Boolean confusion

True and True are treated different depending on whether it is a bool or a np.bool_.
This creates absolute hellish errors. You save date, and load it again. Then bool turned to np.bool_ and things work differently.
I.e. you cannot save your data again.

The following example works perfectly fine:

>>> data = xr.Dataset()
>>> data.attrs['bool_type'] = True
>>> data.to_netcdf()
>>> data
<xarray.Dataset>
Dimensions:  ()
Data variables:
    *empty*
Attributes:
    bool_type:  True

The following example however fails:

>>> data = xr.Dataset()
>>> data.attrs['bool_type'] = np.True_
>>> data.to_netcdf()
TypeError: Invalid value for attr: True must be a number, a string, an ndarray or a list/tuple of numbers/strings for serialization to netCDF files
>>> data
<xarray.Dataset>
Dimensions:  ()
Data variables:
    *empty*
Attributes:
    bool_type:  True

If you load the data again, in both cases np.True_ leading to confusing things, when the seemingly same things fails.

For this routine a simple remedy is probably expanding the instance check in check_attr:

isinstance(value, (str, Number, np.ndarray, np.number, np.bool_, list, tuple))

As numpy warns, np.bool_ are no numbers in contrast to regular Python, see https://numpy.org/doc/stable/reference/arrays.scalars.html#numpy.bool_.

I fear, however, that this problem runs deeper and the same issue might arise at different points.

edit: included TypeError

Start by reproducing the two Dataset.attr examples with data.to_netcdf(), then inspect the check_attr validation mentioned in the issue. Verify that np.bool_ attributes serialize successfully and that the save-and-reload behavior is consistent with regular Python booleans.