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) & ~0x7Fu) == (Base)) 41 42 #define IS_DEVA(u) (IN_HALF_BLOCK (u, 0x0900u)) 43 #define IS_BENG(u) (IN_HALF_BLOCK (u, 0x0980u)) 44 #define IS_GURU(u) (IN_HALF_BLOCK (u, 0x0A00u)) 45 #define IS_GUJR(u) (IN_HALF_BLOCK (u, 0x0A80u)) 46 #define IS_ORYA(u) (IN_HALF_BLOCK (u, 0x0B00u)) 47 #define IS_TAML(u) (IN_HALF_BLOCK (u, 0x0B80u)) 48 #define IS_TELU(u) (IN_HALF_BLOCK (u, 0x0C00u)) 49 #define IS_KNDA(u) (IN_HALF_BLOCK (u, 0x0C80u)) 50 #define IS_MLYM(u) (IN_HALF_BLOCK (u, 0x0D00u)) 51 #define IS_SINH(u) (IN_HALF_BLOCK (u, 0x0D80u)) 52 #define IS_KHMR(u) (IN_HALF_BLOCK (u, 0x1780u)) 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 <= 0x0C42u ? POS_BEFORE_SUB : POS_AFTER_SUB) : \ 64 IS_KNDA(u) ? (u < 0x0CC3u || u > 0xCD6u ? 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 0x0930u, /* Devanagari */ 116 0x09B0u, /* Bengali */ 117 0x09F0u, /* Bengali */ 118 0x0A30u, /* Gurmukhi */ /* No Reph */ 119 0x0AB0u, /* Gujarati */ 120 0x0B30u, /* Oriya */ 121 0x0BB0u, /* Tamil */ /* No Reph */ 122 0x0C30u, /* Telugu */ /* Reph formed only with ZWJ */ 123 0x0CB0u, /* Kannada */ 124 0x0D30u, /* Malayalam */ /* No Reph, Logical Repha */ 125 126 0x0DBBu, /* Sinhala */ /* Reph formed only with ZWJ */ 127 128 0x179Au, /* 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_SAFE (info.indic_category()) & flags); 146 } 147 148 static inline bool 149 is_joiner (const hb_glyph_info_t &info) 150 { 151 return is_one_of (info, JOINER_FLAGS); 152 } 153 154 static inline bool 155 is_consonant (const hb_glyph_info_t &info) 156 { 157 return is_one_of (info, CONSONANT_FLAGS); 158 } 159 160 static inline bool 161 is_halant_or_coeng (const hb_glyph_info_t &info) 162 { 163 return is_one_of (info, HALANT_OR_COENG_FLAGS); 164 } 165 166 static inline void 167 set_indic_properties (hb_glyph_info_t &info) 168 { 169 hb_codepoint_t u = info.codepoint; 170 unsigned int type = hb_indic_get_categories (u); 171 indic_category_t cat = (indic_category_t) (type & 0x7Fu); 172 indic_position_t pos = (indic_position_t) (type >> 8); 173 174 175 /* 176 * Re-assign category 177 */ 178 179 180 /* The spec says U+0952 is OT_A. However, testing shows that Uniscribe 181 * treats a whole bunch of characters similarly. 182 * TESTS: For example, for U+0951: 183 * U+092E,U+0947,U+0952 184 * U+092E,U+0952,U+0947 185 * U+092E,U+0947,U+0951 186 * U+092E,U+0951,U+0947 187 * U+092E,U+0951,U+0952 188 * U+092E,U+0952,U+0951 189 */ 190 if (unlikely (hb_in_ranges (u, 0x0951u, 0x0952u, 191 0x1CD0u, 0x1CD2u, 192 0x1CD4u, 0x1CE1u) || 193 u == 0x1CF4u)) 194 cat = OT_A; 195 /* The following act more like the Bindus. */ 196 else if (unlikely (hb_in_range (u, 0x0953u, 0x0954u))) 197 cat = OT_SM; 198 /* The following act like consonants. */ 199 else if (unlikely (hb_in_ranges (u, 0x0A72u, 0x0A73u, 200 0x1CF5u, 0x1CF6u))) 201 cat = OT_C; 202 /* TODO: The following should only be allowed after a Visarga. 203 * For now, just treat them like regular tone marks. */ 204 else if (unlikely (hb_in_range (u, 0x1CE2u, 0x1CE8u))) 205 cat = OT_A; 206 /* TODO: The following should only be allowed after some of 207 * the nasalization marks, maybe only for U+1CE9..U+1CF1. 208 * For now, just treat them like tone marks. */ 209 else if (unlikely (u == 0x1CEDu)) 210 cat = OT_A; 211 /* The following take marks in standalone clusters, similar to Avagraha. */ 212 else if (unlikely (hb_in_ranges (u, 0xA8F2u, 0xA8F7u, 213 0x1CE9u, 0x1CECu, 214 0x1CEEu, 0x1CF1u))) 215 { 216 cat = OT_Symbol; 217 ASSERT_STATIC ((int) INDIC_SYLLABIC_CATEGORY_AVAGRAHA == OT_Symbol); 218 } 219 else if (unlikely (hb_in_range (u, 0x17CDu, 0x17D1u) || 220 u == 0x17CBu || u == 0x17D3u || u == 0x17DDu)) /* Khmer Various signs */ 221 { 222 /* These are like Top Matras. */ 223 cat = OT_M; 224 pos = POS_ABOVE_C; 225 } 226 else if (unlikely (u == 0x17C6u)) cat = OT_N; /* Khmer Bindu doesn't like to be repositioned. */ 227 else if (unlikely (u == 0x17D2u)) cat = OT_Coeng; /* Khmer coeng */ 228 else if (unlikely (hb_in_range (u, 0x2010u, 0x2011u))) 229 cat = OT_PLACEHOLDER; 230 else if (unlikely (u == 0x25CCu)) cat = OT_DOTTEDCIRCLE; 231 else if (unlikely (u == 0xA982u)) cat = OT_SM; /* Javanese repha. */ 232 else if (unlikely (u == 0xA9BEu)) cat = OT_CM2; /* Javanese medial ya. */ 233 else if (unlikely (u == 0xA9BDu)) { cat = OT_M; pos = POS_POST_C; } /* Javanese vocalic r. */ 234 235 236 /* 237 * Re-assign position. 238 */ 239 240 if ((FLAG_SAFE (cat) & CONSONANT_FLAGS)) 241 { 242 pos = POS_BASE_C; 243 if (is_ra (u)) 244 cat = OT_Ra; 245 } 246 else if (cat == OT_M) 247 { 248 pos = matra_position (u, pos); 249 } 250 else if ((FLAG_SAFE (cat) & (FLAG (OT_SM) | FLAG (OT_VD) | FLAG (OT_A) | FLAG (OT_Symbol)))) 251 { 252 pos = POS_SMVD; 253 } 254 255 if (unlikely (u == 0x0B01u)) pos = POS_BEFORE_SUB; /* Oriya Bindu is BeforeSub in the spec. */ 256 257 258 259 info.indic_category() = cat; 260 info.indic_position() = pos; 261 } 262 263 /* 264 * Things above this line should ideally be moved to the Indic table itself. 265 */ 266 267 268 /* 269 * Indic configurations. Note that we do not want to keep every single script-specific 270 * behavior in these tables necessarily. This should mainly be used for per-script 271 * properties that are cheaper keeping here, than in the code. Ie. if, say, one and 272 * only one script has an exception, that one script can be if'ed directly in the code, 273 * instead of adding a new flag in these structs. 274 */ 275 276 enum base_position_t { 277 BASE_POS_FIRST, 278 BASE_POS_LAST_SINHALA, 279 BASE_POS_LAST 280 }; 281 enum reph_position_t { 282 REPH_POS_AFTER_MAIN = POS_AFTER_MAIN, 283 REPH_POS_BEFORE_SUB = POS_BEFORE_SUB, 284 REPH_POS_AFTER_SUB = POS_AFTER_SUB, 285 REPH_POS_BEFORE_POST = POS_BEFORE_POST, 286 REPH_POS_AFTER_POST = POS_AFTER_POST, 287 REPH_POS_DONT_CARE = POS_RA_TO_BECOME_REPH 288 }; 289 enum reph_mode_t { 290 REPH_MODE_IMPLICIT, /* Reph formed out of initial Ra,H sequence. */ 291 REPH_MODE_EXPLICIT, /* Reph formed out of initial Ra,H,ZWJ sequence. */ 292 REPH_MODE_VIS_REPHA, /* Encoded Repha character, no reordering needed. */ 293 REPH_MODE_LOG_REPHA /* Encoded Repha character, needs reordering. */ 294 }; 295 enum blwf_mode_t { 296 BLWF_MODE_PRE_AND_POST, /* Below-forms feature applied to pre-base and post-base. */ 297 BLWF_MODE_POST_ONLY /* Below-forms feature applied to post-base only. */ 298 }; 299 enum pref_len_t { 300 PREF_LEN_1 = 1, 301 PREF_LEN_2 = 2, 302 PREF_LEN_DONT_CARE = PREF_LEN_2 303 }; 304 struct indic_config_t 305 { 306 hb_script_t script; 307 bool has_old_spec; 308 hb_codepoint_t virama; 309 base_position_t base_pos; 310 reph_position_t reph_pos; 311 reph_mode_t reph_mode; 312 blwf_mode_t blwf_mode; 313 pref_len_t pref_len; 314 }; 315 316 static const indic_config_t indic_configs[] = 317 { 318 /* Default. Should be first. */ 319 {HB_SCRIPT_INVALID, false, 0,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_1}, 320 {HB_SCRIPT_DEVANAGARI,true, 0x094Du,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 321 {HB_SCRIPT_BENGALI, true, 0x09CDu,BASE_POS_LAST, REPH_POS_AFTER_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 322 {HB_SCRIPT_GURMUKHI, true, 0x0A4Du,BASE_POS_LAST, REPH_POS_BEFORE_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 323 {HB_SCRIPT_GUJARATI, true, 0x0ACDu,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 324 {HB_SCRIPT_ORIYA, true, 0x0B4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 325 {HB_SCRIPT_TAMIL, true, 0x0BCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_2}, 326 {HB_SCRIPT_TELUGU, true, 0x0C4Du,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_EXPLICIT, BLWF_MODE_POST_ONLY, PREF_LEN_2}, 327 {HB_SCRIPT_KANNADA, true, 0x0CCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_POST_ONLY, PREF_LEN_2}, 328 {HB_SCRIPT_MALAYALAM, true, 0x0D4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_LOG_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2}, 329 {HB_SCRIPT_SINHALA, false,0x0DCAu,BASE_POS_LAST_SINHALA, 330 REPH_POS_AFTER_MAIN, REPH_MODE_EXPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE}, 331 {HB_SCRIPT_KHMER, false,0x17D2u,BASE_POS_FIRST,REPH_POS_DONT_CARE, REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2}, 332 {HB_SCRIPT_JAVANESE, false,0xA9C0u,BASE_POS_FIRST,REPH_POS_DONT_CARE, REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_1}, 333 }; 334 335 336 337 /* 338 * Indic shaper. 339 */ 340 341 struct feature_list_t { 342 hb_tag_t tag; 343 hb_ot_map_feature_flags_t flags; 344 }; 345 346 static const feature_list_t 347 indic_features[] = 348 { 349 /* 350 * Basic features. 351 * These features are applied in order, one at a time, after initial_reordering. 352 */ 353 {HB_TAG('n','u','k','t'), F_GLOBAL}, 354 {HB_TAG('a','k','h','n'), F_GLOBAL}, 355 {HB_TAG('r','p','h','f'), F_NONE}, 356 {HB_TAG('r','k','r','f'), F_GLOBAL}, 357 {HB_TAG('p','r','e','f'), F_NONE}, 358 {HB_TAG('b','l','w','f'), F_NONE}, 359 {HB_TAG('a','b','v','f'), F_NONE}, 360 {HB_TAG('h','a','l','f'), F_NONE}, 361 {HB_TAG('p','s','t','f'), F_NONE}, 362 {HB_TAG('v','a','t','u'), F_GLOBAL}, 363 {HB_TAG('c','j','c','t'), F_GLOBAL}, 364 {HB_TAG('c','f','a','r'), F_NONE}, 365 /* 366 * Other features. 367 * These features are applied all at once, after final_reordering. 368 * Default Bengali font in Windows for example has intermixed 369 * lookups for init,pres,abvs,blws features. 370 */ 371 {HB_TAG('i','n','i','t'), F_NONE}, 372 {HB_TAG('p','r','e','s'), F_GLOBAL}, 373 {HB_TAG('a','b','v','s'), F_GLOBAL}, 374 {HB_TAG('b','l','w','s'), F_GLOBAL}, 375 {HB_TAG('p','s','t','s'), F_GLOBAL}, 376 {HB_TAG('h','a','l','n'), F_GLOBAL}, 377 /* Positioning features, though we don't care about the types. */ 378 {HB_TAG('d','i','s','t'), F_GLOBAL}, 379 {HB_TAG('a','b','v','m'), F_GLOBAL}, 380 {HB_TAG('b','l','w','m'), F_GLOBAL}, 381 }; 382 383 /* 384 * Must be in the same order as the indic_features array. 385 */ 386 enum { 387 _NUKT, 388 _AKHN, 389 RPHF, 390 _RKRF, 391 PREF, 392 BLWF, 393 ABVF, 394 HALF, 395 PSTF, 396 _VATU, 397 _CJCT, 398 CFAR, 399 400 INIT, 401 _PRES, 402 _ABVS, 403 _BLWS, 404 _PSTS, 405 _HALN, 406 _DIST, 407 _ABVM, 408 _BLWM, 409 410 INDIC_NUM_FEATURES, 411 INDIC_BASIC_FEATURES = INIT /* Don't forget to update this! */ 412 }; 413 414 static void 415 setup_syllables (const hb_ot_shape_plan_t *plan, 416 hb_font_t *font, 417 hb_buffer_t *buffer); 418 static void 419 initial_reordering (const hb_ot_shape_plan_t *plan, 420 hb_font_t *font, 421 hb_buffer_t *buffer); 422 static void 423 final_reordering (const hb_ot_shape_plan_t *plan, 424 hb_font_t *font, 425 hb_buffer_t *buffer); 426 static void 427 clear_syllables (const hb_ot_shape_plan_t *plan, 428 hb_font_t *font, 429 hb_buffer_t *buffer); 430 431 static void 432 collect_features_indic (hb_ot_shape_planner_t *plan) 433 { 434 hb_ot_map_builder_t *map = &plan->map; 435 436 /* Do this before any lookups have been applied. */ 437 map->add_gsub_pause (setup_syllables); 438 439 map->add_global_bool_feature (HB_TAG('l','o','c','l')); 440 /* The Indic specs do not require ccmp, but we apply it here since if 441 * there is a use of it, it's typically at the beginning. */ 442 map->add_global_bool_feature (HB_TAG('c','c','m','p')); 443 444 445 unsigned int i = 0; 446 map->add_gsub_pause (initial_reordering); 447 for (; i < INDIC_BASIC_FEATURES; i++) { 448 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ); 449 map->add_gsub_pause (NULL); 450 } 451 map->add_gsub_pause (final_reordering); 452 for (; i < INDIC_NUM_FEATURES; i++) { 453 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ); 454 } 455 456 map->add_global_bool_feature (HB_TAG('c','a','l','t')); 457 map->add_global_bool_feature (HB_TAG('c','l','i','g')); 458 459 map->add_gsub_pause (clear_syllables); 460 } 461 462 static void 463 override_features_indic (hb_ot_shape_planner_t *plan) 464 { 465 /* Uniscribe does not apply 'kern' in Khmer. */ 466 if (hb_options ().uniscribe_bug_compatible) 467 { 468 switch ((hb_tag_t) plan->props.script) 469 { 470 case HB_SCRIPT_KHMER: 471 plan->map.add_feature (HB_TAG('k','e','r','n'), 0, F_GLOBAL); 472 break; 473 } 474 } 475 476 plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL); 477 } 478 479 480 struct would_substitute_feature_t 481 { 482 inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_) 483 { 484 zero_context = zero_context_; 485 map->get_stage_lookups (0/*GSUB*/, 486 map->get_feature_stage (0/*GSUB*/, feature_tag), 487 &lookups, &count); 488 } 489 490 inline bool would_substitute (const hb_codepoint_t *glyphs, 491 unsigned int glyphs_count, 492 hb_face_t *face) const 493 { 494 for (unsigned int i = 0; i < count; i++) 495 if (hb_ot_layout_lookup_would_substitute_fast (face, lookups[i].index, glyphs, glyphs_count, zero_context)) 496 return true; 497 return false; 498 } 499 500 private: 501 const hb_ot_map_t::lookup_map_t *lookups; 502 unsigned int count; 503 bool zero_context; 504 }; 505 506 struct indic_shape_plan_t 507 { 508 ASSERT_POD (); 509 510 inline bool get_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const 511 { 512 hb_codepoint_t glyph = virama_glyph; 513 if (unlikely (virama_glyph == (hb_codepoint_t) -1)) 514 { 515 if (!config->virama || !font->get_glyph (config->virama, 0, &glyph)) 516 glyph = 0; 517 /* Technically speaking, the spec says we should apply 'locl' to virama too. 518 * Maybe one day... */ 519 520 /* Our get_glyph() function needs a font, so we can't get the virama glyph 521 * during shape planning... Instead, overwrite it here. It's safe. Don't worry! */ 522 (const_cast<indic_shape_plan_t *> (this))->virama_glyph = glyph; 523 } 524 525 *pglyph = glyph; 526 return glyph != 0; 527 } 528 529 const indic_config_t *config; 530 531 bool is_old_spec; 532 hb_codepoint_t virama_glyph; 533 534 would_substitute_feature_t rphf; 535 would_substitute_feature_t pref; 536 would_substitute_feature_t blwf; 537 would_substitute_feature_t pstf; 538 539 hb_mask_t mask_array[INDIC_NUM_FEATURES]; 540 }; 541 542 static void * 543 data_create_indic (const hb_ot_shape_plan_t *plan) 544 { 545 indic_shape_plan_t *indic_plan = (indic_shape_plan_t *) calloc (1, sizeof (indic_shape_plan_t)); 546 if (unlikely (!indic_plan)) 547 return NULL; 548 549 indic_plan->config = &indic_configs[0]; 550 for (unsigned int i = 1; i < ARRAY_LENGTH (indic_configs); i++) 551 if (plan->props.script == indic_configs[i].script) { 552 indic_plan->config = &indic_configs[i]; 553 break; 554 } 555 556 indic_plan->is_old_spec = indic_plan->config->has_old_spec && ((plan->map.chosen_script[0] & 0x000000FFu) != '2'); 557 indic_plan->virama_glyph = (hb_codepoint_t) -1; 558 559 /* Use zero-context would_substitute() matching for new-spec of the main 560 * Indic scripts, and scripts with one spec only, but not for old-specs. */ 561 bool zero_context = !indic_plan->is_old_spec; 562 indic_plan->rphf.init (&plan->map, HB_TAG('r','p','h','f'), zero_context); 563 indic_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), zero_context); 564 indic_plan->blwf.init (&plan->map, HB_TAG('b','l','w','f'), zero_context); 565 indic_plan->pstf.init (&plan->map, HB_TAG('p','s','t','f'), zero_context); 566 567 for (unsigned int i = 0; i < ARRAY_LENGTH (indic_plan->mask_array); i++) 568 indic_plan->mask_array[i] = (indic_features[i].flags & F_GLOBAL) ? 