Fail test suite when GIL is enabled dynamically at runtime

[lysnikolaou]

@ngoldbaum We can get this into 0.6 as well.

Only problem is I don't really know how to test this. I checked manually with a custom package with a compiled extension and the test suite indeed failed.

[ngoldbaum]

This will work, but I have a slightly different approach I was thinking about today.

What do you think about instead of checking at the end of the session, we check after collection is finished and then check after every individual test. If it fails during collection, we say "GIL re-enabled during collection", if it's re-enabled in a test we say "GIL re-enabled during execution of test_foo". That way people get an indication exactly when the GIL is re-enabled.

[lysnikolaou]

Yeah, that makes sense to me as well. I was thinking the warning, that indicates what the module that enabled the GIL was, would be enough but that's lost in the case of a pytest.exit. Let me try to do that in this PR.

[lysnikolaou]

This is a bit more complicated than I initially thought. I'll work on it tomorrow and hopefully do a 0.6 release as well.

[ngoldbaum]

Only problem is I don't really know how to test this.

I think you'd need to create a tiny C extension with e.g. meson and then import it in a test. Seems like a lot of complexity - totally fine with me to punt on testing that for now.

[lysnikolaou]

@ngoldbaum This is realy for another review. What do you think of this approach? It fails early during collection or execution phase.

Collection:

❯ pytest                     
======================================================== test session starts ========================================================
platform darwin -- Python 3.13.5, pytest-8.4.1, pluggy-1.6.0
rootdir: /Users/lysnikolaou/repos/quansight/pytest-run-parallel-test
configfile: pyproject.toml
plugins: run-parallel-0.5.1.dev0
collected 1 item                                                                                                                    
Collected 0 items to run in parallel

======================================================= no tests ran in 0.82s =======================================================
! _pytest.outcomes.Exit: GIL was dynamically re-enabled during test collection to load module 'pkg.t'. When running under a free-threaded interpreter with the GIL initially disabled, the test suite must not cause the GIL to be re-enabled at runtime. Check for compiled extension modules that do not use the 'Py_mod_gil' slot or the 'PyUnstable_Module_SetGIL' API. !

❯ pytest --ignore-gil-enabled
======================================================== test session starts ========================================================
platform darwin -- Python 3.13.5, pytest-8.4.1, pluggy-1.6.0
rootdir: /Users/lysnikolaou/repos/quansight/pytest-run-parallel-test
configfile: pyproject.toml
plugins: run-parallel-0.5.1.dev0
collected 1 item                                                                                                                    
Collected 0 items to run in parallel

tests/test_t.py .                                                                                                             [100%]

========================================================= warnings summary ==========================================================
<frozen importlib._bootstrap>:488
  <frozen importlib._bootstrap>:488: RuntimeWarning: The global interpreter lock (GIL) has been enabled to load module 'pkg.t', which has not declared that it can run safely without the GIL. To override this behavior and keep the GIL disabled (at your own risk), run with PYTHON_GIL=0 or -Xgil=0.

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
=================================================== 1 passed, 1 warning in 0.14s ====================================================

Execution:

❯ pytest                                      
======================================================== test session starts ========================================================
platform darwin -- Python 3.13.5, pytest-8.4.1, pluggy-1.6.0
rootdir: /Users/lysnikolaou/repos/quansight/pytest-run-parallel-test
configfile: pyproject.toml
plugins: run-parallel-0.5.1.dev0
collected 1 item                                                                                                                    
Collected 0 items to run in parallel

tests/test_t.py .

========================================================= 1 passed in 1.18s =========================================================
! _pytest.outcomes.Exit: GIL was dynamically re-enabled during test execution of 'tests/test_t.py::test_number' to load module 'pkg.t'. When running under a free-threaded interpreter with the GIL initially disabled, the test suite must not cause the GIL to be re-enabled at runtime. Check for compiled extension modules that do not use the 'Py_mod_gil' slot or the 'PyUnstable_Module_SetGIL' API. !

❯ pytest --ignore-gil-enabled
======================================================== test session starts ========================================================
platform darwin -- Python 3.13.5, pytest-8.4.1, pluggy-1.6.0
rootdir: /Users/lysnikolaou/repos/quansight/pytest-run-parallel-test
configfile: pyproject.toml
plugins: run-parallel-0.5.1.dev0
collected 1 item                                                                                                                    
Collected 0 items to run in parallel

tests/test_t.py .                                                                                                             [100%]

========================================================= warnings summary ==========================================================
tests/test_t.py::test_number
  <frozen importlib._bootstrap>:488: RuntimeWarning: The global interpreter lock (GIL) has been enabled to load module 'pkg.t', which has not declared that it can run safely without the GIL. To override this behavior and keep the GIL disabled (at your own risk), run with PYTHON_GIL=0 or -Xgil=0.

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
=================================================== 1 passed, 1 warning in 0.13s ====================================================

[ngoldbaum]

LGTM, let's ship it. I might poke at adding tests with a real extension if we ever have issues with this but your mocking with a warning that you raise yourself in a test is probably more than sufficient.

[ngoldbaum]

gemini gave me this explanation for (?P<module>[^']*):

Capture any sequence of zero or more characters that are not a single quote, and name this captured group 'module'.

There might be a better way to do that but this also seems to work... I kind of hate regexes...