Python 3.15 wheels #579

Activity

1. jinzomotsu commented on Oct 12, 2026

   

   jinzomotsu

   on Oct 12, 2026

   Last edited by jinzomotsu

   More actions

   Tested the unmodified NumExpr 2.14.2 sdist locally on CPython 3.15.0, macOS 27.2 arm64 (Apple clang 21.0.0), NumPy 2.5.2 and pytest 9.1.1, standard GIL build, VML unavailable.

   Source build succeeded, and the installed upstream test suite reports 6042 passed, 8 warnings. The warnings are PytestUnknownMarkWarning for thread_unsafe when running from outside the source tree.

   Reproduction in a fresh environment (Python 3.15 already installed):

   uv venv --python python3.15 .venv-numexpr315
   uv pip install --python .venv-numexpr315/bin/python --no-binary numexpr numexpr==2.14.2 numpy==2.5.2 pytest==9.1.1
   env -u NUMEXPR_MAX_THREADS -u NUMEXPR_NUM_THREADS -u OMP_NUM_THREADS .venv-numexpr315/bin/python -m pytest --pyargs numexpr -q

   An initial run with all three thread environment variables forced to 1 had 6039 passes and 3 threading-configuration failures; those tests expect to control the environment themselves. Removing those overrides gave the full pass above, with no source changes.

   This is additional evidence for the macOS arm64 / regular CPython 3.15 build. It does not validate free-threaded Python, wheel portability to older macOS versions, VML, or performance. No compatibility patch was needed for this local build.

   Downstream check: this source build also allowed PyMBAR 4.0.3 to install in the same Python 3.15 environment (NumPy 2.5.2, SciPy 1.18.1, JAX/JAXlib 0.11.1). Its BAR/EXP/utils test subset reports 21 passed and 8 expected failures. A fixed three-state MBAR fixture gave identical free-energy differences, uncertainties, overlap matrix and effective sample counts on Python 3.12 and 3.15. This is a limited downstream smoke test, not the full PyMBAR suite or a performance claim.
2. FrancescAlted commented on Oct 13, 2026

   

   FrancescAlted

   on Oct 13, 2026

   Contributor

   More actions

   Please consider reading this: https://github.com/pydata/numexpr#warning-funding-is-needed