Bug 116839 – Filtered element gets compositing layer unpredictably

• Component: Layout and Rendering
• Status: NEW

[Simon Fraser (smfr)]

Created attachment 202997 [details]
Testcase

Load the attachment with compositing layer borders enabled, and resize the window. Note that the filtered layer alternates between having a compositing layer and not, which should not happen.

[Dirk Schulze]

Can confirm that compositing is enabled and disabled on resize. Alex? Max?

[Max Vujovic]

(In reply to comment #1)
> Can confirm that compositing is enabled and disabled on resize. Alex? Max?

Confirmed as well. I'm playing around with it. 

The compositing state of the filtered layer changes every time the compositing requirements are computed. Compositing always changes like: 0->1->0->1, never 0->0->1->1. Still investigating.

[Max Vujovic]

Figured it out...

When the filtered layer is composited, it uses itself as the clippingRootForPainting. Its local clip rect is infinity, since the filtered layer has no overflow: hidden set. Thus, in RenderLayer::calculateLayerBounds, it returns its own bounds. These bounds do not overlap the composited layer, so the filtered layer changes to not composited.

When the filtered layer is not composited, it uses the nearest compositing container as the clippingRootForPainting. This container uses the local clip rect from the large parent element with overflow: hidden set. For the filtered layer, RenderLayer::calculateLayerBounds incorrectly assumes that this clip rect is smaller than the filtered layer bounds, and returns this clip rect. This large clip rect overlaps the composited layer, so the filtered layer changes back to composited.

I'll start working on a fix Monday next week (6/17) and post what I have then. However, I'm traveling for the rest of that week, so I might not get to finish up the patch until the week after that (week of 6/24).

[Max Vujovic]

Created attachment 204854 [details]
Patch

[Build Bot]

Comment on attachment 204854 [details]
Patch

Attachment 204854 [details] did not pass win-ews (win):
Output: http://webkit-queues.appspot.com/results/929183

New failing tests:
css3/filters/no-compositing-from-clipping-parent-layout-change.html

[Build Bot]

Created attachment 204886 [details]
Archive of layout-test-results from APPLE-EWS-3 for win-future

The attached test failures were seen while running run-webkit-tests on the win-ews.
Bot: APPLE-EWS-3  Port: win-future  Platform: CYGWIN_NT-6.1-WOW64-1.7.20-0.266-5-3-i686-32bit

[Build Bot]

Comment on attachment 204854 [details]
Patch

Attachment 204854 [details] did not pass win-ews (win):
Output: http://webkit-queues.appspot.com/results/921469

New failing tests:
css3/filters/no-compositing-from-clipping-parent-layout-change.html

[Build Bot]

Created attachment 204888 [details]
Archive of layout-test-results from APPLE-EWS-6 for win-future

The attached test failures were seen while running run-webkit-tests on the win-ews.
Bot: APPLE-EWS-6  Port: win-future  Platform: CYGWIN_NT-6.1-WOW64-1.7.20-0.266-5-3-i686-32bit

[Brady Eidson]

Comment on attachment 204854 [details]
Patch

Assuming that patches for review since 2013 are stale, r-

[Simon Fraser (smfr)]

Bug still reproduces.