Ensure HTTP/2 connections are gracefully closed on async cancellation

[karpetrosyan]

Closes #756

Checklist

• [x] I understand that this PR may be closed in case there was no previous discussion. (This doesn't apply to typos!)
• [x] I've added a test for each change that was introduced, and I tried as much as possible to make a single atomic change.
• [ ] I've updated the documentation accordingly.

TODO

• [x] Add failing tests
• [x] Gracefully handle cancellation for unstarted http/2 requests
• [x] Update changelog

[zanieb]

could (async )with self._state_lock: move under _state_housekeeping function?

I agree, it's always used.

[lovelydinosaur]

At thanks! The failing test case is especially helpful here.

Pretty sure this resolution is overcomplicated, tho. I think we actually want here is?...

1. This existing try...except needs to be expanded...

https://github.com/encode/httpcore/blob/21e47a7481fbad3719417899113e1435a340b752/httpcore/_async/http2.py#L106-L113

...so that it instead covers all of this block...

https://github.com/encode/httpcore/blob/21e47a7481fbad3719417899113e1435a340b752/httpcore/_async/http2.py#L106-L127

ie. the except case to move down to line 127 there.

2. This call to acquire()...

https://github.com/encode/httpcore/blob/21e47a7481fbad3719417899113e1435a340b752/httpcore/_async/http2.py#L129

Should also be surrounded with...

try:
    await self._max_streams_semaphore.acquire()
 except BaseException as exc: 
     with AsyncShieldCancellation(): 
         await self.aclose() 
     raise exc

We could actually avoid (1) if we instead addressed the refactoring suggested in https://github.com/encode/httpcore/discussions/738 or refactored self._send_connection_init(**kwargs) so that our initial handshake waits for the settings response before we instantiate our max_streams_semaphore. But let's not be getting into that just now.

[lovelydinosaur]

Ugh. I need to think about this one more clearly, it's making my brain mush.

[karpetrosyan]

I would recommend using this stupid solution for the time being and not waiting for http/2 refactoring from #770

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.