Interesting results. FVWM does pretty well. I was surprised at the good result for E16, and even more with GNOME. The thing that did amaze me was how poorly WindowMaker*, Fluxbox, and XFCE performed, since I always assumed that they were pretty optimised WMs.
(* Amazed but amused, since a work colleague of mine uses WindowMaker on FreeBSD, and often brags about WindowMaker, saying how ancient FVWM is. Needless to say I forwarded this link to him immediately! )
Well, I hate [1] to be so critical of such an obviously well-conceived test, but he doesn’t say which windows he used (just an xterm? I doubt that). How did he time this? If it was the time it took for the window to become visible, then OK… I could just about live with that.
But mapped window (and the timing thereof) are also based around the program that’s loading. Look at a program with QT dependencies. That will take a lot longer to load than say, xmessage. So it’s flawed. I’m hoping he used the same programs on all WMs he tested.
Well, he did make his testsuite available and admits that it doesn’t amount to much yet. Also e17 only performs that fast with about everything disabled, the vanilla e17 performs a lot worse. But all in all I’m not really surprised about e17’s performance. Also I’m wondering about the defaultness of Gnome, I thought they had been using Metacity as a WM for quite some time now?
Also, concerning what I saw on the Fvwm mailinglist, wouldn’t just moving the window on- and offsceen (as in fvwm-cvs) be faster than mapping and unmapping each time (as in fvwm 2.5.12 and before) ? Or am I misinterpreting what he has been testing?
)