On Tue, 20 Jun 2006 22:21:16 +0200, Clemens Eisserer said: > 1.) Not even one comment from a developer. Maybe there are not enough, > maybe nobody cares, ore maybe they can't cope with critic? I don't > know. Maybe none of the developers are seeing your issue, and none of the usual bleeding-edge testers see it either. This makes it hard for them to fix it. > 2.) Some fans respond that on their machines its working fast enough. I'm suspecting you're addressing me with that comment. ;) My point was that it *isn't* the case where every single GTK2 system is seeing horrid performance, and that it always sucks in every case. You seem to be encountering an issue that is *not* present on every system. (On the other hand, if computers and users didn't keep finding new ways to trip over obscure corner cases, I'd be unemployed, so I'm not complaining ;) > I see/saw quite frequently complaints about GTK's performance, I saw > tables comparing GTK-1.2 vs. GTK-2 performance, authors migrating away > from GTK-2 because they were concerned loosing users when switching to > this toolkit. Because usually the person complaining disappears before providing useful info like this: > > So where *does* the time go to, if not pango? The more detail you can > > provide, the easier it is to track down. > I did some sysprof testing back when I tried to analyze why the > Eclipse/GTK port performs like crap (this was FC4 / GTK-2.6 / Default > theme). It was almost a flat profile, but X spiked quite a lot. Doing > the same test (invalidating eclispe's main window) shows about half > (!!!) of the time is spent in X. The test with gftp (moving splitters) > shows X using 67.1% of CPU whereas gftp itself only consumes 31%. > I am using Xorg 6.9 with grayscale-aa fonts (xrender is accalerated). OK, *now* we have something that we can start looking into - we now know that at least part of the problem is that gftp is generating some very sub-optimal X requests. And we have a test case that reproduces easily (although on my box moving the main horizontal splitter in the gftp gui up and down result in about 70% gftp-gtk and 25% in the X server). I'll take a closer look at this later tonight - the kernel I have booted at the moment doesn't have oprofile support. (And no, I don't pretend to know the GTK innards well enough to fix this, but I've been doing systems debugging and tuning for enough decades that I certainly should be able to get a clear and coherent bugzilla out of it so that somebody else can run with it from there...)
Attachment:
pgpYslXlLRlWO.pgp
Description: PGP signature