1/* 2 * Copyright (C) 2013 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17package android.text; 18 19import android.view.View; 20 21import static android.text.TextDirectionHeuristics.FIRSTSTRONG_LTR; 22 23import java.util.Locale; 24 25/** 26 * Utility class for formatting text for display in a potentially opposite-directionality context 27 * without garbling. The directionality of the context is set at formatter creation and the 28 * directionality of the text can be either estimated or passed in when known. 29 * 30 * <p>To support versions lower than {@link android.os.Build.VERSION_CODES#JELLY_BEAN_MR2}, 31 * you can use the support library's {@link android.support.v4.text.BidiFormatter} class. 32 * 33 * <p>These APIs provides the following functionality: 34 * <p> 35 * 1. Bidi Wrapping 36 * When text in one language is mixed into a document in another, opposite-directionality language, 37 * e.g. when an English business name is embedded in some Hebrew text, both the inserted string 38 * and the text surrounding it may be displayed incorrectly unless the inserted string is explicitly 39 * separated from the surrounding text in a "wrapper" that: 40 * <p> 41 * - Declares its directionality so that the string is displayed correctly. This can be done in 42 * Unicode bidi formatting codes by {@link #unicodeWrap} and similar methods. 43 * <p> 44 * - Isolates the string's directionality, so it does not unduly affect the surrounding content. 45 * Currently, this can only be done using invisible Unicode characters of the same direction as 46 * the context (LRM or RLM) in addition to the directionality declaration above, thus "resetting" 47 * the directionality to that of the context. The "reset" may need to be done at both ends of the 48 * string. Without "reset" after the string, the string will "stick" to a number or logically 49 * separate opposite-direction text that happens to follow it in-line (even if separated by 50 * neutral content like spaces and punctuation). Without "reset" before the string, the same can 51 * happen there, but only with more opposite-direction text, not a number. One approach is to 52 * "reset" the direction only after each string, on the theory that if the preceding opposite- 53 * direction text is itself bidi-wrapped, the "reset" after it will prevent the sticking. (Doing 54 * the "reset" only before each string definitely does not work because we do not want to require 55 * bidi-wrapping numbers, and a bidi-wrapped opposite-direction string could be followed by a 56 * number.) Still, the safest policy is to do the "reset" on both ends of each string, since RTL 57 * message translations often contain untranslated Latin-script brand names and technical terms, 58 * and one of these can be followed by a bidi-wrapped inserted value. On the other hand, when one 59 * has such a message, it is best to do the "reset" manually in the message translation itself, 60 * since the message's opposite-direction text could be followed by an inserted number, which we 61 * would not bidi-wrap anyway. Thus, "reset" only after the string is the current default. In an 62 * alternative to "reset", recent additions to the HTML, CSS, and Unicode standards allow the 63 * isolation to be part of the directionality declaration. This form of isolation is better than 64 * "reset" because it takes less space, does not require knowing the context directionality, has a 65 * gentler effect than "reset", and protects both ends of the string. However, we do not yet allow 66 * using it because required platforms do not yet support it. 67 * <p> 68 * Providing these wrapping services is the basic purpose of the bidi formatter. 69 * <p> 70 * 2. Directionality estimation 71 * How does one know whether a string about to be inserted into surrounding text has the same 72 * directionality? Well, in many cases, one knows that this must be the case when writing the code 73 * doing the insertion, e.g. when a localized message is inserted into a localized page. In such 74 * cases there is no need to involve the bidi formatter at all. In some other cases, it need not be 75 * the same as the context, but is either constant (e.g. urls are always LTR) or otherwise known. 