michael@0: /*
michael@0:  * Copyright © 2011,2012  Google, Inc.
michael@0:  *
michael@0:  *  This is part of HarfBuzz, a text shaping library.
michael@0:  *
michael@0:  * Permission is hereby granted, without written agreement and without
michael@0:  * license or royalty fees, to use, copy, modify, and distribute this
michael@0:  * software and its documentation for any purpose, provided that the
michael@0:  * above copyright notice and the following two paragraphs appear in
michael@0:  * all copies of this software.
michael@0:  *
michael@0:  * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
michael@0:  * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
michael@0:  * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
michael@0:  * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
michael@0:  * DAMAGE.
michael@0:  *
michael@0:  * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
michael@0:  * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
michael@0:  * FITNESS FOR A PARTICULAR PURPOSE.  THE SOFTWARE PROVIDED HEREUNDER IS
michael@0:  * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
michael@0:  * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
michael@0:  *
michael@0:  * Google Author(s): Behdad Esfahbod
michael@0:  */
michael@0: 
michael@0: #include "hb-ot-shape-complex-indic-private.hh"
michael@0: #include "hb-ot-layout-private.hh"
michael@0: 
michael@0: /* buffer var allocations */
michael@0: #define indic_category() complex_var_u8_0() /* indic_category_t */
michael@0: #define indic_position() complex_var_u8_1() /* indic_position_t */
michael@0: 
michael@0: 
michael@0: /*
michael@0:  * Indic shaper.
michael@0:  */
michael@0: 
michael@0: 
michael@0: #define IN_HALF_BLOCK(u, Base) (((u) & ~0x7F) == (Base))
michael@0: 
michael@0: #define IS_DEVA(u) (IN_HALF_BLOCK (u, 0x0900))
michael@0: #define IS_BENG(u) (IN_HALF_BLOCK (u, 0x0980))
michael@0: #define IS_GURU(u) (IN_HALF_BLOCK (u, 0x0A00))
michael@0: #define IS_GUJR(u) (IN_HALF_BLOCK (u, 0x0A80))
michael@0: #define IS_ORYA(u) (IN_HALF_BLOCK (u, 0x0B00))
michael@0: #define IS_TAML(u) (IN_HALF_BLOCK (u, 0x0B80))
michael@0: #define IS_TELU(u) (IN_HALF_BLOCK (u, 0x0C00))
michael@0: #define IS_KNDA(u) (IN_HALF_BLOCK (u, 0x0C80))
michael@0: #define IS_MLYM(u) (IN_HALF_BLOCK (u, 0x0D00))
michael@0: #define IS_SINH(u) (IN_HALF_BLOCK (u, 0x0D80))
michael@0: #define IS_KHMR(u) (IN_HALF_BLOCK (u, 0x1780))
michael@0: 
michael@0: 
michael@0: #define MATRA_POS_LEFT(u)	POS_PRE_M
michael@0: #define MATRA_POS_RIGHT(u)	( \
michael@0: 				  IS_DEVA(u) ? POS_AFTER_SUB  : \
michael@0: 				  IS_BENG(u) ? POS_AFTER_POST : \
michael@0: 				  IS_GURU(u) ? POS_AFTER_POST : \
michael@0: 				  IS_GUJR(u) ? POS_AFTER_POST : \
michael@0: 				  IS_ORYA(u) ? POS_AFTER_POST : \
michael@0: 				  IS_TAML(u) ? POS_AFTER_POST : \
michael@0: 				  IS_TELU(u) ? (u <= 0x0C42 ? POS_BEFORE_SUB : POS_AFTER_SUB) : \
michael@0: 				  IS_KNDA(u) ? (u < 0x0CC3 || u > 0xCD6 ? POS_BEFORE_SUB : POS_AFTER_SUB) : \
michael@0: 				  IS_MLYM(u) ? POS_AFTER_POST : \
michael@0: 				  IS_SINH(u) ? POS_AFTER_SUB  : \
michael@0: 				  IS_KHMR(u) ? POS_AFTER_POST : \
michael@0: 				  /*default*/  POS_AFTER_SUB    \
michael@0: 				)
michael@0: #define MATRA_POS_TOP(u)	( /* BENG and MLYM don't have top matras. */ \
michael@0: 				  IS_DEVA(u) ? POS_AFTER_SUB  : \
michael@0: 				  IS_GURU(u) ? POS_AFTER_POST : /* Deviate from spec */ \
michael@0: 				  IS_GUJR(u) ? POS_AFTER_SUB  : \
michael@0: 				  IS_ORYA(u) ? POS_AFTER_MAIN : \
michael@0: 				  IS_TAML(u) ? POS_AFTER_SUB  : \
michael@0: 				  IS_TELU(u) ? POS_BEFORE_SUB : \
michael@0: 				  IS_KNDA(u) ? POS_BEFORE_SUB : \
michael@0: 				  IS_SINH(u) ? POS_AFTER_SUB  : \
michael@0: 				  IS_KHMR(u) ? POS_AFTER_POST : \
michael@0: 				  /*default*/  POS_AFTER_SUB    \
michael@0: 				)
michael@0: #define MATRA_POS_BOTTOM(u)	( \
michael@0: 				  IS_DEVA(u) ? POS_AFTER_SUB  : \
michael@0: 				  IS_BENG(u) ? POS_AFTER_SUB  : \
michael@0: 				  IS_GURU(u) ? POS_AFTER_POST : \
michael@0: 				  IS_GUJR(u) ? POS_AFTER_POST : \
michael@0: 				  IS_ORYA(u) ? POS_AFTER_SUB  : \
michael@0: 				  IS_TAML(u) ? POS_AFTER_POST : \
michael@0: 				  IS_TELU(u) ? POS_BEFORE_SUB : \
michael@0: 				  IS_KNDA(u) ? POS_BEFORE_SUB : \
michael@0: 				  IS_MLYM(u) ? POS_AFTER_POST : \
michael@0: 				  IS_SINH(u) ? POS_AFTER_SUB  : \
michael@0: 				  IS_KHMR(u) ? POS_AFTER_POST : \
michael@0: 				  /*default*/  POS_AFTER_SUB    \
michael@0: 				)
michael@0: 
michael@0: static inline indic_position_t
michael@0: matra_position (hb_codepoint_t u, indic_position_t side)
michael@0: {
michael@0:   switch ((int) side)
michael@0:   {
michael@0:     case POS_PRE_C:	return MATRA_POS_LEFT (u);
michael@0:     case POS_POST_C:	return MATRA_POS_RIGHT (u);
michael@0:     case POS_ABOVE_C:	return MATRA_POS_TOP (u);
michael@0:     case POS_BELOW_C:	return MATRA_POS_BOTTOM (u);
michael@0:   };
michael@0:   return side;
michael@0: }
michael@0: 
michael@0: /* XXX
michael@0:  * This is a hack for now.  We should move this data into the main Indic table.
michael@0:  * Or completely remove it and just check in the tables.
michael@0:  */
michael@0: static const hb_codepoint_t ra_chars[] = {
michael@0:   0x0930, /* Devanagari */
michael@0:   0x09B0, /* Bengali */
michael@0:   0x09F0, /* Bengali */
michael@0:   0x0A30, /* Gurmukhi */	/* No Reph */
michael@0:   0x0AB0, /* Gujarati */
michael@0:   0x0B30, /* Oriya */
michael@0:   0x0BB0, /* Tamil */		/* No Reph */
michael@0:   0x0C30, /* Telugu */		/* Reph formed only with ZWJ */
michael@0:   0x0CB0, /* Kannada */
michael@0:   0x0D30, /* Malayalam */	/* No Reph, Logical Repha */
michael@0: 
michael@0:   0x0DBB, /* Sinhala */		/* Reph formed only with ZWJ */
michael@0: 
michael@0:   0x179A, /* Khmer */		/* No Reph, Visual Repha */
michael@0: };
michael@0: 
michael@0: static inline bool
michael@0: is_ra (hb_codepoint_t u)
michael@0: {
michael@0:   for (unsigned int i = 0; i < ARRAY_LENGTH (ra_chars); i++)
michael@0:     if (u == ra_chars[i])
michael@0:       return true;
michael@0:   return false;
michael@0: }
michael@0: 
michael@0: static inline bool
michael@0: is_one_of (const hb_glyph_info_t &info, unsigned int flags)
michael@0: {
michael@0:   /* If it ligated, all bets are off. */
michael@0:   if (_hb_glyph_info_ligated (&info)) return false;
michael@0:   return !!(FLAG (info.indic_category()) & flags);
michael@0: }
michael@0: 
michael@0: #define JOINER_FLAGS (FLAG (OT_ZWJ) | FLAG (OT_ZWNJ))
michael@0: static inline bool
michael@0: is_joiner (const hb_glyph_info_t &info)
michael@0: {
michael@0:   return is_one_of (info, JOINER_FLAGS);
michael@0: }
michael@0: 
michael@0: #define MEDIAL_FLAGS (FLAG (OT_CM) | FLAG (OT_CM2))
michael@0: 
michael@0: /* Note:
michael@0:  *
michael@0:  * We treat Vowels and placeholders as if they were consonants.  This is safe because Vowels
michael@0:  * cannot happen in a consonant syllable.  The plus side however is, we can call the
michael@0:  * consonant syllable logic from the vowel syllable function and get it all right! */
michael@0: #define CONSONANT_FLAGS (FLAG (OT_C) | FLAG (OT_Ra) | MEDIAL_FLAGS | FLAG (OT_V) | FLAG (OT_NBSP) | FLAG (OT_DOTTEDCIRCLE))
michael@0: static inline bool
michael@0: is_consonant (const hb_glyph_info_t &info)
michael@0: {
michael@0:   return is_one_of (info, CONSONANT_FLAGS);
michael@0: }
michael@0: 
michael@0: #define HALANT_OR_COENG_FLAGS (FLAG (OT_H) | FLAG (OT_Coeng))
michael@0: static inline bool
michael@0: is_halant_or_coeng (const hb_glyph_info_t &info)
michael@0: {
michael@0:   return is_one_of (info, HALANT_OR_COENG_FLAGS);
michael@0: }
michael@0: 
michael@0: static inline void
michael@0: set_indic_properties (hb_glyph_info_t &info)
michael@0: {
michael@0:   hb_codepoint_t u = info.codepoint;
michael@0:   unsigned int type = hb_indic_get_categories (u);
michael@0:   indic_category_t cat = (indic_category_t) (type & 0x7F);
michael@0:   indic_position_t pos = (indic_position_t) (type >> 8);
michael@0: 
michael@0: 
michael@0:   /*
michael@0:    * Re-assign category
michael@0:    */
michael@0: 
michael@0: 
michael@0:   /* The spec says U+0952 is OT_A.  However, testing shows that Uniscribe
michael@0:    * treats U+0951..U+0954 all behave similarly.
