release: 0.21.0-beta.0

[davidhewitt]

Related to #3674

This is a beta release (finally 🎉) for PyO3 0.21. It contains one unreviewed commit, which is #3906 squashed. If that merges before I release this, I'll rebase first, otherwise I'll leave this as-is. Apart from that, this is just the usual version bump.

Given that this is a prerelease, I'm inclined to put this live this evening unless someone can give me a reason to wait longer! 🚀

I have a list of followups to complete after the beta release before final, related to some of the larger ecosystem projects:

• I will update pydantic-core to the beta release. I have been testing for a long time in disable PyO3 "gil refs" feature pydantic/pydantic-core#1085, so I think this should be smooth going.
• I guess we should proceed with the rust-numpy branch to confirm that it's viable to make this update.
• It would be helpful if cryptography can explore updating, cc @alex
• I will also try to reach out to the polars team to get their feedback on how this API change affects them.

I also think we have just a few deprecation warnings to finish implementing in #3684, which @Icxolu has been making great headway into and I'll try to join in when I can find a moment.

Thank you to everyone who has helped get us to this point, in particular:

• @adamreichold, @Icxolu and @LilyFoote for a lot of help writing the implementation
• @MarshalX, @snuderl, @maffoo, @dmatos2012, @mejrs and @alex who have also given ideas, feedback, contributions etc
• @samuelcolvin who has given a fair bit of my working hours at Pydantic to the project so we could make progress here

(Sorry if you also contributed and I have missed your name, this has been a long project with quite a few inputs!)

I hope that we have lots of success making the ecosystem faster and safer with the upcoming releases!

[samuelcolvin]

Amazing work @davidhewitt, thank you for all your work on this.

@ritchie46 maybe you would be interested from the Polaris POV?

[alex]

@reaperhulk has taken a look for cryptography (based on main from a day or two ago) -- I believe the summary is:

• ~20 warnings with the gil-refs feature enabled, almost all related to is_true. Easy to resolve.
• Tests pass with those resolved.
• Building with gil-refs disabled is approximately 90 gazillion warnings, and that's going to take some time :-)

But he can confirm.

[davidhewitt]

Building with gil-refs disabled is approximately 90 gazillion warnings, and that's going to take some time :-)

And we've probably only implemented 90% of the deprecation warnings, so maybe we can hit 100 gazillion warnings with 0.21 final 🙃

[alex]

:D The good news is that it should be possible for us to tackle them incrementally. Once 0.21.0 is out, we'll do an meta issue with each impacted file, and we'll just burn them down one by one.

[reaperhulk]

When I tested on Friday we had 354 warnings without gil-refs (after fixing the initial 18). Frankly that's less than I would have expected, although it sounds like we'll get a few more soon 😄

[davidhewitt]

Great, CI looks happy, and the release is live! Thanks again everybody; I'll keep pressing ahead with polish & testing. Looking forward to baking the final release soon!

[ritchie46]

@ritchie46 maybe you would be interested from the Polaris POV?

Yes, we are definitely interested. Once ready we will do the refactor and keep you posted. :) Great work!