Bug Report

From numpy/numpy#18876:

Reproducing code example:

# example.py
import numpy as np

def f(x: bool) -> bool:
    return x

y = np.bool_(True)

print(f(y))

yields

$ mypy example.py
example.py:9: error: Argument 1 to "f" has incompatible type "bool_"; expected "bool"
Found 1 error in 1 file (checked 1 source file)

The same happens for the inverse:

# example.py
import numpy as np

def f(x: np.bool_) -> np.bool_:
    return x

y = True

print(f(y))

results in

example.py:9: error: Argument 1 to "f" has incompatible type "bool"; expected "bool_"

np.float_ and np.int_ work fine.

As @BvB93 points out, this is not an issue that numpy can fix, as python does not allow creating subclasses of bool:

Runtime:

In [1]: class bool_(bool): pass
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-1-823896cfc52c> in <module>
----> 1 class bool_(bool): pass

TypeError: type 'bool' is not an acceptable base type

mypy:

In [2]: from typing import TYPE_CHECKING

In [3]: if TYPE_CHECKING:
   ...:     class bool_(bool): pass  # error: 'bool' is not a valid base class
   ...:

Is there a possibility to fix this within mypy? E.g. by adding a new Bool or BoolLike type that allows for comparisons. numpy could then possibly create np.bool_ as a subclass of Bool.