Incorrect handling of Concatenate when used with generics and decorators

[mkzeender]

Describe the bug
A clear and concise description of the behavior you are seeing and the expected behavior along with steps to reproduce it.

In the following example, we define a generic decorator that turns a (higher order) function into a MyGeneric (which could be, i.e. a descriptor).

We use that decorator on dumb_thingy, a function that takes a function and injects an int argument into it. The type checker should therefore assume that the ParamSpec **Args has a single mandatory int argument. However, pyright incorrectly infers that there are TWO mandatory int arguments.

Interestingly, this bug does NOT apply to the Callable generic, only user-defined generics. I.e. replacing MyGeneric[Args] with Callable[Args, Any] exhibits the correct behavior. It also does NOT apply unless Concatenate is used. I.e. replacing Concatenate[int, P] with [int] exhibits the correct behavior.

from typing import Any, Concatenate, Callable

class MyGeneric[**Args]: ...

def my_decorator[
    **Args,
](dec_func: Callable[[Callable[Args, Any]], Any]) -> MyGeneric[Args]: ...

@my_decorator
def dumb_thingy[**P](func: Callable[Concatenate[int, P], Any]): ...

_v: MyGeneric[[int]] = dumb_thingy  # false positive type error

_v2: MyGeneric[[int, int]] = dumb_thingy  # passes type check.

Tested on pyright command line version 1.1.392
Also occurs on latest pylance VSCode extension.
mypy has the correct behavior.

[erictraut]

Thanks for the bug report.

On first inspection, there are actually two bugs here. The first bug is a type evaluation bug that results in the int parameter being duplicated (there should be only one). The second is a false negative condition. Both _v and _v2 should result in errors, but _v2 currently does not result in an error. Mypy correctly reports errors in both cases.

[mkzeender]

@erictraut

Sorry for the poor example; I'll take your word that they should both be errors. I don't know enough type theory to have an opinion on whether MyGeneric[[int, ...]] should be assignable to MyGeneric[[int]] (where the ... represents an unknown amount of parameters).

But the duplicated int parameter is definitely a problem.

[erictraut]

This is addressed in pyright 1.1.393.