Align ExtensionArray with abc.Sequence

[WillAyd]

ref #28762 (comment)

Right now the ExtensionArray is missing a few methods that would allow it to align with the abc.Sequence interface, namely count, index, __reversed__ and __contains__

I think we might want to add those items to the interface to aid our type checking and make for more intuitive use cases of the ExtensionArray

[jorisvandenbossche]

@WillAyd do you know how numpy arrays do that?
Because the function in question pd.array also accepts numpy arrays, so wondering how that works for typing (ndarray has no index or count methods)

[WillAyd]

Numpy arrays resolve to Any right now when it comes to typing because they have no annotations in their codebase or typeshed available (AFAIK). So realistically it's almost impossible to cause a type failure with a ndarray in our code base

[jorisvandenbossche]

For this specific case, don't we have some kind of ArrayLike in our typing module to handle this? (I suppose also Series is no Sequence, while that is often accepted as an array like as well)

Personally, I am a bit hesitant to add count and index methods to EA just for this

[WillAyd]

So you wouldn't consider an ExtensionArray to be a type of Sequence, right? Ignoring typing for a second it seems intuitive to me that an ExtensionArray is inherently a Sequence, and if that is true then it would make sense to adhere to the interface as defined by Python

[jorisvandenbossche]

Somewhat relevant discussions in numpy about this: numpy/numpy#2776 and numpy/numpy#7315 (although they have other issues that complicate things).

Yes, I would consider an EA as a sequence. I mainly find it a bit unfortunate that Python defines methods (count, index) that are not essential for sequence as part of the abc.

[WillAyd]

Nice find - I'll give it a deeper look later. FWIW I don't know if we'd necessarily even need to define count or index - I think inheriting from the ABC will take care of that for us (haven't verified)

[WillAyd]

Removing Needs Discussion label and replacing with good first issue as I think we are aligned and can probably be done easy by community, though if you disagree @jorisvandenbossche lmk

[jorisvandenbossche]

cc @TomAugspurger

[jorisvandenbossche]

One other concern that is that adding count might clash with what "count" already means in pandas. We already have a Series.count(), and while adding a EA.count to dispatch to from Series.count was not yet discussed I think, having the same methods with different behaviour is also confusing.