Re: Unaccounted time during containers' lifecycle

["Matt Hoosier" <matt hoosier gmail com>]

On 4/9/07, Kalle Vahlman <kalle vahlman gmail com> wrote:
> (dropping gtk-list)
>
> 2007/4/8, Matt Hoosier <matt hoosier gmail com>:
> > On 4/8/07, Ross Burton <ross burtonini com> wrote:
> > > [Cycle 36]: construction finished at 0.0025 sec
> > > [Cycle 36]: style set at 0.0032 sec
> > > [Cycle 36]: requisition computed at 0.0075 sec
> > > [Cycle 36]: size allocated at 0.0080 sec
> > > [Cycle 36]: realized at 0.0081 sec
> > > [Cycle 36]: mapped at 0.0088 sec
> > > [Cycle 36]: exposed at 0.0726 sec
> > >
> > > Hm, style set isn't that slow.  Hm, for some reason I'm not getting the
> > > second requisition/size-allocated pair after the window was mapped.
> > > Where you using a standard theme, or something custom?  (I'm using
> > > Darkilouche, which is based on Clearlooks).
> >
> > Those tests were run with the stock Gtk 'Default' theme. I also did a
> > quick check just now to make sure that the stripped down version of
> > gnome-settings-daemon isn't somehow producing a spurious 'Default' ->
> > 'Default' theme transition. It wasn't; the timings were the same both
> > with and without that daemon running in the background.
>
> I tested with both Clearlooks and stock theme, and also got just one
> size-request iteration (with a recent gtk+ version if that should
> matter).
>
> One thing I'm wondering is what is your window manager. You said you
> were on ARM so first wm that leaps into mind is Matchbox, which most
> likely will reconfigure the window (as it allows only fullscreen apps)
> when it is mapped. Configure-events seem like good candidates for
> causing size calculation, so you might want to check for these.

Yes, that ends up explaining the second size calculation. With no WM
running, the timings now look like:

[Cycle 7]: construction finished at 0.1061 sec
[Cycle 7]: requisition computed at 0.2181 sec
[Cycle 7]: size allocated at 0.2311 sec
[Cycle 7]: realized at 0.2327 sec
[Cycle 7]: mapped at 0.2590 sec
[Cycle 7]: exposed at 0.3855 sec

(That compares to a .4436 timestamp for the end of exposing while the
WM is running.)

Thanks for the insight on that question.

I'm still keen to figure out where the rest of the work goes with
containers, though.

> The attached version of the program connects to the configure-event
> signal and prints it out, revealing that I actually get two of these
> with metacity (wonder why) between map expose. The size doesn't change
> and I guess matches the widget's idea of it's size so no
> recalculation, but maybe this is not the case for you?
>
> --
> Kalle Vahlman, zuh iki fi
> Powered by http://movial.fi
> Interesting stuff at http://syslog.movial.fi