1. Load http://people.mozilla.com/~nhirata/html_tp/elephants-dream.webm 2. Long press the video and choose "Full Screen" from the context menu. EXPECTED RESULT: The video should play fullscreen. ACTUAL RESULT: In a release build, the page will go gray. In a debug build, Fennec dies from fatal assert in nsLayoutUtils::InflationMinFontSizeFor(): // Check that neither this frame nor any of its ancestors are // currently being reflowed. // It's ok for box frames (but not arbitrary ancestors of box frames) // since they set their size before reflow. ###!!! ABORT: must call nsHTMLReflowState& version during reflow: '!(f->GetStateBits() & NS_FRAME_IN_REFLOW)', file /Users/cpeterson/Code/mozilla/birch/layout/base/nsLayoutUtils.cpp, line 4692 * dbaron says: I presume that the problem is that the nearest XUL frame is in a native anonymous subtree and therefore IsContainerForFontSizeInflation is returning false. Probably I should: (1) just change IsContainerForFontSizeInflation for that case (2) change nsLayoutUtils::InflationMinFontSizeFor(const nsHTMLReflowState&) so that it asserts for this case even when inflation is turned off

Comment on attachment 581086 [details] [diff] [review] patch >+ // ugly exception, but ok because the child should be Could we actually assert that? So (f->GetType() == nsGkAtoms::scrollFrame && IsContainerForFontSizeInflation(f->GetScrollFrame()) (probably with a relevant do_QueryFrame on f or something). r=me.

It turns out it's actually not true, because the whole thing could be inside native-anonymous content, e.g., inside of an input. And we really need nothing to be a container in that case -- but it's also safe in the is-a-reflow-root case. I bet I can revise the assertion to incorporate an NS_FRAME_IN_REFLOW check on the parent of the top of the reflow state chain.