1 /* 2 * Copyright 2011,2012 Google, Inc. 3 * 4 * This is part of HarfBuzz, a text shaping library. 5 * 6 * Permission is hereby granted, without written agreement and without 7 * license or royalty fees, to use, copy, modify, and distribute this 8 * software and its documentation for any purpose, provided that the 9 * above copyright notice and the following two paragraphs appear in 10 * all copies of this software. 11 * 12 * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR 13 * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES 14 * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN 15 * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH 16 * DAMAGE. 17 * 18 * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING, 19 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND 20 * FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS 21 * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO 22 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS. 23 * 24 * Google Author(s): Behdad Esfahbod 25 */ 26 27 #include "hb-ot-shape-complex-indic-private.hh" 28 #include "hb-ot-layout-private.hh" 29 30 /* buffer var allocations */ 31 #define indic_category() complex_var_u8_0() /* indic_category_t */ 32 #define indic_position() complex_var_u8_1() /* indic_position_t */ 33 34 35 /* 36 * Indic shaper. 37 */ 38 39 40 #define IN_HALF_BLOCK(u, Base) (((u) & ~0x7F) == (Base)) 41 42 #define IS_DEVA(u) (IN_HALF_BLOCK (u, 0x0900)) 43 #define IS_BENG(u) (IN_HALF_BLOCK (u, 0x0980)) 44 #define IS_GURU(u) (IN_HALF_BLOCK (u, 0x0A00)) 45 #define IS_GUJR(u) (IN_HALF_BLOCK (u, 0x0A80)) 46 #define IS_ORYA(u) (IN_HALF_BLOCK (u, 0x0B00)) 47 #define IS_TAML(u) (IN_HALF_BLOCK (u, 0x0B80)) 48 #define IS_TELU(u) (IN_HALF_BLOCK (u, 0x0C00)) 49 #define IS_KNDA(u) (IN_HALF_BLOCK (u, 0x0C80)) 50 #define IS_MLYM(u) (IN_HALF_BLOCK (u, 0x0D00)) 51 #define IS_SINH(u) (IN_HALF_BLOCK (u, 0x0D80)) 52 #define IS_KHMR(u) (IN_HALF_BLOCK (u, 0x1780)) 53 54 55 #define MATRA_POS_LEFT(u) POS_PRE_M 56 #define MATRA_POS_RIGHT(u) ( \ 57 IS_DEVA(u) ? POS_AFTER_SUB : \ 58 IS_BENG(u) ? POS_AFTER_POST : \ 59 IS_GURU(u) ? POS_AFTER_POST : \ 60 IS_GUJR(u) ? POS_AFTER_POST : \ 61 IS_ORYA(u) ? POS_AFTER_POST : \ 62 IS_TAML(u) ? POS_AFTER_POST : \ 63 IS_TELU(u) ? (u <= 0x0C42 ? POS_BEFORE_SUB : POS_AFTER_SUB) : \ 64 IS_KNDA(u) ? (u < 0x0CC3 || u > 0xCD6 ? POS_BEFORE_SUB : POS_AFTER_SUB) : \ 65 IS_MLYM(u) ? POS_AFTER_POST : \ 66 IS_SINH(u) ? POS_AFTER_SUB : \ 67 IS_KHMR(u) ? POS_AFTER_POST : \ 68 /*default*/ POS_AFTER_SUB \ 69 ) 70 #define MATRA_POS_TOP(u) ( /* BENG and MLYM don't have top matras. */ \ 71 IS_DEVA(u) ? POS_AFTER_SUB : \ 72 IS_GURU(u) ? POS_AFTER_POST : /* Deviate from spec */ \ 73 IS_GUJR(u) ? POS_AFTER_SUB : \ 74 IS_ORYA(u) ? POS_AFTER_MAIN : \ 75 IS_TAML(u) ? POS_AFTER_SUB : \ 76 IS_TELU(u) ? POS_BEFORE_SUB : \ 77 IS_KNDA(u) ? POS_BEFORE_SUB : \ 78 IS_SINH(u) ? POS_AFTER_SUB : \ 79 IS_KHMR(u) ? POS_AFTER_POST : \ 80 /*default*/ POS_AFTER_SUB \ 81 ) 82 #define MATRA_POS_BOTTOM(u) ( \ 83 IS_DEVA(u) ? POS_AFTER_SUB : \ 84 IS_BENG(u) ? POS_AFTER_SUB : \ 85 IS_GURU(u) ? POS_AFTER_POST : \ 86 IS_GUJR(u) ? POS_AFTER_POST : \ 87 IS_ORYA(u) ? POS_AFTER_SUB : \ 88 IS_TAML(u) ? POS_AFTER_POST : \ 89 IS_TELU(u) ? POS_BEFORE_SUB : \ 90 IS_KNDA(u) ? POS_BEFORE_SUB : \ 91 IS_MLYM(u) ? POS_AFTER_POST : \ 92 IS_SINH(u) ? POS_AFTER_SUB : \ 93 IS_KHMR(u) ? POS_AFTER_POST : \ 94 /*default*/ POS_AFTER_SUB \ 95 ) 96 97 static inline indic_position_t 98 matra_position (hb_codepoint_t u, indic_position_t side) 99 { 100 switch ((int) side) 101 { 102 case POS_PRE_C: return MATRA_POS_LEFT (u); 103 case POS_POST_C: return MATRA_POS_RIGHT (u); 104 case POS_ABOVE_C: return MATRA_POS_TOP (u); 105 case POS_BELOW_C: return MATRA_POS_BOTTOM (u); 106 }; 107 return side; 108 } 109 110 /* XXX 111 * This is a hack for now. We should move this data into the main Indic table. 112 * Or completely remove it and just check in the tables. 113 */ 114 static const hb_codepoint_t ra_chars[] = { 115 0x0930, /* Devanagari */ 116 0x09B0, /* Bengali */ 117 0x09F0, /* Bengali */ 118 0x0A30, /* Gurmukhi */ /* No Reph */ 119 0x0AB0, /* Gujarati */ 120 0x0B30, /* Oriya */ 121 0x0BB0, /* Tamil */ /* No Reph */ 122 0x0C30, /* Telugu */ /* Reph formed only with ZWJ */ 123 0x0CB0, /* Kannada */ 124 0x0D30, /* Malayalam */ /* No Reph, Logical Repha */ 125 126 0x0DBB, /* Sinhala */ /* Reph formed only with ZWJ */ 127 128 0x179A, /* Khmer */ /* No Reph, Visual Repha */ 129 }; 130 131 static inline bool 132 is_ra (hb_codepoint_t u) 133 { 134 for (unsigned int i = 0; i < ARRAY_LENGTH (ra_chars); i++) 135 if (u == ra_chars[i]) 136 return true; 137 return false; 138 } 139 140 static inline bool 141 is_one_of (const hb_glyph_info_t &info, unsigned int flags) 142 { 143 /* If it ligated, all bets are off. */ 144 if (_hb_glyph_info_ligated (&info)) return false; 145 return !!(FLAG (info.indic_category()) & flags); 146 } 147 148 #define JOINER_FLAGS (FLAG (OT_ZWJ) | FLAG (OT_ZWNJ)) 149 static inline bool 150 is_joiner (const hb_glyph_info_t &info) 151 { 152 return is_one_of (info, JOINER_FLAGS); 153 } 154 155 #define MEDIAL_FLAGS (FLAG (OT_CM) | FLAG (OT_CM2)) 156 157 /* Note: 158 * 159 * We treat Vowels and placeholders as if they were consonants. This is safe because Vowels 160 * cannot happen in a consonant syllable. The plus side however is, we can call the 161 * consonant syllable logic from the vowel syllable function and get it all right! */ 162 #define CONSONANT_FLAGS (FLAG (OT_C) | FLAG (OT_Ra) | MEDIAL_FLAGS | FLAG (OT_V) | FLAG (OT_NBSP) | FLAG (OT_DOTTEDCIRCLE)) 163 static inline bool 164 is_consonant (const hb_glyph_info_t &info) 165 { 166 return is_one_of (info, CONSONANT_FLAGS); 167 } 168 169 #define HALANT_OR_COENG_FLAGS (FLAG (OT_H) | FLAG (OT_Coeng)) 170 static inline bool 171 is_halant_or_coeng (const hb_glyph_info_t &info) 172 { 173 return is_one_of (info, HALANT_OR_COENG_FLAGS); 174 } 175 176 static inline void 177 set_indic_properties (hb_glyph_info_t &info) 178 { 179 hb_codepoint_t u = info.codepoint; 180 unsigned int type = hb_indic_get_categories (u); 181 indic_category_t cat = (indic_category_t) (type & 0x7F); 182 indic_position_t pos = (indic_position_t) (type >> 8); 183 184 185 /* 186 * Re-assign category 187 */ 188 189 190 /* The spec says U+0952 is OT_A. However, testing shows that Uniscribe 191 * treats U+0951..U+0954 all behave similarly. 192 * TESTS: 193 * U+092E,U+0947,U+0952 194 * U+092E,U+0952,U+0947 195 * U+092E,U+0947,U+0951 196 * U+092E,U+0951,U+0947 197 */ 198 if (unlikely (hb_in_range<hb_codepoint_t> (u, 0x0951, 0x0954))) 199 cat = OT_A; 200 201 if (unlikely (u == 0x17D1)) 202 cat = OT_X; 203 if (cat == OT_X && 204 unlikely (hb_in_range<hb_codepoint_t> (u, 0x17CB, 0x17D3))) /* Khmer Various signs */ 205 { 206 /* These are like Top Matras. */ 207 cat = OT_M; 208 pos = POS_ABOVE_C; 209 } 210 if (u == 0x17C6) /* Khmer Bindu doesn't like to be repositioned. */ 211 cat = OT_N; 212 213 if (unlikely (u == 0x17D2)) cat = OT_Coeng; /* Khmer coeng */ 214 else if (unlikely (u == 0x200C)) cat = OT_ZWNJ; 215 else if (unlikely (u == 0x200D)) cat = OT_ZWJ; 216 else if (unlikely (u == 0x25CC)) cat = OT_DOTTEDCIRCLE; 217 else if (unlikely (u == 0x0A71)) cat = OT_SM; /* GURMUKHI ADDAK. Move it to the end. */ 218 else if (unlikely (u == 0xA982)) cat = OT_SM; /* Javanese repha. */ 219 else if (unlikely (u == 0xA9BE)) cat = OT_CM2; /* Javanese medial ya. */ 220 else if (unlikely (u == 0xA9BD)) { cat = OT_M; pos = POS_POST_C; } /* Javanese vocalic r. */ 221 222 if (cat == OT_Repha) { 223 /* There are two kinds of characters marked as Repha: 224 * - The ones that are GenCat=Mn are already positioned visually, ie. after base. (eg. Khmer) 225 * - The ones that are GenCat=Lo is encoded logically, ie. beginning of syllable. (eg. Malayalam) 226 * 227 * We recategorize the first kind to look like a Nukta and attached to the base directly. 