Incorrect resolution of type[TypeVar] returning TypeVar with numpy

Bug Report

This came up with pandas-stubs, but can reproduce with just numpy

To Reproduce

from typing import TypeVar
import numpy as np

NDArrayT = TypeVar("NDArrayT", bound=np.ndarray)


def foo(x: type[NDArrayT]) -> NDArrayT:
    return x([1, 2])


reveal_type(foo(np.ndarray))

Expected Behavior

Revealed type should be ndarray[tuple[Any, ...], dtype[Any]], which is what pyright reports.

Actual Behavior

mypy reports numpy.ndarray[Any, Any]

Your Environment

Mypy version used: 1.19.0
Mypy command-line flags: None
Mypy configuration options from mypy.ini (and other config files): None
Python version used: 3.11
numpy version: 2.3.5
pyright version: 1.1.407

I thought this might be related to the PEP 696 type parameter defaults, but that doesn't seem to be the case:

from typing import Any, reveal_type

import numpy as np

def f[T: np.ndarray](t: type[T]) -> T: ...
def g[T: np.ndarray[tuple[Any, ...], np.dtype[Any]]](t: type[T]) -> T: ...

reveal_type(f(np.ndarray))  # Revealed type is "numpy.ndarray[Any, Any]"
reveal_type(g(np.ndarray))  # Revealed type is "numpy.ndarray[Any, Any]"

In fact, any type arguments to np.ndarray seem to be swallowed this way 🤔

Could you explain further why you expect different output? Any is compatible with any bound, type passed as argument is not parameterized hence defaults all tvars to any, why should it not be Any?

Zero-numpy example (playground):

from typing import Any, Generic, TypeVar

T = TypeVar("T")
U = TypeVar("U", bound="Arr[int]")

class Arr(Generic[T]):
    ...

def f(t: type[U]) -> U: ...

reveal_type(f(Arr))  # Revealed type is "__main__.Arr[Any]"

type passed as argument is not parameterized hence defaults all tvars to any,

As my example shows, this also occurs when its fully parametrized.
Also note that the type parameter defaults of np.ndarray are tuple[Any, ...] and np.dtype[Any]. So np.ndarray (as type expression) is therefore exactly equivalent to np.ndarray[tuple[Any, ...], np.dtype[Any]].

Ah, sorry, I missed that numpy already started using tvars defaults. So the problem reduces to the following, right?

from typing import Any, Generic, TypeVar

T = TypeVar("T", default=int)
U = TypeVar("U", bound="Arr")

class Arr(Generic[T]):
    ...

def f(t: type[U]) -> U: ...

reveal_type(f(Arr))  # Revealed type is "__main__.Arr[Any]"

Probably the default should have been applied here.

So the problem reduces to the following, right?

Yea pretty much. There's also where f is parametrized with Arr[int] (so with type arg) that behaves unexpectedly:

# mypy: disable-error-code=empty-body
from typing import Any, Generic, TypeVar

T_co = TypeVar("T_co", default=int)
class Arr(Generic[T_co]): ...

def f[T: Arr](t: type[T]) -> T: ...
def g[T: Arr[int]](t: type[T]) -> T: ...

reveal_type(f(Arr))  # Revealed type is "__main__.Arr[Any]"
reveal_type(g(Arr))  # Revealed type is "__main__.Arr[Any]"

https://mypy-play.net/?mypy=latest&python=3.14&flags=strict&gist=b51972fc19e7c56363428c62dede27e7

Just a hunch but wouldn't this be one of the cases covered by PEP 747?

Nah, this is the correct use for type - TypeForm is not needed when you have an actual class. TypeForm is compatible with things like TypeVar itself, should be irrelevant here - the problem is not evaluating Arr in f(Arr) as if it was in a type context (failure to resolve the typevar to its default). The snippet is correct, rejecting it is a defect in mypy.

Navigation Menu

Repository navigation

Incorrect resolution of `type[TypeVar]` returning `TypeVar` with `numpy` #20354

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Incorrect resolution of type[TypeVar] returning TypeVar with numpy #20354

Description

Activity

jorenham commented on Dec 2, 2025

sterliakov commented on Dec 2, 2025

jorenham commented on Dec 2, 2025

sterliakov commented on Dec 2, 2025

jorenham commented on Dec 2, 2025

jacopoabramo commented on Dec 5, 2025

sterliakov commented on Dec 5, 2025

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions

Incorrect resolution of `type[TypeVar]` returning `TypeVar` with `numpy` #20354