From 8a89d951088af098732f0603ebe00a768aec48bb Mon Sep 17 00:00:00 2001
From: Eric Niebler <eniebler@nvidia.com>
Date: Wed, 29 Apr 2026 17:17:37 -0700
Subject: [PATCH 1/7] try turning on app verifier for msvc

---
 .github/workflows/test-windows.ps1 | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/.github/workflows/test-windows.ps1 b/.github/workflows/test-windows.ps1
index 809542815..4f579700f 100644
--- a/.github/workflows/test-windows.ps1
+++ b/.github/workflows/test-windows.ps1
@@ -27,4 +27,10 @@ Invoke-NativeCommand cmake -B $BuildDirectory -G Ninja `
 	"-DCMAKE_VERBOSE_MAKEFILE:BOOL=ON" `
 	"-DSTDEXEC_BUILD_TESTS:BOOL=TRUE" .
 Invoke-NativeCommand cmake --build $BuildDirectory
+
+# Enable AppVerifier for the test executables
+Invoke-NativeCommand appverif -enable Basics -for test.stdexec.exe
+Invoke-NativeCommand appverif -enable Basics -for test.exec.exe
+
 Invoke-NativeCommand ctest --test-dir $BuildDirectory --output-on-failure --verbose --timeout 60
+

From 39e1d1b9301a71ada84ba54613fa5576ca79f330 Mon Sep 17 00:00:00 2001
From: Eric Niebler <eniebler@nvidia.com>
Date: Wed, 29 Apr 2026 17:41:24 -0700
Subject: [PATCH 2/7] try again

---
 .github/workflows/ci.cpu.yml       | 342 ++++++++++++++---------------
 .github/workflows/test-windows.ps1 |   4 +-
 2 files changed, 173 insertions(+), 173 deletions(-)

diff --git a/.github/workflows/ci.cpu.yml b/.github/workflows/ci.cpu.yml
index 7b353d97f..6f9d34a4a 100644
--- a/.github/workflows/ci.cpu.yml
+++ b/.github/workflows/ci.cpu.yml
@@ -12,133 +12,133 @@ concurrency:
 
 jobs:
 
