I think this also applies to AsyncGenerator

Diff from mypy_primer, showing the effect of this PR on open source code:

tornado (https://github.com/tornadoweb/tornado)
- tornado/gen.py:429: error: Incompatible return value type (got "WaitIterator", expected "AsyncIterator[Any]")
- tornado/gen.py:429: note: Following member(s) of "WaitIterator" have conflicts:
- tornado/gen.py:429: note:     Expected:
- tornado/gen.py:429: note:         def __anext__(self) -> Coroutine[Any, Any, Any]
- tornado/gen.py:429: note:     Got:
- tornado/gen.py:429: note:         def __anext__(self) -> Future[Any]

boostedblob (https://github.com/hauntsaninja/boostedblob)
+ boostedblob/boost.py:547: error: Need type annotation for "task"
+ boostedblob/boost.py:547: error: Argument 1 to "create_task" has incompatible type "Awaitable[T]"; expected "Union[Generator[Any, None, <nothing>], Coroutine[Any, Any, <nothing>]]"

Yep they're all specified as returning Awaitable:

https://peps.python.org/pep-0525/#asynchronous-generator-object

I think this PR needs reverting #7105

I think this PR needs reverting #7105

Yes, I think that's probably right. stubtest is complaining because these are all async def functions at runtime (but wouldn't be in the stub with this PR merged), but the return type for abstract interfaces should be as broad as possible. I've filed python/mypy#12343 to relax that stubtest check.

It would be nice if "real" async def agenfn(...): ... yield had a stricter type which did have Coroutine asend/aclose

It would be nice if "real" async def agenfn(...): ... yield had a stricter type which did have Coroutine asend/aclose

Not sure I understand, sorry

I've proposed a slightly larger PR here: #7491. Would appreciate your thoughts!

It would be nice if "real" async def agenfn(...): ... yield had a stricter type which did have Coroutine asend/aclose

Not sure I understand, sorry

Oh this already exists as types.AsyncGeneratorType

Closing in favour of #7491