Improved closure constructors

[davidhewitt]

Somewhat inspired by the work going on in #5689.

The current capsule constructors PyCapsule::new and PyCapsule::new_with_destructor have some downsides:

• They unconditionally allocate the CapsuleContents type.
• They require an allocated CString for the name, if set (the full name type is Option<CString>)

After we started reworking the capsule methods in 0.27 to require checking a name, I think it would make sense to have a name be required as &'static CStr, to further encourage name checking. Also, if we no longer need to store a CString within the CapsuleContents, for the case of a trivial destructor we don't need to have CapusleContents at all (the only thing to store is T).

This would allow us to store the value as Box<T>, or maybe even store the value without boxing as long as T is no more than pointer sized.

While some experimentation to be had, I'd propose we introduce new constructors for now as

impl PyCapsule {
    pub fn new_with_name<T: 'static + Send + AssertNotZeroSized>(
        py: Python<'_>,
        value: T,
        name: &'static CStr,
    ) -> PyResult<Bound<'_, Self>>;

    pub fn new_with_name_and_destructor<
        T: 'static + Send + AssertNotZeroSized,
        F: FnOnce(T, *mut c_void) + Send,
    >(
        py: Python<'py>,
        value: T,
        name: &'static CStr,
        destructor: F,
    ) -> PyResult<Bound<'_, Self>>;

... and maybe deprecate the existing constructors.

cc @lmmx

[lmmx]

• Also backlinking [Idea] PyCapsule instantiators for raw pointer capsules #5688 here for reference 🔗

This would allow us to store the value as Box<T>, or maybe even store the value without boxing as long as T is no more than pointer sized.

Neat ! 👍

[davidhewitt]

It looks like #5689 will merge and adds new_with_pointer and `new_with_pointer_and_destructor.

Rather than new_with_name, I think maybe the functions proposed above could be new_with_value and new_with_value_and_destructor.