better Some-wrapping for default arguments

[davidhewitt]

Closes #3460

This reworks the Some-wrapping of Option<T> default arguments to decide whether to wrap via a trait rather than searching for the text None. Most clear via the accompanying test; on main both b = Some(2) and d = NONE would be incorrectly double-wrapped.

Refactoring: I moved the ok-wrapping trait alongside this new trait for consistency. Perhaps in the future a refactoring will find a way to harmonise them.

[mejrs]

Thanks, this is much better than the text based solution :)

[mejrs]

No it is not, this works today:

pub struct PyErr;
pub type PyResult<T> = Result<T, PyErr>;

pub trait OkWrap<T> {
    type Error;
    fn wrap(self) -> Result<T, Self::Error>;
}

impl<T> OkWrap<T> for T {
    type Error = PyErr;
    fn wrap(self) -> PyResult<T> {
        Ok(self)
    }
}

impl<T, E> OkWrap<T> for Result<T, E> {
    type Error = E;
    fn wrap(self) -> Result<T, Self::Error> {
        self
    }
}

I suspect the reason for the IntoPy<PyObject> bound is to generate a better error message, but I'm not sure.

[davidhewitt]

Hmm I gave that a go just now and get some error messages like the following:

error[E0283]: type annotations needed
  --> src/test_hygiene/pyfunction.rs:4:1
   |
4  | #[crate::pyfunction]
   | ^^^^^^^^^^^^^^^^^^^^ cannot infer type
   |
note: multiple `impl`s satisfying `Result<i32, err::PyErr>: OkWrap<_>` found
  --> src/impl_/wrap.rs:26:1
   |
26 | impl<T> OkWrap<T> for T {
   | ^^^^^^^^^^^^^^^^^^^^^^^
...
33 | impl<T, E> OkWrap<T> for Result<T, E> {
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   = note: this error originates in the attribute macro `crate::pyfunction` (in Nightly builds, run with -Z macro-backtrace for more info)

So it does seem that this bound avoids a conflict. Maybe in the future we can work out a simplification; I'll proceed to merge now to land the SomeWrap change.