Bug 142181 – REGRESSION: fast/css/acid2-pixel.html and http/tests/misc/acid2-pixel.html became flaky

• Component: Layout and Rendering
• Status: RESOLVED DUPLICATE

[Alexey Proskuryakov]

These tests became flaky on February 26th. First recorded failure in 180720, although the actual culprit could be substantially earlier.

 layer at (84,144) size 144x24
   RenderBlock (positioned) {DIV} at (48,72) size 144x24 [bgcolor=#FF0000]
     RenderBlock {DIV} at (0,0) size 144x0
-      RenderInline {OBJECT} at (0,0) size 131x14
-        RenderInline {OBJECT} at (0,0) size 131x14
-          RenderImage {OBJECT} at (13,0) size 131x24 [border: none (12px solid #000000) none]
+      RenderInline {OBJECT} at (0,0) size 35x14
+        RenderInline {OBJECT} at (0,0) size 35x14
+          RenderImage {OBJECT} at (109,24) size 35x0 [border: none (12px solid #000000) none]
     RenderBlock (floating) {DIV} at (0,0) size 144x24 [border: none (12px solid #FF0000) none (12px solid #000000)]
     RenderBlock {DIV} at (0,0) size 144x24 [border: none (24px solid #FFFF00)]
 layer at (84,216) size 144x24

[Simon Fraser (smfr)]

Possibly caused by

[Simon Fraser (smfr)]

I was thinking of http://trac.webkit.org/changeset/180683

[Alexey Proskuryakov]

Should I try rolling it out to confirm?

[Alexey Proskuryakov]

I tested locally. Got a flaky failure with r180683, not failure with r180682.

run-webkit-tests LayoutTests/fast/css/acid2-pixel.html --repeat 1000 -f --child-processes=50

[Said Abou-Hallawa]

Created attachment 247736 [details]
Patch

[Simon Fraser (smfr)]

Comment on attachment 247736 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=247736&action=review

> Source/WebCore/ChangeLog:13
> +        calls. In a very rare case, the layout does not finish reconstructing the
> +        render tree when the test runner dumps it. This introduced flakiness in

We should fix DRT and WTR to update style before the final layout.

[Antti Koivisto]

Comment on attachment 247736 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=247736&action=review

> Source/WebCore/ChangeLog:14
> +        calls. In a very rare case, the layout does not finish reconstructing the
> +        render tree when the test runner dumps it. This introduced flakiness in
> +        this test.

What case?

> Source/WebCore/html/HTMLObjectElement.cpp:140
>      clearUseFallbackContent();
> -    setNeedsStyleRecalc(SyntheticStyleChange);
> +    setNeedsStyleRecalc(ReconstructRenderTree);

This feels like papering over the actual problem.

[Alexey Proskuryakov]

Said asked me to roll out the offending patch for now.

*** This bug has been marked as a duplicate of bug 142249 ***