build.yml

name: Builds
on:
  push:
    branches:
      - '**'
  pull_request:
  schedule:
    - cron: '0 0 * * *'

permissions:
  contents: read

jobs:
  # For scheduled builds, get the commit ids of the latest versions to build.
  #
  # Doing this as one single job, which then propagates the information further
  # to the other jobs, so that all build jobs in one workflow builds the exact
  # same version. This allows trusting builds without assertions enabled
  # slightly more, when we know a separate build of the same version but with
  # assertions enabled, has passed some amount of tests.
  prepare:
    if: (github.event_name != 'schedule') || (github.repository == 'mstorsjo/llvm-mingw')
    runs-on: ubuntu-latest
    outputs:
      LLVM_VERSION: ${{steps.get-versions.outputs.LLVM_VERSION}}
      MINGW_W64_VERSION: ${{steps.get-versions.outputs.MINGW_W64_VERSION}}
      PYTHON_VERSION_MINGW: ${{steps.get-versions.outputs.PYTHON_VERSION_MINGW}}
      TAG: ${{steps.get-tag.outputs.TAG}}
    steps:
      - name: Select build tag
        id: get-tag
        run: |
          if ${{github.event_name == 'schedule'}}; then
              TAG=nightly
          else
              TAG=$(TZ=UTC date +%Y%m%d)
          fi
          echo TAG=$TAG >> $GITHUB_OUTPUT
          cat $GITHUB_OUTPUT
          cat $GITHUB_OUTPUT >> parameters.txt
      - name: Check latest version
        if: github.event_name == 'schedule'
        id: get-versions
        run: |
          echo LLVM_VERSION=$(git ls-remote https://github.com/llvm/llvm-project.git | grep 'refs/heads/main$' | awk '{print $1}') >> $GITHUB_OUTPUT
          echo MINGW_W64_VERSION=$(git ls-remote https://github.com/mingw-w64/mingw-w64.git | grep 'refs/heads/master$' | awk '{print $1}') >> $GITHUB_OUTPUT
          echo PYTHON_VERSION_MINGW=$(git ls-remote https://github.com/msys2-contrib/cpython-mingw.git | grep 'refs/heads/mingw-v3.10.11$' | awk '{print $1}') >> $GITHUB_OUTPUT
          cat $GITHUB_OUTPUT
          cat $GITHUB_OUTPUT >> parameters.txt
      - uses: actions/upload-artifact@v3
        with:
          name: parameters
          path: |
            parameters.txt
          retention-days: 7

  # Build a cross compiler for Linux, targeting Windows.
  linux:
    if: (github.event_name != 'schedule') || (github.repository == 'mstorsjo/llvm-mingw')
    needs: [prepare]
    runs-on: ubuntu-20.04
    steps:
      - uses: actions/checkout@v3
      - name: Build
        env:
          LLVM_VERSION: ${{needs.prepare.outputs.LLVM_VERSION}}
          MINGW_W64_VERSION: ${{needs.prepare.outputs.MINGW_W64_VERSION}}
          TAG: ${{needs.prepare.outputs.TAG}}
        run: |
          sudo apt-get update && sudo apt-get install ninja-build
          # Skip dynamic library dependencies that might make it harder to
          # run the binaries on other distros (and that have little use within
          # llvm-mingw).
          LLVM_CMAKEFLAGS="-DLLVM_ENABLE_LIBXML2=OFF -DLLVM_ENABLE_TERMINFO=OFF -DLLDB_ENABLE_PYTHON=OFF" ./build-all.sh $(pwd)/install/llvm-mingw --disable-clang-tools-extra
          .github/workflows/store-version.sh install/llvm-mingw/versions.txt
          cd install
          DISTRO=ubuntu-$(grep DISTRIB_RELEASE /etc/lsb-release | cut -f 2 -d =)-$(uname -m)
          NAME=llvm-mingw-$TAG-ucrt-$DISTRO
          mv llvm-mingw $NAME
          tar -Jcf ../$NAME.tar.xz $NAME
      - uses: actions/upload-artifact@v3
        with:
          name: linux-ucrt-x86_64-toolchain
          path: |
            llvm-mingw-*.tar.xz
          retention-days: 7

