Add small test exposing issue from #7794 and suggestion for _wrap_numpy_scalars fix

[huard]

_wrap_numpy_scalars relies on np.isscalar, which incorrectly labels a single cftime object as not a scalar.

import cftime
import numpy as np

c = cftime.datetime(2000, 1, 1, calendar='360_day') 
np.isscalar(c)  # False

The PR adds logic to handle non-numpy objects using the np.ndim function. The logic for built-ins and numpy objects should remain the same.

The function logic could possibly be rewritten more clearly as

    
    if hasattr(array, "dtype"):
        if np.isscalar(array):
            return np.array(array)
        else:
            return array
    
    if np.ndim(array) == 0:
        return np.array(array)
    
    return array

• Closes TypeError for time_bnds variable when calling Dataset.to_netcdf #7794
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

[spencerkclark]

Thanks for digging into this @huard and @aulemahal. This looks reasonable to me based on the NumPy documentation you referenced, though I probably defer to others on whether there are any subtleties we need to worry about.

Is the inclusion of the hasattr(array, "dtype") condition to avoid casting zero-dimensional non-NumPy array types to NumPy arrays?

[huard]

It was rather to avoid casting 0-dimensional non-scalar numpy objects.

>>> a = np.array("abc")
>>> np.isscalar(a)
False
>>> np.ndim(a)
0

[Zeitsperre]

Hi there, I have a library that could benefit from this fix, any updates here?

[keewis]

thanks for the ping, @dcherian, and apologies for saying I would look into this and then promptly forgetting about it. I believe it was good to wait until after numpy=2.1, though, since that appears to have broken is_duck_array.

[aulemahal]

@keewis I applied your suggestion and it seems the issue is fixed! Failures in the tests do not seem related to this PR, I see that the test_where issue is also triggered by the latest nightly upstream CI run.