76 * In the remaining cases, e.g. when the string is user-entered or comes from a database, the 77 * language of the string (and thus its directionality) is not known a priori, and must be 78 * estimated at run-time. The bidi formatter can do this automatically using the default 79 * first-strong estimation algorithm. It can also be configured to use a custom directionality 80 * estimation object. 81 */ 82public final class BidiFormatter { 83 84 /** 85 * The default text direction heuristic. 86 */ 87 private static TextDirectionHeuristic DEFAULT_TEXT_DIRECTION_HEURISTIC = FIRSTSTRONG_LTR; 88 89 /** 90 * Unicode "Left-To-Right Embedding" (LRE) character. 91 */ 92 private static final char LRE = '\u202A'; 93 94 /** 95 * Unicode "Right-To-Left Embedding" (RLE) character. 96 */ 97 private static final char RLE = '\u202B'; 98 99 /** 100 * Unicode "Pop Directional Formatting" (PDF) character. 101 */ 102 private static final char PDF = '\u202C'; 103 104 /** 105 * Unicode "Left-To-Right Mark" (LRM) character. 106 */ 107 private static final char LRM = '\u200E'; 108 109 /* 110 * Unicode "Right-To-Left Mark" (RLM) character. 111 */ 112 private static final char RLM = '\u200F'; 113 114 /* 115 * String representation of LRM 116 */ 117 private static final String LRM_STRING = Character.toString(LRM); 118 119 /* 120 * String representation of RLM 121 */ 122 private static final String RLM_STRING = Character.toString(RLM); 123 124 /** 125 * Empty string constant. 126 */ 127 private static final String EMPTY_STRING = ""; 128 129 /** 130 * A class for building a BidiFormatter with non-default options. 131 */ 132 public static final class Builder { 133 private boolean mIsRtlContext; 134 private int mFlags; 135 private TextDirectionHeuristic mTextDirectionHeuristic; 136 137 /** 138 * Constructor. 139 * 140 */ 141 public Builder() { 142 initialize(isRtlLocale(Locale.getDefault())); 143 } 144 145 /** 146 * Constructor. 147 * 148 * @param rtlContext Whether the context directionality is RTL. 149 */ 150 public Builder(boolean rtlContext) { 151 initialize(rtlContext); 152 } 153 154 /** 155 * Constructor. 156 * 157 * @param locale The context locale. 158 */ 159 public Builder(Locale locale) { 160 initialize(isRtlLocale(locale)); 161 } 162 163 /** 164 * Initializes the builder with the given context directionality and default options. 165 * 166 * @param isRtlContext Whether the context is RTL or not. 167 */ 168 private void initialize(boolean isRtlContext) { 169 mIsRtlContext = isRtlContext; 170 mTextDirectionHeuristic = DEFAULT_TEXT_DIRECTION_HEURISTIC; 171 mFlags = DEFAULT_FLAGS; 172 } 173 174 /** 175 * Specifies whether the BidiFormatter to be built should also "reset" directionality before 176 * a string being bidi-wrapped, not just after it. The default is false. 177 */ 178 public Builder stereoReset(boolean stereoReset) { 179 if (stereoReset) { 180 mFlags |= FLAG_STEREO_RESET; 181 } else { 182 mFlags &= ~FLAG_STEREO_RESET; 183 } 184 return this; 185 } 186 187 /** 188 * Specifies the default directionality estimation algorithm to be used by the BidiFormatter. 189 * By default, uses the first-strong heuristic. 190 * 191 * @param heuristic the {@code TextDirectionHeuristic} to use. 192 * @return the builder itself. 193 */ 194 public Builder setTextDirectionHeuristic(TextDirectionHeuristic heuristic) { 195 mTextDirectionHeuristic = heuristic; 196 return this; 197 } 198 199 private static BidiFormatter getDefaultInstanceFromContext(boolean isRtlContext) { 200 return isRtlContext ? DEFAULT_RTL_INSTANCE : DEFAULT_LTR_INSTANCE; 201 } 202 203 /** 204 * @return A BidiFormatter with the specified options. 205 */ 206 public BidiFormatter build() { 207 if (mFlags == DEFAULT_FLAGS && 208 mTextDirectionHeuristic == DEFAULT_TEXT_DIRECTION_HEURISTIC) { 209 return getDefaultInstanceFromContext(mIsRtlContext); 210 } 211 return new BidiFormatter(mIsRtlContext, mFlags, mTextDirectionHeuristic); 212 } 213 } 214 215 // 216 private static final int FLAG_STEREO_RESET = 2; 217 private static final int DEFAULT_FLAGS = FLAG_STEREO_RESET; 218 219 private static final BidiFormatter DEFAULT_LTR_INSTANCE = new BidiFormatter( 220 false /* LTR context */, 221 DEFAULT_FLAGS, 222 DEFAULT_TEXT_DIRECTION_HEURISTIC); 223 224 private static final BidiFormatter DEFAULT_RTL_INSTANCE = new BidiFormatter( 225 true /* RTL context */, 226 DEFAULT_FLAGS, 227 DEFAULT_TEXT_DIRECTION_HEURISTIC); 228 229 private final boolean mIsRtlContext; 230 private final int mFlags; 231 private final TextDirectionHeuristic mDefaultTextDirectionHeuristic; 232 233 /** 234 * Factory for creating an instance of BidiFormatter for the default locale directionality. 