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 17 package android.text; 18 19 import android.view.View; 20 21 import static android.text.TextDirectionHeuristics.FIRSTSTRONG_LTR; 22 23 import 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 */ 82 public 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 true. 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 367 * that of the context. Unless the formatter was built using 368 * {@link Builder#stereoReset(boolean)} with a {@code false} argument, also prepends a Unicode 369 * bidi mark matching the context directionality when either the overall directionality or the 370 * entry directionality of the string is opposite to that of the context. Note that as opposed 371 * to the overall directionality, the entry and exit directionalities are determined from the 372 * string itself. 373 * <p> 374 * Does *not* do HTML-escaping. 375 * 376 * @param str The input string. 377 * @param heuristic The algorithm to be used to estimate the string's overall direction. 378 * See {@link TextDirectionHeuristics} for pre-defined heuristics. 379 * @param isolate Whether to directionally isolate the string to prevent it from garbling the 380 * content around it 381 * @return Input string after applying the above processing. {@code null} if {@code str} is 382 * {@code null}. 383 */ 384 public String unicodeWrap(String str, TextDirectionHeuristic heuristic, boolean isolate) { 385 if (str == null) return null; 386 final boolean isRtl = heuristic.isRtl(str, 0, str.length()); 387 StringBuilder result = new StringBuilder(); 388 if (getStereoReset() && isolate) { 389 result.append(markBefore(str, 390 isRtl ? TextDirectionHeuristics.RTL : TextDirectionHeuristics.LTR)); 391 } 392 if (isRtl != mIsRtlContext) { 393 result.append(isRtl ? RLE : LRE); 394 result.append(str); 395 result.append(PDF); 396 } else { 397 result.append(str); 398 } 399 if (isolate) { 400 result.append(markAfter(str, 401 isRtl ? TextDirectionHeuristics.RTL : TextDirectionHeuristics.LTR)); 402 } 403 return result.toString(); 404 } 405 406 /** 407 * Operates like {@link #unicodeWrap(String, TextDirectionHeuristic, boolean)}, but assumes 408 * {@code isolate} is true. 409 * 410 * @param str The input string. 411 * @param heuristic The algorithm to be used to estimate the string's overall direction. 412 * See {@link TextDirectionHeuristics} for pre-defined heuristics. 413 * @return Input string after applying the above processing. 414 */ 415 public String unicodeWrap(String str, TextDirectionHeuristic heuristic) { 416 return unicodeWrap(str, heuristic, true /* isolate */); 417 } 418 419 /** 420 * Operates like {@link #unicodeWrap(String, TextDirectionHeuristic, boolean)}, but uses the 421 * formatter's default direction estimation algorithm. 422 * 423 * @param str The input string. 424 * @param isolate Whether to directionally isolate the string to prevent it from garbling the 425 * content around it 426 * @return Input string after applying the above processing. 427 */ 428 public String unicodeWrap(String str, boolean isolate) { 429 return unicodeWrap(str, mDefaultTextDirectionHeuristic, isolate); 430 } 431 432 /** 433 * Operates like {@link #unicodeWrap(String, TextDirectionHeuristic, boolean)}, but uses the 434 * formatter's default direction estimation algorithm and assumes {@code isolate} is true. 435 * 436 * @param str The input string. 437 * @return Input string after applying the above processing. 438 */ 439 public String unicodeWrap(String str) { 440 return unicodeWrap(str, mDefaultTextDirectionHeuristic, true /* isolate */); 441 } 442 443 /** 444 * Helper method to return true if the Locale directionality is RTL. 445 * 446 * @param locale The Locale whose directionality will be checked to be RTL or LTR 447 * @return true if the {@code locale} directionality is RTL. False otherwise. 448 */ 449 private static boolean isRtlLocale(Locale locale) { 450 return (TextUtils.getLayoutDirectionFromLocale(locale) == View.LAYOUT_DIRECTION_RTL); 451 } 452 453 /** 454 * Enum for directionality type. 455 */ 456 private static final int DIR_LTR = -1; 457 private static final int DIR_UNKNOWN = 0; 458 private static final int DIR_RTL = +1; 459 460 /** 461 * Returns the directionality of the last character with strong directionality in the string, or 462 * DIR_UNKNOWN if none was encountered. For efficiency, actually scans backwards from the end of 463 * the string. Treats a non-BN character between an LRE/RLE/LRO/RLO and its matching PDF as a 464 * strong character, LTR after LRE/LRO, and RTL after RLE/RLO. The results are undefined for a 465 * string containing unbalanced LRE/RLE/LRO/RLO/PDF characters. The intended use is to check 466 * whether a logically separate item that starts with a number or a character of the string's 467 * exit directionality and follows this string inline (not counting any neutral characters in 468 * between) would "stick" to it in an opposite-directionality context, thus being displayed in 469 * an incorrect position. An LRM or RLM character (the one of the context's directionality) 470 * between the two will prevent such sticking. 471 * 472 * @param str the string to check. 