-  build-cpu:
-    runs-on: ubuntu-latest
-    name: ${{ matrix.name }}
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - { name: "CPU (clang 16, Debug)",         build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++" }
-          - { name: "CPU (clang 16, Debug, c++23)",  build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "23", cxxflags: "-stdlib=libc++" }
-          - { name: "CPU (clang 16, Debug, TSAN)",   build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-fsanitize=thread" }
-          - { name: "CPU (clang 16, Release)",       build: "Release", tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++" }
-          - { name: "CPU (clang 16, Release, ASAN)", build: "Release", tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++ -fsanitize=address -fsanitize-ignorelist=/home/coder/stdexec/sanitizer-ignorelist.txt" }
-          - { name: "CPU (clang 22, Debug)",         build: "Debug",   tag: llvm22-cuda13.2, cxxstd: "23", cxxflags: "-stdlib=libc++" }
-          - { name: "CPU (clang 22, Release)",       build: "Release", tag: llvm22-cuda13.2, cxxstd: "23", cxxflags: "-stdlib=libc++" }
-          - { name: "CPU (gcc 12, Debug)",           build: "Debug",   tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "", }
-          - { name: "CPU (gcc 12, Release)",         build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "", }
-          # With the following config, 2 tests mysteriously time out, but only in CI and not locally.
-          # - { name: "CPU (gcc 12, Release, ASAN)",   build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
-          - { name: "CPU (gcc 12, Release, TSAN)",   build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=thread" }
-          - { name: "CPU (gcc 13, Debug)",           build: "Debug",   tag: gcc13-cuda12.9,  cxxstd: "20", cxxflags: "", }
-          - { name: "CPU (gcc 14, Debug)",           build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "", }
-          - { name: "CPU (gcc 14, Debug, ASAN)",     build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
-          - { name: "CPU (gcc 14, Debug, TSAN)",     build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=thread" }
-          - { name: "CPU (gcc 14, Release, ASAN)",   build: "Release", tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
-          - { name: "CPU (gcc 14, Release, LEAK)",   build: "Release", tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=leak", }
-          - { name: "CPU (gcc 14, Release, c++23)",  build: "Release", tag: gcc14-cuda12.9,  cxxstd: "23", cxxflags: "", }
-    container:
-      options: -u root
-      image: rapidsai/devcontainers:26.06-cpp-${{ matrix.tag }}
-    permissions:
-      id-token: write # This is required for configure-aws-credentials
-      contents: read  # This is required for actions/checkout
-    defaults:
-      run:
-        shell: su coder {0}
-        working-directory: /home/coder
-    steps:
-      - name: Checkout stdexec
-        uses: actions/checkout@v4
-        with:
-          path: stdexec
-          persist-credentials: false
-      - name: Setup environment
-        run: |
-          echo "ARTIFACT_PREFIX=${{runner.os}}-${{matrix.tag}}-amd64" >> "${GITHUB_ENV}"
-          echo "ARTIFACT_SUFFIX=${{github.run_id}}-${{github.run_attempt}}-$RANDOM" >> "${GITHUB_ENV}"
-      - if: github.repository_owner == 'NVIDIA'
-        name: Get AWS credentials for sccache bucket
-        uses: aws-actions/configure-aws-credentials@v4
-        with:
-          aws-region: us-east-2
-          role-duration-seconds: 28800 # 8 hours
-          role-to-assume: arn:aws:iam::279114543810:role/gha-oidc-NVIDIA
-      - name: Build and test CPU schedulers
-        env:
-          ASAN_OPTIONS: alloc_dealloc_mismatch=0
-          NVCC_APPEND_FLAGS: "-t=100"
-          SCCACHE_BUCKET: "rapids-sccache-devs"
-          SCCACHE_DIST_REQUEST_TIMEOUT: "7140"
-          SCCACHE_DIST_URL: "https://amd64.linux.sccache.rapids.nvidia.com"
-          SCCACHE_IDLE_TIMEOUT: "0"
-          SCCACHE_REGION: "us-east-2"
-          SCCACHE_S3_KEY_PREFIX: "nvidia-stdexec-dev"
-          SCCACHE_S3_PREPROCESSOR_CACHE_KEY_PREFIX: "nvidia-stdexec-dev/preprocessor"
-          SCCACHE_S3_USE_PREPROCESSOR_CACHE_MODE: "true"
-          SCCACHE_SERVER_LOG: "sccache=debug"
-          SCCACHE_SERVER_PORT: "4225"
-        run: |
-          set -e;
-          source /etc/profile
-          set -x;
-
-          devcontainer-utils-install-sccache --version rapids;
-
-          devcontainer-utils-init-sccache-dist                                          \
-              --enable-sccache-dist - <<< "                                             \
-              --auth-type 'token'                                                       \
-              --auth-token '$(                                                          \
-                curl -fsSL -H "Authorization: Bearer $(                                 \
-                  curl -fsSL -H "Authorization: bearer $ACTIONS_ID_TOKEN_REQUEST_TOKEN" \
-                    "${ACTIONS_ID_TOKEN_REQUEST_URL}&audience=token.rapids.nvidia.com"  \
-                | jq -r '.value'                                                        \
-                )" https://token.rapids.nvidia.com/gh/token/exchange                    \
-              | jq -r '.token')'                                                        \
-          ";
-
-          # Copy source folder into ~/stdexec
-          cp -r "${GITHUB_WORKSPACE}"/stdexec ~/;
-          chown -R coder:coder ~/stdexec;
-          cd ~/stdexec;
-
-          # Configure
-          cmake -S . -B build -GNinja \
-            -DCMAKE_BUILD_TYPE=${{ matrix.build }} \
-            -DCMAKE_CXX_FLAGS="${{ matrix.cxxflags }}" \
-            -DSTDEXEC_ENABLE_TBB:BOOL=${{ !contains(matrix.cxxflags, '-fsanitize') }} \
-            -DSTDEXEC_ENABLE_ASIO:BOOL=TRUE \
-            -DSTDEXEC_ASIO_IMPLEMENTATION:STRING=boost \
-            -DCMAKE_CXX_STANDARD:STRING=${{ matrix.cxxstd }} \
-            -DCMAKE_CXX_EXTENSIONS:BOOL=OFF \
-            -DSTDEXEC_BUILD_TESTS:BOOL=ON \
-            ;
-
-          # Compile
-          cmake --build build -v -j 512;
-
-          # Print sccache stats
-          sccache -s;
-
-          # Tests
-          SCCACHE_NO_CACHE=1 SCCACHE_NO_DIST_COMPILE=1 \
-          ctest --test-dir build --verbose --output-on-failure --timeout 60;
-      - if: ${{ !cancelled() }}
-        name: Upload sccache logs
-        uses: actions/upload-artifact@v4
-        with:
-          name: sccache-client-logs-${{env.ARTIFACT_PREFIX}}-${{env.ARTIFACT_SUFFIX}}
-          path: /tmp/sccache*.log
-          compression-level: 9
-
-  ci-cpu:
-    runs-on: ubuntu-latest
-    name: CI (CPU)
-    needs:
-      - build-cpu
-    steps:
-      - run: echo "CI (CPU) success"
+  # build-cpu:
+  #   runs-on: ubuntu-latest
+  #   name: ${{ matrix.name }}
+  #   strategy:
+  #     fail-fast: false
+  #     matrix:
+  #       include:
+  #         - { name: "CPU (clang 16, Debug)",         build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++" }
+  #         - { name: "CPU (clang 16, Debug, c++23)",  build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "23", cxxflags: "-stdlib=libc++" }
+  #         - { name: "CPU (clang 16, Debug, TSAN)",   build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-fsanitize=thread" }
+  #         - { name: "CPU (clang 16, Release)",       build: "Release", tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++" }
+  #         - { name: "CPU (clang 16, Release, ASAN)", build: "Release", tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++ -fsanitize=address -fsanitize-ignorelist=/home/coder/stdexec/sanitizer-ignorelist.txt" }
+  #         - { name: "CPU (clang 22, Debug)",         build: "Debug",   tag: llvm22-cuda13.2, cxxstd: "23", cxxflags: "-stdlib=libc++" }
+  #         - { name: "CPU (clang 22, Release)",       build: "Release", tag: llvm22-cuda13.2, cxxstd: "23", cxxflags: "-stdlib=libc++" }
+  #         - { name: "CPU (gcc 12, Debug)",           build: "Debug",   tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "", }
+  #         - { name: "CPU (gcc 12, Release)",         build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "", }
+  #         # With the following config, 2 tests mysteriously time out, but only in CI and not locally.
+  #         # - { name: "CPU (gcc 12, Release, ASAN)",   build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
+  #         - { name: "CPU (gcc 12, Release, TSAN)",   build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=thread" }
+  #         - { name: "CPU (gcc 13, Debug)",           build: "Debug",   tag: gcc13-cuda12.9,  cxxstd: "20", cxxflags: "", }
+  #         - { name: "CPU (gcc 14, Debug)",           build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "", }
+  #         - { name: "CPU (gcc 14, Debug, ASAN)",     build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
+  #         - { name: "CPU (gcc 14, Debug, TSAN)",     build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=thread" }
+  #         - { name: "CPU (gcc 14, Release, ASAN)",   build: "Release", tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
+  #         - { name: "CPU (gcc 14, Release, LEAK)",   build: "Release", tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=leak", }
+  #         - { name: "CPU (gcc 14, Release, c++23)",  build: "Release", tag: gcc14-cuda12.9,  cxxstd: "23", cxxflags: "", }
+  #   container:
+  #     options: -u root
+  #     image: rapidsai/devcontainers:26.06-cpp-${{ matrix.tag }}
+  #   permissions:
+  #     id-token: write # This is required for configure-aws-credentials
+  #     contents: read  # This is required for actions/checkout
+  #   defaults:
+  #     run:
+  #       shell: su coder {0}
+  #       working-directory: /home/coder
+  #   steps:
+  #     - name: Checkout stdexec
+  #       uses: actions/checkout@v4
+  #       with:
+  #         path: stdexec
+  #         persist-credentials: false
+  #     - name: Setup environment
+  #       run: |
+  #         echo "ARTIFACT_PREFIX=${{runner.os}}-${{matrix.tag}}-amd64" >> "${GITHUB_ENV}"
+  #         echo "ARTIFACT_SUFFIX=${{github.run_id}}-${{github.run_attempt}}-$RANDOM" >> "${GITHUB_ENV}"
+  #     - if: github.repository_owner == 'NVIDIA'
+  #       name: Get AWS credentials for sccache bucket
+  #       uses: aws-actions/configure-aws-credentials@v4
+  #       with:
+  #         aws-region: us-east-2
+  #         role-duration-seconds: 28800 # 8 hours
+  #         role-to-assume: arn:aws:iam::279114543810:role/gha-oidc-NVIDIA
+  #     - name: Build and test CPU schedulers
+  #       env:
+  #         ASAN_OPTIONS: alloc_dealloc_mismatch=0
+  #         NVCC_APPEND_FLAGS: "-t=100"
+  #         SCCACHE_BUCKET: "rapids-sccache-devs"
+  #         SCCACHE_DIST_REQUEST_TIMEOUT: "7140"
+  #         SCCACHE_DIST_URL: "https://amd64.linux.sccache.rapids.nvidia.com"
+  #         SCCACHE_IDLE_TIMEOUT: "0"
+  #         SCCACHE_REGION: "us-east-2"
+  #         SCCACHE_S3_KEY_PREFIX: "nvidia-stdexec-dev"
+  #         SCCACHE_S3_PREPROCESSOR_CACHE_KEY_PREFIX: "nvidia-stdexec-dev/preprocessor"
+  #         SCCACHE_S3_USE_PREPROCESSOR_CACHE_MODE: "true"
+  #         SCCACHE_SERVER_LOG: "sccache=debug"
+  #         SCCACHE_SERVER_PORT: "4225"
+  #       run: |
+  #         set -e;
+  #         source /etc/profile
+  #         set -x;
+
+  #         devcontainer-utils-install-sccache --version rapids;
+
+  #         devcontainer-utils-init-sccache-dist                                          \
+  #             --enable-sccache-dist - <<< "                                             \
+  #             --auth-type 'token'                                                       \
+  #             --auth-token '$(                                                          \
+  #               curl -fsSL -H "Authorization: Bearer $(                                 \
+  #                 curl -fsSL -H "Authorization: bearer $ACTIONS_ID_TOKEN_REQUEST_TOKEN" \
+  #                   "${ACTIONS_ID_TOKEN_REQUEST_URL}&audience=token.rapids.nvidia.com"  \
+  #               | jq -r '.value'                                                        \
+  #               )" https://token.rapids.nvidia.com/gh/token/exchange                    \
+  #             | jq -r '.token')'                                                        \
+  #         ";
+
+  #         # Copy source folder into ~/stdexec
+  #         cp -r "${GITHUB_WORKSPACE}"/stdexec ~/;
+  #         chown -R coder:coder ~/stdexec;
+  #         cd ~/stdexec;
+
+  #         # Configure
+  #         cmake -S . -B build -GNinja \
+  #           -DCMAKE_BUILD_TYPE=${{ matrix.build }} \
+  #           -DCMAKE_CXX_FLAGS="${{ matrix.cxxflags }}" \
+  #           -DSTDEXEC_ENABLE_TBB:BOOL=${{ !contains(matrix.cxxflags, '-fsanitize') }} \
+  #           -DSTDEXEC_ENABLE_ASIO:BOOL=TRUE \
+  #           -DSTDEXEC_ASIO_IMPLEMENTATION:STRING=boost \
+  #           -DCMAKE_CXX_STANDARD:STRING=${{ matrix.cxxstd }} \
+  #           -DCMAKE_CXX_EXTENSIONS:BOOL=OFF \
+  #           -DSTDEXEC_BUILD_TESTS:BOOL=ON \
+  #           ;
+
+  #         # Compile
+  #         cmake --build build -v -j 512;
+
+  #         # Print sccache stats
+  #         sccache -s;
+
+  #         # Tests
+  #         SCCACHE_NO_CACHE=1 SCCACHE_NO_DIST_COMPILE=1 \
+  #         ctest --test-dir build --verbose --output-on-failure --timeout 60;
+  #     - if: ${{ !cancelled() }}
+  #       name: Upload sccache logs
+  #       uses: actions/upload-artifact@v4
+  #       with:
+  #         name: sccache-client-logs-${{env.ARTIFACT_PREFIX}}-${{env.ARTIFACT_SUFFIX}}
+  #         path: /tmp/sccache*.log
+  #         compression-level: 9
+
+  # ci-cpu:
+  #   runs-on: ubuntu-latest
+  #   name: CI (CPU)
+  #   needs:
+  #     - build-cpu
+  #   steps:
+  #     - run: echo "CI (CPU) success"
 
   build-cpu-windows:
     runs-on: windows-2022
@@ -174,47 +174,47 @@ jobs:
     steps:
       - run: echo "CI (CPU) (Windows) success"
 
