Why doesn't `asyncio.as_completed` cancel unfinished tasks if it timeout?

[NewUserHa]

Feature or enhancement

asyncio.as_completed should cancel unfinished tasks if it timeout, or even maybe have the ability to cancel no needed task

Pitch

cpython/Lib/asyncio/tasks.py

Lines 588 to 609 in 0563be2

	def _on_timeout():
	for f in todo:
	f.remove_done_callback(_on_completion)
	done.put_nowait(None) # Queue a dummy value for _wait_for_one().
	todo.clear() # Can't do todo.remove(f) in the loop.
	def _on_completion(f):
	if not todo:
	return # _on_timeout() was here first.
	todo.remove(f)
	done.put_nowait(f)
	if not todo and timeout_handle is not None:
	timeout_handle.cancel()
	async def _wait_for_one():
	f = await done.get()
	if f is None:
	# Dummy value from _on_timeout().
	raise exceptions.TimeoutError
	return f.result() # May raise f.exception().

Why doesn't asyncio.as_completed cancel unfinished tasks if it timeout?
Maybe say there may be many network requests pending, since that already raised exception then future requests should probably be canceled to save resources.
Probably because the concurrent.futures.as_complete didn't cancel unfinished tasks? But it's local resources than asyncio is usually network resources.

More, I know the asyncio.wait() can do cancel the pending job. But asyncio.as_complete() probably should also be able to cancel the next coros if got the wanted result. But currently, the returned is a coroutine that can't cancel.

[gvanrossum]

This is indeed unfortunate, but we cannot change it without breaking backwards compatibility. I recommend trying to work around it using task groups.

[NewUserHa]

But that can definitely waste unnecessary network resources if users input many awaitables to the asyncio.as_completed() isn't it?

Should the document mention that situation for the users who never read the source code? Or recommend using 'task group' in the document of asyncio.as_completed() as well? (currently, only asyncio.gather() recommend using 'task group')

[gvanrossum]

I agree that we could add a note to the docs for as_completed. If we wanted to recommend task groups we'd at least have to give a code sample showing how to do that -- it's not completely trivial.