michael@0:    * TESTS:
michael@0:    * U+092E,U+0947,U+0952
michael@0:    * U+092E,U+0952,U+0947
michael@0:    * U+092E,U+0947,U+0951
michael@0:    * U+092E,U+0951,U+0947
michael@0:    */
michael@0:   if (unlikely (hb_in_range<hb_codepoint_t> (u, 0x0951, 0x0954)))
michael@0:     cat = OT_A;
michael@0: 
michael@0:   if (unlikely (u == 0x17D1))
michael@0:     cat = OT_X;
michael@0:   if (cat == OT_X &&
michael@0:       unlikely (hb_in_range<hb_codepoint_t> (u, 0x17CB, 0x17D3))) /* Khmer Various signs */
michael@0:   {
michael@0:     /* These are like Top Matras. */
michael@0:     cat = OT_M;
michael@0:     pos = POS_ABOVE_C;
michael@0:   }
michael@0:   if (u == 0x17C6) /* Khmer Bindu doesn't like to be repositioned. */
michael@0:     cat = OT_N;
michael@0: 
michael@0:   if (unlikely (u == 0x17D2)) cat = OT_Coeng; /* Khmer coeng */
michael@0:   else if (unlikely (u == 0x200C)) cat = OT_ZWNJ;
michael@0:   else if (unlikely (u == 0x200D)) cat = OT_ZWJ;
michael@0:   else if (unlikely (u == 0x25CC)) cat = OT_DOTTEDCIRCLE;
michael@0:   else if (unlikely (u == 0x0A71)) cat = OT_SM; /* GURMUKHI ADDAK.  Move it to the end. */
michael@0:   else if (unlikely (u == 0xA982)) cat = OT_SM; /* Javanese repha. */
michael@0:   else if (unlikely (u == 0xA9BE)) cat = OT_CM2; /* Javanese medial ya. */
michael@0:   else if (unlikely (u == 0xA9BD)) { cat = OT_M; pos = POS_POST_C; } /* Javanese vocalic r. */
michael@0: 
michael@0:   if (cat == OT_Repha) {
michael@0:     /* There are two kinds of characters marked as Repha:
michael@0:      * - The ones that are GenCat=Mn are already positioned visually, ie. after base. (eg. Khmer)
michael@0:      * - The ones that are GenCat=Lo is encoded logically, ie. beginning of syllable. (eg. Malayalam)
michael@0:      *
michael@0:      * We recategorize the first kind to look like a Nukta and attached to the base directly.
michael@0:      */
michael@0:     if (_hb_glyph_info_get_general_category (&info) == HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK)
michael@0:       cat = OT_N;
michael@0:   }
michael@0: 
michael@0: 
michael@0: 
michael@0:   /*
michael@0:    * Re-assign position.
michael@0:    */
michael@0: 
michael@0:   if ((FLAG (cat) & CONSONANT_FLAGS))
michael@0:   {
michael@0:     pos = POS_BASE_C;
michael@0:     if (is_ra (u))
michael@0:       cat = OT_Ra;
michael@0:   }
michael@0:   else if (cat == OT_M)
michael@0:   {
michael@0:     pos = matra_position (u, pos);
michael@0:   }
michael@0:   else if ((FLAG (cat) & (FLAG (OT_SM) | FLAG (OT_VD) | FLAG (OT_A) | FLAG (OT_Avag))))
michael@0:   {
michael@0:     pos = POS_SMVD;
michael@0:   }
michael@0: 
michael@0:   if (unlikely (u == 0x0B01)) pos = POS_BEFORE_SUB; /* Oriya Bindu is BeforeSub in the spec. */
michael@0: 
michael@0: 
michael@0: 
michael@0:   info.indic_category() = cat;
michael@0:   info.indic_position() = pos;
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Things above this line should ideally be moved to the Indic table itself.
michael@0:  */
michael@0: 
michael@0: 
michael@0: /*
michael@0:  * Indic configurations.  Note that we do not want to keep every single script-specific
michael@0:  * behavior in these tables necessarily.  This should mainly be used for per-script
michael@0:  * properties that are cheaper keeping here, than in the code.  Ie. if, say, one and
michael@0:  * only one script has an exception, that one script can be if'ed directly in the code,
michael@0:  * instead of adding a new flag in these structs.
michael@0:  */
michael@0: 
michael@0: enum base_position_t {
michael@0:   BASE_POS_FIRST,
michael@0:   BASE_POS_LAST_SINHALA,
michael@0:   BASE_POS_LAST
michael@0: };
michael@0: enum reph_position_t {
michael@0:   REPH_POS_AFTER_MAIN  = POS_AFTER_MAIN,
michael@0:   REPH_POS_BEFORE_SUB  = POS_BEFORE_SUB,
michael@0:   REPH_POS_AFTER_SUB   = POS_AFTER_SUB,
michael@0:   REPH_POS_BEFORE_POST = POS_BEFORE_POST,
michael@0:   REPH_POS_AFTER_POST  = POS_AFTER_POST,
michael@0:   REPH_POS_DONT_CARE   = POS_RA_TO_BECOME_REPH
michael@0: };
michael@0: enum reph_mode_t {
michael@0:   REPH_MODE_IMPLICIT,  /* Reph formed out of initial Ra,H sequence. */
michael@0:   REPH_MODE_EXPLICIT,  /* Reph formed out of initial Ra,H,ZWJ sequence. */
michael@0:   REPH_MODE_VIS_REPHA, /* Encoded Repha character, no reordering needed. */
michael@0:   REPH_MODE_LOG_REPHA  /* Encoded Repha character, needs reordering. */
michael@0: };
michael@0: enum blwf_mode_t {
michael@0:   BLWF_MODE_PRE_AND_POST, /* Below-forms feature applied to pre-base and post-base. */
michael@0:   BLWF_MODE_POST_ONLY     /* Below-forms feature applied to post-base only. */
michael@0: };
michael@0: enum pref_len_t {
michael@0:   PREF_LEN_1 = 1,
michael@0:   PREF_LEN_2 = 2,
michael@0:   PREF_LEN_DONT_CARE = PREF_LEN_2
michael@0: };
michael@0: struct indic_config_t
michael@0: {
michael@0:   hb_script_t     script;
michael@0:   bool            has_old_spec;
michael@0:   hb_codepoint_t  virama;
michael@0:   base_position_t base_pos;
michael@0:   reph_position_t reph_pos;
michael@0:   reph_mode_t     reph_mode;
michael@0:   blwf_mode_t     blwf_mode;
michael@0:   pref_len_t      pref_len;
michael@0: };
michael@0: 
michael@0: static const indic_config_t indic_configs[] =
michael@0: {
michael@0:   /* Default.  Should be first. */
michael@0:   {HB_SCRIPT_INVALID,	false,     0,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_1},
michael@0:   {HB_SCRIPT_DEVANAGARI,true, 0x094D,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
michael@0:   {HB_SCRIPT_BENGALI,	true, 0x09CD,BASE_POS_LAST, REPH_POS_AFTER_SUB,  REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
michael@0:   {HB_SCRIPT_GURMUKHI,	true, 0x0A4D,BASE_POS_LAST, REPH_POS_BEFORE_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
michael@0:   {HB_SCRIPT_GUJARATI,	true, 0x0ACD,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
michael@0:   {HB_SCRIPT_ORIYA,	true, 0x0B4D,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
michael@0:   {HB_SCRIPT_TAMIL,	true, 0x0BCD,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
michael@0:   {HB_SCRIPT_TELUGU,	true, 0x0C4D,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_EXPLICIT, BLWF_MODE_POST_ONLY,    PREF_LEN_2},
michael@0:   {HB_SCRIPT_KANNADA,	true, 0x0CCD,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_POST_ONLY,    PREF_LEN_2},
michael@0:   {HB_SCRIPT_MALAYALAM,	true, 0x0D4D,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_LOG_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
michael@0:   {HB_SCRIPT_SINHALA,	false,0x0DCA,BASE_POS_LAST_SINHALA,
michael@0: 						    REPH_POS_AFTER_MAIN, REPH_MODE_EXPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
michael@0:   {HB_SCRIPT_KHMER,	false,0x17D2,BASE_POS_FIRST,REPH_POS_DONT_CARE,  REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
michael@0:   {HB_SCRIPT_JAVANESE,	false,0xA9C0,BASE_POS_FIRST,REPH_POS_DONT_CARE,  REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_1},
michael@0: };
michael@0: 
michael@0: 
michael@0: 
michael@0: /*
michael@0:  * Indic shaper.
michael@0:  */
michael@0: 
michael@0: struct feature_list_t {
michael@0:   hb_tag_t tag;
michael@0:   hb_ot_map_feature_flags_t flags;
michael@0: };
michael@0: 
michael@0: static const feature_list_t
michael@0: indic_features[] =
michael@0: {
michael@0:   /*
michael@0:    * Basic features.
michael@0:    * These features are applied in order, one at a time, after initial_reordering.
michael@0:    */
michael@0:   {HB_TAG('n','u','k','t'), F_GLOBAL},
michael@0:   {HB_TAG('a','k','h','n'), F_GLOBAL},
michael@0:   {HB_TAG('r','p','h','f'), F_NONE},
michael@0:   {HB_TAG('r','k','r','f'), F_GLOBAL},
michael@0:   {HB_TAG('p','r','e','f'), F_NONE},
michael@0:   {HB_TAG('b','l','w','f'), F_NONE},
michael@0:   {HB_TAG('a','b','v','f'), F_NONE},
michael@0:   {HB_TAG('h','a','l','f'), F_NONE},
michael@0:   {HB_TAG('p','s','t','f'), F_NONE},
michael@0:   {HB_TAG('v','a','t','u'), F_GLOBAL},
michael@0:   {HB_TAG('c','j','c','t'), F_GLOBAL},
michael@0:   {HB_TAG('c','f','a','r'), F_NONE},
michael@0:   /*
michael@0:    * Other features.
michael@0:    * These features are applied all at once, after final_reordering.
michael@0:    * Default Bengali font in Windows for example has intermixed
michael@0:    * lookups for init,pres,abvs,blws features.
michael@0:    */
michael@0:   {HB_TAG('i','n','i','t'), F_NONE},
michael@0:   {HB_TAG('p','r','e','s'), F_GLOBAL},
michael@0:   {HB_TAG('a','b','v','s'), F_GLOBAL},
michael@0:   {HB_TAG('b','l','w','s'), F_GLOBAL},
michael@0:   {HB_TAG('p','s','t','s'), F_GLOBAL},
michael@0:   {HB_TAG('h','a','l','n'), F_GLOBAL},
michael@0:   /* Positioning features, though we don't care about the types. */
michael@0:   {HB_TAG('d','i','s','t'), F_GLOBAL},
michael@0:   {HB_TAG('a','b','v','m'), F_GLOBAL},
michael@0:   {HB_TAG('b','l','w','m'), F_GLOBAL},
michael@0: };
michael@0: 
michael@0: /*
michael@0:  * Must be in the same order as the indic_features array.
michael@0:  */
michael@0: enum {
michael@0:   _NUKT,
michael@0:   _AKHN,
michael@0:   RPHF,
michael@0:   _RKRF,
michael@0:   PREF,
michael@0:   BLWF,
michael@0:   ABVF,
michael@0:   HALF,
michael@0:   PSTF,
michael@0:   _VATU,
michael@0:   _CJCT,
michael@0:   CFAR,
michael@0: 
michael@0:   INIT,
michael@0:   _PRES,
michael@0:   _ABVS,
michael@0:   _BLWS,
michael@0:   _PSTS,
michael@0:   _HALN,
michael@0:   _DIST,
michael@0:   _ABVM,
michael@0:   _BLWM,
michael@0: 
michael@0:   INDIC_NUM_FEATURES,
michael@0:   INDIC_BASIC_FEATURES = INIT /* Don't forget to update this! */
michael@0: };
michael@0: 
michael@0: static void
michael@0: setup_syllables (const hb_ot_shape_plan_t *plan,
michael@0: 		 hb_font_t *font,
michael@0: 		 hb_buffer_t *buffer);
michael@0: static void
michael@0: initial_reordering (const hb_ot_shape_plan_t *plan,
michael@0: 		    hb_font_t *font,
michael@0: 		    hb_buffer_t *buffer);
michael@0: static void
michael@0: final_reordering (const hb_ot_shape_plan_t *plan,
michael@0: 		  hb_font_t *font,
michael@0: 		  hb_buffer_t *buffer);
michael@0: static void
michael@0: clear_syllables (const hb_ot_shape_plan_t *plan,
michael@0: 		 hb_font_t *font,
michael@0: 		 hb_buffer_t *buffer);
michael@0: 
michael@0: static void
michael@0: collect_features_indic (hb_ot_shape_planner_t *plan)
michael@0: {
michael@0:   hb_ot_map_builder_t *map = &plan->map;
michael@0: 
michael@0:   /* Do this before any lookups have been applied. */
michael@0:   map->add_gsub_pause (setup_syllables);
michael@0: 
michael@0:   map->add_global_bool_feature (HB_TAG('l','o','c','l'));
michael@0:   /* The Indic specs do not require ccmp, but we apply it here since if
michael@0:    * there is a use of it, it's typically at the beginning. */
michael@0:   map->add_global_bool_feature (HB_TAG('c','c','m','p'));
michael@0: 
michael@0: 
michael@0:   unsigned int i = 0;
michael@0:   map->add_gsub_pause (initial_reordering);
michael@0:   for (; i < INDIC_BASIC_FEATURES; i++) {
michael@0:     map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ);
michael@0:     map->add_gsub_pause (NULL);
michael@0:   }
michael@0:   map->add_gsub_pause (final_reordering);
michael@0:   for (; i < INDIC_NUM_FEATURES; i++) {
michael@0:     map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ);
michael@0:   }
michael@0: 
michael@0:   map->add_global_bool_feature (HB_TAG('c','a','l','t'));
michael@0:   map->add_global_bool_feature (HB_TAG('c','l','i','g'));
michael@0: 
michael@0:   map->add_gsub_pause (clear_syllables);
michael@0: }
michael@0: 
michael@0: static void
michael@0: override_features_indic (hb_ot_shape_planner_t *plan)
michael@0: {
michael@0:   /* Uniscribe does not apply 'kern' in Khmer. */
michael@0:   if (hb_options ().uniscribe_bug_compatible)
michael@0:   {
michael@0:     switch ((hb_tag_t) plan->props.script)
michael@0:     {
michael@0:       case HB_SCRIPT_KHMER:
michael@0: 	plan->map.add_feature (HB_TAG('k','e','r','n'), 0, F_GLOBAL);
michael@0: 	break;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL);
michael@0: }
michael@0: 
michael@0: 
michael@0: struct would_substitute_feature_t
michael@0: {
michael@0:   inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_)
michael@0:   {
michael@0:     zero_context = zero_context_;
michael@0:     map->get_stage_lookups (0/*GSUB*/,
michael@0: 			    map->get_feature_stage (0/*GSUB*/, feature_tag),
michael@0: 			    &lookups, &count);
michael@0:   }
michael@0: 
michael@0:   inline bool would_substitute (const hb_codepoint_t *glyphs,
michael@0: 				unsigned int          glyphs_count,
michael@0: 				hb_face_t            *face) const
michael@0:   {
michael@0:     for (unsigned int i = 0; i < count; i++)
michael@0:       if (hb_ot_layout_lookup_would_substitute_fast (face, lookups[i].index, glyphs, glyphs_count, zero_context))
michael@0: 	return true;
michael@0:     return false;
michael@0:   }
michael@0: 
michael@0:   private:
michael@0:   const hb_ot_map_t::lookup_map_t *lookups;
michael@0:   unsigned int count;
michael@0:   bool zero_context;
michael@0: };
michael@0: 
michael@0: struct indic_shape_plan_t
michael@0: {
michael@0:   ASSERT_POD ();
michael@0: 
michael@0:   inline bool get_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const
michael@0:   {
michael@0:     hb_codepoint_t glyph = virama_glyph;
michael@0:     if (unlikely (virama_glyph == (hb_codepoint_t) -1))
michael@0:     {
michael@0:       if (!config->virama || !font->get_glyph (config->virama, 0, &glyph))
michael@0: 	glyph = 0;
michael@0:       /* Technically speaking, the spec says we should apply 'locl' to virama too.
michael@0:        * Maybe one day... */
michael@0: 
michael@0:       /* Our get_glyph() function needs a font, so we can't get the virama glyph
michael@0:        * during shape planning...  Instead, overwrite it here.  It's safe.  Don't worry! */
michael@0:       (const_cast<indic_shape_plan_t *> (this))->virama_glyph = glyph;
michael@0:     }
michael@0: 
michael@0:     *pglyph = glyph;
michael@0:     return glyph != 0;
michael@0:   }
michael@0: 
michael@0:   const indic_config_t *config;
michael@0: 
michael@0:   bool is_old_spec;
michael@0:   hb_codepoint_t virama_glyph;
michael@0: 
michael@0:   would_substitute_feature_t rphf;
michael@0:   would_substitute_feature_t pref;
michael@0:   would_substitute_feature_t blwf;
michael@0:   would_substitute_feature_t pstf;
michael@0: 
michael@0:   hb_mask_t mask_array[INDIC_NUM_FEATURES];
michael@0: };
michael@0: 
michael@0: static void *
michael@0: data_create_indic (const hb_ot_shape_plan_t *plan)
michael@0: {
michael@0:   indic_shape_plan_t *indic_plan = (indic_shape_plan_t *) calloc (1, sizeof (indic_shape_plan_t));
michael@0:   if (unlikely (!indic_plan))
michael@0:     return NULL;
michael@0: 
michael@0:   indic_plan->config = &indic_configs[0];
michael@0:   for (unsigned int i = 1; i < ARRAY_LENGTH (indic_configs); i++)
michael@0:     if (plan->props.script == indic_configs[i].script) {
michael@0:       indic_plan->config = &indic_configs[i];
michael@0:       break;
michael@0:     }
michael@0: 
michael@0:   indic_plan->is_old_spec = indic_plan->config->has_old_spec && ((plan->map.chosen_script[0] & 0x000000FF) != '2');
michael@0:   indic_plan->virama_glyph = (hb_codepoint_t) -1;
michael@0: 
michael@0:   /* Use zero-context would_substitute() matching for new-spec of the main
michael@0:    * Indic scripts, but not for old-spec or scripts with one spec only. */
michael@0:   bool zero_context = indic_plan->config->has_old_spec || !indic_plan->is_old_spec;
michael@0:   indic_plan->rphf.init (&plan->map, HB_TAG('r','p','h','f'), zero_context);
michael@0:   indic_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), zero_context);
michael@0:   indic_plan->blwf.init (&plan->map, HB_TAG('b','l','w','f'), zero_context);
michael@0:   indic_plan->pstf.init (&plan->map, HB_TAG('p','s','t','f'), zero_context);
michael@0: 
michael@0:   for (unsigned int i = 0; i < ARRAY_LENGTH (indic_plan->mask_array); i++)
michael@0:     indic_plan->mask_array[i] = (indic_features[i].flags & F_GLOBAL) ?
michael@0: 				 0 : plan->map.get_1_mask (indic_features[i].tag);
michael@0: 
michael@0:   return indic_plan;
michael@0: }
michael@0: 
michael@0: static void
michael@0: data_destroy_indic (void *data)
michael@0: {
michael@0:   free (data);
michael@0: }
michael@0: 
michael@0: static indic_position_t
michael@0: consonant_position_from_face (const indic_shape_plan_t *indic_plan,
michael@0: 			      const hb_codepoint_t consonant,
michael@0: 			      const hb_codepoint_t virama,
michael@0: 			      hb_face_t *face)
michael@0: {
michael@0:   /* For old-spec, the order of glyphs is Consonant,Virama,
michael@0:    * whereas for new-spec, it's Virama,Consonant.  However,
michael@0:    * some broken fonts (like Free Sans) simply copied lookups
michael@0:    * from old-spec to new-spec without modification.
michael@0:    * And oddly enough, Uniscribe seems to respect those lookups.