-  build-cpu-macos:
-    runs-on: macos-26-large
-    name: macos-${{ matrix.name }}
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-          - { compiler: "clang++",  build: "Debug",   name: "CPU (MacOS) (clang, Debug)" }
-          - { compiler: "clang++",  build: "Release", name: "CPU (MacOS) (clang, Release)" }
-    steps:
-      - name: Checkout stdexec (MacOS)
-        uses: actions/checkout@v4
-        with:
-          persist-credentials: false
-
-      - name: Install dependencies
-        run: |
-          brew update
-          brew install ninja
-        shell: bash
-
-      - name: Build and test CPU schedulers (MacOS)
-        shell: bash
-        run: |
-          mkdir build
-          cmake -S. -Bbuild -GNinja \
-            -DCMAKE_BUILD_TYPE=${{ matrix.build }} \
-            -DCMAKE_CXX_COMPILER=${{ matrix.compiler }} \
-            -DSTDEXEC_ENABLE_ASIO:BOOL=TRUE \
-            -DSTDEXEC_ASIO_IMPLEMENTATION:STRING=boost \
-            -DCMAKE_CXX_STANDARD:STRING=20 \
-            -DSTDEXEC_BUILD_TESTS:BOOL=ON
-
-          cmake --build build/ -v
-          cd build
-          ctest --output-on-failure
-
-  ci-cpu-macos:
-    runs-on: macos-latest-large
-    name: CI (CPU) (MacOS)
-    needs:
-      - build-cpu-macos
-    steps:
-      - run: echo "CI (CPU) (MacOS) success"
+  # build-cpu-macos:
+  #   runs-on: macos-26-large
+  #   name: macos-${{ matrix.name }}
+  #   strategy:
+  #     fail-fast: false
+  #     matrix:
+  #       include:
+  #         - { compiler: "clang++",  build: "Debug",   name: "CPU (MacOS) (clang, Debug)" }
+  #         - { compiler: "clang++",  build: "Release", name: "CPU (MacOS) (clang, Release)" }
+  #   steps:
+  #     - name: Checkout stdexec (MacOS)
+  #       uses: actions/checkout@v4
+  #       with:
+  #         persist-credentials: false
+
+  #     - name: Install dependencies
+  #       run: |
+  #         brew update
+  #         brew install ninja
+  #       shell: bash
+
+  #     - name: Build and test CPU schedulers (MacOS)
+  #       shell: bash
+  #       run: |
+  #         mkdir build
+  #         cmake -S. -Bbuild -GNinja \
+  #           -DCMAKE_BUILD_TYPE=${{ matrix.build }} \
+  #           -DCMAKE_CXX_COMPILER=${{ matrix.compiler }} \
+  #           -DSTDEXEC_ENABLE_ASIO:BOOL=TRUE \
+  #           -DSTDEXEC_ASIO_IMPLEMENTATION:STRING=boost \
+  #           -DCMAKE_CXX_STANDARD:STRING=20 \
+  #           -DSTDEXEC_BUILD_TESTS:BOOL=ON
+
+  #         cmake --build build/ -v
+  #         cd build
+  #         ctest --output-on-failure
+
+  # ci-cpu-macos:
+  #   runs-on: macos-latest-large
+  #   name: CI (CPU) (MacOS)
+  #   needs:
+  #     - build-cpu-macos
+  #   steps:
+  #     - run: echo "CI (CPU) (MacOS) success"
diff --git a/.github/workflows/test-windows.ps1 b/.github/workflows/test-windows.ps1
index 4f579700f..e6d0ec313 100644
--- a/.github/workflows/test-windows.ps1
+++ b/.github/workflows/test-windows.ps1
@@ -29,8 +29,8 @@ Invoke-NativeCommand cmake -B $BuildDirectory -G Ninja `
 Invoke-NativeCommand cmake --build $BuildDirectory
 
 # Enable AppVerifier for the test executables
-Invoke-NativeCommand appverif -enable Basics -for test.stdexec.exe
-Invoke-NativeCommand appverif -enable Basics -for test.exec.exe
+Invoke-NativeCommand appverif /verify test.stdexec.exe
+Invoke-NativeCommand appverif /verify test.exec.exe
 
 Invoke-NativeCommand ctest --test-dir $BuildDirectory --output-on-failure --verbose --timeout 60
 

From 49f170416643c9f7fc73bd93e635e55ea4160b8c Mon Sep 17 00:00:00 2001
From: Eric Niebler <eniebler@nvidia.com>
Date: Thu, 30 Apr 2026 16:25:44 -0700
Subject: [PATCH 3/7] wrap unhandled_stopped coroutine handles on msvc

---
 .github/workflows/test-windows.ps1            |  8 +--
 include/exec/at_coroutine_exit.hpp            |  4 +-
 include/exec/on_coro_disposition.hpp          |  2 +-
 include/exec/task.hpp                         |  7 +-
 include/stdexec/__detail/__as_awaitable.hpp   | 10 ++-
 include/stdexec/__detail/__config.hpp         |  2 +-
 .../stdexec/__detail/__connect_awaitable.hpp  | 24 +++----
 include/stdexec/__detail/__task.hpp           |  2 +-
 include/stdexec/coroutine.hpp                 | 67 +++++++++++++++----
 9 files changed, 84 insertions(+), 42 deletions(-)

diff --git a/.github/workflows/test-windows.ps1 b/.github/workflows/test-windows.ps1
index e6d0ec313..5f2774ff5 100644
--- a/.github/workflows/test-windows.ps1
+++ b/.github/workflows/test-windows.ps1
@@ -21,16 +21,12 @@ New-Item -ItemType Directory $BuildDirectory | Out-Null
 
 Invoke-NativeCommand cmake -B $BuildDirectory -G Ninja `
 	"-DCMAKE_BUILD_TYPE=$Config" `
+	"-DCMAKE_VERBOSE_MAKEFILE:BOOL=ON" `
 	"-DCMAKE_MSVC_DEBUG_INFORMATION_FORMAT:STRING=Embedded" `
+	"-DCMAKE_CXX_FLAGS:STRING=/fsanitize=address" `
 	"-DSTDEXEC_ENABLE_ASIO:BOOL=TRUE" `
 	"-DSTDEXEC_ASIO_IMPLEMENTATION:STRING=boost" `
-	"-DCMAKE_VERBOSE_MAKEFILE:BOOL=ON" `
 	"-DSTDEXEC_BUILD_TESTS:BOOL=TRUE" .
 Invoke-NativeCommand cmake --build $BuildDirectory
-
-# Enable AppVerifier for the test executables
-Invoke-NativeCommand appverif /verify test.stdexec.exe
-Invoke-NativeCommand appverif /verify test.exec.exe
-
 Invoke-NativeCommand ctest --test-dir $BuildDirectory --output-on-failure --verbose --timeout 60
 
diff --git a/include/exec/at_coroutine_exit.hpp b/include/exec/at_coroutine_exit.hpp
index a54668d63..9a872f67f 100644
--- a/include/exec/at_coroutine_exit.hpp
+++ b/include/exec/at_coroutine_exit.hpp
@@ -141,7 +141,7 @@ namespace experimental::execution
       ~__task()
       {
         if (__coro_)
-          __coro_.destroy();
+          STDEXEC::__coroutine_destroy_nothrow(__coro_);
       }
 
       [[nodiscard]]
@@ -189,7 +189,7 @@ namespace experimental::execution
         {
           auto __cont = __h.promise().continuation();
           auto __coro = __h.promise().__is_stopped_ ? __cont.unhandled_stopped() : __cont.handle();
-          return STDEXEC_CORO_DESTROY_AND_CONTINUE(__h, __coro);
+          return STDEXEC::__coroutine_destroy_and_continue(__h, __coro);
         }
 
         static constexpr void await_resume() noexcept {}
diff --git a/include/exec/on_coro_disposition.hpp b/include/exec/on_coro_disposition.hpp
index bfd8385cd..edb8b2ed1 100644
--- a/include/exec/on_coro_disposition.hpp
+++ b/include/exec/on_coro_disposition.hpp
@@ -121,7 +121,7 @@ namespace experimental::execution
         {
           auto __cont = __h.promise().continuation();
           auto __coro = __h.promise().__is_stopped_ ? __cont.unhandled_stopped() : __cont.handle();
-          return STDEXEC_CORO_DESTROY_AND_CONTINUE(__h, __coro);
+          return STDEXEC::__coroutine_destroy_and_continue(__h, __coro);
         }
 
         void await_resume() const noexcept {}
diff --git a/include/exec/task.hpp b/include/exec/task.hpp
index c73746769..e0a3f4af3 100644
--- a/include/exec/task.hpp
+++ b/include/exec/task.hpp
@@ -531,7 +531,7 @@ namespace experimental::execution
       constexpr ~basic_task()
       {
         if (__coro_)
-          __coro_.destroy();
+          STDEXEC::__coroutine_destroy_nothrow(__coro_);
       }
 
      private:
@@ -653,7 +653,7 @@ namespace experimental::execution
         constexpr ~__task_awaiter()
         {
           if (__coro_)
-            __coro_.destroy();
+            STDEXEC::__coroutine_destroy_nothrow(__coro_);
         }
 
         static constexpr auto await_ready() noexcept -> bool
@@ -672,7 +672,8 @@ namespace experimental::execution
           if constexpr (requires { __coro_.promise().stop_requested() ? 0 : 1; })
           {
             if (__coro_.promise().stop_requested())
-              return __parent.promise().unhandled_stopped();
+              return STDEXEC::__coroutine_destroy_and_continue(
+                __parent.promise().unhandled_stopped());
           }
           return __coro_;
         }
diff --git a/include/stdexec/__detail/__as_awaitable.hpp b/include/stdexec/__detail/__as_awaitable.hpp
index 1d154e29b..e489263c6 100644
--- a/include/stdexec/__detail/__as_awaitable.hpp
+++ b/include/stdexec/__detail/__as_awaitable.hpp
@@ -133,8 +133,14 @@ namespace STDEXEC
         // If the operation was stopped (__result_ is valueless), we should use the
         // unhandled_stopped() continuation. Otherwise, should resume the __continuation_
         // as normal.
-        return __result_.__is_valueless() ? __continuation_.unhandled_stopped()
-                                          : __continuation_.handle();
+        if (__result_.__is_valueless())
+        {
+          return STDEXEC::__coroutine_destroy_and_continue(__continuation_.unhandled_stopped());
+        }
+        else
+        {
+          return __continuation_.handle();
+        }
       }
 
