Repository navigation
Build every Python version on tags only, wait for every build before testing - #271
Conversation
There was a problem hiding this comment.
🟢 Approval recommended
There are no active review issues for the Intel CI matrix and build-dependent test-gating changes.
No open issues 🎉
What changed in this PR
This PR reduces Intel CI runner usage by reserving full Python-version coverage for tags and skipping dependent test jobs when builds fail.
Changes:
- Use the reduced Python matrix for pull requests and pushes; retain all declared versions for tags.
- Standardize matrix events as
pr,tag, andpush. - Require each test job’s corresponding build matrix to succeed, and update the CI documentation.
Root AGENTS.md was unavailable, so additional repository verification was not performed.
| File | Description |
|---|---|
ci/README.md |
Documents tag-only full coverage and the updated event argument. |
ci/matrix.py |
Applies reduced matrices to pr and push and validates event choices. |
.github/workflows/intel-ci.yml |
Passes normalized events and restores success-based test gating. |
Review effort: Balanced
|
The test jobs still start when a build they should wait for fails, so the "wait for every build" goal is not met yet.
Expected, per the title and the description ("no test runners start for a run that can't go green"): no test job starts after any failed build. Proposed fix: Everything else I ran (ci/matrix.py for pr, push and tag, conda and wheel; pre-commit; actionlint; zizmor) is clean. Whether master pushes should keep the full matrix between tags is your call; the PR currently drops 3.12 to 3.14t there. |
…testing Pushes to master now build the versions a pull request builds: the oldest and the newest regular one and the newest free-threaded one. The test jobs start once every build has passed, so a run with a failed build starts no test runners.
022e3ce to
a55dc11
Compare
xaleryb
left a comment
There was a problem hiding this comment.
Checked at a55dc11: the fix is the two needs lines we proposed. test and test-wheel now both need [matrix, build, build-wheel]. The needs graph gives: a failed conda build or a failed wheel build skips both test jobs, and an all-green run runs both. The matrices for pr, push and tag are unchanged from the reviewed head. The workflow parses, and the "Scan GitHub Actions workflows" check passes.
Two changes to
intel-ci.ymlandci/matrix.py, so the workflow starts fewer runners:masternow builds what a pull request builds: the oldest and the newest regular version and the newest free-threaded one.ci/matrix.py --eventtakespr,tagorpush, asci/hooks.pydoes, and the workflow passes it the same value.testandtest-wheeljobs wait for every build again. A failed build already fails the run, so no test runners start for a run that can't go green, and they stay free for other runs in the queue.Validation:
python ci/matrix.py --event prand--event pushlist 3.11, 3.15 and 3.15t, and--event taglists all seven versions, for both the conda packages and the wheels.pre-commitpasses on the changed files. This pull request's own run exercises theprpath; thepushandtagpaths run after merge.