Re: [Utopia] Debian/Ubuntu bug fixes for gnome-volume-manager

["John (J5) Palmieri" <johnp redhat com>]

On Tue, 2005-07-05 at 07:54 +0200, Martin Pitt wrote:
> Hi!
> 
> John (J5) Palmieri [2005-07-04 12:56 -0400]:
> > On Mon, 2005-07-04 at 18:05 +0200, Martin Pitt wrote:
> > 
> > > 04_reconnect_on_dbus_exit.patch:
> > >   - Reconnect to dbus after restarting dbus and hal, instead of just
> > >     crashing.
> > >   - Important in package-based system when you want to continue to use
> > >     your Gnome session while doing a system upgrade.
> > 
> > If this can be handled sanely then great.  There has been a conclusion
> > drawn that for a lot of situations this is a hard problem to solve.  It
> > is safer to consider D-Bus being like the kernel.  If you upgrade it you
> > don't get the new version until you reboot (e.g. don't restart the bus
> > on upgrades).  There are issues like you can't restart any running
> > session bus and you don't even know when libraries will sync up so you
> > might end up being kicked off the bus anyway.
> 
> I see the point here. However, in Debian and Ubuntu we usually design
> our packages in a way that they do not require a reboot in any case
> but a kernel upgrade. Of course this is much more important on servers
> to minimize downtime, but it has become a common standard, Policy, and
> people expect it.
> 
> Restarting hal/dbus on package upgrade has worked very well so far,
> too. Our fine grained dependency system ensures that the matching
> libraries are available, and with the reconnection patch we can get an
> uninterrupted g-v-m session in all cases but HAL API changes (which
> does not occurr very often, so that's not a big deal).
> 
> We do not use the session bus for anything right now, but since it
> would just continue to run, it already behaves the way you described.
> 
> Since I see that different distributions want to handle that
> differently, what about making this optional with a ./configure-time
> option?

I'm not objected to handling reconnects.  Just wanted to point out
roadblocks.  If we can do it without creating a mess in the code then it
is a good thing.
 
-- 
John (J5) Palmieri <johnp redhat com>