Currently if the server is killed the _to_server thread exits and closes both _client_conn and _server_conn. However, this does not cause the _to_client thread to stop. As a result an extra "xpra" process stays listed in the process list and users don't notice that the server has died. Only if you try to interactive with any of the windows will the proxy write something to the server socket and notice the problem.
This patch stops _to_client thread when _to_server thread exits and vice versa. Calling _Thread_stop() is bit ugly but the alternative would probably be to use some sort of polling mechanism instead of blocking read() in _copy_loop.
patch to fix the problem
merged in r98
this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/5