Pyfive integration as backend into h5netcdf #113
Description
I'd like to give an update on the whereabouts of the pyfive backend integration into h5netcdf.
The integration was started by NCAS-CMS team with h5netcdf/h5netcdf#273.
In the first weeks several issues with h5netcdf have been found and fixed and a major refactor of NETCDF4_CLASSIC into h5netcdf has been undertaken:
- Add partial support for NETCDF4_CLASSIC format h5netcdf/h5netcdf#283
- create enums with low-level API to keep order by value, add h5dump-tests h5netcdf/h5netcdf#285
- FIX: interoperability netcdf4 vs h5netcdf h5netcdf/h5netcdf#286
- fixup testsuite and issues to work with pyfive h5netcdf/h5netcdf#279
The changes on the pyfive side, including user type support (enum/compound/vlen) as well as multiple other enhancements and fixes:
- when getting object by address do not fully instantiate when iterating #83
- Enumeration Support #93
- fix Enum and Empty attributes #102
- Fix user datatypes (enum, compound) #105
- fix handling of uninitialized vlen strings #110
- add dataobjects.dtype to DatasetMeta #112
When #110 and #112 are resolved, the h5netcdf testsuite is running smoothly with pyfive as backend. There are still small glitches, which can be ironed out later.
After release of the above features both in pyfive and h5netcdf a pure Python pipeline for reading HDF5 into xarray would be established.
Many thanks to everyone involved! This will certainly draw additional attention to pyfive.