- From: Boris Zbarsky <bzbarsky@MIT.EDU>
- Date: Wed, 27 Jun 2012 12:50:42 -0400
- To: www-style@w3.org
On 6/27/12 12:18 PM, Alex Mogilevsky wrote: > These are real issues the code has to deal with, but there is no evidence that it is a problem for performance. It's a problem only insofar as it disables a certain class of optimizations that are now possible and that some UAs do in fact perform. For example, right now it's possible to share computed display data across nodes that have the same rules applying to them, no matter what their parents are. This is in fact a feature that non-inherited properties in CSS have in common for the most part, and is the basis for a number of optimizations at least in Gecko. Now we can obviously revisit those optimizations, but just disabling them naively is _guaranteed_ to be a problem for both memory usage (because it will disable the sharing, forcing a lot more data to be stored in memory) and for performance (because the data that is shared right now obviously only needs to be computed once). The last time I ran the numbers on a change along these lines, I measured 5-10% difference in pageload performance. If that counts as "evidence". -Boris P.S. I don't have a strong opionion on the matter at hand, since I haven't really been following the discussion, to be honest. Obviously, with enough code complexity we can always make this work on our end...
Received on Wednesday, 27 June 2012 16:51:16 UTC