569 0 : plan->map.get_1_mask (indic_features[i].tag); 570 571 return indic_plan; 572 } 573 574 static void 575 data_destroy_indic (void *data) 576 { 577 free (data); 578 } 579 580 static indic_position_t 581 consonant_position_from_face (const indic_shape_plan_t *indic_plan, 582 const hb_codepoint_t consonant, 583 const hb_codepoint_t virama, 584 hb_face_t *face) 585 { 586 /* For old-spec, the order of glyphs is Consonant,Virama, 587 * whereas for new-spec, it's Virama,Consonant. However, 588 * some broken fonts (like Free Sans) simply copied lookups 589 * from old-spec to new-spec without modification. 590 * And oddly enough, Uniscribe seems to respect those lookups. 591 * Eg. in the sequence U+0924,U+094D,U+0930, Uniscribe finds 592 * base at 0. The font however, only has lookups matching 593 * 930,94D in 'blwf', not the expected 94D,930 (with new-spec 594 * table). As such, we simply match both sequences. Seems 595 * to work. */ 596 hb_codepoint_t glyphs[3] = {virama, consonant, virama}; 597 if (indic_plan->blwf.would_substitute (glyphs , 2, face) || 598 indic_plan->blwf.would_substitute (glyphs+1, 2, face)) 599 return POS_BELOW_C; 600 if (indic_plan->pstf.would_substitute (glyphs , 2, face) || 601 indic_plan->pstf.would_substitute (glyphs+1, 2, face)) 602 return POS_POST_C; 603 unsigned int pref_len = indic_plan->config->pref_len; 604 if ((pref_len == PREF_LEN_2 && 605 (indic_plan->pref.would_substitute (glyphs , 2, face) || 606 indic_plan->pref.would_substitute (glyphs+1, 2, face))) 607 || (pref_len == PREF_LEN_1 && 608 indic_plan->pref.would_substitute (glyphs+1, 1, face))) 609 return POS_POST_C; 610 return POS_BASE_C; 611 } 612 613 614 enum syllable_type_t { 615 consonant_syllable, 616 vowel_syllable, 617 standalone_cluster, 618 symbol_cluster, 619 broken_cluster, 620 non_indic_cluster, 621 }; 622 623 #include "hb-ot-shape-complex-indic-machine.hh" 624 625 626 static void 627 setup_masks_indic (const hb_ot_shape_plan_t *plan HB_UNUSED, 628 hb_buffer_t *buffer, 629 hb_font_t *font HB_UNUSED) 630 { 631 HB_BUFFER_ALLOCATE_VAR (buffer, indic_category); 632 HB_BUFFER_ALLOCATE_VAR (buffer, indic_position); 633 634 /* We cannot setup masks here. We save information about characters 635 * and setup masks later on in a pause-callback. */ 636 637 unsigned int count = buffer->len; 638 hb_glyph_info_t *info = buffer->info; 639 for (unsigned int i = 0; i < count; i++) 640 set_indic_properties (info[i]); 641 } 642 643 static void 644 setup_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED, 645 hb_font_t *font HB_UNUSED, 646 hb_buffer_t *buffer) 647 { 648 find_syllables (buffer); 649 } 650 651 static int 652 compare_indic_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb) 653 { 654 int a = pa->indic_position(); 655 int b = pb->indic_position(); 656 657 return a < b ? -1 : a == b ? 0 : +1; 658 } 659 660 661 662 static void 663 update_consonant_positions (const hb_ot_shape_plan_t *plan, 664 hb_font_t *font, 665 hb_buffer_t *buffer) 666 { 667 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; 668 669 if (indic_plan->config->base_pos != BASE_POS_LAST) 670 return; 671 672 hb_codepoint_t virama; 673 if (indic_plan->get_virama_glyph (font, &virama)) 674 { 675 hb_face_t *face = font->face; 676 unsigned int count = buffer->len; 677 hb_glyph_info_t *info = buffer->info; 678 for (unsigned int i = 0; i < count; i++) 679 if (info[i].indic_position() == POS_BASE_C) 680 { 681 hb_codepoint_t consonant = info[i].codepoint; 682 info[i].indic_position() = consonant_position_from_face (indic_plan, consonant, virama, face); 683 } 684 } 685 } 686 687 688 /* Rules from: 689 * https://www.microsoft.com/typography/otfntdev/devanot/shaping.aspx */ 690 691 static void 692 initial_reordering_consonant_syllable (const hb_ot_shape_plan_t *plan, 693 hb_face_t *face, 694 hb_buffer_t *buffer, 695 unsigned int start, unsigned int end) 696 { 697 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; 698 hb_glyph_info_t *info = buffer->info; 699 700 701 /* 1. Find base consonant: 702 * 703 * The shaping engine finds the base consonant of the syllable, using the 704 * following algorithm: starting from the end of the syllable, move backwards 705 * until a consonant is found that does not have a below-base or post-base 706 * form (post-base forms have to follow below-base forms), or that is not a 707 * pre-base reordering Ra, or arrive at the first consonant. The consonant 708 * stopped at will be the base. 709 * 710 * o If the syllable starts with Ra + Halant (in a script that has Reph) 711 * and has more than one consonant, Ra is excluded from candidates for 712 * base consonants. 713 */ 714 715 unsigned int base = end; 716 bool has_reph = false; 717 718 { 719 /* -> If the syllable starts with Ra + Halant (in a script that has Reph) 720 * and has more than one consonant, Ra is excluded from candidates for 721 * base consonants. */ 722 unsigned int limit = start; 723 if (indic_plan->config->reph_pos != REPH_POS_DONT_CARE && 724 indic_plan->mask_array[RPHF] && 725 start + 3 <= end && 726 ( 727 (indic_plan->config->reph_mode == REPH_MODE_IMPLICIT && !is_joiner (info[start + 2])) || 728 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT && info[start + 2].indic_category() == OT_ZWJ) 729 )) 730 { 731 /* See if it matches the 'rphf' feature. */ 732 hb_codepoint_t glyphs[3] = {info[start].codepoint, 733 info[start + 1].codepoint, 734 indic_plan->config->reph_mode == REPH_MODE_EXPLICIT ? 