The issue that I'm seeing on different platforms is that clients are being terminated
forcibly, and this Jira issue is how we should handle session restart/session shutdown taking clients into consideration.

The way we "rigged" up the start-up/shutdown of the session seems to be bring down forcibly the clients, as their wayland connections gets severed abruptly.

a) any clients started by applaunchd should be terminated by waiting for all of them (guess with a timeout)

b) any user units started by systemd should the terminated/waiting for it (same with a time out)

c) finally, when a) and b) finished, we can shutdown the compositor.

If homescreen service depends on compositor to start, when taking down the session we should not depend on compositor to be alive.

But I don't know how easy would be "engineer" this bit because it seems that this reverse start-up/shutting down might be contradictory:

• at start-up: A depends on B
• at shutdown: to stop B, we need to stop A, but if A depends on B

/cc wminer, scottm is this a thing we should "worry" about?