The member variables of CachedResource could be rearranged for better padding.
Created attachment 56175 [details] Patch
Comment on attachment 56175 [details] Patch I'll do some bitfields to save more space.
Created attachment 56177 [details] Patch
This patch saves 32 bytes per CachedResource
Comment on attachment 56177 [details] Patch r=me
http://trac.webkit.org/changeset/59576
<rdar://problem/7990041>
http://trac.webkit.org/changeset/59576 might have broken Qt Linux ARMv5 Release
Sorry, no. Just the slave was lost. I will fix sherriff bot to understand "slave lost" this week. That or we'll remove the Qt builders since they cause sooooo much spam these days. :(
Comment on attachment 56177 [details] Patch Should we COMPILE_ASSERT that CachedResource is less than a certain size? I'm not sure how many of these objects we typically keep around in process.
We may have 1500+ alive at any one time with lots of windows/tabs open. Yeah, I wish we had a way to identify memory regressions. Maybe we could have some tools that scrape the dSYM info for each build, and track size changes?
(In reply to comment #11) > Yeah, I wish we had a way to identify memory regressions. Maybe we could have some tools that scrape the dSYM info for each build, and track size changes? I think that that, or using COMPILE_ASSERTs on the various classes which we know we need to keep small are excellent ways to help prevent memory usage regressions.