michael@0:    * Eg. in the sequence U+0924,U+094D,U+0930, Uniscribe finds
michael@0:    * base at 0.  The font however, only has lookups matching
michael@0:    * 930,94D in 'blwf', not the expected 94D,930 (with new-spec
michael@0:    * table).  As such, we simply match both sequences.  Seems
michael@0:    * to work. */
michael@0:   hb_codepoint_t glyphs[3] = {virama, consonant, virama};
michael@0:   if (indic_plan->blwf.would_substitute (glyphs  , 2, face) ||
michael@0:       indic_plan->blwf.would_substitute (glyphs+1, 2, face))
michael@0:     return POS_BELOW_C;
michael@0:   if (indic_plan->pstf.would_substitute (glyphs  , 2, face) ||
michael@0:       indic_plan->pstf.would_substitute (glyphs+1, 2, face))
michael@0:     return POS_POST_C;
michael@0:   unsigned int pref_len = indic_plan->config->pref_len;
michael@0:   if ((pref_len == PREF_LEN_2 &&
michael@0:        (indic_plan->pref.would_substitute (glyphs  , 2, face) ||
michael@0:         indic_plan->pref.would_substitute (glyphs+1, 2, face)))
michael@0:    || (pref_len == PREF_LEN_1 &&
michael@0:        indic_plan->pref.would_substitute (glyphs+1, 1, face)))
michael@0:     return POS_POST_C;
michael@0:   return POS_BASE_C;
michael@0: }
michael@0: 
michael@0: 
michael@0: enum syllable_type_t {
michael@0:   consonant_syllable,
michael@0:   vowel_syllable,
michael@0:   standalone_cluster,
michael@0:   avagraha_cluster,
michael@0:   broken_cluster,
michael@0:   non_indic_cluster,
michael@0: };
michael@0: 
michael@0: #include "hb-ot-shape-complex-indic-machine.hh"
michael@0: 
michael@0: 
michael@0: static void
michael@0: setup_masks_indic (const hb_ot_shape_plan_t *plan HB_UNUSED,
michael@0: 		   hb_buffer_t              *buffer,
michael@0: 		   hb_font_t                *font HB_UNUSED)
michael@0: {
michael@0:   HB_BUFFER_ALLOCATE_VAR (buffer, indic_category);
michael@0:   HB_BUFFER_ALLOCATE_VAR (buffer, indic_position);
michael@0: 
michael@0:   /* We cannot setup masks here.  We save information about characters
michael@0:    * and setup masks later on in a pause-callback. */
michael@0: 
michael@0:   unsigned int count = buffer->len;
michael@0:   for (unsigned int i = 0; i < count; i++)
michael@0:     set_indic_properties (buffer->info[i]);
michael@0: }
michael@0: 
michael@0: static void
michael@0: setup_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
michael@0: 		 hb_font_t *font HB_UNUSED,
michael@0: 		 hb_buffer_t *buffer)
michael@0: {
michael@0:   find_syllables (buffer);
michael@0: }
michael@0: 
michael@0: static int
michael@0: compare_indic_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
michael@0: {
michael@0:   int a = pa->indic_position();
michael@0:   int b = pb->indic_position();
michael@0: 
michael@0:   return a < b ? -1 : a == b ? 0 : +1;
michael@0: }
michael@0: 
michael@0: 
michael@0: 
michael@0: static void
michael@0: update_consonant_positions (const hb_ot_shape_plan_t *plan,
michael@0: 			    hb_font_t         *font,
michael@0: 			    hb_buffer_t       *buffer)
michael@0: {
michael@0:   const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
michael@0: 
michael@0:   if (indic_plan->config->base_pos != BASE_POS_LAST)
michael@0:     return;
michael@0: 
michael@0:   hb_codepoint_t virama;
michael@0:   if (indic_plan->get_virama_glyph (font, &virama))
michael@0:   {
michael@0:     hb_face_t *face = font->face;
michael@0:     unsigned int count = buffer->len;
michael@0:     for (unsigned int i = 0; i < count; i++)
michael@0:       if (buffer->info[i].indic_position() == POS_BASE_C) {
michael@0: 	hb_codepoint_t consonant = buffer->info[i].codepoint;
michael@0: 	buffer->info[i].indic_position() = consonant_position_from_face (indic_plan, consonant, virama, face);
michael@0:       }
michael@0:   }
michael@0: }
michael@0: 
michael@0: 
michael@0: /* Rules from:
michael@0:  * https://www.microsoft.com/typography/otfntdev/devanot/shaping.aspx */
michael@0: 
michael@0: static void
michael@0: initial_reordering_consonant_syllable (const hb_ot_shape_plan_t *plan,
michael@0: 				       hb_face_t *face,
michael@0: 				       hb_buffer_t *buffer,
michael@0: 				       unsigned int start, unsigned int end)
michael@0: {
michael@0:   const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
michael@0:   hb_glyph_info_t *info = buffer->info;
michael@0: 
michael@0: 
michael@0:   /* 1. Find base consonant:
michael@0:    *
michael@0:    * The shaping engine finds the base consonant of the syllable, using the
michael@0:    * following algorithm: starting from the end of the syllable, move backwards
michael@0:    * until a consonant is found that does not have a below-base or post-base
michael@0:    * form (post-base forms have to follow below-base forms), or that is not a
michael@0:    * pre-base reordering Ra, or arrive at the first consonant. The consonant
michael@0:    * stopped at will be the base.
michael@0:    *
michael@0:    *   o If the syllable starts with Ra + Halant (in a script that has Reph)
michael@0:    *     and has more than one consonant, Ra is excluded from candidates for
michael@0:    *     base consonants.
michael@0:    */
michael@0: 
michael@0:   unsigned int base = end;
michael@0:   bool has_reph = false;
michael@0: 
michael@0:   {
michael@0:     /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
michael@0:      *    and has more than one consonant, Ra is excluded from candidates for
michael@0:      *    base consonants. */
michael@0:     unsigned int limit = start;
michael@0:     if (indic_plan->config->reph_pos != REPH_POS_DONT_CARE &&
michael@0: 	indic_plan->mask_array[RPHF] &&
michael@0: 	start + 3 <= end &&
michael@0: 	(
michael@0: 	 (indic_plan->config->reph_mode == REPH_MODE_IMPLICIT && !is_joiner (info[start + 2])) ||
michael@0: 	 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT && info[start + 2].indic_category() == OT_ZWJ)
michael@0: 	))
michael@0:     {
michael@0:       /* See if it matches the 'rphf' feature. */
michael@0:       hb_codepoint_t glyphs[2] = {info[start].codepoint, info[start + 1].codepoint};
michael@0:       if (indic_plan->rphf.would_substitute (glyphs, ARRAY_LENGTH (glyphs), face))
michael@0:       {
michael@0: 	limit += 2;
michael@0: 	while (limit < end && is_joiner (info[limit]))
michael@0: 	  limit++;
michael@0: 	base = start;
michael@0: 	has_reph = true;
michael@0:       }
michael@0:     } else if (indic_plan->config->reph_mode == REPH_MODE_LOG_REPHA && info[start].indic_category() == OT_Repha)
michael@0:     {
michael@0: 	limit += 1;
michael@0: 	while (limit < end && is_joiner (info[limit]))
michael@0: 	  limit++;
michael@0: 	base = start;
michael@0: 	has_reph = true;
michael@0:     }
michael@0: 
michael@0:     switch (indic_plan->config->base_pos)
michael@0:     {
michael@0:       default:
michael@0:         assert (false);
michael@0: 	/* fallthrough */
michael@0: 
michael@0:       case BASE_POS_LAST:
michael@0:       {
michael@0: 	/* -> starting from the end of the syllable, move backwards */
michael@0: 	unsigned int i = end;
michael@0: 	bool seen_below = false;
michael@0: 	do {
michael@0: 	  i--;
michael@0: 	  /* -> until a consonant is found */
michael@0: 	  if (is_consonant (info[i]))
michael@0: 	  {
michael@0: 	    /* -> that does not have a below-base or post-base form
michael@0: 	     * (post-base forms have to follow below-base forms), */
michael@0: 	    if (info[i].indic_position() != POS_BELOW_C &&
michael@0: 		(info[i].indic_position() != POS_POST_C || seen_below))
michael@0: 	    {
michael@0: 	      base = i;
michael@0: 	      break;
michael@0: 	    }
michael@0: 	    if (info[i].indic_position() == POS_BELOW_C)
michael@0: 	      seen_below = true;
michael@0: 
michael@0: 	    /* -> or that is not a pre-base reordering Ra,
michael@0: 	     *
michael@0: 	     * IMPLEMENTATION NOTES:
michael@0: 	     *
michael@0: 	     * Our pre-base reordering Ra's are marked POS_POST_C, so will be skipped
michael@0: 	     * by the logic above already.
michael@0: 	     */
michael@0: 
michael@0: 	    /* -> or arrive at the first consonant. The consonant stopped at will
michael@0: 	     * be the base. */
michael@0: 	    base = i;
michael@0: 	  }
michael@0: 	  else
michael@0: 	  {
michael@0: 	    /* A ZWJ after a Halant stops the base search, and requests an explicit
michael@0: 	     * half form.
michael@0: 	     * A ZWJ before a Halant, requests a subjoined form instead, and hence
michael@0: 	     * search continues.  This is particularly important for Bengali
michael@0: 	     * sequence Ra,H,Ya that should form Ya-Phalaa by subjoining Ya. */
michael@0: 	    if (start < i &&
michael@0: 		info[i].indic_category() == OT_ZWJ &&
michael@0: 		info[i - 1].indic_category() == OT_H)
michael@0: 	      break;
michael@0: 	  }
michael@0: 	} while (i > limit);
michael@0:       }
michael@0:       break;
michael@0: 
michael@0:       case BASE_POS_LAST_SINHALA:
michael@0:       {
michael@0:         /* Sinhala base positioning is slightly different from main Indic, in that:
michael@0: 	 * 1. It's ZWJ behavior is different,
michael@0: 	 * 2. We don't need to look into the font for consonant positions.
michael@0: 	 */
michael@0: 
michael@0: 	if (!has_reph)
michael@0: 	  base = limit;
michael@0: 
michael@0: 	/* Find the last base consonant that is not blocked by ZWJ.  If there is
michael@0: 	 * a ZWJ right before a base consonant, that would request a subjoined form. */
michael@0: 	for (unsigned int i = limit; i < end; i++)
michael@0: 	  if (is_consonant (info[i]))
michael@0: 	  {
michael@0: 	    if (limit < i && info[i - 1].indic_category() == OT_ZWJ)
michael@0: 	      break;
michael@0: 	    else
michael@0: 	      base = i;
michael@0: 	  }
michael@0: 
michael@0: 	/* Mark all subsequent consonants as below. */
michael@0: 	for (unsigned int i = base + 1; i < end; i++)
michael@0: 	  if (is_consonant (info[i]))
michael@0: 	    info[i].indic_position() = POS_BELOW_C;
michael@0:       }
michael@0:       break;
michael@0: 
michael@0:       case BASE_POS_FIRST:
michael@0:       {
michael@0: 	/* The first consonant is always the base. */
michael@0: 
michael@0: 	assert (indic_plan->config->reph_mode == REPH_MODE_VIS_REPHA);
michael@0: 	assert (!has_reph);
michael@0: 
michael@0: 	base = start;
michael@0: 
michael@0: 	/* Mark all subsequent consonants as below. */
michael@0: 	for (unsigned int i = base + 1; i < end; i++)
michael@0: 	  if (is_consonant (info[i]))
michael@0: 	    info[i].indic_position() = POS_BELOW_C;
michael@0:       }
michael@0:       break;
michael@0:     }
michael@0: 
michael@0:     /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
michael@0:      *    and has more than one consonant, Ra is excluded from candidates for
michael@0:      *    base consonants.
michael@0:      *
michael@0:      *  Only do this for unforced Reph. (ie. not for Ra,H,ZWJ. */
michael@0:     if (has_reph && base == start && limit - base <= 2) {
michael@0:       /* Have no other consonant, so Reph is not formed and Ra becomes base. */
michael@0:       has_reph = false;
michael@0:     }
michael@0:   }
michael@0: 
michael@0: 
michael@0:   /* 2. Decompose and reorder Matras:
michael@0:    *
michael@0:    * Each matra and any syllable modifier sign in the cluster are moved to the
michael@0:    * appropriate position relative to the consonant(s) in the cluster. The
michael@0:    * shaping engine decomposes two- or three-part matras into their constituent
michael@0:    * parts before any repositioning. Matra characters are classified by which
michael@0:    * consonant in a conjunct they have affinity for and are reordered to the
michael@0:    * following positions:
michael@0:    *
michael@0:    *   o Before first half form in the syllable
michael@0:    *   o After subjoined consonants
michael@0:    *   o After post-form consonant
michael@0:    *   o After main consonant (for above marks)
michael@0:    *
michael@0:    * IMPLEMENTATION NOTES:
michael@0:    *
michael@0:    * The normalize() routine has already decomposed matras for us, so we don't
michael@0:    * need to worry about that.
michael@0:    */
michael@0: 
michael@0: 
michael@0:   /* 3.  Reorder marks to canonical order:
michael@0:    *
michael@0:    * Adjacent nukta and halant or nukta and vedic sign are always repositioned
michael@0:    * if necessary, so that the nukta is first.
michael@0:    *
michael@0:    * IMPLEMENTATION NOTES:
michael@0:    *
michael@0:    * We don't need to do this: the normalize() routine already did this for us.
