I was unable to replicate this bug with valgrind and gr2fonttest (how I normally replicate such bugs). But a visual review of that area of the code has resulted in my tightening some checks. These have been pushed into the upstream repo. This bug will need to be retested when the patch with this repo version in is applied to your code. (repo version: 976 (d74a1988ba08))
What are our plans to update the mozilla version of Graphite2 to the latest upstream?
We've just updated Graphite2 in bug 753230. Christoph, could you re-test this with a build that includes that patch, and confirm whether it's fixed? Thanks!
This bug here is still reproducible.
Is the call stack still identical to comment #1, or have line numbers etc possibly shifted at all in the Graphite update? If you could post a fresh call stack to confirm exactly where it's occurring, I'll try to take a look.
Fix committed to upstream repository. Planning to do a release later today. BTW the latest valgrind has no exp-ptrcheck and exp-sgcheck doesn't trigger on this fault :(
Created attachment 628290 [details] [diff] [review] patch from upstream This is the relevant patch from the upstream repo - Christoph, if you have a chance to try this and confirm that it fixes the issue, that'd be great. (Not requesting review on this for now as I plan to take the complete new release when it's ready in the next day or so, but in the meantime any testing would be welcome.)
Created attachment 628304 [details] [diff] [review] update graphite2 to release 1.1.3 Version 1.1.3 has now been released upstream, so here's the patch to update our copy to the new version, which includes the fix for this bug.
Comment on attachment 628304 [details] [diff] [review] update graphite2 to release 1.1.3 Review of attachment 628304 [details] [diff] [review]: ----------------------------------------------------------------- Tested with the provieded patch, fixed!
Comment on attachment 628304 [details] [diff] [review] update graphite2 to release 1.1.3 If possible, having a crashtest for this would be nice.
https://hg.mozilla.org/integration/mozilla-inbound/rev/1446e5aba066 Not sure about a crashtest - the issue only showed up under ASAN, not as a crash with a normal browser build (though in principle it could've crashed if the out-of-bounds access happened to hit an unmapped page, or something like that). Maybe we should be running unit tests with ASAN builds, at least on a daily basis if not per-push? I expect that'd flush out some issues...
We need these fixes in Beta (Fx14), right? Please request approval on the patch unless it's too risky to consider. (This may not be the only bug or security bug fixed upstream).