235 * 236 */ 237 public static BidiFormatter getInstance() { 238 return new Builder().build(); 239 } 240 241 /** 242 * Factory for creating an instance of BidiFormatter given the context directionality. 243 * 244 * @param rtlContext Whether the context directionality is RTL. 245 */ 246 public static BidiFormatter getInstance(boolean rtlContext) { 247 return new Builder(rtlContext).build(); 248 } 249 250 /** 251 * Factory for creating an instance of BidiFormatter given the context locale. 252 * 253 * @param locale The context locale. 254 */ 255 public static BidiFormatter getInstance(Locale locale) { 256 return new Builder(locale).build(); 257 } 258 259 /** 260 * @param isRtlContext Whether the context directionality is RTL or not. 261 * @param flags The option flags. 262 * @param heuristic The default text direction heuristic. 263 */ 264 private BidiFormatter(boolean isRtlContext, int flags, TextDirectionHeuristic heuristic) { 265 mIsRtlContext = isRtlContext; 266 mFlags = flags; 267 mDefaultTextDirectionHeuristic = heuristic; 268 } 269 270 /** 271 * @return Whether the context directionality is RTL 272 */ 273 public boolean isRtlContext() { 274 return mIsRtlContext; 275 } 276 277 /** 278 * @return Whether directionality "reset" should also be done before a string being 279 * bidi-wrapped, not just after it. 280 */ 281 public boolean getStereoReset() { 282 return (mFlags & FLAG_STEREO_RESET) != 0; 283 } 284 285 /** 286 * Returns a Unicode bidi mark matching the context directionality (LRM or RLM) if either the 287 * overall or the exit directionality of a given string is opposite to the context directionality. 288 * Putting this after the string (including its directionality declaration wrapping) prevents it 289 * from "sticking" to other opposite-directionality text or a number appearing after it inline 290 * with only neutral content in between. Otherwise returns the empty string. While the exit 291 * directionality is determined by scanning the end of the string, the overall directionality is 292 * given explicitly by a heuristic to estimate the {@code str}'s directionality. 293 * 294 * @param str String after which the mark may need to appear. 295 * @param heuristic The text direction heuristic that will be used to estimate the {@code str}'s 296 * directionality. 297 * @return LRM for RTL text in LTR context; RLM for LTR text in RTL context; 298 * else, the empty string. 299 * 300 * @hide 301 */ 302 public String markAfter(String str, TextDirectionHeuristic heuristic) { 303 final boolean isRtl = heuristic.isRtl(str, 0, str.length()); 304 // getExitDir() is called only if needed (short-circuit). 305 if (!mIsRtlContext && (isRtl || getExitDir(str) == DIR_RTL)) { 306 return LRM_STRING; 307 } 308 if (mIsRtlContext && (!isRtl || getExitDir(str) == DIR_LTR)) { 309 return RLM_STRING; 310 } 311 return EMPTY_STRING; 312 } 313 314 /** 315 * Returns a Unicode bidi mark matching the context directionality (LRM or RLM) if either the 316 * overall or the entry directionality of a given string is opposite to the context 317 * directionality. Putting this before the string (including its directionality declaration 318 * wrapping) prevents it from "sticking" to other opposite-directionality text appearing before 319 * it inline with only neutral content in between. Otherwise returns the empty string. While the 320 * entry directionality is determined by scanning the beginning of the string, the overall 321 * directionality is given explicitly by a heuristic to estimate the {@code str}'s directionality. 322 * 323 * @param str String before which the mark may need to appear. 324 * @param heuristic The text direction heuristic that will be used to estimate the {@code str}'s 325 * directionality. 326 * @return LRM for RTL text in LTR context; RLM for LTR text in RTL context; 327 * else, the empty string. 328 * 329 * @hide 330 */ 331 public String markBefore(String str, TextDirectionHeuristic heuristic) { 332 final boolean isRtl = heuristic.isRtl(str, 0, str.length()); 333 // getEntryDir() is called only if needed (short-circuit). 334 if (!mIsRtlContext && (isRtl || getEntryDir(str) == DIR_RTL)) { 335 return LRM_STRING; 336 } 337 if (mIsRtlContext && (!isRtl || getEntryDir(str) == DIR_LTR)) { 338 return RLM_STRING; 339 } 340 return EMPTY_STRING; 341 } 342 343 /** 344 * Estimates the directionality of a string using the default text direction heuristic. 