228 */ 229 if (_hb_glyph_info_get_general_category (&info) == HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK) 230 cat = OT_N; 231 } 232 233 234 235 /* 236 * Re-assign position. 237 */ 238 239 if ((FLAG (cat) & CONSONANT_FLAGS)) 240 { 241 pos = POS_BASE_C; 242 if (is_ra (u)) 243 cat = OT_Ra; 244 } 245 else if (cat == OT_M) 246 { 247 pos = matra_position (u, pos); 248 } 249 else if ((FLAG (cat) & (FLAG (OT_SM) | FLAG (OT_VD) | FLAG (OT_A) | FLAG (OT_Avag)))) 250 { 251 pos = POS_SMVD; 252 } 253 254 if (unlikely (u == 0x0B01)) pos = POS_BEFORE_SUB; /* Oriya Bindu is BeforeSub in the spec. */ 255 256 257 258 info.indic_category() = cat; 259 info.indic_position() = pos; 260 } 261 262 /* 263 * Things above this line should ideally be moved to the Indic table itself. 264 */ 265 266 267 /* 268 * Indic configurations. Note that we do not want to keep every single script-specific 269 * behavior in these tables necessarily. This should mainly be used for per-script 270 * properties that are cheaper keeping here, than in the code. Ie. if, say, one and 271 * only one script has an exception, that one script can be if'ed directly in the code, 272 * instead of adding a new flag in these structs. 273 */ 274 275 enum base_position_t { 276 BASE_POS_FIRST, 277 BASE_POS_LAST_SINHALA, 278 BASE_POS_LAST 279 }; 280 enum reph_position_t { 281 REPH_POS_AFTER_MAIN = POS_AFTER_MAIN, 282 REPH_POS_BEFORE_SUB = POS_BEFORE_SUB, 283 REPH_POS_AFTER_SUB = POS_AFTER_SUB, 284 REPH_POS_BEFORE_POST = POS_BEFORE_POST, 285 REPH_POS_AFTER_POST = POS_AFTER_POST, 286 REPH_POS_DONT_CARE = POS_RA_TO_BECOME_REPH 287 }; 288 enum reph_mode_t { 289 REPH_MODE_IMPLICIT, /* Reph formed out of initial Ra,H sequence. */ 290 REPH_MODE_EXPLICIT, /* Reph formed out of initial Ra,H,ZWJ sequence. */ 291 REPH_MODE_VIS_REPHA, /* Encoded Repha character, no reordering needed. */ 292 REPH_MODE_LOG_REPHA /* Encoded Repha character, needs reordering. */ 293 }; 294 enum blwf_mode_t { 295 BLWF_MODE_PRE_AND_POST, /* Below-forms feature applied to pre-base and post-base. */ 296 BLWF_MODE_POST_ONLY /* Below-forms feature applied to post-base only. */ 297 }; 298 enum pref_len_t { 299 PREF_LEN_1 = 1, 300 PREF_LEN_2 = 2, 301 PREF_LEN_DONT_CARE = PREF_LEN_2 302 }; 303 struct indic_config_t 304 { 305 hb_script_t script; 306 bool has_old_spec; 307 hb_codepoint_t virama; 308 base_position_t base_pos; 309 reph_position_t reph_pos; 310 reph_mode_t reph_mode; 311 blwf_mode_t blwf_mode; 312 pref_len_t pref_len; 313 }; 314 315 static const indic_config_t indic_configs[] = 316 { 317 /* Default. Should be first. */ 318 {HB_SCRIPT_INVALID, false, 0,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_1}, 319 {HB_SCRIPT_DEVANAGARI,true, 0x094D,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 320 {HB_SCRIPT_BENGALI, true, 0x09CD,BASE_POS_LAST, REPH_POS_AFTER_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 321 {HB_SCRIPT_GURMUKHI, true, 0x0A4D,BASE_POS_LAST, REPH_POS_BEFORE_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 322 {HB_SCRIPT_GUJARATI, true, 0x0ACD,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 323 {HB_SCRIPT_ORIYA, true, 0x0B4D,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 324 {HB_SCRIPT_TAMIL, true, 0x0BCD,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_2}, 325 {HB_SCRIPT_TELUGU, true, 0x0C4D,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_EXPLICIT, BLWF_MODE_POST_ONLY, PREF_LEN_2}, 326 {HB_SCRIPT_KANNADA, true, 0x0CCD,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_POST_ONLY, PREF_LEN_2}, 327 {HB_SCRIPT_MALAYALAM, true, 0x0D4D,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_LOG_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2}, 328 {HB_SCRIPT_SINHALA, false,0x0DCA,BASE_POS_LAST_SINHALA, 329 REPH_POS_AFTER_MAIN, REPH_MODE_EXPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 330 {HB_SCRIPT_KHMER, false,0x17D2,BASE_POS_FIRST,REPH_POS_DONT_CARE, REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2}, 331 {HB_SCRIPT_JAVANESE, false,0xA9C0,BASE_POS_FIRST,REPH_POS_DONT_CARE, REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_1}, 332 }; 333 334 335 336 /* 337 * Indic shaper. 338 */ 339 340 struct feature_list_t { 341 hb_tag_t tag; 342 hb_ot_map_feature_flags_t flags; 343 }; 344 345 static const feature_list_t 346 indic_features[] = 347 { 348 /* 349 * Basic features. 350 * These features are applied in order, one at a time, after initial_reordering. 351 */ 352 {HB_TAG('n','u','k','t'), F_GLOBAL}, 353 {HB_TAG('a','k','h','n'), F_GLOBAL}, 354 {HB_TAG('r','p','h','f'), F_NONE}, 355 {HB_TAG('r','k','r','f'), F_GLOBAL}, 356 {HB_TAG('p','r','e','f'), F_NONE}, 357 {HB_TAG('b','l','w','f'), F_NONE}, 358 {HB_TAG('a','b','v','f'), F_NONE}, 359 {HB_TAG('h','a','l','f'), F_NONE}, 360 {HB_TAG('p','s','t','f'), F_NONE}, 361 {HB_TAG('v','a','t','u'), F_GLOBAL}, 362 {HB_TAG('c','j','c','t'), F_GLOBAL}, 363 {HB_TAG('c','f','a','r'), F_NONE}, 364 /* 365 * Other features. 366 * These features are applied all at once, after final_reordering. 367 * Default Bengali font in Windows for example has intermixed 368 * lookups for init,pres,abvs,blws features. 369 */ 370 {HB_TAG('i','n','i','t'), F_NONE}, 371 {HB_TAG('p','r','e','s'), F_GLOBAL}, 372 {HB_TAG('a','b','v','s'), F_GLOBAL}, 373 {HB_TAG('b','l','w','s'), F_GLOBAL}, 374 {HB_TAG('p','s','t','s'), F_GLOBAL}, 375 {HB_TAG('h','a','l','n'), F_GLOBAL}, 376 /* Positioning features, though we don't care about the types. */ 377 {HB_TAG('d','i','s','t'), F_GLOBAL}, 378 {HB_TAG('a','b','v','m'), F_GLOBAL}, 379 {HB_TAG('b','l','w','m'), F_GLOBAL}, 380 }; 381 382 /* 383 * Must be in the same order as the indic_features array. 384 */ 385 enum { 386 _NUKT, 387 _AKHN, 388 RPHF, 389 _RKRF, 390 PREF, 391 BLWF, 392 ABVF, 393 HALF, 394 PSTF, 395 _VATU, 396 _CJCT, 397 CFAR, 398 399 INIT, 400 _PRES, 401 _ABVS, 402 _BLWS, 403 _PSTS, 404 _HALN, 405 _DIST, 406 _ABVM, 407 _BLWM, 408 409 INDIC_NUM_FEATURES, 410 INDIC_BASIC_FEATURES = INIT /* Don't forget to update this! */ 411 }; 412 413 static void 414 setup_syllables (const hb_ot_shape_plan_t *plan, 415 hb_font_t *font, 416 hb_buffer_t *buffer); 417 static void 418 initial_reordering (const hb_ot_shape_plan_t *plan, 419 hb_font_t *font, 420 hb_buffer_t *buffer); 421 static void 422 final_reordering (const hb_ot_shape_plan_t *plan, 423 hb_font_t *font, 424 hb_buffer_t *buffer); 425 static void 426 clear_syllables (const hb_ot_shape_plan_t *plan, 427 hb_font_t *font, 428 hb_buffer_t *buffer); 429 430 static void 431 collect_features_indic (hb_ot_shape_planner_t *plan) 432 { 433 hb_ot_map_builder_t *map = &plan->map; 434 435 /* Do this before any lookups have been applied. */ 436 map->add_gsub_pause (setup_syllables); 437 438 map->add_global_bool_feature (HB_TAG('l','o','c','l')); 439 /* The Indic specs do not require ccmp, but we apply it here since if 440 * there is a use of it, it's typically at the beginning. */ 441 map->add_global_bool_feature (HB_TAG('c','c','m','p')); 442 443 444 unsigned int i = 0; 445 map->add_gsub_pause (initial_reordering); 446 for (; i < INDIC_BASIC_FEATURES; i++) { 447 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ); 448 map->add_gsub_pause (NULL); 449 } 450 map->add_gsub_pause (final_reordering); 451 for (; i < INDIC_NUM_FEATURES; i++) { 452 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ); 453 } 454 455 map->add_global_bool_feature (HB_TAG('c','a','l','t')); 456 map->add_global_bool_feature (HB_TAG('c','l','i','g')); 457 458 map->add_gsub_pause (clear_syllables); 459 } 460 461 static void 462 override_features_indic (hb_ot_shape_planner_t *plan) 463 { 464 /* Uniscribe does not apply 'kern' in Khmer. */ 465 if (hb_options ().uniscribe_bug_compatible) 466 { 467 switch ((hb_tag_t) plan->props.script) 468 { 469 case HB_SCRIPT_KHMER: 470 plan->map.add_feature (HB_TAG('k','e','r','n'), 0, F_GLOBAL); 471 break; 472 } 473 } 474 475 plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL); 476 } 477 478 479 struct would_substitute_feature_t 480 { 481 inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_) 482 { 483 zero_context = zero_context_; 484 map->get_stage_lookups (0/*GSUB*/, 485 map->get_feature_stage (0/*GSUB*/, feature_tag), 486 &lookups, &count); 487 } 488 489 inline bool would_substitute (const hb_codepoint_t *glyphs, 490 unsigned int glyphs_count, 491 hb_face_t *face) const 492 { 493 for (unsigned int i = 0; i < count; i++) 494 if (hb_ot_layout_lookup_would_substitute_fast (face, lookups[i].index, glyphs, glyphs_count, zero_context)) 495 return true; 496 return false; 497 } 498 499 private: 500 const hb_ot_map_t::lookup_map_t *lookups; 501 unsigned int count; 502 bool zero_context; 503 }; 504 505 struct indic_shape_plan_t 506 { 507 ASSERT_POD (); 508 509 inline bool get_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const 510 { 511 hb_codepoint_t glyph = virama_glyph; 512 if (unlikely (virama_glyph == (hb_codepoint_t) -1)) 513 { 514 if (!config->virama || !font->get_glyph (config->virama, 0, &glyph)) 515 glyph = 0; 516 /* Technically speaking, the spec says we should apply 'locl' to virama too. 517 * Maybe one day... */ 518 519 /* Our get_glyph() function needs a font, so we can't get the virama glyph 520 * during shape planning... Instead, overwrite it here. It's safe. Don't worry! */ 521 (const_cast<indic_shape_plan_t *> (this))->virama_glyph = glyph; 522 } 523 524 *pglyph = glyph; 525 return glyph != 0; 526 } 527 528 const indic_config_t *config; 529 530 bool is_old_spec; 531 hb_codepoint_t virama_glyph; 532 533 would_substitute_feature_t rphf; 534 would_substitute_feature_t pref; 535 would_substitute_feature_t blwf; 536 would_substitute_feature_t pstf; 537 538 hb_mask_t mask_array[INDIC_NUM_FEATURES]; 539 }; 540 541 static void * 542 data_create_indic (const hb_ot_shape_plan_t *plan) 543 { 544 indic_shape_plan_t *indic_plan = (indic_shape_plan_t *) calloc (1, sizeof (indic_shape_plan_t)); 545 if (unlikely (!indic_plan)) 546 return NULL; 547 548 indic_plan->config = &indic_configs[0]; 549 for (unsigned int i = 1; i < ARRAY_LENGTH (indic_configs); i++) 550 if (plan->props.script == indic_configs[i].script) { 551 indic_plan->config = &indic_configs[i]; 552 break; 553 } 554 555 indic_plan->is_old_spec = indic_plan->config->has_old_spec && ((plan->map.chosen_script[0] & 0x000000FF) != '2'); 556 indic_plan->virama_glyph = (hb_codepoint_t) -1; 557 558 /* Use zero-context would_substitute() matching for new-spec of the main 559 * Indic scripts, but not for old-spec or scripts with one spec only. */ 560 bool zero_context = indic_plan->config->has_old_spec || !indic_plan->is_old_spec; 561 indic_plan->rphf.init (&plan->map, HB_TAG('r','p','h','f'), zero_context); 562 indic_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), zero_context); 563 indic_plan->blwf.init (&plan->map, HB_TAG('b','l','w','f'), zero_context); 564 indic_plan->pstf.init (&plan->map, HB_TAG('p','s','t','f'), zero_context); 565 566 for (unsigned int i = 0; i < ARRAY_LENGTH (indic_plan->mask_array); i++) 567 indic_plan->mask_array[i] = (indic_features[i].flags & F_GLOBAL) ? 568 0 : plan->map.get_1_mask (indic_features[i].tag); 569 570 return indic_plan; 571 } 572 573 static void 574 data_destroy_indic (void *data) 575 { 576 free (data); 577 } 578 579 static indic_position_t 580 consonant_position_from_face (const indic_shape_plan_t *indic_plan, 581 const hb_codepoint_t consonant, 582 const hb_codepoint_t virama, 583 hb_face_t *face) 584 { 585 /* For old-spec, the order of glyphs is Consonant,Virama, 586 * whereas for new-spec, it's Virama,Consonant. However, 587 * some broken fonts (like Free Sans) simply copied lookups 588 * from old-spec to new-spec without modification. 589 * And oddly enough, Uniscribe seems to respect those lookups. 590 * Eg. in the sequence U+0924,U+094D,U+0930, Uniscribe finds 591 * base at 0. The font however, only has lookups matching 592 * 930,94D in 'blwf', not the expected 94D,930 (with new-spec 593 * table). As such, we simply match both sequences. Seems 594 * to work. */ 595 hb_codepoint_t glyphs[3] = {virama, consonant, virama}; 596 if (indic_plan->blwf.would_substitute (glyphs , 2, face) || 597 indic_plan->blwf.would_substitute (glyphs+1, 2, face)) 598 return POS_BELOW_C; 599 if (indic_plan->pstf.would_substitute (glyphs , 2, face) || 600 indic_plan->pstf.would_substitute (glyphs+1, 2, face)) 601 return POS_POST_C; 602 unsigned int pref_len = indic_plan->config->pref_len; 603 if ((pref_len == PREF_LEN_2 && 604 (indic_plan->pref.would_substitute (glyphs , 2, face) || 605 indic_plan->pref.would_substitute (glyphs+1, 2, face))) 606 || (pref_len == PREF_LEN_1 && 607 indic_plan->pref.would_substitute (glyphs+1, 1, face))) 608 return POS_POST_C; 609 return POS_BASE_C; 610 } 611 612 613 enum syllable_type_t { 614 consonant_syllable, 615 vowel_syllable, 616 standalone_cluster, 617 avagraha_cluster, 618 broken_cluster, 619 non_indic_cluster, 620 }; 621 622 #include "hb-ot-shape-complex-indic-machine.hh" 623 624 625 static void 626 setup_masks_indic (const hb_ot_shape_plan_t *plan HB_UNUSED, 627 hb_buffer_t *buffer, 628 hb_font_t *font HB_UNUSED) 629 { 630 HB_BUFFER_ALLOCATE_VAR (buffer, indic_category); 631 HB_BUFFER_ALLOCATE_VAR (buffer, indic_position); 632 633 /* We cannot setup masks here. We save information about characters 634 * and setup masks later on in a pause-callback. */ 635 636 unsigned int count = buffer->len; 637 for (unsigned int i = 0; i < count; i++) 638 set_indic_properties (buffer->info[i]); 639 } 640 641 static void 642 setup_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED, 643 hb_font_t *font HB_UNUSED, 644 hb_buffer_t *buffer) 645 { 646 find_syllables (buffer); 647 } 648 649 static int 650 compare_indic_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb) 651 { 652 int a = pa->indic_position(); 653 int b = pb->indic_position(); 654 655 return a < b ? -1 : a == b ? 0 : +1; 656 } 657 658 659 660 static void 661 update_consonant_positions (const hb_ot_shape_plan_t *plan, 662 hb_font_t *font, 663 hb_buffer_t *buffer) 664 { 665 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; 666 667 if (indic_plan->config->base_pos != BASE_POS_LAST) 668 return; 669 670 hb_codepoint_t virama; 671 if (indic_plan->get_virama_glyph (font, &virama)) 672 { 673 hb_face_t *face = font->face; 674 unsigned int count = buffer->len; 675 for (unsigned int i = 0; i < count; i++) 676 if (buffer->info[i].indic_position() == POS_BASE_C) { 677 hb_codepoint_t consonant = buffer->info[i].codepoint; 678 buffer->info[i].indic_position() = consonant_position_from_face (indic_plan, consonant, virama, face); 679 } 680 } 681 } 682 683 684 /* Rules from: 685 * https://www.microsoft.com/typography/otfntdev/devanot/shaping.aspx */ 686 687 static void 688 initial_reordering_consonant_syllable (const hb_ot_shape_plan_t *plan, 689 hb_face_t *face, 690 hb_buffer_t *buffer, 691 unsigned int start, unsigned int end) 692 { 693 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; 694 hb_glyph_info_t *info = buffer->info; 695 696 697 /* 1. Find base consonant: 698 * 699 * The shaping engine finds the base consonant of the syllable, using the 700 * following algorithm: starting from the end of the syllable, move backwards 701 * until a consonant is found that does not have a below-base or post-base 702 * form (post-base forms have to follow below-base forms), or that is not a 703 * pre-base reordering Ra, or arrive at the first consonant. The consonant 704 * stopped at will be the base. 705 * 706 * o If the syllable starts with Ra + Halant (in a script that has Reph) 707 * and has more than one consonant, Ra is excluded from candidates for 708 * base consonants. 