Make it possible for faulthandler to terminate the pytest process on timeout

[ogrisel]

Closes #13678.

This would help projects avoid wasting their precious CI resources when running extensive thread-safety tests (which will become more popular in the new free-threading era).

I also changed the logic to mark the test as XFAIL conditionally on the presence of the "CI" environment variable instead of always skipping (related to #7022). This makes it possible to check that the test pass as expected when running locally.

If you think this is too risky, I can remove the condition and always mark the offending params with the xfail mark instead of skipping.

EDIT: I will do the later right away because I have already observed such a random failure on the CI: https://github.com/pytest-dev/pytest/actions/runs/17269674978/job/49010727146

EDIT 2: that does not work because xfail_strict = true is configured in pyproject.toml. Let me skip instead...

[RonnyPfannschmidt]

At first glance this looks nice

There may be some nitpicks but they relate to preexisting patterns

[ogrisel]

I think I understand why the CI var was not properly detected: we probably need to add it to the passenv directive of tox. Let me try that quickly.

[ogrisel]

Ok so many other unrelated test started to fail because they are not written in a way that expect to be ever executed in the presence of the CI environment variable... Let me revert 000876e and 5066d98...

[ogrisel]

Shall I remove the if "CI" in os.environ condition that can never hold as long as pytest's CI is configured to not propagate this env var via its tox.ini config?

[RonnyPfannschmidt]

Let's actually propagate it snd check why tox doesn't automatically

[ogrisel]

CI is not in the list of variables passed by default: https://tox.wiki/en/latest/config.html#pass_env

[ogrisel]

When I tried to add CI to passenv (see 000876e above in this PR) this broke many other unrelated tests:

https://github.com/pytest-dev/pytest/actions/runs/17271882336/job/49018562088

[webknjaz]

In that case, those tests are incorrectly sensitive to said env var. I recommend fixing them all in a separate PR (that would also bypass this env var in tox) by ensuring adequate isolation in them (monkeypatch.delenv('CI')). Once that PR is merged, you'd be able to return to this one.

[webknjaz]

I wonder if the is_ci_env could be computed early, before the test runs instead of having this check within. It could be checked before the tests implement shielding through os.environ manipulation. Perhaps, a session fixture in conftest.py would be useful in general. Thoughts?

[ogrisel]

For this particular case, once the CI variable is properly propagated (after #13684 is merged and this PR synced with main), just increasing the sleep time as originally intended might be enough to make the test stable again and we would not have skip those tests anymore.