Invalid type revealed when type var uses another type var as default

[sobolevn]

from typing import TypeVar, Generic

X = TypeVar('X')
Y = TypeVar('Y', default=X)
Z = TypeVar('Z', default=Y)

class A(Generic[X, Y, Z]):
 x: X
 y: Y
 z: Z

a1: A[int]
reveal_type(a1.x) # N: Revealed type is "builtins.int"
reveal_type(a1.y) # N: Revealed type is "builtins.int"
reveal_type(a1.z) # N: Revealed type is "X`1"

a1.z must be revealed as int as well, not as X'1

Refs #20021

[cdce8p]

Mypy doesn't really support recursive TypeVars with the new syntax just yet. It's one of the open issue left for #14851.

[jorenham]

Mypy doesn't really support recursive TypeVars with the new syntax just yet. It's one of the open issue left for #14851.

Oh, huh, this is news to me...

Since NumPy 2.2.0 it's being used in numpy.complexfloating (numpy/numpy#27420), so that users can write complexfloating[T] instead of complexfloating[T, T]. But if I would have know this, then I probably wouldn't have added it, seeing as complexfloating is used quite frequently.

Is there anyway I could've known this back then? For example, is there some hidden mypy flag that shows a warning if you're something that mypy doesn't yet fully support? Or perhaps an overview of missing or incomplete typing-spec functionality like this?

[erictraut]

Is there anyway I could've known this back then?

The typing spec conformance test results might be of use here. This functionality is covered by the test suite named "generics_defaults_referential".

[jorenham]

Is there anyway I could've known this back then?

The typing spec conformance test results might be of use here. This functionality is covered by the test suite named "generics_defaults_referential".

Great, that helps!