Don't duplicate ParamSpec prefixes and properly substitute Paramspecs

[A5rocks]

I ran into the duplication issue personally, but this also fixes #12734

Also:

• Fixes False postives when decorating functions with signature that uses ParamSpec within a decoarator #12909

[A5rocks]

FYI, while looking through issues to see if this PR fixes anything else, I found that #14169 is untagged (it should have the topic-paramspec tag)

[JukkaL]

Thanks for the PR! I have a few questions.

[JukkaL]

This looks a bit suspicious to me. Why can't both the arg types truthy?

[A5rocks]

I haven't found a case where they were both truthy and yet not equal to each other yet -- I don't know what order to mix their arguments together. My gut says what I wrote next is correct but I would rather have an explicit test case (or be shown that this never happens)

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[A5rocks]

Just a note that this exposes a bug in the constraint finder around paramspec: I've got a local branch that fixes that. (I don't think I can have a PR depend on another?)

PS This is ready for another review :D

[A5rocks]

I have a question about implementation and I'm not sure where to ask it (is there a better spot?), so here goes! Hopefully people who know the implementation of mypy check here :P

Essentially, the goal of this PR (and the followup PR which ATM is just over at https://github.com/A5rocks/mypy/tree/fix-fundamental-paramspecs) is to get this following code to typecheck:

from typing import ParamSpec, Concatenate, Generic, TypeVar, Any
from collections.abc import Callable

P = ParamSpec("P")
T = TypeVar("T")
V = TypeVar("V")
R = TypeVar("R")

# ----- INSIDE BLOOM -----

def command_builder() -> Callable[[Callable[Concatenate[T, P], R]], Callable[P, Callable[[T], R]]]:
    def transformer(f: Callable[Concatenate[T, P], R], /) -> Callable[P, Callable[[T], R]]:
        def returned(*args: P.args, **kwargs: P.kwargs) -> Callable[[T], R]:
            def returned_transformer(z: T) -> R:
                return f(z, *args, **kwargs)

            return returned_transformer
        
        return returned

    return transformer

# ----- TEST CASES -----
class Example(Generic[P]):
    pass

@command_builder()
def test1(ex: Example[P], /) -> Example[Concatenate[int, P]]:
    ...

reveal_type(test1)  # Is OK... But def () -> def [P] (repro.Example[P`-1]) -> repro.Example[[int, **P`-1]]
reveal_type(test1())  # Should be def [P](?) (repro.Example[P`-1]) -> repro.Example[[int, **P`-1]]

ATM on that branch mypy spits out:

repro.py:28: error: Missing return statement  [empty-body]
repro.py:31: note: Revealed type is "def [P] () -> def (repro.Example[P`-1]) -> repro.Example[[builtins.int, **P`-1]]"
repro.py:32: note: Revealed type is "def (repro.Example[<nothing>]) -> repro.Example[<nothing>]"
Found 1 error in 1 file (checked 1 source file)

Obviously that last thing is wrong. (It has bad consequences for actual usage but even just the revealed type is wrong).

The problem here is that I am calling a "def [P] () -> def (repro.Example[P`-1]) -> repro.Example[[builtins.int, **P`-1]]", which in turn tries to infer P. There are no constraints as there are no arguments, meaning that P becomes UninhabitedType which is <nothing>. I'm not sure how to solve this. My preferred outcome is that this turns into "def [P] (repro.Example[P`-1] -> repro.Example[[builtins.int, **P`-1]]".

Basically, how should I solve this? My main idea revolves about making solving for constraints check if that type var appears in the arguments and if it doesn't then it gets "assigned" to the return type (if that's possible, which is only with stuff like callables). However, I'm not sure how to even check whether a type var appears in the arguments, let alone if there's a better way to approach this.

Update: I managed to in A5rocks@8642f3d

[A5rocks]

Hello again! Another review would be much appreciated -- the followup PR fixes a few open issues and I can't open that until this is merged (as far as I know?) without messing up commit history for myself (?).

[hauntsaninja]

Thanks, this looks good to me and am excited to see your future ParamSpec improvements. I'll wait a day or two in case Jukka wants to take another look; if I don't merge soon, just ping me :-)

[hauntsaninja]

Thank you!