Re: Snappy GUI response

On Fri, Jul 04, 2003 at 08:10:15PM -0700, David Jeske wrote: 
> While I'm not against doing full backing store, I want to point out
> that win32 does not do it, and it has display speed and responsiveness
> that far surpasses OS X.

This is where my suggestion comes in to document in detail each piece
of code that runs from the button click to the user-visbile feedback,
along with what process/thread that code is running in and what that
code is doing with hardware.

Then you could line up the X case alongside the Windows case and say
something interesting about the difference between the two.


I do think your benchmark idea is a good one, not so much for the
comparison to Windows, but just to be able to run 5000 iterations of
click-a-menu-item on GTK, with oprofile or another profiler collecting
data. You could also just get a wall clock number on the benchmark and
know if your menu responsiveness is within the desired 50ms or so.  It
is fairly annoying to write benchmark or test suite stuff for
interactive operations but with good knowledge of GTK and the XTest
extension it ought to be possible.

With some care the test could be robust enough not to break with theme
changes and font size changes and such. Then we could even use it to
monitor toolkits (and desktop environments) for performance
regressions.

The exact benchmark probably varies a lot with font, theme, and
configuration, unlike something more constrained such as graphics
demos. But if your goal is profiling rather than comparison that
shouldn't be an issue.

Havoc