Jean Hominal

I think that having a separate capacity limiter makes a lot of sense - if only so that starlette can handle without having its capacity being swallowed by tasks from...

Pushing such a depth of choice to the users of starlette is not a real solution either - it would make the API a lot more fiddly, for what I...

I do not know what @graingert would say, but I would also opine that, on the balance of it, restoring contextvars is not a game that we should play, as...

> I think the issue is that for a FastAPI user this _is_ downstream: > > ```python > def dep1() -> None: > ctx.set("foo") > > def dep2(d1: None =...

As the reporter on #1438, I would say two things about this PR: 1. I would be quite unhappy with a solution that relied on checking the current task name...

@florimondmanca Thank you for your reply! I am going to work on the test cases - at least a few of them will be similar to existing issues/pull requests. But...

I will just comment on the two issues that I believe are still open and would be directly affected by this PR: * #1438 would be fixed by this PR,...

@adriangb: I have looked at your test cases while implementing mine. However, after porting and reviewing a copy of the #919 test, I ~~saw~~ thought that the only substantial difference...

> Hmm good point, looking at my test now I think that may be the case. I'll dig deeper tomorrow, thank you for combing through them. I just reread your...

I have just added a test to check for the issue reported by @kissgyorgy on https://github.com/encode/starlette/issues/1678#issuecomment-1172916042 - about the way that `BaseHTTPMiddleware` cancellation prevents the context manager defined in the...