AsPyRef is stripped down, not in changelog

[birkenfeld]

The AsPyRef trait was stripped down considerably; methods with, with_mut, into_py, into_mut_py are gone.

The changelog should mention how to rewrite code using them, or if removal was an oversight they should be reinstated.

[kngwyu]

Thank you.
Now I'm not sure why they exist so need to check them.
But anyway I will fix it until the non-alpha release.

[birkenfeld]

I was using them to access instances of a Rust object that were behind a Py<T> pointer.

If there is better or newer way, let me know :)

[birkenfeld]

Any news on this?

[davidhewitt]

There's a soundness problem; because Py<T> is a shared pointer, we need to be careful about getting &mut T from a Py<T>.

I think we really need to have RefCell-like semantics, where Py<T> will have .borrow(), .borrow_mut(), .try_borrow() and .try_borrow_mut() methods.

[kngwyu]

Just a reference for readers: https://docs.rs/pyo3/0.8.5/pyo3/trait.AsPyRef.html
@birkenfeld
I'm OK to retrieve with and with_mut. Are they sufficient?
into_py has a confusing name and I think it should be under another trait.

@davidhewitt
Py<T> can handle all Python pointers(including PyDict or so) and I'd rather enhance PyClassShell in that direction.

[birkenfeld]

@davidhewitt you're right; this cannot be allowed like it was. A RefCell like functionality with runtime checking would be very nice.

@kngwyu with is probably also served by as_ref? with_mut needs a safer alternative, "somehow" :)

[kngwyu]

Now AsPyRef::as_ref is changed to return &PyCell.
For PyCell, please see the guide.

[birkenfeld]

Great, thanks!