hb-ot-shaper-arabic.cc (24650B)
1 /* 2 * Copyright © 2010,2012 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.hh" 28 29 #ifndef HB_NO_OT_SHAPE 30 31 #include "hb-ot-shaper-arabic.hh" 32 #include "hb-ot-shape.hh" 33 34 35 /* buffer var allocations */ 36 #define arabic_shaping_action() ot_shaper_var_u8_auxiliary() /* arabic shaping action */ 37 38 #define HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH HB_BUFFER_SCRATCH_FLAG_SHAPER0 39 40 /* See: 41 * https://github.com/harfbuzz/harfbuzz/commit/6e6f82b6f3dde0fc6c3c7d991d9ec6cfff57823d#commitcomment-14248516 */ 42 #define HB_ARABIC_GENERAL_CATEGORY_IS_WORD(gen_cat) \ 43 (FLAG_UNSAFE (gen_cat) & \ 44 (FLAG (HB_UNICODE_GENERAL_CATEGORY_UNASSIGNED) | \ 45 FLAG (HB_UNICODE_GENERAL_CATEGORY_PRIVATE_USE) | \ 46 /*FLAG (HB_UNICODE_GENERAL_CATEGORY_LOWERCASE_LETTER) |*/ \ 47 FLAG (HB_UNICODE_GENERAL_CATEGORY_MODIFIER_LETTER) | \ 48 FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_LETTER) | \ 49 /*FLAG (HB_UNICODE_GENERAL_CATEGORY_TITLECASE_LETTER) |*/ \ 50 /*FLAG (HB_UNICODE_GENERAL_CATEGORY_UPPERCASE_LETTER) |*/ \ 51 FLAG (HB_UNICODE_GENERAL_CATEGORY_SPACING_MARK) | \ 52 FLAG (HB_UNICODE_GENERAL_CATEGORY_ENCLOSING_MARK) | \ 53 FLAG (HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK) | \ 54 FLAG (HB_UNICODE_GENERAL_CATEGORY_DECIMAL_NUMBER) | \ 55 FLAG (HB_UNICODE_GENERAL_CATEGORY_LETTER_NUMBER) | \ 56 FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_NUMBER) | \ 57 FLAG (HB_UNICODE_GENERAL_CATEGORY_CURRENCY_SYMBOL) | \ 58 FLAG (HB_UNICODE_GENERAL_CATEGORY_MODIFIER_SYMBOL) | \ 59 FLAG (HB_UNICODE_GENERAL_CATEGORY_MATH_SYMBOL) | \ 60 FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_SYMBOL))) 61 62 63 /* 64 * Joining types: 65 */ 66 67 /* 68 * Bits used in the joining tables 69 */ 70 enum hb_arabic_joining_type_t { 71 JOINING_TYPE_U = 0, 72 JOINING_TYPE_L = 1, 73 JOINING_TYPE_R = 2, 74 JOINING_TYPE_D = 3, 75 JOINING_TYPE_C = JOINING_TYPE_D, 76 JOINING_GROUP_ALAPH = 4, 77 JOINING_GROUP_DALATH_RISH = 5, 78 NUM_STATE_MACHINE_COLS = 6, 79 80 JOINING_TYPE_T = 7, 81 JOINING_TYPE_X = 8 /* means: use general-category to choose between U or T. */ 82 }; 83 84 #include "hb-ot-shaper-arabic-table.hh" 85 86 static unsigned int get_joining_type (hb_codepoint_t u, hb_unicode_general_category_t gen_cat) 87 { 88 unsigned int j_type = joining_type(u); 89 if (likely (j_type != JOINING_TYPE_X)) 90 return j_type; 91 92 return (FLAG_UNSAFE(gen_cat) & 93 (FLAG(HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK) | 94 FLAG(HB_UNICODE_GENERAL_CATEGORY_ENCLOSING_MARK) | 95 FLAG(HB_UNICODE_GENERAL_CATEGORY_FORMAT)) 96 ) ? JOINING_TYPE_T : JOINING_TYPE_U; 97 } 98 99 #define FEATURE_IS_SYRIAC(tag) hb_in_range<unsigned char> ((unsigned char) (tag), '2', '3') 100 101 static const hb_tag_t arabic_features[] = 102 { 103 HB_TAG('i','s','o','l'), 104 HB_TAG('f','i','n','a'), 105 HB_TAG('f','i','n','2'), 106 HB_TAG('f','i','n','3'), 107 HB_TAG('m','e','d','i'), 108 HB_TAG('m','e','d','2'), 109 HB_TAG('i','n','i','t'), 110 HB_TAG_NONE 111 }; 112 113 114 /* Same order as the feature array */ 115 enum arabic_action_t { 116 ISOL, 117 FINA, 118 FIN2, 119 FIN3, 120 MEDI, 121 MED2, 122 INIT, 123 124 NONE, 125 126 ARABIC_NUM_FEATURES = NONE, 127 128 /* We abuse the same byte for other things... */ 129 STCH_FIXED, 130 STCH_REPEATING, 131 }; 132 133 static const struct arabic_state_table_entry { 134 uint8_t prev_action; 135 uint8_t curr_action; 136 uint16_t next_state; 137 } arabic_state_table[][NUM_STATE_MACHINE_COLS] = 138 { 139 /* jt_U, jt_L, jt_R, jt_D, jg_ALAPH, jg_DALATH_RISH */ 140 141 /* State 0: prev was U, not willing to join. */ 142 { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,6}, }, 143 144 /* State 1: prev was R or ISOL/ALAPH, not willing to join. */ 145 { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,FIN2,5}, {NONE,ISOL,6}, }, 146 147 /* State 2: prev was D/L in ISOL form, willing to join. */ 148 { {NONE,NONE,0}, {NONE,ISOL,2}, {INIT,FINA,1}, {INIT,FINA,3}, {INIT,FINA,4}, {INIT,FINA,6}, }, 149 150 /* State 3: prev was D in FINA form, willing to join. */ 151 { {NONE,NONE,0}, {NONE,ISOL,2}, {MEDI,FINA,1}, {MEDI,FINA,3}, {MEDI,FINA,4}, {MEDI,FINA,6}, }, 152 153 /* State 4: prev was FINA ALAPH, not willing to join. */ 154 { {NONE,NONE,0}, {NONE,ISOL,2}, {MED2,ISOL,1}, {MED2,ISOL,2}, {MED2,FIN2,5}, {MED2,ISOL,6}, }, 155 156 /* State 5: prev was FIN2/FIN3 ALAPH, not willing to join. */ 157 { {NONE,NONE,0}, {NONE,ISOL,2}, {ISOL,ISOL,1}, {ISOL,ISOL,2}, {ISOL,FIN2,5}, {ISOL,ISOL,6}, }, 158 159 /* State 6: prev was DALATH/RISH, not willing to join. */ 160 { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,FIN3,5}, {NONE,ISOL,6}, } 161 }; 162 163 164 static bool 165 arabic_fallback_shape (const hb_ot_shape_plan_t *plan, 166 hb_font_t *font, 167 hb_buffer_t *buffer); 168 169 static bool 170 record_stch (const hb_ot_shape_plan_t *plan, 171 hb_font_t *font, 172 hb_buffer_t *buffer); 173 174 static bool 175 deallocate_buffer_var (const hb_ot_shape_plan_t *plan, 176 hb_font_t *font, 177 hb_buffer_t *buffer) 178 { 179 HB_BUFFER_DEALLOCATE_VAR (buffer, arabic_shaping_action); 180 return false; 181 } 182 183 static void 184 collect_features_arabic (hb_ot_shape_planner_t *plan) 185 { 186 hb_ot_map_builder_t *map = &plan->map; 187 188 /* We apply features according to the Arabic spec, with pauses 189 * in between most. 190 * 191 * The pause between init/medi/... and rlig is required. See eg: 192 * https://bugzilla.mozilla.org/show_bug.cgi?id=644184 193 * 194 * The pauses between init/medi/... themselves are not necessarily 195 * needed as only one of those features is applied to any character. 196 * The only difference it makes is when fonts have contextual 197 * substitutions. We now follow the order of the spec, which makes 198 * for better experience if that's what Uniscribe is doing. 199 * 200 * At least for Arabic, looks like Uniscribe has a pause between 201 * rlig and calt. Otherwise the IranNastaliq's ALLAH ligature won't 202 * work. However, testing shows that rlig and calt are applied 203 * together for Mongolian in Uniscribe. As such, we only add a 204 * pause for Arabic, not other scripts. 