I expect this bug to be fixed by bug 960564, but I wanted to have it on file so that I'd remember to check. Steps to reproduce: Load the attached testcase in a browser that has a lot of busy tabs open. Expected results: The inner box should traverse the four edges of the outer box, always in contact with an edge. Actual results: Sometimes the inner box traverses a path that's not quite along the edge of the outer box; its path to the next corner is starting from somewhere along the previous edge rather than from the previous corner. This is because we don't do anything that flushes before deciding what the starting value for the transition is. The proposal in bug 960564 should fix this.

Er, that should have been bug 960465.

Fixed by bug 960465 (or, very unlikely, something else that landed the same day).