Re: GDM login to a JHBuild session

[Brian Cameron <Brian Cameron Sun COM>]

Owen:

> > We can extend the .desktop file further if this makes it possible to
> > do useful things.  However, if we are going to ask all distros to
> > change the way they start dbus-launch to support DISABLE_DBUS_LAUNCH,
> > then perhaps we should rethink how we start D-Bus.  Perhaps starting
> > this at Xsession time isn't the right place?  Perhaps we should
> > always start dbus-launch in the /usr/share/xsession/*.desktop files
> > as needed?  Or perhaps it should be managed more directly by
> > gnome-session itself rather than needing the login program/process
> > to be aware of D-Bus?
>
> While there are certainly other ways of doing it, having dbus-launch
> be the parent process of the session, ssh-agent style, is a robust way
> of doing things, and I don't see a big argument for changing that and
> forcing everybody to adapt (especially since the dbus for the session
> makes sense in simplified contexts where they may not be a chunk of
> code like gnome-session that could substitute for dbus-launch.)
>
> Similarly, while moving dbus-launch into the desktop file would work,
> that would require adapting all session desktop files in every
> distribution in a way that would break (through double dbus-launches)
> if the Xsession and the session files got out of sync.
>
> My proposal was meant to be surgical - to just address the problem of
> the exceptional case where you didn't want the system D-BUS service -
> rather than to try and redesign things.

I understand, though the temptation to add more "surgical" fixes to
GDM makes it harder to maintain.  I also don't like the fact that your
suggested fix requires that the distro support the feature in their
Xsession script.  Some distros probably won't fix this, and thus
causing confusion and user's thinking GDM is broken.

GDM already has too many surgical workarounds that are only used in
odd corner-cases, and these tend to break on occasion.  I'm not opposed
to adding more if someone wants to provide a patch, but if there is a
way to make things work without hacking GDM further, that would be
better, I think.

Could dbus-launch be made smarter so that if dbus-launch was already
started earlier in the stack, it does something smarter and "just
works"?  If possible, this would be better than adding a hack to GDM
to support a flag that requires the the distro Xsession script support
it.

Brian