Move Tracking_Allocator to its own file.

Make file_console_logger.odin be ignored on freestanding targets
Merge pull request #2978 from thetarnav/patch-2
2026-07-05 11:11:37 -07:00 · 2024-02-01 13:01:38 +00:00 · 2024-01-31 22:01:35 +00:00 · 2024-01-31 21:02:03 +00:00 · 2024-01-31 20:49:30 +00:00 · 2024-01-31 21:38:05 +01:00
1388 changed files with 510035 additions and 104492 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 +1,2 @@
 *.odin linguist-language=Odin
+* text=auto
@@ -1,3 +1,4 @@
 # These are supported funding model platforms

+github: odin-lang
 patreon: gingerbill
@@ -11,8 +11,8 @@ assignees: ''

 Please provide any relevant information about your setup. This is important in case the issue is not reproducible except for under certain conditions.

-* Operating System:
-* Please paste `odin version` output:
+* Operating System & Odin Version:
+* Please paste `odin report` output:

 ## Expected Behavior

@@ -7,6 +7,8 @@ assignees: ''

 ---

+# PLEASE POST THIS IN THE DISCUSSION TAB UNDER "PROPOSALS" OR "IDEAS/REQUESTS"
+
 **Is your feature request related to a problem? Please describe.**
 A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]

@@ -1,64 +1,180 @@
 name: CI
-on: [push, pull_request]
+on: [push, pull_request, workflow_dispatch]

 jobs:
  build_linux:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v1
-      - name: Download LLVM
-        run: sudo apt-get install llvm-11 clang-11 llvm
+      - name: Download LLVM, botan
+        run: sudo apt-get install llvm-11 clang-11 libbotan-2-dev botan
      - name: build odin
-        run: make release
-      - name: Odin run -llvm-api
-        run: ./odin run examples/demo/demo.odin -llvm-api
-      - name: Odin run
-        run: ./odin run examples/demo/demo.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/demo.odin -vet
+        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 internals tests
+        run: |
+          cd tests/internal
+          make
+        timeout-minutes: 10
+      - 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 and setup PATH
+      - name: Download LLVM, botan and setup PATH
        run: |
-          brew install llvm@11
-          echo "/usr/local/opt/llvm/bin" >> $GITHUB_PATH
+          brew install llvm@13 botan
+          echo "/usr/local/opt/llvm@13/bin" >> $GITHUB_PATH
          TMP_PATH=$(xcrun --show-sdk-path)/user/include
          echo "CPATH=$TMP_PATH" >> $GITHUB_ENV
      - name: build odin
-        run: make release
-      - name: Odin run
-        run: |
-          ./odin run examples/demo/demo.odin
-          ./odin run examples/demo/demo.odin -llvm-api
+        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/demo.odin -vet
+        run: ./odin check examples/demo -vet
+        timeout-minutes: 10
+      - name: Odin run
+        run: ./odin run examples/demo
+        timeout-minutes: 10
+      - name: Odin run -debug
+        run: ./odin run examples/demo -debug
+        timeout-minutes: 10
+      - name: Odin check examples/all
+        run: ./odin check examples/all -strict-style
+        timeout-minutes: 10
+      - name: Core library tests
+        run: |
+          cd tests/core
+          make
+        timeout-minutes: 10
+      - name: Odin internals tests
+        run: |
+          cd tests/internal
+          make
+        timeout-minutes: 10
+      - name: Odin check examples/all for Darwin arm64
+        run: ./odin check examples/all -vet -strict-style -target:darwin_arm64
+        timeout-minutes: 10
+      - name: Odin check examples/all for Linux arm64
+        run: ./odin check examples/all -vet -strict-style -target:linux_arm64
+        timeout-minutes: 10
  build_windows:
-    runs-on: windows-latest
+    runs-on: windows-2022
    steps:
      - uses: actions/checkout@v1
-      - name: Download and unpack LLVM bins
-        shell: powershell
-        run: |
-          cd bin
-          $ProgressPreference = "SilentlyContinue";
-          Invoke-WebRequest -Uri https://github.com/odin-lang/Odin/releases/download/llvm-windows/llvm-binaries.zip -OutFile llvm-binaries.zip
-          7z x llvm-binaries.zip > $null
      - 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 run
-        shell: cmd
-        run: |
-          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat
-          odin run examples/demo/demo.odin
+      - 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
-          odin check examples/demo/demo.odin -vet
-
-
+          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\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\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          odin run examples/demo -debug
+        timeout-minutes: 10
+      - name: Odin check examples/all
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          odin check examples/all -strict-style
+        timeout-minutes: 10
+      - name: Core library tests
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          cd tests\core
+          call build.bat
+        timeout-minutes: 10
+      - name: Vendor library tests
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          cd tests\vendor
+          call build.bat
+        timeout-minutes: 10
+      - name: Odin internals tests
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          cd tests\internal
+          call build.bat
+        timeout-minutes: 10
+      - name: Odin documentation tests
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          cd tests\documentation
+          rem call build.bat
+        timeout-minutes: 10
+      - name: core:math/big tests
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          cd tests\core\math\big
+          call build.bat
+        timeout-minutes: 10
+      - name: Odin check examples/all for Windows 32bits
+        shell: cmd
+        run: |
+          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat
+          odin check examples/all -strict-style -target:windows_i386
+        timeout-minutes: 10
@@ -7,36 +7,31 @@ on:

 jobs:
  build_windows:
-    runs-on: windows-latest
+    if: github.repository == 'odin-lang/Odin'
+    runs-on: windows-2022
    steps:
      - uses: actions/checkout@v1
-      - name: Install cURL
-        run: choco install curl
-      - name: Download and unpack LLVM bins
-        shell: cmd
-        run: |
-          cd bin
-          curl -sL https://github.com/odin-lang/Odin/releases/download/llvm-windows/llvm-binaries.zip --output llvm-binaries.zip
-          7z x llvm-binaries.zip > nul
-          rm -f llvm-binaries.zip
      - 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
-          odin run examples/demo/demo.odin
+          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
@@ -45,23 +40,26 @@ jobs:
          name: windows_artifacts
          path: dist
  build_ubuntu:
+    if: github.repository == 'odin-lang/Odin'
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v1
      - name: (Linux) Download LLVM
-        run: sudo apt-get install llvm-11 clang-11 llvm
+        run: sudo apt-get install llvm-11 clang-11
      - name: build odin
        run: make nightly
-      - name: Odin run -llvm-api
-        run: ./odin run examples/demo/demo.odin -llvm-api
      - name: Odin run
-        run: ./odin run examples/demo/demo.odin
+        run: ./odin run examples/demo
      - name: Copy artifacts
        run: |
          mkdir dist
          cp odin dist
+          cp LICENSE dist
+          cp libLLVM* dist
          cp -r shared dist
+          cp -r base dist
          cp -r core dist
+          cp -r vendor dist
          cp -r examples dist
      - name: Upload artifact
        uses: actions/upload-artifact@v1
@@ -69,27 +67,29 @@ jobs:
          name: ubuntu_artifacts
          path: dist
  build_macos:
+    if: github.repository == 'odin-lang/Odin'
    runs-on: macOS-latest
    steps:
      - uses: actions/checkout@v1
      - name: Download LLVM and setup PATH
        run: |
-          brew install llvm
-          echo "/usr/local/opt/llvm/bin" >> $GITHUB_PATH
+          brew install llvm@13
+          echo "/usr/local/opt/llvm@13/bin" >> $GITHUB_PATH
          TMP_PATH=$(xcrun --show-sdk-path)/user/include
          echo "CPATH=$TMP_PATH" >> $GITHUB_ENV
      - name: build odin
        run: make nightly
-      - name: Odin run -llvm-api
-        run: ./odin run examples/demo/demo.odin -llvm-api
      - name: Odin run
-        run: ./odin run examples/demo/demo.odin
+        run: ./odin run examples/demo
      - name: Copy artifacts
        run: |
          mkdir dist
          cp odin dist
+          cp LICENSE dist
          cp -r shared dist
+          cp -r base dist
          cp -r core dist
+          cp -r vendor dist
          cp -r examples dist
      - name: Upload artifact
        uses: actions/upload-artifact@v1
@@ -103,7 +103,7 @@ jobs:
      - uses: actions/checkout@v1
      - uses: actions/setup-python@v2
        with:
-          python-version: '3.x'
+          python-version: '3.8.x'

      - name: Install B2 CLI
        shell: bash
@@ -137,16 +137,23 @@ 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
@@ -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"
@@ -7,6 +7,9 @@
 # User-specific files (MonoDevelop/Xamarin Studio)
 *.userprefs

+# For macOS
+.DS_Store
+
 # Build results
 [Dd]ebug/
 [Dd]ebugPublic/
@@ -19,6 +22,34 @@ bld/
 [Oo]bj/
 [Ll]og/
 ![Cc]ore/[Ll]og/
+tests/documentation/verify/
+tests/documentation/all.odin-doc
+tests/internal/test_map
+tests/internal/test_pow
+tests/internal/test_rtti
+tests/core/test_core_compress
+tests/core/test_core_filepath
+tests/core/test_core_fmt
+tests/core/test_core_i18n
+tests/core/test_core_image
+tests/core/test_core_libc
+tests/core/test_core_match
+tests/core/test_core_math
+tests/core/test_core_net
+tests/core/test_core_os_exit
+tests/core/test_core_reflect
+tests/core/test_core_strings
+tests/core/test_crypto_hash
+tests/core/test_hash
+tests/core/test_hxa
+tests/core/test_json
+tests/core/test_linalg_glsl_math
+tests/core/test_noise
+tests/core/test_varint
+tests/core/test_xml
+tests/core/test_core_slice
+tests/core/test_core_thread
+tests/vendor/vendor_botan
 # Visual Studio 2015 cache/options directory
 .vs/
 # Visual Studio Code options directory
@@ -265,7 +296,11 @@ bin/

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

 # shared collection
 shared/
@@ -276,3 +311,9 @@ shared/
 *.ll

 *.sublime-workspace
+examples/bug/
+build.sh
+!core/debug/
+
+# RAD debugger project file
+*.raddbg
@@ -1,14 +1,18 @@
-Copyright (c) 2016-2020 Ginger Bill. All rights reserved.
+Copyright (c) 2016-2022 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:

-* Redistributions of source code must retain the above copyright notice, this
-  list of conditions and the following disclaimer.
+1. Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.

-* Redistributions in binary form must reproduce the above copyright notice,
-  this list of conditions and the following disclaimer in the documentation
-  and/or other materials provided with the distribution.
+2. Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+3. Neither the name of the copyright holder nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.

 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
@@ -1,34 +1,19 @@
-GIT_SHA=$(shell git rev-parse --short HEAD)
-DISABLED_WARNINGS=-Wno-switch -Wno-pointer-sign -Wno-tautological-constant-out-of-range-compare -Wno-tautological-compare -Wno-macro-redefined
-LDFLAGS=-pthread -ldl -lm -lstdc++
-CFLAGS=-std=c++14 -DGIT_SHA=\"$(GIT_SHA)\"
-CC=clang
-
-OS=$(shell uname)
-
-ifeq ($(OS), Darwin)
-	LDFLAGS:=$(LDFLAGS) -liconv
-	CFLAGS:=$(CFLAGS) $(shell llvm-config --cxxflags --ldflags) -DLLVM_BACKEND_SUPPORT -DUSE_NEW_LLVM_ABI_SYSTEM
-	LDFLAGS:=$(LDFLAGS) -lLLVM-C
-endif
-ifeq ($(OS), Linux)
-	CFLAGS:=$(CFLAGS) $(shell llvm-config-11 --cxxflags --ldflags) -DLLVM_BACKEND_SUPPORT -DUSE_NEW_LLVM_ABI_SYSTEM
-	LDFLAGS:=$(LDFLAGS) $(shell llvm-config-11 --libs core native --system-libs)
-endif
-
-all: debug demo
+all: debug

 demo:
-	./odin run examples/demo/demo.odin
+	./odin run examples/demo/demo.odin -file
+
+report:
+	./odin report

 debug:
-	$(CC) src/main.cpp $(DISABLED_WARNINGS) $(CFLAGS) -g $(LDFLAGS) -o odin
+	./build_odin.sh debug

 release:
-	$(CC) src/main.cpp $(DISABLED_WARNINGS) $(CFLAGS) -O3 -march=native $(LDFLAGS) -o odin
+	./build_odin.sh release
+
+release_native:
+	./build_odin.sh release-native

 nightly:
-	$(CC) src/main.cpp $(DISABLED_WARNINGS) $(CFLAGS) -DNIGHTLY -O3 $(LDFLAGS) -o odin
-
-
-
+	./build_odin.sh nightly
@@ -1,7 +1,7 @@
 <p align="center">
-    <img src="misc/logo-slim.png" alt="Odin logo" height="120">
+    <img src="misc/logo-slim.png" alt="Odin logo" style="width:65%">
    <br/>
-    A fast, concise, readable, pragmatic and open sourced programming language.
+   The Data-Oriented Language for Sane Software Development.
    <br/>
    <br/>
    <a href="https://github.com/odin-lang/odin/releases/latest">
@@ -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/hnwN2Rj">
+    <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">
@@ -21,11 +21,8 @@

 # The Odin Programming Language

-The Odin programming language is fast, concise, readable, pragmatic and open sourced. It is designed with the intent of creating an alternative to C with the following goals:
-* simplicity
-* high performance
-* built for modern systems
-* joy of programming
+
+Odin is a general-purpose programming language with distinct typing, built for high performance, modern systems, and built-in data-oriented data types. The Odin Programming Language, the C alternative for the joy of programming.

 Website: [https://odin-lang.org/](https://odin-lang.org/)

@@ -35,33 +32,36 @@ package main
 import "core:fmt"

 main :: proc() {
-	program := "+ + * 😃 - /";
-	accumulator := 0;
+	program := "+ + * 😃 - /"
+	accumulator := 0

 	for token in program {
 		switch token {
-		case '+': accumulator += 1;
-		case '-': accumulator -= 1;
-		case '*': accumulator *= 2;
-		case '/': accumulator /= 2;
-		case '😃': accumulator *= accumulator;
+		case '+': accumulator += 1
+		case '-': accumulator -= 1
+		case '*': accumulator *= 2
+		case '/': accumulator /= 2
+		case '😃': accumulator *= accumulator
 		case: // Ignore everything else
 		}
 	}

