The timeplot we use on the l10n dashboard doesn't show up correctly on aurora and nightly, but works fine on 10. https://l10n-stage-sj.mozilla.org/dashboard/tree-status/fx_aurora?&bound=7&hideBad shows that, the grey graph seems to be upside down. Filing in canvas as that's what's actually drawing, didn't do a ton of investigation myself so far.
m-c regression range seems to be http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=005488525c43&tochange=402b394b6623 Unfortunately, there's a huge number of changes in there. :(
Axel, are you on Mac? I can't seem to reproduce this on Linux, but I do see it on Mac...
Yes, I'm on mac.
Yeah, definitely not reproducible on Linux. I guess we get to bisect on Mac. :(
The first bad revision is: changeset: 85317:3c47bcef2ce4 user: Jeff Muizelaar <email@example.com> date: Tue Jan 24 15:14:50 2012 -0500 summary: Bug 720721. Properly transform the clip path in DrawTargetCG. r=mwoodrow
Jeff M - this looks like a novel regression in FF12 from some of your work. tracking-firefox12+ and can you tell us the cost of backing out the offending changeset, and/or suggest some other remedy?
Created attachment 599878 [details] [diff] [review] Another rememdy When we fixed transformed clips it caused us to double transform gradients. We fix this by avoiding ::PushClip when drawing gradients. This has the advantage of saving a save and restore pair and it makes Fill() more closely match Stroke()
Verified on nightly. Should this be uplifted?
Comment on attachment 599878 [details] [diff] [review] Another rememdy [Approval Request Comment] Regression caused by (bug #): 720721 User impact if declined: Gradients may draw in the wrong spot Testing completed (on m-c, etc.): on m-c, test case included Risk to taking this patch (and alternatives if risky): Low, the new code is easier to reason about than the old code String changes made by this patch: none
Comment on attachment 599878 [details] [diff] [review] Another rememdy [Triage Comment] Low risk, verified fix for Aurora 12.