1/* 2******************************************************************************* 3* Copyright (C) 2001-2014, International Business Machines 4* Corporation and others. All Rights Reserved. 5******************************************************************************* 6*/ 7/* Written by Simon Montagu, Matitiahu Allouche 8 * (ported from C code written by Markus W. Scherer) 9 */ 10 11package com.ibm.icu.text; 12 13 14import java.util.Arrays; 15 16final class BidiLine { 17 18 /* 19 * General remarks about the functions in this file: 20 * 21 * These functions deal with the aspects of potentially mixed-directional 22 * text in a single paragraph or in a line of a single paragraph 23 * which has already been processed according to 24 * the Unicode 3.0 Bidi algorithm as defined in 25 * http://www.unicode.org/unicode/reports/tr9/ , version 13, 26 * also described in The Unicode Standard, Version 4.0.1 . 27 * 28 * This means that there is a Bidi object with a levels 29 * and a dirProps array. 30 * paraLevel and direction are also set. 31 * Only if the length of the text is zero, then levels==dirProps==NULL. 32 * 33 * The overall directionality of the paragraph 34 * or line is used to bypass the reordering steps if possible. 35 * Even purely RTL text does not need reordering there because 36 * the getLogical/VisualIndex() methods can compute the 37 * index on the fly in such a case. 38 * 39 * The implementation of the access to same-level-runs and of the reordering 40 * do attempt to provide better performance and less memory usage compared to 41 * a direct implementation of especially rule (L2) with an array of 42 * one (32-bit) integer per text character. 43 * 44 * Here, the levels array is scanned as soon as necessary, and a vector of 45 * same-level-runs is created. Reordering then is done on this vector. 46 * For each run of text positions that were resolved to the same level, 47 * only 8 bytes are stored: the first text position of the run and the visual 48 * position behind the run after reordering. 49 * One sign bit is used to hold the directionality of the run. 50 * This is inefficient if there are many very short runs. If the average run 51 * length is <2, then this uses more memory. 52 * 53 * In a further attempt to save memory, the levels array is never changed 54 * after all the resolution rules (Xn, Wn, Nn, In). 55 * Many methods have to consider the field trailingWSStart: 56 * if it is less than length, then there is an implicit trailing run 57 * at the paraLevel, 58 * which is not reflected in the levels array. 59 * This allows a line Bidi object to use the same levels array as 60 * its paragraph parent object. 61 * 62 * When a Bidi object is created for a line of a paragraph, then the 63 * paragraph's levels and dirProps arrays are reused by way of setting 64 * a pointer into them, not by copying. This again saves memory and forbids to 65 * change the now shared levels for (L1). 66 */ 67 68 /* handle trailing WS (L1) -------------------------------------------------- */ 69 70 /* 71 * setTrailingWSStart() sets the start index for a trailing 72 * run of WS in the line. This is necessary because we do not modify 73 * the paragraph's levels array that we just point into. 74 * Using trailingWSStart is another form of performing (L1). 75 * 76 * To make subsequent operations easier, we also include the run 77 * before the WS if it is at the paraLevel - we merge the two here. 78 * 79 * This method is called only from setLine(), so paraLevel is 80 * set correctly for the line even when contextual multiple paragraphs. 81 */ 82 83 static void setTrailingWSStart(Bidi bidi) 84 { 85 byte[] dirProps = bidi.dirProps; 86 byte[] levels = bidi.levels; 87 int start = bidi.length; 88 byte paraLevel = bidi.paraLevel; 89 90 /* If the line is terminated by a block separator, all preceding WS etc... 91 are already set to paragraph level. 92 Setting trailingWSStart to pBidi->length will avoid changing the 93 level of B chars from 0 to paraLevel in getLevels when 94 orderParagraphsLTR==TRUE 95 */ 96 if (dirProps[start - 1] == Bidi.B) { 97 bidi.trailingWSStart = start; /* currently == bidi.length */ 98 return; 99 } 100 /* go backwards across all WS, BN, explicit codes */ 101 while (start > 0 && 102 (Bidi.DirPropFlag(dirProps[start - 1]) & Bidi.MASK_WS) != 0) { 103 --start; 104 } 105 106 /* if the WS run can be merged with the previous run then do so here */ 107 while (start > 0 && levels[start - 1] == paraLevel) { 108 --start; 109 } 110 111 bidi.trailingWSStart=start; 112 } 113 114 static Bidi setLine(Bidi paraBidi, int start, int limit) { 115 int length; 116 117 Bidi lineBidi = new Bidi(); 118 119 /* set the values in lineBidi from its paraBidi parent */ 120 /* class members are already initialized to 0 */ 121 // lineBidi.paraBidi = null; /* mark unfinished setLine */ 122 // lineBidi.flags = 0; 123 // lineBidi.controlCount = 0; 124 125 length = lineBidi.length = lineBidi.originalLength = 126 lineBidi.resultLength = limit - start; 127 128 lineBidi.text = new char[length]; 129 System.arraycopy(paraBidi.text, start, lineBidi.text, 0, length); 130 lineBidi.paraLevel = paraBidi.GetParaLevelAt(start); 131 lineBidi.paraCount = paraBidi.paraCount; 132 lineBidi.runs = new BidiRun[0]; 133 lineBidi.reorderingMode = paraBidi.reorderingMode; 134 lineBidi.reorderingOptions = paraBidi.reorderingOptions; 135 if (paraBidi.controlCount > 0) { 136 int j; 137 for (j = start; j < limit; j++) { 138 if (Bidi.IsBidiControlChar(paraBidi.text[j])) { 139 lineBidi.controlCount++; 140 } 141 } 142 lineBidi.resultLength -= lineBidi.controlCount; 143 } 144 /* copy proper subset of DirProps */ 145 lineBidi.getDirPropsMemory(length); 146 lineBidi.dirProps = lineBidi.dirPropsMemory; 147 System.arraycopy(paraBidi.dirProps, start, lineBidi.dirProps, 0, 148 length); 149 /* copy proper subset of Levels */ 150 lineBidi.getLevelsMemory(length); 151 lineBidi.levels = lineBidi.levelsMemory; 152 System.arraycopy(paraBidi.levels, start, lineBidi.levels, 0, 153 length); 154 lineBidi.runCount = -1; 155 156 if (paraBidi.direction != Bidi.MIXED) { 157 /* the parent is already trivial */ 158 lineBidi.direction = paraBidi.direction; 159 160 /* 161 * The parent's levels are all either 162 * implicitly or explicitly ==paraLevel; 163 * do the same here. 164 */ 165 if (paraBidi.trailingWSStart <= start) { 166 lineBidi.trailingWSStart = 0; 167 } else if (paraBidi.trailingWSStart < limit) { 168 lineBidi.trailingWSStart = paraBidi.trailingWSStart - start; 169 } else { 170 lineBidi.trailingWSStart = length; 171 } 172 } else { 173 byte[] levels = lineBidi.levels; 174 int i, trailingWSStart; 175 byte level; 176 177 setTrailingWSStart(lineBidi); 178 trailingWSStart = lineBidi.trailingWSStart; 179 180 /* recalculate lineBidi.direction */ 181 if (trailingWSStart == 0) { 182 /* all levels are at paraLevel */ 183 lineBidi.direction = (byte)(lineBidi.paraLevel & 1); 184 } else { 185 /* get the level of the first character */ 186 level = (byte)(levels[0] & 1); 187 188 /* if there is anything of a different level, then the line 189 is mixed */ 190 if (trailingWSStart < length && 191 (lineBidi.paraLevel & 1) != level) { 192 /* the trailing WS is at paraLevel, which differs from 193 levels[0] */ 194 lineBidi.direction = Bidi.MIXED; 195 } else { 196 /* see if levels[1..trailingWSStart-1] have the same 197 direction as levels[0] and paraLevel */ 198 for (i = 1; ; i++) { 199 if (i == trailingWSStart) { 200 /* the direction values match those in level */ 201 lineBidi.direction = level; 202 break; 203 } else if ((levels[i] & 1) != level) { 204 lineBidi.direction = Bidi.MIXED; 205 break; 206 } 207 } 208 } 209 } 210 211 switch(lineBidi.direction) { 212 case Bidi.DIRECTION_LEFT_TO_RIGHT: 213 /* make sure paraLevel is even */ 214 lineBidi.paraLevel = (byte) 215 ((lineBidi.paraLevel + 1) & ~1); 216 217 /* all levels are implicitly at paraLevel (important for 218 getLevels()) */ 219 lineBidi.trailingWSStart = 0; 220 break; 221 case Bidi.DIRECTION_RIGHT_TO_LEFT: 222 /* make sure paraLevel is odd */ 223 lineBidi.paraLevel |= 1; 224 225 /* all levels are implicitly at paraLevel (important for 226 getLevels()) */ 227 lineBidi.trailingWSStart = 0; 228 break; 229 default: 230 break; 231 } 232 } 233 lineBidi.paraBidi = paraBidi; /* mark successful setLine */ 234 return lineBidi; 235 } 236 237 static byte getLevelAt(Bidi bidi, int charIndex) 238 { 239 /* return paraLevel if in the trailing WS run, otherwise the real level */ 240 if (bidi.direction != Bidi.MIXED || charIndex >= bidi.trailingWSStart) { 241 return bidi.GetParaLevelAt(charIndex); 242 } else { 243 return bidi.levels[charIndex]; 244 } 245 } 246 247 static byte[] getLevels(Bidi bidi) 248 { 249 int start = bidi.trailingWSStart; 250 int length = bidi.length; 251 252 if (start != length) { 253 /* the current levels array does not reflect the WS run */ 254 /* 255 * After the previous if(), we know that the levels array 256 * has an implicit trailing WS run and therefore does not fully 257 * reflect itself all the levels. 258 * This must be a Bidi object for a line, and 259 * we need to create a new levels array. 