 	fmt.printf("The program \"%s\" calculates the value %d\n",
-	           program, accumulator);
+	           program, accumulator)
 }

 ```

-
 ## Documentation

 #### [Getting Started](https://odin-lang.org/docs/install)

 Instructions for downloading and installing the Odin compiler and libraries.

+#### [Nightly Builds](https://odin-lang.org/docs/nightly/)
+
+Get the latest nightly builds of Odin.
+
 ### Learning Odin

 #### [Overview of Odin](https://odin-lang.org/docs/overview)
@@ -72,102 +72,24 @@ An overview of the Odin programming language.

 Answers to common questions about Odin.

+#### [Packages](https://pkg.odin-lang.org/)
+
+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)

 A wiki maintained by the Odin community.

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

-Get live support and talk with other odiners on the Odin Discord.
-
-### References
-
-#### [Language Specification](https://odin-lang.org/ref/spec)
-
-The official Odin Language specification.
+Get live support and talk with other Odin programmers on the Odin Discord.

 ### Articles

-#### [The Odin Blog](https://odin-lang.org/blog)
+#### [The Odin Blog](https://odin-lang.org/news/)

 The official blog of the Odin programming language, featuring announcements, news, and in-depth articles by the Odin team and guests.

-## Setup
-
-Odin only supports x86-64/amd64 at the moment (64-bit), relies on LLVM for code generation and an external linker.
-
-In addition, the following platform-specific steps are necessary:
-
- Windows
-    * Have Visual Studio installed (MSVC 2010 or later, for the linker)
-    * Have a copy of `opt.exe` and `llc.exe` in `Odin/bin`. Pre-built Windows binaries can be found [here](https://github.com/odin-lang/Odin/releases/tag/llvm-windows) and *must* be explicitly copied
-    * Open a valid command prompt:
-        * **Basic:** run the `x64 Native Tools Command Prompt for VS2017` shortcut bundled with VS 2017, or
-        * **Advanced:** run `vcvarsall.bat x64` from a blank `cmd` session
-
- MacOS
-    * Have LLVM explicitly installed (`brew install llvm`)
-    * Have XCode installed (version X.X or later, for linking)
-    * Make sure the LLVM binaries and the linker are added to your `$PATH` environmental variable
-
- GNU/Linux
-    * Have LLVM installed (opt/llc)
-    * Have Clang installed (version X.X or later, for linking)
-    * Make sure the LLVM binaries and the linker are added to your `$PATH` environmental variable
-
-Then build the compiler by calling `build.bat` (Windows) or `make` (Linux/MacOS). This will automatically run the demo program if successful.
-
-**Notes for Linux:**: The compiler currently relies on the `core` and `shared` library collection being relative to the compiler executable. Installing the compiler in the usual sense (to `/usr/local/bin` or similar) is therefore not as straight forward as you need to make sure the mentioned libraries are available. As a result, it is recommended to simply explicitly invoke the compiler with `/path/to/odin` in your preferred build system, or add `/path/to/odin` to `$PATH`.
-
-Please read the [Getting Started Guide](https://github.com/odin-lang/Odin/wiki#getting-started-with-odin) for more information.
-
-## Requirements to build and run
-
-Please read the [Getting Started Guide](https://github.com/odin-lang/Odin/wiki#getting-started-with-odin).
-
- Windows
-	* x86-64/amd64
-	* MSVC 2010 installed (C++11 support)
-	* [LLVM binaries](https://github.com/odin-lang/Odin/releases/tag/llvm-windows) for `opt.exe`, `llc.exe`, and `lld-link.exe`
-	* Requires MSVC's link.exe as the linker
-		* run `vcvarsall.bat` to setup the path
-
- MacOS
-	* x86-64/amd64
-	* LLVM explicitly installed (`brew install llvm`)
-	* XCode installed (for the linker)
-
- GNU/Linux
-	* x86-64/amd64
-	* Build tools (ld)
-	* LLVM installed
-	* Clang installed (temporary - this is Calling the linker for now)
-
- FreeBSD
-	* x86-64/amd64
-	* Build tools (ld)
-	* LLVM installed
-	* Clang installed (temporary - this is Calling the linker for now)
-
-Other platforms may be supported but are experimental for the time being.
-
 ## Warnings

 * The Odin compiler is still in development.
-
-## Demonstrations:
-* First Talk & Demo
-	- [Talk](https://youtu.be/TMCkT-uASaE?t=338)
-	- [Demo](https://youtu.be/TMCkT-uASaE?t=1800)
-	- [Q&A](https://youtu.be/TMCkT-uASaE?t=5749)
-* [Composition & Refactorability](https://www.youtube.com/watch?v=n1wemZfcbXM)
-* [Introspection, Modules, and Record Layout](https://www.youtube.com/watch?v=UFq8rhWhx4s)
-* [push_allocator & Minimal Dependency Building](https://www.youtube.com/watch?v=f_LGVOAMb78)
-* [when, for & procedure overloading](https://www.youtube.com/watch?v=OzeOekzyZK8)
-* [Context Types, Unexported Entities, Labelled Branches](https://www.youtube.com/watch?v=CkHVwT1Qk-g)
-* [Bit Fields, i128 & u128, Syntax Changes](https://www.youtube.com/watch?v=NlTutcLyF64)
-* [Default and Named Arguments; Explicit Parametric Polymorphism](https://www.youtube.com/watch?v=-XQZE6S6zUU)
-* [Loadsachanges](https://www.youtube.com/watch?v=ar0vFMoMtrI)
-* [Packages, Bit Sets, cstring](https://youtu.be/b8bJbjiXZrQ)
-	- [Q&A](https://youtu.be/5jmxyIfyyTk)
-
@@ -0,0 +1,128 @@
+// This is purely for documentation
+package builtin
+
+nil   :: nil
+false :: 0!=0
+true  :: 0==0
+
+ODIN_OS      :: ODIN_OS
+ODIN_ARCH    :: ODIN_ARCH
+ODIN_ENDIAN  :: ODIN_ENDIAN
+ODIN_VENDOR  :: ODIN_VENDOR
+ODIN_VERSION :: ODIN_VERSION
+ODIN_ROOT    :: ODIN_ROOT
+ODIN_DEBUG   :: ODIN_DEBUG
+
+byte :: u8 // alias
+
+bool          :: bool
+b8            :: b8
+b16           :: b16
+b32           :: b32
+b64           :: b64
+
+i8            :: i8
+u8            :: u8
+i16           :: i16
+u16           :: u16
+i32           :: i32
+u32           :: u32
+i64           :: i64
+u64           :: u64
+
+i128          :: i128
+u128          :: u128
+
+rune          :: rune
+
+f16           :: f16
+f32           :: f32
+f64           :: f64
+
+complex32     :: complex32
+complex64     :: complex64
+complex128    :: complex128
+
+quaternion64  :: quaternion64
+quaternion128 :: quaternion128
+quaternion256 :: quaternion256
+
+int           :: int
+uint          :: uint
+uintptr       :: uintptr
+
+rawptr        :: rawptr
+string        :: string
+cstring       :: cstring
+any           :: any
+
+typeid        :: typeid
+
+// Endian Specific Types
+i16le         :: i16le
+u16le         :: u16le
+i32le         :: i32le
+u32le         :: u32le
+i64le         :: i64le
+u64le         :: u64le
+i128le        :: i128le
+u128le        :: u128le
+
+i16be         :: i16be
+u16be         :: u16be
+i32be         :: i32be
+u32be         :: u32be
+i64be         :: i64be
+u64be         :: u64be
+i128be        :: i128be
+u128be        :: u128be
+
+
+f16le         :: f16le
+f32le         :: f32le
+f64le         :: f64le
+
+f16be         :: f16be
+f32be         :: f32be
+f64be         :: f64be
+
+
+
+// Procedures
+len :: proc(array: Array_Type) -> int ---
+cap :: proc(array: Array_Type) -> int ---
+
+size_of      :: proc($T: typeid) -> int ---
+align_of     :: proc($T: typeid) -> int ---
+
+// 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 ---
+
+swizzle :: proc(x: [N]T, indices: ..int) -> [len(indices)]T ---
+
+complex    :: proc(real, imag: Float) -> Complex_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_values :: proc(value: Struct_Or_Array) -> (A, B, C, ...) ---
+
+min   :: proc(values: ..T) -> T ---
+max   :: proc(values: ..T) -> T ---
+abs   :: proc(value: T) -> T ---
+clamp :: proc(value, minimum, maximum: T) -> T ---
+
+soa_zip :: proc(slices: ...) -> #soa[]Struct ---
+soa_unzip :: proc(value: $S/#soa[]$E) -> (slices: ...) ---
@@ -0,0 +1,322 @@
+// This is purely for documentation
+//+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, $T: typeid) -> type/#soa[N]T
+
+// Volatile
+volatile_load  :: proc(dst: ^$T) -> T ---
+volatile_store :: proc(dst: ^$T, val: T) ---
+
+non_temporal_load  :: proc(dst: ^$T) -> T ---
+non_temporal_store :: proc(dst: ^$T, val: T)  ---
+
+// Trapping
+debug_trap :: proc() ---
+trap       :: proc() -> ! ---
+
+// Instructions
+
+alloca             :: proc(size, align: int) -> [^]u8 ---
+cpu_relax          :: proc() ---
+read_cycle_counter :: proc() -> i64 ---
+
+count_ones           :: proc(x: $T) -> T where type_is_integer(T) || type_is_simd_vector(T) ---
+count_zeros          :: proc(x: $T) -> T where type_is_integer(T) || type_is_simd_vector(T) ---
+count_trailing_zeros :: proc(x: $T) -> T where type_is_integer(T) || type_is_simd_vector(T) ---
+count_leading_zeros  :: proc(x: $T) -> T where type_is_integer(T) || type_is_simd_vector(T) ---
+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 ---
+
+sqrt :: proc(x: $T) -> T where type_is_float(T) || (type_is_simd_vector(T) && type_is_float(type_elem_type(T))) ---
+
+fused_mul_add :: proc(a, b, c: $T) -> T where type_is_float(T) || (type_is_simd_vector(T) && type_is_float(type_elem_type(T))) ---
+
+mem_copy                 :: proc(dst, src: rawptr, len: int) ---
+mem_copy_non_overlapping :: proc(dst, src: rawptr, len: int) ---
+mem_zero                 :: proc(ptr: rawptr, len: int) ---
+mem_zero_volatile        :: proc(ptr: rawptr, len: int) ---
+
+// prefer [^]T operations if possible
+ptr_offset :: proc(ptr: ^$T, offset: int) -> ^T ---
+ptr_sub    :: proc(a, b: ^$T) -> int ---
+
+unaligned_load           :: proc(src: ^$T) -> T ---
+unaligned_store          :: proc(dst: ^$T, val: T) -> T ---
+
+fixed_point_mul     :: proc(lhs, rhs: $T, #const scale: uint) -> T where type_is_integer(T) ---
+fixed_point_div     :: proc(lhs, rhs: $T, #const scale: uint) -> T where type_is_integer(T) ---
+fixed_point_mul_sat :: proc(lhs, rhs: $T, #const scale: uint) -> T where type_is_integer(T) ---
+fixed_point_div_sat :: proc(lhs, rhs: $T, #const scale: uint) -> T where type_is_integer(T) ---
+
+prefetch_read_instruction  :: proc(address: rawptr, #const locality: i32 /* 0..=3 */) ---
+prefetch_read_data         :: proc(address: rawptr, #const locality: i32 /* 0..=3 */) ---
+prefetch_write_instruction :: proc(address: rawptr, #const locality: i32 /* 0..=3 */) ---
+prefetch_write_data        :: proc(address: rawptr, #const locality: i32 /* 0..=3 */) ---
+
+// Compiler Hints
+expect :: proc(val, expected_val: T) -> T ---
+
+// Linux and Darwin Only
+syscall :: proc(id: uintptr, args: ..uintptr) -> uintptr ---
+
+
+// Atomics
+Atomic_Memory_Order :: enum {
+	Relaxed = 0, // Unordered
+	Consume = 1, // Monotonic
+	Acquire = 2,
+	Release = 3,
+	Acq_Rel = 4,
+	Seq_Cst = 5,
+}
+
+atomic_type_is_lock_free :: proc($T: typeid) -> bool ---
+
+atomic_thread_fence :: proc(order: Atomic_Memory_Order) ---
+atomic_signal_fence :: proc(order: Atomic_Memory_Order) ---
+
+atomic_store          :: proc(dst: ^$T, val: T) ---
+atomic_store_explicit :: proc(dst: ^$T, val: T, order: Atomic_Memory_Order) ---
+
+atomic_load           :: proc(dst: ^$T) -> T ---
+atomic_load_explicit  :: proc(dst: ^$T, order: Atomic_Memory_Order) -> T ---
+
+// fetch then operator
+atomic_add               :: proc(dst: ^$T, val: T) -> T ---
+atomic_add_explicit      :: proc(dst: ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_sub               :: proc(dst: ^$T, val: T) -> T ---
+atomic_sub_explicit      :: proc(dst: ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_and               :: proc(dst: ^$T, val: T) -> T ---
+atomic_and_explicit      :: proc(dst: ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_nand              :: proc(dst: ^$T, val: T) -> T ---
+atomic_nand_explicit     :: proc(dst: ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_or                :: proc(dst: ^$T, val: T) -> T ---
+atomic_or_explicit       :: proc(dst: ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_xor               :: proc(dst: ^$T, val: T) -> T ---
+atomic_xor_explicit      :: proc(dst: ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+atomic_exchange          :: proc(dst: ^$T, val: T) -> T ---
+atomic_exchange_explicit :: proc(dst: ^$T, val: T, order: Atomic_Memory_Order) -> T ---
+
+atomic_compare_exchange_strong          :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
+atomic_compare_exchange_strong_explicit :: proc(dst: ^$T, old, new: T, success, failure: Atomic_Memory_Order) -> (T, bool) #optional_ok ---
+atomic_compare_exchange_weak            :: proc(dst: ^$T, old, new: T) -> (T, bool) #optional_ok ---
+atomic_compare_exchange_weak_explicit   :: proc(dst: ^$T, old, new: T, success, failure: Atomic_Memory_Order) -> (T, bool) #optional_ok ---
+
+
+// Constant type tests
+
+type_base_type :: proc($T: typeid) -> type ---
+type_core_type :: proc($T: typeid) -> type ---
+type_elem_type :: proc($T: typeid) -> type ---
+
+type_is_boolean    :: proc($T: typeid) -> bool ---
+type_is_integer    :: proc($T: typeid) -> bool ---
+type_is_rune       :: proc($T: typeid) -> bool ---
+type_is_float      :: proc($T: typeid) -> bool ---
+type_is_complex    :: proc($T: typeid) -> bool ---
+type_is_quaternion :: proc($T: typeid) -> bool ---
+type_is_string     :: proc($T: typeid) -> bool ---
+type_is_typeid     :: proc($T: typeid) -> bool ---
+type_is_any        :: proc($T: typeid) -> bool ---
+
+type_is_endian_platform       :: proc($T: typeid) -> bool ---
+type_is_endian_little         :: proc($T: typeid) -> bool ---
+type_is_endian_big            :: proc($T: typeid) -> bool ---
+type_is_unsigned              :: proc($T: typeid) -> bool ---
+type_is_numeric               :: proc($T: typeid) -> bool ---
+type_is_ordered               :: proc($T: typeid) -> bool ---
+type_is_ordered_numeric       :: proc($T: typeid) -> bool ---
+type_is_indexable             :: proc($T: typeid) -> bool ---
+type_is_sliceable             :: proc($T: typeid) -> bool ---
+type_is_comparable            :: proc($T: typeid) -> bool ---
+type_is_simple_compare        :: proc($T: typeid) -> bool --- // easily compared using memcmp (== and !=)
+type_is_dereferenceable       :: proc($T: typeid) -> bool ---
+type_is_valid_map_key         :: proc($T: typeid) -> bool ---
+type_is_valid_matrix_elements :: proc($T: typeid) -> bool ---
+
+type_is_named            :: proc($T: typeid) -> bool ---
+type_is_pointer          :: proc($T: typeid) -> bool ---
+type_is_multi_pointer    :: proc($T: typeid) -> bool ---
+type_is_array            :: proc($T: typeid) -> bool ---
+type_is_enumerated_array :: proc($T: typeid) -> bool ---
+type_is_slice            :: proc($T: typeid) -> bool ---
+type_is_dynamic_array    :: proc($T: typeid) -> bool ---
+type_is_map              :: proc($T: typeid) -> bool ---
+type_is_struct           :: proc($T: typeid) -> bool ---
+type_is_union            :: proc($T: typeid) -> bool ---
+type_is_enum             :: proc($T: typeid) -> bool ---
+type_is_proc             :: proc($T: typeid) -> bool ---
+type_is_bit_set          :: proc($T: typeid) -> bool ---
+type_is_simd_vector      :: proc($T: typeid) -> bool ---
+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, $V: typeid) -> bool where type_is_union(U) ---
+type_union_tag_type :: proc($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, $V: typeid) -> int where type_is_union(U) ---
+
+type_has_field :: proc($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) ---
+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_polymorphic_record_parameter_count :: proc($T: typeid) -> typeid ---
+type_polymorphic_record_parameter_value :: proc($T: typeid, index: int) -> $V ---
+
+type_is_specialized_polymorphic_record   :: proc($T: typeid) -> bool ---
+type_is_unspecialized_polymorphic_record :: proc($T: typeid) -> bool ---
+
+type_is_subtype_of :: proc($T, $U: typeid) -> bool ---
+
+type_field_index_of :: proc($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, $V: typeid) -> typeid where type_is_union(U), type_is_union(V) ---
+
+constant_utf16_cstring :: proc($literal: string) -> [^]u16 ---
+
+// 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) ---
+
+// 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 ---
+simd_shr :: proc(a: #simd[N]T, b: #simd[N]Unsigned_Integer) -> #simd[N]T ---
+
+// Similar to C's Behaviour
+// x << (y & mask)
+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_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 ---
+
+simd_abs :: proc(a: #simd[N]T) -> #simd[N]T ---
+
+simd_min   :: proc(a, b: #simd[N]T) -> #simd[N]T ---
+simd_max   :: proc(a, b: #simd[N]T) -> #simd[N]T ---
+simd_clamp :: proc(v, min, max: #simd[N]T) -> #simd[N]T ---
+
+// Return an unsigned integer of the same size as the input type
+// NOT A BOOLEAN
+// element-wise:
+//     false => 0x00...00
+//     true  => 0xff...ff
+simd_lanes_eq :: proc(a, b: #simd[N]T) -> #simd[N]Integer ---
+simd_lanes_ne :: proc(a, b: #simd[N]T) -> #simd[N]Integer ---
+simd_lanes_lt :: proc(a, b: #simd[N]T) -> #simd[N]Integer ---
+simd_lanes_le :: proc(a, b: #simd[N]T) -> #simd[N]Integer ---
+simd_lanes_gt :: proc(a, b: #simd[N]T) -> #simd[N]Integer ---
+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_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 ---
+
+// Lane-wise operations
+simd_ceil    :: proc(a: #simd[N]any_float) -> #simd[N]any_float ---
+simd_floor   :: proc(a: #simd[N]any_float) -> #simd[N]any_float ---
+simd_trunc   :: proc(a: #simd[N]any_float) -> #simd[N]any_float ---
+// rounding to the nearest integral value; if two values are equally near, rounds to the even one
+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 ---
+
+simd_rotate_left  :: proc(a: #simd[N]T, $offset: int) -> #simd[N]T ---
+simd_rotate_right :: proc(a: #simd[N]T, $offset: int) -> #simd[N]T ---
+
+
+// WASM targets only
+wasm_memory_grow :: proc(index, delta: uintptr) -> int ---
+wasm_memory_size :: proc(index: uintptr)        -> int ---
+
+// `timeout_ns` is maximum number of nanoseconds the calling thread will be blocked for
+// A negative value will be blocked forever
+// Return value:
+// 0 - indicates that the thread blocked and then was woken up
+// 1 - the loaded value from `ptr` did not match `expected`, the thread did not block
+// 2 - the thread blocked, but the timeout
+wasm_memory_atomic_wait32   :: proc(ptr: ^u32, expected: u32, timeout_ns: i64) -> u32 ---
+wasm_memory_atomic_notify32 :: proc(ptr: ^u32, waiters: u32) -> (waiters_woken_up: u32) ---
+
+// x86 Targets (i386, amd64)
+x86_cpuid  :: proc(ax, cx: u32) -> (eax, ebx, ecx, edx: u32) ---
+x86_xgetbv :: proc(cx: u32) -> (eax, edx: u32) ---
+
+
+// Darwin targets only
+objc_object   :: struct{}
+objc_selector :: struct{}
+objc_class    :: struct{}
+objc_id    :: ^objc_object
+objc_SEL   :: ^objc_selector
+objc_Class :: ^objc_class
+
+objc_find_selector     :: proc($name: string) -> objc_SEL   ---
+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() ---
@@ -18,8 +18,11 @@
 // 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 "base:intrinsics"
+
 // NOTE(bill): This must match the compiler's
 Calling_Convention :: enum u8 {
 	Invalid     = 0,
@@ -30,9 +33,15 @@ Calling_Convention :: enum u8 {
 	Fast_Call   = 5,

 	None        = 6,
+	Naked       = 7,
+
+	_           = 8, // reserved
+
+	Win64       = 9,
+	SysV        = 10,
 }

-Type_Info_Enum_Value :: distinct i64;
+Type_Info_Enum_Value :: distinct i64

 Platform_Endianness :: enum u8 {
 	Platform = 0,
@@ -41,9 +50,9 @@ Platform_Endianness :: enum u8 {
 }

 // Procedure type to test whether two values of the same type are equal
-Equal_Proc :: distinct proc "contextless" (rawptr, rawptr) -> bool;
+Equal_Proc :: distinct proc "contextless" (rawptr, rawptr) -> bool
 // Procedure type to hash a value, default seed value is 0
-Hasher_Proc :: distinct proc "contextless" (data: rawptr, seed: uintptr = 0) -> uintptr;
+Hasher_Proc :: distinct proc "contextless" (data: rawptr, seed: uintptr = 0) -> uintptr

 Type_Info_Struct_Soa_Kind :: enum u8 {
 	None    = 0,
@@ -58,30 +67,33 @@ Type_Info_Named :: struct {
 	base: ^Type_Info,
 	pkg:  string,
 	loc:  Source_Code_Location,
-};
-Type_Info_Integer    :: struct {signed: bool, endianness: Platform_Endianness};
-Type_Info_Rune       :: struct {};
-Type_Info_Float      :: struct {endianness: Platform_Endianness};
-Type_Info_Complex    :: struct {};
-Type_Info_Quaternion :: struct {};
-Type_Info_String     :: struct {is_cstring: bool};
-Type_Info_Boolean    :: struct {};
-Type_Info_Any        :: struct {};
-Type_Info_Type_Id    :: struct {};
+}
+Type_Info_Integer    :: struct {signed: bool, endianness: Platform_Endianness}
+Type_Info_Rune       :: struct {}
+Type_Info_Float      :: struct {endianness: Platform_Endianness}
+Type_Info_Complex    :: struct {}
+Type_Info_Quaternion :: struct {}
+Type_Info_String     :: struct {is_cstring: bool}
+Type_Info_Boolean    :: struct {}
+Type_Info_Any        :: struct {}
+Type_Info_Type_Id    :: struct {}
 Type_Info_Pointer :: struct {
 	elem: ^Type_Info, // nil -> rawptr
-};
+}
+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,
-};
+}
 Type_Info_Array :: struct {
 	elem:      ^Type_Info,
 	elem_size: int,
 	count:     int,
-};
+}
 Type_Info_Enumerated_Array :: struct {
 	elem:      ^Type_Info,
 	index:     ^Type_Info,
@@ -89,13 +101,16 @@ Type_Info_Enumerated_Array :: struct {
 	count:     int,
 	min_value: Type_Info_Enum_Value,
 	max_value: Type_Info_Enum_Value,
-};
-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
+	is_sparse: bool,
+}
+Type_Info_Dynamic_Array :: struct {elem: ^Type_Info, elem_size: int}
+Type_Info_Slice         :: struct {elem: ^Type_Info, elem_size: int}
+
+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 :: struct {
 	types:        []^Type_Info,
@@ -105,6 +120,7 @@ Type_Info_Struct :: struct {
 	tags:         []string,
 	is_packed:    bool,
 	is_raw_union: bool,
+	is_no_copy:   bool,
 	custom_align: bool,

 	equal: Equal_Proc, // set only when the struct has .Comparable set but does not have .Simple_Compare set
@@ -113,52 +129,64 @@ Type_Info_Struct :: struct {
 	soa_kind:      Type_Info_Struct_Soa_Kind,
 	soa_base_type: ^Type_Info,
 	soa_len:       int,
-};
+}
 Type_Info_Union :: struct {
 	variants:     []^Type_Info,
 	tag_offset:   uintptr,
 	tag_type:     ^Type_Info,
+
+	equal: Equal_Proc, // set only when the struct has .Comparable set but does not have .Simple_Compare set
+
 	custom_align: bool,
 	no_nil:       bool,
-	maybe:        bool,
-};
+	shared_nil:   bool,
+}
 Type_Info_Enum :: struct {
 	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,
 	underlying: ^Type_Info, // Possibly nil
 	lower:      i64,
 	upper:      i64,
-};
+}
 Type_Info_Simd_Vector :: struct {
 	elem:       ^Type_Info,
 	elem_size:  int,
 	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 {
+	elem:         ^Type_Info,
+	elem_size:    int,
+	elem_stride:  int, // elem_stride >= row_count
+	row_count:    int,
+	column_count: int,
+	// Total element count = column_count * elem_stride
+}
+Type_Info_Soa_Pointer :: struct {
+	elem: ^Type_Info,
+}

 Type_Info_Flag :: enum u8 {
 	Comparable     = 0,
 	Simple_Compare = 1,
-};
-Type_Info_Flags :: distinct bit_set[Type_Info_Flag; u32];
+}
+Type_Info_Flags :: distinct bit_set[Type_Info_Flag; u32]

 Type_Info :: struct {
 	size:  int,
@@ -178,12 +206,13 @@ Type_Info :: struct {
 		Type_Info_Any,
 		Type_Info_Type_Id,
 		Type_Info_Pointer,
+		Type_Info_Multi_Pointer,
 		Type_Info_Procedure,
 		Type_Info_Array,
 		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,
@@ -191,7 +220,9 @@ 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,
 	},
 }

@@ -208,6 +239,7 @@ Typeid_Kind :: enum u8 {
 	Any,
 	Type_Id,
 	Pointer,
+	Multi_Pointer,
 	Procedure,
 	Array,
 	Enumerated_Array,
@@ -221,11 +253,13 @@ Typeid_Kind :: enum u8 {
 	Bit_Set,
 	Simd_Vector,
 	Relative_Pointer,
-	Relative_Slice,
+	Relative_Multi_Pointer,
+	Matrix,
+	Soa_Pointer,
 }
-#assert(len(Typeid_Kind) < 32);
+#assert(len(Typeid_Kind) < 32)

-// Typeid_Bit_Field :: bit_field #align align_of(uintptr) {
+// Typeid_Bit_Field :: bit_field #align(align_of(uintptr)) {
 // 	index:    8*size_of(uintptr) - 8,
 // 	kind:     5, // Typeid_Kind
 // 	named:    1,
@@ -236,9 +270,22 @@ Typeid_Kind :: enum u8 {

 // 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;
+args__: []cstring
+
+when ODIN_OS == .Windows {
+	// NOTE(Jeroen): If we're a Windows DLL, fwdReason will be populated.
+	// This tells a DLL if it's first loaded, about to be unloaded, or a thread is joining/exiting.
+
+	DLL_Forward_Reason :: enum u32 {
+		Process_Detach = 0, // About to unload DLL
+		Process_Attach = 1, // Entry point
+		Thread_Attach  = 2,
+		Thread_Detach  = 3,
+	}
+	dll_forward_reason: DLL_Forward_Reason
+}

 // IMPORTANT NOTE(bill): Must be in this order (as the compiler relies upon it)

@@ -249,7 +296,7 @@ Source_Code_Location :: struct {
 	procedure:    string,
 }

-Assertion_Failure_Proc :: #type proc(prefix, message: string, loc: Source_Code_Location);
+Assertion_Failure_Proc :: #type proc(prefix, message: string, loc: Source_Code_Location) -> !

 // Allocation Stuff
 Allocator_Mode :: enum byte {
@@ -259,9 +306,11 @@ Allocator_Mode :: enum byte {
 	Resize,
 	Query_Features,
 	Query_Info,
+	Alloc_Non_Zeroed,
+	Resize_Non_Zeroed,
 }

-Allocator_Mode_Set :: distinct bit_set[Allocator_Mode];
+Allocator_Mode_Set :: distinct bit_set[Allocator_Mode]

 Allocator_Query_Info :: struct {
 	pointer:   rawptr,
@@ -270,21 +319,30 @@ Allocator_Query_Info :: struct {
 }

 Allocator_Error :: enum byte {
-	None             = 0,
-	Out_Of_Memory    = 1,
-	Invalid_Pointer  = 2,
-	Invalid_Argument = 3,
+	None                 = 0,
+	Out_Of_Memory        = 1,
+	Invalid_Pointer      = 2,
+	Invalid_Argument     = 3,
+	Mode_Not_Implemented = 4,
 }

 Allocator_Proc :: #type proc(allocator_data: rawptr, mode: Allocator_Mode,
                             size, alignment: int,
                             old_memory: rawptr, old_size: int,
-                             location: Source_Code_Location = #caller_location) -> ([]byte, Allocator_Error);
+                             location: Source_Code_Location = #caller_location) -> ([]byte, Allocator_Error)
 Allocator :: struct {
 	procedure: Allocator_Proc,
 	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 {
@@ -307,8 +365,8 @@ Logger_Option :: enum {
 	Thread_Id,
 }

-Logger_Options :: bit_set[Logger_Option];
-Logger_Proc :: #type proc(data: rawptr, level: Logger_Level, text: string, options: Logger_Options, location := #caller_location);
+Logger_Options :: bit_set[Logger_Option]
+Logger_Proc :: #type proc(data: rawptr, level: Logger_Level, text: string, options: Logger_Options, location := #caller_location)

 Logger :: struct {
 	procedure:    Logger_Proc,
@@ -323,9 +381,6 @@ Context :: struct {
 	assertion_failure_proc: Assertion_Failure_Proc,
 	logger:                 Logger,

-	thread_id:  int,
-
-	user_data:  any,
 	user_ptr:   rawptr,
 	user_index: int,

@@ -335,7 +390,7 @@ Context :: struct {


 Raw_String :: struct {
-	data: ^byte,
+	data: [^]byte,
 	len:  int,
 }

@@ -351,11 +406,126 @@ 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 {
+	data: rawptr,
+	id:   typeid,
+}
+
+Raw_Cstring :: struct {
+	data: [^]byte,
+}
+
+Raw_Soa_Pointer :: struct {
+	data:  rawptr,
+	index: int,
+}
+
+
+
+/*
+	// Defined internally by the compiler
+	Odin_OS_Type :: enum int {
+		Unknown,
+		Windows,
+		Darwin,
+		Linux,
+		Essence,
+		FreeBSD,
+		OpenBSD,
+		WASI,
+		JS,
+		Freestanding,
+	}
+*/
+Odin_OS_Type :: type_of(ODIN_OS)
+
+/*
+	// Defined internally by the compiler
+	Odin_Arch_Type :: enum int {
+		Unknown,
+		amd64,
+		i386,
+		arm32,
+		arm64,
+		wasm32,
+		wasm64p32,
+	}
+*/
+Odin_Arch_Type :: type_of(ODIN_ARCH)
+
+/*
+	// Defined internally by the compiler
+	Odin_Build_Mode_Type :: enum int {
+		Executable,
+		Dynamic,
+		Object,
+		Assembly,
+		LLVM_IR,
+	}
+*/
+Odin_Build_Mode_Type :: type_of(ODIN_BUILD_MODE)
+
+/*
+	// Defined internally by the compiler
+	Odin_Endian_Type :: enum int {
+		Unknown,
+		Little,
+		Big,
+	}
+*/
+Odin_Endian_Type :: type_of(ODIN_ENDIAN)
+
+
+/*
+	// Defined internally by the compiler
+	Odin_Platform_Subtarget_Type :: enum int {
+		Default,
+		iOS,
+	}
+*/
+Odin_Platform_Subtarget_Type :: type_of(ODIN_PLATFORM_SUBTARGET)
+
+/*
+	// Defined internally by the compiler
+	Odin_Sanitizer_Flag :: enum u32 {
+		Address = 0,
+		Memory  = 1,
+		Thread  = 2,
+	}
+	Odin_Sanitizer_Flags :: distinct bitset[Odin_Sanitizer_Flag; u32]
+
+	ODIN_SANITIZER_FLAGS // is a constant
+*/
+Odin_Sanitizer_Flags :: type_of(ODIN_SANITIZER_FLAGS)
+

 /////////////////////////////
 // Init Startup Procedures //
@@ -365,7 +535,14 @@ Raw_Map :: struct {
 // This is probably only useful for freestanding targets
 foreign {
 	@(link_name="__$startup_runtime")
-	_startup_runtime :: proc "contextless" () ---
+	_startup_runtime :: proc "odin" () ---
+	@(link_name="__$cleanup_runtime")
+	_cleanup_runtime :: proc "odin" () ---
+}
+
+_cleanup_runtime_contextless :: proc "contextless" () {
+	context = default_context()
+	_cleanup_runtime()
 }


@@ -376,71 +553,65 @@ foreign {

 type_info_base :: proc "contextless" (info: ^Type_Info) -> ^Type_Info {
 	if info == nil {
-		return nil;
+		return nil
 	}

-	base := info;
+	base := info
 	loop: for {
 		#partial switch i in base.variant {
-		case Type_Info_Named: base = i.base;
-		case: break loop;
+		case Type_Info_Named: base = i.base
+		case: break loop
 		}
 	}
-	return base;
+	return base
 }


 type_info_core :: proc "contextless" (info: ^Type_Info) -> ^Type_Info {
 	if info == nil {
-		return nil;
+		return nil
 	}

-	base := 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: break loop;
+		case Type_Info_Named:  base = i.base
+		case Type_Info_Enum:   base = i.base
+		case: break loop
 		}
 	}
-	return base;
+	return base
 }
-type_info_base_without_enum :: type_info_core;
+type_info_base_without_enum :: type_info_core

-__type_info_of :: proc "contextless" (id: typeid) -> ^Type_Info {
-	MASK :: 1<<(8*size_of(typeid) - 8) - 1;
-	data := transmute(uintptr)id;
-	n := int(data & MASK);
+__type_info_of :: proc "contextless" (id: typeid) -> ^Type_Info #no_bounds_check {
+	MASK :: 1<<(8*size_of(typeid) - 8) - 1
+	data := transmute(uintptr)id
+	n := int(data & MASK)
 	if n < 0 || n >= len(type_table) {
-		n = 0;
+		n = 0
 	}
-	return &type_table[n];
+	return &type_table[n]
 }

-typeid_base :: proc "contextless" (id: typeid) -> typeid {
-	ti := type_info_of(id);
-	ti = type_info_base(ti);
-	return ti.id;
+when !ODIN_NO_RTTI {
+	typeid_base :: proc "contextless" (id: typeid) -> typeid {
+		ti := type_info_of(id)
+		ti = type_info_base(ti)
+		return ti.id
+	}
+	typeid_core :: proc "contextless" (id: typeid) -> typeid {
+		ti := type_info_core(type_info_of(id))
+		return ti.id
+	}
+	typeid_base_without_enum :: typeid_core
 }
-typeid_core :: proc "contextless" (id: typeid) -> typeid {
-	ti := type_info_base_without_enum(type_info_of(id));
-	return ti.id;
-}
-typeid_base_without_enum :: typeid_core;



-@(default_calling_convention = "none")
-foreign {
-	@(link_name="llvm.debugtrap")
-	debug_trap :: proc() ---;
-
-	@(link_name="llvm.trap")
-	trap :: proc() -> ! ---;
-
-	@(link_name="llvm.readcyclecounter")
-	read_cycle_counter :: proc() -> u64 ---;
-}
+debug_trap         :: intrinsics.debug_trap
+trap               :: intrinsics.trap
+read_cycle_counter :: intrinsics.read_cycle_counter



@@ -449,55 +620,62 @@ default_logger_proc :: proc(data: rawptr, level: Logger_Level, text: string, opt
 }

 default_logger :: proc() -> Logger {
-	return Logger{default_logger_proc, nil, Logger_Level.Debug, nil};
+	return Logger{default_logger_proc, nil, Logger_Level.Debug, nil}
 }


 default_context :: proc "contextless" () -> Context {
-	c: Context;
-	__init_context(&c);
-	return c;
+	c: Context
+	__init_context(&c)
+	return c
 }

@private
 __init_context_from_ptr :: proc "contextless" (c: ^Context, other: ^Context) {
 	if c == nil {
-		return;
+		return
 	}
-	c^ = other^;
-	__init_context(c);
+	c^ = other^
+	__init_context(c)
 }

@private
 __init_context :: proc "contextless" (c: ^Context) {
 	if c == nil {
-		return;
+		return
 	}

-	// NOTE(bill): Do not initialize these procedures with a call as they are not defined with the "contexless" calling convention
-	c.allocator.procedure = default_allocator_proc;
-	c.allocator.data = nil;
+	// NOTE(bill): Do not initialize these procedures with a call as they are not defined with the "contextless" calling convention
+	c.allocator.procedure = default_allocator_proc
+	c.allocator.data = nil

-	c.temp_allocator.procedure = default_temp_allocator_proc;
-	c.temp_allocator.data = &global_default_temp_allocator_data;
-
-	c.thread_id = current_thread_id(); // NOTE(bill): This is "contextless" so it is okay to call
-	c.assertion_failure_proc = default_assertion_failure_proc;
-
-	c.logger.procedure = default_logger_proc;
-	c.logger.data = nil;
-}
-
-
-default_assertion_failure_proc :: proc(prefix, message: string, loc: Source_Code_Location) {
-	print_caller_location(loc);
-	print_string(" ");
-	print_string(prefix);
-	if len(message) > 0 {
-		print_string(": ");
-		print_string(message);
+	c.temp_allocator.procedure = default_temp_allocator_proc
+	when !NO_DEFAULT_TEMP_ALLOCATOR {
+		c.temp_allocator.data = &global_default_temp_allocator_data
 	}
-	print_byte('\n');
-	debug_trap();
-	// trap();
+	
+	when !ODIN_DISABLE_ASSERT {
+		c.assertion_failure_proc = default_assertion_failure_proc
+	}
+
+	c.logger.procedure = default_logger_proc
+	c.logger.data = nil
+}
+
+default_assertion_failure_proc :: proc(prefix, message: string, loc: Source_Code_Location) -> ! {
+	when ODIN_OS == .Freestanding {
+		// Do nothing
+	} else {
+		when !ODIN_DISABLE_ASSERT {
+			print_caller_location(loc)
+			print_string(" ")
+		}
+		print_string(prefix)
+		if len(message) > 0 {
+			print_string(": ")
+			print_string(message)
+		}
+		print_byte('\n')
+	}
+	trap()
 }
@@ -0,0 +1,883 @@
+package runtime
+
+import "base:intrinsics"
+
+@builtin
+Maybe :: union($T: typeid) {T}
+
+
+@(builtin, require_results)
+container_of :: #force_inline proc "contextless" (ptr: $P/^$Field_Type, $T: typeid, $field_name: string) -> ^T
+	where intrinsics.type_has_field(T, field_name),
+	      intrinsics.type_field_type(T, field_name) == Field_Type {
+	offset :: offset_of_by_string(T, field_name)
+	return (^T)(uintptr(ptr) - offset) if ptr != nil else nil
+}
+
+
+when !NO_DEFAULT_TEMP_ALLOCATOR {
+	@thread_local global_default_temp_allocator_data: Default_Temp_Allocator
+}
+
+@(builtin, disabled=NO_DEFAULT_TEMP_ALLOCATOR)
+init_global_temporary_allocator :: proc(size: int, backup_allocator := context.allocator) {
+	when !NO_DEFAULT_TEMP_ALLOCATOR {
+		default_temp_allocator_init(&global_default_temp_allocator_data, size, backup_allocator)
+	}
+}
+
+
+// `copy_slice` is a built-in procedure that copies elements from a source slice `src` to a destination slice `dst`.
+// The source and destination may overlap. Copy returns the number of elements copied, which will be the minimum
+// of len(src) and len(dst).
+//
+// Prefer the procedure group `copy`.
+@builtin
+copy_slice :: proc "contextless" (dst, src: $T/[]$E) -> int {
+	n := max(0, min(len(dst), len(src)))
+	if n > 0 {
+		intrinsics.mem_copy(raw_data(dst), raw_data(src), n*size_of(E))
+	}
+	return n
+}
+// `copy_from_string` is a built-in procedure that copies elements from a source slice `src` to a destination string `dst`.
+// The source and destination may overlap. Copy returns the number of elements copied, which will be the minimum
+// of len(src) and len(dst).
+//
+// Prefer the procedure group `copy`.
+@builtin
+copy_from_string :: proc "contextless" (dst: $T/[]$E/u8, src: $S/string) -> int {
+	n := max(0, min(len(dst), len(src)))
+	if n > 0 {
+		intrinsics.mem_copy(raw_data(dst), raw_data(src), n)
+	}
+	return n
+}
+// `copy` is a built-in procedure that copies elements from a source slice `src` to a destination slice/string `dst`.
+// The source and destination may overlap. Copy returns the number of elements copied, which will be the minimum
+// of len(src) and len(dst).
+@builtin
+copy :: proc{copy_slice, copy_from_string}
+
+
+
+// `unordered_remove` removed the element at the specified `index`. It does so by replacing the current end value
+// with the old value, and reducing the length of the dynamic array by 1.
+//
+// Note: This is an O(1) operation.
+// Note: If you the elements to remain in their order, use `ordered_remove`.
+// Note: If the index is out of bounds, this procedure will panic.
+@builtin
+unordered_remove :: proc(array: ^$D/[dynamic]$T, index: int, loc := #caller_location) #no_bounds_check {
+	bounds_check_error_loc(loc, index, len(array))
+	n := len(array)-1
+	if index != n {
+		array[index] = array[n]
+	}
+	(^Raw_Dynamic_Array)(array).len -= 1
+}
+// `ordered_remove` removed the element at the specified `index` whilst keeping the order of the other elements.
+//
+// Note: This is an O(N) operation.
+// Note: If you the elements do not have to remain in their order, prefer `unordered_remove`.
+// Note: If the index is out of bounds, this procedure will panic.
+@builtin
+ordered_remove :: proc(array: ^$D/[dynamic]$T, index: int, loc := #caller_location) #no_bounds_check {
+	bounds_check_error_loc(loc, index, len(array))
+	if index+1 < len(array) {
+		copy(array[index:], array[index+1:])
+	}
+	(^Raw_Dynamic_Array)(array).len -= 1
+}
+
+// `remove_range` removes a range of elements specified by the range `lo` and `hi`, whilst keeping the order of the other elements.
+//
+// Note: This is an O(N) operation.
+// Note: If the range is out of bounds, this procedure will panic.
+@builtin
+remove_range :: proc(array: ^$D/[dynamic]$T, lo, hi: int, loc := #caller_location) #no_bounds_check {
+	slice_expr_error_lo_hi_loc(loc, lo, hi, len(array))
+	n := max(hi-lo, 0)
+	if n > 0 {
+		if hi != len(array) {
+			copy(array[lo:], array[hi:])
+		}
+		(^Raw_Dynamic_Array)(array).len -= n
+	}
+}
+
+
+// `pop` will remove and return the end value of dynamic array `array` and reduces the length of `array` by 1.
+//
+// Note: If the dynamic array has no elements (`len(array) == 0`), this procedure will panic.
+@builtin
+pop :: proc(array: ^$T/[dynamic]$E, loc := #caller_location) -> (res: E) #no_bounds_check {
+	assert(len(array) > 0, loc=loc)
+	res = array[len(array)-1]
+	(^Raw_Dynamic_Array)(array).len -= 1
+	return res
+}
+
+
+// `pop_safe` trys to remove and return the end value of dynamic array `array` and reduces the length of `array` by 1.
+// If the operation is not possible, it will return false.
+@builtin
+pop_safe :: proc(array: ^$T/[dynamic]$E) -> (res: E, ok: bool) #no_bounds_check {
+	if len(array) == 0 {
+		return
+	}
+	res, ok = array[len(array)-1], true
+	(^Raw_Dynamic_Array)(array).len -= 1
+	return
+}
+
+// `pop_front` will remove and return the first value of dynamic array `array` and reduces the length of `array` by 1.
+//
+// Note: If the dynamic array as no elements (`len(array) == 0`), this procedure will panic.
