_to_free_list(a, prev_region)
+			return true
+		}
+
+		start_ptr = start_ptr[-size_of(Region):]
+	} else {
+		create_used_region(start_ptr, size_of(Region))
+
+		new_region_block := (^Root_Region)(start_ptr)
+		new_region_block.next = a.list_of_all_regions
+		new_region_block.end_ptr = end_ptr
+		a.list_of_all_regions = new_region_block
+		start_ptr = start_ptr[size_of(Region):]
+	}
+
+	create_free_region(start_ptr, uint(uintptr(end_sentinel_region)-uintptr(start_ptr)))
+	link_to_free_list(a, (^Region)(start_ptr))
+	return true
+}
+
+@(private="file")
+validate_alloc_size :: proc(size: uint) -> uint {
+	#assert(size_of(uint) >= size_of(uintptr))
+	#assert(size_of(uint)  % size_of(uintptr) == 0)
+
+	// NOTE: emmalloc aligns this forward on pointer size, but I think that is a mistake and will
+	// do bad on wasm64p32.
+
+	validated_size := size > SMALLEST_ALLOCATION_SIZE ? align_forward(size, size_of(uint)) : SMALLEST_ALLOCATION_SIZE
+	assert(validated_size >= size) // Assert we haven't wrapped.
+
+	return validated_size
+}
+
+@(private="file")
+allocate_memory :: proc(a: ^WASM_Allocator, alignment: uint, size: uint, loc := #caller_location) -> rawptr {
+
+	attempt_allocate :: proc(a: ^WASM_Allocator, free_region: ^Region, alignment, size: uint) -> rawptr {
+		assert_locked(a)
+		free_region := free_region
+
+		payload_start_ptr := uintptr(region_payload_start_ptr(free_region))
+		payload_start_ptr_aligned := align_forward(payload_start_ptr, uintptr(alignment))
+		payload_end_ptr := uintptr(region_payload_end_ptr(free_region))
+
+		if payload_start_ptr_aligned + uintptr(size) > payload_end_ptr {
+			return nil
+		}
+
+		// We have enough free space, so the memory allocation will be made into this region. Remove this free region
+		// from the list of free regions: whatever slop remains will be later added back to the free region pool.
+		unlink_from_free_list(free_region)
+
+		// Before we proceed further, fix up the boundary between this and the preceding region,
+		// so that the boundary between the two regions happens at a right spot for the payload to be aligned.
+		if payload_start_ptr != payload_start_ptr_aligned {
+			prev := prev_region(free_region)
+			assert(region_is_in_use(prev))
+			region_boundary_bump_amount := payload_start_ptr_aligned - payload_start_ptr
+			new_this_region_size := free_region.size - uint(region_boundary_bump_amount)
+			create_used_region(prev, prev.size + uint(region_boundary_bump_amount))
+			free_region = (^Region)(uintptr(free_region) + region_boundary_bump_amount)
+			free_region.size = new_this_region_size
+		}
+
+		// Next, we need to decide whether this region is so large that it should be split into two regions,
+		// one representing the newly used memory area, and at the high end a remaining leftover free area.
+		// This splitting to two is done always if there is enough space for the high end to fit a region.
+		// Carve 'size' bytes of payload off this region. So,
+		// [sz prev next sz]
+		// becomes
+		// [sz payload sz] [sz prev next sz]
+		if size_of(Region) + REGION_HEADER_SIZE + size <= free_region.size {
+			new_free_region := (^Region)(uintptr(free_region) + REGION_HEADER_SIZE + uintptr(size))
+			create_free_region(new_free_region, free_region.size - size - REGION_HEADER_SIZE)
+			link_to_free_list(a, new_free_region)
+			create_used_region(free_region, size + REGION_HEADER_SIZE)
+		} else {
+			// There is not enough space to split the free memory region into used+free parts, so consume the whole
+			// region as used memory, not leaving a free memory region behind.
+			// Initialize the free region as used by resetting the ceiling size to the same value as the size at bottom.
+			([^]uint)(uintptr(free_region) + uintptr(free_region.size))[-1] = free_region.size
+		}
+
+		return rawptr(uintptr(free_region) + size_of(uint))
+	}
+
+	assert_locked(a)
+	assert(is_power_of_two(alignment))
+	assert(size <= MAX_ALLOC_SIZE, "allocation too big", loc=loc)
+
+	alignment := alignment
+	alignment  = max(alignment, a.alignment)
+
+	size := size
+	size  = validate_alloc_size(size)
+
+	// Attempt to allocate memory starting from smallest bucket that can contain the required amount of memory.
+	// Under normal alignment conditions this should always be the first or second bucket we look at, but if
+	// performing an allocation with complex alignment, we may need to look at multiple buckets.
+	bucket_index := compute_free_list_bucket(size)
+	bucket_mask := a.free_region_buckets_used >> bucket_index
+
+	// Loop through each bucket that has free regions in it, based on bits set in free_region_buckets_used bitmap.
+	for bucket_mask != 0 {
+		index_add := intrinsics.count_trailing_zeros(bucket_mask)
+		bucket_index += uint(index_add)
+		bucket_mask >>= index_add
+		assert(bucket_index <= NUM_FREE_BUCKETS-1)
+		assert(a.free_region_buckets_used & (BUCKET_BITMASK_T(1) << bucket_index) > 0)
+
+		free_region := a.free_region_buckets[bucket_index].next
+		assert(free_region != nil)
+		if free_region != &a.free_region_buckets[bucket_index] {
+			ptr := attempt_allocate(a, free_region, alignment, size)
+			if ptr != nil {
+				return ptr
+			}
+
+			// We were not able to allocate from the first region found in this bucket, so penalize
+			// the region by cycling it to the end of the doubly circular linked list. (constant time)
+			// This provides a randomized guarantee that when performing allocations of size k to a
+			// bucket of [k-something, k+something] range, we will not always attempt to satisfy the
+			// allocation from the same available region at the front of the list, but we try each
+			// region in turn.
+			unlink_from_free_list(free_region)
+			prepend_to_free_list(free_region, &a.free_region_buckets[bucket_index])
+			// But do not stick around to attempt to look at other regions in this bucket - move
+			// to search the next populated bucket index if this did not fit. This gives a practical
+			// "allocation in constant time" guarantee, since the next higher bucket will only have
+			// regions that are all of strictly larger size than the requested allocation. Only if
+			// there is a difficult alignment requirement we may fail to perform the allocation from
+			// a region in the next bucket, and if so, we keep trying higher buckets until one of them
+			// works.
+			bucket_index += 1
+			bucket_mask >>= 1
+		} else {
+			// This bucket was not populated after all with any regions,
+			// but we just had a stale bit set to mark a populated bucket.
+			// Reset the bit to update latest status so that we do not
+			// redundantly look at this bucket again.
+			a.free_region_buckets_used &~= BUCKET_BITMASK_T(1) << bucket_index
+			bucket_mask ~= 1
+		}
+
+		assert((bucket_index == NUM_FREE_BUCKETS && bucket_mask == 0) || (bucket_mask == a.free_region_buckets_used >> bucket_index))
+	}
+
+	// None of the buckets were able to accommodate an allocation. If this happens we are almost out of memory.
+	// The largest bucket might contain some suitable regions, but we only looked at one region in that bucket, so
+	// as a last resort, loop through more free regions in the bucket that represents the largest allocations available.
+	// But only if the bucket representing largest allocations available is not any of the first thirty buckets,
+	// these represent allocatable areas less than <1024 bytes - which could be a lot of scrap.
+	// In such case, prefer to claim more memory right away.
+	largest_bucket_index := NUM_FREE_BUCKETS - 1 - intrinsics.count_leading_zeros(a.free_region_buckets_used)
+	// free_region will be null if there is absolutely no memory left. (all buckets are 100% used)
+	free_region := a.free_region_buckets_used > 0 ? a.free_region_buckets[largest_bucket_index].next : nil
+	// The 30 first free region buckets cover memory blocks < 2048 bytes, so skip looking at those here (too small)
+	if a.free_region_buckets_used >> 30 > 0 {
+		// Look only at a constant number of regions in this bucket max, to avoid bad worst case behavior.
+		// If this many regions cannot find free space, we give up and prefer to claim more memory instead.
+		max_regions_to_try_before_giving_up :: 99
+		num_tries_left := max_regions_to_try_before_giving_up
+		for ; free_region != &a.free_region_buckets[largest_bucket_index] && num_tries_left > 0; num_tries_left -= 1 {
+			ptr := attempt_allocate(a, free_region, alignment, size)
+			if ptr != nil {
+				return ptr
+			}
+			free_region = free_region.next
+		}
+	}
+
+	// We were unable to find a free memory region. Must claim more memory!
+	num_bytes_to_claim := size+size_of(Region)*3
+	if alignment > a.alignment {
+		num_bytes_to_claim += alignment
+	}
+	success := claim_more_memory(a, num_bytes_to_claim)
+	if (success) {
+		// Try allocate again with the newly available memory.
+		return allocate_memory(a, alignment, size)
+	}
+
+	// also claim_more_memory failed, we are really really constrained :( As a last resort, go back to looking at the
+	// bucket we already looked at above, continuing where the above search left off - perhaps there are
+	// regions we overlooked the first time that might be able to satisfy the allocation.
+	if free_region != nil {
+		for free_region != &a.free_region_buckets[largest_bucket_index] {
+			ptr := attempt_allocate(a, free_region, alignment, size)
+			if ptr != nil {
+				return ptr
+			}
+			free_region = free_region.next
+		}
+	}
+
+	// Fully out of memory.
+	return nil
+}
+
+@(private="file")
+aligned_alloc :: proc(a: ^WASM_Allocator, alignment, size: uint, loc := #caller_location) -> rawptr {
+	lock(a)
+	defer unlock(a)
+
+	return allocate_memory(a, alignment, size, loc)
+}
+
+@(private="file")
+free :: proc(a: ^WASM_Allocator, ptr: rawptr, loc := #caller_location) {
+	if ptr == nil {
+		return
+	}
+
+	region_start_ptr := uintptr(ptr) - size_of(uint)
+	region := (^Region)(region_start_ptr)
+	assert(has_alignment(region_start_ptr, size_of(uint)))
+
+	lock(a)
+	defer unlock(a)
+
+	size := region.size
+	assert(region_is_in_use(region), "double free or corrupt region", loc=loc)
+
+	prev_region_size_field := ([^]uint)(region)[-1]
+	prev_region_size := prev_region_size_field & ~uint(FREE_REGION_FLAG)
+	if prev_region_size_field != prev_region_size {
+		prev_region := (^Region)(uintptr(region) - uintptr(prev_region_size))
+		unlink_from_free_list(prev_region)
+		region_start_ptr = uintptr(prev_region)
+		size += prev_region_size
+	}
+
+	next_reg := next_region(region)
+	size_at_end := (^uint)(region_payload_end_ptr(next_reg))^
+	if next_reg.size != size_at_end {
+		unlink_from_free_list(next_reg)
+		size += next_reg.size
+	}
+
+	create_free_region(rawptr(region_start_ptr), size)
+	link_to_free_list(a, (^Region)(region_start_ptr))
+}
+
+@(private="file")
+aligned_realloc :: proc(a: ^WASM_Allocator, ptr: rawptr, alignment, size: uint, loc := #caller_location) -> rawptr {
+
+	attempt_region_resize :: proc(a: ^WASM_Allocator, region: ^Region, size: uint) -> bool {
+		lock(a)
+		defer unlock(a)
+
+		// First attempt to resize this region, if the next region that follows this one
+		// is a free region.
+		next_reg := next_region(region)
+		next_region_end_ptr := uintptr(next_reg) + uintptr(next_reg.size)
+		size_at_ceiling := ([^]uint)(next_region_end_ptr)[-1]
+		if next_reg.size != size_at_ceiling { // Next region is free?
+			assert(region_is_free(next_reg))
+			new_next_region_start_ptr := uintptr(region) + uintptr(size)
+			assert(has_alignment(new_next_region_start_ptr, size_of(uint)))
+			// Next region does not shrink to too small size?
+			if new_next_region_start_ptr + size_of(Region) <= next_region_end_ptr {
+				unlink_from_free_list(next_reg)
+				create_free_region(rawptr(new_next_region_start_ptr), uint(next_region_end_ptr - new_next_region_start_ptr))
+				link_to_free_list(a, (^Region)(new_next_region_start_ptr))
+				create_used_region(region, uint(new_next_region_start_ptr - uintptr(region)))
+				return true
+			}
+			// If we remove the next region altogether, allocation is satisfied?
+			if new_next_region_start_ptr <= next_region_end_ptr {
+				unlink_from_free_list(next_reg)
+				create_used_region(region, region.size + next_reg.size)
+				return true
+			}
+		} else {
+			// Next region is an used region - we cannot change its starting address. However if we are shrinking the
+			// size of this region, we can create a new free region between this and the next used region.
+			if size + size_of(Region) <= region.size {
+				free_region_size := region.size - size
+				create_used_region(region, size)
+				free_region := (^Region)(uintptr(region) + uintptr(size))
+				create_free_region(free_region, free_region_size)
+				link_to_free_list(a, free_region)
+				return true
+			} else if size <= region.size {
+				// Caller was asking to shrink the size, but due to not being able to fit a full Region in the shrunk
+				// area, we cannot actually do anything. This occurs if the shrink amount is really small. In such case,
+				// just call it success without doing any work.
+				return true
+			}
+		}
+
+		return false
+	}
+
+	if ptr == nil {
+		return aligned_alloc(a, alignment, size, loc)
+	}
+
+	if size == 0 {
+		free(a, ptr, loc)
+		return nil
+	}
+
+	if size > MAX_ALLOC_SIZE {
+		return nil
+	}
+
+	assert(is_power_of_two(alignment))
+	assert(has_alignment(uintptr(ptr), alignment), "realloc on different alignment than original allocation", loc=loc)
+
+	size := size
+	size  = validate_alloc_size(size)
+
+	region := (^Region)(uintptr(ptr) - size_of(uint))
+
+	// Attempt an in-place resize.
+	if attempt_region_resize(a, region, size + REGION_HEADER_SIZE) {
+		return ptr
+	}
+
+	// Can't do it in-place, allocate new region and copy over.
+	newptr := aligned_alloc(a, alignment, size, loc)
+	if newptr != nil {
+		intrinsics.mem_copy(newptr, ptr, min(size, region.size - REGION_HEADER_SIZE))
+		free(a, ptr, loc=loc)
+	}
+
+	return newptr
+}
@@ -2,12 +2,29 @@

