Bug #258099 has regressed. See the testcase in attachment 235766 [details]. The Arabic text should appear as موزيلا in all cases.
WFM on Mac and Windows, apparently a Linux-only regression.
You're going to fix this right? :-)
Created attachment 287652 [details] [diff] [review] Bring back bidi overrides This brings back the bidi overrides removed in bug 362682.
With attachment 287652 [details] [diff] [review] the shaping is mostly correct, but there is one remaining issue: when there is an ltr directional override lam alef are shaped as individual connected characters, not as a lam-alef ligature. I think this may be a bug in the Pango shaper.
Yeah, it's a pango bug. One that is not trivial to fix, but I opened a bug upstream for it anyway: http://bugzilla.gnome.org/show_bug.cgi?id=494486
How are the text offsets being adjusted in this patch? Surely the mapping between glyphs and characters needs to be changed to account for the header?
The last hunk in the patch does that.
I don't see how. For an item which starts after the header, we don't change anything.
So how were the text offsets adjusted before bug 362682?
Comment on attachment 287652 [details] [diff] [review] Bring back bidi overrides Never mind, I get it now. The key is we don't really care what the offset in the UTF8 string is. It's the UTF16 offset that the textrun cares about, and that's tracked separately and is not affected by the header.
Checked in. I don't think that the Linux test boxes have the necessary fonts for Arabic reftests, but I can't be sure until bug 403012 is fixed.