Created attachment 777296 [details] testcase ###!!! ASSERTION: degenerate transform while synthetic bolding: 'm != 0.0', file gfx/thebes/gfxFont.cpp, line 2272
This seems to arise during generation of the page thumbnail for Panorama, not the main page drawing. There's a regression in Nightly where synthetic bolding is no longer handled properly when generating thumbnails; something has changed about the results we're getting when calling gfxSize t = aContext->UserToDevice(gfxSize(1.0, 0.0)) in gfxFont::CalcXScale(), and that's what ends up hitting this assertion (and then a divide-by-zero error) in the case where content is rotated exactly 90°. I'll attach a testcase that demonstrates a visual glitch as a result of this, for rotations that approach 90°. The regression first appears in Nightly of 2013-07-15. OK: Nightly 2013-07-14 Built from http://hg.mozilla.org/mozilla-central/rev/18467a85acf6 Bad: Nightly 2013-07-15 Built from http://hg.mozilla.org/mozilla-central/rev/ff0a372e3170 http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=18467a85acf6&tochange=ff0a372e3170 Given that this seems to be Mac-only, I'm suspicious that bug 892966 may be involved, though I have not tried backing this out locally to confirm.
Created attachment 779288 [details] testcase for Panorama synthetic-bold regression This testcase avoids the exact 90° rotation, so it shouldn't assert or crash, but by opening the Panorama view (Cmd-Shift-E) you can see the synthetic-bolding problem: note the incorrect double-strike of the "a" letters that are rotated to a close-to-vertical orientation.
Yes, it is likely bug 892966. That somewhat unintentionally enables us using CoreGraphics Azure for thumbnail drawing. I expect I have also broken it for the non-drawwindow case with bug 896489.
Bug 906704 is fixing this.