On Mon, Aug 08, 2005 at 03:01:01PM +1000, Peter Moulder wrote:
On Thu, Aug 04, 2005 at 05:26:53PM -0700, Bryce Harrington wrote:
- Several people (including myself) have attempted to do performance
analysis of Inkscape. Typically this seems to really irritate developers; I recall being chastised that I was wasting my time and should be doing coding on the algorithms or something instead. In an inkscape-tester group such work could be conducted without directly irritating developers.
No reference is given to the conversation in question, but I'd guess that the "chastiser" was emphasizing the importance of algorithm choice when trying to improve performance: this is standard advice (second to to "use a profiler before touching the code"), and is mentioned in the comp.lang.c FAQ answer on performance tuning (http://www.eskimo.com/~scs/C-faq/q20.13.html).
Yes, that's accurate according to my memory. I don't recall whether the conversation was in IRC or email, but it probably doesn't matter.
(Ironically, my changed actually also resulted in *simplifying* e.g. by removing duplicate assignments and combining redundant logical checks.)
The main point I took from this comment was that the optimization problems were probably systemic rather than just implementational, and thus that replacement of the code (i.e., with Cairo) would be a better focus of effort than continuing with micro-optimizations.
In any case, I think the important thing is that we really need to develop a game plan for how to tackle performance optimization. This is something that we certainly cannot depend on Cairo to give us; presently the issue is *worse* with Cairo. ;-)
I think that even modest planning for measuring and tackling performance will not only scratch our own itch within Inkscape, but will address a deep need in Cairo and in OSS as a whole.
Does anyone have ideas for how we can approach this?
Bryce