on 32-bit Release only though. Diff: --- /var/lib/buildbot/build/gtk-linux-32-release/build/layout-test-results/tables/mozilla_expected_failures/marvin/table_overflow_dirty_reflow_tbody-expected.txt +++ /var/lib/buildbot/build/gtk-linux-32-release/build/layout-test-results/tables/mozilla_expected_failures/marvin/table_overflow_dirty_reflow_tbody-actual.txt @@ -1,31 +1,18 @@ -layer at (0,0) size 784x728 +layer at (0,0) size 784x612 RenderView at (0,0) size 784x600 -layer at (0,0) size 784x532 - RenderBlock {HTML} at (0,0) size 784x532 - RenderBody {BODY} at (8,8) size 768x516 - RenderTable {TABLE} at (0,0) size 230x516 [border: (1px outset #808080)] +layer at (0,0) size 784x416 + RenderBlock {HTML} at (0,0) size 784x416 + RenderBody {BODY} at (8,8) size 768x400 + RenderTable {TABLE} at (0,0) size 230x400 [border: (1px outset #808080)] RenderBlock {CAPTION} at (0,0) size 230x206 [border: (3px solid #FFA500)] RenderText {#text} at (103,3) size 24x17 text run at (103,3) width 24: "cap" - RenderTableSection {TBODY} at (1,207) size 228x308 - RenderTableRow {TR} at (0,30) size 228x86 - RenderTableCell {TD} at (30,48) size 54x50 [border: (1px inset #808080)] [r=0 c=0 rs=1 cs=1] + RenderTableSection {TBODY} at (1,207) size 228x192 + RenderTableRow {TR} at (0,30) size 228x132 + RenderTableCell {TD} at (30,71) size 54x50 [border: (1px inset #808080)] [r=0 c=0 rs=1 cs=1] RenderText {#text} at (16,16) size 22x17 text run at (16,16) width 22: "foo" - RenderTableCell {TD} at (114,30) size 84x86 [border: (1px inset #808080)] [r=0 c=1 rs=1 cs=1] - RenderText {#text} at (16,16) size 21x17 - text run at (16,16) width 21: "bar" - RenderBR {BR} at (36,16) size 1x17 - RenderText {#text} at (16,34) size 23x17 - text run at (16,34) width 23: "baz" - RenderBR {BR} at (38,34) size 1x17 - RenderText {#text} at (16,52) size 23x17 - text run at (16,52) width 23: "zap" - RenderTableRow {TR} at (0,146) size 228x132 - RenderTableCell {TD} at (30,187) size 54x50 [border: (1px inset #808080)] [r=1 c=0 rs=1 cs=1] - RenderText {#text} at (16,16) size 22x17 - text run at (16,16) width 22: "foo" - RenderTableCell {TD} at (114,194) size 84x84 [border: (1px inset #808080)] [r=1 c=1 rs=1 cs=1] + RenderTableCell {TD} at (114,78) size 84x84 [border: (1px inset #808080)] [r=0 c=1 rs=1 cs=1] RenderBlock {DIV} at (16,16) size 52x52 [border: (1px solid #008000)] RenderBlock {DIV} at (1,1) size 402x302 [border: (1px solid #FF0000)] RenderText {#text} at (1,1) size 8x17
Err 64-bit Debug too.
I suspect this: <body onload="window.setTimeout('domfunc()',10)"> Could be the layout of the table takes more than 10ms in some cases, introducing flakiness. Will increase it to 100ms.
Created attachment 122647 [details] Patch
(In reply to comment #3) > Created an attachment (id=122647) [details] > Patch I checked this fix locally, out of 50 iterations of the test no flakiness!
(In reply to comment #4) > (In reply to comment #3) > > Created an attachment (id=122647) [details] [details] > > Patch > > I checked this fix locally, out of 50 iterations of the test no flakiness! I'll mark the test flaky in the expectations but it'd be great to get this patch reviewed. From what Martin told me, he's not sure how to review patches modifying imported tests.
(In reply to comment #5) > (In reply to comment #4) > > (In reply to comment #3) > > > Created an attachment (id=122647) [details] [details] [details] > > > Patch > > > > I checked this fix locally, out of 50 iterations of the test no flakiness! > > I'll mark the test flaky in the expectations but it'd be great to get this patch reviewed. From what Martin told me, he's not sure how to review patches modifying imported tests. Usually we prefer to point out the issue upstream, have it fixed and merge a newer version of the test or test suite. Here I am not really sure bumping the time taken to run the test is the way to go (especially to such a high value). If the problem is a partial layout, then it sounds like you should force layout at the appropriate time, not hope that WebKit will be done by your new deadline (unless I am missing something).
Comment on attachment 122647 [details] Patch Removing r flag. It seems indeed that updating the test like this is not a great option in this case.
This test(s) has been consistenly failing for the last 4000 revisions, that means the test(s) is no flaky anymore. The test(s) has been gardened and its baseline updated in r269363. Closing bug.