709 */ 710 711 unsigned int base = end; 712 bool has_reph = false; 713 714 { 715 /* -> If the syllable starts with Ra + Halant (in a script that has Reph) 716 * and has more than one consonant, Ra is excluded from candidates for 717 * base consonants. */ 718 unsigned int limit = start; 719 if (indic_plan->config->reph_pos != REPH_POS_DONT_CARE && 720 indic_plan->mask_array[RPHF] && 721 start + 3 <= end && 722 ( 723 (indic_plan->config->reph_mode == REPH_MODE_IMPLICIT && !is_joiner (info[start + 2])) || 724 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT && info[start + 2].indic_category() == OT_ZWJ) 725 )) 726 { 727 /* See if it matches the 'rphf' feature. */ 728 hb_codepoint_t glyphs[2] = {info[start].codepoint, info[start + 1].codepoint}; 729 if (indic_plan->rphf.would_substitute (glyphs, ARRAY_LENGTH (glyphs), face)) 730 { 731 limit += 2; 732 while (limit < end && is_joiner (info[limit])) 733 limit++; 734 base = start; 735 has_reph = true; 736 } 737 } else if (indic_plan->config->reph_mode == REPH_MODE_LOG_REPHA && info[start].indic_category() == OT_Repha) 738 { 739 limit += 1; 740 while (limit < end && is_joiner (info[limit])) 741 limit++; 742 base = start; 743 has_reph = true; 744 } 745 746 switch (indic_plan->config->base_pos) 747 { 748 default: 749 assert (false); 750 /* fallthrough */ 751 752 case BASE_POS_LAST: 753 { 754 /* -> starting from the end of the syllable, move backwards */ 755 unsigned int i = end; 756 bool seen_below = false; 757 do { 758 i--; 759 /* -> until a consonant is found */ 760 if (is_consonant (info[i])) 761 { 762 /* -> that does not have a below-base or post-base form 763 * (post-base forms have to follow below-base forms), */ 764 if (info[i].indic_position() != POS_BELOW_C && 765 (info[i].indic_position() != POS_POST_C || seen_below)) 766 { 767 base = i; 768 break; 769 } 770 if (info[i].indic_position() == POS_BELOW_C) 771 seen_below = true; 772 773 /* -> or that is not a pre-base reordering Ra, 774 * 775 * IMPLEMENTATION NOTES: 776 * 777 * Our pre-base reordering Ra's are marked POS_POST_C, so will be skipped 778 * by the logic above already. 779 */ 780 781 /* -> or arrive at the first consonant. The consonant stopped at will 782 * be the base. */ 783 base = i; 784 } 785 else 786 { 787 /* A ZWJ after a Halant stops the base search, and requests an explicit 788 * half form. 789 * A ZWJ before a Halant, requests a subjoined form instead, and hence 790 * search continues. This is particularly important for Bengali 791 * sequence Ra,H,Ya that should form Ya-Phalaa by subjoining Ya. */ 792 if (start < i && 793 info[i].indic_category() == OT_ZWJ && 794 info[i - 1].indic_category() == OT_H) 795 break; 796 } 797 } while (i > limit); 798 } 799 break; 800 801 case BASE_POS_LAST_SINHALA: 802 { 803 /* Sinhala base positioning is slightly different from main Indic, in that: 804 * 1. It's ZWJ behavior is different, 805 * 2. We don't need to look into the font for consonant positions. 806 */ 807 808 if (!has_reph) 809 base = limit; 810 811 /* Find the last base consonant that is not blocked by ZWJ. If there is 812 * a ZWJ right before a base consonant, that would request a subjoined form. */ 813 for (unsigned int i = limit; i < end; i++) 814 if (is_consonant (info[i])) 815 { 816 if (limit < i && info[i - 1].indic_category() == OT_ZWJ) 817 break; 818 else 819 base = i; 820 } 821 822 /* Mark all subsequent consonants as below. */ 823 for (unsigned int i = base + 1; i < end; i++) 824 if (is_consonant (info[i])) 825 info[i].indic_position() = POS_BELOW_C; 826 } 827 break; 828 829 case BASE_POS_FIRST: 830 { 831 /* The first consonant is always the base. */ 832 833 assert (indic_plan->config->reph_mode == REPH_MODE_VIS_REPHA); 834 assert (!has_reph); 835 836 base = start; 837 838 /* Mark all subsequent consonants as below. */ 839 for (unsigned int i = base + 1; i < end; i++) 840 if (is_consonant (info[i])) 841 info[i].indic_position() = POS_BELOW_C; 842 } 843 break; 844 } 845 846 /* -> If the syllable starts with Ra + Halant (in a script that has Reph) 847 * and has more than one consonant, Ra is excluded from candidates for 848 * base consonants. 849 * 850 * Only do this for unforced Reph. (ie. not for Ra,H,ZWJ. */ 851 if (has_reph && base == start && limit - base <= 2) { 852 /* Have no other consonant, so Reph is not formed and Ra becomes base. */ 853 has_reph = false; 854 } 855 } 856 857 858 /* 2. Decompose and reorder Matras: 859 * 860 * Each matra and any syllable modifier sign in the cluster are moved to the 861 * appropriate position relative to the consonant(s) in the cluster. The 862 * shaping engine decomposes two- or three-part matras into their constituent 863 * parts before any repositioning. Matra characters are classified by which 864 * consonant in a conjunct they have affinity for and are reordered to the 865 * following positions: 866 * 867 * o Before first half form in the syllable 868 * o After subjoined consonants 869 * o After post-form consonant 870 * o After main consonant (for above marks) 871 * 872 * IMPLEMENTATION NOTES: 873 * 874 * The normalize() routine has already decomposed matras for us, so we don't 875 * need to worry about that. 876 */ 877 878 879 /* 3. Reorder marks to canonical order: 880 * 881 * Adjacent nukta and halant or nukta and vedic sign are always repositioned 882 * if necessary, so that the nukta is first. 883 * 884 * IMPLEMENTATION NOTES: 885 * 886 * We don't need to do this: the normalize() routine already did this for us. 887 */ 888 889 890 /* Reorder characters */ 891 892 for (unsigned int i = start; i < base; i++) 893 info[i].indic_position() = MIN (POS_PRE_C, (indic_position_t) info[i].indic_position()); 894 895 if (base < end) 896 info[base].indic_position() = POS_BASE_C; 897 898 /* Mark final consonants. A final consonant is one appearing after a matra, 899 * like in Khmer. */ 900 for (unsigned int i = base + 1; i < end; i++) 901 if (info[i].indic_category() == OT_M) { 902 for (unsigned int j = i + 1; j < end; j++) 903 if (is_consonant (info[j])) { 904 info[j].indic_position() = POS_FINAL_C; 905 break; 906 } 907 break; 908 } 909 910 /* Handle beginning Ra */ 911 if (has_reph) 912 info[start].indic_position() = POS_RA_TO_BECOME_REPH; 913 914 /* For old-style Indic script tags, move the first post-base Halant after 915 * last consonant. Only do this if there is *not* a Halant after last 916 * consonant. Otherwise it becomes messy. */ 917 if (indic_plan->is_old_spec) { 918 for (unsigned int i = base + 1; i < end; i++) 919 if (info[i].indic_category() == OT_H) { 920 unsigned int j; 921 for (j = end - 1; j > i; j--) 922 if (is_consonant (info[j]) || info[j].indic_category() == OT_H) 923 break; 924 if (info[j].indic_category() != OT_H && j > i) { 925 /* Move Halant to after last consonant. */ 926 hb_glyph_info_t t = info[i]; 927 memmove (&info[i], &info[i + 1], (j - i) * sizeof (info[0])); 928 info[j] = t; 929 } 930 break; 931 } 932 } 933 934 /* Attach misc marks to previous char to move with them. */ 935 { 936 indic_position_t last_pos = POS_START; 937 for (unsigned int i = start; i < end; i++) 938 { 939 if ((FLAG (info[i].indic_category()) & (JOINER_FLAGS | FLAG (OT_N) | FLAG (OT_RS) | MEDIAL_FLAGS | HALANT_OR_COENG_FLAGS))) 940 { 941 info[i].indic_position() = last_pos; 942 if (unlikely (info[i].indic_category() == OT_H && 943 info[i].indic_position() == POS_PRE_M)) 944 { 945 /* 946 * Uniscribe doesn't move the Halant with Left Matra. 