Harvesting the Emscripten platform-check cache

A fresh VTK configure runs hundreds of serial try_compile/try_run checks. Each one starts emcc’s Python driver for both compilation and linking, which dominates configure time for the wasm builds. vtkEmscriptenCache-<version>-wasm32.cmake and vtkEmscriptenCache-<version>-wasm64.cmake preload the results of those checks as CACHE INTERNAL entries, so CMake’s check modules see them already answered and skip the probe. They’re consumed by vtkEmscriptenPreloadCache.cmake (included from both vtkEmscriptenToolchain.cmake and the .gitlab/ci/configure_wasm*.cmake CI files) whenever VTK_EMSCRIPTEN_PRELOAD_CACHE is on.

These files are specific to one (emsdk version, architecture) pair. The wasm32 and wasm64 need separate files because their ABI sizes (long, size_t, ptrdiff_t, uintptr_t, …) differ. Regenerate both whenever vtk_emsdk_version in vtkToolVersions.cmake is bumped; a version with no matching cache file is simply not preloaded.

1. Run a real configure and harvest it, inside the CI container

Do this once per architecture, using the CI container so the environment matches what CI actually runs with. The image name and tag are whatever the wasm CI jobs currently use: in .gitlab/os-linux.yml, the .wasm32_emscripten_linux/.wasm64_emscripten_linux jobs extend .fedora44_x86_64, whose image: line names it (e.g. kitware/vtk:ci-fedora44-<tag>). Check there rather than trusting the example below to be current.

git diff .gitlab/ci/configure_wasm_common.cmake   # should be empty before you start
# comment out the VTK_EMSCRIPTEN_PRELOAD_CACHE set() and the two include()s,
# run the harvest below, then:
git checkout .gitlab/ci/configure_wasm_common.cmake

SCRATCH=$HOME/vtk-wasm-cacheprobe
mkdir -p "$SCRATCH"
docker run --rm \
  -v /path/to/vtk:/work:Z \
  -v "$SCRATCH":/output:Z \
  -w /work \
  kitware/vtk:ci-fedora44-<tag> \
  bash -c '
    set -e
    export CMAKE_CONFIGURATION=wasm64_emscripten_linux   # or wasm32_emscripten_linux
    # ... bootstrap cmake/ninja/node/emsdk as needed (installs into /work/.gitlab)
    export PATH="/tmp/shim:$PATH"                        # sccache passthrough shim
    export CMAKE_BUILD_TYPE=Release
    BUILD=/output/build-${CMAKE_CONFIGURATION}-cacheprobe
    emcmake cmake -G Ninja -S /work -B "$BUILD" \
      -C /work/.gitlab/ci/configure_${CMAKE_CONFIGURATION}.cmake \
      -DVTK_BUILD_TESTING=OFF
    python3 /work/CMake/wasm/generate_emscripten_cache.py --build-dir "$BUILD"
  '

2. What generate_emscripten_cache.py does

generate_emscripten_cache.py, next to this document, scans the fresh CMakeCache.txt for the INTERNAL entries CMake’s check modules use to record a probe result, and writes vtkEmscriptenCache-<version>-<arch>.cmake into CMake/wasm/.

With --build-dir it works out <version> and <arch> from the configure itself: the architecture from VTK_WEBASSEMBLY_64_BIT in the cache, and the version from the emscripten-version.txt next to the emsdk toolchain file the cache points at (which is why the script is best run inside the container, where that path exists).

Run the harvest once per architecture. The auto-detected version is whatever emsdk actually installed. Please confirm it matches the vtk_emsdk_version you’re pinning in vtkToolVersions.cmake before committing the result.