fix(checker): avoid false overload-cannot-match with ParamSpec args

[Bahtya]

Problem

When an overload uses ParamSpec-flavored *args (P.args) or **kwargs (P.kwargs), mypy incorrectly reports that a second overload with explicit keyword-only parameters "will never be matched":

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

@overload
def bar(f: Callable[P, T], *a: P.args, **k: P.kwargs) -> T: ...
@overload
def bar(f: Callable[..., T], *a: Any, baz: int, **k: Any) -> T: ...
def bar(f, *a, **k): ...

Output:

test.pyi:9: error: Overloaded function signature 2 will never be matched: signature 1's parameter type(s) are the same or broader  [overload-cannot-match]

Root Cause

overload_can_never_match erases type variables by expanding them (erase_def_to_union_or_bound). For ParamSpec, this erases to Any, so *a: P.args and **k: P.kwargs become *a: Any and **k: Any. This makes the first overload appear to accept all possible arguments — including the baz keyword argument — causing mypy to conclude (incorrectly) that the second overload can never be reached.

In reality, P.kwargs is constrained to the wrapped function f's actual keyword parameters, so a call with baz=5 would NOT match the first overload if f doesn't accept a baz parameter.

Solution

In overload_can_never_match, check if the signature has ParamSpec-flavored variadic arguments (P.args or P.kwargs). If so, skip the can-never-match check, since we cannot reliably determine overlap after erasing ParamSpec to Any.

Testing

• Added testOverloadParamSpecNoFalsePositiveCannotMatch test case in check-parameter-specification.test
• All 463 existing overload-related tests pass (was 462 before — +1 new test)

Fixes #21171

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅

[Bahtya]

Hi mypy maintainers! Friendly ping on this PR.

To summarize for ease of review:

• All CI checks pass (including mypy_primer, mypyc, self-check across Python 3.10–3.14)
• mypy_primer result: No effect on the open source corpus — clean ✅
• The fix is minimal (27 lines added, 0 deleted): it adds an early-exit guard in overload_can_never_match when either signature uses ParamSpec-flavored variadic args (P.args/P.kwargs), since erasing ParamSpec to Any makes overlap analysis unreliable.
• Fixes Overlapping overloads false positive with ParamSpec shenanigans #21171: false overload-cannot-match when combining a ParamSpec-based overload with an explicit keyword-only parameter overload.

Happy to make any adjustments if needed. Thanks for your time!