735 info[start + 2].codepoint : 0}; 736 if (indic_plan->rphf.would_substitute (glyphs, 2, face) || 737 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT && 738 indic_plan->rphf.would_substitute (glyphs, 3, face))) 739 { 740 limit += 2; 741 while (limit < end && is_joiner (info[limit])) 742 limit++; 743 base = start; 744 has_reph = true; 745 } 746 } else if (indic_plan->config->reph_mode == REPH_MODE_LOG_REPHA && info[start].indic_category() == OT_Repha) 747 { 748 limit += 1; 749 while (limit < end && is_joiner (info[limit])) 750 limit++; 751 base = start; 752 has_reph = true; 753 } 754 755 switch (indic_plan->config->base_pos) 756 { 757 default: 758 assert (false); 759 /* fallthrough */ 760 761 case BASE_POS_LAST: 762 { 763 /* -> starting from the end of the syllable, move backwards */ 764 unsigned int i = end; 765 bool seen_below = false; 766 do { 767 i--; 768 /* -> until a consonant is found */ 769 if (is_consonant (info[i])) 770 { 771 /* -> that does not have a below-base or post-base form 772 * (post-base forms have to follow below-base forms), */ 773 if (info[i].indic_position() != POS_BELOW_C && 774 (info[i].indic_position() != POS_POST_C || seen_below)) 775 { 776 base = i; 777 break; 778 } 779 if (info[i].indic_position() == POS_BELOW_C) 780 seen_below = true; 781 782 /* -> or that is not a pre-base reordering Ra, 783 * 784 * IMPLEMENTATION NOTES: 785 * 786 * Our pre-base reordering Ra's are marked POS_POST_C, so will be skipped 787 * by the logic above already. 788 */ 789 790 /* -> or arrive at the first consonant. The consonant stopped at will 791 * be the base. */ 792 base = i; 793 } 794 else 795 { 796 /* A ZWJ after a Halant stops the base search, and requests an explicit 797 * half form. 798 * A ZWJ before a Halant, requests a subjoined form instead, and hence 799 * search continues. This is particularly important for Bengali 800 * sequence Ra,H,Ya that should form Ya-Phalaa by subjoining Ya. */ 801 if (start < i && 802 info[i].indic_category() == OT_ZWJ && 803 info[i - 1].indic_category() == OT_H) 804 break; 805 } 806 } while (i > limit); 807 } 808 break; 809 810 case BASE_POS_LAST_SINHALA: 811 { 812 /* Sinhala base positioning is slightly different from main Indic, in that: 813 * 1. Its ZWJ behavior is different, 814 * 2. We don't need to look into the font for consonant positions. 815 */ 816 817 if (!has_reph) 818 base = limit; 819 820 /* Find the last base consonant that is not blocked by ZWJ. If there is 821 * a ZWJ right before a base consonant, that would request a subjoined form. */ 822 for (unsigned int i = limit; i < end; i++) 823 if (is_consonant (info[i])) 824 { 825 if (limit < i && info[i - 1].indic_category() == OT_ZWJ) 826 break; 827 else 828 base = i; 829 } 830 831 /* Mark all subsequent consonants as below. */ 832 for (unsigned int i = base + 1; i < end; i++) 833 if (is_consonant (info[i])) 834 info[i].indic_position() = POS_BELOW_C; 835 } 836 break; 837 838 case BASE_POS_FIRST: 839 { 840 /* The first consonant is always the base. */ 841 842 assert (indic_plan->config->reph_mode == REPH_MODE_VIS_REPHA); 843 assert (!has_reph); 844 845 base = start; 846 847 /* Mark all subsequent consonants as below. */ 848 for (unsigned int i = base + 1; i < end; i++) 849 if (is_consonant (info[i])) 850 info[i].indic_position() = POS_BELOW_C; 851 } 852 break; 853 } 854 855 /* -> If the syllable starts with Ra + Halant (in a script that has Reph) 856 * and has more than one consonant, Ra is excluded from candidates for 857 * base consonants. 858 * 859 * Only do this for unforced Reph. (ie. not for Ra,H,ZWJ. */ 860 if (has_reph && base == start && limit - base <= 2) { 861 /* Have no other consonant, so Reph is not formed and Ra becomes base. */ 862 has_reph = false; 863 } 864 } 865 866 867 /* 2. Decompose and reorder Matras: 868 * 869 * Each matra and any syllable modifier sign in the cluster are moved to the 870 * appropriate position relative to the consonant(s) in the cluster. The 871 * shaping engine decomposes two- or three-part matras into their constituent 872 * parts before any repositioning. Matra characters are classified by which 873 * consonant in a conjunct they have affinity for and are reordered to the 874 * following positions: 875 * 876 * o Before first half form in the syllable 877 * o After subjoined consonants 878 * o After post-form consonant 879 * o After main consonant (for above marks) 880 * 881 * IMPLEMENTATION NOTES: 882 * 883 * The normalize() routine has already decomposed matras for us, so we don't 884 * need to worry about that. 885 */ 886 887 888 /* 3. Reorder marks to canonical order: 889 * 890 * Adjacent nukta and halant or nukta and vedic sign are always repositioned 891 * if necessary, so that the nukta is first. 892 * 893 * IMPLEMENTATION NOTES: 894 * 895 * We don't need to do this: the normalize() routine already did this for us. 896 */ 897 898 899 /* Reorder characters */ 900 901 for (unsigned int i = start; i < base; i++) 902 info[i].indic_position() = MIN (POS_PRE_C, (indic_position_t) info[i].indic_position()); 903 904 if (base < end) 905 info[base].indic_position() = POS_BASE_C; 906 907 /* Mark final consonants. A final consonant is one appearing after a matra, 908 * like in Khmer. */ 909 for (unsigned int i = base + 1; i < end; i++) 910 if (info[i].indic_category() == OT_M) { 911 for (unsigned int j = i + 1; j < end; j++) 912 if (is_consonant (info[j])) { 913 info[j].indic_position() = POS_FINAL_C; 914 break; 915 } 916 break; 917 } 918 919 /* Handle beginning Ra */ 920 if (has_reph) 921 info[start].indic_position() = POS_RA_TO_BECOME_REPH; 922 923 /* For old-style Indic script tags, move the first post-base Halant after 924 * last consonant. 925 * 926 * Reports suggest that in some scripts Uniscribe does this only if there 927 * is *not* a Halant after last consonant already (eg. Kannada), while it 928 * does it unconditionally in other scripts (eg. Malayalam). We don't 929 * currently know about other scripts, so we single out Malayalam for now. 930 * 931 * Kannada test case: 932 * U+0C9A,U+0CCD,U+0C9A,U+0CCD 933 * With some versions of Lohit Kannada. 934 * https://bugs.freedesktop.org/show_bug.cgi?id=59118 935 * 936 * Malayalam test case: 937 * U+0D38,U+0D4D,U+0D31,U+0D4D,U+0D31,U+0D4D 938 * With lohit-ttf-20121122/Lohit-Malayalam.ttf 939 */ 940 if (indic_plan->is_old_spec) 941 { 942 bool disallow_double_halants = buffer->props.script != HB_SCRIPT_MALAYALAM; 943 for (unsigned int i = base + 1; i < end; i++) 944 if (info[i].indic_category() == OT_H) 945 { 946 unsigned int j; 947 for (j = end - 1; j > i; j--) 948 if (is_consonant (info[j]) || 949 (disallow_double_halants && info[j].indic_category() == OT_H)) 950 break; 951 if (info[j].indic_category() != OT_H && j > i) { 952 /* Move Halant to after last consonant. */ 953 hb_glyph_info_t t = info[i]; 954 memmove (&info[i], &info[i + 1], (j - i) * sizeof (info[0])); 955 info[j] = t; 956 } 957 break; 958 } 959 } 960 961 /* Attach misc marks to previous char to move with them. */ 962 { 963 indic_position_t last_pos = POS_START; 964 for (unsigned int i = start; i < end; i++) 965 { 966 if ((FLAG_SAFE (info[i].indic_category()) & (JOINER_FLAGS | FLAG (OT_N) | FLAG (OT_RS) | MEDIAL_FLAGS | HALANT_OR_COENG_FLAGS))) 967 { 968 info[i].indic_position() = last_pos; 969 if (unlikely (info[i].indic_category() == OT_H && 970 info[i].indic_position() == POS_PRE_M)) 971 { 972 /* 973 * Uniscribe doesn't move the Halant with Left Matra. 974 * TEST: U+092B,U+093F,U+094DE 975 * We follow. This is important for the Sinhala 976 * U+0DDA split matra since it decomposes to U+0DD9,U+0DCA 977 * where U+0DD9 is a left matra and U+0DCA is the virama. 978 * We don't want to move the virama with the left matra. 