The following tests have been failing for GTK since r188947:
imported/w3c/css/css-multicol-1/multicol-basic-001.html [ ImageOnlyFailure ]
imported/w3c/css/css-multicol-1/multicol-basic-002.html [ ImageOnlyFailure ]
imported/w3c/css/css-multicol-1/multicol-basic-003.html [ ImageOnlyFailure ]
imported/w3c/css/css-multicol-1/multicol-basic-004.html [ ImageOnlyFailure ]
imported/w3c/css/css-multicol-1/multicol-rule-002.xht [ ImageOnlyFailure ]
imported/w3c/css/css-multicol-1/multicol-rule-px-001.xht [ ImageOnlyFailure ]
imported/w3c/css/css-multicol-1/multicol-rule-stacking-001.xht [ ImageOnlyFailure ]
imported/w3c/css/css-multicol-1/multicol-shorthand-001.xht [ ImageOnlyFailure ]
imported/w3c/css/css-multicol-1/multicol-span-all-block-sibling-003.xht [ ImageOnlyFailure ]
imported/w3c/css/css-multicol-1/multicol-span-all-margin-nested-firstchild-001.xht [ ImageOnlyFailure ]
I should note that all the .xht tests print the word "FAIL FAIL FAIL" repeatedly, but the expected results include "FAIL FAIL FAIL" as well, so that problem is unrelated to this bug. That is probably that we are treating the .xht files as HTML rather than XHTML, bug #152455. In this bug I am only concerned with the pixel differences introduced by r188947.
I have no clue why r188947 fixed pixel differences on Macs but introduced them on Linux.
Do you mean http://trac.webkit.org/changeset/188947? That's an accessibility change.
The revision number in my first post is a typo. These failures were introduced by r189947, not r188947.
These tests were actually flaky, but only passed rarely. Since enabling threaded compositor, they're passing about half the time. Updating expectations accordingly.