[aat][Regression] 3c3eb5ea9ccf3 breaks accent placement on Times.ttc

[drott]

Bisecting identifies 3c3eb5e as breaking accent placement on an f + fi ligature with Times.ttc. The commit 3c3eb5e was intended as a fix for issue #1357 and did fix it, but also introduced this regression.

Good:

$ util/hb-shape --shaper=ot --font-funcs=ot /System/Library/Fonts/Times.ttc "Af̀fiZ" 
[A=0+1479|f=1+682|gravecmb=1@-480,588+0|fi=3+1139|Z=5+1251]

Bad:

$ util/hb-shape --shaper=ot --font-funcs=ot /System/Library/Fonts/Times.ttc "Af̀fiZ" 
[A=0+1479|f=1+682|gravecmb=1+682|fi=3+1139|Z=5+1251]

[drott]

Same results for Mac OS 10.13 Times.ttc and Mac OS 10.10 Times.ttc.

[khaledhosny]

What does Core Text output look like?

[drott]

$ ./hb-shape --shaper=coretext --font-funcs=ot /System/Library/Fonts/Times.ttc "Af̀fiZ" 
[A=0+1479|f=1+131|gravecmb=1@0,629+551|fi=3+1139|Z=5+1251]

[behdad]

Okay. So, let's tweak further. If morx is available, nor mark advance zeroing, but mark fallback positioning can still happen. That should handle both cases.

[behdad]

It's trickier to NOT adjust mark advances, but do mark positioning. Oh, I know. Okay, let me see. We know to NOT do mark positioning etc for Khmer, Indic, etc shapers. Maybe I reuse that knowledge.

[behdad]

Fixed.