Implement # sage.doctest: flaky marker

[user202729]

The intention is to avoid the annoying failing doctests.

Background on timed out tests

Basically one of the problems with occasional timeout is the following: Sometimes malloc need to hold a lock while doing something. If a signal comes while it is holding the lock, the next time malloc is called it will try to acquire the lock again and deadlock there.

A workaround is to unlock the malloc lock inside the signal handler — but how do you know which lock it is?

I can't reproduce this on my machine (in fact on my machine setting a gdb breakpoint in __lll_lock_wait_private doesn't even hit it during the computation), so I can't figure out a way to fix it.

Workaround

Files starting with # sage.doctest: flaky will be ran once more time if they timeout. Same for segmentation fault. E.g. plural.pyx → #39098

📝 Checklist

• The title is concise and informative.
• The description explains in detail what this PR is about.
• I have linked a relevant issue or discussion.
• I have created tests covering the changes.
• I have updated the documentation and checked the documentation preview.

⌛ Dependencies

[github-actions]

Documentation preview for this PR (built with commit e9b1b35; changes) is ready! 🎉
This preview will update shortly after each push to this PR.

[user202729]

From the log it looks like this was correctly retried once. But it still timeout…? (what's the probability this fails twice?)

[tobiasdiez]

Now that #40814 is merged, what's the plan to go forward here?

[user202729]

I don't know, depends on if tests fail. (like, I hope tests don't fail but…?)

on the other hand the other solution won't work on someone who is not running the CI.

[tobiasdiez]

There are enough doctests that randomly fail and it would be nice to have a solution for this. It's annoying locally as well, but the main headache is CI.

We could add essentially every file that has known flaky tests to the list in #40814. But I got the impression you would like to keep that list reserved for tests that fail with segfaults/timeouts which are hard to catch in python/doctest module. So your plan is to use this PR here for "normal" flaky tests?

[user202729]

I mean, normal flaky test can also be tested with say

sage: for i in range(5):
....:     whatever

ugly but works. (or migrate to pytest, where you probably gain some pytest marker thing (https://pytest-rerunfailures.readthedocs.io/stable/mark.html?), but lose preparsing and need to explicitly import, and make the test far away from the code…)

[tobiasdiez]

I've now added those flaky tests also to the exception list in CI; was the easiest solution for now #41235.