release: update version to 0.36.0

Signed-off-by: deadprogram <ron@hybridgroup.com>

.circleci/config.yml

@@ -10,12 +10,12 @@ commands:
     steps:
       - restore_cache:
           keys:
-            - llvm-source-14-v3
+            - llvm-source-19-v1
       - run:
           name: "Fetch LLVM source"
           command: make llvm-source
       - save_cache:
-          key: llvm-source-14-v3
+          key: llvm-source-19-v1
           paths:
             - llvm-project/clang/lib/Headers
             - llvm-project/clang/include
@@ -33,13 +33,13 @@ commands:
     steps:
       - restore_cache:
           keys:
-            - binaryen-linux-v2
+            - binaryen-linux-v3
       - run:
           name: "Build Binaryen"
           command: |
             make binaryen
       - save_cache:
-          key: binaryen-linux-v2
+          key: binaryen-linux-v3
           paths:
             - build/wasm-opt
   test-linux:
@@ -55,7 +55,7 @@ commands:
       - run:
           name: "Install apt dependencies"
           command: |
-            echo 'deb https://apt.llvm.org/buster/ llvm-toolchain-buster-<<parameters.llvm>> main' > /etc/apt/sources.list.d/llvm.list
+            echo 'deb https://apt.llvm.org/bullseye/ llvm-toolchain-bullseye-<<parameters.llvm>> main' > /etc/apt/sources.list.d/llvm.list
             wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key | apt-key add -
             apt-get update
             apt-get install --no-install-recommends -y \
@@ -69,16 +69,16 @@ commands:
       - build-binaryen-linux
       - restore_cache:
           keys:
-            - go-cache-v3-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
-            - go-cache-v3-{{ checksum "go.mod" }}
+            - go-cache-v4-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
+            - go-cache-v4-{{ checksum "go.mod" }}
       - llvm-source-linux
       - run: go install -tags=llvm<<parameters.llvm>> .
       - restore_cache:
           keys:
-            - wasi-libc-sysroot-systemclang-v6
+            - wasi-libc-sysroot-systemclang-v7
       - run: make wasi-libc
       - save_cache:
-          key: wasi-libc-sysroot-systemclang-v6
+          key: wasi-libc-sysroot-systemclang-v7
           paths:
             - lib/wasi-libc/sysroot
       - when:
@@ -88,22 +88,33 @@ commands:
               # Do this before gen-device so that it doesn't check the
               # formatting of generated files.
               name: Check Go code formatting
-              command: make fmt-check
+              command: make fmt-check lint
       - run: make gen-device -j4
+      # TODO: change this to -skip='TestErrors|TestWasm' with Go 1.20
+      - run: go test -tags=llvm<<parameters.llvm>> -short -run='TestBuild|TestTest|TestGetList|TestTraceback'
       - run: make smoketest XTENSA=0
       - save_cache:
-          key: go-cache-v3-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
+          key: go-cache-v4-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
           paths:
             - ~/.cache/go-build
             - /go/pkg/mod
 
 jobs:
-  test-llvm14-go118:
+  test-llvm15-go119:
     docker:
-      - image: golang:1.18-buster
+      - image: golang:1.19-bullseye
     steps:
       - test-linux:
-          llvm: "14"
+          llvm: "15"
+          # "make lint" fails before go 1.21 because internal/tools/go.mod specifies packages that require go 1.21
+          fmt-check: false
+    resource_class: large
+  test-llvm19-go124:
+    docker:
+      - image: golang:1.24-bullseye
+    steps:
+      - test-linux:
+          llvm: "19"
     resource_class: large
 
 workflows:
@@ -111,4 +122,6 @@ workflows:
     jobs:
       # This tests our lowest supported versions of Go and LLVM, to make sure at
       # least the smoke tests still pass.
-      - test-llvm14-go118
+      - test-llvm15-go119
+      # This tests LLVM 19 support when linking against system libraries.
+      - test-llvm19-go124

.github/FUNDING.yml

@@ -0,0 +1,3 @@
+# These are supported funding model platforms
+
+open_collective: tinygo

.github/workflows/build-macos.yml

@@ -14,26 +14,38 @@ concurrency:
 jobs:
   build-macos:
     name: build-macos
-    runs-on: macos-11
+    strategy:
+      matrix:
+        # macos-13: amd64 (oldest supported version as of 18-10-2024)
+        # macos-14: arm64 (oldest arm64 version)
+        os: [macos-13, macos-14]
+        include:
+          - os: macos-13
+            goarch: amd64
+          - os: macos-14
+            goarch: arm64
+    runs-on: ${{ matrix.os }}
     steps:
       - name: Install Dependencies
-        shell: bash
         run: |
           HOMEBREW_NO_AUTO_UPDATE=1 brew install qemu binaryen
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
         with:
           submodules: true
+      - name: Extract TinyGo version
+        id: version
+        run: ./.github/workflows/tinygo-extract-version.sh | tee -a "$GITHUB_OUTPUT"
       - name: Install Go
-        uses: actions/setup-go@v3
+        uses: actions/setup-go@v5
         with:
-          go-version: '1.20'
+          go-version: '1.24'
           cache: true
       - name: Restore LLVM source cache
-        uses: actions/cache/restore@v3
+        uses: actions/cache/restore@v4
         id: cache-llvm-source
         with:
-          key: llvm-source-15-macos-v3
+          key: llvm-source-19-${{ matrix.os }}-v1
           path: |
             llvm-project/clang/lib/Headers
             llvm-project/clang/include
@@ -44,7 +56,7 @@ jobs:
         if: steps.cache-llvm-source.outputs.cache-hit != 'true'
         run: make llvm-source
       - name: Save LLVM source cache
-        uses: actions/cache/save@v3
+        uses: actions/cache/save@v4
         if: steps.cache-llvm-source.outputs.cache-hit != 'true'
         with:
           key: ${{ steps.cache-llvm-source.outputs.cache-primary-key }}
@@ -55,14 +67,13 @@ jobs:
             llvm-project/lld/include
             llvm-project/llvm/include
       - name: Restore LLVM build cache
-        uses: actions/cache/restore@v3
+        uses: actions/cache/restore@v4
         id: cache-llvm-build
         with:
-          key: llvm-build-15-macos-v4
+          key: llvm-build-19-${{ matrix.os }}-v1
           path: llvm-build
       - name: Build LLVM
         if: steps.cache-llvm-build.outputs.cache-hit != 'true'
-        shell: bash
         run: |
           # fetch LLVM source
           rm -rf llvm-project
@@ -73,16 +84,16 @@ jobs:
           make llvm-build
           find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
       - name: Save LLVM build cache
-        uses: actions/cache/save@v3
+        uses: actions/cache/save@v4
         if: steps.cache-llvm-build.outputs.cache-hit != 'true'
         with:
           key: ${{ steps.cache-llvm-build.outputs.cache-primary-key }}
           path: llvm-build
       - name: Cache wasi-libc sysroot
-        uses: actions/cache@v3
+        uses: actions/cache@v4
         id: cache-wasi-libc
         with:
-          key: wasi-libc-sysroot-v4
+          key: wasi-libc-sysroot-${{ matrix.os }}-v1
           path: lib/wasi-libc/sysroot
       - name: Build wasi-libc
         if: steps.cache-wasi-libc.outputs.cache-hit != 'true'
@@ -90,15 +101,13 @@ jobs:
       - name: make gen-device
         run: make -j3 gen-device
       - name: Test TinyGo
-        shell: bash
         run: make test GOTESTFLAGS="-short"
       - name: Build TinyGo release tarball
         run: make release -j3
       - name: Test stdlib packages
         run: make tinygo-test
       - name: Make release artifact
-        shell: bash
-        run: cp -p build/release.tar.gz build/tinygo.darwin-amd64.tar.gz 
+        run: cp -p build/release.tar.gz build/tinygo${{ steps.version.outputs.version }}.darwin-${{ matrix.goarch }}.tar.gz
       - name: Publish release artifact
         # Note: this release artifact is double-zipped, see:
         # https://github.com/actions/upload-artifact/issues/39
@@ -106,29 +115,43 @@ jobs:
         # - have a double-zipped artifact when downloaded from the UI
         # - have a very slow artifact upload
         # We're doing the former here, to keep artifact uploads fast.
-        uses: actions/upload-artifact@v2
+        uses: actions/upload-artifact@v4
         with:
-          name: release-double-zipped
-          path: build/tinygo.darwin-amd64.tar.gz
+          name: darwin-${{ matrix.goarch }}-double-zipped-${{ steps.version.outputs.version }}
+          path: build/tinygo${{ steps.version.outputs.version }}.darwin-${{ matrix.goarch }}.tar.gz
       - name: Smoke tests
-        shell: bash
         run: make smoketest TINYGO=$(PWD)/build/tinygo
   test-macos-homebrew:
     name: homebrew-install
     runs-on: macos-latest
+    strategy:
+      matrix:
+        version: [16, 17, 18, 19]
     steps:
-      - name: Install LLVM
-        shell: bash
+      - name: Set up Homebrew
+        uses: Homebrew/actions/setup-homebrew@master
+      - name: Fix Python symlinks
         run: |
-          HOMEBREW_NO_AUTO_UPDATE=1 brew install llvm@15
+          # Github runners have broken symlinks, so relink
+          # see: https://github.com/actions/setup-python/issues/577
+          brew list -1 | grep python | while read formula; do brew unlink $formula; brew link --overwrite $formula; done
+      - name: Install LLVM
+        run: |
+          brew install llvm@${{ matrix.version }}
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
       - name: Install Go
-        uses: actions/setup-go@v3
+        uses: actions/setup-go@v5
         with:
-          go-version: '1.20'
+          go-version: '1.24'
           cache: true
-      - name: Build TinyGo
-        run: go install
+      - name: Build TinyGo (LLVM ${{ matrix.version }})
+        run: go install -tags=llvm${{ matrix.version }}
       - name: Check binary
         run: tinygo version
+      - name: Build TinyGo (default LLVM)
+        if: matrix.version == 19
+        run: go install
+      - name: Check binary
+        if: matrix.version == 19
+        run: tinygo version

.github/workflows/docker.yml

@@ -19,15 +19,26 @@ jobs:
       packages: write
       contents: read
     steps:
+      - name: Free Disk space
+        shell: bash
+        run: |
+          df -h
+          sudo rm -rf /opt/hostedtoolcache
+          sudo rm -rf /usr/local/lib/android
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /opt/ghc
+          sudo rm -rf /usr/local/graalvm
+          sudo rm -rf /usr/local/share/boost
+          df -h
       - name: Check out the repo
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
         with:
           submodules: recursive
       - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
+        uses: docker/setup-buildx-action@v3
       - name: Docker meta
         id: meta
-        uses: docker/metadata-action@v4
+        uses: docker/metadata-action@v5
         with:
           images: |
             tinygo/tinygo-dev
@@ -36,24 +47,23 @@ jobs:
             type=sha,format=long
             type=raw,value=latest
       - name: Log in to Docker Hub
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKER_HUB_USERNAME }}
           password: ${{ secrets.DOCKER_HUB_ACCESS_TOKEN }}
       - name: Log in to Github Container Registry
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
         with:
           registry: ghcr.io
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
       - name: Build and push
-        uses: docker/build-push-action@v3
+        uses: docker/build-push-action@v5
         with:
           context: .
           push: true
           tags: ${{ steps.meta.outputs.tags }}
           labels: ${{ steps.meta.outputs.labels }}
-          build-contexts: tinygo-llvm-build=docker-image://tinygo/llvm-15
           cache-from: type=gha
           cache-to: type=gha,mode=max
       - name: Trigger Drivers repo build on Github Actions

.github/workflows/linux.yml

@@ -18,32 +18,37 @@ jobs:
     # statically linked binary.
     runs-on: ubuntu-latest
     container:
-      image: golang:1.20-alpine
+      image: golang:1.24-alpine
+    outputs:
+      version: ${{ steps.version.outputs.version }}
     steps:
       - name: Install apk dependencies
-        # tar: needed for actions/cache@v3
+        # tar: needed for actions/cache@v4
         # git+openssh: needed for checkout (I think?)
         # ruby: needed to install fpm
-        run: apk add tar git openssh make g++ ruby
+        run: apk add tar git openssh make g++ ruby-dev
       - name: Work around CVE-2022-24765
         # We're not on a multi-user machine, so this is safe.
         run: git config --global --add safe.directory "$GITHUB_WORKSPACE"
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
         with:
           submodules: true
+      - name: Extract TinyGo version
+        id: version
+        run: ./.github/workflows/tinygo-extract-version.sh | tee -a "$GITHUB_OUTPUT"
       - name: Cache Go
-        uses: actions/cache@v3
+        uses: actions/cache@v4
         with:
           key: go-cache-linux-alpine-v1-${{ hashFiles('go.mod') }}
           path: |
             ~/.cache/go-build
             ~/go/pkg/mod
       - name: Restore LLVM source cache
-        uses: actions/cache/restore@v3
+        uses: actions/cache/restore@v4
         id: cache-llvm-source
         with:
-          key: llvm-source-15-linux-alpine-v3
+          key: llvm-source-19-linux-alpine-v1
           path: |
             llvm-project/clang/lib/Headers
             llvm-project/clang/include
@@ -54,7 +59,7 @@ jobs:
         if: steps.cache-llvm-source.outputs.cache-hit != 'true'
         run: make llvm-source
       - name: Save LLVM source cache
-        uses: actions/cache/save@v3
+        uses: actions/cache/save@v4
         if: steps.cache-llvm-source.outputs.cache-hit != 'true'
         with:
           key: ${{ steps.cache-llvm-source.outputs.cache-primary-key }}
@@ -65,10 +70,10 @@ jobs:
             llvm-project/lld/include
             llvm-project/llvm/include
       - name: Restore LLVM build cache
-        uses: actions/cache/restore@v3
+        uses: actions/cache/restore@v4
         id: cache-llvm-build
         with:
-          key: llvm-build-15-linux-alpine-v4
+          key: llvm-build-19-linux-alpine-v1
           path: llvm-build
       - name: Build LLVM
         if: steps.cache-llvm-build.outputs.cache-hit != 'true'
@@ -83,13 +88,13 @@ jobs:
           # Remove unnecessary object files (to reduce cache size).
           find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
       - name: Save LLVM build cache
-        uses: actions/cache/save@v3
+        uses: actions/cache/save@v4
         if: steps.cache-llvm-build.outputs.cache-hit != 'true'
         with:
           key: ${{ steps.cache-llvm-build.outputs.cache-primary-key }}
           path: llvm-build
       - name: Cache Binaryen
-        uses: actions/cache@v3
+        uses: actions/cache@v4
         id: cache-binaryen
         with:
           key: binaryen-linux-alpine-v1
@@ -100,10 +105,10 @@ jobs:
           apk add cmake samurai python3
           make binaryen STATIC=1
       - name: Cache wasi-libc
-        uses: actions/cache@v3
+        uses: actions/cache@v4
         id: cache-wasi-libc
         with:
-          key: wasi-libc-sysroot-linux-alpine-v1
+          key: wasi-libc-sysroot-linux-alpine-v2
           path: lib/wasi-libc/sysroot
       - name: Build wasi-libc
         if: steps.cache-wasi-libc.outputs.cache-hit != 'true'
@@ -113,46 +118,53 @@ jobs:
           gem install --version 4.0.7 public_suffix
           gem install --version 2.7.6 dotenv
           gem install --no-document fpm
+      - name: Run linter
+        run: make lint
+      - name: Run spellcheck
+        run: make spell
       - name: Build TinyGo release
         run: |
           make release deb -j3 STATIC=1
-          cp -p build/release.tar.gz /tmp/tinygo.linux-amd64.tar.gz
-          cp -p build/release.deb    /tmp/tinygo_amd64.deb
+          cp -p build/release.tar.gz /tmp/tinygo${{ steps.version.outputs.version }}.linux-amd64.tar.gz
+          cp -p build/release.deb    /tmp/tinygo_${{ steps.version.outputs.version }}_amd64.deb
       - name: Publish release artifact
-        uses: actions/upload-artifact@v3
+        uses: actions/upload-artifact@v4
         with:
-          name: linux-amd64-double-zipped
+          name: linux-amd64-double-zipped-${{ steps.version.outputs.version }}
           path: |
-            /tmp/tinygo.linux-amd64.tar.gz
-            /tmp/tinygo_amd64.deb
+            /tmp/tinygo${{ steps.version.outputs.version }}.linux-amd64.tar.gz
+            /tmp/tinygo_${{ steps.version.outputs.version }}_amd64.deb
   test-linux-build:
     # Test the binaries built in the build-linux job by running the smoke tests.
     runs-on: ubuntu-latest
     needs: build-linux
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
-      - name: Install Go
-        uses: actions/setup-go@v3
+        uses: actions/checkout@v4
         with:
-          go-version: '1.20'
+          submodules: true
+      - name: Install Go
+        uses: actions/setup-go@v5
+        with:
+          go-version: '1.24'
           cache: true
       - name: Install wasmtime
-        run: |
-          mkdir -p $HOME/.wasmtime $HOME/.wasmtime/bin
-          curl https://github.com/bytecodealliance/wasmtime/releases/download/v5.0.0/wasmtime-v5.0.0-x86_64-linux.tar.xz -o wasmtime-v5.0.0-x86_64-linux.tar.xz -SfL
-          tar -C $HOME/.wasmtime/bin --wildcards -xf wasmtime-v5.0.0-x86_64-linux.tar.xz --strip-components=1 wasmtime-v5.0.0-x86_64-linux/*
-          echo "$HOME/.wasmtime/bin" >> $GITHUB_PATH
-      - name: Download release artifact
-        uses: actions/download-artifact@v3
+        uses: bytecodealliance/actions/wasmtime/setup@v1
         with:
-          name: linux-amd64-double-zipped
+          version: "29.0.1"
+      - name: Install wasm-tools
+        uses: bytecodealliance/actions/wasm-tools/setup@v1
+      - name: Download release artifact
+        uses: actions/download-artifact@v4
+        with:
+          name: linux-amd64-double-zipped-${{ needs.build-linux.outputs.version }}
       - name: Extract release tarball
         run: |
           mkdir -p ~/lib
-          tar -C ~/lib -xf tinygo.linux-amd64.tar.gz
+          tar -C ~/lib -xf tinygo${{ needs.build-linux.outputs.version }}.linux-amd64.tar.gz
           ln -s ~/lib/tinygo/bin/tinygo ~/go/bin/tinygo
-      - run: make tinygo-test-wasi-fast
+      - run: make tinygo-test-wasip1-fast
+      - run: make tinygo-test-wasip2-fast
       - run: make smoketest
   assert-test-linux:
     # Run all tests that can run on Linux, with LLVM assertions enabled to catch
@@ -160,7 +172,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
         with:
           submodules: true
       - name: Install apt dependencies
@@ -175,25 +187,25 @@ jobs:
               simavr \
               ninja-build
       - name: Install Go
-        uses: actions/setup-go@v3
+        uses: actions/setup-go@v5
         with:
-          go-version: '1.20'
+          go-version: '1.24'
           cache: true
       - name: Install Node.js
-        uses: actions/setup-node@v3
+        uses: actions/setup-node@v4
         with:
-          node-version: '14'
+          node-version: '18'
       - name: Install wasmtime
-        run: |
-          mkdir -p $HOME/.wasmtime $HOME/.wasmtime/bin
-          curl -L https://github.com/bytecodealliance/wasmtime/releases/download/v5.0.0/wasmtime-v5.0.0-x86_64-linux.tar.xz -o wasmtime-v5.0.0-x86_64-linux.tar.xz -SfL
-          tar -C $HOME/.wasmtime/bin --wildcards -xf wasmtime-v5.0.0-x86_64-linux.tar.xz --strip-components=1 wasmtime-v5.0.0-x86_64-linux/*
-          echo "$HOME/.wasmtime/bin" >> $GITHUB_PATH
+        uses: bytecodealliance/actions/wasmtime/setup@v1
+        with:
+          version: "29.0.1"
+      - name: Setup `wasm-tools`
+        uses: bytecodealliance/actions/wasm-tools/setup@v1
       - name: Restore LLVM source cache
-        uses: actions/cache/restore@v3
+        uses: actions/cache/restore@v4
         id: cache-llvm-source
         with:
-          key: llvm-source-15-linux-asserts-v3
+          key: llvm-source-19-linux-asserts-v1
           path: |
             llvm-project/clang/lib/Headers
             llvm-project/clang/include
@@ -204,7 +216,7 @@ jobs:
         if: steps.cache-llvm-source.outputs.cache-hit != 'true'
         run: make llvm-source
       - name: Save LLVM source cache
-        uses: actions/cache/save@v3
+        uses: actions/cache/save@v4
         if: steps.cache-llvm-source.outputs.cache-hit != 'true'
         with:
           key: ${{ steps.cache-llvm-source.outputs.cache-primary-key }}
@@ -215,10 +227,10 @@ jobs:
             llvm-project/lld/include
             llvm-project/llvm/include
       - name: Restore LLVM build cache
-        uses: actions/cache/restore@v3
+        uses: actions/cache/restore@v4
         id: cache-llvm-build
         with:
-          key: llvm-build-15-linux-asserts-v4
+          key: llvm-build-19-linux-asserts-v1
           path: llvm-build
       - name: Build LLVM
         if: steps.cache-llvm-build.outputs.cache-hit != 'true'
@@ -231,13 +243,13 @@ jobs:
           # Remove unnecessary object files (to reduce cache size).
           find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
       - name: Save LLVM build cache
-        uses: actions/cache/save@v3
+        uses: actions/cache/save@v4
         if: steps.cache-llvm-build.outputs.cache-hit != 'true'
         with:
           key: ${{ steps.cache-llvm-build.outputs.cache-primary-key }}
           path: llvm-build
       - name: Cache Binaryen
-        uses: actions/cache@v3
+        uses: actions/cache@v4
         id: cache-binaryen
         with:
           key: binaryen-linux-asserts-v1
@@ -246,10 +258,10 @@ jobs:
         if: steps.cache-binaryen.outputs.cache-hit != 'true'
         run: make binaryen
       - name: Cache wasi-libc
-        uses: actions/cache@v3
+        uses: actions/cache@v4
         id: cache-wasi-libc
         with:
-          key: wasi-libc-sysroot-linux-asserts-v5
+          key: wasi-libc-sysroot-linux-asserts-v6
           path: lib/wasi-libc/sysroot
       - name: Build wasi-libc
         if: steps.cache-wasi-libc.outputs.cache-hit != 'true'
@@ -266,7 +278,7 @@ jobs:
       - run: make smoketest
       - run: make wasmtest
       - run: make tinygo-baremetal
-  build-linux-arm:
+  build-linux-cross:
     # Build ARM Linux binaries, ready for release.
     # This intentionally uses an older Linux image, so that we compile against
     # an older glibc version and therefore are compatible with a wide range of
@@ -275,28 +287,41 @@ jobs:
     # in that process to avoid doing lots of duplicate work and to avoid
     # complications around precompiled libraries such as compiler-rt shipped as
     # part of the release tarball.
-    runs-on: ubuntu-20.04
+    strategy:
+      matrix:
+        goarch: [ arm, arm64 ]
+        include:
+          - goarch: arm64
+            toolchain: aarch64-linux-gnu
+            libc: arm64
+          - goarch: arm
+            toolchain: arm-linux-gnueabihf
+            libc: armhf
+    runs-on: ubuntu-22.04 # note: use the oldest image available! (see above)
     needs: build-linux
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
+      - name: Get TinyGo version
+        id: version
+        run: ./.github/workflows/tinygo-extract-version.sh | tee -a "$GITHUB_OUTPUT"
       - name: Install apt dependencies
         run: |
           sudo apt-get update
           sudo apt-get install --no-install-recommends \
               qemu-user \
-              g++-arm-linux-gnueabihf \
-              libc6-dev-armhf-cross
+              g++-${{ matrix.toolchain }} \
+              libc6-dev-${{ matrix.libc }}-cross
       - name: Install Go
-        uses: actions/setup-go@v3
+        uses: actions/setup-go@v5
         with:
-          go-version: '1.20'
+          go-version: '1.24'
           cache: true
       - name: Restore LLVM source cache
-        uses: actions/cache/restore@v3
+        uses: actions/cache/restore@v4
         id: cache-llvm-source
         with:
-          key: llvm-source-15-linux-v3
+          key: llvm-source-19-linux-v1
           path: |
             llvm-project/clang/lib/Headers
             llvm-project/clang/include
@@ -307,7 +332,7 @@ jobs:
         if: steps.cache-llvm-source.outputs.cache-hit != 'true'
         run: make llvm-source
       - name: Save LLVM source cache
-        uses: actions/cache/save@v3
+        uses: actions/cache/save@v4
         if: steps.cache-llvm-source.outputs.cache-hit != 'true'
         with:
           key: ${{ steps.cache-llvm-source.outputs.cache-primary-key }}
@@ -318,10 +343,10 @@ jobs:
             llvm-project/lld/include
             llvm-project/llvm/include
       - name: Restore LLVM build cache
-        uses: actions/cache/restore@v3
+        uses: actions/cache/restore@v4
         id: cache-llvm-build
         with:
-          key: llvm-build-15-linux-arm-v4
+          key: llvm-build-19-linux-${{ matrix.goarch }}-v3
           path: llvm-build
       - name: Build LLVM
         if: steps.cache-llvm-build.outputs.cache-hit != 'true'
@@ -332,27 +357,27 @@ jobs:
           # Install build dependencies.
           sudo apt-get install --no-install-recommends ninja-build
           # build!
-          make llvm-build CROSS=arm-linux-gnueabihf
+          make llvm-build CROSS=${{ matrix.toolchain }}
           # Remove unnecessary object files (to reduce cache size).
           find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
       - name: Save LLVM build cache
-        uses: actions/cache/save@v3
+        uses: actions/cache/save@v4
         if: steps.cache-llvm-build.outputs.cache-hit != 'true'
         with:
           key: ${{ steps.cache-llvm-build.outputs.cache-primary-key }}
           path: llvm-build
       - name: Cache Binaryen
-        uses: actions/cache@v3
+        uses: actions/cache@v4
         id: cache-binaryen
         with:
-          key: binaryen-linux-arm-v1
+          key: binaryen-linux-${{ matrix.goarch }}-v4
           path: build/wasm-opt
       - name: Build Binaryen
         if: steps.cache-binaryen.outputs.cache-hit != 'true'
         run: |
           sudo apt-get install --no-install-recommends ninja-build
           git submodule update --init lib/binaryen
-          make CROSS=arm-linux-gnueabihf binaryen
+          make CROSS=${{ matrix.toolchain }} binaryen
       - name: Install fpm
         run: |
           sudo gem install --version 4.0.7 public_suffix
@@ -360,142 +385,28 @@ jobs:
           sudo gem install --no-document fpm
       - name: Build TinyGo binary
         run: |
-          make CROSS=arm-linux-gnueabihf
+          make CROSS=${{ matrix.toolchain }}
       - name: Download amd64 release
-        uses: actions/download-artifact@v3
+        uses: actions/download-artifact@v4
         with:
-          name: linux-amd64-double-zipped
+          name: linux-amd64-double-zipped-${{ needs.build-linux.outputs.version }}
       - name: Extract amd64 release
         run: |
           mkdir -p build/release
-          tar -xf tinygo.linux-amd64.tar.gz -C build/release tinygo
+          tar -xf tinygo${{ needs.build-linux.outputs.version }}.linux-amd64.tar.gz -C build/release tinygo
       - name: Modify release
         run: |
           cp -p build/tinygo   build/release/tinygo/bin
           cp -p build/wasm-opt build/release/tinygo/bin
-      - name: Create arm release
+      - name: Create ${{ matrix.goarch }} release
         run: |
-          make release deb RELEASEONLY=1 DEB_ARCH=armhf
-          cp -p build/release.tar.gz /tmp/tinygo.linux-arm.tar.gz
-          cp -p build/release.deb    /tmp/tinygo_armhf.deb
+          make release deb RELEASEONLY=1 DEB_ARCH=${{ matrix.libc }}
+          cp -p build/release.tar.gz /tmp/tinygo${{ steps.version.outputs.version }}.linux-${{ matrix.goarch }}.tar.gz
+          cp -p build/release.deb    /tmp/tinygo_${{ steps.version.outputs.version }}_${{ matrix.libc }}.deb
       - name: Publish release artifact
-        uses: actions/upload-artifact@v3
+        uses: actions/upload-artifact@v4
         with:
-          name: linux-arm-double-zipped
+          name: linux-${{ matrix.goarch }}-double-zipped-${{ steps.version.outputs.version }}
           path: |
-            /tmp/tinygo.linux-arm.tar.gz
-            /tmp/tinygo_armhf.deb
-  build-linux-arm64:
-    # Build ARM64 Linux binaries, ready for release.
-    # It is set to "needs: build-linux" because it modifies the release created
-    # in that process to avoid doing lots of duplicate work and to avoid
-    # complications around precompiled libraries such as compiler-rt shipped as
-    # part of the release tarball.
-    runs-on: ubuntu-20.04
-    needs: build-linux
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-      - name: Install apt dependencies
-        run: |
-          sudo apt-get update
-          sudo apt-get install --no-install-recommends \
-              qemu-user \
-              g++-aarch64-linux-gnu \
-              libc6-dev-arm64-cross \
-              ninja-build
-      - name: Install Go
-        uses: actions/setup-go@v3
-        with:
-          go-version: '1.20'
-          cache: true
-      - name: Restore LLVM source cache
-        uses: actions/cache/restore@v3
-        id: cache-llvm-source
-        with:
-          key: llvm-source-15-linux-v3
-          path: |
-            llvm-project/clang/lib/Headers
-            llvm-project/clang/include
-            llvm-project/compiler-rt
-            llvm-project/lld/include
-            llvm-project/llvm/include
-      - name: Download LLVM source
-        if: steps.cache-llvm-source.outputs.cache-hit != 'true'
-        run: make llvm-source
-      - name: Save LLVM source cache
-        uses: actions/cache/save@v3
-        if: steps.cache-llvm-source.outputs.cache-hit != 'true'
-        with:
-          key: ${{ steps.cache-llvm-source.outputs.cache-primary-key }}
-          path: |
-            llvm-project/clang/lib/Headers
-            llvm-project/clang/include
-            llvm-project/compiler-rt
-            llvm-project/lld/include
-            llvm-project/llvm/include
-      - name: Restore LLVM build cache
-        uses: actions/cache/restore@v3
-        id: cache-llvm-build
-        with:
-          key: llvm-build-15-linux-arm64-v4
-          path: llvm-build
-      - name: Build LLVM
-        if: steps.cache-llvm-build.outputs.cache-hit != 'true'
-        run: |
-          # fetch LLVM source
-          rm -rf llvm-project
-          make llvm-source
-          # build!
-          make llvm-build CROSS=aarch64-linux-gnu
-          # Remove unnecessary object files (to reduce cache size).
-          find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
-      - name: Save LLVM build cache
-        uses: actions/cache/save@v3
-        if: steps.cache-llvm-build.outputs.cache-hit != 'true'
-        with:
-          key: ${{ steps.cache-llvm-build.outputs.cache-primary-key }}
-          path: llvm-build
-      - name: Cache Binaryen
-        uses: actions/cache@v3
-        id: cache-binaryen
-        with:
-          key: binaryen-linux-arm64-v1
-          path: build/wasm-opt
-      - name: Build Binaryen
-        if: steps.cache-binaryen.outputs.cache-hit != 'true'
-        run: |
-          git submodule update --init lib/binaryen
-          make CROSS=aarch64-linux-gnu binaryen
-      - name: Install fpm
-        run: |
-          sudo gem install --version 4.0.7 public_suffix
-          sudo gem install --version 2.7.6 dotenv
-          sudo gem install --no-document fpm
-      - name: Build TinyGo binary
-        run: |
-          make CROSS=aarch64-linux-gnu
-      - name: Download amd64 release
-        uses: actions/download-artifact@v3
-        with:
-          name: linux-amd64-double-zipped
-      - name: Extract amd64 release
-        run: |
-          mkdir -p build/release
-          tar -xf tinygo.linux-amd64.tar.gz -C build/release tinygo
-      - name: Modify release
-        run: |
-          cp -p build/tinygo   build/release/tinygo/bin
-          cp -p build/wasm-opt build/release/tinygo/bin
-      - name: Create arm64 release
-        run: |
-          make release deb RELEASEONLY=1 DEB_ARCH=arm64
-          cp -p build/release.tar.gz /tmp/tinygo.linux-arm64.tar.gz
-          cp -p build/release.deb    /tmp/tinygo_arm64.deb
-      - name: Publish release artifact
-        uses: actions/upload-artifact@v3
-        with:
-          name: linux-arm64-double-zipped
-          path: |
-            /tmp/tinygo.linux-arm64.tar.gz
-            /tmp/tinygo_arm64.deb
+            /tmp/tinygo${{ steps.version.outputs.version }}.linux-${{ matrix.goarch }}.tar.gz
+            /tmp/tinygo_${{ steps.version.outputs.version }}_${{ matrix.libc }}.deb

.github/workflows/llvm.yml

@@ -25,18 +25,18 @@ jobs:
       contents: read
     steps:
       - name: Check out the repo
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
         with:
           submodules: recursive
       - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
+        uses: docker/setup-buildx-action@v3
       - name: Docker meta
         id: meta
-        uses: docker/metadata-action@v4
+        uses: docker/metadata-action@v5
         with:
           images: |
-            tinygo/llvm-15
-            ghcr.io/${{ github.repository_owner }}/llvm-15
+            tinygo/llvm-19
+            ghcr.io/${{ github.repository_owner }}/llvm-19
           tags: |
             type=sha,format=long
             type=raw,value=latest
@@ -46,13 +46,13 @@ jobs:
           username: ${{ secrets.DOCKER_HUB_USERNAME }}
           password: ${{ secrets.DOCKER_HUB_ACCESS_TOKEN }}
       - name: Log in to Github Container Registry
-        uses: docker/login-action@v2
+        uses: docker/login-action@v3
         with:
           registry: ghcr.io
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
       - name: Build and push
-        uses: docker/build-push-action@v4
+        uses: docker/build-push-action@v5
         with:
           target: tinygo-llvm-build
           context: .

.github/workflows/nix.yml

@@ -0,0 +1,48 @@
+name: Nix
+
+on:
+  pull_request:
+  push:
+    branches:
+      - dev
+      - release
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  nix-test:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Uninstall system LLVM
+        # Hack to work around issue where we still include system headers for
+        # some reason.
+        # See: https://github.com/tinygo-org/tinygo/pull/4516#issuecomment-2416363668
+        run: sudo apt-get remove llvm-18
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Pull musl
+        run: |
+            git submodule update --init lib/musl
+      - name: Restore LLVM source cache
+        uses: actions/cache/restore@v4
+        id: cache-llvm-source
+        with:
+          key: llvm-source-19-linux-nix-v1
+          path: |
+            llvm-project/compiler-rt
+      - name: Download LLVM source
+        if: steps.cache-llvm-source.outputs.cache-hit != 'true'
+        run: make llvm-source
+      - name: Save LLVM source cache
+        uses: actions/cache/save@v4
+        if: steps.cache-llvm-source.outputs.cache-hit != 'true'
+        with:
+          key: ${{ steps.cache-llvm-source.outputs.cache-primary-key }}
+          path: |
+            llvm-project/compiler-rt
+      - uses: cachix/install-nix-action@v22
+      - name: Test
+        run: |
+          nix develop --ignore-environment --keep HOME --command bash -c "go install && ~/go/bin/tinygo version && ~/go/bin/tinygo build -o test ./testdata/cgo"

.github/workflows/sizediff-install-pkgs.sh

@@ -0,0 +1,12 @@
+# Command that's part of sizediff.yml. This is put in a separate file so that it
+# still works after checking out the dev branch (that is, when going from LLVM
+# 16 to LLVM 17 for example, both Clang 16 and Clang 17 are installed).
+
+echo 'deb https://apt.llvm.org/noble/ llvm-toolchain-noble-19 main' | sudo tee /etc/apt/sources.list.d/llvm.list
+wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key | sudo apt-key add -
+sudo apt-get update
+sudo apt-get install --no-install-recommends -y \
+    llvm-19-dev \
+    clang-19 \
+    libclang-19-dev \
+    lld-19

.github/workflows/sizediff.yml

@@ -9,7 +9,9 @@ concurrency:
 
 jobs:
   sizediff:
-    runs-on: ubuntu-22.04
+    # Note: when updating the Ubuntu version, also update the Ubuntu version in
+    # sizediff-install-pkgs.sh
+    runs-on: ubuntu-24.04
     permissions:
       pull-requests: write
     steps:
@@ -18,34 +20,26 @@ jobs:
         run: |
           echo "$HOME/go/bin" >> $GITHUB_PATH
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
         with:
           fetch-depth: 0 # fetch all history (no sparse checkout)
           submodules: true
       - name: Install apt dependencies
-        run: |
-          echo 'deb https://apt.llvm.org/jammy/ llvm-toolchain-jammy-15 main' | sudo tee /etc/apt/sources.list.d/llvm.list
-          wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key | sudo apt-key add -
-          sudo apt-get update
-          sudo apt-get install --no-install-recommends -y \
-              llvm-15-dev \
-              clang-15 \
-              libclang-15-dev \
-              lld-15
+        run: ./.github/workflows/sizediff-install-pkgs.sh
       - name: Restore LLVM source cache
-        uses: actions/cache@v3
+        uses: actions/cache@v4
         id: cache-llvm-source
         with:
-          key: llvm-source-15-sizediff-v1
+          key: llvm-source-19-sizediff-v1
           path: |
             llvm-project/compiler-rt
       - name: Download LLVM source
         if: steps.cache-llvm-source.outputs.cache-hit != 'true'
         run: make llvm-source
       - name: Cache Go
-        uses: actions/cache@v3
+        uses: actions/cache@v4
         with:
-          key: go-cache-linux-sizediff-v1-${{ hashFiles('go.mod') }}
+          key: go-cache-linux-sizediff-v2-${{ hashFiles('go.mod') }}
           path: |
             ~/.cache/go-build
             ~/go/pkg/mod
@@ -55,22 +49,25 @@ jobs:
       - name: Save HEAD
         run: git branch github-actions-saved-HEAD HEAD
 
-      # Compute sizes for the dev branch
-      - name: Checkout dev branch
-        run: git checkout --no-recurse-submodules `git merge-base HEAD origin/dev`
-      - name: Build tinygo binary for the dev branch
-        run: go install
-      - name: Determine binary sizes on the dev branch
-        run: (cd drivers; make smoke-test XTENSA=0 | tee sizes-dev.txt)
-
       # Compute sizes for the PR branch
-      - name: Checkout PR branch
-        run: git checkout --no-recurse-submodules github-actions-saved-HEAD
       - name: Build tinygo binary for the PR branch
         run: go install
       - name: Determine binary sizes on the PR branch
         run: (cd drivers; make smoke-test XTENSA=0 | tee sizes-pr.txt)
 
+      # Compute sizes for the dev branch
+      - name: Checkout dev branch
+        run: |
+          git reset --hard origin/dev
+          git checkout --no-recurse-submodules `git merge-base HEAD origin/dev`
+      - name: Install apt dependencies on the dev branch
+        # this is only needed on a PR that changes the LLVM version
+        run: ./.github/workflows/sizediff-install-pkgs.sh
+      - name: Build tinygo binary for the dev branch
+        run: go install
+      - name: Determine binary sizes on the dev branch
+        run: (cd drivers; make smoke-test XTENSA=0 | tee sizes-dev.txt)
+
       # Create comment
       # TODO: add a summary, something like:
       #  - overall size difference (percent)

.github/workflows/tinygo-extract-version.sh

@@ -0,0 +1,4 @@
+#!/bin/sh
+
+# Extract the version string from the source code, to be stored in a variable.
+grep 'const version' goenv/version.go | sed 's/^const version = "\(.*\)"$/version=\1/g'

.github/workflows/windows.yml

@@ -14,9 +14,11 @@ concurrency:
 jobs:
   build-windows:
     runs-on: windows-2022
+    outputs:
+      version: ${{ steps.version.outputs.version }}
     steps:
       - name: Configure pagefile
-        uses: al-cheb/configure-pagefile-action@v1.3
+        uses: al-cheb/configure-pagefile-action@v1.4
         with:
           minimum-size: 8GB
           maximum-size: 24GB
@@ -29,19 +31,23 @@ jobs:
         run: |
           scoop install ninja binaryen
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
         with:
           submodules: true
+      - name: Extract TinyGo version
+        id: version
+        shell: bash
+        run: ./.github/workflows/tinygo-extract-version.sh | tee -a "$GITHUB_OUTPUT"
       - name: Install Go
-        uses: actions/setup-go@v3
+        uses: actions/setup-go@v5
         with:
-          go-version: '1.20'
+          go-version: '1.24'
           cache: true
       - name: Restore cached LLVM source
-        uses: actions/cache/restore@v3
+        uses: actions/cache/restore@v4
         id: cache-llvm-source
         with:
-          key: llvm-source-15-windows-v4
+          key: llvm-source-19-windows-v1
           path: |
             llvm-project/clang/lib/Headers
             llvm-project/clang/include
@@ -52,7 +58,7 @@ jobs:
         if: steps.cache-llvm-source.outputs.cache-hit != 'true'
         run: make llvm-source
       - name: Save cached LLVM source
-        uses: actions/cache/save@v3
+        uses: actions/cache/save@v4
         if: steps.cache-llvm-source.outputs.cache-hit != 'true'
         with:
           key: ${{ steps.cache-llvm-source.outputs.cache-primary-key }}
@@ -63,10 +69,10 @@ jobs:
             llvm-project/lld/include
             llvm-project/llvm/include
       - name: Restore cached LLVM build
-        uses: actions/cache/restore@v3
+        uses: actions/cache/restore@v4
         id: cache-llvm-build
         with:
-          key: llvm-build-15-windows-v6
+          key: llvm-build-19-windows-v1
           path: llvm-build
       - name: Build LLVM
         if: steps.cache-llvm-build.outputs.cache-hit != 'true'
@@ -80,23 +86,30 @@ jobs:
           # Remove unnecessary object files (to reduce cache size).
           find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
       - name: Save cached LLVM build
-        uses: actions/cache/save@v3
+        uses: actions/cache/save@v4
         if: steps.cache-llvm-build.outputs.cache-hit != 'true'
         with:
           key: ${{ steps.cache-llvm-build.outputs.cache-primary-key }}
           path: llvm-build
       - name: Cache wasi-libc sysroot
-        uses: actions/cache@v3
+        uses: actions/cache@v4
         id: cache-wasi-libc
         with:
-          key: wasi-libc-sysroot-v4
+          key: wasi-libc-sysroot-v5
           path: lib/wasi-libc/sysroot
       - name: Build wasi-libc
         if: steps.cache-wasi-libc.outputs.cache-hit != 'true'
         run: make wasi-libc
+      - name: Cache Go cache
+        uses: actions/cache@v4
+        with:
+          key: go-cache-windows-v1-${{ hashFiles('go.mod') }}
+          path: |
+            C:/Users/runneradmin/AppData/Local/go-build
+            C:/Users/runneradmin/go/pkg/mod
       - name: Install wasmtime
         run: |
-          scoop install wasmtime
+          scoop install wasmtime@29.0.1
       - name: make gen-device
         run: make -j3 gen-device
       - name: Test TinyGo
@@ -108,7 +121,7 @@ jobs:
       - name: Make release artifact
         shell: bash
         working-directory: build/release
-        run: 7z -tzip a release.zip tinygo
+        run: 7z -tzip a tinygo${{ steps.version.outputs.version }}.windows-amd64.zip tinygo
       - name: Publish release artifact
         # Note: this release artifact is double-zipped, see:
         # https://github.com/actions/upload-artifact/issues/39
@@ -116,17 +129,17 @@ jobs:
         # - have a dobule-zipped artifact when downloaded from the UI
         # - have a very slow artifact upload
         # We're doing the former here, to keep artifact uploads fast.
-        uses: actions/upload-artifact@v3
+        uses: actions/upload-artifact@v4
         with:
-          name: release-double-zipped
-          path: build/release/release.zip
+          name: windows-amd64-double-zipped-${{ steps.version.outputs.version }}
+          path: build/release/tinygo${{ steps.version.outputs.version }}.windows-amd64.zip
 
   smoke-test-windows:
     runs-on: windows-2022
     needs: build-windows
     steps:
       - name: Configure pagefile
-        uses: al-cheb/configure-pagefile-action@v1.3
+        uses: al-cheb/configure-pagefile-action@v1.4
         with:
           minimum-size: 8GB
           maximum-size: 24GB
@@ -139,21 +152,21 @@ jobs:
         run: |
           scoop install binaryen
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
       - name: Install Go
-        uses: actions/setup-go@v3
+        uses: actions/setup-go@v5
         with:
-          go-version: '1.20'
+          go-version: '1.24'
           cache: true
       - name: Download TinyGo build
-        uses: actions/download-artifact@v2
+        uses: actions/download-artifact@v4
         with:
-          name: release-double-zipped
+          name: windows-amd64-double-zipped-${{ needs.build-windows.outputs.version }}
           path: build/
       - name: Unzip TinyGo build
         shell: bash
         working-directory: build
-        run: 7z x release.zip -r
+        run: 7z x tinygo*.windows-amd64.zip -r
       - name: Smoke tests
         shell: bash
         run: make smoketest TINYGO=$(PWD)/build/tinygo/bin/tinygo
@@ -163,27 +176,27 @@ jobs:
     needs: build-windows
     steps:
       - name: Configure pagefile
-        uses: al-cheb/configure-pagefile-action@v1.3
+        uses: al-cheb/configure-pagefile-action@v1.4
         with:
           minimum-size: 8GB
           maximum-size: 24GB
           disk-root: "C:"
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
       - name: Install Go
-        uses: actions/setup-go@v3
+        uses: actions/setup-go@v5
         with:
-          go-version: '1.20'
+          go-version: '1.24'
           cache: true
       - name: Download TinyGo build
-        uses: actions/download-artifact@v2
+        uses: actions/download-artifact@v4
         with:
-          name: release-double-zipped
+          name: windows-amd64-double-zipped-${{ needs.build-windows.outputs.version }}
           path: build/
       - name: Unzip TinyGo build
         shell: bash
         working-directory: build
-        run: 7z x release.zip -r
+        run: 7z x tinygo*.windows-amd64.zip -r
       - name: Test stdlib packages
         run: make tinygo-test TINYGO=$(PWD)/build/tinygo/bin/tinygo
 
@@ -192,7 +205,7 @@ jobs:
     needs: build-windows
     steps:
       - name: Configure pagefile
-        uses: al-cheb/configure-pagefile-action@v1.3
+        uses: al-cheb/configure-pagefile-action@v1.4
         with:
           minimum-size: 8GB
           maximum-size: 24GB
@@ -203,22 +216,22 @@ jobs:
       - name: Install Dependencies
         shell: bash
         run: |
-          scoop install binaryen wasmtime
+          scoop install binaryen && scoop install wasmtime@29.0.1
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
       - name: Install Go
-        uses: actions/setup-go@v3
+        uses: actions/setup-go@v5
         with:
-          go-version: '1.20'
+          go-version: '1.24'
           cache: true
       - name: Download TinyGo build
-        uses: actions/download-artifact@v2
+        uses: actions/download-artifact@v4
         with:
-          name: release-double-zipped
+          name: windows-amd64-double-zipped-${{ needs.build-windows.outputs.version }}
           path: build/
       - name: Unzip TinyGo build
         shell: bash
         working-directory: build
-        run: 7z x release.zip -r
-      - name: Test stdlib packages on wasi
-        run: make tinygo-test-wasi-fast TINYGO=$(PWD)/build/tinygo/bin/tinygo
+        run: 7z x tinygo*.windows-amd64.zip -r
+      - name: Test stdlib packages on wasip1
+        run: make tinygo-test-wasip1-fast TINYGO=$(PWD)/build/tinygo/bin/tinygo

.gitignore

@@ -1,3 +1,8 @@
+.DS_Store
+.vscode
+go.work
+go.work.sum
+
 docs/_build
 src/device/avr/*.go
 src/device/avr/*.ld
@@ -15,9 +20,11 @@ src/device/stm32/*.go
 src/device/stm32/*.s
 src/device/kendryte/*.go
 src/device/kendryte/*.s
+src/device/renesas/*.go
+src/device/renesas/*.s
 src/device/rp/*.go
 src/device/rp/*.s
-vendor
+./vendor
 llvm-build
 llvm-project
 build/*
@@ -30,5 +37,9 @@ test.exe
 test.gba
 test.hex
 test.nro
+test.uf2
 test.wasm
 wasm.wasm
+
+*.uf2
+*.elf

.gitmodules

@@ -9,10 +9,11 @@
 	url = https://github.com/avr-rust/avr-mcu.git
 [submodule "lib/cmsis-svd"]
 	path = lib/cmsis-svd
-	url = https://github.com/tinygo-org/cmsis-svd
+	url = https://github.com/cmsis-svd/cmsis-svd-data.git
+	branch = main
 [submodule "lib/wasi-libc"]
 	path = lib/wasi-libc
-	url = https://github.com/CraneStation/wasi-libc
+	url = https://github.com/WebAssembly/wasi-libc
 [submodule "lib/picolibc"]
 	path = lib/picolibc
 	url = https://github.com/keith-packard/picolibc.git
@@ -31,3 +32,10 @@
 [submodule "lib/macos-minimal-sdk"]
 	path = lib/macos-minimal-sdk
 	url = https://github.com/aykevl/macos-minimal-sdk.git
+[submodule "src/net"]
+	path = src/net
+	url = https://github.com/tinygo-org/net.git
+	branch = dev
+[submodule "lib/wasi-cli"]
+	path = lib/wasi-cli
+	url = https://github.com/WebAssembly/wasi-cli

BUILDING.md

@@ -18,7 +18,8 @@ tarball. If you want to help with development of TinyGo itself, you should follo
 LLVM, Clang and LLD are quite light on dependencies, requiring only standard
 build tools to be built. Go is of course necessary to build TinyGo itself.
 
-  * Go (1.18+)
+  * Go (1.19+)
+  * GNU Make
   * Standard build tools (gcc/clang)
   * git
   * CMake
@@ -27,6 +28,21 @@ build tools to be built. Go is of course necessary to build TinyGo itself.
 The rest of this guide assumes you're running Linux, but it should be equivalent
 on a different system like Mac.
 
+## Using GNU Make
+
+The static build of TinyGo is driven by GNUmakefile, which provides a help target for quick reference:
+
+    % make help
+    clean                           Remove build directory
+    fmt                             Reformat source
+    fmt-check                       Warn if any source needs reformatting
+    gen-device                      Generate microcontroller-specific sources
+    llvm-source                     Get LLVM sources
+    llvm-build                      Build LLVM
+    tinygo                          Build the TinyGo compiler
+    lint                            Lint source tree
+    spell                           Spellcheck source tree
+
 ## Download the source
 
 The first step is to download the TinyGo sources (use `--recursive` if you clone

CHANGELOG.md

@@ -1,3 +1,503 @@
+0.36.0
+---
+* **general**
+  - add initial Go 1.24 support
+  - add support for LLVM 19
+  - update license for 2025
+  - make small corrections for README regarding wasm
+  - use GOOS and GOARCH for building wasm simulated boards
+  - only infer target for wasm when GOOS and GOARCH are set correctly, not just based on file extension
+  - add test-corpus-wasip2
+  - use older image for cross-compiling builds
+  - update Linux builds to run on ubuntu-latest since 20.04 is being retired
+  - ensure build output directory is created
+  - add NoSandbox flag to chrome headless that is run during WASM tests, since this is now required for Ubuntu 23+ and we are using Ubuntu 24+ when running Github Actions
+  - update wasmtime used for CI to 29.0.1 to fix issue with install during CI tests
+  - update to use `Get-CimInstance` as `wmic` is being deprecated on WIndows
+  - remove unnecessary executable permissions
+  - `goenv`: update to new v0.36.0 development version
+* **compiler**
+  - `builder`: fix parsing of external ld.lld error messages
+  - `cgo`: mangle identifier names
+  - `interp`: correctly mark functions as modifying memory
+  - add buildmode=wasi-legacy to support existing base of users who expected the older behavior for wasi modules to not return an exit code as if they were reactors
+* **standard library**
+  - `crypto/tls`: add Dialer.DialContext() to fix websocket client
+  - `crypto/tls`: add VersionTLS constants and VersionName(version uint16) method that turns it into a string, copied from big go
+  - `internal/syscall/unix`: use our own version of this package
+  - `machine`: replace hard-coded cpu frequencies on rp2xxx
+  - `machine`: bump rp2350 CPUFrequency to 150 MHz
+  - `machine`: compute rp2 clock dividers from crystal and target frequency
+  - `machine`: remove bytes package dependency in flash code
+  - `machine/usb/descriptor`: avoid bytes package
+  - `net`: update to latest submodule with httptest subpackage and ResolveIPAddress implementation
+  - `os`: add File.Chdir support
+  - `os`: implement stub Chdir for non-OS systems
+  - `os/file`: add file.Chmod
+  - `reflect`: implement Value.Equal
+  - `runtime`: add FIPS helper functions
+  - `runtime`: manually initialize xorshift state
+  - `sync`: move Mutex to internal/task
+  - `syscall`: add wasip1 RandomGet
+  - `testing`: add Chdir
+  - `wasip2`: add stubs to get internal/syscall/unix to work
+* **fixes**
+  - correctly handle calls for GetRNG() when being made from nrf devices with SoftDevice enabled
+  - fix stm32f103 ADC
+  - `wasm`: correctly handle id lookup for finalizeRef call
+  - `wasm`: avoid total failure on wasm finalizer call
+  - `wasm`: convert offset as signed int into unsigned int in syscall/js.stringVal in wasm_exec.js
+* **targets**
+  - rp2350: add pll generalized solution; fix ADC handles; pwm period fix
+  - rp2350: extending support to include the rp2350b
+  - rp2350: cleanup: unexport internal USB and clock package variable, consts and types
+  - nrf: make ADC resolution changeable
+  - turn on GC for TKey1 device, since it does in fact work
+  - match Pico2 stack size to Pico
+* **boards**
+  - add support for Pimoroni Pico Plus2
+  - add target for pico2-w board
+  - add comboat_fw tag for elecrow W5 boards with Combo-AT Wifi firmware
+  - add support for Elecrow Pico rp2350 W5 boards
+  - add support for Elecrow Pico rp2040 W5 boards
+  - add support for NRF51 HW-651
+  - add support for esp32c3-supermini
+  - add support for waveshare-rp2040-tiny
+* **examples**
+  - add naive debouncing for pininterrupt example
+
+
+0.35.0
+---
+* **general**
+  - update cmsis-svd library
+  - use default UART settings in the echo example
+  - `goenv`: also show git hash with custom build of TinyGo
+  - `goenv`: support parsing development versions of Go
+  - `main`: parse extldflags early so we can report the error message
+* **compiler**
+  - `builder`: whitelist temporary directory env var for Clang invocation to fix Windows bug
+  - `builder`: fix cache paths in `-size=full` output
+  - `builder`: work around incorrectly escaped DWARF paths on Windows (Clang bug)
+  - `builder`: fix wasi-libc path names on Windows with `-size=full`
+  - `builder`: write HTML size report
+  - `cgo`: support C identifiers only referred to from within macros
+  - `cgo`: support function-like macros
+  - `cgo`: support errno value as second return parameter
+  - `cgo`: add support for `#cgo noescape` lines
+  - `compiler`: fix bug in interrupt lowering
+  - `compiler`: allow panic directly in `defer`
+  - `compiler`: fix wasmimport -> wasmexport in error message
+  - `compiler`: support `//go:noescape` pragma
+  - `compiler`: report error instead of crashing when instantiating a generic function without body
+  - `interp`: align created globals
+* **standard library**
+  - `machine`: modify i2s interface/implementation to better match specification
+  - `os`: implement `StartProcess`
+  - `reflect`: add `Value.Clear`
+  - `reflect`: add interface support to `NumMethods`
+  - `reflect`: fix `AssignableTo` for named + non-named types
+  - `reflect`: implement `CanConvert`
+  - `reflect`: handle more cases in `Convert`
+  - `reflect`: fix Copy of non-pointer array with size > 64bits
+  - `runtime`: don't call sleepTicks with a negative duration
+  - `runtime`: optimize GC scanning (findHead)
+  - `runtime`: move constants into shared package
+  - `runtime`: add `runtime.fcntl` function for internal/syscall/unix
+  - `runtime`: heapptr only needs to be initialized once
+  - `runtime`: refactor scheduler (this fixes a few bugs with `-scheduler=none`)
+  - `runtime`: rewrite channel implementation to be smaller and more flexible
+  - `runtime`: use `SA_RESTART` when registering a signal for os/signal
+  - `runtime`: implement race-free signals using futexes
+  - `runtime`: run deferred functions in `Goexit`
+  - `runtime`: remove `Cond` which seems to be unused
+  - `runtime`: properly handle unix read on directory
+  - `runtime/trace`: stub all public methods
+  - `sync`: don't use volatile in `Mutex`
+  - `sync`: implement `WaitGroup` using a (pseudo)futex
+  - `sync`: make `Cond` parallelism-safe
+  - `syscall`: use wasi-libc tables for wasm/js target
+* **targets**
+  - `mips`: fix a bug when scanning the stack
+  - `nintendoswitch`: get this target to compile again
+  - `rp2350`: add support for the new RP2350
+  - `rp2040/rp2350` : make I2C implementation shared for rp2040/rp2350
+  - `rp2040/rp2350` : make SPI implementation shared for rp2040/rp2350
+  - `rp2040/rp2350` : make RNG implementation shared for rp2040/rp2350
+  - `wasm`: revise and simplify wasmtime argument handling
+  - `wasm`: support `//go:wasmexport` functions after a call to `time.Sleep`
+  - `wasm`: correctly return from run() in wasm_exec.js
+  - `wasm`: call process.exit() when go.run() returns
+  - `windows`: don't return, exit via exit(0) instead to flush stdout buffer
+* **boards**
+  - add support for the Tillitis TKey
+  - add support for the Raspberry Pi Pico2 (based on the RP2040)
+  - add support for Pimoroni Tiny2350
+
+
+0.34.0
+---
+* **general**
+  - fix `GOOS=wasip1` for `tinygo test`
+  - add `-C DIR` flag
+  - add initial documentation for project governance
+  - add `-ldflags='-extldflags=...'` support
+  - improve usage message with `tinygo help` and when passing invalid parameters
+* **compiler**
+  - `builder`: remove environment variables when invoking Clang, to avoid the environment changing the behavior
+  - `builder`: check for the Go toolchain version used to compile TinyGo
+  - `cgo`: add `C.CBytes` implementation
+  - `compiler`: fix passing weirdly-padded structs as parameters to new goroutines
+  - `compiler`: support pragmas on generic functions
+  - `compiler`: do not let the slice buffer escape when casting a `[]byte` or `[]rune` to a string, to help escape analysis
+  - `compiler`: conform to the latest iteration of the wasm types proposal
+  - `loader`: don't panic when main package is not named 'main'
+  - `loader`: make sure we always return type checker errors even without type errors
+  - `transform`: optimize range over `[]byte(string)`
+* **standard library**
+  - `crypto/x509`: add package stub to build crypto/x509 on macOS
+  - `machine/usb/adc/midi`: fix `PitchBend`
+  - `os`: add `Truncate` stub for baremetal
+  - `os`: add stubs for `os.File` deadlines
+  - `os`: add internal `net.newUnixFile` for the net package
+  - `runtime`: stub runtime_{Before,After}Exec for linkage
+  - `runtime`: randomize map accesses
+  - `runtime`: support `maps.Clone`
+  - `runtime`: add more fields to `MemStats`
+  - `runtime`: implement newcoro, coroswitch to support package iter
+  - `runtime`: disallow defer in interrupts
+  - `runtime`: add support for os/signal on Linux and MacOS
+  - `runtime`: add gc layout info for some basic types to help the precise GC
+  - `runtime`: bump GC mark stack size to avoid excessive heap rescans
+* **targets**
+  - `darwin`: use Go standard library syscall package instead of a custom one
+  - `fe310`: support GPIO `PinInput`
+  - `mips`: fix compiler crash with GOMIPS=softfloat and defer
+  - `mips`: add big-endian (GOARCH=mips) support
+  - `mips`: use MIPS32 (instead of MIPS32R2) as the instruction set for wider compatibility
+  - `wasi`: add relative and absolute --dir options to wasmtime args
+  - `wasip2`: add wasmtime -S args to support network interfaces
+  - `wasm`: add `//go:wasmexport` support (for all WebAssembly targets)
+  - `wasm`: use precise instead of conservative GC for WebAssembly (including WASI)
+  - `wasm-unknown`: add bulk memory flags since basically every runtime has it now
+* **boards**
+  - add RAKwireless RAK4631
+  - add WaveShare ESP-C3-32S-Kit
+
+
+0.33.0
+---
+
+* **general**
+  - use latest version of x/tools
+  - add chromeos 9p support for flashing
+  - sort compiler error messages by source position in a package
+  - don't include prebuilt libraries in the release to simplify packaging and reduce the release tarball size
+  - show runtime panic addresses for `tinygo run`
+  - support Go 1.23 (including all new language features)
+  - `test`: support GOOS/GOARCH pairs in the `-target` flag
+  - `test`: remove message after test binary built
+* **compiler**
+  - remove unused registers for x86_64 linux syscalls
+  - remove old atomics workaround for AVR (not necessary in modern LLVM versions)
+  - support `golang.org/x/sys/unix` syscalls
+  - `builder`: remove workaround for generics race condition
+  - `builder`: add package ID to compiler and optimization error messages
+  - `builder`: show better error messages for some common linker errors
+  - `cgo`: support preprocessor macros passed on the command line
+  - `cgo`: use absolute paths for error messages
+  - `cgo`: add support for printf
+  - `loader`: handle `go list` errors inside TinyGo (for better error messages)
+  - `transform`: fix incorrect alignment of heap-to-stack transform
+  - `transform`: use thinlto-pre-link passes (instead of the full pipeline) to speed up compilation speed slightly
+* **standard library**
+  - `crypto/tls`: add CipherSuiteName and some extra fields to ConnectionSTate
+  - `internal/abi`: implement initial version of this package
+  - `machine`: use new `internal/binary` package
+  - `machine`: rewrite Reply() to fix sending long replies in I2C Target Mode
+  - `machine/usb/descriptor`: Reset joystick physical
+  - `machine/usb/descriptor`: Drop second joystick hat
+  - `machine/usb/descriptor`: Add more HID... functions
+  - `machine/usb/descriptor`: Fix encoding of values
+  - `machine/usb/hid/joystick`: Allow more hat switches
+  - `os`: add `Chown`, `Truncate`
+  - `os/user`: use stdlib version of this package
+  - `reflect`: return correct name for the `unsafe.Pointer` type
+  - `reflect`: implement `Type.Overflow*` functions
+  - `runtime`: implement dummy `getAuxv` to satisfy golang.org/x/sys/
+  - `runtime`: don't zero out new allocations for `-gc=leaking` when they are already zeroed
+  - `runtime`: simplify slice growing/appending code
+  - `runtime`: print a message when a fatal signal like SIGSEGV happens
+  - `runtime/debug`: add `GoVersion` to `debug.BuildInfo`
+  - `sync`: add `Map.Clear()`
+  - `sync/atomic`: add And* and Or* compiler intrinsics needed for Go 1.23
+  - `syscall`: add `Fork` and `Execve`
+  - `syscall`: add all MacOS errno values
+  - `testing`: stub out `T.Deadline`
+  - `unique`: implement custom (naive) version of the unique package
+* **targets**
+  - `arm`: support `GOARM=*,softfloat` (softfloat support for ARM v5, v6, and v7)
+  - `mips`: add linux/mipsle (and experimental linux/mips) support
+  - `mips`: add `GOMIPS=softfloat` support
+  - `wasip2`: add WASI preview 2 support
+  - `wasm/js`: add `node:` prefix in `require()` call of wasm_exec.js
+  - `wasm-unknown`: make sure the `os` package can be imported
+  - `wasm-unknown`: remove import-memory flag
+
+
+0.32.0
+---
+
+* **general**
+  - fix wasi-libc include headers on Nix
+  - apply OpenOCD commands after target configuration
+  - fix a minor race condition when determining the build tags
+  - support UF2 drives with a space in their name on Linux
+  - add LLVM 18 support
+  - drop support for Go 1.18 to be able to stay up to date
+
+* **compiler**
+  - move `-panic=trap` support to the compiler/runtime
+  - fix symbol table index for WebAssembly archives
+  - fix ed25519 build errors by adjusting the alias names
+  - add aliases to generic AES functions
+  - fix race condition by temporarily applying a proposed patch
+  - `builder`: keep un-wasm-opt'd .wasm if -work was passed
+  - `builder`: make sure wasm-opt command line is printed if asked
+  - `cgo`: implement shift operations in preprocessor macros
+  - `interp`: checking for methodset existence
+
+* **standard library**
+  - `machine`: add `__tinygo_spi_tx` function to simulator
+  - `machine`: fix simulator I2C support
+  - `machine`: add GetRNG support to simulator
+  - `machine`: add `TxFifoFreeLevel` for CAN
+  - `os`: add `Link`
+  - `os`: add `FindProcess` for posix
+  - `os`: add `Process.Release` for unix
+  - `os`: add `SetReadDeadline` stub
+  - `os`, `os/signal`: add signal stubs
+  - `os/user`: add stubs for `Lookup{,Group}` and `Group`
+  - `reflect`: use int in `StringHeader` and `SliceHeader` on non-AVR platforms
+  - `reflect`: fix `NumMethods` for Interface type
+  - `runtime`: skip negative sleep durations in sleepTicks
+
+* **targets**
+  - `esp32`: add I2C support
+  - `rp2040`: move UART0 and UART1 to common file
+  - `rp2040`: make all RP2040 boards available for simulation
+  - `rp2040`: fix timeUnit type
+  - `stm32`: add i2c `Frequency` and `SetBaudRate` function for chips that were missing implementation
+  - `wasm-unknown`: add math and memory builtins that LLVM needs
+  - `wasip1`: replace existing `-target=wasi` support with wasip1 as supported in Go 1.21+
+
+* **boards**
+  - `adafruit-esp32-feather-v2`: add the Adafruit ESP32 Feather V2
+  - `badger2040-w`: add support for the Badger2040 W
+  - `feather-nrf52840-sense`: fix lack of LXFO
+  - `m5paper`: add support for the M5 Paper
+  - `mksnanov3`: limit programming speed to 1800 kHz
+  - `nucleol476rg`: add stm32 nucleol476rg support
+  - `pico-w`: add the Pico W (which is near-idential to the pico target)
+  - `thingplus-rp2040`, `waveshare-rp2040-zero`: add WS2812 definition
+  - `pca10059-s140v7`: add this variant to the PCA10059 board
+
+
+0.31.2
+---
+
+* **general**
+  * update the `net` submodule to updated version with `Buffers` implementation
+
+* **compiler**
+  * `syscall`: add wasm_unknown tag to some additional files so it can compile more code
+
+* **standard library**
+  * `runtime`: add Frame.Entry field
+
+
+0.31.1
+---
+
+* **general**
+  * fix Binaryen build in make task
+  * update final build stage of Docker `dev` image to go1.22
+  * only use GHA cache for building Docker `dev` image
+  * update the `net` submodule to latest version
+
+* **compiler**
+  * `interp`: make getelementptr offsets signed
+  * `interp`: return a proper error message when indexing out of range
+
+
+0.31.0
+---
+
+* **general**
+  * remove LLVM 14 support
+  * add LLVM 17 support, and use it by default
+  * add Nix flake support
+  * update bundled Binaryen to version 116
+  * add `ports` subcommand that lists available serial ports for `-port` and `-monitor`
+  * support wasmtime version 14
+  * add `-serial=rtt` for serial output over SWD
+  * add Go 1.22 support and use it by default
+  * change minimum Node.js version from 16 to 18
+* **compiler**
+  * use the new LLVM pass manager
+  * allow systems with more stack space to allocate larger values on the stack
+  * `build`: fix a crash due to sharing GlobalValues between build instances
+  * `cgo`: add `C._Bool` type
+  * `cgo`: fix calling CGo callback inside generic function
+  * `compileopts`: set `purego` build tag by default so that more packages can be built
+  * `compileopts`: force-enable CGo to avoid build issues
+  * `compiler`: fix crash on type assert on interfaces with no methods
+  * `interp`: print LLVM instruction in traceback
+  * `interp`: support runtime times by running them at runtime
+  * `loader`: enforce Go language version in the type checker (this may break existing programs with an incorrect Go version in go.mod)
+  * `transform`: fix bug in StringToBytes optimization pass
+* **standard library**
+  * `crypto/tls`: stub out a lot of functions
+  * `internal/task`, `machine`: make TinyGo code usable with "big Go" CGo
+  * `machine`: implement `I2C.SetBaudRate` consistently across chips
+  * `machine`: implement `SPI.Configure` consistently across chips
+  * `machine`: add `DeviceID` for nrf, rp2040, sam, stm32
+  * `machine`: use smaller UART buffer size on atmega chips
+  * `machine/usb`: allow setting a serial number using a linker flag
+  * `math`: support more math functions on baremetal (picolibc) systems
+  * `net`: replace entire net package with a new one based on the netdev driver
+  * `os/user`: add bare-bones implementation of this package
+  * `reflect`: stub `CallSlice` and `FuncOf`
+  * `reflect`: add `TypeFor[T]`
+  * `reflect`: update `IsZero` to Go 1.22 semantics
+  * `reflect`: move indirect values into interface when setting interfaces
+  * `runtime`: stub `Breakpoint`
+  * `sync`: implement trylock
+* **targets**
+  * `atmega`: use UART double speed mode for fewer errors and higher throughput
+  * `atmega328pb`: refactor to enable extra uart
+  * `avr`: don't compile large parts of picolibc (math, stdio) for LLVM 17 support
+  * `esp32`: switch over to the official SVD file
+  * `esp32c3`: implement USB_SERIAL for USBCDC communication
+  * `esp32c3`: implement I2C
+  * `esp32c3`: implement RNG
+  * `esp32c3`: add more ROM functions and update linker script for the in-progress wifi support
+  * `esp32c3`: update to newer SVD files
+  * `rp2040`: add support for UART hardware flow control
+  * `rp2040`: add definition for `machine.PinToggle`
+  * `rp2040`: set XOSC startup delay multiplier
+  * `samd21`: add support for UART hardware flow control
+  * `samd51`: add support for UART hardware flow control
+  * `wasm`: increase default stack size to 64k for wasi/wasm targets
+  * `wasm`: bump wasi-libc version to SDK 20
+  * `wasm`: remove line of dead code in wasm_exec.js
+* **new targets/boards**
+  * `qtpy-esp32c3`: add Adafruit QT Py ESP32-C3 board
+  * `mksnanov3`: add support for the MKS Robin Nano V3.x
+  * `nrf52840-generic`: add generic nrf52840 chip support
+  * `thumby`: add support for Thumby
+  * `wasm`: add new `wasm-unknown` target that doesn't depend on WASI or a browser
+* **boards**
+  * `arduino-mkrwifi1010`, `arduino-nano33`, `nano-rp2040`, `matrixportal-m4`, `metro-m4-airlift`, `pybadge`, `pyportal`: add `ninafw` build tag and some constants for BLE support
+  * `gopher-badge`: fix typo in USB product name
+  * `nano-rp2040`: add UART1 and correct mappings for NINA via UART
+  * `pico`: bump default stack size from 2kB to 8kB
+  * `wioterminal`: expose UART4
+
+
+0.30.0
+---
+
+* **general**
+  - add LLVM 16 support, use it by default
+* **compiler**
+  - `build`: work around a race condition by building Go SSA serially
+  - `compiler`: fix a crash by not using the LLVM global context types
+  - `interp`: don't copy unknown values in `runtime.sliceCopy` to fix miscompile
+  - `interp`: fix crash in error report by not returning raw LLVM values
+* **standard library**
+  - `machine/usb/adc/midi`: various improvements and API changes
+  - `reflect`: add support for `[...]T` → `[]T` in reflect
+* **targets**
+  - `atsamd21`, `atsamd51`: add support for USB INTERRUPT OUT
+  - `rp2040`: always use the USB device enumeration fix, even in chips that supposedly have the HW fix
+  - `wasm`: increase default stack size to 32k for wasi/wasm
+* **boards**
+  - `gobadge`: add GoBadge target as alias for PyBadge :)
+  - `gemma-m0`: add support for the Adafruit Gemma M0
+
+
+0.29.0
+---
+
+* **general**
+  - Go 1.21 support
+  - use https for renesas submodule #3856
+  - ci: rename release-double-zipped to something more useful
+  - ci: update Node.js from version 14 to version 16
+  - ci: switch GH actions builds to use Go 1.21 final release
+  - docker: update clang to version 15
+  - docker: use Go 1.21 for Docker dev container build
+  - `main`: add target JSON file in `tinygo info` output
+  - `main`: improve detection of filesystems
+  - `main`: use `go env` instead of doing all detection manually
+  - make: add make task to generate Renesas device wrappers
+  - make: add task to check NodeJS version before running tests
+  - add submodule for Renesas SVD file mirror repo
+  - update to go-serial package v1.6.0
+  - `testing`: add Testing function
+  - `tools/gen-device-svd`: small changes needed for Renesas MCUs
+* **compiler**
+  - `builder`: update message for max supported Go version
+  - `compiler,reflect`: NumMethods reports exported methods only
+  - `compiler`: add compiler-rt and wasm symbols to table
+  - `compiler`: add compiler-rt to wasm.json
+  - `compiler`: add min and max builtin support
+  - `compiler`: implement clear builtin for maps
+  - `compiler`: implement clear builtin for slices
+  - `compiler`: improve panic message when a runtime call is unavailable
+  - `compiler`: update .ll test output
+  - `loader`: merge go.env file which is now required starting in Go 1.21 to correctly get required packages
+* **standard library**
+  - `os`: define ErrNoDeadline
+  - `reflect`: Add FieldByNameFunc
+  - `reflect`: add SetZero
+  - `reflect`: fix iterating over maps with interface{} keys
+  - `reflect`: implement Value.Grow
+  - `reflect`: remove unnecessary heap allocations
+  - `reflect`: use .key() instead of a type assert
+  - `sync`: add implementation from upstream Go for OnceFunc, OnceValue, and OnceValues
+* **targets**
+  - `machine`: UART refactor (#3832)
+  - `machine/avr`: pin change interrupt
+  - `machine/macropad_rp2040`: add machine.BUTTON
+  - `machine/nrf`: add I2C timeout
+  - `machine/nrf`: wait for stop condition after reading from the I2C bus
+  - `machine/nRF52`: set SPI TX/RX lengths even data is empty. Fixes #3868 (#3877)
+  - `machine/rp2040`: add missing suffix to CMD_READ_STATUS
+  - `machine/rp2040`: add NoPin support
+  - `machine/rp2040`: move flash related functions into separate file from C imports for correct  - LSP. Fixes #3852
+  - `machine/rp2040`: wait for 1000 us after flash reset to avoid issues with busy USB bus
+  - `machine/samd51,rp2040,nrf528xx,stm32`: implement watchdog
+  - `machine/samd51`: fix i2cTimeout was decreasing due to cache activation
+  - `machine/usb`: Add support for HID Keyboard LEDs
+  - `machine/usb`: allow USB Endpoint settings to be changed externally
+  - `machine/usb`: refactor endpoint configuration
+  - `machine/usb`: remove usbDescriptorConfig
+  - `machine/usb/hid,joystick`: fix hidreport (3) (#3802)
+  - `machine/usb/hid`: add RxHandler interface
+  - `machine/usb/hid`: rename Handler() to TxHandler()
+  - `wasi`: allow zero inodes when reading directories
+  - `wasm`: add support for GOOS=wasip1
+  - `wasm`: fix functions exported through //export
+  - `wasm`: remove i64 workaround, use BigInt instead
+  - `example`: adjust time offset
+  - `example`: simplify pininterrupt
+* **boards**
+  - `targets`: add AKIZUKI DENSHI AE-RP2040
+  - `targets`: adding new uf2 target for PCA10056 (#3765)
+
+
 0.28.0
 ---
 
@@ -1643,7 +2143,7 @@
   - allow packages like github.com/tinygo-org/tinygo/src/\* by aliasing it
   - remove `//go:volatile` support  
     It has been replaced with the runtime/volatile package.
-  - allow poiners in map keys
+  - allow pointers in map keys
   - support non-constant syscall numbers
   - implement non-blocking selects
   - add support for the `-tags` flag

Dockerfile

@@ -1,10 +1,15 @@
 # tinygo-llvm stage obtains the llvm source for TinyGo
-FROM golang:1.20 AS tinygo-llvm
+FROM golang:1.24 AS tinygo-llvm
 
 RUN apt-get update && \
-    apt-get install -y apt-utils make cmake clang-11 ninja-build
+    apt-get install -y apt-utils make cmake clang-15 ninja-build && \
+    rm -rf \
+        /var/lib/apt/lists/* \
+        /var/log/* \
+        /var/tmp/* \
+        /tmp/*
 
-COPY ./Makefile /tinygo/Makefile
+COPY ./GNUmakefile /tinygo/GNUmakefile
 
 RUN cd /tinygo/ && \
     make llvm-source
@@ -15,26 +20,24 @@ FROM tinygo-llvm AS tinygo-llvm-build
 RUN cd /tinygo/ && \
     make llvm-build
 
-# tinygo-compiler stage builds the compiler itself
-FROM tinygo-llvm-build AS tinygo-compiler
+# tinygo-compiler-build stage builds the compiler itself
+FROM tinygo-llvm-build AS tinygo-compiler-build
 
 COPY . /tinygo
 
-# update submodules
+# build the compiler and tools
 RUN cd /tinygo/ && \
-    rm -rf ./lib/*/ && \
-    git submodule sync && \
-    git submodule update --init --recursive --force
-
-RUN cd /tinygo/ && \
-    make
-
-# tinygo-tools stage installs the needed dependencies to compile TinyGo programs for all platforms.
-FROM tinygo-compiler AS tinygo-tools
-
-RUN cd /tinygo/ && \
-    make wasi-libc binaryen && \
+    git submodule update --init && \
     make gen-device -j4 && \
-    cp build/* $GOPATH/bin/
+    make build/release
 
+# tinygo-compiler copies the compiler build over to a base Go container (without
+# all the build tools etc).
+FROM golang:1.24 AS tinygo-compiler
+
+# Copy tinygo build.
+COPY --from=tinygo-compiler-build /tinygo/build/release/tinygo /tinygo
+
+# Configure the container.
+ENV PATH="${PATH}:/tinygo/bin"
 CMD ["tinygo"]

GNUmakefile

@@ -10,7 +10,7 @@ LLD_SRC ?= $(LLVM_PROJECTDIR)/lld
 
 # Try to autodetect LLVM build tools.
 # Versions are listed here in descending priority order.
-LLVM_VERSIONS = 15 14 13 12 11
+LLVM_VERSIONS = 19 18 17 16 15
 errifempty = $(if $(1),$(1),$(error $(2)))
 detect = $(shell which $(call errifempty,$(firstword $(foreach p,$(2),$(shell command -v $(p) 2> /dev/null && echo $(p)))),failed to locate $(1) at any of: $(2)))
 toolSearchPathsVersion = $(1)-$(2)
@@ -31,9 +31,7 @@ export GOROOT = $(shell $(GO) env GOROOT)
 
 # Flags to pass to go test.
 GOTESTFLAGS ?=
-
-# md5sum binary
-MD5SUM = md5sum
+GOTESTPKGS ?= ./builder ./cgo ./compileopts ./compiler ./interp ./transform .
 