+@builtin
+pop_front :: proc(array: ^$T/[dynamic]$E, loc := #caller_location) -> (res: E) #no_bounds_check {
+	assert(len(array) > 0, loc=loc)
+	res = array[0]
+	if len(array) > 1 {
+		copy(array[0:], array[1:])
+	}
+	(^Raw_Dynamic_Array)(array).len -= 1
+	return res
+}
+
+// `pop_front_safe` trys to return and remove the first value of dynamic array `array` and reduces the length of `array` by 1.
+// If the operation is not possible, it will return false.
+@builtin
+pop_front_safe :: proc(array: ^$T/[dynamic]$E) -> (res: E, ok: bool) #no_bounds_check {
+	if len(array) == 0 {
+		return
+	}
+	res, ok = array[0], true
+	if len(array) > 1 {
+		copy(array[0:], array[1:])
+	}
+	(^Raw_Dynamic_Array)(array).len -= 1
+	return
+}
+
+
+// `clear` will set the length of a passed dynamic array or map to `0`
+@builtin
+clear :: proc{clear_dynamic_array, clear_map}
+
+// `reserve` will try to reserve memory of a passed dynamic array or map to the requested element count (setting the `cap`).
+@builtin
+reserve :: proc{reserve_dynamic_array, reserve_map}
+
+@builtin
+non_zero_reserve :: proc{non_zero_reserve_dynamic_array}
+
+// `resize` will try to resize memory of a passed dynamic array to the requested element count (setting the `len`, and possibly `cap`).
+@builtin
+resize :: proc{resize_dynamic_array}
+
+@builtin
+non_zero_resize :: proc{non_zero_resize_dynamic_array}
+
+// Shrinks the capacity of a dynamic array or map down to the current length, or the given capacity.
+@builtin
+shrink :: proc{shrink_dynamic_array, shrink_map}
+
+// `free` will try to free the passed pointer, with the given `allocator` if the allocator supports this operation.
+@builtin
+free :: proc{mem_free}
+
+// `free_all` will try to free/reset all of the memory of the given `allocator` if the allocator supports this operation.
+@builtin
+free_all :: proc{mem_free_all}
+
+
+
+// `delete_string` will try to free the underlying data of the passed string, with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer the procedure group `delete`.
+@builtin
+delete_string :: proc(str: string, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
+	return mem_free_with_size(raw_data(str), len(str), allocator, loc)
+}
+// `delete_cstring` will try to free the underlying data of the passed string, with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer the procedure group `delete`.
+@builtin
+delete_cstring :: proc(str: cstring, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
+	return mem_free((^byte)(str), allocator, loc)
+}
+// `delete_dynamic_array` will try to free the underlying data of the passed dynamic array, with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer the procedure group `delete`.
+@builtin
+delete_dynamic_array :: proc(array: $T/[dynamic]$E, loc := #caller_location) -> Allocator_Error {
+	return mem_free_with_size(raw_data(array), cap(array)*size_of(E), array.allocator, loc)
+}
+// `delete_slice` will try to free the underlying data of the passed sliced, with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer the procedure group `delete`.
+@builtin
+delete_slice :: proc(array: $T/[]$E, allocator := context.allocator, loc := #caller_location) -> Allocator_Error {
+	return mem_free_with_size(raw_data(array), len(array)*size_of(E), allocator, loc)
+}
+// `delete_map` will try to free the underlying data of the passed map, with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer the procedure group `delete`.
+@builtin
+delete_map :: proc(m: $T/map[$K]$V, loc := #caller_location) -> Allocator_Error {
+	return map_free_dynamic(transmute(Raw_Map)m, map_info(T), loc)
+}
+
+
+// `delete` will try to free the underlying data of the passed built-in data structure (string, cstring, dynamic array, slice, or map), with the given `allocator` if the allocator supports this operation.
+//
+// Note: Prefer `delete` over the specific `delete_*` procedures where possible.
+@builtin
+delete :: proc{
+	delete_string,
+	delete_cstring,
+	delete_dynamic_array,
+	delete_slice,
+	delete_map,
+	delete_soa_slice,
+	delete_soa_dynamic_array,
+}
+
+
+// The new built-in procedure allocates memory. The first argument is a type, not a value, and the value
+// return is a pointer to a newly allocated value of that type using the specified allocator, default is context.allocator
+@(builtin, require_results)
+new :: proc($T: typeid, allocator := context.allocator, loc := #caller_location) -> (^T, Allocator_Error) #optional_allocator_error {
+	return new_aligned(T, align_of(T), allocator, loc)
+}
+@(require_results)
+new_aligned :: proc($T: typeid, alignment: int, allocator := context.allocator, loc := #caller_location) -> (t: ^T, err: Allocator_Error) {
+	data := mem_alloc_bytes(size_of(T), alignment, allocator, loc) or_return
+	t = (^T)(raw_data(data))
+	return
+}
+
+@(builtin, require_results)
+new_clone :: proc(data: $T, allocator := context.allocator, loc := #caller_location) -> (t: ^T, err: Allocator_Error) #optional_allocator_error {
+	t_data := mem_alloc_bytes(size_of(T), align_of(T), allocator, loc) or_return
+	t = (^T)(raw_data(t_data))
+	if t != nil {
+		t^ = data
+	}
+	return
+}
+
+DEFAULT_RESERVE_CAPACITY :: 16
+
+@(require_results)
+make_aligned :: proc($T: typeid/[]$E, #any_int len: int, alignment: int, allocator := context.allocator, loc := #caller_location) -> (T, Allocator_Error) #optional_allocator_error {
+	make_slice_error_loc(loc, len)
+	data, err := mem_alloc_bytes(size_of(E)*len, alignment, allocator, loc)
+	if data == nil && size_of(E) != 0 {
+		return nil, err
+	}
+	s := Raw_Slice{raw_data(data), len}
+	return transmute(T)s, err
+}
+
+// `make_slice` allocates and initializes a slice. Like `new`, the first argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(builtin, require_results)
+make_slice :: proc($T: typeid/[]$E, #any_int len: int, allocator := context.allocator, loc := #caller_location) -> (T, Allocator_Error) #optional_allocator_error {
+	return make_aligned(T, len, align_of(E), allocator, loc)
+}
+// `make_dynamic_array` allocates and initializes a dynamic array. Like `new`, the first argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(builtin, require_results)
+make_dynamic_array :: proc($T: typeid/[dynamic]$E, allocator := context.allocator, loc := #caller_location) -> (T, Allocator_Error) #optional_allocator_error {
+	return make_dynamic_array_len_cap(T, 0, DEFAULT_RESERVE_CAPACITY, allocator, loc)
+}
+// `make_dynamic_array_len` allocates and initializes a dynamic array. Like `new`, the first argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(builtin, require_results)
+make_dynamic_array_len :: proc($T: typeid/[dynamic]$E, #any_int len: int, allocator := context.allocator, loc := #caller_location) -> (T, Allocator_Error) #optional_allocator_error {
+	return make_dynamic_array_len_cap(T, len, len, allocator, loc)
+}
+// `make_dynamic_array_len_cap` allocates and initializes a dynamic array. Like `new`, the first argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(builtin, require_results)
+make_dynamic_array_len_cap :: proc($T: typeid/[dynamic]$E, #any_int len: int, #any_int cap: int, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) #optional_allocator_error {
+	make_dynamic_array_error_loc(loc, len, cap)
+	array.allocator = allocator // initialize allocator before just in case it fails to allocate any memory
+	data := mem_alloc_bytes(size_of(E)*cap, align_of(E), allocator, loc) or_return
+	s := Raw_Dynamic_Array{raw_data(data), len, cap, allocator}
+	if data == nil && size_of(E) != 0 {
+		s.len, s.cap = 0, 0
+	}
+	array = transmute(T)s
+	return
+}
+// `make_map` allocates and initializes a dynamic array. Like `new`, the first argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// Note: Prefer using the procedure group `make`.
+@(builtin, require_results)
+make_map :: proc($T: typeid/map[$K]$E, #any_int capacity: int = 1<<MAP_MIN_LOG2_CAPACITY, allocator := context.allocator, loc := #caller_location) -> (m: T, err: Allocator_Error) #optional_allocator_error {
+	make_map_expr_error_loc(loc, capacity)
+	context.allocator = allocator
+
+	err = reserve_map(&m, capacity, loc)
+	return
+}
+// `make_multi_pointer` allocates and initializes a dynamic array. Like `new`, the first argument is a type, not a value.
+// Unlike `new`, `make`'s return value is the same as the type of its argument, not a pointer to it.
+//
+// This is "similar" to doing `raw_data(make([]E, len, allocator))`.
+//
+// Note: Prefer using the procedure group `make`.
+@(builtin, require_results)
+make_multi_pointer :: proc($T: typeid/[^]$E, #any_int len: int, allocator := context.allocator, loc := #caller_location) -> (mp: T, err: Allocator_Error) #optional_allocator_error {
+	make_slice_error_loc(loc, len)
+	data := mem_alloc_bytes(size_of(E)*len, align_of(E), allocator, loc) or_return
+	if data == nil && size_of(E) != 0 {
+		return
+	}
+	mp = cast(T)raw_data(data)
+	return
+}
+
+
+// `make` built-in procedure allocates and initializes a value of type slice, dynamic array, map, or multi-pointer (only).
+//
+// Similar to `new`, the first argument is a type, not a value. Unlike new, make's return type is the same as the
+// type of its argument, not a pointer to it.
+// Make uses the specified allocator, default is context.allocator.
+@builtin
+make :: proc{
+	make_slice,
+	make_dynamic_array,
+	make_dynamic_array_len,
+	make_dynamic_array_len_cap,
+	make_map,
+	make_multi_pointer,
+}
+
+
+
+// `clear_map` will set the length of a passed map to `0`
+//
+// Note: Prefer the procedure group `clear`
+@builtin
+clear_map :: proc "contextless" (m: ^$T/map[$K]$V) {
+	if m == nil {
+		return
+	}
+	map_clear_dynamic((^Raw_Map)(m), map_info(T))
+}
+
+// `reserve_map` will try to reserve memory of a passed map to the requested element count (setting the `cap`).
+//
+// Note: Prefer the procedure group `reserve`
+@builtin
+reserve_map :: proc(m: ^$T/map[$K]$V, capacity: int, loc := #caller_location) -> Allocator_Error {
+	return __dynamic_map_reserve((^Raw_Map)(m), map_info(T), uint(capacity), loc) if m != nil else nil
+}
+
+// Shrinks the capacity of a map down to the current length.
+//
+// Note: Prefer the procedure group `shrink`
+@builtin
+shrink_map :: proc(m: ^$T/map[$K]$V, loc := #caller_location) -> (did_shrink: bool, err: Allocator_Error) {
+	if m != nil {
+		return map_shrink_dynamic((^Raw_Map)(m), map_info(T), loc)
+	}
+	return
+}
+
+// The delete_key built-in procedure deletes the element with the specified key (m[key]) from the map.
+// If m is nil, or there is no such element, this procedure is a no-op
+@builtin
+delete_key :: proc(m: ^$T/map[$K]$V, key: K) -> (deleted_key: K, deleted_value: V) {
+	if m != nil {
+		key := key
+		old_k, old_v, ok := map_erase_dynamic((^Raw_Map)(m), map_info(T), uintptr(&key))
+		if ok {
+			deleted_key   = (^K)(old_k)^
+			deleted_value = (^V)(old_v)^
+		}
+	}
+	return
+}
+
+_append_elem :: #force_inline proc(array: ^$T/[dynamic]$E, arg: E, should_zero: bool, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	if array == nil {
+		return 0, nil
+	}
+	when size_of(E) == 0 {
+		array := (^Raw_Dynamic_Array)(array)
+		array.len += 1
+		return 1, nil
+	} else {
+		if cap(array) < len(array)+1 {
+			cap := 2 * cap(array) + max(8, 1)
+
+			// do not 'or_return' here as it could be a partial success
+			if should_zero {
+				err = reserve(array, cap, loc)
+			} else {
+				err = non_zero_reserve(array, cap, loc) 
+			}
+		}
+		if cap(array)-len(array) > 0 {
+			a := (^Raw_Dynamic_Array)(array)
+			when size_of(E) != 0 {
+				data := ([^]E)(a.data)
+				assert(data != nil, loc=loc)
+				data[a.len] = arg
+			}
+			a.len += 1
+			return 1, err
+		}
+		return 0, err
+	}
+}
+
+@builtin
+append_elem :: proc(array: ^$T/[dynamic]$E, arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_elem(array, arg, true, loc=loc)
+}
+
+@builtin
+non_zero_append_elem :: proc(array: ^$T/[dynamic]$E, arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_elem(array, arg, false, loc=loc)
+}
+
+_append_elems :: #force_inline proc(array: ^$T/[dynamic]$E, should_zero: bool, loc := #caller_location, args: ..E) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	if array == nil {
+		return 0, nil
+	}
+
+	arg_len := len(args)
+	if arg_len <= 0 {
+		return 0, nil
+	}
+
+	when size_of(E) == 0 {
+		array := (^Raw_Dynamic_Array)(array)
+		array.len += arg_len
+		return arg_len, nil
+	} else {
+		if cap(array) < len(array)+arg_len {
+			cap := 2 * cap(array) + max(8, arg_len)
+
+			// do not 'or_return' here as it could be a partial success
+			if should_zero {
+				err = reserve(array, cap, loc)
+			} else {
+				err = non_zero_reserve(array, cap, loc)
+			}
+		}
+		arg_len = min(cap(array)-len(array), arg_len)
+		if arg_len > 0 {
+			a := (^Raw_Dynamic_Array)(array)
+			when size_of(E) != 0 {
+				data := ([^]E)(a.data)
+				assert(data != nil, loc=loc)
+				intrinsics.mem_copy(&data[a.len], raw_data(args), size_of(E) * arg_len)
+			}
+			a.len += arg_len
+		}
+		return arg_len, err
+	}
+}
+
+@builtin
+append_elems :: proc(array: ^$T/[dynamic]$E, args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_elems(array, true, loc, ..args)
+}
+
+@builtin
+non_zero_append_elems :: proc(array: ^$T/[dynamic]$E, args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_elems(array, false, loc, ..args)
+}
+
+// The append_string built-in procedure appends a string to the end of a [dynamic]u8 like type
+_append_elem_string :: proc(array: ^$T/[dynamic]$E/u8, arg: $A/string, should_zero: bool, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	args := transmute([]E)arg
+	if should_zero { 
+		return append_elems(array, ..args, loc=loc)
+	} else {
+		return non_zero_append_elems(array, ..args, loc=loc)
+	}
+}
+
+@builtin
+append_elem_string :: proc(array: ^$T/[dynamic]$E/u8, arg: $A/string, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_elem_string(array, arg, true, loc)
+}
+@builtin
+non_zero_append_elem_string :: proc(array: ^$T/[dynamic]$E/u8, arg: $A/string, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	return _append_elem_string(array, arg, false, loc)
+}
+
+
+// The append_string built-in procedure appends multiple strings to the end of a [dynamic]u8 like type
+@builtin
+append_string :: proc(array: ^$T/[dynamic]$E/u8, args: ..string, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	n_arg: int
+	for arg in args {
+		n_arg, err = append(array, ..transmute([]E)(arg), loc=loc)
+		n += n_arg
+		if err != nil {
+			return
+		}
+	}
+	return
+}
+
+// The append built-in procedure appends elements to the end of a dynamic array
+@builtin append :: proc{append_elem, append_elems, append_elem_string}
+@builtin non_zero_append :: proc{non_zero_append_elem, non_zero_append_elems, non_zero_append_elem_string}
+
+
+@builtin
+append_nothing :: proc(array: ^$T/[dynamic]$E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	if array == nil {
+		return 0, nil
+	}
+	prev_len := len(array)
+	resize(array, len(array)+1, loc) or_return
+	return len(array)-prev_len, nil
+}
+
+
+@builtin
+inject_at_elem :: proc(array: ^$T/[dynamic]$E, index: int, arg: E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	if array == nil {
+		return
+	}
+	n := max(len(array), index)
+	m :: 1
+	new_size := n + m
+
+	resize(array, new_size, loc) or_return
+	when size_of(E) != 0 {
+		copy(array[index + m:], array[index:])
+		array[index] = arg
+	}
+	ok = true
+	return
+}
+
+@builtin
+inject_at_elems :: proc(array: ^$T/[dynamic]$E, index: int, args: ..E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	if array == nil {
+		return
+	}
+	if len(args) == 0 {
+		ok = true
+		return
+	}
+
+	n := max(len(array), index)
+	m := len(args)
+	new_size := n + m
+
+	resize(array, new_size, loc) or_return
+	when size_of(E) != 0 {
+		copy(array[index + m:], array[index:])
+		copy(array[index:], args)
+	}
+	ok = true
+	return
+}
+
+@builtin
+inject_at_elem_string :: proc(array: ^$T/[dynamic]$E/u8, index: int, arg: string, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	if array == nil {
+		return
+	}
+	if len(arg) == 0 {
+		ok = true
+		return
+	}
+
+	n := max(len(array), index)
+	m := len(arg)
+	new_size := n + m
+
+	resize(array, new_size, loc) or_return
+	copy(array[index+m:], array[index:])
+	copy(array[index:], arg)
+	ok = true
+	return
+}
+
+@builtin inject_at :: proc{inject_at_elem, inject_at_elems, inject_at_elem_string}
+
+
+
+@builtin
+assign_at_elem :: proc(array: ^$T/[dynamic]$E, index: int, arg: E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	if index < len(array) {
+		array[index] = arg
+		ok = true
+	} else {
+		resize(array, index+1, loc) or_return
+		array[index] = arg
+		ok = true
+	}
+	return
+}
+
+
+@builtin
+assign_at_elems :: proc(array: ^$T/[dynamic]$E, index: int, args: ..E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	new_size := index + len(args)
+	if len(args) == 0 {
+		ok = true
+	} else if new_size < len(array) {
+		copy(array[index:], args)
+		ok = true
+	} else {
+		resize(array, new_size, loc) or_return
+		copy(array[index:], args)
+		ok = true
+	}
+	return
+}
+
+
+@builtin
+assign_at_elem_string :: proc(array: ^$T/[dynamic]$E/u8, index: int, arg: string, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error {
+	new_size := index + len(arg)
+	if len(arg) == 0 {
+		ok = true
+	} else if new_size < len(array) {
+		copy(array[index:], arg)
+		ok = true
+	} else {
+		resize(array, new_size, loc) or_return
+		copy(array[index:], arg)
+		ok = true
+	}
+	return
+}
+
+@builtin assign_at :: proc{assign_at_elem, assign_at_elems, assign_at_elem_string}
+
+
+
+
+// `clear_dynamic_array` will set the length of a passed dynamic array to `0`
+//
+// Note: Prefer the procedure group `clear`.
+@builtin
+clear_dynamic_array :: proc "contextless" (array: ^$T/[dynamic]$E) {
+	if array != nil {
+		(^Raw_Dynamic_Array)(array).len = 0
+	}
+}
+
+// `reserve_dynamic_array` will try to reserve memory of a passed dynamic array or map to the requested element count (setting the `cap`).
+//
+// Note: Prefer the procedure group `reserve`.
+_reserve_dynamic_array :: #force_inline proc(array: ^$T/[dynamic]$E, capacity: int, should_zero: bool, loc := #caller_location) -> Allocator_Error {
+	if array == nil {
+		return nil
+	}
+	a := (^Raw_Dynamic_Array)(array)
+
+	if capacity <= a.cap {
+		return nil
+	}
+
+	if a.allocator.procedure == nil {
+		a.allocator = context.allocator
+	}
+	assert(a.allocator.procedure != nil)
+
+	old_size  := a.cap * size_of(E)
+	new_size  := capacity * size_of(E)
+	allocator := a.allocator
+
+	new_data: []byte
+	if should_zero {
+		new_data = mem_resize(a.data, old_size, new_size, align_of(E), allocator, loc) or_return
+	} else {
+		new_data = non_zero_mem_resize(a.data, old_size, new_size, align_of(E), allocator, loc) or_return
+	}
+	if new_data == nil && new_size > 0 {
+		return .Out_Of_Memory
+	}
+
+	a.data = raw_data(new_data)
+	a.cap = capacity
+	return nil
+}
+
+@builtin
+reserve_dynamic_array :: proc(array: ^$T/[dynamic]$E, capacity: int, loc := #caller_location) -> Allocator_Error {
+	return _reserve_dynamic_array(array, capacity, true, loc)
+}
+
+@builtin
+non_zero_reserve_dynamic_array :: proc(array: ^$T/[dynamic]$E, capacity: int, loc := #caller_location) -> Allocator_Error {
+	return _reserve_dynamic_array(array, capacity, false, loc)
+}
+
+// `resize_dynamic_array` will try to resize memory of a passed dynamic array or map to the requested element count (setting the `len`, and possibly `cap`).
+//
+// Note: Prefer the procedure group `resize`
+_resize_dynamic_array :: #force_inline proc(array: ^$T/[dynamic]$E, length: int, should_zero: bool, loc := #caller_location) -> Allocator_Error {
+	if array == nil {
+		return nil
+	}
+	a := (^Raw_Dynamic_Array)(array)
+
+	if length <= a.cap {
+		a.len = max(length, 0)
+		return nil
+	}
+
+	if a.allocator.procedure == nil {
+		a.allocator = context.allocator
+	}
+	assert(a.allocator.procedure != nil)
+
+	old_size  := a.cap * size_of(E)
+	new_size  := length * size_of(E)
+	allocator := a.allocator
+
+	new_data : []byte
+	if should_zero {
+		new_data = mem_resize(a.data, old_size, new_size, align_of(E), allocator, loc) or_return
+	} else {
+		new_data = non_zero_mem_resize(a.data, old_size, new_size, align_of(E), allocator, loc) or_return
+	}
+	if new_data == nil && new_size > 0 {
+		return .Out_Of_Memory
+	}
+
+	a.data = raw_data(new_data)
+	a.len = length
+	a.cap = length
+	return nil
+}
+
+@builtin
+resize_dynamic_array :: proc(array: ^$T/[dynamic]$E, length: int, loc := #caller_location) -> Allocator_Error {
+	return _resize_dynamic_array(array, length, true, loc=loc)
+}
+
+@builtin
+non_zero_resize_dynamic_array :: proc(array: ^$T/[dynamic]$E, length: int, loc := #caller_location) -> Allocator_Error {
+	return _resize_dynamic_array(array, length, false, loc=loc)
+}
+
+/*
+	Shrinks the capacity of a dynamic array down to the current length, or the given capacity.
+
+	If `new_cap` is negative, then `len(array)` is used.
+
+	Returns false if `cap(array) < new_cap`, or the allocator report failure.
+
+	If `len(array) < new_cap`, then `len(array)` will be left unchanged.
+
+	Note: Prefer the procedure group `shrink`
+*/
+shrink_dynamic_array :: proc(array: ^$T/[dynamic]$E, new_cap := -1, loc := #caller_location) -> (did_shrink: bool, err: Allocator_Error) {
+	if array == nil {
+		return
+	}
+	a := (^Raw_Dynamic_Array)(array)
+
+	new_cap := new_cap if new_cap >= 0 else a.len
+
+	if new_cap > a.cap {
+		return
+	}
+
+	if a.allocator.procedure == nil {
+		a.allocator = context.allocator
+	}
+	assert(a.allocator.procedure != nil)
+
+	old_size := a.cap * size_of(E)
+	new_size := new_cap * size_of(E)
+
+	new_data := mem_resize(a.data, old_size, new_size, align_of(E), a.allocator, loc) or_return
+
+	a.data = raw_data(new_data)
+	a.len = min(new_cap, a.len)
+	a.cap = new_cap
+	return true, nil
+}
+
+@builtin
+map_insert :: proc(m: ^$T/map[$K]$V, key: K, value: V, loc := #caller_location) -> (ptr: ^V) {
+	key, value := key, value
+	return (^V)(__dynamic_map_set_without_hash((^Raw_Map)(m), map_info(T), rawptr(&key), rawptr(&value), loc))
+}
+
+
+@builtin
+card :: proc(s: $S/bit_set[$E; $U]) -> int {
+	when size_of(S) == 1 {
+		return int(intrinsics.count_ones(transmute(u8)s))
+	} else when size_of(S) == 2 {
+		return int(intrinsics.count_ones(transmute(u16)s))
+	} else when size_of(S) == 4 {
+		return int(intrinsics.count_ones(transmute(u32)s))
+	} else when size_of(S) == 8 {
+		return int(intrinsics.count_ones(transmute(u64)s))
+	} else when size_of(S) == 16 {
+		return int(intrinsics.count_ones(transmute(u128)s))
+	} else {
+		#panic("Unhandled card bit_set size")
+	}
+}
+
+
+
+@builtin
+@(disabled=ODIN_DISABLE_ASSERT)
+assert :: proc(condition: bool, message := "", loc := #caller_location) {
+	if !condition {
+		// NOTE(bill): This is wrapped in a procedure call
+		// to improve performance to make the CPU not
+		// execute speculatively, making it about an order of
+		// magnitude faster
+		@(cold)
+		internal :: proc(message: string, loc: Source_Code_Location) {
+			p := context.assertion_failure_proc
+			if p == nil {
+				p = default_assertion_failure_proc
+			}
+			p("runtime assertion", message, loc)
+		}
+		internal(message, loc)
+	}
+}
+
+@builtin
+panic :: proc(message: string, loc := #caller_location) -> ! {
+	p := context.assertion_failure_proc
+	if p == nil {
+		p = default_assertion_failure_proc
+	}
+	p("panic", message, loc)
+}
+
+@builtin
+unimplemented :: proc(message := "", loc := #caller_location) -> ! {
+	p := context.assertion_failure_proc
+	if p == nil {
+		p = default_assertion_failure_proc
+	}
+	p("not yet implemented", message, loc)
+}
@@ -0,0 +1,429 @@
+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[]$E) -> (footer: ^Raw_SOA_Footer_Slice) {
+	if array == nil {
+		return nil
+	}
+	field_count := uintptr(intrinsics.type_struct_field_count(E))
+	footer = (^Raw_SOA_Footer_Slice)(uintptr(array) + field_count*size_of(rawptr))
+	return
+}
+@(builtin, require_results)
+raw_soa_footer_dynamic_array :: proc(array: ^$T/#soa[dynamic]$E) -> (footer: ^Raw_SOA_Footer_Dynamic_Array) {
+	if array == nil {
+		return nil
+	}
+	field_count: uintptr
+	when intrinsics.type_is_array(E) {
+		field_count = len(E)
+	} else {
+		field_count = uintptr(intrinsics.type_struct_field_count(E))
+	}
+	footer = (^Raw_SOA_Footer_Dynamic_Array)(uintptr(array) + field_count*size_of(rawptr))
+	return
+}
+raw_soa_footer :: proc{
+	raw_soa_footer_slice,
+	raw_soa_footer_dynamic_array,
+}
+
+
+
+@(builtin, require_results)
+make_soa_aligned :: proc($T: typeid/#soa[]$E, length: int, alignment: int, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) #optional_allocator_error {
+	if length <= 0 {
+		return
+	}
+
+	array.allocator = allocator
+	footer := raw_soa_footer(&array)
+	if size_of(E) == 0 {
+		footer.len = length
+		return
+	}
+
+	max_align := max(alignment, align_of(E))
+
+	ti := type_info_of(typeid_of(T))
+	ti = type_info_base(ti)
+	si := &ti.variant.(Type_Info_Struct)
+
+	field_count := uintptr(intrinsics.type_struct_field_count(E))
+
+	total_size := 0
+	for i in 0..<field_count {
+		type := si.types[i].variant.(Type_Info_Pointer).elem
+		total_size += type.size * length
+		total_size = align_forward_int(total_size, max_align)
+	}
+
+	allocator := allocator
+	if allocator.procedure == nil {
+		allocator = context.allocator
+	}
+	assert(allocator.procedure != nil)
+
+	new_bytes: []byte
+	new_bytes, err = allocator.procedure(
+		allocator.data, .Alloc, total_size, max_align,
+		nil, 0, loc,
+	)
+	if new_bytes == nil || err != nil {
+		return
+	}
+	new_data := raw_data(new_bytes)
+
+	data := uintptr(&array)
+	offset := 0
+	for i in 0..<field_count {
+		type := si.types[i].variant.(Type_Info_Pointer).elem
+
+		offset = align_forward_int(offset, max_align)
+
+		(^uintptr)(data)^ = uintptr(new_data) + uintptr(offset)
+		data += size_of(rawptr)
+		offset += type.size * length
+	}
+	footer.len = length
+
+	return
+}
+
+@(builtin, require_results)
+make_soa_slice :: proc($T: typeid/#soa[]$E, length: int, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) #optional_allocator_error {
+	return make_soa_aligned(T, length, align_of(E), allocator, loc)
+}
+
+@(builtin, require_results)
+make_soa_dynamic_array :: proc($T: typeid/#soa[dynamic]$E, allocator := context.allocator, loc := #caller_location) -> (array: T, err: Allocator_Error) #optional_allocator_error {
+	context.allocator = allocator
+	reserve_soa(&array, DEFAULT_RESERVE_CAPACITY, loc) or_return
+	return array, nil
+}
+
+@(builtin, require_results)
+make_soa_dynamic_array_len :: proc($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]$E, length: int, loc := #caller_location) -> Allocator_Error {
+	if array == nil {
+		return nil
+	}
+	reserve_soa(array, length, loc) or_return
+	footer := raw_soa_footer(array)
+	footer.len = length
+	return nil
+}
+
+@builtin
+reserve_soa :: proc(array: ^$T/#soa[dynamic]$E, capacity: int, 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
+	when intrinsics.type_is_array(E) {
+		field_count = len(E)
+	} else {
+		field_count = uintptr(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_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, 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_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]$E, arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error {
+	if array == nil {
+		return 0, nil
+	}
+
+	if cap(array) <= len(array) + 1 {
+		cap := 2 * cap(array) + 8
+		err = reserve_soa(array, cap, loc) // do not 'or_return' here as it could be a partial success
+	}
+
+	footer := raw_soa_footer(array)
+
+	if size_of(E) > 0 && cap(array)-len(array) > 0 {
+		ti := type_info_of(T)
+		ti = type_info_base(ti)
+		si := &ti.variant.(Type_Info_Struct)
+		field_count: uintptr
+		when intrinsics.type_is_array(E) {
+			field_count = len(E)
+		} else {
+			field_count = uintptr(intrinsics.type_struct_field_count(E))
+		}
+
+		data := (^rawptr)(array)^
+
+		soa_offset := 0
+		item_offset := 0
+
+		arg_copy := arg
+		arg_ptr := &arg_copy
+
+		max_align :: align_of(E)
+		for i in 0..<field_count {
+			type := si.types[i].variant.(Type_Info_Pointer).elem
+
+			soa_offset  = align_forward_int(soa_offset, max_align)
+			item_offset = align_forward_int(item_offset, type.align)
+
+			dst := rawptr(uintptr(data) + uintptr(soa_offset) + uintptr(type.size * footer.len))
+			src := rawptr(uintptr(arg_ptr) + uintptr(item_offset))
+			mem_copy(dst, src, type.size)
+
+			soa_offset  += type.size * cap(array)
+			item_offset += type.size
+		}
+		footer.len += 1
+		return 1, err
+	}
+	return 0, err
+}
+
+@builtin
+append_soa_elems :: proc(array: ^$T/#soa[dynamic]$E, 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(8, arg_len)
+		err = reserve_soa(array, cap, loc) // do not 'or_return' here as it could be a partial success
+	}
+	arg_len = min(cap(array)-len(array), arg_len)
+
+	footer := raw_soa_footer(array)
+	if size_of(E) > 0 && arg_len > 0 {
+		ti := type_info_of(typeid_of(T))
+		ti = type_info_base(ti)
+		si := &ti.variant.(Type_Info_Struct)
+		field_count := uintptr(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_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 {
+	when intrinsics.type_struct_field_count(E) != 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 {
+	when intrinsics.type_struct_field_count(E) != 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) {
+	when intrinsics.type_struct_field_count(E) != 0 {
+		footer := raw_soa_footer(array)
+		footer.len = 0
+	}
+}
+
+@builtin
+clear_soa :: proc{
+	clear_soa_dynamic_array,
+}
@@ -0,0 +1,304 @@
+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,
+}
+
+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 usuable 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 +1,12 @@
+package runtime
+
+when ODIN_DEFAULT_TO_NIL_ALLOCATOR {
+	default_allocator_proc :: nil_allocator_proc
+	default_allocator :: nil_allocator
+} else when ODIN_DEFAULT_TO_PANIC_ALLOCATOR {
+	default_allocator_proc :: panic_allocator_proc
+	default_allocator :: panic_allocator
+} else {
+	default_allocator :: heap_allocator
+	default_allocator_proc :: heap_allocator_proc
+}
@@ -0,0 +1,80 @@
+package runtime
+
+nil_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
+                               size, alignment: int,
+                               old_memory: rawptr, old_size: int, loc := #caller_location) -> ([]byte, Allocator_Error) {
+	switch mode {
+	case .Alloc, .Alloc_Non_Zeroed:
+		return nil, .Out_Of_Memory
+	case .Free:
+		return nil, .None
+	case .Free_All:
+		return nil, .Mode_Not_Implemented
+	case .Resize, .Resize_Non_Zeroed:
+		if size == 0 {
+			return nil, .None
+		}
+		return nil, .Out_Of_Memory
+	case .Query_Features:
+		return nil, .Mode_Not_Implemented
+	case .Query_Info:
+		return nil, .Mode_Not_Implemented
+	}
+	return nil, .None
+}
+
+nil_allocator :: proc() -> Allocator {
+	return Allocator{
+		procedure = nil_allocator_proc,
+		data = nil,
+	}
+}
+
+
+panic_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
+                             size, alignment: int,
+                             old_memory: rawptr, old_size: int, loc := #caller_location) -> ([]byte, Allocator_Error) {
+	switch mode {
+	case .Alloc:
+		if size > 0 {
+			panic("panic allocator, .Alloc called", loc=loc)
+		}
+	case .Alloc_Non_Zeroed:
+		if size > 0 {
+			panic("panic allocator, .Alloc_Non_Zeroed called", loc=loc)
+		}
+	case .Resize:
+		if size > 0 {
+			panic("panic allocator, .Resize called", loc=loc)
+		}
+	case .Resize_Non_Zeroed:
+		if size > 0 {
+			panic("panic allocator, .Alloc_Non_Zeroed called", loc=loc)
+		}
+	case .Free:
+		if old_memory != nil {
+			panic("panic allocator, .Free called", loc=loc)
+		}
+	case .Free_All:
+		panic("panic allocator, .Free_All called", loc=loc)
+
+	case .Query_Features:
+		set := (^Allocator_Mode_Set)(old_memory)
+		if set != nil {
+			set^ = {.Query_Features}
+		}
+		return nil, nil
+
+	case .Query_Info:
+		panic("panic allocator, .Query_Info called", loc=loc)
+	}
+
+	return nil, nil
+}
+
+panic_allocator :: proc() -> Allocator {
+	return Allocator{
+		procedure = panic_allocator_proc,
+		data = nil,
+	}
+}
@@ -0,0 +1,79 @@
+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_OS == .JS || 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 +1,179 @@
+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
+umodti3
+udivti3
+modti3
+divti3
+fixdfti
+fixunsdfti
+fixunsdfdi
+floattidf
+floattidf_unsigned
+truncsfhf2
+truncdfhf2
+gnu_h2f_ieee
+gnu_f2h_ieee
+extendhfsf2
+__ashlti3 // wasm specific
+__multi3  // wasm specific
+
+
+
+## Required an entry point is defined (i.e. 'main')
+
+args__
+
+
+## When -no-crt is defined (and not a wasm target) (mostly due to LLVM)
+_tls_index
+_fltused
+
+
+## Bounds checking procedures (when not disabled with -no-bounds-check)
+
+bounds_check_error
+matrix_bounds_check_error
+slice_expr_error_hi
+slice_expr_error_lo_hi
+multi_pointer_slice_expr_error
+
+
+## Type assertion check
+
+type_assertion_check
+type_assertion_check2 // takes in typeid
+
+
+## Arithmetic
+
+quo_complex32
+quo_complex64
+quo_complex128
+
+mul_quaternion64
+mul_quaternion128
+mul_quaternion256
+
+quo_quaternion64
+quo_quaternion128
+quo_quaternion256
+
+abs_complex32
+abs_complex64
+abs_complex128
+
+abs_quaternion64
+abs_quaternion128
+abs_quaternion256
+
+
+## Comparison
+
+memory_equal
+memory_compare
+memory_compare_zero
+
+cstring_eq
+cstring_ne
+cstring_lt
+cstring_gt
+cstring_le
+cstring_gt
+
+string_eq
+string_ne
+string_lt
+string_gt
+string_le
+string_gt
+
+complex32_eq
+complex32_ne
+complex64_eq
+complex64_ne
+complex128_eq
+complex128_ne
+
+quaternion64_eq
+quaternion64_ne
+quaternion128_eq
+quaternion128_ne
+quaternion256_eq
+quaternion256_ne
+
+
+## Map specific calls
+
+map_seed_from_map_data
+__dynamic_map_check_grow // static map calls
+map_insert_hash_dynamic  // static map calls
+__dynamic_map_get // dynamic map calls
+__dynamic_map_set // dynamic map calls
+
+
+## Dynamic literals ([dymamic]T and map[K]V) (can be disabled with -no-dynamic-literals)
+
+__dynamic_array_reserve
+__dynamic_array_append
+
+__dynamic_map_reserve
+
+
+## Objective-C specific
+
+objc_lookUpClass
+sel_registerName
+objc_allocateClassPair
+
+
+## for-in `string` type
+
+string_decode_rune
+string_decode_last_rune // #reverse for
+
+*/
@@ -0,0 +1,138 @@
+package runtime
+
+__dynamic_array_make :: proc(array_: rawptr, elem_size, elem_align: int, len, cap: int, loc := #caller_location) {
+	array := (^Raw_Dynamic_Array)(array_)
+	array.allocator = context.allocator
+	assert(array.allocator.procedure != nil)
+
+	if cap > 0 {
+		__dynamic_array_reserve(array_, elem_size, elem_align, cap, loc)
+		array.len = len
+	}
+}
+
+__dynamic_array_reserve :: proc(array_: rawptr, elem_size, elem_align: int, cap: int, loc := #caller_location) -> bool {
+	array := (^Raw_Dynamic_Array)(array_)
+
+	// NOTE(tetra, 2020-01-26): We set the allocator before earlying-out below, because user code is usually written
+	// assuming that appending/reserving will set the allocator, if it is not already set.
+	if array.allocator.procedure == nil {
+		array.allocator = context.allocator
+	}
+	assert(array.allocator.procedure != nil)
+
+	if cap <= array.cap {
+		return true
+	}
+
+	old_size  := array.cap * elem_size
+	new_size  := cap * elem_size
+	allocator := array.allocator
+
+	new_data, err := mem_resize(array.data, old_size, new_size, elem_align, allocator, loc)
+	if err != nil {
+		return false
+	}
+	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
+	}
+	return false
+}
+
+__dynamic_array_shrink :: proc(array_: rawptr, elem_size, elem_align: int, new_cap: int, loc := #caller_location) -> (did_shrink: bool) {
+	array := (^Raw_Dynamic_Array)(array_)
+
+	// NOTE(tetra, 2020-01-26): We set the allocator before earlying-out below, because user code is usually written
+	// assuming that appending/reserving will set the allocator, if it is not already set.
+	if array.allocator.procedure == nil {
+		array.allocator = context.allocator
+	}
+	assert(array.allocator.procedure != nil)
+
+	if new_cap > array.cap {
+		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 := mem_resize(array.data, old_size, new_size, elem_align, allocator, loc)
+	if err != nil {
+		return
+	}
+
+	array.data = raw_data(new_data)
+	array.len = min(new_cap, array.len)
+	array.cap = new_cap
+	return true
+}
+
+__dynamic_array_resize :: proc(array_: rawptr, elem_size, elem_align: int, len: int, loc := #caller_location) -> bool {
+	array := (^Raw_Dynamic_Array)(array_)
+
+	ok := __dynamic_array_reserve(array_, elem_size, elem_align, len, loc)
+	if ok {
+		array.len = len
+	}
+	return ok
+}
+
+
+__dynamic_array_append :: proc(array_: rawptr, elem_size, elem_align: int,
+                               items: rawptr, item_count: int, loc := #caller_location) -> int {
+	array := (^Raw_Dynamic_Array)(array_)
+
+	if items == nil    {
+		return 0
+	}
+	if item_count <= 0 {
+		return 0
+	}
+
+
+	ok := true
+	if array.cap < array.len+item_count {
+		cap := 2 * array.cap + max(8, item_count)
+		ok = __dynamic_array_reserve(array, elem_size, elem_align, cap, loc)
+	}
+	// TODO(bill): Better error handling for failed reservation
+	if !ok {
+		return array.len
+	}
+
+	assert(array.data != nil)
+	data := uintptr(array.data) + uintptr(elem_size*array.len)
+
+	mem_copy(rawptr(data), items, elem_size * item_count)
+	array.len += item_count
+	return array.len
+}
+
+__dynamic_array_append_nothing :: proc(array_: rawptr, elem_size, elem_align: int, loc := #caller_location) -> int {
+	array := (^Raw_Dynamic_Array)(array_)
+
+	ok := true
+	if array.cap < array.len+1 {
+		cap := 2 * array.cap + max(8, 1)
+		ok = __dynamic_array_reserve(array, elem_size, elem_align, cap, loc)
+	}
+	// TODO(bill): Better error handling for failed reservation
+	if !ok {
+		return array.len
+	}
+
+	assert(array.data != nil)
+	data := uintptr(array.data) + uintptr(elem_size*array.len)
+	mem_zero(rawptr(data), elem_size)
+	array.len += 1
+	return array.len
+}
@@ -0,0 +1,924 @@
+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
+}
+
+
+@(private, 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
+}
+
+// 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.
+@(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 {
+		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 :: 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 :: 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 :: 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 :: 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 :: 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_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
+	}
+}
+
+// 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))
+	m.len += 1
+	return 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 +1,59 @@
