MNT: prepare h5netcdf backend for (coming) change in dimension handling

[kmuehlbauer]

• Closes ⚠️ Nightly upstream-dev CI failed ⚠️ #6197

[kmuehlbauer]

@mathause Although I'm pretty sure that this fixes the upstream-dev run, it would be good to verify. Is it possible to run the upstream-dev with this PR?

[kmuehlbauer]

I'd also like to add fixes to silence new FutureWarnings emitted by h5netcdf (hence the WIP).

[mathause]

Great! Thanks for the fix. Yes the string [test-upstream] has to be part of the commit message (if you have nothing new to commit you can do git commit --allow-empty -m "[test-upstream]")

[kmuehlbauer]

@dionhaefner

Could you please have a look at the chunksizes FutureWarnings (eg. here https://github.com/pydata/xarray/runs/4965336696?check_suite_focus=true#step:8:344).

Are the chunksizes via h5py and via h5netcdf heuristics correct?

[kmuehlbauer]

@mathause The fix is applied and works. We have some issues with chunking over at h5netcdf. This will be fixed after sorting this out.

[mathause]

Looks good. The remaining upstream failure does not seem to be related.

[dionhaefner]

@dionhaefner

Could you please have a look at the chunksizes FutureWarnings (eg. here https://github.com/pydata/xarray/runs/4965336696?check_suite_focus=true#step:8:344).

Are the chunksizes via h5py and via h5netcdf heuristics correct?

Let's take this in h5netcdf/h5netcdf#149.

[mathause]

Thanks @kmuehlbauer!

[mathause]

Sorry @kmuehlbauer I was too fast with merging. I have to revert this (#6208) and we can do it again once #6207 is fixed.

[mathause]

Sorry I was wrong - this was not the cause for the test failure. Let's keep this as is.