 # tinygo binary for tests
 TINYGO ?= $(call detect,tinygo,tinygo $(CURDIR)/build/tinygo)
@@ -113,7 +111,7 @@ endif
 
 .PHONY: all tinygo test $(LLVM_BUILDDIR) llvm-source clean fmt gen-device gen-device-nrf gen-device-nxp gen-device-avr gen-device-rp
 
-LLVM_COMPONENTS = all-targets analysis asmparser asmprinter bitreader bitwriter codegen core coroutines coverage debuginfodwarf debuginfopdb executionengine frontendopenmp instrumentation interpreter ipo irreader libdriver linker lto mc mcjit objcarcopts option profiledata scalaropts support target windowsdriver windowsmanifest
+LLVM_COMPONENTS = all-targets analysis asmparser asmprinter bitreader bitwriter codegen core coroutines coverage debuginfodwarf debuginfopdb executionengine frontenddriver frontendhlsl frontendopenmp instrumentation interpreter ipo irreader libdriver linker lto mc mcjit objcarcopts option profiledata scalaropts support target windowsdriver windowsmanifest
 
 ifeq ($(OS),Windows_NT)
     EXE = .exe
@@ -130,14 +128,14 @@ ifeq ($(OS),Windows_NT)
     USE_SYSTEM_BINARYEN ?= 1
 
 else ifeq ($(shell uname -s),Darwin)
-    MD5SUM = md5
+    MD5SUM ?= md5
 
     CGO_LDFLAGS += -lxar
 
     USE_SYSTEM_BINARYEN ?= 1
 
 else ifeq ($(shell uname -s),FreeBSD)
-    MD5SUM = md5
+    MD5SUM ?= md5
     START_GROUP = -Wl,--start-group
     END_GROUP = -Wl,--end-group
 else
@@ -145,8 +143,11 @@ else
     END_GROUP = -Wl,--end-group
 endif
 
+# md5sum binary default, can be overridden by an environment variable
+MD5SUM ?= md5sum
+
 # Libraries that should be linked in for the statically linked Clang.
-CLANG_LIB_NAMES = clangAnalysis clangAST clangASTMatchers clangBasic clangCodeGen clangCrossTU clangDriver clangDynamicASTMatchers clangEdit clangExtractAPI clangFormat clangFrontend clangFrontendTool clangHandleCXX clangHandleLLVM clangIndex clangLex clangParse clangRewrite clangRewriteFrontend clangSema clangSerialization clangSupport clangTooling clangToolingASTDiff clangToolingCore clangToolingInclusions
+CLANG_LIB_NAMES = clangAnalysis clangAPINotes clangAST clangASTMatchers clangBasic clangCodeGen clangCrossTU clangDriver clangDynamicASTMatchers clangEdit clangExtractAPI clangFormat clangFrontend clangFrontendTool clangHandleCXX clangHandleLLVM clangIndex clangInstallAPI clangLex clangParse clangRewrite clangRewriteFrontend clangSema clangSerialization clangSupport clangTooling clangToolingASTDiff clangToolingCore clangToolingInclusions
 CLANG_LIBS = $(START_GROUP) $(addprefix -l,$(CLANG_LIB_NAMES)) $(END_GROUP) -lstdc++
 
 # Libraries that should be linked in for the statically linked LLD.
@@ -154,7 +155,7 @@ LLD_LIB_NAMES = lldCOFF lldCommon lldELF lldMachO lldMinGW lldWasm
 LLD_LIBS = $(START_GROUP) $(addprefix -l,$(LLD_LIB_NAMES)) $(END_GROUP)
 
 # Other libraries that are needed to link TinyGo.
-EXTRA_LIB_NAMES = LLVMInterpreter LLVMMCA LLVMX86TargetMCA
+EXTRA_LIB_NAMES = LLVMInterpreter LLVMMCA LLVMRISCVTargetMCA LLVMX86TargetMCA
 
 # All libraries to be built and linked with the tinygo binary (lib/lib*.a).
 LIB_NAMES = clang $(CLANG_LIB_NAMES) $(LLD_LIB_NAMES) $(EXTRA_LIB_NAMES)
@@ -165,26 +166,26 @@ LIB_NAMES = clang $(CLANG_LIB_NAMES) $(LLD_LIB_NAMES) $(EXTRA_LIB_NAMES)
 # library path (for ninja).
 # This list also includes a few tools that are necessary as part of the full
 # TinyGo build.
-NINJA_BUILD_TARGETS = clang llvm-config llvm-ar llvm-nm $(addprefix lib/lib,$(addsuffix .a,$(LIB_NAMES)))
+NINJA_BUILD_TARGETS = clang llvm-config llvm-ar llvm-nm lld $(addprefix lib/lib,$(addsuffix .a,$(LIB_NAMES)))
 
 # For static linking.
 ifneq ("$(wildcard $(LLVM_BUILDDIR)/bin/llvm-config*)","")
     CGO_CPPFLAGS+=$(shell $(LLVM_CONFIG_PREFIX) $(LLVM_BUILDDIR)/bin/llvm-config --cppflags) -I$(abspath $(LLVM_BUILDDIR))/tools/clang/include -I$(abspath $(CLANG_SRC))/include -I$(abspath $(LLD_SRC))/include
-    CGO_CXXFLAGS=-std=c++14
+    CGO_CXXFLAGS=-std=c++17
     CGO_LDFLAGS+=-L$(abspath $(LLVM_BUILDDIR)/lib) -lclang $(CLANG_LIBS) $(LLD_LIBS) $(shell $(LLVM_CONFIG_PREFIX) $(LLVM_BUILDDIR)/bin/llvm-config --ldflags --libs --system-libs $(LLVM_COMPONENTS)) -lstdc++ $(CGO_LDFLAGS_EXTRA)
 endif
 
-clean:
+clean: ## Remove build directory
 	@rm -rf build
 
 FMT_PATHS = ./*.go builder cgo/*.go compiler interp loader src transform
-fmt:
+fmt: ## Reformat source
 	@gofmt -l -w $(FMT_PATHS)
-fmt-check:
+fmt-check: ## Warn if any source needs reformatting
 	@unformatted=$$(gofmt -l $(FMT_PATHS)); [ -z "$$unformatted" ] && exit 0; echo "Unformatted:"; for fn in $$unformatted; do echo "  $$fn"; done; exit 1
 
 
-gen-device: gen-device-avr gen-device-esp gen-device-nrf gen-device-sam gen-device-sifive gen-device-kendryte gen-device-nxp gen-device-rp
+gen-device: gen-device-avr gen-device-esp gen-device-nrf gen-device-sam gen-device-sifive gen-device-kendryte gen-device-nxp gen-device-rp gen-device-renesas ## Generate microcontroller-specific sources
 ifneq ($(STM32), 0)
 gen-device: gen-device-stm32
 endif
@@ -232,18 +233,20 @@ gen-device-rp: build/gen-device-svd
 	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/RaspberryPi lib/cmsis-svd/data/RaspberryPi/ src/device/rp/
 	GO111MODULE=off $(GO) fmt ./src/device/rp
 
-# Get LLVM sources.
+gen-device-renesas: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/cmsis-svd/cmsis-svd-data/tree/master/data/Renesas lib/cmsis-svd/data/Renesas/ src/device/renesas/
+	GO111MODULE=off $(GO) fmt ./src/device/renesas
+
 $(LLVM_PROJECTDIR)/llvm:
-	git clone -b xtensa_release_15.x --depth=1 https://github.com/espressif/llvm-project $(LLVM_PROJECTDIR)
-llvm-source: $(LLVM_PROJECTDIR)/llvm
+	git clone -b xtensa_release_19.1.2 --depth=1 https://github.com/espressif/llvm-project $(LLVM_PROJECTDIR)
+llvm-source: $(LLVM_PROJECTDIR)/llvm ## Get LLVM sources
 
 # Configure LLVM.
 TINYGO_SOURCE_DIR=$(shell pwd)
 $(LLVM_BUILDDIR)/build.ninja:
-	mkdir -p $(LLVM_BUILDDIR) && cd $(LLVM_BUILDDIR) && cmake -G Ninja $(TINYGO_SOURCE_DIR)/$(LLVM_PROJECTDIR)/llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;RISCV;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=AVR;Xtensa" -DCMAKE_BUILD_TYPE=Release -DLIBCLANG_BUILD_STATIC=ON -DLLVM_ENABLE_TERMINFO=OFF -DLLVM_ENABLE_ZLIB=OFF -DLLVM_ENABLE_ZSTD=OFF -DLLVM_ENABLE_LIBEDIT=OFF -DLLVM_ENABLE_Z3_SOLVER=OFF -DLLVM_ENABLE_OCAMLDOC=OFF -DLLVM_ENABLE_LIBXML2=OFF -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_TOOL_CLANG_TOOLS_EXTRA_BUILD=OFF -DCLANG_ENABLE_STATIC_ANALYZER=OFF -DCLANG_ENABLE_ARCMT=OFF $(LLVM_OPTION)
+	mkdir -p $(LLVM_BUILDDIR) && cd $(LLVM_BUILDDIR) && cmake -G Ninja $(TINYGO_SOURCE_DIR)/$(LLVM_PROJECTDIR)/llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;AVR;Mips;RISCV;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=Xtensa" -DCMAKE_BUILD_TYPE=Release -DLIBCLANG_BUILD_STATIC=ON -DLLVM_ENABLE_TERMINFO=OFF -DLLVM_ENABLE_ZLIB=OFF -DLLVM_ENABLE_ZSTD=OFF -DLLVM_ENABLE_LIBEDIT=OFF -DLLVM_ENABLE_Z3_SOLVER=OFF -DLLVM_ENABLE_OCAMLDOC=OFF -DLLVM_ENABLE_LIBXML2=OFF -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_TOOL_CLANG_TOOLS_EXTRA_BUILD=OFF -DCLANG_ENABLE_STATIC_ANALYZER=OFF -DCLANG_ENABLE_ARCMT=OFF $(LLVM_OPTION)
 
-# Build LLVM.
-$(LLVM_BUILDDIR): $(LLVM_BUILDDIR)/build.ninja
+$(LLVM_BUILDDIR): $(LLVM_BUILDDIR)/build.ninja ## Build LLVM
 	cd $(LLVM_BUILDDIR) && ninja $(NINJA_BUILD_TARGETS)
 
 ifneq ($(USE_SYSTEM_BINARYEN),1)
@@ -252,7 +255,7 @@ ifneq ($(USE_SYSTEM_BINARYEN),1)
 binaryen: build/wasm-opt$(EXE)
 build/wasm-opt$(EXE):
 	mkdir -p build
-	cd lib/binaryen && cmake -G Ninja . -DBUILD_STATIC_LIB=ON $(BINARYEN_OPTION) && ninja bin/wasm-opt$(EXE)
+	cd lib/binaryen && cmake -G Ninja . -DBUILD_STATIC_LIB=ON -DBUILD_TESTS=OFF -DENABLE_WERROR=OFF $(BINARYEN_OPTION) && ninja bin/wasm-opt$(EXE)
 	cp lib/binaryen/bin/wasm-opt$(EXE) build/wasm-opt$(EXE)
 endif
 
@@ -261,15 +264,36 @@ endif
 wasi-libc: lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a
 lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a:
 	@if [ ! -e lib/wasi-libc/Makefile ]; then echo "Submodules have not been downloaded. Please download them using:\n  git submodule update --init"; exit 1; fi
-	cd lib/wasi-libc && make -j4 EXTRA_CFLAGS="-O2 -g -DNDEBUG -mnontrapping-fptoint -msign-ext" MALLOC_IMPL=none CC=$(CLANG) AR=$(LLVM_AR) NM=$(LLVM_NM)
+	cd lib/wasi-libc && $(MAKE) -j4 EXTRA_CFLAGS="-O2 -g -DNDEBUG -mnontrapping-fptoint -msign-ext" MALLOC_IMPL=none CC="$(CLANG)" AR=$(LLVM_AR) NM=$(LLVM_NM)
 
+# Generate WASI syscall bindings
+WASM_TOOLS_MODULE=github.com/bytecodealliance/wasm-tools-go
+.PHONY: wasi-syscall
+wasi-syscall: wasi-cm
+	go run -modfile ./internal/wasm-tools/go.mod $(WASM_TOOLS_MODULE)/cmd/wit-bindgen-go generate --versioned -o ./src/internal -p internal --cm internal/cm ./lib/wasi-cli/wit
 
-# Build the Go compiler.
-tinygo:
-	@if [ ! -f "$(LLVM_BUILDDIR)/bin/llvm-config" ]; then echo "Fetch and build LLVM first by running:"; echo "  make llvm-source"; echo "  make $(LLVM_BUILDDIR)"; exit 1; fi
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOENVFLAGS) $(GO) build -buildmode exe -o build/tinygo$(EXE) -tags "byollvm osusergo" -ldflags="-X github.com/tinygo-org/tinygo/goenv.GitSha1=`git rev-parse --short HEAD`" .
-test: wasi-libc
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=20m -buildmode exe -tags "byollvm osusergo" ./builder ./cgo ./compileopts ./compiler ./interp ./transform .
+# Copy package cm into src/internal/cm
+.PHONY: wasi-cm
+wasi-cm:
+	# rm -rf ./src/internal/cm
+	rsync -rv --delete --exclude '*_test.go' $(shell go list -modfile ./internal/wasm-tools/go.mod -m -f {{.Dir}} $(WASM_TOOLS_MODULE))/cm ./src/internal/
+
+# Check for Node.js used during WASM tests.
+NODEJS_VERSION := $(word 1,$(subst ., ,$(shell node -v | cut -c 2-)))
+MIN_NODEJS_VERSION=18
+
+.PHONY: check-nodejs-version
+check-nodejs-version:
+ifeq (, $(shell which node))
+	@echo "Install NodeJS version 18+ to run tests."; exit 1;
+endif
+	@if [ $(NODEJS_VERSION) -lt $(MIN_NODEJS_VERSION) ]; then echo "Install NodeJS version 18+ to run tests."; exit 1; fi
+
+tinygo: ## Build the TinyGo compiler
+	@if [ ! -f "$(LLVM_BUILDDIR)/bin/llvm-config" ]; then echo "Fetch and build LLVM first by running:"; echo "  $(MAKE) llvm-source"; echo "  $(MAKE) $(LLVM_BUILDDIR)"; exit 1; fi
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOENVFLAGS) $(GO) build -buildmode exe -o build/tinygo$(EXE) -tags "byollvm osusergo" .
+test: wasi-libc check-nodejs-version
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags "byollvm osusergo" $(GOTESTPKGS)
 
 # Standard library packages that pass tests on darwin, linux, wasi, and windows, but take over a minute in wasi
 TEST_PACKAGES_SLOW = \
@@ -279,26 +303,32 @@ TEST_PACKAGES_SLOW = \
 
 # Standard library packages that pass tests quickly on darwin, linux, wasi, and windows
 TEST_PACKAGES_FAST = \
+	cmp \
 	compress/lzw \
 	compress/zlib \
 	container/heap \
 	container/list \
 	container/ring \
-	crypto/des \
+	crypto/ecdsa \
+	crypto/elliptic \
 	crypto/md5 \
-	crypto/rc4 \
 	crypto/sha1 \
 	crypto/sha256 \
 	crypto/sha512 \
+	database/sql/driver \
 	debug/macho \
 	embed/internal/embedtest \
 	encoding \
 	encoding/ascii85 \
+	encoding/asn1 \
 	encoding/base32 \
 	encoding/base64 \
 	encoding/csv \
 	encoding/hex \
+	go/ast \
+	go/format \
 	go/scanner \
+	go/version \
 	hash \
 	hash/adler32 \
 	hash/crc64 \
@@ -320,49 +350,58 @@ TEST_PACKAGES_FAST = \
 	unicode \
 	unicode/utf16 \
 	unicode/utf8 \
+	unique \
 	$(nil)
 
-# Assume this will go away before Go2, so only check minor version.
-ifeq ($(filter $(shell $(GO) env GOVERSION | cut -f 2 -d.), 16 17 18), )
-TEST_PACKAGES_FAST += crypto/internal/nistec/fiat
-else
-TEST_PACKAGES_FAST += crypto/elliptic/internal/fiat
-endif
-
 # archive/zip requires os.ReadAt, which is not yet supported on windows
 # bytes requires mmap
 # compress/flate appears to hang on wasi
+# crypto/aes fails on wasi, needs panic()/recover()
+# crypto/des fails on wasi, needs panic()/recover()
 # crypto/hmac fails on wasi, it exits with a "slice out of range" panic
 # debug/plan9obj requires os.ReadAt, which is not yet supported on windows
-# image requires recover(), which is not  yet supported on wasi
+# image requires recover(), which is not yet supported on wasi
 # io/ioutil requires os.ReadDir, which is not yet supported on windows or wasi
+# mime: fail on wasi; neds panic()/recover()
+# mime/multipart: needs wasip1 syscall.FDFLAG_NONBLOCK
 # mime/quotedprintable requires syscall.Faccessat
+# net/mail: needs wasip1  syscall.FDFLAG_NONBLOCK
+# net/ntextproto: needs wasip1 syscall.FDFLAG_NONBLOCK
+# regexp/syntax: fails on wasip1; needs panic()/recover()
 # strconv requires recover() which is not yet supported on wasi
-# text/tabwriter requries recover(), which is not  yet supported on wasi
+# text/tabwriter requires recover(), which is not yet supported on wasi
 # text/template/parse requires recover(), which is not yet supported on wasi
 # testing/fstest requires os.ReadDir, which is not yet supported on windows or wasi
 
 # Additional standard library packages that pass tests on individual platforms
 TEST_PACKAGES_LINUX := \
 	archive/zip \
-	bytes \
 	compress/flate \
+	crypto/aes \
+	crypto/des \
 	crypto/hmac \
 	debug/dwarf \
 	debug/plan9obj \
 	image \
 	io/ioutil \
+	mime \
+	mime/multipart \
 	mime/quotedprintable \
 	net \
+	net/mail \
+	net/textproto \
+	os/user \
+	regexp/syntax \
 	strconv \
-	testing/fstest \
 	text/tabwriter \
 	text/template/parse
 
 TEST_PACKAGES_DARWIN := $(TEST_PACKAGES_LINUX)
 
+# os/user requires t.Skip() support
 TEST_PACKAGES_WINDOWS := \
 	compress/flate \
+	crypto/des \
 	crypto/hmac \
 	strconv \
 	text/template/parse \
@@ -412,13 +451,38 @@ tinygo-bench-fast:
 
 # Same thing, except for wasi rather than the current platform.
 tinygo-test-wasi:
-	$(TINYGO) test -target wasi $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_SLOW) ./tests/runtime_wasi
-tinygo-test-wasi-fast:
-	$(TINYGO) test -target wasi $(TEST_PACKAGES_FAST) ./tests/runtime_wasi
-tinygo-bench-wasi:
-	$(TINYGO) test -target wasi -bench . $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_SLOW)
-tinygo-bench-wasi-fast:
-	$(TINYGO) test -target wasi -bench . $(TEST_PACKAGES_FAST)
+	$(TINYGO) test -target wasip1 $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_SLOW) ./tests/runtime_wasi
+tinygo-test-wasip1:
+	GOOS=wasip1 GOARCH=wasm $(TINYGO) test $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_SLOW) ./tests/runtime_wasi
+tinygo-test-wasip1-fast:
+	$(TINYGO) test -target=wasip1 $(TEST_PACKAGES_FAST) ./tests/runtime_wasi
+
+tinygo-test-wasip2-slow:
+	$(TINYGO) test -target=wasip2 $(TEST_PACKAGES_SLOW)
+tinygo-test-wasip2-fast:
+	$(TINYGO) test -target=wasip2 $(TEST_PACKAGES_FAST) ./tests/runtime_wasi
+
+tinygo-test-wasip2-sum-slow:
+	TINYGO=$(TINYGO) \
+	TARGET=wasip2 \
+	TESTOPTS="-x -work" \
+	PACKAGES="$(TEST_PACKAGES_SLOW)" \
+	gotestsum --raw-command -- ./tools/tgtestjson.sh
+tinygo-test-wasip2-sum-fast:
+	TINYGO=$(TINYGO) \
+	TARGET=wasip2 \
+	TESTOPTS="-x -work" \
+	PACKAGES="$(TEST_PACKAGES_FAST)" \
+	gotestsum --raw-command -- ./tools/tgtestjson.sh
+tinygo-bench-wasip1:
+	$(TINYGO) test -target wasip1 -bench . $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_SLOW)
+tinygo-bench-wasip1-fast:
+	$(TINYGO) test -target wasip1 -bench . $(TEST_PACKAGES_FAST)
+
+tinygo-bench-wasip2:
+	$(TINYGO) test -target wasip2 -bench . $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_SLOW)
+tinygo-bench-wasip2-fast:
+	$(TINYGO) test -target wasip2 -bench . $(TEST_PACKAGES_FAST)
 
 # Test external packages in a large corpus.
 test-corpus:
@@ -426,15 +490,26 @@ test-corpus:
 test-corpus-fast:
 	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus -short . -corpus=testdata/corpus.yaml
 test-corpus-wasi: wasi-libc
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus . -corpus=testdata/corpus.yaml -target=wasi
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus . -corpus=testdata/corpus.yaml -target=wasip1
+test-corpus-wasip2: wasi-libc
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus . -corpus=testdata/corpus.yaml -target=wasip2
 
 tinygo-baremetal:
 	# Regression tests that run on a baremetal target and don't fit in either main_test.go or smoketest.
 	# regression test for #2666: e.g. encoding/hex must pass on baremetal
 	$(TINYGO) test -target cortex-m-qemu encoding/hex
 
+.PHONY: testchdir
+testchdir:
+	# test 'build' command with{,out} -C argument
+	$(TINYGO) build -C tests/testing/chdir chdir.go && rm tests/testing/chdir/chdir
+	$(TINYGO) build ./tests/testing/chdir/chdir.go && rm chdir
+	# test 'run' command with{,out} -C argument
+	EXPECT_DIR=$(PWD)/tests/testing/chdir $(TINYGO) run -C tests/testing/chdir chdir.go
+	EXPECT_DIR=$(PWD) $(TINYGO) run ./tests/testing/chdir/chdir.go
+
 .PHONY: smoketest
-smoketest:
+smoketest: testchdir
 	$(TINYGO) version
 	$(TINYGO) targets > /dev/null
 	# regression test for #2892
@@ -444,6 +519,8 @@ smoketest:
 	# regression test for #2563
 	cd tests/os/smoke && $(TINYGO) test -c -target=pybadge && rm smoke.test
 	# test all examples (except pwm)
+	$(TINYGO) build -size short -o test.hex -target=pga2350             examples/echo
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/adc
@@ -472,33 +549,41 @@ smoketest:
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nano-rp2040         examples/rtcinterrupt
 	@$(MD5SUM) test.hex
-	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/serial
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/machinetest
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/systick
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/test
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/time-offset
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=wioterminal         examples/hid-mouse
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=wioterminal         examples/hid-keyboard
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=feather-rp2040      examples/i2c-target
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-rp2040      examples/watchdog
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-rp2040      examples/device-id
+	@$(MD5SUM) test.hex
 	# test simulated boards on play.tinygo.org
 ifneq ($(WASM), 0)
-	$(TINYGO) build -size short -o test.wasm -tags=arduino              examples/blinky1
+	GOOS=js GOARCH=wasm $(TINYGO) build -size short -o test.wasm -tags=arduino              examples/blinky1
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build -size short -o test.wasm -tags=hifive1b             examples/blinky1
+	GOOS=js GOARCH=wasm $(TINYGO) build -size short -o test.wasm -tags=hifive1b             examples/blinky1
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build -size short -o test.wasm -tags=reelboard            examples/blinky1
+	GOOS=js GOARCH=wasm $(TINYGO) build -size short -o test.wasm -tags=reelboard            examples/blinky1
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build -size short -o test.wasm -tags=microbit             examples/microbit-blink
+	GOOS=js GOARCH=wasm $(TINYGO) build -size short -o test.wasm -tags=microbit             examples/microbit-blink
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build -size short -o test.wasm -tags=circuitplay_express  examples/blinky1
+	GOOS=js GOARCH=wasm $(TINYGO) build -size short -o test.wasm -tags=circuitplay_express  examples/blinky1
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build -size short -o test.wasm -tags=circuitplay_bluefruit examples/blinky1
+	GOOS=js GOARCH=wasm $(TINYGO) build -size short -o test.wasm -tags=circuitplay_bluefruit examples/blinky1
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build -size short -o test.wasm -tags=mch2022              examples/serial
+	GOOS=js GOARCH=wasm $(TINYGO) build -size short -o test.wasm -tags=mch2022              examples/machinetest
+	@$(MD5SUM) test.wasm
+	GOOS=js GOARCH=wasm $(TINYGO) build -size short -o test.wasm -tags=gopher_badge         examples/blinky1
 	@$(MD5SUM) test.wasm
 endif
 	# test all targets/boards
@@ -536,6 +621,8 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=trinket-m0          examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=gemma-m0            examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=circuitplay-express examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=circuitplay-bluefruit examples/blinky1
@@ -572,6 +659,8 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10056-s140v7     examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10059-s140v7     examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=reelboard-s140v7    examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=wioterminal         examples/blinky1
@@ -580,6 +669,8 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=xiao                examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=rak4631             examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=circuitplay-express examples/dac
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pyportal            examples/dac
@@ -590,7 +681,7 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=itsybitsy-nrf52840  examples/blinky1
 	@$(MD5SUM) test.hex
-	$(TINYGO) build -size short -o test.hex -target=qtpy                examples/serial
+	$(TINYGO) build -size short -o test.hex -target=qtpy                examples/machinetest
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=teensy41            examples/blinky1
 	@$(MD5SUM) test.hex
@@ -628,6 +719,8 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=badger2040          examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=badger2040-w        examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=tufty2040           examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=thingplus-rp2040    examples/blinky1
@@ -642,6 +735,18 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=gopher-badge      examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=ae-rp2040           examples/echo
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=thumby              examples/echo
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pico2               examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=tiny2350            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pico-plus2          examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=waveshare-rp2040-tiny examples/echo
+	@$(MD5SUM) test.hex
 	# test pwm
 	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m0        examples/pwm
 	@$(MD5SUM) test.hex
@@ -656,6 +761,8 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=feather-nrf52840    examples/usb-midi
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nrf52840-s140v6-uf2-generic	examples/machinetest
+	@$(MD5SUM) test.hex
 ifneq ($(STM32), 0)
 	$(TINYGO) build -size short -o test.hex -target=bluepill            examples/blinky1
 	@$(MD5SUM) test.hex
@@ -671,6 +778,8 @@ ifneq ($(STM32), 0)
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nucleo-l432kc       examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nucleo-l476rg       examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nucleo-l552ze       examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nucleo-wl55jc       examples/blinky1
@@ -689,8 +798,12 @@ ifneq ($(STM32), 0)
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=swan                examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=mksnanov3           examples/blinky1
+	@$(MD5SUM) test.hex
 endif
-	$(TINYGO) build -size short -o test.hex -target=atmega1284p         examples/serial
+	$(TINYGO) build -size short -o test.hex -target=atmega328pb         examples/blinkm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=atmega1284p         examples/machinetest
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=arduino             examples/blinky1
 	@$(MD5SUM) test.hex
@@ -715,32 +828,53 @@ endif
 ifneq ($(XTENSA), 0)
 	$(TINYGO) build -size short -o test.bin -target=esp32-mini32      	examples/blinky1
 	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target=esp32c3-supermini   examples/blinky1
+	@$(MD5SUM) test.bin
 	$(TINYGO) build -size short -o test.bin -target=nodemcu             examples/blinky1
 	@$(MD5SUM) test.bin
-	$(TINYGO) build -size short -o test.bin -target m5stack-core2       examples/serial
+	$(TINYGO) build -size short -o test.bin -target m5stack-core2       examples/machinetest
 	@$(MD5SUM) test.bin
-	$(TINYGO) build -size short -o test.bin -target m5stack             examples/serial
+	$(TINYGO) build -size short -o test.bin -target m5stack             examples/machinetest
 	@$(MD5SUM) test.bin
-	$(TINYGO) build -size short -o test.bin -target m5stick-c           examples/serial
+	$(TINYGO) build -size short -o test.bin -target m5stick-c           examples/machinetest
 	@$(MD5SUM) test.bin
-	$(TINYGO) build -size short -o test.bin -target mch2022             examples/serial
+	$(TINYGO) build -size short -o test.bin -target m5paper             examples/machinetest
+	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target mch2022             examples/machinetest
 	@$(MD5SUM) test.bin
 endif
-	$(TINYGO) build -size short -o test.bin -target=esp32c3             examples/serial
+	$(TINYGO) build -size short -o test.bin -target=esp-c3-32s-kit      examples/blinky1
 	@$(MD5SUM) test.bin
-	$(TINYGO) build -size short -o test.bin -target=esp32c3-12f         examples/serial
+	$(TINYGO) build -size short -o test.bin -target=qtpy-esp32c3        examples/machinetest
 	@$(MD5SUM) test.bin
-	$(TINYGO) build -size short -o test.bin -target=m5stamp-c3          examples/serial
+	$(TINYGO) build -size short -o test.bin -target=m5stamp-c3          examples/machinetest
 	@$(MD5SUM) test.bin
-	$(TINYGO) build -size short -o test.bin -target=xiao-esp32c3        examples/serial
+	$(TINYGO) build -size short -o test.bin -target=xiao-esp32c3        examples/machinetest
+	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target=esp32-c3-devkit-rust-1 examples/blinky1
+	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target=esp32c3-12f         examples/blinky1
+	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target=makerfabs-esp32c3spi35 examples/machinetest
 	@$(MD5SUM) test.bin
 	$(TINYGO) build -size short -o test.hex -target=hifive1b            examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=maixbit             examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=tkey                examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=elecrow-rp2040      examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=elecrow-rp2350      examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=hw-651              examples/machinetest
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=hw-651-s110v8       examples/machinetest
+	@$(MD5SUM) test.hex	
 ifneq ($(WASM), 0)
 	$(TINYGO) build -size short -o wasm.wasm -target=wasm               examples/wasm/export
 	$(TINYGO) build -size short -o wasm.wasm -target=wasm               examples/wasm/main
+	$(TINYGO) build -size short -o wasm.wasm -target=wasm-unknown       examples/hello-wasm-unknown
 endif
 	# test various compiler flags
 	$(TINYGO) build -size short -o test.hex -target=pca10040 -gc=none -scheduler=none examples/blinky1
@@ -749,11 +883,14 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10040 -serial=none examples/echo
 	@$(MD5SUM) test.hex
-	$(TINYGO) build             -o test.nro -target=nintendoswitch      examples/serial
+	$(TINYGO) build -size short -o test.hex -target=pca10040 -serial=rtt examples/echo
+	@$(MD5SUM) test.hex
+	$(TINYGO) build             -o test.nro -target=nintendoswitch      examples/echo2
 	@$(MD5SUM) test.nro
 	$(TINYGO) build -size short -o test.hex -target=pca10040 -opt=0     ./testdata/stdlib.go
 	@$(MD5SUM) test.hex
 	GOOS=linux GOARCH=arm $(TINYGO) build -size short -o test.elf       ./testdata/cgo
+	GOOS=linux GOARCH=mips    $(TINYGO) build -size short -o test.elf   ./testdata/cgo
 	GOOS=windows GOARCH=amd64 $(TINYGO) build -size short -o test.exe   ./testdata/cgo
 	GOOS=windows GOARCH=arm64 $(TINYGO) build -size short -o test.exe   ./testdata/cgo
 	GOOS=darwin GOARCH=amd64 $(TINYGO) build  -size short -o test       ./testdata/cgo
@@ -774,6 +911,7 @@ build/release: tinygo gen-device wasi-libc $(if $(filter 1,$(USE_SYSTEM_BINARYEN
 	@mkdir -p build/release/tinygo/lib/CMSIS/CMSIS
 	@mkdir -p build/release/tinygo/lib/macos-minimal-sdk
 	@mkdir -p build/release/tinygo/lib/mingw-w64/mingw-w64-crt/lib-common
+	@mkdir -p build/release/tinygo/lib/mingw-w64/mingw-w64-crt/stdio
 	@mkdir -p build/release/tinygo/lib/mingw-w64/mingw-w64-headers/defaults
 	@mkdir -p build/release/tinygo/lib/musl/arch
 	@mkdir -p build/release/tinygo/lib/musl/crt
@@ -781,10 +919,10 @@ build/release: tinygo gen-device wasi-libc $(if $(filter 1,$(USE_SYSTEM_BINARYEN
 	@mkdir -p build/release/tinygo/lib/nrfx
 	@mkdir -p build/release/tinygo/lib/picolibc/newlib/libc
 	@mkdir -p build/release/tinygo/lib/picolibc/newlib/libm
-	@mkdir -p build/release/tinygo/lib/wasi-libc
-	@mkdir -p build/release/tinygo/pkg/thumbv6m-unknown-unknown-eabi-cortex-m0
-	@mkdir -p build/release/tinygo/pkg/thumbv6m-unknown-unknown-eabi-cortex-m0plus
-	@mkdir -p build/release/tinygo/pkg/thumbv7em-unknown-unknown-eabi-cortex-m4
+	@mkdir -p build/release/tinygo/lib/wasi-libc/libc-bottom-half/headers
+	@mkdir -p build/release/tinygo/lib/wasi-libc/libc-top-half/musl/arch
+	@mkdir -p build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@mkdir -p build/release/tinygo/lib/wasi-cli/
 	@echo copying source files
 	@cp -p  build/tinygo$(EXE)           build/release/tinygo/bin
 ifneq ($(USE_SYSTEM_BINARYEN),1)
@@ -798,6 +936,7 @@ endif
 	@cp -rp lib/musl/arch/arm            build/release/tinygo/lib/musl/arch
 	@cp -rp lib/musl/arch/generic        build/release/tinygo/lib/musl/arch
 	@cp -rp lib/musl/arch/i386           build/release/tinygo/lib/musl/arch
+	@cp -rp lib/musl/arch/mips           build/release/tinygo/lib/musl/arch
 	@cp -rp lib/musl/arch/x86_64         build/release/tinygo/lib/musl/arch
 	@cp -rp lib/musl/crt/crt1.c          build/release/tinygo/lib/musl/crt
 	@cp -rp lib/musl/COPYRIGHT           build/release/tinygo/lib/musl
@@ -805,21 +944,28 @@ endif
 	@cp -rp lib/musl/src/env             build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/errno           build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/exit            build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/fcntl           build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/include         build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/internal        build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/legacy          build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/locale          build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/linux           build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/malloc          build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/mman            build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/math            build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/misc            build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/multibyte       build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/signal          build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/stdio           build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/string          build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/thread          build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/time            build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/unistd          build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/process         build/release/tinygo/lib/musl/src
 	@cp -rp lib/mingw-w64/mingw-w64-crt/def-include                 build/release/tinygo/lib/mingw-w64/mingw-w64-crt
 	@cp -rp lib/mingw-w64/mingw-w64-crt/lib-common/api-ms-win-crt-* build/release/tinygo/lib/mingw-w64/mingw-w64-crt/lib-common
 	@cp -rp lib/mingw-w64/mingw-w64-crt/lib-common/kernel32.def.in  build/release/tinygo/lib/mingw-w64/mingw-w64-crt/lib-common
+	@cp -rp lib/mingw-w64/mingw-w64-crt/stdio/ucrt_*                build/release/tinygo/lib/mingw-w64/mingw-w64-crt/stdio
 	@cp -rp lib/mingw-w64/mingw-w64-headers/crt/                    build/release/tinygo/lib/mingw-w64/mingw-w64-headers
 	@cp -rp lib/mingw-w64/mingw-w64-headers/defaults/include        build/release/tinygo/lib/mingw-w64/mingw-w64-headers/defaults
 	@cp -rp lib/nrfx/*                   build/release/tinygo/lib/nrfx
@@ -831,17 +977,20 @@ endif
 	@cp -rp lib/picolibc/newlib/libm/common      build/release/tinygo/lib/picolibc/newlib/libm
 	@cp -rp lib/picolibc/newlib/libm/math        build/release/tinygo/lib/picolibc/newlib/libm
 	@cp -rp lib/picolibc-stdio.c         build/release/tinygo/lib
-	@cp -rp lib/wasi-libc/sysroot        build/release/tinygo/lib/wasi-libc/sysroot
+	@cp -rp lib/wasi-libc/libc-bottom-half/headers/public           build/release/tinygo/lib/wasi-libc/libc-bottom-half/headers
+	@cp -rp lib/wasi-libc/libc-top-half/musl/arch/generic           build/release/tinygo/lib/wasi-libc/libc-top-half/musl/arch
+	@cp -rp lib/wasi-libc/libc-top-half/musl/arch/wasm32            build/release/tinygo/lib/wasi-libc/libc-top-half/musl/arch
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/include            build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/internal           build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/math               build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/string             build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/include                build/release/tinygo/lib/wasi-libc/libc-top-half/musl
+	@cp -rp lib/wasi-libc/sysroot                                   build/release/tinygo/lib/wasi-libc/sysroot
+	@cp -rp lib/wasi-cli/wit                                        build/release/tinygo/lib/wasi-cli/wit
 	@cp -rp llvm-project/compiler-rt/lib/builtins build/release/tinygo/lib/compiler-rt-builtins
 	@cp -rp llvm-project/compiler-rt/LICENSE.TXT  build/release/tinygo/lib/compiler-rt-builtins
 	@cp -rp src                          build/release/tinygo/src
 	@cp -rp targets                      build/release/tinygo/targets
-	./build/release/tinygo/bin/tinygo build-library -target=cortex-m0     -o build/release/tinygo/pkg/thumbv6m-unknown-unknown-eabi-cortex-m0/compiler-rt     compiler-rt
-	./build/release/tinygo/bin/tinygo build-library -target=cortex-m0plus -o build/release/tinygo/pkg/thumbv6m-unknown-unknown-eabi-cortex-m0plus/compiler-rt compiler-rt
-	./build/release/tinygo/bin/tinygo build-library -target=cortex-m4     -o build/release/tinygo/pkg/thumbv7em-unknown-unknown-eabi-cortex-m4/compiler-rt    compiler-rt
-	./build/release/tinygo/bin/tinygo build-library -target=cortex-m0     -o build/release/tinygo/pkg/thumbv6m-unknown-unknown-eabi-cortex-m0/picolibc     picolibc
-	./build/release/tinygo/bin/tinygo build-library -target=cortex-m0plus -o build/release/tinygo/pkg/thumbv6m-unknown-unknown-eabi-cortex-m0plus/picolibc picolibc
-	./build/release/tinygo/bin/tinygo build-library -target=cortex-m4     -o build/release/tinygo/pkg/thumbv7em-unknown-unknown-eabi-cortex-m4/picolibc    picolibc
 
 release:
 	tar -czf build/release.tar.gz -C build/release tinygo
@@ -852,9 +1001,40 @@ deb:
 	@mkdir -p build/release-deb/usr/local/lib
 	cp -ar build/release/tinygo build/release-deb/usr/local/lib/tinygo
 	ln -sf ../lib/tinygo/bin/tinygo build/release-deb/usr/local/bin/tinygo
-	fpm -f -s dir -t deb -n tinygo -a $(DEB_ARCH) -v $(shell grep "const Version = " goenv/version.go | awk '{print $$NF}') -m '@tinygo-org' --description='TinyGo is a Go compiler for small places.' --license='BSD 3-Clause' --url=https://tinygo.org/ --deb-changelog CHANGELOG.md -p build/release.deb -C ./build/release-deb
+	fpm -f -s dir -t deb -n tinygo -a $(DEB_ARCH) -v $(shell grep "const version = " goenv/version.go | awk '{print $$NF}') -m '@tinygo-org' --description='TinyGo is a Go compiler for small places.' --license='BSD 3-Clause' --url=https://tinygo.org/ --deb-changelog CHANGELOG.md -p build/release.deb -C ./build/release-deb
 
 ifneq ($(RELEASEONLY), 1)
 release: build/release
 deb: build/release
 endif
+
+.PHONY: tools
+tools:
+	cd internal/tools && go generate -tags tools ./
+
+.PHONY: lint
+lint: tools ## Lint source tree
+	revive -version
+	# TODO: lint more directories!
+	# revive.toml isn't flexible enough to filter out just one kind of error from a checker, so do it with grep here.
+	# Can't use grep with friendly formatter.  Plain output isn't too bad, though.
+	# Use 'grep .' to get rid of stray blank line
+	revive -config revive.toml compiler/... src/{os,reflect}/*.go | grep -v "should have comment or be unexported" | grep '.' | awk '{print}; END {exit NR>0}'
+
+SPELLDIRSCMD=find . -depth 1 -type d  | egrep -wv '.git|lib|llvm|src'; find src -depth 1 | egrep -wv 'device|internal|net|vendor'; find src/internal -depth 1 -type d | egrep -wv src/internal/wasi
+.PHONY: spell
+spell: tools ## Spellcheck source tree
+	misspell -error --dict misspell.csv -i 'ackward,devided,extint,rela' $$( $(SPELLDIRSCMD) ) *.go *.md
+
+.PHONY: spellfix
+spellfix: tools ## Same as spell, but fixes what it finds
+	misspell -w --dict misspell.csv -i 'ackward,devided,extint,rela' $$( $(SPELLDIRSCMD) ) *.go *.md
+
+# https://www.client9.com/self-documenting-makefiles/
+.PHONY: help
+help:
+	@awk -F ':|##' '/^[^\t].+?:.*?##/ {\
+	gsub(/\$$\(LLVM_BUILDDIR\)/, "$(LLVM_BUILDDIR)"); \
+        printf "\033[36m%-30s\033[0m %s\n", $$1, $$NF \
+        }' $(MAKEFILE_LIST)
+#.DEFAULT_GOAL=help

GOVERNANCE.md

@@ -0,0 +1,32 @@
+TinyGo Team Members
+===================
+
+The team of humans who maintain TinyGo.
+
+*   **Purpose**: To maintain the community, code, documentation, and tools for the TinyGo compiler.
+*   **Board**: The group of people who share responsibility for key decisions for the TinyGo organization.  
+    *   **Majority Voting**: The board makes decisions by majority vote.
+    *   **Membership**: The board elects its own members.
+*   **Do-ocracy**: Those who step forward to do a given task propose how it should be done. Then other interested people can make comments.
+*   **Proof of Work**: Power in decision-making is slightly weighted based on a participant's labor for the community.
+*   **Initiation**: We need to establish a procedure for how people join the team of maintainers.
+*   **Transparency**: Important information should be made publicly available, ideally in a way that allows for public comment.
+*   **Code of Conduct**: Participants agree to abide by the current project Code of Conduct.
+
+## Members
+
+* Ayke van Laethem (@aykevl)
+* Daniel Esteban (@conejoninja)
+* Ron Evans (@deadprogram)
+* Damian Gryski (@dgryski)
+* Masaaki Takasago (@sago35)
+* Patricio Whittingslow (@soypat)
+* Yurii Soldak (@ysoldak)
+
+## Experimental
+
+*   **Monthly Meeting**: A monthly meeting for the team and any other interested participants.  
+    Duration: 1 hour  
+    Facilitation: @deadprogram
+    Schedule: See https://github.com/tinygo-org/tinygo/wiki/Meetings for more information
+

LICENSE

@@ -1,7 +1,7 @@
-Copyright (c) 2018-2023 The TinyGo Authors. All rights reserved.
+Copyright (c) 2018-2025 The TinyGo Authors. All rights reserved.
 
 TinyGo includes portions of the Go standard library.
-Copyright (c) 2009-2023 The Go Authors. All rights reserved.
+Copyright (c) 2009-2024 The Go Authors. All rights reserved.
 
 TinyGo includes portions of LLVM, which is under the Apache License v2.0 with
 LLVM Exceptions. See https://llvm.org/LICENSE.txt for license information.

README.md

@@ -1,6 +1,6 @@
 # TinyGo - Go compiler for small places
 
-[![Linux](https://github.com/tinygo-org/tinygo/actions/workflows/linux.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/linux.yml) [![macOS](https://github.com/tinygo-org/tinygo/actions/workflows/build-macos.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/build-macos.yml) [![Windows](https://github.com/tinygo-org/tinygo/actions/workflows/windows.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/windows.yml) [![Docker](https://github.com/tinygo-org/tinygo/actions/workflows/docker.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/docker.yml) [![CircleCI](https://circleci.com/gh/tinygo-org/tinygo/tree/dev.svg?style=svg)](https://circleci.com/gh/tinygo-org/tinygo/tree/dev)
+[![Linux](https://github.com/tinygo-org/tinygo/actions/workflows/linux.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/linux.yml) [![macOS](https://github.com/tinygo-org/tinygo/actions/workflows/build-macos.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/build-macos.yml) [![Windows](https://github.com/tinygo-org/tinygo/actions/workflows/windows.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/windows.yml) [![Docker](https://github.com/tinygo-org/tinygo/actions/workflows/docker.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/docker.yml) [![Nix](https://github.com/tinygo-org/tinygo/actions/workflows/nix.yml/badge.svg?branch=dev)](https://github.com/tinygo-org/tinygo/actions/workflows/nix.yml) [![CircleCI](https://circleci.com/gh/tinygo-org/tinygo/tree/dev.svg?style=svg)](https://circleci.com/gh/tinygo-org/tinygo/tree/dev)
 
 TinyGo is a Go compiler intended for use in small places such as microcontrollers, WebAssembly (wasm/wasi), and command-line tools.
 
@@ -41,27 +41,29 @@ tinygo flash -target arduino examples/blinky1
 
 TinyGo is very useful for compiling programs both for use in browsers (WASM) as well as for use on servers and other edge devices (WASI).
 
-TinyGo programs can run in Fastly Compute@Edge (https://developer.fastly.com/learning/compute/go/), Fermyon Spin (https://developer.fermyon.com/spin/go-components), wazero (https://wazero.io/languages/tinygo/) and many other WebAssembly runtimes.
+TinyGo programs can run in [Fastly Compute](https://www.fastly.com/documentation/guides/compute/go/), [Fermyon Spin](https://developer.fermyon.com/spin/go-components), [wazero](https://wazero.io/languages/tinygo/) and many other WebAssembly runtimes.
 
 Here is a small TinyGo program for use by a WASI host application:
 
 ```go
 package main
 
-//go:wasm-module yourmodulename
-//export add
+//go:wasmexport add
 func add(x, y uint32) uint32 {
 	return x + y
 }
-
-// main is required for the `wasi` target, even if it isn't used.
-func main() {}
 ```
 
-This compiles the above TinyGo program for use on any WASI runtime:
+This compiles the above TinyGo program for use on any WASI Preview 1 runtime:
 
 ```shell
-tinygo build -o main.wasm -target=wasi main.go
+tinygo build -buildmode=c-shared -o add.wasm -target=wasip1 add.go
+```
+
+You can also use the same syntax as Go 1.24+:
+
+```shell
+GOARCH=wasip1 GOOS=wasm tinygo build -buildmode=c-shared -o add.wasm add.go
 ```
 
 ## Installation

builder/ar.go

@@ -12,10 +12,11 @@ import (
 	"path/filepath"
 	"time"
 
+	wasm "github.com/aykevl/go-wasm"
 	"github.com/blakesmith/ar"
 )
 
-// makeArchive creates an arcive for static linking from a list of object files
+// makeArchive creates an archive for static linking from a list of object files
 // given as a parameter. It is equivalent to the following command:
 //
 //	ar -rcs <archivePath> <objs...>
@@ -74,8 +75,35 @@ func makeArchive(arfile *os.File, objs []string) error {
 					fileIndex int
 				}{symbol.Name, i})
 			}
+		} else if dbg, err := wasm.Parse(objfile); err == nil {
+			for _, s := range dbg.Sections {
+				switch section := s.(type) {
+				case *wasm.SectionLinking:
+					for _, symbol := range section.Symbols {
+						if symbol.Flags&wasm.LinkingSymbolFlagUndefined != 0 {
+							// Don't list undefined functions.
+							continue
+						}
+						if symbol.Flags&wasm.LinkingSymbolFlagBindingLocal != 0 {
+							// Don't include local symbols.
+							continue
+						}
+						if symbol.Kind != wasm.LinkingSymbolKindFunction && symbol.Kind != wasm.LinkingSymbolKindData {
+							// Link functions and data symbols.
+							// Some data symbols need to be included, such as
+							// __log_data.
+							continue
+						}
+						// Include in the archive.
+						symbolTable = append(symbolTable, struct {
+							name      string
+							fileIndex int
+						}{symbol.Name, i})
+					}
+				}
+			}
 		} else {
-			return fmt.Errorf("failed to open file %s as ELF or PE/COFF: %w", objpath, err)
+			return fmt.Errorf("failed to open file %s as WASM, ELF or PE/COFF: %w", objpath, err)
 		}
 
 		// Close file, to avoid issues with too many open files (especially on
@@ -122,7 +150,7 @@ func makeArchive(arfile *os.File, objs []string) error {
 	}
 
 	// Keep track of the start of the symbol table.
-	symbolTableStart, err := arfile.Seek(0, os.SEEK_CUR)
+	symbolTableStart, err := arfile.Seek(0, io.SeekCurrent)
 	if err != nil {
 		return err
 	}
@@ -144,7 +172,7 @@ func makeArchive(arfile *os.File, objs []string) error {
 
 		// Store the start index, for when we'll update the symbol table with
 		// the correct file start indices.
-		offset, err := arfile.Seek(0, os.SEEK_CUR)
+		offset, err := arfile.Seek(0, io.SeekCurrent)
 		if err != nil {
 			return err
 		}

builder/build.go

@@ -61,6 +61,10 @@ type BuildResult struct {
 	// correctly printing test results: the import path isn't always the same as
 	// the path listed on the command line.
 	ImportPath string
+
+	// Map from path to package name. It is needed to attribute binary size to
+	// the right Go package.
+	PackagePathMap map[string]string
 }
 
 // packageAction is the struct that is serialized to JSON and hashed, to work as
@@ -83,8 +87,7 @@ type packageAction struct {
 	FileHashes       map[string]string // hash of every file that's part of the package
 	EmbeddedFiles    map[string]string // hash of all the //go:embed files in the package
 	Imports          map[string]string // map from imported package to action ID hash
-	OptLevel         int               // LLVM optimization level (0-3)
-	SizeLevel        int               // LLVM optimization for size level (0-2)
+	OptLevel         string            // LLVM optimization level (O0, O1, O2, Os, Oz)
 	UndefinedGlobals []string          // globals that are left as external globals (no initializer)
 }
 
@@ -115,6 +118,30 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		cacheDir = tmpdir
 	}
 
+	// Create default global values.
+	globalValues := map[string]map[string]string{
+		"runtime": {
+			"buildVersion": goenv.Version(),
+		},
+		"testing": {},
+	}
+	if config.TestConfig.CompileTestBinary {
+		// The testing.testBinary is set to "1" when in a test.
+		// This is needed for testing.Testing() to work correctly.
+		globalValues["testing"]["testBinary"] = "1"
+	}
+
+	// Copy over explicitly set global values, like
+	// -ldflags="-X main.Version="1.0"
+	for pkgPath, vals := range config.Options.GlobalValues {
+		if _, ok := globalValues[pkgPath]; !ok {
+			globalValues[pkgPath] = map[string]string{}
+		}
+		for k, v := range vals {
+			globalValues[pkgPath][k] = v
+		}
+	}
+
 	// Check for a libc dependency.
 	// As a side effect, this also creates the headers for the given libc, if
 	// the libc needs them.
@@ -125,7 +152,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		job := makeDarwinLibSystemJob(config, tmpdir)
 		libcDependencies = append(libcDependencies, job)
 	case "musl":
-		job, unlock, err := Musl.load(config, tmpdir)
+		job, unlock, err := libMusl.load(config, tmpdir)
 		if err != nil {
 			return BuildResult{}, err
 		}
@@ -133,7 +160,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		libcDependencies = append(libcDependencies, dummyCompileJob(filepath.Join(filepath.Dir(job.result), "crt1.o")))
 		libcDependencies = append(libcDependencies, job)
 	case "picolibc":
-		libcJob, unlock, err := Picolibc.load(config, tmpdir)
+		libcJob, unlock, err := libPicolibc.load(config, tmpdir)
 		if err != nil {
 			return BuildResult{}, err
 		}
@@ -145,12 +172,20 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 			return BuildResult{}, errors.New("could not find wasi-libc, perhaps you need to run `make wasi-libc`?")
 		}
 		libcDependencies = append(libcDependencies, dummyCompileJob(path))
-	case "mingw-w64":
-		_, unlock, err := MinGW.load(config, tmpdir)
+	case "wasmbuiltins":
+		libcJob, unlock, err := libWasmBuiltins.load(config, tmpdir)
 		if err != nil {
 			return BuildResult{}, err
 		}
-		unlock()
+		defer unlock()
+		libcDependencies = append(libcDependencies, libcJob)
+	case "mingw-w64":
+		job, unlock, err := libMinGW.load(config, tmpdir)
+		if err != nil {
+			return BuildResult{}, err
+		}
+		defer unlock()
+		libcDependencies = append(libcDependencies, job)
 		libcDependencies = append(libcDependencies, makeMinGWExtraLibs(tmpdir, config.GOARCH())...)
 	case "":
 		// no library specified, so nothing to do
@@ -158,7 +193,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		return BuildResult{}, fmt.Errorf("unknown libc: %s", config.Target.Libc)
 	}
 
-	optLevel, sizeLevel, _ := config.OptLevels()
+	optLevel, speedLevel, sizeLevel := config.OptLevel()
 	compilerConfig := &compiler.Config{
 		Triple:          config.Triple(),
 		CPU:             config.CPU(),
@@ -166,16 +201,19 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		ABI:             config.ABI(),
 		GOOS:            config.GOOS(),
 		GOARCH:          config.GOARCH(),
+		BuildMode:       config.BuildMode(),
 		CodeModel:       config.CodeModel(),
 		RelocationModel: config.RelocationModel(),
 		SizeLevel:       sizeLevel,
-		TinyGoVersion:   goenv.Version,
+		TinyGoVersion:   goenv.Version(),
 
 		Scheduler:          config.Scheduler(),
 		AutomaticStackSize: config.AutomaticStackSize(),
 		DefaultStackSize:   config.StackSize(),
+		MaxStackAlloc:      config.MaxStackAlloc(),
 		NeedsStackObjects:  config.NeedsStackObjects(),
 		Debug:              !config.Options.SkipDWARF, // emit DWARF except when -internal-nodwarf is passed
+		PanicStrategy:      config.PanicStrategy(),
 	}
 
 	// Load the target machine, which is the LLVM object that contains all
@@ -188,7 +226,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 	defer machine.Dispose()
 
 	// Load entire program AST into memory.
-	lprogram, err := loader.Load(config, pkgName, config.ClangHeaders, types.Config{
+	lprogram, err := loader.Load(config, pkgName, types.Config{
 		Sizes: compiler.Sizes(machine),
 	})
 	if err != nil {
@@ -208,6 +246,12 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		return result, err
 	}
 
+	// Store which filesystem paths map to which package name.
+	result.PackagePathMap = make(map[string]string, len(lprogram.Packages))
+	for _, pkg := range lprogram.Sorted() {
+		result.PackagePathMap[pkg.OriginalDir()] = pkg.Pkg.Path()
+	}
+
 	// Create the *ssa.Program. This does not yet build the entire SSA of the
 	// program so it's pretty fast and doesn't need to be parallelized.
 	program := lprogram.LoadSSA()
@@ -217,26 +261,12 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 	var packageJobs []*compileJob
 	packageActionIDJobs := make(map[string]*compileJob)
 
-	if config.Options.GlobalValues["runtime"]["buildVersion"] == "" {
-		version := goenv.Version
-		if strings.HasSuffix(goenv.Version, "-dev") && goenv.GitSha1 != "" {
-			version += "-" + goenv.GitSha1
-		}
-		if config.Options.GlobalValues == nil {
-			config.Options.GlobalValues = make(map[string]map[string]string)
-		}
-		if config.Options.GlobalValues["runtime"] == nil {
-			config.Options.GlobalValues["runtime"] = make(map[string]string)
-		}
-		config.Options.GlobalValues["runtime"]["buildVersion"] = version
-	}
-
 	var embedFileObjects []*compileJob
 	for _, pkg := range lprogram.Sorted() {
 		pkg := pkg // necessary to avoid a race condition
 
 		var undefinedGlobals []string
-		for name := range config.Options.GlobalValues[pkg.Pkg.Path()] {
+		for name := range globalValues[pkg.Pkg.Path()] {
 			undefinedGlobals = append(undefinedGlobals, name)
 		}
 		sort.Strings(undefinedGlobals)
@@ -313,7 +343,6 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 					EmbeddedFiles:    make(map[string]string, len(allFiles)),
 					Imports:          make(map[string]string, len(pkg.Pkg.Imports())),
 					OptLevel:         optLevel,
-					SizeLevel:        sizeLevel,
 					UndefinedGlobals: undefinedGlobals,
 				}
 				for filePath, hash := range pkg.FileHashes {
@@ -358,7 +387,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 				defer mod.Context().Dispose()
 				defer mod.Dispose()
 				if errs != nil {
-					return newMultiError(errs)
+					return newMultiError(errs, pkg.ImportPath)
 				}
 				if err := llvm.VerifyModule(mod, llvm.PrintMessageAction); err != nil {
 					return errors.New("verification error after compiling package " + pkg.ImportPath)
@@ -520,13 +549,13 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 			irbuilder := mod.Context().NewBuilder()
 			defer irbuilder.Dispose()
 			irbuilder.SetInsertPointAtEnd(block)
-			i8ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
+			ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
 			for _, pkg := range lprogram.Sorted() {
 				pkgInit := mod.NamedFunction(pkg.Pkg.Path() + ".init")
 				if pkgInit.IsNil() {
 					panic("init not found for " + pkg.Pkg.Path())
 				}
-				irbuilder.CreateCall(pkgInit.GlobalValueType(), pkgInit, []llvm.Value{llvm.Undef(i8ptrType)}, "")
+				irbuilder.CreateCall(pkgInit.GlobalValueType(), pkgInit, []llvm.Value{llvm.Undef(ptrType)}, "")
 			}
 			irbuilder.CreateRetVoid()
 
@@ -561,7 +590,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 
 			// Run all optimization passes, which are much more effective now
 			// that the optimizer can see the whole program at once.
-			err := optimizeProgram(mod, config)
+			err := optimizeProgram(mod, config, globalValues)
 			if err != nil {
 				return err
 			}
@@ -575,6 +604,11 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		},
 	}
 
+	// Create the output directory, if needed
+	if err := os.MkdirAll(filepath.Dir(outpath), 0777); err != nil {
+		return result, err
+	}
+
 	// Check whether we only need to create an object file.
 	// If so, we don't need to link anything and will be finished quickly.
 	outext := filepath.Ext(outpath)
@@ -631,10 +665,27 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 	result.Binary = result.Executable // final file
 	ldflags := append(config.LDFlags(), "-o", result.Executable)
 
+	if config.Options.BuildMode == "c-shared" {
+		if !strings.HasPrefix(config.Triple(), "wasm32-") {
+			return result, fmt.Errorf("buildmode c-shared is only supported on wasm at the moment")
+		}
+		ldflags = append(ldflags, "--no-entry")
+	}
+
+	if config.Options.BuildMode == "wasi-legacy" {
+		if !strings.HasPrefix(config.Triple(), "wasm32-") {
+			return result, fmt.Errorf("buildmode wasi-legacy is only supported on wasm")
+		}
+
+		if config.Options.Scheduler != "none" {
+			return result, fmt.Errorf("buildmode wasi-legacy only supports scheduler=none")
+		}
+	}
+
 	// Add compiler-rt dependency if needed. Usually this is a simple load from
 	// a cache.
 	if config.Target.RTLib == "compiler-rt" {
-		job, unlock, err := CompilerRT.load(config, tmpdir)
+		job, unlock, err := libCompilerRT.load(config, tmpdir)
 		if err != nil {
 			return result, err
 		}
@@ -650,7 +701,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		job := &compileJob{
 			description: "compile extra file " + path,
 			run: func(job *compileJob) error {
-				result, err := compileAndCacheCFile(abspath, tmpdir, config.CFlags(), config.Options.PrintCommands)
+				result, err := compileAndCacheCFile(abspath, tmpdir, config.CFlags(false), config.Options.PrintCommands)
 				job.result = result
 				return err
 			},
@@ -664,7 +715,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 	for _, pkg := range lprogram.Sorted() {
 		pkg := pkg
 		for _, filename := range pkg.CFiles {
-			abspath := filepath.Join(pkg.Dir, filename)
+			abspath := filepath.Join(pkg.OriginalDir(), filename)
 			job := &compileJob{
 				description: "compile CGo file " + abspath,
 				run: func(job *compileJob) error {
@@ -728,20 +779,21 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 				ldflags = append(ldflags, dependency.result)
 			}
 			ldflags = append(ldflags, "-mllvm", "-mcpu="+config.CPU())
+			ldflags = append(ldflags, "-mllvm", "-mattr="+config.Features()) // needed for MIPS softfloat
 			if config.GOOS() == "windows" {
 				// Options for the MinGW wrapper for the lld COFF linker.
 				ldflags = append(ldflags,
-					"-Xlink=/opt:lldlto="+strconv.Itoa(optLevel),
+					"-Xlink=/opt:lldlto="+strconv.Itoa(speedLevel),
 					"--thinlto-cache-dir="+filepath.Join(cacheDir, "thinlto"))
 			} else if config.GOOS() == "darwin" {
 				// Options for the ld64-compatible lld linker.
 				ldflags = append(ldflags,
-					"--lto-O"+strconv.Itoa(optLevel),
+					"--lto-O"+strconv.Itoa(speedLevel),
 					"-cache_path_lto", filepath.Join(cacheDir, "thinlto"))
 			} else {
 				// Options for the ELF linker.
 				ldflags = append(ldflags,
-					"--lto-O"+strconv.Itoa(optLevel),
+					"--lto-O"+strconv.Itoa(speedLevel),
 					"--thinlto-cache-dir="+filepath.Join(cacheDir, "thinlto"),
 				)
 			}
@@ -752,7 +804,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 			if sizeLevel >= 2 {
 				// Workaround with roughly the same effect as
 				// https://reviews.llvm.org/D119342.
-				// Can hopefully be removed in LLVM 15.
+				// Can hopefully be removed in LLVM 19.
 				ldflags = append(ldflags,
 					"-mllvm", "--rotation-max-header-size=0")
 			}
@@ -761,7 +813,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 			}
 			err = link(config.Target.Linker, ldflags...)
 			if err != nil {
-				return &commandError{"failed to link", result.Executable, err}
+				return err
 			}
 
 			var calculatedStacks []string
@@ -785,6 +837,12 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 					return fmt.Errorf("could not modify stack sizes: %w", err)
 				}
 			}
+
+			// Apply patches of bootloader in the order they appear.
+			if len(config.Target.BootPatches) > 0 {
+				err = applyPatches(result.Executable, config.Target.BootPatches)
+			}
+
 			if config.RP2040BootPatch() {
 				// Patch the second stage bootloader CRC into the .boot2 section
 				err = patchRP2040BootCRC(result.Executable)
@@ -795,21 +853,8 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 
 			// Run wasm-opt for wasm binaries
 			if arch := strings.Split(config.Triple(), "-")[0]; arch == "wasm32" {
-				var opt string
-				switch config.Options.Opt {
-				case "none", "0":
-					opt = "-O0"
-				case "1":
-					opt = "-O1"
-				case "2":
-					opt = "-O2"
-				case "s":
-					opt = "-Os"
-				case "z":
-					opt = "-Oz"
-				default:
-					return fmt.Errorf("unknown opt level: %q", config.Options.Opt)
-				}
+				optLevel, _, _ := config.OptLevel()
+				opt := "-" + optLevel
 
 				var args []string
 
@@ -817,14 +862,20 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 					args = append(args, "--asyncify")
 				}
 
+				inputFile := result.Binary
+				result.Binary = result.Executable + ".wasmopt"
 				args = append(args,
 					opt,
 					"-g",
-					result.Executable,
-					"--output", result.Executable,
+					inputFile,
+					"--output", result.Binary,
 				)
 
-				cmd := exec.Command(goenv.Get("WASMOPT"), args...)
+				wasmopt := goenv.Get("WASMOPT")
+				if config.Options.PrintCommands != nil {
+					config.Options.PrintCommands(wasmopt, args...)
+				}
+				cmd := exec.Command(wasmopt, args...)
 				cmd.Stdout = os.Stdout
 				cmd.Stderr = os.Stderr
 
@@ -834,20 +885,77 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 				}
 			}
 
