[Feature]: attribute reftype list

[bendichter]

What would you like to see added to HDMF?

I am creating an extension where I have a biosample that has a was_derived_from attribute which can point to other biosamples. I have tried do this as follows:

NWBAttributeSpec(
    name="wasDerivedFrom",
    doc="Describes the hierarchy of sample derivation or aggregation.",
    required=False,
    shape=(None, ),
    dtype=NWBRefSpec(
        reftype="object",
        target_type="Biosample",
    )
),

the refspec part works as expected for a single biosample object, but I am unable to use the shape parameter to create a list of there references in the attribute.

1. Is this even possible in HDF5?
2. I am not aware of any place in the schema currently that uses this. Do we want to support it?

I suppose I could store the links in groups and use a MultiContainerInterface for this, so I guess it isn't strictly blocking me.

Is your feature request related to a problem?

No response

What solution would you like?

I would like to be able to store multiple references in a single attribute.

Do you have any interest in helping implement the feature?

Yes, but I would need guidance.

Code of Conduct

• I agree to follow this project's Code of Conduct
• Have you checked the Contributing document?
• Have you ensured this change was not already requested?

[rly]

It is possible to create an HDF5 attribute containing multiple references.

>>> import h5py
>>> myfile = h5py.File('myfile.hdf5', 'w')
>>> myfile.create_group("test_group")
<HDF5 group "/test_group" (0 members)>
>>> myfile.attrs.create("attr_of_refs", data=[myfile.ref, myfile["test_group"].ref], dtype=h5py.ref_dtype)
>>> myfile[myfile.attrs["attr_of_refs"][0]]
<HDF5 group "/" (1 members)>
>>> myfile[myfile.attrs["attr_of_refs"][1]]
<HDF5 group "/test_group" (0 members)>

Could we add this to HDMF? Yes.

Should we? I am concerned about adding complexity to HDMF and the schema language to support a single use case that has another solution. We would also need to support this data type in MatNWB and HDMF-Zarr. There are possible workarounds, but I understand that they are not as elegant:

1. You can define your Biosample data type to contain one or more links to Biosample data types (though see open issue [Bug]: get_class can't work recursively #794)
2. You can use a dataset spec instead of an attribute spec. A dataset of references is supported by HDMF and commonly used.