260 */ 261 /* bidi.paraLevel is ok even if contextual multiple paragraphs, 262 since bidi is a line object */ 263 Arrays.fill(bidi.levels, start, length, bidi.paraLevel); 264 265 /* this new levels array is set for the line and reflects the WS run */ 266 bidi.trailingWSStart = length; 267 } 268 if (length < bidi.levels.length) { 269 byte[] levels = new byte[length]; 270 System.arraycopy(bidi.levels, 0, levels, 0, length); 271 return levels; 272 } 273 return bidi.levels; 274 } 275 276 static BidiRun getLogicalRun(Bidi bidi, int logicalPosition) 277 { 278 /* this is done based on runs rather than on levels since levels have 279 a special interpretation when REORDER_RUNS_ONLY 280 */ 281 BidiRun newRun = new BidiRun(), iRun; 282 getRuns(bidi); 283 int runCount = bidi.runCount; 284 int visualStart = 0, logicalLimit = 0; 285 iRun = bidi.runs[0]; 286 287 for (int i = 0; i < runCount; i++) { 288 iRun = bidi.runs[i]; 289 logicalLimit = iRun.start + iRun.limit - visualStart; 290 if ((logicalPosition >= iRun.start) && 291 (logicalPosition < logicalLimit)) { 292 break; 293 } 294 visualStart = iRun.limit; 295 } 296 newRun.start = iRun.start; 297 newRun.limit = logicalLimit; 298 newRun.level = iRun.level; 299 return newRun; 300 } 301 302 static BidiRun getVisualRun(Bidi bidi, int runIndex) 303 { 304 int start = bidi.runs[runIndex].start; 305 int limit; 306 byte level = bidi.runs[runIndex].level; 307 308 if (runIndex > 0) { 309 limit = start + 310 bidi.runs[runIndex].limit - 311 bidi.runs[runIndex - 1].limit; 312 } else { 313 limit = start + bidi.runs[0].limit; 314 } 315 return new BidiRun(start, limit, level); 316 } 317 318 /* in trivial cases there is only one trivial run; called by getRuns() */ 319 static void getSingleRun(Bidi bidi, byte level) { 320 /* simple, single-run case */ 321 bidi.runs = bidi.simpleRuns; 322 bidi.runCount = 1; 323 324 /* fill and reorder the single run */ 325 bidi.runs[0] = new BidiRun(0, bidi.length, level); 326 } 327 328 /* reorder the runs array (L2) ---------------------------------------------- */ 329 330 /* 331 * Reorder the same-level runs in the runs array. 332 * Here, runCount>1 and maxLevel>=minLevel>=paraLevel. 333 * All the visualStart fields=logical start before reordering. 334 * The "odd" bits are not set yet. 335 * 336 * Reordering with this data structure lends itself to some handy shortcuts: 337 * 338 * Since each run is moved but not modified, and since at the initial maxLevel 339 * each sequence of same-level runs consists of only one run each, we 340 * don't need to do anything there and can predecrement maxLevel. 341 * In many simple cases, the reordering is thus done entirely in the 342 * index mapping. 343 * Also, reordering occurs only down to the lowest odd level that occurs, 344 * which is minLevel|1. However, if the lowest level itself is odd, then 345 * in the last reordering the sequence of the runs at this level or higher 346 * will be all runs, and we don't need the elaborate loop to search for them. 347 * This is covered by ++minLevel instead of minLevel|=1 followed 348 * by an extra reorder-all after the reorder-some loop. 349 * About a trailing WS run: 350 * Such a run would need special treatment because its level is not 351 * reflected in levels[] if this is not a paragraph object. 352 * Instead, all characters from trailingWSStart on are implicitly at 353 * paraLevel. 354 * However, for all maxLevel>paraLevel, this run will never be reordered 355 * and does not need to be taken into account. maxLevel==paraLevel is only reordered 356 * if minLevel==paraLevel is odd, which is done in the extra segment. 357 * This means that for the main reordering loop we don't need to consider 358 * this run and can --runCount. If it is later part of the all-runs 359 * reordering, then runCount is adjusted accordingly. 360 */ 361 private static void reorderLine(Bidi bidi, byte minLevel, byte maxLevel) { 362 363 /* nothing to do? */ 364 if (maxLevel<=(minLevel|1)) { 365 return; 366 } 367 368 BidiRun[] runs; 369 BidiRun tempRun; 370 byte[] levels; 371 int firstRun, endRun, limitRun, runCount; 372 373 /* 374 * Reorder only down to the lowest odd level 375 * and reorder at an odd minLevel in a separate, simpler loop. 376 * See comments above for why minLevel is always incremented. 377 */ 378 ++minLevel; 379 380 runs = bidi.runs; 381 levels = bidi.levels; 382 runCount = bidi.runCount; 383 384 /* do not include the WS run at paraLevel<=old minLevel except in the simple loop */ 385 if (bidi.trailingWSStart < bidi.length) { 386 --runCount; 387 } 388 389 while (--maxLevel >= minLevel) { 390 firstRun = 0; 391 392 /* loop for all sequences of runs */ 393 for ( ; ; ) { 394 /* look for a sequence of runs that are all at >=maxLevel */ 395 /* look for the first run of such a sequence */ 396 while (firstRun < runCount && levels[runs[firstRun].start] < maxLevel) { 397 ++firstRun; 398 } 399 if (firstRun >= runCount) { 400 break; /* no more such runs */ 401 } 402 403 /* look for the limit run of such a sequence (the run behind it) */ 404 for (limitRun = firstRun; ++limitRun < runCount && 405 levels[runs[limitRun].start]>=maxLevel; ) {} 406 407 /* Swap the entire sequence of runs from firstRun to limitRun-1. */ 408 endRun = limitRun - 1; 409 while (firstRun < endRun) { 410 tempRun = runs[firstRun]; 411 runs[firstRun] = runs[endRun]; 412 runs[endRun] = tempRun; 413 ++firstRun; 414 --endRun; 415 } 416 417 if (limitRun == runCount) { 418 break; /* no