Incorrect type/operator error for numpy ArrayLike

[theelderbeever]

PyLance doesn't handle numpy types appropriately and shows them as type errors when attempting to perform operations with them that do exist. Possibly related to numpy/numpy#16515 or #150.

Environment data

• Language Server version: 2021.4.2
• OS and version: XXX
• Python version: 3.8.8 (installed with pyenv)

Expected behaviour

Numpy arrays can be added, subtracted, multiplied, etc and shouldn't show up as an error.

Actual behaviour

Code Snippet / Additional information

from numpy.typing import ArrayLike

def test_func(x: ArrayLike, y: ArrayLike):
    return x + y

[jakebailey]

I think this is a duplicate of #1182, and will likely be fixed in the next release.

[jakebailey]

Hm, maybe not. ArrayLike is a union (not a constrained TypeVar, though numpy.typing does have those available), so I don't think there's a guarantee that the two parameters will be compatible with each other when added (or have +, as far as I can tell).

[jakebailey]

I'm inclined to believe that this is working as intended. Take this call for example:

test_func(1234, [1, 2, 3])

This fails at runtime, but the definition of ArrayLike is:

ArrayLike = Union[
    _ScalarLike,
    Sequence[_ScalarLike],
    Sequence[Sequence[Any]],  # TODO: Wait for support for recursive types
    _SupportsArray,
]

Two things being in this union doesn't imply that they are actually compatible. Under the hood, I believe numpy converts its array-like parameters as needed depending on the operation, but a function that tries to just do a + b isn't the same.

[jakebailey]

If you wanted to be able to handle numpy arrays, you'd probably want:

from numpy import ndarray


def test_func(x: ndarray, y: ndarray) -> ndarray:
    return x + y

This does type check. Or, something like:

import numpy as np
from numpy.typing import ArrayLike

def test_func(x: ArrayLike, y: ArrayLike) -> np.ndarray:
    return np.array(x) + y