Error creating TypeVar using unpacked types as constraints

[LeoCalbi]

Bug Report

Unable to create TypeVar using unpacked types as constraints instead of explicit list of types.

In my specific instance I was trying to constraint a TypeVar using *get_args() of a TypeAlias defined before, but this problems occurs also using a explicit tuple of types.

To Reproduce

from __future__ import annotations

from typing import TypeAlias, TypeVar, get_args, TypeGuard

A: TypeAlias = int | float
d = (int,float)

#T = TypeVar("T", int, float) # Correctly working if uncommented

T = TypeVar("T", *d) 
#T = TypeVar("T", *get_args(A)) # Same results using this tuple of types instead of d

def is_valid(value:A, t:type[T])->TypeGuard[T]:
    return isinstance(A, t)

def is_int(value:A) -> TypeGuard[int]:
    return is_valid(value, int)

def is_float(value:A) -> TypeGuard[float]:
    return is_valid(value, float)

def is_str(value:A) -> TypeGuard[str]: 
    return is_valid(value, str) # Should give error if T follows the type constraints

a: A = 34
reveal_type(a) # Union[builtins.int, builtins.float]
if is_valid(x:= a, int):
    reveal_type(x)  # Should give builtins.int instead gives type is "T?"
    
if is_int(x:=a):
    reveal_type(x) # Gives type is "builtins.int" in both cases

Expected Behavior

No mypy error, with working types constraints.

Actual Behavior

In both cases the following error is identified by mypy:

• Unexpected argument to "TypeVar()" [misc] at T definition
• Variable "__main__.T" is not valid as a type [valid-type] at is_valid definition
  If using the explicit int, float type constraints code works as expected.

Your Environment

• Mypy version used: mypy 0.960
• Python version used: 3.10

[hauntsaninja]

mypy doesn't track values in this way and isn't going to support this — it's too dynamic. No other static Python type checkers would either.