 setlocal EnableDelayedExpansion

-for /f "usebackq tokens=1,2 delims=,=- " %%i in (`wmic os get LocalDateTime /value`) do @if %%i==LocalDateTime (
-	set CURR_DATE_TIME=%%j
+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
 )

+if "%VSCMD_ARG_TGT_ARCH%" neq "x64" (
+	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 %%i in ('powershell get-date -format "{yyyyMMdd}"') do (
+	set CURR_DATE_TIME=%%i
+)
 set curr_year=%CURR_DATE_TIME:~0,4%
 set curr_month=%CURR_DATE_TIME:~4,2%
+set curr_day=%CURR_DATE_TIME:~6,2%

 :: Make sure this is a decent name and not generic
 set exe_name=odin.exe
@@ -28,17 +45,50 @@ if "%2" == "1" (
 	set nightly=0
 )

-set odin_version_raw="dev-%curr_year%-%curr_month%"
+if %release_mode% equ 0 (
+	set V1=%curr_year%
+	set V2=%curr_month%
+	set V3=%curr_day%
+) else (
+	set V1=%curr_year%
+	set V2=%curr_month%
+	set V3=0
+)
+set V4=0
+set odin_version_full="%V1%.%V2%.%V3%.%V4%"
+set odin_version_raw="dev-%V1%-%V2%"
+

 set compiler_flags= -nologo -Oi -TP -fp:precise -Gm- -MP -FC -EHsc- -GR- -GF
+rem Parse source code as utf-8 even on shift-jis and other codepages
+rem See https://learn.microsoft.com/en-us/cpp/build/reference/utf-8-set-source-and-executable-character-sets-to-utf-8?view=msvc-170
+set compiler_flags= %compiler_flags% /utf-8
 set compiler_defines= -DODIN_VERSION_RAW=\"%odin_version_raw%\"

-for /f %%i in ('git rev-parse --short HEAD') do set GIT_SHA=%%i
-if %ERRORLEVEL% equ 0 set compiler_defines=%compiler_defines% -DGIT_SHA=\"%GIT_SHA%\"
+rem fileversion is defined as {Major,Minor,Build,Private: u16} so a bit limited
+set rc_flags=-nologo ^
+-DV1=%V1% -DV2=%V2% -DV3=%V3% -DV4=%V4% ^
+-DVF=%odin_version_full% -DNIGHTLY=%nightly%
+
+where /Q git.exe || goto skip_git_hash
+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%\"
+	set rc_flags=%rc_flags% -DGIT_SHA=%GIT_SHA% -DVP=%odin_version_raw%:%GIT_SHA%
+) else (
+	set rc_flags=%rc_flags% -DVP=%odin_version_raw%
+)
+:skip_git_hash
+
 if %nightly% equ 1 set compiler_defines=%compiler_defines% -DNIGHTLY

 if %release_mode% EQU 0 ( rem Debug
 	set compiler_flags=%compiler_flags% -Od -MDd -Z7
+	set rc_flags=%rc_flags% -D_DEBUG
 ) else ( rem Release
 	set compiler_flags=%compiler_flags% -O2 -MT -Z7
 	set compiler_defines=%compiler_defines% -DNO_ARRAY_BOUNDS_CHECK
@@ -47,35 +97,55 @@ if %release_mode% EQU 0 ( rem Debug
 set compiler_warnings= ^
 	-W4 -WX ^
 	-wd4100 -wd4101 -wd4127 -wd4146 ^
+	-wd4505 ^
 	-wd4456 -wd4457

 set compiler_includes= ^
 	/Isrc\
 set libs= ^
 	kernel32.lib ^
+	Synchronization.lib ^
 	bin\llvm\windows\LLVM-C.lib
+set odin_res=misc\odin.res
+set odin_rc=misc\odin.rc

-set linker_flags= -incremental:no -opt:ref -subsystem:console
+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 -MANIFEST:EMBED

 if %release_mode% EQU 0 ( rem Debug
-	set linker_flags=%linker_flags% -debug
+	set linker_flags=%linker_flags% -debug /NATVIS:src\odin_compiler.natvis
 ) else ( rem Release
 	set linker_flags=%linker_flags% -debug
 )

 set compiler_settings=%compiler_includes% %compiler_flags% %compiler_warnings% %compiler_defines%
-set linker_settings=%libs% %linker_flags%
+set linker_settings=%libs% %odin_res% %linker_flags%

 del *.pdb > NUL 2> NUL
 del *.ilk > NUL 2> NUL

+rc %rc_flags% %odin_rc%
 cl %compiler_settings% "src\main.cpp" "src\libtommath.cpp" /link %linker_settings% -OUT:%exe_name%
+mt -nologo -inputresource:%exe_name%;#1 -manifest misc\odin.manifest -outputresource:%exe_name%;#1 -validate_manifest -identity:"odin, processorArchitecture=amd64, version=%odin_version_full%, type=win32"
 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 -vet -strict-style -resource:examples/demo/demo.rc -- Hellope World
+
+rem Many non-compiler devs seem to run debug build but don't realize.
+if %release_mode% EQU 0 echo: & echo Debug compiler built. Note: run "build.bat release" if you want a faster, release mode compiler.

 del *.obj > NUL 2> NUL

@@ -1,82 +1,125 @@
-#!/bin/bash
+#!/usr/bin/env sh
 set -eu

-GIT_SHA=$(git rev-parse --short HEAD)
-DISABLED_WARNINGS="-Wno-switch -Wno-macro-redefined -Wno-unused-value"
-LDFLAGS="-pthread -lm -lstdc++"
-CFLAGS="-std=c++14 -DGIT_SHA=\"$GIT_SHA\""
-CFLAGS="$CFLAGS -DODIN_VERSION_RAW=\"dev-$(date +"%Y-%m")\""
-CC=clang
-OS=$(uname)
+: ${CPPFLAGS=}
+: ${CXXFLAGS=}
+: ${LDFLAGS=}
+: ${LLVM_CONFIG=}

-panic() {
-	printf "%s\n" "$1"
+CPPFLAGS="$CPPFLAGS -DODIN_VERSION_RAW=\"dev-$(date +"%Y-%m")\""
+CXXFLAGS="$CXXFLAGS -std=c++14"
+DISABLED_WARNINGS="-Wno-switch -Wno-macro-redefined -Wno-unused-value"
+LDFLAGS="$LDFLAGS -pthread -lm -lstdc++"
+OS_ARCH="$(uname -m)"
+OS_NAME="$(uname -s)"
+
+if [ -d ".git" ] && [ -n "$(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); }'; }
+# Brew advises people not to add llvm to their $PATH, so try and use brew to find it.
+if [ -z "$LLVM_CONFIG" ] &&  [ -n "$(command -v brew)" ]; then
+    if   [ -n "$(command -v $(brew --prefix llvm@18)/bin/llvm-config)" ]; then LLVM_CONFIG="$(brew --prefix llvm@18)/bin/llvm-config"
+    elif [ -n "$(command -v $(brew --prefix llvm@17)/bin/llvm-config)" ]; then LLVM_CONFIG="$(brew --prefix llvm@17)/bin/llvm-config"
+    elif [ -n "$(command -v $(brew --prefix llvm@14)/bin/llvm-config)" ]; then LLVM_CONFIG="$(brew --prefix llvm@14)/bin/llvm-config"
+    fi
+fi

-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 begining 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 $MIN_LLVM_VERSION) ]; then
-		if [ ARCH == arm64 ]; then
-			panic "Requirement: llvm-config must be base version 13 for arm64"
-		else
-			panic "Requirement: llvm-config must be base version greater than 11 for amd64/x86"
+if [ -x "$(which clang++)" ]; then
+	: ${CXX="clang++"}
+elif [ -x "$($LLVM_CONFIG --bindir)/clang++" ]; then
+	: ${CXX=$($LLVM_CONFIG --bindir)/clang++}
+else
+	error "No clang++ command found. Set CXX to proceed."
+fi
+
+LLVM_VERSION="$($LLVM_CONFIG --version)"
+LLVM_VERSION_MAJOR="$(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

-	LDFLAGS="$LDFLAGS -liconv -ldl"
-	CFLAGS="$CFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
-	LDFLAGS="$LDFLAGS -lLLVM-C"
-}
-
-config_freebsd() {
-	LLVM_CONFIG=/usr/local/bin/llvm-config11
-
-	CFLAGS="$CFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
-	LDFLAGS="$LDFLAGS $($LLVM_CONFIG --libs core native --system-libs)"
-}
-
-config_openbsd() {
-	LLVM_CONFIG=/usr/local/bin/llvm-config
-
-	LDFLAGS="$LDFLAGS -liconv"
-	CFLAGS="$CFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
-	LDFLAGS="$LDFLAGS $($LLVM_CONFIG --libs core native --system-libs)"
-}
-
-config_linux() {
-	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
+	darwin_sysroot=
+	if [ $(which xcrun) ]; then
+		darwin_sysroot="--sysroot $(xcrun --sdk macosx --show-sdk-path)"
+	elif [[ -e "/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk" ]]; then
+		darwin_sysroot="--sysroot /Library/Developer/CommandLineTools/SDKs/MacOSX.sdk"
 	else
-		panic "Unable to find LLVM-config"
+		echo "Warning: MacOSX.sdk not found."
 	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
-
-	LDFLAGS="$LDFLAGS -ldl"
-	CFLAGS="$CFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
+	CXXFLAGS="$CXXFLAGS $($LLVM_CONFIG --cxxflags --ldflags) ${darwin_sysroot}"
+	LDFLAGS="$LDFLAGS -liconv -ldl -framework System -lLLVM"
+	;;
+FreeBSD)
+	CXXFLAGS="$CXXFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
 	LDFLAGS="$LDFLAGS $($LLVM_CONFIG --libs core native --system-libs)"
-}
+	;;
+NetBSD)
+	CXXFLAGS="$CXXFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
+	LDFLAGS="$LDFLAGS $($LLVM_CONFIG --libs core native --system-libs)"
+	;;
+Linux)
+	CXXFLAGS="$CXXFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
+	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 $1 in
@@ -87,64 +130,50 @@ build_odin() {
 		EXTRAFLAGS="-O3"
 		;;
 	release-native)
-		EXTRAFLAGS="-O3 -march=native"
+		if [ "$OS_ARCH" = "arm64" ] || [ "$OS_ARCH" = "aarch64" ]; 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
-	$CC src/main.cpp src/libtommath.cpp $DISABLED_WARNINGS $CFLAGS $EXTRAFLAGS $LDFLAGS -o odin
+	$CXX src/main.cpp src/libtommath.cpp $DISABLED_WARNINGS $CPPFLAGS $CXXFLAGS $EXTRAFLAGS $LDFLAGS -o odin
 	set +x
 }

 run_demo() {
-	./odin run examples/demo/demo.odin -file
+	if [ $# -eq 0 ] || [ "$1" = "debug" ]; then
+		./odin run examples/demo -vet -strict-style -- Hellope World
+	fi
 }