more such runs */ 419 } else { 420 firstRun = limitRun + 1; 421 } 422 } 423 } 424 425 /* now do maxLevel==old minLevel (==odd!), see above */ 426 if ((minLevel & 1) == 0) { 427 firstRun = 0; 428 429 /* include the trailing WS run in this complete reordering */ 430 if (bidi.trailingWSStart == bidi.length) { 431 --runCount; 432 } 433 434 /* Swap the entire sequence of all runs. (endRun==runCount) */ 435 while (firstRun < runCount) { 436 tempRun = runs[firstRun]; 437 runs[firstRun] = runs[runCount]; 438 runs[runCount] = tempRun; 439 ++firstRun; 440 --runCount; 441 } 442 } 443 } 444 445 /* compute the runs array --------------------------------------------------- */ 446 447 static int getRunFromLogicalIndex(Bidi bidi, int logicalIndex) { 448 BidiRun[] runs = bidi.runs; 449 int runCount = bidi.runCount, visualStart = 0, i, length, logicalStart; 450 451 for (i = 0; i < runCount; i++) { 452 length = runs[i].limit - visualStart; 453 logicalStart = runs[i].start; 454 if ((logicalIndex >= logicalStart) && (logicalIndex < (logicalStart+length))) { 455 return i; 456 } 457 visualStart += length; 458 } 459 ///CLOVER:OFF 460 /* we should never get here */ 461 throw new IllegalStateException("Internal ICU error in getRunFromLogicalIndex"); 462 ///CLOVER:ON 463 } 464 465 /* 466 * Compute the runs array from the levels array. 467 * After getRuns() returns true, runCount is guaranteed to be >0 468 * and the runs are reordered. 469 * Odd-level runs have visualStart on their visual right edge and 470 * they progress visually to the left. 471 * If option OPTION_INSERT_MARKS is set, insertRemove will contain the 472 * sum of appropriate LRM/RLM_BEFORE/AFTER flags. 473 * If option OPTION_REMOVE_CONTROLS is set, insertRemove will contain the 474 * negative number of BiDi control characters within this run. 475 */ 476 static void getRuns(Bidi bidi) { 477 /* 478 * This method returns immediately if the runs are already set. This 479 * includes the case of length==0 (handled in setPara).. 480 */ 481 if (bidi.runCount >= 0) { 482 return; 483 } 484 if (bidi.direction != Bidi.MIXED) { 485 /* simple, single-run case - this covers length==0 */ 486 /* bidi.paraLevel is ok even for contextual multiple paragraphs */ 487 getSingleRun(bidi, bidi.paraLevel); 488 } else /* Bidi.MIXED, length>0 */ { 489 /* mixed directionality */ 490 int length = bidi.length, limit; 491 byte[] levels = bidi.levels; 492 int i, runCount; 493 byte level = -1; /* initialize with no valid level */ 494 /* 495 * If there are WS characters at the end of the line 496 * and the run preceding them has a level different from 497 * paraLevel, then they will form their own run at paraLevel (L1). 498 * Count them separately. 499 * We need some special treatment for this in order to not 500 * modify the levels array which a line Bidi object shares 501 * with its paragraph parent and its other line siblings. 502 * In other words, for the trailing WS, it may be 503 * levels[]!=paraLevel but we have to treat it like it were so. 504 */ 505 limit = bidi.trailingWSStart; 506 /* count the runs, there is at least one non-WS run, and limit>0 */ 507 runCount = 0; 508 for (i = 0; i < limit; ++i) { 509 /* increment runCount at the start of each run */ 510 if (levels[i] != level) { 511 ++runCount; 512 level = levels[i]; 513 } 514 } 515 516 /* 517 * We don't need to see if the last run can be merged with a trailing 518 * WS run because setTrailingWSStart() would have done that. 519 */ 520 if (runCount == 1 && limit == length) { 521 /* There is only one non-WS run and no trailing WS-run. */ 522 getSingleRun(bidi, levels[0]); 523 } else /* runCount>1 || limit<length */ { 524 /* allocate and set the runs */ 525 BidiRun[] runs; 526 int runIndex, start; 527 byte minLevel = Bidi.MAX_EXPLICIT_LEVEL + 1; 528 byte maxLevel=0; 529 530 /* now, count a (non-mergeable) WS run */ 531 if (limit < length) { 532 ++runCount; 533 } 534 535 /* runCount > 1 */ 536 bidi.getRunsMemory(runCount); 537 runs = bidi.runsMemory; 538 539 /* set the runs */ 540 /* FOOD FOR THOUGHT: this could be optimized, e.g.: 541 * 464->444, 484->444, 575->555, 595->555 542 * However, that would take longer. Check also how it would 543 * interact with BiDi control removal and inserting Marks. 