ENH: delegation tracker

[lucascolley]

Where native implementations (more simple than the implementations in _funcs.py) are possible for known array library backends, we should add delegation to those implementations, instead of using the array-agnostic implementation. This is for performance reasons, and it may also allow us to overcome limitations in the "agnostic" implementations if they are not fully agnostic.

See https://data-apis.org/array-api-extra/contributing.html#delegation for how to do this.

• apply_where
• argpartition
• angle
• at
• atleast_nd
• broadcast_shapes
• cov
• create_diagonal
• default_dtype
• expand_dims
• isclose
• isin
• kron
• lazy_apply
• nan_to_num
• nunique
• one_hot
• pad
• partition
• setdiff1d
• sinc
• union1d

[Cyril-36]

Hi, I’d like to work on adding delegation for broadcast_shapes as part of this tracker. I’ve read the contributing section on how to add delegation to a function and will follow that approach. Please let me know if anyone is already working on broadcast_shapes; if not, I’ll start implementing it and open a PR.

[lucascolley]

@Cyril-36 go for it!

[lucascolley]

Every item that was on the list here has been checked off. Remaining work:

• check we haven't missed anything
• check that we're happy that the coverage for functions we've checked off is up-to-date

At that point, I think we can close this and just add delegation on an as-requested basis.

[lucascolley]

looks like angle is still remaining