Repository navigation
Conversation
The windows-2025-vs2026 runner image 20261116.315 replaces MSVC 14.51 with 14.52, so the pinned toolset_version 14.51 fails in setup once a job lands on it. Select the newest toolset matching "14.5" instead, and pass its full version to CMake's -T in both configure steps so MSBuild uses that exact toolset rather than the VS default. Bump the deps cache key, since cached FetchContent sub-builds record the -T value. This replaces build.yml's compiler-toolset-version and vs-path inputs with msvc-toolset-prefix and msvc-platform-toolset; x4 must update its caller accordingly.
The build job always configured with FETCHCONTENT_FULLY_DISCONNECTED=ON, assuming prepare-deps had saved build/_deps under the same key. The key includes the MSVC version, and since the toolset is now selected per job, the two jobs can pick different ones when they land on different runner images during a rollout (14.51 vs 14.52 in PR #120's CI). The build job then missed the cache and configure failed under CMP0170. Set FETCHCONTENT_FULLY_DISCONNECTED only on a cache hit, as prepare-deps already does, so a miss costs a dependency build instead of failing.
|
f5acf42 is actually a good addition; I recall I sometimes encountered that situation and the CI was refusing to build deps. However, please make sure to call |
Decide on FETCHCONTENT_FULLY_DISCONNECTED with the same inline expression as the build job, instead of a separate step that exported it as IRIS_CI_CMAKE_ARGS, so both jobs read the same way.
On a deps cache miss the build job fetched and built Catch2 but never
saved it, so later runs in the same PR rebuilt it whenever their job
landed on that toolset again. Build Catch2WithMain in its own step and
save build/_deps right after configure, as prepare-deps does, so the
cache holds a complete deps build regardless of whether the tests then
compile.
The step is named "Prepare Deps" rather than "Build Deps" because
cpp-warning-notifier links each result to the first successful step
matching STEP_REGEX ("Build .+"), which must remain "Build Tests".
| # Not named "Build ...": the warning notifier links to the first step | ||
| # matching its STEP_REGEX, which must be "Build Tests". |
There was a problem hiding this comment.
I don't get why this caveat is necessary.
I think cppwarningnotifier already checks CPPWARNINGNOTIFIER_LOG_MARKER for parsing, so the restriction on the step name seems quite unnecessary. I'd suggest removing this restriction on cppwarningnotifier's side, because a restriction on a step name is quite surprising for a GitHub Action's design.
If a parse-time check of CPPWARNINGNOTIFIER_LOG_MARKER is going to introduce some runtime overhead, then I think cppwarningnotifier should require the user to specify some key like build_step_name: (or maybe build_step_id: ?) below the uses: parameter. Then STEP_REGEX can be obsolete, as it is too fragile like I mentioned.
Meanwhile, I think we can simply remove this comment, as we have no actual plans to add a step named "Build ..." anyway.
saki7
left a comment
There was a problem hiding this comment.
If all remaining works are done, feel free to self-merge, and please proceed on the x4's companion PR.
No description provided.