       __coroutine_handle<> __continuation_;
diff --git a/include/stdexec/__detail/__config.hpp b/include/stdexec/__detail/__config.hpp
index 3106b08fa..cf895ed44 100644
--- a/include/stdexec/__detail/__config.hpp
+++ b/include/stdexec/__detail/__config.hpp
@@ -321,7 +321,7 @@ namespace STDEXEC::__std
 #  define STDEXEC_ATTR_WHICH_4(_ATTR) __forceinline
 #elif STDEXEC_CLANG()
 #  define STDEXEC_ATTR_WHICH_4(_ATTR)                                                              \
-    inline //__attribute__((__always_inline__, __artificial__, __nodebug__)) inline
+    __attribute__((__always_inline__, __artificial__, __nodebug__)) inline
 #elif STDEXEC_GCC()
 #  define STDEXEC_ATTR_WHICH_4(_ATTR) __attribute__((__always_inline__, __artificial__)) inline
 #else
diff --git a/include/stdexec/__detail/__connect_awaitable.hpp b/include/stdexec/__detail/__connect_awaitable.hpp
index efbe443cf..9b24af9ec 100644
--- a/include/stdexec/__detail/__connect_awaitable.hpp
+++ b/include/stdexec/__detail/__connect_awaitable.hpp
@@ -182,7 +182,7 @@ namespace STDEXEC
 
         ~__state()
         {
-          // make sure to destroy in the reverse order of construction
+          // make sure to __destroy in the reverse order of construction
           __awaiter_.__destroy();
           __awaitable_.__destroy();
         }
@@ -206,12 +206,12 @@ namespace STDEXEC
         : __source_awaitable_(static_cast<_Awaitable2&&>(__awaitable))
       {}
 
-      constexpr void construct(__std::coroutine_handle<_Promise> __coro) noexcept(__is_nothrow)
+      constexpr void __construct(__std::coroutine_handle<_Promise> __coro) noexcept(__is_nothrow)
       {
         __awaiter_.__construct(static_cast<_Awaitable&&>(__source_awaitable_), __coro);
       }
 
-      constexpr void destroy() noexcept
+      constexpr void __destroy() noexcept
       {
         __awaiter_.__destroy();
       }
@@ -263,12 +263,12 @@ namespace STDEXEC
         : __source_awaitable_(static_cast<_Awaitable2&&>(__awaitable))
       {}
 
-      constexpr void construct(__std::coroutine_handle<_Promise> __coro) noexcept(__is_nothrow)
+      constexpr void __construct(__std::coroutine_handle<_Promise> __coro) noexcept(__is_nothrow)
       {
         __awaiter_.__construct(static_cast<_Awaitable&&>(__source_awaitable_), __coro);
       }
 
-      constexpr void destroy() noexcept
+      constexpr void __destroy() noexcept
       {
         __awaiter_.__destroy();
       }
@@ -328,12 +328,12 @@ namespace STDEXEC
         : __source_awaitable_(static_cast<_Awaitable2&&>(__awaitable))
       {}
 
-      constexpr void construct(__std::coroutine_handle<_Promise> __coro) noexcept(__is_nothrow)
+      constexpr void __construct(__std::coroutine_handle<_Promise> __coro) noexcept(__is_nothrow)
       {
         __awaiter_.__construct(static_cast<_Awaitable&&>(__source_awaitable_), __coro);
       }
 
-      constexpr void destroy() noexcept
+      constexpr void __destroy() noexcept
       {
         __awaiter_.__destroy();
       }
@@ -363,12 +363,12 @@ namespace STDEXEC
         __awaiter_.__destroy();
       }
 
-      static constexpr void construct(__std::coroutine_handle<_Promise>) noexcept
+      static constexpr void __construct(__std::coroutine_handle<_Promise>) noexcept
       {
         // no-op
       }
 
-      static constexpr void destroy() noexcept
+      static constexpr void __destroy() noexcept
       {
         // no-op
       }
@@ -389,7 +389,7 @@ namespace STDEXEC
       {
         if (__started_)
         {
-          __awaiter_.destroy();
+          __awaiter_.__destroy();
         }
       }
 
@@ -399,7 +399,7 @@ namespace STDEXEC
 
         STDEXEC_TRY
         {
-          __awaiter_.construct(__coro);
+          __awaiter_.__construct(__coro);
           __started_ = true;
 
           if (!__awaiter_.await_ready())
@@ -439,7 +439,7 @@ namespace STDEXEC
         }
         STDEXEC_CATCH_ALL
         {
-          if constexpr (!noexcept(__awaiter_.construct(__coro))
+          if constexpr (!noexcept(__awaiter_.__construct(__coro))
                         || !noexcept(__awaiter_.await_ready())
                         || !noexcept(__awaiter_.await_suspend(__coro)))
           {
diff --git a/include/stdexec/__detail/__task.hpp b/include/stdexec/__detail/__task.hpp
index 9a8120569..a2350be9e 100644
--- a/include/stdexec/__detail/__task.hpp
+++ b/include/stdexec/__detail/__task.hpp
@@ -287,7 +287,7 @@ namespace STDEXEC
     constexpr ~task()
     {
       if (__coro_)
-        __coro_.destroy();
+        STDEXEC::__coroutine_destroy_nothrow(__coro_);
     }
 
     [[nodiscard]]
diff --git a/include/stdexec/coroutine.hpp b/include/stdexec/coroutine.hpp
index f26378a97..7bc4680af 100644
--- a/include/stdexec/coroutine.hpp
+++ b/include/stdexec/coroutine.hpp
@@ -33,30 +33,28 @@ namespace STDEXEC
     return __std::coroutine_handle<_Tp>::from_address(__h.address());
   }
 
-  inline void __coroutine_resume_nothrow(__std::coroutine_handle<> __h) noexcept
+  STDEXEC_ATTRIBUTE(always_inline)
+  void __coroutine_resume_nothrow(__std::coroutine_handle<> __h) noexcept
   {
     STDEXEC_TRY
     {
-      STDEXEC_ASSERT(__h);
-      __h.resume();
+      __builtin_coro_resume(__h.address());
     }
     STDEXEC_CATCH_ALL
     {
-      STDEXEC_ASSERT(!"Coroutine resume threw an exception!");
       __std::unreachable();
     }
   }
 
-  inline void __coroutine_destroy_nothrow(__std::coroutine_handle<> __h) noexcept
+  STDEXEC_ATTRIBUTE(always_inline)
+  void __coroutine_destroy_nothrow(__std::coroutine_handle<> __h) noexcept
   {
     STDEXEC_TRY
     {
-      STDEXEC_ASSERT(__h);
-      __h.destroy();
+      __builtin_coro_destroy(__h.address());
     }
     STDEXEC_CATCH_ALL
     {
-      STDEXEC_ASSERT(!"Coroutine destroy threw an exception!");
       __std::unreachable();
     }
   }
@@ -174,7 +172,7 @@ namespace STDEXEC
       sizeof(__synthetic_coro_frame));
   }  // namespace __detail
 
-#  if STDEXEC_MSVC() && STDEXEC_MSVC_VERSION <= 1939
+#  if STDEXEC_MSVC() && STDEXEC_MSVC_VERSION < 1950
   // MSVCBUG https://developercommunity.visualstudio.com/t/destroy-coroutine-from-final_suspend-r/10096047
 
   // Prior to Visual Studio 17.9 (Feb, 2024), aka MSVC 19.39, MSVC incorrectly allocates
@@ -212,6 +210,26 @@ namespace STDEXEC
     } __promise_;
   };
 