947 * TEST: U+092B,U+093F,U+094DE 948 * We follow. This is important for the Sinhala 949 * U+0DDA split matra since it decomposes to U+0DD9,U+0DCA 950 * where U+0DD9 is a left matra and U+0DCA is the virama. 951 * We don't want to move the virama with the left matra. 952 * TEST: U+0D9A,U+0DDA 953 */ 954 for (unsigned int j = i; j > start; j--) 955 if (info[j - 1].indic_position() != POS_PRE_M) { 956 info[i].indic_position() = info[j - 1].indic_position(); 957 break; 958 } 959 } 960 } else if (info[i].indic_position() != POS_SMVD) { 961 last_pos = (indic_position_t) info[i].indic_position(); 962 } 963 } 964 } 965 /* For post-base consonants let them own anything before them 966 * since the last consonant or matra. */ 967 { 968 unsigned int last = base; 969 for (unsigned int i = base + 1; i < end; i++) 970 if (is_consonant (info[i])) 971 { 972 for (unsigned int j = last + 1; j < i; j++) 973 if (info[j].indic_position() < POS_SMVD) 974 info[j].indic_position() = info[i].indic_position(); 975 last = i; 976 } else if (info[i].indic_category() == OT_M) 977 last = i; 978 } 979 980 981 { 982 /* Use syllable() for sort accounting temporarily. */ 983 unsigned int syllable = info[start].syllable(); 984 for (unsigned int i = start; i < end; i++) 985 info[i].syllable() = i - start; 986 987 /* Sit tight, rock 'n roll! */ 988 hb_bubble_sort (info + start, end - start, compare_indic_order); 989 /* Find base again */ 990 base = end; 991 for (unsigned int i = start; i < end; i++) 992 if (info[i].indic_position() == POS_BASE_C) 993 { 994 base = i; 995 break; 996 } 997 /* Things are out-of-control for post base positions, they may shuffle 998 * around like crazy. In old-spec mode, we move halants around, so in 999 * that case merge all clusters after base. Otherwise, check the sort 1000 * order and merge as needed. 1001 * For pre-base stuff, we handle cluster issues in final reordering. */ 1002 if (indic_plan->is_old_spec || end - base > 127) 1003 buffer->merge_clusters (base, end); 1004 else 1005 { 1006 /* Note! syllable() is a one-byte field. */ 1007 for (unsigned int i = base; i < end; i++) 1008 if (info[i].syllable() != 255) 1009 { 1010 unsigned int max = i; 1011 unsigned int j = start + info[i].syllable(); 1012 while (j != i) 1013 { 1014 max = MAX (max, j); 1015 unsigned int next = start + info[j].syllable(); 1016 info[j].syllable() = 255; /* So we don't process j later again. */ 1017 j = next; 1018 } 1019 if (i != max) 1020 buffer->merge_clusters (i, max + 1); 1021 } 1022 } 1023 1024 /* Put syllable back in. */ 1025 for (unsigned int i = start; i < end; i++) 1026 info[i].syllable() = syllable; 1027 } 1028 1029 /* Setup masks now */ 1030 1031 { 1032 hb_mask_t mask; 1033 1034 /* Reph */ 1035 for (unsigned int i = start; i < end && info[i].indic_position() == POS_RA_TO_BECOME_REPH; i++) 1036 info[i].mask |= indic_plan->mask_array[RPHF]; 1037 1038 /* Pre-base */ 1039 mask = indic_plan->mask_array[HALF]; 1040 if (!indic_plan->is_old_spec && 1041 indic_plan->config->blwf_mode == BLWF_MODE_PRE_AND_POST) 1042 mask |= indic_plan->mask_array[BLWF]; 1043 for (unsigned int i = start; i < base; i++) 1044 info[i].mask |= mask; 1045 /* Base */ 1046 mask = 0; 1047 if (base < end) 1048 info[base].mask |= mask; 1049 /* Post-base */ 1050 mask = indic_plan->mask_array[BLWF] | indic_plan->mask_array[ABVF] | indic_plan->mask_array[PSTF]; 1051 for (unsigned int i = base + 1; i < end; i++) 1052 info[i].mask |= mask; 1053 } 1054 1055 if (indic_plan->is_old_spec && 1056 buffer->props.script == HB_SCRIPT_DEVANAGARI) 1057 { 1058 /* Old-spec eye-lash Ra needs special handling. From the 1059 * spec: 1060 * 1061 * "The feature 'below-base form' is applied to consonants 1062 * having below-base forms and following the base consonant. 1063 * The exception is vattu, which may appear below half forms 1064 * as well as below the base glyph. The feature 'below-base 1065 * form' will be applied to all such occurrences of Ra as well." 1066 * 1067 * Test case: U+0924,U+094D,U+0930,U+094d,U+0915 1068 * with Sanskrit 2003 font. 1069 * 1070 * However, note that Ra,Halant,ZWJ is the correct way to 1071 * request eyelash form of Ra, so we wouldbn't inhibit it 1072 * in that sequence. 1073 * 1074 * Test case: U+0924,U+094D,U+0930,U+094d,U+200D,U+0915 1075 */ 1076 for (unsigned int i = start; i + 1 < base; i++) 1077 if (info[i ].indic_category() == OT_Ra && 1078 info[i+1].indic_category() == OT_H && 1079 (i + 2 == base || 1080 info[i+2].indic_category() != OT_ZWJ)) 1081 { 1082 info[i ].mask |= indic_plan->mask_array[BLWF]; 1083 info[i+1].mask |= indic_plan->mask_array[BLWF]; 1084 } 1085 } 1086 1087 unsigned int pref_len = indic_plan->config->pref_len; 1088 if (indic_plan->mask_array[PREF] && base + pref_len < end) 1089 { 1090 assert (1 <= pref_len && pref_len <= 2); 1091 /* Find a Halant,Ra sequence and mark it for pre-base reordering processing. */ 1092 for (unsigned int i = base + 1; i + pref_len - 1 < end; i++) { 1093 hb_codepoint_t glyphs[2]; 1094 for (unsigned int j = 0; j < pref_len; j++) 1095 glyphs[j] = info[i + j].codepoint; 1096 if (indic_plan->pref.would_substitute (glyphs, pref_len, face)) 1097 { 1098 for (unsigned int j = 0; j < pref_len; j++) 1099 info[i++].mask |= indic_plan->mask_array[PREF]; 1100 1101 /* Mark the subsequent stuff with 'cfar'. Used in Khmer. 1102 * Read the feature spec. 1103 * This allows distinguishing the following cases with MS Khmer fonts: 1104 * U+1784,U+17D2,U+179A,U+17D2,U+1782 1105 * U+1784,U+17D2,U+1782,U+17D2,U+179A 1106 */ 1107 if (indic_plan->mask_array[CFAR]) 1108 for (; i < end; i++) 1109 info[i].mask |= indic_plan->mask_array[CFAR]; 1110 1111 break; 1112 } 1113 } 1114 } 1115 1116 /* Apply ZWJ/ZWNJ effects */ 1117 for (unsigned int i = start + 1; i < end; i++) 1118 if (is_joiner (info[i])) { 1119 bool non_joiner = info[i].indic_category() == OT_ZWNJ; 1120 unsigned int j = i; 1121 1122 do { 1123 j--; 1124 1125 /* ZWJ/ZWNJ should disable CJCT. They do that by simply 1126 * being there, since we don't skip them for the CJCT 1127 * feature (ie. F_MANUAL_ZWJ) */ 1128 1129 /* A ZWNJ disables HALF. */ 1130 if (non_joiner) 1131 info[j].mask &= ~indic_plan->mask_array[HALF]; 1132 1133 } while (j > start && !is_consonant (info[j])); 1134 } 1135 } 1136 1137 1138 static void 1139 initial_reordering_vowel_syllable (const hb_ot_shape_plan_t *plan, 1140 hb_face_t *face, 1141 hb_buffer_t *buffer, 1142 unsigned int start, unsigned int end) 1143 { 1144 /* We made the vowels look like consonants. So let's call the consonant logic! */ 1145 initial_reordering_consonant_syllable (plan, face, buffer, start, end); 1146 } 1147 1148 static void 1149 initial_reordering_standalone_cluster (const hb_ot_shape_plan_t *plan, 1150 hb_face_t *face, 1151 hb_buffer_t *buffer, 1152 unsigned int start, unsigned int end) 1153 { 1154 /* We treat NBSP/dotted-circle as if they are consonants, so we should just chain. 1155 * Only if not in compatibility mode that is... */ 1156 1157 if (hb_options ().uniscribe_bug_compatible) 1158 { 1159 /* For dotted-circle, this is what Uniscribe does: 1160 * If dotted-circle is the last glyph, it just does nothing. 