-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
+	: ${PROGRAM:=$0}
+	printf "\nDebug compiler built. Note: run \"$PROGRAM release\" or \"$PROGRAM release-native\" if you want a faster, release mode compiler.\n"
+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 $1
 		;;
 	esac
-
 	run_demo
-	exit 0
 else
-	panic "Too many arguments!"
+	error "Too many arguments!"
 fi
@@ -15,3 +15,10 @@ if not exist "vendor\miniaudio\lib\*.lib" (
 		call build.bat
 	popd
 )
+
+
+if not exist "vendor\cgltf\lib\*.lib" (
+	pushd vendor\cgltf\src
+		call build.bat
+	popd
+)
@@ -0,0 +1,19 @@
+#!/usr/bin/env sh
+# Intended for use in Alpine containers, see the "nightly" Github action for a list of dependencies
+
+CXX="clang++-18"
+LLVM_CONFIG="llvm-config-18"
+
+DISABLED_WARNINGS="-Wno-switch -Wno-macro-redefined -Wno-unused-value"
+
+CPPFLAGS="-DODIN_VERSION_RAW=\"dev-$(date +"%Y-%m")\""
+CXXFLAGS="-std=c++14 $($LLVM_CONFIG --cxxflags --ldflags)"
+
+LDFLAGS="-static -lm -lzstd -lz -lffi -pthread -ldl -fuse-ld=mold"
+LDFLAGS="$LDFLAGS $($LLVM_CONFIG --link-static --ldflags --libs --system-libs --libfiles)"
+LDFLAGS="$LDFLAGS -Wl,-rpath=\$ORIGIN"
+
+EXTRAFLAGS="-DNIGHTLY -O3"
+
+set -x
+$CXX src/main.cpp src/libtommath.cpp $DISABLED_WARNINGS $CPPFLAGS $CXXFLAGS $EXTRAFLAGS $LDFLAGS -o odin
@@ -1,51 +0,0 @@
-import subprocess
-import sys
-import json
-import datetime
-import urllib.parse
-import sys
-
-def main():
-    files_by_date = {}
-    bucket = sys.argv[1]
-
-    files_lines = execute_cli(f"b2 ls --long {bucket} nightly").split("\n")
-    for x in files_lines:
-        parts = x.split(" ", 1)
-        if parts[0]:
-            json_str = execute_cli(f"b2 get-file-info {parts[0]}")
-            data = json.loads(json_str)
-            name = remove_prefix(data['fileName'], "nightly/")
-            url = f"https://f001.backblazeb2.com/file/{bucket}/nightly/{urllib.parse.quote_plus(name)}"
-            sha1 = data['contentSha1']
-            size = int(data['size'])
-            ts = int(data['fileInfo']['src_last_modified_millis'])
-            date = datetime.datetime.fromtimestamp(ts/1000).strftime('%Y-%m-%d')
-            
-            if date not in files_by_date.keys():
-                files_by_date[date] = []
-
-            files_by_date[date].append({
-                                            'name': name,
-                                            'url': url,
-                                            'sha1': sha1,
-                                            'sizeInBytes': size,
-                                         })
-
-    now = datetime.datetime.utcnow().isoformat()
-
-    print(json.dumps({
-                        'last_updated' : now,
-                        'files': files_by_date
-                     }, sort_keys=True, indent=4))
-
-def remove_prefix(text, prefix):
-    return text[text.startswith(prefix) and len(prefix):]
-
-def execute_cli(command):
-    sb = subprocess.Popen(command, shell=True, stdout=subprocess.PIPE)
-    return sb.stdout.read().decode("utf-8");
-
-if __name__ == '__main__':
-    sys.exit(main())
-
@@ -1,34 +0,0 @@
-import subprocess
-import sys
-import json
-import datetime
-import urllib.parse
-import sys
-
-def main():
-    files_by_date = {}
-    bucket = sys.argv[1]
-    days_to_keep = int(sys.argv[2])
-    print(f"Looking for binaries to delete older than {days_to_keep} days")
-
-    files_lines = execute_cli(f"b2 ls --long --versions {bucket} nightly").split("\n")
-    for x in files_lines:
-        parts = [y for y in x.split(' ') if y]
-
-        if parts and parts[0]:
-            date = datetime.datetime.strptime(parts[2], '%Y-%m-%d').replace(hour=0, minute=0, second=0, microsecond=0)
-            now = datetime.datetime.utcnow().replace(hour=0, minute=0, second=0, microsecond=0)
-            delta = now - date
-
-            if delta.days > days_to_keep:
-                print(f'Deleting {parts[5]}')
-                execute_cli(f'b2 delete-file-version {parts[0]}')
-
-
-def execute_cli(command):
-    sb = subprocess.Popen(command, shell=True, stdout=subprocess.PIPE)
-    return sb.stdout.read().decode("utf-8");
-
-if __name__ == '__main__':
-    sys.exit(main())
-
@@ -0,0 +1,145 @@
+import os
+import sys
+from zipfile  import ZipFile, ZIP_DEFLATED
+from b2sdk.v2 import InMemoryAccountInfo, B2Api
+from datetime import datetime, timezone
+import json
+
+UPLOAD_FOLDER = "nightly/"
+
+info   = InMemoryAccountInfo()
+b2_api = B2Api(info)
+application_key_id = os.environ['APPID']
+application_key    = os.environ['APPKEY']
+bucket_name        = os.environ['BUCKET']
+days_to_keep       = os.environ['DAYS_TO_KEEP']
+
+def auth() -> bool:
+	try:
+		realm = b2_api.account_info.get_realm()
+		return True # Already authenticated
+	except:
+		pass        # Not yet authenticated
+
+	err = b2_api.authorize_account("production", application_key_id, application_key)
+	return err is None
+
+def get_bucket():
+	if not auth(): sys.exit(1)
+	return b2_api.get_bucket_by_name(bucket_name)
+
+def remove_prefix(text: str, prefix: str) -> str:
+	return text[text.startswith(prefix) and len(prefix):]
+
+def create_and_upload_artifact_zip(platform: str, artifact: str) -> int:
+	now = datetime.now(timezone.utc).replace(hour=0, minute=0, second=0, microsecond=0)
+
+	source_archive: str
+	destination_name = f'odin-{platform}-nightly+{now.strftime("%Y-%m-%d")}'
+
+	if platform.startswith("linux") or platform.startswith("macos"):
+		destination_name += ".tar.gz"
+		source_archive = artifact
+	else:
+		destination_name += ".zip"
+		source_archive = destination_name
+
+		print(f"Creating archive {destination_name} from {artifact} and uploading to {bucket_name}")
+		with ZipFile(source_archive, mode='w', compression=ZIP_DEFLATED, compresslevel=9) as z:
+			for root, directory, filenames in os.walk(artifact):
+				for file in filenames:
+					file_path = os.path.join(root, file)
+					zip_path  = os.path.join("dist", os.path.relpath(file_path, artifact))
+					z.write(file_path, zip_path)
+
+	if not os.path.exists(source_archive):
+		print(f"Error: archive {source_archive} not found.")
+		return 1
+
+	print("Uploading {} to {}".format(source_archive, UPLOAD_FOLDER + destination_name))
+	bucket = get_bucket()
+	res = bucket.upload_local_file(
+		source_archive,               # Local file to upload
+		"nightly/" + destination_name, # B2 destination path
+	)
+	return 0
+
+def prune_artifacts():
+	print(f"Looking for binaries to delete older than {days_to_keep} days")
+
+	bucket = get_bucket()
+	for file, _ in bucket.ls(UPLOAD_FOLDER, latest_only=False):
+		# Timestamp is in milliseconds
+		date  = datetime.fromtimestamp(file.upload_timestamp / 1_000.0, tz=timezone.utc).replace(hour=0, minute=0, second=0, microsecond=0)
+		now   = datetime.now(timezone.utc).replace(hour=0, minute=0, second=0, microsecond=0)
+		delta = now - date
+
+		if delta.days > int(days_to_keep):
+			print("Deleting {}".format(file.file_name))
+			file.delete()
+
+	return 0
+
+def update_nightly_json():
+	print(f"Updating nightly.json with files {days_to_keep} days or newer")
+
+	files_by_date = {}
+
+	bucket = get_bucket()
+
+	for file, _ in bucket.ls(UPLOAD_FOLDER, latest_only=True):
+		# Timestamp is in milliseconds
+		date = datetime.fromtimestamp(file.upload_timestamp / 1_000.0).replace(hour=0, minute=0, second=0, microsecond=0).strftime('%Y-%m-%d')
+		name = remove_prefix(file.file_name, UPLOAD_FOLDER)
+		sha1 = file.content_sha1
+		size = file.size
+		url  = bucket.get_download_url(file.file_name)
+
+		if date not in files_by_date.keys():
+			files_by_date[date] = []
+
+		files_by_date[date].append({
+			'name':        name,
+			'url':         url,
+			'sha1':        sha1,
+			'sizeInBytes': size,
+		})
+
+	now = datetime.now(timezone.utc).isoformat()
+
+	nightly = json.dumps({
+		'last_updated' : now,
+		'files': files_by_date
+	}, sort_keys=True, indent=4, ensure_ascii=False).encode('utf-8')
+
+	res = bucket.upload_bytes(
+		nightly,        # JSON bytes
+		"nightly.json", # B2 destination path
+	)
+	return 0
+
+if __name__ == "__main__":
+	if len(sys.argv) == 1:
+		print("Usage: {} <verb> [arguments]".format(sys.argv[0]))
+		print("\tartifact <platform prefix> <artifact path>\n\t\tCreates and uploads a platform artifact zip.")
+		print("\tprune\n\t\tDeletes old artifacts from bucket")
+		print("\tjson\n\t\tUpdate and upload nightly.json")
+		sys.exit(1)
+	else:
+		command = sys.argv[1].lower()
+		if command == "artifact":
+			if len(sys.argv) != 4:
+				print("Usage: {} artifact <platform prefix> <artifact path>".format(sys.argv[0]))
+				print("Error: Expected artifact command to be given platform prefix and artifact path.\n")
+				sys.exit(1)
+
+			res = create_and_upload_artifact_zip(sys.argv[2], sys.argv[3])
+			sys.exit(res)
+
+		elif command == "prune":
+			res = prune_artifacts()
+			sys.exit(res)
+
+		elif command == "json":
+			res = update_nightly_json()
+			sys.exit(res)
@@ -1,13 +0,0 @@
-#!/bin/bash
-
-bucket=$1
-platform=$2
-artifact=$3
-
-now=$(date +'%Y-%m-%d')
-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"
@@ -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)
 }
@@ -14,55 +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_read_byte = proc(s: io.Stream) -> (c: byte, err: io.Error) {
-		b := (^Read_Writer)(s.stream_data).r
-		return reader_read_byte(b)
-	},
-	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
+}
@@ -29,12 +29,12 @@ MIN_READ_BUFFER_SIZE :: 16
@(private)
 DEFAULT_MAX_CONSECUTIVE_EMPTY_READS :: 128

-reader_init :: proc(b: ^Reader, rd: io.Reader, size: int = DEFAULT_BUF_SIZE, allocator := context.allocator) {
+reader_init :: proc(b: ^Reader, rd: io.Reader, size: int = DEFAULT_BUF_SIZE, allocator := context.allocator, loc := #caller_location) {
 	size := size
 	size = max(size, MIN_READ_BUFFER_SIZE)
 	reader_reset(b, rd)
 	b.buf_allocator = allocator
-	b.buf = make([]byte, size, allocator)
+	b.buf = make([]byte, size, allocator, loc)
 }

 reader_init_with_buf :: proc(b: ^Reader, rd: io.Reader, buf: []byte) {
@@ -81,7 +81,7 @@ _reader_read_new_chunk :: proc(b: ^Reader) -> io.Error {
 	for i := b.max_consecutive_empty_reads; i > 0; i -= 1 {
 		n, err := io.read(b.rd, b.buf[b.w:])
 		if n < 0 {
-			return .Negative_Read
+			return err if err != nil else .Negative_Read
 		}
 		b.w += n
 		if err != nil {
@@ -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 {
@@ -66,7 +66,7 @@ scanner_destroy :: proc(s: ^Scanner) {
 }


-// Returns the first non-EOF error that was encounted by the scanner
+// Returns the first non-EOF error that was encountered by the scanner
 scanner_error :: proc(s: ^Scanner) -> Scanner_Error {
 	switch s._err {
 	case .EOF, nil:
@@ -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
 }
@@ -27,17 +27,22 @@ Read_Op :: enum i8 {
 }