979 * TEST: U+0D9A,U+0DDA 980 */ 981 for (unsigned int j = i; j > start; j--) 982 if (info[j - 1].indic_position() != POS_PRE_M) { 983 info[i].indic_position() = info[j - 1].indic_position(); 984 break; 985 } 986 } 987 } else if (info[i].indic_position() != POS_SMVD) { 988 last_pos = (indic_position_t) info[i].indic_position(); 989 } 990 } 991 } 992 /* For post-base consonants let them own anything before them 993 * since the last consonant or matra. */ 994 { 995 unsigned int last = base; 996 for (unsigned int i = base + 1; i < end; i++) 997 if (is_consonant (info[i])) 998 { 999 for (unsigned int j = last + 1; j < i; j++) 1000 if (info[j].indic_position() < POS_SMVD) 1001 info[j].indic_position() = info[i].indic_position(); 1002 last = i; 1003 } else if (info[i].indic_category() == OT_M) 1004 last = i; 1005 } 1006 1007 1008 { 1009 /* Use syllable() for sort accounting temporarily. */ 1010 unsigned int syllable = info[start].syllable(); 1011 for (unsigned int i = start; i < end; i++) 1012 info[i].syllable() = i - start; 1013 1014 /* Sit tight, rock 'n roll! */ 1015 hb_stable_sort (info + start, end - start, compare_indic_order); 1016 /* Find base again */ 1017 base = end; 1018 for (unsigned int i = start; i < end; i++) 1019 if (info[i].indic_position() == POS_BASE_C) 1020 { 1021 base = i; 1022 break; 1023 } 1024 /* Things are out-of-control for post base positions, they may shuffle 1025 * around like crazy. In old-spec mode, we move halants around, so in 1026 * that case merge all clusters after base. Otherwise, check the sort 1027 * order and merge as needed. 1028 * For pre-base stuff, we handle cluster issues in final reordering. 1029 * 1030 * We could use buffer->sort() for this, if there was no special 1031 * reordering of pre-base stuff happening later... 1032 */ 1033 if (indic_plan->is_old_spec || end - base > 127) 1034 buffer->merge_clusters (base, end); 1035 else 1036 { 1037 /* Note! syllable() is a one-byte field. */ 1038 for (unsigned int i = base; i < end; i++) 1039 if (info[i].syllable() != 255) 1040 { 1041 unsigned int max = i; 1042 unsigned int j = start + info[i].syllable(); 1043 while (j != i) 1044 { 1045 max = MAX (max, j); 1046 unsigned int next = start + info[j].syllable(); 1047 info[j].syllable() = 255; /* So we don't process j later again. */ 1048 j = next; 1049 } 1050 if (i != max) 1051 buffer->merge_clusters (i, max + 1); 1052 } 1053 } 1054 1055 /* Put syllable back in. */ 1056 for (unsigned int i = start; i < end; i++) 1057 info[i].syllable() = syllable; 1058 } 1059 1060 /* Setup masks now */ 1061 1062 { 1063 hb_mask_t mask; 1064 1065 /* Reph */ 1066 for (unsigned int i = start; i < end && info[i].indic_position() == POS_RA_TO_BECOME_REPH; i++) 1067 info[i].mask |= indic_plan->mask_array[RPHF]; 1068 1069 /* Pre-base */ 1070 mask = indic_plan->mask_array[HALF]; 1071 if (!indic_plan->is_old_spec && 1072 indic_plan->config->blwf_mode == BLWF_MODE_PRE_AND_POST) 1073 mask |= indic_plan->mask_array[BLWF]; 1074 for (unsigned int i = start; i < base; i++) 1075 info[i].mask |= mask; 1076 /* Base */ 1077 mask = 0; 1078 if (base < end) 1079 info[base].mask |= mask; 1080 /* Post-base */ 1081 mask = indic_plan->mask_array[BLWF] | indic_plan->mask_array[ABVF] | indic_plan->mask_array[PSTF]; 1082 for (unsigned int i = base + 1; i < end; i++) 1083 info[i].mask |= mask; 1084 } 1085 1086 if (indic_plan->is_old_spec && 1087 buffer->props.script == HB_SCRIPT_DEVANAGARI) 1088 { 1089 /* Old-spec eye-lash Ra needs special handling. From the 1090 * spec: 1091 * 1092 * "The feature 'below-base form' is applied to consonants 1093 * having below-base forms and following the base consonant. 1094 * The exception is vattu, which may appear below half forms 1095 * as well as below the base glyph. The feature 'below-base 1096 * form' will be applied to all such occurrences of Ra as well." 1097 * 1098 * Test case: U+0924,U+094D,U+0930,U+094d,U+0915 1099 * with Sanskrit 2003 font. 1100 * 1101 * However, note that Ra,Halant,ZWJ is the correct way to 1102 * request eyelash form of Ra, so we wouldbn't inhibit it 1103 * in that sequence. 1104 * 1105 * Test case: U+0924,U+094D,U+0930,U+094d,U+200D,U+0915 1106 */ 1107 for (unsigned int i = start; i + 1 < base; i++) 1108 if (info[i ].indic_category() == OT_Ra && 1109 info[i+1].indic_category() == OT_H && 1110 (i + 2 == base || 1111 info[i+2].indic_category() != OT_ZWJ)) 1112 { 1113 info[i ].mask |= indic_plan->mask_array[BLWF]; 1114 info[i+1].mask |= indic_plan->mask_array[BLWF]; 1115 } 1116 } 1117 1118 unsigned int pref_len = indic_plan->config->pref_len; 1119 if (indic_plan->mask_array[PREF] && base + pref_len < end) 1120 { 1121 assert (1 <= pref_len && pref_len <= 2); 1122 /* Find a Halant,Ra sequence and mark it for pre-base reordering processing. */ 1123 for (unsigned int i = base + 1; i + pref_len - 1 < end; i++) { 1124 hb_codepoint_t glyphs[2]; 1125 for (unsigned int j = 0; j < pref_len; j++) 1126 glyphs[j] = info[i + j].codepoint; 1127 if (indic_plan->pref.would_substitute (glyphs, pref_len, face)) 1128 { 1129 for (unsigned int j = 0; j < pref_len; j++) 1130 info[i++].mask |= indic_plan->mask_array[PREF]; 1131 1132 /* Mark the subsequent stuff with 'cfar'. Used in Khmer. 1133 * Read the feature spec. 1134 * This allows distinguishing the following cases with MS Khmer fonts: 1135 * U+1784,U+17D2,U+179A,U+17D2,U+1782 1136 * U+1784,U+17D2,U+1782,U+17D2,U+179A 1137 */ 1138 if (indic_plan->mask_array[CFAR]) 1139 for (; i < end; i++) 1140 info[i].mask |= indic_plan->mask_array[CFAR]; 1141 1142 break; 1143 } 1144 } 1145 } 1146 1147 /* Apply ZWJ/ZWNJ effects */ 1148 for (unsigned int i = start + 1; i < end; i++) 1149 if (is_joiner (info[i])) { 1150 bool non_joiner = info[i].indic_category() == OT_ZWNJ; 1151 unsigned int j = i; 1152 1153 do { 1154 j--; 1155 1156 /* ZWJ/ZWNJ should disable CJCT. They do that by simply 1157 * being there, since we don't skip them for the CJCT 1158 * feature (ie. F_MANUAL_ZWJ) */ 1159 1160 /* A ZWNJ disables HALF. */ 1161 if (non_joiner) 1162 info[j].mask &= ~indic_plan->mask_array[HALF]; 1163 1164 } while (j > start && !is_consonant (info[j])); 1165 } 1166 } 1167 1168 static void 1169 initial_reordering_standalone_cluster (const hb_ot_shape_plan_t *plan, 1170 hb_face_t *face, 1171 hb_buffer_t *buffer, 1172 unsigned int start, unsigned int end) 1173 { 1174 /* We treat placeholder/dotted-circle as if they are consonants, so we 1175 * should just chain. Only if not in compatibility mode that is... */ 1176 1177 if (hb_options ().uniscribe_bug_compatible) 1178 { 1179 /* For dotted-circle, this is what Uniscribe does: 1180 * If dotted-circle is the last glyph, it just does nothing. 