numpy: setting an ndarray shape is deprecated since 2.5.0

[woutdenolf]

• The PR title is formatted as: <Module or Topic>: <Action> <Summary> (see contributing guidelines)

PR summary

AI Disclosure

• No AI used

For reference: #4602 (comment)

[woutdenolf]

Needs the equivalent fix in fabio: silx-kit/fabio#641. Or we could ignore the fabio warnings.

[loichuder]

Looks okay to me.

Do you think we should put copy=False everywhere to be sure that no copy is done?

[woutdenolf]

Do you think we should put copy=False everywhere to be sure that no copy is done?

I was thinking about that but decided not to in the end.

I doubt the array.shape = ... pattern was actually used to ensure data did not get copied. We want to reshape the array and if the array happens to be non-contiguous or has an incompatible stride pattern/view that leads to copying, we still want to reshape. There are many other numpy operations that could lead to copying or not depending on the array/view. We rely on numpy to not copy when not needed.

I always hated the array.shape = ... pattern, it smells and I never used it. Seems like the numpy developers agree.

[loichuder]

We want to reshape the array and if the array happens to be non-contiguous or has an incompatible stride pattern/view that leads to copying, we still want to reshape.

Yeah, ok, I do agree.