Add logical operator support (__and__/__or__ methods on dataframe and column)

[MarcoGorelli]

Also a typo fix for __add__.

[rgommers]

This really was meant to be __add__ I suspect, not __and__, since it came in with the update to Column which matched DataFrame methods, and lives right above __sub__ - and __add__ was part of the arithmetic methods in gh-94. Also, the docstring says "dataframe" and that should be "column".

Adding __and__ and __or__ too seems very reasonable though, I think we can have both __add__ and __and__ here.

[MarcoGorelli]

good catch, thanks, have updated

[rgommers]

LGTM, thanks Marco. The other thing that I thought for a minute should perhaps be different is the error type, however ValueError is also used by any/all when the dtype isn't bool. So this should be uncontroversial I think.

[rgommers]

I plan to merge this within the next couple of days, unless there is a concern about adding __and__ and __or__.

[kkraus14]

Should we clarify whether Kleene logic is used for these or not (to clarify null behavior)?

Also, it looks like for __and__ we're requiring bools, but for __or__ it's more generically Scalar. Should these be consistent?

[MarcoGorelli]

but for or it's more generically Scalar.

I think it's also bools there?

def __or__(self, other: Column[bool] | bool) -> Column:

it's __add__ which can more generically take a Scalar

EDIT: nvm, the docstring did in fact say 'Scalar' - have updated, thanks!

[rgommers]

Should we clarify whether Kleene logic is used for these or not (to clarify null behavior)?

That would be good to add indeed. With the answer being that yes, Kleene logic is used I'd think - I assume that's true for all libraries?

Related: we probably should add __bool__ and make it raise, because truthiness of a column/dataframe is not defined.

[jorisvandenbossche]

I assume that's true for all libraries?

Not for the default (numpy based) data types in pandas. But I would say that's a problem to tackle for pandas / the package implementing this standard on top of pandas. I agree we should specify that it is expected use Kleene logic.

BTW, we should also specify the behaviour of nulls in the comparison methods (__eq__, __lt__, and friends)

[MarcoGorelli]

I've noted that nulls should follow kleene logic - any further requests?

[MarcoGorelli]

merging then - thanks for your reviews!