Skip to content

ci: migrate 14 workflows to needs.setup.outputs.matrix (batch 6/8) - #4922

Open
luhenry wants to merge 14 commits into
mainfrom
fix/matrix-batch-6
Open

luhenry wants to merge 14 commits into
mainfrom
fix/matrix-batch-6

Conversation

@luhenry

@luhenry luhenry commented Nov 18, 2026 •

Copy link
Copy Markdown
Member

Batch 6/8 of the repo-wide migration off hardcoded strategy.matrix python:/libc: arrays onto _setup.yml's needs.setup.outputs.matrix (same bug class as #4914/#4915).

Packages touched (one commit each):

  • pycld2
  • pyfarmhash
  • pyglm
  • pylzss
  • pymetis
  • pymsym
  • pynng
  • pyopenxlsx
  • pyqt5-sip
  • pyqt6-sip
  • pyrodigal
  • pysjef
  • pyswisseph
  • python-sat

For every file: strategy.matrix is now include: ${{ fromJSON(needs.setup.outputs.matrix) }} (dropped the separate version:/python:/libc: arrays), and every remaining ${{ matrix.libc }} in build_wheels (job name:, CIBW_BUILD/only:, artifact name:) is renamed to ${{ matrix.platform }}. gpl_sources jobs (pymetis, pysjef) and publish jobs were untouched except where noted below.

Verified mismatches (step 4)

Ran uv run --quiet ci_scripts/pending_versions.py --python-matrix --python-exclude cp39 <pkg> for each package and compared against the old hardcoded python:/libc: lists for the currently-pending version(s). Every one of the 14 files had a real, pre-existing stale-matrix bug, all the same shape: the hardcoded lists were pinned at cp312+ when the ports were first authored, while docs/packages/<pkg>.yaml has since grown earlier pending entries (cp310/cp311, sometimes with no filename yet) that the hardcoded lists never built, while in several cases the old lists also kept trying to build cp314/cp314t combinations the YAML doesn't even declare. The structural fix resolves all of these as a side effect. Highlights:

  • pyqt5-sip / pyqt6-sip: the only pending target for the current version (12.20.0 / 13.13.0) is cp315:manylinux — a wheel the old hardcoded list (cp310-cp314) could never build. These two versions could never have completed/published under the old matrix; they were also wastefully rebuilding cp310-cp314 (already published) every run.
  • pynng: the hardcoded matrix was manylinux-only with a comment saying musllinux was deliberately dropped because cffi has no musllinux_riscv64 wheel on our registry. docs/packages/pynng.yaml does declare musllinux targets (and cp310/cp311), so per the playbook this conversion now attempts musllinux for the first time. Added the missing MUSLLINUX_RISCV64_IMAGE/CIBW_MUSLLINUX_RISCV64_IMAGE wiring and generalized the publish job's artifact-pattern (was hardcoded to *-manylinux_riscv64). Flagging: the musllinux cffi gap noted in the old comment may still be real — first CI run should be watched, it may need a follow-up if those legs fail.
  • pyopenxlsx: manylinux-only, no musllinux declared in YAML so nothing changes there, but the hardcoded list (cp312-cp314) was missing cp311, which the YAML does declare pending.
  • pymsym / pysjef: both had a comment explaining an intentional cp314t exclusion ("upstream ships no free-threaded wheel"). In both cases docs/packages/<pkg>.yaml simply never declares a 3.14t target, so the generated matrix already omits it with no python-exclude override needed (step 5) — left an updated comment explaining this. Both were also missing cp311, which their YAMLs declare pending.
  • pyqt5-sip / pyqt6-sip: same "no cp314t" situation — YAML declares no 3.14t target at all, so no python-exclude needed.
  • pycld2 / pyfarmhash / pyglm / pylzss / pymetis / pyrodigal / pyswisseph / python-sat: all had the same cp310/cp311 omission (sometimes across multiple pending versions, e.g. pylzss's 0.3.8/0.3.9).

No files were skipped. Not waiting on this PR's CI to go green before reporting back, per instructions.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant