H5PY 3.4.0 causes H5MD writer tests fail with PermissionError on Windows (investigate temporary pin)

[IAlibay]

Expected behavior

All tests pass.

Actual behavior

For the Windows runners TestH5MDWriterWithRealTrajectory.test_write_with_drivers[core] fails due to a permission error in writing the H5MD file.

Code to reproduce the behavior

See: https://dev.azure.com/mdanalysis/mdanalysis/_build/results?buildId=2576&view=logs&j=af49e0b6-b01c-50c5-fac7-b346db9f76f5&t=6e9d53c4-1948-52bf-f8ab-45d3eec15dc0&l=821

[orbeckst]

Do we have any inkling as to why?

@edisj ?

[edisj]

I’ll be home in ~1.5 hrs, can take a look then

[IAlibay]

I wonder if this is linked to the 3.4.0 release (it got released 4 days ago), no issues linked to this issue at this point though.

[IAlibay]

See #3407 - seems like an issue with the 3.4.0 release of h5py. If someone has the spare capacity / time we should report it upstream.

With regards to moving forward, it's probably best to just pin to <3.4.0?

[edisj]

That seems like the best short term fix to me. There was a bug fix listed here https://docs.h5py.org/en/latest/whatsnew/3.4.html relating to the core and stdio drivers and using mode='a', however H5MDWriter only uses mode='w'. I looked through a few of the PRs linked and I haven't found anything to do specifically with Windows or mode='w'

[orbeckst]

Did we ever reproduce the failure locally?

Has anyone the time to bisect h5py releases to pinpoint the ones that fail? We could exclude known bad releases for #3711 (or separately).

[IAlibay]

It doesn't look like this is a thing anymore? Closing.