[petercorke]

Thanks for contributing to spatialgeometry!

Summary

Dry-run the wheel/sdist/Pyodide build on pull requests; only a release publishes.

The wheel matrix only ever ran on release, so a break (e.g. the new cp314 target from #47) would first surface at release time.

• Also trigger the workflow on pull_request, filtered to pyproject.toml, CMakeLists.txt and the workflow file itself, so it only runs when the build configuration changes.
• Gate upload_pypi on github.event_name == 'release', so a dry run can never publish. Releasing is still a single click on "Publish release" and is still the only way to publish; the header comment says so.
• Set fail-fast: false on the wheel matrix so one platform failing doesn't cancel the rest into misleading "cancelled" results.

Cost: 4 platform runners (× CPython 3.10-3.14 each) plus the Pyodide wheel and sdist, a few minutes, only on PRs that touch those files.

This PR touches the workflow file itself, so its own CI run is the first real dry run (including the first cp314 wheel build).

Related issue

Follows up #47 (Python 3.14 support) and #49 (pure Pyodide wheel).

Checklist

Nothing below is checked automatically -- this is a self-check for you before requesting review.

• PR title follows Conventional Commits (type: description)
• Tests pass locally — n/a, workflow-only change
• Added/updated tests for this change, if applicable — n/a; the workflow's own run on this PR is the test
• Type hints / docstrings — n/a
• C++ extension / pure-Python fallback — n/a, extension unchanged
• PR is as small/focused as practical -- if it tackles several unrelated things, consider splitting it so each can be reviewed and accepted independently

🤖 Generated with Claude Code