Context

The handling of socket.timeout errors in receive_message currently looks like this:

    if deadline is not None:
        short_timeout = min(max(deadline - time.monotonic(), 0), _POLL_TIMEOUT)
    else:
        short_timeout = _POLL_TIMEOUT
    conn.set_conn_timeout(short_timeout)
chunk_length = conn.conn.recv_into(mv[bytes_read:])
...
except socket.timeout:
    if conn.cancel_context.cancelled:
        raise _OperationCancelled("operation cancelled") from None
    if _PYPY:
        # We reached the true deadline.
        raise
    continue
...

This results in the driver attempting to read the socket an additional time once deadline - time.monotonic() is less than zero, meaning the deadline has already been reached. This can result in test failures such as PYTHON-4987, where the additional read is performed after an expected timeout instead of the timeout being raised as expected.

Definition of done

The driver should re-raise socket.timeout errors if the deadline has already been exceeded.

Pitfalls

N/A,