problem validating bool

[bendichter]

Running validation on the Neuropixel file found here reveals an issue with the validator when validating a boolean value. It appears there is a confusion between bool and bool_.

$ python -m pynwb.validate /Volumes/easystore5T/data/Allen/neuropixel/sub-699733573_ses-715093703.nwb
Validating /Volumes/easystore5T/data/Allen/neuropixel/sub-699733573_ses-715093703.nwb against cached namespace information using namespace ndx-aibs-ecephys.
 - found the following errors:
EcephysElectrodeGroup/has_lfp_data (general/extracellular_ephys/probeA.has_lfp_data): incorrect type - expected 'bool', got 'bool_'
EcephysElectrodeGroup/has_lfp_data (general/extracellular_ephys/probeB.has_lfp_data): incorrect type - expected 'bool', got 'bool_'
EcephysElectrodeGroup/has_lfp_data (general/extracellular_ephys/probeC.has_lfp_data): incorrect type - expected 'bool', got 'bool_'
EcephysElectrodeGroup/has_lfp_data (general/extracellular_ephys/probeD.has_lfp_data): incorrect type - expected 'bool', got 'bool_'
EcephysElectrodeGroup/has_lfp_data (general/extracellular_ephys/probeE.has_lfp_data): incorrect type - expected 'bool', got 'bool_'
EcephysElectrodeGroup/has_lfp_data (general/extracellular_ephys/probeF.has_lfp_data): incorrect type - expected 'bool', got 'bool_'

[dsleiter]

@bendichter I looked into this, and here's what I believe is going on:

1. the spec for EcephysElectrodeGroup has a dtype of bool for attribute has_lfp_data

   {'name': 'has_lfp_data', 'doc': 'Indicates availability of LFP data', 'dtype': 'bool'}

2. the type of the value for has_lfp_data in this file is np.bool_
3. it looks to me like ObjectMapper always converts bool to np.bool_ when writing to the file (is this correct?)
4. In addition, the user could have explicitly passed in an np.bool_ when creating the EcephysElectrodeGroup object because docval explicitly allows np.bool_ to pass validation when the type is specified as bool

   def __is_bool(value):
       return isinstance(value, bool) or isinstance(value, np.bool_)

5. however, hdmf.validate.validator does not allow np.bool_ to pass validation when the dtype in the spec is bool

It seems like the solution here should be to update the validator in hdmf.validate.validator to allow np.bool_ as a valid dtype when the spec defines the dtype as bool, do you agree? If so, I'll add the issue to hdmf and work on a PR.

[bendichter]

@dsleiter thanks for tracking this down. I agree, it sounds like allowing np.bool_ to count as a bool in hdmf.validate.validator would fix the problem.

[dsleiter]

A fix for this has been implemented here: hdmf-dev/hdmf#505 , and should be available as of the next hdmf version (currently planned to be 2.4.0).