Summary

• validate_conversion_options was updated in Support datetime in conversion options #1139 to JSON-encode conversion_options before schema validation. This broke passing progress_bar_class (a type object) and callback functions in progress_bar_options, as _NWBConversionOptionsEncoder had no handler for these types.
• Add handling for type and callable objects in _NWBConversionOptionsEncoder, serializing them to their qualified module path string for validation purposes. The original objects are still passed through to the actual conversion code unchanged.

How this manifests

NWB GUIDE sets progress_bar_class=TQDMProgressSubscriber (a class, not an instance) in iterator_opts within conversion_options at manage_neuroconv.py:1021:

options_to_update["iterator_opts"] = dict(
    display_progress=True,
    progress_bar_class=TQDMProgressSubscriber,  # <-- type object
    progress_bar_options=progress_bar_options,   # <-- contains a callable
)

This dict ends up in conversion_options, which is passed to NWBConverter.run_conversion(). Inside run_conversion (nwbconverter.py:319):

self.validate_conversion_options(conversion_options=conversion_options)

Which calls (nwbconverter.py:201):

encoded_conversion_options = _NWBConversionOptionsEncoder().encode(conversion_options)

The encoder has no handler for type or callable objects, so it falls through to json.JSONEncoder.default() which raises:

TypeError: Object of type type is not JSON serializable

At runtime (after validation), the conversion_options dict is unpacked as kwargs into each interface's add_to_nwbfile() (nwbconverter.py:253):

data_interface.add_to_nwbfile(nwbfile=nwbfile, metadata=metadata, **conversion_options.get(interface_name, dict()))

This passes iterator_opts as a kwarg, which flows through to SpikeInterfaceRecordingDataChunkIterator / ImagingExtractorDataChunkIterator, then to hdmf's GenericDataChunkIterator, where progress_bar_class is used as a constructor. This runtime path works fine — the issue is only in the validation step.

Before vs after #1139

Before #1139, validation was:

validate(instance=conversion_options or {}, schema=self.get_conversion_options_schema())

jsonschema.validate() traverses the dict but doesn't serialize it, so class objects were fine.

After #1139:

encoded = _NWBConversionOptionsEncoder().encode(conversion_options)
decoded = json.loads(encoded)
validate(instance=decoded, schema=self.get_conversion_options_schema())

The JSON encode/decode round-trip fails on non-serializable types.

Test plan

• Verify that passing progress_bar_class as a type object in conversion_options no longer raises TypeError: Object of type type is not JSON serializable
• Verify that passing callable objects in progress_bar_options works
• Verify that normal conversion options (strings, bools, dicts) still validate correctly

🤖 Generated with Claude Code