Hi everyone, I spent most of today investigating the performance issues everyone(?) of us is experiencing. Heres some interesting things I found out that hopefully will help solving at least some performance issues:
First of all I found out that some performance problems are not connected to kde4 but actually to qt4.4.0. This was obvious since an application I develop, KPorts, is qt-only and start-up-time changed from ~12seconds to ~3minutes upon upgrading qt-4.3.1 to qt-4.4.0. I had a closer look at KPorts startup-functions to see where slowdowns in qt can possily occur. The function that runs during start-up does several things, which I wont explain in detail. It boils down to disk-io(only reading), string-handling (splitting, sectioning..), and creation of new objects. By commenting out relevant passages I could proved that its neither of the first two. So I thought it must something with memory allocation during creation of objects. I wrote a test-app for that and found out that allocating new QObjects is same speed on qt-4.3.1 vs qt-4.4.0. (=> memory allocation isnt the problem) Next things I checked was things going on in the constructors of the objects and I found out that its the connect()-calls which somehow got much slower from qt-4.3.1 to qt-4.4.0. Sounds strange but it is the case. During startup of Kports about 80K objects are created that each create 3 Signal-Slot-Connections to a static object for error handling. After temporarily removing the connect()-calls kports, startup-time went from >3minutes to ~6seconds!!! I wrote a test-application to benchmark creation of signal-slot-connections and it confirms that creating a connection takes at least twice as long with qt-4.4.0. I attached the code, it creates 2million connections and tells you how long it took. (note that you have to ctrl-c to quit the program). Please check whether the time needed to create these connections on your machines differs on qt-4.3.1 vs qt-4.4.0 as well. I have 10seconds on qt43 vs 20seconds on qt44 ( I know that doesnt completely explain the time difference in kports but it seems significant). I'm not sure this problem is responsible for all performance issues, but signal-and-slotting sure is used everywhere, so slowdown there could explain a lot. Greetings Hannes
perftest.tar.bz2
Description: application/tbz
_______________________________________________ kde-freebsd mailing list kde-freebsd@kde.org https://mail.kde.org/mailman/listinfo/kde-freebsd
