Evgeni Burovski

Those in ndimage tests were workarounds for then-current infra, no more.

Ah, right. No, not deliberate. Or I forgot, too. Good catch! Needs a release note at a minimum, needs checking that there's no regression for when the env var is...

(Lucas) >> ndimage.sum_labels(input) # scalar without SCIPY_ARRAY_API, 0-D with SCIPY_ARRAY_API (Evgeni) > No, not deliberate. Or I forgot, too. Actually, it's "I forgot": https://github.com/scipy/scipy/blob/main/scipy/ndimage/_support_alternative_backends.py#L39 The decorator is only applied if...

So what's the status here, can somebody summarize? I tried merging this branch into https://github.com/scipy/scipy/pull/21117 and figure out how to use assertions from this PR. Still baffled, still have 40-odd...

Thanks Lucas. > To clarify your use case: a test module expects 0-d arrays throughout, but you would like to pass expected as a NumPy scalar sometimes? In general yes,...

I would not add yet another check_ argument. If anything, I'd default to `check_0d=False` to make usage lighter and land this.

> why not just call xp.asarray on the reference value? Yes, that. Or check-0d=False. Some 100+ times. > That would need to be done anyway when interpolate gets array API...

Indeed, in the OP example, both `b(x, y)` and `i(x)` functions are going to get the array api support, and will be returning `xp` 0D arrays on all backends. The...

LAPACK should be similar to gh-19970, modulo ILP64 which is not available yet anyhow. Also if you're lucky, BLAS prototypes are already in npy_cblas.h

A short partial answer for now only, sorry. In short, you're missing function prototypes. You can get a bit further by ```diff $ git diff diff --git a/scipy/optimize/__lbfgsb.h b/scipy/optimize/__lbfgsb.h index...