michael@0:    */
michael@0: 
michael@0: 
michael@0:   /* Reorder characters */
michael@0: 
michael@0:   for (unsigned int i = start; i < base; i++)
michael@0:     info[i].indic_position() = MIN (POS_PRE_C, (indic_position_t) info[i].indic_position());
michael@0: 
michael@0:   if (base < end)
michael@0:     info[base].indic_position() = POS_BASE_C;
michael@0: 
michael@0:   /* Mark final consonants.  A final consonant is one appearing after a matra,
michael@0:    * like in Khmer. */
michael@0:   for (unsigned int i = base + 1; i < end; i++)
michael@0:     if (info[i].indic_category() == OT_M) {
michael@0:       for (unsigned int j = i + 1; j < end; j++)
michael@0:         if (is_consonant (info[j])) {
michael@0: 	  info[j].indic_position() = POS_FINAL_C;
michael@0: 	  break;
michael@0: 	}
michael@0:       break;
michael@0:     }
michael@0: 
michael@0:   /* Handle beginning Ra */
michael@0:   if (has_reph)
michael@0:     info[start].indic_position() = POS_RA_TO_BECOME_REPH;
michael@0: 
michael@0:   /* For old-style Indic script tags, move the first post-base Halant after
michael@0:    * last consonant.  Only do this if there is *not* a Halant after last
michael@0:    * consonant.  Otherwise it becomes messy. */
michael@0:   if (indic_plan->is_old_spec) {
michael@0:     for (unsigned int i = base + 1; i < end; i++)
michael@0:       if (info[i].indic_category() == OT_H) {
michael@0:         unsigned int j;
michael@0:         for (j = end - 1; j > i; j--)
michael@0: 	  if (is_consonant (info[j]) || info[j].indic_category() == OT_H)
michael@0: 	    break;
michael@0: 	if (info[j].indic_category() != OT_H && j > i) {
michael@0: 	  /* Move Halant to after last consonant. */
michael@0: 	  hb_glyph_info_t t = info[i];
michael@0: 	  memmove (&info[i], &info[i + 1], (j - i) * sizeof (info[0]));
michael@0: 	  info[j] = t;
michael@0: 	}
michael@0:         break;
michael@0:       }
michael@0:   }
michael@0: 
michael@0:   /* Attach misc marks to previous char to move with them. */
michael@0:   {
michael@0:     indic_position_t last_pos = POS_START;
michael@0:     for (unsigned int i = start; i < end; i++)
michael@0:     {
michael@0:       if ((FLAG (info[i].indic_category()) & (JOINER_FLAGS | FLAG (OT_N) | FLAG (OT_RS) | MEDIAL_FLAGS | HALANT_OR_COENG_FLAGS)))
michael@0:       {
michael@0: 	info[i].indic_position() = last_pos;
michael@0: 	if (unlikely (info[i].indic_category() == OT_H &&
michael@0: 		      info[i].indic_position() == POS_PRE_M))
michael@0: 	{
michael@0: 	  /*
michael@0: 	   * Uniscribe doesn't move the Halant with Left Matra.
michael@0: 	   * TEST: U+092B,U+093F,U+094DE
michael@0: 	   * We follow.  This is important for the Sinhala
michael@0: 	   * U+0DDA split matra since it decomposes to U+0DD9,U+0DCA
michael@0: 	   * where U+0DD9 is a left matra and U+0DCA is the virama.
michael@0: 	   * We don't want to move the virama with the left matra.
michael@0: 	   * TEST: U+0D9A,U+0DDA
michael@0: 	   */
michael@0: 	  for (unsigned int j = i; j > start; j--)
michael@0: 	    if (info[j - 1].indic_position() != POS_PRE_M) {
michael@0: 	      info[i].indic_position() = info[j - 1].indic_position();
michael@0: 	      break;
michael@0: 	    }
michael@0: 	}
michael@0:       } else if (info[i].indic_position() != POS_SMVD) {
michael@0:         last_pos = (indic_position_t) info[i].indic_position();
michael@0:       }
michael@0:     }
michael@0:   }
michael@0:   /* For post-base consonants let them own anything before them
michael@0:    * since the last consonant or matra. */
michael@0:   {
michael@0:     unsigned int last = base;
michael@0:     for (unsigned int i = base + 1; i < end; i++)
michael@0:       if (is_consonant (info[i]))
michael@0:       {
michael@0: 	for (unsigned int j = last + 1; j < i; j++)
michael@0: 	  if (info[j].indic_position() < POS_SMVD)
michael@0: 	    info[j].indic_position() = info[i].indic_position();
michael@0: 	last = i;
michael@0:       } else if (info[i].indic_category() == OT_M)
michael@0:         last = i;
michael@0:   }
michael@0: 
michael@0: 
michael@0:   {
michael@0:     /* Use syllable() for sort accounting temporarily. */
michael@0:     unsigned int syllable = info[start].syllable();
michael@0:     for (unsigned int i = start; i < end; i++)
michael@0:       info[i].syllable() = i - start;
michael@0: 
michael@0:     /* Sit tight, rock 'n roll! */
michael@0:     hb_bubble_sort (info + start, end - start, compare_indic_order);
michael@0:     /* Find base again */
michael@0:     base = end;
michael@0:     for (unsigned int i = start; i < end; i++)
michael@0:       if (info[i].indic_position() == POS_BASE_C)
michael@0:       {
michael@0: 	base = i;
michael@0: 	break;
michael@0:       }
michael@0:     /* Things are out-of-control for post base positions, they may shuffle
michael@0:      * around like crazy.  In old-spec mode, we move halants around, so in
michael@0:      * that case merge all clusters after base.  Otherwise, check the sort
michael@0:      * order and merge as needed.
michael@0:      * For pre-base stuff, we handle cluster issues in final reordering. */
michael@0:     if (indic_plan->is_old_spec || end - base > 127)
michael@0:       buffer->merge_clusters (base, end);
michael@0:     else
michael@0:     {
michael@0:       /* Note!  syllable() is a one-byte field. */
michael@0:       for (unsigned int i = base; i < end; i++)
michael@0:         if (info[i].syllable() != 255)
michael@0: 	{
michael@0: 	  unsigned int max = i;
michael@0: 	  unsigned int j = start + info[i].syllable();
michael@0: 	  while (j != i)
michael@0: 	  {
michael@0: 	    max = MAX (max, j);
michael@0: 	    unsigned int next = start + info[j].syllable();
michael@0: 	    info[j].syllable() = 255; /* So we don't process j later again. */
michael@0: 	    j = next;
michael@0: 	  }
michael@0: 	  if (i != max)
michael@0: 	    buffer->merge_clusters (i, max + 1);
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     /* Put syllable back in. */
michael@0:     for (unsigned int i = start; i < end; i++)
michael@0:       info[i].syllable() = syllable;
michael@0:   }
michael@0: 
michael@0:   /* Setup masks now */
michael@0: 
michael@0:   {
michael@0:     hb_mask_t mask;
michael@0: 
michael@0:     /* Reph */
michael@0:     for (unsigned int i = start; i < end && info[i].indic_position() == POS_RA_TO_BECOME_REPH; i++)
michael@0:       info[i].mask |= indic_plan->mask_array[RPHF];
michael@0: 
michael@0:     /* Pre-base */
michael@0:     mask = indic_plan->mask_array[HALF];
michael@0:     if (!indic_plan->is_old_spec &&
michael@0: 	indic_plan->config->blwf_mode == BLWF_MODE_PRE_AND_POST)
michael@0:       mask |= indic_plan->mask_array[BLWF];
michael@0:     for (unsigned int i = start; i < base; i++)
michael@0:       info[i].mask  |= mask;
michael@0:     /* Base */
michael@0:     mask = 0;
michael@0:     if (base < end)
michael@0:       info[base].mask |= mask;
michael@0:     /* Post-base */
michael@0:     mask = indic_plan->mask_array[BLWF] | indic_plan->mask_array[ABVF] | indic_plan->mask_array[PSTF];
michael@0:     for (unsigned int i = base + 1; i < end; i++)
michael@0:       info[i].mask  |= mask;
michael@0:   }
michael@0: 
michael@0:   if (indic_plan->is_old_spec &&
michael@0:       buffer->props.script == HB_SCRIPT_DEVANAGARI)
michael@0:   {
michael@0:     /* Old-spec eye-lash Ra needs special handling.  From the
michael@0:      * spec:
michael@0:      *
michael@0:      * "The feature 'below-base form' is applied to consonants
michael@0:      * having below-base forms and following the base consonant.
michael@0:      * The exception is vattu, which may appear below half forms
michael@0:      * as well as below the base glyph. The feature 'below-base
michael@0:      * form' will be applied to all such occurrences of Ra as well."
michael@0:      *
michael@0:      * Test case: U+0924,U+094D,U+0930,U+094d,U+0915
michael@0:      * with Sanskrit 2003 font.
michael@0:      *
michael@0:      * However, note that Ra,Halant,ZWJ is the correct way to
michael@0:      * request eyelash form of Ra, so we wouldbn't inhibit it
michael@0:      * in that sequence.
michael@0:      *
michael@0:      * Test case: U+0924,U+094D,U+0930,U+094d,U+200D,U+0915
michael@0:      */
michael@0:     for (unsigned int i = start; i + 1 < base; i++)
michael@0:       if (info[i  ].indic_category() == OT_Ra &&
michael@0: 	  info[i+1].indic_category() == OT_H  &&
michael@0: 	  (i + 2 == base ||
michael@0: 	   info[i+2].indic_category() != OT_ZWJ))
michael@0:       {
michael@0: 	info[i  ].mask |= indic_plan->mask_array[BLWF];
michael@0: 	info[i+1].mask |= indic_plan->mask_array[BLWF];
michael@0:       }
michael@0:   }
michael@0: 
michael@0:   unsigned int pref_len = indic_plan->config->pref_len;
michael@0:   if (indic_plan->mask_array[PREF] && base + pref_len < end)
michael@0:   {
michael@0:     assert (1 <= pref_len && pref_len <= 2);
michael@0:     /* Find a Halant,Ra sequence and mark it for pre-base reordering processing. */
michael@0:     for (unsigned int i = base + 1; i + pref_len - 1 < end; i++) {
michael@0:       hb_codepoint_t glyphs[2];
michael@0:       for (unsigned int j = 0; j < pref_len; j++)
michael@0:         glyphs[j] = info[i + j].codepoint;
michael@0:       if (indic_plan->pref.would_substitute (glyphs, pref_len, face))
michael@0:       {
michael@0: 	for (unsigned int j = 0; j < pref_len; j++)
michael@0: 	  info[i++].mask |= indic_plan->mask_array[PREF];
michael@0: 
michael@0: 	/* Mark the subsequent stuff with 'cfar'.  Used in Khmer.
michael@0: 	 * Read the feature spec.
