API: Categorical constructor shouldn't accept scalar for categories

[TomAugspurger]

Code Sample, a copy-pastable example if possible

I was surprised to find that Categorical accepts scalars for categories.

In [1]: import pandas as pd
In [2]: pd.Categorical([1, 2], categories=2)
Out[2]:
[NaN, 2.0]
Categories (1, int64): [2]

AFAICT, this isn't documented, though we explicitly test it here. Am I missing a use-case where this is desirable, or should we just depreciate it and simplify the constructor?

Anyone want to guess what the categories are on pd.Categorical(['a', 'b'], categories='abcd') :)

cc @JanSchulz

[jreback]

we generally do this, iow, accept scalars in list-like fields. So ok with this (also ok is to raise a ValueError). Maybe that would be better.

[jankatins]

Not sure, I tried to find out if I wrote a message but it's all part of a squashed commit (0165d14), and none of the preserved comit messages show anything about it :-/ I think @jreback explaination is correct: it uses levels = _convert_to_list_like(levels), so it's the same logic as elsewhere...

[jorisvandenbossche]

I would just raise an error on this, no reason to allow it IMO (and I would just regard it as a bug fix without deprecation, or would there be people doing this?)