[mingley]

What

• H2 client request future stays pending forever.
• This happens when the connection closes during send_request.
• Ran into this with a bidi stream hanging for 2h.
• Same race as hyper-util legacy client: an HTTP/1 request can hang forever when the connection closes while the request is being queued #4202

Fix

• I'm not creative, so I basically copied what fix(http1): drain client dispatch receiver on connection shutdown #4222 did.
• ClientTask drains before it resolves.
• Also removed the h1 only gate on Receiver::close and poll_close.

Testing

• cargo test,clippy,fmt

• local reproduction with a 300ms sleep between reserve and publish in Tokio's send.

  • Before: hangs
  • After: Canceled with "connection closed".

• The description, docs, and comments (words meant for humans) are written by a human, not by an LLM. (https://seanmonstar.com/micro/20260729-i-want-your-own-words/)

[dswij]

Why are we removing this?

[mingley]

#4222 added poll_close while fixing the same h1 race condition from #4202.

It was gated to h1 because h1 was it's only caller.

this PR does the same race fix for h2.

hyper/src/proto/h2/client.rs

Line 685 in a319ec5

ready!(self.req_rx.poll_close(cx));

ClientTask calls poll_close, and poll_close calls close.

If we keep these h1 gates, an "h2 only" build fails.