Fedora sound nightmare (Pulseaudio)

[Lennart Poettering]

On Thu, 13.03.08 12:19, Callum Lerwick (seg at haxxed.com) wrote:

> > It seems that sometimes on reboot, pulse gets confused and thinks
> > there already is a copy of the daemon running, and so doesn't start
> > one.
> > 
> > If you have the "sometimes pulse works and sometimes it doesn't on
> > reboot" situation, you may want to jump on the above BZ.  [There is a
> > hack described there that seems to paper over the issue for me.]
> 
> I've run in to this myself. Pulse dies or hangs for some reason, leaving
> the lockfile. Then it refuses to start again. Drove me nuts, it took an
> strace run to figure out the lockfile was in /tmp, and not ~/. This is
> not documented in any obvious place. And there's no command line option
> to force start or force cleanup. Epic fail.
> 
> Don't we have *libraries* that get this lockfile stuff right? Seems like
> something rather fundamental to unix programming. And yet people still
> manage to do it wrong.

The problem here is that /var/run is usually cleaned up during bootup,
so it's usually fine to just do a kill(pid, 0) to detect whether a
daemon is already running. However, PA runs as user daemon and thus
cannot write to /var/run. I thus chose /tmp/ copying X and esd a
bit. Which is admittedly a bad idea, and also introduces a (minor)
security issue.

Recent version of PA check the proc name of the PID they read from the
file before thinking that it is already running. The bug is fixed.

Lennart

-- 
Lennart Poettering                        Red Hat, Inc.
lennart [at] poettering [dot] net         ICQ# 11060553
http://0pointer.net/lennart/           GnuPG 0x1A015CC4