Steps to reproduce: 1. load http://rng.io/?all and let it run 2. open Ring 1 3. open 2D Games 4. open CSS3 MediaQueries 5. look at the results Actual results: second and third tests fail Expected results: all tests pass What's happening here is that we're failing this null-check: http://hg.mozilla.org/mozilla-central/file/6fe7dd2f8f57/dom/base/nsGlobalWindow.cpp#l3817 because presContext is null. If I add a FlushPendingNotifications(Flush_Layout) call earlier in the method, the tests pass. The question is whether there's any weaker flush (e.g., flush on the parent) that would be sufficient to ensure that nsGlobalWindow::GetPresContext gives an up-to-date result.
I _think_ that a Flush_Style on the parent should be sufficient if we don't care about having the prescontext size up to date.
Er, I meant to do Flush_Frames and then I did Flush_Layout for some reason. I think we do need the size up-to-date, though, and I *think* something deeper in the code does a flush to ensure that, though I need to check.
OK. Worst case, a Flush_Layout on the parent would suffice; nsGlobalWindow even has an EnsureSizeUpToDate helper for that.
nsDOMMediaQueryList actually doesn't have any flushes, but almost certainly needs them
Created attachment 639487 [details] [diff] [review] patch Anyway, we should get this in. comment 4 should be a separate bug.
Comment on attachment 639487 [details] [diff] [review] patch r=me