  # Crosscompile the toolchain for running on Linux on a different architecture, bundle the runtime
  # libraries that were built in the 'linux' step above.
  linux-cross-aarch64:
    if: false
    needs: [linux, prepare]
    runs-on: ubuntu-20.04
    steps:
      - uses: actions/checkout@v3
      - name: Build
        env:
          LLVM_VERSION: ${{needs.prepare.outputs.LLVM_VERSION}}
          MINGW_W64_VERSION: ${{needs.prepare.outputs.MINGW_W64_VERSION}}
        run: |
          sudo apt-get update && sudo apt-get install ninja-build g++-aarch64-linux-gnu
          ./build-all.sh $(pwd)/install/llvm-mingw --disable-clang-tools-extra --no-runtimes --host=aarch64-linux-gnu
          .github/workflows/store-version.sh install/llvm-mingw/versions.txt
      - uses: actions/download-artifact@v3
        with:
          name: linux-ucrt-x86_64-toolchain
      - name: Unpack native toolchain
        run: |
          tar -Jxf llvm-mingw-*.tar.xz
          rm llvm-mingw-*.tar.xz
          mv llvm-mingw* llvm-mingw-native
      - name: Assemble the cross-built toolchain
        run: |
          ./prepare-cross-toolchain-unix.sh $(pwd)/llvm-mingw-native $(pwd)/install/llvm-mingw
      - name: Test using the cross-built assembled toolchain
        run: |
          sudo apt-get update && sudo apt-get install qemu-user-static libc6-arm64-cross libstdc++6-arm64-cross
          QEMU_LD_PREFIX=/usr/aarch64-linux-gnu RUN_I686=false RUN_X86_64=false ./run-tests.sh $(pwd)/install/llvm-mingw
      - name: Package the toolchain
        env:
          TAG: ${{needs.prepare.outputs.TAG}}
        run: |
          cd install
          DISTRO=ubuntu-$(grep DISTRIB_RELEASE /etc/lsb-release | cut -f 2 -d =)-aarch64
          NAME=llvm-mingw-$TAG-ucrt-$DISTRO
          mv llvm-mingw $NAME
          tar -Jcf ../$NAME.tar.xz $NAME
      - uses: actions/upload-artifact@v3
        with:
          name: linux-ucrt-aarch64-toolchain
          path: |
            llvm-mingw-*.tar.xz
          retention-days: 7

  # Build a cross compiler for Linux, targeting Windows, with assertions enabled.
  #
  # The scheduled builds are made with the latest git version of llvm-project
  # and mingw-w64. When using random git snapshot builds of llvm, there's
  # always a risk for bugs - thus build such versions with assertions enabled,
  # to better catch such bugs early. This makes the first-stage toolchain built
  # here in scheduled builds somewhat slower.
  linux-asserts:
    if: false
    needs: [prepare]
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v3
      - name: Build
        env:
          LLVM_VERSION: ${{needs.prepare.outputs.LLVM_VERSION}}
          MINGW_W64_VERSION: ${{needs.prepare.outputs.MINGW_W64_VERSION}}
        run: |
          sudo apt-get update && sudo apt-get install ninja-build
          # Skip dynamic library dependencies that might make it harder to
          # run the binaries on other distros (and that have little use within
          # llvm-mingw).
          LLVM_CMAKEFLAGS="-DLLVM_ENABLE_LIBXML2=OFF -DLLVM_ENABLE_TERMINFO=OFF" ./build-all.sh $(pwd)/install/llvm-mingw --disable-clang-tools-extra --disable-lldb --enable-asserts
          .github/workflows/store-version.sh install/llvm-mingw/versions.txt
          cd install
          tar -Jcf ../llvm-mingw-linux.tar.xz llvm-mingw
      - uses: actions/upload-artifact@v3
        with:
          name: linux-asserts-toolchain
          path: |
            llvm-mingw-linux.tar.xz
          retention-days: 7

