Release 1.11

[crusaderky]

Both my work on JAX and @lucascolley's / @lithomas1's work on Dask in scipy depends on the latest changes in array-api-compat.
@ev-br can we have a new release please?

Must have

• BUG: .device attribute inside jax.jit #238
• ENH: Dask: sort and argsort #239

Nice to have

• TST: run parameterized tests on more libraries #227
• BUG: astype: add device kwarg #240

[ev-br]

Yes, the plan is to cut the release with the 2024.12 support right after the spec is finalized.

[crusaderky]

That sounds like a while in the future though. Would it be possible to have a release soon with just the items listed above?

[ev-br]

I'm hoping that the 2024.12 spec is going to get finalized next week or so.

[crusaderky]

It's been a month now, and to my understanding there is a lot of non-trivial work to do, namely for take_along_axis.
Could we have an intermediate release now without waiting for full 2024.12 support?

[ev-br]

Here's the agenda for today's Consortium meeting https://hackmd.io/zn5bvdZTQIeJmb3RW1B-8g?view, which is open all interested parties (hint hint).

Looking at the 2024 milestone, https://github.com/data-apis/array-api/issues?q=is%3Aissue%20state%3Aopen%20milestone%3Av2024, the agenda is dominated by fancy indexing --- where one of the major topics is dask compatibility (hint hint).

The other open issue in the 2024 milestone is around copy, which is also dask-heavy.

Helping push these across the line (or kick the can down the road towards v2025) would be very welcome.

[crusaderky]

None of these things have an impact on the fact that the changes that are already in array-api-compat main are causing attrition in array-api-extra. Nor on take_along_axis needing a lot of work to be implemented.
And if the consortium ends up approving more complicated indexing, I would expect that will take multiple weeks of work to implement in array-api-compat, array-api-strict, and array-api-tests - one more reason to cut a release sooner than that.

[ev-br]

And if the consortium ends up approving more complicated indexing, I would expect that will take multiple weeks of work to implement in array-api-compat, array-api-strict, and array-api-tests - one more reason to cut a release sooner than that.

Fancy indexing in -compat, -strict and -tests are actually draft-ready, so no, it should not take multiple weeks of work.

From where I sit, best ways to speed up the release are

• review Test fancy indexing with integers and index arrays array-api-tests#343 and fancy indexing with ints and integer arrays array-api-strict#127
• review the changelogs: release 1.11 branch #250 and release branch 2.3 array-api-strict#123
• help crafting the spec in feat: add support for integer array indexing array-api#900
• (personal opinion) argue for punting on dask/copy stuff in https://github.com/data-apis/array-api/milestone/4

[ev-br]

https://pypi.org/project/array-api-compat/1.11/ is a thing