python-numpy: Update to 2.4.0

[lazka]

No description provided.

[ognevny]

I've pushed numba update so the check should pass soon

[striezel]

The check fails even with numba 0.63.1:

  Pip Check Failed for mingw-w64-ucrt-x86_64-python-numpy
  Running `pip check` didn't suceed. Maybe missing dependencies?
  Here is what it failed.
  numba 0.63.1 has requirement numpy<2.4,>=1.22, but you have numpy 2.4.0.

[ognevny]

the best way to ensure is trying script with numba. if there is such requirement, ImportError will be raised

update: there is no maximum required version https://github.com/numba/numba/blob/main/numba%2F__init__.py#L34-L40

[striezel]

update: there is no maximum required version https://github.com/numba/numba/blob/main/numba%2F__init__.py#L34-L40

That's only on the main branch. However, if you take a look at the tagged version, then things look different: https://github.com/numba/numba/blob/7484a86bd9ce4d7dbc386da887f62a7bd54be597/numba/__init__.py#L42-L45 There's a hard-coded requirement for NumPy 2.3 or less. This requirement was added in commit numba/numba@be9b583. Looks like the numba folks pin their dependencies before a release is made / tagged to certain upper versions, but they don't seem to do that on the main branch.

We could revert that (it will probably work with NumPy 2.4, too), or we wait for the next numba release which will hopefully support NumPy 2.4.x, or we convince the numba developers to not pin their dependencies in that way anymore.

[lazka]

I tried some small examples and they failed, so looks like it is incompatible. Let's wait a bit then.

[ognevny]

even with patched numpy requirements?

[lazka]

yes

import numpy as np
from numba import jit

# Numba-accelerated function
@jit(nopython=True)
def matrix_multiply_numba(A, B):
    """Matrix multiplication with Numba JIT compilation"""
    m, n = A.shape
    n, p = B.shape
    C = np.zeros((m, p))
    
    for i in range(m):
        for j in range(p):
            for k in range(n):
                C[i, j] += A[i, k] * B[k, j]
    
    return C

A = np.random.rand(500, 500)
B = np.random.rand(500, 500)
matrix_multiply_numba(A[:10, :10], B[:10, :10])

  File "C:/msys64/clang64/lib/python3.12/site-packages/numba/np/arraymath.py", line 2225, in <module>
    @overload(np.trapz)
              ^^^^^^^^
  File "C:/msys64/clang64/lib/python3.12/site-packages/numpy/__init__.py", line 792, in __getattr__
    raise AttributeError(f"module {__name__!r} has no attribute {attr!r}")
AttributeError: module 'numpy' has no attribute 'trapz'. Did you mean: 'trace'?

[lazka]

maybe backport this numba/numba#10329 ?

that just leads to:

  File "C:/msys64/clang64/lib/python3.12/site-packages/numba/np/arraymath.py", line 4957, in <module>
    @overload(np.in1d)
              ^^^^^^^
  File "C:/msys64/clang64/lib/python3.12/site-packages/numpy/__init__.py", line 792, in __getattr__
    raise AttributeError(f"module {__name__!r} has no attribute {attr!r}")
AttributeError: module 'numpy' has no attribute 'in1d'. Did you mean: 'int16'?

edit: upstream issue for 2.4: numba/numba#10263