345 * 346 * @param str String whose directionality is to be estimated. 347 * @return true if {@code str}'s estimated overall directionality is RTL. Otherwise returns 348 * false. 349 */ 350 public boolean isRtl(String str) { 351 return mDefaultTextDirectionHeuristic.isRtl(str, 0, str.length()); 352 } 353 354 /** 355 * Formats a string of given directionality for use in plain-text output of the context 356 * directionality, so an opposite-directionality string is neither garbled nor garbles its 357 * surroundings. This makes use of Unicode bidi formatting characters. 358 * <p> 359 * The algorithm: In case the given directionality doesn't match the context directionality, wraps 360 * the string with Unicode bidi formatting characters: RLE+{@code str}+PDF for RTL text, or 361 * LRE+{@code str}+PDF for LTR text. 362 * <p> 363 * If {@code isolate}, directionally isolates the string so that it does not garble its 364 * surroundings. Currently, this is done by "resetting" the directionality after the string by 365 * appending a trailing Unicode bidi mark matching the context directionality (LRM or RLM) when 366 * either the overall directionality or the exit directionality of the string is opposite to that 367 * of the context. If the formatter was built using {@link Builder#stereoReset(boolean)} and 368 * passing "true" as an argument, also prepends a Unicode bidi mark matching the context 369 * directionality when either the overall directionality or the entry directionality of the 370 * string is opposite to that of the context. Note that as opposed to the overall 371 * directionality, the entry and exit directionalities are determined from the string itself. 372 * <p> 373 * Does *not* do HTML-escaping. 374 * 375 * @param str The input string. 376 * @param heuristic The algorithm to be used to estimate the string's overall direction. 377 * See {@link TextDirectionHeuristics} for pre-defined heuristics. 378 * @param isolate Whether to directionally isolate the string to prevent it from garbling the 379 * content around it 380 * @return Input string after applying the above processing. 381 */ 382 public String unicodeWrap(String str, TextDirectionHeuristic heuristic, boolean isolate) { 383 final boolean isRtl = heuristic.isRtl(str, 0, str.length()); 384 StringBuilder result = new StringBuilder(); 385 if (getStereoReset() && isolate) { 386 result.append(markBefore(str, 387 isRtl ? TextDirectionHeuristics.RTL : TextDirectionHeuristics.LTR)); 388 } 389 if (isRtl != mIsRtlContext) { 390 result.append(isRtl ? RLE : LRE); 391 result.append(str); 392 result.append(PDF); 393 } else { 394 result.append(str); 395 } 396 if (isolate) { 397 result.append(markAfter(str, 398 isRtl ? TextDirectionHeuristics.RTL : TextDirectionHeuristics.LTR)); 399 } 400 return result.toString(); 401 } 402 403 /** 404 * Operates like {@link #unicodeWrap(String, TextDirectionHeuristic, boolean)}, but assumes 405 * {@code isolate} is true. 406 * 407 * @param str The input string. 408 * @param heuristic The algorithm to be used to estimate the string's overall direction. 409 * See {@link TextDirectionHeuristics} for pre-defined heuristics. 410 * @return Input string after applying the above processing. 411 */ 412 public String unicodeWrap(String str, TextDirectionHeuristic heuristic) { 413 return unicodeWrap(str, heuristic, true /* isolate */); 414 } 415 416 /** 417 * Operates like {@link #unicodeWrap(String, TextDirectionHeuristic, boolean)}, but uses the 418 * formatter's default direction estimation algorithm. 419 * 420 * @param str The input string. 421 * @param isolate Whether to directionally isolate the string to prevent it from garbling the 422 * content around it 423 * @return Input string after applying the above processing. 424 */ 425 public String unicodeWrap(String str, boolean isolate) { 426 return unicodeWrap(str, mDefaultTextDirectionHeuristic, isolate); 427 } 428 429 /** 430 * Operates like {@link #unicodeWrap(String, TextDirectionHeuristic, boolean)}, but uses the 431 * formatter's default direction estimation algorithm and assumes {@code isolate} is true. 432 * 433 * @param str The input string. 434 * @return Input string after applying the above processing. 