DEP: deprecate scalar conversions for arrays with ndim>0

[Micky774]

Related Issues/PRs

Resolves #10615 (stalled)

Fixes #10404

Changes

PR #10615: Raises deprecation warning for automatic conversion of size=1 ndarray instances to numpy scalars, e.g.

a = np.array([3.14]) #deprecated
b = np.array([[3.14]]) #deprecated
c = a[0] #preferred

This PR: Reconciles main and fixes tests.

Comments

Co-authored by @nschloe

[mattip]

If I recall correctly the ArrayAPI always requires indexing, so this deprecation brings us closer to the ArrayAPI.

It seems 41 commits is a bit excessive, we could squash-merge it to 1 when merging, or is there a reason to preserve some of the history here?

I think we should try to merge this early in the 1.24 version cycle, to see how much noise it creates downstream.

[Micky774]

It seems 41 commits is a bit excessive, we could squash-merge it to 1 when merging, or is there a reason to preserve some of the history here?

squash-merge is fine with me

[nschloe]

Squash-merge does not sound good to me. I did almost all of the hard work here, and I would not like to see @Micky774 taking all of the credit for it.

Edit: I would like to challenge @Micky774's wording "Co-authored by @nschloe". This PR is not co-authored by me. It is authored by me.

[Micky774]

See relevant comment

@nschloe I'm happy to close my PR. The reason I opened my PR so quickly after asking was because I saw that there was no activity on this PR for ~7 months already, and because I wouldn't mind closing it if you were interested in continuing this one.

I just want to make it clear that this wasn't in interest of claiming credit for your work -- in the main body of my PR I laid out explicitly that the significant contributions were yours and that mine were minor reconciliations. Sorry if this caused you to worry that the work you did was going to be undermined. Again, it really is good work and I apologize that my actions caused an unnecessary stir.