Add special case for integers raised to a power of 2

[jakebailey]

Per: #4352 (comment)

In general, you can't say that the return value of __pow__ is definitely an integer when called on an int, as a negative power means it returns a float. However, it's pretty common to just write x ** 2, moreso than any other literal. Add an overload that recognizes that any int to the second is still an int.

(Sorry to have taken so long to submit this.)

[jakebailey]

Oops, I'm missing the actual pow function, so I'll get that in before this would be merged (though I'm thinking adding it may be annoying, versus __pow__ is rarely directly called).

Two questions I should ask:

• What is the preferred ordering here? Most specific to least specific? I'm thinking it might be strange for Literal[2] to be presented before more useful options.
• Should I leave the optional modulo param? I'm thinking there is no modulo that could be provided that would make a power of 2 not be an int.

[JelleZijlstra]

The more specific overload must be first because the first overload that matches gets picked. If the return type is int regardless of the value of modulo, I don't see why you shouldn't add the modulo param to the Literal[2] overload.

[jakebailey]

Done.

I'm thinking this is okay as-is, if you're willing to let pow() and __pow__ not match. Given things are matched top to bottom, I think it might be annoying for users (with editors showing these overloads) to see the Literal[2] overload as the first choice in their tooltips. Whereas directly accessing the dunder method is very rare (and I was targeting x ** 2).

[JelleZijlstra]

Unfortunately there's a mypy self-test failure because some error message needs to be adjusted. I can deal with it, but if you have time it would be helpful to make a PR to mypy to update the message.

[jakebailey]

Hm, isn't this a chicken/egg problem, where mypy needs to be updated to have this new error, but typeshed can't merge until mypy accepts it?

I can take a look later, I've never touched mypy. 🙂

[JelleZijlstra]

Yes it is, so we'll have to merge the two PRs concurrently, which is why it's annoying to do. :)

[Akuli]

x ** 3 might be somewhat common as well, support that too?

[jakebailey]

It's a slippery slope if more were to be added. When I ran a quick search of a bunch of projects I have checked out for testing, ** 2 was used an order of magnitude more than any other power. #4352 (comment)

[jakebailey]

Forgive me for not updating this; I plan to get back into this / what needs to change in mypy on Friday when I have a bit more time.

[jakebailey]

I've opened python/mypy#9415 to update that self test.