@mhvk did you intend to update the meson submodule here?

Also, can you explain why assert_, which is listed as not recommended in the documentation, is used here? We used to use assert_ in SciPy, but my understanding was that Python assert is preferred now.

Otherwise, the code looks like it avoids turning Python scalars into NumPy types, and it makes some other simplfications. Typically when I set out to take an inactive PR over the finish line, I try to minimize changes to keep the review burden low - but it's great that you've used this as an opportunity to improve the function.

I'd suggest that someone more familiar with the intricacies review the tests when this makes its way into the NumPy PR.