Populate `am_send` slot

[winstxnhdw]

With the introduction of eager_task_factory in Python 3.12, I think it would be a good idea to start looking at populating the am_send slot whenever users implement the send() method.

But since this is a feature that is only available on Python 3.10+, we might have to gate it behind an attribute argument like so.

#[cfg_attr(Py_3_10, pyclass(fast_send))]
#[cfg_attr(not(Py_3_10), pyclass)]
struct Coroutine {
    ...
}

[davidhewitt]

Can you find documentation on the CPython side where implementing am_send is necessary over just .send()? It seems sorely lacking.

I think automatically implementing am_send from send() seems like a scary breaking change, presumably there are lots of cases where users have send() functions which do not conform to the appropriate signature, or relate to sync iterators etc.

But yes, we should probably explore ways to make it possible to fill this slot. There is possibly some overlap with #3949

[winstxnhdw]

Can you find documentation on the CPython side where implementing am_send is necessary over just .send()? It seems sorely lacking.

From what I understand in https://bugs.python.org/issue42085, am_send is meant to be completely compatible with send(), and allows the user to avoid the usual raise StopIteration to return values.

I think automatically implementing am_send from send() seems like a scary breaking change, presumably there are lots of cases where users have send() functions which do not conform to the appropriate signature, or relate to sync iterators etc.

In that case, we should only automatically implement it if the user subclasses Coroutine?

[davidhewitt]

In that case, we should only automatically implement it if the user subclasses Coroutine?

Which Coroutine are you suggesting subclassing?