[HarfBuzz] unbreaking mixed-up khmer fonts

[Jonathan Kew]

On 20/11/12 17:06, Behdad Esfahbod wrote:
> On 12-11-20 10:22 AM, Jonathan Kew wrote:
>>
>> I'm becoming increasingly convinced we need to bite the bullet and reintroduce
>> explicit Ra support for fonts that lack 'pref'. Sigh.
>
> Umm'k.  Walking in this morning I was thinking that maybe checking for 'pref'
> would do it.  But if not, then I'll bring back fallback 'pref'.  May not get
> to it till the weekend though.

Yeah, I got to thinking the same thing, and so far it's looking 
promising... instead of saying we *won't* use indic if there's liga, 
let's say we'll *only* use indic if there's pref.

AFAICT, using the attached patch, all the fonts on my test page render 
correctly (for the few sequences tested there, at least). I'll make a 
test build of Firefox and we can ask people to try it on real sites and 
let us know how it goes.

JK

-------------- next part --------------
From: Jonathan Kew <jkew at mozilla.com>

diff --git a/gfx/harfbuzz/src/hb-ot-shape-complex-private.hh b/gfx/harfbuzz/src/hb-ot-shape-complex-private.hh
--- a/gfx/harfbuzz/src/hb-ot-shape-complex-private.hh
+++ b/gfx/harfbuzz/src/hb-ot-shape-complex-private.hh
@@ -279,26 +279,30 @@ hb_ot_shape_complex_categorize (const hb
 
       /* Only use Indic shaper if the font has Indic tables. */
       if (planner->map.found_script[0])
 	return &_hb_ot_complex_shaper_indic;
       else
 	return &_hb_ot_complex_shaper_default;
 
     case HB_SCRIPT_KHMER:
-      /* If the font has 'liga', let the generic shaper do it. */
-      if (!planner->map.found_script[0] ||
-	  hb_ot_layout_language_find_feature (planner->face, HB_OT_TAG_GSUB,
-					      planner->map.script_index[0],
-					      planner->map.language_index[0],
-					      HB_TAG ('l','i','g','a'), NULL))
+      /* A number of Khmer fonts in the wild don't have a 'pref' feature,
+       * and as such won't shape properly via the Indic shaper;
+       * however, they typically have 'liga' / 'clig' features that implement
+       * the necessary "reordering" by means of ligature substitutions.
+       * So we send such pref-less fonts through the generic shaper instead. */
+      if (planner->map.found_script[0]) &&
+	(hb_ot_layout_language_find_feature (planner->face, HB_OT_TAG_GSUB,
+					     planner->map.script_index[0],
+					     planner->map.language_index[0],
+					     HB_TAG ('p','r','e','f'),
+					     NULL))
+	return &_hb_ot_complex_shaper_indic;
+      else
 	return &_hb_ot_complex_shaper_default;
-      else
-	return &_hb_ot_complex_shaper_indic;
-
 
     case HB_SCRIPT_MYANMAR:
       /* For Myanmar, we only want to use the Indic shaper if the "new" script
        * tag is found.  For "old" script tag we want to use the default shaper. */
       if (planner->map.chosen_script[0] == HB_TAG ('m','y','m','2'))
 	return &_hb_ot_complex_shaper_indic;
       else
 	return &_hb_ot_complex_shaper_default;