Skip to content

ci: migrate 14 workflows to needs.setup.outputs.matrix (batch 3/8) - #4918

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

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

Conversation

@luhenry

@luhenry luhenry commented Nov 18, 2026 •

Copy link
Copy Markdown
Member

Converts the build_wheels job's strategy.matrix in 14 workflows from a
hardcoded python/libc array to include: ${{ fromJSON(needs.setup.outputs.matrix) }},
and renames the remaining matrix.libc references to matrix.platform (job
name:, cibuildwheel only:/CIBW_BUILD, artifact name:). Same shape as
#4914/#4915.

Packages touched: fastcdc, faster-fifo, gemmi, gimli.units, glcontext, gpiod,
guppy3, hdrhistogram, igwn-segments, imgui-bundle, immutables, islpy,
jsonobject, kaldialign.

Verified each conversion with uv run ci_scripts/pending_versions.py --python-matrix --python-exclude cp39 <pkg> against the old hardcoded
python/libc lists. Findings:

  • fastcdc, faster-fifo, glcontext, immutables, jsonobject, kaldialign:
    pre-existing stale-matrix bug, same class as mocpy: drop 3.9/3.10 and musllinux from the 0.20.0 matrix #4914/phono3py: exclude cp310/cp311, phonopy/spglib have no wheel there #4915. The old hardcoded
    list omitted cp310/cp311 (genuinely declared in each package's YAML for the
    pending version) while hardcoding cp314/cp314t (not declared at all for that
    version). This PR's structural fix corrects it as a side effect.
  • gemmi, gpiod, islpy: same stale-list pattern — cp310/cp311 (and, for
    islpy, cp315) were never attempted even though the YAML declares them for
    the pending version(s).
  • gimli.units: cp311 and cp315 (manylinux) were never attempted even
    though declared; fixed as a side effect.
  • hdrhistogram: the old hardcoded list (cp310-cp314, both libc) already
    matched the generated matrix exactly for the pending 0.10.8 version — no
    behavior change.
  • igwn-segments: old list deliberately hardcoded cp314t too (comment cites
    gotcha 127: upstream ships no cp314t wheel, but the extension just
    re-enables the GIL on import rather than failing). docs/packages/igwn-segments.yaml
    declares no 3.14t entries at all, so the generated matrix drops that leg;
    restoring it needs a YAML change, which is out of scope here (not touching
    docs/packages/*.yaml in this branch). cp310/cp311 were also never attempted
    despite being declared — fixed as a side effect.
  • imgui-bundle: the old list intentionally dropped cp311 ("to match this
    repo's numpy-driven cp312 floor"), and docs/packages/imgui-bundle.yaml does
    declare 3.11 entries, so converting to the generated matrix would have
    silently started building cp311 again. Added python-exclude: 'cp39 cp311'
    to the setup: job's with: block to preserve the intentional exclusion.
    Also note cp315 now builds for version 1.93.0, which the old list never
    covered.
  • guppy3: no version is currently pending (fully released), so there is no
    matrix to compare against today; the structural fix is otherwise a no-op
    until a new version lands. The free-threaded exclusion comment was reworded
    since docs/packages/guppy3.yaml simply declares no 3.1[34]t entries (no
    python-exclude change needed).

No files outside .github/workflows/* were touched.

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