+  struct __symmetric_transfer_frame : __detail::__synthetic_coro_frame
+  {
+    constexpr __symmetric_transfer_frame() noexcept
+      : __detail::__synthetic_coro_frame{&__symmetric_transfer_frame::__resume}
+    {}
+
+    static void __resume(void* __address) noexcept
+    {
+      // Make a local copy of the promise to ensure we can safely destroy the suspended
+      // coroutine after resuming the continuation.
+      auto __promise = static_cast<__symmetric_transfer_frame*>(__address)->__promise_;
+      STDEXEC::__coroutine_resume_nothrow(__promise.__continue_);
+    }
+
+    struct __promise
+    {
+      __std::coroutine_handle<> __continue_{};
+    } __promise_;
+  };
+
   inline auto __coroutine_destroy_and_continue(__std::coroutine_handle<> __destroy,            //
                                                __std::coroutine_handle<> __continue) noexcept  //
     -> __std::coroutine_handle<>
@@ -222,12 +240,33 @@ namespace STDEXEC
     return __std::coroutine_handle<>::from_address(&__fr);
   }
 
-#    define STDEXEC_CORO_DESTROY_AND_CONTINUE(__destroy, __continue)                      \
-       ::STDEXEC::__coroutine_destroy_and_continue(__destroy, __continue)
+  inline auto __coroutine_destroy_and_continue(__std::coroutine_handle<> __continue) noexcept  //
+    -> __std::coroutine_handle<>
+  {
+    static constinit thread_local __symmetric_transfer_frame __fr;
+    __fr.__promise_.__continue_ = __continue;
+    return __std::coroutine_handle<>::from_address(&__fr);
+  }
+
 #  else
-#    define STDEXEC_CORO_DESTROY_AND_CONTINUE(__destroy, __continue)                      \
-       (__destroy.destroy(), __continue)
-#  endif
+
+  STDEXEC_ATTRIBUTE(always_inline)
+  auto __coroutine_destroy_and_continue(__std::coroutine_handle<> __destroy,            //
+                                        __std::coroutine_handle<> __continue) noexcept  //
+    -> __std::coroutine_handle<>
+  {
+    ::STDEXEC::__coroutine_destroy_nothrow(__destroy);
+    return __continue;
+  }
+
+  STDEXEC_ATTRIBUTE(always_inline)
+  auto __coroutine_destroy_and_continue(__std::coroutine_handle<> __continue) noexcept  //
+    -> __std::coroutine_handle<>
+  {
+    return __continue;
+  }
+
+#  endif  // STDEXEC_MSVC() && STDEXEC_MSVC_VERSION < 1950
 }  // namespace STDEXEC
 
 #endif  // !STDEXEC_NO_STDCPP_COROUTINES()

From 9b473a38f9d1b0df87b0344e5d5e892ceb483025 Mon Sep 17 00:00:00 2001
From: Eric Niebler <eniebler@nvidia.com>
Date: Thu, 30 Apr 2026 16:57:33 -0700
Subject: [PATCH 4/7] try to convince msvc to constinit the synthetic coro
 frames

---
 include/stdexec/coroutine.hpp | 47 ++++++++++++++++-------------------
 1 file changed, 21 insertions(+), 26 deletions(-)

diff --git a/include/stdexec/coroutine.hpp b/include/stdexec/coroutine.hpp
index 7bc4680af..ba3fbf38b 100644
--- a/include/stdexec/coroutine.hpp
+++ b/include/stdexec/coroutine.hpp
@@ -155,17 +155,12 @@ namespace STDEXEC
   {
     struct __synthetic_coro_frame
     {
-      void (*__resume_)(void*) noexcept;
-      // we never invoke __destroy_ so a no-op implementation is fine; we've chosen the
-      // address of a no-op function rather than nullptr in case some rogue awaitable
-      // *does* invoke destroy on the synthesized handle that it receives in its
-      // await_suspend function
-      void (*__destroy_)(void*) noexcept = &__noop_destroy;
-
-      static void __noop_destroy(void*) noexcept
-      {
-        STDEXEC_ASSERT(!"Attempt to destroy a synthetic coroutine!");
-      }
+      using __callback_fn_t = void(void*) noexcept;
+
+      __callback_fn_t* __resume_  = &__noop_fn;
+      __callback_fn_t* __destroy_ = &__noop_fn;
+
+      static void __noop_fn(void*) noexcept {}
     };
 
     static constexpr std::ptrdiff_t __coro_promise_offset = static_cast<std::ptrdiff_t>(
@@ -190,10 +185,6 @@ namespace STDEXEC
 
   struct __destroy_and_continue_frame : __detail::__synthetic_coro_frame
   {
-    constexpr __destroy_and_continue_frame() noexcept
-      : __detail::__synthetic_coro_frame{&__destroy_and_continue_frame::__resume}
-    {}
-
     static void __resume(void* __address) noexcept
     {
       // Make a local copy of the promise to ensure we can safely destroy the suspended
@@ -208,14 +199,15 @@ namespace STDEXEC
       __std::coroutine_handle<> __destroy_{};
       __std::coroutine_handle<> __continue_{};
     } __promise_;
+
+    static thread_local __destroy_and_continue_frame value;
   };
 
+  constinit inline thread_local __destroy_and_continue_frame __destroy_and_continue_frame::value{
+    {&__destroy_and_continue_frame::__resume}, {}};
+
   struct __symmetric_transfer_frame : __detail::__synthetic_coro_frame
   {
-    constexpr __symmetric_transfer_frame() noexcept
-      : __detail::__synthetic_coro_frame{&__symmetric_transfer_frame::__resume}
-    {}
-
     static void __resume(void* __address) noexcept
     {
       // Make a local copy of the promise to ensure we can safely destroy the suspended
@@ -228,24 +220,27 @@ namespace STDEXEC
     {
       __std::coroutine_handle<> __continue_{};
     } __promise_;
+
+    static thread_local __symmetric_transfer_frame value;
   };
 
+  constinit inline thread_local __symmetric_transfer_frame __symmetric_transfer_frame::value{
+    {&__symmetric_transfer_frame::__resume}, {}};
+
   inline auto __coroutine_destroy_and_continue(__std::coroutine_handle<> __destroy,            //
                                                __std::coroutine_handle<> __continue) noexcept  //
     -> __std::coroutine_handle<>
   {
-    static constinit thread_local __destroy_and_continue_frame __fr;
-    __fr.__promise_.__destroy_  = __destroy;
-    __fr.__promise_.__continue_ = __continue;
-    return __std::coroutine_handle<>::from_address(&__fr);
+    __destroy_and_continue_frame::value.__promise_.__destroy_  = __destroy;
+    __destroy_and_continue_frame::value.__promise_.__continue_ = __continue;
+    return __std::coroutine_handle<>::from_address(&__destroy_and_continue_frame::value);
   }
 
   inline auto __coroutine_destroy_and_continue(__std::coroutine_handle<> __continue) noexcept  //
     -> __std::coroutine_handle<>
   {
-    static constinit thread_local __symmetric_transfer_frame __fr;
-    __fr.__promise_.__continue_ = __continue;
-    return __std::coroutine_handle<>::from_address(&__fr);
+    __symmetric_transfer_frame::value.__promise_.__continue_ = __continue;
+    return __std::coroutine_handle<>::from_address(&__symmetric_transfer_frame::value);
   }
 
 #  else

From df6213ec24993001dc368dade0a3f0097286e5e7 Mon Sep 17 00:00:00 2001
From: Eric Niebler <eniebler@nvidia.com>
Date: Thu, 30 Apr 2026 17:50:22 -0700
Subject: [PATCH 5/7] try again

---
 .github/workflows/test-windows.ps1    |  3 +--
 include/stdexec/__detail/__config.hpp | 17 +++++++++++++
 include/stdexec/coroutine.hpp         | 36 +++++++++++++++++++--------
 3 files changed, 44 insertions(+), 12 deletions(-)

diff --git a/.github/workflows/test-windows.ps1 b/.github/workflows/test-windows.ps1
index 5f2774ff5..c20ecc521 100644
--- a/.github/workflows/test-windows.ps1
+++ b/.github/workflows/test-windows.ps1
@@ -23,10 +23,9 @@ Invoke-NativeCommand cmake -B $BuildDirectory -G Ninja `
 	"-DCMAKE_BUILD_TYPE=$Config" `
 	"-DCMAKE_VERBOSE_MAKEFILE:BOOL=ON" `
 	"-DCMAKE_MSVC_DEBUG_INFORMATION_FORMAT:STRING=Embedded" `
-	"-DCMAKE_CXX_FLAGS:STRING=/fsanitize=address" `
+	"-DCMAKE_CXX_FLAGS:STRING=/fsanitize=address /EHsc" `
 	"-DSTDEXEC_ENABLE_ASIO:BOOL=TRUE" `
 	"-DSTDEXEC_ASIO_IMPLEMENTATION:STRING=boost" `
 	"-DSTDEXEC_BUILD_TESTS:BOOL=TRUE" .
 Invoke-NativeCommand cmake --build $BuildDirectory
 Invoke-NativeCommand ctest --test-dir $BuildDirectory --output-on-failure --verbose --timeout 60
-
diff --git a/include/stdexec/__detail/__config.hpp b/include/stdexec/__detail/__config.hpp
index cf895ed44..ad3643cad 100644
--- a/include/stdexec/__detail/__config.hpp
+++ b/include/stdexec/__detail/__config.hpp
@@ -142,6 +142,13 @@
 #  define STDEXEC_HAS_ATTRIBUTE(...) 0
 #endif
 
+////////////////////////////////////////////////////////////////////////////////////////////////////
+#if defined(__has_cpp_attribute)
+#  define STDEXEC_HAS_CPP_ATTRIBUTE(...) __has_cpp_attribute(__VA_ARGS__)
+#else
+#  define STDEXEC_HAS_CPP_ATTRIBUTE(...) 0
+#endif
+
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 #if STDEXEC_CLANG() && STDEXEC_CUDA_COMPILATION()
 #  define STDEXEC_HOST_DEVICE_DEDUCTION_GUIDE __host__ __device__
@@ -370,6 +377,16 @@ namespace STDEXEC::__std
 #define STDEXEC_ATTR_noinline     STDEXEC_PP_PROBE(~, 9)
 #define STDEXEC_ATTR___noinline__ STDEXEC_PP_PROBE(~, 9)
 
+#if STDEXEC_MSVC() && !STDEXEC_CLANG_CL() && STDEXEC_MSVC_VERSION >= 1950
+#  define STDEXEC_ATTR_WHICH_10(_ATTR) [[msvc::musttail]]
+#elif STDEXEC_HAS_CPP_ATTRIBUTE(gnu::musttail)
+#  define STDEXEC_ATTR_WHICH_10(_ATTR) [[gnu::musttail]]
+#else
+#  define STDEXEC_ATTR_WHICH_10(_ATTR) /*nothing*/
+#endif
+#define STDEXEC_ATTR_musttail     STDEXEC_PP_PROBE(~, 10)
+#define STDEXEC_ATTR___musttail__ STDEXEC_PP_PROBE(~, 10)
+
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // warning push/pop portability macros
 #if STDEXEC_NVCC()
diff --git a/include/stdexec/coroutine.hpp b/include/stdexec/coroutine.hpp
index ba3fbf38b..06d9eda32 100644
--- a/include/stdexec/coroutine.hpp
+++ b/include/stdexec/coroutine.hpp
@@ -34,11 +34,11 @@ namespace STDEXEC
   }
 
   STDEXEC_ATTRIBUTE(always_inline)
-  void __coroutine_resume_nothrow(__std::coroutine_handle<> __h) noexcept
+  void __coroutine_resume_nothrow(void* __address) noexcept
   {
     STDEXEC_TRY
     {
-      __builtin_coro_resume(__h.address());
+      __builtin_coro_resume(__address);
     }
     STDEXEC_CATCH_ALL
     {
@@ -47,11 +47,17 @@ namespace STDEXEC
   }
 
   STDEXEC_ATTRIBUTE(always_inline)
-  void __coroutine_destroy_nothrow(__std::coroutine_handle<> __h) noexcept
+  void __coroutine_resume_nothrow(__std::coroutine_handle<> __h) noexcept
+  {
+    STDEXEC::__coroutine_resume_nothrow(__h.address());
+  }
+
+  STDEXEC_ATTRIBUTE(always_inline)
+  void __coroutine_destroy_nothrow(void* __address) noexcept
   {
     STDEXEC_TRY
     {
-      __builtin_coro_destroy(__h.address());
+      __builtin_coro_destroy(__address);
     }
     STDEXEC_CATCH_ALL
     {
@@ -59,6 +65,12 @@ namespace STDEXEC
     }
   }
 
+  STDEXEC_ATTRIBUTE(always_inline)
+  void __coroutine_destroy_nothrow(__std::coroutine_handle<> __h) noexcept
+  {
+    STDEXEC::__coroutine_destroy_nothrow(__h.address());
+  }
+
   // A coroutine handle that also supports unhandled_stopped() for propagating stop
   // signals through co_awaits of senders.
   template <class _Promise = void>
@@ -191,7 +203,8 @@ namespace STDEXEC
       // coroutine after resuming the continuation.
       auto __promise = static_cast<__destroy_and_continue_frame*>(__address)->__promise_;
       STDEXEC::__coroutine_resume_nothrow(__promise.__continue_);
-      STDEXEC::__coroutine_destroy_nothrow(__promise.__destroy_);
+      STDEXEC_ATTRIBUTE(musttail)
+      return STDEXEC::__coroutine_destroy_nothrow(__promise.__destroy_.address());
     }
 
     struct __promise
@@ -203,8 +216,9 @@ namespace STDEXEC
     static thread_local __destroy_and_continue_frame value;
   };
 
-  constinit inline thread_local __destroy_and_continue_frame __destroy_and_continue_frame::value{
-    {&__destroy_and_continue_frame::__resume}, {}};
+  inline thread_local __destroy_and_continue_frame __destroy_and_continue_frame::value{
+    {&__destroy_and_continue_frame::__resume},
+    {}};
 
   struct __symmetric_transfer_frame : __detail::__synthetic_coro_frame
   {
@@ -213,7 +227,8 @@ namespace STDEXEC
       // Make a local copy of the promise to ensure we can safely destroy the suspended
       // coroutine after resuming the continuation.
       auto __promise = static_cast<__symmetric_transfer_frame*>(__address)->__promise_;
-      STDEXEC::__coroutine_resume_nothrow(__promise.__continue_);
+      STDEXEC_ATTRIBUTE(musttail)
+      return STDEXEC::__coroutine_resume_nothrow(__promise.__continue_.address());
     }
 
     struct __promise
@@ -224,8 +239,9 @@ namespace STDEXEC
     static thread_local __symmetric_transfer_frame value;
   };
 
-  constinit inline thread_local __symmetric_transfer_frame __symmetric_transfer_frame::value{
-    {&__symmetric_transfer_frame::__resume}, {}};
+  inline thread_local __symmetric_transfer_frame __symmetric_transfer_frame::value{
+    {&__symmetric_transfer_frame::__resume},
+    {}};
 
   inline auto __coroutine_destroy_and_continue(__std::coroutine_handle<> __destroy,            //
                                                __std::coroutine_handle<> __continue) noexcept  //

From 43e326d3adca5d4accea6c2d8c96741766eddbb4 Mon Sep 17 00:00:00 2001
From: Eric Niebler <eniebler@nvidia.com>
Date: Thu, 30 Apr 2026 18:08:43 -0700
Subject: [PATCH 6/7] try again

---
 CMakeLists.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index f603401a6..c574a5341 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -252,7 +252,7 @@ target_compile_options(stdexec INTERFACE
 
 # Do you want a preprocessor that works? Picky, picky.
 target_compile_options(stdexec INTERFACE
-                       $<$<COMPILE_LANG_AND_ID:CXX,MSVC>:/Zc:__cplusplus /Zc:preprocessor /Zc:externConstexpr>
+                       $<$<COMPILE_LANG_AND_ID:CXX,MSVC>:/Zc:__cplusplus /Zc:preprocessor /Zc:externConstexpr /bigobj>
                        )
 
 set(STDEXEC_NAMESPACE "stdexec" CACHE STRING "The name of the top-level namespace for stdexec")

From e4caf2fdbaffbb0c59296d0ce9d4983b5f1d8b35 Mon Sep 17 00:00:00 2001
From: Eric Niebler <eniebler@nvidia.com>
Date: Thu, 30 Apr 2026 19:21:34 -0700
Subject: [PATCH 7/7] turn other cpu ci targets back on

---
 .github/workflows/ci.cpu.yml | 342 +++++++++++++++++------------------
 1 file changed, 171 insertions(+), 171 deletions(-)

diff --git a/.github/workflows/ci.cpu.yml b/.github/workflows/ci.cpu.yml
index 6f9d34a4a..7b353d97f 100644
--- a/.github/workflows/ci.cpu.yml
+++ b/.github/workflows/ci.cpu.yml
@@ -12,133 +12,133 @@ concurrency:
 
 jobs:
 