1181 * Ie. It doesn't form Reph. */ 1182 if (buffer->info[end - 1].indic_category() == OT_DOTTEDCIRCLE) 1183 return; 1184 } 1185 1186 initial_reordering_consonant_syllable (plan, face, buffer, start, end); 1187 } 1188 1189 static void 1190 initial_reordering_syllable (const hb_ot_shape_plan_t *plan, 1191 hb_face_t *face, 1192 hb_buffer_t *buffer, 1193 unsigned int start, unsigned int end) 1194 { 1195 syllable_type_t syllable_type = (syllable_type_t) (buffer->info[start].syllable() & 0x0F); 1196 switch (syllable_type) 1197 { 1198 case vowel_syllable: /* We made the vowels look like consonants. So let's call the consonant logic! */ 1199 case consonant_syllable: 1200 initial_reordering_consonant_syllable (plan, face, buffer, start, end); 1201 break; 1202 1203 case broken_cluster: /* We already inserted dotted-circles, so just call the standalone_cluster. */ 1204 case standalone_cluster: 1205 initial_reordering_standalone_cluster (plan, face, buffer, start, end); 1206 break; 1207 1208 case symbol_cluster: 1209 case non_indic_cluster: 1210 break; 1211 } 1212 } 1213 1214 static inline void 1215 insert_dotted_circles (const hb_ot_shape_plan_t *plan HB_UNUSED, 1216 hb_font_t *font, 1217 hb_buffer_t *buffer) 1218 { 1219 /* Note: This loop is extra overhead, but should not be measurable. */ 1220 bool has_broken_syllables = false; 1221 unsigned int count = buffer->len; 1222 hb_glyph_info_t *info = buffer->info; 1223 for (unsigned int i = 0; i < count; i++) 1224 if ((info[i].syllable() & 0x0F) == broken_cluster) 1225 { 1226 has_broken_syllables = true; 1227 break; 1228 } 1229 if (likely (!has_broken_syllables)) 1230 return; 1231 1232 1233 hb_codepoint_t dottedcircle_glyph; 1234 if (!font->get_glyph (0x25CCu, 0, &dottedcircle_glyph)) 1235 return; 1236 1237 hb_glyph_info_t dottedcircle = {0}; 1238 dottedcircle.codepoint = 0x25CCu; 1239 set_indic_properties (dottedcircle); 1240 dottedcircle.codepoint = dottedcircle_glyph; 1241 1242 buffer->clear_output (); 1243 1244 buffer->idx = 0; 1245 unsigned int last_syllable = 0; 1246 while (buffer->idx < buffer->len) 1247 { 1248 unsigned int syllable = buffer->cur().syllable(); 1249 syllable_type_t syllable_type = (syllable_type_t) (syllable & 0x0F); 1250 if (unlikely (last_syllable != syllable && syllable_type == broken_cluster)) 1251 { 1252 last_syllable = syllable; 1253 1254 hb_glyph_info_t info = dottedcircle; 1255 info.cluster = buffer->cur().cluster; 1256 info.mask = buffer->cur().mask; 1257 info.syllable() = buffer->cur().syllable(); 1258 /* TODO Set glyph_props? */ 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 foreach_syllable (buffer, start, end) 1284 initial_reordering_syllable (plan, font->face, buffer, start, end); 1285 } 1286 1287 static void 1288 final_reordering_syllable (const hb_ot_shape_plan_t *plan, 1289 hb_buffer_t *buffer, 1290 unsigned int start, unsigned int end) 1291 { 1292 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; 1293 hb_glyph_info_t *info = buffer->info; 1294 1295 1296 /* This function relies heavily on halant glyphs. Lots of ligation 1297 * and possibly multiplication substitutions happened prior to this 1298 * phase, and that might have messed up our properties. Recover 1299 * from a particular case of that where we're fairly sure that a 1300 * class of OT_H is desired but has been lost. */ 1301 if (indic_plan->virama_glyph) 1302 { 1303 unsigned int virama_glyph = indic_plan->virama_glyph; 1304 for (unsigned int i = start; i < end; i++) 1305 if (info[i].codepoint == virama_glyph && 1306 _hb_glyph_info_ligated (&info[i]) && 1307 _hb_glyph_info_multiplied (&info[i])) 1308 { 1309 /* This will make sure that this glyph passes is_halant_or_coeng() test. */ 1310 info[i].indic_category() = OT_H; 1311 _hb_glyph_info_clear_ligated_and_multiplied (&info[i]); 1312 } 1313 } 1314 1315 1316 /* 4. Final reordering: 1317 * 1318 * After the localized forms and basic shaping forms GSUB features have been 1319 * applied (see below), the shaping engine performs some final glyph 1320 * reordering before applying all the remaining font features to the entire 1321 * cluster. 1322 */ 1323 1324 bool try_pref = !!indic_plan->mask_array[PREF]; 1325 1326 /* Find base again */ 1327 unsigned int base; 1328 for (base = start; base < end; base++) 1329 if (info[base].indic_position() >= POS_BASE_C) 1330 { 1331 if (try_pref && base + 1 < end && indic_plan->config->pref_len == 2) 1332 { 1333 for (unsigned int i = base + 1; i < end; i++) 1334 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0) 1335 { 1336 if (!(_hb_glyph_info_substituted (&info[i]) && 1337 _hb_glyph_info_ligated_and_didnt_multiply (&info[i]))) 1338 { 1339 /* Ok, this was a 'pref' candidate but didn't form any. 1340 * Base is around here... */ 1341 base = i; 1342 while (base < end && is_halant_or_coeng (info[base])) 1343 base++; 1344 info[base].indic_position() = POS_BASE_C; 1345 1346 try_pref = false; 1347 } 1348 break; 1349 } 1350 } 1351 1352 if (start < base && info[base].indic_position() > POS_BASE_C) 1353 base--; 1354 break; 1355 } 1356 if (base == end && start < base && 1357 is_one_of (info[base - 1], FLAG (OT_ZWJ))) 1358 base--; 1359 if (base < end) 1360 while (start < base && 1361 is_one_of (info[base], (FLAG (OT_N) | HALANT_OR_COENG_FLAGS))) 1362 base--; 1363 1364 1365 /* o Reorder matras: 1366 * 1367 * If a pre-base matra character had been reordered before applying basic 1368 * features, the glyph can be moved closer to the main consonant based on 1369 * whether half-forms had been formed. Actual position for the matra is 1370 * defined as “after last standalone halant glyph, after initial matra 1371 * position and before the main consonant”. If ZWJ or ZWNJ follow this 1372 * halant, position is moved after it. 1373 */ 1374 1375 if (start + 1 < end && start < base) /* Otherwise there can't be any pre-base matra characters. */ 1376 { 1377 /* If we lost track of base, alas, position before last thingy. */ 1378 unsigned int new_pos = base == end ? base - 2 : base - 1; 1379 1380 /* Malayalam / Tamil do not have "half" forms or explicit virama forms. 1381 * The glyphs formed by 'half' are Chillus or ligated explicit viramas. 1382 * We want to position matra after them. 1383 */ 1384 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL) 1385 { 1386 while (new_pos > start && 1387 !(is_one_of (info[new_pos], (FLAG (OT_M) | HALANT_OR_COENG_FLAGS)))) 1388 new_pos--; 1389 1390 /* If we found no Halant we are done. 1391 * Otherwise only proceed if the Halant does 1392 * not belong to the Matra itself! */ 1393 if (is_halant_or_coeng (info[new_pos]) && 1394 info[new_pos].indic_position() != POS_PRE_M) 1395 { 1396 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */ 1397 if (new_pos + 1 < end && is_joiner (info[new_pos + 1])) 1398 new_pos++; 1399 } 1400 else 1401 new_pos = start; /* No move. */ 1402 } 1403 1404 if (start < new_pos && info[new_pos].indic_position () != POS_PRE_M) 1405 { 1406 /* Now go see if there's actually any matras... */ 1407 for (unsigned int i = new_pos; i > start; i--) 1408 if (info[i - 1].indic_position () == POS_PRE_M) 1409 { 1410 unsigned int old_pos = i - 1; 1411 if (old_pos < base && base <= new_pos) /* Shouldn't actually happen. */ 1412 base--; 1413 1414 hb_glyph_info_t tmp = info[old_pos]; 1415 memmove (&info[old_pos], &info[old_pos + 1], (new_pos - old_pos) * sizeof (info[0])); 1416 info[new_pos] = tmp; 1417 1418 /* Note: this merge_clusters() is intentionally *after* the reordering. 