-buffer_init :: proc(b: ^Buffer, buf: []byte) {
-	resize(&b.buf, len(buf))
+buffer_init :: proc(b: ^Buffer, buf: []byte, loc := #caller_location) {
+	resize(&b.buf, len(buf), loc=loc)
 	copy(b.buf[:], buf)
 }

-buffer_init_string :: proc(b: ^Buffer, s: string) {
-	resize(&b.buf, len(s))
+buffer_init_string :: proc(b: ^Buffer, s: string, loc := #caller_location) {
+	resize(&b.buf, len(s), loc=loc)
 	copy(b.buf[:], s)
 }

-buffer_init_allocator :: proc(b: ^Buffer, len, cap: int, allocator := context.allocator) {
+buffer_init_allocator :: proc(b: ^Buffer, len, cap: int, allocator := context.allocator, loc := #caller_location) {
+	if b.buf == nil {
+		b.buf = make([dynamic]byte, len, cap, allocator, loc)
+		return
+	}
+
 	b.buf.allocator = allocator
 	reserve(&b.buf, cap)
 	resize(&b.buf, len)
@@ -91,25 +96,28 @@ buffer_truncate :: proc(b: ^Buffer, n: int) {
 }

@(private)
-_buffer_try_grow :: proc(b: ^Buffer, n: int) -> (int, bool) {
+_buffer_try_grow :: proc(b: ^Buffer, n: int, loc := #caller_location) -> (int, bool) {
 	if l := len(b.buf); n <= cap(b.buf)-l {
-		resize(&b.buf, l+n)
+		resize(&b.buf, l+n, loc=loc)
 		return l, true
 	}
 	return 0, false
 }

@(private)
-_buffer_grow :: proc(b: ^Buffer, n: int) -> int {
+_buffer_grow :: proc(b: ^Buffer, n: int, loc := #caller_location) -> int {
 	m := buffer_length(b)
 	if m == 0 && b.off != 0 {
 		buffer_reset(b)
 	}
-	if i, ok := _buffer_try_grow(b, n); ok {
+	if i, ok := _buffer_try_grow(b, n, loc=loc); ok {
 		return i
 	}
+
 	if b.buf == nil && n <= SMALL_BUFFER_SIZE {
-		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, loc=loc)
+		resize(&b.buf, n, loc=loc)
 		return 0
 	}

@@ -119,31 +127,34 @@ _buffer_grow :: proc(b: ^Buffer, n: int) -> int {
 	} else if c > max(int) - c - n {
 		panic("bytes.Buffer: too large")
 	} else {
-		resize(&b.buf, 2*c + n)
+		resize(&b.buf, 2*c + n, loc=loc)
 		copy(b.buf[:], b.buf[b.off:])
 	}
 	b.off = 0
-	resize(&b.buf, m+n)
+	resize(&b.buf, m+n, loc=loc)
 	return m
 }

-buffer_grow :: proc(b: ^Buffer, n: int) {
+buffer_grow :: proc(b: ^Buffer, n: int, loc := #caller_location) {
 	if n < 0 {
 		panic("bytes.buffer_grow: negative count")
 	}
-	m := _buffer_grow(b, n)
-	resize(&b.buf, m)
+	m := _buffer_grow(b, n, loc=loc)
+	resize(&b.buf, m, loc=loc)
 }

-buffer_write_at :: proc(b: ^Buffer, p: []byte, offset: int) -> (n: int, err: io.Error) {
+buffer_write_at :: proc(b: ^Buffer, p: []byte, offset: int, loc := #caller_location) -> (n: int, err: io.Error) {
+	if len(p) == 0 {
+		return 0, nil
+	}
 	b.last_read = .Invalid
 	if offset < 0 {
 		err = .Invalid_Offset
 		return
 	}
-	_, ok := _buffer_try_grow(b, offset+len(p))
+	_, ok := _buffer_try_grow(b, offset+len(p), loc=loc)
 	if !ok {
-		_ = _buffer_grow(b, offset+len(p))
+		_ = _buffer_grow(b, offset+len(p), loc=loc)
 	}
 	if len(b.buf) <= offset {
 		return 0, .Short_Write
@@ -152,43 +163,47 @@ buffer_write_at :: proc(b: ^Buffer, p: []byte, offset: int) -> (n: int, err: io.
 }


-buffer_write :: proc(b: ^Buffer, p: []byte) -> (n: int, err: io.Error) {
+buffer_write :: proc(b: ^Buffer, p: []byte, loc := #caller_location) -> (n: int, err: io.Error) {
 	b.last_read = .Invalid
-	m, ok := _buffer_try_grow(b, len(p))
+	m, ok := _buffer_try_grow(b, len(p), loc=loc)
 	if !ok {
-		m = _buffer_grow(b, len(p))
+		m = _buffer_grow(b, len(p), loc=loc)
 	}
 	return copy(b.buf[m:], p), nil
 }

-buffer_write_string :: proc(b: ^Buffer, s: string) -> (n: int, err: io.Error) {
+buffer_write_ptr :: proc(b: ^Buffer, ptr: rawptr, size: int, loc := #caller_location) -> (n: int, err: io.Error) {
+	return buffer_write(b, ([^]byte)(ptr)[:size], loc=loc)
+}
+
+buffer_write_string :: proc(b: ^Buffer, s: string, loc := #caller_location) -> (n: int, err: io.Error) {
 	b.last_read = .Invalid
-	m, ok := _buffer_try_grow(b, len(s))
+	m, ok := _buffer_try_grow(b, len(s), loc=loc)
 	if !ok {
-		m = _buffer_grow(b, len(s))
+		m = _buffer_grow(b, len(s), loc=loc)
 	}
 	return copy(b.buf[m:], s), nil
 }

-buffer_write_byte :: proc(b: ^Buffer, c: byte) -> io.Error {
+buffer_write_byte :: proc(b: ^Buffer, c: byte, loc := #caller_location) -> io.Error {
 	b.last_read = .Invalid
-	m, ok := _buffer_try_grow(b, 1)
+	m, ok := _buffer_try_grow(b, 1, loc=loc)
 	if !ok {
-		m = _buffer_grow(b, 1)
+		m = _buffer_grow(b, 1, loc=loc)
 	}
 	b.buf[m] = c
 	return nil
 }

-buffer_write_rune :: proc(b: ^Buffer, r: rune) -> (n: int, err: io.Error) {
+buffer_write_rune :: proc(b: ^Buffer, r: rune, loc := #caller_location) -> (n: int, err: io.Error) {
 	if r < utf8.RUNE_SELF {
-		buffer_write_byte(b, byte(r))
+		buffer_write_byte(b, byte(r), loc=loc)
 		return 1, nil
 	}
 	b.last_read = .Invalid
-	m, ok := _buffer_try_grow(b, utf8.UTF_MAX)
+	m, ok := _buffer_try_grow(b, utf8.UTF_MAX, loc=loc)
 	if !ok {
-		m = _buffer_grow(b, utf8.UTF_MAX)
+		m = _buffer_grow(b, utf8.UTF_MAX, loc=loc)
 	}
 	res: [4]byte
 	res, n = utf8.encode_rune(r)
@@ -229,17 +244,21 @@ buffer_read :: proc(b: ^Buffer, p: []byte) -> (n: int, err: io.Error) {
 	return
 }

+buffer_read_ptr :: proc(b: ^Buffer, ptr: rawptr, size: int) -> (n: int, err: io.Error) {
+	return buffer_read(b, ([^]byte)(ptr)[:size])
+}
+
 buffer_read_at :: proc(b: ^Buffer, p: []byte, offset: int) -> (n: int, err: io.Error) {
+	if len(p) == 0 {
+		return 0, nil
+	}
 	b.last_read = .Invalid

-	if offset < 0 || offset >= len(b.buf) {
-		err = .Invalid_Offset
+	if uint(offset) >= len(b.buf) {
+		err = .EOF
 		return
 	}
-
-	if 0 <= offset && offset < len(b.buf) {
-		n = copy(p, b.buf[offset:])
-	}
+	n = copy(p, b.buf[offset:])
 	if n > 0 {
 		b.last_read = .Read
 	}
@@ -297,6 +316,27 @@ buffer_unread_rune :: proc(b: ^Buffer) -> io.Error {
 	return nil
 }

+buffer_seek :: proc(b: ^Buffer, offset: i64, whence: io.Seek_From) -> (i64, io.Error) {
+	abs: i64
+	switch whence {
+	case .Start:
+		abs = offset
+	case .Current:
+		abs = i64(b.off) + offset
+	case .End:
+		abs = i64(len(b.buf)) + offset
+	case:
+		return 0, .Invalid_Whence
+	}
+
+	abs_int := int(abs)
+	if abs_int < 0 {
+		return 0, .Invalid_Offset
+	}
+	b.last_read = .Invalid
+	b.off = abs_int
+	return abs, nil
+}

 buffer_read_bytes :: proc(b: ^Buffer, delim: byte) -> (line: []byte, err: io.Error) {
 	i := index_byte(b.buf[b.off:], delim)
@@ -346,7 +386,7 @@ 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
 		}

@@ -365,69 +405,34 @@ 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 .Seek:
+		n, err = buffer_seek(b, offset, whence)
+		return
+	case .Size:
+		n = i64(buffer_length(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, .Seek, .Size, .Destroy, .Query})
+	}
+	return 0, .Empty
 }
-
@@ -1,9 +1,38 @@
 package bytes

+import "base:intrinsics"
 import "core:mem"
+import "core:simd"
 import "core:unicode"
 import "core:unicode/utf8"

+when ODIN_ARCH == .amd64 && intrinsics.has_target_feature("avx2") {
+	@(private)
+	SCANNER_INDICES_256 : simd.u8x32 : {
+		0,  1,  2,  3,  4,  5,  6,  7,
+		8,  9, 10, 11, 12, 13, 14, 15,
+		16, 17, 18, 19, 20, 21, 22, 23,
+		24, 25, 26, 27, 28, 29, 30, 31,
+	}
+	@(private)
+	SCANNER_SENTINEL_MAX_256: simd.u8x32 : u8(0x00)
+	@(private)
+	SCANNER_SENTINEL_MIN_256: simd.u8x32 : u8(0xff)
+	@(private)
+	SIMD_REG_SIZE_256 :: 32
+}
+@(private)
+SCANNER_INDICES_128 : simd.u8x16 : {
+	0,  1,  2,  3,  4,  5,  6,  7,
+	8,  9, 10, 11, 12, 13, 14, 15,
+}
+@(private)
+SCANNER_SENTINEL_MAX_128: simd.u8x16 : u8(0x00)
+@(private)
+SCANNER_SENTINEL_MIN_128: simd.u8x16 : u8(0xff)
+@(private)
+SIMD_REG_SIZE_128 :: 16
+
 clone :: proc(s: []byte, allocator := context.allocator, loc := #caller_location) -> []byte {
 	c := make([]byte, len(s), allocator, loc)
 	copy(c, s)
@@ -293,28 +322,277 @@ split_after_iterator :: proc(s: ^[]byte, sep: []byte) -> ([]byte, bool) {
 	return _split_iterator(s, sep, len(sep))
 }

+/*
+Scan a slice of bytes for a specific byte.

