Created attachment 172254 [details] Patch

Comment on attachment 172254 [details] Patch Clearing flags on attachment: 172254 Committed r133429: <http://trac.webkit.org/changeset/133429>

All reviewed patches have been landed. Closing bug.

This patch looks like it caused a bunch of perf/memory regressions: http://build.chromium.org/f/chromium/perf/mac-release-10.6-webkit-latest/dom_perf/report.html?rev=166198&graph=CreateNodes&history=100 http://build.chromium.org/f/chromium/perf/mac-release-10.6-webkit-latest/intl1/report.html?rev=166182&graph=V8.MemoryHeapSampleTotalUsed_0.75&trace=V8.MemoryHeapSampleTotalUsed_0.75&history=150 http://build.chromium.org/f/chromium/perf/chromium-rel-win7-webkit/intl2/report.html?rev=165925&graph=ws_single_peak_r&history=100 The only other patch in the regression range was rolled out in https://bugs.webkit.org/show_bug.cgi?id=101533 and the regression didn't go away. It's hard for me to see where the regression would come from. The only thing I can think of is maybe the implementation of V8EnabledPerContext versus V8EnabledAtRuntime? That's just a wild guess. Is it possible at this point to do a rollback so we can be 100% sure it was this patch?

Can we think that regression comes from chromium side instead of WebKit side? This is also a wild guess, though.

(In reply to comment #5) > Can we think that regression comes from chromium side instead of WebKit side? > This is also a wild guess, though. I considered that, but on the non-webkit canary bots, the same regression happened when this webkit revision was pulled into chromium deps and not at the same chromium revisions as on the webkit canary bots. So, I don't see how it could be a chromium side change. It's certainly possible I got confused and misread the graphs though. Feel free to double-check my conclusions. :)

Hmm... I would like to hear morrita's comment.

Reverted r133429 for reason: Speculative roll out, investigating perf regression. Committed r133904: <http://trac.webkit.org/changeset/133904>

Looks like this was the cause of the regression. All the graphs listed have recovered their losses.

This was rolled out at r133904 and we have no plan to land this again. Closing.