Constructing a FieldArray from a list of ndarray objects fails typechecking

[maksverver]

Consider the following example:

import galois
import numpy as np

GF = galois.GF(16)

a: galois.FieldArray = GF([[1,2,3]])            # OK
b: galois.FieldArray = GF(np.array([[1,2,3]]))  # OK
c: galois.FieldArray = GF([np.array([1,2,3])])  # ERROR (see below)

print(a)  # [[1, 2, 3]]
print(b)  # [[1, 2, 3]]
print(c)  # [[1, 2, 3]]

This runs just fine, but typechecking with pyright fails only for the third assignment:

test.py:8:28 - error: Argument of type "list[NDArray[Any]]" cannot be assigned to parameter "x" of type "ElementLike | ArrayLike" in function "__new__"
    Type "NDArray[Any]" is not assignable to type "IterableLike"
      "ndarray[_Shape, dtype[Any]]" is not assignable to "Sequence[ElementLike]"
      "ndarray[_Shape, dtype[Any]]" is not assignable to "Sequence[IterableLike]" (reportArgumentType)
test.py:8:28 - error: Argument of type "list[NDArray[Any]]" cannot be assigned to parameter "x" of type "ElementLike | ArrayLike" in function "__init__"
    Type "NDArray[Any]" is not assignable to type "IterableLike"
      "ndarray[_Shape, dtype[Any]]" is not assignable to "Sequence[ElementLike]"
      "ndarray[_Shape, dtype[Any]]" is not assignable to "Sequence[IterableLike]" (reportArgumentType)
2 errors, 0 warnings, 0 informations 

(The two errors are exactly the same, except one is about __new__ and the other about __init__.)

So the question is: why is an NDArray considered not IterableLike? An ndarray instance is definitely iterable. Is this a bug in the type definitions for galois?

For reference: I'm using Python 3.13.3, galois 0.4.5, pyright 1.1.400.

[mhostetter]

I suppose we could add NDArray to IterableLike.

The typing in NumPy, and consequently this package, have been challenging to say the least. See #486.

[maksverver]

I suppose we could add NDArray to IterableLike.

That would fix my problem.

It also seems like the right thing to do since the initializer already passes the “iterable” off to np.array() here, so in practice the FieldArray constructor already accepts anything that's acceptable by np.array(), at least so long as the values can be converted to field elements.

[mhostetter]

Released in v0.4.6.