[systemd-devel] dbus user services using the session bus

[Will Stephenson]

On 03/04/13 18:48, Kok, Auke-jan H wrote:
> On Wed, Apr 3, 2013 at 6:22 AM, Will Stephenson <wstephenson at kde.org> wrote:
>> I'm joining the flock of desktop people investigating using systemd to start a desktop session.  So far I've checked out user-session-units and have created some more granular unit files that will do a native systemd start rather than u-s-u's wrapper around the upstream start script.
> As Tom asked, I'm interested in these changes as well - do you have
> your code posted somewhere?

I got a bit more time for this project and got things working.

https://github.com/wstephenson/klyde/tree/master/startup/kde-session-units
>
> Also, you're saying there's a "start script" somewhere, can you tell
> me what you mean by that?

http://quickgit.kde.org/?p=kde-workspace.git&a=blob&h=21c117492268db828990ee7cc232d991ee757126&hb=a7d0ef49cb7839d42646334459f1140ed32c0a41&f=startkde.cmake

>
> I don't understand what you refer to by "wrapper", either. Technically
> user-session-units doesn't wrap anything, it just provides straight
> unit files. One of them is derivative of user at .service from systemd,
> sure.
I meant that kde.service is a wrapper around the startkde shell script.


>> I'm stuck now, because I want to start a service with Type=dbus that puts a service on the session bus, however, I can't see a way to specify the bus in the unit file.
> you can insert variables into the user session by `systemctl --user
> set-environment DBUS_S....`. This is the only way to globally assign a
> dbus address to the session, but you'll have to do it before any
> significant service runs in the first place, which includes before
> dbus starts.
>
>> I don't want to use dbus activation here to start the service as that invokes the chicken-egg problem that the service in question (kdeinit) job is to start all the processes that will be calling it via dbus later. Long term I would like to do away with this and perform this task via systemd too, but Type=dbus on the session bus seems like a valid use case that should work.
>>
>> I see there's a way to get the dbus session bus address into systemd because of the patch to fix this in user-session-units; would the right approach be to add a BusInstance= field to service and use the provided session bus address when watching for Type=dbus services that depend on user/dbus.service?
> all of this stuff seems to work just fine with user-session-units - of
> course the login/DM integration is wholly missing right now, but could
> you start with user-session-units and make modifications to get where
> you want?
>
> Cheers,
>
>
> Auke