544 */ 545 runIndex = 0; 546 547 /* search for the run limits and initialize visualLimit values with the run lengths */ 548 i = 0; 549 do { 550 /* prepare this run */ 551 start = i; 552 level = levels[i]; 553 if (level < minLevel) { 554 minLevel = level; 555 } 556 if (level > maxLevel) { 557 maxLevel = level; 558 } 559 560 /* look for the run limit */ 561 while (++i < limit && levels[i] == level) {} 562 563 /* i is another run limit */ 564 runs[runIndex] = new BidiRun(start, i - start, level); 565 ++runIndex; 566 } while (i < limit); 567 568 if (limit < length) { 569 /* there is a separate WS run */ 570 runs[runIndex] = new BidiRun(limit, length - limit, bidi.paraLevel); 571 /* For the trailing WS run, bidi.paraLevel is ok even 572 if contextual multiple paragraphs. */ 573 if (bidi.paraLevel < minLevel) { 574 minLevel = bidi.paraLevel; 575 } 576 } 577 578 /* set the object fields */ 579 bidi.runs = runs; 580 bidi.runCount = runCount; 581 582 reorderLine(bidi, minLevel, maxLevel); 583 584 /* now add the direction flags and adjust the visualLimit's to be just that */ 585 /* this loop will also handle the trailing WS run */ 586 limit = 0; 587 for (i = 0; i < runCount; ++i) { 588 runs[i].level = levels[runs[i].start]; 589 limit = (runs[i].limit += limit); 590 } 591 592 /* Set the embedding level for the trailing WS run. */ 593 /* For a RTL paragraph, it will be the *first* run in visual order. */ 594 /* For the trailing WS run, bidi.paraLevel is ok even if 595 contextual multiple paragraphs. */ 596 if (runIndex < runCount) { 597 int trailingRun = ((bidi.paraLevel & 1) != 0)? 0 : runIndex; 598 runs[trailingRun].level = bidi.paraLevel; 599 } 600 } 601 } 602 603 /* handle insert LRM/RLM BEFORE/AFTER run */ 604 if (bidi.insertPoints.size > 0) { 605 Bidi.Point point; 606 int runIndex, ip; 607 for (ip = 0; ip < bidi.insertPoints.size; ip++) { 608 point = bidi.insertPoints.points[ip]; 609 runIndex = getRunFromLogicalIndex(bidi, point.pos); 610 bidi.runs[runIndex].insertRemove |= point.flag; 611 } 612 } 613 614 /* handle remove BiDi control characters */ 615 if (bidi.controlCount > 0) { 616 int runIndex, ic; 617 char c; 618 for (ic = 0; ic < bidi.length; ic++) { 619 c = bidi.text[ic]; 620 if (Bidi.IsBidiControlChar(c)) { 621 runIndex = getRunFromLogicalIndex(bidi, ic); 622 bidi.runs[runIndex].insertRemove--; 623 } 624 } 625 } 626 } 627 628 static int[] prepareReorder(byte[] levels, byte[] pMinLevel, byte[] pMaxLevel) 629 { 630 int start; 631 byte level, minLevel, maxLevel; 632 633 if (levels == null || levels.length <= 0) { 634 return null; 635 } 636 637 /* determine minLevel and maxLevel */ 638 minLevel = Bidi.MAX_EXPLICIT_LEVEL + 1; 639 maxLevel = 0; 640 for (start = levels.length; start>0; ) { 641 level = levels[--start]; 642 if (level < 0) { 643 return null; 644 } 645 if (level > (Bidi.MAX_EXPLICIT_LEVEL + 1)) { 646 return null; 647 } 648 if (level < minLevel) { 649 minLevel = level; 650 } 651 if (level > maxLevel) { 652 maxLevel = level; 653 } 654 } 655 pMinLevel[0] = minLevel; 656 pMaxLevel[0] = maxLevel; 657 658 /* initialize the index map */ 659 int[] indexMap = new int[levels.length]; 660 for (start = levels.length; start > 0; ) { 661 --start; 662 indexMap[start] = start; 663 } 664 665 return indexMap; 666 } 667 668 static int[] reorderLogical(byte[] levels) 669 { 670 byte[] aMinLevel = new byte[1]; 671 byte[] aMaxLevel = new byte[1]; 672 int start, limit, sumOfSosEos; 673 byte minLevel, maxLevel; 674 int[] indexMap = prepareReorder(levels, aMinLevel, aMaxLevel); 675 if (indexMap == null) { 676 return null; 677 } 678 679 minLevel = aMinLevel[0]; 680 maxLevel = aMaxLevel[0]; 681 682 /* nothing to do? */ 683 if (minLevel == maxLevel && (minLevel & 1) == 0) { 684 return indexMap; 685 } 686 687 /* reorder only down to the lowest odd level */ 688 minLevel |= 1; 689 690 /* loop maxLevel..minLevel */ 691 do { 692 start = 0; 693 694 /* loop for all sequences of levels to reorder at the current maxLevel */ 695 for ( ; ; ) { 696 /* look for a sequence of levels that are all at >=maxLevel */ 697 /* look for the first index of such a sequence */ 698 while (start < levels.length && levels[start] < maxLevel) { 699 ++start; 700 } 701 if (start >= levels.length) { 702 break; /* no more such sequences */ 703 } 704 705 /* look for the limit of such a sequence (the index behind it) */ 706 for (limit = start; ++limit < levels.length && levels[limit] >= maxLevel; ) {} 707 708 /* 709 * sos=start of sequence, eos=end of sequence 710 * 711 * The closed (inclusive) interval from sos to eos includes all the logical 712 * and visual indexes within this sequence. They are logically and 713 * visually contiguous and in the same range. 714 * 715 * For each run, the new visual index=sos+eos-old visual index; 716 * we pre-add sos+eos into sumOfSosEos -> 717 * new visual index=sumOfSosEos-old visual index; 718 */ 719 sumOfSosEos = start + limit - 1; 720 721 /* reorder each index in the sequence */ 722 do { 723 indexMap[start] = sumOfSosEos - indexMap[start]; 724 } while (++start < limit); 725 726 /* start==limit */ 727 if (limit == levels.length) { 728 break; /* no more such sequences */ 729 } else { 730 start = limit + 1; 731 } 732 } 733 } while (--maxLevel >= minLevel); 734 return indexMap; 735 } 736 737 static int[] reorderVisual(byte[] levels) 738 { 739 byte[] aMinLevel = new byte[1]; 740 byte[] aMaxLevel = new byte[1]; 741 int start, end, limit, temp; 742 byte minLevel, maxLevel; 743 744 int[] indexMap = prepareReorder(levels, aMinLevel, aMaxLevel); 745 if (indexMap == null) { 746 return null; 747 } 748 749 minLevel = aMinLevel[0]; 750 maxLevel = aMaxLevel[0]; 751 752 /* nothing to do? */ 753 if (minLevel == maxLevel && (minLevel & 1) == 0) { 754 return indexMap; 755 } 756 757 /* reorder only down to the lowest odd level */ 758 minLevel |= 1; 759 760 /* loop maxLevel..minLevel */ 761 do { 762 start = 0; 763 764 /* loop for all sequences of levels to reorder at the current maxLevel */ 765 for ( ; ; ) { 766 /* look for a sequence of levels that are all at >=maxLevel */ 767 /* look for the first index of such a sequence */ 768 while (start < levels.length && levels[start] < maxLevel) { 769 ++start; 770 } 771 if (start >= levels.length) { 772 break; /* no more such runs */ 773 } 774 775 /* look for the limit of such a sequence (the index behind it) */ 776 for (limit = start; ++limit < levels.length && levels[limit] >= maxLevel; ) {} 777 778 /* 779 * Swap the entire interval of indexes from start to limit-1. 780 * We don't need to swap the levels for the purpose of this 781 * algorithm: the sequence of levels that we look at does not 782 * move anyway. 783 */ 784 end = limit - 1; 785 while (start < end) { 786 temp = indexMap[start]; 787 indexMap[start] = indexMap[end]; 788 indexMap[end] = temp; 789 790 ++start; 791 --end; 792 } 793 794 if (limit == levels.length) { 795 break; /* no more such sequences */ 796 } else { 797 start = limit + 1; 798 } 799 } 800 } while (--maxLevel >= minLevel); 801 802 return indexMap; 803 } 804 805 static int getVisualIndex(Bidi bidi, int logicalIndex) 806 { 807 int visualIndex = Bidi.MAP_NOWHERE; 808 809 /* we can do the trivial cases without the runs array */ 810 switch(bidi.direction) { 811 case Bidi.LTR: 812 visualIndex = logicalIndex; 813 break; 814 case Bidi.RTL: 815 visualIndex = bidi.length - logicalIndex - 1; 816 break; 817 default: 818 getRuns(bidi); 819 BidiRun[] runs = bidi.runs; 820 int i, visualStart = 0, offset, length; 821 822 /* linear search for the run, search on the visual runs */ 823 for (i = 0; i < bidi.runCount; ++i) { 824 length = runs[i].limit - visualStart; 825 offset = logicalIndex - runs[i].start; 826 if (offset >= 0 && offset < length) { 827 if (runs[i].isEvenRun()) { 828 /* LTR */ 829 visualIndex = visualStart + offset; 830 } else { 831 /* RTL */ 832 visualIndex = visualStart + length - offset - 1; 833 } 834 break; /* exit for loop */ 835 } 836 visualStart += length; 837 } 838 if (i >= bidi.runCount) { 839 return Bidi.MAP_NOWHERE; 840 } 841 } 842 843 if (bidi.insertPoints.size > 0) { 844 /* add the number of added marks until the calculated visual index */ 845 BidiRun runs[] = bidi.runs; 846 int i, length, insertRemove; 847 int visualStart = 0, markFound = 0; 848 for (i = 0; ; i++, visualStart += length) { 849 length = runs[i].limit - visualStart; 850 insertRemove = runs[i].insertRemove; 851 if ((insertRemove & (Bidi.LRM_BEFORE|Bidi.RLM_BEFORE)) > 0) { 852 markFound++; 853 } 854 /* is it the run containing the visual index? */ 855 if (visualIndex < runs[i].limit) { 856 return visualIndex + markFound; 857 } 858 if ((insertRemove & (Bidi.LRM_AFTER|Bidi.RLM_AFTER)) > 0) { 859 markFound++; 860 } 861 } 862 } 863 else if (bidi.controlCount > 0) { 864 /* subtract the number of controls until the calculated visual index */ 865 BidiRun[] runs = bidi.runs; 866 int i, j, start, limit, length, insertRemove; 867 int visualStart = 0, controlFound = 0; 868 char uchar = bidi.text[logicalIndex]; 869 /* is the logical index pointing to a control ? */ 870 if (Bidi.IsBidiControlChar(uchar)) { 871 return Bidi.MAP_NOWHERE; 872 } 873 /* loop on runs */ 874 for (i = 0; ; i++, visualStart += length) { 875 length = runs[i].limit - visualStart; 876 insertRemove = runs[i].insertRemove; 877 /* calculated visual index is beyond this run? */ 878 if (visualIndex >= runs[i].limit) { 879 controlFound -= insertRemove; 880 continue; 881 } 882 /* calculated visual index must be within current run */ 883 if (insertRemove == 0) { 884 return visualIndex - controlFound; 885 } 886 if (runs[i].isEvenRun()) { 887 /* LTR: check from run start to logical index */ 888 start = runs[i].start; 889 limit = logicalIndex; 890 } else { 891 /* RTL: check from logical index to run end */ 892 start = logicalIndex + 1; 893 limit = runs[i].start + length; 894 } 895 for (j = start; j < limit; j++) { 896 uchar = bidi.text[j]; 897 if (Bidi.IsBidiControlChar(uchar)) { 898 controlFound++; 899 } 900 } 901 return visualIndex - controlFound; 902 } 903 } 904 905 return visualIndex; 906 } 907 908 static int getLogicalIndex(Bidi bidi, int visualIndex) 909 { 910 BidiRun[] runs; 911 int i, runCount, start; 912 913 runs = bidi.runs; 914 runCount = bidi.runCount; 915 if (bidi.insertPoints.size > 0) { 916 /* handle inserted LRM/RLM */ 917 int markFound = 0, insertRemove; 918 int visualStart = 0, length; 919 /* subtract number of marks until visual index */ 920 for (i = 0; ; i++, visualStart += length) { 921 length = runs[i].limit - visualStart; 922 insertRemove = runs[i].insertRemove; 923 if ((insertRemove & (Bidi.LRM_BEFORE|Bidi.RLM_BEFORE)) > 0) { 924 if (visualIndex <= (visualStart+markFound)) { 925 return Bidi.MAP_NOWHERE; 926 } 927 markFound++; 928 } 929 /* is adjusted visual index within this run? */ 930 if (visualIndex < (runs[i].limit + markFound)) { 931 visualIndex -= markFound; 932 break; 933 } 934 if ((insertRemove & (Bidi.LRM_AFTER|Bidi.RLM_AFTER)) > 0) { 935 if (visualIndex == (visualStart + length + markFound)) { 936 return Bidi.MAP_NOWHERE; 937 } 938 markFound++; 939 } 940 } 941 } 942 else if (bidi.controlCount > 0) { 943 /* handle removed BiDi control characters */ 944 int controlFound = 0, insertRemove, length; 945 int logicalStart, logicalEnd, visualStart = 0, j, k; 946 char uchar; 947 boolean evenRun; 948 /* add number of controls until visual index */ 949 for (i = 0; ; i++, visualStart += length) { 950 length = runs[i].limit - visualStart; 951 insertRemove = runs[i].insertRemove; 952 /* is adjusted visual index beyond current run? */ 953 if (visualIndex >= (runs[i].limit - controlFound + insertRemove)) { 954 controlFound -= insertRemove; 955 continue; 956 } 957 /* adjusted visual index is within current run */ 958 if (insertRemove == 0) { 959 visualIndex += controlFound; 960 break; 961 } 962 /* count non-control chars until visualIndex */ 963 logicalStart = runs[i].start; 964 evenRun = runs[i].isEvenRun(); 965 logicalEnd = logicalStart + length - 1; 966 for (j = 0; j < length; j++) { 967 k= evenRun ? logicalStart+j : logicalEnd-j; 968 uchar = bidi.text[k]; 969 if (Bidi.IsBidiControlChar(uchar)) { 970 controlFound++; 971 } 972 if ((visualIndex + controlFound) == (visualStart + j)) { 973 break; 974 } 975 } 976 visualIndex += controlFound; 977 break; 978 } 979 } 980 /* handle all cases */ 981 if (runCount <= 10) { 982 /* linear search for the run */ 983 for (i = 0; visualIndex >= runs[i].limit; ++i) {} 984 } else { 985 /* binary search for the run */ 986 int begin = 0, limit = runCount; 987 988 /* the middle if() is guaranteed to find the run, we don't need a loop limit */ 989 for ( ; ; ) { 990 i = (begin + limit) >>> 1; 991 if (visualIndex >= runs[i].limit) { 992 begin = i + 1; 993 } else if (i==0 || visualIndex >= runs[i-1].limit) { 994 break; 995 } else { 996 limit = i; 997 } 998 } 999 } 1000 1001 start= runs[i].start; 1002 if (runs[i].isEvenRun()) { 1003 /* LTR */ 1004 /* the offset in runs[i] is visualIndex-runs[i-1].visualLimit */ 1005 if (i > 0) { 1006 visualIndex -= runs[i - 1].limit; 1007 } 1008 return start + visualIndex; 1009 } else { 1010 /* RTL */ 1011 return start + runs[i].limit - visualIndex - 1; 1012 } 1013 } 1014 1015 static int[] getLogicalMap(Bidi bidi) 1016 { 1017 /* fill a logical-to-visual index map using the runs[] */ 1018 BidiRun[] runs = bidi.runs; 1019 int logicalStart, visualStart, logicalLimit, visualLimit; 1020 int[] indexMap = new int[bidi.length]; 1021 if (bidi.length > bidi.resultLength) { 1022 Arrays.fill(indexMap, Bidi.MAP_NOWHERE); 1023 } 1024 1025 visualStart = 0; 1026 for (int j = 0; j < bidi.runCount; ++j) { 1027 logicalStart = runs[j].start; 1028 visualLimit = runs[j].limit; 1029 if (runs[j].isEvenRun()) { 1030 do { /* LTR */ 1031 indexMap[logicalStart++] = visualStart++; 1032 } while (visualStart < visualLimit); 1033 } else { 1034 logicalStart += visualLimit - visualStart; /* logicalLimit */ 1035 do { /* RTL */ 1036 indexMap[--logicalStart] = visualStart++; 1037 } while (visualStart < visualLimit); 1038 } 1039 /* visualStart==visualLimit; */ 1040 } 1041 1042 if (bidi.insertPoints.size > 0) { 1043 int markFound = 0, runCount = bidi.runCount; 1044 int length, insertRemove, i, j; 1045 runs = bidi.runs; 1046 visualStart = 0; 1047 /* add number of marks found until each index */ 1048 for (i = 0; i < runCount; i++, visualStart += length) { 1049 length = runs[i].limit - visualStart; 1050 insertRemove = runs[i].insertRemove; 1051 if ((insertRemove & (Bidi.LRM_BEFORE|Bidi.RLM_BEFORE)) > 0) { 1052 markFound++; 1053 } 1054 if (markFound > 0) { 1055 logicalStart = runs[i].start; 1056 logicalLimit = logicalStart + length; 1057 for (j = logicalStart; j < logicalLimit; j++) { 1058 indexMap[j] += markFound; 1059 } 1060 } 1061 if ((insertRemove & (Bidi.LRM_AFTER|Bidi.RLM_AFTER)) > 0) { 1062 markFound++; 1063 } 1064 } 1065 } 1066 else if (bidi.controlCount > 0) { 1067 int controlFound = 0, runCount = bidi.runCount; 1068 int length, insertRemove, i, j, k; 1069 boolean evenRun; 1070 char uchar; 1071 runs = bidi.runs; 1072 visualStart = 0; 1073 /* subtract number of controls found until each index */ 1074 for (i = 0; i < runCount; i++, visualStart += length) { 1075 length = runs[i].limit - visualStart; 1076 insertRemove = runs[i].insertRemove; 