435 */ 436 public String unicodeWrap(String str) { 437 return unicodeWrap(str, mDefaultTextDirectionHeuristic, true /* isolate */); 438 } 439 440 /** 441 * Helper method to return true if the Locale directionality is RTL. 442 * 443 * @param locale The Locale whose directionality will be checked to be RTL or LTR 444 * @return true if the {@code locale} directionality is RTL. False otherwise. 445 */ 446 private static boolean isRtlLocale(Locale locale) { 447 return (TextUtils.getLayoutDirectionFromLocale(locale) == View.LAYOUT_DIRECTION_RTL); 448 } 449 450 /** 451 * Enum for directionality type. 452 */ 453 private static final int DIR_LTR = -1; 454 private static final int DIR_UNKNOWN = 0; 455 private static final int DIR_RTL = +1; 456 457 /** 458 * Returns the directionality of the last character with strong directionality in the string, or 459 * DIR_UNKNOWN if none was encountered. For efficiency, actually scans backwards from the end of 460 * the string. Treats a non-BN character between an LRE/RLE/LRO/RLO and its matching PDF as a 461 * strong character, LTR after LRE/LRO, and RTL after RLE/RLO. The results are undefined for a 462 * string containing unbalanced LRE/RLE/LRO/RLO/PDF characters. The intended use is to check 463 * whether a logically separate item that starts with a number or a character of the string's 464 * exit directionality and follows this string inline (not counting any neutral characters in 465 * between) would "stick" to it in an opposite-directionality context, thus being displayed in 466 * an incorrect position. An LRM or RLM character (the one of the context's directionality) 467 * between the two will prevent such sticking. 468 * 469 * @param str the string to check. 470 */ 471 private static int getExitDir(String str) { 472 return new DirectionalityEstimator(str, false /* isHtml */).getExitDir(); 473 } 474 475 /** 476 * Returns the directionality of the first character with strong directionality in the string, 477 * or DIR_UNKNOWN if none was encountered. Treats a non-BN character between an 478 * LRE/RLE/LRO/RLO and its matching PDF as a strong character, LTR after LRE/LRO, and RTL after 479 * RLE/RLO. The results are undefined for a string containing unbalanced LRE/RLE/LRO/RLO/PDF 480 * characters. The intended use is to check whether a logically separate item that ends with a 481 * character of the string's entry directionality and precedes the string inline (not counting 482 * any neutral characters in between) would "stick" to it in an opposite-directionality context, 483 * thus being displayed in an incorrect position. An LRM or RLM character (the one of the 484 * context's directionality) between the two will prevent such sticking. 485 * 486 * @param str the string to check. 487 */ 488 private static int getEntryDir(String str) { 489 return new DirectionalityEstimator(str, false /* isHtml */).getEntryDir(); 490 } 491 492 /** 493 * An object that estimates the directionality of a given string by various methods. 494 * 495 */ 496 private static class DirectionalityEstimator { 497 498 // Internal static variables and constants. 499 500 /** 501 * Size of the bidi character class cache. The results of the Character.getDirectionality() 502 * calls on the lowest DIR_TYPE_CACHE_SIZE codepoints are kept in an array for speed. 503 * The 0x700 value is designed to leave all the European and Near Eastern languages in the 504 * cache. It can be reduced to 0x180, restricting the cache to the Western European 505 * languages. 506 */ 507 private static final int DIR_TYPE_CACHE_SIZE = 0x700; 508 509 /** 510 * The bidi character class cache. 511 */ 512 private static final byte DIR_TYPE_CACHE[]; 513 514 static { 515 DIR_TYPE_CACHE = new byte[DIR_TYPE_CACHE_SIZE]; 516 for (int i = 0; i < DIR_TYPE_CACHE_SIZE; i++) { 517 DIR_TYPE_CACHE[i] = Character.getDirectionality(i); 518 } 519 } 520 521 // Internal instance variables. 522 523 /** 524 * The text to be scanned. 525 */ 526 private final String text; 527 528 /** 529 * Whether the text to be scanned is to be treated as HTML, i.e. skipping over tags and 530 * entities when looking for the next / preceding dir type. 531 */ 532 private final boolean isHtml; 533 534 /** 535 * The length of the text in chars. 536 */ 537 private final int length; 538 539 /** 540 * The current position in the text. 541 */ 542 private int charIndex; 543 544 /** 545 * The char encountered by the last dirTypeForward or dirTypeBackward call. If it 546 * encountered a supplementary codepoint, this contains a char that is not a valid 547 * codepoint. This is ok, because this member is only used to detect some well-known ASCII 548 * syntax, e.g. "http://" and the beginning of an HTML tag or entity. 