  # Build a cross compiler for macOS, targeting Windows.
  macos:
    if: false
    needs: [prepare]
    runs-on: macos-latest
    steps:
      - uses: actions/checkout@v3
      - name: Build
        env:
          LLVM_VERSION: ${{needs.prepare.outputs.LLVM_VERSION}}
          MINGW_W64_VERSION: ${{needs.prepare.outputs.MINGW_W64_VERSION}}
          TAG: ${{needs.prepare.outputs.TAG}}
        run: |
          brew install ninja
          # Disable zstd and python. Both are available on the runners, but
          # installed with homebrew, and only available in the native (x86_64)
          # form. Therefore, autodetection will pick them up, but linking
          # universal binaries fails as those libraries are unavailable in the
          # other (arm64) architecture form.
          MACOS_REDIST=1 LLVM_CMAKEFLAGS="-DLLVM_ENABLE_ZSTD=OFF -DLLDB_ENABLE_PYTHON=OFF" ./build-all.sh $(pwd)/install/llvm-mingw --disable-clang-tools-extra
          .github/workflows/store-version.sh install/llvm-mingw/versions.txt
          cd install
          NAME=llvm-mingw-$TAG-ucrt-macos-universal
          mv llvm-mingw $NAME
          tar -Jcf ../$NAME.tar.xz $NAME
      - uses: actions/upload-artifact@v3
        with:
          name: macos-ucrt-toolchain
          path: |
            llvm-mingw-*.tar.xz
          retention-days: 7

  # Test building the toolchain on msys2 (in the mingw64 and clang64
  # environments). The binaries built here rely on the runtime libraries from
  # the host environment (libstdc++ or libc++). No artifacts are stored from
  # these builds, but llvm-mingw's tests are run.
  msys2:
    if: false
    needs: [prepare]
    runs-on: windows-latest
    defaults:
      run:
        shell: msys2 {0}
    strategy:
      fail-fast: false
      matrix:
        sys:
          - mingw64
          - clang64
    steps:
      - uses: msys2/setup-msys2@v2
        with:
          msystem: ${{matrix.sys}}
          install: >-
            git
            make
            diffutils
          pacboy: >-
            toolchain:p
            cmake:p
            ninja:p
      - uses: actions/checkout@v3
      - name: Build
        env:
          LLVM_VERSION: ${{needs.prepare.outputs.LLVM_VERSION}}
          MINGW_W64_VERSION: ${{needs.prepare.outputs.MINGW_W64_VERSION}}
        run: |
          ./build-all.sh $(pwd)/install/llvm-mingw --disable-clang-tools-extra --disable-lldb
          .github/workflows/store-version.sh install/llvm-mingw/versions.txt
          ./run-tests.sh $(pwd)/install/llvm-mingw

  # Use the Linux cross compilers built in the first step to cross compile
  # llvm and make a proper standalone toolchain for Windows (for all 4
  # architectures). The binaries built here match actual releases quite closely.
  linux-cross-windows:
    if: (github.event_name != 'schedule') || (github.repository == 'mstorsjo/llvm-mingw')
    needs: [linux, prepare]
    runs-on: ubuntu-latest
    strategy:
      fail-fast: false
      matrix:
        include:
          - { arch: i686, crt: ucrt }
          - { arch: x86_64, crt: ucrt }
          - { arch: armv7, crt: ucrt }
          - { arch: aarch64, crt: ucrt }
    steps:
      - uses: actions/download-artifact@v3
        with:
          name: linux-${{matrix.crt}}-x86_64-toolchain
      - name: Unpack cross toolchain
        run: |
          tar -Jxf llvm-mingw-*.tar.xz
          rm llvm-mingw-*.tar.xz
          sudo mv llvm-mingw* /opt/llvm-mingw
          echo /opt/llvm-mingw/bin >> $GITHUB_PATH
      - uses: actions/checkout@v3
      - name: Build
        env:
          LLVM_VERSION: ${{needs.prepare.outputs.LLVM_VERSION}}
          MINGW_W64_VERSION: ${{needs.prepare.outputs.MINGW_W64_VERSION}}
          PYTHON_VERSION_MINGW: ${{needs.prepare.outputs.PYTHON_VERSION_MINGW}}
          TAG: ${{needs.prepare.outputs.TAG}}
        run: |
          sudo apt-get update && sudo apt-get install autoconf-archive ninja-build
          ./build-cross-tools.sh /opt/llvm-mingw $(pwd)/install/llvm-mingw ${{matrix.arch}} --with-python
          .github/workflows/store-version.sh install/llvm-mingw/versions.txt
          cd install
          NAME=llvm-mingw-$TAG-${{matrix.crt}}-${{matrix.arch}}
          mv llvm-mingw $NAME
          zip -9rq ../$NAME.zip $NAME
      - uses: actions/upload-artifact@v3
        with:
          name: windows-${{matrix.crt}}-${{matrix.arch}}-toolchain
          path: |
            llvm-mingw-*.zip
          retention-days: 7