1161 * Ie. It doesn't form Reph. */ 1162 if (buffer->info[end - 1].indic_category() == OT_DOTTEDCIRCLE) 1163 return; 1164 } 1165 1166 initial_reordering_consonant_syllable (plan, face, buffer, start, end); 1167 } 1168 1169 static void 1170 initial_reordering_broken_cluster (const hb_ot_shape_plan_t *plan, 1171 hb_face_t *face, 1172 hb_buffer_t *buffer, 1173 unsigned int start, unsigned int end) 1174 { 1175 /* We already inserted dotted-circles, so just call the standalone_cluster. */ 1176 initial_reordering_standalone_cluster (plan, face, buffer, start, end); 1177 } 1178 1179 static void 1180 initial_reordering_avagraha_cluster (const hb_ot_shape_plan_t *plan HB_UNUSED, 1181 hb_face_t *face HB_UNUSED, 1182 hb_buffer_t *buffer HB_UNUSED, 1183 unsigned int start HB_UNUSED, unsigned int end HB_UNUSED) 1184 { 1185 /* Nothing to do right now. If we ever switch to using the output 1186 * buffer in the reordering process, we'd need to next_glyph() here. */ 1187 } 1188 1189 static void 1190 initial_reordering_non_indic_cluster (const hb_ot_shape_plan_t *plan HB_UNUSED, 1191 hb_face_t *face HB_UNUSED, 1192 hb_buffer_t *buffer HB_UNUSED, 1193 unsigned int start HB_UNUSED, unsigned int end HB_UNUSED) 1194 { 1195 /* Nothing to do right now. If we ever switch to using the output 1196 * buffer in the reordering process, we'd need to next_glyph() here. */ 1197 } 1198 1199 1200 static void 1201 initial_reordering_syllable (const hb_ot_shape_plan_t *plan, 1202 hb_face_t *face, 1203 hb_buffer_t *buffer, 1204 unsigned int start, unsigned int end) 1205 { 1206 syllable_type_t syllable_type = (syllable_type_t) (buffer->info[start].syllable() & 0x0F); 1207 switch (syllable_type) { 1208 case consonant_syllable: initial_reordering_consonant_syllable (plan, face, buffer, start, end); return; 1209 case vowel_syllable: initial_reordering_vowel_syllable (plan, face, buffer, start, end); return; 1210 case standalone_cluster: initial_reordering_standalone_cluster (plan, face, buffer, start, end); return; 1211 case avagraha_cluster: initial_reordering_avagraha_cluster (plan, face, buffer, start, end); return; 1212 case broken_cluster: initial_reordering_broken_cluster (plan, face, buffer, start, end); return; 1213 case non_indic_cluster: initial_reordering_non_indic_cluster (plan, face, buffer, start, end); return; 1214 } 1215 } 1216 1217 static inline void 1218 insert_dotted_circles (const hb_ot_shape_plan_t *plan HB_UNUSED, 1219 hb_font_t *font, 1220 hb_buffer_t *buffer) 1221 { 1222 /* Note: This loop is extra overhead, but should not be measurable. */ 1223 bool has_broken_syllables = false; 1224 unsigned int count = buffer->len; 1225 for (unsigned int i = 0; i < count; i++) 1226 if ((buffer->info[i].syllable() & 0x0F) == broken_cluster) { 1227 has_broken_syllables = true; 1228 break; 1229 } 1230 if (likely (!has_broken_syllables)) 1231 return; 1232 1233 1234 hb_codepoint_t dottedcircle_glyph; 1235 if (!font->get_glyph (0x25CC, 0, &dottedcircle_glyph)) 1236 return; 1237 1238 hb_glyph_info_t dottedcircle = {0}; 1239 dottedcircle.codepoint = 0x25CC; 1240 set_indic_properties (dottedcircle); 1241 dottedcircle.codepoint = dottedcircle_glyph; 1242 1243 buffer->clear_output (); 1244 1245 buffer->idx = 0; 1246 unsigned int last_syllable = 0; 1247 while (buffer->idx < buffer->len) 1248 { 1249 unsigned int syllable = buffer->cur().syllable(); 1250 syllable_type_t syllable_type = (syllable_type_t) (syllable & 0x0F); 1251 if (unlikely (last_syllable != syllable && syllable_type == broken_cluster)) 1252 { 1253 last_syllable = syllable; 1254 1255 hb_glyph_info_t info = dottedcircle; 1256 info.cluster = buffer->cur().cluster; 1257 info.mask = buffer->cur().mask; 1258 info.syllable() = buffer->cur().syllable(); 1259 1260 /* Insert dottedcircle after possible Repha. */ 1261 while (buffer->idx < buffer->len && 1262 last_syllable == buffer->cur().syllable() && 1263 buffer->cur().indic_category() == OT_Repha) 1264 buffer->next_glyph (); 1265 1266 buffer->output_info (info); 1267 } 1268 else 1269 buffer->next_glyph (); 1270 } 1271 1272 buffer->swap_buffers (); 1273 } 1274 1275 static void 1276 initial_reordering (const hb_ot_shape_plan_t *plan, 1277 hb_font_t *font, 1278 hb_buffer_t *buffer) 1279 { 1280 update_consonant_positions (plan, font, buffer); 1281 insert_dotted_circles (plan, font, buffer); 1282 1283 hb_glyph_info_t *info = buffer->info; 1284 unsigned int count = buffer->len; 1285 if (unlikely (!count)) return; 1286 unsigned int last = 0; 1287 unsigned int last_syllable = info[0].syllable(); 1288 for (unsigned int i = 1; i < count; i++) 1289 if (last_syllable != info[i].syllable()) { 1290 initial_reordering_syllable (plan, font->face, buffer, last, i); 1291 last = i; 1292 last_syllable = info[last].syllable(); 1293 } 1294 initial_reordering_syllable (plan, font->face, buffer, last, count); 1295 } 1296 1297 static void 1298 final_reordering_syllable (const hb_ot_shape_plan_t *plan, 1299 hb_buffer_t *buffer, 1300 unsigned int start, unsigned int end) 1301 { 1302 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; 1303 hb_glyph_info_t *info = buffer->info; 1304 1305 /* 4. Final reordering: 1306 * 1307 * After the localized forms and basic shaping forms GSUB features have been 1308 * applied (see below), the shaping engine performs some final glyph 1309 * reordering before applying all the remaining font features to the entire 1310 * cluster. 1311 */ 1312 1313 /* Find base again */ 1314 unsigned int base; 1315 for (base = start; base < end; base++) 1316 if (info[base].indic_position() >= POS_BASE_C) { 1317 if (start < base && info[base].indic_position() > POS_BASE_C) 1318 base--; 1319 break; 1320 } 1321 if (base == end && start < base && 1322 info[base - 1].indic_category() != OT_ZWJ) 1323 base--; 1324 while (start < base && 1325 (info[base].indic_category() == OT_H || 1326 info[base].indic_category() == OT_N)) 1327 base--; 1328 1329 1330 /* o Reorder matras: 1331 * 1332 * If a pre-base matra character had been reordered before applying basic 1333 * features, the glyph can be moved closer to the main consonant based on 1334 * whether half-forms had been formed. Actual position for the matra is 1335 * defined as after last standalone halant glyph, after initial matra 1336 * position and before the main consonant. If ZWJ or ZWNJ follow this 1337 * halant, position is moved after it. 1338 */ 1339 1340 if (start + 1 < end && start < base) /* Otherwise there can't be any pre-base matra characters. */ 1341 { 1342 /* If we lost track of base, alas, position before last thingy. */ 1343 unsigned int new_pos = base == end ? base - 2 : base - 1; 1344 1345 /* Malayalam / Tamil do not have "half" forms or explicit virama forms. 1346 * The glyphs formed by 'half' are Chillus or ligated explicit viramas. 1347 * We want to position matra after them. 1348 */ 1349 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL) 1350 { 1351 while (new_pos > start && 1352 !