The landing of bug 653649 caused some kind invalidation regression which is affecting my work on the full-screen API (bug 545812). Before the landing of bug 653649, when an element requests full-screen, the window will go full-screen and the browser's frame will hide, and the toolbar will "slide up" out of sight. After the landing of bug 653649, the "slide up" doesn't invalidate properly (I assume), and you're left with artefacts/noise on the content that has "slid up". For an example, point an up-to-date nightly at http://pearce.org.nz/full-screen/ , set the full-screen-api.enabled pref to true, and hit the <button> labelled "Fullscreen". Observe toolbar sliding away, leaving behind rendering artefacts. [Mozilla/5.0 (Windows NT 6.1; WOW64; rv:9.0a1) Gecko/20110906 Firefox/9.0a1, GPU Accelerated Windows: 2/2 Direct3D 10]

I should also mention, the bug definitely manifests in revision d9797d99f5f7 (which is part 4/5 of the patches for Bug 653649), but not in revision 3fbd9eaf9deb (the changeset preceding Bug 653649's patches). Part 2/5 and 3/5 wouldn't build by themselves for me, and I didn't try building Part 1/5 as it looks like it doesn't change any existing code.

Does the patch in bug 684574 fix it?

Yes! Awesome!