  # Run llvm-mingw's tests on x86_64 and i686 with the cross-built corresponding
  # toolchains from above.
  test-toolchain:
    if: (github.event_name != 'schedule') || (github.repository == 'mstorsjo/llvm-mingw')
    needs: [linux-cross-windows]
    runs-on: windows-latest
    defaults:
      run:
        shell: msys2 {0}
    strategy:
      fail-fast: false
      matrix:
        arch:
          - x86_64
          - i686
    steps:
      - uses: msys2/setup-msys2@v2
        with:
          msystem: mingw64
          install: >-
            unzip
            make
      - uses: actions/download-artifact@v3
        with:
          name: windows-ucrt-${{matrix.arch}}-toolchain
      - name: Unpack toolchain
        run: |
          unzip -q llvm-mingw-*.zip
          rm llvm-mingw-*.zip
          mv llvm-mingw-* /llvm-mingw
          echo /llvm-mingw/bin >> $GITHUB_PATH
      - uses: actions/checkout@v3
      - name: Run tests
        run: |
          ./run-tests.sh /llvm-mingw
          ./run-lldb-tests.sh /llvm-mingw
      - name: Run tests with make directly
        run: |
          mkdir test/local
          cd test/local
          export PATH=/llvm-mingw/bin:$PATH
          make -f ../Makefile -j$(nproc) HAVE_SANITIZERS=1 HAVE_OPENMP=1 NATIVE=1 test

  test-toolchain-cmd:
    if: (github.event_name != 'schedule') || (github.repository == 'mstorsjo/llvm-mingw')
    needs: [linux-cross-windows]
    runs-on: windows-latest
    defaults:
      run:
        shell: cmd
    strategy:
      fail-fast: false
      matrix:
        arch:
          - x86_64
          - i686
    steps:
      - uses: actions/download-artifact@v3
        with:
          name: windows-ucrt-${{matrix.arch}}-toolchain
      - name: Unpack toolchain
        run: |
          powershell Expand-Archive llvm-mingw-*.zip -DestinationPath .
          del llvm-mingw-*.zip
          move llvm-mingw-* ..\toolchain
          echo %cd%\..\toolchain\bin>> %GITHUB_PATH%
      - uses: actions/checkout@v3
      - name: Run tests
        run: |
          mkdir test\local
          cd test\local
          mingw32-make -f ../Makefile HAVE_SANITIZERS=1 HAVE_OPENMP=1 test