-			// Print code size if requested.
-			if config.Options.PrintSizes == "short" || config.Options.PrintSizes == "full" {
-				packagePathMap := make(map[string]string, len(lprogram.Packages))
-				for _, pkg := range lprogram.Sorted() {
-					packagePathMap[pkg.OriginalDir()] = pkg.Pkg.Path()
+			// Run wasm-tools for component-model binaries
+			witPackage := strings.ReplaceAll(config.Target.WITPackage, "{root}", goenv.Get("TINYGOROOT"))
+			if config.Options.WITPackage != "" {
+				witPackage = config.Options.WITPackage
+			}
+			witWorld := config.Target.WITWorld
+			if config.Options.WITWorld != "" {
+				witWorld = config.Options.WITWorld
+			}
+			if witPackage != "" && witWorld != "" {
+
+				// wasm-tools component embed -w wasi:cli/command
+				// 		$$(tinygo env TINYGOROOT)/lib/wasi-cli/wit/ main.wasm -o embedded.wasm
+				componentEmbedInputFile := result.Binary
+				result.Binary = result.Executable + ".wasm-component-embed"
+				args := []string{
+					"component",
+					"embed",
+					"-w", witWorld,
+					witPackage,
+					componentEmbedInputFile,
+					"-o", result.Binary,
 				}
-				sizes, err := loadProgramSize(result.Executable, packagePathMap)
+
+				wasmtools := goenv.Get("WASMTOOLS")
+				if config.Options.PrintCommands != nil {
+					config.Options.PrintCommands(wasmtools, args...)
+				}
+				cmd := exec.Command(wasmtools, args...)
+				cmd.Stdout = os.Stdout
+				cmd.Stderr = os.Stderr
+
+				err := cmd.Run()
+				if err != nil {
+					return fmt.Errorf("`wasm-tools component embed` failed: %w", err)
+				}
+
+				// wasm-tools component new embedded.wasm -o component.wasm
+				componentNewInputFile := result.Binary
+				result.Binary = result.Executable + ".wasm-component-new"
+				args = []string{
+					"component",
+					"new",
+					componentNewInputFile,
+					"-o", result.Binary,
+				}
+
+				if config.Options.PrintCommands != nil {
+					config.Options.PrintCommands(wasmtools, args...)
+				}
+				cmd = exec.Command(wasmtools, args...)
+				cmd.Stdout = os.Stdout
+				cmd.Stderr = os.Stderr
+
+				err = cmd.Run()
+				if err != nil {
+					return fmt.Errorf("`wasm-tools component new` failed: %w", err)
+				}
+			}
+
+			// Print code size if requested.
+			if config.Options.PrintSizes != "" {
+				sizes, err := loadProgramSize(result.Executable, result.PackagePathMap)
 				if err != nil {
 					return err
 				}
-				if config.Options.PrintSizes == "short" {
+				switch config.Options.PrintSizes {
+				case "short":
 					fmt.Printf("   code    data     bss |   flash     ram\n")
 					fmt.Printf("%7d %7d %7d | %7d %7d\n", sizes.Code+sizes.ROData, sizes.Data, sizes.BSS, sizes.Flash(), sizes.RAM())
-				} else {
+				case "full":
 					if !config.Debug() {
 						fmt.Println("warning: data incomplete, remove the -no-debug flag for more detail")
 					}
@@ -859,6 +967,13 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 					}
 					fmt.Printf("------------------------------- | --------------- | -------\n")
 					fmt.Printf("%7d %7d %7d %7d | %7d %7d | total\n", sizes.Code, sizes.ROData, sizes.Data, sizes.BSS, sizes.Code+sizes.ROData+sizes.Data, sizes.Data+sizes.BSS)
+				case "html":
+					const filename = "size-report.html"
+					err := writeSizeReport(sizes, filename, pkgName)
+					if err != nil {
+						return err
+					}
+					fmt.Println("Wrote size report to", filename)
 				}
 			}
 
@@ -1030,7 +1145,7 @@ func createEmbedObjectFile(data, hexSum, sourceFile, sourceDir, tmpdir string, c
 // optimizeProgram runs a series of optimizations and transformations that are
 // needed to convert a program to its final form. Some transformations are not
 // optional and must be run as the compiler expects them to run.
-func optimizeProgram(mod llvm.Module, config *compileopts.Config) error {
+func optimizeProgram(mod llvm.Module, config *compileopts.Config, globalValues map[string]map[string]string) error {
 	err := interp.Run(mod, config.Options.InterpTimeout, config.DumpSSA())
 	if err != nil {
 		return err
@@ -1049,28 +1164,16 @@ func optimizeProgram(mod llvm.Module, config *compileopts.Config) error {
 	}
 
 	// Insert values from -ldflags="-X ..." into the IR.
-	err = setGlobalValues(mod, config.Options.GlobalValues)
+	err = setGlobalValues(mod, globalValues)
 	if err != nil {
 		return err
 	}
 
-	// Browsers cannot handle external functions that have type i64 because it
-	// cannot be represented exactly in JavaScript (JS only has doubles). To
-	// keep functions interoperable, pass int64 types as pointers to
-	// stack-allocated values.
-	if config.WasmAbi() == "js" {
-		err := transform.ExternalInt64AsPtr(mod, config)
-		if err != nil {
-			return err
-		}
-	}
-
-	// Optimization levels here are roughly the same as Clang, but probably not
-	// exactly.
-	optLevel, sizeLevel, inlinerThreshold := config.OptLevels()
-	errs := transform.Optimize(mod, config, optLevel, sizeLevel, inlinerThreshold)
+	// Run most of the whole-program optimizations (including the whole
+	// O0/O1/O2/Os/Oz optimization pipeline).
+	errs := transform.Optimize(mod, config)
 	if len(errs) > 0 {
-		return newMultiError(errs)
+		return newMultiError(errs, "")
 	}
 	if err := llvm.VerifyModule(mod, llvm.PrintMessageAction); err != nil {
 		return errors.New("verification failure after LLVM optimization passes")
@@ -1197,7 +1300,7 @@ func determineStackSizes(mod llvm.Module, executable string) ([]string, map[stri
 	}
 
 	// Goroutines need to be started and finished and take up some stack space
-	// that way. This can be measured by measuing the stack size of
+	// that way. This can be measured by measuring the stack size of
 	// tinygo_startTask.
 	if numFuncs := len(functions["tinygo_startTask"]); numFuncs != 1 {
 		return nil, nil, fmt.Errorf("expected exactly one definition of tinygo_startTask, got %d", numFuncs)
@@ -1360,6 +1463,23 @@ func printStacks(calculatedStacks []string, stackSizes map[string]functionStackS
 	}
 }
 
+func applyPatches(executable string, bootPatches []string) (err error) {
+	for _, patch := range bootPatches {
+		switch patch {
+		case "rp2040":
+			err = patchRP2040BootCRC(executable)
+		// case "rp2350":
+		// 	err = patchRP2350BootIMAGE_DEF(executable)
+		default:
+			err = errors.New("undefined boot patch name")
+		}
+		if err != nil {
+			return fmt.Errorf("apply boot patch %q: %w", patch, err)
+		}
+	}
+	return nil
+}
+
 // RP2040 second stage bootloader CRC32 calculation
 //
 // Spec: https://datasheets.raspberrypi.org/rp2040/rp2040-datasheet.pdf
@@ -1371,7 +1491,7 @@ func patchRP2040BootCRC(executable string) error {
 	}
 
 	if len(bytes) != 256 {
-		return fmt.Errorf("rp2040 .boot2 section must be exactly 256 bytes")
+		return fmt.Errorf("rp2040 .boot2 section must be exactly 256 bytes, got %d", len(bytes))
 	}
 
 	// From the 'official' RP2040 checksum script:
@@ -1410,3 +1530,10 @@ func lock(path string) func() {
 
 	return func() { flock.Close() }
 }
+
+func b2u8(b bool) uint8 {
+	if b {
+		return 1
+	}
+	return 0
+}

builder/builder_test.go

@@ -8,7 +8,6 @@ import (
 	"testing"
 
 	"github.com/tinygo-org/tinygo/compileopts"
-	"github.com/tinygo-org/tinygo/goenv"
 	"tinygo.org/x/go-llvm"
 )
 
@@ -34,8 +33,11 @@ func TestClangAttributes(t *testing.T) {
 		"k210",
 		"nintendoswitch",
 		"riscv-qemu",
-		"wasi",
+		"tkey",
+		"wasip1",
+		"wasip2",
 		"wasm",
+		"wasm-unknown",
 	}
 	if hasBuiltinTools {
 		// hasBuiltinTools is set when TinyGo is statically linked with LLVM,
@@ -52,10 +54,17 @@ func TestClangAttributes(t *testing.T) {
 	for _, options := range []*compileopts.Options{
 		{GOOS: "linux", GOARCH: "386"},
 		{GOOS: "linux", GOARCH: "amd64"},
-		{GOOS: "linux", GOARCH: "arm", GOARM: "5"},
-		{GOOS: "linux", GOARCH: "arm", GOARM: "6"},
-		{GOOS: "linux", GOARCH: "arm", GOARM: "7"},
+		{GOOS: "linux", GOARCH: "arm", GOARM: "5,softfloat"},
+		{GOOS: "linux", GOARCH: "arm", GOARM: "6,softfloat"},
+		{GOOS: "linux", GOARCH: "arm", GOARM: "7,softfloat"},
+		{GOOS: "linux", GOARCH: "arm", GOARM: "5,hardfloat"},
+		{GOOS: "linux", GOARCH: "arm", GOARM: "6,hardfloat"},
+		{GOOS: "linux", GOARCH: "arm", GOARM: "7,hardfloat"},
 		{GOOS: "linux", GOARCH: "arm64"},
+		{GOOS: "linux", GOARCH: "mips", GOMIPS: "hardfloat"},
+		{GOOS: "linux", GOARCH: "mipsle", GOMIPS: "hardfloat"},
+		{GOOS: "linux", GOARCH: "mips", GOMIPS: "softfloat"},
+		{GOOS: "linux", GOARCH: "mipsle", GOMIPS: "softfloat"},
 		{GOOS: "darwin", GOARCH: "amd64"},
 		{GOOS: "darwin", GOARCH: "arm64"},
 		{GOOS: "windows", GOARCH: "amd64"},
@@ -65,6 +74,9 @@ func TestClangAttributes(t *testing.T) {
 		if options.GOARCH == "arm" {
 			name += ",GOARM=" + options.GOARM
 		}
+		if options.GOARCH == "mips" || options.GOARCH == "mipsle" {
+			name += ",GOMIPS=" + options.GOMIPS
+		}
 		t.Run(name, func(t *testing.T) {
 			testClangAttributes(t, options)
 		})
@@ -73,7 +85,6 @@ func TestClangAttributes(t *testing.T) {
 
 func testClangAttributes(t *testing.T, options *compileopts.Options) {
 	testDir := t.TempDir()
-	clangHeaderPath := getClangHeaderPath(goenv.Get("TINYGOROOT"))
 
 	ctx := llvm.NewContext()
 	defer ctx.Dispose()
@@ -83,9 +94,8 @@ func testClangAttributes(t *testing.T, options *compileopts.Options) {
 		t.Fatalf("could not load target: %s", err)
 	}
 	config := compileopts.Config{
-		Options:      options,
-		Target:       target,
-		ClangHeaders: clangHeaderPath,
+		Options: options,
+		Target:  target,
 	}
 
 	// Create a very simple C input file.
@@ -97,7 +107,7 @@ func testClangAttributes(t *testing.T, options *compileopts.Options) {
 
 	// Compile this file using Clang.
 	outpath := filepath.Join(testDir, "test.bc")
-	flags := append([]string{"-c", "-emit-llvm", "-o", outpath, srcpath}, config.CFlags()...)
+	flags := append([]string{"-c", "-emit-llvm", "-o", outpath, srcpath}, config.CFlags(false)...)
 	if config.GOOS() == "darwin" {
 		// Silence some warnings that happen when testing GOOS=darwin on
 		// something other than MacOS.

builder/buildid.go

@@ -77,6 +77,15 @@ func ReadBuildID() ([]byte, error) {
 			}
 			return raw[4:], nil
 		}
+
+		// Normally we would have found a build ID by now. But not on Nix,
+		// unfortunately, because Nix adds -no_uuid for some reason:
+		// https://github.com/NixOS/nixpkgs/issues/178366
+		// Fall back to the same implementation that we use for Windows.
+		id, err := readRawGoBuildID(f, 32*1024)
+		if len(id) != 0 || err != nil {
+			return id, err
+		}
 	default:
 		// On other platforms (such as Windows) there isn't such a convenient
 		// build ID. Luckily, Go does have an equivalent of the build ID, which
@@ -88,16 +97,31 @@ func ReadBuildID() ([]byte, error) {
 		// directly. Luckily the build ID is always at the start of the file.
 		// For details, see:
 		// https://github.com/golang/go/blob/master/src/cmd/internal/buildid/buildid.go
-		fileStart := make([]byte, 4096)
-		_, err := io.ReadFull(f, fileStart)
-		index := bytes.Index(fileStart, []byte("\xff Go build ID: \""))
-		if index < 0 || index > len(fileStart)-103 {
-			return nil, fmt.Errorf("could not find build id in %s", err)
-		}
-		buf := fileStart[index : index+103]
-		if bytes.HasPrefix(buf, []byte("\xff Go build ID: \"")) && bytes.HasSuffix(buf, []byte("\"\n \xff")) {
-			return buf[len("\xff Go build ID: \"") : len(buf)-1], nil
+		id, err := readRawGoBuildID(f, 4096)
+		if len(id) != 0 || err != nil {
+			return id, err
 		}
 	}
-	return nil, fmt.Errorf("could not find build ID in %s", executable)
+	return nil, fmt.Errorf("could not find build ID in %v", executable)
+}
+
+// The Go toolchain stores a build ID in the binary that we can use, as a
+// fallback if binary file specific build IDs can't be obtained.
+// This function reads that build ID from the binary.
+func readRawGoBuildID(f *os.File, prefixSize int) ([]byte, error) {
+	fileStart := make([]byte, prefixSize)
+	_, err := io.ReadFull(f, fileStart)
+	if err != nil {
+		return nil, fmt.Errorf("could not read build id from %s: %v", f.Name(), err)
+	}
+	index := bytes.Index(fileStart, []byte("\xff Go build ID: \""))
+	if index < 0 || index > len(fileStart)-103 {
+		return nil, fmt.Errorf("could not find build id in %s", f.Name())
+	}
+	buf := fileStart[index : index+103]
+	if bytes.HasPrefix(buf, []byte("\xff Go build ID: \"")) && bytes.HasSuffix(buf, []byte("\"\n \xff")) {
+		return buf[len("\xff Go build ID: \"") : len(buf)-1], nil
+	}
+
+	return nil, nil
 }

builder/builtins.go

@@ -3,19 +3,19 @@ package builder
 import (
 	"os"
 	"path/filepath"
-	"strings"
 
+	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/goenv"
 )
 
-// These are the GENERIC_SOURCES according to CMakeList.txt.
+// These are the GENERIC_SOURCES according to CMakeList.txt except for
+// divmodsi4.c and udivmodsi4.c.
 var genericBuiltins = []string{
 	"absvdi2.c",
 	"absvsi2.c",
 	"absvti2.c",
 	"adddf3.c",
 	"addsf3.c",
-	"addtf3.c",
 	"addvdi3.c",
 	"addvsi3.c",
 	"addvti3.c",
@@ -40,12 +40,12 @@ var genericBuiltins = []string{
 	"divdf3.c",
 	"divdi3.c",
 	"divmoddi4.c",
+	//"divmodsi4.c",
+	"divmodti4.c",
 	"divsc3.c",
 	"divsf3.c",
 	"divsi3.c",
-	"divtc3.c",
 	"divti3.c",
-	"divtf3.c",
 	"extendsfdf2.c",
 	"extendhfsf2.c",
 	"ffsdi2.c",
@@ -91,7 +91,6 @@ var genericBuiltins = []string{
 	"mulsc3.c",
 	"mulsf3.c",
 	"multi3.c",
-	"multf3.c",
 	"mulvdi3.c",
 	"mulvsi3.c",
 	"mulvti3.c",
@@ -111,13 +110,11 @@ var genericBuiltins = []string{
 	"popcountti2.c",
 	"powidf2.c",
 	"powisf2.c",
-	"powitf2.c",
 	"subdf3.c",
 	"subsf3.c",
 	"subvdi3.c",
 	"subvsi3.c",
 	"subvti3.c",
-	"subtf3.c",
 	"trampoline_setup.c",
 	"truncdfhf2.c",
 	"truncdfsf2.c",
@@ -126,6 +123,7 @@ var genericBuiltins = []string{
 	"ucmpti2.c",
 	"udivdi3.c",
 	"udivmoddi4.c",
+	//"udivmodsi4.c",
 	"udivmodti4.c",
 	"udivsi3.c",
 	"udivti3.c",
@@ -134,6 +132,38 @@ var genericBuiltins = []string{
 	"umodti3.c",
 }
 
+// These are the GENERIC_TF_SOURCES as of LLVM 18.
+// They are not needed on all platforms (32-bit platforms usually don't need
+// these) but they seem to compile fine so it's easier to include them.
+var genericBuiltins128 = []string{
+	"addtf3.c",
+	"comparetf2.c",
+	"divtc3.c",
+	"divtf3.c",
+	"extenddftf2.c",
+	"extendhftf2.c",
+	"extendsftf2.c",
+	"fixtfdi.c",
+	"fixtfsi.c",
+	"fixtfti.c",
+	"fixunstfdi.c",
+	"fixunstfsi.c",
+	"fixunstfti.c",
+	"floatditf.c",
+	"floatsitf.c",
+	"floattitf.c",
+	"floatunditf.c",
+	"floatunsitf.c",
+	"floatuntitf.c",
+	"multc3.c",
+	"multf3.c",
+	"powitf2.c",
+	"subtf3.c",
+	"trunctfdf2.c",
+	"trunctfhf2.c",
+	"trunctfsf2.c",
+}
+
 var aeabiBuiltins = []string{
 	"arm/aeabi_cdcmp.S",
 	"arm/aeabi_cdcmpeq_check_nan.c",
@@ -171,12 +201,12 @@ var avrBuiltins = []string{
 	"avr/udivmodqi4.S",
 }
 
-// CompilerRT is a library with symbols required by programs compiled with LLVM.
-// These symbols are for operations that cannot be emitted with a single
+// libCompilerRT is a library with symbols required by programs compiled with
+// LLVM. These symbols are for operations that cannot be emitted with a single
 // instruction or a short sequence of instructions for that target.
 //
 // For more information, see: https://compiler-rt.llvm.org/
-var CompilerRT = Library{
+var libCompilerRT = Library{
 	name: "compiler-rt",
 	cflags: func(target, headerPath string) []string {
 		return []string{"-Werror", "-Wall", "-std=c11", "-nostdlibinc"}
@@ -192,11 +222,13 @@ var CompilerRT = Library{
 	},
 	librarySources: func(target string) ([]string, error) {
 		builtins := append([]string{}, genericBuiltins...) // copy genericBuiltins
-		if strings.HasPrefix(target, "arm") || strings.HasPrefix(target, "thumb") {
+		switch compileopts.CanonicalArchName(target) {
+		case "arm":
 			builtins = append(builtins, aeabiBuiltins...)
-		}
-		if strings.HasPrefix(target, "avr") {
+		case "avr":
 			builtins = append(builtins, avrBuiltins...)
+		case "x86_64", "aarch64", "riscv64": // any 64-bit arch
+			builtins = append(builtins, genericBuiltins128...)
 		}
 		return builtins, nil
 	},

builder/cc1as.cpp

@@ -1,5 +1,12 @@
 //go:build byollvm
 
+// Source: https://github.com/llvm/llvm-project/blob/main/clang/tools/driver/cc1as_main.cpp
+// This file needs to be updated each LLVM release.
+// There are a few small modifications to make, like:
+//   * ExecuteAssembler is made non-static.
+//   * The struct AssemblerImplementation is moved to cc1as.h so it can be
+//     included elsewhere.
+
 //===-- cc1as.cpp - Clang Assembler  --------------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
@@ -21,8 +28,8 @@
 #include "clang/Frontend/TextDiagnosticPrinter.h"
 #include "clang/Frontend/Utils.h"
 #include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringSwitch.h"
-#include "llvm/ADT/Triple.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/MC/MCAsmBackend.h"
 #include "llvm/MC/MCAsmInfo.h"
@@ -46,7 +53,6 @@
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/FormattedStream.h"
-#include "llvm/Support/Host.h"
 #include "llvm/Support/MemoryBuffer.h"
 #include "llvm/Support/Path.h"
 #include "llvm/Support/Process.h"
@@ -55,7 +61,10 @@
 #include "llvm/Support/TargetSelect.h"
 #include "llvm/Support/Timer.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/TargetParser/Host.h"
+#include "llvm/TargetParser/Triple.h"
 #include <memory>
+#include <optional>
 #include <system_error>
 using namespace clang;
 using namespace clang::driver;
@@ -73,10 +82,10 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
   // Parse the arguments.
   const OptTable &OptTbl = getDriverOptTable();
 
-  const unsigned IncludedFlagsBitmask = options::CC1AsOption;
+  llvm::opt::Visibility VisibilityMask(options::CC1AsOption);
   unsigned MissingArgIndex, MissingArgCount;
-  InputArgList Args = OptTbl.ParseArgs(Argv, MissingArgIndex, MissingArgCount,
-                                       IncludedFlagsBitmask);
+  InputArgList Args =
+      OptTbl.ParseArgs(Argv, MissingArgIndex, MissingArgCount, VisibilityMask);
 
   // Check for missing argument error.
   if (MissingArgCount) {
@@ -89,7 +98,7 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
   for (const Arg *A : Args.filtered(OPT_UNKNOWN)) {
     auto ArgString = A->getAsString(Args);
     std::string Nearest;
-    if (OptTbl.findNearest(ArgString, Nearest, IncludedFlagsBitmask) > 1)
+    if (OptTbl.findNearest(ArgString, Nearest, VisibilityMask) > 1)
       Diags.Report(diag::err_drv_unknown_argument) << ArgString;
     else
       Diags.Report(diag::err_drv_unknown_argument_with_suggestion)
@@ -103,6 +112,14 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
   Opts.Triple = llvm::Triple::normalize(Args.getLastArgValue(OPT_triple));
   if (Arg *A = Args.getLastArg(options::OPT_darwin_target_variant_triple))
     Opts.DarwinTargetVariantTriple = llvm::Triple(A->getValue());
+  if (Arg *A = Args.getLastArg(OPT_darwin_target_variant_sdk_version_EQ)) {
+    VersionTuple Version;
+    if (Version.tryParse(A->getValue()))
+      Diags.Report(diag::err_drv_invalid_value)
+          << A->getAsString(Args) << A->getValue();
+    else
+      Opts.DarwinTargetVariantSDKVersion = Version;
+  }
 
   Opts.CPU = std::string(Args.getLastArgValue(OPT_target_cpu));
   Opts.Features = Args.getAllArgValues(OPT_target_feature);
@@ -122,11 +139,13 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
     Opts.CompressDebugSections =
         llvm::StringSwitch<llvm::DebugCompressionType>(A->getValue())
             .Case("none", llvm::DebugCompressionType::None)
-            .Case("zlib", llvm::DebugCompressionType::Z)
+            .Case("zlib", llvm::DebugCompressionType::Zlib)
+            .Case("zstd", llvm::DebugCompressionType::Zstd)
             .Default(llvm::DebugCompressionType::None);
   }
 
-  Opts.RelaxELFRelocations = Args.hasArg(OPT_mrelax_relocations);
+  Opts.RelaxELFRelocations = !Args.hasArg(OPT_mrelax_relocations_no);
+  Opts.SSE2AVX = Args.hasArg(OPT_msse2avx);
   if (auto *DwarfFormatArg = Args.getLastArg(OPT_gdwarf64, OPT_gdwarf32))
     Opts.Dwarf64 = DwarfFormatArg->getOption().matches(OPT_gdwarf64);
   Opts.DwarfVersion = getLastArgIntValue(Args, OPT_dwarf_version_EQ, 2, Diags);
@@ -141,8 +160,7 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
 
   for (const auto &Arg : Args.getAllArgValues(OPT_fdebug_prefix_map_EQ)) {
     auto Split = StringRef(Arg).split('=');
-    Opts.DebugPrefixMap.insert(
-        {std::string(Split.first), std::string(Split.second)});
+    Opts.DebugPrefixMap.emplace_back(Split.first, Split.second);
   }
 
   // Frontend Options
@@ -189,6 +207,7 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
   Opts.NoExecStack = Args.hasArg(OPT_mno_exec_stack);
   Opts.FatalWarnings = Args.hasArg(OPT_massembler_fatal_warnings);
   Opts.NoWarn = Args.hasArg(OPT_massembler_no_warn);
+  Opts.NoTypeCheck = Args.hasArg(OPT_mno_type_check);
   Opts.RelocationModel =
       std::string(Args.getLastArgValue(OPT_mrelocation_model, "pic"));
   Opts.TargetABI = std::string(Args.getLastArgValue(OPT_target_abi));
@@ -214,6 +233,12 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
             .Case("default", EmitDwarfUnwindType::Default);
   }
 
+  Opts.EmitCompactUnwindNonCanonical =
+      Args.hasArg(OPT_femit_compact_unwind_non_canonical);
+  Opts.Crel = Args.hasArg(OPT_crel);
+
+  Opts.AsSecureLogFile = Args.getLastArgValue(OPT_as_secure_log_file);
+
   return Success;
 }
 
@@ -247,8 +272,8 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
       MemoryBuffer::getFileOrSTDIN(Opts.InputFile, /*IsText=*/true);
 
   if (std::error_code EC = Buffer.getError()) {
-    Error = EC.message();
-    return Diags.Report(diag::err_fe_error_reading) << Opts.InputFile;
+    return Diags.Report(diag::err_fe_error_reading)
+           << Opts.InputFile << EC.message();
   }
 
   SourceMgr SrcMgr;
@@ -264,7 +289,15 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
   assert(MRI && "Unable to create target register info!");
 
   MCTargetOptions MCOptions;
+  MCOptions.MCRelaxAll = Opts.RelaxAll;
   MCOptions.EmitDwarfUnwind = Opts.EmitDwarfUnwind;
+  MCOptions.EmitCompactUnwindNonCanonical = Opts.EmitCompactUnwindNonCanonical;
+  MCOptions.MCSaveTempLabels = Opts.SaveTemporaryLabels;
+  MCOptions.Crel = Opts.Crel;
+  MCOptions.X86RelaxRelocations = Opts.RelaxELFRelocations;
+  MCOptions.X86Sse2Avx = Opts.SSE2AVX;
+  MCOptions.CompressDebugSections = Opts.CompressDebugSections;
+  MCOptions.AsSecureLogFile = Opts.AsSecureLogFile;
 
   std::unique_ptr<MCAsmInfo> MAI(
       TheTarget->createMCAsmInfo(*MRI, Opts.Triple, MCOptions));
@@ -272,9 +305,7 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
 
   // Ensure MCAsmInfo initialization occurs before any use, otherwise sections
   // may be created with a combination of default and explicit settings.
-  MAI->setCompressDebugSections(Opts.CompressDebugSections);
 
-  MAI->setRelaxELFRelocations(Opts.RelaxELFRelocations);
 
   bool IsBinary = Opts.OutputType == AssemblerInvocation::FT_Obj;
   if (Opts.OutputPath.empty())
@@ -314,10 +345,10 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
       TheTarget->createMCObjectFileInfo(Ctx, PIC));
   if (Opts.DarwinTargetVariantTriple)
     MOFI->setDarwinTargetVariantTriple(*Opts.DarwinTargetVariantTriple);
+  if (!Opts.DarwinTargetVariantSDKVersion.empty())
+    MOFI->setDarwinTargetVariantSDKVersion(Opts.DarwinTargetVariantSDKVersion);
   Ctx.setObjectFileInfo(MOFI.get());
 
-  if (Opts.SaveTemporaryLabels)
-    Ctx.setAllowTemporaryLabels(false);
   if (Opts.GenDwarfForAssembly)
     Ctx.setGenDwarfForAssembly(true);
   if (!Opts.DwarfDebugFlags.empty())
@@ -353,6 +384,10 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
 
   MCOptions.MCNoWarn = Opts.NoWarn;
   MCOptions.MCFatalWarnings = Opts.FatalWarnings;
+  MCOptions.MCNoTypeCheck = Opts.NoTypeCheck;
+  MCOptions.ShowMCInst = Opts.ShowInst;
+  MCOptions.AsmVerbose = true;
+  MCOptions.MCUseDwarfDirectory = MCTargetOptions::EnableDwarfDirectory;
   MCOptions.ABIName = Opts.TargetABI;
 
   // FIXME: There is a bit of code duplication with addPassesToEmitFile.
@@ -367,10 +402,8 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
         TheTarget->createMCAsmBackend(*STI, *MRI, MCOptions));
 
     auto FOut = std::make_unique<formatted_raw_ostream>(*Out);
-    Str.reset(TheTarget->createAsmStreamer(
-        Ctx, std::move(FOut), /*asmverbose*/ true,
-        /*useDwarfDirectory*/ true, IP, std::move(CE), std::move(MAB),
-        Opts.ShowInst));
+    Str.reset(TheTarget->createAsmStreamer(Ctx, std::move(FOut), IP,
+                                           std::move(CE), std::move(MAB)));
   } else if (Opts.OutputType == AssemblerInvocation::FT_Null) {
     Str.reset(createNullStreamer(Ctx));
   } else {
@@ -393,9 +426,7 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
 
     Triple T(Opts.Triple);
     Str.reset(TheTarget->createMCObjectStreamer(
-        T, Ctx, std::move(MAB), std::move(OW), std::move(CE), *STI,
-        Opts.RelaxAll, Opts.IncrementalLinkerCompatible,
-        /*DWARFMustBeAtTheEnd*/ true));
+        T, Ctx, std::move(MAB), std::move(OW), std::move(CE), *STI));
     Str.get()->initSections(Opts.NoExecStack, *STI);
   }
 
@@ -408,9 +439,6 @@ static bool ExecuteAssemblerImpl(AssemblerInvocation &Opts,
     Str.get()->emitZeros(1);
   }
 
-  // Assembly to object compilation should leverage assembly info.
-  Str->setUseAssemblerInfoForParsing(true);
-
   bool Failed = false;
 
   std::unique_ptr<MCAsmParser> Parser(
@@ -493,9 +521,10 @@ int cc1as_main(ArrayRef<const char *> Argv, const char *Argv0, void *MainAddr) {
   if (Asm.ShowHelp) {
     getDriverOptTable().printHelp(
         llvm::outs(), "clang -cc1as [options] file...",
-        "Clang Integrated Assembler",
-        /*Include=*/driver::options::CC1AsOption, /*Exclude=*/0,
-        /*ShowAllAliases=*/false);
+        "Clang Integrated Assembler", /*ShowHidden=*/false,
+        /*ShowAllAliases=*/false,
+        llvm::opt::Visibility(driver::options::CC1AsOption));
+
     return 0;
   }
 

builder/cc1as.h

@@ -1,3 +1,6 @@
+// Source: https://github.com/llvm/llvm-project/blob/main/clang/tools/driver/cc1as_main.cpp
+// See cc1as.cpp for details.
+
 //===-- cc1as.h - Clang Assembler  ----------------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
@@ -35,16 +38,23 @@ struct AssemblerInvocation {
   /// @{
 
   std::vector<std::string> IncludePaths;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned NoInitialTextSection : 1;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned SaveTemporaryLabels : 1;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned GenDwarfForAssembly : 1;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned RelaxELFRelocations : 1;
+  LLVM_PREFERRED_TYPE(bool)
+  unsigned SSE2AVX : 1;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned Dwarf64 : 1;
   unsigned DwarfVersion;
   std::string DwarfDebugFlags;
   std::string DwarfDebugProducer;
   std::string DebugCompilationDir;
-  std::map<const std::string, const std::string> DebugPrefixMap;
+  llvm::SmallVector<std::pair<std::string, std::string>, 0> DebugPrefixMap;
   llvm::DebugCompressionType CompressDebugSections =
       llvm::DebugCompressionType::None;
   std::string MainFileName;
@@ -63,7 +73,9 @@ struct AssemblerInvocation {
     FT_Obj   ///< Object file output.
   };
   FileType OutputType;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned ShowHelp : 1;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned ShowVersion : 1;
 
   /// @}
@@ -71,23 +83,41 @@ struct AssemblerInvocation {
   /// @{
 
   unsigned OutputAsmVariant;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned ShowEncoding : 1;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned ShowInst : 1;
 
   /// @}
   /// @name Assembler Options
   /// @{
 
+  LLVM_PREFERRED_TYPE(bool)
   unsigned RelaxAll : 1;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned NoExecStack : 1;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned FatalWarnings : 1;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned NoWarn : 1;
+  LLVM_PREFERRED_TYPE(bool)
+  unsigned NoTypeCheck : 1;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned IncrementalLinkerCompatible : 1;
+  LLVM_PREFERRED_TYPE(bool)
   unsigned EmbedBitcode : 1;
 
   /// Whether to emit DWARF unwind info.
   EmitDwarfUnwindType EmitDwarfUnwind;
 
+  // Whether to emit compact-unwind for non-canonical entries.
+  // Note: maybe overriden by other constraints.
+  LLVM_PREFERRED_TYPE(bool)
+  unsigned EmitCompactUnwindNonCanonical : 1;
+
+  LLVM_PREFERRED_TYPE(bool)
+  unsigned Crel : 1;
+
   /// The name of the relocation model to use.
   std::string RelocationModel;
 
@@ -97,7 +127,14 @@ struct AssemblerInvocation {
 
   /// Darwin target variant triple, the variant of the deployment target
   /// for which the code is being compiled.
-  llvm::Optional<llvm::Triple> DarwinTargetVariantTriple;
+  std::optional<llvm::Triple> DarwinTargetVariantTriple;
+
+  /// The version of the darwin target variant SDK which was used during the
+  /// compilation
+  llvm::VersionTuple DarwinTargetVariantSDKVersion;
+
+  /// The name of a file to use with \c .secure_log_unique directives.
+  std::string AsSecureLogFile;
   /// @}
 
 public:
@@ -111,14 +148,18 @@ public:
     ShowInst = 0;
     ShowEncoding = 0;
     RelaxAll = 0;
+    SSE2AVX = 0;
     NoExecStack = 0;
     FatalWarnings = 0;
     NoWarn = 0;
+    NoTypeCheck = 0;
     IncrementalLinkerCompatible = 0;
     Dwarf64 = 0;
     DwarfVersion = 0;
     EmbedBitcode = 0;
     EmitDwarfUnwind = EmitDwarfUnwindType::Default;
+    EmitCompactUnwindNonCanonical = false;
+    Crel = false;
   }
 
   static bool CreateFromArgs(AssemblerInvocation &Res,

builder/clang.cpp

@@ -11,7 +11,7 @@
 #include <clang/FrontendTool/Utils.h>
 #include <llvm/ADT/IntrusiveRefCntPtr.h>
 #include <llvm/Option/Option.h>
-#include <llvm/Support/Host.h>
+#include <llvm/TargetParser/Host.h>
 
 using namespace llvm;
 using namespace clang;

builder/commands.go

@@ -3,7 +3,6 @@ package builder
 import (
 	"errors"
 	"fmt"
-	"os"
 	"os/exec"
 	"runtime"
 	"strings"
@@ -76,14 +75,3 @@ func LookupCommand(name string) (string, error) {
 	}
 	return "", errors.New("none of these commands were found in your $PATH: " + strings.Join(commands[name], " "))
 }
-
-func execCommand(name string, args ...string) error {
-	name, err := LookupCommand(name)
-	if err != nil {
-		return err
-	}
-	cmd := exec.Command(name, args...)
-	cmd.Stdout = os.Stdout
-	cmd.Stderr = os.Stderr
-	return cmd.Run()
-}

builder/config.go

@@ -1,8 +1,8 @@
 package builder
 
 import (
-	"errors"
 	"fmt"
+	"runtime"
 
 	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/goenv"
@@ -24,28 +24,37 @@ func NewConfig(options *compileopts.Options) (*compileopts.Config, error) {
 		spec.OpenOCDCommands = options.OpenOCDCommands
 	}
 
-	goroot := goenv.Get("GOROOT")
-	if goroot == "" {
-		return nil, errors.New("cannot locate $GOROOT, please set it manually")
-	}
+	// Version range supported by TinyGo.
+	const minorMin = 19
+	const minorMax = 24
 
-	major, minor, err := goenv.GetGorootVersion(goroot)
+	// Check that we support this Go toolchain version.
+	gorootMajor, gorootMinor, err := goenv.GetGorootVersion()
 	if err != nil {
-		return nil, fmt.Errorf("could not read version from GOROOT (%v): %v", goroot, err)
+		return nil, err
 	}
-	if major != 1 || minor < 18 || minor > 20 {
+	if gorootMajor != 1 || gorootMinor < minorMin || gorootMinor > minorMax {
 		// Note: when this gets updated, also update the Go compatibility matrix:
 		// https://github.com/tinygo-org/tinygo-site/blob/dev/content/docs/reference/go-compat-matrix.md
-		return nil, fmt.Errorf("requires go version 1.18 through 1.20, got go%d.%d", major, minor)
+		return nil, fmt.Errorf("requires go version 1.%d through 1.%d, got go%d.%d", minorMin, minorMax, gorootMajor, gorootMinor)
 	}
 
-	clangHeaderPath := getClangHeaderPath(goenv.Get("TINYGOROOT"))
+	// Check that the Go toolchain version isn't too new, if we haven't been
+	// compiled with the latest Go version.
+	// This may be a bit too aggressive: if the newer version doesn't change the
+	// Go language we will most likely be able to compile it.
+	buildMajor, buildMinor, _, err := goenv.Parse(runtime.Version())
+	if err != nil {
+		return nil, err
+	}
+	if buildMajor != 1 || buildMinor < gorootMinor {
+		return nil, fmt.Errorf("cannot compile with Go toolchain version go%d.%d (TinyGo was built using toolchain version %s)", gorootMajor, gorootMinor, runtime.Version())
+	}
 
 	return &compileopts.Config{
 		Options:        options,
 		Target:         spec,
-		GoMinorVersion: minor,
-		ClangHeaders:   clangHeaderPath,
+		GoMinorVersion: gorootMinor,
 		TestConfig:     options.TestConfig,
 	}, nil
 }

builder/env.go

@@ -1,105 +0,0 @@
-package builder
-
-import (
-	"errors"
-	"io/fs"
-	"io/ioutil"
-	"os"
-	"os/exec"
-	"path/filepath"
-	"sort"
-	"strings"
-
-	"tinygo.org/x/go-llvm"
-)
-
-// getClangHeaderPath returns the path to the built-in Clang headers. It tries
-// multiple locations, which should make it find the directory when installed in
-// various ways.
-func getClangHeaderPath(TINYGOROOT string) string {
-	// Check whether we're running from the source directory.
-	path := filepath.Join(TINYGOROOT, "llvm-project", "clang", "lib", "Headers")
-	if _, err := os.Stat(path); !errors.Is(err, fs.ErrNotExist) {
-		return path
-	}
-
-	// Check whether we're running from the installation directory.
-	path = filepath.Join(TINYGOROOT, "lib", "clang", "include")
-	if _, err := os.Stat(path); !errors.Is(err, fs.ErrNotExist) {
-		return path
-	}
-
-	// It looks like we are built with a system-installed LLVM. Do a last
-	// attempt: try to use Clang headers relative to the clang binary.
-	llvmMajor := strings.Split(llvm.Version, ".")[0]
-	for _, cmdName := range commands["clang"] {
-		binpath, err := exec.LookPath(cmdName)
-		if err == nil {
-			// This should be the command that will also be used by
-			// execCommand. To avoid inconsistencies, make sure we use the
-			// headers relative to this command.
-			binpath, err = filepath.EvalSymlinks(binpath)
-			if err != nil {
-				// Unexpected.
-				return ""
-			}
-			// Example executable:
-			//     /usr/lib/llvm-9/bin/clang
-			// Example include path:
-			//     /usr/lib/llvm-9/lib64/clang/9.0.1/include/
-			llvmRoot := filepath.Dir(filepath.Dir(binpath))
-			clangVersionRoot := filepath.Join(llvmRoot, "lib64", "clang")
-			dirs64, err64 := ioutil.ReadDir(clangVersionRoot)
-			// Example include path:
-			//     /usr/lib/llvm-9/lib/clang/9.0.1/include/
-			clangVersionRoot = filepath.Join(llvmRoot, "lib", "clang")
-			dirs32, err32 := ioutil.ReadDir(clangVersionRoot)
-			if err64 != nil && err32 != nil {
-				// Unexpected.
-				continue
-			}
-			dirnames := make([]string, len(dirs64)+len(dirs32))
-			dirCount := 0
-			for _, d := range dirs32 {
-				name := d.Name()
-				if name == llvmMajor || strings.HasPrefix(name, llvmMajor+".") {
-					dirnames[dirCount] = filepath.Join(llvmRoot, "lib", "clang", name)
-					dirCount++
-				}
-			}
-			for _, d := range dirs64 {
-				name := d.Name()
-				if name == llvmMajor || strings.HasPrefix(name, llvmMajor+".") {
-					dirnames[dirCount] = filepath.Join(llvmRoot, "lib64", "clang", name)
-					dirCount++
-				}
-			}
-			sort.Strings(dirnames)
-			// Check for the highest version first.
-			for i := dirCount - 1; i >= 0; i-- {
-				path := filepath.Join(dirnames[i], "include")
-				_, err := os.Stat(filepath.Join(path, "stdint.h"))
-				if err == nil {
-					return path
-				}
-			}
-		}
-	}
-
-	// On Arch Linux, the clang executable is stored in /usr/bin rather than being symlinked from there.
-	// Search directly in /usr/lib for clang.
-	if matches, err := filepath.Glob("/usr/lib/clang/" + llvmMajor + ".*.*"); err == nil {
-		// Check for the highest version first.
-		sort.Strings(matches)
-		for i := len(matches) - 1; i >= 0; i-- {
-			path := filepath.Join(matches[i], "include")
-			_, err := os.Stat(filepath.Join(path, "stdint.h"))
-			if err == nil {
-				return path
-			}
-		}
-	}
-
-	// Could not find it.
-	return ""
-}

builder/error.go

@@ -3,7 +3,8 @@ package builder
 // MultiError is a list of multiple errors (actually: diagnostics) returned
 // during LLVM IR generation.
 type MultiError struct {
-	Errs []error
+	ImportPath string
+	Errs       []error
 }
 
 func (e *MultiError) Error() string {
@@ -14,15 +15,16 @@ func (e *MultiError) Error() string {
 
 // newMultiError returns a *MultiError if there is more than one error, or
 // returns that error directly when there is only one. Passing an empty slice
-// will lead to a panic.
-func newMultiError(errs []error) error {
+// will return nil (because there is no error).
+// The importPath may be passed if this error is for a single package.
+func newMultiError(errs []error, importPath string) error {
 	switch len(errs) {
 	case 0:
-		panic("attempted to create empty MultiError")
+		return nil
 	case 1:
 		return errs[0]
 	default:
-		return &MultiError{errs}
+		return &MultiError{importPath, errs}
 	}
 }
 

builder/jobs.go

@@ -17,14 +17,6 @@ import (
 // concurrency or performance issues.
 const jobRunnerDebug = false
 
-type jobState uint8
-
-const (
-	jobStateQueued   jobState = iota // not yet running
-	jobStateRunning                  // running
-	jobStateFinished                 // finished running
-)
-
 // compileJob is a single compiler job, comparable to a single Makefile target.
 // It is used to orchestrate various compiler tasks that can be run in parallel
 // but that have dependencies and thus have limitations in how they can be run.
@@ -55,12 +47,11 @@ func dummyCompileJob(result string) *compileJob {
 // ordered as such in the job dependencies.
 func runJobs(job *compileJob, sema chan struct{}) error {
 	if sema == nil {
-		// Have a default, if the semaphore isn't set. This is useful for
-		// tests.
+		// Have a default, if the semaphore isn't set. This is useful for tests.
 		sema = make(chan struct{}, runtime.NumCPU())
 	}
 	if cap(sema) == 0 {
-		return errors.New("cannot 0 jobs at a time")
+		return errors.New("cannot run 0 jobs at a time")
 	}
 
 	// Create a slice of jobs to run, where all dependencies are run in order.
@@ -81,10 +72,10 @@ func runJobs(job *compileJob, sema chan struct{}) error {
 
 	waiting := make(map[*compileJob]map[*compileJob]struct{}, len(jobs))
 	dependents := make(map[*compileJob][]*compileJob, len(jobs))
-	jidx := make(map[*compileJob]int)
+	compileJobs := make(map[*compileJob]int)
 	var ready intHeap
 	for i, job := range jobs {
-		jidx[job] = i
+		compileJobs[job] = i
 		if len(job.dependencies) == 0 {
 			// This job is ready to run.
 			ready.Push(i)
@@ -105,8 +96,7 @@ func runJobs(job *compileJob, sema chan struct{}) error {
 	// Create a channel to accept notifications of completion.
 	doneChan := make(chan *compileJob)
 
-	// Send each job in the jobs slice to a worker, taking care of job
-	// dependencies.
+	// Send each job in the jobs slice to a worker, taking care of job dependencies.
 	numRunningJobs := 0
 	var totalTime time.Duration
 	start := time.Now()
@@ -156,7 +146,7 @@ func runJobs(job *compileJob, sema chan struct{}) error {
 			delete(wait, completed)
 			if len(wait) == 0 {
 				// This job is now ready to run.
-				ready.Push(jidx[j])
+				ready.Push(compileJobs[j])
 				delete(waiting, j)
 			}
 		}

builder/library.go

@@ -35,19 +35,6 @@ type Library struct {
 	crt1Source string
 }
 
-// Load the library archive, possibly generating and caching it if needed.
-// The resulting directory may be stored in the provided tmpdir, which is
-// expected to be removed after the Load call.
-func (l *Library) Load(config *compileopts.Config, tmpdir string) (dir string, err error) {
-	job, unlock, err := l.load(config, tmpdir)
-	if err != nil {
-		return "", err
-	}
-	defer unlock()
-	err = runJobs(job, config.Options.Semaphore)
-	return filepath.Dir(job.result), err
-}
-
 // load returns a compile job to build this library file for the given target
 // and CPU. It may return a dummy compileJob if the library build is already
 // cached. The path is stored as job.result but is only valid after the job has
@@ -143,6 +130,10 @@ func (l *Library) load(config *compileopts.Config, tmpdir string) (job *compileJ
 	// reproducible. Otherwise the temporary directory is stored in the archive
 	// itself, which varies each run.
 	args := append(l.cflags(target, headerPath), "-c", "-Oz", "-gdwarf-4", "-ffunction-sections", "-fdata-sections", "-Wno-macro-redefined", "--target="+target, "-fdebug-prefix-map="+dir+"="+remapDir)
+	resourceDir := goenv.ClangResourceDir(false)
+	if resourceDir != "" {
+		args = append(args, "-resource-dir="+resourceDir)
+	}
 	cpu := config.CPU()
 	if cpu != "" {
 		// X86 has deprecated the -mcpu flag, so we need to use -march instead.
@@ -158,31 +149,37 @@ func (l *Library) load(config *compileopts.Config, tmpdir string) (job *compileJ
 	if config.ABI() != "" {
 		args = append(args, "-mabi="+config.ABI())
 	}
-	if strings.HasPrefix(target, "arm") || strings.HasPrefix(target, "thumb") {
+	switch compileopts.CanonicalArchName(target) {
+	case "arm":
 		if strings.Split(target, "-")[2] == "linux" {
 			args = append(args, "-fno-unwind-tables", "-fno-asynchronous-unwind-tables")
 		} else {
 			args = append(args, "-fshort-enums", "-fomit-frame-pointer", "-mfloat-abi=soft", "-fno-unwind-tables", "-fno-asynchronous-unwind-tables")
 		}
-	}
-	if strings.HasPrefix(target, "avr") {
+	case "avr":
 		// AVR defaults to C float and double both being 32-bit. This deviates
 		// from what most code (and certainly compiler-rt) expects. So we need
 		// to force the compiler to use 64-bit floating point numbers for
 		// double.
 		args = append(args, "-mdouble=64")
-	}
-	if strings.HasPrefix(target, "riscv32-") {
+	case "riscv32":
 		args = append(args, "-march=rv32imac", "-fforce-enable-int128")
-	}
-	if strings.HasPrefix(target, "riscv64-") {
+	case "riscv64":
 		args = append(args, "-march=rv64gc")
+	case "mips":
+		args = append(args, "-fno-pic")
 	}
-	if strings.HasPrefix(target, "xtensa") {
-		// Hack to work around an issue in the Xtensa port:
-		// https://github.com/espressif/llvm-project/issues/52
-		// Hopefully this will be fixed soon (LLVM 14).
-		args = append(args, "-D__ELF__")
+	if config.Target.SoftFloat {
+		// Use softfloat instead of floating point instructions. This is
+		// supported on many architectures.
+		args = append(args, "-msoft-float")
+	} else {
+		if strings.HasPrefix(target, "armv5") {
+			// On ARMv5 we need to explicitly enable hardware floating point
+			// instructions: Clang appears to assume the hardware doesn't have a
+			// FPU otherwise.
+			args = append(args, "-mfpu=vfpv2")
+		}
 	}
 
 	var once sync.Once

builder/lld.cpp

@@ -5,7 +5,11 @@
 #include <lld/Common/Driver.h>
 #include <llvm/Support/Parallel.h>
 
-extern "C" {
+LLD_HAS_DRIVER(coff)
+LLD_HAS_DRIVER(elf)
+LLD_HAS_DRIVER(mingw)
+LLD_HAS_DRIVER(macho)
+LLD_HAS_DRIVER(wasm)
 
 static void configure() {
 #if _WIN64
@@ -16,28 +20,13 @@ static void configure() {
 #endif
 }
 
-bool tinygo_link_elf(int argc, char **argv) {
-	configure();
-	std::vector<const char*> args(argv, argv + argc);
-	return lld::elf::link(args, llvm::outs(), llvm::errs(), false, false);
-}
+extern "C" {
 
-bool tinygo_link_macho(int argc, char **argv) {
+bool tinygo_link(int argc, char **argv) {
 	configure();
 	std::vector<const char*> args(argv, argv + argc);
-	return lld::macho::link(args, llvm::outs(), llvm::errs(), false, false);
-}
-
-bool tinygo_link_mingw(int argc, char **argv) {
-	configure();
-	std::vector<const char*> args(argv, argv + argc);
-	return lld::mingw::link(args, llvm::outs(), llvm::errs(), false, false);
-}
-
-bool tinygo_link_wasm(int argc, char **argv) {
-	configure();
-	std::vector<const char*> args(argv, argv + argc);
-	return lld::wasm::link(args, llvm::outs(), llvm::errs(), false, false);
+	lld::Result r = lld::lldMain(args, llvm::outs(), llvm::errs(), LLD_ALL_DRIVERS);
+	return !r.retCode;
 }
 
 } // external "C"

builder/mingw-w64.go

@@ -10,7 +10,7 @@ import (
 	"github.com/tinygo-org/tinygo/goenv"
 )
 
-var MinGW = Library{
+var libMinGW = Library{
 	name: "mingw-w64",
 	makeHeaders: func(target, includeDir string) error {
 		// copy _mingw.h
@@ -27,14 +27,30 @@ var MinGW = Library{
 		_, err = io.Copy(outf, inf)
 		return err
 	},
-	sourceDir: func() string { return "" }, // unused
+	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/mingw-w64") },
 	cflags: func(target, headerPath string) []string {
-		// No flags necessary because there are no files to compile.
-		return nil
+		mingwDir := filepath.Join(goenv.Get("TINYGOROOT"), "lib/mingw-w64")
+		return []string{
+			"-nostdlibinc",
+			"-isystem", mingwDir + "/mingw-w64-headers/crt",
+			"-I", mingwDir + "/mingw-w64-headers/defaults/include",
+			"-I" + headerPath,
+		}
 	},
 	librarySources: func(target string) ([]string, error) {
-		// We only use the UCRT DLL file. No source files necessary.
-		return nil, nil
+		// These files are needed so that printf and the like are supported.
+		sources := []string{
+			"mingw-w64-crt/stdio/ucrt_fprintf.c",
+			"mingw-w64-crt/stdio/ucrt_fwprintf.c",
+			"mingw-w64-crt/stdio/ucrt_printf.c",
+			"mingw-w64-crt/stdio/ucrt_snprintf.c",
+			"mingw-w64-crt/stdio/ucrt_sprintf.c",
+			"mingw-w64-crt/stdio/ucrt_vfprintf.c",
+			"mingw-w64-crt/stdio/ucrt_vprintf.c",
+			"mingw-w64-crt/stdio/ucrt_vsnprintf.c",
+			"mingw-w64-crt/stdio/ucrt_vsprintf.c",
+		}
+		return sources, nil
 	},
 }
 

builder/musl.go

@@ -6,15 +6,13 @@ import (
 	"os"
 	"path/filepath"
 	"regexp"
-	"strconv"
 	"strings"
 
 	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/goenv"
-	"tinygo.org/x/go-llvm"
 )
 
-var Musl = Library{
+var libMusl = Library{
 	name: "musl",
 	makeHeaders: func(target, includeDir string) error {
 		bits := filepath.Join(includeDir, "bits")
@@ -93,6 +91,9 @@ var Musl = Library{
 			"-Wno-string-plus-int",
 			"-Wno-ignored-pragmas",
 			"-Wno-tautological-constant-out-of-range-compare",
+			"-Wno-deprecated-non-prototype",
+			"-Wno-format",
+			"-Wno-parentheses",
 			"-Qunused-arguments",
 			// Select include dirs. Don't include standard library includes
 			// (that would introduce host dependencies and other complications),
@@ -106,11 +107,6 @@ var Musl = Library{
 			"-I" + muslDir + "/include",
 			"-fno-stack-protector",
 		}
-		llvmMajor, _ := strconv.Atoi(strings.SplitN(llvm.Version, ".", 2)[0])
-		if llvmMajor >= 15 {
-			// This flag was added in Clang 15. It is not present in LLVM 14.
-			cflags = append(cflags, "-Wno-deprecated-non-prototype")
-		}
 		return cflags
 	},
 	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/musl/src") },
@@ -120,15 +116,21 @@ var Musl = Library{
 			"env/*.c",
 			"errno/*.c",
 			"exit/*.c",
+			"fcntl/*.c",
 			"internal/defsysinfo.c",
 			"internal/libc.c",
 			"internal/syscall_ret.c",
 			"internal/vdso.c",
 			"legacy/*.c",
+			"locale/*.c",
+			"linux/*.c",
 			"malloc/*.c",
 			"malloc/mallocng/*.c",
 			"mman/*.c",
 			"math/*.c",
+			"misc/*.c",
+			"multibyte/*.c",
+			"signal/" + arch + "/*.s",
 			"signal/*.c",
 			"stdio/*.c",
 			"string/*.c",
@@ -136,12 +138,20 @@ var Musl = Library{
 			"thread/*.c",
 			"time/*.c",
 			"unistd/*.c",
+			"process/*.c",
 		}
+
 		if arch == "arm" {
 			// These files need to be added to the start for some reason.
 			globs = append([]string{"thread/arm/*.c"}, globs...)
 		}
 
+		if arch != "aarch64" && arch != "mips" {
+			//aarch64 and mips have no architecture specific code, either they
+			// are not supported or don't need any?
+			globs = append([]string{"process/" + arch + "/*.s"}, globs...)
+		}
+
 		var sources []string
 		seenSources := map[string]struct{}{}
 		basepath := goenv.Get("TINYGOROOT") + "/lib/musl/src/"

builder/picolibc.go

@@ -3,13 +3,14 @@ package builder
 import (
 	"os"
 	"path/filepath"
+	"strings"
 
 	"github.com/tinygo-org/tinygo/goenv"
 )
 
-// Picolibc is a C library for bare metal embedded devices. It was originally
+// libPicolibc is a C library for bare metal embedded devices. It was originally
 // based on newlib.
-var Picolibc = Library{
+var libPicolibc = Library{
 	name: "picolibc",
 	makeHeaders: func(target, includeDir string) error {
 		f, err := os.Create(filepath.Join(includeDir, "picolibc.h"))
@@ -28,10 +29,12 @@ var Picolibc = Library{
 			"-D_HAVE_ALIAS_ATTRIBUTE",
 			"-DTINY_STDIO",
 			"-DPOSIX_IO",
+			"-DFORMAT_DEFAULT_INTEGER", // use __i_vfprintf and __i_vfscanf by default
 			"-D_IEEE_LIBM",
 			"-D__OBSOLETE_MATH_FLOAT=1", // use old math code that doesn't expect a FPU
 			"-D__OBSOLETE_MATH_DOUBLE=0",
 			"-D_WANT_IO_C99_FORMATS",
+			"-D__PICOLIBC_ERRNO_FUNCTION=__errno_location",
 			"-nostdlibinc",
 			"-isystem", newlibDir + "/libc/include",
 			"-I" + newlibDir + "/libc/tinystdio",
@@ -41,91 +44,23 @@ var Picolibc = Library{
 	},
 	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/picolibc/newlib") },
 	librarySources: func(target string) ([]string, error) {
-		return picolibcSources, nil
+		sources := append([]string(nil), picolibcSources...)
+		if !strings.HasPrefix(target, "avr") {
+			// Small chips without long jumps can't compile many files (printf,
+			// pow, etc). Therefore exclude those source files for those chips.
+			// Unfortunately it's difficult to exclude only some chips, so this
+			// excludes those files on all AVR chips for now.
+			// More information:
+			// https://github.com/llvm/llvm-project/issues/67042
+			sources = append(sources, picolibcSourcesLarge...)
+		}
+		return sources, nil
 	},
 }
 
 var picolibcSources = []string{
 	"../../picolibc-stdio.c",
 
-	// srcs_tinystdio
-	"libc/tinystdio/asprintf.c",
-	"libc/tinystdio/bufio.c",
-	"libc/tinystdio/clearerr.c",
-	"libc/tinystdio/ecvt_r.c",
-	"libc/tinystdio/ecvt.c",
-	"libc/tinystdio/ecvtf_r.c",
-	"libc/tinystdio/ecvtf.c",
-	"libc/tinystdio/fcvt.c",
-	"libc/tinystdio/fcvt_r.c",
-	"libc/tinystdio/fcvtf.c",
-	"libc/tinystdio/fcvtf_r.c",
-	"libc/tinystdio/gcvt.c",
-	"libc/tinystdio/gcvtf.c",
-	"libc/tinystdio/fclose.c",
-	"libc/tinystdio/fdevopen.c",
-	"libc/tinystdio/feof.c",
-	"libc/tinystdio/ferror.c",
-	"libc/tinystdio/fflush.c",
-	"libc/tinystdio/fgetc.c",
-	"libc/tinystdio/fgets.c",
-	"libc/tinystdio/fileno.c",
-	"libc/tinystdio/filestrget.c",
-	"libc/tinystdio/filestrput.c",
-	"libc/tinystdio/filestrputalloc.c",
-	"libc/tinystdio/fmemopen.c",
-	"libc/tinystdio/fprintf.c",
-	"libc/tinystdio/fputc.c",
-	"libc/tinystdio/fputs.c",
-	"libc/tinystdio/fread.c",
-	//"libc/tinystdio/freopen.c", // crashes with AVR, see: https://github.com/picolibc/picolibc/pull/369
-	"libc/tinystdio/fscanf.c",
-	"libc/tinystdio/fseek.c",
-	"libc/tinystdio/fseeko.c",
-	"libc/tinystdio/ftell.c",
-	"libc/tinystdio/ftello.c",
-	"libc/tinystdio/fwrite.c",
-	"libc/tinystdio/getchar.c",
-	"libc/tinystdio/gets.c",
-	"libc/tinystdio/matchcaseprefix.c",
-	"libc/tinystdio/mktemp.c",
-	"libc/tinystdio/perror.c",
-	"libc/tinystdio/printf.c",
-	"libc/tinystdio/putchar.c",
-	"libc/tinystdio/puts.c",
-	"libc/tinystdio/rewind.c",
-	"libc/tinystdio/scanf.c",
-	"libc/tinystdio/setbuf.c",
-	"libc/tinystdio/setbuffer.c",
-	"libc/tinystdio/setlinebuf.c",
-	"libc/tinystdio/setvbuf.c",
-	"libc/tinystdio/snprintf.c",
-	"libc/tinystdio/sprintf.c",
-	"libc/tinystdio/snprintfd.c",
-	"libc/tinystdio/snprintff.c",
-	"libc/tinystdio/sprintff.c",
-	"libc/tinystdio/sprintfd.c",
-	"libc/tinystdio/sscanf.c",
-	"libc/tinystdio/strfromf.c",
-	"libc/tinystdio/strfromd.c",
-	"libc/tinystdio/strtof.c",
-	"libc/tinystdio/strtof_l.c",
-	"libc/tinystdio/strtod.c",
-	"libc/tinystdio/strtod_l.c",
-	"libc/tinystdio/ungetc.c",
-	"libc/tinystdio/vasprintf.c",
-	"libc/tinystdio/vfiprintf.c",
-	"libc/tinystdio/vfprintf.c",
-	"libc/tinystdio/vfprintff.c",
-	"libc/tinystdio/vfscanf.c",
-	"libc/tinystdio/vfiscanf.c",
-	"libc/tinystdio/vfscanff.c",
-	"libc/tinystdio/vprintf.c",
-	"libc/tinystdio/vscanf.c",
-	"libc/tinystdio/vsscanf.c",
-	"libc/tinystdio/vsnprintf.c",
-	"libc/tinystdio/vsprintf.c",
-
 	"libc/string/bcmp.c",
 	"libc/string/bcopy.c",
 	"libc/string/bzero.c",
@@ -229,6 +164,87 @@ var picolibcSources = []string{
 	"libc/string/wmempcpy.c",
 	"libc/string/wmemset.c",
 	"libc/string/xpg_strerror_r.c",
+}
+
+// Parts of picolibc that are too large for small AVRs.
+var picolibcSourcesLarge = []string{
+	// srcs_tinystdio
+	"libc/tinystdio/asprintf.c",
+	"libc/tinystdio/bufio.c",
+	"libc/tinystdio/clearerr.c",
+	"libc/tinystdio/ecvt_r.c",
+	"libc/tinystdio/ecvt.c",
+	"libc/tinystdio/ecvtf_r.c",
+	"libc/tinystdio/ecvtf.c",
+	"libc/tinystdio/fcvt.c",
+	"libc/tinystdio/fcvt_r.c",
+	"libc/tinystdio/fcvtf.c",
+	"libc/tinystdio/fcvtf_r.c",
+	"libc/tinystdio/gcvt.c",
+	"libc/tinystdio/gcvtf.c",
+	"libc/tinystdio/fclose.c",
+	"libc/tinystdio/fdevopen.c",
+	"libc/tinystdio/feof.c",
+	"libc/tinystdio/ferror.c",
+	"libc/tinystdio/fflush.c",
+	"libc/tinystdio/fgetc.c",
+	"libc/tinystdio/fgets.c",
+	"libc/tinystdio/fileno.c",
+	"libc/tinystdio/filestrget.c",
+	"libc/tinystdio/filestrput.c",
+	"libc/tinystdio/filestrputalloc.c",
+	"libc/tinystdio/fmemopen.c",
+	"libc/tinystdio/fprintf.c",
+	"libc/tinystdio/fputc.c",
+	"libc/tinystdio/fputs.c",
+	"libc/tinystdio/fread.c",
+	//"libc/tinystdio/freopen.c", // crashes with AVR, see: https://github.com/picolibc/picolibc/pull/369
+	"libc/tinystdio/fscanf.c",
+	"libc/tinystdio/fseek.c",
+	"libc/tinystdio/fseeko.c",
+	"libc/tinystdio/ftell.c",
+	"libc/tinystdio/ftello.c",
+	"libc/tinystdio/fwrite.c",
+	"libc/tinystdio/getchar.c",
+	"libc/tinystdio/gets.c",
+	"libc/tinystdio/matchcaseprefix.c",
+	"libc/tinystdio/mktemp.c",
+	"libc/tinystdio/perror.c",
+	"libc/tinystdio/printf.c",
+	"libc/tinystdio/putchar.c",
+	"libc/tinystdio/puts.c",
+	"libc/tinystdio/rewind.c",
+	"libc/tinystdio/scanf.c",
+	"libc/tinystdio/setbuf.c",
+	"libc/tinystdio/setbuffer.c",
+	"libc/tinystdio/setlinebuf.c",
+	"libc/tinystdio/setvbuf.c",
+	"libc/tinystdio/snprintf.c",
+	"libc/tinystdio/sprintf.c",
+	"libc/tinystdio/snprintfd.c",
+	"libc/tinystdio/snprintff.c",
+	"libc/tinystdio/sprintff.c",
+	"libc/tinystdio/sprintfd.c",
+	"libc/tinystdio/sscanf.c",
+	"libc/tinystdio/strfromf.c",
+	"libc/tinystdio/strfromd.c",
+	"libc/tinystdio/strtof.c",
+	"libc/tinystdio/strtof_l.c",
+	"libc/tinystdio/strtod.c",
+	"libc/tinystdio/strtod_l.c",
+	"libc/tinystdio/ungetc.c",
+	"libc/tinystdio/vasprintf.c",
+	"libc/tinystdio/vfiprintf.c",
+	"libc/tinystdio/vfprintf.c",
+	"libc/tinystdio/vfprintff.c",
+	"libc/tinystdio/vfscanf.c",
+	"libc/tinystdio/vfiscanf.c",
+	"libc/tinystdio/vfscanff.c",
+	"libc/tinystdio/vprintf.c",
+	"libc/tinystdio/vscanf.c",
+	"libc/tinystdio/vsscanf.c",
+	"libc/tinystdio/vsnprintf.c",
+	"libc/tinystdio/vsprintf.c",
 
 	"libm/common/sf_finite.c",
 	"libm/common/sf_copysign.c",
@@ -323,6 +339,12 @@ var picolibcSources = []string{
 	"libm/common/math_err_may_uflow.c",
 	"libm/common/math_err_check_uflow.c",
 	"libm/common/math_err_check_oflow.c",
+	"libm/common/math_errf_divzerof.c",
+	"libm/common/math_errf_invalidf.c",
+	"libm/common/math_errf_may_uflowf.c",
+	"libm/common/math_errf_oflowf.c",
+	"libm/common/math_errf_uflowf.c",
+	"libm/common/math_errf_with_errnof.c",
 	"libm/common/math_inexact.c",
 	"libm/common/math_inexactf.c",
 	"libm/common/log.c",

builder/size-report.go

@@ -0,0 +1,56 @@
+package builder
+
+import (
+	_ "embed"
+	"fmt"
+	"html/template"
+	"os"
+)
+
+//go:embed size-report.html
+var sizeReportBase string
+
+func writeSizeReport(sizes *programSize, filename, pkgName string) error {
+	tmpl, err := template.New("report").Parse(sizeReportBase)
+	if err != nil {
+		return err
+	}
+
+	f, err := os.Create(filename)
+	if err != nil {
+		return fmt.Errorf("could not open report file: %w", err)
+	}
+	defer f.Close()
+
+	// Prepare data for the report.
+	type sizeLine struct {
+		Name string
+		Size *packageSize
+	}
+	programData := []sizeLine{}
+	for _, name := range sizes.sortedPackageNames() {
+		pkgSize := sizes.Packages[name]
+		programData = append(programData, sizeLine{
+			Name: name,
+			Size: pkgSize,
+		})
+	}
+	sizeTotal := map[string]uint64{
+		"code":   sizes.Code,
+		"rodata": sizes.ROData,
+		"data":   sizes.Data,
+		"bss":    sizes.BSS,
+		"flash":  sizes.Flash(),
+	}
+
+	// Write the report.
+	err = tmpl.Execute(f, map[string]any{
+		"pkgName":   pkgName,
+		"sizes":     programData,
+		"sizeTotal": sizeTotal,
+	})
+	if err != nil {
+		return fmt.Errorf("could not create report file: %w", err)
+	}
+	return nil
+}

builder/size-report.html

@@ -0,0 +1,109 @@
+<!DOCTYPE html>
+<html lang="en">
+  <head>
+    <title>Size Report for {{.pkgName}}</title>
+    <meta charset="utf-8"/>
+    <meta name="viewport" content="width=device-width, initial-scale=1">
+    <link href="https://cdn.jsdelivr.net/npm/bootstrap@5.3.3/dist/css/bootstrap.min.css" rel="stylesheet" integrity="sha384-QWTKZyjpPEjISv5WaRU9OFeRpok6YctnYmDr5pNlyT2bRjXh0JMhjY6hW+ALEwIH" crossorigin="anonymous">
+    <style>
+
+.table-vertical-border {
+  border-left: calc(var(--bs-border-width) * 2) solid currentcolor;
+}
+
+/* Hover on only the rows that are clickable. */
+.row-package:hover > * {
+  --bs-table-color-state: var(--bs-table-hover-color);
+  --bs-table-bg-state: var(--bs-table-hover-bg);
+}
+
+    </style>
+  </head>
+  <body>
+    <div class="container-xxl">
+      <h1>Size Report for {{.pkgName}}</h1>
+
+      <p>How much space is used by Go packages, C libraries, and other bits to set up the program environment.</p>
+
+      <ul>
+        <li><strong>Code</strong> is the actual program code (machine code instructions).</li>
+        <li><strong>Read-only data</strong> are read-only global variables. On most microcontrollers, these are stored in flash and do not take up any RAM.</li>
+        <li><strong>Data</strong> are writable global variables with a non-zero initializer. On microcontrollers, they are copied from flash to RAM on reset.</li>
+        <li><strong>BSS</strong> are writable global variables that are zero initialized. They do not take up any space in the binary, but do take up RAM. On microcontrollers, this area is zeroed on reset.</li>
+      </ul>
+
+      <p>The binary size consists of code, read-only data, and data. On microcontrollers, this is exactly the size of the firmware image. On other systems, there is some extra overhead: binary metadata (headers of the ELF/MachO/COFF file), debug information, exception tables, symbol names, etc. Using <code>-no-debug</code> strips most of those.</p>
+
+      <h2>Program breakdown</h2>
+
+      <p>You can click on the rows below to see which files contribute to the binary size.</p>
+
+      <div class="table-responsive">
+        <table class="table w-auto">
+          <thead>
+            <tr>
+              <th>Package</th>
+              <th class="table-vertical-border">Code</th>
+              <th>Read-only data</th>
+              <th>Data</th>
+              <th title="zero-initialized data">BSS</th>
+              <th class="table-vertical-border" style="min-width: 16em">Binary size</th>
+            </tr>
+          </thead>
+          <tbody class="table-group-divider">
+            {{range $i, $pkg := .sizes}}
+            <tr class="row-package" data-collapse=".collapse-row-{{$i}}">
+              <td>{{.Name}}</td>
+              <td class="table-vertical-border">{{.Size.Code}}</td>
+              <td>{{.Size.ROData}}</td>
+              <td>{{.Size.Data}}</td>
+              <td>{{.Size.BSS}}</td>
+              <td class="table-vertical-border" style="background: linear-gradient(to right, var(--bs-info-bg-subtle) {{.Size.FlashPercent}}%, var(--bs-table-bg) {{.Size.FlashPercent}}%)">
+                {{.Size.Flash}}
+              </td>
+            </tr>
+            {{range $filename, $sizes := .Size.Sub}}
+            <tr class="table-secondary collapse collapse-row-{{$i}}">
+              <td class="ps-4">
+                {{if eq $filename ""}}
+                  (unknown file)
+                {{else}}
+                  {{$filename}}
+                {{end}}
+              </td>
+              <td class="table-vertical-border">{{$sizes.Code}}</td>
+              <td>{{$sizes.ROData}}</td>
+              <td>{{$sizes.Data}}</td>
+              <td>{{$sizes.BSS}}</td>
+              <td class="table-vertical-border" style="background: linear-gradient(to right, var(--bs-info-bg-subtle) {{$sizes.FlashPercent}}%, var(--bs-table-bg) {{$sizes.FlashPercent}}%)">
+                {{$sizes.Flash}}
+              </td>
+            </tr>
+            {{end}}
+            {{end}}
+          </tbody>
+          <tfoot class="table-group-divider">
+            <tr>
+              <th>Total</th>
+              <td class="table-vertical-border">{{.sizeTotal.code}}</td>
+              <td>{{.sizeTotal.rodata}}</td>
+              <td>{{.sizeTotal.data}}</td>
+              <td>{{.sizeTotal.bss}}</td>
+              <td class="table-vertical-border">{{.sizeTotal.flash}}</td>
+            </tr>
+          </tfoot>
+        </table>
+      </div>
+    </div>
+    <script>
+// Make table rows toggleable to show filenames.
+for (let clickable of document.querySelectorAll('.row-package')) {
+  clickable.addEventListener('click', e => {
+    for (let row of document.querySelectorAll(clickable.dataset.collapse)) {
+      row.classList.toggle('show');
+    }
+  });
+}
+    </script>
+  </body>
+</html>

builder/sizes.go

@@ -12,6 +12,7 @@ import (
 	"os"
 	"path/filepath"
 	"regexp"
+	"runtime"
 	"sort"
 	"strings"
 
@@ -24,7 +25,7 @@ const sizesDebug = false
 
 // programSize contains size statistics per package of a compiled program.
 type programSize struct {
-	Packages map[string]packageSize
+	Packages map[string]*packageSize
 	Code     uint64
 	ROData   uint64
 	Data     uint64
@@ -52,13 +53,29 @@ func (ps *programSize) RAM() uint64 {
 	return ps.Data + ps.BSS
 }
 