549 */ 550 private char lastChar; 551 552 /** 553 * Constructor. 554 * 555 * @param text The string to scan. 556 * @param isHtml Whether the text to be scanned is to be treated as HTML, i.e. skipping over 557 * tags and entities. 558 */ 559 DirectionalityEstimator(String text, boolean isHtml) { 560 this.text = text; 561 this.isHtml = isHtml; 562 length = text.length(); 563 } 564 565 /** 566 * Returns the directionality of the first character with strong directionality in the 567 * string, or DIR_UNKNOWN if none was encountered. Treats a non-BN character between an 568 * LRE/RLE/LRO/RLO and its matching PDF as a strong character, LTR after LRE/LRO, and RTL 569 * after RLE/RLO. The results are undefined for a string containing unbalanced 570 * LRE/RLE/LRO/RLO/PDF characters. 571 */ 572 int getEntryDir() { 573 // The reason for this method name, as opposed to getFirstStrongDir(), is that 574 // "first strong" is a commonly used description of Unicode's estimation algorithm, 575 // but the two must treat formatting characters quite differently. Thus, we are staying 576 // away from both "first" and "last" in these method names to avoid confusion. 577 charIndex = 0; 578 int embeddingLevel = 0; 579 int embeddingLevelDir = DIR_UNKNOWN; 580 int firstNonEmptyEmbeddingLevel = 0; 581 while (charIndex < length && firstNonEmptyEmbeddingLevel == 0) { 582 switch (dirTypeForward()) { 583 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_EMBEDDING: 584 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_OVERRIDE: 585 ++embeddingLevel; 586 embeddingLevelDir = DIR_LTR; 587 break; 588 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_EMBEDDING: 589 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_OVERRIDE: 590 ++embeddingLevel; 591 embeddingLevelDir = DIR_RTL; 592 break; 593 case Character.DIRECTIONALITY_POP_DIRECTIONAL_FORMAT: 594 --embeddingLevel; 595 // To restore embeddingLevelDir to its previous value, we would need a 596 // stack, which we want to avoid. Thus, at this point we do not know the 597 // current embedding's directionality. 598 embeddingLevelDir = DIR_UNKNOWN; 599 break; 600 case Character.DIRECTIONALITY_BOUNDARY_NEUTRAL: 601 break; 602 case Character.DIRECTIONALITY_LEFT_TO_RIGHT: 603 if (embeddingLevel == 0) { 604 return DIR_LTR; 605 } 606 firstNonEmptyEmbeddingLevel = embeddingLevel; 607 break; 608 case Character.DIRECTIONALITY_RIGHT_TO_LEFT: 609 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_ARABIC: 610 if (embeddingLevel == 0) { 611 return DIR_RTL; 612 } 613 firstNonEmptyEmbeddingLevel = embeddingLevel; 614 break; 615 default: 616 firstNonEmptyEmbeddingLevel = embeddingLevel; 617 break; 618 } 619 } 620 621 // We have either found a non-empty embedding or scanned the entire string finding 622 // neither a non-empty embedding nor a strong character outside of an embedding. 623 if (firstNonEmptyEmbeddingLevel == 0) { 624 // We have not found a non-empty embedding. Thus, the string contains neither a 625 // non-empty embedding nor a strong character outside of an embedding. 626 return DIR_UNKNOWN; 627 } 628 629 // We have found a non-empty embedding. 630 if (embeddingLevelDir != DIR_UNKNOWN) { 631 // We know the directionality of the non-empty embedding. 632 return embeddingLevelDir; 633 } 634 635 // We do not remember the directionality of the non-empty embedding we found. So, we go 636 // backwards to find the start of the non-empty embedding and get its directionality. 637 while (charIndex > 0) { 638 switch (dirTypeBackward()) { 639 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_EMBEDDING: 640 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_OVERRIDE: 641 if (firstNonEmptyEmbeddingLevel == embeddingLevel) { 642 return DIR_LTR; 643 } 644 --embeddingLevel; 645 break; 646 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_EMBEDDING: 647 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_OVERRIDE: 648 if (firstNonEmptyEmbeddingLevel == embeddingLevel) { 649 return DIR_RTL; 650 } 651 --embeddingLevel; 652 break; 653 case Character.DIRECTIONALITY_POP_DIRECTIONAL_FORMAT: 654 ++embeddingLevel; 655 break; 656 } 657 } 658 // We should never get here. 659 return DIR_UNKNOWN; 660 } 661 662 /** 663 * Returns the directionality of the last character with strong directionality in the 664 * string, or DIR_UNKNOWN if none was encountered. For efficiency, actually scans backwards 665 * from the end of the string. Treats a non-BN character between an LRE/RLE/LRO/RLO and its 666 * matching PDF as a strong character, LTR after LRE/LRO, and RTL after RLE/RLO. The results 667 * are undefined for a string containing unbalanced LRE/RLE/LRO/RLO/PDF characters. 