Cannot infer generic parameter types within `match` statements

[msto]

Bug Report

Related to #13612.

#13618 fixed the error reported in #13612, so mypy correctly infers the type of the unpacked value.

However, mypy does not correctly infer the type parameter of the class containing the generic attribute within the scope of the match statement, despite correct inference outside the statement.

To Reproduce

Extending the reprex from #13612:

from typing import Generic, TypeVar

T = TypeVar("T")


class A(Generic[T]):
    x: T

    __match_args__ = ("x",)

    def __init__(self, x: T):
        self.x = x


a = A("foo")
reveal_type(a)    # A[builtins.str] (correct)
reveal_type(a.x)  # builtins.str (correct)

match a:
    case A(x) as a_match:
        reveal_type(x)          # builtins.str (correct)
        reveal_type(a_match)    # A[Any]  **(incorrect! should be A[str])**

Expected Behavior

Within the match statement, A(x) should be inferred to be of type A[str].

Revealed type is "test_generic.A[builtins.str]"
Revealed type is "builtins.str"
Revealed type is "builtins.str"
Revealed type is "test_generic.A[builtins.str]"

I have found that pyright, or at least the pylance VS code extension, infers this correctly. See below:

Actual Behavior

The final line does not reveal A[str].

Revealed type is "test_generic.A[builtins.str]"
Revealed type is "builtins.str"
Revealed type is "builtins.str"
Revealed type is "test_generic.A[Any]"

Your Environment

• Mypy version used: 1.14.1
• Mypy command-line flags: none
• Mypy configuration options from mypy.ini (and other config files): none
• Python version used: 3.12.8

Motivation

This becomes relevant when trying to match on a union of generic types, as it creates an unreachable error.

e.g.

type Union_AB[T] = A[T] | B[T]

def func(val: Union_AB[int]) -> None:
    match val:
        case A(x):
            pass
        case B(x):  #  error: Subclass of "A[int]" and "B[Any]" cannot exist: would have incompatible method signatures  [unreachable]
            pass

[sterliakov]

That happens here and is not caused by some weird type erasure:

mypy/mypy/checkpattern.py

Lines 546 to 556 in 1affabe

	if isinstance(type_info, TypeInfo):
	any_type = AnyType(TypeOfAny.implementation_artifact)
	args: list[Type] = []
	for tv in type_info.defn.type_vars:
	if isinstance(tv, TypeVarTupleType):
	args.append(
	UnpackType(self.chk.named_generic_type("builtins.tuple", [any_type]))
	)
	else:
	args.append(any_type)
	typ: Type = Instance(type_info, args)

When we encounter a class pattern, there is no relationship between its constructor (the primary way to infer type args as of now) and the pattern, so the best we can do is to fill all typevars with any.

Solving all typevars may be impossible at all: what if one of generics defines a type of some attribute not listed in __match_args__ and not captured as a kwarg? So defaulting all of them to Any is reasonable.

If a pattern matches, we may want to use that information to resolve some of generics. I don't know how complicated that would be since it's similar to regular inference, but can't rely on __init__ callable.

However, we are already able to do similar backpropagation of generics to solve generic protocols. So it may be feasible to do something similar here: collect all constraints from the pattern, map them to type variables, solve the resulting system, and apply the result to the typeinfo falling back to Any for unresolved vars.