-  # build-cpu:
-  #   runs-on: ubuntu-latest
-  #   name: ${{ matrix.name }}
-  #   strategy:
-  #     fail-fast: false
-  #     matrix:
-  #       include:
-  #         - { name: "CPU (clang 16, Debug)",         build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++" }
-  #         - { name: "CPU (clang 16, Debug, c++23)",  build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "23", cxxflags: "-stdlib=libc++" }
-  #         - { name: "CPU (clang 16, Debug, TSAN)",   build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-fsanitize=thread" }
-  #         - { name: "CPU (clang 16, Release)",       build: "Release", tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++" }
-  #         - { name: "CPU (clang 16, Release, ASAN)", build: "Release", tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++ -fsanitize=address -fsanitize-ignorelist=/home/coder/stdexec/sanitizer-ignorelist.txt" }
-  #         - { name: "CPU (clang 22, Debug)",         build: "Debug",   tag: llvm22-cuda13.2, cxxstd: "23", cxxflags: "-stdlib=libc++" }
-  #         - { name: "CPU (clang 22, Release)",       build: "Release", tag: llvm22-cuda13.2, cxxstd: "23", cxxflags: "-stdlib=libc++" }
-  #         - { name: "CPU (gcc 12, Debug)",           build: "Debug",   tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "", }
-  #         - { name: "CPU (gcc 12, Release)",         build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "", }
-  #         # With the following config, 2 tests mysteriously time out, but only in CI and not locally.
-  #         # - { name: "CPU (gcc 12, Release, ASAN)",   build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
-  #         - { name: "CPU (gcc 12, Release, TSAN)",   build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=thread" }
-  #         - { name: "CPU (gcc 13, Debug)",           build: "Debug",   tag: gcc13-cuda12.9,  cxxstd: "20", cxxflags: "", }
-  #         - { name: "CPU (gcc 14, Debug)",           build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "", }
-  #         - { name: "CPU (gcc 14, Debug, ASAN)",     build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
-  #         - { name: "CPU (gcc 14, Debug, TSAN)",     build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=thread" }
-  #         - { name: "CPU (gcc 14, Release, ASAN)",   build: "Release", tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
-  #         - { name: "CPU (gcc 14, Release, LEAK)",   build: "Release", tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=leak", }
-  #         - { name: "CPU (gcc 14, Release, c++23)",  build: "Release", tag: gcc14-cuda12.9,  cxxstd: "23", cxxflags: "", }
-  #   container:
-  #     options: -u root
-  #     image: rapidsai/devcontainers:26.06-cpp-${{ matrix.tag }}
-  #   permissions:
-  #     id-token: write # This is required for configure-aws-credentials
-  #     contents: read  # This is required for actions/checkout
-  #   defaults:
-  #     run:
-  #       shell: su coder {0}
-  #       working-directory: /home/coder
-  #   steps:
-  #     - name: Checkout stdexec
-  #       uses: actions/checkout@v4
-  #       with:
-  #         path: stdexec
-  #         persist-credentials: false
-  #     - name: Setup environment
-  #       run: |
-  #         echo "ARTIFACT_PREFIX=${{runner.os}}-${{matrix.tag}}-amd64" >> "${GITHUB_ENV}"
-  #         echo "ARTIFACT_SUFFIX=${{github.run_id}}-${{github.run_attempt}}-$RANDOM" >> "${GITHUB_ENV}"
-  #     - if: github.repository_owner == 'NVIDIA'
-  #       name: Get AWS credentials for sccache bucket
-  #       uses: aws-actions/configure-aws-credentials@v4
-  #       with:
-  #         aws-region: us-east-2
-  #         role-duration-seconds: 28800 # 8 hours
-  #         role-to-assume: arn:aws:iam::279114543810:role/gha-oidc-NVIDIA
-  #     - name: Build and test CPU schedulers
-  #       env:
-  #         ASAN_OPTIONS: alloc_dealloc_mismatch=0
-  #         NVCC_APPEND_FLAGS: "-t=100"
-  #         SCCACHE_BUCKET: "rapids-sccache-devs"
-  #         SCCACHE_DIST_REQUEST_TIMEOUT: "7140"
-  #         SCCACHE_DIST_URL: "https://amd64.linux.sccache.rapids.nvidia.com"
-  #         SCCACHE_IDLE_TIMEOUT: "0"
-  #         SCCACHE_REGION: "us-east-2"
-  #         SCCACHE_S3_KEY_PREFIX: "nvidia-stdexec-dev"
-  #         SCCACHE_S3_PREPROCESSOR_CACHE_KEY_PREFIX: "nvidia-stdexec-dev/preprocessor"
-  #         SCCACHE_S3_USE_PREPROCESSOR_CACHE_MODE: "true"
-  #         SCCACHE_SERVER_LOG: "sccache=debug"
-  #         SCCACHE_SERVER_PORT: "4225"
-  #       run: |
-  #         set -e;
-  #         source /etc/profile
-  #         set -x;
-
-  #         devcontainer-utils-install-sccache --version rapids;
-
-  #         devcontainer-utils-init-sccache-dist                                          \
-  #             --enable-sccache-dist - <<< "                                             \
-  #             --auth-type 'token'                                                       \
-  #             --auth-token '$(                                                          \
-  #               curl -fsSL -H "Authorization: Bearer $(                                 \
-  #                 curl -fsSL -H "Authorization: bearer $ACTIONS_ID_TOKEN_REQUEST_TOKEN" \
-  #                   "${ACTIONS_ID_TOKEN_REQUEST_URL}&audience=token.rapids.nvidia.com"  \
-  #               | jq -r '.value'                                                        \
-  #               )" https://token.rapids.nvidia.com/gh/token/exchange                    \
-  #             | jq -r '.token')'                                                        \
-  #         ";
-
-  #         # Copy source folder into ~/stdexec
-  #         cp -r "${GITHUB_WORKSPACE}"/stdexec ~/;
-  #         chown -R coder:coder ~/stdexec;
-  #         cd ~/stdexec;
-
-  #         # Configure
-  #         cmake -S . -B build -GNinja \
-  #           -DCMAKE_BUILD_TYPE=${{ matrix.build }} \
-  #           -DCMAKE_CXX_FLAGS="${{ matrix.cxxflags }}" \
-  #           -DSTDEXEC_ENABLE_TBB:BOOL=${{ !contains(matrix.cxxflags, '-fsanitize') }} \
-  #           -DSTDEXEC_ENABLE_ASIO:BOOL=TRUE \
-  #           -DSTDEXEC_ASIO_IMPLEMENTATION:STRING=boost \
-  #           -DCMAKE_CXX_STANDARD:STRING=${{ matrix.cxxstd }} \
-  #           -DCMAKE_CXX_EXTENSIONS:BOOL=OFF \
-  #           -DSTDEXEC_BUILD_TESTS:BOOL=ON \
-  #           ;
-
-  #         # Compile
-  #         cmake --build build -v -j 512;
-
-  #         # Print sccache stats
-  #         sccache -s;
-
-  #         # Tests
-  #         SCCACHE_NO_CACHE=1 SCCACHE_NO_DIST_COMPILE=1 \
-  #         ctest --test-dir build --verbose --output-on-failure --timeout 60;
-  #     - if: ${{ !cancelled() }}
-  #       name: Upload sccache logs
-  #       uses: actions/upload-artifact@v4
-  #       with:
-  #         name: sccache-client-logs-${{env.ARTIFACT_PREFIX}}-${{env.ARTIFACT_SUFFIX}}
-  #         path: /tmp/sccache*.log
-  #         compression-level: 9
-
-  # ci-cpu:
-  #   runs-on: ubuntu-latest
-  #   name: CI (CPU)
-  #   needs:
-  #     - build-cpu
-  #   steps:
-  #     - run: echo "CI (CPU) success"
+  build-cpu:
+    runs-on: ubuntu-latest
+    name: ${{ matrix.name }}
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - { name: "CPU (clang 16, Debug)",         build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++" }
+          - { name: "CPU (clang 16, Debug, c++23)",  build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "23", cxxflags: "-stdlib=libc++" }
+          - { name: "CPU (clang 16, Debug, TSAN)",   build: "Debug",   tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-fsanitize=thread" }
+          - { name: "CPU (clang 16, Release)",       build: "Release", tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++" }
+          - { name: "CPU (clang 16, Release, ASAN)", build: "Release", tag: llvm16-cuda12.9, cxxstd: "20", cxxflags: "-stdlib=libc++ -fsanitize=address -fsanitize-ignorelist=/home/coder/stdexec/sanitizer-ignorelist.txt" }
+          - { name: "CPU (clang 22, Debug)",         build: "Debug",   tag: llvm22-cuda13.2, cxxstd: "23", cxxflags: "-stdlib=libc++" }
+          - { name: "CPU (clang 22, Release)",       build: "Release", tag: llvm22-cuda13.2, cxxstd: "23", cxxflags: "-stdlib=libc++" }
+          - { name: "CPU (gcc 12, Debug)",           build: "Debug",   tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "", }
+          - { name: "CPU (gcc 12, Release)",         build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "", }
+          # With the following config, 2 tests mysteriously time out, but only in CI and not locally.
+          # - { name: "CPU (gcc 12, Release, ASAN)",   build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
+          - { name: "CPU (gcc 12, Release, TSAN)",   build: "Release", tag: gcc12-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=thread" }
+          - { name: "CPU (gcc 13, Debug)",           build: "Debug",   tag: gcc13-cuda12.9,  cxxstd: "20", cxxflags: "", }
+          - { name: "CPU (gcc 14, Debug)",           build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "", }
+          - { name: "CPU (gcc 14, Debug, ASAN)",     build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
+          - { name: "CPU (gcc 14, Debug, TSAN)",     build: "Debug",   tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=thread" }
+          - { name: "CPU (gcc 14, Release, ASAN)",   build: "Release", tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=address" }
+          - { name: "CPU (gcc 14, Release, LEAK)",   build: "Release", tag: gcc14-cuda12.9,  cxxstd: "20", cxxflags: "-fsanitize=leak", }
+          - { name: "CPU (gcc 14, Release, c++23)",  build: "Release", tag: gcc14-cuda12.9,  cxxstd: "23", cxxflags: "", }
+    container:
+      options: -u root
+      image: rapidsai/devcontainers:26.06-cpp-${{ matrix.tag }}
+    permissions:
+      id-token: write # This is required for configure-aws-credentials
+      contents: read  # This is required for actions/checkout
+    defaults:
+      run:
+        shell: su coder {0}
+        working-directory: /home/coder
+    steps:
+      - name: Checkout stdexec
+        uses: actions/checkout@v4
+        with:
+          path: stdexec
+          persist-credentials: false
+      - name: Setup environment
+        run: |
+          echo "ARTIFACT_PREFIX=${{runner.os}}-${{matrix.tag}}-amd64" >> "${GITHUB_ENV}"
+          echo "ARTIFACT_SUFFIX=${{github.run_id}}-${{github.run_attempt}}-$RANDOM" >> "${GITHUB_ENV}"
+      - if: github.repository_owner == 'NVIDIA'
+        name: Get AWS credentials for sccache bucket
+        uses: aws-actions/configure-aws-credentials@v4
+        with:
+          aws-region: us-east-2
+          role-duration-seconds: 28800 # 8 hours
+          role-to-assume: arn:aws:iam::279114543810:role/gha-oidc-NVIDIA
+      - name: Build and test CPU schedulers
+        env:
+          ASAN_OPTIONS: alloc_dealloc_mismatch=0
+          NVCC_APPEND_FLAGS: "-t=100"
+          SCCACHE_BUCKET: "rapids-sccache-devs"
+          SCCACHE_DIST_REQUEST_TIMEOUT: "7140"
+          SCCACHE_DIST_URL: "https://amd64.linux.sccache.rapids.nvidia.com"
+          SCCACHE_IDLE_TIMEOUT: "0"
+          SCCACHE_REGION: "us-east-2"
+          SCCACHE_S3_KEY_PREFIX: "nvidia-stdexec-dev"
+          SCCACHE_S3_PREPROCESSOR_CACHE_KEY_PREFIX: "nvidia-stdexec-dev/preprocessor"
+          SCCACHE_S3_USE_PREPROCESSOR_CACHE_MODE: "true"
+          SCCACHE_SERVER_LOG: "sccache=debug"
+          SCCACHE_SERVER_PORT: "4225"
+        run: |
+          set -e;
+          source /etc/profile
+          set -x;
+
+          devcontainer-utils-install-sccache --version rapids;
+
+          devcontainer-utils-init-sccache-dist                                          \
+              --enable-sccache-dist - <<< "                                             \
+              --auth-type 'token'                                                       \
+              --auth-token '$(                                                          \
+                curl -fsSL -H "Authorization: Bearer $(                                 \
+                  curl -fsSL -H "Authorization: bearer $ACTIONS_ID_TOKEN_REQUEST_TOKEN" \
+                    "${ACTIONS_ID_TOKEN_REQUEST_URL}&audience=token.rapids.nvidia.com"  \
+                | jq -r '.value'                                                        \
+                )" https://token.rapids.nvidia.com/gh/token/exchange                    \
+              | jq -r '.token')'                                                        \
+          ";
+
+          # Copy source folder into ~/stdexec
+          cp -r "${GITHUB_WORKSPACE}"/stdexec ~/;
+          chown -R coder:coder ~/stdexec;
+          cd ~/stdexec;
+
+          # Configure
+          cmake -S . -B build -GNinja \
+            -DCMAKE_BUILD_TYPE=${{ matrix.build }} \
+            -DCMAKE_CXX_FLAGS="${{ matrix.cxxflags }}" \
+            -DSTDEXEC_ENABLE_TBB:BOOL=${{ !contains(matrix.cxxflags, '-fsanitize') }} \
+            -DSTDEXEC_ENABLE_ASIO:BOOL=TRUE \
+            -DSTDEXEC_ASIO_IMPLEMENTATION:STRING=boost \
+            -DCMAKE_CXX_STANDARD:STRING=${{ matrix.cxxstd }} \
+            -DCMAKE_CXX_EXTENSIONS:BOOL=OFF \
+            -DSTDEXEC_BUILD_TESTS:BOOL=ON \
+            ;
+
+          # Compile
+          cmake --build build -v -j 512;
+
+          # Print sccache stats
+          sccache -s;
+
+          # Tests
+          SCCACHE_NO_CACHE=1 SCCACHE_NO_DIST_COMPILE=1 \
+          ctest --test-dir build --verbose --output-on-failure --timeout 60;
+      - if: ${{ !cancelled() }}
+        name: Upload sccache logs
+        uses: actions/upload-artifact@v4
+        with:
+          name: sccache-client-logs-${{env.ARTIFACT_PREFIX}}-${{env.ARTIFACT_SUFFIX}}
+          path: /tmp/sccache*.log
+          compression-level: 9
+
+  ci-cpu:
+    runs-on: ubuntu-latest
+    name: CI (CPU)
+    needs:
+      - build-cpu
+    steps:
+      - run: echo "CI (CPU) success"
 