668 */ 669 int getExitDir() { 670 // The reason for this method name, as opposed to getLastStrongDir(), is that "last 671 // strong" sounds like the exact opposite of "first strong", which is a commonly used 672 // description of Unicode's estimation algorithm (getUnicodeDir() above), but the two 673 // must treat formatting characters quite differently. Thus, we are staying away from 674 // both "first" and "last" in these method names to avoid confusion. 675 charIndex = length; 676 int embeddingLevel = 0; 677 int lastNonEmptyEmbeddingLevel = 0; 678 while (charIndex > 0) { 679 switch (dirTypeBackward()) { 680 case Character.DIRECTIONALITY_LEFT_TO_RIGHT: 681 if (embeddingLevel == 0) { 682 return DIR_LTR; 683 } 684 if (lastNonEmptyEmbeddingLevel == 0) { 685 lastNonEmptyEmbeddingLevel = embeddingLevel; 686 } 687 break; 688 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_EMBEDDING: 689 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_OVERRIDE: 690 if (lastNonEmptyEmbeddingLevel == embeddingLevel) { 691 return DIR_LTR; 692 } 693 --embeddingLevel; 694 break; 695 case Character.DIRECTIONALITY_RIGHT_TO_LEFT: 696 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_ARABIC: 697 if (embeddingLevel == 0) { 698 return DIR_RTL; 699 } 700 if (lastNonEmptyEmbeddingLevel == 0) { 701 lastNonEmptyEmbeddingLevel = embeddingLevel; 702 } 703 break; 704 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_EMBEDDING: 705 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_OVERRIDE: 706 if (lastNonEmptyEmbeddingLevel == embeddingLevel) { 707 return DIR_RTL; 708 } 709 --embeddingLevel; 710 break; 711 case Character.DIRECTIONALITY_POP_DIRECTIONAL_FORMAT: 712 ++embeddingLevel; 713 break; 714 case Character.DIRECTIONALITY_BOUNDARY_NEUTRAL: 715 break; 716 default: 717 if (lastNonEmptyEmbeddingLevel == 0) { 718 lastNonEmptyEmbeddingLevel = embeddingLevel; 719 } 720 break; 721 } 722 } 723 return DIR_UNKNOWN; 724 } 725 726 // Internal methods 727 728 /** 729 * Gets the bidi character class, i.e. Character.getDirectionality(), of a given char, using 730 * a cache for speed. Not designed for supplementary codepoints, whose results we do not 731 * cache. 732 */ 733 private static byte getCachedDirectionality(char c) { 734 return c < DIR_TYPE_CACHE_SIZE ? DIR_TYPE_CACHE[c] : Character.getDirectionality(c); 735 } 736 737 /** 738 * Returns the Character.DIRECTIONALITY_... value of the next codepoint and advances 739 * charIndex. If isHtml, and the codepoint is '<' or '&', advances through the tag/entity, 740 * and returns Character.DIRECTIONALITY_WHITESPACE. For an entity, it would be best to 741 * figure out the actual character, and return its dirtype, but treating it as whitespace is 742 * good enough for our purposes. 743 * 744 * @throws java.lang.IndexOutOfBoundsException if called when charIndex >= length or < 0. 745 */ 746 byte dirTypeForward() { 747 lastChar = text.charAt(charIndex); 748 if (Character.isHighSurrogate(lastChar)) { 749 int codePoint = Character.codePointAt(text, charIndex); 750 charIndex += Character.charCount(codePoint); 751 return Character.getDirectionality(codePoint); 752 } 753 charIndex++; 754 byte dirType = getCachedDirectionality(lastChar); 755 if (isHtml) { 756 // Process tags and entities. 757 if (lastChar == '<') { 758 dirType = skipTagForward(); 759 } else if (lastChar == '&') { 760 dirType = skipEntityForward(); 761 } 762 } 763 return dirType; 764 } 765 766 /** 767 * Returns the Character.DIRECTIONALITY_... value of the preceding codepoint and advances 768 * charIndex backwards. If isHtml, and the codepoint is the end of a complete HTML tag or 769 * entity, advances over the whole tag/entity and returns 770 * Character.DIRECTIONALITY_WHITESPACE. For an entity, it would be best to figure out the 771 * actual character, and return its dirtype, but treating it as whitespace is good enough 772 * for our purposes. 773 * 774 * @throws java.lang.IndexOutOfBoundsException if called when charIndex > length or <= 0. 