-index_byte :: proc(s: []byte, c: byte) -> int {
-	for i := 0; i < len(s); i += 1 {
+This procedure safely handles slices of any length, including empty slices.
+
+Inputs:
+- data: A slice of bytes.
+- c: The byte to search for.
+
+Returns:
+- index: The index of the byte `c`, or -1 if it was not found.
+*/
+index_byte :: proc "contextless" (s: []byte, c: byte) -> (index: int) #no_bounds_check {
+	i, l := 0, len(s)
+
+	// Guard against small strings.  On modern systems, it is ALWAYS
+	// worth vectorizing assuming there is a hardware vector unit, and
+	// the data size is large enough.
+	if l < SIMD_REG_SIZE_128 {
+		for /**/; i < l; i += 1 {
+			if s[i] == c {
+				return i
+			}
+		}
+		return -1
+	}
+
+	c_vec: simd.u8x16 = c
+	when !simd.IS_EMULATED {
+		// Note: While this is something that could also logically take
+		// advantage of AVX512, the various downclocking and power
+		// consumption related woes make premature to have a dedicated
+		// code path.
+		when ODIN_ARCH == .amd64 && intrinsics.has_target_feature("avx2") {
+			c_vec_256: simd.u8x32 = c
+
+			s_vecs: [4]simd.u8x32 = ---
+			c_vecs: [4]simd.u8x32 = ---
+			m_vec: [4]u8 = ---
+
+			// Scan 128-byte chunks, using 256-bit SIMD.
+			for nr_blocks := l / (4 * SIMD_REG_SIZE_256); nr_blocks > 0; nr_blocks -= 1 {
+				#unroll for j in 0..<4 {
+					s_vecs[j] = intrinsics.unaligned_load(cast(^simd.u8x32)raw_data(s[i+j*SIMD_REG_SIZE_256:]))
+					c_vecs[j] = simd.lanes_eq(s_vecs[j], c_vec_256)
+					m_vec[j] = simd.reduce_or(c_vecs[j])
+				}
+				if m_vec[0] | m_vec[1] | m_vec[2] | m_vec[3] > 0 {
+					#unroll for j in 0..<4 {
+						if m_vec[j] > 0 {
+							sel := simd.select(c_vecs[j], SCANNER_INDICES_256, SCANNER_SENTINEL_MIN_256)
+							off := simd.reduce_min(sel)
+							return i + j * SIMD_REG_SIZE_256 + int(off)
+						}
+					}
+				}
+
+				i += 4 * SIMD_REG_SIZE_256
+			}
+
+			// Scan 64-byte chunks, using 256-bit SIMD.
+			for nr_blocks := (l - i) / (2 * SIMD_REG_SIZE_256); nr_blocks > 0; nr_blocks -= 1 {
+				#unroll for j in 0..<2 {
+					s_vecs[j] = intrinsics.unaligned_load(cast(^simd.u8x32)raw_data(s[i+j*SIMD_REG_SIZE_256:]))
+					c_vecs[j] = simd.lanes_eq(s_vecs[j], c_vec_256)
+					m_vec[j] = simd.reduce_or(c_vecs[j])
+				}
+				if m_vec[0] | m_vec[1] > 0 {
+					#unroll for j in 0..<2 {
+						if m_vec[j] > 0 {
+							sel := simd.select(c_vecs[j], SCANNER_INDICES_256, SCANNER_SENTINEL_MIN_256)
+							off := simd.reduce_min(sel)
+							return i + j * SIMD_REG_SIZE_256 + int(off)
+						}
+					}
+				}
+
+				i += 2 * SIMD_REG_SIZE_256
+			}
+		} else {
+			s_vecs: [4]simd.u8x16 = ---
+			c_vecs: [4]simd.u8x16 = ---
+			m_vecs: [4]u8 = ---
+
+			// Scan 64-byte chunks, using 128-bit SIMD.
+			for nr_blocks := l / (4 * SIMD_REG_SIZE_128); nr_blocks > 0; nr_blocks -= 1 {
+				#unroll for j in 0..<4 {
+					s_vecs[j]= intrinsics.unaligned_load(cast(^simd.u8x16)raw_data(s[i+j*SIMD_REG_SIZE_128:]))
+					c_vecs[j] = simd.lanes_eq(s_vecs[j], c_vec)
+					m_vecs[j] = simd.reduce_or(c_vecs[j])
+				}
+				if m_vecs[0] | m_vecs[1] | m_vecs[2] | m_vecs[3] > 0 {
+					#unroll for j in 0..<4 {
+						if m_vecs[j] > 0 {
+							sel := simd.select(c_vecs[j], SCANNER_INDICES_128, SCANNER_SENTINEL_MIN_128)
+							off := simd.reduce_min(sel)
+							return i + j * SIMD_REG_SIZE_128 + int(off)
+						}
+					}
+				}
+
+				i += 4 * SIMD_REG_SIZE_128
+			}
+		}
+	}
+
+	// Scan the remaining SIMD register sized chunks.
+	//
+	// Apparently LLVM does ok with 128-bit SWAR, so this path is also taken
+	// on potato targets.  Scanning more at a time when LLVM is emulating SIMD
+	// likely does not buy much, as all that does is increase GP register
+	// pressure.
+	for nr_blocks := (l - i) / SIMD_REG_SIZE_128; nr_blocks > 0; nr_blocks -= 1 {
+		s0 := intrinsics.unaligned_load(cast(^simd.u8x16)raw_data(s[i:]))
+		c0 := simd.lanes_eq(s0, c_vec)
+		if simd.reduce_or(c0) > 0 {
+			sel := simd.select(c0, SCANNER_INDICES_128, SCANNER_SENTINEL_MIN_128)
+			off := simd.reduce_min(sel)
+			return i + int(off)
+		}
+
+		i += SIMD_REG_SIZE_128
+	}
+
+	// Scan serially for the remainder.
+	for /**/; i < l; i += 1 {
 		if s[i] == c {
 			return i
 		}
 	}
+
 	return -1
 }

-// Returns -1 if c is not present
-last_index_byte :: proc(s: []byte, c: byte) -> int {
-	for i := len(s)-1; i >= 0; i -= 1 {
+/*
+Scan a slice of bytes for a specific byte, starting from the end and working
+backwards to the start.
+
+This procedure safely handles slices of any length, including empty slices.
+
+Inputs:
+- data: A slice of bytes.
+- c: The byte to search for.
+
+Returns:
+- index: The index of the byte `c`, or -1 if it was not found.
+*/
+last_index_byte :: proc "contextless" (s: []byte, c: byte) -> int #no_bounds_check {
+	i := len(s)
+
+	// Guard against small strings.  On modern systems, it is ALWAYS
+	// worth vectorizing assuming there is a hardware vector unit, and
+	// the data size is large enough.
+	if i < SIMD_REG_SIZE_128 {
+		#reverse for ch, j in s {
+			if ch == c {
+				return j
+			}
+		}
+		return -1
+	}
+
+	c_vec: simd.u8x16 = c
+	when !simd.IS_EMULATED {
+		// Note: While this is something that could also logically take
+		// advantage of AVX512, the various downclocking and power
+		// consumption related woes make premature to have a dedicated
+		// code path.
+		when ODIN_ARCH == .amd64 && intrinsics.has_target_feature("avx2") {
+			c_vec_256: simd.u8x32 = c
+
+			s_vecs: [4]simd.u8x32 = ---
+			c_vecs: [4]simd.u8x32 = ---
+			m_vec: [4]u8 = ---
+
+			// Scan 128-byte chunks, using 256-bit SIMD.
+			for i >= 4 * SIMD_REG_SIZE_256 {
+				i -= 4 * SIMD_REG_SIZE_256
+
+				#unroll for j in 0..<4 {
+					s_vecs[j] = intrinsics.unaligned_load(cast(^simd.u8x32)raw_data(s[i+j*SIMD_REG_SIZE_256:]))
+					c_vecs[j] = simd.lanes_eq(s_vecs[j], c_vec_256)
+					m_vec[j] = simd.reduce_or(c_vecs[j])
+				}
+				if m_vec[0] | m_vec[1] | m_vec[2] | m_vec[3] > 0 {
+					#unroll for j in 0..<4 {
+						if m_vec[3-j] > 0 {
+							sel := simd.select(c_vecs[3-j], SCANNER_INDICES_256, SCANNER_SENTINEL_MAX_256)
+							off := simd.reduce_max(sel)
+							return i + (3-j) * SIMD_REG_SIZE_256 + int(off)
+						}
+					}
+				}
+			}
+
+			// Scan 64-byte chunks, using 256-bit SIMD.
+			for i >= 2 * SIMD_REG_SIZE_256 {
+				i -= 2 * SIMD_REG_SIZE_256
+
+				#unroll for j in 0..<2 {
+					s_vecs[j] = intrinsics.unaligned_load(cast(^simd.u8x32)raw_data(s[i+j*SIMD_REG_SIZE_256:]))
+					c_vecs[j] = simd.lanes_eq(s_vecs[j], c_vec_256)
+					m_vec[j] = simd.reduce_or(c_vecs[j])
+				}
+				if m_vec[0] | m_vec[1] > 0 {
+					#unroll for j in 0..<2 {
+						if m_vec[1-j] > 0 {
+							sel := simd.select(c_vecs[1-j], SCANNER_INDICES_256, SCANNER_SENTINEL_MAX_256)
+							off := simd.reduce_max(sel)
+							return i + (1-j) * SIMD_REG_SIZE_256 + int(off)
+						}
+					}
+				}
+			}
+		} else {
+			s_vecs: [4]simd.u8x16 = ---
+			c_vecs: [4]simd.u8x16 = ---
+			m_vecs: [4]u8 = ---
+
+			// Scan 64-byte chunks, using 128-bit SIMD.
+			for i >= 4 * SIMD_REG_SIZE_128 {
+				i -= 4 * SIMD_REG_SIZE_128
+
+				#unroll for j in 0..<4 {
+					s_vecs[j] = intrinsics.unaligned_load(cast(^simd.u8x16)raw_data(s[i+j*SIMD_REG_SIZE_128:]))
+					c_vecs[j] = simd.lanes_eq(s_vecs[j], c_vec)
+					m_vecs[j] = simd.reduce_or(c_vecs[j])
+				}
+				if m_vecs[0] | m_vecs[1] | m_vecs[2] | m_vecs[3] > 0 {
+					#unroll for j in 0..<4 {
+						if m_vecs[3-j] > 0 {
+							sel := simd.select(c_vecs[3-j], SCANNER_INDICES_128, SCANNER_SENTINEL_MAX_128)
+							off := simd.reduce_max(sel)
+							return i + (3-j) * SIMD_REG_SIZE_128 + int(off)
+						}
+					}
+				}
+			}
+		}
+	}
+
+	// Scan the remaining SIMD register sized chunks.
+	//
+	// Apparently LLVM does ok with 128-bit SWAR, so this path is also taken
+	// on potato targets.  Scanning more at a time when LLVM is emulating SIMD
+	// likely does not buy much, as all that does is increase GP register
+	// pressure.
+	for i >= SIMD_REG_SIZE_128 {
+		i -= SIMD_REG_SIZE_128
+
+		s0 := intrinsics.unaligned_load(cast(^simd.u8x16)raw_data(s[i:]))
+		c0 := simd.lanes_eq(s0, c_vec)
+		if simd.reduce_or(c0) > 0 {
+			sel := simd.select(c0, SCANNER_INDICES_128, SCANNER_SENTINEL_MAX_128)
+			off := simd.reduce_max(sel)
+			return i + int(off)
+		}
+	}
+
+	// Scan serially for the remainder.
+	for i > 0 {
+		i -= 1
 		if s[i] == c {
 			return i
 		}
 	}
+
 	return -1
 }


-
@private PRIME_RABIN_KARP :: 16777619

 index :: proc(s, substr: []byte) -> int {
@@ -638,7 +916,7 @@ trim_left_proc :: proc(s: []byte, p: proc(rune) -> bool) -> []byte {

 index_rune :: proc(s: []byte, r: rune) -> int {
 	switch {
-	case 0 <= r && r < utf8.RUNE_SELF:
+	case u32(r) < utf8.RUNE_SELF:
 		return index_byte(s, byte(r))

 	case r == utf8.RUNE_ERROR:
@@ -895,7 +1173,7 @@ 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
@@ -1167,3 +1445,28 @@ fields_proc :: proc(s: []byte, f: proc(rune) -> bool, allocator := context.alloc

 	return subslices[:]
 }
+
+// alias returns true iff a and b have a non-zero length, and any part of
+// a overlaps with b.
+alias :: proc "contextless" (a, b: []byte) -> bool {
+	a_len, b_len := len(a), len(b)
+	if a_len == 0 || b_len == 0 {
+		return false
+	}
+
+	a_start, b_start := uintptr(raw_data(a)), uintptr(raw_data(b))
+	a_end, b_end := a_start + uintptr(a_len-1), b_start + uintptr(b_len-1)
+
+	return a_start <= b_end && b_start <= a_end
+}
+
+// alias_inexactly returns true iff a and b have a non-zero length,
+// the base pointer of a and b are NOT equal, and any part of a overlaps
+// with b (ie: `alias(a, b)` with an exception that returns false for
+// `a == b`, `b = a[:len(a)-69]` and similar conditions).
+alias_inexactly :: proc "contextless" (a, b: []byte) -> bool {
+	if raw_data(a) == raw_data(b) {
+		return false
+	}
+	return alias(a, b)
+}
@@ -9,15 +9,16 @@ Reader :: struct {
 	prev_rune: int,    // previous reading index of rune or < 0
 }

-reader_init :: proc(r: ^Reader, s: []byte) {
+reader_init :: proc(r: ^Reader, s: []byte) -> io.Stream {
 	r.s = s
 	r.i = 0
 	r.prev_rune = -1
+	return reader_to_stream(r)
 }

