Tanu Kaskinen wrote on 27/10/14 08:51: > On Sun, 2014-10-26 at 19:55 +0200, Tanu Kaskinen wrote: >> On Fri, 2014-10-24 at 13:02 +0100, Colin Guthrie wrote: >>> Tanu Kaskinen wrote on 24/10/14 12:08: >>>> On Fri, 2014-10-24 at 11:14 +0100, Colin Guthrie wrote: >>>>> Tanu Kaskinen wrote on 24/10/14 10:42: >>>>>> "Disable the manual override" doesn't sound like a good idea... Does >>>>>> this mean that "systemctl --user disable pulseaudio.socket" doesn't >>>>>> work? >>>>> >>>>> Yes, but it also has the advantage that every single user on the system >>>>> doesn't have to run "systemctl --user enable pulseaudio.socket" before >>>>> their sound will work. If distros ship this, they will definitely ship >>>>> this symlink or something similar to it, so I think doing this by >>>>> default makes sense for us too. >>>> >>>> I expect distros to enable the service only on first install, not on >>>> package updates. >>> >>> I expect distros to completely forget this step and then complain - >>> especially so because this is a user socket unit, not a system on and >>> the packaging guidelines for such things are still in their relative >>> infancy. >>> >>> But I'm not strongly against it, but I just don't like the fact that >>> this enabling, if done in packaging, would hence be in /etc tree and >>> thus also suffer from a "factory reset" unless corresponding tmpfiles >>> rules were also put in place to recreate the symlink. >> >> Good point, enabling the service under /etc is not optimal either. In >> the other thread you were convinced that we shouldn't install the >> default.target.wants link, but I'm starting to be convinced that we >> should :) I'm still annoyed by the fact that users can't use "systemctl >> --user disable pulseaudio", but there doesn't seem to be any better way >> for upstreams/distros to enable services by default (this seems very odd >> to me, since all other configuration in systemd can be cleanly >> overridden at multiple levels). > > Above I said "default.target.wants link", but I actually meant the > sockets.target.wants link. Sorry this caused any confusion. Hehe! Yeah, don't worry, I've done that in some of my emails too - they maybe didn't last a readthrough before sending, but it's easy to get confused here :) Feel free to ping me on IRC and we can thrash out the best option if you like? Both leaving it to packagers as a "systemctl --global enable" option and doing a hard-coded /usr/lib symlink on install have benefits. I could potentially write up some wiki based documentation about it from a packagers perspective, regardless of what option we go for in the end. As either approach will have drawbacks, it will be nice to document it properly (and I'm not 100% sure it's right for the "Running" page which is more aimed at users... although does go a bit more low level than is desirable too at times! Col -- Colin Guthrie http://colin.guthr.ie/
