Handle async cancelled error explicitly

[T-256]

Summary

Discussed in #805

Checklist

• I understand that this PR may be closed in case there was no previous discussion. (This doesn't apply to typos!)
• 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.

[Tunglies]

We'd need to change in both httpcore/_async and httpcore/_sync and ensure them match.

[T-256]

We'd need to change in both httpcore/_async and httpcore/_sync and ensure them match.

As what's said in #806, I think this could be ignored in unasync.py to avoid extra unuseful load. (Task behaviors not defined on threads.)
But still I'm waiting to get proper response from team to choose final implementation for sync part.

[karpetrosyan]

Is this solution respects such cases?

because the user may catch the keyboardinterrupt in his code and continue to use httpcore.ConnectionPool.

[T-256]

Is this solution respects such cases?

No, #812 will respect.

Actually this PR can expose why current cleanup system is not well implemented.

[karpetrosyan]

No, #812 will respect.

Do we want to marge this PR when we know it will break programs that handle BaseExceptions like KeyboardInterrupt?

[T-256]

Do we want to marge this PR when we know it will break programs that handle BaseExceptions like KeyboardInterrupt?

Yes. because still some of cleanups doesn't handle BaseException in codebase:

• httpcore/httpcore/_async/connection.py

  Lines 97 to 100 in 94ffb33

  	exceptExceptionasexc:
  	self._connect_failed=True
  	raiseexc
  	elifnotself._connection.is_available():

• httpcore/httpcore/_async/http2.py

  Lines 441 to 453 in 94ffb33

  	exceptExceptionasexc:
  	# If we get a network error we should:
  	#
  	# 1. Save the exception and just raise it immediately on any future reads.
  	# (For example, this means that a single read timeout or disconnect will
  	# immediately close all pending streams. Without requiring multiple
  	# sequential timeouts.)
  	# 2. Mark the connection as errored, so that we don't accept any other
  	# incoming requests.
  	self._read_exception=exc
  	self._connection_error=True
  	raiseexc

• httpcore/httpcore/_async/http2.py

  Lines 470 to 482 in 94ffb33

  	exceptExceptionasexc: # pragma: nocover
  	# If we get a network error we should:
  	#
  	# 1. Save the exception and just raise it immediately on any future write.
  	# (For example, this means that a single write timeout or disconnect will
  	# immediately close all pending streams. Without requiring multiple
  	# sequential timeouts.)
  	# 2. Mark the connection as errored, so that we don't accept any other
  	# incoming requests.
  	self._write_exception=exc
  	self._connection_error=True
  	raiseexc

• httpcore/httpcore/_async/socks_proxy.py

  Lines 294 to 297 in 94ffb33

  	exceptExceptionasexc:
  	self._connect_failed=True
  	raiseexc
  	elifnotself._connection.is_available(): # pragma: nocover

• httpcore/httpcore/_backends/anyio.py

  Lines 76 to 79 in 94ffb33

  	exceptExceptionasexc: # pragma: nocover
  	awaitself.aclose()
  	raiseexc
  	returnAnyIOStream(ssl_stream)

• httpcore/httpcore/_backends/trio.py

  Lines 78 to 81 in 94ffb33

  	exceptExceptionasexc: # pragma: nocover
  	awaitself.aclose()
  	raiseexc
  	returnTrioStream(ssl_stream)

• plus, same things in _sync part.

And also notice few BaseException was introduced in #726, so before that we were not supporting KeyboardInterrupt:

• First, Before: no error handling, After: BaseException handling.
• Second, Before: Exception handling, After: BaseException handling.

[T-256]

Should we consider convert exception handlings in #811 (comment) to EXCEPTION_OR_CANCELLED?

[stale]

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.