Type error with np.dtype

[person142]

Minimal reproducer

Running mypy on the following code:

import numpy as np

object_dtype = [('field1', object)]
np.dtype(object_dtype)

Results in the following error:

$ mypy test_dtype.py
test_dtype.py:4: error: Argument 1 to "dtype" has incompatible type "List[Tuple[str, Type[object]]]"; expected "Union[dtype, None, type, str, Tuple[Any, int], Tuple[Any, Union[int, Sequence[int]]], List[Union[Tuple[Union[str, Tuple[str, str]], Any], Tuple[Union[str, Tuple[str, str]], Any, Union[int, Sequence[int]]]]], Dict[str, Union[Sequence[str], Sequence[Any], Sequence[int], Sequence[Union[bytes, str, None]], int]], Dict[str, Tuple[Any, int]], Tuple[Any, Any]]"

Note that putting object_dtype on the previous line is important; otherwise the context of the function signature:

https://mypy.readthedocs.io/en/latest/type_inference_and_annotations.html#context-in-type-inference

will broaden the type to List[Tuple[str, Any]] (I think.)

Context

Stumbled across this while working on typing in SciPy, with the current version of the NumPy stubs there's an error because of this line:

https://github.com/scipy/scipy/blob/master/scipy/io/matlab/tests/test_mio.py#L140

Speculation

I think this is happening because List is invariant and object is a subtype of Any, so despite the fact that _DtypeLikeNested is typed as any here:

https://github.com/numpy/numpy-stubs/blob/master/numpy-stubs/__init__.pyi#L41

object still isn't let through here:

https://github.com/numpy/numpy-stubs/blob/master/numpy-stubs/__init__.pyi#L61