Simon McVittie wrote: > On Wed, 27 Jul 2016 at 23:48:42 -0700, Josh Triplett wrote: > > I can think of another reason that this might change: if we introduce an > > (experimental, and eventually non-experimental) variant based on a > > future stable version of kdbus. > > The usual difficulties of depending on kernel features aside, I would hope > that that would just be dbus-user-session, operating in a faster mode. > > There are two main reasons why dbus-user-session isn't just part of dbus: > > * it changes the semantics of the session bus, from per-login-session to > per-user-session > * it depends on systemd > > I haven't seen any proposals for kdbus that don't share both of those > properties, and the version that was prototyped in systemd using an > out-of-tree kernel module behaves a lot like dbus-user-session.
kdbus doesn't inherently imply the per-user-session semantics, and tying those two concepts together doesn't seem inherent. The kdbus transport shouldn't care which busses userspace creates or what semantics userspace associates with them. Note that the planned rewrite of kdbus (which prompted withdrawing the previous version and no longer pushing for its inclusion) will move more of the dbus-daemon compatibility glue from the kernel into userspace. Also, I can imagine wanting to choose between a kdbus-based session bus or a traditional session bus, especially while still experimenting with kdbus. So that might not want to live in the dbus-user-session package. > > If we need to go through this transition anyway, could we please go > > ahead and introduce a dbus-session-bus package for other packages to > > depend on, to allow for potential future transitions or experiments? > > Do you think that's a worthwhile use of a Packages entry, even if > we come to the conclusion that only desktop environments are expected > to depend on it, and random apps that rely on it (like Empathy) are just > expected to assume its presence? (See my reply to Laurent) Yes, I still do even in that case. But I also don't think it's reasonable for applications that expect a session bus to just assume that the desktop environment will supply one; applications should work in any desktop environment, even none at all. - Josh Triplett
