Humm. Thinking about this more, and reading the original summary: http://lists.freedesktop.org/archives/harfbuzz/2013-April/003101.html
I'm NOT sure what we were thinking. The email message above clearly suggests that Uniscribe DOESN'T zero width while at the time we were. The Uniscribe part is consistent with my testing. I think we misunderstood the issue during the hackfest. The problem is, I couldn't reproduce the original problem before our faulty commit... bisecting and yeah, looks like we tried to "fix" the wrong problem at the hackfest. Reverted the faulty commit and all looks good. See commit log for details. behdad On 13-10-20 07:35 PM, Jonathan Kew wrote: > Turns out that zeroing marks by GDEF in Tibetan (commit > d5bd0590ae2fbc7b0dee86385a565aef00ffb835) has significantly regressed our > numbers for the Windows himalaya.ttf font. :( > > The main problem seems to be the handling of > > 0F3C;TIBETAN MARK ANG KHANG GYON;Ps;0;ON;;;;;Y;TIBETAN LEFT BRACE;;;; > 0F3D;TIBETAN MARK ANG KHANG GYAS;Pe;0;ON;;;;;Y;TIBETAN RIGHT BRACE;;;; > > both of which the GDEF table in himalaya.ttf classifies (incorrectly) as class > 3. Result: > > http://ec2-54-226-13-158.compute-1.amazonaws.com/testcase-view.html?file=out/fonts/win8/himalaya.ttf/texts/tibt/wikipedia/bo.txt.diffs#P_1_1_77 > > > and lots of similar examples where these characters are used. > > Sigh. > > JK > -- behdad http://behdad.org/ _______________________________________________ HarfBuzz mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/harfbuzz