+// Return the package size information for a given package path, creating it if
+// it doesn't exist yet.
+func (ps *programSize) getPackage(path string) *packageSize {
+	if field, ok := ps.Packages[path]; ok {
+		return field
+	}
+	field := &packageSize{
+		Program: ps,
+		Sub:     map[string]*packageSize{},
+	}
+	ps.Packages[path] = field
+	return field
+}
+
 // packageSize contains the size of a package, calculated from the linked object
 // file.
 type packageSize struct {
-	Code   uint64
-	ROData uint64
-	Data   uint64
-	BSS    uint64
+	Program *programSize
+	Code    uint64
+	ROData  uint64
+	Data    uint64
+	BSS     uint64
+	Sub     map[string]*packageSize
 }
 
 // Flash usage in regular microcontrollers.
@@ -71,6 +88,31 @@ func (ps *packageSize) RAM() uint64 {
 	return ps.Data + ps.BSS
 }
 
+// Flash usage in regular microcontrollers, as a percentage of the total flash
+// usage of the program.
+func (ps *packageSize) FlashPercent() float64 {
+	return float64(ps.Flash()) / float64(ps.Program.Flash()) * 100
+}
+
+// Add a single size data point to this package.
+// This must only be called while calculating package size, not afterwards.
+func (ps *packageSize) addSize(getField func(*packageSize, bool) *uint64, filename string, size uint64, isVariable bool) {
+	if size == 0 {
+		return
+	}
+
+	// Add size for the package.
+	*getField(ps, isVariable) += size
+
+	// Add size for file inside package.
+	sub, ok := ps.Sub[filename]
+	if !ok {
+		sub = &packageSize{Program: ps.Program}
+		ps.Sub[filename] = sub
+	}
+	*getField(sub, isVariable) += size
+}
+
 // A mapping of a single chunk of code or data to a file path.
 type addressLine struct {
 	Address    uint64
@@ -194,11 +236,22 @@ func readProgramSizeFromDWARF(data *dwarf.Data, codeOffset, codeAlignment uint64
 				if !prevLineEntry.EndSequence {
 					// The chunk describes the code from prevLineEntry to
 					// lineEntry.
+					path := prevLineEntry.File.Name
+					if runtime.GOOS == "windows" {
+						// Work around a Clang bug on Windows:
+						// https://github.com/llvm/llvm-project/issues/117317
+						path = strings.ReplaceAll(path, "\\\\", "\\")
+
+						// wasi-libc likes to use forward slashes, but we
+						// canonicalize everything to use backwards slashes as
+						// is common on Windows.
+						path = strings.ReplaceAll(path, "/", "\\")
+					}
 					line := addressLine{
 						Address: prevLineEntry.Address + codeOffset,
 						Length:  lineEntry.Address - prevLineEntry.Address,
 						Align:   codeAlignment,
-						File:    prevLineEntry.File.Name,
+						File:    path,
 					}
 					if line.Length != 0 {
 						addresses = append(addresses, line)
@@ -773,49 +826,40 @@ func loadProgramSize(path string, packagePathMap map[string]string) (*programSiz
 
 	// Now finally determine the binary/RAM size usage per package by going
 	// through each allocated section.
-	sizes := make(map[string]packageSize)
+	sizes := make(map[string]*packageSize)
+	program := &programSize{
+		Packages: sizes,
+	}
 	for _, section := range sections {
 		switch section.Type {
 		case memoryCode:
-			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
-				field := sizes[path]
+			readSection(section, addresses, program, func(ps *packageSize, isVariable bool) *uint64 {
 				if isVariable {
-					field.ROData += size
-				} else {
-					field.Code += size
+					return &ps.ROData
 				}
-				sizes[path] = field
+				return &ps.Code
 			}, packagePathMap)
 		case memoryROData:
-			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
-				field := sizes[path]
-				field.ROData += size
-				sizes[path] = field
+			readSection(section, addresses, program, func(ps *packageSize, isVariable bool) *uint64 {
+				return &ps.ROData
 			}, packagePathMap)
 		case memoryData:
-			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
-				field := sizes[path]
-				field.Data += size
-				sizes[path] = field
+			readSection(section, addresses, program, func(ps *packageSize, isVariable bool) *uint64 {
+				return &ps.Data
 			}, packagePathMap)
 		case memoryBSS:
-			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
-				field := sizes[path]
-				field.BSS += size
-				sizes[path] = field
+			readSection(section, addresses, program, func(ps *packageSize, isVariable bool) *uint64 {
+				return &ps.BSS
 			}, packagePathMap)
 		case memoryStack:
 			// We store the C stack as a pseudo-package.
-			sizes["C stack"] = packageSize{
-				BSS: section.Size,
-			}
+			program.getPackage("C stack").addSize(func(ps *packageSize, isVariable bool) *uint64 {
+				return &ps.BSS
+			}, "", section.Size, false)
 		}
 	}
 
 	// ...and summarize the results.
-	program := &programSize{
-		Packages: sizes,
-	}
 	for _, pkg := range sizes {
 		program.Code += pkg.Code
 		program.ROData += pkg.ROData
@@ -826,8 +870,8 @@ func loadProgramSize(path string, packagePathMap map[string]string) (*programSiz
 }
 
 // readSection determines for each byte in this section to which package it
-// belongs. It reports this usage through the addSize callback.
-func readSection(section memorySection, addresses []addressLine, addSize func(string, uint64, bool), packagePathMap map[string]string) {
+// belongs.
+func readSection(section memorySection, addresses []addressLine, program *programSize, getField func(*packageSize, bool) *uint64, packagePathMap map[string]string) {
 	// The addr variable tracks at which address we are while going through this
 	// section. We start at the beginning.
 	addr := section.Address
@@ -849,9 +893,9 @@ func readSection(section memorySection, addresses []addressLine, addSize func(st
 			addrAligned := (addr + line.Align - 1) &^ (line.Align - 1)
 			if line.Align > 1 && addrAligned >= line.Address {
 				// It is, assume that's what causes the gap.
-				addSize("(padding)", line.Address-addr, true)
+				program.getPackage("(padding)").addSize(getField, "", line.Address-addr, true)
 			} else {
-				addSize("(unknown)", line.Address-addr, false)
+				program.getPackage("(unknown)").addSize(getField, "", line.Address-addr, false)
 				if sizesDebug {
 					fmt.Printf("%08x..%08x %5d:  unknown (gap), alignment=%d\n", addr, line.Address, line.Address-addr, line.Align)
 				}
@@ -873,7 +917,8 @@ func readSection(section memorySection, addresses []addressLine, addSize func(st
 			length = line.Length - (addr - line.Address)
 		}
 		// Finally, mark this chunk of memory as used by the given package.
-		addSize(findPackagePath(line.File, packagePathMap), length, line.IsVariable)
+		packagePath, filename := findPackagePath(line.File, packagePathMap)
+		program.getPackage(packagePath).addSize(getField, filename, length, line.IsVariable)
 		addr = line.Address + line.Length
 	}
 	if addr < sectionEnd {
@@ -882,9 +927,9 @@ func readSection(section memorySection, addresses []addressLine, addSize func(st
 		if section.Align > 1 && addrAligned >= sectionEnd {
 			// The gap is caused by the section alignment.
 			// For example, if a .rodata section ends with a non-aligned string.
-			addSize("(padding)", sectionEnd-addr, true)
+			program.getPackage("(padding)").addSize(getField, "", sectionEnd-addr, true)
 		} else {
-			addSize("(unknown)", sectionEnd-addr, false)
+			program.getPackage("(unknown)").addSize(getField, "", sectionEnd-addr, false)
 			if sizesDebug {
 				fmt.Printf("%08x..%08x %5d:  unknown (end), alignment=%d\n", addr, sectionEnd, sectionEnd-addr, section.Align)
 			}
@@ -894,17 +939,25 @@ func readSection(section memorySection, addresses []addressLine, addSize func(st
 
 // findPackagePath returns the Go package (or a pseudo package) for the given
 // path. It uses some heuristics, for example for some C libraries.
-func findPackagePath(path string, packagePathMap map[string]string) string {
+func findPackagePath(path string, packagePathMap map[string]string) (packagePath, filename string) {
 	// Check whether this path is part of one of the compiled packages.
 	packagePath, ok := packagePathMap[filepath.Dir(path)]
-	if !ok {
+	if ok {
+		// Directory is known as a Go package.
+		// Add the file itself as well.
+		filename = filepath.Base(path)
+	} else {
 		if strings.HasPrefix(path, filepath.Join(goenv.Get("TINYGOROOT"), "lib")) {
 			// Emit C libraries (in the lib subdirectory of TinyGo) as a single
-			// package, with a "C" prefix. For example: "C compiler-rt" for the
-			// compiler runtime library from LLVM.
-			packagePath = "C " + strings.Split(strings.TrimPrefix(path, filepath.Join(goenv.Get("TINYGOROOT"), "lib")), string(os.PathSeparator))[1]
-		} else if strings.HasPrefix(path, filepath.Join(goenv.Get("TINYGOROOT"), "llvm-project")) {
+			// package, with a "C" prefix. For example: "C picolibc" for the
+			// baremetal libc.
+			libPath := strings.TrimPrefix(path, filepath.Join(goenv.Get("TINYGOROOT"), "lib")+string(os.PathSeparator))
+			parts := strings.SplitN(libPath, string(os.PathSeparator), 2)
+			packagePath = "C " + parts[0]
+			filename = parts[1]
+		} else if prefix := filepath.Join(goenv.Get("TINYGOROOT"), "llvm-project", "compiler-rt"); strings.HasPrefix(path, prefix) {
 			packagePath = "C compiler-rt"
+			filename = strings.TrimPrefix(path, prefix+string(os.PathSeparator))
 		} else if packageSymbolRegexp.MatchString(path) {
 			// Parse symbol names like main$alloc or runtime$string.
 			packagePath = path[:strings.LastIndex(path, "$")]
@@ -927,9 +980,11 @@ func findPackagePath(path string, packagePathMap map[string]string) string {
 			// fixed in the compiler.
 			packagePath = "-"
 		} else {
-			// This is some other path. Not sure what it is, so just emit its directory.
-			packagePath = filepath.Dir(path) // fallback
+			// This is some other path. Not sure what it is, so just emit its
+			// directory as a fallback.
+			packagePath = filepath.Dir(path)
+			filename = filepath.Base(path)
 		}
 	}
-	return packagePath
+	return
 }

builder/sizes_test.go

@@ -1,6 +1,7 @@
 package builder
 
 import (
+	"regexp"
 	"runtime"
 	"testing"
 	"time"
@@ -41,9 +42,9 @@ func TestBinarySize(t *testing.T) {
 	// This is a small number of very diverse targets that we want to test.
 	tests := []sizeTest{
 		// microcontrollers
-		{"hifive1b", "examples/echo", 4612, 280, 0, 2252},
-		{"microbit", "examples/serial", 2724, 388, 8, 2256},
-		{"wioterminal", "examples/pininterrupt", 6159, 1485, 116, 6816},
+		{"hifive1b", "examples/echo", 4560, 280, 0, 2268},
+		{"microbit", "examples/serial", 2916, 388, 8, 2272},
+		{"wioterminal", "examples/pininterrupt", 7359, 1489, 116, 6912},
 
 		// TODO: also check wasm. Right now this is difficult, because
 		// wasm binaries are run through wasm-opt and therefore the
@@ -55,26 +56,7 @@ func TestBinarySize(t *testing.T) {
 			t.Parallel()
 
 			// Build the binary.
-			options := compileopts.Options{
-				Target:        tc.target,
-				Opt:           "z",
-				Semaphore:     sema,
-				InterpTimeout: 60 * time.Second,
-				Debug:         true,
-				VerifyIR:      true,
-			}
-			target, err := compileopts.LoadTarget(&options)
-			if err != nil {
-				t.Fatal("could not load target:", err)
-			}
-			config := &compileopts.Config{
-				Options: &options,
-				Target:  target,
-			}
-			result, err := Build(tc.path, "", t.TempDir(), config)
-			if err != nil {
-				t.Fatal("could not build:", err)
-			}
+			result := buildBinary(t, tc.target, tc.path)
 
 			// Check whether the size of the binary matches the expected size.
 			sizes, err := loadProgramSize(result.Executable, nil)
@@ -90,3 +72,69 @@ func TestBinarySize(t *testing.T) {
 		})
 	}
 }
+
+// Check that the -size=full flag attributes binary size to the correct package
+// without filesystem paths and things like that.
+func TestSizeFull(t *testing.T) {
+	tests := []string{
+		"microbit",
+		"wasip1",
+	}
+
+	libMatch := regexp.MustCompile(`^C [a-z -]+$`) // example: "C interrupt vector"
+	pkgMatch := regexp.MustCompile(`^[a-z/]+$`)    // example: "internal/task"
+
+	for _, target := range tests {
+		target := target
+		t.Run(target, func(t *testing.T) {
+			t.Parallel()
+
+			// Build the binary.
+			result := buildBinary(t, target, "examples/serial")
+
+			// Check whether the binary doesn't contain any unexpected package
+			// names.
+			sizes, err := loadProgramSize(result.Executable, result.PackagePathMap)
+			if err != nil {
+				t.Fatal("could not read program size:", err)
+			}
+			for _, pkg := range sizes.sortedPackageNames() {
+				if pkg == "(padding)" || pkg == "(unknown)" {
+					// TODO: correctly attribute all unknown binary size.
+					continue
+				}
+				if libMatch.MatchString(pkg) {
+					continue
+				}
+				if pkgMatch.MatchString(pkg) {
+					continue
+				}
+				t.Error("unexpected package name in size output:", pkg)
+			}
+		})
+	}
+}
+
+func buildBinary(t *testing.T, targetString, pkgName string) BuildResult {
+	options := compileopts.Options{
+		Target:        targetString,
+		Opt:           "z",
+		Semaphore:     sema,
+		InterpTimeout: 60 * time.Second,
+		Debug:         true,
+		VerifyIR:      true,
+	}
+	target, err := compileopts.LoadTarget(&options)
+	if err != nil {
+		t.Fatal("could not load target:", err)
+	}
+	config := &compileopts.Config{
+		Options: &options,
+		Target:  target,
+	}
+	result, err := Build(pkgName, "", t.TempDir(), config)
+	if err != nil {
+		t.Fatal("could not build:", err)
+	}
+	return result
+}

builder/tools-builtin.go

@@ -12,10 +12,7 @@ import (
 #include <stdbool.h>
 #include <stdlib.h>
 bool tinygo_clang_driver(int argc, char **argv);
-bool tinygo_link_elf(int argc, char **argv);
-bool tinygo_link_macho(int argc, char **argv);
-bool tinygo_link_mingw(int argc, char **argv);
-bool tinygo_link_wasm(int argc, char **argv);
+bool tinygo_link(int argc, char **argv);
 */
 import "C"
 
@@ -26,16 +23,7 @@ const hasBuiltinTools = true
 // This version actually runs the tools because TinyGo was compiled while
 // linking statically with LLVM (with the byollvm build tag).
 func RunTool(tool string, args ...string) error {
-	linker := "elf"
-	if tool == "ld.lld" && len(args) >= 2 {
-		if args[0] == "-m" && (args[1] == "i386pep" || args[1] == "arm64pe") {
-			linker = "mingw"
-		} else if args[0] == "-flavor" {
-			linker = args[1]
-			args = args[2:]
-		}
-	}
-	args = append([]string{"tinygo:" + tool}, args...)
+	args = append([]string{tool}, args...)
 
 	var cflag *C.char
 	buf := C.calloc(C.size_t(len(args)), C.size_t(unsafe.Sizeof(cflag)))
@@ -51,19 +39,8 @@ func RunTool(tool string, args ...string) error {
 	switch tool {
 	case "clang":
 		ok = C.tinygo_clang_driver(C.int(len(args)), (**C.char)(buf))
-	case "ld.lld":
-		switch linker {
-		case "darwin":
-			ok = C.tinygo_link_macho(C.int(len(args)), (**C.char)(buf))
-		case "elf":
-			ok = C.tinygo_link_elf(C.int(len(args)), (**C.char)(buf))
-		case "mingw":
-			ok = C.tinygo_link_mingw(C.int(len(args)), (**C.char)(buf))
-		default:
-			return errors.New("unknown linker: " + linker)
-		}
-	case "wasm-ld":
-		ok = C.tinygo_link_wasm(C.int(len(args)), (**C.char)(buf))
+	case "ld.lld", "wasm-ld":
+		ok = C.tinygo_link(C.int(len(args)), (**C.char)(buf))
 	default:
 		return errors.New("unknown tool: " + tool)
 	}

builder/tools.go

@@ -1,50 +1,191 @@
 package builder
 
 import (
-	"errors"
+	"bytes"
+	"fmt"
+	"go/scanner"
+	"go/token"
 	"os"
 	"os/exec"
-
-	"github.com/tinygo-org/tinygo/goenv"
+	"regexp"
+	"strconv"
+	"strings"
 )
 
 // runCCompiler invokes a C compiler with the given arguments.
 func runCCompiler(flags ...string) error {
+	// Find the right command to run Clang.
+	var cmd *exec.Cmd
 	if hasBuiltinTools {
 		// Compile this with the internal Clang compiler.
-		headerPath := getClangHeaderPath(goenv.Get("TINYGOROOT"))
-		if headerPath == "" {
-			return errors.New("could not locate Clang headers")
+		cmd = exec.Command(os.Args[0], append([]string{"clang"}, flags...)...)
+	} else {
+		// Compile this with an external invocation of the Clang compiler.
+		name, err := LookupCommand("clang")
+		if err != nil {
+			return err
 		}
-		flags = append(flags, "-I"+headerPath)
-		cmd := exec.Command(os.Args[0], append([]string{"clang"}, flags...)...)
-		cmd.Stdout = os.Stdout
-		cmd.Stderr = os.Stderr
-		return cmd.Run()
+		cmd = exec.Command(name, flags...)
 	}
 
-	// Compile this with an external invocation of the Clang compiler.
-	return execCommand("clang", flags...)
+	cmd.Stdout = os.Stdout
+	cmd.Stderr = os.Stderr
+
+	// Make sure the command doesn't use any environmental variables.
+	// Most importantly, it should not use C_INCLUDE_PATH and the like.
+	cmd.Env = []string{}
+
+	// Let some environment variables through. One important one is the
+	// temporary directory, especially on Windows it looks like Clang breaks if
+	// the temporary directory has not been set.
+	// See: https://github.com/tinygo-org/tinygo/issues/4557
+	// Also see: https://github.com/llvm/llvm-project/blob/release/18.x/llvm/lib/Support/Unix/Path.inc#L1435
+	for _, env := range os.Environ() {
+		// We could parse the key and look it up in a map, but since there are
+		// only a few keys iterating through them is easier and maybe even
+		// faster.
+		for _, prefix := range []string{"TMPDIR=", "TMP=", "TEMP=", "TEMPDIR="} {
+			if strings.HasPrefix(env, prefix) {
+				cmd.Env = append(cmd.Env, env)
+				break
+			}
+		}
+	}
+
+	return cmd.Run()
 }
 
 // link invokes a linker with the given name and flags.
 func link(linker string, flags ...string) error {
-	if hasBuiltinTools && (linker == "ld.lld" || linker == "wasm-ld") {
-		// Run command with internal linker.
-		cmd := exec.Command(os.Args[0], append([]string{linker}, flags...)...)
-		cmd.Stdout = os.Stdout
-		cmd.Stderr = os.Stderr
-		return cmd.Run()
+	// We only support LLD.
+	if linker != "ld.lld" && linker != "wasm-ld" {
+		return fmt.Errorf("unexpected: linker %s should be ld.lld or wasm-ld", linker)
 	}
 
-	// Fall back to external command.
-	if _, ok := commands[linker]; ok {
-		return execCommand(linker, flags...)
+	var cmd *exec.Cmd
+	if hasBuiltinTools {
+		cmd = exec.Command(os.Args[0], append([]string{linker}, flags...)...)
+	} else {
+		name, err := LookupCommand(linker)
+		if err != nil {
+			return err
+		}
+		cmd = exec.Command(name, flags...)
 	}
-
-	cmd := exec.Command(linker, flags...)
+	var buf bytes.Buffer
 	cmd.Stdout = os.Stdout
-	cmd.Stderr = os.Stderr
-	cmd.Dir = goenv.Get("TINYGOROOT")
-	return cmd.Run()
+	cmd.Stderr = &buf
+	err := cmd.Run()
+	if err != nil {
+		if buf.Len() == 0 {
+			// The linker failed but there was no output.
+			// Therefore, show some output anyway.
+			return fmt.Errorf("failed to run linker: %w", err)
+		}
+		return parseLLDErrors(buf.String())
+	}
+	return nil
+}
+
+// Split LLD errors into individual erros (including errors that continue on the
+// next line, using a ">>>" prefix). If possible, replace the raw errors with a
+// more user-friendly version (and one that's more in a Go style).
+func parseLLDErrors(text string) error {
+	// Split linker output in separate error messages.
+	lines := strings.Split(text, "\n")
+	var errorLines []string // one or more line (belonging to a single error) per line
+	for _, line := range lines {
+		line = strings.TrimRight(line, "\r") // needed for Windows
+		if len(errorLines) != 0 && strings.HasPrefix(line, ">>> ") {
+			errorLines[len(errorLines)-1] += "\n" + line
+			continue
+		}
+		if line == "" {
+			continue
+		}
+		errorLines = append(errorLines, line)
+	}
+
+	// Parse error messages.
+	var linkErrors []error
+	var flashOverflow, ramOverflow uint64
+	for _, message := range errorLines {
+		parsedError := false
+
+		// Check for undefined symbols.
+		// This can happen in some cases like with CGo and //go:linkname tricker.
+		if matches := regexp.MustCompile(`^ld.lld(-[0-9]+)?: error: undefined symbol: (.*)\n`).FindStringSubmatch(message); matches != nil {
+			symbolName := matches[2]
+			for _, line := range strings.Split(message, "\n") {
+				matches := regexp.MustCompile(`referenced by .* \(((.*):([0-9]+))\)`).FindStringSubmatch(line)
+				if matches != nil {
+					parsedError = true
+					line, _ := strconv.Atoi(matches[3])
+					// TODO: detect common mistakes like -gc=none?
+					linkErrors = append(linkErrors, scanner.Error{
+						Pos: token.Position{
+							Filename: matches[2],
+							Line:     line,
+						},
+						Msg: "linker could not find symbol " + symbolName,
+					})
+				}
+			}
+		}
+
+		// Check for flash/RAM overflow.
+		if matches := regexp.MustCompile(`^ld.lld(-[0-9]+)?: error: section '(.*?)' will not fit in region '(.*?)': overflowed by ([0-9]+) bytes$`).FindStringSubmatch(message); matches != nil {
+			region := matches[3]
+			n, err := strconv.ParseUint(matches[4], 10, 64)
+			if err != nil {
+				// Should not happen at all (unless it overflows an uint64 for some reason).
+				continue
+			}
+
+			// Check which area overflowed.
+			// Some chips use differently named memory areas, but these are by
+			// far the most common.
+			switch region {
+			case "FLASH_TEXT":
+				if n > flashOverflow {
+					flashOverflow = n
+				}
+				parsedError = true
+			case "RAM":
+				if n > ramOverflow {
+					ramOverflow = n
+				}
+				parsedError = true
+			}
+		}
+
+		// If we couldn't parse the linker error: show the error as-is to
+		// the user.
+		if !parsedError {
+			linkErrors = append(linkErrors, LinkerError{message})
+		}
+	}
+
+	if flashOverflow > 0 {
+		linkErrors = append(linkErrors, LinkerError{
+			Msg: fmt.Sprintf("program too large for this chip (flash overflowed by %d bytes)\n\toptimization guide: https://tinygo.org/docs/guides/optimizing-binaries/", flashOverflow),
+		})
+	}
+	if ramOverflow > 0 {
+		linkErrors = append(linkErrors, LinkerError{
+			Msg: fmt.Sprintf("program uses too much static RAM on this chip (RAM overflowed by %d bytes)", ramOverflow),
+		})
+	}
+
+	return newMultiError(linkErrors, "")
+}
+
+// LLD linker error that could not be parsed or doesn't refer to a source
+// location.
+type LinkerError struct {
+	Msg string
+}
+
+func (e LinkerError) Error() string {
+	return e.Msg
 }

builder/wasmbuiltins.go

@@ -0,0 +1,81 @@
+package builder
+
+import (
+	"os"
+	"path/filepath"
+
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+var libWasmBuiltins = Library{
+	name: "wasmbuiltins",
+	makeHeaders: func(target, includeDir string) error {
+		if err := os.Mkdir(includeDir+"/bits", 0o777); err != nil {
+			return err
+		}
+		f, err := os.Create(includeDir + "/bits/alltypes.h")
+		if err != nil {
+			return err
+		}
+		if _, err := f.Write([]byte(wasmAllTypes)); err != nil {
+			return err
+		}
+		return f.Close()
+	},
+	cflags: func(target, headerPath string) []string {
+		libcDir := filepath.Join(goenv.Get("TINYGOROOT"), "lib/wasi-libc")
+		return []string{
+			"-Werror",
+			"-Wall",
+			"-std=gnu11",
+			"-nostdlibinc",
+			"-isystem", libcDir + "/libc-top-half/musl/arch/wasm32",
+			"-isystem", libcDir + "/libc-top-half/musl/arch/generic",
+			"-isystem", libcDir + "/libc-top-half/musl/src/internal",
+			"-isystem", libcDir + "/libc-top-half/musl/src/include",
+			"-isystem", libcDir + "/libc-top-half/musl/include",
+			"-isystem", libcDir + "/libc-bottom-half/headers/public",
+			"-I" + headerPath,
+		}
+	},
+	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/wasi-libc") },
+	librarySources: func(target string) ([]string, error) {
+		return []string{
+			// memory builtins needed for llvm.memcpy.*, llvm.memmove.*, and
+			// llvm.memset.* LLVM intrinsics.
+			"libc-top-half/musl/src/string/memcpy.c",
+			"libc-top-half/musl/src/string/memmove.c",
+			"libc-top-half/musl/src/string/memset.c",
+
+			// exp, exp2, and log are needed for LLVM math builtin functions
+			// like llvm.exp.*.
+			"libc-top-half/musl/src/math/__math_divzero.c",
+			"libc-top-half/musl/src/math/__math_invalid.c",
+			"libc-top-half/musl/src/math/__math_oflow.c",
+			"libc-top-half/musl/src/math/__math_uflow.c",
+			"libc-top-half/musl/src/math/__math_xflow.c",
+			"libc-top-half/musl/src/math/exp.c",
+			"libc-top-half/musl/src/math/exp_data.c",
+			"libc-top-half/musl/src/math/exp2.c",
+			"libc-top-half/musl/src/math/log.c",
+			"libc-top-half/musl/src/math/log_data.c",
+		}, nil
+	},
+}
+
+// alltypes.h for wasm-libc, using the types as defined inside Clang.
+const wasmAllTypes = `
+typedef __SIZE_TYPE__    size_t;
+typedef __INT8_TYPE__    int8_t;
+typedef __INT16_TYPE__   int16_t;
+typedef __INT32_TYPE__   int32_t;
+typedef __INT64_TYPE__   int64_t;
+typedef __UINT8_TYPE__   uint8_t;
+typedef __UINT16_TYPE__  uint16_t;
+typedef __UINT32_TYPE__  uint32_t;
+typedef __UINT64_TYPE__  uint64_t;
+typedef __UINTPTR_TYPE__ uintptr_t;
+
+// This type is used internally in wasi-libc.
+typedef double double_t;
+`

cgo/cgo.go

@@ -18,6 +18,7 @@ import (
 	"go/scanner"
 	"go/token"
 	"path/filepath"
+	"sort"
 	"strconv"
 	"strings"
 
@@ -25,9 +26,15 @@ import (
 	"golang.org/x/tools/go/ast/astutil"
 )
 
+// Function that's only defined in Go 1.22.
+var setASTFileFields = func(f *ast.File, start, end token.Pos) {
+}
+
 // cgoPackage holds all CGo-related information of a package.
 type cgoPackage struct {
 	generated       *ast.File
+	packageName     string
+	cgoFiles        []*ast.File
 	generatedPos    token.Pos
 	errors          []error
 	currentDir      string // current working directory
@@ -36,6 +43,7 @@ type cgoPackage struct {
 	fset            *token.FileSet
 	tokenFiles      map[string]*token.File
 	definedGlobally map[string]ast.Node
+	noescapingFuncs map[string]*noescapingFunc // #cgo noescape lines
 	anonDecls       map[interface{}]string
 	cflags          []string // CFlags from #cgo lines
 	ldflags         []string // LDFlags from #cgo lines
@@ -74,20 +82,28 @@ type bitfieldInfo struct {
 	endBit   int64 // may be 0 meaning "until the end of the field"
 }
 
+// Information about a #cgo noescape line in the source code.
+type noescapingFunc struct {
+	name string
+	pos  token.Pos
+	used bool // true if used somewhere in the source (for proper error reporting)
+}
+
 // cgoAliases list type aliases between Go and C, for types that are equivalent
 // in both languages. See addTypeAliases.
 var cgoAliases = map[string]string{
-	"C.int8_t":    "int8",
-	"C.int16_t":   "int16",
-	"C.int32_t":   "int32",
-	"C.int64_t":   "int64",
-	"C.uint8_t":   "uint8",
-	"C.uint16_t":  "uint16",
-	"C.uint32_t":  "uint32",
-	"C.uint64_t":  "uint64",
-	"C.uintptr_t": "uintptr",
-	"C.float":     "float32",
-	"C.double":    "float64",
+	"_Cgo_int8_t":    "int8",
+	"_Cgo_int16_t":   "int16",
+	"_Cgo_int32_t":   "int32",
+	"_Cgo_int64_t":   "int64",
+	"_Cgo_uint8_t":   "uint8",
+	"_Cgo_uint16_t":  "uint16",
+	"_Cgo_uint32_t":  "uint32",
+	"_Cgo_uint64_t":  "uint64",
+	"_Cgo_uintptr_t": "uintptr",
+	"_Cgo_float":     "float32",
+	"_Cgo_double":    "float64",
+	"_Cgo__Bool":     "bool",
 }
 
 // builtinAliases are handled specially because they only exist on the Go side
@@ -129,31 +145,105 @@ typedef unsigned long long  _Cgo_ulonglong;
 // The string/bytes functions below implement C.CString etc. To make sure the
 // runtime doesn't need to know the C int type, lengths are converted to uintptr
 // first.
-// These functions will be modified to get a "C." prefix, so the source below
-// doesn't reflect the final AST.
-const generatedGoFilePrefix = `
+const generatedGoFilePrefixBase = `
+import "syscall"
 import "unsafe"
 
 var _ unsafe.Pointer
 
-//go:linkname C.CString runtime.cgo_CString
-func CString(string) *C.char
+//go:linkname _Cgo_CString runtime.cgo_CString
+func _Cgo_CString(string) *_Cgo_char
 
-//go:linkname C.GoString runtime.cgo_GoString
-func GoString(*C.char) string
+//go:linkname _Cgo_GoString runtime.cgo_GoString
+func _Cgo_GoString(*_Cgo_char) string
 
-//go:linkname C.__GoStringN runtime.cgo_GoStringN
-func __GoStringN(*C.char, uintptr) string
+//go:linkname _Cgo___GoStringN runtime.cgo_GoStringN
+func _Cgo___GoStringN(*_Cgo_char, uintptr) string
 
-func GoStringN(cstr *C.char, length C.int) string {
-	return C.__GoStringN(cstr, uintptr(length))
+func _Cgo_GoStringN(cstr *_Cgo_char, length _Cgo_int) string {
+	return _Cgo___GoStringN(cstr, uintptr(length))
 }
 
-//go:linkname C.__GoBytes runtime.cgo_GoBytes
-func __GoBytes(unsafe.Pointer, uintptr) []byte
+//go:linkname _Cgo___GoBytes runtime.cgo_GoBytes
+func _Cgo___GoBytes(unsafe.Pointer, uintptr) []byte
 
-func GoBytes(ptr unsafe.Pointer, length C.int) []byte {
-	return C.__GoBytes(ptr, uintptr(length))
+func _Cgo_GoBytes(ptr unsafe.Pointer, length _Cgo_int) []byte {
+	return _Cgo___GoBytes(ptr, uintptr(length))
+}
+
+//go:linkname _Cgo___CBytes runtime.cgo_CBytes
+func _Cgo___CBytes([]byte) unsafe.Pointer
+
+func _Cgo_CBytes(b []byte) unsafe.Pointer {
+	return _Cgo___CBytes(b)
+}
+
+//go:linkname _Cgo___get_errno_num runtime.cgo_errno
+func _Cgo___get_errno_num() uintptr
+`
+
+const generatedGoFilePrefixOther = generatedGoFilePrefixBase + `
+func _Cgo___get_errno() error {
+	return syscall.Errno(_Cgo___get_errno_num())
+}
+`
+
+// Windows uses fake errno values in the syscall package.
+// See for example: https://github.com/golang/go/issues/23468
+// TinyGo uses mingw-w64 though, which does have defined errno values. Since the
+// syscall package is the standard library one we can't change it, but we can
+// map the errno values to match the values in the syscall package.
+// Source of the errno values: lib/mingw-w64/mingw-w64-headers/crt/errno.h
+const generatedGoFilePrefixWindows = generatedGoFilePrefixBase + `
+var _Cgo___errno_mapping = [...]syscall.Errno{
+	1:  syscall.EPERM,
+	2:  syscall.ENOENT,
+	3:  syscall.ESRCH,
+	4:  syscall.EINTR,
+	5:  syscall.EIO,
+	6:  syscall.ENXIO,
+	7:  syscall.E2BIG,
+	8:  syscall.ENOEXEC,
+	9:  syscall.EBADF,
+	10: syscall.ECHILD,
+	11: syscall.EAGAIN,
+	12: syscall.ENOMEM,
+	13: syscall.EACCES,
+	14: syscall.EFAULT,
+	16: syscall.EBUSY,
+	17: syscall.EEXIST,
+	18: syscall.EXDEV,
+	19: syscall.ENODEV,
+	20: syscall.ENOTDIR,
+	21: syscall.EISDIR,
+	22: syscall.EINVAL,
+	23: syscall.ENFILE,
+	24: syscall.EMFILE,
+	25: syscall.ENOTTY,
+	27: syscall.EFBIG,
+	28: syscall.ENOSPC,
+	29: syscall.ESPIPE,
+	30: syscall.EROFS,
+	31: syscall.EMLINK,
+	32: syscall.EPIPE,
+	33: syscall.EDOM,
+	34: syscall.ERANGE,
+	36: syscall.EDEADLK,
+	38: syscall.ENAMETOOLONG,
+	39: syscall.ENOLCK,
+	40: syscall.ENOSYS,
+	41: syscall.ENOTEMPTY,
+	42: syscall.EILSEQ,
+}
+
+func _Cgo___get_errno() error {
+	num := _Cgo___get_errno_num()
+	if num < uintptr(len(_Cgo___errno_mapping)) {
+		if mapped := _Cgo___errno_mapping[num]; mapped != 0 {
+			return mapped
+		}
+	}
+	return syscall.Errno(num)
 }
 `
 
@@ -164,13 +254,15 @@ func GoBytes(ptr unsafe.Pointer, length C.int) []byte {
 // functions), the CFLAGS and LDFLAGS found in #cgo lines, and a map of file
 // hashes of the accessed C header files. If there is one or more error, it
 // returns these in the []error slice but still modifies the AST.
-func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cflags []string, clangHeaders string) (*ast.File, []string, []string, []string, map[string][]byte, []error) {
+func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cflags []string, goos string) ([]*ast.File, []string, []string, []string, map[string][]byte, []error) {
 	p := &cgoPackage{
+		packageName:     files[0].Name.Name,
 		currentDir:      dir,
 		importPath:      importPath,
 		fset:            fset,
 		tokenFiles:      map[string]*token.File{},
 		definedGlobally: map[string]ast.Node{},
+		noescapingFuncs: map[string]*noescapingFunc{},
 		anonDecls:       map[interface{}]string{},
 		visitedFiles:    map[string][]byte{},
 	}
@@ -195,32 +287,21 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 	// Construct a new in-memory AST for CGo declarations of this package.
 	// The first part is written as Go code that is then parsed, but more code
 	// is added later to the AST to declare functions, globals, etc.
-	goCode := "package " + files[0].Name.Name + "\n\n" + generatedGoFilePrefix
+	goCode := "package " + files[0].Name.Name + "\n\n"
+	if goos == "windows" {
+		goCode += generatedGoFilePrefixWindows
+	} else {
+		goCode += generatedGoFilePrefixOther
+	}
 	p.generated, err = parser.ParseFile(fset, dir+"/!cgo.go", goCode, parser.ParseComments)
 	if err != nil {
 		// This is always a bug in the cgo package.
 		panic("unexpected error: " + err.Error())
 	}
+	p.cgoFiles = append(p.cgoFiles, p.generated)
 	// If the Comments field is not set to nil, the go/format package will get
 	// confused about where comments should go.
 	p.generated.Comments = nil
-	// Adjust some of the functions in there.
-	for _, decl := range p.generated.Decls {
-		switch decl := decl.(type) {
-		case *ast.FuncDecl:
-			switch decl.Name.Name {
-			case "CString", "GoString", "GoStringN", "__GoStringN", "GoBytes", "__GoBytes":
-				// Adjust the name to have a "C." prefix so it is correctly
-				// resolved.
-				decl.Name.Name = "C." + decl.Name.Name
-			}
-		}
-	}
-	// Patch some types, for example *C.char in C.CString.
-	cf := p.newCGoFile(nil, -1) // dummy *cgoFile for the walker
-	astutil.Apply(p.generated, func(cursor *astutil.Cursor) bool {
-		return cf.walker(cursor, nil)
-	}, nil)
 
 	// Find `import "C"` C fragments in the file.
 	p.cgoHeaders = make([]string, len(files)) // combined CGo header fragment for each file
@@ -291,9 +372,6 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 	// have better alternatives anyway.
 	cflagsForCGo := append([]string{"-D_FORTIFY_SOURCE=0"}, cflags...)
 	cflagsForCGo = append(cflagsForCGo, p.cflags...)
-	if clangHeaders != "" {
-		cflagsForCGo = append(cflagsForCGo, "-isystem", clangHeaders)
-	}
 
 	// Retrieve types such as C.int, C.longlong, etc from C.
 	p.newCGoFile(nil, -1).readNames(builtinAliasTypedefs, cflagsForCGo, "", func(names map[string]clangCursor) {
@@ -302,7 +380,7 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 			Tok:    token.TYPE,
 		}
 		for _, name := range builtinAliases {
-			typeSpec := p.getIntegerType("C."+name, names["_Cgo_"+name])
+			typeSpec := p.getIntegerType("_Cgo_"+name, names["_Cgo_"+name])
 			gen.Specs = append(gen.Specs, typeSpec)
 		}
 		p.generated.Decls = append(p.generated.Decls, gen)
@@ -311,10 +389,11 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 	// Process CGo imports for each file.
 	for i, f := range files {
 		cf := p.newCGoFile(f, i)
-		// Float and double are aliased, meaning that C.float is the same thing
-		// as float32 in Go.
+		// These types are aliased with the corresponding types in C. For
+		// example, float in C is always float32 in Go.
 		cf.names["float"] = clangCursor{}
 		cf.names["double"] = clangCursor{}
+		cf.names["_Bool"] = clangCursor{}
 		// Now read all the names (identifies) that C defines in the header
 		// snippet.
 		cf.readNames(p.cgoHeaders[i], cflagsForCGo, filepath.Base(fset.File(f.Pos()).Name()), func(names map[string]clangCursor) {
@@ -330,10 +409,26 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 		})
 	}
 
+	// Show an error when a #cgo noescape line isn't used in practice.
+	// This matches upstream Go. I think the goal is to avoid issues with
+	// misspelled function names, which seems very useful.
+	var unusedNoescapeLines []*noescapingFunc
+	for _, value := range p.noescapingFuncs {
+		if !value.used {
+			unusedNoescapeLines = append(unusedNoescapeLines, value)
+		}
+	}
+	sort.SliceStable(unusedNoescapeLines, func(i, j int) bool {
+		return unusedNoescapeLines[i].pos < unusedNoescapeLines[j].pos
+	})
+	for _, value := range unusedNoescapeLines {
+		p.addError(value.pos, fmt.Sprintf("function %#v in #cgo noescape line is not used", value.name))
+	}
+
 	// Print the newly generated in-memory AST, for debugging.
 	//ast.Print(fset, p.generated)
 
-	return p.generated, p.cgoHeaders, p.cflags, p.ldflags, p.visitedFiles, p.errors
+	return p.cgoFiles, p.cgoHeaders, p.cflags, p.ldflags, p.visitedFiles, p.errors
 }
 
 func (p *cgoPackage) newCGoFile(file *ast.File, index int) *cgoFile {
@@ -395,6 +490,33 @@ func (p *cgoPackage) parseCGoPreprocessorLines(text string, pos token.Pos) strin
 		}
 		text = text[:lineStart] + string(spaces) + text[lineEnd:]
 
+		allFields := strings.Fields(line[4:])
+		switch allFields[0] {
+		case "noescape":
+			// The code indicates that pointer parameters will not be captured
+			// by the called C function.
+			if len(allFields) < 2 {
+				p.addErrorAfter(pos, text[:lineStart], "missing function name in #cgo noescape line")
+				continue
+			}
+			if len(allFields) > 2 {
+				p.addErrorAfter(pos, text[:lineStart], "multiple function names in #cgo noescape line")
+				continue
+			}
+			name := allFields[1]
+			p.noescapingFuncs[name] = &noescapingFunc{
+				name: name,
+				pos:  pos,
+				used: false,
+			}
+			continue
+		case "nocallback":
+			// We don't do anything special when calling a C function, so there
+			// appears to be no optimization that we can do here.
+			// Accept, but ignore the parameter for compatibility.
+			continue
+		}
+
 		// Get the text before the colon in the #cgo directive.
 		colon := strings.IndexByte(line, ':')
 		if colon < 0 {
@@ -1131,22 +1253,22 @@ func (p *cgoPackage) getUnnamedDeclName(prefix string, itf interface{}) string {
 func (f *cgoFile) getASTDeclName(name string, found clangCursor, iscall bool) string {
 	// Some types are defined in stdint.h and map directly to a particular Go
 	// type.
-	if alias := cgoAliases["C."+name]; alias != "" {
+	if alias := cgoAliases["_Cgo_"+name]; alias != "" {
 		return alias
 	}
-	node := f.getASTDeclNode(name, found, iscall)
+	node := f.getASTDeclNode(name, found)
 	if node, ok := node.(*ast.FuncDecl); ok {
 		if !iscall {
 			return node.Name.Name + "$funcaddr"
 		}
 		return node.Name.Name
 	}
-	return "C." + name
+	return "_Cgo_" + name
 }
 
 // getASTDeclNode will declare the given C AST node (if not already defined) and
 // returns it.
-func (f *cgoFile) getASTDeclNode(name string, found clangCursor, iscall bool) ast.Node {
+func (f *cgoFile) getASTDeclNode(name string, found clangCursor) ast.Node {
 	if node, ok := f.defined[name]; ok {
 		// Declaration was found in the current file, so return it immediately.
 		return node
@@ -1241,8 +1363,8 @@ extern __typeof(%s) %s __attribute__((alias(%#v)));
 	case *elaboratedTypeInfo:
 		// Add struct bitfields.
 		for _, bitfield := range elaboratedType.bitfields {
-			f.createBitfieldGetter(bitfield, "C."+name)
-			f.createBitfieldSetter(bitfield, "C."+name)
+			f.createBitfieldGetter(bitfield, "_Cgo_"+name)
+			f.createBitfieldSetter(bitfield, "_Cgo_"+name)
 		}
 		if elaboratedType.unionSize != 0 {
 			// Create union getters/setters.
@@ -1251,7 +1373,7 @@ extern __typeof(%s) %s __attribute__((alias(%#v)));
 					f.addError(elaboratedType.pos, fmt.Sprintf("union must have field with a single name, it has %d names", len(field.Names)))
 					continue
 				}
-				f.createUnionAccessor(field, "C."+name)
+				f.createUnionAccessor(field, "_Cgo_"+name)
 			}
 		}
 	}
@@ -1265,6 +1387,45 @@ extern __typeof(%s) %s __attribute__((alias(%#v)));
 // separate namespace (no _Cgo_ hacks like in gc).
 func (f *cgoFile) walker(cursor *astutil.Cursor, names map[string]clangCursor) bool {
 	switch node := cursor.Node().(type) {
+	case *ast.AssignStmt:
+		// An assign statement could be something like this:
+		//
+		//   val, errno := C.some_func()
+		//
+		// Check whether it looks like that, and if so, read the errno value and
+		// return it as the second return value. The call will be transformed
+		// into something like this:
+		//
+		//   val, errno := C.some_func(), C.__get_errno()
+		if len(node.Lhs) != 2 || len(node.Rhs) != 1 {
+			return true
+		}
+		rhs, ok := node.Rhs[0].(*ast.CallExpr)
+		if !ok {
+			return true
+		}
+		fun, ok := rhs.Fun.(*ast.SelectorExpr)
+		if !ok {
+			return true
+		}
+		x, ok := fun.X.(*ast.Ident)
+		if !ok {
+			return true
+		}
+		if found, ok := names[fun.Sel.Name]; ok && x.Name == "C" {
+			// Replace "C"."some_func" into "C.somefunc".
+			rhs.Fun = &ast.Ident{
+				NamePos: x.NamePos,
+				Name:    f.getASTDeclName(fun.Sel.Name, found, true),
+			}
+			// Add the errno value as the second value in the statement.
+			node.Rhs = append(node.Rhs, &ast.CallExpr{
+				Fun: &ast.Ident{
+					NamePos: node.Lhs[1].End(),
+					Name:    "_Cgo___get_errno",
+				},
+			})
+		}
 	case *ast.CallExpr:
 		fun, ok := node.Fun.(*ast.SelectorExpr)
 		if !ok {
@@ -1286,7 +1447,7 @@ func (f *cgoFile) walker(cursor *astutil.Cursor, names map[string]clangCursor) b
 			return true
 		}
 		if x.Name == "C" {
-			name := "C." + node.Sel.Name
+			name := "_Cgo_" + node.Sel.Name
 			if found, ok := names[node.Sel.Name]; ok {
 				name = f.getASTDeclName(node.Sel.Name, found, false)
 			}

cgo/cgo_go122.go

@@ -0,0 +1,17 @@
+//go:build go1.22
+
+package cgo
+
+// Code specifically for Go 1.22.
+
+import (
+	"go/ast"
+	"go/token"
+)
+
+func init() {
+	setASTFileFields = func(f *ast.File, start, end token.Pos) {
+		f.FileStart = start
+		f.FileEnd = end
+	}
+}

cgo/cgo_test.go

@@ -7,6 +7,7 @@ import (
 	"go/ast"
 	"go/format"
 	"go/parser"
+	"go/scanner"
 	"go/token"
 	"go/types"
 	"os"
@@ -55,7 +56,7 @@ func TestCGo(t *testing.T) {
 			}
 
 			// Process the AST with CGo.
-			cgoAST, _, _, _, _, cgoErrors := Process([]*ast.File{f}, "testdata", "main", fset, cflags, "")
+			cgoFiles, _, _, _, _, cgoErrors := Process([]*ast.File{f}, "testdata", "main", fset, cflags, "linux")
 
 			// Check the AST for type errors.
 			var typecheckErrors []error
@@ -63,10 +64,10 @@ func TestCGo(t *testing.T) {
 				Error: func(err error) {
 					typecheckErrors = append(typecheckErrors, err)
 				},
-				Importer: simpleImporter{},
+				Importer: newSimpleImporter(),
 				Sizes:    types.SizesFor("gccgo", "arm"),
 			}
-			_, err = config.Check("", fset, []*ast.File{f, cgoAST}, nil)
+			_, err = config.Check("", fset, append([]*ast.File{f}, cgoFiles...), nil)
 			if err != nil && len(typecheckErrors) == 0 {
 				// Only report errors when no type errors are found (an
 				// unexpected condition).
@@ -91,7 +92,7 @@ func TestCGo(t *testing.T) {
 				}
 				buf.WriteString("\n")
 			}
-			err = format.Node(buf, fset, cgoAST)
+			err = format.Node(buf, fset, cgoFiles[0])
 			if err != nil {
 				t.Errorf("could not write out CGo AST: %v", err)
 			}
@@ -201,14 +202,33 @@ func Test_cgoPackage_isEquivalentAST(t *testing.T) {
 }
 
 // simpleImporter implements the types.Importer interface, but only allows
-// importing the unsafe package.
+// importing the syscall and unsafe packages.
 type simpleImporter struct {
+	syscallPkg *types.Package
+}
+
+func newSimpleImporter() *simpleImporter {
+	i := &simpleImporter{}
+
+	// Implement a dummy syscall package with the Errno type.
+	i.syscallPkg = types.NewPackage("syscall", "syscall")
+	obj := types.NewTypeName(token.NoPos, i.syscallPkg, "Errno", nil)
+	named := types.NewNamed(obj, nil, nil)
+	i.syscallPkg.Scope().Insert(obj)
+	named.SetUnderlying(types.Typ[types.Uintptr])
+	sig := types.NewSignatureType(nil, nil, nil, types.NewTuple(), types.NewTuple(types.NewParam(token.NoPos, i.syscallPkg, "", types.Typ[types.String])), false)
+	named.AddMethod(types.NewFunc(token.NoPos, i.syscallPkg, "Error", sig))
+	i.syscallPkg.MarkComplete()
+
+	return i
 }
 
 // Import implements the Importer interface. For testing usage only: it only
 // supports importing the unsafe package.
-func (i simpleImporter) Import(path string) (*types.Package, error) {
+func (i *simpleImporter) Import(path string) (*types.Package, error) {
 	switch path {
+	case "syscall":
+		return i.syscallPkg, nil
 	case "unsafe":
 		return types.Unsafe, nil
 	default:
@@ -216,10 +236,16 @@ func (i simpleImporter) Import(path string) (*types.Package, error) {
 	}
 }
 
-// formatDiagnostics formats the error message to be an indented comment. It
+// formatDiagnostic formats the error message to be an indented comment. It
 // also fixes Windows path name issues (backward slashes).
 func formatDiagnostic(err error) string {
-	msg := err.Error()
+	var msg string
+	switch err := err.(type) {
+	case scanner.Error:
+		msg = err.Pos.String() + ": " + err.Msg
+	default:
+		msg = err.Error()
+	}
 	if runtime.GOOS == "windows" {
 		// Fix Windows path slashes.
 		msg = strings.ReplaceAll(msg, "testdata\\", "testdata/")

cgo/const.go

@@ -17,6 +17,8 @@ var (
 		token.OR:  precedenceOr,
 		token.XOR: precedenceXor,
 		token.AND: precedenceAnd,
+		token.SHL: precedenceShift,
+		token.SHR: precedenceShift,
 		token.ADD: precedenceAdd,
 		token.SUB: precedenceAdd,
 		token.MUL: precedenceMul,
@@ -25,11 +27,13 @@ var (
 	}
 )
 
+// See: https://en.cppreference.com/w/c/language/operator_precedence
 const (
 	precedenceLowest = iota + 1
 	precedenceOr
 	precedenceXor
 	precedenceAnd
+	precedenceShift
 	precedenceAdd
 	precedenceMul
 	precedencePrefix
@@ -50,8 +54,72 @@ func init() {
 }
 
 // parseConst parses the given string as a C constant.
-func parseConst(pos token.Pos, fset *token.FileSet, value string) (ast.Expr, *scanner.Error) {
-	t := newTokenizer(pos, fset, value)
+func parseConst(pos token.Pos, fset *token.FileSet, value string, params []ast.Expr, callerPos token.Pos, f *cgoFile) (ast.Expr, *scanner.Error) {
+	t := newTokenizer(pos, fset, value, f)
+
+	// If params is non-nil (could be a zero length slice), this const is
+	// actually a function-call like expression from another macro.
+	// This means we have to parse a string like "(a, b) (a+b)".
+	// We do this by parsing the parameters at the start and then treating the
+	// following like a normal constant expression.
+	if params != nil {
+		// Parse opening paren.
+		if t.curToken != token.LPAREN {
+			return nil, unexpectedToken(t, token.LPAREN)
+		}
+		t.Next()
+
+		// Parse parameters (identifiers) and closing paren.
+		var paramIdents []string
+		for i := 0; ; i++ {
+			if i == 0 && t.curToken == token.RPAREN {
+				// No parameters, break early.
+				t.Next()
+				break
+			}
+
+			// Read the parameter name.
+			if t.curToken != token.IDENT {
+				return nil, unexpectedToken(t, token.IDENT)
+			}
+			paramIdents = append(paramIdents, t.curValue)
+			t.Next()
+
+			// Read the next token: either a continuation (comma) or end of list
+			// (rparen).
+			if t.curToken == token.RPAREN {
+				// End of parameter list.
+				t.Next()
+				break
+			} else if t.curToken == token.COMMA {
+				// Comma, so there will be another parameter name.
+				t.Next()
+			} else {
+				return nil, &scanner.Error{
+					Pos: t.fset.Position(t.curPos),
+					Msg: "unexpected token " + t.curToken.String() + " inside macro parameters, expected ',' or ')'",
+				}
+			}
+		}
+
+		// Report an error if there is a mismatch in parameter length.
+		// The error is reported at the location of the closing paren from the
+		// caller location.
+		if len(params) != len(paramIdents) {
+			return nil, &scanner.Error{
+				Pos: t.fset.Position(callerPos),
+				Msg: fmt.Sprintf("unexpected number of parameters: expected %d, got %d", len(paramIdents), len(params)),
+			}
+		}
+
+		// Assign values to the parameters.
+		// These parameter names are closer in 'scope' than other identifiers so
+		// will be used first when parsing an identifier.
+		for i, name := range paramIdents {
+			t.params[name] = params[i]
+		}
+	}
+
 	expr, err := parseConstExpr(t, precedenceLowest)
 	t.Next()
 	if t.curToken != token.EOF {
@@ -82,7 +150,7 @@ func parseConstExpr(t *tokenizer, precedence int) (ast.Expr, *scanner.Error) {
 
 	for t.peekToken != token.EOF && precedence < precedences[t.peekToken] {
 		switch t.peekToken {
-		case token.OR, token.XOR, token.AND, token.ADD, token.SUB, token.MUL, token.QUO, token.REM:
+		case token.OR, token.XOR, token.AND, token.SHL, token.SHR, token.ADD, token.SUB, token.MUL, token.QUO, token.REM:
 			t.Next()
 			leftExpr, err = parseBinaryExpr(t, leftExpr)
 		}
@@ -92,6 +160,68 @@ func parseConstExpr(t *tokenizer, precedence int) (ast.Expr, *scanner.Error) {
 }
 
 func parseIdent(t *tokenizer) (ast.Expr, *scanner.Error) {
+	// If the identifier is one of the parameters of this function-like macro,
+	// use the parameter value.
+	if val, ok := t.params[t.curValue]; ok {
+		return val, nil
+	}
+
+	if t.f != nil {
+		// Check whether this identifier is actually a macro "call" with
+		// parameters. In that case, we should parse the parameters and pass it
+		// on to a new invocation of parseConst.
+		if t.peekToken == token.LPAREN {
+			if cursor, ok := t.f.names[t.curValue]; ok && t.f.isFunctionLikeMacro(cursor) {
+				// We know the current and peek tokens (the peek one is the '('
+				// token). So skip ahead until the current token is the first
+				// unknown token.
+				t.Next()
+				t.Next()
+
+				// Parse the list of parameters until ')' (rparen) is found.
+				params := []ast.Expr{}
+				for i := 0; ; i++ {
+					if i == 0 && t.curToken == token.RPAREN {
+						break
+					}
+					x, err := parseConstExpr(t, precedenceLowest)
+					if err != nil {
+						return nil, err
+					}
+					params = append(params, x)
+					t.Next()
+					if t.curToken == token.COMMA {
+						t.Next()
+					} else if t.curToken == token.RPAREN {
+						break
+					} else {
+						return nil, &scanner.Error{
+							Pos: t.fset.Position(t.curPos),
+							Msg: "unexpected token " + t.curToken.String() + ", ',' or ')'",
+						}
+					}
+				}
+
+				// Evaluate the macro value and use it as the identifier value.
+				rparen := t.curPos
+				pos, text := t.f.getMacro(cursor)
+				return parseConst(pos, t.fset, text, params, rparen, t.f)
+			}
+		}
+
+		// Normally the name is something defined in the file (like another
+		// macro) which we get the declaration from using getASTDeclName.
+		// This ensures that names that are only referenced inside a macro are
+		// still getting defined.
+		if cursor, ok := t.f.names[t.curValue]; ok {
+			return &ast.Ident{
+				NamePos: t.curPos,
+				Name:    t.f.getASTDeclName(t.curValue, cursor, false),
+			}, nil
+		}
+	}
+
+	// t.f is nil during testing. This is a fallback.
 	return &ast.Ident{
 		NamePos: t.curPos,
 		Name:    "C." + t.curValue,
@@ -160,21 +290,25 @@ func unexpectedToken(t *tokenizer, expected token.Token) *scanner.Error {
 
 // tokenizer reads C source code and converts it to Go tokens.
 type tokenizer struct {
+	f                   *cgoFile
 	curPos, peekPos     token.Pos
 	fset                *token.FileSet
 	curToken, peekToken token.Token
 	curValue, peekValue string
 	buf                 string
+	params              map[string]ast.Expr
 }
 
 // newTokenizer initializes a new tokenizer, positioned at the first token in
 // the string.
-func newTokenizer(start token.Pos, fset *token.FileSet, buf string) *tokenizer {
+func newTokenizer(start token.Pos, fset *token.FileSet, buf string, f *cgoFile) *tokenizer {
 	t := &tokenizer{
+		f:         f,
 		peekPos:   start,
 		fset:      fset,
 		buf:       buf,
 		peekToken: token.ILLEGAL,
+		params:    make(map[string]ast.Expr),
 	}
 	// Parse the first two tokens (cur and peek).
 	t.Next()
@@ -191,7 +325,9 @@ func (t *tokenizer) Next() {
 	t.curValue = t.peekValue
 
 	// Parse the next peek token.
-	t.peekPos += token.Pos(len(t.curValue))
+	if t.peekPos != token.NoPos {
+		t.peekPos += token.Pos(len(t.curValue))
+	}
 	for {
 		if len(t.buf) == 0 {
 			t.peekToken = token.EOF
@@ -203,20 +339,28 @@ func (t *tokenizer) Next() {
 			// Skip whitespace.
 			// Based on this source, not sure whether it represents C whitespace:
 			// https://en.cppreference.com/w/cpp/string/byte/isspace
-			t.peekPos++
+			if t.peekPos != token.NoPos {
+				t.peekPos++
+			}
 			t.buf = t.buf[1:]
-		case len(t.buf) >= 2 && (string(t.buf[:2]) == "||" || string(t.buf[:2]) == "&&"):
+		case len(t.buf) >= 2 && (string(t.buf[:2]) == "||" || string(t.buf[:2]) == "&&" || string(t.buf[:2]) == "<<" || string(t.buf[:2]) == ">>"):
 			// Two-character tokens.
 			switch c {
 			case '&':
 				t.peekToken = token.LAND
 			case '|':
 				t.peekToken = token.LOR
+			case '<':
+				t.peekToken = token.SHL
+			case '>':
+				t.peekToken = token.SHR
+			default:
+				panic("unreachable")
 			}
 			t.peekValue = t.buf[:2]
 			t.buf = t.buf[2:]
 			return
-		case c == '(' || c == ')' || c == '+' || c == '-' || c == '*' || c == '/' || c == '%' || c == '&' || c == '|' || c == '^':
+		case c == '(' || c == ')' || c == ',' || c == '+' || c == '-' || c == '*' || c == '/' || c == '%' || c == '&' || c == '|' || c == '^':
 			// Single-character tokens.
 			// TODO: ++ (increment) and -- (decrement) operators.
 			switch c {
@@ -224,6 +368,8 @@ func (t *tokenizer) Next() {
 				t.peekToken = token.LPAREN
 			case ')':
 				t.peekToken = token.RPAREN
+			case ',':
+				t.peekToken = token.COMMA
 			case '+':
 				t.peekToken = token.ADD
 			case '-':

cgo/const_test.go

@@ -40,6 +40,10 @@ func TestParseConst(t *testing.T) {
 		{`5&5`, `5 & 5`},
 		{`5|5`, `5 | 5`},
 		{`5^5`, `5 ^ 5`},
+		{`5<<5`, `5 << 5`},
+		{`5>>5`, `5 >> 5`},
+		{`5>>5 + 3`, `5 >> (5 + 3)`},
+		{`5>>5 ^ 3`, `5>>5 ^ 3`},
 		{`5||5`, `error: 1:2: unexpected token ||, expected end of expression`}, // logical binops aren't supported yet
 		{`(5/5)`, `(5 / 5)`},
 		{`1 - 2`, `1 - 2`},
@@ -55,7 +59,7 @@ func TestParseConst(t *testing.T) {
 	} {
 		fset := token.NewFileSet()
 		startPos := fset.AddFile("", -1, 1000).Pos(0)
-		expr, err := parseConst(startPos, fset, tc.C)
+		expr, err := parseConst(startPos, fset, tc.C, nil, token.NoPos, nil)
 		s := "<invalid>"
 		if err != nil {
 			if !strings.HasPrefix(tc.Go, "error: ") {

cgo/libclang.go

@@ -4,6 +4,7 @@ package cgo
 // modification. It does not touch the AST itself.
 
 import (
+	"bytes"
 	"crypto/sha256"
 	"crypto/sha512"
 	"encoding/hex"
@@ -60,11 +61,26 @@ CXSourceRange tinygo_clang_getCursorExtent(GoCXCursor c);
 CXTranslationUnit tinygo_clang_Cursor_getTranslationUnit(GoCXCursor c);
 long long tinygo_clang_getEnumConstantDeclValue(GoCXCursor c);
 CXType tinygo_clang_getEnumDeclIntegerType(GoCXCursor c);
+unsigned tinygo_clang_Cursor_isAnonymous(GoCXCursor c);
 unsigned tinygo_clang_Cursor_isBitField(GoCXCursor c);
+unsigned tinygo_clang_Cursor_isMacroFunctionLike(GoCXCursor c);
 
+// Fix some warnings on Windows ARM. Without the __declspec(dllexport), it gives warnings like this:
+//     In file included from _cgo_export.c:4:
+//     cgo-gcc-export-header-prolog:49:34: warning: redeclaration of 'tinygo_clang_globals_visitor' should not add 'dllexport' attribute [-Wdll-attribute-on-redeclaration]
+//     libclang.go:68:5: note: previous declaration is here
+// See: https://github.com/golang/go/issues/49721
+#if defined(_WIN32)
+#define CGO_DECL // __declspec(dllexport)
+#else
+#define CGO_DECL
+#endif
+
+CGO_DECL
 int tinygo_clang_globals_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
+CGO_DECL
 int tinygo_clang_struct_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
-int tinygo_clang_enum_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
+CGO_DECL
 void tinygo_clang_inclusion_visitor(CXFile included_file, CXSourceLocation *inclusion_stack, unsigned include_len, CXClientData client_data);
 */
 import "C"
@@ -203,7 +219,7 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 		numArgs := int(C.tinygo_clang_Cursor_getNumArguments(c))
 		obj := &ast.Object{
 			Kind: ast.Fun,
-			Name: "C." + name,
+			Name: "_Cgo_" + name,
 		}
 		exportName := name
 		localName := name
@@ -241,7 +257,7 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 			},
 			Name: &ast.Ident{
 				NamePos: pos,
-				Name:    "C." + localName,
+				Name:    "_Cgo_" + localName,
 				Obj:     obj,
 			},
 			Type: &ast.FuncType{
@@ -253,10 +269,18 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 				},
 			},
 		}
+		var doc []string
 		if C.clang_isFunctionTypeVariadic(cursorType) != 0 {
+			doc = append(doc, "//go:variadic")
+		}
+		if _, ok := f.noescapingFuncs[name]; ok {
+			doc = append(doc, "//go:noescape")
+			f.noescapingFuncs[name].used = true
+		}
+		if len(doc) != 0 {
 			decl.Doc.List = append(decl.Doc.List, &ast.Comment{
 				Slash: pos - 1,
-				Text:  "//go:variadic",
+				Text:  strings.Join(doc, "\n"),
 			})
 		}
 		for i := 0; i < numArgs; i++ {
@@ -295,7 +319,7 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 		return decl, stringSignature
 	case C.CXCursor_StructDecl, C.CXCursor_UnionDecl:
 		typ := f.makeASTRecordType(c, pos)
-		typeName := "C." + name
+		typeName := "_Cgo_" + name
 		typeExpr := typ.typeExpr
 		if typ.unionSize != 0 {
 			// Convert to a single-field struct type.
@@ -316,7 +340,7 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 		obj.Decl = typeSpec
 		return typeSpec, typ
 	case C.CXCursor_TypedefDecl:
-		typeName := "C." + name
+		typeName := "_Cgo_" + name
 		underlyingType := C.tinygo_clang_getTypedefDeclUnderlyingType(c)
 		obj := &ast.Object{
 			Kind: ast.Typ,
@@ -354,12 +378,12 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 		}
 		obj := &ast.Object{
 			Kind: ast.Var,
-			Name: "C." + name,
+			Name: "_Cgo_" + name,
 		}
 		valueSpec := &ast.ValueSpec{
 			Names: []*ast.Ident{{
 				NamePos: pos,
-				Name:    "C." + name,
+				Name:    "_Cgo_" + name,
 				Obj:     obj,
 			}},
 			Type: typeExpr,
@@ -368,42 +392,8 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 		gen.Specs = append(gen.Specs, valueSpec)
 		return gen, nil
 	case C.CXCursor_MacroDefinition:
-		sourceRange := C.tinygo_clang_getCursorExtent(c)
-		start := C.clang_getRangeStart(sourceRange)
-		end := C.clang_getRangeEnd(sourceRange)
-		var file, endFile C.CXFile
-		var startOffset, endOffset C.unsigned
-		C.clang_getExpansionLocation(start, &file, nil, nil, &startOffset)
-		if file == nil {
-			f.addError(pos, "internal error: could not find file where macro is defined")
-			return nil, nil
-		}
-		C.clang_getExpansionLocation(end, &endFile, nil, nil, &endOffset)
-		if file != endFile {
-			f.addError(pos, "internal error: expected start and end location of a macro to be in the same file")
-			return nil, nil
-		}
-		if startOffset > endOffset {
-			f.addError(pos, "internal error: start offset of macro is after end offset")
-			return nil, nil
-		}
-
-		// read file contents and extract the relevant byte range
-		tu := C.tinygo_clang_Cursor_getTranslationUnit(c)
-		var size C.size_t
-		sourcePtr := C.clang_getFileContents(tu, file, &size)
-		if endOffset >= C.uint(size) {
-			f.addError(pos, "internal error: end offset of macro lies after end of file")
-			return nil, nil
-		}
-		source := string(((*[1 << 28]byte)(unsafe.Pointer(sourcePtr)))[startOffset:endOffset:endOffset])
-		if !strings.HasPrefix(source, name) {
-			f.addError(pos, fmt.Sprintf("internal error: expected macro value to start with %#v, got %#v", name, source))
-			return nil, nil
-		}
-		value := source[len(name):]
-		// Try to convert this #define into a Go constant expression.
-		expr, scannerError := parseConst(pos+token.Pos(len(name)), f.fset, value)
+		tokenPos, value := f.getMacro(c)
+		expr, scannerError := parseConst(tokenPos, f.fset, value, nil, token.NoPos, f)
 		if scannerError != nil {
 			f.errors = append(f.errors, *scannerError)
 			return nil, nil
@@ -417,12 +407,12 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 		}
 		obj := &ast.Object{
 			Kind: ast.Con,
-			Name: "C." + name,
+			Name: "_Cgo_" + name,
 		}
 		valueSpec := &ast.ValueSpec{
 			Names: []*ast.Ident{{
 				NamePos: pos,
-				Name:    "C." + name,
+				Name:    "_Cgo_" + name,
 				Obj:     obj,
 			}},
 			Values: []ast.Expr{expr},
@@ -433,7 +423,7 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 	case C.CXCursor_EnumDecl:
 		obj := &ast.Object{
 			Kind: ast.Typ,
-			Name: "C." + name,
+			Name: "_Cgo_" + name,
 		}
 		underlying := C.tinygo_clang_getEnumDeclIntegerType(c)
 		// TODO: gc's CGo implementation uses types such as `uint32` for enums
@@ -441,7 +431,7 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 		typeSpec := &ast.TypeSpec{
 			Name: &ast.Ident{
 				NamePos: pos,
-				Name:    "C." + name,
+				Name:    "_Cgo_" + name,
 				Obj:     obj,
 			},
 			Assign: pos,
@@ -464,12 +454,12 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 		}
 		obj := &ast.Object{
 			Kind: ast.Con,
-			Name: "C." + name,
+			Name: "_Cgo_" + name,
 		}
 		valueSpec := &ast.ValueSpec{
 			Names: []*ast.Ident{{
 				NamePos: pos,
-				Name:    "C." + name,
+				Name:    "_Cgo_" + name,
 				Obj:     obj,
 			}},
 			Values: []ast.Expr{expr},
@@ -483,6 +473,62 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 	}
 }
 
+// Return whether this is a macro that's also function-like, like this:
+//
+//	#define add(a, b) (a+b)
+func (f *cgoFile) isFunctionLikeMacro(c clangCursor) bool {
+	if C.tinygo_clang_getCursorKind(c) != C.CXCursor_MacroDefinition {
+		return false
+	}
+	return C.tinygo_clang_Cursor_isMacroFunctionLike(c) != 0
+}
+
+// Get the macro value: the position in the source file and the string value of
+// the macro.
+func (f *cgoFile) getMacro(c clangCursor) (pos token.Pos, value string) {
+	// Extract tokens from the Clang tokenizer.
+	// See: https://stackoverflow.com/a/19074846/559350
+	sourceRange := C.tinygo_clang_getCursorExtent(c)
+	tu := C.tinygo_clang_Cursor_getTranslationUnit(c)
+	var rawTokens *C.CXToken
+	var numTokens C.unsigned
+	C.clang_tokenize(tu, sourceRange, &rawTokens, &numTokens)
+	tokens := unsafe.Slice(rawTokens, numTokens)
+	defer C.clang_disposeTokens(tu, rawTokens, numTokens)
+
+	// Convert this range of tokens back to source text.
+	// Ugly, but it works well enough.
+	sourceBuf := &bytes.Buffer{}
+	var startOffset int
+	for i, token := range tokens {
+		spelling := getString(C.clang_getTokenSpelling(tu, token))
+		location := C.clang_getTokenLocation(tu, token)
+		var tokenOffset C.unsigned
+		C.clang_getExpansionLocation(location, nil, nil, nil, &tokenOffset)
+		if i == 0 {
+			// The first token is the macro name itself.
+			// Skip it (after using its location).
+			startOffset = int(tokenOffset)
+		} else {
+			// Later tokens are the macro contents.
+			for int(tokenOffset) > (startOffset + sourceBuf.Len()) {
+				// Pad the source text with whitespace (that must have been
+				// present in the original source as well).
+				sourceBuf.WriteByte(' ')
+			}
+			sourceBuf.WriteString(spelling)
+		}
+	}
+	value = sourceBuf.String()
+
+	// Obtain the position of this token. This is the position of the first
+	// character in the 'value' string and is used to report errors at the
+	// correct location in the source file.
+	pos = f.getCursorPosition(c)
+
+	return
+}
+
 func getString(clangString C.CXString) (s string) {
 	rawString := C.clang_getCString(clangString)
 	s = C.GoString(rawString)
@@ -588,6 +634,13 @@ func (p *cgoPackage) getClangLocationPosition(location C.CXSourceLocation, tu C.
 		f := p.fset.AddFile(filename, -1, int(size))
 		f.SetLines(lines)
 		p.tokenFiles[filename] = f
+		// Add dummy file AST, to satisfy the type checker.
+		astFile := &ast.File{
+			Package: f.Pos(0),
+			Name:    ast.NewIdent(p.packageName),
+		}
+		setASTFileFields(astFile, f.Pos(0), f.Pos(int(size)))
+		p.cgoFiles = append(p.cgoFiles, astFile)
 	}
 	positionFile := p.tokenFiles[filename]
 
@@ -634,13 +687,6 @@ func (p *cgoPackage) addErrorAfter(pos token.Pos, after, msg string) {
 
