Bug 83400

Summary: [chromium] Hundreds of mac failures are due to a slightly different color repaint background
Product: WebKit Reporter: Ojan Vafai <ojan>
Component: Tools / TestsAssignee: Nobody <webkit-unassigned>
Status: RESOLVED WONTFIX    
Severity: Normal CC: caryclark, danakj, epoger, reed, senorblanco
Priority: P2    
Version: 528+ (Nightly build)   
Hardware: Mac   
OS: Unspecified   

Description Ojan Vafai 2012-04-06 15:09:11 PDT
For example, see http://test-results.appspot.com/dashboards/flakiness_dashboard.html#showExpectations=true&tests=fast%2Ftable%2Fempty-section-crash.html.

The repaint background on the platform/chromium-win and platform/chromium-linux results is #575757. On the Chromium mac bot, some pixels are #575757 and some are #565656. Not really sure how image_diff translates that to the whole background not matching.

The platform/mac result has a background that is #6a6a6a, with a color profile. Diffing the chromium win/linux result with the platform/mac result only shows a diff on the text though (i.e. the backgrounds match). I assume image-diff accounts for the color-profile somehow.

Also, curiously, on some of the snowleopard bots, we sometimes get a result that matches the platform/mac result: http://test-results.appspot.com/dashboards/flakiness_dashboard.html#group=%40ToT%20-%20webkit.org&showExpectations=true&tests=fast%2Ftable%2Fempty-section-crash.html
Comment 1 Dana Jansens 2012-04-06 15:51:41 PDT
Probably this https://codereview.appspot.com/5758043
Comment 2 Dana Jansens 2012-04-06 15:52:06 PDT
Here's the skia bug http://code.google.com/p/skia/issues/detail?id=420
Comment 3 epoger 2012-04-09 12:04:22 PDT
Ojan, can you please tell me how to get a list of the "hundreds" of tests that fall into this category?  (Or at least a good chunk of them?)

Taking the example given, empty-section-crash, I see (via svn blame) that it was marked as expected to IMAGE fail at least 16 months ago (http://trac.webkit.org/changeset/74070/trunk/LayoutTests/platform/chromium/test_expectations.txt).  Its expectations line refers to http://crbug.com/23489 , which was filed in Sept 2009...
Comment 4 Ojan Vafai 2012-04-09 15:02:36 PDT
(In reply to comment #3)
> Ojan, can you please tell me how to get a list of the "hundreds" of tests that fall into this category?  (Or at least a good chunk of them?)

1. load "webkit-patch garden-o-matic"
2. go to the expected failures tab
3. examine the fast/repaint tests

Fixing the color issue probably won't make all these tests pass, but it will make it so that the pixel diffs are just text-rendering/antialiasing. So, they would be straightforward rebaselines.

I think a large percentage of these have also be rebaselined. I know I rebaselined a lot of these for the Lion port before I realized that it was a more widespread problem. So, fixing this would cause a number of tests to "fail" that would need rebaselining.
Comment 5 Stephen Chenney 2013-04-09 17:05:54 PDT
Marked LayoutTest bugs, bugs with Chromium IDs, and some others as WontFix. Test failure bugs still are trackable via TestExpectations or disabled unit tests.