BUG: Overloaded return types with np.cross NoReturn leads to type-checking issues

[danielpmorton]

Describe the issue:

np.cross appears to have a return type of NoReturn when dealing with _ArrayLikeBool_co type, and this NoReturn type is incompatible with type checking in Pylance, and VSCode determines that all code after a call to np.cross is unreachable.

This has been mentioned in a Pylance issue here (microsoft/pylance-release#3195), but they recommended mentioning it with NumPy.

Reproduce the code example:

import numpy as np

x = np.array([1,2,3])
y = np.array([2,3,4])

z = np.cross(x, y)
# This code is determined as unreachable by VSCode due to a NoReturn type
print(z)

Error message:

No response

NumPy/Python version information:

1.23.2 3.9.12 (main, Jun 12 2022, 13:39:05)
[GCC 9.4.0]

Context for the issue:

My code uses np.cross in many places, as well as extensive type-checking. It would be very nice if the return types could be cleaned up to improve compatibility with VSCode/Pylance/Pyright, because right now all of my code is grayed-out

[LeonaTaric]

in the code, only type ArrayLikeBool will noreturn, ArrayLikeInt or ArrayLikeFloat will return type Array correctly

[wzy-99]

I have the same problem, and i just commented out the NoReturn function.

[nimagitac]

I have also the same problem. Pylance marked some lines as unreachable. (The temporary remedy could be the usage of "dtype" in numpy.cross function)

[danielpmorton]

Yeah, I would prefer to not have to manually modify my own installation of NumPy to adjust this, and this should be a relatively quick fix for the development team

[wolfrayet]

I also have the same problem. I cannot solve it even if I specify the array type.

[ShapingView]

Reopen this issue?

--> #22359 (comment)

[Liquidmasl]

this issue returned. for a while now.
please re open the issue!

[mattip]

@Liquidmasl please tell us a bit more: how are you seeing this? Is it exactly like in the original description? What version of NumPy are you using?

[Liquidmasl]

EDIT: I just found this issue: #24999
sorryyy
will leave rest of comment for another lost developer like me...

@Liquidmasl please tell us a bit more: how are you seeing this? Is it exactly like in the original description? What version of NumPy are you using?

i use 1.36.3
I do not use pylance but pycharm, but the issue is (i believe) the same.
although you did make me less sure about it

when checking the overloads there is this one overload:

@overload
def cross(
    a: _ArrayLikeBool_co,
    b: _ArrayLikeBool_co,
    axisa: int = ...,
    axisb: int = ...,
    axisc: int = ...,
    axis: None | int = ...,
) -> NoReturn: ...

which seams to be the issue. although i dont quite understand why the arrays are assumed to be dtype bool

even declaring dtypes on the input arrays does not help

But maybe this is a completely different issue?

maybe i should comment there? #22359 (comment)

(sorry i am somewhat new to the open source world.)

EDIT: I just found this issue: #24999