Consistently. Diff: --- /home/slave/webkitgtk/gtk-linux-64-release/build/layout-test-results/fast/events/autoscroll-in-textfield-expected.txt +++ /home/slave/webkitgtk/gtk-linux-64-release/build/layout-test-results/fast/events/autoscroll-in-textfield-actual.txt @@ -1,3 +1,3 @@ https://bugs.webkit.org/show_bug.cgi?id=20201 To do the test manually you have to try triggering the autoscroll by starting the dragging from within the text field and moving to the right. If the autoscroll occurs the test has PASSED. -PASSED +FAILED the textfield should have been scrolled I'll skip this one for now.
This is likely an issue where we don't allocate a RenderLayer and just lose the scroll offset information in this case. A render tree dump of the failing test case would definitely help here!
Would you have any advice on where to break in gdb? I'm not very familiar with the scroll code :(
(In reply to comment #2) > Would you have any advice on where to break in gdb? I'm not very familiar with the scroll code :( A failing tree dump would be super helpful regardless of debugging. That said, my theory is that we don't allocate the layer as we should. To confirm that, you could break into RenderBox::scrollLeft - as this is called from the test page - and check that hasOverflowClipWithLayer() is true. If it isn't, I would dive into RenderBlock::addOverflowFromChildren and see why RenderBox::addLayoutOverflow doesn't allocate the layer.
The tests covered by this bug were found to be passing consistently after moving from using the Skipped list to using test_expectations.txt. Their expectations were removed in r116122[1] (covered by bug #85591). Closing the bug. 1: http://trac.webkit.org/changeset/116122