Test failures in test_xdr.py test_offset_lock_created method

[PardhavMaradani]

A lot of tests are currently failing on develop with the following two failures:

FAILED testsuite/MDAnalysisTests/coordinates/test_xdr.py::TestXTCReader_offsets::test_offset_lock_created - AssertionError: assert False
FAILED testsuite/MDAnalysisTests/coordinates/test_xdr.py::TestTRRReader_offsets::test_offset_lock_created - AssertionError: assert False

The reason for these failures is a change in the filelock version. I was able to reproduce this on a local setup. filelock up until 3.20.3 didn't cause any problem. Anything above, causes these errors - see filelock release history. A quick search of the changes after 3.20.3 (see diff), seem to indicate that this could be due to changes in their PR 408 that deletes the lock file on release, but we aren't handling it and still checking for it's presence. Thanks

ps. filelock 3.20.4 was released on Feb 12, 2026 and this checks out with when we started having these test failures.

[jeremyleung521]

Thanks for tracking down the cause. #4800 seems relevant.

To be honest, the test (test_offset_lock_created) does not make sense to me since it was testing if we ever locked a file before and not actual current behavior (and was apparently never tested in windows due to filelock's behavioral differences across OSes). Now that UNIX behavior is same as Windows (orphaned lock file automatically deleted after unlock), it seems ok to remove the test. Or if we want a replacement test, it seems like the best way around it is to open a lock --> check --> close it --> check it's gone.

I'll post a PR shortly. This seems kind of important to fix since it's breaking every CI run.

[talagayev]

Ah I will quickly jump in into the discussion, since I was the one doing the PR back then when the transfer from fasteners to filelock was happening. Yes there was quite some problem with windows and I did run it back then so the code was modified a couple of times to work for windows and azure, but yes the new release of filelock broke probably the go arounds that we did back then. @jeremyleung521 if you would need an additional pair of eyes for the PR I would be happy to take a look and run it also locally :)

[jeremyleung521]

Ah I will quickly jump in into the discussion, since I was the one doing the PR back then when the transfer from fasteners to filelock was happening. Yes there was quite some problem with windows and I did run it back then so the code was modified a couple of times to work for windows and azure, but yes the new release of filelock broke probably the go arounds that we did back then. @jeremyleung521 if you would need an additional pair of eyes for the PR I would be happy to take a look and run it also locally :)

Feel free to review! The more the merrier. :) (Esp. since the main maintainers are quite busy right now)