On 04/12/2012 07:55 AM, Jonathan Kew wrote: > On 12/4/12 02:47, Behdad Esfahbod wrote: >> As a crude test, I profiled the Indic shaping, and am conjecturing that about >> 10 to 20 percent of the time can be saved pre-planning the shaping process. >> My testing showed no measurable saving for skipping the sanitizing process.
I should note that my comment re sanitize() having no measurable effect was due to the fact that sanitize happens once per face creation and not per hb_shape() call. Which still means that we can postpone optimizing that. >> Maybe both can wait (and not block a 1.0 release) since neither one seems to >> be hugely effective. > > A saving of 10-20% sounds pretty worthwhile to me - I should note that my test case was with mostly one-word strings, so for real world data the savings may be much less. But I'll give it a try. > and if 1.0 is supposed to > provide a long-term stable API, then perhaps this should be done sooner rather > than later. > > Otherwise, there'll be a strong temptation to rev the API again shortly after > 1.0 in order to achieve this performance boost. Well, API *additions* can happen with every new release. It's API changes / deprecations we have to avoid (contractually). But since the code is already organized to allow it, I'll give this a try now. b > JK > _______________________________________________ > HarfBuzz mailing list > [email protected] > http://lists.freedesktop.org/mailman/listinfo/harfbuzz > _______________________________________________ HarfBuzz mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/harfbuzz
