Comment 4 for bug 789389

Attaching a patch against lightdm-0.3.7-0ubuntu2 that implements the third idea above. When lightdm receives a TERM signal, right before ending the mainloop it calls a cleanup function that just calls g_hash_table_remove_all on the processes table. This works on its own, but in rare instances I was able to trigger a SIGSEGV with from_child_cb still being called on a finalized child process. To fix this, the patch saves the event source ids for g_io_add_watch and g_child_watch_add and removes them when the child process is finalized.

This works for service start/stop but restart still isn't working. This seems to be because the lightdm process is restarted so quickly that the previous X server instance hasn't had time to completely exit, so the new X server instance exits with an error (that is what the log seems to indicate:

[+0.00s] DEBUG: Logging to /var/log/lightdm/lightdm.log
[+0.00s] DEBUG: Starting Light Display Manager 0.3.7, PID=6574
[+0.00s] DEBUG: Loaded configuration from /etc/lightdm.conf
[+0.00s] DEBUG: Loading display default-display
[+0.00s] DEBUG: Logging to /var/log/lightdm/:0.log
[+0.00s] DEBUG: Starting on /dev/tty7
[+0.01s] DEBUG: Launching X Server
[+0.01s] DEBUG: Launching process 6578: /usr/bin/X :0 -auth /var/run/lightdm/authority/0 -nolisten tcp vt7
[+0.01s] DEBUG: Waiting for signal from X server :0
[+0.01s] DEBUG: Process 6578 exited with return value 1
[+0.01s] WARNING: X server exited with value 1)

Not really related to the issue in this bug, but I thought I would mention it here since I stumbled on it, and it is partially related.