 // addErrorAt is a utility function to add an error to the list of errors.
 func (p *cgoPackage) addErrorAt(position token.Position, msg string) {
-	if filepath.IsAbs(position.Filename) {
-		// Relative paths for readability, like other Go parser errors.
-		relpath, err := filepath.Rel(p.currentDir, position.Filename)
-		if err == nil {
-			position.Filename = relpath
-		}
-	}
 	p.errors = append(p.errors, scanner.Error{
 		Pos: position,
 		Msg: msg,
@@ -653,8 +699,19 @@ func (p *cgoPackage) addErrorAt(position token.Position, msg string) {
 func (f *cgoFile) makeDecayingASTType(typ C.CXType, pos token.Pos) ast.Expr {
 	// Strip typedefs, if any.
 	underlyingType := typ
+	if underlyingType.kind == C.CXType_Elaborated {
+		// Starting with LLVM 16, the elaborated type is used for more types.
+		// According to the Clang documentation, the elaborated type has no
+		// semantic meaning so can be stripped (it is used to better convey type
+		// name information).
+		// Source:
+		// https://clang.llvm.org/doxygen/classclang_1_1ElaboratedType.html#details
+		// > The type itself is always "sugar", used to express what was written
+		// > in the source code but containing no additional semantic information.
+		underlyingType = C.clang_Type_getNamedType(underlyingType)
+	}
 	if underlyingType.kind == C.CXType_Typedef {
-		c := C.tinygo_clang_getTypeDeclaration(typ)
+		c := C.tinygo_clang_getTypeDeclaration(underlyingType)
 		underlyingType = C.tinygo_clang_getTypedefDeclUnderlyingType(c)
 		// TODO: support a chain of typedefs. At the moment, it seems to get
 		// stuck in an endless loop when trying to get to the most underlying
@@ -688,27 +745,27 @@ func (f *cgoFile) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 	var typeName string
 	switch typ.kind {
 	case C.CXType_Char_S, C.CXType_Char_U:
-		typeName = "C.char"
+		typeName = "_Cgo_char"
 	case C.CXType_SChar:
-		typeName = "C.schar"
+		typeName = "_Cgo_schar"
 	case C.CXType_UChar:
-		typeName = "C.uchar"
+		typeName = "_Cgo_uchar"
 	case C.CXType_Short:
-		typeName = "C.short"
+		typeName = "_Cgo_short"
 	case C.CXType_UShort:
-		typeName = "C.ushort"
+		typeName = "_Cgo_ushort"
 	case C.CXType_Int:
-		typeName = "C.int"
+		typeName = "_Cgo_int"
 	case C.CXType_UInt:
-		typeName = "C.uint"
+		typeName = "_Cgo_uint"
 	case C.CXType_Long:
-		typeName = "C.long"
+		typeName = "_Cgo_long"
 	case C.CXType_ULong:
-		typeName = "C.ulong"
+		typeName = "_Cgo_ulong"
 	case C.CXType_LongLong:
-		typeName = "C.longlong"
+		typeName = "_Cgo_longlong"
 	case C.CXType_ULongLong:
-		typeName = "C.ulonglong"
+		typeName = "_Cgo_ulonglong"
 	case C.CXType_Bool:
 		typeName = "bool"
 	case C.CXType_Float, C.CXType_Double, C.CXType_LongDouble:
@@ -788,6 +845,8 @@ func (f *cgoFile) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 			return f.makeASTType(underlying, pos)
 		case C.CXType_Enum:
 			return f.makeASTType(underlying, pos)
+		case C.CXType_Typedef:
+			return f.makeASTType(underlying, pos)
 		default:
 			typeKindSpelling := getString(C.clang_getTypeKindSpelling(underlying.kind))
 			f.addError(pos, fmt.Sprintf("unknown elaborated type (libclang type kind %s)", typeKindSpelling))
@@ -806,7 +865,7 @@ func (f *cgoFile) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 			// makeASTRecordType will create an appropriate error.
 			cgoRecordPrefix = "record_"
 		}
-		if name == "" {
+		if name == "" || C.tinygo_clang_Cursor_isAnonymous(cursor) != 0 {
 			// Anonymous record, probably inside a typedef.
 			location := f.getUniqueLocationID(pos, cursor)
 			name = f.getUnnamedDeclName("_Ctype_"+cgoRecordPrefix+"__", location)
@@ -837,7 +896,7 @@ func (f *cgoFile) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 		typeSpelling := getString(C.clang_getTypeSpelling(typ))
 		typeKindSpelling := getString(C.clang_getTypeKindSpelling(typ.kind))
 		f.addError(pos, fmt.Sprintf("unknown C type: %v (libclang type kind %s)", typeSpelling, typeKindSpelling))
-		typeName = "C.<unknown>"
+		typeName = "_Cgo_<unknown>"
 	}
 	return &ast.Ident{
 		NamePos: pos,
@@ -854,7 +913,7 @@ func (p *cgoPackage) getIntegerType(name string, cursor clangCursor) *ast.TypeSp
 	var goName string
 	typeSize := C.clang_Type_getSizeOf(underlyingType)
 	switch name {
-	case "C.char":
+	case "_Cgo_char":
 		if typeSize != 1 {
 			// This happens for some very special purpose architectures
 			// (DSPs etc.) that are not currently targeted.
@@ -867,7 +926,7 @@ func (p *cgoPackage) getIntegerType(name string, cursor clangCursor) *ast.TypeSp
 		case C.CXType_Char_U:
 			goName = "uint8"
 		}
-	case "C.schar", "C.short", "C.int", "C.long", "C.longlong":
+	case "_Cgo_schar", "_Cgo_short", "_Cgo_int", "_Cgo_long", "_Cgo_longlong":
 		switch typeSize {
 		case 1:
 			goName = "int8"
@@ -878,7 +937,7 @@ func (p *cgoPackage) getIntegerType(name string, cursor clangCursor) *ast.TypeSp
 		case 8:
 			goName = "int64"
 		}
-	case "C.uchar", "C.ushort", "C.uint", "C.ulong", "C.ulonglong":
+	case "_Cgo_uchar", "_Cgo_ushort", "_Cgo_uint", "_Cgo_ulong", "_Cgo_ulonglong":
 		switch typeSize {
 		case 1:
 			goName = "uint8"

cgo/libclang_config_llvm14.go

@@ -1,15 +0,0 @@
-//go:build !byollvm && llvm14
-
-package cgo
-
-/*
-#cgo linux        CFLAGS:  -I/usr/lib/llvm-14/include
-#cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@14/include
-#cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@14/include
-#cgo freebsd      CFLAGS:  -I/usr/local/llvm14/include
-#cgo linux        LDFLAGS: -L/usr/lib/llvm-14/lib -lclang
-#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@14/lib -lclang -lffi
-#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@14/lib -lclang -lffi
-#cgo freebsd      LDFLAGS: -L/usr/local/llvm14/lib -lclang
-*/
-import "C"

cgo/libclang_config_llvm15.go

@@ -1,4 +1,4 @@
-//go:build !byollvm && !llvm14
+//go:build !byollvm && llvm15
 
 package cgo
 
@@ -8,8 +8,8 @@ package cgo
 #cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@15/include
 #cgo freebsd      CFLAGS:  -I/usr/local/llvm15/include
 #cgo linux        LDFLAGS: -L/usr/lib/llvm-15/lib -lclang
-#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@15/lib -lclang -lffi
-#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@15/lib -lclang -lffi
+#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@15/lib -lclang
+#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@15/lib -lclang
 #cgo freebsd      LDFLAGS: -L/usr/local/llvm15/lib -lclang
 */
 import "C"

cgo/libclang_config_llvm16.go

@@ -0,0 +1,21 @@
+//go:build !byollvm && llvm16
+
+package cgo
+
+// As of 2023-05-05, there is a packaging issue with LLVM 16 on Debian:
+// https://github.com/llvm/llvm-project/issues/62199
+// A workaround is to fix this locally, using something like this:
+//
+//   ln -sf ../../x86_64-linux-gnu/libclang-16.so.1 /usr/lib/llvm-16/lib/libclang.so
+
+/*
+#cgo linux        CFLAGS:  -I/usr/lib/llvm-16/include
+#cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@16/include
+#cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@16/include
+#cgo freebsd      CFLAGS:  -I/usr/local/llvm16/include
+#cgo linux        LDFLAGS: -L/usr/lib/llvm-16/lib -lclang
+#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@16/lib -lclang
+#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@16/lib -lclang
+#cgo freebsd      LDFLAGS: -L/usr/local/llvm16/lib -lclang
+*/
+import "C"

cgo/libclang_config_llvm17.go

@@ -0,0 +1,15 @@
+//go:build !byollvm && llvm17
+
+package cgo
+
+/*
+#cgo linux        CFLAGS:  -I/usr/include/llvm-17 -I/usr/include/llvm-c-17 -I/usr/lib/llvm-17/include
+#cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@17/include
+#cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@17/include
+#cgo freebsd      CFLAGS:  -I/usr/local/llvm17/include
+#cgo linux        LDFLAGS: -L/usr/lib/llvm-17/lib -lclang
+#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@17/lib -lclang
+#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@17/lib -lclang
+#cgo freebsd      LDFLAGS: -L/usr/local/llvm17/lib -lclang
+*/
+import "C"

cgo/libclang_config_llvm18.go

@@ -0,0 +1,15 @@
+//go:build !byollvm && llvm18
+
+package cgo
+
+/*
+#cgo linux        CFLAGS:  -I/usr/include/llvm-18 -I/usr/include/llvm-c-18 -I/usr/lib/llvm-18/include
+#cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@18/include
+#cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@18/include
+#cgo freebsd      CFLAGS:  -I/usr/local/llvm18/include
+#cgo linux        LDFLAGS: -L/usr/lib/llvm-18/lib -lclang
+#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@18/lib -lclang
+#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@18/lib -lclang
+#cgo freebsd      LDFLAGS: -L/usr/local/llvm18/lib -lclang
+*/
+import "C"

cgo/libclang_config_llvm19.go

@@ -0,0 +1,15 @@
+//go:build !byollvm && !llvm15 && !llvm16 && !llvm17 && !llvm18
+
+package cgo
+
+/*
+#cgo linux        CFLAGS:  -I/usr/include/llvm-19 -I/usr/include/llvm-c-19 -I/usr/lib/llvm-19/include
+#cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@19/include
+#cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@19/include
+#cgo freebsd      CFLAGS:  -I/usr/local/llvm19/include
+#cgo linux        LDFLAGS: -L/usr/lib/llvm-19/lib -lclang
+#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@19/lib -lclang
+#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@19/lib -lclang
+#cgo freebsd      LDFLAGS: -L/usr/local/llvm19/lib -lclang
+*/
+import "C"

cgo/libclang_stubs.c

@@ -77,6 +77,14 @@ CXType tinygo_clang_getEnumDeclIntegerType(CXCursor c) {
 	return clang_getEnumDeclIntegerType(c);
 }
 
+unsigned tinygo_clang_Cursor_isAnonymous(CXCursor c) {
+	return clang_Cursor_isAnonymous(c);
+}
+
 unsigned tinygo_clang_Cursor_isBitField(CXCursor c) {
 	return clang_Cursor_isBitField(c);
-}
+}
+
+unsigned tinygo_clang_Cursor_isMacroFunctionLike(CXCursor c) {
+	return clang_Cursor_isMacroFunctionLike(c);
+}

cgo/testdata/basic.out.go

@@ -1,39 +1,54 @@
 package main
 
+import "syscall"
 import "unsafe"
 
 var _ unsafe.Pointer
 
-//go:linkname C.CString runtime.cgo_CString
-func C.CString(string) *C.char
+//go:linkname _Cgo_CString runtime.cgo_CString
+func _Cgo_CString(string) *_Cgo_char
 
-//go:linkname C.GoString runtime.cgo_GoString
-func C.GoString(*C.char) string
+//go:linkname _Cgo_GoString runtime.cgo_GoString
+func _Cgo_GoString(*_Cgo_char) string
 
-//go:linkname C.__GoStringN runtime.cgo_GoStringN
-func C.__GoStringN(*C.char, uintptr) string
+//go:linkname _Cgo___GoStringN runtime.cgo_GoStringN
+func _Cgo___GoStringN(*_Cgo_char, uintptr) string
 
-func C.GoStringN(cstr *C.char, length C.int) string {
-	return C.__GoStringN(cstr, uintptr(length))
+func _Cgo_GoStringN(cstr *_Cgo_char, length _Cgo_int) string {
+	return _Cgo___GoStringN(cstr, uintptr(length))
 }
 
-//go:linkname C.__GoBytes runtime.cgo_GoBytes
-func C.__GoBytes(unsafe.Pointer, uintptr) []byte
+//go:linkname _Cgo___GoBytes runtime.cgo_GoBytes
+func _Cgo___GoBytes(unsafe.Pointer, uintptr) []byte
 
-func C.GoBytes(ptr unsafe.Pointer, length C.int) []byte {
-	return C.__GoBytes(ptr, uintptr(length))
+func _Cgo_GoBytes(ptr unsafe.Pointer, length _Cgo_int) []byte {
+	return _Cgo___GoBytes(ptr, uintptr(length))
+}
+
+//go:linkname _Cgo___CBytes runtime.cgo_CBytes
+func _Cgo___CBytes([]byte) unsafe.Pointer
+
+func _Cgo_CBytes(b []byte) unsafe.Pointer {
+	return _Cgo___CBytes(b)
+}
+
+//go:linkname _Cgo___get_errno_num runtime.cgo_errno
+func _Cgo___get_errno_num() uintptr
+
+func _Cgo___get_errno() error {
+	return syscall.Errno(_Cgo___get_errno_num())
 }
 
 type (
-	C.char      uint8
-	C.schar     int8
-	C.uchar     uint8
-	C.short     int16
-	C.ushort    uint16
-	C.int       int32
-	C.uint      uint32
-	C.long      int32
-	C.ulong     uint32
-	C.longlong  int64
-	C.ulonglong uint64
+	_Cgo_char      uint8
+	_Cgo_schar     int8
+	_Cgo_uchar     uint8
+	_Cgo_short     int16
+	_Cgo_ushort    uint16
+	_Cgo_int       int32
+	_Cgo_uint      uint32
+	_Cgo_long      int32
+	_Cgo_ulong     uint32
+	_Cgo_longlong  int64
+	_Cgo_ulonglong uint64
 )

cgo/testdata/const.go

@@ -3,10 +3,26 @@ package main
 /*
 #define foo 3
 #define bar foo
+
+#define unreferenced 4
+#define referenced unreferenced
+
+#define fnlike() 5
+#define fnlike_val fnlike()
+#define square(n) (n*n)
+#define square_val square(20)
+#define add(a, b) (a + b)
+#define add_val add(3, 5)
 */
 import "C"
 
 const (
 	Foo = C.foo
 	Bar = C.bar
+
+	Baz = C.referenced
+
+	fnlike = C.fnlike_val
+	square = C.square_val
+	add    = C.add_val
 )

cgo/testdata/const.out.go

@@ -1,42 +1,62 @@
 package main
 
+import "syscall"
 import "unsafe"
 
 var _ unsafe.Pointer
 
-//go:linkname C.CString runtime.cgo_CString
-func C.CString(string) *C.char
+//go:linkname _Cgo_CString runtime.cgo_CString
+func _Cgo_CString(string) *_Cgo_char
 
-//go:linkname C.GoString runtime.cgo_GoString
-func C.GoString(*C.char) string
+//go:linkname _Cgo_GoString runtime.cgo_GoString
+func _Cgo_GoString(*_Cgo_char) string
 
-//go:linkname C.__GoStringN runtime.cgo_GoStringN
-func C.__GoStringN(*C.char, uintptr) string
+//go:linkname _Cgo___GoStringN runtime.cgo_GoStringN
+func _Cgo___GoStringN(*_Cgo_char, uintptr) string
 
-func C.GoStringN(cstr *C.char, length C.int) string {
-	return C.__GoStringN(cstr, uintptr(length))
+func _Cgo_GoStringN(cstr *_Cgo_char, length _Cgo_int) string {
+	return _Cgo___GoStringN(cstr, uintptr(length))
 }
 
-//go:linkname C.__GoBytes runtime.cgo_GoBytes
-func C.__GoBytes(unsafe.Pointer, uintptr) []byte
+//go:linkname _Cgo___GoBytes runtime.cgo_GoBytes
+func _Cgo___GoBytes(unsafe.Pointer, uintptr) []byte
 
-func C.GoBytes(ptr unsafe.Pointer, length C.int) []byte {
-	return C.__GoBytes(ptr, uintptr(length))
+func _Cgo_GoBytes(ptr unsafe.Pointer, length _Cgo_int) []byte {
+	return _Cgo___GoBytes(ptr, uintptr(length))
+}
+
+//go:linkname _Cgo___CBytes runtime.cgo_CBytes
+func _Cgo___CBytes([]byte) unsafe.Pointer
+
+func _Cgo_CBytes(b []byte) unsafe.Pointer {
+	return _Cgo___CBytes(b)
+}
+
+//go:linkname _Cgo___get_errno_num runtime.cgo_errno
+func _Cgo___get_errno_num() uintptr
+
+func _Cgo___get_errno() error {
+	return syscall.Errno(_Cgo___get_errno_num())
 }
 
 type (
-	C.char      uint8
-	C.schar     int8
-	C.uchar     uint8
-	C.short     int16
-	C.ushort    uint16
-	C.int       int32
-	C.uint      uint32
-	C.long      int32
-	C.ulong     uint32
-	C.longlong  int64
-	C.ulonglong uint64
+	_Cgo_char      uint8
+	_Cgo_schar     int8
+	_Cgo_uchar     uint8
+	_Cgo_short     int16
+	_Cgo_ushort    uint16
+	_Cgo_int       int32
+	_Cgo_uint      uint32
+	_Cgo_long      int32
+	_Cgo_ulong     uint32
+	_Cgo_longlong  int64
+	_Cgo_ulonglong uint64
 )
 
-const C.foo = 3
-const C.bar = C.foo
+const _Cgo_foo = 3
+const _Cgo_bar = _Cgo_foo
+const _Cgo_unreferenced = 4
+const _Cgo_referenced = _Cgo_unreferenced
+const _Cgo_fnlike_val = 5
+const _Cgo_square_val = (20 * 20)
+const _Cgo_add_val = (3 + 5)

cgo/testdata/errors.go

@@ -10,20 +10,35 @@ typedef struct {
 
 typedef someType noType; // undefined type
 
+// Some invalid noescape lines
+#cgo noescape
+#cgo noescape foo bar
+#cgo noescape unusedFunction
+
 #define SOME_CONST_1 5) // invalid const syntax
 #define SOME_CONST_2 6) // const not used (so no error)
 #define SOME_CONST_3 1234 // const too large for byte
+#define   SOME_CONST_b      3   ) // const with lots of weird whitespace (to test error locations)
+#  define SOME_CONST_startspace 3)
 */
 //
 //
 // #define SOME_CONST_4 8) // after some empty lines
+// #cgo CFLAGS: -DSOME_PARAM_CONST_invalid=3/+3
+// #cgo CFLAGS: -DSOME_PARAM_CONST_valid=3+4
 import "C"
 
 // #warning another warning
 import "C"
 
+// #define add(a, b) (a+b)
+// #define add_toomuch add(1, 2, 3)
+// #define add_toolittle add(1)
+import "C"
+
 // Make sure that errors for the following lines won't change with future
 // additions to the CGo preamble.
+//
 //line errors.go:100
 var (
 	// constant too large
@@ -38,4 +53,15 @@ var (
 	_ byte = C.SOME_CONST_3
 
 	_ = C.SOME_CONST_4
+
+	_ = C.SOME_CONST_b
+
+	_ = C.SOME_CONST_startspace
+
+	// constants passed by a command line parameter
+	_ = C.SOME_PARAM_CONST_invalid
+	_ = C.SOME_PARAM_CONST_valid
+
+	_ = C.add_toomuch
+	_ = C.add_toolittle
 )

cgo/testdata/errors.out.go

@@ -1,60 +1,89 @@
 // CGo errors:
+//     testdata/errors.go:14:1: missing function name in #cgo noescape line
+//     testdata/errors.go:15:1: multiple function names in #cgo noescape line
 //     testdata/errors.go:4:2: warning: some warning
 //     testdata/errors.go:11:9: error: unknown type name 'someType'
-//     testdata/errors.go:22:5: warning: another warning
-//     testdata/errors.go:13:23: unexpected token ), expected end of expression
-//     testdata/errors.go:19:26: unexpected token ), expected end of expression
+//     testdata/errors.go:31:5: warning: another warning
+//     testdata/errors.go:18:23: unexpected token ), expected end of expression
+//     testdata/errors.go:26:26: unexpected token ), expected end of expression
+//     testdata/errors.go:21:33: unexpected token ), expected end of expression
+//     testdata/errors.go:22:34: unexpected token ), expected end of expression
+//     -: unexpected token INT, expected end of expression
+//     testdata/errors.go:35:35: unexpected number of parameters: expected 2, got 3
+//     testdata/errors.go:36:31: unexpected number of parameters: expected 2, got 1
+//     testdata/errors.go:3:1: function "unusedFunction" in #cgo noescape line is not used
 
 // Type checking errors after CGo processing:
-//     testdata/errors.go:102: cannot use 2 << 10 (untyped int constant 2048) as C.char value in variable declaration (overflows)
+//     testdata/errors.go:102: cannot use 2 << 10 (untyped int constant 2048) as _Cgo_char value in variable declaration (overflows)
 //     testdata/errors.go:105: unknown field z in struct literal
-//     testdata/errors.go:108: undefined: C.SOME_CONST_1
-//     testdata/errors.go:110: cannot use C.SOME_CONST_3 (untyped int constant 1234) as byte value in variable declaration (overflows)
-//     testdata/errors.go:112: undefined: C.SOME_CONST_4
+//     testdata/errors.go:108: undefined: _Cgo_SOME_CONST_1
+//     testdata/errors.go:110: cannot use _Cgo_SOME_CONST_3 (untyped int constant 1234) as byte value in variable declaration (overflows)
+//     testdata/errors.go:112: undefined: _Cgo_SOME_CONST_4
+//     testdata/errors.go:114: undefined: _Cgo_SOME_CONST_b
+//     testdata/errors.go:116: undefined: _Cgo_SOME_CONST_startspace
+//     testdata/errors.go:119: undefined: _Cgo_SOME_PARAM_CONST_invalid
+//     testdata/errors.go:122: undefined: _Cgo_add_toomuch
+//     testdata/errors.go:123: undefined: _Cgo_add_toolittle
 
 package main
 
+import "syscall"
 import "unsafe"
 
 var _ unsafe.Pointer
 
-//go:linkname C.CString runtime.cgo_CString
-func C.CString(string) *C.char
+//go:linkname _Cgo_CString runtime.cgo_CString
+func _Cgo_CString(string) *_Cgo_char
 
-//go:linkname C.GoString runtime.cgo_GoString
-func C.GoString(*C.char) string
+//go:linkname _Cgo_GoString runtime.cgo_GoString
+func _Cgo_GoString(*_Cgo_char) string
 
-//go:linkname C.__GoStringN runtime.cgo_GoStringN
-func C.__GoStringN(*C.char, uintptr) string
+//go:linkname _Cgo___GoStringN runtime.cgo_GoStringN
+func _Cgo___GoStringN(*_Cgo_char, uintptr) string
 
-func C.GoStringN(cstr *C.char, length C.int) string {
-	return C.__GoStringN(cstr, uintptr(length))
+func _Cgo_GoStringN(cstr *_Cgo_char, length _Cgo_int) string {
+	return _Cgo___GoStringN(cstr, uintptr(length))
 }
 
-//go:linkname C.__GoBytes runtime.cgo_GoBytes
-func C.__GoBytes(unsafe.Pointer, uintptr) []byte
+//go:linkname _Cgo___GoBytes runtime.cgo_GoBytes
+func _Cgo___GoBytes(unsafe.Pointer, uintptr) []byte
 
-func C.GoBytes(ptr unsafe.Pointer, length C.int) []byte {
-	return C.__GoBytes(ptr, uintptr(length))
+func _Cgo_GoBytes(ptr unsafe.Pointer, length _Cgo_int) []byte {
+	return _Cgo___GoBytes(ptr, uintptr(length))
+}
+
+//go:linkname _Cgo___CBytes runtime.cgo_CBytes
+func _Cgo___CBytes([]byte) unsafe.Pointer
+
+func _Cgo_CBytes(b []byte) unsafe.Pointer {
+	return _Cgo___CBytes(b)
+}
+
+//go:linkname _Cgo___get_errno_num runtime.cgo_errno
+func _Cgo___get_errno_num() uintptr
+
+func _Cgo___get_errno() error {
+	return syscall.Errno(_Cgo___get_errno_num())
 }
 
 type (
-	C.char      uint8
-	C.schar     int8
-	C.uchar     uint8
-	C.short     int16
-	C.ushort    uint16
-	C.int       int32
-	C.uint      uint32
-	C.long      int32
-	C.ulong     uint32
-	C.longlong  int64
-	C.ulonglong uint64
+	_Cgo_char      uint8
+	_Cgo_schar     int8
+	_Cgo_uchar     uint8
+	_Cgo_short     int16
+	_Cgo_ushort    uint16
+	_Cgo_int       int32
+	_Cgo_uint      uint32
+	_Cgo_long      int32
+	_Cgo_ulong     uint32
+	_Cgo_longlong  int64
+	_Cgo_ulonglong uint64
 )
-type C._Ctype_struct___0 struct {
-	x C.int
-	y C.int
+type _Cgo_struct_point_t struct {
+	x _Cgo_int
+	y _Cgo_int
 }
-type C.point_t = C._Ctype_struct___0
+type _Cgo_point_t = _Cgo_struct_point_t
 
-const C.SOME_CONST_3 = 1234
+const _Cgo_SOME_CONST_3 = 1234
+const _Cgo_SOME_PARAM_CONST_valid = 3 + 4

cgo/testdata/flags.out.go

@@ -5,43 +5,58 @@
 
 package main
 
+import "syscall"
 import "unsafe"
 
 var _ unsafe.Pointer
 
-//go:linkname C.CString runtime.cgo_CString
-func C.CString(string) *C.char
+//go:linkname _Cgo_CString runtime.cgo_CString
+func _Cgo_CString(string) *_Cgo_char
 
-//go:linkname C.GoString runtime.cgo_GoString
-func C.GoString(*C.char) string
+//go:linkname _Cgo_GoString runtime.cgo_GoString
+func _Cgo_GoString(*_Cgo_char) string
 
-//go:linkname C.__GoStringN runtime.cgo_GoStringN
-func C.__GoStringN(*C.char, uintptr) string
+//go:linkname _Cgo___GoStringN runtime.cgo_GoStringN
+func _Cgo___GoStringN(*_Cgo_char, uintptr) string
 
-func C.GoStringN(cstr *C.char, length C.int) string {
-	return C.__GoStringN(cstr, uintptr(length))
+func _Cgo_GoStringN(cstr *_Cgo_char, length _Cgo_int) string {
+	return _Cgo___GoStringN(cstr, uintptr(length))
 }
 
-//go:linkname C.__GoBytes runtime.cgo_GoBytes
-func C.__GoBytes(unsafe.Pointer, uintptr) []byte
+//go:linkname _Cgo___GoBytes runtime.cgo_GoBytes
+func _Cgo___GoBytes(unsafe.Pointer, uintptr) []byte
 
-func C.GoBytes(ptr unsafe.Pointer, length C.int) []byte {
-	return C.__GoBytes(ptr, uintptr(length))
+func _Cgo_GoBytes(ptr unsafe.Pointer, length _Cgo_int) []byte {
+	return _Cgo___GoBytes(ptr, uintptr(length))
+}
+
+//go:linkname _Cgo___CBytes runtime.cgo_CBytes
+func _Cgo___CBytes([]byte) unsafe.Pointer
+
+func _Cgo_CBytes(b []byte) unsafe.Pointer {
+	return _Cgo___CBytes(b)
+}
+
+//go:linkname _Cgo___get_errno_num runtime.cgo_errno
+func _Cgo___get_errno_num() uintptr
+
+func _Cgo___get_errno() error {
+	return syscall.Errno(_Cgo___get_errno_num())
 }
 
 type (
-	C.char      uint8
-	C.schar     int8
-	C.uchar     uint8
-	C.short     int16
-	C.ushort    uint16
-	C.int       int32
-	C.uint      uint32
-	C.long      int32
-	C.ulong     uint32
-	C.longlong  int64
-	C.ulonglong uint64
+	_Cgo_char      uint8
+	_Cgo_schar     int8
+	_Cgo_uchar     uint8
+	_Cgo_short     int16
+	_Cgo_ushort    uint16
+	_Cgo_int       int32
+	_Cgo_uint      uint32
+	_Cgo_long      int32
+	_Cgo_ulong     uint32
+	_Cgo_longlong  int64
+	_Cgo_ulonglong uint64
 )
 
-const C.BAR = 3
-const C.FOO_H = 1
+const _Cgo_BAR = 3
+const _Cgo_FOO_H = 1

cgo/testdata/symbols.go

@@ -9,6 +9,10 @@ static void staticfunc(int x);
 
 // Global variable signatures.
 extern int someValue;
+
+void notEscapingFunction(int *a);
+
+#cgo noescape notEscapingFunction
 */
 import "C"
 
@@ -18,6 +22,7 @@ func accessFunctions() {
 	C.variadic0()
 	C.variadic2(3, 5)
 	C.staticfunc(3)
+	C.notEscapingFunction(nil)
 }
 
 func accessGlobals() {

cgo/testdata/symbols.out.go

@@ -1,64 +1,84 @@
 package main
 
+import "syscall"
 import "unsafe"
 
 var _ unsafe.Pointer
 
-//go:linkname C.CString runtime.cgo_CString
-func C.CString(string) *C.char
+//go:linkname _Cgo_CString runtime.cgo_CString
+func _Cgo_CString(string) *_Cgo_char
 
-//go:linkname C.GoString runtime.cgo_GoString
-func C.GoString(*C.char) string
+//go:linkname _Cgo_GoString runtime.cgo_GoString
+func _Cgo_GoString(*_Cgo_char) string
 
-//go:linkname C.__GoStringN runtime.cgo_GoStringN
-func C.__GoStringN(*C.char, uintptr) string
+//go:linkname _Cgo___GoStringN runtime.cgo_GoStringN
+func _Cgo___GoStringN(*_Cgo_char, uintptr) string
 
-func C.GoStringN(cstr *C.char, length C.int) string {
-	return C.__GoStringN(cstr, uintptr(length))
+func _Cgo_GoStringN(cstr *_Cgo_char, length _Cgo_int) string {
+	return _Cgo___GoStringN(cstr, uintptr(length))
 }
 
-//go:linkname C.__GoBytes runtime.cgo_GoBytes
-func C.__GoBytes(unsafe.Pointer, uintptr) []byte
+//go:linkname _Cgo___GoBytes runtime.cgo_GoBytes
+func _Cgo___GoBytes(unsafe.Pointer, uintptr) []byte
 
-func C.GoBytes(ptr unsafe.Pointer, length C.int) []byte {
-	return C.__GoBytes(ptr, uintptr(length))
+func _Cgo_GoBytes(ptr unsafe.Pointer, length _Cgo_int) []byte {
+	return _Cgo___GoBytes(ptr, uintptr(length))
+}
+
+//go:linkname _Cgo___CBytes runtime.cgo_CBytes
+func _Cgo___CBytes([]byte) unsafe.Pointer
+
+func _Cgo_CBytes(b []byte) unsafe.Pointer {
+	return _Cgo___CBytes(b)
+}
+
+//go:linkname _Cgo___get_errno_num runtime.cgo_errno
+func _Cgo___get_errno_num() uintptr
+
+func _Cgo___get_errno() error {
+	return syscall.Errno(_Cgo___get_errno_num())
 }
 
 type (
-	C.char      uint8
-	C.schar     int8
-	C.uchar     uint8
-	C.short     int16
-	C.ushort    uint16
-	C.int       int32
-	C.uint      uint32
-	C.long      int32
-	C.ulong     uint32
-	C.longlong  int64
-	C.ulonglong uint64
+	_Cgo_char      uint8
+	_Cgo_schar     int8
+	_Cgo_uchar     uint8
+	_Cgo_short     int16
+	_Cgo_ushort    uint16
+	_Cgo_int       int32
+	_Cgo_uint      uint32
+	_Cgo_long      int32
+	_Cgo_ulong     uint32
+	_Cgo_longlong  int64
+	_Cgo_ulonglong uint64
 )
 
 //export foo
-func C.foo(a C.int, b C.int) C.int
+func _Cgo_foo(a _Cgo_int, b _Cgo_int) _Cgo_int
 
-var C.foo$funcaddr unsafe.Pointer
+var _Cgo_foo$funcaddr unsafe.Pointer
 
 //export variadic0
 //go:variadic
-func C.variadic0()
+func _Cgo_variadic0()
 
-var C.variadic0$funcaddr unsafe.Pointer
+var _Cgo_variadic0$funcaddr unsafe.Pointer
 
 //export variadic2
 //go:variadic
-func C.variadic2(x C.int, y C.int)
+func _Cgo_variadic2(x _Cgo_int, y _Cgo_int)
 
-var C.variadic2$funcaddr unsafe.Pointer
+var _Cgo_variadic2$funcaddr unsafe.Pointer
 
 //export _Cgo_static_173c95a79b6df1980521_staticfunc
-func C.staticfunc!symbols.go(x C.int)
+func _Cgo_staticfunc!symbols.go(x _Cgo_int)
 
-var C.staticfunc!symbols.go$funcaddr unsafe.Pointer
+var _Cgo_staticfunc!symbols.go$funcaddr unsafe.Pointer
 
+//export notEscapingFunction
+//go:noescape
+func _Cgo_notEscapingFunction(a *_Cgo_int)
+
+var _Cgo_notEscapingFunction$funcaddr unsafe.Pointer
 //go:extern someValue
-var C.someValue C.int
+var _Cgo_someValue _Cgo_int

cgo/testdata/types.out.go

@@ -1,149 +1,166 @@
 package main
 
+import "syscall"
 import "unsafe"
 
 var _ unsafe.Pointer
 
-//go:linkname C.CString runtime.cgo_CString
-func C.CString(string) *C.char
+//go:linkname _Cgo_CString runtime.cgo_CString
+func _Cgo_CString(string) *_Cgo_char
 
-//go:linkname C.GoString runtime.cgo_GoString
-func C.GoString(*C.char) string
+//go:linkname _Cgo_GoString runtime.cgo_GoString
+func _Cgo_GoString(*_Cgo_char) string
 
-//go:linkname C.__GoStringN runtime.cgo_GoStringN
-func C.__GoStringN(*C.char, uintptr) string
+//go:linkname _Cgo___GoStringN runtime.cgo_GoStringN
+func _Cgo___GoStringN(*_Cgo_char, uintptr) string
 
-func C.GoStringN(cstr *C.char, length C.int) string {
-	return C.__GoStringN(cstr, uintptr(length))
+func _Cgo_GoStringN(cstr *_Cgo_char, length _Cgo_int) string {
+	return _Cgo___GoStringN(cstr, uintptr(length))
 }
 
-//go:linkname C.__GoBytes runtime.cgo_GoBytes
-func C.__GoBytes(unsafe.Pointer, uintptr) []byte
+//go:linkname _Cgo___GoBytes runtime.cgo_GoBytes
+func _Cgo___GoBytes(unsafe.Pointer, uintptr) []byte
 
-func C.GoBytes(ptr unsafe.Pointer, length C.int) []byte {
-	return C.__GoBytes(ptr, uintptr(length))
+func _Cgo_GoBytes(ptr unsafe.Pointer, length _Cgo_int) []byte {
+	return _Cgo___GoBytes(ptr, uintptr(length))
+}
+
+//go:linkname _Cgo___CBytes runtime.cgo_CBytes
+func _Cgo___CBytes([]byte) unsafe.Pointer
+
+func _Cgo_CBytes(b []byte) unsafe.Pointer {
+	return _Cgo___CBytes(b)
+}
+
+//go:linkname _Cgo___get_errno_num runtime.cgo_errno
+func _Cgo___get_errno_num() uintptr
+
+func _Cgo___get_errno() error {
+	return syscall.Errno(_Cgo___get_errno_num())
 }
 
 type (
-	C.char      uint8
-	C.schar     int8
-	C.uchar     uint8
-	C.short     int16
-	C.ushort    uint16
-	C.int       int32
-	C.uint      uint32
-	C.long      int32
-	C.ulong     uint32
-	C.longlong  int64
-	C.ulonglong uint64
+	_Cgo_char      uint8
+	_Cgo_schar     int8
+	_Cgo_uchar     uint8
+	_Cgo_short     int16
+	_Cgo_ushort    uint16
+	_Cgo_int       int32
+	_Cgo_uint      uint32
+	_Cgo_long      int32
+	_Cgo_ulong     uint32
+	_Cgo_longlong  int64
+	_Cgo_ulonglong uint64
 )
-type C.myint = C.int
-type C._Ctype_struct___0 struct {
-	x C.int
-	y C.int
+type _Cgo_myint = _Cgo_int
+type _Cgo_struct_point2d_t struct {
+	x _Cgo_int
+	y _Cgo_int
 }
-type C.point2d_t = C._Ctype_struct___0
-type C.struct_point3d struct {
-	x C.int
-	y C.int
-	z C.int
+type _Cgo_point2d_t = _Cgo_struct_point2d_t
+type _Cgo_struct_point3d struct {
+	x _Cgo_int
+	y _Cgo_int
+	z _Cgo_int
 }
-type C.point3d_t = C.struct_point3d
-type C.struct_type1 struct {
-	_type   C.int
-	__type  C.int
-	___type C.int
+type _Cgo_point3d_t = _Cgo_struct_point3d
+type _Cgo_struct_type1 struct {
+	_type   _Cgo_int
+	__type  _Cgo_int
+	___type _Cgo_int
 }
-type C.struct_type2 struct{ _type C.int }
-type C._Ctype_union___1 struct{ i C.int }
-type C.union1_t = C._Ctype_union___1
-type C._Ctype_union___2 struct{ $union uint64 }
+type _Cgo_struct_type2 struct{ _type _Cgo_int }
+type _Cgo_union_union1_t struct{ i _Cgo_int }
+type _Cgo_union1_t = _Cgo_union_union1_t
+type _Cgo_union_union3_t struct{ $union uint64 }
 
-func (union *C._Ctype_union___2) unionfield_i() *C.int {
-	return (*C.int)(unsafe.Pointer(&union.$union))
+func (union *_Cgo_union_union3_t) unionfield_i() *_Cgo_int {
+	return (*_Cgo_int)(unsafe.Pointer(&union.$union))
 }
-func (union *C._Ctype_union___2) unionfield_d() *float64 {
+func (union *_Cgo_union_union3_t) unionfield_d() *float64 {
 	return (*float64)(unsafe.Pointer(&union.$union))
 }
-func (union *C._Ctype_union___2) unionfield_s() *C.short {
-	return (*C.short)(unsafe.Pointer(&union.$union))
+func (union *_Cgo_union_union3_t) unionfield_s() *_Cgo_short {
+	return (*_Cgo_short)(unsafe.Pointer(&union.$union))
 }
 
-type C.union3_t = C._Ctype_union___2
-type C.union_union2d struct{ $union [2]uint64 }
+type _Cgo_union3_t = _Cgo_union_union3_t
+type _Cgo_union_union2d struct{ $union [2]uint64 }
 
-func (union *C.union_union2d) unionfield_i() *C.int { return (*C.int)(unsafe.Pointer(&union.$union)) }
-func (union *C.union_union2d) unionfield_d() *[2]float64 {
+func (union *_Cgo_union_union2d) unionfield_i() *_Cgo_int {
+	return (*_Cgo_int)(unsafe.Pointer(&union.$union))
+}
+func (union *_Cgo_union_union2d) unionfield_d() *[2]float64 {
 	return (*[2]float64)(unsafe.Pointer(&union.$union))
 }
 
-type C.union2d_t = C.union_union2d
-type C._Ctype_union___3 struct{ arr [10]C.uchar }
-type C.unionarray_t = C._Ctype_union___3
-type C._Ctype_union___5 struct{ $union [3]uint32 }
+type _Cgo_union2d_t = _Cgo_union_union2d
+type _Cgo_union_unionarray_t struct{ arr [10]_Cgo_uchar }
+type _Cgo_unionarray_t = _Cgo_union_unionarray_t
+type _Cgo__Ctype_union___0 struct{ $union [3]uint32 }
 
-func (union *C._Ctype_union___5) unionfield_area() *C.point2d_t {
-	return (*C.point2d_t)(unsafe.Pointer(&union.$union))
+func (union *_Cgo__Ctype_union___0) unionfield_area() *_Cgo_point2d_t {
+	return (*_Cgo_point2d_t)(unsafe.Pointer(&union.$union))
 }
-func (union *C._Ctype_union___5) unionfield_solid() *C.point3d_t {
-	return (*C.point3d_t)(unsafe.Pointer(&union.$union))
+func (union *_Cgo__Ctype_union___0) unionfield_solid() *_Cgo_point3d_t {
+	return (*_Cgo_point3d_t)(unsafe.Pointer(&union.$union))
 }
 
-type C._Ctype_struct___4 struct {
-	begin C.point2d_t
-	end   C.point2d_t
-	tag   C.int
+type _Cgo_struct_struct_nested_t struct {
+	begin _Cgo_point2d_t
+	end   _Cgo_point2d_t
+	tag   _Cgo_int
 
-	coord C._Ctype_union___5
+	coord _Cgo__Ctype_union___0
 }
-type C.struct_nested_t = C._Ctype_struct___4
-type C._Ctype_union___6 struct{ $union [2]uint64 }
+type _Cgo_struct_nested_t = _Cgo_struct_struct_nested_t
+type _Cgo_union_union_nested_t struct{ $union [2]uint64 }
 
-func (union *C._Ctype_union___6) unionfield_point() *C.point3d_t {
-	return (*C.point3d_t)(unsafe.Pointer(&union.$union))
+func (union *_Cgo_union_union_nested_t) unionfield_point() *_Cgo_point3d_t {
+	return (*_Cgo_point3d_t)(unsafe.Pointer(&union.$union))
 }
-func (union *C._Ctype_union___6) unionfield_array() *C.unionarray_t {
-	return (*C.unionarray_t)(unsafe.Pointer(&union.$union))
+func (union *_Cgo_union_union_nested_t) unionfield_array() *_Cgo_unionarray_t {
+	return (*_Cgo_unionarray_t)(unsafe.Pointer(&union.$union))
 }
-func (union *C._Ctype_union___6) unionfield_thing() *C.union3_t {
-	return (*C.union3_t)(unsafe.Pointer(&union.$union))
+func (union *_Cgo_union_union_nested_t) unionfield_thing() *_Cgo_union3_t {
+	return (*_Cgo_union3_t)(unsafe.Pointer(&union.$union))
 }
 
-type C.union_nested_t = C._Ctype_union___6
-type C.enum_option = C.int
-type C.option_t = C.enum_option
-type C._Ctype_enum___7 = C.uint
-type C.option2_t = C._Ctype_enum___7
-type C._Ctype_struct___8 struct {
+type _Cgo_union_nested_t = _Cgo_union_union_nested_t
+type _Cgo_enum_option = _Cgo_int
+type _Cgo_option_t = _Cgo_enum_option
+type _Cgo_enum_option2_t = _Cgo_uint
+type _Cgo_option2_t = _Cgo_enum_option2_t
+type _Cgo_struct_types_t struct {
 	f   float32
 	d   float64
-	ptr *C.int
+	ptr *_Cgo_int
 }
-type C.types_t = C._Ctype_struct___8
-type C.myIntArray = [10]C.int
-type C._Ctype_struct___9 struct {
-	start        C.uchar
-	__bitfield_1 C.uchar
+type _Cgo_types_t = _Cgo_struct_types_t
+type _Cgo_myIntArray = [10]_Cgo_int
+type _Cgo_struct_bitfield_t struct {
+	start        _Cgo_uchar
+	__bitfield_1 _Cgo_uchar
 
-	d C.uchar
-	e C.uchar
+	d _Cgo_uchar
+	e _Cgo_uchar
 }
 
-func (s *C._Ctype_struct___9) bitfield_a() C.uchar { return s.__bitfield_1 & 0x1f }
-func (s *C._Ctype_struct___9) set_bitfield_a(value C.uchar) {
+func (s *_Cgo_struct_bitfield_t) bitfield_a() _Cgo_uchar { return s.__bitfield_1 & 0x1f }
+func (s *_Cgo_struct_bitfield_t) set_bitfield_a(value _Cgo_uchar) {
 	s.__bitfield_1 = s.__bitfield_1&^0x1f | value&0x1f<<0
 }
-func (s *C._Ctype_struct___9) bitfield_b() C.uchar {
+func (s *_Cgo_struct_bitfield_t) bitfield_b() _Cgo_uchar {
 	return s.__bitfield_1 >> 5 & 0x1
 }
-func (s *C._Ctype_struct___9) set_bitfield_b(value C.uchar) {
+func (s *_Cgo_struct_bitfield_t) set_bitfield_b(value _Cgo_uchar) {
 	s.__bitfield_1 = s.__bitfield_1&^0x20 | value&0x1<<5
 }
-func (s *C._Ctype_struct___9) bitfield_c() C.uchar {
+func (s *_Cgo_struct_bitfield_t) bitfield_c() _Cgo_uchar {
 	return s.__bitfield_1 >> 6
 }
-func (s *C._Ctype_struct___9) set_bitfield_c(value C.uchar,
+func (s *_Cgo_struct_bitfield_t) set_bitfield_c(value _Cgo_uchar,
 
 ) { s.__bitfield_1 = s.__bitfield_1&0x3f | value<<6 }
 
-type C.bitfield_t = C._Ctype_struct___9
+type _Cgo_bitfield_t = _Cgo_struct_bitfield_t

compileopts/config.go

@@ -19,7 +19,6 @@ type Config struct {
 	Options        *Options
 	Target         *TargetSpec
 	GoMinorVersion int
-	ClangHeaders   string // Clang built-in header include path
 	TestConfig     TestConfig
 }
 
@@ -34,6 +33,17 @@ func (c *Config) CPU() string {
 	return c.Target.CPU
 }
 
+// The current build mode (like the `-buildmode` command line flag).
+func (c *Config) BuildMode() string {
+	if c.Options.BuildMode != "" {
+		return c.Options.BuildMode
+	}
+	if c.Target.BuildMode != "" {
+		return c.Target.BuildMode
+	}
+	return "default"
+}
+
 // Features returns a list of features this CPU supports. For example, for a
 // RISC-V processor, that could be "+a,+c,+m". For many targets, an empty list
 // will be returned.
@@ -61,7 +71,7 @@ func (c *Config) GOOS() string {
 }
 
 // GOARCH returns the GOARCH of the target. This might not always be the actual
-// archtecture: for example, the AVR target is not supported by the Go standard
+// architecture: for example, the AVR target is not supported by the Go standard
 // library so such targets will usually pretend to be linux/arm.
 func (c *Config) GOARCH() string {
 	return c.Target.GOARCH
@@ -73,9 +83,22 @@ func (c *Config) GOARM() string {
 	return c.Options.GOARM
 }
 
+// GOMIPS will return the GOMIPS environment variable given to the compiler when
+// building a program.
+func (c *Config) GOMIPS() string {
+	return c.Options.GOMIPS
+}
+
 // BuildTags returns the complete list of build tags used during this build.
 func (c *Config) BuildTags() []string {
-	tags := append(c.Target.BuildTags, []string{"tinygo", "math_big_pure_go", "gc." + c.GC(), "scheduler." + c.Scheduler(), "serial." + c.Serial()}...)
+	tags := append([]string(nil), c.Target.BuildTags...) // copy slice (avoid a race)
+	tags = append(tags, []string{
+		"tinygo",                                     // that's the compiler
+		"purego",                                     // to get various crypto packages to work
+		"osusergo",                                   // to get os/user to work
+		"math_big_pure_go",                           // to get math/big to work
+		"gc." + c.GC(), "scheduler." + c.Scheduler(), // used inside the runtime package
+		"serial." + c.Serial()}...) // used inside the machine package
 	for i := 1; i <= c.GoMinorVersion; i++ {
 		tags = append(tags, fmt.Sprintf("go1.%d", i))
 	}
@@ -83,12 +106,6 @@ func (c *Config) BuildTags() []string {
 	return tags
 }
 
-// CgoEnabled returns true if (and only if) CGo is enabled. It is true by
-// default and false if CGO_ENABLED is set to "0".
-func (c *Config) CgoEnabled() bool {
-	return goenv.Get("CGO_ENABLED") == "1"
-}
-
 // GC returns the garbage collection strategy in use on this platform. Valid
 // values are "none", "leaking", "conservative" and "precise".
 func (c *Config) GC() string {
@@ -145,18 +162,18 @@ func (c *Config) Serial() string {
 
 // OptLevels returns the optimization level (0-2), size level (0-2), and inliner
 // threshold as used in the LLVM optimization pipeline.
-func (c *Config) OptLevels() (optLevel, sizeLevel int, inlinerThreshold uint) {
+func (c *Config) OptLevel() (level string, speedLevel, sizeLevel int) {
 	switch c.Options.Opt {
 	case "none", "0":
-		return 0, 0, 0 // -O0
+		return "O0", 0, 0
 	case "1":
-		return 1, 0, 0 // -O1
+		return "O1", 1, 0
 	case "2":
-		return 2, 0, 225 // -O2
+		return "O2", 2, 0
 	case "s":
-		return 2, 1, 225 // -Os
+		return "Os", 2, 1
 	case "z":
-		return 2, 2, 5 // -Oz, default
+		return "Oz", 2, 2 // default
 	default:
 		// This is not shown to the user: valid choices are already checked as
 		// part of Options.Verify(). It is here as a sanity check.
@@ -190,6 +207,15 @@ func (c *Config) StackSize() uint64 {
 	return c.Target.DefaultStackSize
 }
 
+// MaxStackAlloc returns the size of the maximum allocation to put on the stack vs heap.
+func (c *Config) MaxStackAlloc() uint64 {
+	if c.StackSize() > 32*1024 {
+		return 1024
+	}
+
+	return 256
+}
+
 // RP2040BootPatch returns whether the RP2040 boot patch should be applied that
 // calculates and patches in the checksum for the 2nd stage bootloader.
 func (c *Config) RP2040BootPatch() bool {
@@ -199,14 +225,26 @@ func (c *Config) RP2040BootPatch() bool {
 	return false
 }
 
+// Return a canonicalized architecture name, so we don't have to deal with arm*
+// vs thumb* vs arm64.
+func CanonicalArchName(triple string) string {
+	arch := strings.Split(triple, "-")[0]
+	if arch == "arm64" {
+		return "aarch64"
+	}
+	if strings.HasPrefix(arch, "arm") || strings.HasPrefix(arch, "thumb") {
+		return "arm"
+	}
+	if arch == "mipsel" {
+		return "mips"
+	}
+	return arch
+}
+
 // MuslArchitecture returns the architecture name as used in musl libc. It is
 // usually the same as the first part of the LLVM triple, but not always.
 func MuslArchitecture(triple string) string {
-	arch := strings.Split(triple, "-")[0]
-	if strings.HasPrefix(arch, "arm") || strings.HasPrefix(arch, "thumb") {
-		arch = "arm"
-	}
-	return arch
+	return CanonicalArchName(triple)
 }
 
 // LibcPath returns the path to the libc directory. The libc path will be either
@@ -220,6 +258,9 @@ func (c *Config) LibcPath(name string) (path string, precompiled bool) {
 	if c.ABI() != "" {
 		archname += "-" + c.ABI()
 	}
+	if c.Target.SoftFloat {
+		archname += "-softfloat"
+	}
 
 	// Try to load a precompiled library.
 	precompiledDir := filepath.Join(goenv.Get("TINYGOROOT"), "pkg", archname, name)
@@ -259,26 +300,38 @@ func (c *Config) DefaultBinaryExtension() string {
 
 // CFlags returns the flags to pass to the C compiler. This is necessary for CGo
 // preprocessing.
-func (c *Config) CFlags() []string {
+func (c *Config) CFlags(libclang bool) []string {
 	var cflags []string
 	for _, flag := range c.Target.CFlags {
 		cflags = append(cflags, strings.ReplaceAll(flag, "{root}", goenv.Get("TINYGOROOT")))
 	}
+	resourceDir := goenv.ClangResourceDir(libclang)
+	if resourceDir != "" {
+		// The resource directory contains the built-in clang headers like
+		// stdbool.h, stdint.h, float.h, etc.
+		// It is left empty if we're using an external compiler (that already
+		// knows these headers).
+		cflags = append(cflags,
+			"-resource-dir="+resourceDir,
+		)
+	}
 	switch c.Target.Libc {
 	case "darwin-libSystem":
 		root := goenv.Get("TINYGOROOT")
 		cflags = append(cflags,
-			"--sysroot="+filepath.Join(root, "lib/macos-minimal-sdk/src"),
+			"-nostdlibinc",
+			"-isystem", filepath.Join(root, "lib/macos-minimal-sdk/src/usr/include"),
 		)
 	case "picolibc":
 		root := goenv.Get("TINYGOROOT")
 		picolibcDir := filepath.Join(root, "lib", "picolibc", "newlib", "libc")
 		path, _ := c.LibcPath("picolibc")
 		cflags = append(cflags,
-			"--sysroot="+path,
-			"-isystem", filepath.Join(path, "include"), // necessary for Xtensa
+			"-nostdlibinc",
+			"-isystem", filepath.Join(path, "include"),
 			"-isystem", filepath.Join(picolibcDir, "include"),
 			"-isystem", filepath.Join(picolibcDir, "tinystdio"),
+			"-D__PICOLIBC_ERRNO_FUNCTION=__errno_location",
 		)
 	case "musl":
 		root := goenv.Get("TINYGOROOT")
@@ -288,16 +341,22 @@ func (c *Config) CFlags() []string {
 			"-nostdlibinc",
 			"-isystem", filepath.Join(path, "include"),
 			"-isystem", filepath.Join(root, "lib", "musl", "arch", arch),
+			"-isystem", filepath.Join(root, "lib", "musl", "arch", "generic"),
 			"-isystem", filepath.Join(root, "lib", "musl", "include"),
 		)
 	case "wasi-libc":
 		root := goenv.Get("TINYGOROOT")
-		cflags = append(cflags, "--sysroot="+root+"/lib/wasi-libc/sysroot")
+		cflags = append(cflags,
+			"-nostdlibinc",
+			"-isystem", root+"/lib/wasi-libc/sysroot/include")
+	case "wasmbuiltins":
+		// nothing to add (library is purely for builtins)
 	case "mingw-w64":
 		root := goenv.Get("TINYGOROOT")
 		path, _ := c.LibcPath("mingw-w64")
 		cflags = append(cflags,
-			"--sysroot="+path,
+			"-nostdlibinc",
+			"-isystem", filepath.Join(path, "include"),
 			"-isystem", filepath.Join(root, "lib", "mingw-w64", "mingw-w64-headers", "crt"),
 			"-isystem", filepath.Join(root, "lib", "mingw-w64", "mingw-w64-headers", "defaults", "include"),
 			"-D_UCRT",
@@ -349,6 +408,8 @@ func (c *Config) LDFlags() []string {
 	if c.Target.LinkerScript != "" {
 		ldflags = append(ldflags, "-T", c.Target.LinkerScript)
 	}
+	ldflags = append(ldflags, c.Options.ExtLDFlags...)
+
 	return ldflags
 }
 
@@ -416,7 +477,7 @@ func (c *Config) BinaryFormat(ext string) string {
 
 // Programmer returns the flash method and OpenOCD interface name given a
 // particular configuration. It may either be all configured in the target JSON
-// file or be modified using the -programmmer command-line option.
+// file or be modified using the -programmer command-line option.
 func (c *Config) Programmer() (method, openocdInterface string) {
 	switch c.Options.Programmer {
 	case "":
@@ -456,9 +517,6 @@ func (c *Config) OpenOCDConfiguration() (args []string, err error) {
 		return nil, fmt.Errorf("unknown OpenOCD transport: %#v", c.Target.OpenOCDTransport)
 	}
 	args = []string{"-f", "interface/" + openocdInterface + ".cfg"}
-	for _, cmd := range c.Target.OpenOCDCommands {
-		args = append(args, "-c", cmd)
-	}
 	if c.Target.OpenOCDTransport != "" {
 		transport := c.Target.OpenOCDTransport
 		if transport == "swd" {
@@ -470,6 +528,9 @@ func (c *Config) OpenOCDConfiguration() (args []string, err error) {
 		args = append(args, "-c", "transport select "+transport)
 	}
 	args = append(args, "-f", "target/"+c.Target.OpenOCDTarget+".cfg")
+	for _, cmd := range c.Target.OpenOCDCommands {
+		args = append(args, "-c", cmd)
+	}
 	return args, nil
 }
 
@@ -492,11 +553,6 @@ func (c *Config) RelocationModel() string {
 	return "static"
 }
 
-// WasmAbi returns the WASM ABI which is specified in the target JSON file.
-func (c *Config) WasmAbi() string {
-	return c.Target.WasmAbi
-}
-
 // EmulatorName is a shorthand to get the command for this emulator, something
 // like qemu-system-arm or simavr.
 func (c *Config) EmulatorName() string {

compileopts/options.go

@@ -8,10 +8,11 @@ import (
 )
 
 var (
+	validBuildModeOptions     = []string{"default", "c-shared", "wasi-legacy"}
 	validGCOptions            = []string{"none", "leaking", "conservative", "custom", "precise"}
 	validSchedulerOptions     = []string{"none", "tasks", "asyncify"}
-	validSerialOptions        = []string{"none", "uart", "usb"}
-	validPrintSizeOptions     = []string{"none", "short", "full"}
+	validSerialOptions        = []string{"none", "uart", "usb", "rtt"}
+	validPrintSizeOptions     = []string{"none", "short", "full", "html"}
 	validPanicStrategyOptions = []string{"print", "trap"}
 	validOptOptions           = []string{"none", "0", "1", "2", "s", "z"}
 )
@@ -23,7 +24,10 @@ type Options struct {
 	GOOS            string // environment variable
 	GOARCH          string // environment variable
 	GOARM           string // environment variable (only used with GOARCH=arm)
+	GOMIPS          string // environment variable (only used with GOARCH=mips and GOARCH=mipsle)
+	Directory       string // working dir, leave it unset to use the current working dir
 	Target          string
+	BuildMode       string // -buildmode flag
 	Opt             string
 	GC              string
 	PanicStrategy   string
@@ -48,15 +52,25 @@ type Options struct {
 	Programmer      string
 	OpenOCDCommands []string
 	LLVMFeatures    string
-	Directory       string
 	PrintJSON       bool
 	Monitor         bool
 	BaudRate        int
 	Timeout         time.Duration
+	WITPackage      string // pass through to wasm-tools component embed invocation
+	WITWorld        string // pass through to wasm-tools component embed -w option
+	ExtLDFlags      []string
 }
 
 // Verify performs a validation on the given options, raising an error if options are not valid.
 func (o *Options) Verify() error {
+	if o.BuildMode != "" {
+		valid := isInArray(validBuildModeOptions, o.BuildMode)
+		if !valid {
+			return fmt.Errorf(`invalid buildmode option '%s': valid values are %s`,
+				o.BuildMode,
+				strings.Join(validBuildModeOptions, ", "))
+		}
+	}
 	if o.GC != "" {
 		valid := isInArray(validGCOptions, o.GC)
 		if !valid {

compileopts/options_test.go

@@ -11,7 +11,7 @@ func TestVerifyOptions(t *testing.T) {
 
 	expectedGCError := errors.New(`invalid gc option 'incorrect': valid values are none, leaking, conservative, custom, precise`)
 	expectedSchedulerError := errors.New(`invalid scheduler option 'incorrect': valid values are none, tasks, asyncify`)
-	expectedPrintSizeError := errors.New(`invalid size option 'incorrect': valid values are none, short, full`)
+	expectedPrintSizeError := errors.New(`invalid size option 'incorrect': valid values are none, short, full, html`)
 	expectedPanicStrategyError := errors.New(`invalid panic option 'incorrect': valid values are print, trap`)
 
 	testCases := []struct {

compileopts/target.go

@@ -23,46 +23,50 @@ import (
 // https://doc.rust-lang.org/nightly/nightly-rustc/rustc_target/spec/struct.TargetOptions.html
 // https://github.com/shepmaster/rust-arduino-blink-led-no-core-with-cargo/blob/master/blink/arduino.json
 type TargetSpec struct {
-	Inherits         []string `json:"inherits"`
-	Triple           string   `json:"llvm-target"`
-	CPU              string   `json:"cpu"`
-	ABI              string   `json:"target-abi"` // rougly equivalent to -mabi= flag
-	Features         string   `json:"features"`
-	GOOS             string   `json:"goos"`
-	GOARCH           string   `json:"goarch"`
-	BuildTags        []string `json:"build-tags"`
-	GC               string   `json:"gc"`
-	Scheduler        string   `json:"scheduler"`
-	Serial           string   `json:"serial"` // which serial output to use (uart, usb, none)
-	Linker           string   `json:"linker"`
-	RTLib            string   `json:"rtlib"` // compiler runtime library (libgcc, compiler-rt)
-	Libc             string   `json:"libc"`
-	AutoStackSize    *bool    `json:"automatic-stack-size"` // Determine stack size automatically at compile time.
-	DefaultStackSize uint64   `json:"default-stack-size"`   // Default stack size if the size couldn't be determined at compile time.
-	CFlags           []string `json:"cflags"`
-	LDFlags          []string `json:"ldflags"`
-	LinkerScript     string   `json:"linkerscript"`
-	ExtraFiles       []string `json:"extra-files"`
-	RP2040BootPatch  *bool    `json:"rp2040-boot-patch"` // Patch RP2040 2nd stage bootloader checksum
-	Emulator         string   `json:"emulator"`
-	FlashCommand     string   `json:"flash-command"`
-	GDB              []string `json:"gdb"`
-	PortReset        string   `json:"flash-1200-bps-reset"`
-	SerialPort       []string `json:"serial-port"` // serial port IDs in the form "vid:pid"
-	FlashMethod      string   `json:"flash-method"`
-	FlashVolume      []string `json:"msd-volume-name"`
-	FlashFilename    string   `json:"msd-firmware-name"`
-	UF2FamilyID      string   `json:"uf2-family-id"`
-	BinaryFormat     string   `json:"binary-format"`
-	OpenOCDInterface string   `json:"openocd-interface"`
-	OpenOCDTarget    string   `json:"openocd-target"`
-	OpenOCDTransport string   `json:"openocd-transport"`
-	OpenOCDCommands  []string `json:"openocd-commands"`
-	OpenOCDVerify    *bool    `json:"openocd-verify"` // enable verify when flashing with openocd
-	JLinkDevice      string   `json:"jlink-device"`
-	CodeModel        string   `json:"code-model"`
-	RelocationModel  string   `json:"relocation-model"`
-	WasmAbi          string   `json:"wasm-abi"`
+	Inherits         []string `json:"inherits,omitempty"`
+	Triple           string   `json:"llvm-target,omitempty"`
+	CPU              string   `json:"cpu,omitempty"`
+	ABI              string   `json:"target-abi,omitempty"` // roughly equivalent to -mabi= flag
+	Features         string   `json:"features,omitempty"`
+	GOOS             string   `json:"goos,omitempty"`
+	GOARCH           string   `json:"goarch,omitempty"`
+	SoftFloat        bool     // used for non-baremetal systems (GOMIPS=softfloat etc)
+	BuildTags        []string `json:"build-tags,omitempty"`
+	BuildMode        string   `json:"buildmode,omitempty"` // default build mode (if nothing specified)
+	GC               string   `json:"gc,omitempty"`
+	Scheduler        string   `json:"scheduler,omitempty"`
+	Serial           string   `json:"serial,omitempty"` // which serial output to use (uart, usb, none)
+	Linker           string   `json:"linker,omitempty"`
+	RTLib            string   `json:"rtlib,omitempty"` // compiler runtime library (libgcc, compiler-rt)
+	Libc             string   `json:"libc,omitempty"`
+	AutoStackSize    *bool    `json:"automatic-stack-size,omitempty"` // Determine stack size automatically at compile time.
+	DefaultStackSize uint64   `json:"default-stack-size,omitempty"`   // Default stack size if the size couldn't be determined at compile time.
+	CFlags           []string `json:"cflags,omitempty"`
+	LDFlags          []string `json:"ldflags,omitempty"`
+	LinkerScript     string   `json:"linkerscript,omitempty"`
+	ExtraFiles       []string `json:"extra-files,omitempty"`
+	RP2040BootPatch  *bool    `json:"rp2040-boot-patch,omitempty"` // Patch RP2040 2nd stage bootloader checksum
+	BootPatches      []string `json:"boot-patches,omitempty"`      // Bootloader patches to be applied in the order they appear.
+	Emulator         string   `json:"emulator,omitempty"`
+	FlashCommand     string   `json:"flash-command,omitempty"`
+	GDB              []string `json:"gdb,omitempty"`
+	PortReset        string   `json:"flash-1200-bps-reset,omitempty"`
+	SerialPort       []string `json:"serial-port,omitempty"` // serial port IDs in the form "vid:pid"
+	FlashMethod      string   `json:"flash-method,omitempty"`
+	FlashVolume      []string `json:"msd-volume-name,omitempty"`
+	FlashFilename    string   `json:"msd-firmware-name,omitempty"`
+	UF2FamilyID      string   `json:"uf2-family-id,omitempty"`
+	BinaryFormat     string   `json:"binary-format,omitempty"`
+	OpenOCDInterface string   `json:"openocd-interface,omitempty"`
+	OpenOCDTarget    string   `json:"openocd-target,omitempty"`
+	OpenOCDTransport string   `json:"openocd-transport,omitempty"`
+	OpenOCDCommands  []string `json:"openocd-commands,omitempty"`
+	OpenOCDVerify    *bool    `json:"openocd-verify,omitempty"` // enable verify when flashing with openocd
+	JLinkDevice      string   `json:"jlink-device,omitempty"`
+	CodeModel        string   `json:"code-model,omitempty"`
+	RelocationModel  string   `json:"relocation-model,omitempty"`
+	WITPackage       string   `json:"wit-package,omitempty"`
+	WITWorld         string   `json:"wit-world,omitempty"`
 }
 
 // overrideProperties overrides all properties that are set in child into itself using reflection.
@@ -85,6 +89,10 @@ func (spec *TargetSpec) overrideProperties(child *TargetSpec) error {
 			if src.Uint() != 0 {
 				dst.Set(src)
 			}
+		case reflect.Bool:
+			if src.Bool() {
+				dst.Set(src)
+			}
 		case reflect.Ptr: // for pointers, copy if not nil
 			if !src.IsNil() {
 				dst.Set(src)
@@ -171,54 +179,7 @@ func (spec *TargetSpec) resolveInherits() error {
 // Load a target specification.
 func LoadTarget(options *Options) (*TargetSpec, error) {
 	if options.Target == "" {
-		// Configure based on GOOS/GOARCH environment variables (falling back to
-		// runtime.GOOS/runtime.GOARCH), and generate a LLVM target based on it.
-		var llvmarch string
-		switch options.GOARCH {
-		case "386":
-			llvmarch = "i386"
-		case "amd64":
-			llvmarch = "x86_64"
-		case "arm64":
-			llvmarch = "aarch64"
-		case "arm":
-			switch options.GOARM {
-			case "5":
-				llvmarch = "armv5"
-			case "6":
-				llvmarch = "armv6"
-			case "7":
-				llvmarch = "armv7"
-			default:
-				return nil, fmt.Errorf("invalid GOARM=%s, must be 5, 6, or 7", options.GOARM)
-			}
-		default:
-			llvmarch = options.GOARCH
-		}
-		llvmvendor := "unknown"
-		llvmos := options.GOOS
-		if llvmos == "darwin" {
-			// Use macosx* instead of darwin, otherwise darwin/arm64 will refer
-			// to iOS!
-			llvmos = "macosx10.12.0"
-			if llvmarch == "aarch64" {
-				// Looks like Apple prefers to call this architecture ARM64
-				// instead of AArch64.
-				llvmarch = "arm64"
-				llvmos = "macosx11.0.0"
-			}
-			llvmvendor = "apple"
-		}
-		// Target triples (which actually have four components, but are called
-		// triples for historical reasons) have the form:
-		//   arch-vendor-os-environment
-		target := llvmarch + "-" + llvmvendor + "-" + llvmos
-		if options.GOOS == "windows" {
-			target += "-gnu"
-		} else if options.GOARCH == "arm" {
-			target += "-gnueabihf"
-		}
-		return defaultTarget(options.GOOS, options.GOARCH, target)
+		return defaultTarget(options)
 	}
 
 	// See whether there is a target specification for this target (e.g.
@@ -242,14 +203,50 @@ func LoadTarget(options *Options) (*TargetSpec, error) {
 	return spec, nil
 }
 
