dtype is not defined when using ArrayLike with get_type_hints

[thewtex]

dtype is not defined when using numpy.typing.ArrayLike with typing.get_type_hints and numpy==1.21.0.

Reproducing code example:

from typing import get_type_hints

import numpy as np

from numpy.typing import ArrayLike

def give_me_an_array(array: ArrayLike):
    pass

type_hints = get_type_hints(give_me_an_array)

Error message:

Traceback (most recent call last):
  File "/tmp/test.py", line 10, in <module>
    type_hints = get_type_hints(gimme_an_array)
  File "/home/matt/bin/miniconda/envs/numpy/lib/python3.8/typing.py", line 1264, in get_type_hints
    value = _eval_type(value, globalns, localns)
  File "/home/matt/bin/miniconda/envs/numpy/lib/python3.8/typing.py", line 272, in _eval_type
    ev_args = tuple(_eval_type(a, globalns, localns) for a in t.__args__)
  File "/home/matt/bin/miniconda/envs/numpy/lib/python3.8/typing.py", line 272, in <genexpr>
    ev_args = tuple(_eval_type(a, globalns, localns) for a in t.__args__)
  File "/home/matt/bin/miniconda/envs/numpy/lib/python3.8/typing.py", line 272, in _eval_type
    ev_args = tuple(_eval_type(a, globalns, localns) for a in t.__args__)
  File "/home/matt/bin/miniconda/envs/numpy/lib/python3.8/typing.py", line 272, in <genexpr>
    ev_args = tuple(_eval_type(a, globalns, localns) for a in t.__args__)
  File "/home/matt/bin/miniconda/envs/numpy/lib/python3.8/typing.py", line 270, in _eval_type
    return t._evaluate(globalns, localns)
  File "/home/matt/bin/miniconda/envs/numpy/lib/python3.8/typing.py", line 518, in _evaluate
    eval(self.__forward_code__, globalns, localns),
  File "<string>", line 1, in <module>
NameError: name 'dtype' is not defined

NumPy/Python version information:

This occurs with numpy 1.21.0. It did not occur with 1.20.3 or previous versions.

[BvB93]

So this is admittedly a rather annoying issue that highlights some serious limitations of typing.get_type_hints.

dtype is currently expressed as a string in the relevant union, which is fine for static type checking, but means
that you'll have to passed the stringified objects via either globals or locals. Except that this simply introduces
a new issue as dtype is not subscriptable during runtime, so you'll get a TypeError and are back to square one.

numpy/numpy/typing/_array_like.py

Lines 73 to 79 in 9b19257

	ArrayLike = Union[
	_RecursiveSequence,
	_ArrayLike[
	"dtype[Any]",
	Union[bool, int, float, complex, str, bytes]
	],
	]

In any case, I've just created #19328 which will resolve the issue once it's backported for 1.21.1.

EDIT: xref python/typing#819

[thewtex]

@BvB93 thank you!! 🙏