should we use QT_NO_GLIB=1 like kwin?

[Andreas Pakulat]

On 07.10.10 00:07:33, Milian Wolff wrote:
> On Wednesday 06 October 2010 23:03:30 Aleix Pol wrote:
> > On Wed, Oct 6, 2010 at 10:51 PM, Milian Wolff <mail at milianw.de> wrote:
> > > Hey all,
> > > 
> > > since I spent some time today to investigate the excessive wakeups of
> > > KDevelop
> > > while it's supposedly doing *nothing* I got a suggestion by pinotree in
> > > #kde-devel:
> > > 
> > > He proposed to use QT_NO_GLIB for KDevelop. Indeed that removes most
> > > calls to
> > > poll() when comparing a strace -tt -p $(pidof kdevelop).
> > > 
> > > See e.g. here how to implement:
> > > 
> > > 
> > > http://websvn.kde.org/trunk/KDE/kdebase/workspace/kwin/main.cpp?revision=
> > > 1177865&view=markup search for QT_NO_GLIB.
> > > 
> > > It's very easy and for kwin it also fixed high CPU usage bugs. Didn't we
> > > have
> > > some like these as well?
> > > 
> > > Anyhow, the draw back would be: It's not possible to use gstreamer or
> > > phonon
> > > in KDevelop anymore as these require that glib eventloop. I personally
> > > don't
> > > see a reason (now) for using phonon in KDevelop. In the far future one
> > > might
> > > want to play screencasts or other media in embedded documentation...
> > > 
> > > So what do you guys think: Should we try it out? Or leave it as-is?
> > > 
> > > Bye
> 
> > As I said on the IRC I wouldn't be against it but I'd like to make it sure
> > that we _need_ that to work properly. Isn't there a reason for these poll
> > calls for instance? If you open a small app and leave it open it doesn't
> > take CPU constantly, at least...
> 
> Ah indeed, I didn't make that clear in my first mail:
> 
> 1) according to powertop the wakeups are not noticeably less with QT_NO_GLIB
> 2) there are still quite a few stats etc. going on according to strace which 
> I'll investigate sooner or later.
> 
> It's just that pinotree (someone whom I trust code-wise) suggested it and that 
> kwin already uses it... I hoped for a freebee...

Then what changes without it? If the change has no (almost no) effect I
don't see a reason to do it. It'll probably also bite you (or someone
else) if they do want to use something that needs the
glib-loop-integration as by then everybody will have forgotten about
this and spends time debugging why is feature doesn't work...

Andreas

-- 
A long-forgotten loved one will appear soon.

Buy the negatives at any price.