BTW I forgot to comment that we currently cache the XDOM but it's cached before transformations are executed.
We could improve this and define the notion of cacheable transformations vs non-cacheable transformations. For example Macros would be non-cacheable transformations (since they contain script and you need dynamism) but WikiMacros transformation is a cacheable transformation. Thanks -Vincent On Dec 12, 2011, at 9:59 AM, Vincent Massol wrote: > Hi, > > On Dec 12, 2011, at 9:16 AM, Hamster wrote: > >> Hi Vincent, >> >> Just thinking out loud about the performance-hit when WikiWords are >> enabled... >> >> I guess that there's performance-hit because the pages are processed each >> time a user opens that page, right? >> >> If so, maybe we should use a different mechanisme for the WikiWords? There >> are two cases when WikiWords should be processed: (1) When a user edits a >> page and adds a WikiWord to that page, and (2) when a user adds a new page >> to the Wiki (all other pages should be updated). If we have a "database" of >> all WikiWords in the Wiki, we would know up-font which pages we need to >> update. >> >> In both cases, the WikiWords are processed only once, and further >> performance-hits should not occur(???) > > Yes you're completely right. > > Actually I code the WikiWord Rendering Transformation as an exercise just to > prove that it worked and as a stop gap solution to offer this feature quickly. > > You're right that a better solution would be what you describe. However the > problem still remains that you'll need to map the "table of links" to the > document content being rendered. Imagine for example that for document X the > table returns 2 links called A and B. When the rendering engin renders the > content as HML (for ex), our code would need to extend/override the onWord() > event and compare it to A and B to see if they match. > > So the performance gain will be the different between a contains() in a List > (for checking if it matches A and B) vs the time to do a regex match with a > compiled pattern. > > I have the feeling the regex match could actually be faster or very similar > since if it's implemented correctly it'll check the first letter and return > if they don't match and if it's not an uppercase letter, then check the > second letter, etc. > > WDYT? > > Thanks > -Vincent > >> WDYT? > _______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs
