test failures related to hdf5 lib

[orbeckst]

Expected behavior

The tests all pass at all times.

Actual behavior

Scheduled CI failed, see https://github.com/Becksteinlab/zarrtraj/actions/runs/13220516432/job/36904931628 with errors pointing at the Python hdf5 package:

On macOS

E   ImportError: dlopen(/Users/runner/miniconda3/envs/zarrtraj-test/lib/python3.10/site-packages/h5py/defs.cpython-310-darwin.so, 0x0002): Symbol not found: _H5Pget_fapl_ros3
E     Referenced from: <0BB57518-61A3-3E6E-B925-56055D74DA28> /Users/runner/miniconda3/envs/zarrtraj-test/lib/python3.10/site-packages/h5py/defs.cpython-310-darwin.so
E     Expected in:     <AD1C1EA3-9334-3911-B862-F17F58553E3B> /Users/runner/miniconda3/envs/zarrtraj-test/lib/libhdf5.310.5.0.dylib

On Ubuntu

 ImportError while loading conftest '/home/runner/work/zarrtraj/zarrtraj/zarrtraj/tests/conftest.py'.
zarrtraj/__init__.py:7: in <module>
    from .ZARR import ZARRH5MDReader, ZARRMDWriter
zarrtraj/ZARR.py:139: in <module>
    from .utils import *
zarrtraj/utils.py:8: in <module>
    from kerchunk.hdf import SingleHdf5ToZarr
/usr/share/miniconda3/envs/zarrtraj-test/lib/python3.10/site-packages/kerchunk/hdf.py:17: in <module>
    import h5py
/usr/share/miniconda3/envs/zarrtraj-test/lib/python3.10/site-packages/h5py/__init__.py:33: in <module>
    from . import version
/usr/share/miniconda3/envs/zarrtraj-test/lib/python3.10/site-packages/h5py/version.py:15: in <module>
    from . import h5 as _h5
h5py/h5.pyx:1: in init h5py.h5
    ???
E   ImportError: /usr/share/miniconda3/envs/zarrtraj-test/lib/python3.10/site-packages/h5py/defs.cpython-310-x86_64-linux-gnu.so: undefined symbol: H5Pset_fapl_ros3

Code to reproduce the behavior

See https://github.com/Becksteinlab/zarrtraj/actions/runs/13220516432

What to do?

Not sure – possibly we need to pin hdf5 ??

[orbeckst]

@ljwoods2 @hmacdope any ideas what's happening?

For the review of the JOSS paper we should have at least green CI...

[orbeckst]

Is this a temporary fix? If so, when can we revert it?

Do we need to cut a new release?

[hmacdope]

See #79, not sure about new release, need to confirm it is not from mixing conda-forge and defaults as suggested on feedstock issue refd in #79