1419 * Indic matra reordering is special and tricky... */ 1420 buffer->merge_clusters (new_pos, MIN (end, base + 1)); 1421 1422 new_pos--; 1423 } 1424 } else { 1425 for (unsigned int i = start; i < base; i++) 1426 if (info[i].indic_position () == POS_PRE_M) { 1427 buffer->merge_clusters (i, MIN (end, base + 1)); 1428 break; 1429 } 1430 } 1431 } 1432 1433 1434 /* o Reorder reph: 1435 * 1436 * Reph’s original position is always at the beginning of the syllable, 1437 * (i.e. it is not reordered at the character reordering stage). However, 1438 * it will be reordered according to the basic-forms shaping results. 1439 * Possible positions for reph, depending on the script, are; after main, 1440 * before post-base consonant forms, and after post-base consonant forms. 1441 */ 1442 1443 /* Two cases: 1444 * 1445 * - If repha is encoded as a sequence of characters (Ra,H or Ra,H,ZWJ), then 1446 * we should only move it if the sequence ligated to the repha form. 1447 * 1448 * - If repha is encoded separately and in the logical position, we should only 1449 * move it if it did NOT ligate. If it ligated, it's probably the font trying 1450 * to make it work without the reordering. 1451 */ 1452 if (start + 1 < end && 1453 info[start].indic_position() == POS_RA_TO_BECOME_REPH && 1454 ((info[start].indic_category() == OT_Repha) ^ 1455 _hb_glyph_info_ligated_and_didnt_multiply (&info[start]))) 1456 { 1457 unsigned int new_reph_pos; 1458 reph_position_t reph_pos = indic_plan->config->reph_pos; 1459 1460 assert (reph_pos != REPH_POS_DONT_CARE); 1461 1462 /* 1. If reph should be positioned after post-base consonant forms, 1463 * proceed to step 5. 1464 */ 1465 if (reph_pos == REPH_POS_AFTER_POST) 1466 { 1467 goto reph_step_5; 1468 } 1469 1470 /* 2. If the reph repositioning class is not after post-base: target 1471 * position is after the first explicit halant glyph between the 1472 * first post-reph consonant and last main consonant. If ZWJ or ZWNJ 1473 * are following this halant, position is moved after it. If such 1474 * position is found, this is the target position. Otherwise, 1475 * proceed to the next step. 1476 * 1477 * Note: in old-implementation fonts, where classifications were 1478 * fixed in shaping engine, there was no case where reph position 1479 * will be found on this step. 1480 */ 1481 { 1482 new_reph_pos = start + 1; 1483 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos])) 1484 new_reph_pos++; 1485 1486 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos])) 1487 { 1488 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */ 1489 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1])) 1490 new_reph_pos++; 1491 goto reph_move; 1492 } 1493 } 1494 1495 /* 3. If reph should be repositioned after the main consonant: find the 1496 * first consonant not ligated with main, or find the first 1497 * consonant that is not a potential pre-base reordering Ra. 1498 */ 1499 if (reph_pos == REPH_POS_AFTER_MAIN) 1500 { 1501 new_reph_pos = base; 1502 while (new_reph_pos + 1 < end && info[new_reph_pos + 1].indic_position() <= POS_AFTER_MAIN) 1503 new_reph_pos++; 1504 if (new_reph_pos < end) 1505 goto reph_move; 1506 } 1507 1508 /* 4. If reph should be positioned before post-base consonant, find 1509 * first post-base classified consonant not ligated with main. If no 1510 * consonant is found, the target position should be before the 1511 * first matra, syllable modifier sign or vedic sign. 1512 */ 1513 /* This is our take on what step 4 is trying to say (and failing, BADLY). */ 1514 if (reph_pos == REPH_POS_AFTER_SUB) 1515 { 1516 new_reph_pos = base; 1517 while (new_reph_pos < end && 1518 !( FLAG_SAFE (info[new_reph_pos + 1].indic_position()) & (FLAG (POS_POST_C) | FLAG (POS_AFTER_POST) | FLAG (POS_SMVD)))) 1519 new_reph_pos++; 1520 if (new_reph_pos < end) 1521 goto reph_move; 1522 } 1523 1524 /* 5. If no consonant is found in steps 3 or 4, move reph to a position 1525 * immediately before the first post-base matra, syllable modifier 1526 * sign or vedic sign that has a reordering class after the intended 1527 * reph position. For example, if the reordering position for reph 1528 * is post-main, it will skip above-base matras that also have a 1529 * post-main position. 1530 */ 1531 reph_step_5: 1532 { 1533 /* Copied from step 2. */ 1534 new_reph_pos = start + 1; 1535 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos])) 1536 new_reph_pos++; 1537 1538 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos])) 1539 { 1540 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */ 1541 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1])) 1542 new_reph_pos++; 1543 goto reph_move; 1544 } 1545 } 1546 1547 /* 6. Otherwise, reorder reph to the end of the syllable. 1548 */ 1549 { 1550 new_reph_pos = end - 1; 1551 while (new_reph_pos > start && info[new_reph_pos].indic_position() == POS_SMVD) 1552 new_reph_pos--; 1553 1554 /* 1555 * If the Reph is to be ending up after a Matra,Halant sequence, 1556 * position it before that Halant so it can interact with the Matra. 1557 * However, if it's a plain Consonant,Halant we shouldn't do that. 1558 * Uniscribe doesn't do this. 1559 * TEST: U+0930,U+094D,U+0915,U+094B,U+094D 1560 */ 1561 if (!hb_options ().uniscribe_bug_compatible && 1562 unlikely (is_halant_or_coeng (info[new_reph_pos]))) { 1563 for (unsigned int i = base + 1; i < new_reph_pos; i++) 1564 if (info[i].indic_category() == OT_M) { 1565 /* Ok, got it. */ 1566 new_reph_pos--; 1567 } 1568 } 1569 goto reph_move; 1570 } 1571 1572 reph_move: 1573 { 1574 /* Move */ 1575 buffer->merge_clusters (start, new_reph_pos + 1); 1576 hb_glyph_info_t reph = info[start]; 1577 memmove (&info[start], &info[start + 1], (new_reph_pos - start) * sizeof (info[0])); 1578 info[new_reph_pos] = reph; 1579 1580 if (start < base && base <= new_reph_pos) 1581 base--; 1582 } 1583 } 1584 1585 1586 /* o Reorder pre-base reordering consonants: 1587 * 1588 * If a pre-base reordering consonant is found, reorder it according to 1589 * the following rules: 1590 */ 1591 1592 if (try_pref && base + 1 < end) /* Otherwise there can't be any pre-base reordering Ra. */ 1593 { 1594 unsigned int pref_len = indic_plan->config->pref_len; 1595 for (unsigned int i = base + 1; i < end; i++) 1596 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0) 1597 { 1598 /* 1. Only reorder a glyph produced by substitution during application 1599 * of the <pref> feature. (Note that a font may shape a Ra consonant with 1600 * the feature generally but block it in certain contexts.) 1601 */ 1602 /* Note: We just check that something got substituted. We don't check that 1603 * the <pref> feature actually did it... 1604 * 1605 * If pref len is longer than one, then only reorder if it ligated. If 1606 * pref len is one, only reorder if it didn't ligate with other things. */ 1607 if (_hb_glyph_info_substituted (&info[i]) && 1608 ((pref_len == 1) ^ _hb_glyph_info_ligated_and_didnt_multiply (&info[i]))) 1609 { 1610 /* 1611 * 2. Try to find a target position the same way as for pre-base matra. 1612 * If it is found, reorder pre-base consonant glyph. 1613 * 1614 * 3. If position is not found, reorder immediately before main 1615 * consonant. 1616 */ 1617 1618 unsigned int new_pos = base; 1619 /* Malayalam / Tamil do not have "half" forms or explicit virama forms. 1620 * The glyphs formed by 'half' are Chillus or ligated explicit viramas. 