[rkingsbury]

Summary

Supersedes #1151. It includes that PR's commit (9782d35 Add Python 3.14 to supported versions, unchanged) plus a fix for the Python 3.14 CI hang that kept #1151 from going green.

Root cause: Python 3.14 changed the default multiprocessing start method on Linux from fork to forkserver. Pool workers no longer inherit the parent's sys.modules, so they have to re-import every pickled builder by name. Two builder sources couldn't be imported:

• Builders defined in test modules (tests.cli.test_distributed). --import-mode=importlib leaves the repo root off sys.path when pytest is launched through its console script, as CI does.
• Builders that mrun loads from a .py/.ipynb file. These are re-imported through ScriptFinder, which used regex.match. regex is only installed with the notebook_runner extra.

When that happened, every pool worker died with ModuleNotFoundError/NameError. AsyncUnorderedMap swallowed the resulting BrokenProcessPool, so multi() waited forever. The stuck non-daemon threads then kept the interpreter from exiting, and CI ran until it hit the 6-hour job limit.

Major changes:

• fix 1: pyproject.toml: add pythonpath = ["."] to [tool.pytest.ini_options] so forkserver children can import builders defined in test modules. Bump minversion to 7.0, since pythonpath needs pytest 7 or later.
• fix 2: src/maggma/cli/source_loader.py: use the stdlib re.match instead of regex.match, so ScriptFinder works without the notebook_runner extra.

Todos

• Watch the two 3.14 jobs, test (ubuntu-latest, 3.14) and test-min-deps (ubuntu-latest, 3.14).
• Known pre-existing issue, not addressed here: tests/cli/test_distributed.py has an intermittent fork-at-exit deadlock on Python 3.11–3.13. It reproduced in about 1 of 8 runs on unmodified main and is caused by leaked ProcessPoolExecutors in multi(). If a 3.11–3.13 job hangs in that file, it is not a regression from this PR.

Checklist

• Google format doc strings added. (n/a: no new functions)
• Code linted with ruff. (pre-commit run passed: ruff, black, codespell, file hygiene hooks)
• Type annotations included. Check with mypy. (n/a: no new signatures)
• Tests added for new features/fixes. (existing tests/cli tests cover the hang on 3.14)
• I have run the tests locally and they passed. On Python 3.14.6 with requirements/ubuntu-latest_py3.14_extras.txt, pytest tests/cli/test_distributed.py tests/cli/test_init.py::test_python_source --timeout 30 (console script) gave 6 passed, 2 xfailed (pre-existing markers), and exited in about 12 s. MongoDB-backed tests were not run locally (no MongoDB available); they are left to CI.

🤖 Generated with Claude Code

https://claude.ai/code/session_013Pt7k8oDSGSVZDB41JuKtk

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 69.77%. Comparing base (9a98aef) to head (0d144b4).
⚠️ Report is 30 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1152      +/-   ##
==========================================
+ Coverage   69.63%   69.77%   +0.14%     
==========================================
  Files          47       47              
  Lines        4113     4106       -7     
==========================================
+ Hits         2864     2865       +1     
+ Misses       1249     1241       -8     

☔ View full report in Codecov by Harness.
📢 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.