gh-163: full support of NumPy v2

[paddyroddy]

Fixes #163. Think the only change is required is np.trapz to np.trapezoid, verified with NPY201. Also fixes #189.

[paddyroddy]

Have had to drop 3.8-3.9 support for full v2 support. Fixes #198.

[Saransh-cpp]

Looks good! But should this be 2 PRs? Or at least should be 2 separate CHANGELOG notes (dropping Python versions should go in breaking changes?)?

[paddyroddy]

Looks good! But should this be 2 PRs? Or at least should be 2 separate CHANGELOG notes (dropping Python versions should go in breaking changes?)?

Yeah I guess the Python changes (#198) should be done first, then NumPy. I hadn't realised they were necessary at the time.

[ntessore]

AFAICT, this would make numpy v2 required, not merely compatible without deprecations. I find that a bit of a strong requirement when the code works perfectly well with numpy<2 as well. Why not conditionally import trapezoid or trapz as trapezoid somewhere in our utility modules, then use that?

[paddyroddy]

AFAICT, this would make numpy v2 required, not merely compatible without deprecations. I find that a bit of a strong requirement when the code works perfectly well with numpy<2 as well. Why not conditionally import trapezoid or trapz as trapezoid somewhere in our utility modules, then use that?

Yep, I had thought about that. I think this will become a piece of frustrating maintenance burden to be both API compatible, whilst still maintaining numpy<2. What do you think? Happy to do as you suggest, but I think it will quickly become a bit of a barrier

[ntessore]

I would probably keep it as long as it's one single change.

To be evaluated again if we start implementing the Array API, since NumPy 2 should be compatible with that out of the box, so we could forego the compatibility layer. Not sure if that's desirable, though, so probably something to discuss soon.

[paddyroddy]

I would probably keep it as long as it's one single change.

To be evaluated again if we start implementing the Array API, since NumPy 2 should be compatible with that out of the box, so we could forego the compatibility layer. Not sure if that's desirable, though, so probably something to discuss soon.

Let's discuss this on Monday. Making sure I understand your POV.

[paddyroddy]

Closing in favour of #326. Can be revisited when we are happy to drop 3.9 support (#198).