API: Clean up dtypes/common.py

[gfyoung]

When I was developing #16237, I noticed that the API for checking dtype is quite messy. We appear to have duplicate functionality, and the interface is not clear as to what we should accept.

I propose that for all such dtype checking functions, we should accept ANY array-like and ANY dtype OR type object and cast any list-like object to np.array.

Also, I added several TODO in that same file that I was hoping to get some feedback on. They stood out to me because they seemed either non-intuitive OR symptomatic of the API redundancy that I was referring to.

• is_datetimetz seems like a duplicate of is_datetime64tz_dtype. Do we need it?
• is_period seems like a duplicate of is_period_arraylike. Do we need it?
• Should Period be an instance of PeriodDtype?
• Should Interval be an instance of IntervalDtype?

[jreback]

I propose that for all such dtype checking functions, we should accept ANY array-like and ANY dtype OR type object and cast any list-like object to np.array.

this is likely not desired. These are for unknown array-like only (e.g. Index,ndarray,Series), which can simply be introspected with NO computation. IOW anything that needs element-by-element detection is a no-no here.

[jreback]

can you post the TODO's here.

[gfyoung]

this is likely not desired. These are for unknown array-like only (e.g. Index,ndarray,Series), which can simply be introspected with NO computation.

I'm confused, something like:

def is_*_dtype(arr_or_dtype):
    if is_list_like(arr_or_dtype) and not hasattr(arr_or_dtype, dtype):
       arr_or_dtype = np.asarray(arr_or_dtype)
...

[gfyoung]

can you post the TODO's here.

I edited the original issue to add them.

[jreback]

Should Period be an instance of PeriodDtype?
Should Interval be an instance of IntervalDtype?

yes but these would require quite some work
they only exists as Index currently

[gfyoung]

they only exists as Index currently

What do you mean?

[jreback]

welll see what happens when you store a PeriodIndex as a column

[jbrockmendel]

@gfyoung closeable? most of the topics mentioned in the OP look like they've been handled

[mroeschke]

Yup I think all the initial issues have been address so closing. We can open follow up issues targeting specific function if still necessary