Update h5py to 3.3

Description

There are various minor fixes and improvements in h5py since 2.10. Probably none of them make a massive difference to us straight away, though we could look at reading/writing data while computation is going on now that it releases the GIL. But I've spent a fair bit of time working on h5py, and I'd like to see us using a recent version.

How Has This Been Tested?

Run AGIPD correction through xfel-calibrate (output in /gpfs/exfel/data/scratch/kluyvert/agipd-calib-900201-203-h5py-3.3).

I haven't tested all notebooks. But the biggest API change from 2.x is that h5py.File('abc.h5') opens files read-only by default. I've searched through the project for uses of File() without an explicit mode: there are only 2 (in src/cal_tools/dssclib.py and notebooks/LPD/playground/metroLib.py) and in both cases the file is only used for reading.

Reviewers

@calibration

Thanks for putting in the work. Do you think we should do further tests or are we reasonably safe to just use it?

added 1 commit

• cad08c7e - Replace h5py dataset.value with dataset[()]

Compare with previous version

I did some more searching, and found one bit of code using the (removed) Dataset.value property. I've replaced this with the newer equivalent. But the notebook which uses this (PlotFromCalDB_NBC.ipynb) seems to be broken anyway - I think it wasn't updated for the karabo_id/karabo_da change. So I can't easily test it.

I think this is reasonably safe. The changes in h5py 3.0 weren't massive. The biggest changes I can't easily search for are to how strings are handled, but we don't do much with HDF5 strings. I believe that all the strings in raw data files are marked as ASCII, not UTF-8, and reading ASCII strings still behaves the same.

Geez, can't wait to no longer manipulate files manually

changed milestone to %3.4.1

No reason to wait then, LGTM.

merged

mentioned in commit c29ba0ef