+//+private
+//+build linux, darwin, freebsd, openbsd
+//+no-instrumentation
+package runtime
+
+import "base:intrinsics"
+
+when ODIN_BUILD_MODE == .Dynamic {
+	@(link_name="_odin_entry_point", linkage="strong", require/*, link_section=".init"*/)
+	_odin_entry_point :: proc "c" () {
+		context = default_context()
+		#force_no_inline _startup_runtime()
+		intrinsics.__entry_point()
+	}
+	@(link_name="_odin_exit_point", linkage="strong", require/*, link_section=".fini"*/)
+	_odin_exit_point :: proc "c" () {
+		context = default_context()
+		#force_no_inline _cleanup_runtime()
+	}
+	@(link_name="main", linkage="strong", require)
+	main :: proc "c" (argc: i32, argv: [^]cstring) -> i32 {
+		return 0
+	}
+} else when !ODIN_TEST && !ODIN_NO_ENTRY_POINT {
+	when ODIN_NO_CRT {
+		// NOTE(flysand): We need to start from assembly because we need
+		// to retrieve argc and argv from the stack
+		when ODIN_ARCH == .amd64 {
+			@require foreign import entry "entry_unix_no_crt_amd64.asm"
+			SYS_exit :: 60
+		} else when ODIN_ARCH == .i386 {
+			@require foreign import entry "entry_unix_no_crt_i386.asm"
+			SYS_exit :: 1
+		} else when ODIN_OS == .Darwin && ODIN_ARCH == .arm64 {
+			@require foreign import entry "entry_unix_no_crt_darwin_arm64.asm"
+			SYS_exit :: 1
+		}
+		@(link_name="_start_odin", linkage="strong", require)
+		_start_odin :: proc "c" (argc: i32, argv: [^]cstring) -> ! {
+			args__ = argv[:argc]
+			context = default_context()
+			#force_no_inline _startup_runtime()
+			intrinsics.__entry_point()
+			#force_no_inline _cleanup_runtime()
+			intrinsics.syscall(SYS_exit, 0)
+			unreachable()
+		}
+	} else {
+		@(link_name="main", linkage="strong", require)
+		main :: proc "c" (argc: i32, argv: [^]cstring) -> i32 {
+			args__ = argv[:argc]
+			context = default_context()
+			#force_no_inline _startup_runtime()
+			intrinsics.__entry_point()
+			#force_no_inline _cleanup_runtime()
+			return 0
+		}
+	}
+}
@@ -0,0 +1,43 @@
+bits 64
+
+extern _start_odin
+global _start
+
+section .text
+
+;; Entry point for programs that specify -no-crt option
+;; This entry point should be compatible with dynamic loaders on linux
+;; The parameters the dynamic loader passes to the _start function:
+;;    RDX = pointer to atexit function
+;; The stack layout is as follows:
+;;    +-------------------+
+;;            NULL
+;;    +-------------------+
+;;           envp[m]
+;;    +-------------------+
+;;            ...
+;;    +-------------------+
+;;           envp[0]
+;;    +-------------------+
+;;            NULL
+;;    +-------------------+
+;;           argv[n]
+;;    +-------------------+
+;;            ...
+;;    +-------------------+
+;;           argv[0]
+;;    +-------------------+
+;;            argc
+;;    +-------------------+ <------ RSP
+;;
+_start:
+    ;; Mark stack frame as the top of the stack
+    xor rbp, rbp
+    ;; Load argc into 1st param reg, argv into 2nd param reg
+    pop rdi
+    mov rdx, rsi
+    ;; Align stack pointer down to 16-bytes (sysv calling convention)
+    and rsp, -16
+    ;; Call into odin entry point
+    call _start_odin
+    jmp $$
@@ -0,0 +1,20 @@
+	.section __TEXT,__text
+
+	; NOTE(laytan): this should ideally be the -minimum-os-version flag but there is no nice way of preprocessing assembly in Odin.
+	; 10 seems to be the lowest it goes and I don't see it mess with any targeted os version so this seems fine.
+	.build_version macos, 10, 0
+
+	.extern __start_odin
+
+	.global _main
+	.align 2
+_main:
+	mov x5, sp       ; use x5 as the stack pointer
+
+	str x0, [x5]     ; get argc into x0 (kernel passes 32-bit int argc as 64-bits on stack to keep alignment)
+	str x1, [x5, #8] ; get argv into x1
+
+	and sp, x5, #~15 ; force 16-byte alignment of the stack
+	
+	bl __start_odin  ; call into Odin entry point
+	ret              ; should never get here
@@ -0,0 +1,18 @@
+bits 32
+
+extern _start_odin
+global _start
+
+section .text
+
+;; NOTE(flysand): For description see the corresponding *_amd64.asm file
+;; also I didn't test this on x86-32
+_start:
+    xor ebp, rbp
+    pop ecx
+    mov eax, esp
+    and esp, -16
+    push eax
+    push ecx
+    call _start_odin
+    jmp $$
@@ -0,0 +1,20 @@
+//+private
+//+build wasm32, wasm64p32
+//+no-instrumentation
+package runtime
+
+import "base:intrinsics"
+
+when !ODIN_TEST && !ODIN_NO_ENTRY_POINT {
+	@(link_name="_start", linkage="strong", require, export)
+	_start :: proc "c" () {
+		context = default_context()
+		#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()
+	}
+}
@@ -0,0 +1,50 @@
+//+private
+//+build windows
+//+no-instrumentation
+package runtime
+
+import "base:intrinsics"
+
+when ODIN_BUILD_MODE == .Dynamic {
+	@(link_name="DllMain", linkage="strong", require)
+	DllMain :: proc "system" (hinstDLL: rawptr, fdwReason: u32, lpReserved: rawptr) -> b32 {
+		context = default_context()
+
+		// Populate Windows DLL-specific global
+		dll_forward_reason = DLL_Forward_Reason(fdwReason)
+
+		switch dll_forward_reason {
+		case .Process_Attach:
+			#force_no_inline _startup_runtime()
+			intrinsics.__entry_point()
+		case .Process_Detach:
+			#force_no_inline _cleanup_runtime()
+		case .Thread_Attach:
+			break
+		case .Thread_Detach:
+			break
+		}
+		return true
+	}
+} else when !ODIN_TEST && !ODIN_NO_ENTRY_POINT {
+	when ODIN_ARCH == .i386 || ODIN_NO_CRT {
+		@(link_name="mainCRTStartup", linkage="strong", require)
+		mainCRTStartup :: proc "system" () -> i32 {
+			context = default_context()
+			#force_no_inline _startup_runtime()
+			intrinsics.__entry_point()
+			#force_no_inline _cleanup_runtime()
+			return 0
+		}
+	} 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 +1,292 @@
+package runtime
+
+@(no_instrumentation)
+bounds_trap :: proc "contextless" () -> ! {
+	when ODIN_OS == .Windows {
+		windows_trap_array_bounds()
+	} else {
+		trap()
+	}
+}
+
+@(no_instrumentation)
+type_assertion_trap :: proc "contextless" () -> ! {
+	when ODIN_OS == .Windows {
+		windows_trap_type_assertion()
+	} else {
+		trap()
+	}
+}
+
+
+bounds_check_error :: proc "contextless" (file: string, line, column: i32, index, count: int) {
+	if uint(index) < uint(count) {
+		return
+	}
+	@(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))
+		print_string(" is out of range 0..<")
+		print_i64(i64(count))
+		print_byte('\n')
+		bounds_trap()
+	}
+	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 ")
+	print_i64(i64(lo))
+	print_string(":")
+	print_i64(i64(hi))
+	print_string(" is out of range 0..<")
+	print_i64(i64(len))
+	print_byte('\n')
+	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 ")
+	print_i64(i64(lo))
+	print_string(":")
+	print_i64(i64(hi))
+	print_byte('\n')
+	bounds_trap()
+}
+
+
+multi_pointer_slice_expr_error :: proc "contextless" (file: string, line, column: i32, lo, hi: int) {
+	if lo <= hi {
+		return
+	}
+	multi_pointer_slice_handle_error(file, line, column, lo, hi)
+}
+
+slice_expr_error_hi :: proc "contextless" (file: string, line, column: i32, hi: int, len: int) {
+	if 0 <= hi && hi <= len {
+		return
+	}
+	slice_handle_error(file, line, column, 0, hi, len)
+}
+
+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
+	}
+	slice_handle_error(file, line, column, lo, hi, len)
+}
+
+dynamic_array_expr_error :: proc "contextless" (file: string, line, column: i32, low, high, max: int) {
+	if 0 <= low && low <= high && high <= max {
+		return
+	}
+	@(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))
+		print_string(":")
+		print_i64(i64(high))
+		print_string(" is out of range 0..<")
+		print_i64(i64(max))
+		print_byte('\n')
+		bounds_trap()
+	}
+	handle_error(file, line, column, low, high, max)
+}
+
+
+matrix_bounds_check_error :: proc "contextless" (file: string, line, column: i32, row_index, column_index, row_count, column_count: int) {
+	if uint(row_index) < uint(row_count) &&
+	   uint(column_index) < uint(column_count) {
+		return
+	}
+	@(cold, 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))
+		print_string(", ")
+		print_i64(i64(column_index))
+		print_string(" is out of range [0..<")
+		print_i64(i64(row_count))
+		print_string(", 0..<")
+		print_i64(i64(column_count))
+		print_string("]")
+		print_byte('\n')
+		bounds_trap()
+	}
+	handle_error(file, line, column, row_index, column_index, row_count, column_count)
+}
+
+
+when ODIN_NO_RTTI {
+	type_assertion_check :: proc "contextless" (ok: bool, file: string, line, column: i32) {
+		if ok {
+			return
+		}
+		@(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()
+		}
+		handle_error(file, line, column)
+	}
+
+	type_assertion_check2 :: proc "contextless" (ok: bool, file: string, line, column: i32) {
+		if ok {
+			return
+		}
+		@(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()
+		}
+		handle_error(file, line, column)
+	}
+} else {
+	type_assertion_check :: proc "contextless" (ok: bool, file: string, line, column: i32, from, to: typeid) {
+		if ok {
+			return
+		}
+		@(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)
+			print_string(" to ")
+			print_typeid(to)
+			print_byte('\n')
+			type_assertion_trap()
+		}
+		handle_error(file, line, column, from, to)
+	}
+
+	type_assertion_check2 :: proc "contextless" (ok: bool, file: string, line, column: i32, from, to: typeid, from_data: rawptr) {
+		if ok {
+			return
+		}
+
+		variant_type :: proc "contextless" (id: typeid, data: rawptr) -> typeid {
+			if id == nil || data == nil {
+				return id
+			}
+			ti := type_info_base(type_info_of(id))
+			#partial switch v in ti.variant {
+			case Type_Info_Any:
+				return (^any)(data).id
+			case Type_Info_Union:
+				tag_ptr := uintptr(data) + v.tag_offset
+				idx := 0
+				switch v.tag_type.size {
+				case 1:  idx = int((^u8)(tag_ptr)^)   - 1
+				case 2:  idx = int((^u16)(tag_ptr)^)  - 1
+				case 4:  idx = int((^u32)(tag_ptr)^)  - 1
+				case 8:  idx = int((^u64)(tag_ptr)^)  - 1
+				case 16: idx = int((^u128)(tag_ptr)^) - 1
+				}
+				if idx < 0 {
+					return nil
+				} else if idx < len(v.variants) {
+					return v.variants[idx].id
+				}
+			}
+			return id
+		}
+
+		@(cold, no_instrumentation)
+		handle_error :: proc "contextless" (file: string, line, column: i32, from, to: typeid, from_data: rawptr) -> ! {
+
+			actual := variant_type(from, from_data)
+
+			print_caller_location(Source_Code_Location{file, line, column, ""})
+			print_string(" Invalid type assertion from ")
+			print_typeid(from)
+			print_string(" to ")
+			print_typeid(to)
+			if actual != from {
+				print_string(", actual type: ")
+				print_typeid(actual)
+			}
+			print_byte('\n')
+			type_assertion_trap()
+		}
+		handle_error(file, line, column, from, to, from_data)
+	}
+}
+
+
+make_slice_error_loc :: #force_inline proc "contextless" (loc := #caller_location, len: int) {
+	if 0 <= len {
+		return
+	}
+	@(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))
+		print_byte('\n')
+		bounds_trap()
+	}
+	handle_error(loc, len)
+}
+
+make_dynamic_array_error_loc :: #force_inline proc "contextless" (loc := #caller_location, len, cap: int) {
+	if 0 <= len && len <= cap {
+		return
+	}
+	@(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))
+		print_byte(':')
+		print_i64(i64(cap))
+		print_byte('\n')
+		bounds_trap()
+	}
+	handle_error(loc, len, cap)
+}
+
+make_map_expr_error_loc :: #force_inline proc "contextless" (loc := #caller_location, cap: int) {
+	if 0 <= cap {
+		return
+	}
+	@(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))
+		print_byte('\n')
+		bounds_trap()
+	}
+	handle_error(loc, cap)
+}
+
+
+
+
+
+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" (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" (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" (loc := #caller_location, low, high, max: int) {
+	dynamic_array_expr_error(loc.file_path, loc.line, loc.column, low, high, max)
+}
@@ -0,0 +1,110 @@
+package runtime
+
+import "base:intrinsics"
+
+heap_allocator :: proc() -> Allocator {
+	return Allocator{
+		procedure = heap_allocator_proc,
+		data = nil,
+	}
+}
+
+heap_allocator_proc :: proc(allocator_data: rawptr, mode: Allocator_Mode,
+                            size, alignment: int,
+                            old_memory: rawptr, old_size: int, loc := #caller_location) -> ([]byte, Allocator_Error) {
+	//
+	// NOTE(tetra, 2020-01-14): The heap doesn't respect alignment.
+	// Instead, we overallocate by `alignment + size_of(rawptr) - 1`, and insert
+	// padding. We also store the original pointer returned by heap_alloc right before
+	// the pointer we return to the user.
+	//
+
+	aligned_alloc :: proc(size, alignment: int, old_ptr: rawptr = nil, zero_memory := true) -> ([]byte, Allocator_Error) {
+		a := max(alignment, align_of(rawptr))
+		space := size + a - 1
+
+		allocated_mem: rawptr
+		if old_ptr != nil {
+			original_old_ptr := ([^]rawptr)(old_ptr)[-1]
+			allocated_mem = heap_resize(original_old_ptr, space+size_of(rawptr))
+		} else {
+			allocated_mem = heap_alloc(space+size_of(rawptr), zero_memory)
+		}
+		aligned_mem := rawptr(([^]u8)(allocated_mem)[size_of(rawptr):])
+
+		ptr := uintptr(aligned_mem)
+		aligned_ptr := (ptr - 1 + uintptr(a)) & -uintptr(a)
+		diff := int(aligned_ptr - ptr)
+		if (size + diff) > space || allocated_mem == nil {
+			return nil, .Out_Of_Memory
+		}
+
+		aligned_mem = rawptr(aligned_ptr)
+		([^]rawptr)(aligned_mem)[-1] = allocated_mem
+
+		return byte_slice(aligned_mem, size), nil
+	}
+
+	aligned_free :: proc(p: rawptr) {
+		if p != nil {
+			heap_free(([^]rawptr)(p)[-1])
+		}
+	}
+
+	aligned_resize :: proc(p: rawptr, old_size: int, new_size: int, new_alignment: int, zero_memory := true) -> (new_memory: []byte, err: Allocator_Error) {
+		if p == nil {
+			return nil, nil
+		}
+
+		new_memory = aligned_alloc(new_size, new_alignment, p, zero_memory) or_return
+
+		// NOTE: heap_resize does not zero the new memory, so we do it
+		if zero_memory && new_size > old_size {
+			new_region := raw_data(new_memory[old_size:])
+			intrinsics.mem_zero(new_region, new_size - old_size)
+		}
+		return
+	}
+
+	switch mode {
+	case .Alloc, .Alloc_Non_Zeroed:
+		return aligned_alloc(size, alignment, nil, mode == .Alloc)
+
+	case .Free:
+		aligned_free(old_memory)
+
+	case .Free_All:
+		return nil, .Mode_Not_Implemented
+
+	case .Resize, .Resize_Non_Zeroed:
+		if old_memory == nil {
+			return aligned_alloc(size, alignment, nil, mode == .Resize)
+		}
+		return aligned_resize(old_memory, old_size, size, alignment, mode == .Resize)
+
+	case .Query_Features:
+		set := (^Allocator_Mode_Set)(old_memory)
+		if set != nil {
+			set^ = {.Alloc, .Alloc_Non_Zeroed, .Free, .Resize, .Resize_Non_Zeroed, .Query_Features}
+		}
+		return nil, nil
+
+	case .Query_Info:
+		return nil, .Mode_Not_Implemented
+	}
+
+	return nil, nil
+}
+
+
+heap_alloc :: proc(size: int, zero_memory := true) -> rawptr {
+	return _heap_alloc(size, zero_memory)
+}
+
+heap_resize :: proc(ptr: rawptr, new_size: int) -> rawptr {
+	return _heap_resize(ptr, new_size)
+}
+
+heap_free :: proc(ptr: rawptr) {
+	_heap_free(ptr)
+}
@@ -0,0 +1,15 @@
+//+build js, wasi, freestanding, essence
+//+private
+package runtime
+
+_heap_alloc :: proc(size: int, zero_memory := true) -> rawptr {
+	unimplemented("base:runtime 'heap_alloc' procedure is not supported on this platform")
+}
+
+_heap_resize :: proc(ptr: rawptr, new_size: int) -> rawptr {
+	unimplemented("base:runtime 'heap_resize' procedure is not supported on this platform")
+}
+
+_heap_free :: proc(ptr: rawptr) {
+	unimplemented("base:runtime 'heap_free' procedure is not supported on this platform")
+}
@@ -0,0 +1,38 @@
+//+build linux, darwin, freebsd, openbsd
+//+private
+package runtime
+
+when ODIN_OS == .Darwin {
+	foreign import libc "system:System.framework"
+} else {
+	foreign import libc "system:c"
+}
+
+@(default_calling_convention="c")
+foreign libc {
+	@(link_name="malloc")   _unix_malloc   :: proc(size: int) -> rawptr ---
+	@(link_name="calloc")   _unix_calloc   :: proc(num, size: int) -> rawptr ---
+	@(link_name="free")     _unix_free     :: proc(ptr: rawptr) ---
+	@(link_name="realloc")  _unix_realloc  :: proc(ptr: rawptr, size: int) -> rawptr ---
+}
+
+_heap_alloc :: proc(size: int, zero_memory := true) -> rawptr {
+	if size <= 0 {
+		return nil
+	}
+	if zero_memory {
+		return _unix_calloc(1, size)
+	} else {
+		return _unix_malloc(size)
+	}
+}
+
+_heap_resize :: proc(ptr: rawptr, new_size: int) -> rawptr {
+	// NOTE: _unix_realloc doesn't guarantee new memory will be zeroed on
+	// POSIX platforms. Ensure your caller takes this into account.
+	return _unix_realloc(ptr, new_size)
+}
+
+_heap_free :: proc(ptr: rawptr) {
+	_unix_free(ptr)
+}
@@ -0,0 +1,39 @@
+package runtime
+
+foreign import kernel32 "system:Kernel32.lib"
+
+@(private="file")
+@(default_calling_convention="system")
+foreign kernel32 {
+	// NOTE(bill): The types are not using the standard names (e.g. DWORD and LPVOID) to just minimizing the dependency
+
+	// default_allocator
+	GetProcessHeap :: proc() -> rawptr ---
+	HeapAlloc      :: proc(hHeap: rawptr, dwFlags: u32, dwBytes: uint) -> rawptr ---
+	HeapReAlloc    :: proc(hHeap: rawptr, dwFlags: u32, lpMem: rawptr, dwBytes: uint) -> rawptr ---
+	HeapFree       :: proc(hHeap: rawptr, dwFlags: u32, lpMem: rawptr) -> b32 ---
+}
+
+_heap_alloc :: proc(size: int, zero_memory := true) -> rawptr {
+	HEAP_ZERO_MEMORY :: 0x00000008
+	return HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY if zero_memory else 0, uint(size))
+}
+_heap_resize :: proc(ptr: rawptr, new_size: int) -> rawptr {
+	if new_size == 0 {
+		_heap_free(ptr)
+		return nil
+	}
+	if ptr == nil {
+		return _heap_alloc(new_size)
+	}
+
+	HEAP_ZERO_MEMORY :: 0x00000008
+	return HeapReAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, ptr, uint(new_size))
+}
+_heap_free :: proc(ptr: rawptr) {
+	if ptr == nil {
+		return
+	}
+	HeapFree(GetProcessHeap(), 0, ptr)
+}
+
@@ -0,0 +1,7 @@
+package runtime
+
+_OS_Errno :: distinct int
+
+stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	return _stderr_write(data)
+}
@@ -0,0 +1,22 @@
+//+build freebsd, openbsd
+//+private
+package runtime
+
+foreign import libc "system:c"
+
+@(default_calling_convention="c")
+foreign libc {
+	@(link_name="write")
+	_unix_write :: proc(fd: i32, buf: rawptr, size: int) -> int ---
+
+	__error :: proc() -> ^i32 ---
+}
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	ret := _unix_write(2, raw_data(data), len(data))
+	if ret < len(data) {
+		err := __error()
+		return int(ret), _OS_Errno(err^ if err != nil else 0)
+	}
+	return int(ret), 0
+}
@@ -0,0 +1,26 @@
+//+build darwin
+//+private
+package runtime
+
+foreign import libc "system:System.framework"
+
+@(default_calling_convention="c")
+foreign libc {
+	@(link_name="__stderrp")
+	_stderr: rawptr
+
+	@(link_name="fwrite")
+	_fwrite :: proc(ptr: rawptr, size: uint, nmemb: uint, stream: rawptr) -> uint ---
+
+	@(link_name="__error")
+	_get_errno :: proc() -> ^i32 ---
+}
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	ret := _fwrite(raw_data(data), 1, len(data), _stderr)
+	if ret < len(data) {
+		err := _get_errno()
+		return int(ret), _OS_Errno(err^ if err != nil else 0)
+	}
+	return int(ret), 0
+}
@@ -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 +1,13 @@
+//+build js
+//+private
+package runtime
+
+foreign import "odin_env"
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	foreign odin_env {
+		write :: proc "contextless" (fd: u32, p: []byte) ---
+	}
+	write(1, data)
+	return len(data), 0
+}
@@ -0,0 +1,24 @@
+//+private
+package runtime
+
+import "base:intrinsics"
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	when ODIN_ARCH == .amd64 {
+		SYS_write :: uintptr(1)
+	} else when ODIN_ARCH == .arm64 {
+		SYS_write :: uintptr(64)
+	} else when ODIN_ARCH == .i386 {
+		SYS_write :: uintptr(4)
+	} else when ODIN_ARCH == .arm32 {
+		SYS_write :: uintptr(4)
+	}
+
+	stderr :: 2
+
+	ret := int(intrinsics.syscall(SYS_write, uintptr(stderr), uintptr(raw_data(data)), uintptr(len(data))))
+	if ret < 0 && ret > -4096 {
+		return 0, _OS_Errno(-ret)
+	}
+	return ret, 0
+}
@@ -0,0 +1,11 @@
+//+build wasi
+//+private
+package runtime
+
+import "core:sys/wasm/wasi"
+
+_stderr_write :: proc "contextless" (data: []byte) -> (int, _OS_Errno) {
+	data := (wasi.ciovec_t)(data)
+	n, err := wasi.fd_write(1, {data})
+	return int(n), _OS_Errno(err)
+}
@@ -0,0 +1,51 @@
+//+build windows
+//+private
+package runtime
+
+foreign import kernel32 "system:Kernel32.lib"
+
+@(private="file")
+@(default_calling_convention="system")
+foreign kernel32 {
+	// NOTE(bill): The types are not using the standard names (e.g. DWORD and LPVOID) to just minimizing the dependency
+
+	// stderr_write
+	GetStdHandle         :: proc(which: u32) -> rawptr ---
+	SetHandleInformation :: proc(hObject: rawptr, dwMask: u32, dwFlags: u32) -> b32 ---
+	WriteFile            :: proc(hFile: rawptr, lpBuffer: rawptr, nNumberOfBytesToWrite: u32, lpNumberOfBytesWritten: ^u32, lpOverlapped: rawptr) -> b32 ---
+	GetLastError         :: proc() -> u32 ---
+}
+
+_stderr_write :: proc "contextless" (data: []byte) -> (n: int, err: _OS_Errno) #no_bounds_check {
+	if len(data) == 0 {
+		return 0, 0
+	}
+
+	STD_ERROR_HANDLE :: ~u32(0) -12 + 1
+	HANDLE_FLAG_INHERIT :: 0x00000001
+	MAX_RW :: 1<<30
+
+	h := GetStdHandle(STD_ERROR_HANDLE)
+	when size_of(uintptr) == 8 {
+		SetHandleInformation(h, HANDLE_FLAG_INHERIT, 0)
+	}
+
+	single_write_length: u32
+	total_write: i64
+	length := i64(len(data))
+
+	for total_write < length {
+		remaining := length - total_write
+		to_write := u32(min(i32(remaining), MAX_RW))
+
+		e := WriteFile(h, &data[total_write], to_write, &single_write_length, nil)
+		if single_write_length <= 0 || !e {
+			err = _OS_Errno(GetLastError())
+			n = int(total_write)
+			return
+		}
+		total_write += i64(single_write_length)
+	}
+	n = int(total_write)
+	return
+}
@@ -0,0 +1,489 @@
+package runtime
+
+_INTEGER_DIGITS :: "0123456789abcdefghijklmnopqrstuvwxyz"
+
+@(private="file")
+_INTEGER_DIGITS_VAR := _INTEGER_DIGITS
+
+when !ODIN_NO_RTTI {
+	print_any_single :: 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 :: 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
+
+	buf: [4]u8
+	i := u32(r)
+	mask :: u8(0x3f)
+	if i <= 1<<7-1 {
+		buf[0] = u8(r)
+		return buf, 1
+	}
+	if i <= 1<<11-1 {
+		buf[0] = 0xc0 | u8(r>>6)
+		buf[1] = 0x80 | u8(r) & mask
+		return buf, 2
+	}
+
+	// Invalid or Surrogate range
+	if i > 0x0010ffff ||
+	   (0xd800 <= i && i <= 0xdfff) {
+		r = 0xfffd
+	}
+
+	if i <= 1<<16-1 {
+		buf[0] = 0xe0 | u8(r>>12)
+		buf[1] = 0x80 | u8(r>>6) & mask
+		buf[2] = 0x80 | u8(r)    & mask
+		return buf, 3
+	}
+
+	buf[0] = 0xf0 | u8(r>>18)
+	buf[1] = 0x80 | u8(r>>12) & mask
+	buf[2] = 0x80 | u8(r>>6)  & mask
+	buf[3] = 0x80 | u8(r)     & mask
+	return buf, 4
+}
+
+print_string :: proc "contextless" (str: string) -> (n: int) {
+	n, _ = stderr_write(transmute([]byte)str)
+	return
+}
+
+print_strings :: proc "contextless" (args: ..string) -> (n: int) {
+	for str in args {
+		m, err := stderr_write(transmute([]byte)str)
+		n += m
+		if err != 0 {
+			break
+		}
+	}
+	return
+}
+
+print_byte :: proc "contextless" (b: byte) -> (n: int) {
+	n, _ = stderr_write([]byte{b})
+	return
+}
+
+print_encoded_rune :: proc "contextless" (r: rune) {
+	print_byte('\'')
+
+	switch r {
+	case '\a': print_string("\\a")
+	case '\b': print_string("\\b")
+	case '\e': print_string("\\e")
+	case '\f': print_string("\\f")
+	case '\n': print_string("\\n")
+	case '\r': print_string("\\r")
+	case '\t': print_string("\\t")
+	case '\v': print_string("\\v")
+	case:
+		if r <= 0 {
+			print_string("\\x00")
+		} else if r < 32 {
+			n0, n1 := u8(r) >> 4, u8(r) & 0xf
+			print_string("\\x")
+			print_byte(_INTEGER_DIGITS_VAR[n0])
+			print_byte(_INTEGER_DIGITS_VAR[n1])
+		} else {
+			print_rune(r)
+		}
+	}
+	print_byte('\'')
+}
+
+print_rune :: proc "contextless" (r: rune) -> int #no_bounds_check {
+	RUNE_SELF :: 0x80
+
+	if r < RUNE_SELF {
+		return print_byte(byte(r))
+	}
+
+	b, n := encode_rune(r)
+	m, _ := stderr_write(b[:n])
+	return m
+}
+
+
+print_u64 :: 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] = _INTEGER_DIGITS_VAR[u % b]
+		u /= b
+	}
+	i -= 1; a[i] = _INTEGER_DIGITS_VAR[u % b]
+
+	stderr_write(a[i:])
+}
+
+
+print_i64 :: proc "contextless" (x: i64) #no_bounds_check {
+	b :: i64(10)
+
+	u := x
+	neg := u < 0
+	u = abs(u)
+
+	a: [129]byte
+	i := len(a)
+	for u >= b {
+		i -= 1; a[i] = _INTEGER_DIGITS_VAR[u % b]
+		u /= b
+	}
+	i -= 1; a[i] = _INTEGER_DIGITS_VAR[u % b]
+	if neg {
+		i -= 1; 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" (loc: Source_Code_Location) {
+	print_string(loc.file_path)
+	when ODIN_ERROR_POS_STYLE == .Default {
+		print_byte('(')
+		print_u64(u64(loc.line))
+		print_byte(':')
+		print_u64(u64(loc.column))
+		print_byte(')')
+	} else when ODIN_ERROR_POS_STYLE == .Unix {
+		print_byte(':')
+		print_u64(u64(loc.line))
+		print_byte(':')
+		print_u64(u64(loc.column))
+		print_byte(':')
+	} else {
+		#panic("unhandled ODIN_ERROR_POS_STYLE")
+	}
+}
+print_typeid :: proc "contextless" (id: typeid) {
+	when ODIN_NO_RTTI {
+		if id == nil {
+			print_string("nil")
+		} else {
+			print_string("<unknown type>")
+		}
+	} else {
+		if id == nil {
+			print_string("nil")
+		} else {
+			ti := type_info_of(id)
+			print_type(ti)
+		}
+	}
+}
+print_type :: proc "contextless" (ti: ^Type_Info) {
+	if ti == nil {
+		print_string("nil")
+		return
+	}
+
+	switch info in ti.variant {
+	case Type_Info_Named:
+		print_string(info.name)
+	case Type_Info_Integer:
+		switch ti.id {
+		case int:     print_string("int")
+		case uint:    print_string("uint")
+		case uintptr: print_string("uintptr")
+		case:
+			print_byte('i' if info.signed else 'u')
+			print_u64(u64(8*ti.size))
+		}
+	case Type_Info_Rune:
+		print_string("rune")
+	case Type_Info_Float:
+		print_byte('f')
+		print_u64(u64(8*ti.size))
+	case Type_Info_Complex:
+		print_string("complex")
+		print_u64(u64(8*ti.size))
+	case Type_Info_Quaternion:
+		print_string("quaternion")
+		print_u64(u64(8*ti.size))
+	case Type_Info_String:
+		print_string("string")
+	case Type_Info_Boolean:
+		switch ti.id {
+		case bool: print_string("bool")
+		case:
+			print_byte('b')
+			print_u64(u64(8*ti.size))
+		}
+	case Type_Info_Any:
+		print_string("any")
+	case Type_Info_Type_Id:
+		print_string("typeid")
+
+	case Type_Info_Pointer:
+		if info.elem == nil {
+			print_string("rawptr")
+		} else {
+			print_string("^")
+			print_type(info.elem)
+		}
+	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_Parameters)
+			print_byte('(')
+			for t, i in t.types {
+				if i > 0 { print_string(", ") }
+				print_type(t)
+			}
+			print_string(")")
+		}
+		if info.results != nil {
+			print_string(" -> ")
+			print_type(info.results)
+		}
+	case Type_Info_Parameters:
+		count := len(info.names)
+		if count != 1 { print_byte('(') }
+		for name, i in info.names {
+			if i > 0 { print_string(", ") }
+
+			t := info.types[i]
+
+			if len(name) > 0 {
+				print_string(name)
+				print_string(": ")
+			}
+			print_type(t)
+		}
+		if count != 1 { print_string(")") }
+
+	case Type_Info_Array:
+		print_byte('[')
+		print_u64(u64(info.count))
+		print_byte(']')
+		print_type(info.elem)
+
+	case Type_Info_Enumerated_Array:
+		if info.is_sparse {
+			print_string("#sparse")
+		}
+		print_byte('[')
+		print_type(info.index)
+		print_byte(']')
+		print_type(info.elem)
+
+
+	case Type_Info_Dynamic_Array:
+		print_string("[dynamic]")
+		print_type(info.elem)
+	case Type_Info_Slice:
+		print_string("[]")
+		print_type(info.elem)
+
+	case Type_Info_Map:
+		print_string("map[")
+		print_type(info.key)
+		print_byte(']')
+		print_type(info.value)
+
+	case Type_Info_Struct:
+		switch info.soa_kind {
+		case .None: // Ignore
+		case .Fixed:
+			print_string("#soa[")
+			print_u64(u64(info.soa_len))
+			print_byte(']')
+			print_type(info.soa_base_type)
+			return
+		case .Slice:
+			print_string("#soa[]")
+			print_type(info.soa_base_type)
+			return
+		case .Dynamic:
+			print_string("#soa[dynamic]")
+			print_type(info.soa_base_type)
+			return
+		}
+
+		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(")
+			print_u64(u64(ti.align))
+			print_string(") ")
+		}
+		print_byte('{')
+		for name, i in info.names {
+			if i > 0 { print_string(", ") }
+			print_string(name)
+			print_string(": ")
+			print_type(info.types[i])
+		}
+		print_byte('}')
+
+	case Type_Info_Union:
+		print_string("union ")
+		if info.custom_align {
+			print_string("#align(")
+			print_u64(u64(ti.align))
+			print_string(") ")
+		}
+		if info.no_nil {
+			print_string("#no_nil ")
+		}
+		print_byte('{')
+		for variant, i in info.variants {
+			if i > 0 { print_string(", ") }
+			print_type(variant)
+		}
+		print_string("}")
+
+	case Type_Info_Enum:
+		print_string("enum ")
+		print_type(info.base)
+		print_string(" {")
+		for name, i in info.names {
+			if i > 0 { print_string(", ") }
+			print_string(name)
+		}
+		print_string("}")
+
+	case Type_Info_Bit_Set:
+		print_string("bit_set[")
+
+		#partial switch elem in type_info_base(info.elem).variant {
+		case Type_Info_Enum:
+			print_type(info.elem)
+		case Type_Info_Rune:
+			print_encoded_rune(rune(info.lower))
+			print_string("..")
+			print_encoded_rune(rune(info.upper))
+		case:
+			print_i64(info.lower)
+			print_string("..")
+			print_i64(info.upper)
+		}
+		if info.underlying != nil {
+			print_string("; ")
+			print_type(info.underlying)
+		}
+		print_byte(']')
+
+
+	case Type_Info_Simd_Vector:
+		print_string("#simd[")
+		print_u64(u64(info.count))
+		print_byte(']')
+		print_type(info.elem)
+
+	case Type_Info_Relative_Pointer:
+		print_string("#relative(")
+		print_type(info.base_integer)
+		print_string(") ")
+		print_type(info.pointer)
+
+	case Type_Info_Relative_Multi_Pointer:
+		print_string("#relative(")
+		print_type(info.base_integer)
+		print_string(") ")
+		print_type(info.pointer)
+		
+	case Type_Info_Matrix:
+		print_string("matrix[")
+		print_u64(u64(info.row_count))
+		print_string(", ")
+		print_u64(u64(info.column_count))
+		print_string("]")
+		print_type(info.elem)
+	}
+}
@@ -0,0 +1,95 @@
+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_ARCH == .wasm32 || ODIN_ARCH == .wasm64p32) {
+	@(link_name="memset", linkage="strong", require)
+	memset :: proc "c" (ptr: rawptr, val: i32, len: int) -> rawptr {
+		if ptr != nil && len != 0 {
+			b := byte(val)
+			p := ([^]byte)(ptr)
+			for i := 0; i < len; i += 1 {
+				p[i] = b
+			}
+		}
+		return ptr
+	}
+
+	@(link_name="bzero", linkage="strong", require)
+	bzero :: proc "c" (ptr: rawptr, len: int) -> rawptr {
+		if ptr != nil && len != 0 {
+			p := ([^]byte)(ptr)
+			for i := 0; i < len; i += 1 {
+				p[i] = 0
+			}
+		}
+		return ptr
+	}
+
+	@(link_name="memmove", linkage="strong", require)
+	memmove :: proc "c" (dst, src: rawptr, len: int) -> rawptr {
+		d, s := ([^]byte)(dst), ([^]byte)(src)
+		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 := 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, len: int) -> rawptr {
+		d, s := ([^]byte)(dst), ([^]byte)(src)
+		if d != s {
+			for i := 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
+	}
+}
@@ -0,0 +1,21 @@
+//+private
+package runtime
+
+foreign import "system:Foundation.framework"
+
+import "base:intrinsics"
+
+objc_id :: ^intrinsics.objc_object
+objc_Class :: ^intrinsics.objc_class
+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_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 ---
+	objc_msgSend_fp2ret :: proc "c" (self: objc_id, op: objc_SEL, #c_vararg args: ..any) -> complex128 ---
+	objc_msgSend_stret  :: proc "c" (self: objc_id, op: objc_SEL, #c_vararg args: ..any) ---
+}
@@ -0,0 +1,15 @@
+//+build js
+package runtime
+
+init_default_context_for_js: Context
+@(init, private="file")
+init_default_context :: proc() {
+	init_default_context_for_js = context
+}
+
+@(export)
+@(link_name="default_context_ptr")
+default_context_ptr :: proc "contextless" () -> ^Context {
+	return &init_default_context_for_js
+}
+
@@ -0,0 +1,40 @@
+//+build wasm32, wasm64p32
+package runtime
+
+@(private="file")
+ti_int :: struct #raw_union {
+	using s: struct { lo, hi: u64 },
+	all: i128,
+}
+
+@(link_name="__ashlti3", linkage="strong")
+__ashlti3 :: proc "contextless" (a: i128, b_: u32) -> i128 {
+	bits_in_dword :: size_of(u32)*8
+	b := u32(b_)
+	
+	input, result: ti_int
+	input.all = a
+	if b & bits_in_dword != 0 {
+		result.lo = 0
+		result.hi = input.lo << (b-bits_in_dword)
+	} else {
+		if b == 0 {
+			return a
+		}
+		result.lo = input.lo<<b
+		result.hi = (input.hi<<b) | (input.lo>>(bits_in_dword-b))
+	}
+	return result.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 = i128(x.lo * y.lo) // TODO this is incorrect
+	r.hi += x.hi*y.lo + x.lo*y.hi
+	return r.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
@@ -0,0 +1,26 @@
+//+private
+//+no-instrumentation
+package runtime
+
+foreign import kernel32 "system:Kernel32.lib"
+
+@(private)
+foreign kernel32 {
+	RaiseException :: proc "system" (dwExceptionCode, dwExceptionFlags, nNumberOfArguments: u32, lpArguments: ^uint) -> ! ---
+}
+
+windows_trap_array_bounds :: proc "contextless" () -> ! {
+	EXCEPTION_ARRAY_BOUNDS_EXCEEDED :: 0xC000008C
+
+
+	RaiseException(EXCEPTION_ARRAY_BOUNDS_EXCEEDED, 0, 0, nil)
+}
+
+windows_trap_type_assertion :: proc "contextless" () -> ! {
+	windows_trap_array_bounds()
+}
+
+when ODIN_NO_CRT {
+	@(require)
+	foreign import crt_lib "procs_windows_amd64.asm"
+}
@@ -0,0 +1,29 @@
+//+private
+//+no-instrumentation
+package runtime
+
+@require foreign import "system:int64.lib"
+
+foreign import kernel32 "system:Kernel32.lib"
+
+windows_trap_array_bounds :: proc "contextless" () -> ! {
+	DWORD :: u32
+	ULONG_PTR :: uint
+
+	EXCEPTION_ARRAY_BOUNDS_EXCEEDED :: 0xC000008C
+
+	foreign kernel32 {
+		RaiseException :: proc "system" (dwExceptionCode, dwExceptionFlags, nNumberOfArguments: DWORD, lpArguments: ^ULONG_PTR) -> ! ---
+	}
+
+	RaiseException(EXCEPTION_ARRAY_BOUNDS_EXCEEDED, 0, 0, nil)
+}
+
+windows_trap_type_assertion :: proc "contextless" () -> ! {
+	windows_trap_array_bounds()
+}
+
+@(private, export, link_name="_fltused") _fltused: i32 = 0x9875
+
+@(private, export, link_name="_tls_index") _tls_index: u32
+@(private, export, link_name="_tls_array") _tls_array: u32
@@ -0,0 +1,156 @@
+package runtime
+
+import "base:intrinsics"
+
+udivmod128 :: proc "c" (a, b: u128, rem: ^u128) -> u128 {
+	_ctz :: intrinsics.count_trailing_zeros
+	_clz :: intrinsics.count_leading_zeros
+
+	n := transmute([2]u64)a
+	d := transmute([2]u64)b
+	q, r: [2]u64
+	sr: u32 = 0
+
+	low  :: 1 when ODIN_ENDIAN == .Big else 0
+	high :: 1 - low
+	U64_BITS :: 8*size_of(u64)
+	U128_BITS :: 8*size_of(u128)
+
+	// Special Cases
+
+	if n[high] == 0 {
+		if d[high] == 0 {
+			if rem != nil {
+				res := n[low] % d[low]
+				rem^ = u128(res)
+			}
+			return u128(n[low] / d[low])
+		}
+
+		if rem != nil {
+			rem^ = u128(n[low])
+		}
+		return 0
+	}
+
+	if d[low] == 0 {
+		if d[high] == 0 {
+			if rem != nil {
+				rem^ = u128(n[high] % d[low])
+			}
+			return u128(n[high] / d[low])
+		}
+		if n[low] == 0 {
+			if rem != nil {
+				r[high] = n[high] % d[high]
+				r[low] = 0
+				rem^ = transmute(u128)r
+			}
+			return u128(n[high] / d[high])
+		}
+
+		if d[high] & (d[high]-1) == 0 {
+			if rem != nil {
+				r[low] = n[low]
+				r[high] = n[high] & (d[high] - 1)
+				rem^ = transmute(u128)r
+			}
+			return u128(n[high] >> _ctz(d[high]))
+		}
+
+		sr = transmute(u32)(i32(_clz(d[high])) - i32(_clz(n[high])))
+		if sr > U64_BITS - 2 {
+			if rem != nil {
+				rem^ = a
+			}
+			return 0
+		}
+
+		sr += 1
+
+		q[low]  = 0
+		q[high] = n[low] << u64(U64_BITS - sr)
+		r[high] = n[high] >> sr
+		r[low]  = (n[high] << (U64_BITS - sr)) | (n[low] >> sr)
+	} else {
+		if d[high] == 0 {
+			if d[low] & (d[low] - 1) == 0 {
+				if rem != nil {
+					rem^ = u128(n[low] & (d[low] - 1))
+				}
+				if d[low] == 1 {
+					return a
+				}
+				sr = u32(_ctz(d[low]))
+				q[high] = n[high] >> sr
+				q[low] = (n[high] << (U64_BITS-sr)) | (n[low] >> sr)
+				return transmute(u128)q
+			}
+
+			sr = 1 + U64_BITS + u32(_clz(d[low])) - u32(_clz(n[high]))
+
+			switch {
+			case sr == U64_BITS:
+				q[low]  = 0
+				q[high] = n[low]
+				r[high] = 0
+				r[low]  = n[high]
+			case sr < U64_BITS:
+				q[low]  = 0
+				q[high] = n[low] << (U64_BITS - sr)
+				r[high] = n[high] >> sr
+				r[low]  = (n[high] << (U64_BITS - sr)) | (n[low] >> sr)
+			case:
+				q[low]  = n[low] << (U128_BITS - sr)
+				q[high] = (n[high] << (U128_BITS - sr)) | (n[low] >> (sr - U64_BITS))
+				r[high] = 0
+				r[low]  = n[high] >> (sr - U64_BITS)
+			}
+		} else {
+			sr = transmute(u32)(i32(_clz(d[high])) - i32(_clz(n[high])))
+
+			if sr > U64_BITS - 1 {
+				if rem != nil {
+					rem^ = a
+				}
+				return 0
+			}
+
+			sr += 1
+
+			q[low] = 0
+			if sr == U64_BITS {
+				q[high] = n[low]
+				r[high] = 0
+				r[low]  = n[high]
+			} else {
+				r[high] = n[high] >> sr
+				r[low]  = (n[high] << (U64_BITS - sr)) | (n[low] >> sr)
+				q[high] = n[low] << (U64_BITS - sr)
+			}
+		}
+	}
+
+	carry: u32 = 0
+	r_all: u128
+
+	for ; sr > 0; sr -= 1 {
+		r[high] = (r[high] << 1) | (r[low]  >> (U64_BITS - 1))
+		r[low]  = (r[low]  << 1) | (q[high] >> (U64_BITS - 1))
+		q[high] = (q[high] << 1) | (q[low]  >> (U64_BITS - 1))
+		q[low]  = (q[low]  << 1) | u64(carry)
+
+		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 = transmute([2]u64)r_all
+	}
+
+	q_all := ((transmute(u128)q) << 1) | u128(carry)
+	if rem != nil {
+		rem^ = r_all
+	}
+
+	return q_all
+}
@@ -2,13 +2,12 @@

 ## Setup

-Odin only supports x86-64 at the moment (64-bit), relies on LLVM for code generation and an external linker.
+Odin currently supports x86-64 and ARM64 at the moment (64-bit), relies on LLVM for code generation and an external linker.

 In addition, the following platform-specific steps are necessary:

 - Windows
    * Have Visual Studio installed (MSVC 2010 or later, for the linker)
-    * Have a copy of `opt.exe` and `llc.exe` in `Odin/bin`. Pre-built Windows binaries can be found [here](https://github.com/odin-lang/Odin/releases/tag/llvm-windows) and *must* be explicitly copied
    * Open a valid command prompt:
        * **Basic:** run the `x64 Native Tools Command Prompt for VS2017` shortcut bundled with VS 2017, or
        * **Advanced:** run `vcvarsall.bat x64` from a blank `cmd` session
@@ -19,12 +18,12 @@ In addition, the following platform-specific steps are necessary:
    * Make sure the LLVM binaries and the linker are added to your `$PATH` environmental variable

 - GNU/Linux
-    * Have LLVM installed (opt/llc)
    * Have Clang installed (version X.X or later, for linking)
    * Make sure the LLVM binaries and the linker are added to your `$PATH` environmental variable

 Then build the compiler by calling `build.bat` (Windows) or `make` (Linux/MacOS). This will automatically run the demo program if successful.