   build-cpu-windows:
     runs-on: windows-2022
@@ -174,47 +174,47 @@ jobs:
     steps:
       - run: echo "CI (CPU) (Windows) success"
 
-  # build-cpu-macos:
-  #   runs-on: macos-26-large
-  #   name: macos-${{ matrix.name }}
-  #   strategy:
-  #     fail-fast: false
-  #     matrix:
-  #       include:
-  #         - { compiler: "clang++",  build: "Debug",   name: "CPU (MacOS) (clang, Debug)" }
-  #         - { compiler: "clang++",  build: "Release", name: "CPU (MacOS) (clang, Release)" }
-  #   steps:
-  #     - name: Checkout stdexec (MacOS)
-  #       uses: actions/checkout@v4
-  #       with:
-  #         persist-credentials: false
-
-  #     - name: Install dependencies
-  #       run: |
-  #         brew update
-  #         brew install ninja
-  #       shell: bash
-
-  #     - name: Build and test CPU schedulers (MacOS)
-  #       shell: bash
-  #       run: |
-  #         mkdir build
-  #         cmake -S. -Bbuild -GNinja \
-  #           -DCMAKE_BUILD_TYPE=${{ matrix.build }} \
-  #           -DCMAKE_CXX_COMPILER=${{ matrix.compiler }} \
-  #           -DSTDEXEC_ENABLE_ASIO:BOOL=TRUE \
-  #           -DSTDEXEC_ASIO_IMPLEMENTATION:STRING=boost \
-  #           -DCMAKE_CXX_STANDARD:STRING=20 \
-  #           -DSTDEXEC_BUILD_TESTS:BOOL=ON
-
-  #         cmake --build build/ -v
-  #         cd build
-  #         ctest --output-on-failure
-
-  # ci-cpu-macos:
-  #   runs-on: macos-latest-large
-  #   name: CI (CPU) (MacOS)
-  #   needs:
-  #     - build-cpu-macos
-  #   steps:
-  #     - run: echo "CI (CPU) (MacOS) success"
+  build-cpu-macos:
+    runs-on: macos-26-large
+    name: macos-${{ matrix.name }}
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - { compiler: "clang++",  build: "Debug",   name: "CPU (MacOS) (clang, Debug)" }
+          - { compiler: "clang++",  build: "Release", name: "CPU (MacOS) (clang, Release)" }
+    steps:
+      - name: Checkout stdexec (MacOS)
+        uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+
+      - name: Install dependencies
+        run: |
+          brew update
+          brew install ninja
+        shell: bash
+
+      - name: Build and test CPU schedulers (MacOS)
+        shell: bash
+        run: |
+          mkdir build
+          cmake -S. -Bbuild -GNinja \
+            -DCMAKE_BUILD_TYPE=${{ matrix.build }} \
+            -DCMAKE_CXX_COMPILER=${{ matrix.compiler }} \
+            -DSTDEXEC_ENABLE_ASIO:BOOL=TRUE \
+            -DSTDEXEC_ASIO_IMPLEMENTATION:STRING=boost \
+            -DCMAKE_CXX_STANDARD:STRING=20 \
+            -DSTDEXEC_BUILD_TESTS:BOOL=ON
+
+          cmake --build build/ -v
+          cd build
+          ctest --output-on-failure
+
+  ci-cpu-macos:
+    runs-on: macos-latest-large
+    name: CI (CPU) (MacOS)
+    needs:
+      - build-cpu-macos
+    steps:
+      - run: echo "CI (CPU) (MacOS) success"