(is_one_of (info[new_pos], (FLAG (OT_M) | FLAG (OT_H) | FLAG (OT_Coeng))))) 1353 new_pos--; 1354 1355 /* If we found no Halant we are done. 1356 * Otherwise only proceed if the Halant does 1357 * not belong to the Matra itself! */ 1358 if (is_halant_or_coeng (info[new_pos]) && 1359 info[new_pos].indic_position() != POS_PRE_M) 1360 { 1361 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */ 1362 if (new_pos + 1 < end && is_joiner (info[new_pos + 1])) 1363 new_pos++; 1364 } 1365 else 1366 new_pos = start; /* No move. */ 1367 } 1368 1369 if (start < new_pos && info[new_pos].indic_position () != POS_PRE_M) 1370 { 1371 /* Now go see if there's actually any matras... */ 1372 for (unsigned int i = new_pos; i > start; i--) 1373 if (info[i - 1].indic_position () == POS_PRE_M) 1374 { 1375 unsigned int old_pos = i - 1; 1376 hb_glyph_info_t tmp = info[old_pos]; 1377 memmove (&info[old_pos], &info[old_pos + 1], (new_pos - old_pos) * sizeof (info[0])); 1378 info[new_pos] = tmp; 1379 if (old_pos < base && base <= new_pos) /* Shouldn't actually happen. */ 1380 base--; 1381 buffer->merge_clusters (new_pos, MIN (end, base + 1)); 1382 new_pos--; 1383 } 1384 } else { 1385 for (unsigned int i = start; i < base; i++) 1386 if (info[i].indic_position () == POS_PRE_M) { 1387 buffer->merge_clusters (i, MIN (end, base + 1)); 1388 break; 1389 } 1390 } 1391 } 1392 1393 1394 /* o Reorder reph: 1395 * 1396 * Rephs original position is always at the beginning of the syllable, 1397 * (i.e. it is not reordered at the character reordering stage). However, 1398 * it will be reordered according to the basic-forms shaping results. 1399 * Possible positions for reph, depending on the script, are; after main, 1400 * before post-base consonant forms, and after post-base consonant forms. 1401 */ 1402 1403 /* Two cases: 1404 * 1405 * - If repha is encoded as a sequence of characters (Ra,H or Ra,H,ZWJ), then 1406 * we should only move it if the sequence ligated to the repha form. 1407 * 1408 * - If repha is encoded separately and in the logical position, we should only 1409 * move it if it did NOT ligate. If it ligated, it's probably the font trying 1410 * to make it work without the reordering. 1411 */ 1412 if (start + 1 < end && 1413 info[start].indic_position() == POS_RA_TO_BECOME_REPH && 1414 ((info[start].indic_category() == OT_Repha) ^ 1415 _hb_glyph_info_ligated (&info[start]))) 1416 { 1417 unsigned int new_reph_pos; 1418 reph_position_t reph_pos = indic_plan->config->reph_pos; 1419 1420 assert (reph_pos != REPH_POS_DONT_CARE); 1421 1422 /* 1. If reph should be positioned after post-base consonant forms, 1423 * proceed to step 5. 1424 */ 1425 if (reph_pos == REPH_POS_AFTER_POST) 1426 { 1427 goto reph_step_5; 1428 } 1429 1430 /* 2. If the reph repositioning class is not after post-base: target 1431 * position is after the first explicit halant glyph between the 1432 * first post-reph consonant and last main consonant. If ZWJ or ZWNJ 1433 * are following this halant, position is moved after it. If such 1434 * position is found, this is the target position. Otherwise, 1435 * proceed to the next step. 1436 * 1437 * Note: in old-implementation fonts, where classifications were 1438 * fixed in shaping engine, there was no case where reph position 1439 * will be found on this step. 1440 */ 1441 { 1442 new_reph_pos = start + 1; 1443 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos])) 1444 new_reph_pos++; 1445 1446 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos])) 1447 { 1448 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */ 1449 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1])) 1450 new_reph_pos++; 1451 goto reph_move; 1452 } 1453 } 1454 1455 /* 3. If reph should be repositioned after the main consonant: find the 1456 * first consonant not ligated with main, or find the first 1457 * consonant that is not a potential pre-base reordering Ra. 1458 */ 1459 if (reph_pos == REPH_POS_AFTER_MAIN) 1460 { 1461 new_reph_pos = base; 1462 while (new_reph_pos + 1 < end && info[new_reph_pos + 1].indic_position() <= POS_AFTER_MAIN) 1463 new_reph_pos++; 1464 if (new_reph_pos < end) 1465 goto reph_move; 1466 } 1467 1468 /* 4. If reph should be positioned before post-base consonant, find 1469 * first post-base classified consonant not ligated with main. If no 1470 * consonant is found, the target position should be before the 1471 * first matra, syllable modifier sign or vedic sign. 1472 */ 1473 /* This is our take on what step 4 is trying to say (and failing, BADLY). */ 1474 if (reph_pos == REPH_POS_AFTER_SUB) 1475 { 1476 new_reph_pos = base; 1477 while (new_reph_pos < end && 1478 !( FLAG (info[new_reph_pos + 1].indic_position()) & (FLAG (POS_POST_C) | FLAG (POS_AFTER_POST) | FLAG (POS_SMVD)))) 1479 new_reph_pos++; 1480 if (new_reph_pos < end) 1481 goto reph_move; 1482 } 1483 1484 /* 5. If no consonant is found in steps 3 or 4, move reph to a position 1485 * immediately before the first post-base matra, syllable modifier 1486 * sign or vedic sign that has a reordering class after the intended 1487 * reph position. For example, if the reordering position for reph 1488 * is post-main, it will skip above-base matras that also have a 1489 * post-main position. 1490 */ 1491 reph_step_5: 1492 { 1493 /* Copied from step 2. */ 1494 new_reph_pos = start + 1; 1495 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos])) 1496 new_reph_pos++; 1497 1498 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos])) 1499 { 1500 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */ 1501 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1])) 1502 new_reph_pos++; 1503 goto reph_move; 1504 } 1505 } 1506 1507 /* 6. Otherwise, reorder reph to the end of the syllable. 1508 */ 1509 { 1510 new_reph_pos = end - 1; 1511 while (new_reph_pos > start && info[new_reph_pos].indic_position() == POS_SMVD) 1512 new_reph_pos--; 1513 1514 /* 1515 * If the Reph is to be ending up after a Matra,Halant sequence, 1516 * position it before that Halant so it can interact with the Matra. 1517 * However, if it's a plain Consonant,Halant we shouldn't do that. 1518 * Uniscribe doesn't do this. 1519 * TEST: U+0930,U+094D,U+0915,U+094B,U+094D 1520 */ 1521 if (!hb_options ().uniscribe_bug_compatible && 1522 unlikely (is_halant_or_coeng (info[new_reph_pos]))) { 1523 for (unsigned int i = base + 1; i < new_reph_pos; i++) 1524 if (info[i].indic_category() == OT_M) { 1525 /* Ok, got it. */ 1526 new_reph_pos--; 1527 } 1528 } 1529 goto reph_move; 1530 } 1531 1532 reph_move: 1533 { 1534 buffer->merge_clusters (start, new_reph_pos + 1); 1535 1536 /* Move */ 1537 hb_glyph_info_t reph = info[start]; 1538 memmove (&info[start], &info[start + 1], (new_reph_pos - start) * sizeof (info[0])); 1539 info[new_reph_pos] = reph; 1540 if (start < base && base <= new_reph_pos) 1541 base--; 1542 } 1543 } 1544 1545 1546 /* o Reorder pre-base reordering consonants: 1547 * 1548 * If a pre-base reordering consonant is found, reorder it according to 1549 * the following rules: 1550 */ 1551 1552 if (indic_plan->mask_array[PREF] && base + 1 < end) /* Otherwise there can't be any pre-base reordering Ra. */ 1553 { 1554 unsigned int pref_len = indic_plan->config->pref_len; 1555 for (unsigned int i = base + 1; i < end; i++) 1556 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0) 1557 { 1558 /* 1. Only reorder a glyph produced by substitution during application 1559 * of the <pref> feature. (Note that a font may shape a Ra consonant with 1560 * the feature generally but block it in certain contexts.) 1561 */ 1562 /* Note: We just check that something got substituted. We don't check that 1563 * the <pref> feature actually did it... 1564 * 1565 * If pref len is longer than one, then only reorder if it ligated. If 1566 * pref len is one, only reorder if it didn't ligate with other things. */ 1567 if (_hb_glyph_info_substituted (&info[i]) && 1568 ((pref_len == 1) ^ _hb_glyph_info_ligated (&info[i]))) 1569 { 1570 /* 1571 * 2. Try to find a target position the same way as for pre-base matra. 1572 * If it is found, reorder pre-base consonant glyph. 1573 * 1574 * 3. If position is not found, reorder immediately before main 1575 * consonant. 1576 */ 1577 1578 unsigned int new_pos = base; 1579 /* Malayalam / Tamil do not have "half" forms or explicit virama forms. 1580 * The glyphs formed by 'half' are Chillus or ligated explicit viramas. 