Power devices from session to system

[David Zeuthen]

On Fri, 2008-12-19 at 12:47 +0000, Richard Hughes wrote:
> I've recently converted gnome-power-manager to DeviceKit-power. The
> amount of code I've deleted is simply staggering, and apart from a few
> typos it seems to have gone down quietly with users.

Cool.

> One feature I've yet to convert is cell phones, where the data is
> presented from gnome-phone-manager (session).
> 
> In the old g-p-m model we had this, where g-p-m processed a lot of
> logic, handled a few quirks, and aggregated data sources:
> 
> H      {  G  }
> A ---> {  P  } ---> User
> L      {  M  }
>           ^
>           |
> gnome-phone-manager
> 
> In the new model we have this, where g-p-m is just a thin shim, and is
> used for presentation and policy:
> 
> D      {G}
> K ---> {P} ---> User
> P      {M}
> 
> In the new model, we pass around object paths, and don't build up an
> cached array of devices. Needless to say, it's much simpler, and it is a
> much better model.
> 
> Where we fail in this new model is taking data from the session and
> pushing it down into DeviceKit-power. In the example with
> gnome-phone-manager it connects over bluetooth in userspace, and polls
> for the battery state, and other network information. We can't have
> gnome-phone-manager connect to the same device as DeviceKit-power.
> 
> In an ideal world we could send untrusted data from the session to
> DeviceKit-power (protected by a PolicyKit method to active local
> sessions). We could get gnome-phone-manager to call into DeviceKit-power
> with a SetupDataPipe() which returns a fd for gnome-phone-manager to
> write into. Or something like that.
> 
> This means that any session device whether it's a cell phone,
> presentation clicker or a bluetooth mouse battery can be supported
> without having to open it system level.
> 
> I know, in an ideal world the device would sprout a power_supply class
> and we could just key off that, but some stuff like this has to be done
> in the session. Unless I'm missing the obvious.

I'm not sure we want to let sessions inject data into DeviceKit-power;
seems very complicated for handling the rather edge cases like this.

Why can't gnome-power-manager (or whatever component that is drawing the
UI) simply provide a very simple D-Bus interface that things like
gnome-phone-manager can hook into to post data?

(FWIW, I'd like to use this kind of interface too for my gphoto2 GVfs
backend; it's possible to read how much juice on these devices)

     David