hmacht at suse.de
Thu Mar 22 06:33:18 PDT 2007
On Thu 22. Mar - 13:21:01, Richard Hughes wrote:
> On 22/03/07, Rob Taylor <rob.taylor at codethink.co.uk> wrote:
> >Richard Hughes wrote:
> >> On Wed, 2007-03-21 at 15:58 -0400, Havoc Pennington wrote:
> >>> Though dbus does allow you to do these async (fire them all off then
> >>> wait for all results), which I would recommend for anything an app is
> >>> doing at startup, even though it's sort of painful. Assuming most
> >>> apps
> >>> will be doing a variety of calls on startup it probably will speed up
> >>> app launch if say 20-30 round trips are all done async instead of
> >>> with
> >>> blocking round trips in serial.
> >> Sure, agree.
> >Umm, i was actually talking about context switches, nothing to do with
> >async versus sync. I'd expect an app to do async calls on startup.
> >I guess its theoretically possible at the libdbus level to queue up a
> >number of messages and do one context switch to send them all...
> Gotcha. What about doing both?
> We keep the CanSuspend for the trivial desktop case, and also provide
> a GetCapabilities method that returns a bitfield for embedded use, or
> for languages like python that can unwrap a bitfield easily.
> It's one extra method, and one extra signal, so it's hardly much API
I disagree. In which case is another application querying this information
more than once? And in which case does the other application need more
than one specific information. I don't think application developers
understand why they should duplicate this functionality for those easy and
rare methods. It makes the whole thing unnecessarily complicated. It
should be as simple and transparent as possible.
More information about the xdg