Reading custom opaque types

To assist reproducing bugs, please include the following:

• Operating System: Ubuntu 18.04
• Python version: 3.6.9
• Where Python was acquired: System python
• h5py version: 3.1.0
• HDF5 version: 1.12.0

I have a h5-file which contains an array of a custom opaque 2-byte datatype (bfloat16). The data is written by a non-python application and should now be read with h5py.

I try to read the data as 2-byte-blobs and then to convert them to a float32 numpy-array. Unfortunately, I could not find a way to just read the blobs (the conversion should then be trivial).

I always get this error:

h5py/_objects.pyx in h5py._objects.with_phil.wrapper()

h5py/_objects.pyx in h5py._objects.with_phil.wrapper()

/mnt/c/git/test/.venv/lib/python3.6/site-packages/h5py/_hl/dataset.py in __getitem__(self, args)
    571         mspace = h5s.create_simple(mshape)
    572         fspace = selection.id
--> 573         self.id.read(mspace, fspace, arr, mtype, dxpl=self._dxpl)
    574 
    575         # Patch up the output for NumPy

h5py/_objects.pyx in h5py._objects.with_phil.wrapper()

h5py/_objects.pyx in h5py._objects.with_phil.wrapper()

h5py/h5d.pyx in h5py.h5d.DatasetID.read()

h5py/_proxy.pyx in h5py._proxy.dset_rw()

h5py/_proxy.pyx in h5py._proxy.H5PY_H5Dread()

OSError: Can't read data (no appropriate function for conversion path)

Is there a possibility to register a custom function for reading opaque data? Alternatively it would also be okay to read the data as (u)int16 if this is possible.

Thanks

[kutoga]

Thank for this very quick response. In general: All responses were always quick. Thanks a lot for this 😃!

If I could, I would provide a PR which fixes this, but I don’t know the code base (so my solution would be hacky). Until this is fixed, I’ll go for the not-so-nice uint16-solution.

Thanks:-)!

[aragilar]

@kutoga The exponential bias (set by H5Tset_ebias) you have for the bfloat16 looks wrong, it should be 127 not 15 (which is from half/IEEE-16 bit float), which would explain the large values.