michael@0: 	 * This allows distinguishing the following cases with MS Khmer fonts:
michael@0: 	 * U+1784,U+17D2,U+179A,U+17D2,U+1782
michael@0: 	 * U+1784,U+17D2,U+1782,U+17D2,U+179A
michael@0: 	 */
michael@0: 	if (indic_plan->mask_array[CFAR])
michael@0: 	  for (; i < end; i++)
michael@0: 	    info[i].mask |= indic_plan->mask_array[CFAR];
michael@0: 
michael@0: 	break;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   /* Apply ZWJ/ZWNJ effects */
michael@0:   for (unsigned int i = start + 1; i < end; i++)
michael@0:     if (is_joiner (info[i])) {
michael@0:       bool non_joiner = info[i].indic_category() == OT_ZWNJ;
michael@0:       unsigned int j = i;
michael@0: 
michael@0:       do {
michael@0: 	j--;
michael@0: 
michael@0: 	/* ZWJ/ZWNJ should disable CJCT.  They do that by simply
michael@0: 	 * being there, since we don't skip them for the CJCT
michael@0: 	 * feature (ie. F_MANUAL_ZWJ) */
michael@0: 
michael@0: 	/* A ZWNJ disables HALF. */
michael@0: 	if (non_joiner)
michael@0: 	  info[j].mask &= ~indic_plan->mask_array[HALF];
michael@0: 
michael@0:       } while (j > start && !is_consonant (info[j]));
michael@0:     }
michael@0: }
michael@0: 
michael@0: 
michael@0: static void
michael@0: initial_reordering_vowel_syllable (const hb_ot_shape_plan_t *plan,
michael@0: 				   hb_face_t *face,
michael@0: 				   hb_buffer_t *buffer,
michael@0: 				   unsigned int start, unsigned int end)
michael@0: {
michael@0:   /* We made the vowels look like consonants.  So let's call the consonant logic! */
michael@0:   initial_reordering_consonant_syllable (plan, face, buffer, start, end);
michael@0: }
michael@0: 
michael@0: static void
michael@0: initial_reordering_standalone_cluster (const hb_ot_shape_plan_t *plan,
michael@0: 				       hb_face_t *face,
michael@0: 				       hb_buffer_t *buffer,
michael@0: 				       unsigned int start, unsigned int end)
michael@0: {
michael@0:   /* We treat NBSP/dotted-circle as if they are consonants, so we should just chain.
michael@0:    * Only if not in compatibility mode that is... */
michael@0: 
michael@0:   if (hb_options ().uniscribe_bug_compatible)
michael@0:   {
michael@0:     /* For dotted-circle, this is what Uniscribe does:
michael@0:      * If dotted-circle is the last glyph, it just does nothing.
michael@0:      * Ie. It doesn't form Reph. */
michael@0:     if (buffer->info[end - 1].indic_category() == OT_DOTTEDCIRCLE)
michael@0:       return;
michael@0:   }
michael@0: 
michael@0:   initial_reordering_consonant_syllable (plan, face, buffer, start, end);
michael@0: }
michael@0: 
michael@0: static void
michael@0: initial_reordering_broken_cluster (const hb_ot_shape_plan_t *plan,
michael@0: 				   hb_face_t *face,
michael@0: 				   hb_buffer_t *buffer,
michael@0: 				   unsigned int start, unsigned int end)
michael@0: {
michael@0:   /* We already inserted dotted-circles, so just call the standalone_cluster. */
michael@0:   initial_reordering_standalone_cluster (plan, face, buffer, start, end);
michael@0: }
michael@0: 
michael@0: static void
michael@0: initial_reordering_avagraha_cluster (const hb_ot_shape_plan_t *plan HB_UNUSED,
michael@0: 				     hb_face_t *face HB_UNUSED,
michael@0: 				     hb_buffer_t *buffer HB_UNUSED,
michael@0: 				     unsigned int start HB_UNUSED, unsigned int end HB_UNUSED)
michael@0: {
michael@0:   /* Nothing to do right now.  If we ever switch to using the output
michael@0:    * buffer in the reordering process, we'd need to next_glyph() here. */
michael@0: }
michael@0: 
michael@0: static void
michael@0: initial_reordering_non_indic_cluster (const hb_ot_shape_plan_t *plan HB_UNUSED,
michael@0: 				      hb_face_t *face HB_UNUSED,
michael@0: 				      hb_buffer_t *buffer HB_UNUSED,
michael@0: 				      unsigned int start HB_UNUSED, unsigned int end HB_UNUSED)
michael@0: {
michael@0:   /* Nothing to do right now.  If we ever switch to using the output
michael@0:    * buffer in the reordering process, we'd need to next_glyph() here. */
michael@0: }
michael@0: 
michael@0: 
michael@0: static void
michael@0: initial_reordering_syllable (const hb_ot_shape_plan_t *plan,
michael@0: 			     hb_face_t *face,
michael@0: 			     hb_buffer_t *buffer,
michael@0: 			     unsigned int start, unsigned int end)
michael@0: {
michael@0:   syllable_type_t syllable_type = (syllable_type_t) (buffer->info[start].syllable() & 0x0F);
michael@0:   switch (syllable_type) {
michael@0:   case consonant_syllable:	initial_reordering_consonant_syllable (plan, face, buffer, start, end); return;
michael@0:   case vowel_syllable:		initial_reordering_vowel_syllable     (plan, face, buffer, start, end); return;
michael@0:   case standalone_cluster:	initial_reordering_standalone_cluster (plan, face, buffer, start, end); return;
michael@0:   case avagraha_cluster:	initial_reordering_avagraha_cluster   (plan, face, buffer, start, end); return;
michael@0:   case broken_cluster:		initial_reordering_broken_cluster     (plan, face, buffer, start, end); return;
michael@0:   case non_indic_cluster:	initial_reordering_non_indic_cluster  (plan, face, buffer, start, end); return;
michael@0:   }
michael@0: }
michael@0: 
michael@0: static inline void
michael@0: insert_dotted_circles (const hb_ot_shape_plan_t *plan HB_UNUSED,
michael@0: 		       hb_font_t *font,
michael@0: 		       hb_buffer_t *buffer)
michael@0: {
michael@0:   /* Note: This loop is extra overhead, but should not be measurable. */
michael@0:   bool has_broken_syllables = false;
michael@0:   unsigned int count = buffer->len;
michael@0:   for (unsigned int i = 0; i < count; i++)
michael@0:     if ((buffer->info[i].syllable() & 0x0F) == broken_cluster) {
michael@0:       has_broken_syllables = true;
michael@0:       break;
michael@0:     }
michael@0:   if (likely (!has_broken_syllables))
michael@0:     return;
michael@0: 
michael@0: 
michael@0:   hb_codepoint_t dottedcircle_glyph;
michael@0:   if (!font->get_glyph (0x25CC, 0, &dottedcircle_glyph))
michael@0:     return;
michael@0: 
michael@0:   hb_glyph_info_t dottedcircle = {0};
michael@0:   dottedcircle.codepoint = 0x25CC;
michael@0:   set_indic_properties (dottedcircle);
michael@0:   dottedcircle.codepoint = dottedcircle_glyph;
michael@0: 
michael@0:   buffer->clear_output ();
michael@0: 
michael@0:   buffer->idx = 0;
michael@0:   unsigned int last_syllable = 0;
michael@0:   while (buffer->idx < buffer->len)
michael@0:   {
michael@0:     unsigned int syllable = buffer->cur().syllable();
michael@0:     syllable_type_t syllable_type = (syllable_type_t) (syllable & 0x0F);
michael@0:     if (unlikely (last_syllable != syllable && syllable_type == broken_cluster))
michael@0:     {
michael@0:       last_syllable = syllable;
michael@0: 
michael@0:       hb_glyph_info_t info = dottedcircle;
michael@0:       info.cluster = buffer->cur().cluster;
michael@0:       info.mask = buffer->cur().mask;
michael@0:       info.syllable() = buffer->cur().syllable();
michael@0: 
michael@0:       /* Insert dottedcircle after possible Repha. */
michael@0:       while (buffer->idx < buffer->len &&
michael@0: 	     last_syllable == buffer->cur().syllable() &&
michael@0: 	     buffer->cur().indic_category() == OT_Repha)
michael@0:         buffer->next_glyph ();
michael@0: 
michael@0:       buffer->output_info (info);
michael@0:     }
michael@0:     else
michael@0:       buffer->next_glyph ();
michael@0:   }
michael@0: 
michael@0:   buffer->swap_buffers ();
michael@0: }
michael@0: 
michael@0: static void
michael@0: initial_reordering (const hb_ot_shape_plan_t *plan,
michael@0: 		    hb_font_t *font,
michael@0: 		    hb_buffer_t *buffer)
michael@0: {
michael@0:   update_consonant_positions (plan, font, buffer);
michael@0:   insert_dotted_circles (plan, font, buffer);
michael@0: 
michael@0:   hb_glyph_info_t *info = buffer->info;
michael@0:   unsigned int count = buffer->len;
michael@0:   if (unlikely (!count)) return;
michael@0:   unsigned int last = 0;
michael@0:   unsigned int last_syllable = info[0].syllable();
michael@0:   for (unsigned int i = 1; i < count; i++)
michael@0:     if (last_syllable != info[i].syllable()) {
michael@0:       initial_reordering_syllable (plan, font->face, buffer, last, i);
michael@0:       last = i;
michael@0:       last_syllable = info[last].syllable();
michael@0:     }
michael@0:   initial_reordering_syllable (plan, font->face, buffer, last, count);
michael@0: }
michael@0: 
michael@0: static void
michael@0: final_reordering_syllable (const hb_ot_shape_plan_t *plan,
michael@0: 			   hb_buffer_t *buffer,
michael@0: 			   unsigned int start, unsigned int end)
michael@0: {
michael@0:   const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
michael@0:   hb_glyph_info_t *info = buffer->info;
michael@0: 
michael@0:   /* 4. Final reordering:
michael@0:    *
michael@0:    * After the localized forms and basic shaping forms GSUB features have been
michael@0:    * applied (see below), the shaping engine performs some final glyph
michael@0:    * reordering before applying all the remaining font features to the entire
michael@0:    * cluster.
michael@0:    */
michael@0: 
michael@0:   /* Find base again */
michael@0:   unsigned int base;
michael@0:   for (base = start; base < end; base++)
michael@0:     if (info[base].indic_position() >= POS_BASE_C) {
michael@0:       if (start < base && info[base].indic_position() > POS_BASE_C)
michael@0:         base--;
michael@0:       break;
michael@0:     }
michael@0:   if (base == end && start < base &&
michael@0:       info[base - 1].indic_category() != OT_ZWJ)
michael@0:     base--;
michael@0:   while (start < base &&
michael@0: 	 (info[base].indic_category() == OT_H ||
michael@0: 	  info[base].indic_category() == OT_N))
michael@0:     base--;
michael@0: 
michael@0: 
michael@0:   /*   o Reorder matras:
michael@0:    *
michael@0:    *     If a pre-base matra character had been reordered before applying basic
michael@0:    *     features, the glyph can be moved closer to the main consonant based on
michael@0:    *     whether half-forms had been formed. Actual position for the matra is
michael@0:    *     defined as “after last standalone halant glyph, after initial matra
michael@0:    *     position and before the main consonant”. If ZWJ or ZWNJ follow this
michael@0:    *     halant, position is moved after it.
