0.28 Release

[davidhewitt]

We already have the MSRV bump and the changes to use multi-phase-init on main, plus a couple of deprecations.

I guess we'll release this all around the end of the year, subject to exactly when we have a meaningful set of changes to ship.

Some areas that I'm going to focus on:

• type hints (ongoing as usual)
• cleaning up #[new] and __init__ support (somewhat related to the above)
  • Allow #[new] to return a SubClass of Class #5560
  • Support automatic __new__ generation in #[pyclass] with set_all #5416
  • refactor #[new] code generation #5551
  • Implement #[init] method attribute in #[pymethods] #4951
• finish off remove remaining private _Py symbols from pyo3-ffi #5592
• free-threading clean-up, e.g. Disallow races in initializing types #5211
• some items which have come up in recent conversations I've had, which I'd love to get resolved
  • Controlling PyBuffer requests #5245
  • Making docs accessible to Python only and / or Rust only. #5269

Help welcome on all of the above, just let me know what anyone picks up so I know not to race with you. I'm sure there will also be plenty of CI and maintenance work, plus a bunch of optimizations and other bugfixing.

We had a bunch of threads we haven't yet wrapped up, such as how to move forward with PyClassGuard. We probably need to kick off design issues for some of those and work out the next steps.

There's also plenty of bugs and other tasks on the backlog which we can try to burn down too :)

[Icxolu]

I could look into refactoring #4951 and/or #5421 on top of #5551 if you haven't done so already.

[davidhewitt]

Please feel free to take over #4951; I haven't implemented anything for it other than reading it in detail and discovering I wanted to do #5551. Following on from my last comment in that PR I decided it's probably good enough to just proceed to introduce support for __init__.

#5421 - I think the author might still be around to handle it as soon as we've got #5551 merged.

Note that #5551 is itself waiting to rebased after #5586 is merged (and should hopefully make #5551 a much smaller and more reviewable diff).

[Icxolu]

Great, then I'll look into #4951. I must have missed #5586, just left a comment there.

[davidhewitt]

Some suggestions if anyone is looking for relatively self-contained issues which I've wanted to do for ages but just keep failing to prioritize myself:

• Contribution opportunity: use Python 3.12 managed dict and weakref #4253 would be a nice way to drop some of the complexity of our pyclass internals (on newer Python versions) by pushing the contents of those fields into state managed by the Python interpreter.
• PEP 688 buffer methods #3148 would be a great to allow us to replace __getbuffer__ and __releasebuffer__ with the standard forms.

[davidhewitt]

Once #5739 and #5740 have merged, I think we're in a relatively good position to release.

I will take a look again to see if I can help unblock #5416 so it makes it over the line. There are a few other items in the original post and also #5599 (comment) which could land if anyone has a chance, also not a problem if they slip.

[ngoldbaum]

I extracted #5764 from my open PR to fix some bugs in the FFI definitions I added so they don't get released in a buggy state.