Fix tests and workflows

[phaarnes]

Fix CI workflow failures in release.yaml

Several issues are causing CI failures in the release workflow:

1. sdist job fails with "twine: error: invalid choice: 'check'"

Cause: conda-forge may install twine 1.11.0, which lacks the check subcommand (added in 1.12.0).

Fix: Change twine to twine>=1.12 in .github/workflows/release.yaml:

create-args: >-
  python-build
  twine>=1.12
  cython
  proj=${{ env.PROJ_VERSION }}

2. sdist job uses undefined matrix.proj-version

Cause: The make_sdist job references ${{ matrix.proj-version }} but has no matrix defined, causing proj= (empty version) to be passed to micromamba.

Fix: Change to ${{ env.PROJ_VERSION }} to use the workflow-level environment variable.

3. Windows 32-bit build fails with "Found GNU link.exe instead of MSVC link.exe"

Cause: Git's link.exe at C:\Program Files\Git\usr\bin\ is found before MSVC's link.exe during meson configuration.

Fix: Add CIBW_BEFORE_BUILD_WINDOWS to rename Git's link.exe before building:

CIBW_BEFORE_BUILD_WINDOWS: >
  python -m pip install delvewheel &&
  if exist "C:\Program Files\Git\usr\bin\link.exe" (rename "C:\Program Files\Git\usr\bin\link.exe" "link.exe.bak")

4. macOS builds take 1.5+ hours due to single-threaded PROJ compilation

Cause: ci/proj-compile-wheels.sh uses nproc which doesn't exist on macOS, causing cmake to default to single-threaded build.

Fix: Add cross-platform CPU detection in ci/proj-compile-wheels.sh:

# Detect number of CPU cores (cross-platform)
if command -v nproc &> /dev/null; then
    NPROC=$(nproc)
elif command -v sysctl &> /dev/null; then
    NPROC=$(sysctl -n hw.ncpu)
else
    NPROC=2
fi
cmake --build . -j${NPROC}

5. Free-threaded Python (3.13t/3.14t) builds segfault during tests

Cause: Cython extension crashes with segfault in __repr__ methods on free-threaded Python. This affects both Linux aarch64 and macOS x86_64.

Fix: Skip free-threaded builds with CIBW_SKIP until Cython's free-threading support matures:

CIBW_SKIP: "*t-*"

Note: I'm a bit on thin ice with skipping the free-threaded builds entirely. However, this was the only solution I found to avoid the segfaults. Open to suggestions for a better approach!

---

✅ Closes #1570

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.85%. Comparing base (b78a50d) to head (e81e5e5).
⚠️ Report is 44 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1571      +/-   ##
==========================================
- Coverage   96.55%   95.85%   -0.71%     
==========================================
  Files          20       20              
  Lines        1827     1880      +53     
==========================================
+ Hits         1764     1802      +38     
- Misses         63       78      +15     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[snowman2]

Thanks @phaarnes 👍. Mind creating an issue to track re-enabling multithreaded python?

[phaarnes]

Thanks @phaarnes 👍. Mind creating an issue to track re-enabling multithreaded python?

Will do👍 Is there anything else that has to be fixed now before making a new release @snowman2 ?

[snowman2]

Is there anything else that has to be fixed now before making a new release?

The main thing is having time to make a release. But, wouldn't hurt to update PROJ in the wheels before the next release.

[phaarnes]

Is there anything else that has to be fixed now before making a new release?

The main thing is having time to make a release. But, wouldn't hurt to update PROJ in the wheels before the next release.

Yes, I understand. I can look into updating to PROJ 9.7.1 tomorrow and opening a PR👍