-func defaultTarget(goos, goarch, triple string) (*TargetSpec, error) {
-	// No target spec available. Use the default one, useful on most systems
-	// with a regular OS.
+// GetTargetSpecs retrieves target specifications from the TINYGOROOT targets
+// directory.  Only valid target JSON files are considered, and the function
+// returns a map of target names to their respective TargetSpec.
+func GetTargetSpecs() (map[string]*TargetSpec, error) {
+	dir := filepath.Join(goenv.Get("TINYGOROOT"), "targets")
+	entries, err := os.ReadDir(dir)
+	if err != nil {
+		return nil, fmt.Errorf("could not list targets: %w", err)
+	}
+
+	maps := map[string]*TargetSpec{}
+	for _, entry := range entries {
+		entryInfo, err := entry.Info()
+		if err != nil {
+			return nil, fmt.Errorf("could not get entry info: %w", err)
+		}
+		if !entryInfo.Mode().IsRegular() || !strings.HasSuffix(entry.Name(), ".json") {
+			// Only inspect JSON files.
+			continue
+		}
+		path := filepath.Join(dir, entry.Name())
+		spec, err := LoadTarget(&Options{Target: path})
+		if err != nil {
+			return nil, fmt.Errorf("could not list target: %w", err)
+		}
+		if spec.FlashMethod == "" && spec.FlashCommand == "" && spec.Emulator == "" {
+			// This doesn't look like a regular target file, but rather like
+			// a parent target (such as targets/cortex-m.json).
+			continue
+		}
+		name := entry.Name()
+		name = name[:len(name)-5]
+		maps[name] = spec
+	}
+	return maps, nil
+}
+
+// Load a target from environment variables (which default to
+// runtime.GOOS/runtime.GOARCH).
+func defaultTarget(options *Options) (*TargetSpec, error) {
 	spec := TargetSpec{
-		Triple:           triple,
-		GOOS:             goos,
-		GOARCH:           goarch,
-		BuildTags:        []string{goos, goarch},
+		GOOS:             options.GOOS,
+		GOARCH:           options.GOARCH,
+		BuildTags:        []string{options.GOOS, options.GOARCH},
 		GC:               "precise",
 		Scheduler:        "tasks",
 		Linker:           "cc",
@@ -257,45 +254,163 @@ func defaultTarget(goos, goarch, triple string) (*TargetSpec, error) {
 		GDB:              []string{"gdb"},
 		PortReset:        "false",
 	}
-	switch goarch {
+
+	// Configure target based on GOARCH.
+	var llvmarch string
+	switch options.GOARCH {
 	case "386":
+		llvmarch = "i386"
 		spec.CPU = "pentium4"
-		spec.Features = "+cx8,+fxsr,+mmx,+sse,+sse2,+x87"
+		spec.Features = "+cmov,+cx8,+fxsr,+mmx,+sse,+sse2,+x87"
 	case "amd64":
+		llvmarch = "x86_64"
 		spec.CPU = "x86-64"
-		spec.Features = "+cx8,+fxsr,+mmx,+sse,+sse2,+x87"
+		spec.Features = "+cmov,+cx8,+fxsr,+mmx,+sse,+sse2,+x87"
 	case "arm":
 		spec.CPU = "generic"
 		spec.CFlags = append(spec.CFlags, "-fno-unwind-tables", "-fno-asynchronous-unwind-tables")
-		switch strings.Split(triple, "-")[0] {
-		case "armv5":
-			spec.Features = "+armv5t,+strict-align,-aes,-bf16,-d32,-dotprod,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-mve.fp,-neon,-sha2,-thumb-mode,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp"
-		case "armv6":
-			spec.Features = "+armv6,+dsp,+fp64,+strict-align,+vfp2,+vfp2sp,-aes,-d32,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fullfp16,-neon,-sha2,-thumb-mode,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp"
-		case "armv7":
-			spec.Features = "+armv7-a,+d32,+dsp,+fp64,+neon,+vfp2,+vfp2sp,+vfp3,+vfp3d16,+vfp3d16sp,+vfp3sp,-aes,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fullfp16,-sha2,-thumb-mode,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp"
+		subarch := strings.Split(options.GOARM, ",")
+		if len(subarch) > 2 {
+			return nil, fmt.Errorf("invalid GOARM=%s, must be of form <num>,[hardfloat|softfloat]", options.GOARM)
+		}
+		archLevel := subarch[0]
+		var fpu string
+		if len(subarch) >= 2 {
+			fpu = subarch[1]
+		} else {
+			// Pick the default fpu value: softfloat for armv5 and hardfloat
+			// above that.
+			if archLevel == "5" {
+				fpu = "softfloat"
+			} else {
+				fpu = "hardfloat"
+			}
+		}
+		switch fpu {
+		case "softfloat":
+			spec.CFlags = append(spec.CFlags, "-msoft-float")
+			spec.SoftFloat = true
+		case "hardfloat":
+			// Hardware floating point support is the default everywhere except
+			// on ARMv5 where it needs to be enabled explicitly.
+			if archLevel == "5" {
+				spec.CFlags = append(spec.CFlags, "-mfpu=vfpv2")
+			}
+		default:
+			return nil, fmt.Errorf("invalid extension GOARM=%s, must be softfloat or hardfloat", options.GOARM)
+		}
+		switch archLevel {
+		case "5":
+			llvmarch = "armv5"
+			if spec.SoftFloat {
+				spec.Features = "+armv5t,+soft-float,+strict-align,-aes,-bf16,-d32,-dotprod,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-mve,-mve.fp,-neon,-sha2,-thumb-mode,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp"
+			} else {
+				spec.Features = "+armv5t,+fp64,+strict-align,+vfp2,+vfp2sp,-aes,-d32,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fullfp16,-neon,-sha2,-thumb-mode,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp"
+			}
+		case "6":
+			llvmarch = "armv6"
+			if spec.SoftFloat {
+				spec.Features = "+armv6,+dsp,+soft-float,+strict-align,-aes,-bf16,-d32,-dotprod,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-mve,-mve.fp,-neon,-sha2,-thumb-mode,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp"
+			} else {
+				spec.Features = "+armv6,+dsp,+fp64,+strict-align,+vfp2,+vfp2sp,-aes,-d32,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fullfp16,-neon,-sha2,-thumb-mode,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp"
+			}
+		case "7":
+			llvmarch = "armv7"
+			if spec.SoftFloat {
+				spec.Features = "+armv7-a,+dsp,+soft-float,-aes,-bf16,-d32,-dotprod,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-mve,-mve.fp,-neon,-sha2,-thumb-mode,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp"
+			} else {
+				spec.Features = "+armv7-a,+d32,+dsp,+fp64,+neon,+vfp2,+vfp2sp,+vfp3,+vfp3d16,+vfp3d16sp,+vfp3sp,-aes,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fullfp16,-sha2,-thumb-mode,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp"
+			}
+		default:
+			return nil, fmt.Errorf("invalid GOARM=%s, must be of form <num>,[hardfloat|softfloat] where num is 5, 6, or 7", options.GOARM)
 		}
 	case "arm64":
 		spec.CPU = "generic"
-		spec.Features = "+neon"
+		llvmarch = "aarch64"
+		if options.GOOS == "darwin" {
+			spec.Features = "+ete,+fp-armv8,+neon,+trbe,+v8a"
+			// Looks like Apple prefers to call this architecture ARM64
+			// instead of AArch64.
+			llvmarch = "arm64"
+		} else if options.GOOS == "windows" {
+			spec.Features = "+ete,+fp-armv8,+neon,+trbe,+v8a,-fmv"
+		} else { // linux
+			spec.Features = "+ete,+fp-armv8,+neon,+trbe,+v8a,-fmv,-outline-atomics"
+		}
+	case "mips", "mipsle":
+		spec.CPU = "mips32"
+		spec.CFlags = append(spec.CFlags, "-fno-pic")
+		if options.GOARCH == "mips" {
+			llvmarch = "mips" // big endian
+		} else {
+			llvmarch = "mipsel" // little endian
+		}
+		switch options.GOMIPS {
+		case "hardfloat":
+			spec.Features = "+fpxx,+mips32,+nooddspreg,-noabicalls"
+		case "softfloat":
+			spec.SoftFloat = true
+			spec.Features = "+mips32,+soft-float,-noabicalls"
+			spec.CFlags = append(spec.CFlags, "-msoft-float")
+		default:
+			return nil, fmt.Errorf("invalid GOMIPS=%s: must be hardfloat or softfloat", options.GOMIPS)
+		}
+	case "wasm":
+		return nil, fmt.Errorf("GOARCH=wasm but GOOS is unset. Please set GOOS to wasm, wasip1, or wasip2.")
+	default:
+		return nil, fmt.Errorf("unknown GOARCH=%s", options.GOARCH)
 	}
-	if goos == "darwin" {
+
+	// Configure target based on GOOS.
+	llvmos := options.GOOS
+	llvmvendor := "unknown"
+	switch options.GOOS {
+	case "darwin":
+		platformVersion := "10.12.0"
+		if options.GOARCH == "arm64" {
+			platformVersion = "11.0.0" // first macosx platform with arm64 support
+		}
+		llvmvendor = "apple"
 		spec.Linker = "ld.lld"
 		spec.Libc = "darwin-libSystem"
-		arch := strings.Split(triple, "-")[0]
-		platformVersion := strings.TrimPrefix(strings.Split(triple, "-")[2], "macosx")
+		// Use macosx* instead of darwin, otherwise darwin/arm64 will refer to
+		// iOS!
+		llvmos = "macosx" + platformVersion
 		spec.LDFlags = append(spec.LDFlags,
 			"-flavor", "darwin",
 			"-dead_strip",
-			"-arch", arch,
+			"-arch", llvmarch,
 			"-platform_version", "macos", platformVersion, platformVersion,
 		)
-	} else if goos == "linux" {
+		spec.ExtraFiles = append(spec.ExtraFiles,
+			"src/internal/futex/futex_darwin.c",
+			"src/runtime/os_darwin.c",
+			"src/runtime/runtime_unix.c",
+			"src/runtime/signal.c")
+	case "linux":
 		spec.Linker = "ld.lld"
 		spec.RTLib = "compiler-rt"
 		spec.Libc = "musl"
 		spec.LDFlags = append(spec.LDFlags, "--gc-sections")
-	} else if goos == "windows" {
+		if options.GOARCH == "arm64" {
+			// Disable outline atomics. For details, see:
+			// https://cpufun.substack.com/p/atomics-in-aarch64
+			// A better way would be to fully support outline atomics, which
+			// makes atomics slightly more efficient on systems with many cores.
+			// But the instructions are only supported on newer aarch64 CPUs, so
+			// this feature is normally put in a system library which does
+			// feature detection for you.
+			// We take the lazy way out and simply disable this feature, instead
+			// of enabling it in compiler-rt (which is a bit more complicated).
+			// We don't really need this feature anyway as we don't even support
+			// proper threading.
+			spec.CFlags = append(spec.CFlags, "-mno-outline-atomics")
+		}
+		spec.ExtraFiles = append(spec.ExtraFiles,
+			"src/internal/futex/futex_linux.c",
+			"src/runtime/runtime_unix.c",
+			"src/runtime/signal.c")
+	case "windows":
 		spec.Linker = "ld.lld"
 		spec.Libc = "mingw-w64"
 		// Note: using a medium code model, low image base and no ASLR
@@ -304,7 +419,7 @@ func defaultTarget(goos, goarch, triple string) (*TargetSpec, error) {
 		// normally present in Go (without explicitly opting in).
 		// For more discussion:
 		// https://groups.google.com/g/Golang-nuts/c/Jd9tlNc6jUE/m/Zo-7zIP_m3MJ?pli=1
-		switch goarch {
+		switch options.GOARCH {
 		case "amd64":
 			spec.LDFlags = append(spec.LDFlags,
 				"-m", "i386pep",
@@ -321,24 +436,51 @@ func defaultTarget(goos, goarch, triple string) (*TargetSpec, error) {
 			"--no-insert-timestamp",
 			"--no-dynamicbase",
 		)
-	} else {
-		spec.LDFlags = append(spec.LDFlags, "-no-pie", "-Wl,--gc-sections") // WARNING: clang < 5.0 requires -nopie
+	case "wasm", "wasip1", "wasip2":
+		return nil, fmt.Errorf("GOOS=%s but GOARCH is unset. Please set GOARCH to wasm", options.GOOS)
+	default:
+		return nil, fmt.Errorf("unknown GOOS=%s", options.GOOS)
 	}
-	if goarch != "wasm" {
+
+	// Target triples (which actually have four components, but are called
+	// triples for historical reasons) have the form:
+	//   arch-vendor-os-environment
+	spec.Triple = llvmarch + "-" + llvmvendor + "-" + llvmos
+	if options.GOOS == "windows" {
+		spec.Triple += "-gnu"
+	} else if options.GOOS == "linux" {
+		// We use musl on Linux (not glibc) so we should use -musleabi* instead
+		// of -gnueabi*.
+		// The *hf suffix selects between soft/hard floating point ABI.
+		if spec.SoftFloat {
+			spec.Triple += "-musleabi"
+		} else {
+			spec.Triple += "-musleabihf"
+		}
+	}
+
+	// Add extra assembly files (needed for the scheduler etc).
+	if options.GOARCH != "wasm" {
 		suffix := ""
-		if goos == "windows" && goarch == "amd64" {
+		if options.GOOS == "windows" && options.GOARCH == "amd64" {
 			// Windows uses a different calling convention on amd64 from other
 			// operating systems so we need separate assembly files.
 			suffix = "_windows"
 		}
-		spec.ExtraFiles = append(spec.ExtraFiles, "src/runtime/asm_"+goarch+suffix+".S")
-		spec.ExtraFiles = append(spec.ExtraFiles, "src/internal/task/task_stack_"+goarch+suffix+".S")
+		asmGoarch := options.GOARCH
+		if options.GOARCH == "mips" || options.GOARCH == "mipsle" {
+			asmGoarch = "mipsx"
+		}
+		spec.ExtraFiles = append(spec.ExtraFiles, "src/runtime/asm_"+asmGoarch+suffix+".S")
+		spec.ExtraFiles = append(spec.ExtraFiles, "src/internal/task/task_stack_"+asmGoarch+suffix+".S")
 	}
-	if goarch != runtime.GOARCH {
+
+	// Configure the emulator.
+	if options.GOARCH != runtime.GOARCH {
 		// Some educated guesses as to how to invoke helper programs.
 		spec.GDB = []string{"gdb-multiarch"}
-		if goos == "linux" {
-			switch goarch {
+		if options.GOOS == "linux" {
+			switch options.GOARCH {
 			case "386":
 				// amd64 can _usually_ run 32-bit programs, so skip the emulator in that case.
 				if runtime.GOARCH != "amd64" {
@@ -350,14 +492,19 @@ func defaultTarget(goos, goarch, triple string) (*TargetSpec, error) {
 				spec.Emulator = "qemu-arm {}"
 			case "arm64":
 				spec.Emulator = "qemu-aarch64 {}"
+			case "mips":
+				spec.Emulator = "qemu-mips {}"
+			case "mipsle":
+				spec.Emulator = "qemu-mipsel {}"
 			}
 		}
 	}
-	if goos != runtime.GOOS {
-		if goos == "windows" {
+	if options.GOOS != runtime.GOOS {
+		if options.GOOS == "windows" {
 			spec.Emulator = "wine {}"
 		}
 	}
+
 	return &spec, nil
 }
 

compiler/alias.go

@@ -16,13 +16,18 @@ import "tinygo.org/x/go-llvm"
 
 var stdlibAliases = map[string]string{
 	// crypto packages
-	"crypto/ed25519/internal/edwards25519/field.feMul":    "crypto/ed25519/internal/edwards25519/field.feMulGeneric",
-	"crypto/ed25519/internal/edwards25519/field.feSquare": "crypto/ed25519/internal/edwards25519/field.feSquareGeneric",
-	"crypto/md5.block":         "crypto/md5.blockGeneric",
-	"crypto/sha1.block":        "crypto/sha1.blockGeneric",
-	"crypto/sha1.blockAMD64":   "crypto/sha1.blockGeneric",
-	"crypto/sha256.block":      "crypto/sha256.blockGeneric",
-	"crypto/sha512.blockAMD64": "crypto/sha512.blockGeneric",
+	"crypto/ed25519/internal/edwards25519/field.feMul": "crypto/ed25519/internal/edwards25519/field.feMulGeneric",
+	"crypto/internal/edwards25519/field.feSquare":      "crypto/ed25519/internal/edwards25519/field.feSquareGeneric",
+	"crypto/md5.block":           "crypto/md5.blockGeneric",
+	"crypto/sha1.block":          "crypto/sha1.blockGeneric",
+	"crypto/sha1.blockAMD64":     "crypto/sha1.blockGeneric",
+	"crypto/sha256.block":        "crypto/sha256.blockGeneric",
+	"crypto/sha512.blockAMD64":   "crypto/sha512.blockGeneric",
+	"internal/chacha8rand.block": "internal/chacha8rand.block_generic",
+
+	// AES
+	"crypto/aes.decryptBlockAsm": "crypto/aes.decryptBlock",
+	"crypto/aes.encryptBlockAsm": "crypto/aes.encryptBlock",
 
 	// math package
 	"math.archHypot": "math.hypot",

compiler/asserts.go

@@ -99,7 +99,7 @@ func (b *builder) createUnsafeSliceStringCheck(name string, ptr, len llvm.Value,
 	// However, in practice, it is also necessary to check that the length is
 	// not too big that a GEP wouldn't be possible without wrapping the pointer.
 	// These two checks (non-negative and not too big) can be merged into one
-	// using an unsiged greater than.
+	// using an unsigned greater than.
 
 	// Make sure the len value is at least as big as a uintptr.
 	len = b.extendInteger(len, lenType, b.uintptrType)
@@ -135,7 +135,7 @@ func (b *builder) createChanBoundsCheck(elementSize uint64, bufSize llvm.Value,
 
 	// Calculate (^uintptr(0)) >> 1, which is the max value that fits in an
 	// uintptr if uintptrs were signed.
-	maxBufSize := llvm.ConstLShr(llvm.ConstNot(llvm.ConstInt(b.uintptrType, 0, false)), llvm.ConstInt(b.uintptrType, 1, false))
+	maxBufSize := b.CreateLShr(llvm.ConstNot(llvm.ConstInt(b.uintptrType, 0, false)), llvm.ConstInt(b.uintptrType, 1, false), "")
 	if elementSize > maxBufSize.ZExtValue() {
 		b.addError(pos, fmt.Sprintf("channel element type is too big (%v bytes)", elementSize))
 		return
@@ -150,7 +150,7 @@ func (b *builder) createChanBoundsCheck(elementSize uint64, bufSize llvm.Value,
 
 	// Make sure maxBufSize has the same type as bufSize.
 	if maxBufSize.Type() != bufSize.Type() {
-		maxBufSize = llvm.ConstZExt(maxBufSize, bufSize.Type())
+		maxBufSize = b.CreateZExt(maxBufSize, bufSize.Type(), "")
 	}
 
 	// Do the check for a too large (or negative) buffer size.

compiler/atomic.go

@@ -1,9 +1,6 @@
 package compiler
 
 import (
-	"fmt"
-	"strings"
-
 	"tinygo.org/x/go-llvm"
 )
 
@@ -15,37 +12,23 @@ func (b *builder) createAtomicOp(name string) llvm.Value {
 	case "AddInt32", "AddInt64", "AddUint32", "AddUint64", "AddUintptr":
 		ptr := b.getValue(b.fn.Params[0], getPos(b.fn))
 		val := b.getValue(b.fn.Params[1], getPos(b.fn))
-		if strings.HasPrefix(b.Triple, "avr") {
-			// AtomicRMW does not work on AVR as intended:
-			// - There are some register allocation issues (fixed by https://reviews.llvm.org/D97127 which is not yet in a usable LLVM release)
-			// - The result is the new value instead of the old value
-			vType := val.Type()
-			name := fmt.Sprintf("__sync_fetch_and_add_%d", vType.IntTypeWidth()/8)
-			fn := b.mod.NamedFunction(name)
-			if fn.IsNil() {
-				fn = llvm.AddFunction(b.mod, name, llvm.FunctionType(vType, []llvm.Type{ptr.Type(), vType}, false))
-			}
-			oldVal := b.createCall(fn.GlobalValueType(), fn, []llvm.Value{ptr, val}, "")
-			// Return the new value, not the original value returned.
-			return b.CreateAdd(oldVal, val, "")
-		}
 		oldVal := b.CreateAtomicRMW(llvm.AtomicRMWBinOpAdd, ptr, val, llvm.AtomicOrderingSequentiallyConsistent, true)
 		// Return the new value, not the original value returned by atomicrmw.
 		return b.CreateAdd(oldVal, val, "")
+	case "AndInt32", "AndInt64", "AndUint32", "AndUint64", "AndUintptr":
+		ptr := b.getValue(b.fn.Params[0], getPos(b.fn))
+		val := b.getValue(b.fn.Params[1], getPos(b.fn))
+		oldVal := b.CreateAtomicRMW(llvm.AtomicRMWBinOpAnd, ptr, val, llvm.AtomicOrderingSequentiallyConsistent, true)
+		return oldVal
+	case "OrInt32", "OrInt64", "OrUint32", "OrUint64", "OrUintptr":
+		ptr := b.getValue(b.fn.Params[0], getPos(b.fn))
+		val := b.getValue(b.fn.Params[1], getPos(b.fn))
+		oldVal := b.CreateAtomicRMW(llvm.AtomicRMWBinOpOr, ptr, val, llvm.AtomicOrderingSequentiallyConsistent, true)
+		return oldVal
 	case "SwapInt32", "SwapInt64", "SwapUint32", "SwapUint64", "SwapUintptr", "SwapPointer":
 		ptr := b.getValue(b.fn.Params[0], getPos(b.fn))
 		val := b.getValue(b.fn.Params[1], getPos(b.fn))
-		isPointer := val.Type().TypeKind() == llvm.PointerTypeKind
-		if isPointer {
-			// atomicrmw only supports integers, so cast to an integer.
-			// TODO: this is fixed in LLVM 15.
-			val = b.CreatePtrToInt(val, b.uintptrType, "")
-			ptr = b.CreateBitCast(ptr, llvm.PointerType(val.Type(), 0), "")
-		}
 		oldVal := b.CreateAtomicRMW(llvm.AtomicRMWBinOpXchg, ptr, val, llvm.AtomicOrderingSequentiallyConsistent, true)
-		if isPointer {
-			oldVal = b.CreateIntToPtr(oldVal, b.i8ptrType, "")
-		}
 		return oldVal
 	case "CompareAndSwapInt32", "CompareAndSwapInt64", "CompareAndSwapUint32", "CompareAndSwapUint64", "CompareAndSwapUintptr", "CompareAndSwapPointer":
 		ptr := b.getValue(b.fn.Params[0], getPos(b.fn))
@@ -63,24 +46,6 @@ func (b *builder) createAtomicOp(name string) llvm.Value {
 	case "StoreInt32", "StoreInt64", "StoreUint32", "StoreUint64", "StoreUintptr", "StorePointer":
 		ptr := b.getValue(b.fn.Params[0], getPos(b.fn))
 		val := b.getValue(b.fn.Params[1], getPos(b.fn))
-		if strings.HasPrefix(b.Triple, "avr") {
-			// SelectionDAGBuilder is currently missing the "are unaligned atomics allowed" check for stores.
-			vType := val.Type()
-			isPointer := vType.TypeKind() == llvm.PointerTypeKind
-			if isPointer {
-				// libcalls only supports integers, so cast to an integer.
-				vType = b.uintptrType
-				val = b.CreatePtrToInt(val, vType, "")
-				ptr = b.CreateBitCast(ptr, llvm.PointerType(vType, 0), "")
-			}
-			name := fmt.Sprintf("__atomic_store_%d", vType.IntTypeWidth()/8)
-			fn := b.mod.NamedFunction(name)
-			if fn.IsNil() {
-				fn = llvm.AddFunction(b.mod, name, llvm.FunctionType(vType, []llvm.Type{ptr.Type(), vType, b.uintptrType}, false))
-			}
-			b.createCall(fn.GlobalValueType(), fn, []llvm.Value{ptr, val, llvm.ConstInt(b.uintptrType, 5, false)}, "")
-			return llvm.Value{}
-		}
 		store := b.CreateStore(val, ptr)
 		store.SetOrdering(llvm.AtomicOrderingSequentiallyConsistent)
 		store.SetAlignment(b.targetData.PrefTypeAlignment(val.Type())) // required

compiler/calls.go

@@ -19,8 +19,9 @@ const maxFieldsPerParam = 3
 // useful while declaring or defining a function.
 type paramInfo struct {
 	llvmType llvm.Type
-	name     string // name, possibly with suffixes for e.g. struct fields
-	elemSize uint64 // size of pointer element type, or 0 if this isn't a pointer
+	name     string     // name, possibly with suffixes for e.g. struct fields
+	elemSize uint64     // size of pointer element type, or 0 if this isn't a pointer
+	flags    paramFlags // extra flags for this parameter
 }
 
 // paramFlags identifies parameter attributes for flags. Most importantly, it
@@ -28,20 +29,24 @@ type paramInfo struct {
 type paramFlags uint8
 
 const (
-	// Parameter may have the deferenceable_or_null attribute. This attribute
-	// cannot be applied to unsafe.Pointer and to the data pointer of slices.
-	paramIsDeferenceableOrNull = 1 << iota
+	// Whether this is a full or partial Go parameter (int, slice, etc).
+	// The extra context parameter is not a Go parameter.
+	paramIsGoParam = 1 << iota
 )
 
 // createRuntimeCallCommon creates a runtime call. Use createRuntimeCall or
 // createRuntimeInvoke instead.
 func (b *builder) createRuntimeCallCommon(fnName string, args []llvm.Value, name string, isInvoke bool) llvm.Value {
-	fn := b.program.ImportedPackage("runtime").Members[fnName].(*ssa.Function)
+	member := b.program.ImportedPackage("runtime").Members[fnName]
+	if member == nil {
+		panic("unknown runtime call: " + fnName)
+	}
+	fn := member.(*ssa.Function)
 	fnType, llvmFn := b.getFunction(fn)
 	if llvmFn.IsNil() {
 		panic("trying to call non-existent function: " + fn.RelString(nil))
 	}
-	args = append(args, llvm.Undef(b.i8ptrType)) // unused context parameter
+	args = append(args, llvm.Undef(b.dataPtrType)) // unused context parameter
 	if isInvoke {
 		return b.createInvoke(fnType, llvmFn, args, name)
 	}
@@ -191,6 +196,7 @@ func (c *compilerContext) getParamInfo(t llvm.Type, name string, goType types.Ty
 	info := paramInfo{
 		llvmType: t,
 		name:     name,
+		flags:    paramIsGoParam,
 	}
 	if goType != nil {
 		switch underlying := goType.Underlying().(type) {

compiler/channel.go

@@ -4,7 +4,9 @@ package compiler
 // or pseudo-operations that are lowered during goroutine lowering.
 
 import (
+	"fmt"
 	"go/types"
+	"math"
 
 	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"golang.org/x/tools/go/ssa"
@@ -33,28 +35,27 @@ func (b *builder) createChanSend(instr *ssa.Send) {
 	// store value-to-send
 	valueType := b.getLLVMType(instr.X.Type())
 	isZeroSize := b.targetData.TypeAllocSize(valueType) == 0
-	var valueAlloca, valueAllocaCast, valueAllocaSize llvm.Value
+	var valueAlloca, valueAllocaSize llvm.Value
 	if isZeroSize {
-		valueAlloca = llvm.ConstNull(llvm.PointerType(valueType, 0))
-		valueAllocaCast = llvm.ConstNull(b.i8ptrType)
+		valueAlloca = llvm.ConstNull(b.dataPtrType)
 	} else {
-		valueAlloca, valueAllocaCast, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
+		valueAlloca, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
 		b.CreateStore(chanValue, valueAlloca)
 	}
 
-	// Allocate blockedlist buffer.
-	channelBlockedList := b.getLLVMRuntimeType("channelBlockedList")
-	channelBlockedListAlloca, channelBlockedListAllocaCast, channelBlockedListAllocaSize := b.createTemporaryAlloca(channelBlockedList, "chan.blockedList")
+	// Allocate buffer for the channel operation.
+	channelOp := b.getLLVMRuntimeType("channelOp")
+	channelOpAlloca, channelOpAllocaSize := b.createTemporaryAlloca(channelOp, "chan.op")
 
 	// Do the send.
-	b.createRuntimeCall("chanSend", []llvm.Value{ch, valueAllocaCast, channelBlockedListAlloca}, "")
+	b.createRuntimeCall("chanSend", []llvm.Value{ch, valueAlloca, channelOpAlloca}, "")
 
 	// End the lifetime of the allocas.
 	// This also works around a bug in CoroSplit, at least in LLVM 8:
 	// https://bugs.llvm.org/show_bug.cgi?id=41742
-	b.emitLifetimeEnd(channelBlockedListAllocaCast, channelBlockedListAllocaSize)
+	b.emitLifetimeEnd(channelOpAlloca, channelOpAllocaSize)
 	if !isZeroSize {
-		b.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
+		b.emitLifetimeEnd(valueAlloca, valueAllocaSize)
 	}
 }
 
@@ -66,28 +67,27 @@ func (b *builder) createChanRecv(unop *ssa.UnOp) llvm.Value {
 
 	// Allocate memory to receive into.
 	isZeroSize := b.targetData.TypeAllocSize(valueType) == 0
-	var valueAlloca, valueAllocaCast, valueAllocaSize llvm.Value
+	var valueAlloca, valueAllocaSize llvm.Value
 	if isZeroSize {
-		valueAlloca = llvm.ConstNull(llvm.PointerType(valueType, 0))
-		valueAllocaCast = llvm.ConstNull(b.i8ptrType)
+		valueAlloca = llvm.ConstNull(b.dataPtrType)
 	} else {
-		valueAlloca, valueAllocaCast, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
+		valueAlloca, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
 	}
 
-	// Allocate blockedlist buffer.
-	channelBlockedList := b.getLLVMRuntimeType("channelBlockedList")
-	channelBlockedListAlloca, channelBlockedListAllocaCast, channelBlockedListAllocaSize := b.createTemporaryAlloca(channelBlockedList, "chan.blockedList")
+	// Allocate buffer for the channel operation.
+	channelOp := b.getLLVMRuntimeType("channelOp")
+	channelOpAlloca, channelOpAllocaSize := b.createTemporaryAlloca(channelOp, "chan.op")
 
 	// Do the receive.
-	commaOk := b.createRuntimeCall("chanRecv", []llvm.Value{ch, valueAllocaCast, channelBlockedListAlloca}, "")
+	commaOk := b.createRuntimeCall("chanRecv", []llvm.Value{ch, valueAlloca, channelOpAlloca}, "")
 	var received llvm.Value
 	if isZeroSize {
 		received = llvm.ConstNull(valueType)
 	} else {
 		received = b.CreateLoad(valueType, valueAlloca, "chan.received")
-		b.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
+		b.emitLifetimeEnd(valueAlloca, valueAllocaSize)
 	}
-	b.emitLifetimeEnd(channelBlockedListAllocaCast, channelBlockedListAllocaSize)
+	b.emitLifetimeEnd(channelOpAlloca, channelOpAllocaSize)
 
 	if unop.CommaOk {
 		tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{valueType, b.ctx.Int1Type()}, false))
@@ -126,6 +126,20 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 		}
 	}
 
+	const maxSelectStates = math.MaxUint32 >> 2
+	if len(expr.States) > maxSelectStates {
+		// The runtime code assumes that the number of state must fit in 30 bits
+		// (so the select index can be stored in a uint32 with two bits reserved
+		// for other purposes). It seems unlikely that a real program would have
+		// that many states, but we check for this case anyway to be sure.
+		// We use a uint32 (and not a uintptr or uint64) to avoid 64-bit atomic
+		// operations which aren't available everywhere.
+		b.addError(expr.Pos(), fmt.Sprintf("too many select states: got %d but the maximum supported number is %d", len(expr.States), maxSelectStates))
+
+		// Continue as usual (we'll generate broken code but the error will
+		// prevent the compilation to complete).
+	}
+
 	// This code create a (stack-allocated) slice containing all the select
 	// cases and then calls runtime.chanSelect to perform the actual select
 	// statement.
@@ -159,8 +173,7 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 			sendValue := b.getValue(state.Send, state.Pos)
 			alloca := llvmutil.CreateEntryBlockAlloca(b.Builder, sendValue.Type(), "select.send.value")
 			b.CreateStore(sendValue, alloca)
-			ptr := b.CreateBitCast(alloca, b.i8ptrType, "")
-			selectState = b.CreateInsertValue(selectState, ptr, 1, "")
+			selectState = b.CreateInsertValue(selectState, alloca, 1, "")
 		default:
 			panic("unreachable")
 		}
@@ -168,10 +181,10 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 	}
 
 	// Create a receive buffer, where the received value will be stored.
-	recvbuf := llvm.Undef(b.i8ptrType)
+	recvbuf := llvm.Undef(b.dataPtrType)
 	if recvbufSize != 0 {
 		allocaType := llvm.ArrayType(b.ctx.Int8Type(), int(recvbufSize))
-		recvbufAlloca, _, _ := b.createTemporaryAlloca(allocaType, "select.recvbuf.alloca")
+		recvbufAlloca, _ := b.createTemporaryAlloca(allocaType, "select.recvbuf.alloca")
 		recvbufAlloca.SetAlignment(recvbufAlign)
 		recvbuf = b.CreateGEP(allocaType, recvbufAlloca, []llvm.Value{
 			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
@@ -181,7 +194,7 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 
 	// Create the states slice (allocated on the stack).
 	statesAllocaType := llvm.ArrayType(chanSelectStateType, len(selectStates))
-	statesAlloca, statesI8, statesSize := b.createTemporaryAlloca(statesAllocaType, "select.states.alloca")
+	statesAlloca, statesSize := b.createTemporaryAlloca(statesAllocaType, "select.states.alloca")
 	for i, state := range selectStates {
 		// Set each slice element to the appropriate channel.
 		gep := b.CreateGEP(statesAllocaType, statesAlloca, []llvm.Value{
@@ -201,10 +214,10 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 	if expr.Blocking {
 		// Stack-allocate operation structures.
 		// If these were simply created as a slice, they would heap-allocate.
-		chBlockAllocaType := llvm.ArrayType(b.getLLVMRuntimeType("channelBlockedList"), len(selectStates))
-		chBlockAlloca, chBlockAllocaPtr, chBlockSize := b.createTemporaryAlloca(chBlockAllocaType, "select.block.alloca")
-		chBlockLen := llvm.ConstInt(b.uintptrType, uint64(len(selectStates)), false)
-		chBlockPtr := b.CreateGEP(chBlockAllocaType, chBlockAlloca, []llvm.Value{
+		opsAllocaType := llvm.ArrayType(b.getLLVMRuntimeType("channelOp"), len(selectStates))
+		opsAlloca, opsSize := b.createTemporaryAlloca(opsAllocaType, "select.block.alloca")
+		opsLen := llvm.ConstInt(b.uintptrType, uint64(len(selectStates)), false)
+		opsPtr := b.CreateGEP(opsAllocaType, opsAlloca, []llvm.Value{
 			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
 			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
 		}, "select.block")
@@ -212,20 +225,23 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 		results = b.createRuntimeCall("chanSelect", []llvm.Value{
 			recvbuf,
 			statesPtr, statesLen, statesLen, // []chanSelectState
-			chBlockPtr, chBlockLen, chBlockLen, // []channelBlockList
+			opsPtr, opsLen, opsLen, // []channelOp
 		}, "select.result")
 
 		// Terminate the lifetime of the operation structures.
-		b.emitLifetimeEnd(chBlockAllocaPtr, chBlockSize)
+		b.emitLifetimeEnd(opsAlloca, opsSize)
 	} else {
-		results = b.createRuntimeCall("tryChanSelect", []llvm.Value{
+		opsPtr := llvm.ConstNull(b.dataPtrType)
+		opsLen := llvm.ConstInt(b.uintptrType, 0, false)
+		results = b.createRuntimeCall("chanSelect", []llvm.Value{
 			recvbuf,
 			statesPtr, statesLen, statesLen, // []chanSelectState
+			opsPtr, opsLen, opsLen, // []channelOp (nil slice)
 		}, "select.result")
 	}
 
 	// Terminate the lifetime of the states alloca.
-	b.emitLifetimeEnd(statesI8, statesSize)
+	b.emitLifetimeEnd(statesAlloca, statesSize)
 
 	// The result value does not include all the possible received values,
 	// because we can't load them in advance. Instead, the *ssa.Extract
@@ -265,7 +281,6 @@ func (b *builder) getChanSelectResult(expr *ssa.Extract) llvm.Value {
 		// it to the correct type, and dereference it.
 		recvbuf := b.selectRecvBuf[expr.Tuple.(*ssa.Select)]
 		typ := b.getLLVMType(expr.Type())
-		ptr := b.CreateBitCast(recvbuf, llvm.PointerType(typ, 0), "")
-		return b.CreateLoad(typ, ptr, "")
+		return b.CreateLoad(typ, recvbuf, "")
 	}
 }

compiler/compiler.go

@@ -17,6 +17,7 @@ import (
 
 	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"github.com/tinygo-org/tinygo/loader"
+	"github.com/tinygo-org/tinygo/src/tinygo"
 	"golang.org/x/tools/go/ssa"
 	"golang.org/x/tools/go/types/typeutil"
 	"tinygo.org/x/go-llvm"
@@ -44,6 +45,7 @@ type Config struct {
 	ABI             string
 	GOOS            string
 	GOARCH          string
+	BuildMode       string
 	CodeModel       string
 	RelocationModel string
 	SizeLevel       int
@@ -53,8 +55,10 @@ type Config struct {
 	Scheduler          string
 	AutomaticStackSize bool
 	DefaultStackSize   uint64
+	MaxStackAlloc      uint64
 	NeedsStackObjects  bool
 	Debug              bool // Whether to emit debug information in the LLVM module.
+	PanicStrategy      string
 }
 
 // compilerContext contains function-independent data that should still be
@@ -75,10 +79,9 @@ type compilerContext struct {
 	machine          llvm.TargetMachine
 	targetData       llvm.TargetData
 	intType          llvm.Type
-	i8ptrType        llvm.Type // for convenience
-	rawVoidFuncType  llvm.Type // for convenience
+	dataPtrType      llvm.Type // pointer in address space 0
+	funcPtrType      llvm.Type // pointer in function address space (1 for AVR, 0 elsewhere)
 	funcPtrAddrSpace int
-	hasTypedPointers bool // for LLVM 14 backwards compatibility
 	uintptrType      llvm.Type
 	program          *ssa.Program
 	diagnostics      []error
@@ -123,13 +126,12 @@ func newCompilerContext(moduleName string, machine llvm.TargetMachine, config *C
 	} else {
 		panic("unknown pointer size")
 	}
-	c.i8ptrType = llvm.PointerType(c.ctx.Int8Type(), 0)
+	c.dataPtrType = llvm.PointerType(c.ctx.Int8Type(), 0)
 
 	dummyFuncType := llvm.FunctionType(c.ctx.VoidType(), nil, false)
 	dummyFunc := llvm.AddFunction(c.mod, "tinygo.dummy", dummyFuncType)
 	c.funcPtrAddrSpace = dummyFunc.Type().PointerAddressSpace()
-	c.hasTypedPointers = c.i8ptrType != llvm.PointerType(c.ctx.Int16Type(), 0) // with opaque pointers, all pointers are the same type (LLVM 15+)
-	c.rawVoidFuncType = dummyFunc.Type()
+	c.funcPtrType = dummyFunc.Type()
 	dummyFunc.EraseFromParentAsFunction()
 
 	return c
@@ -242,7 +244,7 @@ func NewTargetMachine(config *Config) (llvm.TargetMachine, error) {
 }
 
 // Sizes returns a types.Sizes appropriate for the given target machine. It
-// includes the correct int size and aligment as is necessary for the Go
+// includes the correct int size and alignment as is necessary for the Go
 // typechecker.
 func Sizes(machine llvm.TargetMachine) types.Sizes {
 	targetData := machine.CreateTargetData()
@@ -417,16 +419,14 @@ func (c *compilerContext) makeLLVMType(goType types.Type) llvm.Type {
 		case types.Uintptr:
 			return c.uintptrType
 		case types.UnsafePointer:
-			return c.i8ptrType
+			return c.dataPtrType
 		default:
 			panic("unknown basic type: " + typ.String())
 		}
-	case *types.Chan:
-		return llvm.PointerType(c.getLLVMRuntimeType("channel"), 0)
+	case *types.Chan, *types.Map, *types.Pointer:
+		return c.dataPtrType // all pointers are the same
 	case *types.Interface:
 		return c.getLLVMRuntimeType("_interface")
-	case *types.Map:
-		return llvm.PointerType(c.getLLVMRuntimeType("hashmap"), 0)
 	case *types.Named:
 		if st, ok := typ.Underlying().(*types.Struct); ok {
 			// Structs are a special case. While other named types are ignored
@@ -441,21 +441,11 @@ func (c *compilerContext) makeLLVMType(goType types.Type) llvm.Type {
 			return llvmType
 		}
 		return c.getLLVMType(typ.Underlying())
-	case *types.Pointer:
-		if c.hasTypedPointers {
-			ptrTo := c.getLLVMType(typ.Elem())
-			return llvm.PointerType(ptrTo, 0)
-		}
-		return c.i8ptrType // all pointers are the same
 	case *types.Signature: // function value
 		return c.getFuncType(typ)
 	case *types.Slice:
-		ptrType := c.i8ptrType
-		if c.hasTypedPointers {
-			ptrType = llvm.PointerType(c.getLLVMType(typ.Elem()), 0)
-		}
 		members := []llvm.Type{
-			ptrType,
+			c.dataPtrType,
 			c.uintptrType, // len
 			c.uintptrType, // cap
 		}
@@ -545,8 +535,8 @@ func (c *compilerContext) createDIType(typ types.Type) llvm.Metadata {
 				Elements: []llvm.Metadata{
 					c.dibuilder.CreateMemberType(llvm.Metadata{}, llvm.DIMemberType{
 						Name:         "ptr",
-						SizeInBits:   c.targetData.TypeAllocSize(c.i8ptrType) * 8,
-						AlignInBits:  uint32(c.targetData.ABITypeAlignment(c.i8ptrType)) * 8,
+						SizeInBits:   c.targetData.TypeAllocSize(c.dataPtrType) * 8,
+						AlignInBits:  uint32(c.targetData.ABITypeAlignment(c.dataPtrType)) * 8,
 						OffsetInBits: 0,
 						Type:         c.getDIType(types.NewPointer(types.Typ[types.Byte])),
 					}),
@@ -1396,6 +1386,11 @@ func (b *builder) createFunction() {
 		b.llvmFn.SetLinkage(llvm.InternalLinkage)
 		b.createFunction()
 	}
+
+	// Create wrapper function that can be called externally.
+	if b.info.wasmExport != "" {
+		b.createWasmExport()
+	}
 }
 
 // posser is an interface that's implemented by both ssa.Value and
@@ -1548,21 +1543,18 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		src := argValues[0]
 		elems := argValues[1]
 		srcBuf := b.CreateExtractValue(src, 0, "append.srcBuf")
-		srcPtr := b.CreateBitCast(srcBuf, b.i8ptrType, "append.srcPtr")
 		srcLen := b.CreateExtractValue(src, 1, "append.srcLen")
 		srcCap := b.CreateExtractValue(src, 2, "append.srcCap")
 		elemsBuf := b.CreateExtractValue(elems, 0, "append.elemsBuf")
-		elemsPtr := b.CreateBitCast(elemsBuf, b.i8ptrType, "append.srcPtr")
 		elemsLen := b.CreateExtractValue(elems, 1, "append.elemsLen")
 		elemType := b.getLLVMType(argTypes[0].Underlying().(*types.Slice).Elem())
 		elemSize := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(elemType), false)
-		result := b.createRuntimeCall("sliceAppend", []llvm.Value{srcPtr, elemsPtr, srcLen, srcCap, elemsLen, elemSize}, "append.new")
+		result := b.createRuntimeCall("sliceAppend", []llvm.Value{srcBuf, elemsBuf, srcLen, srcCap, elemsLen, elemSize}, "append.new")
 		newPtr := b.CreateExtractValue(result, 0, "append.newPtr")
-		newBuf := b.CreateBitCast(newPtr, srcBuf.Type(), "append.newBuf")
 		newLen := b.CreateExtractValue(result, 1, "append.newLen")
 		newCap := b.CreateExtractValue(result, 2, "append.newCap")
 		newSlice := llvm.Undef(src.Type())
-		newSlice = b.CreateInsertValue(newSlice, newBuf, 0, "")
+		newSlice = b.CreateInsertValue(newSlice, newPtr, 0, "")
 		newSlice = b.CreateInsertValue(newSlice, newLen, 1, "")
 		newSlice = b.CreateInsertValue(newSlice, newCap, 2, "")
 		return newSlice, nil
@@ -1600,6 +1592,42 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		cplx = b.CreateInsertValue(cplx, r, 0, "")
 		cplx = b.CreateInsertValue(cplx, i, 1, "")
 		return cplx, nil
+	case "clear":
+		value := argValues[0]
+		switch typ := argTypes[0].Underlying().(type) {
+		case *types.Slice:
+			elementType := b.getLLVMType(typ.Elem())
+			elementSize := b.targetData.TypeAllocSize(elementType)
+			elementAlign := b.targetData.ABITypeAlignment(elementType)
+
+			// The pointer to the data to be cleared.
+			llvmBuf := b.CreateExtractValue(value, 0, "buf")
+
+			// The length (in bytes) to be cleared.
+			llvmLen := b.CreateExtractValue(value, 1, "len")
+			llvmLen = b.CreateMul(llvmLen, llvm.ConstInt(llvmLen.Type(), elementSize, false), "")
+
+			// Do the clear operation using the LLVM memset builtin.
+			// This is also correct for nil slices: in those cases, len will be
+			// 0 which means the memset call is a no-op (according to the LLVM
+			// LangRef).
+			memset := b.getMemsetFunc()
+			call := b.createCall(memset.GlobalValueType(), memset, []llvm.Value{
+				llvmBuf, // dest
+				llvm.ConstInt(b.ctx.Int8Type(), 0, false), // val
+				llvmLen, // len
+				llvm.ConstInt(b.ctx.Int1Type(), 0, false), // isVolatile
+			}, "")
+			call.AddCallSiteAttribute(1, b.ctx.CreateEnumAttribute(llvm.AttributeKindID("align"), uint64(elementAlign)))
+
+			return llvm.Value{}, nil
+		case *types.Map:
+			m := argValues[0]
+			b.createMapClear(m)
+			return llvm.Value{}, nil
+		default:
+			return llvm.Value{}, b.makeError(pos, "unsupported type in clear builtin: "+typ.String())
+		}
 	case "copy":
 		dst := argValues[0]
 		src := argValues[1]
@@ -1608,8 +1636,6 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		dstBuf := b.CreateExtractValue(dst, 0, "copy.dstArray")
 		srcBuf := b.CreateExtractValue(src, 0, "copy.srcArray")
 		elemType := b.getLLVMType(argTypes[0].Underlying().(*types.Slice).Elem())
-		dstBuf = b.CreateBitCast(dstBuf, b.i8ptrType, "copy.dstPtr")
-		srcBuf = b.CreateBitCast(srcBuf, b.i8ptrType, "copy.srcPtr")
 		elemSize := llvm.ConstInt(b.uintptrType, b.targetData.TypeAllocSize(elemType), false)
 		return b.createRuntimeCall("sliceCopy", []llvm.Value{dstBuf, srcBuf, dstLen, srcLen, elemSize}, "copy.n"), nil
 	case "delete":
@@ -1637,7 +1663,30 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 			llvmLen = b.CreateZExt(llvmLen, b.intType, "len.int")
 		}
 		return llvmLen, nil
+	case "min", "max":
+		// min and max builtins, added in Go 1.21.
+		// We can simply reuse the existing binop comparison code, which has all
+		// the edge cases figured out already.
+		tok := token.LSS
+		if callName == "max" {
+			tok = token.GTR
+		}
+		result := argValues[0]
+		typ := argTypes[0]
+		for _, arg := range argValues[1:] {
+			cmp, err := b.createBinOp(tok, typ, typ, result, arg, pos)
+			if err != nil {
+				return result, err
+			}
+			result = b.CreateSelect(cmp, result, arg, "")
+		}
+		return result, nil
+	case "panic":
+		// This is rare, but happens in "defer panic()".
+		b.createRuntimeInvoke("_panic", argValues, "")
+		return llvm.Value{}, nil
 	case "print", "println":
+		b.createRuntimeCall("printlock", nil, "")
 		for i, value := range argValues {
 			if i >= 1 && callName == "println" {
 				b.createRuntimeCall("printspace", nil, "")
@@ -1698,6 +1747,7 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		if callName == "println" {
 			b.createRuntimeCall("printnl", nil, "")
 		}
+		b.createRuntimeCall("printunlock", nil, "")
 		return llvm.Value{}, nil // print() or println() returns void
 	case "real":
 		cplx := argValues[0]
@@ -1809,9 +1859,11 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 			return b.emitSV64Call(instr.Args, getPos(instr))
 		case strings.HasPrefix(name, "(device/riscv.CSR)."):
 			return b.emitCSROperation(instr)
-		case strings.HasPrefix(name, "syscall.Syscall") || strings.HasPrefix(name, "syscall.RawSyscall"):
-			return b.createSyscall(instr)
-		case strings.HasPrefix(name, "syscall.rawSyscallNoError"):
+		case strings.HasPrefix(name, "syscall.Syscall") || strings.HasPrefix(name, "syscall.RawSyscall") || strings.HasPrefix(name, "golang.org/x/sys/unix.Syscall") || strings.HasPrefix(name, "golang.org/x/sys/unix.RawSyscall"):
+			if b.GOOS != "darwin" {
+				return b.createSyscall(instr)
+			}
+		case strings.HasPrefix(name, "syscall.rawSyscallNoError") || strings.HasPrefix(name, "golang.org/x/sys/unix.RawSyscallNoError"):
 			return b.createRawSyscallNoError(instr)
 		case name == "runtime.supportsRecover":
 			supportsRecover := uint64(0)
@@ -1819,8 +1871,19 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 				supportsRecover = 1
 			}
 			return llvm.ConstInt(b.ctx.Int1Type(), supportsRecover, false), nil
+		case name == "runtime.panicStrategy":
+			panicStrategy := map[string]uint64{
+				"print": tinygo.PanicStrategyPrint,
+				"trap":  tinygo.PanicStrategyTrap,
+			}[b.Config.PanicStrategy]
+			return llvm.ConstInt(b.ctx.Int8Type(), panicStrategy, false), nil
 		case name == "runtime/interrupt.New":
 			return b.createInterruptGlobal(instr)
+		case name == "internal/abi.FuncPCABI0":
+			retval := b.createDarwinFuncPCABI0Call(instr)
+			if !retval.IsNil() {
+				return retval, nil
+			}
 		}
 
 		calleeType, callee = b.getFunction(fn)
@@ -1831,14 +1894,13 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 		switch value := instr.Value.(type) {
 		case *ssa.Function:
 			// Regular function call. No context is necessary.
-			context = llvm.Undef(b.i8ptrType)
+			context = llvm.Undef(b.dataPtrType)
 			if info.variadic && len(fn.Params) == 0 {
 				// This matches Clang, see: https://godbolt.org/z/Gqv49xKMq
 				// Eventually we might be able to eliminate this special case
 				// entirely. For details, see:
 				// https://discourse.llvm.org/t/rfc-enabling-wstrict-prototypes-by-default-in-c/60521
 				calleeType = llvm.FunctionType(callee.GlobalValueType().ReturnType(), nil, false)
-				callee = llvm.ConstBitCast(callee, llvm.PointerType(calleeType, b.funcPtrAddrSpace))
 			}
 		case *ssa.MakeClosure:
 			// A call on a func value, but the callee is trivial to find. For
@@ -1866,13 +1928,14 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 		params = append(params, typecode)
 		callee = b.getInvokeFunction(instr)
 		calleeType = callee.GlobalValueType()
-		context = llvm.Undef(b.i8ptrType)
+		context = llvm.Undef(b.dataPtrType)
 	} else {
 		// Function pointer.
 		value := b.getValue(instr.Value, getPos(instr))
 		// This is a func value, which cannot be called directly. We have to
 		// extract the function pointer and context first from the func value.
-		calleeType, callee, context = b.decodeFuncValue(value, instr.Value.Type().Underlying().(*types.Signature))
+		callee, context = b.decodeFuncValue(value)
+		calleeType = b.getLLVMFunctionType(instr.Value.Type().Underlying().(*types.Signature))
 		b.createNilCheck(instr.Value, callee, "fpcall")
 	}
 
@@ -1905,7 +1968,7 @@ func (b *builder) getValue(expr ssa.Value, pos token.Pos) llvm.Value {
 			return llvm.Undef(b.getLLVMType(expr.Type()))
 		}
 		_, fn := b.getFunction(expr)
-		return b.createFuncValue(fn, llvm.Undef(b.i8ptrType), expr.Signature)
+		return b.createFuncValue(fn, llvm.Undef(b.dataPtrType), expr.Signature)
 	case *ssa.Global:
 		value := b.getGlobal(expr)
 		if value.IsNil() {
@@ -1919,7 +1982,7 @@ func (b *builder) getValue(expr ssa.Value, pos token.Pos) llvm.Value {
 			return value
 		} else {
 			// indicates a compiler bug
-			panic("local has not been parsed: " + expr.String())
+			panic("SSA value not previously found in function: " + expr.String())
 		}
 	}
 }
@@ -1962,8 +2025,8 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 	case *ssa.Alloc:
 		typ := b.getLLVMType(expr.Type().Underlying().(*types.Pointer).Elem())
 		size := b.targetData.TypeAllocSize(typ)
-		// Move all "large" allocations to the heap.  This value is also transform.maxStackAlloc.
-		if expr.Heap || size > 256 {
+		// Move all "large" allocations to the heap.
+		if expr.Heap || size > b.MaxStackAlloc {
 			// Calculate ^uintptr(0)
 			maxSize := llvm.ConstNot(llvm.ConstInt(b.uintptrType, 0, false)).ZExtValue()
 			if size > maxSize {
@@ -1973,7 +2036,8 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 			sizeValue := llvm.ConstInt(b.uintptrType, size, false)
 			layoutValue := b.createObjectLayout(typ, expr.Pos())
 			buf := b.createRuntimeCall("alloc", []llvm.Value{sizeValue, layoutValue}, expr.Comment)
-			buf = b.CreateBitCast(buf, llvm.PointerType(typ, 0), "")
+			align := b.targetData.ABITypeAlignment(typ)
+			buf.AddCallSiteAttribute(0, b.ctx.CreateEnumAttribute(llvm.AttributeKindID("align"), uint64(align)))
 			return buf, nil
 		} else {
 			buf := llvmutil.CreateEntryBlockAlloca(b.Builder, typ, expr.Comment)
@@ -2019,10 +2083,6 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 				value = b.CreateInsertValue(value, field, i, "changetype.struct")
 			}
 			return value, nil
-		case llvm.PointerTypeKind:
-			// This can happen with pointers to structs. This case is easy:
-			// simply bitcast the pointer to the destination type.
-			return b.CreateBitCast(x, llvmType, "changetype.pointer"), nil
 		default:
 			return llvm.Value{}, errors.New("todo: unknown ChangeType type: " + expr.X.Type().String())
 		}
@@ -2099,12 +2159,12 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 			// Can't load directly from array (as index is non-constant), so
 			// have to do it using an alloca+gep+load.
 			arrayType := collection.Type()
-			alloca, allocaPtr, allocaSize := b.createTemporaryAlloca(arrayType, "index.alloca")
+			alloca, allocaSize := b.createTemporaryAlloca(arrayType, "index.alloca")
 			b.CreateStore(collection, alloca)
 			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
 			ptr := b.CreateInBoundsGEP(arrayType, alloca, []llvm.Value{zero, index}, "index.gep")
 			result := b.CreateLoad(arrayType.ElementType(), ptr, "index.load")
-			b.emitLifetimeEnd(allocaPtr, allocaSize)
+			b.emitLifetimeEnd(alloca, allocaSize)
 			return result, nil
 		default:
 			panic("unknown *ssa.Index type")
@@ -2142,7 +2202,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 			return llvm.Value{}, b.makeError(expr.Pos(), "todo: indexaddr: "+ptrTyp.String())
 		}
 
-		// Make sure index is at least the size of uintptr becuase getelementptr
+		// Make sure index is at least the size of uintptr because getelementptr
 		// assumes index is a signed integer.
 		index = b.extendInteger(index, expr.Index.Type(), b.uintptrType)
 
@@ -2184,6 +2244,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		sliceType := expr.Type().Underlying().(*types.Slice)
 		llvmElemType := b.getLLVMType(sliceType.Elem())
 		elemSize := b.targetData.TypeAllocSize(llvmElemType)
+		elemAlign := b.targetData.ABITypeAlignment(llvmElemType)
 		elemSizeValue := llvm.ConstInt(b.uintptrType, elemSize, false)
 
 		maxSize := b.maxSliceSize(llvmElemType)
@@ -2207,7 +2268,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		sliceSize := b.CreateBinOp(llvm.Mul, elemSizeValue, sliceCapCast, "makeslice.cap")
 		layoutValue := b.createObjectLayout(llvmElemType, expr.Pos())
 		slicePtr := b.createRuntimeCall("alloc", []llvm.Value{sliceSize, layoutValue}, "makeslice.buf")
-		slicePtr = b.CreateBitCast(slicePtr, llvm.PointerType(llvmElemType, 0), "makeslice.array")
+		slicePtr.AddCallSiteAttribute(0, b.ctx.CreateEnumAttribute(llvm.AttributeKindID("align"), uint64(elemAlign)))
 
 		// Extend or truncate if necessary. This is safe as we've already done
 		// the bounds check.
@@ -2253,7 +2314,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		default:
 			panic("unknown type in range: " + typ.String())
 		}
-		it, _, _ := b.createTemporaryAlloca(iteratorType, "range.it")
+		it, _ := b.createTemporaryAlloca(iteratorType, "range.it")
 		b.CreateStore(llvm.ConstNull(iteratorType), it)
 		return it, nil
 	case *ssa.Select:
@@ -2421,7 +2482,6 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		arrayLen := expr.Type().Underlying().(*types.Pointer).Elem().Underlying().(*types.Array).Len()
 		b.createSliceToArrayPointerCheck(sliceLen, arrayLen)
 		ptr := b.CreateExtractValue(slice, 0, "")
-		ptr = b.CreateBitCast(ptr, b.getLLVMType(expr.Type()), "")
 		return ptr, nil
 	case *ssa.TypeAssert:
 		return b.createTypeAssert(expr), nil
@@ -2520,7 +2580,7 @@ func (b *builder) createBinOp(op token.Token, typ, ytyp types.Type, x, y llvm.Va
 				sizeY := b.targetData.TypeAllocSize(y.Type())
 
 				// Check if the shift is bigger than the bit-width of the shifted value.
-				// This is UB in LLVM, so it needs to be handled seperately.
+				// This is UB in LLVM, so it needs to be handled separately.
 				// The Go spec indirectly defines the result as 0.
 				// Negative shifts are handled earlier, so we can treat y as unsigned.
 				overshifted := b.CreateICmp(llvm.IntUGE, y, llvm.ConstInt(y.Type(), 8*sizeX, false), "shift.overflow")
@@ -2887,16 +2947,16 @@ func (c *compilerContext) createConst(expr *ssa.Const, pos token.Pos) llvm.Value
 				zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 				strPtr = llvm.ConstInBoundsGEP(globalType, global, []llvm.Value{zero, zero})
 			} else {
-				strPtr = llvm.ConstNull(c.i8ptrType)
+				strPtr = llvm.ConstNull(c.dataPtrType)
 			}
 			strObj := llvm.ConstNamedStruct(c.getLLVMRuntimeType("_string"), []llvm.Value{strPtr, strLen})
 			return strObj
 		} else if typ.Kind() == types.UnsafePointer {
 			if !expr.IsNil() {
 				value, _ := constant.Uint64Val(constant.ToInt(expr.Value))
-				return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, value, false), c.i8ptrType)
+				return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, value, false), c.dataPtrType)
 			}
-			return llvm.ConstNull(c.i8ptrType)
+			return llvm.ConstNull(c.dataPtrType)
 		} else if typ.Info()&types.IsUnsigned != 0 {
 			n, _ := constant.Uint64Val(constant.ToInt(expr.Value))
 			return llvm.ConstInt(llvmType, n, false)
@@ -2940,7 +3000,7 @@ func (c *compilerContext) createConst(expr *ssa.Const, pos token.Pos) llvm.Value
 		// Create a generic nil interface with no dynamic type (typecode=0).
 		fields := []llvm.Value{
 			llvm.ConstInt(c.uintptrType, 0, false),
-			llvm.ConstPointerNull(c.i8ptrType),
+			llvm.ConstPointerNull(c.dataPtrType),
 		}
 		return llvm.ConstNamedStruct(c.getLLVMRuntimeType("_interface"), fields)
 	case *types.Pointer:
@@ -2948,12 +3008,16 @@ func (c *compilerContext) createConst(expr *ssa.Const, pos token.Pos) llvm.Value
 			panic("expected nil pointer constant")
 		}
 		return llvm.ConstPointerNull(c.getLLVMType(typ))
+	case *types.Array:
+		if expr.Value != nil {
+			panic("expected nil array constant")
+		}
+		return llvm.ConstNull(c.getLLVMType(expr.Type()))
 	case *types.Slice:
 		if expr.Value != nil {
 			panic("expected nil slice constant")
 		}
-		elemType := c.getLLVMType(typ.Elem())
-		llvmPtr := llvm.ConstPointerNull(llvm.PointerType(elemType, 0))
+		llvmPtr := llvm.ConstPointerNull(c.dataPtrType)
 		llvmLen := llvm.ConstInt(c.uintptrType, 0, false)
 		slice := c.ctx.ConstStruct([]llvm.Value{
 			llvmPtr, // backing array
@@ -2961,6 +3025,11 @@ func (c *compilerContext) createConst(expr *ssa.Const, pos token.Pos) llvm.Value
 			llvmLen, // cap
 		}, false)
 		return slice
+	case *types.Struct:
+		if expr.Value != nil {
+			panic("expected nil struct constant")
+		}
+		return llvm.ConstNull(c.getLLVMType(expr.Type()))
 	case *types.Map:
 		if !expr.IsNil() {
 			// I believe this is not allowed by the Go spec.
@@ -2989,7 +3058,7 @@ func (b *builder) createConvert(typeFrom, typeTo types.Type, value llvm.Value, p
 
 	// Conversion between pointers and unsafe.Pointer.
 	if isPtrFrom && isPtrTo {
-		return b.CreateBitCast(value, llvmTypeTo, ""), nil
+		return value, nil
 	}
 
 	switch typeTo := typeTo.Underlying().(type) {
@@ -3224,11 +3293,15 @@ func (b *builder) createUnOp(unop *ssa.UnOp) (llvm.Value, error) {
 			// Instead of a load from the global, create a bitcast of the
 			// function pointer itself.
 			name := strings.TrimSuffix(unop.X.(*ssa.Global).Name(), "$funcaddr")
-			_, fn := b.getFunction(b.fn.Pkg.Members[name].(*ssa.Function))
+			pkg := b.fn.Pkg
+			if pkg == nil {
+				pkg = b.fn.Origin().Pkg
+			}
+			_, fn := b.getFunction(pkg.Members[name].(*ssa.Function))
 			if fn.IsNil() {
 				return llvm.Value{}, b.makeError(unop.Pos(), "cgo function not found: "+name)
 			}
-			return b.CreateBitCast(fn, b.i8ptrType, ""), nil
+			return fn, nil
 		} else {
 			b.createNilCheck(unop.X, x, "deref")
 			load := b.CreateLoad(valueType, x, "")

compiler/compiler_test.go

@@ -29,7 +29,7 @@ func TestCompiler(t *testing.T) {
 	t.Parallel()
 
 	// Determine Go minor version (e.g. 16 in go1.16.3).
-	_, goMinor, err := goenv.GetGorootVersion(goenv.Get("GOROOT"))
+	_, goMinor, err := goenv.GetGorootVersion()
 	if err != nil {
 		t.Fatal("could not read Go version:", err)
 	}
@@ -54,6 +54,9 @@ func TestCompiler(t *testing.T) {
 	if goMinor >= 20 {
 		tests = append(tests, testCase{"go1.20.go", "", ""})
 	}
+	if goMinor >= 21 {
+		tests = append(tests, testCase{"go1.21.go", "", ""})
+	}
 
 	for _, tc := range tests {
 		name := tc.file
@@ -88,14 +91,12 @@ func TestCompiler(t *testing.T) {
 			}
 
 			// Optimize IR a little.
-			funcPasses := llvm.NewFunctionPassManagerForModule(mod)
-			defer funcPasses.Dispose()
-			funcPasses.AddInstructionCombiningPass()
-			funcPasses.InitializeFunc()
-			for fn := mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-				funcPasses.RunFunc(fn)
+			passOptions := llvm.NewPassBuilderOptions()
+			defer passOptions.Dispose()
+			err = mod.RunPasses("instcombine", llvm.TargetMachine{}, passOptions)
+			if err != nil {
+				t.Error(err)
 			}
-			funcPasses.FinalizeFunc()
 
 			outFilePrefix := tc.file[:len(tc.file)-3]
 			if tc.target != "" {
@@ -166,9 +167,9 @@ func filterIrrelevantIRLines(lines []string) []string {
 		if strings.HasPrefix(line, "source_filename = ") {
 			continue
 		}
-		if llvmVersion < 14 && strings.HasPrefix(line, "target datalayout = ") {
+		if llvmVersion < 15 && strings.HasPrefix(line, "target datalayout = ") {
 			// The datalayout string may vary betewen LLVM versions.
-			// Right now test outputs are for LLVM 14 and higher.
+			// Right now test outputs are for LLVM 15 and higher.
 			continue
 		}
 		out = append(out, line)
@@ -242,7 +243,7 @@ func testCompilePackage(t *testing.T, options *compileopts.Options, file string)
 	defer machine.Dispose()
 
 	// Load entire program AST into memory.
-	lprogram, err := loader.Load(config, "./testdata/"+file, config.ClangHeaders, types.Config{
+	lprogram, err := loader.Load(config, "./testdata/"+file, types.Config{
 		Sizes: Sizes(machine),
 	})
 	if err != nil {

compiler/defer.go

@@ -16,6 +16,7 @@ package compiler
 import (
 	"go/types"
 	"strconv"
+	"strings"
 
 	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"golang.org/x/tools/go/ssa"
@@ -60,9 +61,8 @@ func (b *builder) deferInitFunc() {
 	b.deferBuiltinFuncs = make(map[ssa.Value]deferBuiltin)
 
 	// Create defer list pointer.
-	deferType := llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)
-	b.deferPtr = b.CreateAlloca(deferType, "deferPtr")
-	b.CreateStore(llvm.ConstPointerNull(deferType), b.deferPtr)
+	b.deferPtr = b.CreateAlloca(b.dataPtrType, "deferPtr")
+	b.CreateStore(llvm.ConstPointerNull(b.dataPtrType), b.deferPtr)
 
 	if b.hasDeferFrame() {
 		// Set up the defer frame with the current stack pointer.
@@ -161,7 +161,7 @@ str x2, [x1, #8]
 mov x0, #0
 1:
 `
-		constraints = "={x0},{x1},~{x1},~{x2},~{x3},~{x4},~{x5},~{x6},~{x7},~{x8},~{x9},~{x10},~{x11},~{x12},~{x13},~{x14},~{x15},~{x16},~{x17},~{x19},~{x20},~{x21},~{x22},~{x23},~{x24},~{x25},~{x26},~{x27},~{x28},~{lr},~{q0},~{q1},~{q2},~{q3},~{q4},~{q5},~{q6},~{q7},~{q8},~{q9},~{q10},~{q11},~{q12},~{q13},~{q14},~{q15},~{q16},~{q17},~{q18},~{q19},~{q20},~{q21},~{q22},~{q23},~{q24},~{q25},~{q26},~{q27},~{q28},~{q29},~{q30},~{nzcv},~{ffr},~{vg},~{memory}"
+		constraints = "={x0},{x1},~{x1},~{x2},~{x3},~{x4},~{x5},~{x6},~{x7},~{x8},~{x9},~{x10},~{x11},~{x12},~{x13},~{x14},~{x15},~{x16},~{x17},~{x19},~{x20},~{x21},~{x22},~{x23},~{x24},~{x25},~{x26},~{x27},~{x28},~{lr},~{q0},~{q1},~{q2},~{q3},~{q4},~{q5},~{q6},~{q7},~{q8},~{q9},~{q10},~{q11},~{q12},~{q13},~{q14},~{q15},~{q16},~{q17},~{q18},~{q19},~{q20},~{q21},~{q22},~{q23},~{q24},~{q25},~{q26},~{q27},~{q28},~{q29},~{q30},~{nzcv},~{ffr},~{memory}"
 		if b.GOOS != "darwin" && b.GOOS != "windows" {
 			// These registers cause the following warning when compiling for
 			// MacOS and Windows:
@@ -188,6 +188,24 @@ std z+5, r29
 ldi r24, 0
 1:`
 		constraints = "={r24},z,~{r0},~{r2},~{r3},~{r4},~{r5},~{r6},~{r7},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15},~{r16},~{r17},~{r18},~{r19},~{r20},~{r21},~{r22},~{r23},~{r25},~{r26},~{r27}"
+	case "mips":
+		// $4 flag (zero or non-zero)
+		// $5 defer frame
+		asmString = `
+.set noat
+move $$4, $$zero
+jal 1f
+1:
+addiu $$ra, 8
+sw $$ra, 4($$5)
+.set at`
+		constraints = "={$4},{$5},~{$1},~{$2},~{$3},~{$5},~{$6},~{$7},~{$8},~{$9},~{$10},~{$11},~{$12},~{$13},~{$14},~{$15},~{$16},~{$17},~{$18},~{$19},~{$20},~{$21},~{$22},~{$23},~{$24},~{$25},~{$26},~{$27},~{$28},~{$29},~{$30},~{$31},~{memory}"
+		if !strings.Contains(b.Features, "+soft-float") {
+			// Using floating point registers together with GOMIPS=softfloat
+			// results in a crash: "This value type is not natively supported!"
+			// So only add them when using hardfloat.
+			constraints += ",~{$f0},~{$f1},~{$f2},~{$f3},~{$f4},~{$f5},~{$f6},~{$f7},~{$f8},~{$f9},~{$f10},~{$f11},~{$f12},~{$f13},~{$f14},~{$f15},~{$f16},~{$f17},~{$f18},~{$f19},~{$f20},~{$f21},~{$f22},~{$f23},~{$f24},~{$f25},~{$f26},~{$f27},~{$f28},~{$f29},~{$f30},~{$f31}"
+		}
 	case "riscv32":
 		asmString = `
 la a2, 1f
@@ -249,8 +267,7 @@ func isInLoop(start *ssa.BasicBlock) bool {
 func (b *builder) createDefer(instr *ssa.Defer) {
 	// The pointer to the previous defer struct, which we will replace to
 	// make a linked list.
-	deferType := llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)
-	next := b.CreateLoad(deferType, b.deferPtr, "defer.next")
+	next := b.CreateLoad(b.dataPtrType, b.deferPtr, "defer.next")
 
 	var values []llvm.Value
 	valueTypes := []llvm.Type{b.uintptrType, next.Type()}
@@ -271,7 +288,7 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 		typecode := b.CreateExtractValue(itf, 0, "invoke.func.typecode")
 		receiverValue := b.CreateExtractValue(itf, 1, "invoke.func.receiver")
 		values = []llvm.Value{callback, next, typecode, receiverValue}
-		valueTypes = append(valueTypes, b.i8ptrType, b.i8ptrType)
+		valueTypes = append(valueTypes, b.dataPtrType, b.dataPtrType)
 		for _, arg := range instr.Call.Args {
 			val := b.getValue(arg, getPos(instr))
 			values = append(values, val)
@@ -391,9 +408,8 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 		// This may be hit a variable number of times, so use a heap allocation.
 		size := b.targetData.TypeAllocSize(deferredCallType)
 		sizeValue := llvm.ConstInt(b.uintptrType, size, false)
-		nilPtr := llvm.ConstNull(b.i8ptrType)
-		allocCall := b.createRuntimeCall("alloc", []llvm.Value{sizeValue, nilPtr}, "defer.alloc.call")
-		alloca = b.CreateBitCast(allocCall, llvm.PointerType(deferredCallType, 0), "defer.alloc")
+		nilPtr := llvm.ConstNull(b.dataPtrType)
+		alloca = b.createRuntimeCall("alloc", []llvm.Value{sizeValue, nilPtr}, "defer.alloc.call")
 	}
 	if b.NeedsStackObjects {
 		b.trackPointer(alloca)
@@ -401,14 +417,12 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 	b.CreateStore(deferredCall, alloca)
 
 	// Push it on top of the linked list by replacing deferPtr.
-	allocaCast := b.CreateBitCast(alloca, next.Type(), "defer.alloca.cast")
-	b.CreateStore(allocaCast, b.deferPtr)
+	b.CreateStore(alloca, b.deferPtr)
 }
 
 // createRunDefers emits code to run all deferred functions.
 func (b *builder) createRunDefers() {
 	deferType := b.getLLVMRuntimeType("_defer")
-	deferPtrType := llvm.PointerType(deferType, 0)
 
 	// Add a loop like the following:
 	//     for stack != nil {
@@ -435,7 +449,7 @@ func (b *builder) createRunDefers() {
 	// Create loop head:
 	//     for stack != nil {
 	b.SetInsertPointAtEnd(loophead)
-	deferData := b.CreateLoad(deferPtrType, b.deferPtr, "")
+	deferData := b.CreateLoad(b.dataPtrType, b.deferPtr, "")
 	stackIsNil := b.CreateICmp(llvm.IntEQ, deferData, llvm.ConstPointerNull(deferData.Type()), "stackIsNil")
 	b.CreateCondBr(stackIsNil, end, loop)
 
@@ -448,7 +462,7 @@ func (b *builder) createRunDefers() {
 		llvm.ConstInt(b.ctx.Int32Type(), 0, false),
 		llvm.ConstInt(b.ctx.Int32Type(), 1, false), // .next field
 	}, "stack.next.gep")
-	nextStack := b.CreateLoad(deferPtrType, nextStackGEP, "stack.next")
+	nextStack := b.CreateLoad(b.dataPtrType, nextStackGEP, "stack.next")
 	b.CreateStore(nextStack, b.deferPtr)
 	gep := b.CreateInBoundsGEP(deferType, deferData, []llvm.Value{
 		llvm.ConstInt(b.ctx.Int32Type(), 0, false),
@@ -469,28 +483,26 @@ func (b *builder) createRunDefers() {
 			// Call on an value or interface value.
 
 			// Get the real defer struct type and cast to it.
-			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
+			valueTypes := []llvm.Type{b.uintptrType, b.dataPtrType}
 
 			if !callback.IsInvoke() {
 				//Expect funcValue to be passed through the deferred call.
 				valueTypes = append(valueTypes, b.getFuncType(callback.Signature()))
 			} else {
 				//Expect typecode
-				valueTypes = append(valueTypes, b.i8ptrType, b.i8ptrType)
+				valueTypes = append(valueTypes, b.dataPtrType, b.dataPtrType)
 			}
 
 			for _, arg := range callback.Args {
 				valueTypes = append(valueTypes, b.getLLVMType(arg.Type()))
 			}
 
-			deferredCallType := b.ctx.StructType(valueTypes, false)
-			deferredCallPtr := b.CreateBitCast(deferData, llvm.PointerType(deferredCallType, 0), "defercall")
-
 			// Extract the params from the struct (including receiver).
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+			deferredCallType := b.ctx.StructType(valueTypes, false)
 			for i := 2; i < len(valueTypes); i++ {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferredCallPtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "gep")
+				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "gep")
 				forwardParam := b.CreateLoad(valueTypes[i], gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -505,7 +517,8 @@ func (b *builder) createRunDefers() {
 
 				//Get function pointer and context
 				var context llvm.Value
-				fnType, fnPtr, context = b.decodeFuncValue(funcValue, callback.Signature())
+				fnPtr, context = b.decodeFuncValue(funcValue)
+				fnType = b.getLLVMFunctionType(callback.Signature())
 
 				//Pass context
 				forwardParams = append(forwardParams, context)
@@ -519,7 +532,7 @@ func (b *builder) createRunDefers() {
 				// Add the context parameter. An interface call cannot also be a
 				// closure but we have to supply the parameter anyway for platforms
 				// with a strict calling convention.
-				forwardParams = append(forwardParams, llvm.Undef(b.i8ptrType))
+				forwardParams = append(forwardParams, llvm.Undef(b.dataPtrType))
 			}
 
 			b.createCall(fnType, fnPtr, forwardParams, "")
@@ -528,18 +541,17 @@ func (b *builder) createRunDefers() {
 			// Direct call.
 