 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
 }

@@ -33,6 +34,9 @@ reader_size :: proc(r: ^Reader) -> i64 {
 }

 reader_read :: proc(r: ^Reader, p: []byte) -> (n: int, err: io.Error) {
+	if len(p) == 0 {
+		return 0, nil
+	}
 	if r.i >= i64(len(r.s)) {
 		return 0, .EOF
 	}
@@ -42,6 +46,9 @@ reader_read :: proc(r: ^Reader, p: []byte) -> (n: int, err: io.Error) {
 	return
 }
 reader_read_at :: proc(r: ^Reader, p: []byte, off: i64) -> (n: int, err: io.Error) {
+	if len(p) == 0 {
+		return 0, nil
+	}
 	if off < 0 {
 		return 0, .Invalid_Offset
 	}
@@ -97,7 +104,6 @@ reader_unread_rune :: proc(r: ^Reader) -> io.Error {
 	return nil
 }
 reader_seek :: proc(r: ^Reader, offset: i64, whence: io.Seek_From) -> (i64, io.Error) {
-	r.prev_rune = -1
 	abs: i64
 	switch whence {
 	case .Start:
@@ -114,6 +120,7 @@ reader_seek :: proc(r: ^Reader, offset: i64, whence: io.Seek_From) -> (i64, io.E
 		return 0, .Invalid_Offset
 	}
 	r.i = abs
+	r.prev_rune = -1
 	return abs, nil
 }
 reader_write_to :: proc(r: ^Reader, w: io.Writer) -> (n: i64, err: io.Error) {
@@ -137,41 +144,22 @@ 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,
+}
@@ -519,7 +519,7 @@ join_adjacent_string_literals :: proc(cpp: ^Preprocessor, initial_tok: ^Token) {


 quote_string :: proc(s: string) -> []byte {
-	b := strings.make_builder(0, len(s)+2)
+	b := strings.builder_make(0, len(s)+2)
 	io.write_quoted_string(strings.to_writer(&b), s, '"')
 	return b.buf[:]
 }
@@ -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)
 	}
 }
@@ -291,7 +291,7 @@ scan_escape :: proc(t: ^Tokenizer) -> bool {
 		n -= 1
 	}

