Thank you for opening this pull request! It may take us a few days to respond here, so thank you for being patient.
If you have questions, some answers may be found in ourcontributing guidelines.

Looks like this is failing due to dask not having implementednp.astype. Not familiar with dask. Shall I just close this, or should this be amended somehow to only usenp.astype in the numpy, non-dask case?

it's not just that: we're still maintaining backwards compatibility withnumpy<2 (see the bare-minimum CI), which means that we still need the special-casing for that, as well.

Either way, thanks for the PR. Unfortunately, I think this might not be the right time, yet.

Edit: but maybe you can repurpose the PR to make the comment more accurate? I.e. that this is for backwards-compat anddask?

@keewis Done.
I still don't know why this is needed for dask - doesn't it use its own namespace? I've no experience with dask.
Perhaps you'd like to make that clearer in the comment and/or commit message (besides also telling me).

I think you force-pushed back to the original state (I think we need the changes fromfeb81db).

dask doesn't follow the array API (yet?), which explains the missing function

Oops, sorry about that, pushed again.

dask doesn't follow the array API (yet?), which explains the missing function

but I didn't get why it's under the conditionif xp == np

Thanks! This looks good to me.

but I didn't get why it's under the conditionif xp == np

dask doesn't implement__array_namespace__, so we're falling back tonp to dispatch through__array_function__ and__array_ufunc__.

dask doesn't implementarray_namespace, so we're falling back to np to dispatch througharray_function andarray_ufunc.

Ah, I see. Thanks!