michael@0:    */
michael@0: 
michael@0:   if (start + 1 < end && start < base) /* Otherwise there can't be any pre-base matra characters. */
michael@0:   {
michael@0:     /* If we lost track of base, alas, position before last thingy. */
michael@0:     unsigned int new_pos = base == end ? base - 2 : base - 1;
michael@0: 
michael@0:     /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
michael@0:      * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
michael@0:      * We want to position matra after them.
michael@0:      */
michael@0:     if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
michael@0:     {
michael@0:       while (new_pos > start &&
michael@0: 	     !(is_one_of (info[new_pos], (FLAG (OT_M) | FLAG (OT_H) | FLAG (OT_Coeng)))))
michael@0: 	new_pos--;
michael@0: 
michael@0:       /* If we found no Halant we are done.
michael@0:        * Otherwise only proceed if the Halant does
michael@0:        * not belong to the Matra itself! */
michael@0:       if (is_halant_or_coeng (info[new_pos]) &&
michael@0: 	  info[new_pos].indic_position() != POS_PRE_M)
michael@0:       {
michael@0: 	/* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
michael@0: 	if (new_pos + 1 < end && is_joiner (info[new_pos + 1]))
michael@0: 	  new_pos++;
michael@0:       }
michael@0:       else
michael@0:         new_pos = start; /* No move. */
michael@0:     }
michael@0: 
michael@0:     if (start < new_pos && info[new_pos].indic_position () != POS_PRE_M)
michael@0:     {
michael@0:       /* Now go see if there's actually any matras... */
michael@0:       for (unsigned int i = new_pos; i > start; i--)
michael@0: 	if (info[i - 1].indic_position () == POS_PRE_M)
michael@0: 	{
michael@0: 	  unsigned int old_pos = i - 1;
michael@0: 	  hb_glyph_info_t tmp = info[old_pos];
michael@0: 	  memmove (&info[old_pos], &info[old_pos + 1], (new_pos - old_pos) * sizeof (info[0]));
michael@0: 	  info[new_pos] = tmp;
michael@0: 	  if (old_pos < base && base <= new_pos) /* Shouldn't actually happen. */
michael@0: 	    base--;
michael@0: 	  buffer->merge_clusters (new_pos, MIN (end, base + 1));
michael@0: 	  new_pos--;
michael@0: 	}
michael@0:     } else {
michael@0:       for (unsigned int i = start; i < base; i++)
michael@0: 	if (info[i].indic_position () == POS_PRE_M) {
michael@0: 	  buffer->merge_clusters (i, MIN (end, base + 1));
michael@0: 	  break;
michael@0: 	}
michael@0:     }
michael@0:   }
michael@0: 
michael@0: 
michael@0:   /*   o Reorder reph:
michael@0:    *
michael@0:    *     Reph’s original position is always at the beginning of the syllable,
michael@0:    *     (i.e. it is not reordered at the character reordering stage). However,
michael@0:    *     it will be reordered according to the basic-forms shaping results.
michael@0:    *     Possible positions for reph, depending on the script, are; after main,
michael@0:    *     before post-base consonant forms, and after post-base consonant forms.
michael@0:    */
michael@0: 
michael@0:   /* Two cases:
michael@0:    *
michael@0:    * - If repha is encoded as a sequence of characters (Ra,H or Ra,H,ZWJ), then
michael@0:    *   we should only move it if the sequence ligated to the repha form.
michael@0:    *
michael@0:    * - If repha is encoded separately and in the logical position, we should only
michael@0:    *   move it if it did NOT ligate.  If it ligated, it's probably the font trying
michael@0:    *   to make it work without the reordering.
michael@0:    */
michael@0:   if (start + 1 < end &&
michael@0:       info[start].indic_position() == POS_RA_TO_BECOME_REPH &&
michael@0:       ((info[start].indic_category() == OT_Repha) ^
michael@0:        _hb_glyph_info_ligated (&info[start])))
michael@0:   {
michael@0:     unsigned int new_reph_pos;
michael@0:     reph_position_t reph_pos = indic_plan->config->reph_pos;
michael@0: 
michael@0:     assert (reph_pos != REPH_POS_DONT_CARE);
michael@0: 
michael@0:     /*       1. If reph should be positioned after post-base consonant forms,
michael@0:      *          proceed to step 5.
michael@0:      */
michael@0:     if (reph_pos == REPH_POS_AFTER_POST)
michael@0:     {
michael@0:       goto reph_step_5;
michael@0:     }
michael@0: 
michael@0:     /*       2. If the reph repositioning class is not after post-base: target
michael@0:      *          position is after the first explicit halant glyph between the
michael@0:      *          first post-reph consonant and last main consonant. If ZWJ or ZWNJ
michael@0:      *          are following this halant, position is moved after it. If such
michael@0:      *          position is found, this is the target position. Otherwise,
michael@0:      *          proceed to the next step.
michael@0:      *
michael@0:      *          Note: in old-implementation fonts, where classifications were
michael@0:      *          fixed in shaping engine, there was no case where reph position
michael@0:      *          will be found on this step.
michael@0:      */
michael@0:     {
michael@0:       new_reph_pos = start + 1;
michael@0:       while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos]))
michael@0: 	new_reph_pos++;
michael@0: 
michael@0:       if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos]))
michael@0:       {
michael@0: 	/* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
michael@0: 	if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
michael@0: 	  new_reph_pos++;
michael@0: 	goto reph_move;
michael@0:       }
michael@0:     }
michael@0: 
michael@0:     /*       3. If reph should be repositioned after the main consonant: find the
michael@0:      *          first consonant not ligated with main, or find the first
michael@0:      *          consonant that is not a potential pre-base reordering Ra.
michael@0:      */
michael@0:     if (reph_pos == REPH_POS_AFTER_MAIN)
michael@0:     {
michael@0:       new_reph_pos = base;
michael@0:       while (new_reph_pos + 1 < end && info[new_reph_pos + 1].indic_position() <= POS_AFTER_MAIN)
michael@0: 	new_reph_pos++;
michael@0:       if (new_reph_pos < end)
michael@0:         goto reph_move;
michael@0:     }
michael@0: 
michael@0:     /*       4. If reph should be positioned before post-base consonant, find
michael@0:      *          first post-base classified consonant not ligated with main. If no
michael@0:      *          consonant is found, the target position should be before the
michael@0:      *          first matra, syllable modifier sign or vedic sign.
michael@0:      */
michael@0:     /* This is our take on what step 4 is trying to say (and failing, BADLY). */
michael@0:     if (reph_pos == REPH_POS_AFTER_SUB)
michael@0:     {
michael@0:       new_reph_pos = base;
michael@0:       while (new_reph_pos < end &&
michael@0: 	     !( FLAG (info[new_reph_pos + 1].indic_position()) & (FLAG (POS_POST_C) | FLAG (POS_AFTER_POST) | FLAG (POS_SMVD))))
michael@0: 	new_reph_pos++;
michael@0:       if (new_reph_pos < end)
michael@0:         goto reph_move;
michael@0:     }
michael@0: 
michael@0:     /*       5. If no consonant is found in steps 3 or 4, move reph to a position
michael@0:      *          immediately before the first post-base matra, syllable modifier
michael@0:      *          sign or vedic sign that has a reordering class after the intended
michael@0:      *          reph position. For example, if the reordering position for reph
michael@0:      *          is post-main, it will skip above-base matras that also have a
michael@0:      *          post-main position.
michael@0:      */
michael@0:     reph_step_5:
michael@0:     {
michael@0:       /* Copied from step 2. */
michael@0:       new_reph_pos = start + 1;
michael@0:       while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos]))
michael@0: 	new_reph_pos++;
michael@0: 
michael@0:       if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos]))
michael@0:       {
michael@0: 	/* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
michael@0: 	if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
michael@0: 	  new_reph_pos++;
michael@0: 	goto reph_move;
michael@0:       }
michael@0:     }
michael@0: 
michael@0:     /*       6. Otherwise, reorder reph to the end of the syllable.
michael@0:      */
michael@0:     {
michael@0:       new_reph_pos = end - 1;
michael@0:       while (new_reph_pos > start && info[new_reph_pos].indic_position() == POS_SMVD)
michael@0: 	new_reph_pos--;
michael@0: 
michael@0:       /*
michael@0:        * If the Reph is to be ending up after a Matra,Halant sequence,
michael@0:        * position it before that Halant so it can interact with the Matra.
michael@0:        * However, if it's a plain Consonant,Halant we shouldn't do that.
michael@0:        * Uniscribe doesn't do this.
michael@0:        * TEST: U+0930,U+094D,U+0915,U+094B,U+094D
michael@0:        */
michael@0:       if (!hb_options ().uniscribe_bug_compatible &&
michael@0: 	  unlikely (is_halant_or_coeng (info[new_reph_pos]))) {
michael@0: 	for (unsigned int i = base + 1; i < new_reph_pos; i++)
michael@0: 	  if (info[i].indic_category() == OT_M) {
michael@0: 	    /* Ok, got it. */
michael@0: 	    new_reph_pos--;
michael@0: 	  }
michael@0:       }
michael@0:       goto reph_move;
michael@0:     }
michael@0: 
michael@0:     reph_move:
michael@0:     {
michael@0:       buffer->merge_clusters (start, new_reph_pos + 1);
michael@0: 
michael@0:       /* Move */
michael@0:       hb_glyph_info_t reph = info[start];
michael@0:       memmove (&info[start], &info[start + 1], (new_reph_pos - start) * sizeof (info[0]));
michael@0:       info[new_reph_pos] = reph;
michael@0:       if (start < base && base <= new_reph_pos)
michael@0: 	base--;
michael@0:     }
michael@0:   }
michael@0: 
michael@0: 
michael@0:   /*   o Reorder pre-base reordering consonants:
michael@0:    *
michael@0:    *     If a pre-base reordering consonant is found, reorder it according to
michael@0:    *     the following rules:
michael@0:    */
michael@0: 
michael@0:   if (indic_plan->mask_array[PREF] && base + 1 < end) /* Otherwise there can't be any pre-base reordering Ra. */
michael@0:   {
michael@0:     unsigned int pref_len = indic_plan->config->pref_len;
michael@0:     for (unsigned int i = base + 1; i < end; i++)
michael@0:       if ((info[i].mask & indic_plan->mask_array[PREF]) != 0)
michael@0:       {
michael@0: 	/*       1. Only reorder a glyph produced by substitution during application
michael@0: 	 *          of the <pref> feature. (Note that a font may shape a Ra consonant with
michael@0: 	 *          the feature generally but block it in certain contexts.)
michael@0: 	 */
michael@0:         /* Note: We just check that something got substituted.  We don't check that
michael@0: 	 * the <pref> feature actually did it...
michael@0: 	 *
michael@0: 	 * If pref len is longer than one, then only reorder if it ligated.  If
michael@0: 	 * pref len is one, only reorder if it didn't ligate with other things. */
michael@0: 	if (_hb_glyph_info_substituted (&info[i]) &&
michael@0: 	    ((pref_len == 1) ^ _hb_glyph_info_ligated (&info[i])))
michael@0: 	{
michael@0: 	  /*
michael@0: 	   *       2. Try to find a target position the same way as for pre-base matra.
