Hello @valtron! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

• In the file pandas/core/indexes/category.py:

Line 442:89: E501 line too long (92 > 88 characters)

Comment last updated at 2019-08-24 05:29:33 UTC

I think I'm -1 on this as it adds complexity for a rather niche case. What real advantage are you hoping to get out of replacing NA with None?

I'm also initially against this. We have a few things to work out with NA values (#28095). It's not clear how None will be handled there.

If we wanted to support this, the easier way would be

default_fill_value = object()
def fillna(..., value=default_fill_value, ...):

If the user passes None, it'd be up to the dtype to determine whether or not that's a valid fill value. But we'll need to settle on whether this is desirable before moving forward.

@valtron if you have use cases where None is a useful value, you may want to speak up in #28095.

@WillAyd I've had object columns that contain NaN (and possibly a mixture of None and Nan) and currently I do c.loc[c.isna()] = None to standardize things and simplify stuff further down the line. I prefer None to NaN (otherwise I could use c.fillna(np.nan) to standardize) because the values have to be compared later and I don't want to write (c1 == c2) or (pd.isna() and pd.isna()).

BTW, currently method=None means "fill with constant given by value" (cf. numpy.pad) so this isn't a "complex" change.

@TomAugspurger I tried implementing it that way (using a sentinel for the default value) at first, but it requires more changes, and wouldn't be backward-compatible (e.g. fillna(None, method="pad") would become an error).

@TomAugspurger I tried implementing it that way (using a sentinel for the default value) at first, but it requires more changes, and wouldn't be backward-compatible (e.g. fillna(None, method="pad") would become an error).

Can you expand on what "more changes" entails? Understood concern on the latter piece but generally this seems like a more ideal approach to get what you are after

I spent a bit more time on the alternate implementation (https://github.com/valtron/pandas/commit/66e987faca4a677928479e402302ddefb5099398); I don't have all the tests passing yet, but so far I'm guessing the remaining changes would all involve tracking down fill_values in other functions and changing them to default to MISSING, change the logic from checking from None -> MISSING, etc.

Thanks for the contribution! I think this is a hold for now though, and as mentioned before probably worth voicing thoughts in #28095 to support this feature