  # Run libcxx's tests with the cross-built i686/x86_64 toolchains from above.
  # (This builds its own copy of libcxx, but it should be pretty much
  # identical to the one shipped - and tests that the toolchain works
  # for running the libcxx tests.)
  # This also forces testing the bundled python executables.
  test-libcxx:
    # This test is kinda slow, and kinda redundant (libcxx does contain
    # llvm-mingw configurations in its own CI, so the only difference with
    # testing here is different runners with different OS setups, and
    # testing with the latest compiler instead of an older release).
    # Therefore, keep the test disabled by default; it's easy to enable
    # temporarily on a branch for testing.
    if: (github.event_name != 'schedule') || (github.repository == 'mstorsjo/llvm-mingw')
    needs: [linux-cross-windows, prepare]
    runs-on: windows-latest
    strategy:
      fail-fast: false
      matrix:
        arch:
          - x86_64
          - i686
    steps:
      - name: Install dependencies
        run: |
          choco install ninja
      - uses: actions/download-artifact@v3
        with:
          name: windows-ucrt-${{matrix.arch}}-toolchain
      - name: Unpack toolchain
        run: |
          Expand-Archive llvm-mingw-*.zip -DestinationPath .
          del llvm-mingw-*.zip
          mv llvm-mingw-* c:\llvm-mingw
          echo "c:\llvm-mingw\bin" | Out-File -FilePath $Env:GITHUB_PATH -Encoding utf8 -Append
          echo "c:\llvm-mingw\python\bin" | Out-File -FilePath $Env:GITHUB_PATH -Encoding utf8 -Append
          echo "PYTHON_EXE=c:/llvm-mingw/python/bin/python3.exe" | Out-File -FilePath $Env:GITHUB_ENV -Encoding utf8 -Append
      - uses: actions/checkout@v3
      - name: Checkout llvm-project
        env:
          LLVM_VERSION: ${{needs.prepare.outputs.LLVM_VERSION}}
        run: |
          bash -c "CHECKOUT_ONLY=1 ./build-llvm.sh"
      - name: Build and test libcxx
        run: |
          cd llvm-project
          mkdir build
          cd build
          cmake ../runtimes `
            -G Ninja `
            -DCMAKE_BUILD_TYPE=Release `
            -DPython3_EXECUTABLE="$Env:PYTHON_EXE" `
            -DLIBCXX_ENABLE_WERROR=YES `
            -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi;libunwind" `
            -DLIBCXX_CXX_ABI=libcxxabi `
            -DCMAKE_C_COMPILER=clang `
            -DCMAKE_CXX_COMPILER=clang++ `
            -DLIBCXXABI_ENABLE_SHARED=NO `
            -DLIBCXX_ENABLE_STATIC_ABI_LIBRARY=YES `
            -DLIBCXX_USE_COMPILER_RT=YES `
            -DLIBCXXABI_USE_COMPILER_RT=YES `
            -DLIBUNWIND_USE_COMPILER_RT=YES `
            -DLIBCXXABI_USE_LLVM_UNWINDER=YES `
            -DLIBCXX_EXTRA_SITE_DEFINES="__USE_MINGW_ANSI_STDIO=1;TEST_IS_EXECUTED_IN_A_SLOW_ENVIRONMENT"
          ninja
          ninja check-cxx check-cxxabi check-unwind

  # Run the OpenMP tests with the cross-built i686/x86_64 toolchains from above.
  # This also forces testing the bundled python executables.
  test-openmp:
    if: (github.event_name != 'schedule') || (github.repository == 'mstorsjo/llvm-mingw')
    needs: [linux-cross-windows, prepare]
    runs-on: windows-latest
    strategy:
      fail-fast: false
      matrix:
        include:
          - { arch: i686, asmflag: }
          - { arch: x86_64, asmflag: -m64 }
    steps:
      - name: Install dependencies
        run: |
          choco install ninja
      - uses: actions/download-artifact@v3
        with:
          name: windows-ucrt-${{matrix.arch}}-toolchain
      - name: Unpack toolchain
        run: |
          Expand-Archive llvm-mingw-*.zip -DestinationPath .
          del llvm-mingw-*.zip
          mv llvm-mingw-* c:\llvm-mingw
          echo "c:\llvm-mingw\bin" | Out-File -FilePath $Env:GITHUB_PATH -Encoding utf8 -Append
          echo "c:\llvm-mingw\python\bin" | Out-File -FilePath $Env:GITHUB_PATH -Encoding utf8 -Append
          echo "PYTHON_EXE=c:/llvm-mingw/python/bin/python3.exe" | Out-File -FilePath $Env:GITHUB_ENV -Encoding utf8 -Append
      - uses: actions/checkout@v3
      - name: Checkout llvm-project
        env:
          LLVM_VERSION: ${{needs.prepare.outputs.LLVM_VERSION}}
        run: |
          bash -c "CHECKOUT_ONLY=1 ./build-llvm.sh"
      - name: Build test executables
        run: |
          cd llvm-project
          mkdir build-tools
          cd build-tools
          cmake ../llvm `
            -G Ninja `
            -DCMAKE_BUILD_TYPE=Release `
            -DLLVM_TARGETS_TO_BUILD=X86 `
            -DCMAKE_C_COMPILER=clang `
            -DCMAKE_CXX_COMPILER=clang++
          ninja not FileCheck
          echo "TOOLS=$PWD\bin" | Out-File -FilePath $Env:GITHUB_ENV -Encoding utf8 -Append
      - name: Build and test OpenMP
        run: |
          cd llvm-project
          mkdir build-openmp
          cd build-openmp
          cmake ../openmp `
            -G Ninja `
            -DCMAKE_BUILD_TYPE=Release `
            -DPython3_EXECUTABLE="$Env:PYTHON_EXE" `
            -DOPENMP_LLVM_TOOLS_DIR="$Env:TOOLS" `
            -DCMAKE_C_COMPILER=clang `
            -DCMAKE_CXX_COMPILER=clang++ `
            -DCMAKE_ASM_MASM_COMPILER=llvm-ml `
            -DLIBOMP_ASMFLAGS=${{matrix.asmflag}}
          ninja
          ninja check-openmp

