Re: Re: how to write snappy benchmark?

On Sat, 2003-07-05 at 12:52, Marko Macek wrote:
> Owen Taylor wrote:
> > On Sat, 2003-07-05 at 11:56, Marko Macek wrote:
> 
> >>$ while true ; do xrefresh ; sleep 0.1 ; done
> >>1) observe repaints. Ideally, nothing on the screen will blink
> >>(the only app I can find that does this right is Mozilla,
> >>mostly by my efforts).
> > 
> > Note that the reason that Mozilla doesn't blink is the same
> > reason:
> > 
> >  A) When you switch to the desktop with Mozilla on it and
> >     Mozilla is swapped out, you see Mozilla slowly appear
> >     out of whatever you had before, often first as a blinking
> >     cursor in the URL line.
> 
> Yes, this happens with unresponsive applications. But we shouldn't
> penalize the responsive ones.

Any application can be unresponsive in some circumstances - 
it's swapped out to disk or whatever. Background none makes
a bad situation worse by entirely breaking the user's model
of what windows are - why do I see the desktop / another
app here - that's supposed to be Mozillla.

> A solution would be an extension to X server/protocol that repainted
> the window after a specified timeout expired (>0.5 seconds)

0.5 seconds is an eternity ... I think you'd have a hard time
picking such a timeout usefully.

> >  B) Windows dragged over Mozilla's content area leave huge
> >     ugly trails of window border bits.
> 
> This also only happens when the application can't keep up with 
> repainting. This is either on slow hardware, unfair scheduling in
> X/kernel or inefficient repaint algorithm in the app or toolkit.
> 
> (it's quite visible in mozilla mailnews folder pane, but not
> at all in the browser).

Depends a lot for me on what the web page and how big the
window I'm dragging is. It's seems pretty easy to get past
the threshold where the trails are obnoxious.

> > Some of us consider not blinking in Mozilla's way a distinctly
> > bad thing. 
> 
> It's a tradeoff. My way, the repaints are faster and they happen
> only once (no blinking), but they are slightly delayed on slower
> machines.

What works best is really a question of the physiology of the
nervous system ... if the background color is a good approximation
to the contents of the page, it isn't clear to me that having the
window border briefly drawn at two pages is preferrable to a brief
clearing to the background color.

Regards,
						Owen