Inconsistent handling of unsolved TypeVars

PEP 484 doesn't provide much clarity about the situation where a TypeVar is provided with no value in a generic function call, so I'm not sure if there is a "correct" behavior. However, I'd expect that the behavior would at least be consistent for a type checker.

from typing import Optional, TypeVar

T1 = TypeVar("T1")
def func1(suffix: Optional[T1] = ...) -> T1:
    ...

T2 = TypeVar("T2", bound=str)
def func2(suffix: Optional[T2] = ...) -> T2:
    ...

T3 = TypeVar("T3", str, bytes)
def func3(suffix: Optional[T3] = ...) -> T3:
    ...

T4 = TypeVar("T4", bytes, str)
def func4(suffix: Optional[T4] = ...) -> T4:
    ...

reveal_type(func1()) # <nothing>
reveal_type(func2()) # <nothing>
reveal_type(func3()) # str
reveal_type(func4()) # bytes

Note that mypy's results depend on whether the TypeVar is bound or constrained. If it is constrained, mypy appears to (arbitrarily?) choose the first constrained type.

Note that this situation arises in some common situations like the following:

from tempfile import mkstemp
_, temp_path = mkstemp()
reveal_type(temp_path) # str?

Like I said, I'm not sure what the "correct" or intended behavior is in this case, but the current behavior strikes me as inconsistent.

No files or tests are named. Start by reproducing the four generic function examples and the mkstemp() case, then compare the inferred types for unconstrained, bound, and constrained TypeVars. Done requires an agreed and consistent behavior, followed by implementation and regression coverage.