set_item consumes, but uses ToPyObject instead of IntoPyObject

[m-ou-se]

PyDict::set_item consumes the key and value, but uses ToPyObject. Shouldn't that be IntoPyObject?

For ToPyObject, taking a reference would be sufficient. But it seems more efficient to consume and use IntoPyObject.

(Same for PyList::set_item, etc.)

[PetrGlad]

Also set_item borrows mutates the dict. Should it borrow mutable?

[m-ou-se]

@PetrGlad No, that's unreleated, and on purpose. Python objects can always have multiple owners, so Rust's mut is not used for mutability in PyO3: See https://docs.rs/pyo3/0.7.0/pyo3/#ownership-and-lifetimes

[davidhewitt]

Marking to evaluate this as part of possible collections improvements for 0.15.

[aviramha]

@davidhewitt What would be the change here? I played a bit (tried changing ToPyObject -> IntoPy) and got into some errors. Is this supposed to be a tricky thing to solve?

[davidhewitt]

Changing to IntoPy sounds like the right idea, but as it's breaking I have basically been sitting on this as I'm not sure how much it really helps. The semantics of IntoPy and ToPyObject are a little blurred...

[davidhewitt]

Finally solved by the new IntoPyObject trait in #4493