-**Notes for Linux:**: The compiler currently relies on the `core` and `shared` library collection being relative to the compiler executable. Installing the compiler in the usual sense (to `/usr/local/bin` or similar) is therefore not as straight forward as you need to make sure the mentioned libraries are available. As a result, it is recommended to simply explicitly invoke the compiler with `/path/to/odin` in your preferred build system, or add `/path/to/odin` to `$PATH`.
+**Notes for \*Nix Systems:**: The compiler currently relies on the `core` and `shared` library collection being relative to the compiler executable, by default. Installing the compiler in the usual sense (to `/usr/local/bin` or similar) is therefore not as straight forward as you need to make sure the mentioned libraries are available. As a result, it is recommended to either simply explicitly invoke the compiler with `/path/to/odin` in your preferred build system, or `set ODIN_ROOT=/path/to/odin_root`.
+

 Please read the [Getting Started Guide](https://github.com/odin-lang/Odin/wiki#getting-started-with-odin) for more information.
@@ -0,0 +1,29 @@
+NASM is now licensed under the 2-clause BSD license, also known as the
+simplified BSD license.
+
+    Copyright 1996-2010 the NASM Authors - All rights reserved.
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following
+    conditions are met:
+
+    * Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+      copyright notice, this list of conditions and the following
+      disclaimer in the documentation and/or other materials provided
+      with the distribution.
+      
+      THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
+      CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
+      INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+      MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+      DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
+      CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+      SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+      NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+      LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+      HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+      CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
+      OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
+      EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
@@ -1,37 +0,0 @@
-#!/usr/bin/env bash
-
-release_mode=$1
-
-warnings_to_disable="-std=c++11 -Wno-switch"
-
-libraries="-pthread -ldl -lm -lstdc++ -lz -lcurses -lxml2"
-other_args="-DLLVM_BACKEND_SUPPORT"
-compiler="clang"
-
-if [ -z "$release_mode" ]; then release_mode="0"; fi
-
-if [ "$release_mode" -eq "0" ]; then
-	other_args="${other_args} -g"
-fi
-if [ "$release_mode" -eq "1" ]; then
-	other_args="${other_args} -O3 -march=native"
-fi
-
-if [[ "$(uname)" == "Darwin" ]]; then
-
-	# Set compiler to clang on MacOS
-	# MacOS provides a symlink to clang called gcc, but it's nice to be explicit here.
-	compiler="clang"
-
-	llvm_config_flags="--cxxflags --ldflags"
-	# llvm_config_flags="${llvm_config_flags} --link-static"
-	llvm_config="llvm-config ${llvm_config_flags}"
-
-	other_args="${other_args} -liconv"
-	other_args="${other_args} `${llvm_config}` -lLLVM-C"
-elif [[ "$(uname)" == "FreeBSD" ]]; then
-	compiler="clang"
-fi
-
-${compiler} src/main.cpp ${warnings_to_disable} ${libraries} ${other_args} -o odin
-	# && ./odin run examples/demo/demo.odin -llvm-api
@@ -1,5 +1,31 @@
@echo off

+setlocal EnableDelayedExpansion
+
+where /Q cl.exe || (
+	set __VSCMD_ARG_NO_LOGO=1
+	for /f "tokens=*" %%i in ('"C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe" -latest -requires Microsoft.VisualStudio.Workload.NativeDesktop -property installationPath') do set VS=%%i
+	if "!VS!" equ "" (
+		echo ERROR: Visual Studio installation not found
+		exit /b 1
+	)
+	call "!VS!\VC\Auxiliary\Build\vcvarsall.bat" amd64 || exit /b 1
+)
+
+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 "usebackq tokens=1,2 delims=,=- " %%i in (`wmic os get LocalDateTime /value`) do @if %%i==LocalDateTime (
+	set CURR_DATE_TIME=%%j
+)
+
+set curr_year=%CURR_DATE_TIME:~0,4%
+set curr_month=%CURR_DATE_TIME:~4,2%
+
 :: Make sure this is a decent name and not generic
 set exe_name=odin.exe

@@ -14,16 +40,24 @@ if "%1" == "1" (

 :: Normal = 0, CI Nightly = 1
 if "%2" == "1" (
-    set nightly=1
+	set nightly=1
 ) else (
-    set nightly=0
+	set nightly=0
 )

-set compiler_flags= -nologo -Oi -TP -fp:precise -Gm- -MP -FC -EHsc- -GR- -GF
-set compiler_defines= -DLLVM_BACKEND_SUPPORT -DUSE_NEW_LLVM_ABI_SYSTEM
+set odin_version_raw="dev-%curr_year%-%curr_month%"

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

 if %release_mode% EQU 0 ( rem Debug
@@ -35,20 +69,31 @@ if %release_mode% EQU 0 ( rem Debug

 set compiler_warnings= ^
 	-W4 -WX ^
-	-wd4100 -wd4101 -wd4127 -wd4189 ^
-	-wd4201 -wd4204 ^
-	-wd4456 -wd4457 -wd4480 ^
-	-wd4512
+	-wd4100 -wd4101 -wd4127 -wd4146 ^
+	-wd4505 ^
+	-wd4456 -wd4457

-set compiler_includes=
+set compiler_includes= ^
+	/Isrc\
 set libs= ^
 	kernel32.lib ^
+	Synchronization.lib ^
 	bin\llvm\windows\LLVM-C.lib

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

 if %release_mode% EQU 0 ( rem Debug
-	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
 )
@@ -59,11 +104,15 @@ set linker_settings=%libs% %linker_flags%
 del *.pdb > NUL 2> NUL
 del *.ilk > NUL 2> NUL

-cl %compiler_settings% "src\main.cpp" /link %linker_settings% -OUT:%exe_name%
-
+cl %compiler_settings% "src\main.cpp" "src\libtommath.cpp" /link %linker_settings% -OUT:%exe_name%
 if %errorlevel% neq 0 goto end_of_build
-if %release_mode% EQU 0 odin run examples/demo/demo.odin
+
+call build_vendor.bat
+if %errorlevel% neq 0 goto end_of_build
+
+rem If the demo doesn't run for you and your CPU is more than a decade old, try -microarch:native
+if %release_mode% EQU 0 odin run examples/demo -- Hellope World

 del *.obj > NUL 2> NUL

-:end_of_build
+:end_of_build
@@ -1,30 +0,0 @@
-#!/usr/bin/env bash
-
-release_mode=$1
-
-warnings_to_disable="-std=c++11 -Wno-switch -Wno-pointer-sign -Wno-tautological-constant-out-of-range-compare -Wno-tautological-compare -Wno-macro-redefined"
-libraries="-pthread -ldl -lm -lstdc++"
-other_args=""
-compiler="clang"
-
-if [ -z "$release_mode" ]; then release_mode="0"; fi
-
-if [ "$release_mode" -eq "0" ]; then
-	other_args="${other_args} -g"
-fi
-if [ "$release_mode" -eq "1" ]; then
-	other_args="${other_args} -O3 -march=native"
-fi
-
-if [[ "$(uname)" == "Darwin" ]]; then
-
-	# Set compiler to clang on MacOS
-	# MacOS provides a symlink to clang called gcc, but it's nice to be explicit here.
-	compiler="clang"
-
-	other_args="${other_args} -liconv"
-elif [[ "$(uname)" == "FreeBSD" ]]; then
-	compiler="clang"
-fi
-
-${compiler} src/main.cpp ${warnings_to_disable} ${libraries} ${other_args} -o odin && ./odin run examples/demo/demo.odin
@@ -0,0 +1,141 @@
+#!/usr/bin/env sh
+set -eu
+
+: ${CPPFLAGS=}
+: ${CXX=clang++}
+: ${CXXFLAGS=}
+: ${LDFLAGS=}
+: ${LLVM_CONFIG=}
+
+CPPFLAGS="$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" "$1"
+	exit 1
+}
+
+if [ -z "$LLVM_CONFIG" ]; then
+	# darwin, linux, openbsd
+	if   [ -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-config17)" ]; then  LLVM_CONFIG="llvm-config-17"
+	elif [ -n "$(command -v llvm-config14)" ]; then  LLVM_CONFIG="llvm-config-14"
+	elif [ -n "$(command -v llvm-config13)" ]; then  LLVM_CONFIG="llvm-config-13"
+	elif [ -n "$(command -v llvm-config12)" ]; then  LLVM_CONFIG="llvm-config-12"
+	elif [ -n "$(command -v llvm-config11)" ]; then  LLVM_CONFIG="llvm-config-11"
+	# fallback
+	elif [ -n "$(command -v llvm-config)" ]; then LLVM_CONFIG="llvm-config"
+	else
+		error "No llvm-config command found. Set LLVM_CONFIG to proceed."
+	fi
+fi
+
+LLVM_VERSION="$($LLVM_CONFIG --version)"
+LLVM_VERSION_MAJOR="$(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 ]); then
+	error "Invalid LLVM version $LLVM_VERSION: must be 11, 12, 13, 14 or 17"
+fi
+
+case "$OS_NAME" in
+Darwin)
+	if [ "$OS_ARCH" == "arm64" ]; then
+		if [ $LLVM_VERSION_MAJOR -lt 13 ] || [ $LLVM_VERSION_MAJOR -gt 17 ]; then
+			error "Darwin Arm64 requires LLVM 13, 14 or 17"
+		fi
+	fi
+
+	CXXFLAGS="$CXXFLAGS $($LLVM_CONFIG --cxxflags --ldflags)"
+	LDFLAGS="$LDFLAGS -liconv -ldl -framework System"
+	LDFLAGS="$LDFLAGS -lLLVM-C"
+	;;
+FreeBSD)
+	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)"
+	;;
+*)
+	error "Platform \"$OS_NAME\" unsupported"
+	;;
+esac
+
+build_odin() {
+	case $1 in
+	debug)
+		EXTRAFLAGS="-g"
+		;;
+	release)
+		EXTRAFLAGS="-O3"
+		;;
+	release-native)
+		if [ "$OS_ARCH" == "arm64" ]; then
+			# Use preferred flag for Arm (ie arm64 / aarch64 / etc)
+			EXTRAFLAGS="-O3 -mcpu=native"
+		else
+			# Use preferred flag for x86 / amd64
+			EXTRAFLAGS="-O3 -march=native"
+		fi
+		;;
+	nightly)
+		EXTRAFLAGS="-DNIGHTLY -O3"
+		;;
+	*)
+		error "Build mode \"$1\" unsupported!"
+		;;
+	esac
+
+	set -x
+	$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 -- Hellope World
+}
+
+if [ $# -eq 0 ]; then
+	build_odin debug
+	run_demo
+elif [ $# -eq 1 ]; then
+	case $1 in
+	report)
+		[ ! -f "./odin" ] && build_odin debug
+		./odin report
+		;;
+	*)
+		build_odin $1
+		;;
+	esac
+	run_demo
+else
+	error "Too many arguments!"
+fi
@@ -0,0 +1,24 @@
+@echo off
+
+setlocal EnableDelayedExpansion
+
+rem build the .lib files already exist
+
+if not exist "vendor\stb\lib\*.lib" (
+	pushd vendor\stb\src
+		call build.bat
+	popd
+)
+
+if not exist "vendor\miniaudio\lib\*.lib" (
+	pushd vendor\miniaudio\src
+		call build.bat
+	popd
+)
+
+
+if not exist "vendor\cgltf\lib\*.lib" (
+	pushd vendor\cgltf\src
+		call build.bat
+	popd
+)
@@ -18,7 +18,7 @@ def main():
            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['contentLength'])
+            size = int(data['size'])
            ts = int(data['fileInfo']['src_last_modified_millis'])
            date = datetime.datetime.fromtimestamp(ts/1000).strftime('%Y-%m-%d')
            
@@ -0,0 +1,83 @@
+package bufio
+
+import "core:io"
+
+// Lookahead_Reader provides io lookahead.
+// This is useful for tokenizers/parsers.
+// 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.
+Lookahead_Reader :: struct {
+	r:   io.Reader,
+	buf: []byte,
+	n:   int,
+}
+
+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: ^Lookahead_Reader) -> []byte {
+	return lr.buf[:lr.n]
+}
+
+
+// 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: ^Lookahead_Reader, n: int) -> ([]byte, io.Error) {
+	switch {
+	case n < 0:
+		return nil, .Negative_Read
+	case n > len(lr.buf):
+		return nil, .Buffer_Full
+	}
+
+	n := n
+	err: io.Error
+	read_count: int
+
+	if lr.n < n {
+		read_count, err = io.read_at_least(lr.r, lr.buf[lr.n:], n-lr.n)
+		if err == .Unexpected_EOF {
+			err = .EOF
+		}
+	}
+
+	lr.n += read_count
+
+	if n > lr.n {
+		n = lr.n
+	}
+	return lr.buf[:n], err
+}
+
+// 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: ^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: ^Lookahead_Reader, n: int) -> io.Error {
+	switch {
+	case n == 0:
+		return nil
+	case n < 0:
+		return .Negative_Read
+	case lr.n < n:
+		return .Short_Buffer
+	}
+	copy(lr.buf, lr.buf[n:lr.n])
+	lr.n -= n
+	return nil
+}
+
+lookahead_reader_consume_all :: proc(lr: ^Lookahead_Reader) -> io.Error {
+	return lookahead_reader_consume(lr, lr.n)
+}
@@ -10,59 +10,33 @@ Read_Writer :: struct {


 read_writer_init :: proc(rw: ^Read_Writer, r: ^Reader, w: ^Writer) {
-	rw.r, rw.w = r, w;
+	rw.r, rw.w = r, w
 }

 read_writer_to_stream :: proc(rw: ^Read_Writer) -> (s: io.Stream) {
-	s.stream_data = rw;
-	s.stream_vtable = _read_writer_vtable;
-	return;
+	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
+}
@@ -17,84 +17,90 @@ Reader :: struct {

 	last_byte:      int, // last byte read, invalid is -1
 	last_rune_size: int, // size of last rune read, invalid is -1
+
+	max_consecutive_empty_reads: int,
 }


-DEFAULT_BUF_SIZE :: 4096;
+DEFAULT_BUF_SIZE :: 4096

@(private)
-MIN_READ_BUFFER_SIZE :: 16;
+MIN_READ_BUFFER_SIZE :: 16
@(private)
-MAX_CONSECUTIVE_EMPTY_READS :: 128;
+DEFAULT_MAX_CONSECUTIVE_EMPTY_READS :: 128

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

 reader_init_with_buf :: proc(b: ^Reader, rd: io.Reader, buf: []byte) {
-	reader_reset(b, rd);
-	b.buf_allocator = {};
-	b.buf = buf;
+	reader_reset(b, rd)
+	b.buf_allocator = {}
+	b.buf = buf
 }

 // reader_destroy destroys the underlying buffer with its associated allocator IFF that allocator has been set
 reader_destroy :: proc(b: ^Reader) {
-	delete(b.buf, b.buf_allocator);
-	b^ = {};
+	delete(b.buf, b.buf_allocator)
+	b^ = {}
 }

 reader_size :: proc(b: ^Reader) -> int {
-	return len(b.buf);
+	return len(b.buf)
 }

 reader_reset :: proc(b: ^Reader, r: io.Reader) {
-	b.rd = r;
-	b.r, b.w = 0, 0;
-	b.err = nil;
-	b.last_byte      = -1;
-	b.last_rune_size = -1;
+	b.rd = r
+	b.r, b.w = 0, 0
+	b.err = nil
+	b.last_byte      = -1
+	b.last_rune_size = -1
 }

@(private)
 _reader_read_new_chunk :: proc(b: ^Reader) -> io.Error {
 	if b.r > 0 {
-		copy(b.buf, b.buf[b.r:b.w]);
-		b.w -= b.r;
-		b.r = 0;
+		copy(b.buf, b.buf[b.r:b.w])
+		b.w -= b.r
+		b.r = 0
 	}

 	if b.w >= len(b.buf) {
-		return .Buffer_Full;
+		return .Buffer_Full
+	}
+
+	if b.max_consecutive_empty_reads <= 0 {
+		b.max_consecutive_empty_reads = DEFAULT_MAX_CONSECUTIVE_EMPTY_READS
 	}

 	// read new data, and try a limited number of times
-	for i := MAX_CONSECUTIVE_EMPTY_READS; i > 0; i -= 1 {
-		n, err := io.read(b.rd, b.buf[b.w:]);
+	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 .Negative_Read
 		}
-		b.w += n;
+		b.w += n
 		if err != nil {
-			b.err = err;
-			return nil;
+			b.err = err
+			return nil
 		}
 		if n > 0 {
-			return nil;
+			return nil
 		}
 	}
-	b.err = .No_Progress;
-	return nil;
+	b.err = .No_Progress
+	return nil
 }

@(private)
 _reader_consume_err :: proc(b: ^Reader) -> io.Error {
-	err := b.err;
-	b.err = nil;
-	return err;
+	err := b.err
+	b.err = nil
+	return err
 }

 // reader_peek returns the next n bytes without advancing the reader
@@ -103,151 +109,145 @@ _reader_consume_err :: proc(b: ^Reader) -> io.Error {
 // explaining why the read is short
 // The error will be .Buffer_Full if n is larger than the internal buffer size
 reader_peek :: proc(b: ^Reader, n: int) -> (data: []byte, err: io.Error) {
-	n := n;
+	n := n

 	if n < 0 {
-		return nil, .Negative_Count;
+		return nil, .Negative_Count
 	}
-	b.last_byte = -1;
-	b.last_rune_size = -1;
+	b.last_byte = -1
+	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) {
-		return b.buf[b.r : b.w], .Buffer_Full;
+		return b.buf[b.r : b.w], .Buffer_Full
 	}

 	if available := b.w - b.r; available < n {
-		n = available;
-		err = _reader_consume_err(b);
+		n = available
+		err = _reader_consume_err(b)
 		if err == nil {
-			err = .Buffer_Full;
+			err = .Buffer_Full
 		}
 	}

-	return b.buf[b.r : b.r+n], err;
+	return b.buf[b.r : b.r+n], err
 }

 // reader_buffered returns the number of bytes that can be read from the current buffer
 reader_buffered :: proc(b: ^Reader) -> int {
-	return b.w - b.r;
+	return b.w - b.r
 }

 // reader_discard skips the next n bytes, and returns the number of bytes that were discarded
 reader_discard :: proc(b: ^Reader, n: int) -> (discarded: int, err: io.Error) {
 	if n < 0 {
-		return 0, .Negative_Count;
+		return 0, .Negative_Count
 	}
 	if n == 0 {
-		return;
+		return
 	}

-	remaining := n;
+	remaining := n
 	for {
-		skip := reader_buffered(b);
+		skip := reader_buffered(b)
 		if skip == 0 {
-			if fill_err := _reader_read_new_chunk(b); fill_err != nil {
-				return 0, fill_err;
-			}
-			skip = reader_buffered(b);
+			_reader_read_new_chunk(b) or_return
+			skip = reader_buffered(b)
 		}
-		skip = min(skip, remaining);
-		b.r += skip;
-		remaining -= skip;
+		skip = min(skip, remaining)
+		b.r += skip
+		remaining -= skip
 		if remaining == 0 {
-			return n, nil;
+			return n, nil
 		}
 		if b.err != nil {
-			return n - remaining, _reader_consume_err(b);
+			return n - remaining, _reader_consume_err(b)
 		}
 	}

-	return;
+	return
 }

 // reader_read reads data into p
 // The bytes are taken from at most one read on the underlying Reader, which means n may be less than len(p)
 reader_read :: proc(b: ^Reader, p: []byte) -> (n: int, err: io.Error) {
-	n = len(p);
+	n = len(p)
 	if n == 0 {
 		if reader_buffered(b) > 0 {
-			return 0, nil;
+			return 0, nil
 		}
-		return 0, _reader_consume_err(b);
+		return 0, _reader_consume_err(b)
 	}
 	if b.r == b.w {
 		if b.err != nil {
-			return 0, _reader_consume_err(b);
+			return 0, _reader_consume_err(b)
 		}

 		if len(p) >= len(b.buf) {
-			n, b.err = io.read(b.rd, p);
+			n, b.err = io.read(b.rd, p)
 			if n < 0 {
-				return 0, .Negative_Read;
+				return 0, .Negative_Read
 			}

 			if n > 0 {
-				b.last_byte = int(p[n-1]);
-				b.last_rune_size = -1;
+				b.last_byte = int(p[n-1])
+				b.last_rune_size = -1
 			}
-			return n, _reader_consume_err(b);
+			return n, _reader_consume_err(b)
 		}

-		b.r, b.w = 0, 0;
-		n, b.err = io.read(b.rd, b.buf);
+		b.r, b.w = 0, 0
+		n, b.err = io.read(b.rd, b.buf)
 		if n < 0 {
-			return 0, .Negative_Read;
+			return 0, .Negative_Read
 		}
 		if n == 0 {
-			return 0, _reader_consume_err(b);
+			return 0, _reader_consume_err(b)
 		}
-		b.w += n;
+		b.w += n
 	}

-	n = copy(p, b.buf[b.r:b.w]);
-	b.r += n;
-	b.last_byte = int(b.buf[b.r-1]);
-	b.last_rune_size = -1;
-	return n, nil;
+	n = copy(p, b.buf[b.r:b.w])
+	b.r += n
+	b.last_byte = int(b.buf[b.r-1])
+	b.last_rune_size = -1
+	return n, nil
 }

 // 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) {
-	b.last_rune_size = -1;
+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;
+			return 0, _reader_consume_err(b)
 		}
+		_reader_read_new_chunk(b) or_return
 	}
-	c := b.buf[b.r];
-	b.r += 1;
-	b.last_byte = int(c);
-	return c, nil;
+	c = b.buf[b.r]
+	b.r += 1
+	b.last_byte = int(c)
+	return
 }

 // reader_unread_byte unreads the last byte. Only the most recently read byte can be unread
 reader_unread_byte :: proc(b: ^Reader) -> io.Error {
 	if b.last_byte < 0 || b.r == 0 && b.w > 0 {
-		return .Invalid_Unread;
+		return .Invalid_Unread
 	}
 	if b.r > 0 {
-		b.r -= 1;
+		b.r -= 1
 	} else {
 		// b.r == 0 && b.w == 0
-		b.w = 1;
+		b.w = 1
 	}
-	b.buf[b.r] = byte(b.last_byte);
-	b.last_byte = -1;
-	b.last_rune_size = -1;
-	return nil;
+	b.buf[b.r] = byte(b.last_byte)
+	b.last_byte = -1
+	b.last_rune_size = -1
+	return nil
 }

 // reader_read_rune reads a single UTF-8 encoded unicode character
@@ -258,135 +258,93 @@ 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;
+	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;
+	r, size = rune(b.buf[b.r]), 1
 	if r >= utf8.RUNE_SELF {
-		r, size = utf8.decode_rune(b.buf[b.r : b.w]);
+		r, size = utf8.decode_rune(b.buf[b.r : b.w])
 	}
-	b.r += size;
-	b.last_byte = int(b.buf[b.r-1]);
-	b.last_rune_size = size;
-	return;
+	b.r += size
+	b.last_byte = int(b.buf[b.r-1])
+	b.last_rune_size = size
+	return
 }

 // reader_unread_rune unreads the last rune. Only the most recently read rune can be unread
 reader_unread_rune :: proc(b: ^Reader) -> io.Error {
 	if b.last_rune_size < 0 || b.r < b.last_rune_size {
-		return .Invalid_Unread;
+		return .Invalid_Unread
 	}
-	b.r -= b.last_rune_size;
-	b.last_byte = -1;
-	b.last_rune_size = -1;
-	return nil;
+	b.r -= b.last_rune_size
+	b.last_byte = -1
+	b.last_rune_size = -1
+	return nil
 }

 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]);
+		n, err := io.write(w, b.buf[b.r:b.w])
 		if n < 0 {
-			return 0, .Negative_Write;
+			return 0, .Negative_Write
 		}
-		b.r += n;
-		return i64(n), err;
+		b.r += n
+		return i64(n), err
 	}

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

+	m: i64
 	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;
+		m, err = write_buf(b, w)
+		n += m // this needs to be done before returning
 		if err != nil {
-			return;
-		}
-		if err = _reader_read_new_chunk(b); err != nil {
-			return;
+			return
 		}
+		_reader_read_new_chunk(b) or_return
 	}

 	if b.err == .EOF {
-		b.err = nil;
+		b.err = nil
 	}

-	err = _reader_consume_err(b);
-	return;
+	err = _reader_consume_err(b)
+	return
 }



 // 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;
-	return;
+	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_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);
-	},
-};
-
-
+_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
+	case .Query:
+		return io.query_utility({.Read, .Destroy, .Query})
+	}
+	return 0, .Empty
+}

 //
 // Utility procedures
@@ -404,71 +362,69 @@ _reader_vtable := &io.Stream_VTable{
 // reader_read_slice returns err != nil if and only if line does not end in delim
 //
 reader_read_slice :: proc(b: ^Reader, delim: byte) -> (line: []byte, err: io.Error) {
-	s := 0;
+	s := 0
 	for {
 		if i := bytes.index_byte(b.buf[b.r+s : b.w], delim); i >= 0 {
-			i += s;
-			line = b.buf[b.r:][:i+1];
-			b.r += i + 1;
-			break;
+			i += s
+			line = b.buf[b.r:][:i+1]
+			b.r += i + 1
+			break
 		}

 		if b.err != nil {
-			line = b.buf[b.r : b.w];
-			b.r = b.w;
-			err = _reader_consume_err(b);
-			break;
+			line = b.buf[b.r : b.w]
+			b.r = b.w
+			err = _reader_consume_err(b)
+			break
 		}

 		if reader_buffered(b) >= len(b.buf) {
-			b.r = b.w;
-			line = b.buf;
-			err = .Buffer_Full;
-			break;
+			b.r = b.w
+			line = b.buf
+			err = .Buffer_Full
+			break
 		}

-		s = b.w - b.r;
+		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 {
-		b.last_byte = int(line[i]);
-		b.last_rune_size = -1;
+		b.last_byte = int(line[i])
+		b.last_rune_size = -1
 	}

-	return;
+	return
 }

 // reader_read_bytes reads until the first occurrence of delim from the Reader
 // It returns an allocated slice containing the data up to and including the delimiter
 reader_read_bytes :: proc(b: ^Reader, delim: byte, allocator := context.allocator) -> (buf: []byte, err: io.Error) {
-	full: [dynamic]byte;
-	full.allocator = allocator;
+	full: [dynamic]byte
+	full.allocator = allocator

-	frag: []byte;
+	frag: []byte
 	for {
-		e: io.Error;
-		frag, e = reader_read_slice(b, delim);
+		e: io.Error
+		frag, e = reader_read_slice(b, delim)
 		if e == nil {
-			break;
+			break
 		}
 		if e != .Buffer_Full {
-			err = e;
-			break;
+			err = e
+			break
 		}

-		append(&full, ..frag);
+		append(&full, ..frag)
 	}
-	append(&full, ..frag);
-	return full[:], err;
+	append(&full, ..frag)
+	return full[:], err
 }

 // reader_read_string reads until the first occurrence of delim from the Reader
 // It returns an allocated string containing the data up to and including the delimiter
 reader_read_string :: proc(b: ^Reader, delim: byte, allocator := context.allocator) -> (string, io.Error) {
-	buf, err := reader_read_bytes(b, delim, allocator);
-	return string(buf), err;
+	buf, err := reader_read_bytes(b, delim, allocator)
+	return string(buf), err
 }
@@ -0,0 +1,337 @@
+package bufio
+
+import "core:bytes"
+import "core:io"
+import "core:mem"
+import "core:unicode/utf8"
+import "base:intrinsics"
+
+// Extra errors returns by scanning procedures
+Scanner_Extra_Error :: enum i32 {
+	None,
+	Negative_Advance,
+	Advanced_Too_Far,
+	Bad_Read_Count,
+	Too_Long,
+	Too_Short,
+}
+
+Scanner_Error :: union #shared_nil {
+	io.Error,
+	Scanner_Extra_Error,
+}
+
+// Split_Proc is the signature of the split procedure used to tokenize the input.
+Split_Proc :: proc(data: []byte, at_eof: bool) -> (advance: int, token: []byte, err: Scanner_Error, final_token: bool)
+
+Scanner :: struct {
+	r:              io.Reader,
+	split:          Split_Proc,
+
+	buf:            [dynamic]byte,
+	max_token_size: int,
+	start:          int,
+	end:            int,
+	token:          []byte,
+
+	_err: Scanner_Error,
+	max_consecutive_empty_reads:  int,
+	successive_empty_token_count: int,
+	scan_called: bool,
+	done:        bool,
+}
+
+DEFAULT_MAX_SCAN_TOKEN_SIZE :: 1<<16
+
+@(private)
+_INIT_BUF_SIZE :: 4096
+
+scanner_init :: proc(s: ^Scanner, r: io.Reader, buf_allocator := context.allocator) -> ^Scanner {
+	s.r = r
+	s.split = scan_lines
+	s.max_token_size = DEFAULT_MAX_SCAN_TOKEN_SIZE
+	s.buf.allocator = buf_allocator
+	return s
+}
+scanner_init_with_buffer :: proc(s: ^Scanner, r: io.Reader, buf: []byte) -> ^Scanner {
+	s.r = r
+	s.split = scan_lines
+	s.max_token_size = DEFAULT_MAX_SCAN_TOKEN_SIZE
+	s.buf = mem.buffer_from_slice(buf)
+	resize(&s.buf, cap(s.buf))
+	return s
+}
+scanner_destroy :: proc(s: ^Scanner) {
+	delete(s.buf)
+}
+
+
+// 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:
+		return nil
+	}
+	return s._err
+}
+
+// Returns the most recent token created by scanner_scan.
+// The underlying array may point to data that may be overwritten
+// by another call to scanner_scan.
+// Treat the returned value as if it is immutable.
+scanner_bytes :: proc(s: ^Scanner) -> []byte {
+	return s.token
+}
+
+// Returns the most recent token created by scanner_scan.
+// The underlying array may point to data that may be overwritten
+// by another call to scanner_scan.
+// Treat the returned value as if it is immutable.
+scanner_text :: proc(s: ^Scanner) -> string {
+	return string(s.token)
+}
+
+// scanner_scan advances the scanner
+scanner_scan :: proc(s: ^Scanner) -> bool {
+	set_err :: proc(s: ^Scanner, err: Scanner_Error) {
+		switch s._err {
+		case nil, .EOF:
+			s._err = err
+		}
+	}
+
+	if s.done {
+		return false
+	}
+	s.scan_called = true
+
+	for {
+		// Check if a token is possible with what is available
+		// Allow the split procedure to recover if it fails
+		if s.start < s.end || s._err != nil {
+			advance, token, err, final_token := s.split(s.buf[s.start:s.end], s._err != nil)
+			if final_token {
+				s.token = token
+				s.done = true
+				return true
+			}
+			if err != nil {
+				set_err(s, err)
+				return false
+			}
+
+			// Do advance
+			if advance < 0 {
+				set_err(s, .Negative_Advance)
+				return false
+			}
+			if advance > s.end-s.start {
+				set_err(s, .Advanced_Too_Far)
+				return false
+			}
+			s.start += advance
+
+			s.token = token
+			if s.token != nil {
+				if s._err == nil || advance > 0 {
+					s.successive_empty_token_count = 0
+				} else {
+					s.successive_empty_token_count += 1
+
+					if s.max_consecutive_empty_reads <= 0 {
+						s.max_consecutive_empty_reads = DEFAULT_MAX_CONSECUTIVE_EMPTY_READS
+					}
+					if s.successive_empty_token_count > s.max_consecutive_empty_reads {
+						set_err(s, .No_Progress)
+						return false
+					}
+				}
+				return true
+			}
+		}
+
+		// If an error is hit, no token can be created
+		if s._err != nil {
+			s.start = 0
+			s.end = 0
+			return false
+		}
+
+		// More data must be required to be read
+		if s.start > 0 && (s.end == len(s.buf) || s.start > len(s.buf)/2) {
+			copy(s.buf[:], s.buf[s.start:s.end])
+			s.end -= s.start
+			s.start = 0
+		}
+
+		could_be_too_short := false
+
+		// Resize the buffer if full
+		if s.end == len(s.buf) {
+			if s.max_token_size <= 0 {
+				s.max_token_size = DEFAULT_MAX_SCAN_TOKEN_SIZE
+			}
+			if len(s.buf) >= s.max_token_size {
+				set_err(s, .Too_Long)
+				return false
+			}
+			// overflow check
+			new_size := _INIT_BUF_SIZE
+			if len(s.buf) > 0 {
+				overflowed: bool
+				if new_size, overflowed = intrinsics.overflow_mul(len(s.buf), 2); overflowed {
+					set_err(s, .Too_Long)
+					return false
+				}
+			}
+
+			old_size := len(s.buf)
+			new_size = min(new_size, s.max_token_size)
+			resize(&s.buf, new_size)
+			s.end -= s.start
+			s.start = 0
+
+			could_be_too_short = old_size >= len(s.buf)
+
+		}
+
+		// Read data into the buffer
+		loop := 0
+		for {
+			n, err := io.read(s.r, s.buf[s.end:len(s.buf)])
+			if n < 0 || len(s.buf)-s.end < n {
+				set_err(s, .Bad_Read_Count)
+				break
+			}
+			s.end += n
+			if err != nil {
+				set_err(s, err)
+				break
+			}
+			if n > 0 {
+				s.successive_empty_token_count = 0
+				break
+			}
+			loop += 1
+
+			if s.max_consecutive_empty_reads <= 0 {
+				s.max_consecutive_empty_reads = DEFAULT_MAX_CONSECUTIVE_EMPTY_READS
+			}
+			if loop > s.max_consecutive_empty_reads {
+				if could_be_too_short {
+					set_err(s, .Too_Short)
+				} else {
+					set_err(s, .No_Progress)
+				}
+				break
+			}
+		}
+	}
+}
+
+scan_bytes :: proc(data: []byte, at_eof: bool) -> (advance: int, token: []byte, err: Scanner_Error, final_token: bool) {
+	if at_eof && len(data) == 0 {
+		return
+	}
+	return 1, data[0:1], nil, false
+}
+
+scan_runes :: proc(data: []byte, at_eof: bool) -> (advance: int, token: []byte, err: Scanner_Error, final_token: bool) {
+	if at_eof && len(data) == 0 {
+		return
+	}
+
+	if data[0] < utf8.RUNE_SELF {
+		advance = 1
+		token = data[0:1]
+		return
+	}
+
+	_, width := utf8.decode_rune(data)
+	if width > 1 {
+		advance = width
+		token = data[0:width]
+		return
+	}
+
+	if !at_eof && !utf8.full_rune(data) {
+		return
+	}
+
+	@thread_local ERROR_RUNE := []byte{0xef, 0xbf, 0xbd}
+
+	advance = 1
+	token = ERROR_RUNE
+	return
+}
+
+scan_words :: proc(data: []byte, at_eof: bool) -> (advance: int, token: []byte, err: Scanner_Error, final_token: bool) {
+	is_space :: proc "contextless" (r:  rune) -> bool {
+		switch r {
+		// lower ones
+		case ' ', '\t', '\n', '\v', '\f', '\r':
+			return true
+		case '\u0085', '\u00a0':
+			return true
+		// higher ones
+		case '\u2000' ..= '\u200a':
+			return true
+		case '\u1680', '\u2028', '\u2029', '\u202f', '\u205f', '\u3000':
+			return true
+		}
+		return false
+	}
+
+	// skip spaces at the beginning
+	start := 0
+	for width := 0; start < len(data); start += width {
+		r: rune
+		r, width = utf8.decode_rune(data[start:])
+		if !is_space(r) {
+			break
+		}
+	}
+
+	for width, i := 0, start; i < len(data); i += width {
+		r: rune
+		r, width = utf8.decode_rune(data[i:])
+		if is_space(r) {
+			advance = i+width
+			token = data[start:i]
+			return
+		}
+	}
+
+	if at_eof && len(data) > start {
+		advance = len(data)
+		token = data[start:]
+		return
+	}
+
+	advance = start
+	return
+}
+
+scan_lines :: proc(data: []byte, at_eof: bool) -> (advance: int, token: []byte, err: Scanner_Error, final_token: bool) {
+	trim_carriage_return :: proc "contextless" (data: []byte) -> []byte {
+		if len(data) > 0 && data[len(data)-1] == '\r' {
+			return data[0:len(data)-1]
+		}
+		return data
+	}
+
+	if at_eof && len(data) == 0 {
+		return
+	}
+	if i := bytes.index_byte(data, '\n'); i >= 0 {
+		advance = i+1
+		token = trim_carriage_return(data[0:i])
+		return
+	}
+
+	if at_eof {
+		advance = len(data)
+		token = trim_carriage_return(data)
+	}
+	return
+}
@@ -15,153 +15,155 @@ Writer :: struct {

 	err: io.Error,

+	max_consecutive_empty_writes: int,
+
 }

 writer_init :: proc(b: ^Writer, wr: io.Writer, size: int = DEFAULT_BUF_SIZE, allocator := context.allocator) {
-	size := size;
-	size = max(size, MIN_READ_BUFFER_SIZE);
-	writer_reset(b, wr);
-	b.buf_allocator = allocator;
-	b.buf = make([]byte, size, allocator);
+	size := size
+	size = max(size, MIN_READ_BUFFER_SIZE)
+	writer_reset(b, wr)
+	b.buf_allocator = allocator
+	b.buf = make([]byte, size, allocator)
 }

 writer_init_with_buf :: proc(b: ^Writer, wr: io.Writer, buf: []byte) {
-	writer_reset(b, wr);
-	b.buf_allocator = {};
-	b.buf = buf;
+	writer_reset(b, wr)
+	b.buf_allocator = {}
+	b.buf = buf
 }

 // writer_destroy destroys the underlying buffer with its associated allocator IFF that allocator has been set
 writer_destroy :: proc(b: ^Writer) {
-	delete(b.buf, b.buf_allocator);
-	b^ = {};
+	delete(b.buf, b.buf_allocator)
+	b^ = {}
 }

 // writer_size returns the size of underlying buffer in bytes
 writer_size :: proc(b: ^Writer) -> int {
-	return len(b.buf);
+	return len(b.buf)
 }

 writer_reset :: proc(b: ^Writer, w: io.Writer) {
-	b.wr = w;
-	b.n = 0;
-	b.err = nil;
+	b.wr = w
+	b.n = 0
+	b.err = nil
 }


 // writer_flush writes any buffered data into the underlying io.Writer
 writer_flush :: proc(b: ^Writer) -> io.Error {
 	if b.err != nil {
-		return b.err;
+		return b.err
 	}
 	if b.n == 0 {
-		return nil;
+		return nil
 	}

-	n, err := io.write(b.wr, b.buf[0:b.n]);
+	n, err := io.write(b.wr, b.buf[0:b.n])
 	if n < b.n && err == nil {
-		err = .Short_Write;
+		err = .Short_Write
 	}
 	if err != nil {
 		if n > 0 && n < b.n {
-			copy(b.buf[:b.n-n], b.buf[n : b.n]);
+			copy(b.buf[:b.n-n], b.buf[n : b.n])
 		}
-		b.n -= n;
-		b.err = err;
-		return err;
+		b.n -= n
+		b.err = err
+		return err
 	}
-	b.n = 0;
-	return nil;
+	b.n = 0
+	return nil
 }

 // writer_available returns how many bytes are unused in the buffer
 writer_available :: proc(b: ^Writer) -> int {
-	return len(b.buf) - b.n;
+	return len(b.buf) - b.n
 }

 // writer_buffered returns the number of bytes that have been writted into the current buffer
 writer_buffered :: proc(b: ^Writer) -> int {
-	return b.n;
+	return b.n
 }

 // writer_write writes the contents of p 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 :: proc(b: ^Writer, p: []byte) -> (n: int, err: io.Error) {
-	p := p;
+	p := p
 	for len(p) > writer_available(b) && b.err == nil {
-		m: int;
+		m: int
 		if writer_buffered(b) == 0 {
-			m, b.err = io.write(b.wr, p);
+			m, b.err = io.write(b.wr, p)
 		} else {
-			m = copy(b.buf[b.n:], p);
-			b.n += m;
-			writer_flush(b);
+			m = copy(b.buf[b.n:], p)
+			b.n += m
+			writer_flush(b)
 		}
-		n += m;
-		p = p[m:];
+		n += m
+		p = p[m:]
 	}
 	if b.err != nil {
-		return n, b.err;
+		return n, b.err
 	}
-	m := copy(b.buf[b.n:], p);
-	b.n += m;
-	m += n;
-	return m, nil;
+	m := copy(b.buf[b.n:], p)
+	b.n += m
+	m += n
+	return m, nil
 }

 // writer_write_byte writes a single byte
 writer_write_byte :: proc(b: ^Writer, c: byte) -> io.Error {
 	if b.err != nil {
-		return b.err;
+		return b.err
 	}
 	if writer_available(b) <= 0 && writer_flush(b) != nil {
-		return b.err;
+		return b.err
 	}
-	b.buf[b.n] = c;
-	b.n += 1;
-	return nil;
+	b.buf[b.n] = c
+	b.n += 1
+	return nil
 }

 // writer_write_rune writes a single unicode code point, and returns the number of bytes written with any error
 writer_write_rune :: proc(b: ^Writer, r: rune) -> (size: int, err: io.Error) {
 	if r < utf8.RUNE_SELF {
-		err = writer_write_byte(b, byte(r));
-		size = 0 if err != nil else 1;
-		return;
+		err = writer_write_byte(b, byte(r))
+		size = 0 if err != nil else 1
+		return
 	}
 	if b.err != nil {
-		return 0, b.err;
+		return 0, b.err
 	}

-	buf: [4]u8;
+	buf: [4]u8

-	n := writer_available(b);
+	n := writer_available(b)
 	if n < utf8.UTF_MAX {
-		writer_flush(b);
+		writer_flush(b)
 		if b.err != nil {
-			return 0, b.err;
+			return 0, b.err
 		}
-		n = writer_available(b);
+		n = writer_available(b)
 		if n < utf8.UTF_MAX {
 			// this only happens if the buffer is very small
-			w: int;
-			buf, w = utf8.encode_rune(r);
-			return writer_write(b, buf[:w]);
+			w: int
+			buf, w = utf8.encode_rune(r)
+			return writer_write(b, buf[:w])
 		}
 	}

-	buf, size = utf8.encode_rune(r);
-	copy(b.buf[b.n:], buf[:size]);
-	b.n += size;
-	return;
+	buf, size = utf8.encode_rune(r)
+	copy(b.buf[b.n:], buf[:size])
+	b.n += size
+	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) {
-	return writer_write(b, transmute([]byte)s);
+	return writer_write(b, transmute([]byte)s)
 }

 // writer_read_from is to support io.Reader_From types
@@ -169,87 +171,78 @@ writer_write_string :: proc(b: ^Writer, s: string) -> (int, io.Error) {
 // this procedure calls the underlying read_from implementation without buffering
 writer_read_from :: proc(b: ^Writer, r: io.Reader) -> (n: i64, err: io.Error) {
 	if b.err != nil {
-		return 0, b.err;
+		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 {
-			if ferr := writer_flush(b); ferr != nil {
-				return n, ferr;
-			}
+			writer_flush(b) or_return
 		}
-		m: int;
-		nr := 0;
-		for nr < MAX_CONSECUTIVE_EMPTY_READS {
-			m, err = io.read(r, b.buf[b.n:]);
+		if b.max_consecutive_empty_writes <= 0 {
+			b.max_consecutive_empty_writes = DEFAULT_MAX_CONSECUTIVE_EMPTY_READS
+		}
+
+		m: int
+		nr := 0
+		for nr < b.max_consecutive_empty_writes {
+			m, err = io.read(r, b.buf[b.n:])
 			if m != 0 || err != nil {
-				break;
+				break
 			}
-			nr += 1;
+			nr += 1
 		}
-		if nr == MAX_CONSECUTIVE_EMPTY_READS {
-			return n, .No_Progress;
+		if nr == b.max_consecutive_empty_writes {
+			return n, .No_Progress
 		}
-		b.n += m;
-		n += i64(m);
+		b.n += m
+		n += i64(m)
 		if err != nil {
-			break;
+			break
 		}
 	}

 	if err == .EOF {
 		if writer_available(b) == 0 {
-			err = writer_flush(b);
+			err = writer_flush(b)
 		} else {
-			err = nil;
+			err = nil
 		}
 	}
-	return;
+	return
 }



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



@(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_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
+}
@@ -1,120 +0,0 @@
-// This is purely for documentation
-package builtin
-
-nil   :: nil;
-false :: 0!==0;
-true  :: 0==0;
-
-ODIN_OS      :: ODIN_OS;
-ODIN_ARCH    :: ODIN_ARCH;
-ODIN_ENDIAN  :: ODIN_ENDIAN;
-ODIN_VENDOR  :: ODIN_VENDOR;
-ODIN_VERSION :: ODIN_VERSION;
-ODIN_ROOT    :: ODIN_ROOT;
-ODIN_DEBUG   :: ODIN_DEBUG;
-
-byte :: u8; // alias
-
-bool          :: bool;
-b8            :: b8;
-b16           :: b16;
-b32           :: b32;
-b64           :: b64;
-
-i8            :: i8;
-u8            :: u8;
-i16           :: i16;
-u16           :: u16;
-i32           :: i32;
-u32           :: u32;
-i64           :: i64;
-u64           :: u64;
-
-i128          :: i128;
-u128          :: u128;
-
-rune          :: rune;
-
-f16           :: f16;
-f32           :: f32;
-f64           :: f64;
-
-complex32     :: complex32;
-complex64     :: complex64;
-complex128    :: complex128;
-
-quaternion64  :: quaternion64;
-quaternion128 :: quaternion128;
-quaternion256 :: quaternion256;
-
-int           :: int;
-uint          :: uint;
-uintptr       :: uintptr;
-
-rawptr        :: rawptr;
-string        :: string;
-cstring       :: cstring;
-any           :: any;
-
-typeid        :: typeid;
-
-// Endian Specific Types
-i16le         :: i16le;
-u16le         :: u16le;
-i32le         :: i32le;
-u32le         :: u32le;
-i64le         :: i64le;
-u64le         :: u64le;
-i128le        :: i128le;
-u128le        :: u128le;
-
-i16be         :: i16be;
-u16be         :: u16be;
-i32be         :: i32be;
-u32be         :: u32be;
-i64be         :: i64be;
-u64be         :: u64be;
-i128be        :: i128be;
-u128be        :: u128be;
-
-
-f16le         :: f16le;
-f32le         :: f32le;
-f64le         :: f64le;
-
-f16be         :: f16be;
-f32be         :: f32be;
-f64be         :: f64be;
-
-
-
-// Procedures
-len :: proc(array: Array_Type) -> int ---
-cap :: proc(array: Array_Type) -> int ---
-
-size_of      :: proc($T: typeid) -> int ---
-align_of     :: proc($T: typeid) -> int ---
-offset_of    :: proc($T: typeid) -> uintptr ---
-type_of      :: proc(x: expr) -> type ---
-type_info_of :: proc($T: typeid) -> ^runtime.Type_Info ---
-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 ---