775 */ 776 byte dirTypeBackward() { 777 lastChar = text.charAt(charIndex - 1); 778 if (Character.isLowSurrogate(lastChar)) { 779 int codePoint = Character.codePointBefore(text, charIndex); 780 charIndex -= Character.charCount(codePoint); 781 return Character.getDirectionality(codePoint); 782 } 783 charIndex--; 784 byte dirType = getCachedDirectionality(lastChar); 785 if (isHtml) { 786 // Process tags and entities. 787 if (lastChar == '>') { 788 dirType = skipTagBackward(); 789 } else if (lastChar == ';') { 790 dirType = skipEntityBackward(); 791 } 792 } 793 return dirType; 794 } 795 796 /** 797 * Advances charIndex forward through an HTML tag (after the opening < has already been 798 * read) and returns Character.DIRECTIONALITY_WHITESPACE. If there is no matching >, 799 * does not change charIndex and returns Character.DIRECTIONALITY_OTHER_NEUTRALS (for the 800 * < that hadn't been part of a tag after all). 801 */ 802 private byte skipTagForward() { 803 int initialCharIndex = charIndex; 804 while (charIndex < length) { 805 lastChar = text.charAt(charIndex++); 806 if (lastChar == '>') { 807 // The end of the tag. 808 return Character.DIRECTIONALITY_WHITESPACE; 809 } 810 if (lastChar == '"' || lastChar == '\'') { 811 // Skip over a quoted attribute value inside the tag. 812 char quote = lastChar; 813 while (charIndex < length && (lastChar = text.charAt(charIndex++)) != quote) {} 814 } 815 } 816 // The original '<' wasn't the start of a tag after all. 817 charIndex = initialCharIndex; 818 lastChar = '<'; 819 return Character.DIRECTIONALITY_OTHER_NEUTRALS; 820 } 821 822 /** 823 * Advances charIndex backward through an HTML tag (after the closing > has already been 824 * read) and returns Character.DIRECTIONALITY_WHITESPACE. If there is no matching <, does 825 * not change charIndex and returns Character.DIRECTIONALITY_OTHER_NEUTRALS (for the > 826 * that hadn't been part of a tag after all). Nevertheless, the running time for calling 827 * skipTagBackward() in a loop remains linear in the size of the text, even for a text like 828 * ">>>>", because skipTagBackward() also stops looking for a matching < 829 * when it encounters another >. 830 */ 831 private byte skipTagBackward() { 832 int initialCharIndex = charIndex; 833 while (charIndex > 0) { 834 lastChar = text.charAt(--charIndex); 835 if (lastChar == '<') { 836 // The start of the tag. 837 return Character.DIRECTIONALITY_WHITESPACE; 838 } 839 if (lastChar == '>') { 840 break; 841 } 842 if (lastChar == '"' || lastChar == '\'') { 843 // Skip over a quoted attribute value inside the tag. 844 char quote = lastChar; 845 while (charIndex > 0 && (lastChar = text.charAt(--charIndex)) != quote) {} 846 } 847 } 848 // The original '>' wasn't the end of a tag after all. 849 charIndex = initialCharIndex; 850 lastChar = '>'; 851 return Character.DIRECTIONALITY_OTHER_NEUTRALS; 852 } 853 854 /** 855 * Advances charIndex forward through an HTML character entity tag (after the opening 856 * & has already been read) and returns Character.DIRECTIONALITY_WHITESPACE. It would be 857 * best to figure out the actual character and return its dirtype, but this is good enough. 858 */ 859 private byte skipEntityForward() { 860 while (charIndex < length && (lastChar = text.charAt(charIndex++)) != ';') {} 861 return Character.DIRECTIONALITY_WHITESPACE; 862 } 863 864 /** 865 * Advances charIndex backward through an HTML character entity tag (after the closing ; 866 * has already been read) and returns Character.DIRECTIONALITY_WHITESPACE. It would be best 867 * to figure out the actual character and return its dirtype, but this is good enough. 868 * If there is no matching &, does not change charIndex and returns 869 * Character.DIRECTIONALITY_OTHER_NEUTRALS (for the ';' that did not start an entity after 870 * all). Nevertheless, the running time for calling skipEntityBackward() in a loop remains 871 * linear in the size of the text, even for a text like ";;;;;;;", because skipTagBackward() 872 * also stops looking for a matching & when it encounters another ;. 873 */ 874 private byte skipEntityBackward() { 875 int initialCharIndex = charIndex; 876 while (charIndex > 0) { 877 lastChar = text.charAt(--charIndex); 878 if (lastChar == '&') { 879 return Character.DIRECTIONALITY_WHITESPACE; 880 } 881 if (lastChar == ';') { 882 break; 883 } 884 } 885 charIndex = initialCharIndex; 886 lastChar = ';'; 887 return Character.DIRECTIONALITY_OTHER_NEUTRALS; 888 } 889 } 890}