-	if x > max || 0xd800 <= x && x <= 0xe000 {
+	if x > max || 0xd800 <= x && x <= 0xdfff {
 		error_offset(t, offset, "escape sequence is an invalid Unicode code point")
 		return false
 	}
@@ -14,7 +14,7 @@ The following is a mostly-complete projection of the C11 standard library as def
 | `<inttypes.h>`    | Fully projected                                    |
 | `<iso646.h>`      | Not applicable, use Odin's operators               |
 | `<limits.h>`      | Not projected                                      |
-| `<locale.h>`      | Not projected                                      |
+| `<locale.h>`      | Fully projected                                    |
 | `<math.h>`        | Mostly projected, see [limitations](#Limitations)  |
 | `<setjmp.h>`      | Fully projected                                    |
 | `<signal.h>`      | Fully projected                                    |
@@ -70,4 +70,4 @@ with the following copyright.

 ```
 Copyright 2021 Dale Weiler <weilercdale@gmail.com>.
-```
+```
@@ -47,10 +47,10 @@ foreign libc {
 	clogf   :: proc(z: complex_float) -> complex_float ---

 	// 7.3.8 Power and absolute-value functions
-	cabs    :: proc(z: complex_double) -> complex_double ---
-	cabsf   :: proc(z: complex_float) -> complex_float ---
-	cpow    :: proc(z: complex_double) -> complex_double ---
-	cpowf   :: proc(z: complex_float) -> complex_float ---
+	cabs    :: proc(z: complex_double) -> double ---
+	cabsf   :: proc(z: complex_float) -> float ---
+	cpow    :: proc(x, y: complex_double) -> complex_double ---
+	cpowf   :: proc(x, y: complex_float) -> complex_float ---
 	csqrt   :: proc(z: complex_double) -> complex_double ---
 	csqrtf  :: proc(z: complex_float) -> complex_float ---

@@ -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
@@ -14,11 +14,11 @@ when ODIN_OS == .Windows {
 //	EDOM,
 //	EILSEQ
 //	ERANGE
-when ODIN_OS == .Linux || ODIN_OS == .FreeBSD {
+when ODIN_OS == .Linux {
 	@(private="file")
 	@(default_calling_convention="c")
 	foreign libc {
-		@(link_name="__libc_errno_location")
+		@(link_name="__errno_location")
 		_get_errno :: proc() -> ^int ---
 	}

@@ -27,7 +27,20 @@ when ODIN_OS == .Linux || ODIN_OS == .FreeBSD {
 	ERANGE :: 34
 }

-when ODIN_OS == .OpenBSD {
+when ODIN_OS == .FreeBSD {
+	@(private="file")
+	@(default_calling_convention="c")
+	foreign libc {
+		@(link_name="__error")
+		_get_errno :: proc() -> ^int ---
+	}
+
+	EDOM   :: 33
+	EILSEQ :: 84
+	ERANGE :: 34
+}
+
+when ODIN_OS == .OpenBSD || ODIN_OS == .NetBSD {
 	@(private="file")
 	@(default_calling_convention="c")
 	foreign libc {
@@ -67,10 +80,36 @@ 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
+}
+
+when ODIN_OS == .JS {
+	_ :: libc
+	_get_errno :: proc "c" () -> ^int {
+		@(static) errno: int
+		return &errno
+	}
+}
+
 // 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
 // it actually is.
-errno :: #force_inline proc() -> ^int {
+errno :: #force_inline proc "contextless" () -> ^int {
 	return _get_errno()
 }
@@ -0,0 +1,133 @@
+package libc
+
+import "core:c"
+
+when ODIN_OS == .Windows {
+	foreign import libc "system:libucrt.lib"
+} else when ODIN_OS == .Darwin {
+	foreign import libc "system:System.framework"
+} else {
+	foreign import libc "system:c"
+}
+
+// locale.h - category macros
+
+foreign libc {
+	/*
+	Sets the components of an object with the type lconv with the values appropriate for the
+	formatting of numeric quantities (monetary and otherwise) according to the rules of the current
+	locale.
+
+	Returns: a pointer to the lconv structure, might be invalidated by subsequent calls to localeconv() and setlocale()
+
+	[[ More; https://pubs.opengroup.org/onlinepubs/9699919799/functions/localeconv.html ]]
+	*/
+	localeconv :: proc() -> ^lconv ---
+
+	/*
+	Selects the appropriate piece of the global locale, as specified by the category and locale arguments,
+	and can be used to change or query the entire global locale or portions thereof.
+
+	Returns: the current locale if `locale` is `nil`, the set locale otherwise
+
+	[[ More; https://pubs.opengroup.org/onlinepubs/9699919799/functions/setlocale.html ]]
+	*/
+	@(link_name=LSETLOCALE)
+	setlocale :: proc(category: Locale_Category, locale: cstring) -> cstring ---
+}
+
+Locale_Category :: enum c.int {
+	ALL      = LC_ALL,
+	COLLATE  = LC_COLLATE,
+	CTYPE    = LC_CTYPE,
+	MESSAGES = LC_MESSAGES,
+	MONETARY = LC_MONETARY,
+	NUMERIC  = LC_NUMERIC,
+	TIME     = LC_TIME,
+}
+
+when ODIN_OS == .NetBSD {
+	@(private) LSETLOCALE :: "__setlocale50"
+} else {
+	@(private) LSETLOCALE :: "setlocale"
+}
+
+when ODIN_OS == .Windows {
+	lconv :: struct {
+		decimal_point:        cstring,
+		thousand_sep:         cstring,
+		grouping:             cstring,
+		int_curr_symbol:      cstring,
+		currency_symbol:      cstring,
+		mon_decimal_points:   cstring,
+		mon_thousands_sep:    cstring,
+		mon_grouping:         cstring,
+		positive_sign:        cstring,
+		negative_sign:        cstring,
+		int_frac_digits:      c.char,
+		frac_digits:          c.char,
+		p_cs_precedes:        c.char,
+		p_sep_by_space:       c.char,
+		n_cs_precedes:        c.char,
+		n_sep_by_space:       c.char,
+		p_sign_posn:          c.char,
+		n_sign_posn:          c.char,
+		_W_decimal_point:     [^]u16 `fmt:"s,0"`,
+		_W_thousands_sep:     [^]u16 `fmt:"s,0"`,
+		_W_int_curr_symbol:   [^]u16 `fmt:"s,0"`,
+		_W_currency_symbol:   [^]u16 `fmt:"s,0"`,
+		_W_mon_decimal_point: [^]u16 `fmt:"s,0"`,
+		_W_mon_thousands_sep: [^]u16 `fmt:"s,0"`,
+		_W_positive_sign:     [^]u16 `fmt:"s,0"`,
+		_W_negative_sign:     [^]u16 `fmt:"s,0"`,
+	}
+} else {
+	lconv :: struct {
+		decimal_point:       cstring,
+		thousand_sep:        cstring,
+		grouping:            cstring,
+		int_curr_symbol:     cstring,
+		currency_symbol:     cstring,
+		mon_decimal_points:  cstring,
+		mon_thousands_sep:   cstring,
+		mon_grouping:        cstring,
+		positive_sign:       cstring,
+		negative_sign:       cstring,
+		int_frac_digits:     c.char,
+		frac_digits:         c.char,
+		p_cs_precedes:       c.char,
+		p_sep_by_space:      c.char,
+		n_cs_precedes:       c.char,
+		n_sep_by_space:      c.char,
+		p_sign_posn:         c.char,
+		n_sign_posn:         c.char,
+		_int_p_cs_precedes:  c.char,
+		_int_n_cs_precedes:  c.char,
+		_int_p_sep_by_space: c.char,
+		_int_n_sep_by_space: c.char,
+		_int_p_sign_posn:    c.char,
+		_int_n_sign_posn:    c.char,
+	}
+}
+
+when ODIN_OS == .Darwin || ODIN_OS == .FreeBSD || ODIN_OS == .NetBSD  || ODIN_OS == .OpenBSD || ODIN_OS == .Windows {
+
+	LC_ALL      :: 0
+	LC_COLLATE  :: 1
+	LC_CTYPE    :: 2
+	LC_MESSAGES :: 6
+	LC_MONETARY :: 3
+	LC_NUMERIC  :: 4
+	LC_TIME     :: 5
+
+} else when ODIN_OS == .Linux {
+
+	LC_CTYPE    :: 0
+	LC_NUMERIC  :: 1
+	LC_TIME     :: 2
+	LC_COLLATE  :: 3
+	LC_MONETARY :: 4
+	LC_MESSAGES :: 5
+	LC_ALL      :: 6
+
+}
@@ -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"
@@ -331,7 +331,7 @@ fmin       :: proc{libc_fmin, libc_fminf}
 fma        :: proc{libc_fma, libc_fmaf}

 // But retain the 'f' suffix-variant functions as well so they can be used,
-// a trick is used here where we use explicit procedrual overloading of one
+// a trick is used here where we use explicit procedural overloading of one
 // procedure. This is done because the foreign block is marked @(private) and
 // aliasing functions does not remove privateness from the entity.
 acosf      :: proc{libc_acosf}
@@ -32,24 +32,21 @@ when ODIN_OS == .Windows {
 		// the RDX register will contain zero and correctly set the flag to disable
 		// stack unwinding.
 		@(link_name="_setjmp")
-		setjmp  :: proc(env: ^jmp_buf, hack: rawptr = nil) -> int ---
+		setjmp :: proc(env: ^jmp_buf, hack: rawptr = nil) -> int ---
 	}
 } else {
 	@(default_calling_convention="c")
 	foreign libc {
 		// 7.13.1 Save calling environment
-		//
-		// NOTE(dweiler): C11 requires setjmp be a macro, which means it won't
-		// necessarily export a symbol named setjmp but rather _setjmp in the case
-		// of musl, glibc, BSD libc, and msvcrt.
-		@(link_name="_setjmp")
-		setjmp  :: proc(env: ^jmp_buf) -> int ---
+		@(link_name=LSETJMP)
+		setjmp :: proc(env: ^jmp_buf) -> int ---
 	}
 }

@(default_calling_convention="c")
 foreign libc {
 	// 7.13.2 Restore calling environment
+	@(link_name=LLONGJMP)
 	longjmp :: proc(env: ^jmp_buf, val: int) -> ! ---
 }

@@ -63,4 +60,12 @@ 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, }
+
+when ODIN_OS == .NetBSD {
+	@(private) LSETJMP  :: "__setjmp14"
+	@(private) LLONGJMP :: "__longjmp14"
+} else {
+	@(private) LSETJMP  :: "setjmp"
+	@(private) LLONGJMP :: "longjmp"
+}
@@ -34,20 +34,7 @@ when ODIN_OS == .Windows {
 	SIGTERM :: 15
 }

-when ODIN_OS == .Linux || ODIN_OS == .FreeBSD {
-	SIG_ERR  :: rawptr(~uintptr(0))
-	SIG_DFL  :: rawptr(uintptr(0))
-	SIG_IGN  :: rawptr(uintptr(1)) 
-
-	SIGABRT  :: 6
-	SIGFPE   :: 8
-	SIGILL   :: 4
-	SIGINT   :: 2
-	SIGSEGV  :: 11
-	SIGTERM  :: 15
-}
-
-when ODIN_OS == .Darwin {
+when ODIN_OS == .Linux || ODIN_OS == .FreeBSD || ODIN_OS == .Haiku || ODIN_OS == .OpenBSD || ODIN_OS == .NetBSD || ODIN_OS == .Darwin {
 	SIG_ERR  :: rawptr(~uintptr(0))
 	SIG_DFL  :: rawptr(uintptr(0))
 	SIG_IGN  :: rawptr(uintptr(1)) 
@@ -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
@@ -235,7 +235,7 @@ atomic_compare_exchange_weak :: #force_inline proc(object, expected: ^$T, desire
 	return ok
 }

-atomic_compare_exchange_weak_explicit :: #force_inline proc(object, expected: ^$T, desited: T, success, failure: memory_order) -> bool {
+atomic_compare_exchange_weak_explicit :: #force_inline proc(object, expected: ^$T, desired: T, success, failure: memory_order) -> bool {
 	assert(failure != .release)
 	assert(failure != .acq_rel)

@@ -1,7 +1,12 @@
 package libc

+import "core:io"
+
 when ODIN_OS == .Windows {
-	foreign import libc "system:libucrt.lib"
+	foreign import libc {
+		"system:libucrt.lib",
+		"system:legacy_stdio_definitions.lib",
+	}
 } else when ODIN_OS == .Darwin {
 	foreign import libc "system:System.framework"
 } else {
@@ -12,6 +17,12 @@ when ODIN_OS == .Windows {

 FILE :: struct {}

+Whence :: enum int {
+	SET = SEEK_SET,
+	CUR = SEEK_CUR,
+	END = SEEK_END,
+}
+
 // MSVCRT compatible.
 when ODIN_OS == .Windows {
 	_IOFBF       :: 0x0000
@@ -78,7 +89,31 @@ when ODIN_OS == .Linux {
 	}
 }

-when ODIN_OS == .OpenBSD {
+when ODIN_OS == .JS {
+	fpos_t        :: struct #raw_union { _: [16]char, _: longlong, _: double, }
+
+	_IOFBF        :: 0
+	_IOLBF        :: 1
+	_IONBF        :: 2
+
+	BUFSIZ        :: 1024
+
+	EOF           :: int(-1)
+
+	FOPEN_MAX     :: 1000
+
+	FILENAME_MAX  :: 4096
+
+	L_tmpnam      :: 20
+
+	SEEK_SET      :: 0
+	SEEK_CUR      :: 1
+	SEEK_END      :: 2
+
+	TMP_MAX       :: 308915776
+}
+
+when ODIN_OS == .OpenBSD || ODIN_OS == .NetBSD {
 	fpos_t :: distinct i64

 	_IOFBF :: 0
@@ -96,11 +131,15 @@ when ODIN_OS == .OpenBSD {
 	SEEK_CUR :: 1
 	SEEK_END :: 2

+	TMP_MAX :: 308915776
+
 	foreign libc {
-		stderr: ^FILE
-		stdin:  ^FILE
-		stdout: ^FILE
+		__sF: [3]FILE
 	}
+
+	stdin:  ^FILE = &__sF[0]
+	stdout: ^FILE = &__sF[1]
+	stderr: ^FILE = &__sF[2]
 }

 when ODIN_OS == .FreeBSD {
@@ -121,10 +160,12 @@ when ODIN_OS == .FreeBSD {
 	SEEK_CUR :: 1
 	SEEK_END :: 2

+	TMP_MAX :: 308915776
+
 	foreign libc {
-		stderr: ^FILE
-		stdin:  ^FILE
-		stdout: ^FILE
+		@(link_name="__stderrp") stderr: ^FILE
+		@(link_name="__stdinp")  stdin:  ^FILE
+		@(link_name="__stdoutp") stdout: ^FILE
 	}
 }

@@ -158,10 +199,51 @@ 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
+	}
+}
+
+when ODIN_OS == .NetBSD {
+	@(private) LRENAME  :: "__posix_rename"
+	@(private) LFGETPOS :: "__fgetpos50"
+	@(private) LFSETPOS :: "__fsetpos50"
+} else {
+	@(private) LRENAME  :: "rename"
+	@(private) LFGETPOS :: "fgetpos"
+	@(private) LFSETPOS :: "fsetpos"
+}
+
@(default_calling_convention="c")
 foreign libc {
 	// 7.21.4 Operations on files
 	remove    :: proc(filename: cstring) -> int ---
+	@(link_name=LRENAME)
 	rename    :: proc(old, new: cstring) -> int ---
 	tmpfile   :: proc() -> ^FILE ---
 	tmpnam    :: proc(s: [^]char) -> [^]char ---
@@ -179,7 +261,7 @@ 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 ---
@@ -196,15 +278,17 @@ foreign libc {
 	getc      :: proc(stream: ^FILE) -> int ---
 	getchar   :: proc() -> int ---
 	putc      :: proc(c: int, stream: ^FILE) -> int ---
-	putchar   :: proc() -> int ---
+	putchar   :: proc(c: int) -> int ---
 	puts      :: proc(s: cstring) -> int ---
 	ungetc    :: proc(c: int, stream: ^FILE) -> int ---
 	fread     :: proc(ptr: rawptr, size: size_t, nmemb: size_t, stream: ^FILE) -> size_t ---
 	fwrite    :: proc(ptr: rawptr, size: size_t, nmemb: size_t, stream: ^FILE) -> size_t ---

 	// 7.21.9 File positioning functions
+	@(link_name=LFGETPOS)
 	fgetpos   :: proc(stream: ^FILE, pos: ^fpos_t) -> int ---
-	fseek     :: proc(stream: ^FILE, offset: long, whence: int) -> int ---
+	fseek     :: proc(stream: ^FILE, offset: long, whence: Whence) -> int ---
+	@(link_name=LFSETPOS)
 	fsetpos   :: proc(stream: ^FILE, pos: ^fpos_t) -> int ---
 	ftell     :: proc(stream: ^FILE) -> long ---
 	rewind    :: proc(stream: ^FILE) ---
@@ -215,3 +299,106 @@ 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), .SET) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+			defer fseek(file, long(curr), .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), .SET) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+			defer fseek(file, long(curr), .SET)
+
+			n = i64(fwrite(raw_data(p), size_of(byte), len(p), file))
+			if n == 0 { err = unknown_or_eof(file) }
+
+		case .Seek:
+			#assert(int(Whence.SET) == int(io.Seek_From.Start))
+			#assert(int(Whence.CUR) == int(io.Seek_From.Current))
+			#assert(int(Whence.END) == int(io.Seek_From.End))
+
+			if fseek(file, long(offset), Whence(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, .SET)
+
+			if fseek(file, 0, .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, .Query })
+		}
+		return
+	}
+
+	return {
+		data      = file,
+		procedure = stream_proc,
+	}
+}
@@ -10,6 +10,9 @@ when ODIN_OS == .Windows {
 	foreign import libc "system:c"
 }

+@(require)
+import "base:runtime"
+
 when ODIN_OS == .Windows {
 	RAND_MAX :: 0x7fff

@@ -40,10 +43,9 @@ when ODIN_OS == .Linux {
 }


-when ODIN_OS == .Darwin {
+when ODIN_OS == .Darwin || ODIN_OS == .FreeBSD || ODIN_OS == .OpenBSD {
 	RAND_MAX :: 0x7fffffff

-	// GLIBC and MUSL only
 	@(private="file")
 	@(default_calling_convention="c")
 	foreign libc {
@@ -55,6 +57,20 @@ when ODIN_OS == .Darwin {
 	}
 }

+when ODIN_OS == .NetBSD {
+	RAND_MAX :: 0x7fffffff
+
+	@(private="file")
+	@(default_calling_convention="c")
+	foreign libc {
+		__mb_cur_max: size_t
+	}
+
+	MB_CUR_MAX :: #force_inline proc() -> size_t {
+		return __mb_cur_max
+	}
