Go to: http://www.cnn.com/SHOWBIZ/ The categories in the sidebar should have a solid dark blue background. Instead there's some weird lighter blue mixed in. The sidebar looks totally garbled as a result. Shipping Safari looks fine.
This problem is also showing up on digg.com (using the latest nightly).
Created attachment 7851 [details] Reduction I have reduced this to: <head> <style> ul#nav li { background:url(http://i.a.cnn.net/cnn/.element/img/1.3/nav/nav.blue.gif) no-repeat; } </style> </head> <body> <ul id="nav"> <li></li> </ul> </body> If you look at the actual image that is used here, it is clear that we are squishing the whole thing to fit in the list marker, whereas before we would just display anything that fit.
Created attachment 7863 [details] Patch This patch fixes the regression and maintains the correct behavior for backgrounds with both background-size and no-repeat set. It passes all of the layout tests.
Comment on attachment 7863 [details] Patch For now I recommend we remove the new NO_REPEAT clause you added. I think we need way more tests in the tree to baseline against before we stop calling drawTiledImage in the NO_REPEAT case. I noticed scaledWidth and scaledHeight initialized to bogus values earlier in the code. You can remove those and just declare scaledWidth/Height right above your code that sets them to the right values. There's a bug on the Mac only in tileInRect where oneTileRect is being used to set the fromRect's size and destination should be used instead.
Disregard my comment about adjusting scaledWidth/Height. Beth explained how that worked on IRC, and I was wrong.
Created attachment 7864 [details] Still not right Can't figure out what is going wrong here...
Comment on attachment 7864 [details] Still not right This is right enough to fix the regressions though, and the tests that break don't have correct results even before the patch. We can land this and then file a followup bug about the failing test cases.
Created attachment 7866 [details] I think this works... Even though the other patch was review+ed, i think this one takes care of the background-size regressions that one would cause...
Comment on attachment 7866 [details] I think this works... Yes. That's great. You should make a background-origin/size combined test.
I committed the fix and added a few layout tests.
*** Bug 8518 has been marked as a duplicate of this bug. ***
*** Bug 8508 has been marked as a duplicate of this bug. ***