205 */ 206 207 208 map->enable_feature (HB_TAG('s','t','c','h')); 209 map->add_gsub_pause (record_stch); 210 211 map->enable_feature (HB_TAG('c','c','m','p'), F_MANUAL_ZWJ); 212 map->enable_feature (HB_TAG('l','o','c','l'), F_MANUAL_ZWJ); 213 214 map->add_gsub_pause (nullptr); 215 216 for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++) 217 { 218 bool has_fallback = plan->props.script == HB_SCRIPT_ARABIC && !FEATURE_IS_SYRIAC (arabic_features[i]); 219 map->add_feature (arabic_features[i], F_MANUAL_ZWJ | (has_fallback ? F_HAS_FALLBACK : F_NONE)); 220 map->add_gsub_pause (nullptr); 221 } 222 map->add_gsub_pause (deallocate_buffer_var); 223 224 /* Normally, Unicode says a ZWNJ means "don't ligate". In Arabic script 225 * however, it says a ZWJ should also mean "don't ligate". So we run 226 * the main ligating features as MANUAL_ZWJ. */ 227 228 map->enable_feature (HB_TAG('r','l','i','g'), F_MANUAL_ZWJ | F_HAS_FALLBACK); 229 230 if (plan->props.script == HB_SCRIPT_ARABIC) 231 map->add_gsub_pause (arabic_fallback_shape); 232 233 map->enable_feature (HB_TAG('c','a','l','t'), F_MANUAL_ZWJ); 234 /* https://github.com/harfbuzz/harfbuzz/issues/1573 */ 235 if (!map->has_feature (HB_TAG('r','c','l','t'))) 236 map->add_gsub_pause (nullptr); 237 238 map->enable_feature (HB_TAG('l','i','g','a'), F_MANUAL_ZWJ); 239 map->enable_feature (HB_TAG('c','l','i','g'), F_MANUAL_ZWJ); 240 241 /* The spec includes 'cswh'. Earlier versions of Windows 242 * used to enable this by default, but testing suggests 243 * that Windows 8 and later do not enable it by default, 244 * and spec now says 'Off by default'. 245 * We disabled this in ae23c24c32. 246 * Note that IranNastaliq uses this feature extensively 247 * to fixup broken glyph sequences. Oh well... 248 * Test case: U+0643,U+0640,U+0631. */ 249 //map->enable_feature (HB_TAG('c','s','w','h'), F_MANUAL_ZWJ); 250 map->enable_feature (HB_TAG('m','s','e','t'), F_MANUAL_ZWJ); 251 } 252 253 #include "hb-ot-shaper-arabic-fallback.hh" 254 255 struct arabic_shape_plan_t 256 { 257 /* The "+ 1" in the next array is to accommodate for the "NONE" command, 258 * which is not an OpenType feature, but this simplifies the code by not 259 * having to do a "if (... < NONE) ..." and just rely on the fact that 260 * mask_array[NONE] == 0. */ 261 hb_mask_t mask_array[ARABIC_NUM_FEATURES + 1]; 262 263 mutable hb_atomic_t<arabic_fallback_plan_t *> fallback_plan; 264 265 unsigned int do_fallback : 1; 266 unsigned int has_stch : 1; 267 }; 268 269 void * 270 data_create_arabic (const hb_ot_shape_plan_t *plan) 271 { 272 arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) hb_calloc (1, sizeof (arabic_shape_plan_t)); 273 if (unlikely (!arabic_plan)) 274 return nullptr; 275 276 arabic_plan->do_fallback = plan->props.script == HB_SCRIPT_ARABIC; 277 arabic_plan->has_stch = !!plan->map.get_1_mask (HB_TAG ('s','t','c','h')); 278 for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++) { 279 arabic_plan->mask_array[i] = plan->map.get_1_mask (arabic_features[i]); 280 arabic_plan->do_fallback = arabic_plan->do_fallback && 281 (FEATURE_IS_SYRIAC (arabic_features[i]) || 282 plan->map.needs_fallback (arabic_features[i])); 283 } 284 285 return arabic_plan; 286 } 287 288 void 289 data_destroy_arabic (void *data) 290 { 291 arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) data; 292 293 arabic_fallback_plan_destroy (arabic_plan->fallback_plan); 294 295 hb_free (data); 296 } 297 298 static void 299 arabic_joining (hb_buffer_t *buffer) 300 { 301 unsigned int count = buffer->len; 302 hb_glyph_info_t *info = buffer->info; 303 unsigned int prev = UINT_MAX, state = 0; 304 305 /* Check pre-context */ 306 for (unsigned int i = 0; i < buffer->context_len[0]; i++) 307 { 308 unsigned int this_type = get_joining_type (buffer->context[0][i], buffer->unicode->general_category (buffer->context[0][i])); 309 310 if (unlikely (this_type == JOINING_TYPE_T)) 311 continue; 312 313 const arabic_state_table_entry *entry = &arabic_state_table[state][this_type]; 314 state = entry->next_state; 315 break; 316 } 317 318 for (unsigned int i = 0; i < count; i++) 319 { 320 unsigned int this_type = get_joining_type (info[i].codepoint, _hb_glyph_info_get_general_category (&info[i])); 321 322 if (unlikely (this_type == JOINING_TYPE_T)) { 323 info[i].arabic_shaping_action() = NONE; 324 continue; 325 } 326 327 const arabic_state_table_entry *entry = &arabic_state_table[state][this_type]; 328 329 if (entry->prev_action != NONE && prev != UINT_MAX) 330 { 331 info[prev].arabic_shaping_action() = entry->prev_action; 332 buffer->safe_to_insert_tatweel (prev, i + 1); 333 } 334 else 335 { 336 if (prev == UINT_MAX) 337 { 338 if (this_type >= JOINING_TYPE_R) 339 buffer->unsafe_to_concat_from_outbuffer (0, i + 1); 340 } 341 else 342 { 343 if (this_type >= JOINING_TYPE_R || 344 (2 <= state && state <= 5) /* States that have a possible prev_action. */) 345 buffer->unsafe_to_concat (prev, i + 1); 346 } 347 } 348 349 info[i].arabic_shaping_action() = entry->curr_action; 350 351 prev = i; 352 state = entry->next_state; 353 } 354 355 for (unsigned int i = 0; i < buffer->context_len[1]; i++) 356 { 357 unsigned int this_type = get_joining_type (buffer->context[1][i], buffer->unicode->general_category (buffer->context[1][i])); 358 359 if (unlikely (this_type == JOINING_TYPE_T)) 360 continue; 361 362 const arabic_state_table_entry *entry = &arabic_state_table[state][this_type]; 363 if (entry->prev_action != NONE && prev != UINT_MAX) 364 { 365 info[prev].arabic_shaping_action() = entry->prev_action; 366 buffer->safe_to_insert_tatweel (prev, buffer->len); 367 } 368 else if (2 <= state && state <= 5) /* States that have a possible prev_action. */ 369 { 370 buffer->unsafe_to_concat (prev, buffer->len); 371 } 372 break; 373 } 374 } 375 376 static void 377 mongolian_variation_selectors (hb_buffer_t *buffer) 378 { 379 /* Copy arabic_shaping_action() from base to Mongolian variation selectors. */ 380 unsigned int count = buffer->len; 381 hb_glyph_info_t *info = buffer->info; 382 for (unsigned