I've had more of a look @ the transforms - the main problem would seem to be iterating through the whole transform stack every time an object is drawn, surely we could do some caching ?
I guess I need to have a look at the older branch that didn't do that to understand how that worked...
What would be really useful when messing with the transforms would be a bot that tests transforms/rotations etc, so it is easy to visually check everything is in the right place.
----- Original Message ----
From: ricardo lafuente <***@sollec.org>
To: shoebot-devel-***@public.gmane.org
Sent: Thursday, September 3, 2009 12:17:21 PM
Subject: Re: [shoebot-devel] [joke] One way to make ShoeBot run faster...
Post by Stuart AxonIn all seriousness though, the lib2geom longterm plan seems fine...
I've been vaguely thinking of making a benchmark bot... I did some initial testing and for some reason 'rotate' seems to really slow things down - from about 30fps to about 8fps on my computer IRCC
yeah, that's the transform issue we've been having since the beginning and that the 0.3 branch was supposed to solve.
In the end the old transform approach had to go back in because of paths with draw=False.
The main issue is that having the transform data included in each path (through the Mixin class) really takes its toll on the CPU -- would be good to have a smart way to have it work only when the path is not being drawn immediately to the context... i wonder.