`read_spikeglx` broken on the latest release (0.104.6)

[h-mayorquin]

neo latest release 0.14.5 removed the load_sync_channel argument from SpikeGLXRawIO (NeuralEnsemble/python-neo#1683). We drop it here as well (#4560), but the latest release did not include it. The result is that read_spikeglx raises a TypeError immediately under neo 0.14.5, which a fresh install now resolves to since the floor is only neo>=0.14.4.

This is breaking neuroconv (catalystneuro/neuroconv#1763) and will break any other users of spikeglx that do not have a cap on python-neo. Can we cut a release that includes #4560 on it?

[h-mayorquin]

Fixed

[ecobost]

This still breaks (in the current stable release) when a recording saved with SI<0.104.7 is loaded from a dictionary because the load_sync_channel kwarg is saved in the .json/.pickle:

In [2]: import spikeinterface.core.loading as sl

In [3]: sl.load('./01_preprocessing/ap.json')
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[3], line 1
----> 1 sl.load('/pre-edi-pipeline/orcapod/pipeline_output/spike_sorting/Malachi/20260316T193257Z/probe_imec0/01_preprocessing/ap.json')

File ~/enigma-ephys/.venv/lib/python3.14/site-packages/spikeinterface/core/loading.py:99, in load(file_or_folder_or_dict, **kwargs)
     97     if object_type is None:
     98         raise ValueError(_error_msg.format(file_path=file_path))
---> 99     return _load_object_from_dict(d, object_type, base_folder=base_folder)
    101 elif is_local and file_path.is_dir():
    103     folder = file_path

File ~/enigma-ephys/.venv/lib/python3.14/site-packages/spikeinterface/core/loading.py:165, in _load_object_from_dict(d, object_type, base_folder)
    163 def _load_object_from_dict(d, object_type, base_folder=None):
    164     if object_type in ("Recording", "Sorting", "Recording|Sorting"):
--> 165         return BaseExtractor.from_dict(d, base_folder=base_folder)
    167     elif object_type == "Templates":
    168         from spikeinterface.core import Templates

File ~/enigma-ephys/.venv/lib/python3.14/site-packages/spikeinterface/core/base.py:593, in BaseExtractor.from_dict(dictionary, base_folder)
    591     assert base_folder is not None, "When  relative_paths=True, need to provide base_folder"
    592     dictionary = make_paths_absolute(dictionary, base_folder)
--> 593 extractor = _load_extractor_from_dict(dictionary)
    594 folder_metadata = dictionary.get("folder_metadata", None)
    595 if folder_metadata is not None:

File ~/enigma-ephys/.venv/lib/python3.14/site-packages/spikeinterface/core/base.py:1128, in _load_extractor_from_dict(dic)
   1126 for name, value in dic["kwargs"].items():
   1127     if is_dict_extractor(value):
-> 1128         new_kwargs[name] = _load_extractor_from_dict(value)
   1129     elif isinstance(value, dict):
   1130         new_kwargs[name] = {k: transform_dict_to_extractor(v) for k, v in value.items()}

File ~/enigma-ephys/.venv/lib/python3.14/site-packages/spikeinterface/core/base.py:1128, in _load_extractor_from_dict(dic)
   1126 for name, value in dic["kwargs"].items():
   1127     if is_dict_extractor(value):
-> 1128         new_kwargs[name] = _load_extractor_from_dict(value)
   1129     elif isinstance(value, dict):
   1130         new_kwargs[name] = {k: transform_dict_to_extractor(v) for k, v in value.items()}

    [... skipping similar frames: _load_extractor_from_dict at line 1128 (2 times)]

File ~/enigma-ephys/.venv/lib/python3.14/site-packages/spikeinterface/core/base.py:1128, in _load_extractor_from_dict(dic)
   1126 for name, value in dic["kwargs"].items():
   1127     if is_dict_extractor(value):
-> 1128         new_kwargs[name] = _load_extractor_from_dict(value)
   1129     elif isinstance(value, dict):
   1130         new_kwargs[name] = {k: transform_dict_to_extractor(v) for k, v in value.items()}

File ~/enigma-ephys/.venv/lib/python3.14/site-packages/spikeinterface/core/base.py:1145, in _load_extractor_from_dict(dic)
   1142     new_kwargs = extractor_class._handle_backward_compatibility(new_kwargs, dic)
   1144 # Initialize the extractor
-> 1145 extractor = extractor_class(**new_kwargs)
   1147 extractor._annotations.update(dic["annotations"])
   1148 for k, v in dic["properties"].items():

TypeError: SpikeGLXRecordingExtractor.__init__() got an unexpected keyword argument 'load_sync_channel'

There is a established path to handle backwards compatibility here

spikeinterface/src/spikeinterface/core/base.py

Lines 1140 to 1142 in 76c4184

	is_old_version = not _check_same_version(class_name, dic["version"])
	if is_old_version and hasattr(extractor_class, "_handle_backward_compatibility"):
	new_kwargs = extractor_class._handle_backward_compatibility(new_kwargs, dic)

but that does not work for the stable version because:

1. is_old_version can be False (because it only checks for major&minor version so for instance if you saved something with 0.104.x and reload it with 0.104.x but the new neo, then the backward compatibility check won't happen)
2. even if is_old_version is True (say I am trying to reload a extractor saved with 0.103.x), the code to drop load_sync_channel as kwarg only exists in 0.105.0, so it still won't get handled in any SI<0.105.0. See: Fix backward compatibility for binary recording/openephys/spikeglx #4666

I think rather than changing _check_same_version and adding the _handle_kwargs_backward_compatibility in 0.104.9, we add the neo<=0.14.4 as restriction and drop it on main. Essentially say, SI<0.105.0 only works for neo<0.14.4.