-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, ...) ---
-
-min   :: proc(values: ..T) -> T ---
-max   :: proc(values: ..T) -> T ---
-abs   :: proc(value: T) -> T ---
-clamp :: proc(value, minimum, maximum: T) -> T ---
-
-soa_zip :: proc(slices: ...) -> #soa[]Struct ---
-soa_unzip :: proc(value: $S/#soa[]$E) -> (slices: ...) ---
@@ -3,10 +3,10 @@ package bytes
 import "core:io"
 import "core:unicode/utf8"

-MIN_READ :: 512;
+MIN_READ :: 512

@(private)
-SMALL_BUFFER_SIZE :: 64;
+SMALL_BUFFER_SIZE :: 64

 // A Buffer is a variable-sized buffer of bytes with a io.Stream interface
 // The zero value for Buffer is an empty buffer ready to use.
@@ -28,406 +28,381 @@ Read_Op :: enum i8 {


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

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

 buffer_init_allocator :: proc(b: ^Buffer, len, cap: int, allocator := context.allocator) {
-	b.buf.allocator = allocator;
-	reserve(&b.buf, cap);
-	resize(&b.buf, len);
+	if b.buf == nil {
+		b.buf = make([dynamic]byte, len, cap, allocator)
+		return
+	}
+
+	b.buf.allocator = allocator
+	reserve(&b.buf, cap)
+	resize(&b.buf, len)
 }

 buffer_destroy :: proc(b: ^Buffer) {
-	delete(b.buf);
-	buffer_reset(b);
+	delete(b.buf)
+	buffer_reset(b)
 }

 buffer_to_bytes :: proc(b: ^Buffer) -> []byte {
-	return b.buf[b.off:];
+	return b.buf[b.off:]
 }

 buffer_to_string :: proc(b: ^Buffer) -> string {
 	if b == nil {
-		return "<nil>";
+		return "<nil>"
 	}
-	return string(b.buf[b.off:]);
+	return string(b.buf[b.off:])
 }

 buffer_is_empty :: proc(b: ^Buffer) -> bool {
-	return len(b.buf) <= b.off;
+	return len(b.buf) <= b.off
 }

 buffer_length :: proc(b: ^Buffer) -> int {
-	return len(b.buf) - b.off;
+	return len(b.buf) - b.off
 }

 buffer_capacity :: proc(b: ^Buffer) -> int {
-	return cap(b.buf);
+	return cap(b.buf)
 }

 buffer_reset :: proc(b: ^Buffer) {
-	clear(&b.buf);
-	b.off = 0;
-	b.last_read = .Invalid;
+	clear(&b.buf)
+	b.off = 0
+	b.last_read = .Invalid
 }


 buffer_truncate :: proc(b: ^Buffer, n: int) {
 	if n == 0 {
-		buffer_reset(b);
-		return;
+		buffer_reset(b)
+		return
 	}
-	b.last_read = .Invalid;
+	b.last_read = .Invalid
 	if n < 0 || n > buffer_length(b) {
-		panic("bytes.truncate: truncation out of range");
+		panic("bytes.truncate: truncation out of range")
 	}
-	resize(&b.buf, b.off+n);
+	resize(&b.buf, b.off+n)
 }

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

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

-	c := cap(b.buf);
-	if n <= c/2 - m {
-		copy(b.buf[:], b.buf[b.off:]);
-	} else if c > max(int) - c - n {
-		panic("bytes.Buffer: too large");
-	} else {
-		resize(&b.buf, 2*c + n);
-		copy(b.buf[:], b.buf[b.off:]);
+	if b.buf == nil && n <= SMALL_BUFFER_SIZE {
+		// Fixes #2756 by preserving allocator if already set on Buffer via init_buffer_allocator
+		reserve(&b.buf, SMALL_BUFFER_SIZE)
+		resize(&b.buf, n)
+		return 0
 	}
-	b.off = 0;
-	resize(&b.buf, m+n);
-	return m;
+
+	c := cap(b.buf)
+	if n <= c/2 - m {
+		copy(b.buf[:], b.buf[b.off:])
+	} else if c > max(int) - c - n {
+		panic("bytes.Buffer: too large")
+	} else {
+		resize(&b.buf, 2*c + n)
+		copy(b.buf[:], b.buf[b.off:])
+	}
+	b.off = 0
+	resize(&b.buf, m+n)
+	return m
 }

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

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


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

 buffer_write_string :: proc(b: ^Buffer, s: string) -> (n: int, err: io.Error) {
-	b.last_read = .Invalid;
-	m, ok := _buffer_try_grow(b, len(s));
+	b.last_read = .Invalid
+	m, ok := _buffer_try_grow(b, len(s))
 	if !ok {
-		m = _buffer_grow(b, len(s));
+		m = _buffer_grow(b, len(s))
 	}
-	return copy(b.buf[m:], s), nil;
+	return copy(b.buf[m:], s), nil
 }

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

 buffer_write_rune :: proc(b: ^Buffer, r: rune) -> (n: int, err: io.Error) {
 	if r < utf8.RUNE_SELF {
-		buffer_write_byte(b, byte(r));
-		return 1, nil;
+		buffer_write_byte(b, byte(r))
+		return 1, nil
 	}
-	b.last_read = .Invalid;
-	m, ok := _buffer_try_grow(b, utf8.UTF_MAX);
+	b.last_read = .Invalid
+	m, ok := _buffer_try_grow(b, utf8.UTF_MAX)
 	if !ok {
-		m = _buffer_grow(b, utf8.UTF_MAX);
+		m = _buffer_grow(b, utf8.UTF_MAX)
 	}
-	res: [4]byte;
-	res, n = utf8.encode_rune(r);
-	copy(b.buf[m:][:utf8.UTF_MAX], res[:n]);
-	resize(&b.buf, m+n);
-	return;
+	res: [4]byte
+	res, n = utf8.encode_rune(r)
+	copy(b.buf[m:][:utf8.UTF_MAX], res[:n])
+	resize(&b.buf, m+n)
+	return
 }

 buffer_next :: proc(b: ^Buffer, n: int) -> []byte {
-	n := n;
-	b.last_read = .Invalid;
-	m := buffer_length(b);
+	n := n
+	b.last_read = .Invalid
+	m := buffer_length(b)
 	if n > m {
-		n = m;
+		n = m
 	}
-	data := b.buf[b.off : b.off + n];
-	b.off += n;
+	data := b.buf[b.off : b.off + n]
+	b.off += n
 	if n > 0 {
-		b.last_read = .Read;
+		b.last_read = .Read
 	}
-	return data;
+	return data
 }

 buffer_read :: proc(b: ^Buffer, p: []byte) -> (n: int, err: io.Error) {
-	b.last_read = .Invalid;
+	b.last_read = .Invalid
 	if buffer_is_empty(b) {
-		buffer_reset(b);
+		buffer_reset(b)
 		if len(p) == 0 {
-			return 0, nil;
+			return 0, nil
 		}
-		return 0, .EOF;
+		return 0, .EOF
 	}
-	n = copy(p, b.buf[b.off:]);
-	b.off += n;
+	n = copy(p, b.buf[b.off:])
+	b.off += n
 	if n > 0 {
-		b.last_read = .Read;
+		b.last_read = .Read
 	}
-	return;
+	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) {
-	b.last_read = .Invalid;
+	b.last_read = .Invalid

-	if offset < 0 || offset >= len(b.buf) {
-		err = .Invalid_Offset;
-		return;
-	}
-
-	if 0 <= offset && offset < len(b.buf) {
-		n = copy(p, b.buf[offset:]);
+	if uint(offset) >= len(b.buf) {
+		err = .Invalid_Offset
+		return
 	}
+	n = copy(p, b.buf[offset:])
 	if n > 0 {
-		b.last_read = .Read;
+		b.last_read = .Read
 	}
-	return;
+	return
 }


 buffer_read_byte :: proc(b: ^Buffer) -> (byte, io.Error) {
 	if buffer_is_empty(b) {
-		buffer_reset(b);
-		return 0, .EOF;
+		buffer_reset(b)
+		return 0, .EOF
 	}
-	c := b.buf[b.off];
-	b.off += 1;
-	b.last_read = .Read;
-	return c, nil;
+	c := b.buf[b.off]
+	b.off += 1
+	b.last_read = .Read
+	return c, nil
 }

 buffer_read_rune :: proc(b: ^Buffer) -> (r: rune, size: int, err: io.Error) {
 	if buffer_is_empty(b) {
-		buffer_reset(b);
-		return 0, 0, .EOF;
+		buffer_reset(b)
+		return 0, 0, .EOF
 	}
-	c := b.buf[b.off];
+	c := b.buf[b.off]
 	if c < utf8.RUNE_SELF {
-		b.off += 1;
-		b.last_read = .Read_Rune1;
-		return rune(c), 1, nil;
+		b.off += 1
+		b.last_read = .Read_Rune1
+		return rune(c), 1, nil
 	}
-	r, size = utf8.decode_rune(b.buf[b.off:]);
-	b.off += size;
-	b.last_read = Read_Op(i8(size));
-	return;
+	r, size = utf8.decode_rune(b.buf[b.off:])
+	b.off += size
+	b.last_read = Read_Op(i8(size))
+	return
 }

 buffer_unread_byte :: proc(b: ^Buffer) -> io.Error {
 	if b.last_read == .Invalid {
-		return .Invalid_Unread;
+		return .Invalid_Unread
 	}
-	b.last_read = .Invalid;
+	b.last_read = .Invalid
 	if b.off > 0 {
-		b.off -= 1;
+		b.off -= 1
 	}
-	return nil;
+	return nil
 }

 buffer_unread_rune :: proc(b: ^Buffer) -> io.Error {
 	if b.last_read <= .Invalid {
-		return .Invalid_Unread;
+		return .Invalid_Unread
 	}
 	if b.off >= int(b.last_read) {
-		b.off -= int(i8(b.last_read));
+		b.off -= int(i8(b.last_read))
 	}
-	b.last_read = .Invalid;
-	return nil;
+	b.last_read = .Invalid
+	return nil
 }


 buffer_read_bytes :: proc(b: ^Buffer, delim: byte) -> (line: []byte, err: io.Error) {
-	i := index_byte(b.buf[b.off:], delim);
-	end := b.off + i + 1;
+	i := index_byte(b.buf[b.off:], delim)
+	end := b.off + i + 1
 	if i < 0 {
-		end = len(b.buf);
-		err = .EOF;
+		end = len(b.buf)
+		err = .EOF
 	}
-	line = b.buf[b.off:end];
-	b.off = end;
-	b.last_read = .Read;
-	return;
+	line = b.buf[b.off:end]
+	b.off = end
+	b.last_read = .Read
+	return
 }

 buffer_read_string :: proc(b: ^Buffer, delim: byte) -> (line: string, err: io.Error) {
-	slice: []byte;
-	slice, err = buffer_read_bytes(b, delim);
-	return string(slice), err;
+	slice: []byte
+	slice, err = buffer_read_bytes(b, delim)
+	return string(slice), err
 }

 buffer_write_to :: proc(b: ^Buffer, w: io.Writer) -> (n: i64, err: io.Error) {
-	b.last_read = .Invalid;
+	b.last_read = .Invalid
 	if byte_count := buffer_length(b); byte_count > 0 {
-		m, e := io.write(w, b.buf[b.off:]);
+		m, e := io.write(w, b.buf[b.off:])
 		if m > byte_count {
-			panic("bytes.buffer_write_to: invalid io.write count");
+			panic("bytes.buffer_write_to: invalid io.write count")
 		}
-		b.off += m;
-		n = i64(m);
+		b.off += m
+		n = i64(m)
 		if e != nil {
-			err = e;
-			return;
+			err = e
+			return
 		}
 		if m != byte_count {
-			err = .Short_Write;
-			return;
+			err = .Short_Write
+			return
 		}
 	}
-	buffer_reset(b);
-	return;
+	buffer_reset(b)
+	return
 }

 buffer_read_from :: proc(b: ^Buffer, r: io.Reader) -> (n: i64, err: io.Error) #no_bounds_check {
-	b.last_read = .Invalid;
+	b.last_read = .Invalid
 	for {
-		i := _buffer_grow(b, MIN_READ);
-		resize(&b.buf, i);
-		m, e := io.read(r, b.buf[i:cap(b.buf)]);
+		i := _buffer_grow(b, MIN_READ)
+		resize(&b.buf, i)
+		m, e := io.read(r, b.buf[i:cap(b.buf)])
 		if m < 0 {
-			err = .Negative_Read;
-			return;
+			err = .Negative_Read
+			return
 		}

-		resize(&b.buf, i+m);
-		n += i64(m);
+		resize(&b.buf, i+m)
+		n += i64(m)
 		if e == .EOF {
-			return;
+			return
 		}
 		if e != nil {
-			err = e;
-			return;
+			err = e
+			return
 		}
 	}
-	return;
+	return
 }


 buffer_to_stream :: proc(b: ^Buffer) -> (s: io.Stream) {
-	s.stream_data = b;
-	s.stream_vtable = _buffer_vtable;
-	return;
+	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_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);
-	},
-};
-
+_buffer_proc :: proc(stream_data: rawptr, mode: io.Stream_Mode, p: []byte, offset: i64, whence: io.Seek_From) -> (n: i64, err: io.Error) {
+	b := (^Buffer)(stream_data)
+	#partial switch mode {
+	case .Read:
+		return io._i64_err(buffer_read(b, p))
+	case .Read_At:
+		return io._i64_err(buffer_read_at(b, p, int(offset)))
+	case .Write:
+		return io._i64_err(buffer_write(b, p))
+	case .Write_At:
+		return io._i64_err(buffer_write_at(b, p, int(offset)))
+	case .Size:
+		n = i64(buffer_capacity(b))
+		return
+	case .Destroy:
+		buffer_destroy(b)
+		return
+	case .Query:
+		return io.query_utility({.Read, .Read_At, .Write, .Write_At, .Size, .Destroy})
+	}
+	return 0, .Empty
+}
@@ -10,168 +10,149 @@ Reader :: struct {
 }

 reader_init :: proc(r: ^Reader, s: []byte) {
-	r.s = s;
-	r.i = 0;
-	r.prev_rune = -1;
+	r.s = s
+	r.i = 0
+	r.prev_rune = -1
 }

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

 reader_length :: proc(r: ^Reader) -> int {
 	if r.i >= i64(len(r.s)) {
-		return 0;
+		return 0
 	}
-	return int(i64(len(r.s)) - r.i);
+	return int(i64(len(r.s)) - r.i)
 }

 reader_size :: proc(r: ^Reader) -> i64 {
-	return i64(len(r.s));
+	return i64(len(r.s))
 }

 reader_read :: proc(r: ^Reader, p: []byte) -> (n: int, err: io.Error) {
 	if r.i >= i64(len(r.s)) {
-		return 0, .EOF;
+		return 0, .EOF
 	}
-	r.prev_rune = -1;
-	n = copy(p, r.s[r.i:]);
-	r.i += i64(n);
-	return;
+	r.prev_rune = -1
+	n = copy(p, r.s[r.i:])
+	r.i += i64(n)
+	return
 }
 reader_read_at :: proc(r: ^Reader, p: []byte, off: i64) -> (n: int, err: io.Error) {
 	if off < 0 {
-		return 0, .Invalid_Offset;
+		return 0, .Invalid_Offset
 	}
 	if off >= i64(len(r.s)) {
-		return 0, .EOF;
+		return 0, .EOF
 	}
-	n = copy(p, r.s[off:]);
+	n = copy(p, r.s[off:])
 	if n < len(p) {
-		err = .EOF;
+		err = .EOF
 	}
-	return;
+	return
 }
 reader_read_byte :: proc(r: ^Reader) -> (byte, io.Error) {
-	r.prev_rune = -1;
+	r.prev_rune = -1
 	if r.i >= i64(len(r.s)) {
-		return 0, .EOF;
+		return 0, .EOF
 	}
-	b := r.s[r.i];
-	r.i += 1;
-	return b, nil;
+	b := r.s[r.i]
+	r.i += 1
+	return b, nil
 }
 reader_unread_byte :: proc(r: ^Reader) -> io.Error {
 	if r.i <= 0 {
-		return .Invalid_Unread;
+		return .Invalid_Unread
 	}
-	r.prev_rune = -1;
-	r.i -= 1;
-	return nil;
+	r.prev_rune = -1
+	r.i -= 1
+	return nil
 }
 reader_read_rune :: proc(r: ^Reader) -> (ch: rune, size: int, err: io.Error) {
 	if r.i >= i64(len(r.s)) {
-		r.prev_rune = -1;
-		return 0, 0, .EOF;
+		r.prev_rune = -1
+		return 0, 0, .EOF
 	}
-	r.prev_rune = int(r.i);
+	r.prev_rune = int(r.i)
 	if c := r.s[r.i]; c < utf8.RUNE_SELF {
-		r.i += 1;
-		return rune(c), 1, nil;
+		r.i += 1
+		return rune(c), 1, nil
 	}
-	ch, size = utf8.decode_rune(r.s[r.i:]);
-	r.i += i64(size);
-	return;
+	ch, size = utf8.decode_rune(r.s[r.i:])
+	r.i += i64(size)
+	return
 }
 reader_unread_rune :: proc(r: ^Reader) -> io.Error {
 	if r.i <= 0 {
-		return .Invalid_Unread;
+		return .Invalid_Unread
 	}
 	if r.prev_rune < 0 {
-		return .Invalid_Unread;
+		return .Invalid_Unread
 	}
-	r.i = i64(r.prev_rune);
-	r.prev_rune = -1;
-	return nil;
+	r.i = i64(r.prev_rune)
+	r.prev_rune = -1
+	return nil
 }
 reader_seek :: proc(r: ^Reader, offset: i64, whence: io.Seek_From) -> (i64, io.Error) {
-	r.prev_rune = -1;
-	abs: i64;
+	r.prev_rune = -1
+	abs: i64
 	switch whence {
 	case .Start:
-		abs = offset;
+		abs = offset
 	case .Current:
-		abs = r.i + offset;
+		abs = r.i + offset
 	case .End:
-		abs = i64(len(r.s)) + offset;
+		abs = i64(len(r.s)) + offset
 	case:
-		return 0, .Invalid_Whence;
+		return 0, .Invalid_Whence
 	}

 	if abs < 0 {
-		return 0, .Invalid_Offset;
+		return 0, .Invalid_Offset
 	}
-	r.i = abs;
-	return abs, nil;
+	r.i = abs
+	return abs, nil
 }
 reader_write_to :: proc(r: ^Reader, w: io.Writer) -> (n: i64, err: io.Error) {
-	r.prev_rune = -1;
+	r.prev_rune = -1
 	if r.i >= i64(len(r.s)) {
-		return 0, nil;
+		return 0, nil
 	}
-	s := r.s[r.i:];
-	m: int;
-	m, err = io.write(w, s);
+	s := r.s[r.i:]
+	m: int
+	m, err = io.write(w, s)
 	if m > len(s) {
-		panic("bytes.Reader.write_to: invalid io.write_string count");
+		panic("bytes.Reader.write_to: invalid io.write_string count")
 	}
-	r.i += i64(m);
-	n = i64(m);
+	r.i += i64(m)
+	n = i64(m)
 	if m != len(s) && err == nil {
-		err = .Short_Write;
+		err = .Short_Write
 	}
-	return;
+	return
 }


@(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
+}
+
@@ -1,35 +1,106 @@
 package c

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

-CHAR_BIT :: 8;
+char           :: builtin.u8  // assuming -funsigned-char

-bool   :: b.bool;
-char   :: b.u8;
-byte   :: b.byte;
-schar  :: b.i8;
-uchar  :: b.u8;
-short  :: b.i16;
-ushort :: b.u16;
-int    :: b.i32;
-uint   :: b.u32;
+schar          :: builtin.i8
+short          :: builtin.i16
+int            :: builtin.i32
+long           :: builtin.i32 when (ODIN_OS == .Windows || size_of(builtin.rawptr) == 4) else builtin.i64
+longlong       :: builtin.i64

-long  :: b.i32 when (ODIN_OS == "windows" || size_of(b.rawptr) == 4) else b.i64;
-ulong :: b.u32 when (ODIN_OS == "windows" || size_of(b.rawptr) == 4) else b.u64;
+uchar          :: builtin.u8
+ushort         :: builtin.u16
+uint           :: builtin.u32
+ulong          :: builtin.u32 when (ODIN_OS == .Windows || size_of(builtin.rawptr) == 4) else builtin.u64
+ulonglong      :: builtin.u64

-longlong       :: b.i64;
-ulonglong      :: b.u64;
-float          :: b.f32;
-double         :: b.f64;
-complex_float  :: b.complex64;
-complex_double :: b.complex128;
+bool           :: builtin.bool

-#assert(size_of(b.uintptr) == size_of(b.int));
+size_t         :: builtin.uint
+ssize_t        :: builtin.int
+wchar_t        :: builtin.u16 when (ODIN_OS == .Windows) else builtin.u32

-size_t    :: b.uint;
-ssize_t   :: b.int;
-ptrdiff_t :: b.int;
-uintptr_t :: b.uintptr;
-intptr_t  :: b.int;
+float          :: builtin.f32
+double         :: builtin.f64
+complex_float  :: builtin.complex64
+complex_double :: builtin.complex128

-wchar_t :: b.u16 when (ODIN_OS == "windows") else b.u32;
+// 7.20.1 Integer types
+int8_t         :: builtin.i8
+uint8_t        :: builtin.u8
+int16_t        :: builtin.i16
+uint16_t       :: builtin.u16
+int32_t        :: builtin.i32
+uint32_t       :: builtin.u32
+int64_t        :: builtin.i64
+uint64_t       :: builtin.u64
+
+// These are all the same in multiple libc's for multiple architectures.
+int_least8_t   :: builtin.i8
+uint_least8_t  :: builtin.u8
+int_least16_t  :: builtin.i16
+uint_least16_t :: builtin.u16
+int_least32_t  :: builtin.i32
+uint_least32_t :: builtin.u32
+int_least64_t  :: builtin.i64
+uint_least64_t :: builtin.u64
+
+// Same on Windows, Linux, and FreeBSD
+when ODIN_ARCH == .i386 || ODIN_ARCH == .amd64 {
+	int_fast8_t    :: builtin.i8
+	uint_fast8_t   :: builtin.u8
+	int_fast16_t   :: builtin.i32
+	uint_fast16_t  :: builtin.u32
+	int_fast32_t   :: builtin.i32
+	uint_fast32_t  :: builtin.u32
+	int_fast64_t   :: builtin.i64
+	uint_fast64_t  :: builtin.u64
+} else {
+	int_fast8_t    :: builtin.i8
+	uint_fast8_t   :: builtin.u8
+	int_fast16_t   :: builtin.i16
+	uint_fast16_t  :: builtin.u16
+	int_fast32_t   :: builtin.i32
+	uint_fast32_t  :: builtin.u32
+	int_fast64_t   :: builtin.i64
+	uint_fast64_t  :: builtin.u64
+}
+
+intptr_t       :: builtin.int
+uintptr_t      :: builtin.uintptr
+ptrdiff_t      :: distinct intptr_t
+
+intmax_t       :: builtin.i64
+uintmax_t      :: builtin.u64
+
+// Copy C's rules for type promotion here by forcing the type on the literals.
+INT8_MAX       :: int(0x7f)
+INT16_MAX      :: int(0x7fff)
+INT32_MAX      :: int(0x7fffffff)
+INT64_MAX      :: longlong(0x7fffffffffffffff)
+
+UINT8_MAX      :: int(0xff)
+UINT16_MAX     :: int(0xffff)
+UINT32_MAX     :: uint(0xffffffff)
+UINT64_MAX     :: ulonglong(0xffffffffffffffff)
+
+INT8_MIN       :: ~INT8_MAX
+INT16_MIN      :: ~INT16_MAX
+INT32_MIN      :: ~INT32_MAX
+INT64_MIN      :: ~INT64_MAX
+
+SIZE_MAX       :: max(size_t)
+
+PTRDIFF_MIN    :: min(ptrdiff_t)
+PTRDIFF_MAX    :: max(ptrdiff_t)
+
+WCHAR_MIN      :: min(wchar_t)
+WCHAR_MAX      :: max(wchar_t)
+
+NULL           :: rawptr(uintptr(0))
+
+NDEBUG         :: !ODIN_DEBUG
+
+CHAR_BIT :: 8
@@ -6,20 +6,20 @@ const_expr :: proc(rest: ^^Token, tok: ^Token) -> i64 {
 	// TODO(bill): Handle const_expr correctly
 	// This is effectively a mini-parser

-	assert(rest != nil);
-	assert(tok != nil);
-	rest^ = tokenizer.new_eof(tok);
+	assert(rest != nil)
+	assert(tok != nil)
+	rest^ = tokenizer.new_eof(tok)
 	switch v in tok.val {
 	case i64:
-		return v;
+		return v
 	case f64:
-		return i64(v);
+		return i64(v)
 	case string:
-		return 0;
+		return 0
 	case []u16:
 		// TODO
 	case []u32:
 		// TODO
 	}
-	return 0;
+	return 0
 }
@@ -5,150 +5,150 @@ import "core:unicode/utf8"
 unquote_char :: proc(str: string, quote: byte) -> (r: rune, multiple_bytes: bool, tail_string: string, success: bool) {
 	hex_to_int :: proc(c: byte) -> int {
 		switch c {
-		case '0'..'9': return int(c-'0');
-		case 'a'..'f': return int(c-'a')+10;
-		case 'A'..'F': return int(c-'A')+10;
+		case '0'..='9': return int(c-'0')
+		case 'a'..='f': return int(c-'a')+10
+		case 'A'..='F': return int(c-'A')+10
 		}
-		return -1;
+		return -1
 	}
-	w: int;
+	w: int

 	if str[0] == quote && quote == '"' {
-		return;
+		return
 	} else if str[0] >= 0x80 {
-		r, w = utf8.decode_rune_in_string(str);
-		return r, true, str[w:], true;
+		r, w = utf8.decode_rune_in_string(str)
+		return r, true, str[w:], true
 	} else if str[0] != '\\' {
-		return rune(str[0]), false, str[1:], true;
+		return rune(str[0]), false, str[1:], true
 	}

 	if len(str) <= 1 {
-		return;
+		return
 	}
-	s := str;
-	c := s[1];
-	s = s[2:];
+	s := str
+	c := s[1]
+	s = s[2:]

 	switch c {
-	case: r = rune(c);
+	case: r = rune(c)

-	case 'a':  r = '\a';
-	case 'b':  r = '\b';
-	case 'e':  r = '\e';
-	case 'f':  r = '\f';
-	case 'n':  r = '\n';
-	case 'r':  r = '\r';
-	case 't':  r = '\t';
-	case 'v':  r = '\v';
-	case '\\': r = '\\';
+	case 'a':  r = '\a'
+	case 'b':  r = '\b'
+	case 'e':  r = '\e'
+	case 'f':  r = '\f'
+	case 'n':  r = '\n'
+	case 'r':  r = '\r'
+	case 't':  r = '\t'
+	case 'v':  r = '\v'
+	case '\\': r = '\\'

-	case '"':  r = '"';
-	case '\'': r = '\'';
+	case '"':  r = '"'
+	case '\'': r = '\''

-	case '0'..'7':
-		v := int(c-'0');
+	case '0'..='7':
+		v := int(c-'0')
 		if len(s) < 2 {
-			return;
+			return
 		}
 		for i in 0..<len(s) {
-			d := int(s[i]-'0');
+			d := int(s[i]-'0')
 			if d < 0 || d > 7 {
-				return;
+				return
 			}
-			v = (v<<3) | d;
+			v = (v<<3) | d
 		}
-		s = s[2:];
+		s = s[2:]
 		if v > 0xff {
-			return;
+			return
 		}
-		r = rune(v);
+		r = rune(v)

 	case 'x', 'u', 'U':
-		count: int;
+		count: int
 		switch c {
-		case 'x': count = 2;
-		case 'u': count = 4;
-		case 'U': count = 8;
+		case 'x': count = 2
+		case 'u': count = 4
+		case 'U': count = 8
 		}

 		if len(s) < count {
-			return;
+			return
 		}

 		for i in 0..<count {
-			d := hex_to_int(s[i]);
+			d := hex_to_int(s[i])
 			if d < 0 {
-				return;
+				return
 			}
-			r = (r<<4) | rune(d);
+			r = (r<<4) | rune(d)
 		}
-		s = s[count:];
+		s = s[count:]
 		if c == 'x' {
-			break;
+			break
 		}
 		if r > utf8.MAX_RUNE {
-			return;
+			return
 		}
-		multiple_bytes = true;
+		multiple_bytes = true
 	}

-	success = true;
-	tail_string = s;
-	return;
+	success = true
+	tail_string = s
+	return
 }

 unquote_string :: proc(lit: string, allocator := context.allocator) -> (res: string, allocated, success: bool) {
 	contains_rune :: proc(s: string, r: rune) -> int {
 		for c, offset in s {
 			if c == r {
-				return offset;
+				return offset
 			}
 		}
-		return -1;
+		return -1
 	}

-	assert(len(lit) >= 2);
+	assert(len(lit) >= 2)

-	s := lit;
-	quote := '"';
+	s := lit
+	quote := '"'

 	if s == `""` {
-		return "", false, true;
+		return "", false, true
 	}

 	if contains_rune(s, '\n') >= 0 {
-		return s, false, false;
+		return s, false, false
 	}

 	if contains_rune(s, '\\') < 0 && contains_rune(s, quote) < 0 {
 		if quote == '"' {
-			return s, false, true;
+			return s, false, true
 		}
 	}
-	s = s[1:len(s)-1];
+	s = s[1:len(s)-1]


-	buf_len := 3*len(s) / 2;
-	buf := make([]byte, buf_len, allocator);
-	offset := 0;
+	buf_len := 3*len(s) / 2
+	buf := make([]byte, buf_len, allocator)
+	offset := 0
 	for len(s) > 0 {
-		r, multiple_bytes, tail_string, ok := unquote_char(s, byte(quote));
+		r, multiple_bytes, tail_string, ok := unquote_char(s, byte(quote))
 		if !ok {
-			delete(buf);
-			return s, false, false;
+			delete(buf)
+			return s, false, false
 		}
-		s = tail_string;
+		s = tail_string
 		if r < 0x80 || !multiple_bytes {
-			buf[offset] = byte(r);
-			offset += 1;
+			buf[offset] = byte(r)
+			offset += 1
 		} else {
-			b, w := utf8.encode_rune(r);
-			copy(buf[offset:], b[:w]);
-			offset += w;
+			b, w := utf8.encode_rune(r)
+			copy(buf[offset:], b[:w])
+			offset += w
 		}
 	}

-	new_string := string(buf[:offset]);
+	new_string := string(buf[:offset])

-	return new_string, true, true;
+	return new_string, true, true
 }
@@ -11,58 +11,58 @@ Hide_Set :: struct {


 new_hide_set :: proc(name: string) -> ^Hide_Set {
-	hs := new(Hide_Set);
-	hs.name = name;
-	return hs;
+	hs := new(Hide_Set)
+	hs.name = name
+	return hs
 }

 hide_set_contains :: proc(hs: ^Hide_Set, name: string) -> bool {
 	for h := hs; h != nil; h = h.next {
 		if h.name == name {
-			return true;
+			return true
 		}
 	}
-	return false;
+	return false
 }


 hide_set_union :: proc(a, b: ^Hide_Set) -> ^Hide_Set {
-	head: Hide_Set;
-	curr := &head;
+	head: Hide_Set
+	curr := &head

 	for h := a; h != nil; h = h.next {
-		curr.next = new_hide_set(h.name);
-		curr = curr.next;
+		curr.next = new_hide_set(h.name)
+		curr = curr.next
 	}
-	curr.next = b;
-	return head.next;
+	curr.next = b
+	return head.next
 }


 hide_set_intersection :: proc(a, b: ^Hide_Set) -> ^Hide_Set {
-	head: Hide_Set;
-	curr := &head;
+	head: Hide_Set
+	curr := &head

 	for h := a; h != nil; h = h.next {
 		if hide_set_contains(b, h.name) {
-			curr.next = new_hide_set(h.name);
-			curr = curr.next;
+			curr.next = new_hide_set(h.name)
+			curr = curr.next
 		}
 	}
-	return head.next;
+	return head.next
 }


 add_hide_set :: proc(tok: ^Token, hs: ^Hide_Set) -> ^Token {
-	head: Token;
-	curr := &head;
+	head: Token
+	curr := &head

-	tok := tok;
+	tok := tok
 	for ; tok != nil; tok = tok.next {
-		t := copy_token(tok);
-		t.hide_set = hide_set_union(t.hide_set, hs);
-		curr.next = t;
-		curr = curr.next;
+		t := copy_token(tok)
+		t.hide_set = hide_set_union(t.hide_set, hs)
+		curr.next = t
+		curr = curr.next
 	}
-	return head.next;
+	return head.next
 }
@@ -80,29 +80,29 @@ Token :: struct {
 	origin:   ^Token,
 }

-Is_Keyword_Proc :: #type proc(tok: ^Token) -> bool;
+Is_Keyword_Proc :: #type proc(tok: ^Token) -> bool

 copy_token :: proc(tok: ^Token) -> ^Token {
-	t := new_clone(tok^);
-	t.next = nil;
-	return t;
+	t, _ := new_clone(tok^)
+	t.next = nil
+	return t
 }

 new_eof :: proc(tok: ^Token) -> ^Token {
-	t := new_clone(tok^);
-	t.kind = .EOF;
-	t.lit = "";
-	return t;
+	t, _ := new_clone(tok^)
+	t.kind = .EOF
+	t.lit = ""
+	return t
 }

 default_is_keyword :: proc(tok: ^Token) -> bool {
 	if tok.kind == .Keyword {
-		return true;
+		return true
 	}
 	if len(tok.lit) > 0 {
-		return default_keyword_set[tok.lit];
+		return default_keyword_set[tok.lit]
 	}
-	return false;
+	return false
 }


@@ -117,7 +117,7 @@ token_name := [Token_Kind]string {
 	.PP_Number = "preprocessor number",
 	.Comment   = "comment",
 	.EOF       = "eof",
-};
+}

 default_keyword_set := map[string]bool{
 	"auto"          = true,
@@ -166,4 +166,4 @@ default_keyword_set := map[string]bool{
 	"__restrict__"  = true,
 	"__thread"      = true,
 	"__attribute__" = true,
-};
+}
@@ -6,7 +6,7 @@ import "core:strings"
 import "core:unicode/utf8"


-Error_Handler :: #type proc(pos: Pos, fmt: string, args: ..any);
+Error_Handler :: #type proc(pos: Pos, fmt: string, args: ..any)


 Tokenizer :: struct {
@@ -34,415 +34,415 @@ Tokenizer :: struct {
 }

 init_defaults :: proc(t: ^Tokenizer, err: Error_Handler = default_error_handler, warn: Error_Handler = default_warn_handler) {
-	t.err = err;
-	t.warn = warn;
+	t.err = err
+	t.warn = warn
 }


@(private)
 offset_to_pos :: proc(t: ^Tokenizer, offset: int) -> (pos: Pos) {
-	pos.file = t.path;
-	pos.offset = offset;
-	pos.line = t.line_count;
-	pos.column = offset - t.line_offset + 1;
-	return;
+	pos.file = t.path
+	pos.offset = offset
+	pos.line = t.line_count
+	pos.column = offset - t.line_offset + 1
+	return
 }

 default_error_handler :: proc(pos: Pos, msg: string, args: ..any) {
-	fmt.eprintf("%s(%d:%d) ", pos.file, pos.line, pos.column);
-	fmt.eprintf(msg, ..args);
-	fmt.eprintf("\n");
+	fmt.eprintf("%s(%d:%d) ", pos.file, pos.line, pos.column)
+	fmt.eprintf(msg, ..args)
+	fmt.eprintf("\n")
 }

 default_warn_handler :: proc(pos: Pos, msg: string, args: ..any) {
-	fmt.eprintf("%s(%d:%d) warning: ", pos.file, pos.line, pos.column);
-	fmt.eprintf(msg, ..args);
-	fmt.eprintf("\n");
+	fmt.eprintf("%s(%d:%d) warning: ", pos.file, pos.line, pos.column)
+	fmt.eprintf(msg, ..args)
+	fmt.eprintf("\n")
 }

 error_offset :: proc(t: ^Tokenizer, offset: int, msg: string, args: ..any) {
-	pos := offset_to_pos(t, offset);
+	pos := offset_to_pos(t, offset)
 	if t.err != nil {
-		t.err(pos, msg, ..args);
+		t.err(pos, msg, ..args)
 	}
-	t.error_count += 1;
+	t.error_count += 1
 }

 warn_offset :: proc(t: ^Tokenizer, offset: int, msg: string, args: ..any) {
-	pos := offset_to_pos(t, offset);
+	pos := offset_to_pos(t, offset)
 	if t.warn != nil {
-		t.warn(pos, msg, ..args);
+		t.warn(pos, msg, ..args)
 	}
-	t.warning_count += 1;
+	t.warning_count += 1
 }

 error :: proc(t: ^Tokenizer, tok: ^Token, msg: string, args: ..any) {
-	pos := tok.pos;
+	pos := tok.pos
 	if t.err != nil {
-		t.err(pos, msg, ..args);
+		t.err(pos, msg, ..args)
 	}
-	t.error_count += 1;
+	t.error_count += 1
 }

 warn :: proc(t: ^Tokenizer, tok: ^Token, msg: string, args: ..any) {
-	pos := tok.pos;
+	pos := tok.pos
 	if t.warn != nil {
-		t.warn(pos, msg, ..args);
+		t.warn(pos, msg, ..args)
 	}
-	t.warning_count += 1;
+	t.warning_count += 1
 }


 advance_rune :: proc(t: ^Tokenizer) {
 	if t.read_offset < len(t.src) {
-		t.offset = t.read_offset;
+		t.offset = t.read_offset
 		if t.ch == '\n' {
-			t.at_bol = true;
-			t.line_offset = t.offset;
-			t.line_count += 1;
+			t.at_bol = true
+			t.line_offset = t.offset
+			t.line_count += 1
 		}
-		r, w := rune(t.src[t.read_offset]), 1;
+		r, w := rune(t.src[t.read_offset]), 1
 		switch {
 		case r == 0:
-			error_offset(t, t.offset, "illegal character NUL");
+			error_offset(t, t.offset, "illegal character NUL")
 		case r >= utf8.RUNE_SELF:
-			r, w = utf8.decode_rune(t.src[t.read_offset:]);
+			r, w = utf8.decode_rune(t.src[t.read_offset:])
 			if r == utf8.RUNE_ERROR && w == 1 {
-				error_offset(t, t.offset, "illegal UTF-8 encoding");
+				error_offset(t, t.offset, "illegal UTF-8 encoding")
 			} else if r == utf8.RUNE_BOM && t.offset > 0 {
-				error_offset(t, t.offset, "illegal byte order mark");
+				error_offset(t, t.offset, "illegal byte order mark")
 			}
 		}
-		t.read_offset += w;
-		t.ch = r;
+		t.read_offset += w
+		t.ch = r
 	} else {
-		t.offset = len(t.src);
+		t.offset = len(t.src)
 		if t.ch == '\n' {
-			t.at_bol = true;
-			t.line_offset = t.offset;
-			t.line_count += 1;
+			t.at_bol = true
+			t.line_offset = t.offset
+			t.line_count += 1
 		}
-		t.ch = -1;
+		t.ch = -1
 	}
 }

 advance_rune_n :: proc(t: ^Tokenizer, n: int) {
-	for in 0..<n {
-		advance_rune(t);
+	for _ in 0..<n {
+		advance_rune(t)
 	}
 }

 is_digit :: proc(r: rune) -> bool {
-	return '0' <= r && r <= '9';
+	return '0' <= r && r <= '9'
 }

 skip_whitespace :: proc(t: ^Tokenizer) {
 	for {
 		switch t.ch {
 		case ' ', '\t', '\r', '\v', '\f', '\n':
-			t.has_space = true;
-			advance_rune(t);
+			t.has_space = true
+			advance_rune(t)
 		case:
-			return;
+			return
 		}
 	}
 }

 scan_comment :: proc(t: ^Tokenizer) -> string {
-	offset := t.offset-1;
-	next := -1;
+	offset := t.offset-1
+	next := -1
 	general: {
 		if t.ch == '/'{ // line comments
-			advance_rune(t);
+			advance_rune(t)
 			for t.ch != '\n' && t.ch >= 0 {
-				advance_rune(t);
+				advance_rune(t)
 			}

-			next = t.offset;
+			next = t.offset
 			if t.ch == '\n' {
-				next += 1;
+				next += 1
 			}
-			break general;
+			break general
 		}

 		/* style comment */
-		advance_rune(t);
+		advance_rune(t)
 		for t.ch >= 0 {
-			ch := t.ch;
-			advance_rune(t);
+			ch := t.ch
+			advance_rune(t)
 			if ch == '*' && t.ch == '/' {
-				advance_rune(t);
-				next = t.offset;
-				break general;
+				advance_rune(t)
+				next = t.offset
+				break general
 			}
 		}

-		error_offset(t, offset, "comment not terminated");
+		error_offset(t, offset, "comment not terminated")
 	}

-	lit := t.src[offset : t.offset];
+	lit := t.src[offset : t.offset]

 	// NOTE(bill): Strip CR for line comments
 	for len(lit) > 2 && lit[1] == '/' && lit[len(lit)-1] == '\r' {
-		lit = lit[:len(lit)-1];
+		lit = lit[:len(lit)-1]
 	}


-	return string(lit);
+	return string(lit)
 }

 scan_identifier :: proc(t: ^Tokenizer) -> string {
-	offset := t.offset;
+	offset := t.offset

 	for is_ident1(t.ch) {
-		advance_rune(t);
+		advance_rune(t)
 	}

-	return string(t.src[offset : t.offset]);
+	return string(t.src[offset : t.offset])
 }

 scan_string :: proc(t: ^Tokenizer) -> string {
-	offset := t.offset-1;
+	offset := t.offset-1

 	for {
-		ch := t.ch;
+		ch := t.ch
 		if ch == '\n' || ch < 0 {
-			error_offset(t, offset, "string literal was not terminated");
-			break;
+			error_offset(t, offset, "string literal was not terminated")
+			break
 		}
-		advance_rune(t);
+		advance_rune(t)
 		if ch == '"' {
-			break;
+			break
 		}
 		if ch == '\\' {
-			scan_escape(t);
+			scan_escape(t)
 		}
 	}

-	return string(t.src[offset : t.offset]);
+	return string(t.src[offset : t.offset])
 }

 digit_val :: proc(r: rune) -> int {
 	switch r {
-	case '0'..'9':
-		return int(r-'0');
-	case 'A'..'F':
-		return int(r-'A' + 10);
-	case 'a'..'f':
-		return int(r-'a' + 10);
+	case '0'..='9':
+		return int(r-'0')
+	case 'A'..='F':
+		return int(r-'A' + 10)
+	case 'a'..='f':
+		return int(r-'a' + 10)
 	}
-	return 16;
+	return 16
 }

 scan_escape :: proc(t: ^Tokenizer) -> bool {
-	offset := t.offset;
+	offset := t.offset

-	esc := t.ch;
-	n: int;
-	base, max: u32;
+	esc := t.ch
+	n: int
+	base, max: u32
 	switch esc {
 	case 'a', 'b', 'e', 'f', 'n', 't', 'v', 'r', '\\', '\'', '"':
-		advance_rune(t);
-		return true;
+		advance_rune(t)
+		return true

-	case '0'..'7':
+	case '0'..='7':
 		for digit_val(t.ch) < 8 {
-			advance_rune(t);
+			advance_rune(t)
 		}
-		return true;
+		return true
 	case 'x':
-		advance_rune(t);
+		advance_rune(t)
 		for digit_val(t.ch) < 16 {
-			advance_rune(t);
+			advance_rune(t)
 		}
-		return true;
+		return true
 	case 'u':
-		advance_rune(t);
-		n, base, max = 4, 16, utf8.MAX_RUNE;
+		advance_rune(t)
+		n, base, max = 4, 16, utf8.MAX_RUNE
 	case 'U':
-		advance_rune(t);
-		n, base, max = 8, 16, utf8.MAX_RUNE;
+		advance_rune(t)
+		n, base, max = 8, 16, utf8.MAX_RUNE
 	case:
 		if t.ch < 0 {
-			error_offset(t, offset, "escape sequence was not terminated");
+			error_offset(t, offset, "escape sequence was not terminated")
 		} else {
-			break;
+			break
 		}
-		return false;
+		return false
 	}

-	x: u32;
+	x: u32
 	main_loop: for n > 0 {
-		d := u32(digit_val(t.ch));
+		d := u32(digit_val(t.ch))
 		if d >= base {
 			if t.ch == '"' || t.ch == '\'' {
-				break main_loop;
+				break main_loop
 			}
 			if t.ch < 0 {
-				error_offset(t, t.offset, "escape sequence was not terminated");
+				error_offset(t, t.offset, "escape sequence was not terminated")
 			} else {
-				error_offset(t, t.offset, "illegal character '%r' : %d in escape sequence", t.ch, t.ch);
+				error_offset(t, t.offset, "illegal character '%r' : %d in escape sequence", t.ch, t.ch)
 			}
-			return false;
+			return false
 		}

-		x = x*base + d;
-		advance_rune(t);
-		n -= 1;
+		x = x*base + d
+		advance_rune(t)
+		n -= 1
 	}

 	if x > max || 0xd800 <= x && x <= 0xe000 {
-		error_offset(t, offset, "escape sequence is an invalid Unicode code point");
-		return false;
+		error_offset(t, offset, "escape sequence is an invalid Unicode code point")
+		return false
 	}
-	return true;
+	return true
 }

 scan_rune :: proc(t: ^Tokenizer) -> string {
-	offset := t.offset-1;
-	valid := true;
-	n := 0;
+	offset := t.offset-1
+	valid := true
+	n := 0
 	for {
-		ch := t.ch;
+		ch := t.ch
 		if ch == '\n' || ch < 0 {
 			if valid {
-				error_offset(t, offset, "rune literal not terminated");
-				valid = false;
+				error_offset(t, offset, "rune literal not terminated")
+				valid = false
 			}
-			break;
+			break
 		}
-		advance_rune(t);
+		advance_rune(t)
 		if ch == '\'' {
-			break;
+			break
 		}
-		n += 1;
+		n += 1
 		if ch == '\\' {
 			if !scan_escape(t)  {
-				valid = false;
+				valid = false
 			}
 		}
 	}

 	if valid && n != 1 {
-		error_offset(t, offset, "illegal rune literal");
+		error_offset(t, offset, "illegal rune literal")
 	}

-	return string(t.src[offset : t.offset]);
+	return string(t.src[offset : t.offset])
 }

 scan_number :: proc(t: ^Tokenizer, seen_decimal_point: bool) -> (Token_Kind, string) {
 	scan_mantissa :: proc(t: ^Tokenizer, base: int) {
 		for digit_val(t.ch) < base {
-			advance_rune(t);
+			advance_rune(t)
 		}
 	}
 	scan_exponent :: proc(t: ^Tokenizer) {
 		if t.ch == 'e' || t.ch == 'E' || t.ch == 'p' || t.ch == 'P' {
-			advance_rune(t);
+			advance_rune(t)
 			if t.ch == '-' || t.ch == '+' {
-				advance_rune(t);
+				advance_rune(t)
 			}
 			if digit_val(t.ch) < 10 {
-				scan_mantissa(t, 10);
+				scan_mantissa(t, 10)
 			} else {
-				error_offset(t, t.offset, "illegal floating-point exponent");
+				error_offset(t, t.offset, "illegal floating-point exponent")
 			}
 		}
 	}
 	scan_fraction :: proc(t: ^Tokenizer) -> (early_exit: bool) {
 		if t.ch == '.' && peek(t) == '.' {
-			return true;
+			return true
 		}
 		if t.ch == '.' {
-			advance_rune(t);
-			scan_mantissa(t, 10);
+			advance_rune(t)
+			scan_mantissa(t, 10)
 		}
-		return false;
+		return false
 	}

-	check_end := true;
+	check_end := true


-	offset := t.offset;
-	seen_point := seen_decimal_point;
+	offset := t.offset
+	seen_point := seen_decimal_point

 	if seen_point {
-		offset -= 1;
-		scan_mantissa(t, 10);
-		scan_exponent(t);
+		offset -= 1
+		scan_mantissa(t, 10)
+		scan_exponent(t)
 	} else {
 		if t.ch == '0' {
 			int_base :: proc(t: ^Tokenizer, base: int, msg: string) {
-				prev := t.offset;
-				advance_rune(t);
-				scan_mantissa(t, base);
+				prev := t.offset
+				advance_rune(t)
+				scan_mantissa(t, base)
 				if t.offset - prev <= 1 {
-					error_offset(t, t.offset, msg);
+					error_offset(t, t.offset, msg)
 				}
 			}

-			advance_rune(t);
+			advance_rune(t)
 			switch t.ch {
 			case 'b', 'B':
-				int_base(t, 2, "illegal binary integer");
+				int_base(t, 2, "illegal binary integer")
 			case 'x', 'X':
-				int_base(t, 16, "illegal hexadecimal integer");
+				int_base(t, 16, "illegal hexadecimal integer")
 			case:
-				seen_point = false;
-				scan_mantissa(t, 10);
+				seen_point = false
+				scan_mantissa(t, 10)
 				if t.ch == '.' {
-					seen_point = true;
+					seen_point = true
 					if scan_fraction(t) {
-						check_end = false;
+						check_end = false
 					}
 				}
 				if check_end {
-					scan_exponent(t);
-					check_end = false;
+					scan_exponent(t)
+					check_end = false
 				}
 			}
 		}
 	}

 	if check_end {
-		scan_mantissa(t, 10);
+		scan_mantissa(t, 10)

 		if !scan_fraction(t) {
-			scan_exponent(t);
+			scan_exponent(t)
 		}
 	}

-	return .Number, string(t.src[offset : t.offset]);
+	return .Number, string(t.src[offset : t.offset])
 }

 scan_punct :: proc(t: ^Tokenizer, ch: rune) -> (kind: Token_Kind) {
-	kind = .Punct;
+	kind = .Punct
 	switch ch {
 	case:
-		kind = .Invalid;
+		kind = .Invalid

 	case '<', '>':
 		if t.ch == ch {
-			advance_rune(t);
+			advance_rune(t)
 		}
 		if t.ch == '=' {
-			advance_rune(t);
+			advance_rune(t)
 		}
 	case '!', '+', '-', '*', '/', '%', '^', '=':
 		if t.ch == '=' {
-			advance_rune(t);
+			advance_rune(t)
 		}
 	case '#':
 		if t.ch == '#' {
-			advance_rune(t);
+			advance_rune(t)
 		}
 	case '&':
 		if t.ch == '=' || t.ch == '&' {
-			advance_rune(t);
+			advance_rune(t)
 		}
 	case '|':
 		if t.ch == '=' || t.ch == '|' {
-			advance_rune(t);
+			advance_rune(t)
 		}
 	case '(', ')', '[', ']', '{', '}':
 		// okay
@@ -452,216 +452,216 @@ scan_punct :: proc(t: ^Tokenizer, ch: rune) -> (kind: Token_Kind) {
 		// okay
 	case '.':
 		if t.ch == '.' && peek(t) == '.' {
-			advance_rune(t);
-			advance_rune(t); // consume last '.'
+			advance_rune(t)
+			advance_rune(t) // consume last '.'
 		}
 	}
-	return;
+	return
 }

 peek :: proc(t: ^Tokenizer) -> byte {
 	if t.read_offset < len(t.src) {
-		return t.src[t.read_offset];
+		return t.src[t.read_offset]
 	}
-	return 0;
+	return 0
 }
 peek_str :: proc(t: ^Tokenizer, str: string) -> bool {
 	if t.read_offset < len(t.src) {