  # Run the compiler-rt tests with the cross-built i686/x86_64 toolchains from
  # above. This also forces testing the bundled python executables.
  test-compiler-rt:
    if: (github.event_name != 'schedule') || (github.repository == 'mstorsjo/llvm-mingw')
    needs: [linux-cross-windows, prepare]
    runs-on: windows-latest
    strategy:
      fail-fast: false
      matrix:
        arch:
          - x86_64
          - i686
    steps:
      - name: Install dependencies
        run: |
          choco install ninja
      - uses: actions/download-artifact@v3
        with:
          name: windows-ucrt-${{matrix.arch}}-toolchain
      - name: Unpack toolchain
        run: |
          Expand-Archive llvm-mingw-*.zip -DestinationPath .
          del llvm-mingw-*.zip
          mv llvm-mingw-* c:\llvm-mingw
          echo "c:\llvm-mingw\bin" | Out-File -FilePath $Env:GITHUB_PATH -Encoding utf8 -Append
          echo "c:\llvm-mingw\python\bin" | Out-File -FilePath $Env:GITHUB_PATH -Encoding utf8 -Append
          echo "PYTHON_EXE=c:/llvm-mingw/python/bin/python3.exe" | Out-File -FilePath $Env:GITHUB_ENV -Encoding utf8 -Append
      - uses: actions/checkout@v3
      - name: Checkout llvm-project
        env:
          LLVM_VERSION: ${{needs.prepare.outputs.LLVM_VERSION}}
        run: |
          git config --global core.autocrlf false
          bash -c "CHECKOUT_ONLY=1 ./build-llvm.sh"
      - name: Build test executables
        run: |
          cd llvm-project
          mkdir build-tools
          cd build-tools
          cmake ../llvm `
            -G Ninja `
            -DCMAKE_BUILD_TYPE=Release `
            -DLLVM_TARGETS_TO_BUILD=X86 `
            -DCMAKE_C_COMPILER=clang `
            -DCMAKE_CXX_COMPILER=clang++
          ninja not FileCheck llvm-config KillTheDoctor count split-file
          echo "LLVM_CONFIG=$PWD\bin\llvm-config.exe" | Out-File -FilePath $Env:GITHUB_ENV -Encoding utf8 -Append
          echo "CMAKE_DIR=$PWD\lib\cmake\llvm" | Out-File -FilePath $Env:GITHUB_ENV -Encoding utf8 -Append
      - name: Build and test compiler-rt
        # Skipping libfuzzer; all tests fail since the clang mingw driver
        # doesn't support -fsanitize=fuzzer.
        # Skipping ORC; the tests rely on llvm-jitlink being available,
        # which requires building a large part of LLVM among the test
        # tools above, and jitlink requires an MSVC toolchain to be available.
        run: |
          cd llvm-project
          mkdir build-compiler-rt
          cd build-compiler-rt
          cmake ../compiler-rt `
            -G Ninja `
            -DCMAKE_BUILD_TYPE=Release `
            -DPython3_EXECUTABLE="$Env:PYTHON_EXE" `
            -DCMAKE_C_COMPILER=clang `
            -DCMAKE_CXX_COMPILER=clang++ `
            -DCMAKE_C_COMPILER_TARGET=${{matrix.arch}}-w64-windows-gnu `
            -DCOMPILER_RT_DEFAULT_TARGET_ONLY=TRUE `
            -DCOMPILER_RT_USE_BUILTINS_LIBRARY=TRUE `
            -DLLVM_CONFIG_PATH="$Env:LLVM_CONFIG" `
            -DLLVM_CMAKE_DIR="$Env:CMAKE_DIR" `
            -DSANITIZER_CXX_ABI=libc++ `
            -DCOMPILER_RT_INCLUDE_TESTS=ON `
            -DCOMPILER_RT_BUILD_LIBFUZZER=OFF `
            -DCOMPILER_RT_BUILD_ORC=OFF
          ninja
          ninja check-compiler-rt

