Am 10.04.2017 um 07:56 schrieb Yale Zhang:

I've noticed dragging objects has become very slow (the refresh rate
is very low) in trunk.

I did a binary search and found the culprit is revision 15531 (Disable
GTK3 "client side decorations" on Windows by setting "GTK_CSD=0").

Eduard, are you seeing this? Do you think it's fundamentally caused by
disabling CSD or incidentally by some bug in GTK+?

-Yale

Hi Yale,

yes, I noticed slow dragging in trunk, but didn't make the connection to 
"GTK_CSD=0" (thought it was a general gtk3 issue).

I hope it's not a fundamental issue, as it would mean we'd be unable to 
achieve an even remotely native look on Windows (that's the very reason 
why I set that option in the first place).

I'll check this evening if re-enabling client side decorations also 
resolves the other performance issues I'm seeing in trunk compared to 
0.92.x and in any case I hope there's a possibility to get this fixed 
upstream.

Regards,
Eduard

• Setting "GTK_CSD=1" in fact improves performance a lot compared to the current default of "GTK_CSD=0". There definitely seems to be a GTK+ 3 issue. (If anything I'd expected performance to increase if GTK does not need to draw window decorations itself but can leave it to the window manager :-/)
  
• However even with "GTK_CSD=1" performance in trunk (r15618, gtk3) is  still *a lot* worse compared to 0.92.x (r15371, gtk2). This is particularly visible when zooming, but the overall perceived performance is also much better in the stable branch.

Yale: Are you (or anybody else) able to compile a working version of the stable branch with "WITH_GTK3_EXPERIMENTAL" enabled? I wanted to check if the remaining performance regressions might be related to the changes implemented during the last hackfest (which got reverted in 0.92.x in r15024), but my gtk3 build does not show the canvas at all (making a bit hard to compare drawing performance ;-) ).

Regards,
Eduard