1621 * We want to position matra after them. 1622 */ 1623 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL) 1624 { 1625 while (new_pos > start && 1626 !(is_one_of (info[new_pos - 1], FLAG(OT_M) | HALANT_OR_COENG_FLAGS))) 1627 new_pos--; 1628 1629 /* In Khmer coeng model, a H,Ra can go *after* matras. If it goes after a 1630 * split matra, it should be reordered to *before* the left part of such matra. */ 1631 if (new_pos > start && info[new_pos - 1].indic_category() == OT_M) 1632 { 1633 unsigned int old_pos = i; 1634 for (unsigned int i = base + 1; i < old_pos; i++) 1635 if (info[i].indic_category() == OT_M) 1636 { 1637 new_pos--; 1638 break; 1639 } 1640 } 1641 } 1642 1643 if (new_pos > start && is_halant_or_coeng (info[new_pos - 1])) 1644 { 1645 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */ 1646 if (new_pos < end && is_joiner (info[new_pos])) 1647 new_pos++; 1648 } 1649 1650 { 1651 unsigned int old_pos = i; 1652 1653 buffer->merge_clusters (new_pos, old_pos + 1); 1654 hb_glyph_info_t tmp = info[old_pos]; 1655 memmove (&info[new_pos + 1], &info[new_pos], (old_pos - new_pos) * sizeof (info[0])); 1656 info[new_pos] = tmp; 1657 1658 if (new_pos <= base && base < old_pos) 1659 base++; 1660 } 1661 } 1662 1663 break; 1664 } 1665 } 1666 1667 1668 /* Apply 'init' to the Left Matra if it's a word start. */ 1669 if (info[start].indic_position () == POS_PRE_M && 1670 (!start || 1671 !(FLAG_SAFE (_hb_glyph_info_get_general_category (&info[start - 1])) & 1672 FLAG_RANGE (HB_UNICODE_GENERAL_CATEGORY_FORMAT, HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK)))) 1673 info[start].mask |= indic_plan->mask_array[INIT]; 1674 1675 1676 /* 1677 * Finish off the clusters and go home! 1678 */ 1679 if (hb_options ().uniscribe_bug_compatible) 1680 { 1681 switch ((hb_tag_t) plan->props.script) 1682 { 1683 case HB_SCRIPT_TAMIL: 1684 case HB_SCRIPT_SINHALA: 1685 break; 1686 1687 default: 1688 /* Uniscribe merges the entire cluster... Except for Tamil & Sinhala. 1689 * This means, half forms are submerged into the main consonants cluster. 1690 * This is unnecessary, and makes cursor positioning harder, but that's what 1691 * Uniscribe does. */ 1692 buffer->merge_clusters (start, end); 1693 break; 1694 } 1695 } 1696 } 1697 1698 1699 static void 1700 final_reordering (const hb_ot_shape_plan_t *plan, 1701 hb_font_t *font HB_UNUSED, 1702 hb_buffer_t *buffer) 1703 { 1704 unsigned int count = buffer->len; 1705 if (unlikely (!count)) return; 1706 1707 foreach_syllable (buffer, start, end) 1708 final_reordering_syllable (plan, buffer, start, end); 1709 1710 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_category); 1711 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_position); 1712 } 1713 1714 1715 static void 1716 clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED, 1717 hb_font_t *font HB_UNUSED, 1718 hb_buffer_t *buffer) 1719 { 1720 hb_glyph_info_t *info = buffer->info; 1721 unsigned int count = buffer->len; 1722 for (unsigned int i = 0; i < count; i++) 1723 info[i].syllable() = 0; 1724 } 1725 1726 1727 static bool 1728 decompose_indic (const hb_ot_shape_normalize_context_t *c, 1729 hb_codepoint_t ab, 1730 hb_codepoint_t *a, 1731 hb_codepoint_t *b) 1732 { 1733 switch (ab) 1734 { 1735 /* Don't decompose these. */ 1736 case 0x0931u : return false; 1737 case 0x0B94u : return false; 1738 1739 1740 /* 1741 * Decompose split matras that don't have Unicode decompositions. 1742 */ 1743 1744 case 0x0F77u : *a = 0x0FB2u; *b= 0x0F81u; return true; 1745 case 0x0F79u : *a = 0x0FB3u; *b= 0x0F81u; return true; 1746 case 0x17BEu : *a = 0x17C1u; *b= 0x17BEu; return true; 1747 case 0x17BFu : *a = 0x17C1u; *b= 0x17BFu; return true; 1748 case 0x17C0u : *a = 0x17C1u; *b= 0x17C0u; return true; 1749 case 0x17C4u : *a = 0x17C1u; *b= 0x17C4u; return true; 1750 case 0x17C5u : *a = 0x17C1u; *b= 0x17C5u; return true; 1751 case 0x1925u : *a = 0x1920u; *b= 0x1923u; return true; 1752 case 0x1926u : *a = 0x1920u; *b= 0x1924u; return true; 1753 case 0x1B3Cu : *a = 0x1B42u; *b= 0x1B3Cu; return true; 1754 case 0x1112Eu : *a = 0x11127u; *b= 0x11131u; return true; 1755 case 0x1112Fu : *a = 0x11127u; *b= 0x11132u; return true; 1756 #if 0 1757 /* This one has no decomposition in Unicode, but needs no decomposition either. */ 1758 /* case 0x0AC9u : return false; */ 1759 case 0x0B57u : *a = no decomp, -> RIGHT; return true; 1760 case 0x1C29u : *a = no decomp, -> LEFT; return true; 1761 case 0xA9C0u : *a = no decomp, -> RIGHT; return true; 1762 case 0x111BuF : *a = no decomp, -> ABOVE; return true; 1763 #endif 1764 } 1765 1766 if ((ab == 0x0DDAu || hb_in_range (ab, 0x0DDCu, 0x0DDEu))) 1767 { 1768 /* 1769 * Sinhala split matras... Let the fun begin. 1770 * 1771 * These four characters have Unicode decompositions. However, Uniscribe 1772 * decomposes them "Khmer-style", that is, it uses the character itself to 1773 * get the second half. The first half of all four decompositions is always 1774 * U+0DD9. 1775 * 1776 * Now, there are buggy fonts, namely, the widely used lklug.ttf, that are 1777 * broken with Uniscribe. But we need to support them. As such, we only 1778 * do the Uniscribe-style decomposition if the character is transformed into 1779 * its "sec.half" form by the 'pstf' feature. Otherwise, we fall back to 1780 * Unicode decomposition. 1781 * 1782 * Note that we can't unconditionally use Unicode decomposition. That would 1783 * break some other fonts, that are designed to work with Uniscribe, and 1784 * don't have positioning features for the Unicode-style decomposition. 1785 * 1786 * Argh... 1787 * 1788 * The Uniscribe behavior is now documented in the newly published Sinhala 1789 * spec in 2012: 1790 * 1791 * http://www.microsoft.com/typography/OpenTypeDev/sinhala/intro.htm#shaping 1792 */ 1793 1794 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) c->plan->data; 1795 1796 hb_codepoint_t glyph; 1797 1798 if (hb_options ().uniscribe_bug_compatible || 1799 (c->font->get_glyph (ab, 0, &glyph) && 1800 indic_plan->pstf.would_substitute (&glyph, 1, c->font->face))) 1801 { 1802 /* Ok, safe to use Uniscribe-style decomposition. */ 1803 *a = 0x0DD9u; 1804 *b = ab; 1805 return true; 1806 } 1807 } 1808 1809 return c->unicode->decompose (ab, a, b); 1810 } 1811 1812 static bool 1813 compose_indic (const hb_ot_shape_normalize_context_t *c, 1814 hb_codepoint_t a, 1815 hb_codepoint_t b, 1816 hb_codepoint_t *ab) 1817 { 1818 /* Avoid recomposing split matras. */ 1819 if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a))) 1820 return false; 1821 1822 /* Composition-exclusion exceptions that we want to recompose. */ 1823 if (a == 0x09AFu && b == 0x09BCu) { *ab = 0x09DFu; return true; } 1824 1825 return c->unicode->compose (a, b, ab); 1826 } 1827 1828 1829 const hb_ot_complex_shaper_t _hb_ot_complex_shaper_indic = 1830 { 1831 "indic", 1832 collect_features_indic, 1833 override_features_indic, 1834 data_create_indic, 1835 data_destroy_indic, 1836 NULL, /* preprocess_text */ 1837 HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT, 1838 decompose_indic, 1839 compose_indic, 1840 setup_masks_indic, 1841 HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE, 1842 false, /* fallback_position */ 1843 };