Created attachment 570608 [details] testcase (hangs Firefox) ###!!! ASSERTION: Invalid offset: 'aOffset <= mSkipChars->mCharCount', file gfx/thebes/gfxSkipChars.cpp, line 92 ###!!! ASSERTION: Text run does not map enough text for our reflow: 'gfxSkipCharsIterator(iter).ConvertOriginalToSkipped(offset + length) <= mTextRun->GetLength()', file layout/generic/nsTextFrameThebes.cpp, line 7109 ###!!! ASSERTION: Invalid offset: 'aOffset <= mSkipChars->mCharCount', file gfx/thebes/gfxSkipChars.cpp, line 92 ###!!! ASSERTION: unconstrained height on totally empty line: 'NS_UNCONSTRAINEDSIZE != aFloatAvailableSpace.mRect.height', file layout/generic/nsBlockFrame.cpp, line 3686 ###!!! ASSERTION: redo line on totally empty line with non-empty band...: 'aFloatAvailableSpace.mHasFloats', file layout/generic/nsBlockFrame.cpp, line 3691
Is this exploitable or just a DoS/hang? would more/fewer/different characters change the symptoms?
smontagu - any update here?
Created attachment 588753 [details] [diff] [review] Patch This is apparently regression from making a preformatted newline a new bidi paragraph: we need to test whether there are any fluid continuations even beyond the current paragraph, but children of the same block element and convert them to non-fluid.
Do we need to land this on Aurora as a security bug or did it turn out benign and we can unhide the bug? If it really is a potentially sg:critical bug we maybe shouldn't have called it out in the Platform meeting, especially with a checked-in crashtest.
Regression window is http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=a666b4f809f0&tochange=953f9620f395, which is not what I expected: possibly bug 578977?
Comment on attachment 588753 [details] [diff] [review] Patch This doesn't seem exploitable to me, but since we have a simple patch let's fix it on aurora and beta. [Approval Request Comment] Regression caused by (bug #): 578977(? see comment 8) User impact if declined: hangs and possible crashes Testing completed (on m-c, etc.): baked 6 days on m-c, has crashtest. Risk to taking this patch (and alternatives if risky): minimal.
(In reply to Simon Montagu from comment #8) > possibly bug 578977? Confirmed by reverting bug 578977 in an Aurora debug build on Linux. Bug 578977 did change some things that affects next-in-flows of first-letter frames, so it makes sense.
given comment 9 I think we can hold off on this given timing for 10. But we should be able to land this on Aurora, we can cover that in the next triage/channel mtg.
Comment on attachment 588753 [details] [diff] [review] Patch [Triage Comment] We don't have any evidence that this is exploitable or causing significant user pain - let's let this ride the train.
Verified using linux debug build on Ubuntu 11.10 with beta debug build (18.4.2012)- no hang/assertion when loading the test case from comment 0.