1581 * We want to position matra after them. 1582 */ 1583 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL) 1584 { 1585 while (new_pos > start && 1586 !(is_one_of (info[new_pos - 1], FLAG(OT_M) | HALANT_OR_COENG_FLAGS))) 1587 new_pos--; 1588 1589 /* In Khmer coeng model, a H,Ra can go *after* matras. If it goes after a 1590 * split matra, it should be reordered to *before* the left part of such matra. */ 1591 if (new_pos > start && info[new_pos - 1].indic_category() == OT_M) 1592 { 1593 unsigned int old_pos = i; 1594 for (unsigned int i = base + 1; i < old_pos; i++) 1595 if (info[i].indic_category() == OT_M) 1596 { 1597 new_pos--; 1598 break; 1599 } 1600 } 1601 } 1602 1603 if (new_pos > start && is_halant_or_coeng (info[new_pos - 1])) 1604 { 1605 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */ 1606 if (new_pos < end && is_joiner (info[new_pos])) 1607 new_pos++; 1608 } 1609 1610 { 1611 unsigned int old_pos = i; 1612 buffer->merge_clusters (new_pos, old_pos + 1); 1613 hb_glyph_info_t tmp = info[old_pos]; 1614 memmove (&info[new_pos + 1], &info[new_pos], (old_pos - new_pos) * sizeof (info[0])); 1615 info[new_pos] = tmp; 1616 if (new_pos <= base && base < old_pos) 1617 base++; 1618 } 1619 } 1620 1621 break; 1622 } 1623 } 1624 1625 1626 /* Apply 'init' to the Left Matra if it's a word start. */ 1627 if (info[start].indic_position () == POS_PRE_M && 1628 (!start || 1629 !(FLAG (_hb_glyph_info_get_general_category (&info[start - 1])) & 1630 FLAG_RANGE (HB_UNICODE_GENERAL_CATEGORY_FORMAT, HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK)))) 1631 info[start].mask |= indic_plan->mask_array[INIT]; 1632 1633 1634 /* 1635 * Finish off the clusters and go home! 1636 */ 1637 if (hb_options ().uniscribe_bug_compatible) 1638 { 1639 switch ((hb_tag_t) plan->props.script) 1640 { 1641 case HB_SCRIPT_TAMIL: 1642 case HB_SCRIPT_SINHALA: 1643 break; 1644 1645 default: 1646 /* Uniscribe merges the entire cluster... Except for Tamil & Sinhala. 1647 * This means, half forms are submerged into the main consonants cluster. 1648 * This is unnecessary, and makes cursor positioning harder, but that's what 1649 * Uniscribe does. */ 1650 buffer->merge_clusters (start, end); 1651 break; 1652 } 1653 } 1654 } 1655 1656 1657 static void 1658 final_reordering (const hb_ot_shape_plan_t *plan, 1659 hb_font_t *font HB_UNUSED, 1660 hb_buffer_t *buffer) 1661 { 1662 unsigned int count = buffer->len; 1663 if (unlikely (!count)) return; 1664 1665 hb_glyph_info_t *info = buffer->info; 1666 unsigned int last = 0; 1667 unsigned int last_syllable = info[0].syllable(); 1668 for (unsigned int i = 1; i < count; i++) 1669 if (last_syllable != info[i].syllable()) { 1670 final_reordering_syllable (plan, buffer, last, i); 1671 last = i; 1672 last_syllable = info[last].syllable(); 1673 } 1674 final_reordering_syllable (plan, buffer, last, count); 1675 1676 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_category); 1677 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_position); 1678 } 1679 1680 1681 static void 1682 clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED, 1683 hb_font_t *font HB_UNUSED, 1684 hb_buffer_t *buffer) 1685 { 1686 hb_glyph_info_t *info = buffer->info; 1687 unsigned int count = buffer->len; 1688 for (unsigned int i = 0; i < count; i++) 1689 info[i].syllable() = 0; 1690 } 1691 1692 1693 static hb_ot_shape_normalization_mode_t 1694 normalization_preference_indic (const hb_segment_properties_t *props HB_UNUSED) 1695 { 1696 return HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT; 1697 } 1698 1699 static bool 1700 decompose_indic (const hb_ot_shape_normalize_context_t *c, 1701 hb_codepoint_t ab, 1702 hb_codepoint_t *a, 1703 hb_codepoint_t *b) 1704 { 1705 switch (ab) 1706 { 1707 /* Don't decompose these. */ 1708 case 0x0931 : return false; 1709 case 0x0B94 : return false; 1710 1711 1712 /* 1713 * Decompose split matras that don't have Unicode decompositions. 1714 */ 1715 1716 case 0x0F77 : *a = 0x0FB2; *b= 0x0F81; return true; 1717 case 0x0F79 : *a = 0x0FB3; *b= 0x0F81; return true; 1718 case 0x17BE : *a = 0x17C1; *b= 0x17BE; return true; 1719 case 0x17BF : *a = 0x17C1; *b= 0x17BF; return true; 1720 case 0x17C0 : *a = 0x17C1; *b= 0x17C0; return true; 1721 case 0x17C4 : *a = 0x17C1; *b= 0x17C4; return true; 1722 case 0x17C5 : *a = 0x17C1; *b= 0x17C5; return true; 1723 case 0x1925 : *a = 0x1920; *b= 0x1923; return true; 1724 case 0x1926 : *a = 0x1920; *b= 0x1924; return true; 1725 case 0x1B3C : *a = 0x1B42; *b= 0x1B3C; return true; 1726 case 0x1112E : *a = 0x11127; *b= 0x11131; return true; 1727 case 0x1112F : *a = 0x11127; *b= 0x11132; return true; 1728 #if 0 1729 /* This one has no decomposition in Unicode, but needs no decomposition either. */ 1730 /* case 0x0AC9 : return false; */ 1731 case 0x0B57 : *a = no decomp, -> RIGHT; return true; 1732 case 0x1C29 : *a = no decomp, -> LEFT; return true; 1733 case 0xA9C0 : *a = no decomp, -> RIGHT; return true; 1734 case 0x111BF : *a = no decomp, -> ABOVE; return true; 1735 #endif 1736 } 1737 1738 if ((ab == 0x0DDA || hb_in_range<hb_codepoint_t> (ab, 0x0DDC, 0x0DDE))) 1739 { 1740 /* 1741 * Sinhala split matras... Let the fun begin. 1742 * 1743 * These four characters have Unicode decompositions. However, Uniscribe 1744 * decomposes them "Khmer-style", that is, it uses the character itself to 1745 * get the second half. The first half of all four decompositions is always 1746 * U+0DD9. 1747 * 1748 * Now, there are buggy fonts, namely, the widely used lklug.ttf, that are 1749 * broken with Uniscribe. But we need to support them. As such, we only 1750 * do the Uniscribe-style decomposition if the character is transformed into 1751 * its "sec.half" form by the 'pstf' feature. Otherwise, we fall back to 1752 * Unicode decomposition. 1753 * 1754 * Note that we can't unconditionally use Unicode decomposition. That would 1755 * break some other fonts, that are designed to work with Uniscribe, and 1756 * don't have positioning features for the Unicode-style decomposition. 1757 * 1758 * Argh... 1759 * 1760 * The Uniscribe behavior is now documented in the newly published Sinhala 1761 * spec in 2012: 1762 * 1763 * http://www.microsoft.com/typography/OpenTypeDev/sinhala/intro.htm#shaping 1764 */ 1765 1766 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) c->plan->data; 1767 1768 hb_codepoint_t glyph; 1769 1770 if (hb_options ().uniscribe_bug_compatible || 1771 (c->font->get_glyph (ab, 0, &glyph) && 1772 indic_plan->pstf.would_substitute (&glyph, 1, c->font->face))) 1773 { 1774 /* Ok, safe to use Uniscribe-style decomposition. */ 1775 *a = 0x0DD9; 1776 *b = ab; 1777 return true; 1778 } 1779 } 1780 1781 return c->unicode->decompose (ab, a, b); 1782 } 1783 1784 static bool 1785 compose_indic (const hb_ot_shape_normalize_context_t *c, 1786 hb_codepoint_t a, 1787 hb_codepoint_t b, 1788 hb_codepoint_t *ab) 1789 { 1790 /* Avoid recomposing split matras. */ 1791 if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a))) 1792 return false; 1793 1794 /* Composition-exclusion exceptions that we want to recompose. */ 1795 if (a == 0x09AF && b == 0x09BC) { *ab = 0x09DF; return true; } 1796 1797 return c->unicode->compose (a, b, ab); 1798 } 1799 1800 1801 const hb_ot_complex_shaper_t _hb_ot_complex_shaper_indic = 1802 { 1803 "indic", 1804 collect_features_indic, 1805 override_features_indic, 1806 data_create_indic, 1807 data_destroy_indic, 1808 NULL, /* preprocess_text */ 1809 normalization_preference_indic, 1810 decompose_indic, 1811 compose_indic, 1812 setup_masks_indic, 1813 HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE, 1814 false, /* fallback_position */ 1815 }; 1816