 			// Get the real defer struct type and cast to it.
-			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
+			valueTypes := []llvm.Type{b.uintptrType, b.dataPtrType}
 			for _, param := range getParams(callback.Signature) {
 				valueTypes = append(valueTypes, b.getLLVMType(param.Type()))
 			}
 			deferredCallType := b.ctx.StructType(valueTypes, false)
-			deferredCallPtr := b.CreateBitCast(deferData, llvm.PointerType(deferredCallType, 0), "defercall")
 
 			// Extract the params from the struct.
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
 			for i := range getParams(callback.Signature) {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferredCallPtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
+				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
 				forwardParam := b.CreateLoad(valueTypes[i+2], gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -549,7 +561,7 @@ func (b *builder) createRunDefers() {
 			if !b.getFunctionInfo(callback).exported {
 				// Add the context parameter. We know it is ignored by the receiving
 				// function, but we have to pass one anyway.
-				forwardParams = append(forwardParams, llvm.Undef(b.i8ptrType))
+				forwardParams = append(forwardParams, llvm.Undef(b.dataPtrType))
 			}
 
 			// Call real function.
@@ -559,20 +571,19 @@ func (b *builder) createRunDefers() {
 		case *ssa.MakeClosure:
 			// Get the real defer struct type and cast to it.
 			fn := callback.Fn.(*ssa.Function)
-			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
+			valueTypes := []llvm.Type{b.uintptrType, b.dataPtrType}
 			params := fn.Signature.Params()
 			for i := 0; i < params.Len(); i++ {
 				valueTypes = append(valueTypes, b.getLLVMType(params.At(i).Type()))
 			}
-			valueTypes = append(valueTypes, b.i8ptrType) // closure
+			valueTypes = append(valueTypes, b.dataPtrType) // closure
 			deferredCallType := b.ctx.StructType(valueTypes, false)
-			deferredCallPtr := b.CreateBitCast(deferData, llvm.PointerType(deferredCallType, 0), "defercall")
 
 			// Extract the params from the struct.
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
 			for i := 2; i < len(valueTypes); i++ {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferredCallPtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "")
+				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "")
 				forwardParam := b.CreateLoad(valueTypes[i], gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -584,7 +595,7 @@ func (b *builder) createRunDefers() {
 			db := b.deferBuiltinFuncs[callback]
 
 			//Get parameter types
-			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
+			valueTypes := []llvm.Type{b.uintptrType, b.dataPtrType}
 
 			//Get signature from call results
 			params := callback.Type().Underlying().(*types.Signature).Params()
@@ -593,13 +604,12 @@ func (b *builder) createRunDefers() {
 			}
 
 			deferredCallType := b.ctx.StructType(valueTypes, false)
-			deferredCallPtr := b.CreateBitCast(deferData, llvm.PointerType(deferredCallType, 0), "defercall")
 
 			// Extract the params from the struct.
 			var argValues []llvm.Value
 			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
 			for i := 0; i < params.Len(); i++ {
-				gep := b.CreateInBoundsGEP(deferredCallType, deferredCallPtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
+				gep := b.CreateInBoundsGEP(deferredCallType, deferData, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
 				forwardParam := b.CreateLoad(valueTypes[i+2], gep, "param")
 				argValues = append(argValues, forwardParam)
 			}

compiler/func.go

@@ -13,35 +13,14 @@ import (
 // createFuncValue creates a function value from a raw function pointer with no
 // context.
 func (b *builder) createFuncValue(funcPtr, context llvm.Value, sig *types.Signature) llvm.Value {
-	return b.compilerContext.createFuncValue(b.Builder, funcPtr, context, sig)
-}
-
-// createFuncValue creates a function value from a raw function pointer with no
-// context.
-func (c *compilerContext) createFuncValue(builder llvm.Builder, funcPtr, context llvm.Value, sig *types.Signature) llvm.Value {
 	// Closure is: {context, function pointer}
-	funcValueScalar := llvm.ConstBitCast(funcPtr, c.rawVoidFuncType)
-	funcValueType := c.getFuncType(sig)
+	funcValueType := b.getFuncType(sig)
 	funcValue := llvm.Undef(funcValueType)
-	funcValue = builder.CreateInsertValue(funcValue, context, 0, "")
-	funcValue = builder.CreateInsertValue(funcValue, funcValueScalar, 1, "")
+	funcValue = b.CreateInsertValue(funcValue, context, 0, "")
+	funcValue = b.CreateInsertValue(funcValue, funcPtr, 1, "")
 	return funcValue
 }
 
-// getFuncSignatureID returns a new external global for a given signature. This
-// global reference is not real, it is only used during func lowering to assign
-// signature types to functions and will then be removed.
-func (c *compilerContext) getFuncSignatureID(sig *types.Signature) llvm.Value {
-	s, _ := getTypeCodeName(sig)
-	sigGlobalName := "reflect/types.funcid:" + s
-	sigGlobal := c.mod.NamedGlobal(sigGlobalName)
-	if sigGlobal.IsNil() {
-		sigGlobal = llvm.AddGlobal(c.mod, c.ctx.Int8Type(), sigGlobalName)
-		sigGlobal.SetGlobalConstant(true)
-	}
-	return sigGlobal
-}
-
 // extractFuncScalar returns some scalar that can be used in comparisons. It is
 // a cheap operation.
 func (b *builder) extractFuncScalar(funcValue llvm.Value) llvm.Value {
@@ -55,28 +34,20 @@ func (b *builder) extractFuncContext(funcValue llvm.Value) llvm.Value {
 }
 
 // decodeFuncValue extracts the context and the function pointer from this func
-// value. This may be an expensive operation.
-func (b *builder) decodeFuncValue(funcValue llvm.Value, sig *types.Signature) (funcType llvm.Type, funcPtr, context llvm.Value) {
+// value.
+func (b *builder) decodeFuncValue(funcValue llvm.Value) (funcPtr, context llvm.Value) {
 	context = b.CreateExtractValue(funcValue, 0, "")
 	funcPtr = b.CreateExtractValue(funcValue, 1, "")
-	if !funcPtr.IsAConstantExpr().IsNil() && funcPtr.Opcode() == llvm.BitCast {
-		funcPtr = funcPtr.Operand(0) // needed for LLVM 14 (no opaque pointers)
-	}
-	if sig != nil {
-		funcType = b.getRawFuncType(sig)
-		llvmSig := llvm.PointerType(funcType, b.funcPtrAddrSpace)
-		funcPtr = b.CreateBitCast(funcPtr, llvmSig, "")
-	}
 	return
 }
 
 // getFuncType returns the type of a func value given a signature.
 func (c *compilerContext) getFuncType(typ *types.Signature) llvm.Type {
-	return c.ctx.StructType([]llvm.Type{c.i8ptrType, c.rawVoidFuncType}, false)
+	return c.ctx.StructType([]llvm.Type{c.dataPtrType, c.funcPtrType}, false)
 }
 
-// getRawFuncType returns a LLVM function type for a given signature.
-func (c *compilerContext) getRawFuncType(typ *types.Signature) llvm.Type {
+// getLLVMFunctionType returns a LLVM function type for a given signature.
+func (c *compilerContext) getLLVMFunctionType(typ *types.Signature) llvm.Type {
 	// Get the return type.
 	var returnType llvm.Type
 	switch typ.Results().Len() {
@@ -104,7 +75,7 @@ func (c *compilerContext) getRawFuncType(typ *types.Signature) llvm.Type {
 		if recv.StructName() == "runtime._interface" {
 			// This is a call on an interface, not a concrete type.
 			// The receiver is not an interface, but a i8* type.
-			recv = c.i8ptrType
+			recv = c.dataPtrType
 		}
 		for _, info := range c.expandFormalParamType(recv, "", nil) {
 			paramTypes = append(paramTypes, info.llvmType)
@@ -117,7 +88,7 @@ func (c *compilerContext) getRawFuncType(typ *types.Signature) llvm.Type {
 		}
 	}
 	// All functions take these parameters at the end.
-	paramTypes = append(paramTypes, c.i8ptrType) // context
+	paramTypes = append(paramTypes, c.dataPtrType) // context
 
 	// Make a func type out of the signature.
 	return llvm.FunctionType(returnType, paramTypes, false)

compiler/gc.go

@@ -78,12 +78,9 @@ func (b *builder) trackValue(value llvm.Value) {
 	}
 }
 
-// trackPointer creates a call to runtime.trackPointer, bitcasting the poitner
+// trackPointer creates a call to runtime.trackPointer, bitcasting the pointer
 // first if needed. The input value must be of LLVM pointer type.
 func (b *builder) trackPointer(value llvm.Value) {
-	if value.Type() != b.i8ptrType {
-		value = b.CreateBitCast(value, b.i8ptrType, "")
-	}
 	b.createRuntimeCall("trackPointer", []llvm.Value{value, b.stackChainAlloca}, "")
 }
 

compiler/goroutine.go

@@ -7,43 +7,14 @@ import (
 	"go/token"
 	"go/types"
 
+	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
 
 // createGo emits code to start a new goroutine.
 func (b *builder) createGo(instr *ssa.Go) {
-	// Get all function parameters to pass to the goroutine.
-	var params []llvm.Value
-	for _, param := range instr.Call.Args {
-		params = append(params, b.getValue(param, getPos(instr)))
-	}
-
-	var prefix string
-	var funcPtr llvm.Value
-	var funcPtrType llvm.Type
-	hasContext := false
-	if callee := instr.Call.StaticCallee(); callee != nil {
-		// Static callee is known. This makes it easier to start a new
-		// goroutine.
-		var context llvm.Value
-		switch value := instr.Call.Value.(type) {
-		case *ssa.Function:
-			// Goroutine call is regular function call. No context is necessary.
-		case *ssa.MakeClosure:
-			// A goroutine call on a func value, but the callee is trivial to find. For
-			// example: immediately applied functions.
-			funcValue := b.getValue(value, getPos(instr))
-			context = b.extractFuncContext(funcValue)
-		default:
-			panic("StaticCallee returned an unexpected value")
-		}
-		if !context.IsNil() {
-			params = append(params, context) // context parameter
-			hasContext = true
-		}
-		funcPtrType, funcPtr = b.getFunction(callee)
-	} else if builtin, ok := instr.Call.Value.(*ssa.Builtin); ok {
+	if builtin, ok := instr.Call.Value.(*ssa.Builtin); ok {
 		// We cheat. None of the builtins do any long or blocking operation, so
 		// we might as well run these builtins right away without the program
 		// noticing the difference.
@@ -74,13 +45,45 @@ func (b *builder) createGo(instr *ssa.Go) {
 		}
 		b.createBuiltin(argTypes, argValues, builtin.Name(), instr.Pos())
 		return
+	}
+
+	// Get all function parameters to pass to the goroutine.
+	var params []llvm.Value
+	for _, param := range instr.Call.Args {
+		params = append(params, b.expandFormalParam(b.getValue(param, getPos(instr)))...)
+	}
+
+	var prefix string
+	var funcPtr llvm.Value
+	var funcType llvm.Type
+	hasContext := false
+	if callee := instr.Call.StaticCallee(); callee != nil {
+		// Static callee is known. This makes it easier to start a new
+		// goroutine.
+		var context llvm.Value
+		switch value := instr.Call.Value.(type) {
+		case *ssa.Function:
+			// Goroutine call is regular function call. No context is necessary.
+		case *ssa.MakeClosure:
+			// A goroutine call on a func value, but the callee is trivial to find. For
+			// example: immediately applied functions.
+			funcValue := b.getValue(value, getPos(instr))
+			context = b.extractFuncContext(funcValue)
+		default:
+			panic("StaticCallee returned an unexpected value")
+		}
+		if !context.IsNil() {
+			params = append(params, context) // context parameter
+			hasContext = true
+		}
+		funcType, funcPtr = b.getFunction(callee)
 	} else if instr.Call.IsInvoke() {
 		// This is a method call on an interface value.
 		itf := b.getValue(instr.Call.Value, getPos(instr))
 		itfTypeCode := b.CreateExtractValue(itf, 0, "")
 		itfValue := b.CreateExtractValue(itf, 1, "")
 		funcPtr = b.getInvokeFunction(&instr.Call)
-		funcPtrType = funcPtr.GlobalValueType()
+		funcType = funcPtr.GlobalValueType()
 		params = append([]llvm.Value{itfValue}, params...) // start with receiver
 		params = append(params, itfTypeCode)               // end with typecode
 	} else {
@@ -90,7 +93,8 @@ func (b *builder) createGo(instr *ssa.Go) {
 		//   * The function context, for closures.
 		//   * The function pointer (for tasks).
 		var context llvm.Value
-		funcPtrType, funcPtr, context = b.decodeFuncValue(b.getValue(instr.Call.Value, getPos(instr)), instr.Call.Value.Type().Underlying().(*types.Signature))
+		funcPtr, context = b.decodeFuncValue(b.getValue(instr.Call.Value, getPos(instr)))
+		funcType = b.getLLVMFunctionType(instr.Call.Value.Type().Underlying().(*types.Signature))
 		params = append(params, context, funcPtr)
 		hasContext = true
 		prefix = b.fn.RelString(nil)
@@ -98,14 +102,14 @@ func (b *builder) createGo(instr *ssa.Go) {
 
 	paramBundle := b.emitPointerPack(params)
 	var stackSize llvm.Value
-	callee := b.createGoroutineStartWrapper(funcPtrType, funcPtr, prefix, hasContext, instr.Pos())
+	callee := b.createGoroutineStartWrapper(funcType, funcPtr, prefix, hasContext, false, instr.Pos())
 	if b.AutomaticStackSize {
 		// The stack size is not known until after linking. Call a dummy
 		// function that will be replaced with a load from a special ELF
 		// section that contains the stack size (and is modified after
 		// linking).
 		stackSizeFnType, stackSizeFn := b.getFunction(b.program.ImportedPackage("internal/task").Members["getGoroutineStackSize"].(*ssa.Function))
-		stackSize = b.createCall(stackSizeFnType, stackSizeFn, []llvm.Value{callee, llvm.Undef(b.i8ptrType)}, "stacksize")
+		stackSize = b.createCall(stackSizeFnType, stackSizeFn, []llvm.Value{callee, llvm.Undef(b.dataPtrType)}, "stacksize")
 	} else {
 		// The stack size is fixed at compile time. By emitting it here as a
 		// constant, it can be optimized.
@@ -115,7 +119,148 @@ func (b *builder) createGo(instr *ssa.Go) {
 		stackSize = llvm.ConstInt(b.uintptrType, b.DefaultStackSize, false)
 	}
 	fnType, start := b.getFunction(b.program.ImportedPackage("internal/task").Members["start"].(*ssa.Function))
-	b.createCall(fnType, start, []llvm.Value{callee, paramBundle, stackSize, llvm.Undef(b.i8ptrType)}, "")
+	b.createCall(fnType, start, []llvm.Value{callee, paramBundle, stackSize, llvm.Undef(b.dataPtrType)}, "")
+}
+
+// Create an exported wrapper function for functions with the //go:wasmexport
+// pragma. This wrapper function is quite complex when the scheduler is enabled:
+// it needs to start a new goroutine each time the exported function is called.
+func (b *builder) createWasmExport() {
+	pos := b.info.wasmExportPos
+	if b.info.exported {
+		// //export really shouldn't be used anymore when //go:wasmexport is
+		// available, because //go:wasmexport is much better defined.
+		b.addError(pos, "cannot use //export and //go:wasmexport at the same time")
+		return
+	}
+
+	const suffix = "#wasmexport"
+
+	// Declare the exported function.
+	paramTypes := b.llvmFnType.ParamTypes()
+	exportedFnType := llvm.FunctionType(b.llvmFnType.ReturnType(), paramTypes[:len(paramTypes)-1], false)
+	exportedFn := llvm.AddFunction(b.mod, b.fn.RelString(nil)+suffix, exportedFnType)
+	b.addStandardAttributes(exportedFn)
+	llvmutil.AppendToGlobal(b.mod, "llvm.used", exportedFn)
+	exportedFn.AddFunctionAttr(b.ctx.CreateStringAttribute("wasm-export-name", b.info.wasmExport))
+
+	// Create a builder for this wrapper function.
+	builder := newBuilder(b.compilerContext, b.ctx.NewBuilder(), b.fn)
+	defer builder.Dispose()
+
+	// Define this function as a separate function in DWARF
+	if b.Debug {
+		if b.fn.Syntax() != nil {
+			// Create debug info file if needed.
+			pos := b.program.Fset.Position(pos)
+			builder.difunc = builder.attachDebugInfoRaw(b.fn, exportedFn, suffix, pos.Filename, pos.Line)
+		}
+		builder.setDebugLocation(pos)
+	}
+
+	// Create a single basic block inside of it.
+	bb := llvm.AddBasicBlock(exportedFn, "entry")
+	builder.SetInsertPointAtEnd(bb)
+
+	// Insert an assertion to make sure this //go:wasmexport function is not
+	// called at a time when it is not allowed (for example, before the runtime
+	// is initialized).
+	builder.createRuntimeCall("wasmExportCheckRun", nil, "")
+
+	if b.Scheduler == "none" {
+		// When the scheduler has been disabled, this is really trivial: just
+		// call the function.
+		params := exportedFn.Params()
+		params = append(params, llvm.ConstNull(b.dataPtrType)) // context parameter
+		retval := builder.CreateCall(b.llvmFnType, b.llvmFn, params, "")
+		if b.fn.Signature.Results() == nil {
+			builder.CreateRetVoid()
+		} else {
+			builder.CreateRet(retval)
+		}
+
+	} else {
+		// The scheduler is enabled, so we need to start a new goroutine, wait
+		// for it to complete, and read the result value.
+
+		// Build a function that looks like this:
+		//
+		//   func foo#wasmexport(param0, param1, ..., paramN) {
+		//       var state *stateStruct
+		//
+		//       // 'done' must be explicitly initialized ('state' is not zeroed)
+		//       state.done = false
+		//
+		//       // store the parameters in the state object
+		//       state.param0 = param0
+		//       state.param1 = param1
+		//       ...
+		//       state.paramN = paramN
+		//
+		//       // create a goroutine and push it to the runqueue
+		//       task.start(uintptr(gowrapper), &state)
+		//
+		//       // run the scheduler
+		//       runtime.wasmExportRun(&state.done)
+		//
+		//       // if there is a return value, load it and return
+		//       return state.result
+		//   }
+
+		hasReturn := b.fn.Signature.Results() != nil
+
+		// Build the state struct type.
+		// It stores the function parameters, the 'done' flag, and reserves
+		// space for a return value if needed.
+		stateFields := exportedFnType.ParamTypes()
+		numParams := len(stateFields)
+		stateFields = append(stateFields, b.ctx.Int1Type()) // 'done' field
+		if hasReturn {
+			stateFields = append(stateFields, b.llvmFnType.ReturnType())
+		}
+		stateStruct := b.ctx.StructType(stateFields, false)
+
+		// Allocate the state struct on the stack.
+		statePtr := builder.CreateAlloca(stateStruct, "status")
+
+		// Initialize the 'done' field.
+		doneGEP := builder.CreateInBoundsGEP(stateStruct, statePtr, []llvm.Value{
+			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(b.ctx.Int32Type(), uint64(numParams), false),
+		}, "done.gep")
+		builder.CreateStore(llvm.ConstNull(b.ctx.Int1Type()), doneGEP)
+
+		// Store all parameters in the state object.
+		for i, param := range exportedFn.Params() {
+			gep := builder.CreateInBoundsGEP(stateStruct, statePtr, []llvm.Value{
+				llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+				llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false),
+			}, "")
+			builder.CreateStore(param, gep)
+		}
+
+		// Create a new goroutine and add it to the runqueue.
+		wrapper := b.createGoroutineStartWrapper(b.llvmFnType, b.llvmFn, "", false, true, pos)
+		stackSize := llvm.ConstInt(b.uintptrType, b.DefaultStackSize, false)
+		taskStartFnType, taskStartFn := builder.getFunction(b.program.ImportedPackage("internal/task").Members["start"].(*ssa.Function))
+		builder.createCall(taskStartFnType, taskStartFn, []llvm.Value{wrapper, statePtr, stackSize, llvm.Undef(b.dataPtrType)}, "")
+
+		// Run the scheduler.
+		builder.createRuntimeCall("wasmExportRun", []llvm.Value{doneGEP}, "")
+
+		// Read the return value (if any) and return to the caller of the
+		// //go:wasmexport function.
+		if hasReturn {
+			gep := builder.CreateInBoundsGEP(stateStruct, statePtr, []llvm.Value{
+				llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+				llvm.ConstInt(b.ctx.Int32Type(), uint64(numParams)+1, false),
+			}, "")
+			retval := builder.CreateLoad(b.llvmFnType.ReturnType(), gep, "retval")
+			builder.CreateRet(retval)
+		} else {
+			builder.CreateRetVoid()
+		}
+	}
 }
 
 // createGoroutineStartWrapper creates a wrapper for the task-based
@@ -141,7 +286,7 @@ func (b *builder) createGo(instr *ssa.Go) {
 // to last parameter of the function) is used for this wrapper. If hasContext is
 // false, the parameter bundle is assumed to have no context parameter and undef
 // is passed instead.
-func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.Value, prefix string, hasContext bool, pos token.Pos) llvm.Value {
+func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.Value, prefix string, hasContext, isWasmExport bool, pos token.Pos) llvm.Value {
 	var wrapper llvm.Value
 
 	b := &builder{
@@ -159,14 +304,18 @@ func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.
 	if !fn.IsAFunction().IsNil() {
 		// See whether this wrapper has already been created. If so, return it.
 		name := fn.Name()
-		wrapper = c.mod.NamedFunction(name + "$gowrapper")
+		wrapperName := name + "$gowrapper"
+		if isWasmExport {
+			wrapperName += "-wasmexport"
+		}
+		wrapper = c.mod.NamedFunction(wrapperName)
 		if !wrapper.IsNil() {
 			return llvm.ConstPtrToInt(wrapper, c.uintptrType)
 		}
 
 		// Create the wrapper.
-		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.i8ptrType}, false)
-		wrapper = llvm.AddFunction(c.mod, name+"$gowrapper", wrapperType)
+		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.dataPtrType}, false)
+		wrapper = llvm.AddFunction(c.mod, wrapperName, wrapperType)
 		c.addStandardAttributes(wrapper)
 		wrapper.SetLinkage(llvm.LinkOnceODRLinkage)
 		wrapper.SetUnnamedAddr(true)
@@ -196,23 +345,110 @@ func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.
 			b.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), difunc, llvm.Metadata{})
 		}
 
-		// Create the list of params for the call.
-		paramTypes := fnType.ParamTypes()
-		if !hasContext {
-			paramTypes = paramTypes[:len(paramTypes)-1] // strip context parameter
-		}
-		params := b.emitPointerUnpack(wrapper.Param(0), paramTypes)
-		if !hasContext {
-			params = append(params, llvm.Undef(c.i8ptrType)) // add dummy context parameter
-		}
+		if !isWasmExport {
+			// Regular 'go' instruction.
 
-		// Create the call.
-		b.CreateCall(fnType, fn, params, "")
+			// Create the list of params for the call.
+			paramTypes := fnType.ParamTypes()
+			if !hasContext {
+				paramTypes = paramTypes[:len(paramTypes)-1] // strip context parameter
+			}
 
-		if c.Scheduler == "asyncify" {
-			b.CreateCall(deadlockType, deadlock, []llvm.Value{
-				llvm.Undef(c.i8ptrType),
-			}, "")
+			params := b.emitPointerUnpack(wrapper.Param(0), paramTypes)
+			if !hasContext {
+				params = append(params, llvm.Undef(c.dataPtrType)) // add dummy context parameter
+			}
+
+			// Create the call.
+			b.CreateCall(fnType, fn, params, "")
+
+			if c.Scheduler == "asyncify" {
+				b.CreateCall(deadlockType, deadlock, []llvm.Value{
+					llvm.Undef(c.dataPtrType),
+				}, "")
+			}
+		} else {
+			// Goroutine started from a //go:wasmexport pragma.
+			// The function looks like this:
+			//
+			//   func foo$gowrapper-wasmexport(state *stateStruct) {
+			//       // load values
+			//       param0 := state.params[0]
+			//       param1 := state.params[1]
+			//
+			//       // call wrapped functions
+			//       result := foo(param0, param1, ...)
+			//
+			//       // store result value (if there is any)
+			//       state.result = result
+			//
+			//       // finish exported function
+			//       state.done = true
+			//       runtime.wasmExportExit()
+			//   }
+			//
+			// The state object here looks like:
+			//
+			//   struct state {
+			//       param0
+			//       param1
+			//       param* // etc
+			//       done bool
+			//       result returnType
+			//   }
+
+			returnType := fnType.ReturnType()
+			hasReturn := returnType != b.ctx.VoidType()
+			statePtr := wrapper.Param(0)
+
+			// Create the state struct (it must match the type in createWasmExport).
+			stateFields := fnType.ParamTypes()
+			numParams := len(stateFields) - 1
+			stateFields = stateFields[:numParams:numParams]     // strip 'context' parameter
+			stateFields = append(stateFields, c.ctx.Int1Type()) // 'done' bool
+			if hasReturn {
+				stateFields = append(stateFields, returnType)
+			}
+			stateStruct := b.ctx.StructType(stateFields, false)
+
+			// Extract parameters from the state object, and call the function
+			// that's being wrapped.
+			var callParams []llvm.Value
+			for i := 0; i < numParams; i++ {
+				gep := b.CreateInBoundsGEP(stateStruct, statePtr, []llvm.Value{
+					llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+					llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false),
+				}, "")
+				param := b.CreateLoad(stateFields[i], gep, "")
+				callParams = append(callParams, param)
+			}
+			callParams = append(callParams, llvm.ConstNull(c.dataPtrType)) // add 'context' parameter
+			result := b.CreateCall(fnType, fn, callParams, "")
+
+			// Store the return value back into the shared state.
+			// Unlike regular goroutines, these special //go:wasmexport
+			// goroutines can return a value.
+			if hasReturn {
+				gep := b.CreateInBoundsGEP(stateStruct, statePtr, []llvm.Value{
+					llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+					llvm.ConstInt(c.ctx.Int32Type(), uint64(numParams)+1, false),
+				}, "result.ptr")
+				b.CreateStore(result, gep)
+			}
+
+			// Mark this function as having finished executing.
+			// This is important so the runtime knows the exported function
+			// didn't block.
+			doneGEP := b.CreateInBoundsGEP(stateStruct, statePtr, []llvm.Value{
+				llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+				llvm.ConstInt(c.ctx.Int32Type(), uint64(numParams), false),
+			}, "done.gep")
+			b.CreateStore(llvm.ConstInt(b.ctx.Int1Type(), 1, false), doneGEP)
+
+			// Call back into the runtime. This will exit the goroutine, switch
+			// back to the scheduler, which will in turn return from the
+			// //go:wasmexport function.
+			b.createRuntimeCall("wasmExportExit", nil, "")
 		}
 
 	} else {
@@ -234,7 +470,7 @@ func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.
 		// merged into one.
 
 		// Create the wrapper.
-		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.i8ptrType}, false)
+		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.dataPtrType}, false)
 		wrapper = llvm.AddFunction(c.mod, prefix+".gowrapper", wrapperType)
 		c.addStandardAttributes(wrapper)
 		wrapper.SetLinkage(llvm.LinkOnceODRLinkage)
@@ -279,7 +515,7 @@ func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.
 
 		if c.Scheduler == "asyncify" {
 			b.CreateCall(deadlockType, deadlock, []llvm.Value{
-				llvm.Undef(c.i8ptrType),
+				llvm.Undef(c.dataPtrType),
 			}, "")
 		}
 	}
@@ -294,5 +530,5 @@ func (c *compilerContext) createGoroutineStartWrapper(fnType llvm.Type, fn llvm.
 	}
 
 	// Return a ptrtoint of the wrapper, not the function itself.
-	return b.CreatePtrToInt(wrapper, c.uintptrType, "")
+	return llvm.ConstPtrToInt(wrapper, c.uintptrType)
 }

compiler/inlineasm.go

@@ -99,8 +99,8 @@ func (b *builder) createInlineAsmFull(instr *ssa.CallCommon) (llvm.Value, error)
 			case llvm.IntegerTypeKind:
 				constraints = append(constraints, "r")
 			case llvm.PointerTypeKind:
-				// Memory references require a type in LLVM 14, probably as a
-				// preparation for opaque pointers.
+				// Memory references require a type starting with LLVM 14,
+				// probably as a preparation for opaque pointers.
 				err = b.makeError(instr.Pos(), "support for pointer operands was dropped in TinyGo 0.23")
 				return s
 			default:

compiler/interface.go

@@ -86,7 +86,7 @@ func (b *builder) createMakeInterface(val llvm.Value, typ types.Type, pos token.
 
 // extractValueFromInterface extract the value from an interface value
 // (runtime._interface) under the assumption that it is of the type given in
-// llvmType. The behavior is undefied if the interface is nil or llvmType
+// llvmType. The behavior is undefined if the interface is nil or llvmType
 // doesn't match the underlying type of the interface.
 func (b *builder) extractValueFromInterface(itf llvm.Value, llvmType llvm.Type) llvm.Value {
 	valuePtr := b.CreateExtractValue(itf, 1, "typeassert.value.ptr")
@@ -124,10 +124,22 @@ func (c *compilerContext) pkgPathPtr(pkgpath string) llvm.Value {
 func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 	ms := c.program.MethodSets.MethodSet(typ)
 	hasMethodSet := ms.Len() != 0
-	if _, ok := typ.Underlying().(*types.Interface); ok {
+	_, isInterface := typ.Underlying().(*types.Interface)
+	if isInterface {
 		hasMethodSet = false
 	}
 
+	// As defined in https://pkg.go.dev/reflect#Type:
+	// NumMethod returns the number of methods accessible using Method.
+	// For a non-interface type, it returns the number of exported methods.
+	// For an interface type, it returns the number of exported and unexported methods.
+	var numMethods int
+	for i := 0; i < ms.Len(); i++ {
+		if isInterface || ms.At(i).Obj().Exported() {
+			numMethods++
+		}
+	}
+
 	// Short-circuit all the global pointer logic here for pointers to pointers.
 	if typ, ok := typ.(*types.Pointer); ok {
 		if _, ok := typ.Elem().(*types.Pointer); ok {
@@ -138,7 +150,7 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 				c.addError(token.NoPos, fmt.Sprintf("too many levels of pointers for typecode: %s", typstr))
 			}
 			return llvm.ConstGEP(c.ctx.Int8Type(), ptr, []llvm.Value{
-				llvm.ConstInt(llvm.Int32Type(), 1, false),
+				llvm.ConstInt(c.ctx.Int32Type(), 1, false),
 			})
 		}
 	}
@@ -209,6 +221,7 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 				types.NewVar(token.NoPos, nil, "ptrTo", types.Typ[types.UnsafePointer]),
 				types.NewVar(token.NoPos, nil, "elementType", types.Typ[types.UnsafePointer]),
 				types.NewVar(token.NoPos, nil, "length", types.Typ[types.Uintptr]),
+				types.NewVar(token.NoPos, nil, "sliceOf", types.Typ[types.UnsafePointer]),
 			)
 		case *types.Map:
 			typeFieldTypes = append(typeFieldTypes,
@@ -277,11 +290,11 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 			}
 			pkgPathPtr := c.pkgPathPtr(pkgpath)
 			typeFields = []llvm.Value{
-				llvm.ConstInt(c.ctx.Int16Type(), uint64(ms.Len()), false), // numMethods
-				c.getTypeCode(types.NewPointer(typ)),                      // ptrTo
-				c.getTypeCode(typ.Underlying()),                           // underlying
-				pkgPathPtr,                                                // pkgpath pointer
-				c.ctx.ConstString(pkgname+"."+name+"\x00", false),         // name
+				llvm.ConstInt(c.ctx.Int16Type(), uint64(numMethods), false), // numMethods
+				c.getTypeCode(types.NewPointer(typ)),                        // ptrTo
+				c.getTypeCode(typ.Underlying()),                             // underlying
+				pkgPathPtr,                                                  // pkgpath pointer
+				c.ctx.ConstString(pkgname+"."+name+"\x00", false),           // name
 			}
 			metabyte |= 1 << 5 // "named" flag
 		case *types.Chan:
@@ -308,7 +321,7 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 			}
 		case *types.Pointer:
 			typeFields = []llvm.Value{
-				llvm.ConstInt(c.ctx.Int16Type(), uint64(ms.Len()), false), // numMethods
+				llvm.ConstInt(c.ctx.Int16Type(), uint64(numMethods), false), // numMethods
 				c.getTypeCode(typ.Elem()),
 			}
 		case *types.Array:
@@ -317,6 +330,7 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 				c.getTypeCode(types.NewPointer(typ)),                   // ptrTo
 				c.getTypeCode(typ.Elem()),                              // elementType
 				llvm.ConstInt(c.uintptrType, uint64(typ.Len()), false), // length
+				c.getTypeCode(types.NewSlice(typ.Elem())),              // slicePtr
 			}
 		case *types.Map:
 			typeFields = []llvm.Value{
@@ -337,8 +351,8 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 			llvmStructType := c.getLLVMType(typ)
 			size := c.targetData.TypeStoreSize(llvmStructType)
 			typeFields = []llvm.Value{
-				llvm.ConstInt(c.ctx.Int16Type(), uint64(ms.Len()), false), // numMethods
-				c.getTypeCode(types.NewPointer(typ)),                      // ptrTo
+				llvm.ConstInt(c.ctx.Int16Type(), uint64(numMethods), false), // numMethods
+				c.getTypeCode(types.NewPointer(typ)),                        // ptrTo
 				pkgPathPtr,
 				llvm.ConstInt(c.ctx.Int32Type(), uint64(size), false),            // size
 				llvm.ConstInt(c.ctx.Int16Type(), uint64(typ.NumFields()), false), // numFields
@@ -403,10 +417,10 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 		}, typeFields...)
 		if hasMethodSet {
 			typeFields = append([]llvm.Value{
-				llvm.ConstBitCast(c.getTypeMethodSet(typ), c.i8ptrType),
+				c.getTypeMethodSet(typ),
 			}, typeFields...)
 		}
-		alignment := c.targetData.TypeAllocSize(c.i8ptrType)
+		alignment := c.targetData.TypeAllocSize(c.dataPtrType)
 		if alignment < 4 {
 			alignment = 4
 		}
@@ -441,8 +455,8 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 		offset = 1
 	}
 	return llvm.ConstGEP(global.GlobalValueType(), global, []llvm.Value{
-		llvm.ConstInt(llvm.Int32Type(), 0, false),
-		llvm.ConstInt(llvm.Int32Type(), offset, false),
+		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+		llvm.ConstInt(c.ctx.Int32Type(), offset, false),
 	})
 }
 
@@ -500,8 +514,7 @@ var basicTypeNames = [...]string{
 func getTypeCodeName(t types.Type) (string, bool) {
 	switch t := t.(type) {
 	case *types.Named:
-		// Note: check for `t.Obj().Pkg() != nil` for Go 1.18 only.
-		if t.Obj().Pkg() != nil && t.Obj().Parent() != t.Obj().Pkg().Scope() {
+		if t.Obj().Parent() != t.Obj().Pkg().Scope() {
 			return "named:" + t.String() + "$local", true
 		}
 		return "named:" + t.String(), false
@@ -617,7 +630,7 @@ func (c *compilerContext) getTypeMethodSet(typ types.Type) llvm.Value {
 		// Construct global value.
 		globalValue := c.ctx.ConstStruct([]llvm.Value{
 			llvm.ConstInt(c.uintptrType, uint64(ms.Len()), false),
-			llvm.ConstArray(c.i8ptrType, signatures),
+			llvm.ConstArray(c.dataPtrType, signatures),
 			c.ctx.ConstStruct(wrappers, false),
 		}, false)
 		global = llvm.AddGlobal(c.mod, globalValue.Type(), globalName)
@@ -673,19 +686,25 @@ func (b *builder) createTypeAssert(expr *ssa.TypeAssert) llvm.Value {
 
 	actualTypeNum := b.CreateExtractValue(itf, 0, "interface.type")
 	commaOk := llvm.Value{}
-	if _, ok := expr.AssertedType.Underlying().(*types.Interface); ok {
-		// Type assert on interface type.
-		// This is a call to an interface type assert function.
-		// The interface lowering pass will define this function by filling it
-		// with a type switch over all concrete types that implement this
-		// interface, and returning whether it's one of the matched types.
-		// This is very different from how interface asserts are implemented in
-		// the main Go compiler, where the runtime checks whether the type
-		// implements each method of the interface. See:
-		// https://research.swtch.com/interfaces
-		fn := b.getInterfaceImplementsFunc(expr.AssertedType)
-		commaOk = b.CreateCall(fn.GlobalValueType(), fn, []llvm.Value{actualTypeNum}, "")
 
+	if intf, ok := expr.AssertedType.Underlying().(*types.Interface); ok {
+		if intf.Empty() {
+			// intf is the empty interface => no methods
+			// This type assertion always succeeds, so we can just set commaOk to true.
+			commaOk = llvm.ConstInt(b.ctx.Int1Type(), 1, true)
+		} else {
+			// Type assert on interface type with methods.
+			// This is a call to an interface type assert function.
+			// The interface lowering pass will define this function by filling it
+			// with a type switch over all concrete types that implement this
+			// interface, and returning whether it's one of the matched types.
+			// This is very different from how interface asserts are implemented in
+			// the main Go compiler, where the runtime checks whether the type
+			// implements each method of the interface. See:
+			// https://research.swtch.com/interfaces
+			fn := b.getInterfaceImplementsFunc(expr.AssertedType)
+			commaOk = b.CreateCall(fn.GlobalValueType(), fn, []llvm.Value{actualTypeNum}, "")
+		}
 	} else {
 		name, _ := getTypeCodeName(expr.AssertedType)
 		globalName := "reflect/types.typeid:" + name
@@ -768,7 +787,7 @@ func (c *compilerContext) getInterfaceImplementsFunc(assertedType types.Type) ll
 	fnName := s + ".$typeassert"
 	llvmFn := c.mod.NamedFunction(fnName)
 	if llvmFn.IsNil() {
-		llvmFnType := llvm.FunctionType(c.ctx.Int1Type(), []llvm.Type{c.i8ptrType}, false)
+		llvmFnType := llvm.FunctionType(c.ctx.Int1Type(), []llvm.Type{c.dataPtrType}, false)
 		llvmFn = llvm.AddFunction(c.mod, fnName, llvmFnType)
 		c.addStandardDeclaredAttributes(llvmFn)
 		methods := c.getMethodsString(assertedType.Underlying().(*types.Interface))
@@ -791,7 +810,7 @@ func (c *compilerContext) getInvokeFunction(instr *ssa.CallCommon) llvm.Value {
 			paramTuple = append(paramTuple, sig.Params().At(i))
 		}
 		paramTuple = append(paramTuple, types.NewVar(token.NoPos, nil, "$typecode", types.Typ[types.UnsafePointer]))
-		llvmFnType := c.getRawFuncType(types.NewSignature(sig.Recv(), types.NewTuple(paramTuple...), sig.Results(), false))
+		llvmFnType := c.getLLVMFunctionType(types.NewSignature(sig.Recv(), types.NewTuple(paramTuple...), sig.Results(), false))
 		llvmFn = llvm.AddFunction(c.mod, fnName, llvmFnType)
 		c.addStandardDeclaredAttributes(llvmFn)
 		llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("tinygo-invoke", c.getMethodSignatureName(instr.Method)))
@@ -831,7 +850,7 @@ func (c *compilerContext) getInterfaceInvokeWrapper(fn *ssa.Function, llvmFnType
 	}
 
 	// create wrapper function
-	paramTypes := append([]llvm.Type{c.i8ptrType}, llvmFnType.ParamTypes()[len(expandedReceiverType):]...)
+	paramTypes := append([]llvm.Type{c.dataPtrType}, llvmFnType.ParamTypes()[len(expandedReceiverType):]...)
 	wrapFnType := llvm.FunctionType(llvmFnType.ReturnType(), paramTypes, false)
 	wrapper = llvm.AddFunction(c.mod, wrapperName, wrapFnType)
 	c.addStandardAttributes(wrapper)

compiler/interrupt.go

@@ -36,11 +36,13 @@ func (b *builder) createInterruptGlobal(instr *ssa.CallCommon) (llvm.Value, erro
 		// Fall back to a generic error.
 		return llvm.Value{}, b.makeError(instr.Pos(), "interrupt function must be constant")
 	}
-	_, funcRawPtr, funcContext := b.decodeFuncValue(funcValue, nil)
+	funcRawPtr, funcContext := b.decodeFuncValue(funcValue)
 	funcPtr := llvm.ConstPtrToInt(funcRawPtr, b.uintptrType)
 
 	// Create a new global of type runtime/interrupt.handle. Globals of this
 	// type are lowered in the interrupt lowering pass.
+	// It must have an alignment of 1, otherwise LLVM thinks a ptrtoint of the
+	// global has the lower bits unset.
 	globalType := b.program.ImportedPackage("runtime/interrupt").Type("handle").Type()
 	globalLLVMType := b.getLLVMType(globalType)
 	globalName := b.fn.Package().Pkg.Path() + "$interrupt" + strconv.FormatInt(id.Int64(), 10)
@@ -48,6 +50,7 @@ func (b *builder) createInterruptGlobal(instr *ssa.CallCommon) (llvm.Value, erro
 	global.SetVisibility(llvm.HiddenVisibility)
 	global.SetGlobalConstant(true)
 	global.SetUnnamedAddr(true)
+	global.SetAlignment(1)
 	initializer := llvm.ConstNull(globalLLVMType)
 	initializer = b.CreateInsertValue(initializer, funcContext, 0, "")
 	initializer = b.CreateInsertValue(initializer, funcPtr, 1, "")

compiler/intrinsics.go

@@ -7,7 +7,6 @@ import (
 	"strconv"
 	"strings"
 
-	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"tinygo.org/x/go-llvm"
 )
 
@@ -24,6 +23,8 @@ func (b *builder) defineIntrinsicFunction() {
 		b.createMemoryCopyImpl()
 	case name == "runtime.memzero":
 		b.createMemoryZeroImpl()
+	case name == "runtime.stacksave":
+		b.createStackSaveImpl()
 	case name == "runtime.KeepAlive":
 		b.createKeepAliveImpl()
 	case strings.HasPrefix(name, "runtime/volatile.Load"):
@@ -48,12 +49,9 @@ func (b *builder) defineIntrinsicFunction() {
 func (b *builder) createMemoryCopyImpl() {
 	b.createFunctionStart(true)
 	fnName := "llvm." + b.fn.Name() + ".p0.p0.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
-	if llvmutil.Major() < 15 { // compatibility with LLVM 14
-		fnName = "llvm." + b.fn.Name() + ".p0i8.p0i8.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
-	}
 	llvmFn := b.mod.NamedFunction(fnName)
 	if llvmFn.IsNil() {
-		fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.i8ptrType, b.i8ptrType, b.uintptrType, b.ctx.Int1Type()}, false)
+		fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.dataPtrType, b.dataPtrType, b.uintptrType, b.ctx.Int1Type()}, false)
 		llvmFn = llvm.AddFunction(b.mod, fnName, fnType)
 	}
 	var params []llvm.Value
@@ -70,15 +68,7 @@ func (b *builder) createMemoryCopyImpl() {
 // regular libc memset calls if they aren't optimized out in a different way.
 func (b *builder) createMemoryZeroImpl() {
 	b.createFunctionStart(true)
-	fnName := "llvm.memset.p0.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
-	if llvmutil.Major() < 15 { // compatibility with LLVM 14
-		fnName = "llvm.memset.p0i8.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
-	}
-	llvmFn := b.mod.NamedFunction(fnName)
-	if llvmFn.IsNil() {
-		fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.i8ptrType, b.ctx.Int8Type(), b.uintptrType, b.ctx.Int1Type()}, false)
-		llvmFn = llvm.AddFunction(b.mod, fnName, fnType)
-	}
+	llvmFn := b.getMemsetFunc()
 	params := []llvm.Value{
 		b.getValue(b.fn.Params[0], getPos(b.fn)),
 		llvm.ConstInt(b.ctx.Int8Type(), 0, false),
@@ -89,6 +79,25 @@ func (b *builder) createMemoryZeroImpl() {
 	b.CreateRetVoid()
 }
 
+// createStackSaveImpl creates a call to llvm.stacksave.p0 to read the current
+// stack pointer.
+func (b *builder) createStackSaveImpl() {
+	b.createFunctionStart(true)
+	sp := b.readStackPointer()
+	b.CreateRet(sp)
+}
+
+// Return the llvm.memset.p0.i8 function declaration.
+func (c *compilerContext) getMemsetFunc() llvm.Value {
+	fnName := "llvm.memset.p0.i" + strconv.Itoa(c.uintptrType.IntTypeWidth())
+	llvmFn := c.mod.NamedFunction(fnName)
+	if llvmFn.IsNil() {
+		fnType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.dataPtrType, c.ctx.Int8Type(), c.uintptrType, c.ctx.Int1Type()}, false)
+		llvmFn = llvm.AddFunction(c.mod, fnName, fnType)
+	}
+	return llvmFn
+}
+
 // createKeepAlive creates the runtime.KeepAlive function. It is implemented
 // using inline assembly.
 func (b *builder) createKeepAliveImpl() {
@@ -105,7 +114,7 @@ func (b *builder) createKeepAliveImpl() {
 	//
 	// It should be portable to basically everything as the "r" register type
 	// exists basically everywhere.
-	asmType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.i8ptrType}, false)
+	asmType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.dataPtrType}, false)
 	asmFn := llvm.InlineAsm(asmType, "", "r", true, false, 0, false)
 	b.createCall(asmType, asmFn, []llvm.Value{pointerValue}, "")
 

compiler/llvm.go

@@ -7,6 +7,7 @@ import (
 	"math/big"
 	"strings"
 
+	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"tinygo.org/x/go-llvm"
 )
@@ -20,7 +21,7 @@ import (
 //
 // This is useful for creating temporary allocas for intrinsics. Don't forget to
 // end the lifetime using emitLifetimeEnd after you're done with it.
-func (b *builder) createTemporaryAlloca(t llvm.Type, name string) (alloca, bitcast, size llvm.Value) {
+func (b *builder) createTemporaryAlloca(t llvm.Type, name string) (alloca, size llvm.Value) {
 	return llvmutil.CreateTemporaryAlloca(b.Builder, b.mod, t, name)
 }
 
@@ -63,47 +64,45 @@ func (b *builder) emitPointerPack(values []llvm.Value) llvm.Value {
 	// Allocate memory for the packed data.
 	size := b.targetData.TypeAllocSize(packedType)
 	if size == 0 {
-		return llvm.ConstPointerNull(b.i8ptrType)
+		return llvm.ConstPointerNull(b.dataPtrType)
 	} else if len(values) == 1 && values[0].Type().TypeKind() == llvm.PointerTypeKind {
-		return b.CreateBitCast(values[0], b.i8ptrType, "pack.ptr")
-	} else if size <= b.targetData.TypeAllocSize(b.i8ptrType) {
+		return values[0]
+	} else if size <= b.targetData.TypeAllocSize(b.dataPtrType) {
 		// Packed data fits in a pointer, so store it directly inside the
 		// pointer.
 		if len(values) == 1 && values[0].Type().TypeKind() == llvm.IntegerTypeKind {
 			// Try to keep this cast in SSA form.
-			return b.CreateIntToPtr(values[0], b.i8ptrType, "pack.int")
+			return b.CreateIntToPtr(values[0], b.dataPtrType, "pack.int")
 		}
 
 		// Because packedType is a struct and we have to cast it to a *i8, store
 		// it in a *i8 alloca first and load the *i8 value from there. This is
 		// effectively a bitcast.
-		packedAlloc, _, _ := b.createTemporaryAlloca(b.i8ptrType, "")
+		packedAlloc, _ := b.createTemporaryAlloca(b.dataPtrType, "")
 
-		if size < b.targetData.TypeAllocSize(b.i8ptrType) {
+		if size < b.targetData.TypeAllocSize(b.dataPtrType) {
 			// The alloca is bigger than the value that will be stored in it.
 			// To avoid having some bits undefined, zero the alloca first.
 			// Hopefully this will get optimized away.
-			b.CreateStore(llvm.ConstNull(b.i8ptrType), packedAlloc)
+			b.CreateStore(llvm.ConstNull(b.dataPtrType), packedAlloc)
 		}
 
 		// Store all values in the alloca.
-		packedAllocCast := b.CreateBitCast(packedAlloc, llvm.PointerType(packedType, 0), "")
 		for i, value := range values {
 			indices := []llvm.Value{
 				llvm.ConstInt(b.ctx.Int32Type(), 0, false),
 				llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false),
 			}
-			gep := b.CreateInBoundsGEP(packedType, packedAllocCast, indices, "")
+			gep := b.CreateInBoundsGEP(packedType, packedAlloc, indices, "")
 			b.CreateStore(value, gep)
 		}
 
 		// Load value (the *i8) from the alloca.
-		result := b.CreateLoad(b.i8ptrType, packedAlloc, "")
+		result := b.CreateLoad(b.dataPtrType, packedAlloc, "")
 
 		// End the lifetime of the alloca, to help the optimizer.
-		packedPtr := b.CreateBitCast(packedAlloc, b.i8ptrType, "")
 		packedSize := llvm.ConstInt(b.ctx.Int64Type(), b.targetData.TypeAllocSize(packedAlloc.Type()), false)
-		b.emitLifetimeEnd(packedPtr, packedSize)
+		b.emitLifetimeEnd(packedAlloc, packedSize)
 
 		return result
 	} else {
@@ -124,21 +123,22 @@ func (b *builder) emitPointerPack(values []llvm.Value) llvm.Value {
 			global.SetGlobalConstant(true)
 			global.SetUnnamedAddr(true)
 			global.SetLinkage(llvm.InternalLinkage)
-			return llvm.ConstBitCast(global, b.i8ptrType)
+			return global
 		}
 
 		// Packed data is bigger than a pointer, so allocate it on the heap.
 		sizeValue := llvm.ConstInt(b.uintptrType, size, false)
+		align := b.targetData.ABITypeAlignment(packedType)
 		alloc := b.mod.NamedFunction("runtime.alloc")
-		packedHeapAlloc := b.CreateCall(alloc.GlobalValueType(), alloc, []llvm.Value{
+		packedAlloc := b.CreateCall(alloc.GlobalValueType(), alloc, []llvm.Value{
 			sizeValue,
-			llvm.ConstNull(b.i8ptrType),
-			llvm.Undef(b.i8ptrType), // unused context parameter
+			llvm.ConstNull(b.dataPtrType),
+			llvm.Undef(b.dataPtrType), // unused context parameter
 		}, "")
+		packedAlloc.AddCallSiteAttribute(0, b.ctx.CreateEnumAttribute(llvm.AttributeKindID("align"), uint64(align)))
 		if b.NeedsStackObjects {
-			b.trackPointer(packedHeapAlloc)
+			b.trackPointer(packedAlloc)
 		}
-		packedAlloc := b.CreateBitCast(packedHeapAlloc, llvm.PointerType(packedType, 0), "")
 
 		// Store all values in the heap pointer.
 		for i, value := range values {
@@ -151,7 +151,7 @@ func (b *builder) emitPointerPack(values []llvm.Value) llvm.Value {
 		}
 
 		// Return the original heap allocation pointer, which already is an *i8.
-		return packedHeapAlloc
+		return packedAlloc
 	}
 }
 
@@ -160,28 +160,27 @@ func (b *builder) emitPointerUnpack(ptr llvm.Value, valueTypes []llvm.Type) []ll
 	packedType := b.ctx.StructType(valueTypes, false)
 
 	// Get a correctly-typed pointer to the packed data.
-	var packedAlloc, packedRawAlloc llvm.Value
+	var packedAlloc llvm.Value
+	needsLifetimeEnd := false
 	size := b.targetData.TypeAllocSize(packedType)
 	if size == 0 {
 		// No data to unpack.
 	} else if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.PointerTypeKind {
 		// A single pointer is always stored directly.
-		return []llvm.Value{b.CreateBitCast(ptr, valueTypes[0], "unpack.ptr")}
-	} else if size <= b.targetData.TypeAllocSize(b.i8ptrType) {
+		return []llvm.Value{ptr}
+	} else if size <= b.targetData.TypeAllocSize(b.dataPtrType) {
 		// Packed data stored directly in pointer.
 		if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.IntegerTypeKind {
 			// Keep this cast in SSA form.
 			return []llvm.Value{b.CreatePtrToInt(ptr, valueTypes[0], "unpack.int")}
 		}
 		// Fallback: load it using an alloca.
-		packedRawAlloc, _, _ = b.createTemporaryAlloca(llvm.PointerType(b.i8ptrType, 0), "unpack.raw.alloc")
-		packedRawValue := b.CreateBitCast(ptr, llvm.PointerType(b.i8ptrType, 0), "unpack.raw.value")
-		b.CreateStore(packedRawValue, packedRawAlloc)
-		packedAlloc = b.CreateBitCast(packedRawAlloc, llvm.PointerType(packedType, 0), "unpack.alloc")
+		packedAlloc, _ = b.createTemporaryAlloca(b.dataPtrType, "unpack.raw.alloc")
+		b.CreateStore(ptr, packedAlloc)
+		needsLifetimeEnd = true
 	} else {
-		// Packed data stored on the heap. Bitcast the passed-in pointer to the
-		// correct pointer type.
-		packedAlloc = b.CreateBitCast(ptr, llvm.PointerType(packedType, 0), "unpack.raw.ptr")
+		// Packed data stored on the heap.
+		packedAlloc = ptr
 	}
 	// Load each value from the packed data.
 	values := make([]llvm.Value, len(valueTypes))
@@ -198,10 +197,9 @@ func (b *builder) emitPointerUnpack(ptr llvm.Value, valueTypes []llvm.Type) []ll
 		gep := b.CreateInBoundsGEP(packedType, packedAlloc, indices, "")
 		values[i] = b.CreateLoad(valueType, gep, "")
 	}
-	if !packedRawAlloc.IsNil() {
-		allocPtr := b.CreateBitCast(packedRawAlloc, b.i8ptrType, "")
+	if needsLifetimeEnd {
 		allocSize := llvm.ConstInt(b.ctx.Int64Type(), b.targetData.TypeAllocSize(b.uintptrType), false)
-		b.emitLifetimeEnd(allocPtr, allocSize)
+		b.emitLifetimeEnd(packedAlloc, allocSize)
 	}
 	return values
 }
@@ -253,12 +251,12 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 	// Do a few checks to see whether we need to generate any object layout
 	// information at all.
 	objectSizeBytes := c.targetData.TypeAllocSize(t)
-	pointerSize := c.targetData.TypeAllocSize(c.i8ptrType)
-	pointerAlignment := c.targetData.PrefTypeAlignment(c.i8ptrType)
+	pointerSize := c.targetData.TypeAllocSize(c.dataPtrType)
+	pointerAlignment := c.targetData.PrefTypeAlignment(c.dataPtrType)
 	if objectSizeBytes < pointerSize {
 		// Too small to contain a pointer.
 		layout := (uint64(1) << 1) | 1
-		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.i8ptrType)
+		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.dataPtrType)
 	}
 	bitmap := c.getPointerBitmap(t, pos)
 	if bitmap.BitLen() == 0 {
@@ -266,13 +264,13 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 		// TODO: this can be done in many other cases, e.g. when allocating an
 		// array (like [4][]byte, which repeats a slice 4 times).
 		layout := (uint64(1) << 1) | 1
-		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.i8ptrType)
+		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.dataPtrType)
 	}
 	if objectSizeBytes%uint64(pointerAlignment) != 0 {
 		// This shouldn't happen except for packed structs, which aren't
 		// currently used.
 		c.addError(pos, "internal error: unexpected object size for object with pointer field")
-		return llvm.ConstNull(c.i8ptrType)
+		return llvm.ConstNull(c.dataPtrType)
 	}
 	objectSizeWords := objectSizeBytes / uint64(pointerAlignment)
 
@@ -297,7 +295,7 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 		// The runtime knows that if the least significant bit of the pointer is
 		// set, the pointer contains the value itself.
 		layout := bitmap.Uint64()<<(sizeFieldBits+1) | (objectSizeWords << 1) | 1
-		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.i8ptrType)
+		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.dataPtrType)
 	}
 
 	// Unfortunately, the object layout is too big to fit in a pointer-sized
@@ -308,7 +306,7 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 	globalName := "runtime/gc.layout:" + fmt.Sprintf("%d-%0*x", objectSizeWords, (objectSizeWords+15)/16, bitmap)
 	global := c.mod.NamedGlobal(globalName)
 	if !global.IsNil() {
-		return llvm.ConstBitCast(global, c.i8ptrType)
+		return global
 	}
 
 	// Create the global initializer.
@@ -359,13 +357,13 @@ func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Va
 		global.AddMetadata(0, diglobal)
 	}
 
-	return llvm.ConstBitCast(global, c.i8ptrType)
+	return global
 }
 
 // getPointerBitmap scans the given LLVM type for pointers and sets bits in a
 // bigint at the word offset that contains a pointer. This scan is recursive.
 func (c *compilerContext) getPointerBitmap(typ llvm.Type, pos token.Pos) *big.Int {
-	alignment := c.targetData.PrefTypeAlignment(c.i8ptrType)
+	alignment := c.targetData.PrefTypeAlignment(c.dataPtrType)
 	switch typ.TypeKind() {
 	case llvm.IntegerTypeKind, llvm.FloatTypeKind, llvm.DoubleTypeKind:
 		return big.NewInt(0)
@@ -378,7 +376,7 @@ func (c *compilerContext) getPointerBitmap(typ llvm.Type, pos token.Pos) *big.In
 			// of type uintptr, but before the LowerFuncValues pass it actually
 			// contains a pointer (ptrtoint) to a global. This trips up the
 			// interp package. Therefore, make the id field a pointer for now.
-			typ = c.ctx.StructType([]llvm.Type{c.i8ptrType, c.i8ptrType}, false)
+			typ = c.ctx.StructType([]llvm.Type{c.dataPtrType, c.dataPtrType}, false)
 		}
 		for i, subtyp := range typ.StructElementTypes() {
 			subptrs := c.getPointerBitmap(subtyp, pos)
@@ -421,18 +419,11 @@ func (c *compilerContext) getPointerBitmap(typ llvm.Type, pos token.Pos) *big.In
 	}
 }
 
-// archFamily returns the archtecture from the LLVM triple but with some
+// archFamily returns the architecture from the LLVM triple but with some
 // architecture names ("armv6", "thumbv7m", etc) merged into a single
 // architecture name ("arm").
 func (c *compilerContext) archFamily() string {
-	arch := strings.Split(c.Triple, "-")[0]
-	if strings.HasPrefix(arch, "arm64") {
-		return "aarch64"
-	}
-	if strings.HasPrefix(arch, "arm") || strings.HasPrefix(arch, "thumb") {
-		return "arm"
-	}
-	return arch
+	return compileopts.CanonicalArchName(c.Triple)
 }
 
 // isThumb returns whether we're in ARM or in Thumb mode. It panics if the
@@ -456,10 +447,14 @@ func (c *compilerContext) isThumb() bool {
 // readStackPointer emits a LLVM intrinsic call that returns the current stack
 // pointer as an *i8.
 func (b *builder) readStackPointer() llvm.Value {
-	stacksave := b.mod.NamedFunction("llvm.stacksave")
+	name := "llvm.stacksave.p0"
+	if llvmutil.Version() < 18 {
+		name = "llvm.stacksave" // backwards compatibility with LLVM 17 and below
+	}
+	stacksave := b.mod.NamedFunction(name)
 	if stacksave.IsNil() {
-		fnType := llvm.FunctionType(b.i8ptrType, nil, false)
-		stacksave = llvm.AddFunction(b.mod, "llvm.stacksave", fnType)
+		fnType := llvm.FunctionType(b.dataPtrType, nil, false)
+		stacksave = llvm.AddFunction(b.mod, name, fnType)
 	}
 	return b.CreateCall(stacksave.GlobalValueType(), stacksave, nil, "")
 }

compiler/llvmutil/llvm.go

@@ -1,5 +1,5 @@
 // Package llvmutil contains utility functions used across multiple compiler
-// packages. For example, they may be used by both the compiler pacakge and
+// packages. For example, they may be used by both the compiler package and
 // transformation packages.
 //
 // Normally, utility packages are avoided. However, in this case, the utility
@@ -8,22 +8,13 @@
 package llvmutil
 
 import (
+	"encoding/binary"
 	"strconv"
 	"strings"
 
 	"tinygo.org/x/go-llvm"
 )
 
-// Major returns the LLVM major version.
-func Major() int {
-	llvmMajor, err := strconv.Atoi(strings.SplitN(llvm.Version, ".", 2)[0])
-	if err != nil {
-		// sanity check, should be unreachable
-		panic("could not parse LLVM version: " + err.Error())
-	}
-	return llvmMajor
-}
-
 // CreateEntryBlockAlloca creates a new alloca in the entry block, even though
 // the IR builder is located elsewhere. It assumes that the insert point is
 // at the end of the current block.
@@ -41,21 +32,19 @@ func CreateEntryBlockAlloca(builder llvm.Builder, t llvm.Type, name string) llvm
 }
 
 // CreateTemporaryAlloca creates a new alloca in the entry block and adds
-// lifetime start infromation in the IR signalling that the alloca won't be used
+// lifetime start information in the IR signalling that the alloca won't be used
 // before this point.
 //
 // This is useful for creating temporary allocas for intrinsics. Don't forget to
 // end the lifetime using emitLifetimeEnd after you're done with it.
-func CreateTemporaryAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type, name string) (alloca, bitcast, size llvm.Value) {
+func CreateTemporaryAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type, name string) (alloca, size llvm.Value) {
 	ctx := t.Context()
 	targetData := llvm.NewTargetData(mod.DataLayout())
 	defer targetData.Dispose()
-	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
 	alloca = CreateEntryBlockAlloca(builder, t, name)
-	bitcast = builder.CreateBitCast(alloca, i8ptrType, name+".bitcast")
 	size = llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(t), false)
 	fnType, fn := getLifetimeStartFunc(mod)
-	builder.CreateCall(fnType, fn, []llvm.Value{size, bitcast}, "")
+	builder.CreateCall(fnType, fn, []llvm.Value{size, alloca}, "")
 	return
 }
 
@@ -64,21 +53,19 @@ func CreateInstructionAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type,
 	ctx := mod.Context()
 	targetData := llvm.NewTargetData(mod.DataLayout())
 	defer targetData.Dispose()
-	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
 
 	alloca := CreateEntryBlockAlloca(builder, t, name)
 	builder.SetInsertPointBefore(inst)
-	bitcast := builder.CreateBitCast(alloca, i8ptrType, name+".bitcast")
 	size := llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(t), false)
 	fnType, fn := getLifetimeStartFunc(mod)
-	builder.CreateCall(fnType, fn, []llvm.Value{size, bitcast}, "")
+	builder.CreateCall(fnType, fn, []llvm.Value{size, alloca}, "")
 	if next := llvm.NextInstruction(inst); !next.IsNil() {
 		builder.SetInsertPointBefore(next)
 	} else {
 		builder.SetInsertPointAtEnd(inst.InstructionParent())
 	}
 	fnType, fn = getLifetimeEndFunc(mod)
-	builder.CreateCall(fnType, fn, []llvm.Value{size, bitcast}, "")
+	builder.CreateCall(fnType, fn, []llvm.Value{size, alloca}, "")
 	return alloca
 }
 
@@ -94,13 +81,10 @@ func EmitLifetimeEnd(builder llvm.Builder, mod llvm.Module, ptr, size llvm.Value
 // first if it doesn't exist yet.
 func getLifetimeStartFunc(mod llvm.Module) (llvm.Type, llvm.Value) {
 	fnName := "llvm.lifetime.start.p0"
-	if Major() < 15 { // compatibility with LLVM 14
-		fnName = "llvm.lifetime.start.p0i8"
-	}
 	fn := mod.NamedFunction(fnName)
 	ctx := mod.Context()
-	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
-	fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), i8ptrType}, false)
+	ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+	fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), ptrType}, false)
 	if fn.IsNil() {
 		fn = llvm.AddFunction(mod, fnName, fnType)
 	}
@@ -111,13 +95,10 @@ func getLifetimeStartFunc(mod llvm.Module) (llvm.Type, llvm.Value) {
 // first if it doesn't exist yet.
 func getLifetimeEndFunc(mod llvm.Module) (llvm.Type, llvm.Value) {
 	fnName := "llvm.lifetime.end.p0"
-	if Major() < 15 {
-		fnName = "llvm.lifetime.end.p0i8"
-	}
 	fn := mod.NamedFunction(fnName)
 	ctx := mod.Context()
-	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
-	fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), i8ptrType}, false)
+	ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+	fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), ptrType}, false)
 	if fn.IsNil() {
 		fn = llvm.AddFunction(mod, fnName, fnType)
 	}
@@ -196,7 +177,7 @@ func SplitBasicBlock(builder llvm.Builder, afterInst llvm.Value, insertAfter llv
 	return newBlock
 }
 
-// Append the given values to a global array like llvm.used. The global might
+// AppendToGlobal appends the given values to a global array like llvm.used. The global might
 // not exist yet. The values can be any pointer type, they will be cast to i8*.
 func AppendToGlobal(mod llvm.Module, globalName string, values ...llvm.Value) {
 	// Read the existing values in the llvm.used array (if it exists).
@@ -213,14 +194,36 @@ func AppendToGlobal(mod llvm.Module, globalName string, values ...llvm.Value) {
 	}
 
 	// Add the new values.
-	i8ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
+	ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
 	for _, value := range values {
-		usedValues = append(usedValues, llvm.ConstPointerCast(value, i8ptrType))
+		// Note: the bitcast is necessary to cast AVR function pointers to
+		// address space 0 pointer types.
+		usedValues = append(usedValues, llvm.ConstPointerCast(value, ptrType))
 	}
 
 	// Create a new array (with the old and new values).
-	usedInitializer := llvm.ConstArray(i8ptrType, usedValues)
+	usedInitializer := llvm.ConstArray(ptrType, usedValues)
 	used := llvm.AddGlobal(mod, usedInitializer.Type(), globalName)
 	used.SetInitializer(usedInitializer)
 	used.SetLinkage(llvm.AppendingLinkage)
 }
+
+// Version returns the LLVM major version.
+func Version() int {
+	majorStr := strings.Split(llvm.Version, ".")[0]
+	major, err := strconv.Atoi(majorStr)
+	if err != nil {
+		panic("unexpected error while parsing LLVM version: " + err.Error()) // should not happen
+	}
+	return major
+}
+
+// Return the byte order for the given target triple. Most targets are little
+// endian, but for example MIPS can be big-endian.
+func ByteOrder(target string) binary.ByteOrder {
+	if strings.HasPrefix(target, "mips-") {
+		return binary.BigEndian
+	} else {
+		return binary.LittleEndian
+	}
+}

compiler/map.go

@@ -6,17 +6,11 @@ import (
 	"go/token"
 	"go/types"
 
+	"github.com/tinygo-org/tinygo/src/tinygo"
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
 
-// constants for hashmap algorithms; must match src/runtime/hashmap.go
-const (
-	hashmapAlgorithmBinary = iota
-	hashmapAlgorithmString
-	hashmapAlgorithmInterface
-)
-
 // createMakeMap creates a new map object (runtime.hashmap) by allocating and
 // initializing an appropriately sized object.
 func (b *builder) createMakeMap(expr *ssa.MakeMap) (llvm.Value, error) {
@@ -24,20 +18,20 @@ func (b *builder) createMakeMap(expr *ssa.MakeMap) (llvm.Value, error) {
 	keyType := mapType.Key().Underlying()
 	llvmValueType := b.getLLVMType(mapType.Elem().Underlying())
 	var llvmKeyType llvm.Type
-	var alg uint64 // must match values in src/runtime/hashmap.go
+	var alg uint64
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// String keys.
 		llvmKeyType = b.getLLVMType(keyType)
-		alg = hashmapAlgorithmString
+		alg = uint64(tinygo.HashmapAlgorithmString)
 	} else if hashmapIsBinaryKey(keyType) {
 		// Trivially comparable keys.
 		llvmKeyType = b.getLLVMType(keyType)
-		alg = hashmapAlgorithmBinary
+		alg = uint64(tinygo.HashmapAlgorithmBinary)
 	} else {
 		// All other keys. Implemented as map[interface{}]valueType for ease of
 		// implementation.
 		llvmKeyType = b.getLLVMRuntimeType("_interface")
-		alg = hashmapAlgorithmInterface
+		alg = uint64(tinygo.HashmapAlgorithmInterface)
 	}
 	keySize := b.targetData.TypeAllocSize(llvmKeyType)
 	valueSize := b.targetData.TypeAllocSize(llvmValueType)
@@ -65,7 +59,7 @@ func (b *builder) createMapLookup(keyType, valueType types.Type, m, key llvm.Val
 	// Allocate the memory for the resulting type. Do not zero this memory: it
 	// will be zeroed by the hashmap get implementation if the key is not
 	// present in the map.
-	mapValueAlloca, mapValuePtr, mapValueAllocaSize := b.createTemporaryAlloca(llvmValueType, "hashmap.value")
+	mapValueAlloca, mapValueAllocaSize := b.createTemporaryAlloca(llvmValueType, "hashmap.value")
 
 	// We need the map size (with type uintptr) to pass to the hashmap*Get
 	// functions. This is necessary because those *Get functions are valid on
@@ -82,19 +76,19 @@ func (b *builder) createMapLookup(keyType, valueType types.Type, m, key llvm.Val
 	keyType = keyType.Underlying()
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
-		params := []llvm.Value{m, key, mapValuePtr, mapValueSize}
+		params := []llvm.Value{m, key, mapValueAlloca, mapValueSize}
 		commaOkValue = b.createRuntimeCall("hashmapStringGet", params, "")
 	} else if hashmapIsBinaryKey(keyType) {
 		// key can be compared with runtime.memequal
 		// Store the key in an alloca, in the entry block to avoid dynamic stack
 		// growth.
-		mapKeyAlloca, mapKeyPtr, mapKeySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
+		mapKeyAlloca, mapKeySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
 		b.CreateStore(key, mapKeyAlloca)
 		b.zeroUndefBytes(b.getLLVMType(keyType), mapKeyAlloca)
 		// Fetch the value from the hashmap.
-		params := []llvm.Value{m, mapKeyPtr, mapValuePtr, mapValueSize}
+		params := []llvm.Value{m, mapKeyAlloca, mapValueAlloca, mapValueSize}
 		commaOkValue = b.createRuntimeCall("hashmapBinaryGet", params, "")
-		b.emitLifetimeEnd(mapKeyPtr, mapKeySize)
+		b.emitLifetimeEnd(mapKeyAlloca, mapKeySize)
 	} else {
 		// Not trivially comparable using memcmp. Make it an interface instead.
 		itfKey := key
@@ -102,14 +96,14 @@ func (b *builder) createMapLookup(keyType, valueType types.Type, m, key llvm.Val
 			// Not already an interface, so convert it to an interface now.
 			itfKey = b.createMakeInterface(key, origKeyType, pos)
 		}
-		params := []llvm.Value{m, itfKey, mapValuePtr, mapValueSize}
+		params := []llvm.Value{m, itfKey, mapValueAlloca, mapValueSize}
 		commaOkValue = b.createRuntimeCall("hashmapInterfaceGet", params, "")
 	}
 