michael@0: 	   *          If it is found, reorder pre-base consonant glyph.
michael@0: 	   *
michael@0: 	   *       3. If position is not found, reorder immediately before main
michael@0: 	   *          consonant.
michael@0: 	   */
michael@0: 
michael@0: 	  unsigned int new_pos = base;
michael@0: 	  /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
michael@0: 	   * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
michael@0: 	   * We want to position matra after them.
michael@0: 	   */
michael@0: 	  if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
michael@0: 	  {
michael@0: 	    while (new_pos > start &&
michael@0: 		   !(is_one_of (info[new_pos - 1], FLAG(OT_M) | HALANT_OR_COENG_FLAGS)))
michael@0: 	      new_pos--;
michael@0: 
michael@0: 	    /* In Khmer coeng model, a H,Ra can go *after* matras.  If it goes after a
michael@0: 	     * split matra, it should be reordered to *before* the left part of such matra. */
michael@0: 	    if (new_pos > start && info[new_pos - 1].indic_category() == OT_M)
michael@0: 	    {
michael@0: 	      unsigned int old_pos = i;
michael@0: 	      for (unsigned int i = base + 1; i < old_pos; i++)
michael@0: 		if (info[i].indic_category() == OT_M)
michael@0: 		{
michael@0: 		  new_pos--;
michael@0: 		  break;
michael@0: 		}
michael@0: 	    }
michael@0: 	  }
michael@0: 
michael@0: 	  if (new_pos > start && is_halant_or_coeng (info[new_pos - 1]))
michael@0: 	  {
michael@0: 	    /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
michael@0: 	    if (new_pos < end && is_joiner (info[new_pos]))
michael@0: 	      new_pos++;
michael@0: 	  }
michael@0: 
michael@0: 	  {
michael@0: 	    unsigned int old_pos = i;
michael@0: 	    buffer->merge_clusters (new_pos, old_pos + 1);
michael@0: 	    hb_glyph_info_t tmp = info[old_pos];
michael@0: 	    memmove (&info[new_pos + 1], &info[new_pos], (old_pos - new_pos) * sizeof (info[0]));
michael@0: 	    info[new_pos] = tmp;
michael@0: 	    if (new_pos <= base && base < old_pos)
michael@0: 	      base++;
michael@0: 	  }
michael@0: 	}
michael@0: 
michael@0:         break;
michael@0:       }
michael@0:   }
michael@0: 
michael@0: 
michael@0:   /* Apply 'init' to the Left Matra if it's a word start. */
michael@0:   if (info[start].indic_position () == POS_PRE_M &&
michael@0:       (!start ||
michael@0:        !(FLAG (_hb_glyph_info_get_general_category (&info[start - 1])) &
michael@0: 	 FLAG_RANGE (HB_UNICODE_GENERAL_CATEGORY_FORMAT, HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK))))
michael@0:     info[start].mask |= indic_plan->mask_array[INIT];
michael@0: 
michael@0: 
michael@0:   /*
michael@0:    * Finish off the clusters and go home!
michael@0:    */
michael@0:   if (hb_options ().uniscribe_bug_compatible)
michael@0:   {
michael@0:     switch ((hb_tag_t) plan->props.script)
michael@0:     {
michael@0:       case HB_SCRIPT_TAMIL:
michael@0:       case HB_SCRIPT_SINHALA:
michael@0:         break;
michael@0: 
michael@0:       default:
michael@0: 	/* Uniscribe merges the entire cluster... Except for Tamil & Sinhala.
michael@0: 	 * This means, half forms are submerged into the main consonants cluster.
michael@0: 	 * This is unnecessary, and makes cursor positioning harder, but that's what
michael@0: 	 * Uniscribe does. */
michael@0: 	buffer->merge_clusters (start, end);
michael@0: 	break;
michael@0:     }
michael@0:   }
michael@0: }
michael@0: 
michael@0: 
michael@0: static void
michael@0: final_reordering (const hb_ot_shape_plan_t *plan,
michael@0: 		  hb_font_t *font HB_UNUSED,
michael@0: 		  hb_buffer_t *buffer)
michael@0: {
michael@0:   unsigned int count = buffer->len;
michael@0:   if (unlikely (!count)) return;
michael@0: 
michael@0:   hb_glyph_info_t *info = buffer->info;
michael@0:   unsigned int last = 0;
michael@0:   unsigned int last_syllable = info[0].syllable();
michael@0:   for (unsigned int i = 1; i < count; i++)
michael@0:     if (last_syllable != info[i].syllable()) {
michael@0:       final_reordering_syllable (plan, buffer, last, i);
michael@0:       last = i;
michael@0:       last_syllable = info[last].syllable();
michael@0:     }
michael@0:   final_reordering_syllable (plan, buffer, last, count);
michael@0: 
michael@0:   HB_BUFFER_DEALLOCATE_VAR (buffer, indic_category);
michael@0:   HB_BUFFER_DEALLOCATE_VAR (buffer, indic_position);
michael@0: }
michael@0: 
michael@0: 
michael@0: static void
michael@0: clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
michael@0: 		 hb_font_t *font HB_UNUSED,
michael@0: 		 hb_buffer_t *buffer)
michael@0: {
michael@0:   hb_glyph_info_t *info = buffer->info;
michael@0:   unsigned int count = buffer->len;
michael@0:   for (unsigned int i = 0; i < count; i++)
michael@0:     info[i].syllable() = 0;
michael@0: }
michael@0: 
michael@0: 
michael@0: static bool
michael@0: decompose_indic (const hb_ot_shape_normalize_context_t *c,
michael@0: 		 hb_codepoint_t  ab,
michael@0: 		 hb_codepoint_t *a,
michael@0: 		 hb_codepoint_t *b)
michael@0: {
michael@0:   switch (ab)
michael@0:   {
michael@0:     /* Don't decompose these. */
michael@0:     case 0x0931  : return false;
michael@0:     case 0x0B94  : return false;
michael@0: 
michael@0: 
michael@0:     /*
michael@0:      * Decompose split matras that don't have Unicode decompositions.
michael@0:      */
michael@0: 
michael@0:     case 0x0F77  : *a = 0x0FB2; *b= 0x0F81; return true;
michael@0:     case 0x0F79  : *a = 0x0FB3; *b= 0x0F81; return true;
michael@0:     case 0x17BE  : *a = 0x17C1; *b= 0x17BE; return true;
michael@0:     case 0x17BF  : *a = 0x17C1; *b= 0x17BF; return true;
michael@0:     case 0x17C0  : *a = 0x17C1; *b= 0x17C0; return true;
michael@0:     case 0x17C4  : *a = 0x17C1; *b= 0x17C4; return true;
michael@0:     case 0x17C5  : *a = 0x17C1; *b= 0x17C5; return true;
michael@0:     case 0x1925  : *a = 0x1920; *b= 0x1923; return true;
michael@0:     case 0x1926  : *a = 0x1920; *b= 0x1924; return true;
michael@0:     case 0x1B3C  : *a = 0x1B42; *b= 0x1B3C; return true;
michael@0:     case 0x1112E  : *a = 0x11127; *b= 0x11131; return true;
michael@0:     case 0x1112F  : *a = 0x11127; *b= 0x11132; return true;
michael@0: #if 0
michael@0:     /* This one has no decomposition in Unicode, but needs no decomposition either. */
michael@0:     /* case 0x0AC9  : return false; */
michael@0:     case 0x0B57  : *a = no decomp, -> RIGHT; return true;
michael@0:     case 0x1C29  : *a = no decomp, -> LEFT; return true;
michael@0:     case 0xA9C0  : *a = no decomp, -> RIGHT; return true;
michael@0:     case 0x111BF  : *a = no decomp, -> ABOVE; return true;
michael@0: #endif
michael@0:   }
michael@0: 
michael@0:   if ((ab == 0x0DDA || hb_in_range<hb_codepoint_t> (ab, 0x0DDC, 0x0DDE)))
michael@0:   {
michael@0:     /*
michael@0:      * Sinhala split matras...  Let the fun begin.
michael@0:      *
michael@0:      * These four characters have Unicode decompositions.  However, Uniscribe
michael@0:      * decomposes them "Khmer-style", that is, it uses the character itself to
michael@0:      * get the second half.  The first half of all four decompositions is always
michael@0:      * U+0DD9.
michael@0:      *
michael@0:      * Now, there are buggy fonts, namely, the widely used lklug.ttf, that are
michael@0:      * broken with Uniscribe.  But we need to support them.  As such, we only
michael@0:      * do the Uniscribe-style decomposition if the character is transformed into
michael@0:      * its "sec.half" form by the 'pstf' feature.  Otherwise, we fall back to
michael@0:      * Unicode decomposition.
michael@0:      *
michael@0:      * Note that we can't unconditionally use Unicode decomposition.  That would
michael@0:      * break some other fonts, that are designed to work with Uniscribe, and
michael@0:      * don't have positioning features for the Unicode-style decomposition.
michael@0:      *
michael@0:      * Argh...
michael@0:      *
michael@0:      * The Uniscribe behavior is now documented in the newly published Sinhala
michael@0:      * spec in 2012:
michael@0:      *
michael@0:      *   http://www.microsoft.com/typography/OpenTypeDev/sinhala/intro.htm#shaping
michael@0:      */
michael@0: 
michael@0:     const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) c->plan->data;
michael@0: 
michael@0:     hb_codepoint_t glyph;
michael@0: 
michael@0:     if (hb_options ().uniscribe_bug_compatible ||
michael@0: 	(c->font->get_glyph (ab, 0, &glyph) &&
michael@0: 	 indic_plan->pstf.would_substitute (&glyph, 1, c->font->face)))
michael@0:     {
michael@0:       /* Ok, safe to use Uniscribe-style decomposition. */
michael@0:       *a = 0x0DD9;
michael@0:       *b = ab;
michael@0:       return true;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   return c->unicode->decompose (ab, a, b);
michael@0: }
michael@0: 
michael@0: static bool
michael@0: compose_indic (const hb_ot_shape_normalize_context_t *c,
michael@0: 	       hb_codepoint_t  a,
michael@0: 	       hb_codepoint_t  b,
michael@0: 	       hb_codepoint_t *ab)
michael@0: {
michael@0:   /* Avoid recomposing split matras. */
michael@0:   if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a)))
michael@0:     return false;
michael@0: 
michael@0:   /* Composition-exclusion exceptions that we want to recompose. */
michael@0:   if (a == 0x09AF && b == 0x09BC) { *ab = 0x09DF; return true; }
michael@0: 
michael@0:   return c->unicode->compose (a, b, ab);
michael@0: }
michael@0: 
michael@0: 
michael@0: const hb_ot_complex_shaper_t _hb_ot_complex_shaper_indic =
michael@0: {
michael@0:   "indic",
michael@0:   collect_features_indic,
michael@0:   override_features_indic,
michael@0:   data_create_indic,
michael@0:   data_destroy_indic,
michael@0:   NULL, /* preprocess_text */
michael@0:   HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT,
michael@0:   decompose_indic,
michael@0:   compose_indic,
michael@0:   setup_masks_indic,
michael@0:   HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE,
michael@0:   false, /* fallback_position */
michael@0: };