1077 /* no control found within previous runs nor within this run */ 1078 if ((controlFound - insertRemove) == 0) { 1079 continue; 1080 } 1081 logicalStart = runs[i].start; 1082 evenRun = runs[i].isEvenRun(); 1083 logicalLimit = logicalStart + length; 1084 /* if no control within this run */ 1085 if (insertRemove == 0) { 1086 for (j = logicalStart; j < logicalLimit; j++) { 1087 indexMap[j] -= controlFound; 1088 } 1089 continue; 1090 } 1091 for (j = 0; j < length; j++) { 1092 k = evenRun ? logicalStart + j : logicalLimit - j - 1; 1093 uchar = bidi.text[k]; 1094 if (Bidi.IsBidiControlChar(uchar)) { 1095 controlFound++; 1096 indexMap[k] = Bidi.MAP_NOWHERE; 1097 continue; 1098 } 1099 indexMap[k] -= controlFound; 1100 } 1101 } 1102 } 1103 return indexMap; 1104 } 1105 1106 static int[] getVisualMap(Bidi bidi) 1107 { 1108 /* fill a visual-to-logical index map using the runs[] */ 1109 BidiRun[] runs = bidi.runs; 1110 int logicalStart, visualStart, visualLimit; 1111 int allocLength = bidi.length > bidi.resultLength ? bidi.length 1112 : bidi.resultLength; 1113 int[] indexMap = new int[allocLength]; 1114 1115 visualStart = 0; 1116 int idx = 0; 1117 for (int j = 0; j < bidi.runCount; ++j) { 1118 logicalStart = runs[j].start; 1119 visualLimit = runs[j].limit; 1120 if (runs[j].isEvenRun()) { 1121 do { /* LTR */ 1122 indexMap[idx++] = logicalStart++; 1123 } while (++visualStart < visualLimit); 1124 } else { 1125 logicalStart += visualLimit - visualStart; /* logicalLimit */ 1126 do { /* RTL */ 1127 indexMap[idx++] = --logicalStart; 1128 } while (++visualStart < visualLimit); 1129 } 1130 /* visualStart==visualLimit; */ 1131 } 1132 1133 if (bidi.insertPoints.size > 0) { 1134 int markFound = 0, runCount = bidi.runCount; 1135 int insertRemove, i, j, k; 1136 runs = bidi.runs; 1137 /* count all inserted marks */ 1138 for (i = 0; i < runCount; i++) { 1139 insertRemove = runs[i].insertRemove; 1140 if ((insertRemove & (Bidi.LRM_BEFORE|Bidi.RLM_BEFORE)) > 0) { 1141 markFound++; 1142 } 1143 if ((insertRemove & (Bidi.LRM_AFTER|Bidi.RLM_AFTER)) > 0) { 1144 markFound++; 1145 } 1146 } 1147 /* move back indexes by number of preceding marks */ 1148 k = bidi.resultLength; 1149 for (i = runCount - 1; i >= 0 && markFound > 0; i--) { 1150 insertRemove = runs[i].insertRemove; 1151 if ((insertRemove & (Bidi.LRM_AFTER|Bidi.RLM_AFTER)) > 0) { 1152 indexMap[--k] = Bidi.MAP_NOWHERE; 1153 markFound--; 1154 } 1155 visualStart = i > 0 ? runs[i-1].limit : 0; 1156 for (j = runs[i].limit - 1; j >= visualStart && markFound > 0; j--) { 1157 indexMap[--k] = indexMap[j]; 1158 } 1159 if ((insertRemove & (Bidi.LRM_BEFORE|Bidi.RLM_BEFORE)) > 0) { 1160 indexMap[--k] = Bidi.MAP_NOWHERE; 1161 markFound--; 1162 } 1163 } 1164 } 1165 else if (bidi.controlCount > 0) { 1166 int runCount = bidi.runCount, logicalEnd; 1167 int insertRemove, length, i, j, k, m; 1168 char uchar; 1169 boolean evenRun; 1170 runs = bidi.runs; 1171 visualStart = 0; 1172 /* move forward indexes by number of preceding controls */ 1173 k = 0; 1174 for (i = 0; i < runCount; i++, visualStart += length) { 1175 length = runs[i].limit - visualStart; 1176 insertRemove = runs[i].insertRemove; 1177 /* if no control found yet, nothing to do in this run */ 1178 if ((insertRemove == 0) && (k == visualStart)) { 1179 k += length; 1180 continue; 1181 } 1182 /* if no control in this run */ 1183 if (insertRemove == 0) { 1184 visualLimit = runs[i].limit; 1185 for (j = visualStart; j < visualLimit; j++) { 1186 indexMap[k++] = indexMap[j]; 1187 } 1188 continue; 1189 } 1190 logicalStart = runs[i].start; 1191 evenRun = runs[i].isEvenRun(); 1192 logicalEnd = logicalStart + length - 1; 1193 for (j = 0; j < length; j++) { 1194 m = evenRun ? logicalStart + j : logicalEnd - j; 1195 uchar = bidi.text[m]; 1196 if (!Bidi.IsBidiControlChar(uchar)) { 1197 indexMap[k++] = m; 1198 } 1199 } 1200 } 1201 } 1202 if (allocLength == bidi.resultLength) { 1203 return indexMap; 1204 } 1205 int[] newMap = new int[bidi.resultLength]; 1206 System.arraycopy(indexMap, 0, newMap, 0, bidi.resultLength); 1207 return newMap; 1208 } 1209 1210 static int[] invertMap(int[] srcMap) 1211 { 1212 int srcLength = srcMap.length; 1213 int destLength = -1, count = 0, i, srcEntry; 1214 1215 /* find highest value and count positive indexes in srcMap */ 1216 for (i = 0; i < srcLength; i++) { 1217 srcEntry = srcMap[i]; 1218 if (srcEntry > destLength) { 1219 destLength = srcEntry; 1220 } 1221 if (srcEntry >= 0) { 1222 count++; 1223 } 1224 } 1225 destLength++; /* add 1 for origin 0 */ 1226 int[] destMap = new int[destLength]; 1227 if (count < destLength) { 1228 /* we must fill unmatched destMap entries with -1 */ 1229 Arrays.fill(destMap, Bidi.MAP_NOWHERE); 1230 } 1231 for (i = 0; i < srcLength; i++) { 1232 srcEntry = srcMap[i]; 1233 if (srcEntry >= 0) { 1234 destMap[srcEntry] = i; 1235 } 1236 } 1237 return destMap; 1238 } 1239} 1240