This is far from ideal: we end up with 2 extra X11 servers (the Xvfb and Xnest running inside it) and performance is not going to be great. But this is fairly easy to add, so why not?

Server must tell the clients that it supports this mode (==it has Xnest installed and this is enabled in config). Then on app start we run Xnest and use the old port probing code to launch the real app against the Xnest display. (not going to do pre-launch mode for this - not worth it)

When starting a process and watching its output, we shouldn't do this in a separate thread but instead integrate the code with twisted main-loop (gtk/glib's actually) using ProcessProtocol?, see ​here

Also, when waiting for a process by pid (one we may not have launched ourselves), we should use gobject.child_watch_add(pid) which will also integrate with the main loop.

Places of interest:

• 1 ssh_key_util.generate_new_key: simple call_when_done callback
• 2 config_server.do_tunnel_connect_test: simple call_when_done callback (needs output and returncode)
• 3 ssh_command_monitor: needs every line to be parsed as they are received...
• 4 server_link.do_connect_to_server: re-implements wait_for_socket!!!

Should be a pretty straight forward swap. No reason to use yet another network library (raw sockets in this case)