[devanshc777]

On Windows, if the peer resets the connection after sending FIN, sock.shutdown(SHUT_RDWR) in _ProactorBasePipeTransport._call_connection_lost() raises ConnectionResetError (WinError 10054). The rest of the cleanup is skipped: the socket stays open, the transport is never detached from its server, and Server.wait_closed() never returns.

The connection is already gone at that point, so the reset is ignored, the same way _loop_reading() and _loop_writing() already treat ConnectionResetError as a normal end of the connection. Other shutdown() errors propagate as before.

Reproduced with real sockets on Windows 11 against a debug build of main: the server writes a response, the client half-closes without reading it, then closes, which sends a RST. Without the fix, 10 of 10 runs left the socket open (the debug build reports it as an unclosed-socket ResourceWarning) and wait_closed() hung; with the fix, 10 of 10 cleaned up. The added test uses a mocked socket, like test_call_connection_lost. test_asyncio passes on the same build.

gh-160600 was an earlier draft for this issue, closed by its author before review.

• Issue: asyncio: Windows socket reset during Proactor cleanup skips socket closure and transport detachment #160429

[python-cla-bot]

All commit authors signed the Contributor License Agreement.