[systemd-devel] [PATCH] Delay dbus connection for user session

[Kok, Auke-jan H]

On Tue, Jun 4, 2013 at 10:31 AM, Lennart Poettering
<lennart at poettering.net> wrote:
> On Wed, 29.05.13 08:59, Yang Chengwei (chengwei.yang at intel.com) wrote:
>
>> > > diff --git a/src/core/manager.c b/src/core/manager.c
>> > > index 0508628..467b5ba 100644
>> > > --- a/src/core/manager.c
>> > > +++ b/src/core/manager.c
>> > > @@ -503,7 +503,7 @@ int manager_new(SystemdRunningAs running_as, Manager **_m) {
>> > >                  goto fail;
>> > >
>> > >          /* Try to connect to the busses, if possible. */
>> > > -        r = bus_init(m, running_as != SYSTEMD_SYSTEM);
>> > > +        r = bus_init(m, running_as != SYSTEMD_SYSTEM && running_as != SYSTEMD_USER);
>> > >          if (r < 0)
>> > >                  goto fail;
>> >
>> > So, I've looked at this code again as I've spotted this problem over a
>> > year ago, and I'm wondering if this entire section should just be
>> > removed, since running_as only has 2 values (_SYSTEM and _USER)
>> > anyway...
>> >
>> > Would there be any objection to just dropping these 4 lines entirely?
>> > Lennart, Kay? I don't see how we'd have a 3rd manager type here.
>>
>> Yes, just in that case. If all of you agreed, I'd like submit a V2 to
>> drop these lines.
>
> Hmm, this would break my usual testing routine, where i run user systemd
> against a bus daemon that is already running... I have no doubt that
> dropping the check entirely is the right way to go eventually, but maybe
> we can find a better way so that this continues to work for me until
> GNOME has switched over to systemd as service manager.

I'm kinda dumb for not thinking this over, but, if you already have a
dbus-daemon running, you have DBUS_SESSION_BUS_ADDRESS set.

So, the whole idea of connection to a session bus if
DBUS_SESSION_BUS_ADDRESS is unset makes no sense whatsoever.

Lennart, would you agree that we could just do something like:

if (getenv("DBUS_SESSION_BUS_ADDRESS"))
    // attempt to init dbus early
else
   // debug message printing out that an attempt to connect to an
existing session bus was not done

I think that solves your use case, and will remove the whole dbus init
attempt for folks running under a user session where this is not
needed.

Thanks,


Auke