-		return strings.has_prefix(string(t.src[t.offset:]), str);
+		return strings.has_prefix(string(t.src[t.offset:]), str)
 	}
-	return false;
+	return false
 }

 scan_literal_prefix :: proc(t: ^Tokenizer, str: string, prefix: ^string) -> bool {
 	if peek_str(t, str) {
-		offset := t.offset;
+		offset := t.offset
 		for _ in str {
-			advance_rune(t);
+			advance_rune(t)
 		}
-		prefix^ = string(t.src[offset:][:len(str)-1]);
-		return true;
+		prefix^ = string(t.src[offset:][:len(str)-1])
+		return true
 	}
-	return false;
+	return false
 }


 allow_next_to_be_newline :: proc(t: ^Tokenizer) -> bool {
 	if t.ch == '\n' {
-		advance_rune(t);
-		return true;
+		advance_rune(t)
+		return true
 	} else if t.ch == '\r' && peek(t) == '\n' { // allow for MS-DOS style line endings
-		advance_rune(t); // \r
-		advance_rune(t); // \n
-		return true;
+		advance_rune(t) // \r
+		advance_rune(t) // \n
+		return true
 	}
-	return false;
+	return false
 }

 scan :: proc(t: ^Tokenizer, f: ^File) -> ^Token {
-	skip_whitespace(t);
+	skip_whitespace(t)

-	offset := t.offset;
+	offset := t.offset

-	kind: Token_Kind;
-	lit: string;
-	prefix: string;
+	kind: Token_Kind
+	lit: string
+	prefix: string

 	switch ch := t.ch; {
 	case scan_literal_prefix(t, `u8"`, &prefix):
-		kind = .String;
-		lit = scan_string(t);
+		kind = .String
+		lit = scan_string(t)
 	case scan_literal_prefix(t, `u"`, &prefix):
-		kind = .String;
-		lit = scan_string(t);
+		kind = .String
+		lit = scan_string(t)
 	case scan_literal_prefix(t, `L"`, &prefix):
-		kind = .String;
-		lit = scan_string(t);
+		kind = .String
+		lit = scan_string(t)
 	case scan_literal_prefix(t, `U"`, &prefix):
-		kind = .String;
-		lit = scan_string(t);
+		kind = .String
+		lit = scan_string(t)
 	case scan_literal_prefix(t, `u'`, &prefix):
-		kind = .Char;
-		lit = scan_rune(t);
+		kind = .Char
+		lit = scan_rune(t)
 	case scan_literal_prefix(t, `L'`, &prefix):
-		kind = .Char;
-		lit = scan_rune(t);
+		kind = .Char
+		lit = scan_rune(t)
 	case scan_literal_prefix(t, `U'`, &prefix):
-		kind = .Char;
-		lit = scan_rune(t);
+		kind = .Char
+		lit = scan_rune(t)

 	case is_ident0(ch):
-		lit = scan_identifier(t);
-		kind = .Ident;
+		lit = scan_identifier(t)
+		kind = .Ident
 	case '0' <= ch && ch <= '9':
-		kind, lit = scan_number(t, false);
+		kind, lit = scan_number(t, false)
 	case:
-		advance_rune(t);
+		advance_rune(t)
 		switch ch {
 		case -1:
-			kind = .EOF;
+			kind = .EOF
 		case '\\':
-			kind = .Punct;
+			kind = .Punct
 			if allow_next_to_be_newline(t) {
-				t.at_bol = true;
-				t.has_space = false;
-				return scan(t, f);
+				t.at_bol = true
+				t.has_space = false
+				return scan(t, f)
 			}

 		case '.':
 			if is_digit(t.ch) {
-				kind, lit = scan_number(t, true);
+				kind, lit = scan_number(t, true)
 			} else {
-				kind = scan_punct(t, ch);
+				kind = scan_punct(t, ch)
 			}
 		case '"':
-			kind = .String;
-			lit = scan_string(t);
+			kind = .String
+			lit = scan_string(t)
 		case '\'':
-			kind = .Char;
-			lit = scan_rune(t);
+			kind = .Char
+			lit = scan_rune(t)
 		case '/':
 			if t.ch == '/' || t.ch == '*' {
-				kind = .Comment;
-				lit = scan_comment(t);
-				t.has_space = true;
-				break;
+				kind = .Comment
+				lit = scan_comment(t)
+				t.has_space = true
+				break
 			}
-			fallthrough;
+			fallthrough
 		case:
-			kind = scan_punct(t, ch);
+			kind = scan_punct(t, ch)
 			if kind == .Invalid && ch != utf8.RUNE_BOM {
-				error_offset(t, t.offset, "illegal character '%r': %d", ch, ch);
+				error_offset(t, t.offset, "illegal character '%r': %d", ch, ch)
 			}
 		}
 	}

 	if lit == "" {
-		lit = string(t.src[offset : t.offset]);
+		lit = string(t.src[offset : t.offset])
 	}

 	if kind == .Comment {
-		return scan(t, f);
+		return scan(t, f)
 	}

-	tok := new(Token);
-	tok.kind = kind;
-	tok.lit = lit;
-	tok.pos = offset_to_pos(t, offset);
-	tok.file = f;
-	tok.prefix = prefix;
-	tok.at_bol = t.at_bol;
-	tok.has_space = t.has_space;
+	tok := new(Token)
+	tok.kind = kind
+	tok.lit = lit
+	tok.pos = offset_to_pos(t, offset)
+	tok.file = f
+	tok.prefix = prefix
+	tok.at_bol = t.at_bol
+	tok.has_space = t.has_space

-	t.at_bol, t.has_space = false, false;
+	t.at_bol, t.has_space = false, false

-	return tok;
+	return tok
 }

 tokenize :: proc(t: ^Tokenizer, f: ^File) -> ^Token {
 	setup_tokenizer: {
-		t.src = f.src;
-		t.ch = ' ';
-		t.offset = 0;
-		t.read_offset = 0;
-		t.line_offset = 0;
-		t.line_count = len(t.src) > 0 ? 1 : 0;
-		t.error_count = 0;
-		t.path = f.name;
+		t.src = f.src
+		t.ch = ' '
+		t.offset = 0
+		t.read_offset = 0
+		t.line_offset = 0
+		t.line_count = len(t.src) > 0 ? 1 : 0
+		t.error_count = 0
+		t.path = f.name


-		advance_rune(t);
+		advance_rune(t)
 		if t.ch == utf8.RUNE_BOM {
-			advance_rune(t);
+			advance_rune(t)
 		}
 	}


-	t.at_bol = true;
-	t.has_space = false;
+	t.at_bol = true
+	t.has_space = false

-	head: Token;
-	curr := &head;
+	head: Token
+	curr := &head
 	for {
-		tok := scan(t, f);
+		tok := scan(t, f)
 		if tok == nil {
-			break;
+			break
 		}
-		curr.next = tok;
-		curr = curr.next;
+		curr.next = tok
+		curr = curr.next
 		if tok.kind == .EOF {
-			break;
+			break
 		}
 	}

-	return head.next;
+	return head.next
 }

 add_new_file :: proc(t: ^Tokenizer, name: string, src: []byte, id: int) -> ^File {
-	file := new(File);
-	file.id = id;
-	file.src = src;
-	file.name = name;
-	file.display_name = name;
-	return file;
+	file := new(File)
+	file.id = id
+	file.src = src
+	file.name = name
+	file.display_name = name
+	return file
 }

 tokenize_file :: proc(t: ^Tokenizer, path: string, id: int, loc := #caller_location) -> ^Token {
-	src, ok := os.read_entire_file(path);
+	src, ok := os.read_entire_file(path)
 	if !ok {
-		return nil;
+		return nil
 	}
-	return tokenize(t, add_new_file(t, path, src, id));
+	return tokenize(t, add_new_file(t, path, src, id))
 }


 inline_tokenize :: proc(t: ^Tokenizer, tok: ^Token, src: []byte) -> ^Token {
-	file := new(File);
-	file.src = src;
+	file := new(File)
+	file.src = src
 	if tok.file != nil {
-		file.id = tok.file.id;
-		file.name = tok.file.name;
-		file.display_name = tok.file.name;
+		file.id = tok.file.id
+		file.name = tok.file.name
+		file.display_name = tok.file.name
 	}

-	return tokenize(t, file);
+	return tokenize(t, file)
 }
@@ -4,10 +4,10 @@ package c_frontend_tokenizer
 in_range :: proc(range: []rune, c: rune) -> bool #no_bounds_check {
 	for i := 0; range[i] != -1; i += 2 {
 		if range[i] <= c && c <= range[i+1] {
-			return true;
+			return true
 		}
 	}
-	return false;
+	return false
 }


@@ -15,11 +15,11 @@ in_range :: proc(range: []rune, c: rune) -> bool #no_bounds_check {
 //
 // is_ident0 returns true if a given character is acceptable as the first character of an identifier.
 is_ident0 :: proc(c: rune) -> bool {
-	return in_range(_range_ident0, c);
+	return in_range(_range_ident0, c)
 }
 // is_ident0 returns true if a given character is acceptable as a non-first character of an identifier.
 is_ident1 :: proc(c: rune) -> bool {
-	return is_ident0(c) || in_range(_range_ident1, c);
+	return is_ident0(c) || in_range(_range_ident1, c)
 }

 // Returns the number of columns needed to display a given character in a fixed-width font.
@@ -27,18 +27,18 @@ is_ident1 :: proc(c: rune) -> bool {
 char_width :: proc(c: rune) -> int {
 	switch {
 	case in_range(_range_width0, c):
-		return 0;
+		return 0
 	case in_range(_range_width2, c):
-		return 2;
+		return 2
 	}
-	return 1;
+	return 1
 }

 display_width :: proc(str: string) -> (w: int) {
 	for c in str {
-		w += char_width(c);
+		w += char_width(c)
 	}
-	return;
+	return
 }


@@ -59,12 +59,12 @@ _range_ident0 := []rune{
 	0x90000, 0x9FFFD, 0xA0000, 0xAFFFD, 0xB0000, 0xBFFFD, 0xC0000, 0xCFFFD,
 	0xD0000, 0xDFFFD, 0xE0000, 0xEFFFD,
 	-1,
-};
+}

 _range_ident1 := []rune{
 	'0', '9', '$', '$', 0x0300, 0x036F, 0x1DC0, 0x1DFF, 0x20D0, 0x20FF, 0xFE20, 0xFE2F,
 	-1,
-};
+}


 _range_width0 := []rune{
@@ -105,7 +105,7 @@ _range_width0 := []rune{
 	0x1D167, 0x1D169, 0x1D173, 0x1D182, 0x1D185, 0x1D18B, 0x1D1AA, 0x1D1AD,
 	0x1D242, 0x1D244, 0xE0001, 0xE0001, 0xE0020, 0xE007F, 0xE0100, 0xE01EF,
 	-1,
-};
+}

 _range_width2 := []rune{
 	0x1100, 0x115F, 0x2329, 0x2329, 0x232A, 0x232A, 0x2E80, 0x303E,
@@ -113,4 +113,4 @@ _range_width2 := []rune{
 	0xFE30, 0xFE6F, 0xFF00, 0xFF60, 0xFFE0, 0xFFE6, 0x1F000, 0x1F644,
 	0x20000, 0x2FFFD, 0x30000, 0x3FFFD,
 	-1,
-};
+}
@@ -0,0 +1,73 @@
+# C support
+
+The following is a mostly-complete projection of the C11 standard library as defined by the C11 specification: N1570, or ISO/IEC 9899:2011. Only the macros, types, and functions as required by the standard are projected. Extensions to C, such as POSIX are not handled by these bindings, this is otherwise portable to any implementation which can support a hosted C runtime.
+
+## Support matrix
+| Header            | Status                                             |
+|:------------------|:---------------------------------------------------|
+| `<assert.h>`      | Not applicable, use Odin's `#assert`               |
+| `<complex.h>`     | Mostly projected, see [limitations](#Limitations)  |
+| `<ctype.h>`       | Fully projected                                    |
+| `<errno.h>`       | Fully projected                                    |
+| `<fenv.h>`        | Not projected                                      |
+| `<float.h>`       | Not projected                                      |
+| `<inttypes.h>`    | Fully projected                                    |
+| `<iso646.h>`      | Not applicable, use Odin's operators               |
+| `<limits.h>`      | Not projected                                      |
+| `<locale.h>`      | Not projected                                      |
+| `<math.h>`        | Mostly projected, see [limitations](#Limitations)  |
+| `<setjmp.h>`      | Fully projected                                    |
+| `<signal.h>`      | Fully projected                                    |
+| `<stdalign.h>`    | Not applicable, use Odin's `#align`                |
+| `<stdarg.h>`      | Mostly projected, see [limitations](#Limitations)  |
+| `<stdatomic.h>`   | Fully projected                                    |
+| `<stdbool.h>`     | Not applicable, use Odin's `b32`                   |
+| `<stddef.h>`      | Mostly projected, see [limitations](#Limitations)  |
+| `<stdint.h>`      | Fully projected                                    |
+| `<stdio.h>`       | Fully projected                                    |
+| `<stdlib.h>`      | Fully projected                                    |
+| `<stdnoreturn.h>` | Not applicable, use Odin's divergent return `!`    |
+| `<string.h>`      | Fully projected                                    |
+| `<tgmath.h>`      | Mostly projected, see [limitations](#Limitations)  |
+| `<threads.h>`     | Fully projected                                    |
+| `<time.h>`        | Fully projected                                    |
+| `<uchar.h>`       | Fully projected                                    |
+| `<wchar.h>`       | Fully projected                                    |
+| `<wctype.h>`      | Fully projected                                    |
+
+## Limitations
+Not all C standard library functionality can be fully projected due to language differences. These limitations are listed here.
+
+### `long double`
+As Odin lacks a means to interact with `long double` in it's foreign interface, this projection effort does not bind or define anything requiring `long double` which is permitted by the C standard.
+
+### `<complex.h>`
+The special values `_Complex_I`, `_Imaginary_I` and the appropriate definition of `I` cannot be realized with the same type in Odin as it would be in C. The literal `1i` is tempting to use for these definitions but the semantics differ from C and would be confusing to use.
+
+### `<math.h>`
+The classification functions, e.g: `fpclassify` are required by C to be implemented as macros, meaning no implementation would expose functions in their library we could bind. Instead, we provide native Odin implementations with functionally equivalent semantics and behavior as the C ones. Unfortunately, since classification returns unspecified constant values this may be an ABI break where the value of those constants enter and exit native C code.
+
+### `<stdarg.h>`
+While Odin can interact with variable argument C functions through the use of the `#c_vararg` attribute within a foreign block, it's not actually possible to create procedures in Odin with bodies that have the same ABI as that of variable argument C functions, as a result `va_arg` is not projected.
+
+### `<stddef.h>`
+`offsetof` is not realizable in Odin, however you can use `offset_of` instead.
+
+### `<tgmath.h>`
+C has some strange promotion and type-coercion behavior for `<tgmath.h>` which isn't correctly handled by this projection, specifically involving the use of complex arithmetic and kernels. We do mostly support type-generic math through the use of Odin's explicit procedure overloading, however the semantic behavior of that doesn't match C and so literal expressions of complex type in C may not call the same underlying math kernel functions as they do in Odin through this projection.
+
+## Caveats
+
+In addition to limitations, there are some minor caveats you should be aware when using this projection.
+
+* `errno()` is a function which returns `^int` rather than a macro.
+* `MB_CUR_MAX()` is a function which return `size_t` rather than a macro.
+* Currently only works on Windows (MSVCRT) and Linux (GLIBC or MUSL)
+
+## License
+Every file within this directory is made available under Odin's BSD-2 license
+with the following copyright.
+
+```
+Copyright 2021 Dale Weiler <weilercdale@gmail.com>.
+```
@@ -0,0 +1,84 @@
+package libc
+
+// 7.3 Complex arithmetic
+
+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"
+}
+
+@(default_calling_convention="c")
+foreign libc {
+	// 7.3.5 Trigonometric functions
+	cacos   :: proc(z: complex_double) -> complex_double ---
+	cacosf  :: proc(z: complex_float) -> complex_float ---
+	casin   :: proc(z: complex_double) -> complex_double ---
+	casinf  :: proc(z: complex_float) -> complex_float ---
+	catan   :: proc(z: complex_double) -> complex_double ---
+	catanf  :: proc(z: complex_float) -> complex_float ---
+	ccos    :: proc(z: complex_double) -> complex_double ---
+	ccosf   :: proc(z: complex_float) -> complex_float ---
+	csin    :: proc(z: complex_double) -> complex_double ---
+	csinf   :: proc(z: complex_float) -> complex_float ---
+	ctan    :: proc(z: complex_double) -> complex_double ---
+	ctanf   :: proc(z: complex_float) -> complex_float ---
+
+	// 7.3.6 Hyperbolic functions
+	cacosh  :: proc(z: complex_double) -> complex_double ---
+	cacoshf :: proc(z: complex_float) -> complex_float ---
+	casinh  :: proc(z: complex_double) -> complex_double ---
+	casinhf :: proc(z: complex_float) -> complex_float ---
+	catanh  :: proc(z: complex_double) -> complex_double ---
+	catanhf :: proc(z: complex_float) -> complex_float ---
+	ccosh   :: proc(z: complex_double) -> complex_double ---
+	ccoshf  :: proc(z: complex_float) -> complex_float ---
+	csinh   :: proc(z: complex_double) -> complex_double ---
+	csinhf  :: proc(z: complex_float) -> complex_float ---
+	ctanh   :: proc(z: complex_double) -> complex_double ---
+	ctanhf  :: proc(z: complex_float) -> complex_float ---
+
+	// 7.3.7 Exponential and logarithmic functions
+	cexp    :: proc(z: complex_double) -> complex_double ---
+	cexpf   :: proc(z: complex_float) -> complex_float ---
+	clog    :: proc(z: complex_double) -> complex_double ---
+	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(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 ---
+
+	// 7.3.9 Manipulation functions
+	carg    :: proc(z: complex_double) -> double ---
+	cargf   :: proc(z: complex_float) -> float ---
+	cimag   :: proc(z: complex_double) -> double ---
+	cimagf  :: proc(z: complex_float) -> float ---
+	conj    :: proc(z: complex_double) -> complex_double ---
+	conjf   :: proc(z: complex_float) -> complex_float ---
+	cproj   :: proc(z: complex_double) -> complex_double ---
+	cprojf  :: proc(z: complex_float) -> complex_float ---
+	creal   :: proc(z: complex_double) -> double ---
+	crealf  :: proc(z: complex_float) -> float ---
+}
+
+import builtin "base:builtin"
+
+complex_float  :: distinct builtin.complex64
+complex_double :: distinct builtin.complex128
+
+// Cannot implement _Complex_I or _Imaginary_I in Odin, thus
+// complex and imaginary cannot be implement either.
+
+CMPLX :: #force_inline proc(x, y: double) -> complex_double {
+	return builtin.complex(x, y)
+}
+
+CMPLXF :: #force_inline proc(x, y: float) -> complex_float {
+	return builtin.complex(x, y)
+}
@@ -0,0 +1,32 @@
+package libc
+
+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"
+}
+
+// 7.4 Character handling
+
+@(default_calling_convention="c")
+foreign libc {
+	// 7.4.1 Character classification functions
+	isalnum  :: proc(c: int) -> int ---
+	isalpha  :: proc(c: int) -> int ---
+	isblank  :: proc(c: int) -> int ---
+	iscntrl  :: proc(c: int) -> int ---
+	isdigit  :: proc(c: int) -> int ---
+	isgraph  :: proc(c: int) -> int ---
+	islower  :: proc(c: int) -> int ---
+	isprint  :: proc(c: int) -> int ---
+	ispunct  :: proc(c: int) -> int ---
+	isspace  :: proc(c: int) -> int ---
+	isupper  :: proc(c: int) -> int ---
+	isxdigit :: proc(c: int) -> int ---
+
+	// 7.4.2 Character case mapping functions
+	tolower  :: proc(c: int) -> int ---
+	toupper  :: proc(c: int) -> int ---
+}
@@ -0,0 +1,89 @@
+package libc
+
+// 7.5 Errors
+
+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"
+}
+
+// C11 standard only requires the definition of:
+//	EDOM,
+//	EILSEQ
+//	ERANGE
+when ODIN_OS == .Linux {
+	@(private="file")
+	@(default_calling_convention="c")
+	foreign libc {
+		@(link_name="__errno_location")
+		_get_errno :: proc() -> ^int ---
+	}
+
+	EDOM   :: 33
+	EILSEQ :: 84
+	ERANGE :: 34
+}
+
+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 {
+	@(private="file")
+	@(default_calling_convention="c")
+	foreign libc {
+		@(link_name="__errno")
+		_get_errno :: proc() -> ^int ---
+	}
+
+	EDOM   :: 33
+	EILSEQ :: 84
+	ERANGE :: 34
+}
+
+when ODIN_OS == .Windows {
+	@(private="file")
+	@(default_calling_convention="c")
+	foreign libc {
+		@(link_name="_errno")
+		_get_errno :: proc() -> ^int ---
+	}
+
+	EDOM   :: 33
+	EILSEQ :: 42
+	ERANGE :: 34
+}
+
+when ODIN_OS == .Darwin {
+	@(private="file")
+	@(default_calling_convention="c")
+	foreign libc {
+		@(link_name="__error")
+		_get_errno :: proc() -> ^int ---
+	}
+
+	// Unknown
+	EDOM   :: 33
+	EILSEQ :: 92
+	ERANGE :: 34
+}
+
+// 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 {
+	return _get_errno()
+}
@@ -0,0 +1,404 @@
+package libc
+
+// 7.12 Mathematics
+
+import "base:intrinsics"
+
+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"
+}
+
+// To support C's tgmath behavior we use Odin's explicit procedure overloading,
+// but we cannot use the same names as exported by libc so use @(link_name)
+// and keep them as private symbols of name "libc_"
+@(private="file")
+@(default_calling_convention="c")
+foreign libc {
+	// 7.12.4 Trigonometric functions
+	@(link_name="acos")       libc_acos       :: proc(x: double) -> double ---
+	@(link_name="acosf")      libc_acosf      :: proc(x: float) -> float ---
+	@(link_name="asin")       libc_asin       :: proc(x: double) -> double ---
+	@(link_name="asinf")      libc_asinf      :: proc(x: float) -> float ---
+	@(link_name="atan")       libc_atan       :: proc(x: double) -> double ---
+	@(link_name="atanf")      libc_atanf      :: proc(x: float) -> float ---
+	@(link_name="atan2")      libc_atan2      :: proc(y: double, x: double) -> double ---
+	@(link_name="atan2f")     libc_atan2f     :: proc(y: float, x: float) -> float ---
+	@(link_name="cos")        libc_cos        :: proc(x: double) -> double ---
+	@(link_name="cosf")       libc_cosf       :: proc(x: float) -> float ---
+	@(link_name="sin")        libc_sin        :: proc(x: double) -> double ---
+	@(link_name="sinf")       libc_sinf       :: proc(x: float) -> float ---
+	@(link_name="tan")        libc_tan        :: proc(x: double) -> double ---
+	@(link_name="tanf")       libc_tanf       :: proc(x: float) -> float ---
+
+	// 7.12.5 Hyperbolic functions
+	@(link_name="acosh")      libc_acosh      :: proc(x: double) -> double ---
+	@(link_name="acoshf")     libc_acoshf     :: proc(x: float) -> float ---
+	@(link_name="asinh")      libc_asinh      :: proc(x: double) -> double ---
+	@(link_name="asinhf")     libc_asinhf     :: proc(x: float) -> float ---
+	@(link_name="atanh")      libc_atanh      :: proc(x: double) -> double ---
+	@(link_name="atanhf")     libc_atanhf     :: proc(x: float) -> float ---
+	@(link_name="cosh")       libc_cosh       :: proc(x: double) -> double ---
+	@(link_name="coshf")      libc_coshf      :: proc(x: float) -> float ---
+	@(link_name="sinh")       libc_sinh       :: proc(x: double) -> double ---
+	@(link_name="sinhf")      libc_sinhf      :: proc(x: float) -> float ---
+	@(link_name="tanh")       libc_tanh       :: proc(x: double) -> double ---
+	@(link_name="tanhf")      libc_tanhf      :: proc(x: float) -> float ---
+
+	// 7.12.6 Exponential and logarithmic functions
+	@(link_name="exp")        libc_exp        :: proc(x: double) -> double ---
+	@(link_name="expf")       libc_expf       :: proc(x: float) -> float ---
+	@(link_name="exp2")       libc_exp2       :: proc(x: double) -> double ---
+	@(link_name="exp2f")      libc_exp2f      :: proc(x: float) -> float ---
+	@(link_name="expm1")      libc_expm1      :: proc(x: double) -> double ---
+	@(link_name="expm1f")     libc_expm1f     :: proc(x: float) -> float ---
+	@(link_name="frexp")      libc_frexp      :: proc(value: double, exp: ^int) -> double ---
+	@(link_name="frexpf")     libc_frexpf     :: proc(value: float, exp: ^int) -> float ---
+	@(link_name="ilogb")      libc_ilogb      :: proc(x: double) -> int ---
+	@(link_name="ilogbf")     libc_ilogbf     :: proc(x: float) -> int ---
+	@(link_name="ldexp")      libc_ldexp      :: proc(x: double, exp: int) -> double ---
+	@(link_name="ldexpf")     libc_ldexpf     :: proc(x: float, exp: int) -> float ---
+	@(link_name="log")        libc_log        :: proc(x: double) -> double ---
+	@(link_name="logf")       libc_logf       :: proc(x: float) -> float ---
+	@(link_name="log10")      libc_log10      :: proc(x: double) -> double ---
+	@(link_name="log10f")     libc_log10f     :: proc(x: float) -> float ---
+	@(link_name="log1p")      libc_log1p      :: proc(x: double) -> double ---
+	@(link_name="log1pf")     libc_log1pf     :: proc(x: float) -> float ---
+	@(link_name="log2")       libc_log2       :: proc(x: double) -> double ---
+	@(link_name="log2f")      libc_log2f      :: proc(x: float) -> float ---
+	@(link_name="logb")       libc_logb       :: proc(x: double) -> double ---
+	@(link_name="logbf")      libc_logbf      :: proc(x: float) -> float ---
+	@(link_name="modf")       libc_modf       :: proc(value: double, iptr: ^double) -> double ---
+	@(link_name="modff")      libc_modff      :: proc(value: float, iptr: ^float) -> float ---
+	@(link_name="scalbn")     libc_scalbn     :: proc(x: double, n: int) -> double ---
+	@(link_name="scalbnf")    libc_scalbnf    :: proc(x: float, n: int) -> float ---
+	@(link_name="scalbln")    libc_scalbln    :: proc(x: double, n: long) -> double ---
+	@(link_name="scalblnf")   libc_scalblnf   :: proc(x: float, n: long) -> float ---
+
+	// 7.12.7 Power and absolute-value functions
+	@(link_name="cbrt")       libc_cbrt       :: proc(x: double) -> double ---
+	@(link_name="cbrtf")      libc_cbrtf      :: proc(x: float) -> float ---
+	@(link_name="fabs")       libc_fabs       :: proc(x: double) -> double ---
+	@(link_name="fabsf")      libc_fabsf      :: proc(x: float) -> float ---
+	@(link_name="hypot")      libc_hypot      :: proc(x: double, y: double) -> double ---
+	@(link_name="hypotf")     libc_hypotf     :: proc(x: float, y: float) -> float ---
+	@(link_name="pow")        libc_pow        :: proc(x: double, y: double) -> double ---
+	@(link_name="powf")       libc_powf       :: proc(x: float, y: float) -> float ---
+	@(link_name="sqrt")       libc_sqrt       :: proc(x: double) -> double ---
+	@(link_name="sqrtf")      libc_sqrtf      :: proc(x: float) -> float ---
+
+	// 7.12.8 Error and gamma functions
+	@(link_name="erf")        libc_erf        :: proc(x: double) -> double ---
+	@(link_name="erff")       libc_erff       :: proc(x: float) -> float ---
+	@(link_name="erfc")       libc_erfc       :: proc(x: double) -> double ---
+	@(link_name="erfcf")      libc_erfcf      :: proc(x: float) -> float ---
+	@(link_name="lgamma")     libc_lgamma     :: proc(x: double) -> double ---
+	@(link_name="lgammaf")    libc_lgammaf    :: proc(x: float) -> float ---
+	@(link_name="tgamma")     libc_tgamma     :: proc(x: double) -> double ---
+	@(link_name="tgammaf")    libc_tgammaf    :: proc(x: float) -> float ---
+
+	// 7.12.9 Nearest integer functions
+	@(link_name="ceil")       libc_ceil       :: proc(x: double) -> double ---
+	@(link_name="ceilf")      libc_ceilf      :: proc(x: float) -> float ---
+	@(link_name="floor")      libc_floor      :: proc(x: double) -> double ---
+	@(link_name="floorf")     libc_floorf     :: proc(x: float) -> float ---
+	@(link_name="nearbyint")  libc_nearbyint  :: proc(x: double) -> double ---
+	@(link_name="nearbyintf") libc_nearbyintf :: proc(x: float) -> float ---
+	@(link_name="rint")       libc_rint       :: proc(x: double) -> double ---
+	@(link_name="rintf")      libc_rintf      :: proc(x: float) -> float ---
+	@(link_name="lrint")      libc_lrint      :: proc(x: double) -> long ---
+	@(link_name="lrintf")     libc_lrintf     :: proc(x: float) -> long ---
+	@(link_name="llrint")     libc_llrint     :: proc(x: double) -> longlong ---
+	@(link_name="llrintf")    libc_llrintf    :: proc(x: float) -> longlong ---
+	@(link_name="round")      libc_round      :: proc(x: double) -> double ---
+	@(link_name="roundf")     libc_roundf     :: proc(x: float) -> float ---
+	@(link_name="lround")     libc_lround     :: proc(x: double) -> long ---
+	@(link_name="lroundf")    libc_lroundf    :: proc(x: float) -> long ---
+	@(link_name="llround")    libc_llround    :: proc(x: double) -> longlong ---
+	@(link_name="llroundf")   libc_llroundf   :: proc(x: float) -> longlong ---
+	@(link_name="trunc")      libc_trunc      :: proc(x: double) -> double ---
+	@(link_name="truncf")     libc_truncf     :: proc(x: float) -> float ---
+
+	// 7.12.10 Remainder functions
+	@(link_name="fmod")       libc_fmod       :: proc(x: double, y: double) -> double ---
+	@(link_name="fmodf")      libc_fmodf      :: proc(x: float, y: float) -> float ---
+	@(link_name="remainder")  libc_remainder  :: proc(x: double, y: double) -> double ---
+	@(link_name="remainderf") libc_remainderf :: proc(x: float, y: float) -> float ---
+	@(link_name="remquo")     libc_remquo     :: proc(x: double, y: double, quo: ^int) -> double ---
+	@(link_name="remquof")    libc_remquof    :: proc(x: float, y: float, quo: ^int) -> float ---
+
+	// 7.12.11 Manipulation functions
+	@(link_name="copysign")   libc_copysign   :: proc(x: double, y: double) -> double ---
+	@(link_name="copysignf")  libc_copysignf  :: proc(x: float, y: float) -> float ---
+	@(link_name="nan")        libc_nan        :: proc(tagp: cstring) -> double ---
+	@(link_name="nanf")       libc_nanf       :: proc(tagp: cstring) -> float ---
+	@(link_name="nextafter")  libc_nextafter  :: proc(x: double, y: double) -> double ---
+	@(link_name="nextafterf") libc_nextafterf :: proc(x: float, y: float) -> float ---
+
+	// 7.12.12 Maximum, minimum, and positive difference functions
+	@(link_name="fdim")       libc_fdim       :: proc(x: double, y: double) -> double ---
+	@(link_name="fdimf")      libc_fdimf      :: proc(x: float, y: float) -> float ---
+	@(link_name="fmax")       libc_fmax       :: proc(x: double, y: double) -> double ---
+	@(link_name="fmaxf")      libc_fmaxf      :: proc(x: float, y: float) -> float ---
+	@(link_name="fmin")       libc_fmin       :: proc(x: double, y: double) -> double ---
+	@(link_name="fminf")      libc_fminf      :: proc(x: float, y: float) -> float ---
+	@(link_name="fma")        libc_fma        :: proc(x, y, z: double) -> double ---
+	@(link_name="fmaf")       libc_fmaf       :: proc(x, y, z: float) -> float ---
+}
+
+@(private="file")
+_nan_bit_pattern := ~u64(0)
+
+// On amd64 Windows and Linux, float_t and double_t are respectively both
+// their usual types. On x86 it's not possible to define these types correctly
+// since they would be long double which Odin does have support for.
+float_t          :: float
+double_t         :: double
+
+NAN              := transmute(double)(_nan_bit_pattern)
+INFINITY         :: 1e5000
+
+HUGE_VALF        :: INFINITY
+HUGE_VAL         :: double(INFINITY)
+
+MATH_ERRNO       :: 1
+MATH_ERREXCEPT   :: 2
+
+math_errhandling :: 2 // Windows, Linux, macOS all use this mode.
+
+FP_ILOGBNAN      :: -1 - int((~uint(0)) >> 1)
+FP_ILOGB0        :: FP_ILOGBNAN
+
+// Number classification constants. These do not have to match libc since we
+// implement our own classification functions as libc requires they be macros,
+// which means libc does not export standard functions for them.
+FP_NAN           :: 0
+FP_INFINITE      :: 1
+FP_ZERO          :: 2
+FP_NORMAL        :: 3
+FP_SUBNORMAL     :: 4
+
+@(private)
+_fpclassify :: #force_inline proc(x: double) -> int {
+	u := transmute(uint64_t)x
+	switch e := u >> 52 & 0x7ff; e {
+	case 0:     return FP_SUBNORMAL if (u << 1)  != 0 else FP_ZERO
+	case 0x7ff: return FP_NAN       if (u << 12) != 0 else FP_INFINITE
+	}
+	return FP_NORMAL
+}
+
+@(private)
+_fpclassifyf :: #force_inline proc(x: float) -> int {
+	u := transmute(uint32_t)x
+	switch e := u >> 23 & 0xff; e {
+	case 0:    return FP_SUBNORMAL if (u << 1)  != 0 else FP_ZERO
+	case 0xff: return FP_NAN       if (u << 9)  != 0 else FP_INFINITE
+	}
+	return FP_NORMAL
+}
+
+@(private)
+_signbit :: #force_inline proc(x: double) -> int {
+	return int(transmute(uint64_t)x >> 63)
+}
+
+@(private)
+_signbitf :: #force_inline proc(x: float) -> int {
+	return int(transmute(uint32_t)x >> 31)
+}
+
+isfinite :: #force_inline proc(x: $T) -> bool where intrinsics.type_is_float(T) {
+	return fpclassify(x) == FP_INFINITE
+}
+
+isinf :: #force_inline proc(x: $T) -> bool where intrinsics.type_is_float(T) {
+	return fpclassify(x) > FP_INFINITE
+}
+
+isnan :: #force_inline proc(x: $T) -> bool where intrinsics.type_is_float(T) {
+	return fpclassify(x) == FP_NAN
+}
+
+isnormal :: #force_inline proc(x: $T) -> bool where intrinsics.type_is_float(T) {
+	return fpclassify(x) == FP_NORMAL
+}
+
+// These are special in that they avoid float exceptions. They cannot just be
+// implemented as the relational comparisons, as that would produce an invalid
+// "sticky" state that propagates and affects maths results. These need
+// to be implemented natively in Odin assuming isunordered to prevent that.
+isgreater :: #force_inline proc(x, y: $T) -> bool where intrinsics.type_is_float(T) {
+	return !isunordered(x, y) && x > y
+}
+
+isgreaterequal :: #force_inline proc(x, y: $T) -> bool where intrinsics.type_is_float(T) {
+	return !isunordered(x, y) && x >= y
+}
+
+isless :: #force_inline proc(x, y: $T) -> bool where intrinsics.type_is_float(T) {
+	return !isunordered(x, y) && x < y
+}
+
+islessequal :: #force_inline proc(x, y: $T) -> bool where intrinsics.type_is_float(T) {
+	return !isunordered(x, y) && x <= y
+}
+
+islessgreater :: #force_inline proc(x, y: $T) -> bool where intrinsics.type_is_float(T) {
+	return !isunordered(x, y) && x <= y
+}
+
+isunordered :: #force_inline proc(x, y: $T) -> bool where intrinsics.type_is_float(T) {
+	if isnan(x) {
+		// Force evaluation of y to propagate exceptions for ordering semantics.
+		// To ensure correct semantics of IEEE 754 this cannot be compiled away.
+		sink: T
+		intrinsics.volatile_store(&sink, intrinsics.volatile_load(&y))
+		return true
+	}
+	return isnan(y)
+}
+
+fpclassify :: proc{_fpclassify, _fpclassifyf}
+signbit    :: proc{_signbit, _signbitf}
+
+// Emulate tgmath.h behavior with explicit procedure overloading here.
+acos       :: proc{libc_acos, libc_acosf, cacos, cacosf}
+asin       :: proc{libc_asin, libc_asinf, casin, casinf}
+atan       :: proc{libc_atan, libc_atanf, catan, catanf}
+atan2      :: proc{libc_atan2, libc_atan2f}
+cos        :: proc{libc_cos, libc_cosf, ccos, ccosf}
+sin        :: proc{libc_sin, libc_sinf, csin, csinf}
+tan        :: proc{libc_tan, libc_tanf, ctan, ctanf}
+
+acosh      :: proc{libc_acosh, libc_acoshf, cacosh, cacoshf}
+asinh      :: proc{libc_asinh, libc_asinhf, casinh, casinhf}
+atanh      :: proc{libc_atanh, libc_atanhf, catanh, catanhf}
+cosh       :: proc{libc_cosh, libc_coshf, ccosh, ccoshf}
+sinh       :: proc{libc_sinh, libc_sinhf, csinh, csinhf}
+tanh       :: proc{libc_tanh, libc_tanhf, ctanh, ctanhf}
+
+exp        :: proc{libc_exp, libc_expf, cexp, cexpf}
+exp2       :: proc{libc_exp2, libc_exp2f} 
+expm1      :: proc{libc_expm1, libc_expm1f}
+frexp      :: proc{libc_frexp, libc_frexpf}
+ilogb      :: proc{libc_ilogb, libc_ilogbf}
+ldexp      :: proc{libc_ldexp, libc_ldexpf}
+log        :: proc{libc_log, libc_logf, clog, clogf}
+log10      :: proc{libc_log10, libc_log10f}
+log1p      :: proc{libc_log1p, libc_log1pf}
+log2       :: proc{libc_log2, libc_log2f}
+logb       :: proc{libc_logb, libc_logbf}
+modf       :: proc{libc_modf, libc_modff}
+scalbn     :: proc{libc_scalbn, libc_scalbnf}
+scalbln    :: proc{libc_scalbln, libc_scalblnf}
+
+cbrt       :: proc{libc_cbrt, libc_cbrtf}
+fabs       :: proc{libc_fabs, libc_fabsf, cabs, cabsf}
+hypot      :: proc{libc_hypot, libc_hypotf}
+pow        :: proc{libc_pow, libc_powf, cpow, cpowf}
+sqrt       :: proc{libc_sqrt, libc_sqrtf, csqrt, csqrtf}
+
+erf        :: proc{libc_erf, libc_erff}
+erfc       :: proc{libc_erfc, libc_erfcf}
+lgamma     :: proc{libc_lgamma, libc_lgammaf}
+tgamma     :: proc{libc_tgamma, libc_tgammaf}
+
+ceil       :: proc{libc_ceil, libc_ceilf}
+floor      :: proc{libc_floor, libc_floorf}
+nearbyint  :: proc{libc_nearbyint, libc_nearbyintf}
+rint       :: proc{libc_rint, libc_rintf}
+lrint      :: proc{libc_lrint, libc_lrintf}
+llrint     :: proc{libc_llrint, libc_llrintf}
+round      :: proc{libc_round, libc_roundf}
+lround     :: proc{libc_lround, libc_lroundf}
+llround    :: proc{libc_llround, libc_llroundf}
+trunc      :: proc{libc_trunc, libc_truncf}
+
+fmod       :: proc{libc_fmod, libc_fmodf}
+remainder  :: proc{libc_remainder, libc_remainderf}
+remquo     :: proc{libc_remquo, libc_remquof}
+
+copysign   :: proc{libc_copysign, libc_copysignf}
+nextafter  :: proc{libc_nextafter, libc_nextafterf}
+
+fdim       :: proc{libc_fdim, libc_fdimf}
+fmax       :: proc{libc_fmax, libc_fmaxf}
+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 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}
+asinf      :: proc{libc_asinf}
+atanf      :: proc{libc_atanf}
+atan2f     :: proc{libc_atan2f}
+cosf       :: proc{libc_cosf}
+sinf       :: proc{libc_sinf}
+tanf       :: proc{libc_tanf}
+
+acoshf     :: proc{libc_acoshf}
+asinhf     :: proc{libc_asinhf}
+atanhf     :: proc{libc_atanhf}
+coshf      :: proc{libc_coshf}
+sinhf      :: proc{libc_sinhf}
+tanhf      :: proc{libc_tanhf}
+
+expf       :: proc{libc_expf}
+exp2f      :: proc{libc_exp2f}
+expm1f     :: proc{libc_expm1f}
+frexpf     :: proc{libc_frexpf}
+ilogbf     :: proc{libc_ilogbf}
+ldexpf     :: proc{libc_ldexpf}
+logf       :: proc{libc_logf}
+log10f     :: proc{libc_log10f}
+log1pf     :: proc{libc_log1pf}
+log2f      :: proc{libc_log2f}
+logbf      :: proc{libc_logbf}
+modff      :: proc{libc_modff}
+scalbnf    :: proc{libc_scalbnf}
+scalblnf   :: proc{libc_scalblnf}
+
+cbrtf      :: proc{libc_cbrtf}
+fabsf      :: proc{libc_fabsf}
+hypotf     :: proc{libc_hypotf}
+powf       :: proc{libc_powf}
+sqrtf      :: proc{libc_sqrtf}
+
+erff       :: proc{libc_erff}
+erfcf      :: proc{libc_erfcf}
+lgammaf    :: proc{libc_lgammaf}
+tgammaf    :: proc{libc_tgammaf}
+
+ceilf      :: proc{libc_ceilf}
+floorf     :: proc{libc_floorf}
+nearbyintf :: proc{libc_nearbyintf}
+rintf      :: proc{libc_rintf}
+lrintf     :: proc{libc_lrintf}
+llrintf    :: proc{libc_llrintf}
+roundf     :: proc{libc_roundf}
+lroundf    :: proc{libc_lroundf}
+llroundf   :: proc{libc_llroundf}
+truncf     :: proc{libc_truncf}
+
+fmodf      :: proc{libc_fmodf}
+remainderf :: proc{libc_remainderf}
+remquof    :: proc{libc_remquof}
+
+copysignf  :: proc{libc_copysignf}
+nextafterf :: proc{libc_nextafterf}
+
+fdimf      :: proc{libc_fdimf} 
+fmaxf      :: proc{libc_fmaxf}
+fminf      :: proc{libc_fminf}
+fmaf       :: proc{libc_fmaf}
+
+// These two functions are special and not made type generic in tgmath.h since
+// they only differ by their return type.
+nan        :: proc{libc_nan}
+nanf       :: proc{libc_nanf}
@@ -0,0 +1,66 @@
+package libc
+
+// 7.13 Nonlocal jumps
+
+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"
+}
+when ODIN_OS == .Windows {
+	@(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.
+		//
+		/// NOTE(dweiler): UCRT has two implementations of longjmp. One that performs
+		// stack unwinding and one that doesn't. The choice of which to use depends on a
+		// flag which is set inside the jmp_buf structure given to setjmp. The default
+		// behavior is to unwind the stack. Within Odin, we cannot use the stack
+		// unwinding version as the unwinding information isn't present. To opt-in to
+		// the regular non-unwinding version we need a way to set this flag. Since the
+		// location of the flag within the struct is not defined or part of the ABI and
+		// can change between versions of UCRT, we must rely on setjmp to set it. It
+		// turns out that setjmp receives this flag in the RDX register on Win64, this
+		// just so happens to coincide with the second argument of a function in the
+		// Win64 ABI. By giving our setjmp a second argument with the value of zero,
+		// 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 ---
+	}
+} 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 ---
+	}
+}
+
+@(default_calling_convention="c")
+foreign libc {
+	// 7.13.2 Restore calling environment
+	longjmp :: proc(env: ^jmp_buf, val: int) -> ! ---
+}
+
+// The C99 Rationale describes jmp_buf as being an array type for backward
+// compatibility. Odin does not need to honor this and couldn't as arrays in
+// Odin don't decay to pointers. It is somewhat easy for us to bind this, we
+// just need to ensure the structure contains enough storage with appropriate
+// alignment. 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.
+jmp_buf :: struct #align(16) { _: [4096]char, }
