543 bytes, text/html
Created attachment 8650979 [details] testcase that depends on intrinsic size of missing-image frame See testcase. Reload it repeatedly; the frame with the missing image will sometimes appear immediately outside its containing block, instead of inside it. When this happens, zooming in or out will fix the rendering. The <img> elements use position:absolute with no offsets, so they should remain at the "hypothetical box position" where they would be if they appeared within the flow. Because the containing block uses vertical-rl writing mode, the computation of the hypothetical box position depends on knowing the intrinsic width of the image frame. It seems that for the broken-image icon, we sometimes don't get that early enough, or don't reflow after the width becomes available. (First noticed in an Android reftest, where this led to a consistent failure; see bug 1183431 comments 37, 40. But the erratic behavior with this testcase occurs on desktop as well.)
(In reply to Jonathan Kew (:jfkthame) from comment #0) > See testcase. Reload it repeatedly; the frame with the missing image will > sometimes appear immediately outside its containing block, instead of inside > it. On my system, this happens more than "sometimes" -- it's nearly 100% of the time. (all but maybe 1 load, out of ~25 page loads that I just now did.) > When this happens, zooming in or out will fix the rendering. (I can confirm this, too.)
You need to log in before you can comment on or make changes to this bug.