@EpicWink Just letting you know your PR might be reverted. Don't worry, we can resubmit we disable the offending test. We may also be able to avoid reverting the whole PR by just disabling that test, if we're convinced the problem is restricted to the combination of free-threading, Windows, and that specific test.

After reviewing the added code, I suggest immediately reverting the failing test on Windows free-thread builds to stop interfering with good PRs, and monitor for other CI/buildbot failures, while investigating. I was thinking about debug prints but then saw the note on the original PR that they themselves cause failures.

EDIT: side note line 241: n_items is not used in this function as _qsize is re-called each loop.

Closing in favor of#115269 which disables the offending tests on Windows.