pyproto: deprecate protocol traits

[davidhewitt]

The final step before I think we're ready to release 0.16!

[davidhewitt]

@birkenfeld to answer your question in #2159 (comment) - I started this branch last night. If you want to work on it please do take it over and push further commits! I'll get working on the 0.16 release branch instead.

I guess the remaining work here is to go through documentation, generally making #[pyproto] less obvious, and wherever it does exist point users to #[pymethods] instead?

[birkenfeld]

Ok, I'll get on it. BTW, it is sad to see all the #![allow(deprecated)]; I'd have thought that deprecations in the same crate would be allowed, as this will be easy to overlook needing removal later.

[davidhewitt]

Agreed, though I think most of these files can simply be deleted when we finally remove #[pyproto]

[birkenfeld]

So, questions that came up while working on this:

1. Is __getattr__ actually equivalent to Python __getattr__ or rather __getattribute__?
2. In pyproto's __getitem__ etc., negative indices were handled specially. I assume I was correct that this is not done for pymethods?
3. There doesn't seem to be support for __inplace_concat__ and __inplace_repeat__ sequence methods?

[davidhewitt]

1. Yes, the two methods share the one slot, however we have this code which makes the slot behave like __getattr__. I think in principle we could support both methods like CPython.

2. Yes, the summary of the issue is that if the Py_sq_len slot is implemented then CPython will add that length to a negative index before accessing the sequence slots. The mapping slots are always tried first, so in practice this means that only in weird edge cases (from C) would CPython do this adjustment to negative indices. I felt this was hard for users to understand and impossible to correct for, so I chose for our __len__ to not implement the Py_sq_len slot and thus minimize potential confusion.

3. Ooh, good spot. I think I missed because these magic methods don't exist in pure Python; they're just lumped in with __iadd__. We could copy the existing design straight from #[pyproto], seems clear enough?

[davidhewitt]

Pushed a commit to add __inplace_concat__ and __inplace_repeat__ support.

[davidhewitt]

I'll merge this tonight if nobody has any concerns!

[birkenfeld]

Nice, __getattribute__ can then be another PR.