+}
+
 // C does not declare what these values should be, as an implementation is free
 // to use any two distinct values it wants to indicate success or failure.
 // However, nobody actually does and everyone appears to have agreed upon these
@@ -88,7 +104,6 @@ foreign libc {
 	srand         :: proc(seed: uint) ---

 	// 7.22.3 Memory management functions
-	aligned_alloc :: proc(aligment, size: size_t) -> rawptr ---
 	calloc        :: proc(nmemb, size: size_t) -> rawptr ---
 	free          :: proc(ptr: rawptr) ---
 	malloc        :: proc(size: size_t) -> rawptr ---
@@ -100,7 +115,7 @@ foreign libc {
 	at_quick_exit :: proc(func: proc "c" ()) -> int ---
 	exit          :: proc(status: int) -> ! ---
 	_Exit         :: proc(status: int) -> ! ---
-	getenv        :: proc(name: cstring) -> [^]char ---
+	getenv        :: proc(name: cstring) -> cstring ---
 	quick_exit    :: proc(status: int) -> ! ---
 	system        :: proc(cmd: cstring) -> int ---

@@ -125,3 +140,37 @@ foreign libc {
 	mbstowcs      :: proc(pwcs: ^wchar_t, s: cstring, n: size_t) -> size_t ---
 	wcstombs      :: proc(s: [^]char, pwcs: ^wchar_t, n: size_t) -> size_t ---
 }
+
+
+aligned_alloc :: #force_inline proc "c" (alignment, size: size_t) -> rawptr {
+	when ODIN_OS == .Windows {
+		foreign libc {
+			_aligned_malloc :: proc(size, alignment: size_t) -> rawptr ---
+		}
+		return _aligned_malloc(size=size, alignment=alignment)
+	} else when ODIN_ARCH == .wasm32 || ODIN_ARCH == .wasm64p32 {
+		context = runtime.default_context()
+		data, _ := runtime.mem_alloc_bytes(auto_cast size, auto_cast alignment)
+		return raw_data(data)
+	} else {
+		foreign libc {
+			aligned_alloc :: proc(alignment, size: size_t) -> rawptr ---
+		}
+		return aligned_alloc(alignment=alignment, size=size)
+	}
+}
+
+
+aligned_free :: #force_inline proc "c" (ptr: rawptr) {
+	when ODIN_OS == .Windows {
+		foreign libc {
+			_aligned_free :: proc(ptr: rawptr) ---
+		}
+		_aligned_free(ptr)
+	} else when ODIN_ARCH == .wasm32 || ODIN_ARCH == .wasm64p32 {
+		context = runtime.default_context()
+		runtime.mem_free(ptr)
+	} else {
+		free(ptr)
+	}
+}
@@ -1,6 +1,6 @@
 package libc

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

 // 7.24 String handling

@@ -12,6 +12,7 @@ when ODIN_OS == .Windows {
 	foreign import libc "system:c"
 }

+@(default_calling_convention="c")
 foreign libc {
 	// 7.24.2 Copying functions
 	memcpy   :: proc(s1, s2: rawptr, n: size_t) -> rawptr ---
@@ -40,7 +41,7 @@ foreign libc {
 	strtok   :: proc(s1: [^]char, s2: cstring) -> [^]char ---

 	// 7.24.6 Miscellaneous functions
-	strerror :: proc(errnum: int) -> [^]char ---
+	strerror :: proc(errnum: int) -> cstring ---
 	strlen   :: proc(s: cstring) -> size_t ---
 }
 memset :: proc "c" (s: rawptr, c: int, n: size_t) -> rawptr {
@@ -44,7 +44,7 @@ when ODIN_OS == .Windows {
 		@(link_name="_Cnd_destroy")   cnd_destroy   :: proc(cond: ^cnd_t) ---
 		@(link_name="_Cnd_init")      cnd_init      :: proc(cond: ^cnd_t) -> int ---
 		@(link_name="_Cnd_signal")    cnd_signal    :: proc(cond: ^cnd_t) -> int ---
-		@(link_name="_Cnd_timedwait") cnd_timedwait :: proc(cond: ^cnd_t, ts: ^timespec) -> int ---
+		@(link_name="_Cnd_timedwait") cnd_timedwait :: proc(cond: ^cnd_t, mtx: ^mtx_t, ts: ^timespec) -> int ---
 		@(link_name="_Cnd_wait")      cnd_wait      :: proc(cond: ^cnd_t, mtx: ^mtx_t) -> int ---
 		
 		// 7.26.4 Mutex functions
@@ -108,7 +108,7 @@ when ODIN_OS == .Linux {
 		cnd_destroy   :: proc(cond: ^cnd_t) ---
 		cnd_init      :: proc(cond: ^cnd_t) -> int ---
 		cnd_signal    :: proc(cond: ^cnd_t) -> int ---
-		cnd_timedwait :: proc(cond: ^cnd_t, ts: ^timespec) -> int ---
+		cnd_timedwait :: proc(cond: ^cnd_t, mtx: ^mtx_t, ts: ^timespec) -> int ---
 		cnd_wait      :: proc(cond: ^cnd_t, mtx: ^mtx_t) -> int ---
 		
 		// 7.26.4 Mutex functions
@@ -35,45 +35,71 @@ 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 == .NetBSD || ODIN_OS == .Haiku || ODIN_OS == .JS  {
 	@(default_calling_convention="c")
 	foreign libc {
 		// 7.27.2 Time manipulation functions
 		clock        :: proc() -> clock_t ---
+		@(link_name=LDIFFTIME)
 		difftime     :: proc(time1, time2: time_t) -> double ---
+		@(link_name=LMKTIME)
 		mktime       :: proc(timeptr: ^tm) -> time_t ---
+		@(link_name=LTIME)
 		time         :: proc(timer: ^time_t) -> time_t ---
 		timespec_get :: proc(ts: ^timespec, base: int) -> int ---

 		// 7.27.3 Time conversion functions
 		asctime      :: proc(timeptr: ^tm) -> [^]char ---
+		@(link_name=LCTIME)
 		ctime        :: proc(timer: ^time_t) -> [^]char ---
+		@(link_name=LGMTIME)
 		gmtime       :: proc(timer: ^time_t) -> ^tm ---
+		@(link_name=LLOCALTIME)
 		localtime    :: proc(timer: ^time_t) -> ^tm ---
 		strftime     :: proc(s: [^]char, maxsize: size_t, format: cstring, timeptr: ^tm) -> size_t ---
 	}

+	when ODIN_OS == .NetBSD {
+		@(private) LDIFFTIME  :: "__difftime50"
+		@(private) LMKTIME    :: "__mktime50"
+		@(private) LTIME      :: "__time50"
+		@(private) LCTIME     :: "__ctime50"
+		@(private) LGMTIME    :: "__gmtime50"
+		@(private) LLOCALTIME :: "__localtime50"
+	} else {
+		@(private) LDIFFTIME  :: "difftime"
+		@(private) LMKTIME    :: "mktime"
+		@(private) LTIME      :: "time"
+		@(private) LCTIME     :: "ctime"
+		@(private) LGMTIME    :: "gmtime"
+		@(private) LLOCALTIME :: "localtime"
+	}
+
 	when ODIN_OS == .OpenBSD {
 		CLOCKS_PER_SEC :: 100
 	} else {
 		CLOCKS_PER_SEC :: 1000000
 	}

-	TIME_UTC       :: 1
+	TIME_UTC :: 1

-	time_t         :: distinct i64
+	time_t :: distinct i64

-	clock_t        :: long
+	when ODIN_OS == .FreeBSD || ODIN_OS == .NetBSD {
+		clock_t :: distinct int32_t
+	} else {
+		clock_t :: distinct long
+	}

 	timespec :: struct {
 		tv_sec:  time_t,
@@ -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.
@@ -14,7 +14,7 @@ when ODIN_OS == .Windows {
 	wctrans_t :: distinct wchar_t
 	wctype_t  :: distinct ushort

-} else when ODIN_OS == .Linux {
+} else when ODIN_OS == .Linux || ODIN_OS == .JS {
 	wctrans_t :: distinct intptr_t
 	wctype_t  :: distinct ulong

@@ -22,14 +22,18 @@ when ODIN_OS == .Windows {
 	wctrans_t :: distinct int
 	wctype_t  :: distinct u32

-} else when ODIN_OS == .OpenBSD {
+} else when ODIN_OS == .OpenBSD || ODIN_OS == .NetBSD {
 	wctrans_t :: distinct rawptr
 	wctype_t  :: distinct rawptr

 } 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))

 /*
@@ -34,15 +33,13 @@ COMPRESS_OUTPUT_ALLOCATE_MIN :: int(#config(COMPRESS_OUTPUT_ALLOCATE_MIN, 1 << 2

 */
 when size_of(uintptr) == 8 {
-	/*
-		For 64-bit platforms, we set the default max buffer size to 4 GiB,
-		which is GZIP and PKZIP's max payload size.
-	*/	
+
+	// For 64-bit platforms, we set the default max buffer size to 4 GiB,
+	// which is GZIP and PKZIP's max payload size.
 	COMPRESS_OUTPUT_ALLOCATE_MAX :: int(#config(COMPRESS_OUTPUT_ALLOCATE_MAX, 1 << 32))
 } else {
-	/*
-		For 32-bit platforms, we set the default max buffer size to 512 MiB.
-	*/
+	
+	// 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,
@@ -188,12 +180,13 @@ 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}

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_slice_from_memory :: #force_inline proc(z: ^Context_Memory_Input, size: int) -> (res: []u8, err: io.Error) {
 	#no_bounds_check {
 		if len(z.input_data) >= size {
@@ -210,30 +203,24 @@ read_slice_from_memory :: #force_inline proc(z: ^Context_Memory_Input, size: int
 	}
 }

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_slice_from_stream :: #force_inline proc(z: ^Context_Stream_Input, size: int) -> (res: []u8, err: io.Error) {
+	// TODO: REMOVE ALL USE OF context.temp_allocator here
+	// there is literally no need for it
 	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")
+@(optimization_mode="favor_size")
 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")
+@(optimization_mode="favor_size")
 read_u8_from_memory :: #force_inline proc(z: ^Context_Memory_Input) -> (res: u8, err: io.Error) {
 	#no_bounds_check {
 		if len(z.input_data) >= 1 {
@@ -245,23 +232,17 @@ read_u8_from_memory :: #force_inline proc(z: ^Context_Memory_Input) -> (res: u8,
 	return 0, .EOF
 }

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_u8_from_stream :: #force_inline proc(z: ^Context_Stream_Input) -> (res: u8, err: io.Error) {
-	b, 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.
-*/
-@(optimization_mode="speed")
+// You would typically only use this at the end of Inflate, to drain bits from the code buffer
+// preferentially.
+@(optimization_mode="favor_size")
 read_u8_prefer_code_buffer_lsb :: #force_inline proc(z: ^$C) -> (res: u8, err: io.Error) {
 	if z.num_bits >= 8 {
 		res = u8(read_bits_no_refill_lsb(z, 8))
@@ -276,7 +257,7 @@ read_u8_prefer_code_buffer_lsb :: #force_inline proc(z: ^$C) -> (res: u8, err: i
 	return
 }

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_data_from_memory :: #force_inline proc(z: ^Context_Memory_Input, $T: typeid) -> (res: T, err: io.Error) {
 	size :: size_of(T)

@@ -294,17 +275,32 @@ peek_data_from_memory :: #force_inline proc(z: ^Context_Memory_Input, $T: typeid
 	}
 }

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
+peek_data_at_offset_from_memory :: #force_inline proc(z: ^Context_Memory_Input, $T: typeid, #any_int offset: int) -> (res: T, err: io.Error) {
+	size :: size_of(T)
+
+	#no_bounds_check {
+		if len(z.input_data) >= size + offset {
+			buf := z.input_data[offset:][:size]
+			return (^T)(&buf[0])^, .None
+		}
+	}
+
+	if len(z.input_data) == 0 {
+		return T{}, .EOF
+	} else {
+		return T{}, .Short_Buffer
+	}
+}
+
+@(optimization_mode="favor_size")
 peek_data_from_stream :: #force_inline proc(z: ^Context_Stream_Input, $T: typeid) -> (res: T, err: io.Error) {
 	size :: size_of(T)

 	// 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 {
@@ -312,8 +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
 	}

@@ -321,19 +317,49 @@ peek_data_from_stream :: #force_inline proc(z: ^Context_Stream_Input, $T: typeid
 	return res, .None
 }

-peek_data :: proc{peek_data_from_memory, peek_data_from_stream}
+@(optimization_mode="favor_size")
+peek_data_at_offset_from_stream :: #force_inline proc(z: ^Context_Stream_Input, $T: typeid, #any_int offset: int) -> (res: T, err: io.Error) {
+	size :: size_of(T)
+
+	// Get current position to return to.
+	cur_pos := z.input->impl_seek(0, .Current) or_return
+	// Seek to offset.
+	pos := z.input->impl_seek(offset, .Start) or_return
+
+	r, e3 := io.to_reader_at(z.input)
+	if !e3 {
+		return T{}, .Empty
+	}
+	when size <= 128 {
+		b: [size]u8
+	} else {
+		b := make([]u8, size, context.temp_allocator)
+	}
+	_, e4 := io.read_at(r, b[:], pos)
+	if e4 != .None {
+		return T{}, .Empty
+	}
+
+	// Return read head to original position.
+	z.input->impl_seek(cur_pos, .Start)
+
+	res = (^T)(&b[0])^
+	return res, .None
+}
+
+peek_data :: proc{peek_data_from_memory, peek_data_from_stream, peek_data_at_offset_from_memory, peek_data_at_offset_from_stream}



 // Sliding window read back
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_back_byte :: #force_inline proc(z: ^$C, offset: i64) -> (res: u8, err: io.Error) {
 	// Look back into the sliding window.
 	return z.output.buf[z.bytes_written - offset], .None
 }

 // Generalized bit reader LSB
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 refill_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width := i8(48)) {
 	refill := u64(width)
 	b      := u64(0)
@@ -359,7 +385,7 @@ refill_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width :=
 }

 // Generalized bit reader LSB
-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 refill_lsb_from_stream :: proc(z: ^Context_Stream_Input, width := i8(24)) {
 	refill := u64(width)

@@ -388,13 +414,13 @@ refill_lsb_from_stream :: proc(z: ^Context_Stream_Input, width := i8(24)) {
 refill_lsb :: proc{refill_lsb_from_memory, refill_lsb_from_stream}


-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 consume_bits_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width: u8) {
 	z.code_buffer >>= width
 	z.num_bits -= u64(width)
 }

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 consume_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width: u8) {
 	z.code_buffer >>= width
 	z.num_bits -= u64(width)
@@ -402,46 +428,46 @@ consume_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, wid

 consume_bits_lsb :: proc{consume_bits_lsb_from_memory, consume_bits_lsb_from_stream}

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_bits_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width: u8) -> u32 {
 	if z.num_bits < u64(width) {
 		refill_lsb(z)
 	}
-	return u32(z.code_buffer & ~(~u64(0) << width))
+	return u32(z.code_buffer &~ (~u64(0) << width))
 }

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 peek_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width: u8) -> u32 {
 	if z.num_bits < u64(width) {
 		refill_lsb(z)
 	}
-	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}

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

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

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_bits_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width: u8) -> u32 {
 	k := #force_inline peek_bits_lsb(z, width)
 	#force_inline consume_bits_lsb(z, width)
 	return k
 }

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width: u8) -> u32 {
 	k := peek_bits_lsb(z, width)
 	consume_bits_lsb(z, width)
@@ -450,14 +476,14 @@ read_bits_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width:

 read_bits_lsb :: proc{read_bits_lsb_from_memory, read_bits_lsb_from_stream}

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_bits_no_refill_lsb_from_memory :: #force_inline proc(z: ^Context_Memory_Input, width: u8) -> u32 {
 	k := #force_inline peek_bits_no_refill_lsb(z, width)
 	#force_inline consume_bits_lsb(z, width)
 	return k
 }

-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 read_bits_no_refill_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Input, width: u8) -> u32 {
 	k := peek_bits_no_refill_lsb(z, width)
 	consume_bits_lsb(z, width)
@@ -467,14 +493,14 @@ read_bits_no_refill_lsb_from_stream :: #force_inline proc(z: ^Context_Stream_Inp
 read_bits_no_refill_lsb :: proc{read_bits_no_refill_lsb_from_memory, read_bits_no_refill_lsb_from_stream}


-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 discard_to_next_byte_lsb_from_memory :: proc(z: ^Context_Memory_Input) {
 	discard := u8(z.num_bits & 7)
 	#force_inline consume_bits_lsb(z, discard)
 }


-@(optimization_mode="speed")
+@(optimization_mode="favor_size")
 discard_to_next_byte_lsb_from_stream :: proc(z: ^Context_Stream_Input) {
 	discard := u8(z.num_bits & 7)
 	consume_bits_lsb(z, discard)
--- a/Show More
+++ b/Show More