Incorrect error for `numpy.iinfo(numpy.int32)`

[jakevdp]

Describe the Bug

Pyrefly incorrectly errors for the following, with NumPy v2.4.2:

import numpy as np
np.iinfo(np.int32)

$ pyrefly check test.py
ERROR No matching overload found for function `numpy._core.getlimits.iinfo.__init__` called with arguments: (type[signedinteger[_32Bit]]) [no-matching-overload]
 --> test.py:3:9
  |
3 | np.iinfo(np.int32)
  |         ^^^^^^^^^^
  |
  Possible overloads:
  (int_type: _HasDType[dtype[_IntegerT_co]] | _HasNumPyDType[dtype[_IntegerT_co]] | dtype[_IntegerT_co] | _IntegerT_co | type[_IntegerT_co]) -> None [closest match]
  (int_type: Literal['<n', '=n', '>n', 'int', 'int_', 'intp', 'n', '|n'] | int | type[int]) -> None
  (int_type: _Int8Codes) -> None
  (int_type: _UInt8Codes) -> None
  (int_type: _Int16Codes) -> None
  (int_type: _UInt16Codes) -> None
  (int_type: _Int32Codes) -> None
  (int_type: _UInt32Codes) -> None
  (int_type: _Int64Codes) -> None
  (int_type: _UInt64Codes) -> None
  (int_type: str) -> None
 INFO 1 error

This code passes with mypy, pytype, ty, and pyright; I'm not sure exactly what the issue is with pyrefly, and I have so far been unable to reduce this to a smaller/self-contained reproduction.

Sandbox Link

Sandbox link not possible because of NumPy requirement.

(Only applicable for extension issues) IDE Information

No response

[kinto0]

removing the other overloads leads to a clearer picture of the issue:

Argument `type[signedinteger[_32Bit]]` is not assignable to parameter `int_type` with type `Literal['<i4', '=i4', '>i4', 'i4', 'int32', '|i4']` in function `numpy._core.getlimits.iinfo.__init__`Pyrefly[bad-argument-type](https://pyrefly.org/en/docs/error-kinds/#bad-argument-type)

Pyright also raises an error here:

Argument of type "int32" cannot be assigned to parameter "int_type" of type "_Int32Codes" in function "__init__"
  Type "int32" is not assignable to type "_Int32Codes"
    Type "int32" is not assignable to type "Literal['int32']"
    Type "int32" is not assignable to type "Literal['i4']"
    Type "int32" is not assignable to type "Literal['|i4']"
    Type "int32" is not assignable to type "Literal['=i4']"
    Type "int32" is not assignable to type "Literal['<i4']"
    Type "int32" is not assignable to type "Literal['>i4']"Pylance[reportArgumentType](https://github.com/microsoft/pyright/blob/main/docs/configuration.md#reportArgumentType)

it looks like the stubs need to be changed to support signedinteger[_32Bit] if that's a common use case either by adding to int32codes here or by making np.int32 compatible with those codes

[jorenham]

It's the first overload that is relevant here:

    @overload
    def __init__(self, /, int_type: _IntegerT_co | _DTypeLike[_IntegerT_co]) -> None: ...

where

_IntegerT_co = TypeVar("_IntegerT_co", bound=np.integer, default=np.integer, covariant=True)

and _DTypeLike[T] is a union including type[T].

So since np.int32 <: np.integer, the type[int32] should be assignable to _DTypeLike[_IntegerT_co], which matches the first overload.

[kinto0]

thanks, I missed that overload. I've got an isolated sandbox example here. I've narrowed it down further here