Session daemon quitting

[Havoc Pennington]

On Sat, 2004-04-10 at 12:01, Thomas Leonard wrote:
> 
> Yes, but what happens if dbus-launch is killed (OOM, etc)?
> 

It could trivially be made to handle OOM I think. I'm not sure I care
about users doing "kill -9" on the babysitter - there are lots of things
you can manually kill that break the system, and it's your own dumb
fault ;-) you can manually kill the bus daemon afterward too, no?

Are there other situations where the babysitter breaks? Would like to
know about that independently from what you're doing here.

> As I say, it generally works, but it's not robust, which is what worries
> me. Checking stdin seems fairly fool-proof, since ROX-Session will quit if
> it loses its X connection or dies in any other way, and then dbus would
> quit when it lost its pipe from ROX-Session. It also saves one process and
> one X connection.

dbus-launch babysitter does look at stdin already... 

I guess you might be saying, replace the kill(daemon_pid) in the
babysitter with a pipe to the daemon. So instead of killing the daemon
with kill(), the babysitter just closes the pipe. We could add a
--with-kill-pipe=FD option to the daemon I suppose and both the
dbus-launch babysitter and ROX session could use it.

Havoc