[Indic] Unbreak Khmer
For Khmer, all consonants are subjoining. No need to look in the font. We were looking in the wrong order anyway.
This commit is contained in:
parent
e0475345d5
commit
5249f3aee1
|
@ -116,9 +116,11 @@ would_substitute (hb_codepoint_t *glyphs, unsigned int glyphs_count,
|
|||
static indic_position_t
|
||||
consonant_position (hb_codepoint_t u, hb_ot_map_t *map, hb_font_t *font)
|
||||
{
|
||||
if ((u & ~0x007F) == 0x1780)
|
||||
return POS_BELOW_C; /* In Khmer coeng model, all are subjoining. */
|
||||
|
||||
hb_codepoint_t virama = (u & ~0x007F) | 0x004D;
|
||||
if ((u & ~0x007F) == 0x0D80) virama = 0x0DCA; /* Sinahla */
|
||||
if ((u & ~0x007F) == 0x1780) virama = 0x17D2; /* Khmer */
|
||||
hb_codepoint_t glyphs[2];
|
||||
|
||||
hb_font_get_glyph (font, virama, 0, &glyphs[0]);
|
||||
|
|
Loading…
Reference in New Issue