On Fri, 16 Sep 2005 16:18:11 -0500 (CDT) "Joe Lapp" <joe.lapp@xxxxxxxxx> wrote: > > (4) When a page is rendered, we create an md5 file with extension .f > > containing the RSS description for the page, unless the .f file > > already exists. When saving a page's new instructions, we delete > > the page's .f file. (Drawback: the description reflects the current > > file and not the cited revision.) I'm not sure if I like this idea for just the drawback you mentioned. > This eliminates the biggest remaining resource drain, after the > rsort(): instead of deserializing all of the instruction files into > object trees, concatenating them together into HTML, and stripping > tags, we just load this stuff already predone. 1) If the page was cached already (second stage) we have a xhtml plain file, no objects, no deserialization involved. 2) Did you try and prove the deserialisation/object creation is the problem? Any benchmarks? I just want to be sure we're not fixing the wrong stuff. There was a request some time ago to use diffs in the RSS summaries which might be the faster way anyway. Even directly accessing the main source could be an idea - ignoring the problems that may arise (eg. unprocessed plugins). BTW: I'd like to remind you on the main usage of DokuWiki as wiki for (company internal) documentation - so people usually are not interested in what the page is generally about but what has changed (this is only different when the page is created). Andi -- http://www.splitbrain.org -- DokuWiki mailing list - more info at http://wiki.splitbrain.org/wiki:mailinglist