@@ -0,0 +1,61 @@
+package libc
+
+// 7.14 Signal handling
+
+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"
+}
+
+sig_atomic_t :: distinct atomic_int
+
+SIG_ATOMIC_MIN :: min(sig_atomic_t)
+SIG_ATOMIC_MAX :: max(sig_atomic_t)
+
+@(default_calling_convention="c")
+foreign libc {
+	signal :: proc(sig: int, func: proc "c" (int)) -> proc "c" (int) ---
+	raise  :: proc(sig: int) -> int ---
+}
+
+when ODIN_OS == .Windows {
+	SIG_ERR :: rawptr(~uintptr(0)) 
+	SIG_DFL :: rawptr(uintptr(0))
+	SIG_IGN :: rawptr(uintptr(1))
+
+	SIGABRT :: 22
+	SIGFPE  :: 8
+	SIGILL  :: 4
+	SIGINT  :: 2
+	SIGSEGV :: 11
+	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 {
+	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
+}
@@ -0,0 +1,41 @@
+package libc
+
+// 7.16 Variable arguments
+
+import "base:intrinsics"
+
+@(private="file")
+@(default_calling_convention="none")
+foreign _ {
+	@(link_name="llvm.va_start") _va_start :: proc(arglist: ^i8) ---
+	@(link_name="llvm.va_end")   _va_end   :: proc(arglist: ^i8) ---
+	@(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_start :: #force_inline proc(ap: ^va_list, _: any) {
+	_va_start(cast(^i8)ap)
+}
+
+va_end :: #force_inline proc(ap: ^va_list) {
+	_va_end(cast(^i8)ap)
+}
+
+va_copy :: #force_inline proc(dst, src: ^va_list) {
+	_va_copy(cast(^i8)dst, cast(^i8)src)
+}
+
+// We cannot provide va_arg as there is no way to create "C" style procedures
+// in Odin which take variable arguments the C way. The #c_vararg attribute only
+// exists for foreign imports. That being said, being able to copy a va_list,
+// as well as start and end one is necessary in some functions, the va_list
+// taking functions in libc as an example.
@@ -0,0 +1,381 @@
+package libc
+
+// 7.17 Atomics
+
+import "base:intrinsics"
+
+ATOMIC_BOOL_LOCK_FREE     :: true
+ATOMIC_CHAR_LOCK_FREE     :: true
+ATOMIC_CHAR16_T_LOCK_FREE :: true
+ATOMIC_CHAR32_T_LOCK_FREE :: true
+ATOMIC_WCHAR_T_LOCK_FREE  :: true
+ATOMIC_SHORT_LOCK_FREE    :: true
+ATOMIC_INT_LOCK_FREE      :: true
+ATOMIC_LONG_LOCK_FREE     :: true
+ATOMIC_LLONG_LOCK_FREE    :: true
+ATOMIC_POINTER_LOCK_FREE  :: true
+
+// 7.17.3 Order and consistency
+memory_order :: enum int {
+	relaxed,
+	consume,
+	acquire,
+	release,
+	acq_rel,
+	seq_cst,
+}
+
+memory_order_relaxed :: memory_order.relaxed
+memory_order_consume :: memory_order.consume
+memory_order_acquire :: memory_order.acquire
+memory_order_release :: memory_order.release
+memory_order_acq_rel :: memory_order.acq_rel
+memory_order_seq_cst :: memory_order.seq_cst
+
+// 7.17.2 Initialization
+ATOMIC_VAR_INIT :: #force_inline proc(value: $T) -> T {
+	return value
+}
+
+atomic_init :: #force_inline proc(obj: ^$T, value: T) {
+	intrinsics.atomic_store(obj, value)
+}
+
+kill_dependency :: #force_inline proc(value: $T) -> T {
+	return value
+}
+
+// 7.17.4 Fences
+atomic_thread_fence :: #force_inline proc(order: memory_order) {
+	assert(order != .relaxed)
+	assert(order != .consume)
+	#partial switch order {
+	case .acquire: intrinsics.atomic_thread_fence(.Acquire)
+	case .release: intrinsics.atomic_thread_fence(.Release)
+	case .acq_rel: intrinsics.atomic_thread_fence(.Acq_Rel)
+	case .seq_cst: intrinsics.atomic_thread_fence(.Seq_Cst)
+	}
+}
+
+atomic_signal_fence :: #force_inline proc(order: memory_order) {
+	assert(order != .relaxed)
+	assert(order != .consume)
+	#partial switch order {
+	case .acquire: intrinsics.atomic_signal_fence(.Acquire)
+	case .release: intrinsics.atomic_signal_fence(.Release)
+	case .acq_rel: intrinsics.atomic_signal_fence(.Acq_Rel)
+	case .seq_cst: intrinsics.atomic_signal_fence(.Seq_Cst)
+	}
+}
+
+// 7.17.5 Lock-free property
+atomic_is_lock_free :: #force_inline proc(obj: ^$T) -> bool {
+	return intrinsics.atomic_type_is_lock_free(T)
+}
+
+// 7.17.6 Atomic integer types
+atomic_bool           :: distinct bool
+atomic_char           :: distinct char
+atomic_schar          :: distinct char
+atomic_uchar          :: distinct uchar
+atomic_short          :: distinct short
+atomic_ushort         :: distinct ushort
+atomic_int            :: distinct int
+atomic_uint           :: distinct uint
+atomic_long           :: distinct long
+atomic_ulong          :: distinct ulong
+atomic_llong          :: distinct longlong
+atomic_ullong         :: distinct ulonglong
+atomic_char16_t       :: distinct char16_t
+atomic_char32_t       :: distinct char32_t
+atomic_wchar_t        :: distinct wchar_t
+atomic_int_least8_t   :: distinct int_least8_t
+atomic_uint_least8_t  :: distinct uint_least8_t
+atomic_int_least16_t  :: distinct int_least16_t
+atomic_uint_least16_t :: distinct uint_least16_t
+atomic_int_least32_t  :: distinct int_least32_t
+atomic_uint_least32_t :: distinct uint_least32_t
+atomic_int_least64_t  :: distinct int_least64_t
+atomic_uint_least64_t :: distinct uint_least64_t
+atomic_int_fast8_t    :: distinct int_fast8_t
+atomic_uint_fast8_t   :: distinct uint_fast8_t
+atomic_int_fast16_t   :: distinct int_fast16_t
+atomic_uint_fast16_t  :: distinct uint_fast16_t
+atomic_int_fast32_t   :: distinct int_fast32_t
+atomic_uint_fast32_t  :: distinct uint_fast32_t
+atomic_int_fast64_t   :: distinct int_fast64_t
+atomic_uint_fast64_t  :: distinct uint_fast64_t
+atomic_intptr_t       :: distinct intptr_t
+atomic_uintptr_t      :: distinct uintptr_t
+atomic_size_t         :: distinct size_t
+atomic_ptrdiff_t      :: distinct ptrdiff_t
+atomic_intmax_t       :: distinct intmax_t
+atomic_uintmax_t      :: distinct uintmax_t
+
+// 7.17.7 Operations on atomic types
+atomic_store :: #force_inline proc(object: ^$T, desired: T) {
+	intrinsics.atomic_store(object, desired)
+}
+
+atomic_store_explicit :: #force_inline proc(object: ^$T, desired: T, order: memory_order) {
+	assert(order != .consume)
+	assert(order != .acquire)
+	assert(order != .acq_rel)
+
+	#partial switch order {
+	case .relaxed: intrinsics.atomic_store_explicit(object, desired, .Relaxed)
+	case .release: intrinsics.atomic_store_explicit(object, desired, .Release)
+	case .seq_cst: intrinsics.atomic_store_explicit(object, desired, .Seq_Cst)
+	}
+}
+
+atomic_load :: #force_inline proc(object: ^$T) -> T {
+	return intrinsics.atomic_load(object)
+}
+
+atomic_load_explicit :: #force_inline proc(object: ^$T, order: memory_order) {
+	assert(order != .release)
+	assert(order != .acq_rel)
+
+	#partial switch order {
+	case .relaxed: return intrinsics.atomic_load_explicit(object, .Relaxed)
+	case .consume: return intrinsics.atomic_load_explicit(object, .Consume)
+	case .acquire: return intrinsics.atomic_load_explicit(object, .Acquire)
+	case .seq_cst: return intrinsics.atomic_load_explicit(object, .Seq_Cst)
+	}
+}
+
+atomic_exchange :: #force_inline proc(object: ^$T, desired: T) -> T {
+	return intrinsics.atomic_exchange(object, desired)
+}
+
+atomic_exchange_explicit :: #force_inline proc(object: ^$T, desired: T, order: memory_order) -> T {
+	switch order {
+	case .relaxed: return intrinsics.atomic_exchange_explicit(object, desired, .Relaxed)
+	case .consume: return intrinsics.atomic_exchange_explicit(object, desired, .Consume)
+	case .acquire: return intrinsics.atomic_exchange_explicit(object, desired, .Acquire)
+	case .release: return intrinsics.atomic_exchange_explicit(object, desired, .Release)
+	case .acq_rel: return intrinsics.atomic_exchange_explicit(object, desired, .Acq_Rel)
+	case .seq_cst: return intrinsics.atomic_exchange_explicit(object, desired, .Seq_Cst)
+	}
+	return false
+}
+
+// C does not allow failure memory order to be order_release or acq_rel.
+// Similarly, it does not allow the failure order to be stronger than success
+// order. Since consume and acquire are both monotonic, we can count them as
+// one, for a total of three memory orders that are relevant in compare exchange.
+// 	relaxed, acquire (consume), seq_cst.
+// The requirement that the failure order cannot be stronger than success limits
+// the valid combinations for the failure order to this table:
+// 	[success = seq_cst, failure = seq_cst] => _
+// 	[success = acquire, failure = seq_cst] => acq
+// 	[success = release, failure = seq_cst] => rel
+// 	[success = acq_rel, failure = seq_cst] => acqrel
+// 	[success = relaxed, failure = relaxed] => relaxed
+// 	[success = seq_cst, failure = relaxed] => failrelaxed
+// 	[success = seq_cst, failure = acquire] => failacq
+// 	[success = acquire, failure = relaxed] => acq_failrelaxed
+// 	[success = acq_rel, failure = relaxed] => acqrel_failrelaxed
+atomic_compare_exchange_strong :: #force_inline proc(object, expected: ^$T, desired: T) -> bool {
+	value, ok := intrinsics.atomic_compare_exchange_strong(object, expected^, desired)
+	if !ok { expected^ = value } 
+	return ok
+}
+
+atomic_compare_exchange_strong_explicit :: #force_inline proc(object, expected: ^$T, desired: T, success, failure: memory_order) -> bool {
+	assert(failure != .release)
+	assert(failure != .acq_rel)
+
+	value: T; ok: bool
+	#partial switch failure {
+	case .seq_cst:
+		assert(success != .relaxed)
+		#partial switch success {
+		case .seq_cst:
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Seq_Cst, .Seq_Cst)
+		case .acquire:
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Acquire, .Seq_Cst)
+		case .consume:
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Consume, .Seq_Cst)
+		case .release:
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Release, .Seq_Cst)
+		case .acq_rel:
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Acq_Rel, .Seq_Cst)
+		}
+	case .relaxed:
+		assert(success != .release)
+		#partial switch success {
+		case .relaxed:
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Relaxed, .Relaxed)
+		case .seq_cst:
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Seq_Cst, .Relaxed)
+		case .acquire:
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Acquire, .Relaxed)
+		case .consume:
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Consume, .Relaxed)
+		case .acq_rel:
+			value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Acq_Rel, .Relaxed)
+		}
+	case .consume:
+		assert(success == .seq_cst)
+		value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Seq_Cst, .Consume)
+	case .acquire:
+		assert(success == .seq_cst)
+		value, ok = intrinsics.atomic_compare_exchange_strong_explicit(object, expected^, desired, .Seq_Cst, .Acquire)
+
+	}
+	if !ok { expected^ = value }
+	return ok
+}
+
+atomic_compare_exchange_weak :: #force_inline proc(object, expected: ^$T, desired: T) -> bool {
+	value, ok := intrinsics.atomic_compare_exchange_weak(object, expected^, desired)
+	if !ok { expected^ = value }
+	return ok
+}
+
+atomic_compare_exchange_weak_explicit :: #force_inline proc(object, expected: ^$T, desited: T, success, failure: memory_order) -> bool {
+	assert(failure != .release)
+	assert(failure != .acq_rel)
+
+	value: T; ok: bool
+	#partial switch failure {
+	case .seq_cst:
+		assert(success != .relaxed)
+		#partial switch success {
+		case .seq_cst:
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Seq_Cst, .Seq_Cst)
+		case .acquire:
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Acquire, .Seq_Cst)
+		case .consume:
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Consume, .Seq_Cst)
+		case .release:
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Release, .Seq_Cst)
+		case .acq_rel:
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Acq_Rel, .Seq_Cst)
+		}
+	case .relaxed:
+		assert(success != .release)
+		#partial switch success {
+		case .relaxed:
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Relaxed, .Relaxed)
+		case .seq_cst:
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Seq_Cst, .Relaxed)
+		case .acquire:
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Acquire, .Relaxed)
+		case .consume:
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Consume, .Relaxed)
+		case .acq_rel:
+			value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Acq_Rel, .Relaxed)
+		}
+	case .consume:
+		assert(success == .seq_cst)
+		value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Seq_Cst, .Consume)
+	case .acquire:
+		assert(success == .seq_cst)
+		value, ok = intrinsics.atomic_compare_exchange_weak_explicit(object, expected^, desired, .Seq_Cst, .Acquire)
+
+	}
+	if !ok { expected^ = value }
+	return ok
+}
+
+// 7.17.7.5 The atomic_fetch and modify generic functions
+atomic_fetch_add :: #force_inline proc(object: ^$T, operand: T) -> T {
+	return intrinsics.atomic_add(object, operand)
+}
+
+atomic_fetch_add_explicit :: #force_inline proc(object: ^$T, operand: T, order: memory_order) -> T {
+	switch order {
+	case .relaxed: return intrinsics.atomic_add_explicit(object, operand, .Relaxed)
+	case .consume: return intrinsics.atomic_add_explicit(object, operand, .Consume)
+	case .acquire: return intrinsics.atomic_add_explicit(object, operand, .Acquire)
+	case .release: return intrinsics.atomic_add_explicit(object, operand, .Release)
+	case .acq_rel: return intrinsics.atomic_add_explicit(object, operand, .Acq_Rel)
+	case: fallthrough
+	case .seq_cst: return intrinsics.atomic_add_explicit(object, operand, .Seq_Cst)
+	}
+}
+
+atomic_fetch_sub :: #force_inline proc(object: ^$T, operand: T) -> T {
+	return intrinsics.atomic_sub(object, operand)
+}
+
+atomic_fetch_sub_explicit :: #force_inline proc(object: ^$T, operand: T, order: memory_order) -> T {
+	switch order {
+	case .relaxed: return intrinsics.atomic_sub_explicit(object, operand, .Relaxed)
+	case .consume: return intrinsics.atomic_sub_explicit(object, operand, .Consume)
+	case .acquire: return intrinsics.atomic_sub_explicit(object, operand, .Acquire)
+	case .release: return intrinsics.atomic_sub_explicit(object, operand, .Release)
+	case .acq_rel: return intrinsics.atomic_sub_explicit(object, operand, .Acq_Rel)
+	case: fallthrough
+	case .seq_cst: return intrinsics.atomic_sub_explicit(object, operand, .Seq_Cst)
+	}
+}
+
+atomic_fetch_or :: #force_inline proc(object: ^$T, operand: T) -> T {
+	return intrinsics.atomic_or(object, operand)
+}
+
+atomic_fetch_or_explicit :: #force_inline proc(object: ^$T, operand: T, order: memory_order) -> T {
+	switch order {
+	case .relaxed: return intrinsics.atomic_or_explicit(object, operand, .Relaxed)
+	case .consume: return intrinsics.atomic_or_explicit(object, operand, .Consume)
+	case .acquire: return intrinsics.atomic_or_explicit(object, operand, .Acquire)
+	case .release: return intrinsics.atomic_or_explicit(object, operand, .Release)
+	case .acq_rel: return intrinsics.atomic_or_explicit(object, operand, .Acq_Rel)
+	case: fallthrough
+	case .seq_cst: return intrinsics.atomic_or_explicit(object, operand, .Seq_Cst)
+	}
+}
+
+atomic_fetch_xor :: #force_inline proc(object: ^$T, operand: T) -> T {
+	return intrinsics.atomic_xor(object, operand)
+}
+
+atomic_fetch_xor_explicit :: #force_inline proc(object: ^$T, operand: T, order: memory_order) -> T {
+	switch order {
+	case .relaxed: return intrinsics.atomic_xor_explicit(object, operand, .Relaxed)
+	case .consume: return intrinsics.atomic_xor_explicit(object, operand, .Consume)
+	case .acquire: return intrinsics.atomic_xor_explicit(object, operand, .Acquire)
+	case .release: return intrinsics.atomic_xor_explicit(object, operand, .Release)
+	case .acq_rel: return intrinsics.atomic_xor_explicit(object, operand, .Acq_Rel)
+	case: fallthrough
+	case .seq_cst: return intrinsics.atomic_xor_explicit(object, operand, .Seq_Cst)
+	}
+}
+
+atomic_fetch_and :: #force_inline proc(object: ^$T, operand: T) -> T {
+	return intrinsics.atomic_and(object, operand)
+}
+atomic_fetch_and_explicit :: #force_inline proc(object: ^$T, operand: T, order: memory_order) -> T {
+	switch order {
+	case .relaxed: return intrinsics.atomic_and_explicit(object, operand, .Relaxed)
+	case .consume: return intrinsics.atomic_and_explicit(object, operand, .Consume)
+	case .acquire: return intrinsics.atomic_and_explicit(object, operand, .Acquire)
+	case .release: return intrinsics.atomic_and_explicit(object, operand, .Release)
+	case .acq_rel: return intrinsics.atomic_and_explicit(object, operand, .Acq_Rel)
+	case: fallthrough
+	case .seq_cst: return intrinsics.atomic_and_explicit(object, operand, .Seq_Cst)
+	}
+}
+
+// 7.17.8 Atomic flag type and operations
+atomic_flag :: distinct atomic_bool
+
+atomic_flag_test_and_set :: #force_inline proc(flag: ^atomic_flag) -> bool {
+	return bool(atomic_exchange(flag, atomic_flag(true)))
+}
+
+atomic_flag_test_and_set_explicit :: #force_inline proc(flag: ^atomic_flag, order: memory_order) -> bool {
+	return bool(atomic_exchange_explicit(flag, atomic_flag(true), order))
+}
+
+atomic_flag_clear :: #force_inline proc(flag: ^atomic_flag) {
+	atomic_store(flag, atomic_flag(false))
+}
+
+atomic_flag_clear_explicit :: #force_inline proc(flag: ^atomic_flag, order: memory_order) {
+	atomic_store_explicit(flag, atomic_flag(false), order)
+}
@@ -0,0 +1,321 @@
+package libc
+
+import "core:io"
+
+when ODIN_OS == .Windows {
+	foreign import libc {
+		"system:libucrt.lib",
+		"system:legacy_stdio_definitions.lib",
+	}
+} else when ODIN_OS == .Darwin {
+	foreign import libc "system:System.framework"
+} else {
+	foreign import libc "system:c"
+}
+
+// 7.21 Input/output
+
+FILE :: struct {}
+
+// MSVCRT compatible.
+when ODIN_OS == .Windows {
+	_IOFBF       :: 0x0000
+	_IONBF       :: 0x0004
+	_IOLBF       :: 0x0040
+
+	BUFSIZ       :: 512
+
+	EOF          :: int(-1)
+
+	FOPEN_MAX    :: 20
+
+	FILENAME_MAX :: 260
+
+	L_tmpnam     :: 15 // "\\" + 12 + NUL
+
+	SEEK_SET     :: 0
+	SEEK_CUR     :: 1
+	SEEK_END     :: 2
+
+	TMP_MAX      :: 32767 // SHRT_MAX
+
+	fpos_t       :: distinct i64
+
+	@(private="file")
+	@(default_calling_convention="c")
+	foreign libc {
+		__acrt_iob_func :: proc (index: uint) -> ^FILE ---
+	}
+
+	stdin  := __acrt_iob_func(0)
+	stdout := __acrt_iob_func(1)
+	stderr := __acrt_iob_func(2)
+}
+
+// GLIBC and MUSL compatible.
+when ODIN_OS == .Linux {
+	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
+
+	foreign libc {
+		stderr: ^FILE
+		stdin:  ^FILE
+		stdout: ^FILE
+	}
+}
+
+when ODIN_OS == .OpenBSD {
+	fpos_t :: distinct i64
+
+	_IOFBF :: 0
+	_IOLBF :: 1
+	_IONBF :: 1
+
+	BUFSIZ :: 1024
+
+	EOF :: int(-1)
+
+	FOPEN_MAX	:: 20
+	FILENAME_MAX	:: 1024
+
+	SEEK_SET :: 0
+	SEEK_CUR :: 1
+	SEEK_END :: 2
+
+	foreign libc {
+		stderr: ^FILE
+		stdin:  ^FILE
+		stdout: ^FILE
+	}
+}
+
+when ODIN_OS == .FreeBSD {
+	fpos_t :: distinct i64
+
+	_IOFBF :: 0
+	_IOLBF :: 1
+	_IONBF :: 1
+
+	BUFSIZ :: 1024
+
+	EOF :: int(-1)
+
+	FOPEN_MAX	:: 20
+	FILENAME_MAX	:: 1024
+
+	SEEK_SET :: 0
+	SEEK_CUR :: 1
+	SEEK_END :: 2
+
+	foreign libc {
+		stderr: ^FILE
+		stdin:  ^FILE
+		stdout: ^FILE
+	}
+}
+
+when ODIN_OS == .Darwin {
+	fpos_t :: distinct i64
+	
+	_IOFBF        :: 0
+	_IOLBF        :: 1
+	_IONBF        :: 2
+
+	BUFSIZ        :: 1024
+
+	EOF           :: int(-1)
+
+	FOPEN_MAX     :: 20
+
+	FILENAME_MAX  :: 1024
+
+	L_tmpnam      :: 1024
+
+	SEEK_SET      :: 0
+	SEEK_CUR      :: 1
+	SEEK_END      :: 2
+
+	TMP_MAX       :: 308915776
+
+	foreign libc {
+		@(link_name="__stderrp") stderr: ^FILE
+		@(link_name="__stdinp")  stdin:  ^FILE
+		@(link_name="__stdoutp") stdout: ^FILE
+	}
+}
+
+@(default_calling_convention="c")
+foreign libc {
+	// 7.21.4 Operations on files
+	remove    :: proc(filename: cstring) -> int ---
+	rename    :: proc(old, new: cstring) -> int ---
+	tmpfile   :: proc() -> ^FILE ---
+	tmpnam    :: proc(s: [^]char) -> [^]char ---
+
+	// 7.21.5 File access functions
+	fclose    :: proc(stream: ^FILE) -> int ---
+	fflush    :: proc(stream: ^FILE) -> int ---
+	fopen     :: proc(filename, mode: cstring) -> ^FILE ---
+	freopen   :: proc(filename, mode: cstring, stream: ^FILE) -> ^FILE ---
+	setbuf    :: proc(stream: ^FILE, buf: [^]char) ---
+	setvbuf   :: proc(stream: ^FILE, buf: [^]char, mode: int, size: size_t) -> int ---
+
+	// 7.21.6 Formatted input/output functions
+	fprintf   :: proc(stream: ^FILE, format: cstring, #c_vararg args: ..any) -> int ---
+	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, 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 ---
+	vprintf   :: proc(format: cstring, arg: ^va_list) -> int ---
+	vscanf    :: proc(format: cstring, arg: ^va_list) -> int ---
+	vsnprintf :: proc(s: [^]char, n: size_t, format: cstring, arg: ^va_list) -> int ---
+	vsprintf  :: proc(s: [^]char, format: cstring, arg: ^va_list) -> int ---
+	vsscanf   :: proc(s, format: cstring, arg: ^va_list) -> int ---
+
+	// 7.21.7 Character input/output functions
+	fgetc     :: proc(stream: ^FILE) -> int ---
+	fgets     :: proc(s: [^]char, n: int, stream: ^FILE) -> [^]char ---
+	fputc     :: proc(s: cstring, stream: ^FILE) -> int ---
+	getc      :: proc(stream: ^FILE) -> int ---
+	getchar   :: proc() -> int ---
+	putc      :: proc(c: int, stream: ^FILE) -> 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
+	fgetpos   :: proc(stream: ^FILE, pos: ^fpos_t) -> int ---
+	fseek     :: proc(stream: ^FILE, offset: long, whence: int) -> int ---
+	fsetpos   :: proc(stream: ^FILE, pos: ^fpos_t) -> int ---
+	ftell     :: proc(stream: ^FILE) -> long ---
+	rewind    :: proc(stream: ^FILE) ---
+
+	// 7.21.10 Error-handling functions
+	clearerr  :: proc(stream: ^FILE) ---
+	feof      :: proc(stream: ^FILE) -> int ---
+	ferror    :: proc(stream: ^FILE) -> int ---
+	perror    :: proc(s: cstring) ---
+}
+
+to_stream :: proc(file: ^FILE) -> io.Stream {
+	stream_proc :: proc(stream_data: rawptr, mode: io.Stream_Mode, p: []byte, offset: i64, whence: io.Seek_From) -> (n: i64, err: io.Error) {
+		unknown_or_eof :: proc(f: ^FILE) -> io.Error {
+			switch {
+			case ferror(f) != 0:
+				return .Unknown
+			case feof(f) != 0:
+				return .EOF
+			case:
+				return nil
+			}
+		}
+
+		file := (^FILE)(stream_data)
+		switch mode {
+		case .Close:
+			if fclose(file) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+		case .Flush:
+			if fflush(file) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+		case .Read:
+			n = i64(fread(raw_data(p), size_of(byte), len(p), file))
+			if n == 0 { err = unknown_or_eof(file) }
+
+		case .Read_At:
+			curr := ftell(file)
+			if curr == -1 {
+				return 0, unknown_or_eof(file)
+			}
+
+			if fseek(file, long(offset), SEEK_SET) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+			defer fseek(file, long(curr), SEEK_SET)
+
+			n = i64(fread(raw_data(p), size_of(byte), len(p), file))
+			if n == 0 { err = unknown_or_eof(file) }
+		
+		case .Write:
+			n = i64(fwrite(raw_data(p), size_of(byte), len(p), file))
+			if n == 0 { err = unknown_or_eof(file) }
+
+		case .Write_At:
+			curr := ftell(file)
+			if curr == -1 {
+				return 0, unknown_or_eof(file)
+			}
+
+			if fseek(file, long(offset), SEEK_SET) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+			defer fseek(file, long(curr), SEEK_SET)
+
+			n = i64(fwrite(raw_data(p), size_of(byte), len(p), file))
+			if n == 0 { err = unknown_or_eof(file) }
+
+		case .Seek:
+			if fseek(file, long(offset), int(whence)) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+		
+		case .Size:
+			curr := ftell(file)
+			if curr == -1 {
+				return 0, unknown_or_eof(file)
+			}
+			defer fseek(file, curr, SEEK_SET)
+
+			if fseek(file, 0, SEEK_END) != 0 {
+				return 0, unknown_or_eof(file)
+			}
+
+			n = i64(ftell(file))
+			if n == -1 {
+				return 0, unknown_or_eof(file)
+			}
+
+		case .Destroy:
+			return 0, .Empty
+		
+		case .Query:
+			return io.query_utility({ .Close, .Flush, .Read, .Read_At, .Write, .Write_At, .Seek, .Size })
+		}
+		return
+	}
+
+	return {
+		data      = file,
+		procedure = stream_proc,
+	}
+}
@@ -0,0 +1,153 @@
+package libc
+
+// 7.22 General utilities
+
+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"
+}
+
+when ODIN_OS == .Windows {
+	RAND_MAX :: 0x7fff
+
+	@(private="file")
+	@(default_calling_convention="c")
+	foreign libc {
+		___mb_cur_max_func :: proc() -> int ---
+	}
+
+	MB_CUR_MAX :: #force_inline proc() -> size_t {
+		return size_t(___mb_cur_max_func())
+	}
+}
+
+when ODIN_OS == .Linux {
+	RAND_MAX :: 0x7fffffff
+
+	// GLIBC and MUSL only
+	@(private="file")
+	@(default_calling_convention="c")
+	foreign libc {
+		__ctype_get_mb_cur_max :: proc() -> size_t ---
+	}
+
+	MB_CUR_MAX :: #force_inline proc() -> size_t {
+		return size_t(__ctype_get_mb_cur_max())
+	}
+}
+
+
+when ODIN_OS == .Darwin {
+	RAND_MAX :: 0x7fffffff
+
+	// GLIBC and MUSL only
+	@(private="file")
+	@(default_calling_convention="c")
+	foreign libc {
+		___mb_cur_max :: proc() -> int ---
+	}
+
+	MB_CUR_MAX :: #force_inline proc() -> size_t {
+		return size_t(___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
+// values.
+EXIT_SUCCESS :: 0
+EXIT_FAILURE :: 1
+
+// C does not declare which order 'quot' and 'rem' should be for the divide
+// structures. An implementation could put 'rem' first. However, nobody actually
+// does and everyone appears to have agreed upon this layout.
+div_t   :: struct { quot, rem: int, }
+ldiv_t  :: struct { quot, rem: long, }
+lldiv_t :: struct { quot, rem: longlong, }
+
+@(default_calling_convention="c")
+foreign libc {
+	// 7.22.1 Numeric conversion functions
+	atof          :: proc(nptr: cstring) -> double ---
+	atoi          :: proc(nptr: cstring) -> int ---
+	atol          :: proc(nptr: cstring) -> long ---
+	atoll         :: proc(nptr: cstring) -> longlong ---
+	strtod        :: proc(nptr: cstring, endptr: ^[^]char) -> double ---
+	strtof        :: proc(nptr: cstring, endptr: ^[^]char) -> float ---
+	strtol        :: proc(nptr: cstring, endptr: ^[^]char, base: int) -> long ---
+	strtoll       :: proc(nptr: cstring, endptr: ^[^]char, base: int) -> longlong ---
+	strtoul       :: proc(nptr: cstring, endptr: ^[^]char, base: int) -> ulong ---
+	strtoull      :: proc(nptr: cstring, endptr: ^[^]char, base: int) -> ulonglong ---
+
+	// 7.22.2 Pseudo-random sequence generation functions
+	rand          :: proc() -> int ---
+	srand         :: proc(seed: uint) ---
+
+	// 7.22.3 Memory management functions
+	calloc        :: proc(nmemb, size: size_t) -> rawptr ---
+	free          :: proc(ptr: rawptr) ---
+	malloc        :: proc(size: size_t) -> rawptr ---
+	realloc       :: proc(ptr: rawptr, size: size_t) -> rawptr ---
+
+	// 7.22.4 Communication with the environment
+	abort         :: proc() -> ! ---
+	atexit        :: proc(func: proc "c" ()) -> int ---
+	at_quick_exit :: proc(func: proc "c" ()) -> int ---
+	exit          :: proc(status: int) -> ! ---
+	_Exit         :: proc(status: int) -> ! ---
+	getenv        :: proc(name: cstring) -> [^]char ---
+	quick_exit    :: proc(status: int) -> ! ---
+	system        :: proc(cmd: cstring) -> int ---
+
+	// 7.22.5 Searching and sorting utilities
+	bsearch       :: proc(key, base: rawptr, nmemb, size: size_t, compar: proc "c" (lhs, rhs: rawptr) -> int) -> rawptr ---
+	qsort         :: proc(base: rawptr, nmemb, size: size_t, compar: proc "c" (lhs, rhs: rawptr) -> int) ---
+
+	// 7.22.6 Integer arithmetic functions
+	abs           :: proc(j: int) -> int ---
+	labs          :: proc(j: long) -> long ---
+	llabs         :: proc(j: longlong) -> longlong ---
+	div           :: proc(numer, denom: int) -> div_t ---
+	ldiv          :: proc(numer, denom: long) -> ldiv_t ---
+	lldiv         :: proc(numer, denom: longlong) -> lldiv_t ---
+
+	// 7.22.7 Multibyte/wide character conversion functions
+	mblen         :: proc(s: cstring, n: size_t) -> int ---
+	mbtowc        :: proc(pwc: ^wchar_t, s: cstring, n: size_t) -> int ---
+	wctomb        :: proc(s: [^]char, wc: wchar_t) -> int ---
+
+	// 7.22.8 Multibyte/wide string conversion functions
+	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 {
+		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 {
+		free(ptr)
+	}
+}
@@ -0,0 +1,48 @@
+package libc
+
+import "base:runtime"
+
+// 7.24 String handling
+
+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"
+}
+
+foreign libc {
+	// 7.24.2 Copying functions
+	memcpy   :: proc(s1, s2: rawptr, n: size_t) -> rawptr ---
+	memmove  :: proc(s1, s2: rawptr, n: size_t) -> rawptr ---
+	strcpy   :: proc(s1: [^]char, s2: cstring) -> [^]char ---
+	strncpy  :: proc(s1: [^]char, s2: cstring, n: size_t) -> [^]char ---
+
+	// 7.24.3 Concatenation functions
+	strcat   :: proc(s1: [^]char, s2: cstring) -> [^]char ---
+	strncat  :: proc(s1: [^]char, s2: cstring, n: size_t) -> [^]char ---
+
+	// 7.24.4 Comparison functions
+	memcmp   :: proc(s1, s2: rawptr, n: size_t) -> int ---
+	strcmp   :: proc(s1, s2: cstring) -> int ---
+	strcoll  :: proc(s1, s2: cstring) -> int ---
+	strncmp  :: proc(s1, s2: cstring, n: size_t) -> int ---
+	strxfrm  :: proc(s1: [^]char, s2: cstring, n: size_t) -> size_t ---
+
+	// 7.24.5 Search functions
+	memchr   :: proc(s: rawptr, c: int, n: size_t) -> rawptr ---
+	strchr   :: proc(s: cstring, c: int) -> [^]char ---
+	strcspn  :: proc(s1, s2: cstring) -> size_t ---
+	strpbrk  :: proc(s1, s2: cstring) -> [^]char ---
+	strrchr  :: proc(s: [^]char, c: int) -> [^]char ---
+	strcpn   :: proc(s1, s2: cstring) -> [^]char ---
+	strtok   :: proc(s1: [^]char, s2: cstring) -> [^]char ---
+
+	// 7.24.6 Miscellaneous functions
+	strerror :: proc(errnum: int) -> [^]char ---
+	strlen   :: proc(s: cstring) -> size_t ---
+}
+memset :: proc "c" (s: rawptr, c: int, n: size_t) -> rawptr {
+	return runtime.memset(s, c, auto_cast n)
+}
@@ -0,0 +1,50 @@
+package libc_tests
+
+import "core:c/libc"
+
+test_stdio :: proc() {
+    c: libc.char = 'C';
+    libc.puts("Hello from puts");
+    libc.printf("Hello from printf in %c\n", c);
+}
+test_thread :: proc() {
+    thread_proc :: proc "c" (rawptr) -> libc.int {
+        libc.printf("Hello from thread");
+        return 42;
+    }
+    thread: libc.thrd_t;
+    libc.thrd_create(&thread, thread_proc, nil);
+    result: libc.int;
+    libc.thrd_join(thread, &result);
+    libc.printf(" %d\n", result);
+}
+
+jmp: libc.jmp_buf;
+test_sjlj :: proc() {
+    if libc.setjmp(&jmp) != 0 {
+        libc.printf("Hello from longjmp\n");
+        return;
+    }
+    libc.printf("Hello from setjmp\n");
+    libc.longjmp(&jmp, 1);
+}
+test_signal :: proc() {
+    handler :: proc "c" (sig: libc.int) {
+        libc.printf("Hello from signal handler\n");
+    }
+    libc.signal(libc.SIGABRT, handler);
+    libc.raise(libc.SIGABRT);
+}
+test_atexit :: proc() {
+    handler :: proc "c" () {
+        libc.printf("Hello from atexit\n");
+    }
+    libc.atexit(handler);
+}
+main :: proc() {
+    test_stdio();
+    test_thread();
+    test_sjlj();
+    test_signal();
+    test_atexit();
+}
@@ -0,0 +1,143 @@
+package libc
+
+// 7.26 Threads
+
+thrd_start_t :: proc "c" (rawptr) -> int
+tss_dtor_t   :: proc "c" (rawptr)
+
+when ODIN_OS == .Windows {
+	foreign import libc {
+		"system:libucrt.lib", 
+		"system:msvcprt.lib",
+	}
+
+	thrd_success        :: 0                             // _Thrd_success
+	thrd_nomem          :: 1                             // _Thrd_nomem
+	thrd_timedout       :: 2                             // _Thrd_timedout
+	thrd_busy           :: 3                             // _Thrd_busy
+	thrd_error          :: 4                             // _Thrd_error
+
+	mtx_plain           :: 1                             // _Mtx_plain
+	mtx_recursive       :: 0x100                         // _Mtx_recursive
+	mtx_timed           :: 4                             // _Mtx_timed
+
+	TSS_DTOR_ITERATIONS :: 4                             // _TSS_DTOR_ITERATIONS_IMP
+
+	once_flag           :: distinct i8                   // _Once_flag_imp_t
+	thrd_t              :: struct { _: rawptr, _: uint, } // _Thrd_t
+	tss_t               :: distinct int                  // _Tss_imp_t
+	cnd_t               :: distinct rawptr               // _Cnd_imp_t
+	mtx_t               :: distinct rawptr               // _Mtx_imp_t
+
+	// MSVCRT does not expose the C11 symbol names as what they are in C11
+	// because they held off implementing <threads.h> and C11 support for so
+	// long that people started implementing their own. To prevent symbol
+	// conflict with existing customers code they had to namespace them
+	// differently. Thus we need to alias the correct symbol names with Odin's
+	// link_name attribute.
+	@(default_calling_convention="c")
+	foreign libc {
+		// 7.26.2 Initialization functions
+		@(link_name="_Call_once")     call_once     :: proc(flag: ^once_flag, func: proc "c" ()) ---
+		// 7.26.3 Condition variable functions
+		@(link_name="_Cnd_broadcast") cnd_broadcast :: proc(cond: ^cnd_t) -> int ---
+		@(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, 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
+		@(link_name="_Mtx_destroy")   mtx_destroy   :: proc(mtx: ^mtx_t) ---
+		@(link_name="_Mtx_init")      mtx_init      :: proc(mtx: ^mtx_t, type: int) -> int ---
+		@(link_name="_Mtx_lock")      mtx_lock      :: proc(mtx: ^mtx_t) -> int ---
+		@(link_name="_Mtx_timedlock") mtx_timedlock :: proc(mtx: ^mtx_t, ts: ^timespec) -> int ---
+		@(link_name="_Mtx_trylock")   mtx_trylock   :: proc(mtx: ^mtx_t) -> int ---
+		@(link_name="_Mtx_unlock")    mtx_unlock    :: proc(mtx: ^mtx_t) -> int ---
+
+		// 7.26.5 Thread functions
+		@(link_name="_Thrd_create")   thrd_create   :: proc(thr: ^thrd_t, func: thrd_start_t, arg: rawptr) -> int ---
+		@(link_name="_Thrd_current")  thrd_current  :: proc() -> thrd_t ---
+		@(link_name="_Thrd_detach")   thrd_detach   :: proc(thr: thrd_t) -> int ---
+		@(link_name="_Thrd_equal")    thrd_equal    :: proc(lhs, rhs: thrd_t) -> int ---
+		@(link_name="_Thrd_exit")     thrd_exit     :: proc(res: int) -> ! ---
+		@(link_name="_Thrd_join")     thrd_join     :: proc(thr: thrd_t, res: ^int) -> int ---
+		@(link_name="_Thrd_sleep")    thrd_sleep    :: proc(duration, remaining: ^timespec) -> int ---
+		@(link_name="_Thrd_yield")    thrd_yield    :: proc() ---
+
+		// 7.26.6 Thread-specific storage functions
+		@(link_name="_Tss_create")    tss_create    :: proc(key: ^tss_t, dtor: tss_dtor_t) -> int ---
+		@(link_name="_Tss_delete")    tss_delete    :: proc(key: tss_t) ---
+		@(link_name="_Tss_get")       tss_get       :: proc(key: tss_t) -> rawptr ---
+		@(link_name="_Tss_set")       tss_set       :: proc(key: tss_t, val: rawptr) -> int ---
+	}
+}
+
+// GLIBC and MUSL compatible constants and types.
+when ODIN_OS == .Linux {
+	foreign import libc {
+		"system:c",
+		"system:pthread",
+	}
+
+	thrd_success        :: 0
+	thrd_busy           :: 1
+	thrd_error          :: 2
+	thrd_nomem          :: 3
+	thrd_timedout       :: 4
+
+	mtx_plain           :: 0
+	mtx_recursive       :: 1
+	mtx_timed           :: 2
+
+	TSS_DTOR_ITERATIONS :: 4
+
+	once_flag           :: distinct int
+	thrd_t              :: distinct ulong
+	tss_t               :: distinct uint
+	cnd_t               :: struct #raw_union { _: [12]int, _: [12 * size_of(int) / size_of(rawptr)]rawptr, }
+	mtx_t               :: struct #raw_union { _: [10 when size_of(long) == 8 else 6]int, _: [5 when size_of(long) == 8 else 6]rawptr, }
+
+	@(default_calling_convention="c")
+	foreign libc {
+		// 7.26.2 Initialization functions
+		call_once     :: proc(flag: ^once_flag, func: proc "c" ()) ---
+
+		// 7.26.3 Condition variable functions
+		cnd_broadcast :: proc(cond: ^cnd_t) -> int ---
+		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, mtx: ^mtx_t, ts: ^timespec) -> int ---
+		cnd_wait      :: proc(cond: ^cnd_t, mtx: ^mtx_t) -> int ---
+		
+		// 7.26.4 Mutex functions
+		mtx_destroy   :: proc(mtx: ^mtx_t) ---
+		mtx_init      :: proc(mtx: ^mtx_t, type: int) -> int ---
+		mtx_lock      :: proc(mtx: ^mtx_t) -> int ---
+		mtx_timedlock :: proc(mtx: ^mtx_t, ts: ^timespec) -> int ---
+		mtx_trylock   :: proc(mtx: ^mtx_t) -> int ---
+		mtx_unlock    :: proc(mtx: ^mtx_t) -> int ---
+
+		// 7.26.5 Thread functions
+		thrd_create   :: proc(thr: ^thrd_t, func: thrd_start_t, arg: rawptr) -> int ---
+		thrd_current  :: proc() -> thrd_t ---
+		thrd_detach   :: proc(thr: thrd_t) -> int ---
+		thrd_equal    :: proc(lhs, rhs: thrd_t) -> int ---
+		thrd_exit     :: proc(res: int) -> ! ---
+		thrd_join     :: proc(thr: thrd_t, res: ^int) -> int ---
+		thrd_sleep    :: proc(duration, remaining: ^timespec) -> int ---
+		thrd_yield    :: proc() ---
+
+		// 7.26.6 Thread-specific storage functions
+		tss_create    :: proc(key: ^tss_t, dtor: tss_dtor_t) -> int ---
+		tss_delete    :: proc(key: tss_t) ---
+		tss_get       :: proc(key: tss_t) -> rawptr ---
+		tss_set       :: proc(key: tss_t, val: rawptr) -> int ---
+	}
+}
+
+
+when ODIN_OS == .Darwin {
+	// TODO: find out what this is meant to be!
+}
@@ -0,0 +1,88 @@
+package libc
+
+// 7.27 Date and time
+
+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"
+}
+
+// We enforce 64-bit time_t and timespec as there is no reason to use 32-bit as
+// we approach the 2038 problem. Windows has defaulted to this since VC8 (2005).
+when ODIN_OS == .Windows {
+	foreign libc {
+		// 7.27.2 Time manipulation functions
+		                               clock        :: proc() -> clock_t ---
+		@(link_name="_difftime64")     difftime     :: proc(time1, time2: time_t) -> double ---
+		                               mktime       :: proc(timeptr: ^tm) -> time_t ---
+		@(link_name="_time64")         time         :: proc(timer: ^time_t) -> time_t ---
+		@(link_name="_timespec64_get") timespec_get :: proc(ts: ^timespec, base: int) -> int ---
+
+		// 7.27.3 Time conversion functions
+		                               asctime      :: proc(timeptr: ^tm) -> [^]char ---
+		@(link_name="_ctime64")        ctime        :: proc(timer: ^time_t) -> [^]char ---
+		@(link_name="_gmtime64")       gmtime       :: proc(timer: ^time_t) -> ^tm ---
+		@(link_name="_localtime64")    localtime    :: proc(timer: ^time_t) -> ^tm ---
+		                               strftime     :: proc(s: [^]char, maxsize: size_t, format: cstring, timeptr: ^tm) -> size_t ---
+	}
+
+	CLOCKS_PER_SEC :: 1000
+	TIME_UTC       :: 1
+
+	clock_t        :: distinct long
+	time_t         :: distinct i64
+
+	timespec :: struct #align(8) {
+		tv_sec:  time_t,
+		tv_nsec: long,
+	}
+
+	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 {
+	@(default_calling_convention="c")
+	foreign libc {
+		// 7.27.2 Time manipulation functions
+		clock        :: proc() -> clock_t ---
+		difftime     :: proc(time1, time2: time_t) -> double ---
+		mktime       :: proc(timeptr: ^tm) -> time_t ---
+		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 ---
+		ctime        :: proc(timer: ^time_t) -> [^]char ---
+		gmtime       :: proc(timer: ^time_t) -> ^tm ---
+		localtime    :: proc(timer: ^time_t) -> ^tm ---
+		strftime     :: proc(s: [^]char, maxsize: size_t, format: cstring, timeptr: ^tm) -> size_t ---
+	}
+
+	when ODIN_OS == .OpenBSD {
+		CLOCKS_PER_SEC :: 100
+	} else {
+		CLOCKS_PER_SEC :: 1000000
+	}
+
+	TIME_UTC       :: 1
+
+	time_t         :: distinct i64
+
+	clock_t        :: long
+
+	timespec :: struct {
+		tv_sec:  time_t,
+		tv_nsec: long,
+	}
+
+	tm :: struct {
+		tm_sec, tm_min, tm_hour, tm_mday, tm_mon, tm_year, tm_wday, tm_yday, tm_isdst: int,
+		tm_gmtoff: long,
+		tm_zone: rawptr,
+	}
+}
--- a/Show More
+++ b/Show More