int i = 1; i < count; i++) 383 if (unlikely (hb_in_ranges<hb_codepoint_t> (info[i].codepoint, 0x180Bu, 0x180Du, 0x180Fu, 0x180Fu))) 384 info[i].arabic_shaping_action() = info[i - 1].arabic_shaping_action(); 385 } 386 387 void 388 setup_masks_arabic_plan (const arabic_shape_plan_t *arabic_plan, 389 hb_buffer_t *buffer, 390 hb_script_t script) 391 { 392 HB_BUFFER_ALLOCATE_VAR (buffer, arabic_shaping_action); 393 394 arabic_joining (buffer); 395 if (script == HB_SCRIPT_MONGOLIAN) 396 mongolian_variation_selectors (buffer); 397 398 unsigned int count = buffer->len; 399 hb_glyph_info_t *info = buffer->info; 400 for (unsigned int i = 0; i < count; i++) 401 info[i].mask |= arabic_plan->mask_array[info[i].arabic_shaping_action()]; 402 } 403 404 static void 405 setup_masks_arabic (const hb_ot_shape_plan_t *plan, 406 hb_buffer_t *buffer, 407 hb_font_t *font HB_UNUSED) 408 { 409 const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data; 410 setup_masks_arabic_plan (arabic_plan, buffer, plan->props.script); 411 } 412 413 static bool 414 arabic_fallback_shape (const hb_ot_shape_plan_t *plan, 415 hb_font_t *font, 416 hb_buffer_t *buffer) 417 { 418 #ifdef HB_NO_OT_SHAPER_ARABIC_FALLBACK 419 return false; 420 #endif 421 422 const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data; 423 424 if (!arabic_plan->do_fallback) 425 return false; 426 427 retry: 428 arabic_fallback_plan_t *fallback_plan = arabic_plan->fallback_plan; 429 if (unlikely (!fallback_plan)) 430 { 431 /* This sucks. We need a font to build the fallback plan... */ 432 fallback_plan = arabic_fallback_plan_create (plan, font); 433 if (unlikely (!arabic_plan->fallback_plan.cmpexch (nullptr, fallback_plan))) 434 { 435 arabic_fallback_plan_destroy (fallback_plan); 436 goto retry; 437 } 438 } 439 440 arabic_fallback_plan_shape (fallback_plan, font, buffer); 441 return true; 442 } 443 444 /* 445 * Stretch feature: "stch". 446 * See example here: 447 * https://docs.microsoft.com/en-us/typography/script-development/syriac 448 * We implement this in a generic way, such that the Arabic subtending 449 * marks can use it as well. 450 */ 451 452 static bool 453 record_stch (const hb_ot_shape_plan_t *plan, 454 hb_font_t *font HB_UNUSED, 455 hb_buffer_t *buffer) 456 { 457 const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data; 458 if (!arabic_plan->has_stch) 459 return false; 460 461 /* 'stch' feature was just applied. Look for anything that multiplied, 462 * and record it for stch treatment later. Note that rtlm, frac, etc 463 * are applied before stch, but we assume that they didn't result in 464 * anything multiplying into 5 pieces, so it's safe-ish... */ 465 466 unsigned int count = buffer->len; 467 hb_glyph_info_t *info = buffer->info; 468 for (unsigned int i = 0; i < count; i++) 469 if (unlikely (_hb_glyph_info_multiplied (&info[i]))) 470 { 471 unsigned int comp = _hb_glyph_info_get_lig_comp (&info[i]); 472 info[i].arabic_shaping_action() = comp % 2 ? STCH_REPEATING : STCH_FIXED; 473 buffer->scratch_flags |= HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH; 474 } 475 return false; 476 } 477 478 static void 479 apply_stch (const hb_ot_shape_plan_t *plan HB_UNUSED, 480 hb_buffer_t *buffer, 481 hb_font_t *font) 482 { 483 if (likely (!