 	// Load the resulting value from the hashmap. The value is set to the zero
 	// value if the key doesn't exist in the hashmap.
 	mapValue := b.CreateLoad(llvmValueType, mapValueAlloca, "")
-	b.emitLifetimeEnd(mapValuePtr, mapValueAllocaSize)
+	b.emitLifetimeEnd(mapValueAlloca, mapValueAllocaSize)
 
 	if commaOk {
 		tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{llvmValueType, b.ctx.Int1Type()}, false))
@@ -124,22 +118,22 @@ func (b *builder) createMapLookup(keyType, valueType types.Type, m, key llvm.Val
 // createMapUpdate updates a map key to a given value, by creating an
 // appropriate runtime call.
 func (b *builder) createMapUpdate(keyType types.Type, m, key, value llvm.Value, pos token.Pos) {
-	valueAlloca, valuePtr, valueSize := b.createTemporaryAlloca(value.Type(), "hashmap.value")
+	valueAlloca, valueSize := b.createTemporaryAlloca(value.Type(), "hashmap.value")
 	b.CreateStore(value, valueAlloca)
 	origKeyType := keyType
 	keyType = keyType.Underlying()
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
-		params := []llvm.Value{m, key, valuePtr}
+		params := []llvm.Value{m, key, valueAlloca}
 		b.createRuntimeCall("hashmapStringSet", params, "")
 	} else if hashmapIsBinaryKey(keyType) {
 		// key can be compared with runtime.memequal
-		keyAlloca, keyPtr, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
+		keyAlloca, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
 		b.CreateStore(key, keyAlloca)
 		b.zeroUndefBytes(b.getLLVMType(keyType), keyAlloca)
-		params := []llvm.Value{m, keyPtr, valuePtr}
+		params := []llvm.Value{m, keyAlloca, valueAlloca}
 		b.createRuntimeCall("hashmapBinarySet", params, "")
-		b.emitLifetimeEnd(keyPtr, keySize)
+		b.emitLifetimeEnd(keyAlloca, keySize)
 	} else {
 		// Key is not trivially comparable, so compare it as an interface instead.
 		itfKey := key
@@ -147,10 +141,10 @@ func (b *builder) createMapUpdate(keyType types.Type, m, key, value llvm.Value,
 			// Not already an interface, so convert it to an interface first.
 			itfKey = b.createMakeInterface(key, origKeyType, pos)
 		}
-		params := []llvm.Value{m, itfKey, valuePtr}
+		params := []llvm.Value{m, itfKey, valueAlloca}
 		b.createRuntimeCall("hashmapInterfaceSet", params, "")
 	}
-	b.emitLifetimeEnd(valuePtr, valueSize)
+	b.emitLifetimeEnd(valueAlloca, valueSize)
 }
 
 // createMapDelete deletes a key from a map by calling the appropriate runtime
@@ -164,12 +158,12 @@ func (b *builder) createMapDelete(keyType types.Type, m, key llvm.Value, pos tok
 		b.createRuntimeCall("hashmapStringDelete", params, "")
 		return nil
 	} else if hashmapIsBinaryKey(keyType) {
-		keyAlloca, keyPtr, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
+		keyAlloca, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
 		b.CreateStore(key, keyAlloca)
 		b.zeroUndefBytes(b.getLLVMType(keyType), keyAlloca)
-		params := []llvm.Value{m, keyPtr}
+		params := []llvm.Value{m, keyAlloca}
 		b.createRuntimeCall("hashmapBinaryDelete", params, "")
-		b.emitLifetimeEnd(keyPtr, keySize)
+		b.emitLifetimeEnd(keyAlloca, keySize)
 		return nil
 	} else {
 		// Key is not trivially comparable, so compare it as an interface
@@ -185,6 +179,11 @@ func (b *builder) createMapDelete(keyType types.Type, m, key llvm.Value, pos tok
 	}
 }
 
+// Clear the given map.
+func (b *builder) createMapClear(m llvm.Value) {
+	b.createRuntimeCall("hashmapClear", []llvm.Value{m}, "")
+}
+
 // createMapIteratorNext lowers the *ssa.Next instruction for iterating over a
 // map. It returns a tuple of {bool, key, value} with the result of the
 // iteration.
@@ -220,9 +219,9 @@ func (b *builder) createMapIteratorNext(rangeVal ssa.Value, llvmRangeVal, it llv
 	}
 
 	// Extract the key and value from the map.
-	mapKeyAlloca, mapKeyPtr, mapKeySize := b.createTemporaryAlloca(llvmStoredKeyType, "range.key")
-	mapValueAlloca, mapValuePtr, mapValueSize := b.createTemporaryAlloca(llvmValueType, "range.value")
-	ok := b.createRuntimeCall("hashmapNext", []llvm.Value{llvmRangeVal, it, mapKeyPtr, mapValuePtr}, "range.next")
+	mapKeyAlloca, mapKeySize := b.createTemporaryAlloca(llvmStoredKeyType, "range.key")
+	mapValueAlloca, mapValueSize := b.createTemporaryAlloca(llvmValueType, "range.value")
+	ok := b.createRuntimeCall("hashmapNext", []llvm.Value{llvmRangeVal, it, mapKeyAlloca, mapValueAlloca}, "range.next")
 	mapKey := b.CreateLoad(llvmStoredKeyType, mapKeyAlloca, "")
 	mapValue := b.CreateLoad(llvmValueType, mapValueAlloca, "")
 
@@ -233,8 +232,8 @@ func (b *builder) createMapIteratorNext(rangeVal ssa.Value, llvmRangeVal, it llv
 	}
 
 	// End the lifetimes of the allocas, because we're done with them.
-	b.emitLifetimeEnd(mapKeyPtr, mapKeySize)
-	b.emitLifetimeEnd(mapValuePtr, mapValueSize)
+	b.emitLifetimeEnd(mapKeyAlloca, mapKeySize)
+	b.emitLifetimeEnd(mapValueAlloca, mapValueSize)
 
 	// Construct the *ssa.Next return value: {ok, mapKey, mapValue}
 	tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{b.ctx.Int1Type(), llvmKeyType, llvmValueType}, false))
@@ -321,21 +320,14 @@ func (b *builder) zeroUndefBytes(llvmType llvm.Type, ptr llvm.Value) error {
 			if i < numFields-1 {
 				nextOffset = b.targetData.ElementOffset(llvmStructType, i+1)
 			} else {
-				// Last field?  Next offset is the total size of the allcoate struct.
+				// Last field?  Next offset is the total size of the allocate struct.
 				nextOffset = b.targetData.TypeAllocSize(llvmStructType)
 			}
 
 			if fieldEndOffset != nextOffset {
 				n := llvm.ConstInt(b.uintptrType, nextOffset-fieldEndOffset, false)
 				llvmStoreSize := llvm.ConstInt(b.uintptrType, storeSize, false)
-				gepPtr := elemPtr
-				if gepPtr.Type() != b.i8ptrType {
-					gepPtr = b.CreateBitCast(gepPtr, b.i8ptrType, "") // LLVM 14
-				}
-				paddingStart := b.CreateInBoundsGEP(b.ctx.Int8Type(), gepPtr, []llvm.Value{llvmStoreSize}, "")
-				if paddingStart.Type() != b.i8ptrType {
-					paddingStart = b.CreateBitCast(paddingStart, b.i8ptrType, "") // LLVM 14
-				}
+				paddingStart := b.CreateInBoundsGEP(b.ctx.Int8Type(), elemPtr, []llvm.Value{llvmStoreSize}, "")
 				b.createRuntimeCall("memzero", []llvm.Value{paddingStart, n}, "")
 			}
 		}

compiler/symbol.go

@@ -12,6 +12,7 @@ import (
 	"strings"
 
 	"github.com/tinygo-org/tinygo/compiler/llvmutil"
+	"github.com/tinygo-org/tinygo/goenv"
 	"github.com/tinygo-org/tinygo/loader"
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
@@ -23,15 +24,18 @@ import (
 // The linkName value contains a valid link name, even if //go:linkname is not
 // present.
 type functionInfo struct {
-	module     string     // go:wasm-module
-	importName string     // go:linkname, go:export - The name the developer assigns
-	linkName   string     // go:linkname, go:export - The name that we map for the particular module -> importName
-	section    string     // go:section - object file section name
-	exported   bool       // go:export, CGo
-	interrupt  bool       // go:interrupt
-	nobounds   bool       // go:nobounds
-	variadic   bool       // go:variadic (CGo only)
-	inline     inlineType // go:inline
+	wasmModule    string     // go:wasm-module
+	wasmName      string     // wasm-export-name or wasm-import-name in the IR
+	wasmExport    string     // go:wasmexport is defined (export is unset, this adds an exported wrapper)
+	wasmExportPos token.Pos  // position of //go:wasmexport comment
+	linkName      string     // go:linkname, go:export - the IR function name
+	section       string     // go:section - object file section name
+	exported      bool       // go:export, CGo
+	interrupt     bool       // go:interrupt
+	nobounds      bool       // go:nobounds
+	noescape      bool       // go:noescape
+	variadic      bool       // go:variadic (CGo only)
+	inline        inlineType // go:inline
 }
 
 type inlineType int
@@ -96,7 +100,7 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 	// Add an extra parameter as the function context. This context is used in
 	// closures and bound methods, but should be optimized away when not used.
 	if !info.exported {
-		paramInfos = append(paramInfos, paramInfo{llvmType: c.i8ptrType, name: "context", elemSize: 0})
+		paramInfos = append(paramInfos, paramInfo{llvmType: c.dataPtrType, name: "context", elemSize: 0})
 	}
 
 	var paramTypes []llvm.Type
@@ -124,11 +128,20 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 	c.addStandardDeclaredAttributes(llvmFn)
 
 	dereferenceableOrNullKind := llvm.AttributeKindID("dereferenceable_or_null")
-	for i, info := range paramInfos {
-		if info.elemSize != 0 {
-			dereferenceableOrNull := c.ctx.CreateEnumAttribute(dereferenceableOrNullKind, info.elemSize)
+	for i, paramInfo := range paramInfos {
+		if paramInfo.elemSize != 0 {
+			dereferenceableOrNull := c.ctx.CreateEnumAttribute(dereferenceableOrNullKind, paramInfo.elemSize)
 			llvmFn.AddAttributeAtIndex(i+1, dereferenceableOrNull)
 		}
+		if info.noescape && paramInfo.flags&paramIsGoParam != 0 && paramInfo.llvmType.TypeKind() == llvm.PointerTypeKind {
+			// Parameters to functions with a //go:noescape parameter should get
+			// the nocapture attribute. However, the context parameter should
+			// not.
+			// (It may be safe to add the nocapture parameter to the context
+			// parameter, but I'd like to stay on the safe side here).
+			nocapture := c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0)
+			llvmFn.AddAttributeAtIndex(i+1, nocapture)
+		}
 	}
 
 	// Set a number of function or parameter attributes, depending on the
@@ -139,26 +152,26 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		// On *nix systems, the "abort" functuion in libc is used to handle fatal panics.
 		// Mark it as noreturn so LLVM can optimize away code.
 		llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("noreturn"), 0))
+	case "internal/abi.NoEscape":
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
 	case "runtime.alloc":
 		// Tell the optimizer that runtime.alloc is an allocator, meaning that it
 		// returns values that are never null and never alias to an existing value.
 		for _, attrName := range []string{"noalias", "nonnull"} {
 			llvmFn.AddAttributeAtIndex(0, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(attrName), 0))
 		}
-		if llvmutil.Major() >= 15 { // allockind etc are not available in LLVM 14
-			// Add attributes to signal to LLVM that this is an allocator
-			// function. This enables a number of optimizations.
-			llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("allockind"), allocKindAlloc|allocKindZeroed))
-			llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("alloc-family", "runtime.alloc"))
-			// Use a special value to indicate the first parameter:
-			// > allocsize has two integer arguments, but because they're both 32 bits, we can
-			// > pack them into one 64-bit value, at the cost of making said value
-			// > nonsensical.
-			// >
-			// > In order to do this, we need to reserve one value of the second (optional)
-			// > allocsize argument to signify "not present."
-			llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("allocsize"), 0x0000_0000_ffff_ffff))
-		}
+		// Add attributes to signal to LLVM that this is an allocator function.
+		// This enables a number of optimizations.
+		llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("allockind"), allocKindAlloc|allocKindZeroed))
+		llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("alloc-family", "runtime.alloc"))
+		// Use a special value to indicate the first parameter:
+		// > allocsize has two integer arguments, but because they're both 32 bits, we can
+		// > pack them into one 64-bit value, at the cost of making said value
+		// > nonsensical.
+		// >
+		// > In order to do this, we need to reserve one value of the second (optional)
+		// > allocsize argument to signify "not present."
+		llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("allocsize"), 0x0000_0000_ffff_ffff))
 	case "runtime.sliceAppend":
 		// Appending a slice will only read the to-be-appended slice, it won't
 		// be modified.
@@ -170,6 +183,12 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
 		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
 		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+	case "runtime.stringFromBytes":
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
+	case "runtime.stringFromRunes":
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
 	case "runtime.trackPointer":
 		// This function is necessary for tracking pointers on the stack in a
 		// portable way (see gc_stack_portable.go). Indicate to the optimizer
@@ -180,7 +199,7 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		if strings.Split(c.Triple, "-")[0] == "avr" {
 			// These functions are compiler-rt/libgcc functions that are
 			// currently implemented in Go. Assembly versions should appear in
-			// LLVM 16 hopefully. Until then, they need to be made available to
+			// LLVM 17 hopefully. Until then, they need to be made available to
 			// the linker and the best way to do that is llvm.compiler.used.
 			// I considered adding a pragma for this, but the LLVM language
 			// reference explicitly says that this feature should not be exposed
@@ -194,20 +213,14 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 	// External/exported functions may not retain pointer values.
 	// https://golang.org/cmd/cgo/#hdr-Passing_pointers
 	if info.exported {
-		if c.archFamily() == "wasm32" {
+		if c.archFamily() == "wasm32" && len(fn.Blocks) == 0 {
 			// We need to add the wasm-import-module and the wasm-import-name
 			// attributes.
-			module := info.module
-			if module == "" {
-				module = "env"
+			if info.wasmModule != "" {
+				llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("wasm-import-module", info.wasmModule))
 			}
-			llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("wasm-import-module", module))
 
-			name := info.importName
-			if name == "" {
-				name = info.linkName
-			}
-			llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("wasm-import-name", name))
+			llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("wasm-import-name", info.wasmName))
 		}
 		nocaptureKind := llvm.AttributeKindID("nocapture")
 		nocapture := c.ctx.CreateEnumAttribute(nocaptureKind, 0)
@@ -218,13 +231,26 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		}
 	}
 
+	// Build the function if needed.
+	c.maybeCreateSyntheticFunction(fn, llvmFn)
+
+	return fnType, llvmFn
+}
+
+// If this is a synthetic function (such as a generic function or a wrapper),
+// create it now.
+func (c *compilerContext) maybeCreateSyntheticFunction(fn *ssa.Function, llvmFn llvm.Value) {
 	// Synthetic functions are functions that do not appear in the source code,
 	// they are artificially constructed. Usually they are wrapper functions
 	// that are not referenced anywhere except in a SSA call instruction so
 	// should be created right away.
 	// The exception is the package initializer, which does appear in the
 	// *ssa.Package members and so shouldn't be created here.
-	if fn.Synthetic != "" && fn.Synthetic != "package initializer" && fn.Synthetic != "generic function" {
+	if fn.Synthetic != "" && fn.Synthetic != "package initializer" && fn.Synthetic != "generic function" && fn.Synthetic != "range-over-func yield" {
+		if len(fn.Blocks) == 0 {
+			c.addError(fn.Pos(), "missing function body")
+			return
+		}
 		irbuilder := c.ctx.NewBuilder()
 		b := newBuilder(c, irbuilder, fn)
 		b.createFunction()
@@ -232,8 +258,6 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		llvmFn.SetLinkage(llvm.LinkOnceODRLinkage)
 		llvmFn.SetUnnamedAddr(true)
 	}
-
-	return fnType, llvmFn
 }
 
 // getFunctionInfo returns information about a function that is not directly
@@ -247,8 +271,27 @@ func (c *compilerContext) getFunctionInfo(f *ssa.Function) functionInfo {
 		// Pick the default linkName.
 		linkName: f.RelString(nil),
 	}
+
+	// Check for a few runtime functions that are treated specially.
+	if info.linkName == "runtime.wasmEntryReactor" && c.BuildMode == "c-shared" {
+		info.linkName = "_initialize"
+		info.wasmName = "_initialize"
+		info.exported = true
+	}
+	if info.linkName == "runtime.wasmEntryCommand" && c.BuildMode == "default" {
+		info.linkName = "_start"
+		info.wasmName = "_start"
+		info.exported = true
+	}
+	if info.linkName == "runtime.wasmEntryLegacy" && c.BuildMode == "wasi-legacy" {
+		info.linkName = "_start"
+		info.wasmName = "_start"
+		info.exported = true
+	}
+
 	// Check for //go: pragmas, which may change the link name (among others).
 	c.parsePragmas(&info, f)
+
 	c.functionInfos[f] = info
 	return info
 }
@@ -256,163 +299,242 @@ func (c *compilerContext) getFunctionInfo(f *ssa.Function) functionInfo {
 // parsePragmas is used by getFunctionInfo to parse function pragmas such as
 // //export or //go:noinline.
 func (c *compilerContext) parsePragmas(info *functionInfo, f *ssa.Function) {
-	if f.Syntax() == nil {
+	syntax := f.Syntax()
+	if f.Origin() != nil {
+		syntax = f.Origin().Syntax()
+	}
+	if syntax == nil {
 		return
 	}
-	if decl, ok := f.Syntax().(*ast.FuncDecl); ok && decl.Doc != nil {
-
-		// Our importName for a wasm module (if we are compiling to wasm), or llvm link name
-		var importName string
 
+	// Read all pragmas of this function.
+	var pragmas []*ast.Comment
+	hasWasmExport := false
+	if decl, ok := syntax.(*ast.FuncDecl); ok && decl.Doc != nil {
 		for _, comment := range decl.Doc.List {
 			text := comment.Text
-			if strings.HasPrefix(text, "//export ") {
-				// Rewrite '//export' to '//go:export' for compatibility with
-				// gc.
-				text = "//go:" + text[2:]
+			if strings.HasPrefix(text, "//go:") || strings.HasPrefix(text, "//export ") {
+				pragmas = append(pragmas, comment)
+				if strings.HasPrefix(comment.Text, "//go:wasmexport ") {
+					hasWasmExport = true
+				}
 			}
-			if !strings.HasPrefix(text, "//go:") {
+		}
+	}
+
+	// Parse each pragma.
+	for _, comment := range pragmas {
+		parts := strings.Fields(comment.Text)
+		switch parts[0] {
+		case "//export", "//go:export":
+			if len(parts) != 2 {
+				continue
+			}
+			if hasWasmExport {
+				// //go:wasmexport overrides //export.
 				continue
 			}
-			parts := strings.Fields(text)
-			switch parts[0] {
-			case "//go:export":
-				if len(parts) != 2 {
-					continue
-				}
 
-				importName = parts[1]
-				info.exported = true
-			case "//go:interrupt":
-				if hasUnsafeImport(f.Pkg.Pkg) {
-					info.interrupt = true
-				}
-			case "//go:wasm-module":
-				// Alternative comment for setting the import module.
-				if len(parts) != 2 {
-					continue
-				}
-				info.module = parts[1]
-			case "//go:wasmimport":
-				// Import a WebAssembly function, for example a WASI function.
-				// Original proposal: https://github.com/golang/go/issues/38248
-				// Allow globally: https://github.com/golang/go/issues/59149
-				if len(parts) != 3 {
-					continue
-				}
-				c.checkWasmImport(f, comment.Text)
-				info.exported = true
-				info.module = parts[1]
-				info.importName = parts[2]
-			case "//go:inline":
-				info.inline = inlineHint
-			case "//go:noinline":
+			info.linkName = parts[1]
+			info.wasmName = info.linkName
+			info.exported = true
+		case "//go:interrupt":
+			if hasUnsafeImport(f.Pkg.Pkg) {
+				info.interrupt = true
+			}
+		case "//go:wasm-module":
+			// Alternative comment for setting the import module.
+			// This is deprecated, use //go:wasmimport instead.
+			if len(parts) != 2 {
+				continue
+			}
+			info.wasmModule = parts[1]
+		case "//go:wasmimport":
+			// Import a WebAssembly function, for example a WASI function.
+			// Original proposal: https://github.com/golang/go/issues/38248
+			// Allow globally: https://github.com/golang/go/issues/59149
+			if len(parts) != 3 {
+				continue
+			}
+			if f.Blocks != nil {
+				// Defined functions cannot be exported.
+				c.addError(f.Pos(), "can only use //go:wasmimport on declarations")
+				continue
+			}
+			c.checkWasmImportExport(f, comment.Text)
+			info.exported = true
+			info.wasmModule = parts[1]
+			info.wasmName = parts[2]
+		case "//go:wasmexport":
+			if f.Blocks == nil {
+				c.addError(f.Pos(), "can only use //go:wasmexport on definitions")
+				continue
+			}
+			if len(parts) != 2 {
+				c.addError(f.Pos(), fmt.Sprintf("expected one parameter to //go:wasmexport, not %d", len(parts)-1))
+				continue
+			}
+			name := parts[1]
+			if name == "_start" || name == "_initialize" {
+				c.addError(f.Pos(), fmt.Sprintf("//go:wasmexport does not allow %#v", name))
+				continue
+			}
+			if c.BuildMode != "c-shared" && f.RelString(nil) == "main.main" {
+				c.addError(f.Pos(), fmt.Sprintf("//go:wasmexport does not allow main.main to be exported with -buildmode=%s", c.BuildMode))
+				continue
+			}
+			if c.archFamily() != "wasm32" {
+				c.addError(f.Pos(), "//go:wasmexport is only supported on wasm")
+			}
+			c.checkWasmImportExport(f, comment.Text)
+			info.wasmExport = name
+			info.wasmExportPos = comment.Slash
+		case "//go:inline":
+			info.inline = inlineHint
+		case "//go:noinline":
+			info.inline = inlineNone
+		case "//go:linkname":
+			if len(parts) != 3 || parts[1] != f.Name() {
+				continue
+			}
+			// Only enable go:linkname when the package imports "unsafe".
+			// This is a slightly looser requirement than what gc uses: gc
+			// requires the file to import "unsafe", not the package as a
+			// whole.
+			if hasUnsafeImport(f.Pkg.Pkg) {
+				info.linkName = parts[2]
+			}
+		case "//go:section":
+			// Only enable go:section when the package imports "unsafe".
+			// go:section also implies go:noinline since inlining could
+			// move the code to a different section than that requested.
+			if len(parts) == 2 && hasUnsafeImport(f.Pkg.Pkg) {
+				info.section = parts[1]
 				info.inline = inlineNone
-			case "//go:linkname":
-				if len(parts) != 3 || parts[1] != f.Name() {
-					continue
-				}
-				// Only enable go:linkname when the package imports "unsafe".
-				// This is a slightly looser requirement than what gc uses: gc
-				// requires the file to import "unsafe", not the package as a
-				// whole.
-				if hasUnsafeImport(f.Pkg.Pkg) {
-					info.linkName = parts[2]
-				}
-			case "//go:section":
-				// Only enable go:section when the package imports "unsafe".
-				// go:section also implies go:noinline since inlining could
-				// move the code to a different section than that requested.
-				if len(parts) == 2 && hasUnsafeImport(f.Pkg.Pkg) {
-					info.section = parts[1]
-					info.inline = inlineNone
-				}
-			case "//go:nobounds":
-				// Skip bounds checking in this function. Useful for some
-				// runtime functions.
-				// This is somewhat dangerous and thus only imported in packages
-				// that import unsafe.
-				if hasUnsafeImport(f.Pkg.Pkg) {
-					info.nobounds = true
-				}
-			case "//go:variadic":
-				// The //go:variadic pragma is emitted by the CGo preprocessing
-				// pass for C variadic functions. This includes both explicit
-				// (with ...) and implicit (no parameters in signature)
-				// functions.
-				if strings.HasPrefix(f.Name(), "C.") {
-					// This prefix cannot naturally be created, it must have
-					// been created as a result of CGo preprocessing.
-					info.variadic = true
-				}
+			}
+		case "//go:nobounds":
+			// Skip bounds checking in this function. Useful for some
+			// runtime functions.
+			// This is somewhat dangerous and thus only imported in packages
+			// that import unsafe.
+			if hasUnsafeImport(f.Pkg.Pkg) {
+				info.nobounds = true
+			}
+		case "//go:noescape":
+			// Don't let pointer parameters escape.
+			// Following the upstream Go implementation, we only do this for
+			// declarations, not definitions.
+			if len(f.Blocks) == 0 {
+				info.noescape = true
+			}
+		case "//go:variadic":
+			// The //go:variadic pragma is emitted by the CGo preprocessing
+			// pass for C variadic functions. This includes both explicit
+			// (with ...) and implicit (no parameters in signature)
+			// functions.
+			if strings.HasPrefix(f.Name(), "_Cgo_") {
+				// This prefix was created as a result of CGo preprocessing.
+				info.variadic = true
 			}
 		}
-
-		// Set the importName for our exported function if we have one
-		if importName != "" {
-			if info.module == "" {
-				info.linkName = importName
-			} else {
-				// WebAssembly import
-				info.importName = importName
-			}
-		}
-
 	}
 }
 
-// Check whether this function cannot be used in //go:wasmimport. It will add an
-// error if this is the case.
+// Check whether this function can be used in //go:wasmimport or
+// //go:wasmexport. It will add an error if this is not the case.
 //
 // The list of allowed types is based on this proposal:
 // https://github.com/golang/go/issues/59149
-func (c *compilerContext) checkWasmImport(f *ssa.Function, pragma string) {
-	if c.pkg.Path() == "runtime" {
+func (c *compilerContext) checkWasmImportExport(f *ssa.Function, pragma string) {
+	if c.pkg.Path() == "runtime" || c.pkg.Path() == "syscall/js" || c.pkg.Path() == "syscall" {
 		// The runtime is a special case. Allow all kinds of parameters
 		// (importantly, including pointers).
 		return
 	}
-	if f.Blocks != nil {
-		// Defined functions cannot be exported.
-		c.addError(f.Pos(), fmt.Sprintf("can only use //go:wasmimport on declarations"))
-		return
-	}
 	if f.Signature.Results().Len() > 1 {
 		c.addError(f.Signature.Results().At(1).Pos(), fmt.Sprintf("%s: too many return values", pragma))
 	} else if f.Signature.Results().Len() == 1 {
 		result := f.Signature.Results().At(0)
-		if !isValidWasmType(result.Type(), true) {
+		if !c.isValidWasmType(result.Type(), siteResult) {
 			c.addError(result.Pos(), fmt.Sprintf("%s: unsupported result type %s", pragma, result.Type().String()))
 		}
 	}
 	for _, param := range f.Params {
 		// Check whether the type is allowed.
 		// Only a very limited number of types can be mapped to WebAssembly.
-		if !isValidWasmType(param.Type(), false) {
+		if !c.isValidWasmType(param.Type(), siteParam) {
 			c.addError(param.Pos(), fmt.Sprintf("%s: unsupported parameter type %s", pragma, param.Type().String()))
 		}
 	}
 }
 
-// Check whether the type maps directly to a WebAssembly type, according to:
+// Check whether the type maps directly to a WebAssembly type.
+//
+// This reflects the relaxed type restrictions proposed here (except for structs.HostLayout):
+// https://github.com/golang/go/issues/66984
+//
+// This previously reflected the additional restrictions documented here:
 // https://github.com/golang/go/issues/59149
-func isValidWasmType(typ types.Type, isReturn bool) bool {
+func (c *compilerContext) isValidWasmType(typ types.Type, site wasmSite) bool {
 	switch typ := typ.Underlying().(type) {
 	case *types.Basic:
 		switch typ.Kind() {
+		case types.Bool:
+			return true
+		case types.Int8, types.Uint8, types.Int16, types.Uint16:
+			return site == siteIndirect
 		case types.Int32, types.Uint32, types.Int64, types.Uint64:
 			return true
 		case types.Float32, types.Float64:
 			return true
-		case types.UnsafePointer:
-			if !isReturn {
-				return true
+		case types.Uintptr, types.UnsafePointer:
+			return true
+		case types.String:
+			// string flattens to two values, so disallowed as a result
+			return site == siteParam || site == siteIndirect
+		}
+	case *types.Array:
+		return site == siteIndirect && c.isValidWasmType(typ.Elem(), siteIndirect)
+	case *types.Struct:
+		if site != siteIndirect {
+			return false
+		}
+		// Structs with no fields do not need structs.HostLayout
+		if typ.NumFields() == 0 {
+			return true
+		}
+		hasHostLayout := true // default to true before detecting Go version
+		// (*types.Package).GoVersion added in go1.21
+		if gv, ok := any(c.pkg).(interface{ GoVersion() string }); ok {
+			if goenv.Compare(gv.GoVersion(), "go1.23") >= 0 {
+				hasHostLayout = false // package structs added in go1.23
 			}
 		}
+		for i := 0; i < typ.NumFields(); i++ {
+			ftyp := typ.Field(i).Type()
+			if ftyp.String() == "structs.HostLayout" {
+				hasHostLayout = true
+				continue
+			}
+			if !c.isValidWasmType(ftyp, siteIndirect) {
+				return false
+			}
+		}
+		return hasHostLayout
+	case *types.Pointer:
+		return c.isValidWasmType(typ.Elem(), siteIndirect)
 	}
 	return false
 }
 
+type wasmSite int
+
+const (
+	siteParam wasmSite = iota
+	siteResult
+	siteIndirect // pointer or field
+)
+
 // getParams returns the function parameters, including the receiver at the
 // start. This is an alternative to the Params member of *ssa.Function, which is
 // not yet populated when the package has not yet been built.
@@ -464,15 +586,10 @@ func (c *compilerContext) addStandardDefinedAttributes(llvmFn llvm.Value) {
 	llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nounwind"), 0))
 	if strings.Split(c.Triple, "-")[0] == "x86_64" {
 		// Required by the ABI.
-		if llvmutil.Major() < 15 {
-			// Needed for LLVM 14 support.
-			llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("uwtable"), 0))
-		} else {
-			// The uwtable has two possible values: sync (1) or async (2). We
-			// use sync because we currently don't use async unwind tables.
-			// For details, see: https://llvm.org/docs/LangRef.html#function-attributes
-			llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("uwtable"), 1))
-		}
+		// The uwtable has two possible values: sync (1) or async (2). We use
+		// sync because we currently don't use async unwind tables.
+		// For details, see: https://llvm.org/docs/LangRef.html#function-attributes
+		llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("uwtable"), 1))
 	}
 }
 

compiler/syscall.go

@@ -5,6 +5,7 @@ package compiler
 
 import (
 	"strconv"
+	"strings"
 
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
@@ -12,7 +13,8 @@ import (
 
 // createRawSyscall creates a system call with the provided system call number
 // and returns the result as a single integer (the system call result). The
-// result is not further interpreted.
+// result is not further interpreted (with the exception of MIPS to use the same
+// return value everywhere).
 func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 	num := b.getValue(call.Args[0], getPos(call))
 	switch {
@@ -33,18 +35,17 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 				"{r10}",
 				"{r8}",
 				"{r9}",
-				"{r11}",
-				"{r12}",
-				"{r13}",
 			}[i]
 			llvmValue := b.getValue(arg, getPos(call))
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
+		// rcx and r11 are clobbered by the syscall, so make sure they are not used
 		constraints += ",~{rcx},~{r11}"
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "syscall", constraints, true, false, llvm.InlineAsmDialectIntel, false)
 		return b.CreateCall(fnType, target, args, ""), nil
+
 	case b.GOARCH == "386" && b.GOOS == "linux":
 		// Sources:
 		//   syscall(2) man page
@@ -71,6 +72,7 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "int 0x80", constraints, true, false, llvm.InlineAsmDialectIntel, false)
 		return b.CreateCall(fnType, target, args, ""), nil
+
 	case b.GOARCH == "arm" && b.GOOS == "linux":
 		// Implement the EABI system call convention for Linux.
 		// Source: syscall(2) man page.
@@ -103,6 +105,7 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "svc #0", constraints, true, false, 0, false)
 		return b.CreateCall(fnType, target, args, ""), nil
+
 	case b.GOARCH == "arm64" && b.GOOS == "linux":
 		// Source: syscall(2) man page.
 		args := []llvm.Value{}
@@ -135,6 +138,98 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "svc #0", constraints, true, false, 0, false)
 		return b.CreateCall(fnType, target, args, ""), nil
+
+	case (b.GOARCH == "mips" || b.GOARCH == "mipsle") && b.GOOS == "linux":
+		// Implement the system call convention for Linux.
+		// Source: syscall(2) man page and musl:
+		// https://git.musl-libc.org/cgit/musl/tree/arch/mips/syscall_arch.h
+		// Also useful:
+		// https://web.archive.org/web/20220529105937/https://www.linux-mips.org/wiki/Syscall
+		// The syscall number goes in r2, the result also in r2.
+		// Register r7 is both an input parameter and an output parameter: if it
+		// is non-zero, the system call failed and r2 is the error code.
+		// The code below implements the O32 syscall ABI, not the N32 ABI. It
+		// could implement both at the same time if needed (like what appears to
+		// be done in musl) by forcing arg5-arg7 into the right registers but
+		// letting the compiler decide the registers should result in _slightly_
+		// faster and smaller code.
+		args := []llvm.Value{num}
+		argTypes := []llvm.Type{b.uintptrType}
+		constraints := "={$2},={$7},0"
+		syscallParams := call.Args[1:]
+		if len(syscallParams) > 7 {
+			// There is one syscall that uses 7 parameters: sync_file_range.
+			// But only 7, not more. Go however only has Syscall6 and Syscall9.
+			// Therefore, we can ignore the remaining parameters.
+			syscallParams = syscallParams[:7]
+		}
+		for i, arg := range syscallParams {
+			constraints += "," + [...]string{
+				"{$4}", // arg1
+				"{$5}", // arg2
+				"{$6}", // arg3
+				"1",    // arg4, error return
+				"r",    // arg5 on the stack
+				"r",    // arg6 on the stack
+				"r",    // arg7 on the stack
+			}[i]
+			llvmValue := b.getValue(arg, getPos(call))
+			args = append(args, llvmValue)
+			argTypes = append(argTypes, llvmValue.Type())
+		}
+		// Create assembly code.
+		// Parameters beyond the first 4 are passed on the stack instead of in
+		// registers in the O32 syscall ABI.
+		// We need ".set noat" because LLVM might pick register $1 ($at) as the
+		// register for a parameter and apparently this is not allowed on MIPS
+		// unless you use this specific pragma.
+		asm := "syscall"
+		switch len(syscallParams) {
+		case 5:
+			asm = "" +
+				".set noat\n" +
+				"subu $$sp, $$sp, 32\n" +
+				"sw $7, 16($$sp)\n" + // arg5
+				"syscall\n" +
+				"addu $$sp, $$sp, 32\n" +
+				".set at\n"
+		case 6:
+			asm = "" +
+				".set noat\n" +
+				"subu $$sp, $$sp, 32\n" +
+				"sw $7, 16($$sp)\n" + // arg5
+				"sw $8, 20($$sp)\n" + // arg6
+				"syscall\n" +
+				"addu $$sp, $$sp, 32\n" +
+				".set at\n"
+		case 7:
+			asm = "" +
+				".set noat\n" +
+				"subu $$sp, $$sp, 32\n" +
+				"sw $7, 16($$sp)\n" + // arg5
+				"sw $8, 20($$sp)\n" + // arg6
+				"sw $9, 24($$sp)\n" + // arg7
+				"syscall\n" +
+				"addu $$sp, $$sp, 32\n" +
+				".set at\n"
+		}
+		constraints += ",~{$3},~{$4},~{$5},~{$6},~{$8},~{$9},~{$10},~{$11},~{$12},~{$13},~{$14},~{$15},~{$24},~{$25},~{hi},~{lo},~{memory}"
+		returnType := b.ctx.StructType([]llvm.Type{b.uintptrType, b.uintptrType}, false)
+		fnType := llvm.FunctionType(returnType, argTypes, false)
+		target := llvm.InlineAsm(fnType, asm, constraints, true, true, 0, false)
+		call := b.CreateCall(fnType, target, args, "")
+		resultCode := b.CreateExtractValue(call, 0, "") // r2
+		errorFlag := b.CreateExtractValue(call, 1, "")  // r7
+		// Pseudocode to return the result with the same convention as other
+		// archs:
+		//    return (errorFlag != 0) ? -resultCode : resultCode;
+		// At least on QEMU with the O32 ABI, the error code is always positive.
+		zero := llvm.ConstInt(b.uintptrType, 0, false)
+		isError := b.CreateICmp(llvm.IntNE, errorFlag, zero, "")
+		negativeResult := b.CreateSub(zero, resultCode, "")
+		result := b.CreateSelect(isError, negativeResult, resultCode, "")
+		return result, nil
+
 	default:
 		return llvm.Value{}, b.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+b.GOOS+"/"+b.GOARCH)
 	}
@@ -183,7 +278,7 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		}
 		llvmType := llvm.FunctionType(b.uintptrType, paramTypes, false)
 		fn := b.getValue(call.Args[0], getPos(call))
-		fnPtr := b.CreateIntToPtr(fn, llvm.PointerType(llvmType, 0), "")
+		fnPtr := b.CreateIntToPtr(fn, b.dataPtrType, "")
 
 		// Prepare some functions that will be called later.
 		setLastError := b.mod.NamedFunction("SetLastError")
@@ -217,6 +312,7 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		retval = b.CreateInsertValue(retval, syscallResult, 0, "")
 		retval = b.CreateInsertValue(retval, errResult, 2, "")
 		return retval, nil
+
 	default:
 		return llvm.Value{}, b.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+b.GOOS+"/"+b.GOARCH)
 	}
@@ -234,3 +330,64 @@ func (b *builder) createRawSyscallNoError(call *ssa.CallCommon) (llvm.Value, err
 	retval = b.CreateInsertValue(retval, llvm.ConstInt(b.uintptrType, 0, false), 1, "")
 	return retval, nil
 }
+
+// Lower a call to internal/abi.FuncPCABI0 on MacOS.
+// This function is called like this:
+//
+//	syscall(abi.FuncPCABI0(libc_mkdir_trampoline), uintptr(unsafe.Pointer(_p0)), uintptr(mode), 0)
+//
+// So we'll want to return a function pointer (as uintptr) that points to the
+// libc function. Specifically, we _don't_ want to point to the trampoline
+// function (which is implemented in Go assembly which we can't read), but
+// rather to the actually intended function. For this we're going to assume that
+// all the functions follow a specific pattern: libc_<functionname>_trampoline.
+//
+// The return value is the function pointer as an uintptr, or a nil value if
+// this isn't possible (and a regular call should be made as fallback).
+func (b *builder) createDarwinFuncPCABI0Call(instr *ssa.CallCommon) llvm.Value {
+	if b.GOOS != "darwin" {
+		// This has only been tested on MacOS (and only seems to be used there).
+		return llvm.Value{}
+	}
+
+	// Check that it uses a function call like syscall.libc_*_trampoline
+	itf := instr.Args[0].(*ssa.MakeInterface)
+	calledFn := itf.X.(*ssa.Function)
+	if pkgName := calledFn.Pkg.Pkg.Path(); pkgName != "syscall" && pkgName != "internal/syscall/unix" {
+		return llvm.Value{}
+	}
+	if !strings.HasPrefix(calledFn.Name(), "libc_") || !strings.HasSuffix(calledFn.Name(), "_trampoline") {
+
+		return llvm.Value{}
+	}
+
+	// Extract the libc function name.
+	name := strings.TrimPrefix(strings.TrimSuffix(calledFn.Name(), "_trampoline"), "libc_")
+	if name == "open" {
+		// Special case: open() is a variadic function and can't be called like
+		// a regular function. Therefore, we need to use a wrapper implemented
+		// in C.
+		name = "syscall_libc_open"
+	}
+	if b.GOARCH == "amd64" {
+		if name == "fdopendir" || name == "readdir_r" {
+			// Hack to support amd64, which needs the $INODE64 suffix.
+			// This is also done in upstream Go:
+			// https://github.com/golang/go/commit/096ab3c21b88ccc7d411379d09fe6274e3159467
+			name += "$INODE64"
+		}
+	}
+
+	// Obtain the C function.
+	// Use a simple function (no parameters or return value) because all we need
+	// is the address of the function.
+	llvmFn := b.mod.NamedFunction(name)
+	if llvmFn.IsNil() {
+		llvmFnType := llvm.FunctionType(b.ctx.VoidType(), nil, false)
+		llvmFn = llvm.AddFunction(b.mod, name, llvmFnType)
+	}
+
+	// Cast the function pointer to a uintptr (because that's what
+	// abi.FuncPCABI0 returns).
+	return b.CreatePtrToInt(llvmFn, b.uintptrType, "")
+}

compiler/testdata/basic.ll

@@ -196,10 +196,6 @@ entry:
 ; Function Attrs: nounwind
 define hidden void @main.foo(ptr %context) unnamed_addr #2 {
 entry:
-  %complit = alloca %main.kv.0, align 8
-  %stackalloc = alloca i8, align 1
-  store %main.kv.0 zeroinitializer, ptr %complit, align 8
-  call void @runtime.trackPointer(ptr nonnull %complit, ptr nonnull %stackalloc, ptr undef) #3
   call void @"main.foo$1"(%main.kv.0 zeroinitializer, ptr undef)
   ret void
 }
@@ -207,15 +203,10 @@ entry:
 ; Function Attrs: nounwind
 define internal void @"main.foo$1"(%main.kv.0 %b, ptr %context) unnamed_addr #2 {
 entry:
-  %b1 = alloca %main.kv.0, align 8
-  %stackalloc = alloca i8, align 1
-  store %main.kv.0 zeroinitializer, ptr %b1, align 8
-  call void @runtime.trackPointer(ptr nonnull %b1, ptr nonnull %stackalloc, ptr undef) #3
-  store %main.kv.0 %b, ptr %b1, align 8
   ret void
 }
 
-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #1 = { "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
+attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
 attributes #3 = { nounwind }

compiler/testdata/channel.ll

@@ -3,7 +3,7 @@ source_filename = "channel.go"
 target datalayout = "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-n32:64-S128-ni:1:10:20"
 target triple = "wasm32-unknown-wasi"
 
-%runtime.channelBlockedList = type { ptr, ptr, ptr, { ptr, i32, i32 } }
+%runtime.channelOp = type { ptr, ptr, i32, ptr }
 %runtime.chanSelectState = type { ptr, ptr }
 
 ; Function Attrs: allockind("alloc,zeroed") allocsize(0)
@@ -18,80 +18,77 @@ entry:
 }
 
 ; Function Attrs: nounwind
-define hidden void @main.chanIntSend(ptr dereferenceable_or_null(32) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanIntSend(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
 entry:
-  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.op = alloca %runtime.channelOp, align 8
   %chan.value = alloca i32, align 4
   call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %chan.value)
   store i32 3, ptr %chan.value, align 4
-  call void @llvm.lifetime.start.p0(i64 24, ptr nonnull %chan.blockedList)
-  call void @runtime.chanSend(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.blockedList, ptr undef) #4
-  call void @llvm.lifetime.end.p0(i64 24, ptr nonnull %chan.blockedList)
+  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
+  call void @runtime.chanSend(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.op, ptr undef) #4
+  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
   call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %chan.value)
   ret void
 }
 
-; Function Attrs: argmemonly nocallback nofree nosync nounwind willreturn
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
 declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #3
 
-declare void @runtime.chanSend(ptr dereferenceable_or_null(32), ptr, ptr dereferenceable_or_null(24), ptr) #1
+declare void @runtime.chanSend(ptr dereferenceable_or_null(36), ptr, ptr dereferenceable_or_null(16), ptr) #1
 
-; Function Attrs: argmemonly nocallback nofree nosync nounwind willreturn
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
 declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #3
 
 ; Function Attrs: nounwind
-define hidden void @main.chanIntRecv(ptr dereferenceable_or_null(32) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanIntRecv(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
 entry:
-  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.op = alloca %runtime.channelOp, align 8
   %chan.value = alloca i32, align 4
   call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %chan.value)
-  call void @llvm.lifetime.start.p0(i64 24, ptr nonnull %chan.blockedList)
-  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.blockedList, ptr undef) #4
+  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
+  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.op, ptr undef) #4
   call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %chan.value)
-  call void @llvm.lifetime.end.p0(i64 24, ptr nonnull %chan.blockedList)
+  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
   ret void
 }
 
-declare i1 @runtime.chanRecv(ptr dereferenceable_or_null(32), ptr, ptr dereferenceable_or_null(24), ptr) #1
+declare i1 @runtime.chanRecv(ptr dereferenceable_or_null(36), ptr, ptr dereferenceable_or_null(16), ptr) #1
 
 ; Function Attrs: nounwind
-define hidden void @main.chanZeroSend(ptr dereferenceable_or_null(32) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanZeroSend(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
 entry:
-  %complit = alloca {}, align 8
-  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
-  %stackalloc = alloca i8, align 1
-  call void @runtime.trackPointer(ptr nonnull %complit, ptr nonnull %stackalloc, ptr undef) #4
-  call void @llvm.lifetime.start.p0(i64 24, ptr nonnull %chan.blockedList)
-  call void @runtime.chanSend(ptr %ch, ptr null, ptr nonnull %chan.blockedList, ptr undef) #4
-  call void @llvm.lifetime.end.p0(i64 24, ptr nonnull %chan.blockedList)
+  %chan.op = alloca %runtime.channelOp, align 8
+  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
+  call void @runtime.chanSend(ptr %ch, ptr null, ptr nonnull %chan.op, ptr undef) #4
+  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
   ret void
 }
 
 ; Function Attrs: nounwind
-define hidden void @main.chanZeroRecv(ptr dereferenceable_or_null(32) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanZeroRecv(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
 entry:
-  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
-  call void @llvm.lifetime.start.p0(i64 24, ptr nonnull %chan.blockedList)
-  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr null, ptr nonnull %chan.blockedList, ptr undef) #4
-  call void @llvm.lifetime.end.p0(i64 24, ptr nonnull %chan.blockedList)
+  %chan.op = alloca %runtime.channelOp, align 8
+  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
+  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr null, ptr nonnull %chan.op, ptr undef) #4
+  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
   ret void
 }
 
 ; Function Attrs: nounwind
-define hidden void @main.selectZeroRecv(ptr dereferenceable_or_null(32) %ch1, ptr dereferenceable_or_null(32) %ch2, ptr %context) unnamed_addr #2 {
+define hidden void @main.selectZeroRecv(ptr dereferenceable_or_null(36) %ch1, ptr dereferenceable_or_null(36) %ch2, ptr %context) unnamed_addr #2 {
 entry:
   %select.states.alloca = alloca [2 x %runtime.chanSelectState], align 8
   %select.send.value = alloca i32, align 4
   store i32 1, ptr %select.send.value, align 4
   call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %select.states.alloca)
-  store ptr %ch1, ptr %select.states.alloca, align 8
-  %select.states.alloca.repack1 = getelementptr inbounds %runtime.chanSelectState, ptr %select.states.alloca, i32 0, i32 1
+  store ptr %ch1, ptr %select.states.alloca, align 4
+  %select.states.alloca.repack1 = getelementptr inbounds i8, ptr %select.states.alloca, i32 4
   store ptr %select.send.value, ptr %select.states.alloca.repack1, align 4
-  %0 = getelementptr inbounds [2 x %runtime.chanSelectState], ptr %select.states.alloca, i32 0, i32 1
-  store ptr %ch2, ptr %0, align 8
-  %.repack3 = getelementptr inbounds [2 x %runtime.chanSelectState], ptr %select.states.alloca, i32 0, i32 1, i32 1
+  %0 = getelementptr inbounds i8, ptr %select.states.alloca, i32 8
+  store ptr %ch2, ptr %0, align 4
+  %.repack3 = getelementptr inbounds i8, ptr %select.states.alloca, i32 12
   store ptr null, ptr %.repack3, align 4
-  %select.result = call { i32, i1 } @runtime.tryChanSelect(ptr undef, ptr nonnull %select.states.alloca, i32 2, i32 2, ptr undef) #4
+  %select.result = call { i32, i1 } @runtime.chanSelect(ptr undef, ptr nonnull %select.states.alloca, i32 2, i32 2, ptr null, i32 0, i32 0, ptr undef) #4
   call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %select.states.alloca)
   %1 = extractvalue { i32, i1 } %select.result, 0
   %2 = icmp eq i32 %1, 0
@@ -108,10 +105,10 @@ select.body:                                      ; preds = %select.next
   br label %select.done
 }
 
-declare { i32, i1 } @runtime.tryChanSelect(ptr, ptr, i32, i32, ptr) #1
+declare { i32, i1 } @runtime.chanSelect(ptr, ptr, i32, i32, ptr, i32, i32, ptr) #1
 
-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #1 = { "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #3 = { argmemonly nocallback nofree nosync nounwind willreturn }
+attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #3 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
 attributes #4 = { nounwind }

compiler/testdata/defer-cortex-m-qemu.ll

@@ -3,9 +3,8 @@ source_filename = "defer.go"
 target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
 target triple = "thumbv7m-unknown-unknown-eabi"
 
-%runtime.deferFrame = type { ptr, ptr, [0 x ptr], ptr, i1, %runtime._interface }
+%runtime.deferFrame = type { ptr, ptr, [0 x ptr], ptr, i8, %runtime._interface }
 %runtime._interface = type { ptr, ptr }
-%runtime._defer = type { i32, ptr }
 
 ; Function Attrs: allockind("alloc,zeroed") allocsize(0)
 declare noalias nonnull ptr @runtime.alloc(i32, ptr, ptr) #0
@@ -25,10 +24,10 @@ entry:
   %deferPtr = alloca ptr, align 4
   store ptr null, ptr %deferPtr, align 4
   %deferframe.buf = alloca %runtime.deferFrame, align 4
-  %0 = call ptr @llvm.stacksave()
+  %0 = call ptr @llvm.stacksave.p0()
   call void @runtime.setupDeferFrame(ptr nonnull %deferframe.buf, ptr %0, ptr undef) #4
   store i32 0, ptr %defer.alloca, align 4
-  %defer.alloca.repack15 = getelementptr inbounds { i32, ptr }, ptr %defer.alloca, i32 0, i32 1
+  %defer.alloca.repack15 = getelementptr inbounds i8, ptr %defer.alloca, i32 4
   store ptr null, ptr %defer.alloca.repack15, align 4
   store ptr %defer.alloca, ptr %deferPtr, align 4
   %setjmp = call i32 asm "\0Amovs r0, #0\0Amov r2, pc\0Astr r2, [r1, #4]", "={r0},{r1},~{r1},~{r2},~{r3},~{r4},~{r5},~{r6},~{r7},~{r8},~{r9},~{r10},~{r11},~{r12},~{lr},~{q0},~{q1},~{q2},~{q3},~{q4},~{q5},~{q6},~{q7},~{q8},~{q9},~{q10},~{q11},~{q12},~{q13},~{q14},~{q15},~{cpsr},~{memory}"(ptr nonnull %deferframe.buf) #5
@@ -52,7 +51,7 @@ rundefers.loophead:                               ; preds = %3, %rundefers.block
   br i1 %stackIsNil, label %rundefers.end, label %rundefers.loop
 
 rundefers.loop:                                   ; preds = %rundefers.loophead
-  %stack.next.gep = getelementptr inbounds %runtime._defer, ptr %2, i32 0, i32 1
+  %stack.next.gep = getelementptr inbounds i8, ptr %2, i32 4
   %stack.next = load ptr, ptr %stack.next.gep, align 4
   store ptr %stack.next, ptr %deferPtr, align 4
   %callback = load i32, ptr %2, align 4
@@ -88,7 +87,7 @@ rundefers.loophead6:                              ; preds = %5, %lpad
   br i1 %stackIsNil7, label %rundefers.end3, label %rundefers.loop5
 
 rundefers.loop5:                                  ; preds = %rundefers.loophead6
-  %stack.next.gep8 = getelementptr inbounds %runtime._defer, ptr %4, i32 0, i32 1
+  %stack.next.gep8 = getelementptr inbounds i8, ptr %4, i32 4
   %stack.next9 = load ptr, ptr %stack.next.gep8, align 4
   store ptr %stack.next9, ptr %deferPtr, align 4
   %callback11 = load i32, ptr %4, align 4
@@ -113,7 +112,7 @@ rundefers.end3:                                   ; preds = %rundefers.loophead6
 }
 
 ; Function Attrs: nocallback nofree nosync nounwind willreturn
-declare ptr @llvm.stacksave() #3
+declare ptr @llvm.stacksave.p0() #3
 
 declare void @runtime.setupDeferFrame(ptr dereferenceable_or_null(24), ptr, ptr) #2
 
@@ -122,12 +121,18 @@ declare void @runtime.destroyDeferFrame(ptr dereferenceable_or_null(24), ptr) #2
 ; Function Attrs: nounwind
 define internal void @"main.deferSimple$1"(ptr %context) unnamed_addr #1 {
 entry:
+  call void @runtime.printlock(ptr undef) #4
   call void @runtime.printint32(i32 3, ptr undef) #4
+  call void @runtime.printunlock(ptr undef) #4
   ret void
 }
 
+declare void @runtime.printlock(ptr) #2
+
 declare void @runtime.printint32(i32, ptr) #2
 
+declare void @runtime.printunlock(ptr) #2
+
 ; Function Attrs: nounwind
 define hidden void @main.deferMultiple(ptr %context) unnamed_addr #1 {
 entry:
@@ -136,14 +141,14 @@ entry:
   %deferPtr = alloca ptr, align 4
   store ptr null, ptr %deferPtr, align 4
   %deferframe.buf = alloca %runtime.deferFrame, align 4
-  %0 = call ptr @llvm.stacksave()
+  %0 = call ptr @llvm.stacksave.p0()
   call void @runtime.setupDeferFrame(ptr nonnull %deferframe.buf, ptr %0, ptr undef) #4
   store i32 0, ptr %defer.alloca, align 4
-  %defer.alloca.repack22 = getelementptr inbounds { i32, ptr }, ptr %defer.alloca, i32 0, i32 1
+  %defer.alloca.repack22 = getelementptr inbounds i8, ptr %defer.alloca, i32 4
   store ptr null, ptr %defer.alloca.repack22, align 4
   store ptr %defer.alloca, ptr %deferPtr, align 4
   store i32 1, ptr %defer.alloca2, align 4
-  %defer.alloca2.repack23 = getelementptr inbounds { i32, ptr }, ptr %defer.alloca2, i32 0, i32 1
+  %defer.alloca2.repack23 = getelementptr inbounds i8, ptr %defer.alloca2, i32 4
   store ptr %defer.alloca, ptr %defer.alloca2.repack23, align 4
   store ptr %defer.alloca2, ptr %deferPtr, align 4
   %setjmp = call i32 asm "\0Amovs r0, #0\0Amov r2, pc\0Astr r2, [r1, #4]", "={r0},{r1},~{r1},~{r2},~{r3},~{r4},~{r5},~{r6},~{r7},~{r8},~{r9},~{r10},~{r11},~{r12},~{lr},~{q0},~{q1},~{q2},~{q3},~{q4},~{q5},~{q6},~{q7},~{q8},~{q9},~{q10},~{q11},~{q12},~{q13},~{q14},~{q15},~{cpsr},~{memory}"(ptr nonnull %deferframe.buf) #5
@@ -167,7 +172,7 @@ rundefers.loophead:                               ; preds = %4, %3, %rundefers.b
   br i1 %stackIsNil, label %rundefers.end, label %rundefers.loop
 
 rundefers.loop:                                   ; preds = %rundefers.loophead
-  %stack.next.gep = getelementptr inbounds %runtime._defer, ptr %2, i32 0, i32 1
+  %stack.next.gep = getelementptr inbounds i8, ptr %2, i32 4
   %stack.next = load ptr, ptr %stack.next.gep, align 4
   store ptr %stack.next, ptr %deferPtr, align 4
   %callback = load i32, ptr %2, align 4
@@ -213,7 +218,7 @@ rundefers.loophead10:                             ; preds = %7, %6, %lpad
   br i1 %stackIsNil11, label %rundefers.end7, label %rundefers.loop9
 
 rundefers.loop9:                                  ; preds = %rundefers.loophead10
-  %stack.next.gep12 = getelementptr inbounds %runtime._defer, ptr %5, i32 0, i32 1
+  %stack.next.gep12 = getelementptr inbounds i8, ptr %5, i32 4
   %stack.next13 = load ptr, ptr %stack.next.gep12, align 4
   store ptr %stack.next13, ptr %deferPtr, align 4
   %callback15 = load i32, ptr %5, align 4
@@ -250,20 +255,24 @@ rundefers.end7:                                   ; preds = %rundefers.loophead1
 ; Function Attrs: nounwind
 define internal void @"main.deferMultiple$1"(ptr %context) unnamed_addr #1 {
 entry:
+  call void @runtime.printlock(ptr undef) #4
   call void @runtime.printint32(i32 3, ptr undef) #4
+  call void @runtime.printunlock(ptr undef) #4
   ret void
 }
 
 ; Function Attrs: nounwind
 define internal void @"main.deferMultiple$2"(ptr %context) unnamed_addr #1 {
 entry:
+  call void @runtime.printlock(ptr undef) #4
   call void @runtime.printint32(i32 5, ptr undef) #4
+  call void @runtime.printunlock(ptr undef) #4
   ret void
 }
 
-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+armv7-m,+hwdiv,+soft-float,+strict-align,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
-attributes #1 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+strict-align,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
-attributes #2 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+strict-align,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
+attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
+attributes #1 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
+attributes #2 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
 attributes #3 = { nocallback nofree nosync nounwind willreturn }
 attributes #4 = { nounwind }
 attributes #5 = { nounwind returns_twice }

compiler/testdata/errors.go

@@ -1,6 +1,9 @@
 package main
 
-import "unsafe"
+import (
+	"structs"
+	"unsafe"
+)
 
 //go:wasmimport modulename empty
 func empty()
@@ -13,31 +16,92 @@ func implementation() {
 
 type Uint uint32
 
-//go:wasmimport modulename validparam
-func validparam(a int32, b uint64, c float64, d unsafe.Pointer, e Uint)
+type S struct {
+	_ structs.HostLayout
+	a [4]uint32
+	b uintptr
+	d float32
+	e float64
+}
 
-// ERROR: //go:wasmimport modulename invalidparam: unsupported parameter type int
-// ERROR: //go:wasmimport modulename invalidparam: unsupported parameter type string
+//go:wasmimport modulename validparam
+func validparam(a int32, b uint64, c float64, d unsafe.Pointer, e Uint, f uintptr, g string, h *int32, i *S, j *struct{}, k *[8]uint8)
+
+// ERROR: //go:wasmimport modulename invalidparam: unsupported parameter type [4]uint32
 // ERROR: //go:wasmimport modulename invalidparam: unsupported parameter type []byte
-// ERROR: //go:wasmimport modulename invalidparam: unsupported parameter type *int32
+// ERROR: //go:wasmimport modulename invalidparam: unsupported parameter type struct{a int}
+// ERROR: //go:wasmimport modulename invalidparam: unsupported parameter type chan struct{}
+// ERROR: //go:wasmimport modulename invalidparam: unsupported parameter type func()
+// ERROR: //go:wasmimport modulename invalidparam: unsupported parameter type int
+// ERROR: //go:wasmimport modulename invalidparam: unsupported parameter type uint
+// ERROR: //go:wasmimport modulename invalidparam: unsupported parameter type [8]int
 //
 //go:wasmimport modulename invalidparam
-func invalidparam(a int, b string, c []byte, d *int32)
+func invalidparam(a [4]uint32, b []byte, c struct{ a int }, d chan struct{}, e func(), f int, g uint, h [8]int)
 
-//go:wasmimport modulename validreturn
-func validreturn() int32
+// ERROR: //go:wasmimport modulename invalidparam_no_hostlayout: unsupported parameter type *struct{int}
+// ERROR: //go:wasmimport modulename invalidparam_no_hostlayout: unsupported parameter type *struct{string}
+//
+//go:wasmimport modulename invalidparam_no_hostlayout
+func invalidparam_no_hostlayout(a *struct{ int }, b *struct{ string })
+
+//go:wasmimport modulename validreturn_int32
+func validreturn_int32() int32
+
+//go:wasmimport modulename validreturn_ptr_int32
+func validreturn_ptr_int32() *int32
+
+//go:wasmimport modulename validreturn_ptr_string
+func validreturn_ptr_string() *string
+
+//go:wasmimport modulename validreturn_ptr_struct
+func validreturn_ptr_struct() *S
+
+//go:wasmimport modulename validreturn_ptr_struct
+func validreturn_ptr_empty_struct() *struct{}
+
+//go:wasmimport modulename validreturn_ptr_array
+func validreturn_ptr_array() *[8]uint8
+
+//go:wasmimport modulename validreturn_unsafe_pointer
+func validreturn_unsafe_pointer() unsafe.Pointer
 
 // ERROR: //go:wasmimport modulename manyreturns: too many return values
 //
 //go:wasmimport modulename manyreturns
 func manyreturns() (int32, int32)
 
-// ERROR: //go:wasmimport modulename invalidreturn: unsupported result type int
+// ERROR: //go:wasmimport modulename invalidreturn_int: unsupported result type int
 //
-//go:wasmimport modulename invalidreturn
-func invalidreturn() int
+//go:wasmimport modulename invalidreturn_int
+func invalidreturn_int() int
 
-// ERROR: //go:wasmimport modulename invalidUnsafePointerReturn: unsupported result type unsafe.Pointer
+// ERROR: //go:wasmimport modulename invalidreturn_int: unsupported result type uint
 //
-//go:wasmimport modulename invalidUnsafePointerReturn
-func invalidUnsafePointerReturn() unsafe.Pointer
+//go:wasmimport modulename invalidreturn_int
+func invalidreturn_uint() uint
+
+// ERROR: //go:wasmimport modulename invalidreturn_func: unsupported result type func()
+//
+//go:wasmimport modulename invalidreturn_func
+func invalidreturn_func() func()
+
+// ERROR: //go:wasmimport modulename invalidreturn_pointer_array_int: unsupported result type *[8]int
+//
+//go:wasmimport modulename invalidreturn_pointer_array_int
+func invalidreturn_pointer_array_int() *[8]int
+
+// ERROR: //go:wasmimport modulename invalidreturn_slice_byte: unsupported result type []byte
+//
+//go:wasmimport modulename invalidreturn_slice_byte
+func invalidreturn_slice_byte() []byte
+
+// ERROR: //go:wasmimport modulename invalidreturn_chan_int: unsupported result type chan int
+//
+//go:wasmimport modulename invalidreturn_chan_int
+func invalidreturn_chan_int() chan int
+
+// ERROR: //go:wasmimport modulename invalidreturn_string: unsupported result type string
+//
+//go:wasmimport modulename invalidreturn_string
+func invalidreturn_string() string

compiler/testdata/float.ll

@@ -93,6 +93,6 @@ entry:
   ret i8 %0
 }
 
-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #1 = { "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
+attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }

compiler/testdata/func.ll

@@ -44,7 +44,7 @@ entry:
   ret void
 }
 
-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #1 = { "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
+attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
 attributes #3 = { nounwind }

compiler/testdata/gc.ll

@@ -16,9 +16,9 @@ target triple = "wasm32-unknown-wasi"
 @main.struct2 = hidden global ptr null, align 4
 @main.struct3 = hidden global ptr null, align 4
 @main.struct4 = hidden global ptr null, align 4
-@main.slice1 = hidden global { ptr, i32, i32 } zeroinitializer, align 8
-@main.slice2 = hidden global { ptr, i32, i32 } zeroinitializer, align 8
-@main.slice3 = hidden global { ptr, i32, i32 } zeroinitializer, align 8
+@main.slice1 = hidden global { ptr, i32, i32 } zeroinitializer, align 4
+@main.slice2 = hidden global { ptr, i32, i32 } zeroinitializer, align 4
+@main.slice3 = hidden global { ptr, i32, i32 } zeroinitializer, align 4
 @"runtime/gc.layout:62-2000000000000001" = linkonce_odr unnamed_addr constant { i32, [8 x i8] } { i32 62, [8 x i8] c"\01\00\00\00\00\00\00 " }
 @"runtime/gc.layout:62-0001" = linkonce_odr unnamed_addr constant { i32, [8 x i8] } { i32 62, [8 x i8] c"\01\00\00\00\00\00\00\00" }
 @"reflect/types.type:basic:complex128" = linkonce_odr constant { i8, ptr } { i8 80, ptr @"reflect/types.type:pointer:basic:complex128" }, align 4
@@ -39,16 +39,16 @@ entry:
 define hidden void @main.newScalar(ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %new = call dereferenceable(1) ptr @runtime.alloc(i32 1, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %new = call align 1 dereferenceable(1) ptr @runtime.alloc(i32 1, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new, ptr @main.scalar1, align 4
-  %new1 = call dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %new1 = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new1, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new1, ptr @main.scalar2, align 4
-  %new2 = call dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %new2 = call align 8 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new2, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new2, ptr @main.scalar3, align 4
-  %new3 = call dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %new3 = call align 4 dereferenceable(4) ptr @runtime.alloc(i32 4, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new3, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new3, ptr @main.scalar4, align 4
   ret void
@@ -58,13 +58,13 @@ entry:
 define hidden void @main.newArray(ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %new = call dereferenceable(3) ptr @runtime.alloc(i32 3, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %new = call align 1 dereferenceable(3) ptr @runtime.alloc(i32 3, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new, ptr @main.array1, align 4
-  %new1 = call dereferenceable(71) ptr @runtime.alloc(i32 71, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %new1 = call align 1 dereferenceable(71) ptr @runtime.alloc(i32 71, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new1, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new1, ptr @main.array2, align 4
-  %new2 = call dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 67 to ptr), ptr undef) #3
+  %new2 = call align 4 dereferenceable(12) ptr @runtime.alloc(i32 12, ptr nonnull inttoptr (i32 67 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new2, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new2, ptr @main.array3, align 4
   ret void
@@ -74,16 +74,16 @@ entry:
 define hidden void @main.newStruct(ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %new = call ptr @runtime.alloc(i32 0, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %new = call align 1 ptr @runtime.alloc(i32 0, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new, ptr @main.struct1, align 4
-  %new1 = call dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %new1 = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new1, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new1, ptr @main.struct2, align 4
-  %new2 = call dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-2000000000000001", ptr undef) #3
+  %new2 = call align 4 dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-2000000000000001", ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new2, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new2, ptr @main.struct3, align 4
-  %new3 = call dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-0001", ptr undef) #3
+  %new3 = call align 4 dereferenceable(248) ptr @runtime.alloc(i32 248, ptr nonnull @"runtime/gc.layout:62-0001", ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new3, ptr nonnull %stackalloc, ptr undef) #3
   store ptr %new3, ptr @main.struct4, align 4
   ret void
@@ -93,7 +93,7 @@ entry:
 define hidden ptr @main.newFuncValue(ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %new = call dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 197 to ptr), ptr undef) #3
+  %new = call align 4 dereferenceable(8) ptr @runtime.alloc(i32 8, ptr nonnull inttoptr (i32 197 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %new, ptr nonnull %stackalloc, ptr undef) #3
   ret ptr %new
 }
@@ -102,21 +102,21 @@ entry:
 define hidden void @main.makeSlice(ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %makeslice = call dereferenceable(5) ptr @runtime.alloc(i32 5, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
+  %makeslice = call align 1 dereferenceable(5) ptr @runtime.alloc(i32 5, ptr nonnull inttoptr (i32 3 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %makeslice, ptr nonnull %stackalloc, ptr undef) #3
-  store ptr %makeslice, ptr @main.slice1, align 8
-  store i32 5, ptr getelementptr inbounds ({ ptr, i32, i32 }, ptr @main.slice1, i32 0, i32 1), align 4
-  store i32 5, ptr getelementptr inbounds ({ ptr, i32, i32 }, ptr @main.slice1, i32 0, i32 2), align 8
-  %makeslice1 = call dereferenceable(20) ptr @runtime.alloc(i32 20, ptr nonnull inttoptr (i32 67 to ptr), ptr undef) #3
+  store ptr %makeslice, ptr @main.slice1, align 4
+  store i32 5, ptr getelementptr inbounds (i8, ptr @main.slice1, i32 4), align 4
+  store i32 5, ptr getelementptr inbounds (i8, ptr @main.slice1, i32 8), align 4
+  %makeslice1 = call align 4 dereferenceable(20) ptr @runtime.alloc(i32 20, ptr nonnull inttoptr (i32 67 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %makeslice1, ptr nonnull %stackalloc, ptr undef) #3
-  store ptr %makeslice1, ptr @main.slice2, align 8
-  store i32 5, ptr getelementptr inbounds ({ ptr, i32, i32 }, ptr @main.slice2, i32 0, i32 1), align 4
-  store i32 5, ptr getelementptr inbounds ({ ptr, i32, i32 }, ptr @main.slice2, i32 0, i32 2), align 8
-  %makeslice3 = call dereferenceable(60) ptr @runtime.alloc(i32 60, ptr nonnull inttoptr (i32 71 to ptr), ptr undef) #3
+  store ptr %makeslice1, ptr @main.slice2, align 4
+  store i32 5, ptr getelementptr inbounds (i8, ptr @main.slice2, i32 4), align 4
+  store i32 5, ptr getelementptr inbounds (i8, ptr @main.slice2, i32 8), align 4
+  %makeslice3 = call align 4 dereferenceable(60) ptr @runtime.alloc(i32 60, ptr nonnull inttoptr (i32 71 to ptr), ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %makeslice3, ptr nonnull %stackalloc, ptr undef) #3
-  store ptr %makeslice3, ptr @main.slice3, align 8
-  store i32 5, ptr getelementptr inbounds ({ ptr, i32, i32 }, ptr @main.slice3, i32 0, i32 1), align 4
-  store i32 5, ptr getelementptr inbounds ({ ptr, i32, i32 }, ptr @main.slice3, i32 0, i32 2), align 8
+  store ptr %makeslice3, ptr @main.slice3, align 4
+  store i32 5, ptr getelementptr inbounds (i8, ptr @main.slice3, i32 4), align 4
+  store i32 5, ptr getelementptr inbounds (i8, ptr @main.slice3, i32 8), align 4
   ret void
 }
 
@@ -124,10 +124,10 @@ entry:
 define hidden %runtime._interface @main.makeInterface(double %v.r, double %v.i, ptr %context) unnamed_addr #2 {
 entry:
   %stackalloc = alloca i8, align 1
-  %0 = call dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #3
+  %0 = call align 8 dereferenceable(16) ptr @runtime.alloc(i32 16, ptr null, ptr undef) #3
   call void @runtime.trackPointer(ptr nonnull %0, ptr nonnull %stackalloc, ptr undef) #3
   store double %v.r, ptr %0, align 8
-  %.repack1 = getelementptr inbounds { double, double }, ptr %0, i32 0, i32 1
+  %.repack1 = getelementptr inbounds i8, ptr %0, i32 8
   store double %v.i, ptr %.repack1, align 8
   %1 = insertvalue %runtime._interface { ptr @"reflect/types.type:basic:complex128", ptr undef }, ptr %0, 1
   call void @runtime.trackPointer(ptr nonnull @"reflect/types.type:basic:complex128", ptr nonnull %stackalloc, ptr undef) #3
@@ -135,7 +135,7 @@ entry:
   ret %runtime._interface %1
 }
 
-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #1 = { "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
+attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
 attributes #3 = { nounwind }

compiler/testdata/go1.20.ll

@@ -36,7 +36,7 @@ entry:
   br i1 %4, label %unsafe.String.throw, label %unsafe.String.next
 
 unsafe.String.next:                               ; preds = %entry
-  %5 = zext i16 %len to i32
+  %5 = zext nneg i16 %len to i32
   %6 = insertvalue %runtime._string undef, ptr %ptr, 0
   %7 = insertvalue %runtime._string %6, i32 %5, 1
   call void @runtime.trackPointer(ptr %ptr, ptr nonnull %stackalloc, ptr undef) #3
@@ -57,7 +57,7 @@ entry:
   ret ptr %s.data
 }
 
-attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #1 = { "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
-attributes #2 = { nounwind "target-features"="+bulk-memory,+nontrapping-fptoint,+sign-ext" }
+attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #1 = { "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
+attributes #2 = { nounwind "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext,-multivalue,-reference-types" }
 attributes #3 = { nounwind }

compiler/testdata/go1.21.go

@@ -0,0 +1,65 @@
+package main
+
+func min1(a int) int {
+	return min(a)
+}
+
+func min2(a, b int) int {
+	return min(a, b)
+}
+
+func min3(a, b, c int) int {
+	return min(a, b, c)
+}
+
+func min4(a, b, c, d int) int {
+	return min(a, b, c, d)
+}
+
+func minUint8(a, b uint8) uint8 {
+	return min(a, b)
+}
+
+func minUnsigned(a, b uint) uint {
+	return min(a, b)
+}
+
+func minFloat32(a, b float32) float32 {
+	return min(a, b)
+}
+
+func minFloat64(a, b float64) float64 {
+	return min(a, b)
+}
+
+func minString(a, b string) string {
+	return min(a, b)
+}
+
+func maxInt(a, b int) int {
+	return max(a, b)
+}
+
+func maxUint(a, b uint) uint {
+	return max(a, b)
+}
+
+func maxFloat32(a, b float32) float32 {
+	return max(a, b)
+}
+
+func maxString(a, b string) string {
+	return max(a, b)
+}
+
+func clearSlice(s []int) {
+	clear(s)
+}
+
+func clearZeroSizedSlice(s []struct{}) {
+	clear(s)
+}
+
+func clearMap(m map[string]int) {
+	clear(m)
+}