Okay, this is looking decent yup!

I happened to inroduce a run_until_first_complete since I was looking at a similar pattern for working with websockets.

I reckon we ought to use that here, and have .listen_for_disconnect and .stream_response_body methods. That'd cover us in any wierd broken cases where disconnect is never sent for some reason.

@tomchristie Sure, I'll adjust the code this evening.

Just to be clear, you meant something like this:

await run_until_first_complete(self.stream_response(send), self.listen_for_disconnect(receive))

Right?

@Kamforka Yup, I'll be along these lines...

await run_until_first_complete(
    (self.stream_response, {"send": send}),
    (self.listen_for_disconnect, {"receive": receive})
)

@tomchristie added the changes.

However I have one observation. Is there any particular reason for the signature of run_until_first_complete?

async def run_until_first_complete(*args: typing.Tuple[typing.Callable, dict]) -> None

I think the following signature would be more intuitive:

async def run_until_first_complete(tasks: typing.Sequence[typing.Awaitable]) -> None:

Note: typing.Awaitable might not be the best type here but you get my point.

So in our particular case one could call it like:

await run_until_first_complete(
    [self.stream_response(send), self.listen_for_disconnect(receive)]
)

As far as I can tell the functionality is the same, however we get less boilerplate with better typing and auto-completion experience. What do you think?

Good suggestion yup, although I’m also keeping half an eye here on trio compatibility, which it wouldn’t play so well with.

What do you mean by trio compatibility?

What do you mean by trio compatibility?

We might want a future version of Starlette to support being run on Trio https://trio.readthedocs.io/en/stable/ as an alternative to Asyncio. The API that Trio uses for starting tasks doesn't allow passing instantiated coroutines (for some good reasons) so keeping this constraint in our own codebase would make it easier for us support Trio at some point.

And, great stuff, thank you!

What do you mean by trio compatibility?

We might want a future version of Starlette to support being run on Trio https://trio.readthedocs.io/en/stable/ as an alternative to Asyncio. The API that Trio uses for starting tasks doesn't allow passing instantiated coroutines (for some good reasons) so keeping this constraint in our own codebase would make it easier for us support Trio at some point.

Got it, thanks for the clarification!