[HarfBuzz] harfbuzz: Branch 'master'

[Behdad Esfahbod]

 src/hb-ot-layout-gsubgpos-private.hh |   24 +++++++++++++++++++++++-
 1 file changed, 23 insertions(+), 1 deletion(-)

New commits:
commit 9ac9af725c781f4597e8db46cf330d595c072739
Author: Behdad Esfahbod <behdad at behdad.org>
Date:   Sun Mar 5 13:51:01 2017 -0800

    Add TODO item

diff --git a/src/hb-ot-layout-gsubgpos-private.hh b/src/hb-ot-layout-gsubgpos-private.hh
index 244c395..005ed21 100644
--- a/src/hb-ot-layout-gsubgpos-private.hh
+++ b/src/hb-ot-layout-gsubgpos-private.hh
@@ -996,7 +996,29 @@ static inline bool apply_lookup (hb_apply_context_t *c,
     if (!delta)
         continue;
 
-    /* Recursed lookup changed buffer len.  Adjust. */
+    /* Recursed lookup changed buffer len.  Adjust.
+     *
+     * TODO:
+     *
+     * Right now, if buffer length increased by n, we assume n new glyphs
+     * were added right after the current position, and if buffer length
+     * was decreased by n, we assume n match positions after the current
+     * one where removed.  The former (buffer length increased) case is
+     * fine, but the decrease case can be improved in at least two ways,
+     * both of which are significant:
+     *
+     *   - If recursed-to lookup is MultipleSubst and buffer length
+     *     decreased, then it's current match position that was deleted,
+     *     NOT the one after it.
+     *
+     *   - If buffer length was decreased by n, it does not necessarily
+     *     mean that n match positions where removed, as there might
+     *     have been marks and default-ignorables in the sequence.  We
+     *     should instead drop match positions between current-position
+     *     and current-position + n instead.
+     *
+     * It should be possible to construct tests for both of these cases.
+     */
 
     end += delta;
     if (end < int (match_positions[idx]))