[HarfBuzz] harfbuzz: Branch 'master'

[Behdad Esfahbod]

 src/hb-ot-shape-complex-indic.cc |   11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

New commits:
commit 508cc3d3cfcfb0383df0fe795cc28db4e0fd5729
Author: Behdad Esfahbod <behdad at behdad.org>
Date:   Thu Dec 17 17:31:17 2015 +0000

    [indic] Allow context when matching for Malayalam new-spec
    
    Test sequence:
    	U+0995,U+09CD,U+09B0
    
    With Nirmala shipped on Windows 10, this failed to form the below form.
    Works now.
    
    Reported by Sairus.

diff --git a/src/hb-ot-shape-complex-indic.cc b/src/hb-ot-shape-complex-indic.cc
index 5354897..a630419 100644
--- a/src/hb-ot-shape-complex-indic.cc
+++ b/src/hb-ot-shape-complex-indic.cc
@@ -557,8 +557,15 @@ data_create_indic (const hb_ot_shape_plan_t *plan)
   indic_plan->virama_glyph = (hb_codepoint_t) -1;
 
   /* Use zero-context would_substitute() matching for new-spec of the main
-   * Indic scripts, and scripts with one spec only, but not for old-specs. */
-  bool zero_context = !indic_plan->is_old_spec;
+   * Indic scripts, and scripts with one spec only, but not for old-specs.
+   * The new-spec for all dual-spec scripts says zero-context matching happens.
+   *
+   * However, testing with Malayalam shows that old and new spec both allow
+   * context.  Testing with Bengali new-spec however shows that it doesn't.
+   * So, the heuristic here is the way it is.  It should *only* be changed,
+   * as we discover more cases of what Windows does.  DON'T TOUCH OTHERWISE.
+   */
+  bool zero_context = !indic_plan->is_old_spec && plan->props.script != HB_SCRIPT_MALAYALAM;
   indic_plan->rphf.init (&plan->map, HB_TAG('r','p','h','f'), zero_context);
   indic_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), zero_context);
   indic_plan->blwf.init (&plan->map, HB_TAG('b','l','w','f'), zero_context);