Consider removing the blanket `FromPyObject for T: PyClass + Clone` ?

[davidhewitt]

I wonder, should we remove the FromPyObject implementation for T: PyClass + Clone while we're making all the changes in 0.27? (Orginally introduced in #730)

I feel like there are several drawbacks to having that implementation:

• it makes the error messages for .extract() always error that X: PyClass is not satisfied due to the blanket (maybe we can fix this with a better diagnostic::on_unimplemented for FromPyObject)
• it makes it easy for users to accidentally clone their data when receiving it as an argument from Python
• it makes it impossible to add users' own FromPyObject implementation for #[pyclass] types which implement Clone, due to the blanket. (Unable to implement FromPyObject for #[pyclass] with Clone #4337)

While we're breaking FromPyObject, should we consider removing this implementation?

I can also see an argument for keeping the blanket and instead just documenting it better to mitigate the downsides, to some users it may be convenient and it would be more breakage to remove it.

[Icxolu]

Maybe we can try to move this to macro code and provide an opt-out (and possibly switch to an opt-in at some later point)?

[Tpt]

This blanket impl is quite convenient to be able to do things like:

#[derive(FromPyObject)]
enum AOrB {
    A(A),
    B(B)
}

and use it in function arguments.

What about making it an opt-in using #[pyclass(from_py_object)] similar to #[pyclass(eq)]? I am a bit afraid that making it opt-out then opt-in will create more churn than making it opt-in from the start.

[Icxolu]

The idea with the opt-out was that that everything kind of keeps working as is, so there should be essentially no churn from that. The switch to opt-in (if we were to choose it) could then potentially more graceful, by guiding users with deprecation messages. I haven't thought this completely through or checked what would be possible, but this was what I had in mind.

[davidhewitt]

Implemented in #5550, will close this issue.