remove internal APIs from pyo3-ffi

[Cheukting]

Using methods from PyAny to replace calling _Py internal APIS from pyo3-ffi.

see #3762

[alex]

Looks like CI is failing due to a warning from an unused macro.

[Cheukting]

Sure, this PR is still WIP. I will work on it more soon.

[davidhewitt]

Thanks very much for working on this! I think there's a little bit more we can do as part of this one:

• I see a use of _Py_c_neg around line 154 which we can update
• Also a use of _Py_c_abs around line 288 to update
• Let's add newsfragments/4201.changed.md to note that the implementation has changed here even if the API remained the same, as there will be a performance impact.

I also wonder if we should add a TODO here or in a follow-up issue to discuss whether in the long term these operations should be removed? Given that CPython doesn't guarantee these methods won't fail I somewhat wonder if we should peel these away and encourage users to fall back to the PyAnyMethods functions. Panics are never nice.

For a first step though, it's very good for us to get off the CPython private APIs!

[Cheukting]

@davidhewitt happy to do it either here or open a new PR for the suggestion

[davidhewitt]

Given this PR is still relatively self-contained, I'd be happy for the additional changes to be pushed here 👍

[davidhewitt]

Looking good, thanks, just a final thought (which might create some more ideas... 👀)

The idea to add to PyAnyMethods - I'm fine for it to be either part of this PR, a separate PR which we merge first, or a follow-up.

[davidhewitt]

Super, this looks great, thanks!

Just a couple of final suggestions to tidy up the PyAnyMethods additions, and then let's merge 🚀

[davidhewitt]

Let's also add 4102.added.md please with something like "Added PyAnyMethods::{abs, pos, neg}."

[Cheukting]

Sure!

[davidhewitt]

Here I think the Python equivalent is abs(self).

[davidhewitt]

Thanks, this looks good to me! Really pleased to get this dependency on the private API removed 👍

[adamreichold]

@Cheukting Could you have a look into the build error on GraalPy from our CI? If you'd like, we can also label this PR so the cull CI suite is run on every push so you do not have to wait for the merge queue to validate your changes?

[Cheukting]

@adamreichold it is due to an unused import it seems: https://github.com/PyO3/pyo3/actions/runs/9336538447/job/25697227447#step:10:132

I will have a look into it, thanks!

[Cheukting]

@adamreichold do you mind running the check on the merge queue again to see if my fix works?