(buffer->scratch_flags & HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH))) 484 return; 485 486 bool rtl = buffer->props.direction == HB_DIRECTION_RTL; 487 488 if (!rtl) 489 buffer->reverse (); 490 491 /* We do a two pass implementation: 492 * First pass calculates the exact number of extra glyphs we need, 493 * We then enlarge buffer to have that much room, 494 * Second pass applies the stretch, copying things to the end of buffer. 495 */ 496 497 int sign = font->x_scale < 0 ? -1 : +1; 498 unsigned int extra_glyphs_needed = 0; // Set during MEASURE, used during CUT 499 enum { MEASURE, CUT } /* step_t */; 500 501 for (unsigned int step = MEASURE; step <= CUT; step = step + 1) 502 { 503 unsigned int count = buffer->len; 504 hb_glyph_info_t *info = buffer->info; 505 hb_glyph_position_t *pos = buffer->pos; 506 unsigned int new_len = count + extra_glyphs_needed; // write head during CUT 507 unsigned int j = new_len; 508 for (unsigned int i = count; i; i--) 509 { 510 if (!hb_in_range<uint8_t> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING)) 511 { 512 if (step == CUT) 513 { 514 --j; 515 info[j] = info[i - 1]; 516 pos[j] = pos[i - 1]; 517 } 518 continue; 519 } 520 521 /* Yay, justification! */ 522 523 hb_position_t w_total = 0; // Total to be filled 524 hb_position_t w_fixed = 0; // Sum of fixed tiles 525 hb_position_t w_repeating = 0; // Sum of repeating tiles 526 int n_fixed = 0; 527 int n_repeating = 0; 528 529 unsigned int end = i; 530 while (i && 531 hb_in_range<uint8_t> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING)) 532 { 533 i--; 534 hb_position_t width = font->get_glyph_h_advance (info[i].codepoint); 535 if (info[i].arabic_shaping_action() == STCH_FIXED) 536 { 537 w_fixed += width; 538 n_fixed++; 539 } 540 else 541 { 542 w_repeating += width; 543 n_repeating++; 544 } 545 } 546 unsigned int start = i; 547 unsigned int context = i; 548 while (context && 549 !hb_in_range<uint8_t> (info[context - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING) && 550 (_hb_glyph_info_is_default_ignorable (&info[context - 1]) || 551 HB_ARABIC_GENERAL_CATEGORY_IS_WORD (_hb_glyph_info_get_general_category (&info[context - 1])))) 552 { 553 context--; 554 w_total += pos[context].x_advance; 555 } 556 i++; // Don't touch i again. 557 558 DEBUG_MSG (ARABIC, nullptr, "%s stretch at (%u,%u,%u)", 559 step == MEASURE ? "measuring" : "cutting", context, start, end); 560 DEBUG_MSG (ARABIC, nullptr, "rest of word: count=%u width %" PRId32, start - context, w_total); 561 DEBUG_MSG (ARABIC, nullptr, "fixed tiles: count=%d width=%" PRId32, n_fixed, w_fixed); 562 DEBUG_MSG (ARABIC, nullptr, "repeating tiles: count=%d width=%" PRId32, n_repeating, w_repeating); 563 564 /* Number of additional times to repeat each repeating tile. */ 565 int n_copies = 0; 566 567 hb_position_t w_remaining = w_total - w_fixed; 568 if (sign * w_remaining > sign * w_repeating && sign * w_repeating > 0) 569 n_copies = (sign * w_remaining) / (sign * w_repeating) - 1; 570 571 /* See if we can improve the fit by adding an extra repeat and squeezing them together a bit. */ 572 hb_position_t extra_repeat_overlap = 0; 573 hb_position_t shortfall = sign * w_remaining - sign * w_repeating * (n_copies + 1); 574 if (shortfall > 0 && n_repeating > 0) 575 { 576 ++n_copies; 577 hb_position_t excess = (n_copies + 1) * sign * w_repeating - sign * w_remaining; 578 if (excess > 0) 579 { 580 extra_repeat_overlap = excess / (n_copies * n_repeating); 581 w_remaining = 0; 582 } 583 } 584 585 if (step == MEASURE) 586 { 587 extra_glyphs_needed += n_copies * n_repeating; 588 DEBUG_MSG (ARABIC, nullptr, "will add extra %d copies of repeating tiles", n_copies); 589 } 590 else 591 { 592 buffer->unsafe_to_break (context, end); 593 hb_position_t x_offset = w_remaining / 2; 594 for (unsigned int k = end; k > start; k--) 595 { 596 hb_position_t width = font->get_glyph_h_advance (info[k - 1].codepoint); 597 598 unsigned int repeat = 1; 599 if (info[k - 1].arabic_shaping_action() == STCH_REPEATING) 600 repeat += n_copies; 601 602 DEBUG_MSG (ARABIC, nullptr, "appending %u copies of glyph %" PRIu32 "; j=%u", 603 repeat, info[k - 1].codepoint, j); 604 pos[k - 1].x_advance = 0; 605 for (unsigned int n = 0; n < repeat; n++) 606 { 607 if (rtl) 608 { 609 x_offset -= width; 610 if (n > 0) 611 x_offset += extra_repeat_overlap; 612 } 613 pos[k - 1].x_offset = x_offset; 614 /* Append copy. */ 615 --j; 616 info[j] = info[k - 1]; 617 pos[j] = pos[k - 1]; 618 619 if (!rtl) 620 { 621 x_offset += width; 622 if (n > 0) 623 x_offset -= extra_repeat_overlap; 624 } 625 } 626 } 627 } 628 } 629 630 if (step == MEASURE) 631 { 632 if (unlikely (!buffer->ensure (count + extra_glyphs_needed))) 633 break; 634 } 635 else 636 { 637 assert (j == 0); 638 buffer->len = new_len; 639 } 640 } 641 642 if (!rtl) 643 buffer->reverse (); 644 } 645 646 647 static void 648 postprocess_glyphs_arabic (const hb_ot_shape_plan_t *plan, 649 hb_buffer_t *buffer, 650 hb_font_t *font) 651 { 652 apply_stch (plan, buffer, font); 653 } 654 655 /* https://www.unicode.org/reports/tr53/ */ 656 657 static const hb_codepoint_t 658 modifier_combining_marks[] = 659 { 660 0x0654u, /* ARABIC HAMZA ABOVE */ 661 0x0655u, /* ARABIC HAMZA BELOW */ 662 0x0658u, /* ARABIC MARK NOON GHUNNA */ 663 0x06DCu, /* ARABIC SMALL HIGH SEEN */ 664 0x06E3u, /* ARABIC SMALL LOW SEEN */ 665 0x06E7u, /* ARABIC SMALL HIGH YEH */ 666 0x06E8u, /* ARABIC SMALL HIGH NOON */ 667 0x08CAu, /* ARABIC SMALL HIGH FARSI YEH */ 668 0x08CBu, /* ARABIC SMALL HIGH YEH BARREE WITH TWO DOTS BELOW */ 669 0x08CDu, /* ARABIC SMALL HIGH ZAH */ 670 0x08CEu, /* ARABIC LARGE ROUND DOT ABOVE */ 671 0x08CFu, /* ARABIC LARGE ROUND DOT BELOW */ 672 0x08D3u, /* ARABIC SMALL LOW WAW */ 673 0x08F3u, /* ARABIC SMALL HIGH WAW */ 674 }; 675 676 static inline bool 677 info_is_mcm (const hb_glyph_info_t &info) 678 { 679 hb_codepoint_t u = info.codepoint; 680 for (unsigned int i = 0; i < ARRAY_LENGTH (modifier_combining_marks); i++) 681 if (u == modifier_combining_marks[i]) 682 return true; 683 return false; 684 } 685 686 static void 687 reorder_marks_arabic (const hb_ot_shape_plan_t *plan HB_UNUSED, 688 hb_buffer_t *buffer, 689 unsigned int start, 690 unsigned int end) 691 { 692 hb_glyph_info_t *info = buffer->info; 693 694 DEBUG_MSG (ARABIC, buffer, "Reordering marks from %u to %u", start, end); 695 696 unsigned int i = start; 697 for (unsigned int cc = 220; cc <= 230; cc += 10) 698 { 699 DEBUG_MSG (ARABIC, buffer, "Looking for %u's starting at %u", cc, i); 700 while (i < end && info_cc(info[i]) < cc) 701 i++; 702 DEBUG_MSG (ARABIC, buffer, "Looking for %u's stopped at %u", cc, i); 703 704 if (i == end) 705 break; 706 707 if (info_cc(info[i]) > cc) 708 continue; 709 710 unsigned int j = i; 711 while (j < end && info_cc(info[j]) == cc && info_is_mcm (info[j])) 712 j++; 713 714 if (i == j) 715 continue; 716 717 DEBUG_MSG (ARABIC, buffer, "Found %u's from %u to %u", cc, i, j); 718 719 /* Shift it! */ 720 DEBUG_MSG (ARABIC, buffer, "Shifting %u's: %u %u", cc, i, j); 721 hb_glyph_info_t temp[HB_OT_SHAPE_MAX_COMBINING_MARKS]; 722 assert (j - i <= ARRAY_LENGTH (temp)); 723 buffer->merge_clusters (start, j); 724 memmove (temp, &info[i], (j - i) * sizeof (hb_glyph_info_t)); 725 memmove (&info[start + j - i], &info[start], (i - start) * sizeof (hb_glyph_info_t)); 726 memmove (&info[start], temp, (j - i) * sizeof (hb_glyph_info_t)); 727 728 /* Renumber CC such that the reordered sequence is still sorted. 729 * 22 and 26 are chosen because they are smaller than all Arabic categories, 730 * and are folded back to 220/230 respectively during fallback mark positioning. 731 * 732 * We do this because the CGJ-handling logic in the normalizer relies on 733 * mark sequences having an increasing order even after this reordering. 734 * https://github.com/harfbuzz/harfbuzz/issues/554 735 * This, however, does break some obscure sequences, where the normalizer 736 * might compose a sequence that it should not. For example, in the seequence 737 * ALEF, HAMZAH, MADDAH, we should NOT try to compose ALEF+MADDAH, but with this 738 * renumbering, we will. 739 */ 740 unsigned int new_start = start + j - i; 741 unsigned int new_cc = cc == 220 ? HB_MODIFIED_COMBINING_CLASS_CCC22 : HB_MODIFIED_COMBINING_CLASS_CCC26; 742 while (start < new_start) 743 { 744 _hb_glyph_info_set_modified_combining_class (&info[start], new_cc); 745 start++; 746 } 747 748 i = j; 749 } 750 } 751 752 const hb_ot_shaper_t _hb_ot_shaper_arabic = 753 { 754 collect_features_arabic, 755 nullptr, /* override_features */ 756 data_create_arabic, 757 data_destroy_arabic, 758 nullptr, /* preprocess_text */ 759 postprocess_glyphs_arabic, 760 nullptr, /* decompose */ 761 nullptr, /* compose */ 762 setup_masks_arabic, 763 reorder_marks_arabic, 764 HB_TAG_NONE, /* gpos_tag */ 765 HB_OT_SHAPE_NORMALIZATION_MODE_DEFAULT, 766 HB_OT_SHAPE_ZERO_WIDTH_MARKS_BY_GDEF_LATE, 767 true, /* fallback_position */ 768 }; 769 770 771 #endif