  # Test cross-building ffmpeg for all 4 targeted architectures from linux.
  # This is done only on scheduled builds, with a toolchain with assertions
  # enabled, to catch code generation bugs that might trigger asserts, to
  # find such regressions early.
  linux-test-cross-build-ffmpeg:
    if: false
    needs: [linux-asserts]
    runs-on: ubuntu-latest
    strategy:
      fail-fast: false
      matrix:
        arch:
          - i686
          - x86_64
          - armv7
          - aarch64
    steps:
      - uses: actions/download-artifact@v3
        with:
          name: linux-asserts-toolchain
      - name: Unpack cross toolchain
        run: |
          tar -Jxf llvm-mingw-*.tar.xz
          rm llvm-mingw-*.tar.xz
          sudo mv llvm-mingw* /opt/llvm-mingw
          echo /opt/llvm-mingw/bin >> $GITHUB_PATH
      - name: Checkout ffmpeg
        uses: actions/checkout@v3
        with:
          repository: ffmpeg/ffmpeg
          ref: n6.0
          path: ffmpeg
      - name: Build ffmpeg
        run: |
          sudo apt-get update && sudo apt-get install nasm
          mkdir ffmpeg-build
          cd ffmpeg-build
          # Set a limit of 3 minutes per process/object file; with unproven
          # versions from latest git, the compiler may occasionally hang.
          # Make sure we exit in a reasonable time instead of waiting for
          # the whole github actions timeout of 6h before exiting.
          ulimit -t 180
          ../ffmpeg/configure --arch=${{matrix.arch}} --target-os=mingw32 --cross-prefix=${{matrix.arch}}-w64-mingw32- --enable-gpl
          make -j$(nproc)

  # Test building ffmpeg for natively on x86_64 Windows and run its tests,
  # to find cases of compiler bugs don't show up as failed asserts in the
  # compiler itself, but that only show up at runtime. This is only done
  # for scheduled builds.
  test-ffmpeg:
    if: false
    needs: [linux-cross-windows]
    runs-on: windows-latest
    defaults:
      run:
        shell: msys2 {0}
    steps:
      - name: Avoid git checking out files with CRLF
        shell: cmd
        run: |
          git config --global core.autocrlf false
      - uses: msys2/setup-msys2@v2
        with:
          msystem: mingw64
          install: >-
            unzip
            make
            rsync
            diffutils
          pacboy: >-
            nasm:p
      - uses: actions/download-artifact@v3
        with:
          name: windows-ucrt-x86_64-toolchain
      - name: Unpack toolchain
        run: |
          unzip -q llvm-mingw-*.zip
          rm llvm-mingw-*.zip
          mv llvm-mingw-* /llvm-mingw
          echo /llvm-mingw/bin >> $GITHUB_PATH
      - name: Checkout ffmpeg
        uses: actions/checkout@v3
        with:
          repository: ffmpeg/ffmpeg
          ref: n6.0
          path: ffmpeg
      - name: Build & test ffmpeg
        run: |
          export PATH=/llvm-mingw/bin:$PATH
          mkdir ffmpeg-build
          cd ffmpeg-build
          ../ffmpeg/configure --samples=../fate-samples --enable-gpl
          make -j$(nproc)
          make fate-rsync
          make -j$(nproc) fate

  upload-nightly:
    if: false
    permissions:
      contents: write
    needs: [linux, linux-cross-aarch64, macos, linux-cross-windows, test-toolchain, linux-test-cross-build-ffmpeg, test-ffmpeg]
    runs-on: ubuntu-latest
    steps:
      - name: Download all artifacts
        uses: actions/download-artifact@v3
      - name: Rearrange files
        run: |
          rm -rf linux-asserts*
          mv *-toolchain/*.zip *-toolchain/*.tar.xz .
      - name: Upload binaries
        env:
          GITHUB_TOKEN: ${{github.token}}
        run: |
          gh release upload nightly *.tar.xz *.zip --clobber -R ${{github.repository}}