Summary: | DOM/DOMWalk.html is flaky in Chromium Win Release Perf bot | ||||||
---|---|---|---|---|---|---|---|
Product: | WebKit | Reporter: | Kentaro Hara <haraken> | ||||
Component: | Tools / Tests | Assignee: | Dominic Cooney <dominicc> | ||||
Status: | RESOLVED FIXED | ||||||
Severity: | Normal | CC: | dominicc, noel.gordon, rniwa, webkit.review.bot | ||||
Priority: | P2 | ||||||
Version: | 528+ (Nightly build) | ||||||
Hardware: | Unspecified | ||||||
OS: | Unspecified | ||||||
Attachments: |
|
Description
Kentaro Hara
2012-11-16 06:40:46 PST
*** Bug 104866 has been marked as a duplicate of this bug. *** Created attachment 182715 [details]
Patch
I am assuming that zero is a plausible result for some of the (micro) benchmarks in DOM/DOMWalk.html. One thing that makes me suspicious is that Dromaeo results on the Windows bots are typically numerically larger than those of the Mac bots, yet the Mac bot does not exhibit this failure. Perhaps the timer works differently on those platforms. If this warrants more investigation, the alternative is to add more instrumentation to Dromaeo so we can see which specific test functions are failing (it is a flaky set, usually >1 but from zero to a few tests fail this way.) For now proposing that we assume the result is correct and allow zero as a (very fast) result to green the Windows perf bot. Comment on attachment 182715 [details]
Patch
Looks reasonable to me as an immediate fix to suppress the win bot errors, though we want to improve the test so it won't return zero (zero means that we're not testing something meaningful).
Comment on attachment 182715 [details] Patch Thank you for the quick review. I filed bug 106882 to improve the test. Comment on attachment 182715 [details] Patch Clearing flags on attachment: 182715 Committed r139728: <http://trac.webkit.org/changeset/139728> All reviewed patches have been landed. Closing bug. |