Thanks @Mousius. Let's let CI complete here, and then a rebase should make the aarch64 wheel builds on Cirrus run too. I'd be curious to see whether those pass with this update-to-latest, even if we decide to go back to the 1.0.7 release instead in the end.

Thanks @Mousius. Let's let CI complete here, and then a rebase should make the aarch64 wheel builds on Cirrus run too. I'd be curious to see whether those pass with this update-to-latest, even if we decide to go back to the 1.0.7 release instead in the end.

I don't think we can go back to 1.0.7, it's quite old and we've bumped a few times for fixes already. Potentially worth getting this in and then latching onto the next version release?

Everything seems happy here, so I've disabled non-Cirrus jobs in the rerun.

I don't think we can go back to 1.0.7, it's quite old and we've bumped a few times for fixes already. Potentially worth getting this in and then latching onto the next version release?

Sounds reasonable to me.

Looks like less errors, but still some errors. I'll try and take a look at this tomorrow.

Thanks!

Building locally on a manylinux aarch64 container will probably be more helpful that in CI. For macOS that's possible too with the Tart VM, see #25012 (comment).

@rgommers I checked out 1.0.7 on a c7g, and it actually seems ok. I think what happened is we started from a hash when VQSortStatic was introduced and didn't get the Highway release locked in, but it seems going back is alright.

I got mega test failures in ciwheelbuild on the instance but ran tests outside it, and they passed, so I think they're ok. Unsure what the difference is 🤔

Apologies for the CI resource usage 🙀 I had to re-rebase; hopefully, those jobs get shut down.

Apologies for the CI resource usage 🙀 I had to re-rebase; hopefully, those jobs get shut down.

No worries about this one at all. We try to be careful, but it's secondary to getting things to work:) We burned $15 in ~48 hours, which is fine - it's more the admin hassle of it than anything else.