Summary: | REGRESSION(r145305) Performance: 1.3% mac-release-10.6-webkit-latest/intl2/times/t change after rev 145300 | ||||||
---|---|---|---|---|---|---|---|
Product: | WebKit | Reporter: | Nils Barth <nbarth> | ||||
Component: | WebCore Misc. | Assignee: | Robert Hogan <robert> | ||||
Status: | RESOLVED FIXED | ||||||
Severity: | Normal | CC: | eric, esprehn+autocc, jchaffraix, ojan.autocc, robert, webkit.review.bot | ||||
Priority: | P1 | ||||||
Version: | 528+ (Nightly build) | ||||||
Hardware: | Unspecified | ||||||
OS: | Unspecified | ||||||
Attachments: |
|
Description
Nils Barth
2013-03-12 03:10:53 PDT
Link added below: (In reply to comment #0) > There looks like there's a performance regression caused by r145305 fixing > Bug 108357 REGRESSION(r140907): Incorrect baseline for cells with media content during load r145305: http://trac.webkit.org/changeset/145305 Bug 108357 REGRESSION(r140907): Incorrect baseline for cells with media content during load How do I access the source of the test that observed the regression? When I access the results page I can't seem to get anything other than a bunch of tabs. I don't get a graph or any other information. I'm using Chrome on Linux. Any suggestions on what I need to do to view the regression? Added Julien to CC list, in response to discussion at Chromium Issue 173154: https://code.google.com/p/chromium/issues/detail?id=173154 Opened bug at Chromium: Issue 189102: Performance regression 1.3% in table layout baseline fix http://crbug.com/189102 (In reply to comment #2) > How do I access the source of the test that observed the regression? Command line looks to be: /b/build/slave/Mac10_6_Perf/build/src/out/Release/performance_ui_tests --gtest_filter=PageCycler*.Intl2File:PageCycler*.*_Intl2File So (on Mac 10.6) should be able to reproduce by running: out/Release/performance_ui_tests --gtest_filter=PageCycler*.Intl2File:PageCycler*.*_Intl2File (In reply to comment #3) > When I access the results page I can't seem to get anything other than a bunch of tabs. I don't get a graph or any other information. I'm using Chrome on Linux. > > Any suggestions on what I need to do to view the regression? The graph takes a little while to load, and there's no feedback that it's loading. I'm using Chrome on Linux as well, and this takes about 7 seconds to load: http://chromium-perf.appspot.com/?tab=mac-release-10.6-webkit-latest&graph=times&trace=t&rev=187433&history=150&master=ChromiumWebkit&testSuite=page_cycler_intl2&details=true Could you try viewing again? The regression is about 80% of the way across; if you mouseover you'll see the Chromium and WebKit revision numbers. (In reply to comment #6) > (In reply to comment #2) > > How do I access the source of the test that observed the regression? > > Command line looks to be: > /b/build/slave/Mac10_6_Perf/build/src/out/Release/performance_ui_tests --gtest_filter=PageCycler*.Intl2File:PageCycler*.*_Intl2File BTW, this is determined by: * clicking through to build bot waterfall (from graph page), * search for test suite (on waterfall page) * click on stdio * eyeballing the stdio for a likely command line Created attachment 192880 [details]
Patch
I'm pretty sure this will solve the regression and feel a bit dumb for letting it through in the first place. My attempt to detect cases where the delayed load of an image would push the baseline of the row down incorrectly unless it is allowed to move up into the headroom above did not consider the case where we're laying out the row for the first time so the row doesn't have a baseline yet. In that situation it was laying out the cell a second time unnecessarily. The page cycler test suite is a set of popular websites as far as I can tell, so there's no specific use-case to test. If Julien is OK with the patch I guess we just watch the Performance bots after it lands. Comment on attachment 192880 [details] Patch Clearing flags on attachment: 192880 Committed r145822: <http://trac.webkit.org/changeset/145822> All reviewed patches have been landed. Closing bug. Fingers crossed, that appears to have done the trick: http://chromium-perf.appspot.com/?tab=mac-release-10.6-webkit-latest&graph=times&trace=t&rev=188265&history=150&master=ChromiumWebkit&testSuite=page_cycler_intl2&details=true |