473 */ 474 private static int getExitDir(String str) { 475 return new DirectionalityEstimator(str, false /* isHtml */).getExitDir(); 476 } 477 478 /** 479 * Returns the directionality of the first character with strong directionality in the string, 480 * or DIR_UNKNOWN if none was encountered. Treats a non-BN character between an 481 * LRE/RLE/LRO/RLO and its matching PDF as a strong character, LTR after LRE/LRO, and RTL after 482 * RLE/RLO. The results are undefined for a string containing unbalanced LRE/RLE/LRO/RLO/PDF 483 * characters. The intended use is to check whether a logically separate item that ends with a 484 * character of the string's entry directionality and precedes the string inline (not counting 485 * any neutral characters in between) would "stick" to it in an opposite-directionality context, 486 * thus being displayed in an incorrect position. An LRM or RLM character (the one of the 487 * context's directionality) between the two will prevent such sticking. 488 * 489 * @param str the string to check. 490 */ 491 private static int getEntryDir(String str) { 492 return new DirectionalityEstimator(str, false /* isHtml */).getEntryDir(); 493 } 494 495 /** 496 * An object that estimates the directionality of a given string by various methods. 497 * 498 */ 499 private static class DirectionalityEstimator { 500 501 // Internal static variables and constants. 502 503 /** 504 * Size of the bidi character class cache. The results of the Character.getDirectionality() 505 * calls on the lowest DIR_TYPE_CACHE_SIZE codepoints are kept in an array for speed. 506 * The 0x700 value is designed to leave all the European and Near Eastern languages in the 507 * cache. It can be reduced to 0x180, restricting the cache to the Western European 508 * languages. 509 */ 510 private static final int DIR_TYPE_CACHE_SIZE = 0x700; 511 512 /** 513 * The bidi character class cache. 514 */ 515 private static final byte DIR_TYPE_CACHE[]; 516 517 static { 518 DIR_TYPE_CACHE = new byte[DIR_TYPE_CACHE_SIZE]; 519 for (int i = 0; i < DIR_TYPE_CACHE_SIZE; i++) { 520 DIR_TYPE_CACHE[i] = Character.getDirectionality(i); 521 } 522 } 523 524 // Internal instance variables. 525 526 /** 527 * The text to be scanned. 528 */ 529 private final String text; 530 531 /** 532 * Whether the text to be scanned is to be treated as HTML, i.e. skipping over tags and 533 * entities when looking for the next / preceding dir type. 534 */ 535 private final boolean isHtml; 536 537 /** 538 * The length of the text in chars. 539 */ 540 private final int length; 541 542 /** 543 * The current position in the text. 544 */ 545 private int charIndex; 546 547 /** 548 * The char encountered by the last dirTypeForward or dirTypeBackward call. If it 549 * encountered a supplementary codepoint, this contains a char that is not a valid 550 * codepoint. This is ok, because this member is only used to detect some well-known ASCII 551 * syntax, e.g. "http://" and the beginning of an HTML tag or entity. 552 */ 553 private char lastChar; 554 555 /** 556 * Constructor. 557 * 558 * @param text The string to scan. 559 * @param isHtml Whether the text to be scanned is to be treated as HTML, i.e. skipping over 560 * tags and entities. 561 */ 562 DirectionalityEstimator(String text, boolean isHtml) { 563 this.text = text; 564 this.isHtml = isHtml; 565 length = text.length(); 566 } 567 568 /** 569 * Returns the directionality of the first character with strong directionality in the 570 * string, or DIR_UNKNOWN if none was encountered. Treats a non-BN character between an 571 * LRE/RLE/LRO/RLO and its matching PDF as a strong character, LTR after LRE/LRO, and RTL 572 * after RLE/RLO. The results are undefined for a string containing unbalanced 573 * LRE/RLE/LRO/RLO/PDF characters. 574 */ 575 int getEntryDir() { 576 // The reason for this method name, as opposed to getFirstStrongDir(), is that 577 // "first strong" is a commonly used description of Unicode's estimation algorithm, 578 // but the two must treat formatting characters quite differently. Thus, we are staying 579 // away from both "first" and "last" in these method names to avoid confusion. 580 charIndex = 0; 581 int embeddingLevel = 0; 582 int embeddingLevelDir = DIR_UNKNOWN; 583 int firstNonEmptyEmbeddingLevel = 0; 584 while (charIndex < length && firstNonEmptyEmbeddingLevel == 0) { 585 switch (dirTypeForward()) { 586 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_EMBEDDING: 587 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_OVERRIDE: 588 ++embeddingLevel; 589 embeddingLevelDir = DIR_LTR; 590 break; 591 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_EMBEDDING: 592 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_OVERRIDE: 593 ++embeddingLevel; 594 embeddingLevelDir = DIR_RTL; 595 break; 596 case Character.DIRECTIONALITY_POP_DIRECTIONAL_FORMAT: 597 --embeddingLevel; 598 // To restore embeddingLevelDir to its previous value, we would need a 599 // stack, which we want to avoid. Thus, at this point we do not know the 600 // current embedding's directionality. 601 embeddingLevelDir = DIR_UNKNOWN; 602 break; 603 case Character.DIRECTIONALITY_BOUNDARY_NEUTRAL: 604 break; 605 case Character.DIRECTIONALITY_LEFT_TO_RIGHT: 606 if (embeddingLevel == 0) { 607 return DIR_LTR; 608 } 609 firstNonEmptyEmbeddingLevel = embeddingLevel; 610 break; 611 case Character.DIRECTIONALITY_RIGHT_TO_LEFT: 612 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_ARABIC: 613 if (embeddingLevel == 0) { 614 return DIR_RTL; 615 } 616 firstNonEmptyEmbeddingLevel = embeddingLevel; 617 break; 618 default: 619 firstNonEmptyEmbeddingLevel = embeddingLevel; 620 break; 621 } 622 } 623 624 // We have either found a non-empty embedding or scanned the entire string finding 625 // neither a non-empty embedding nor a strong character outside of an embedding. 626 if (firstNonEmptyEmbeddingLevel == 0) { 627 // We have not found a non-empty embedding. Thus, the string contains neither a 628 // non-empty embedding nor a strong character outside of an embedding. 629 return DIR_UNKNOWN; 630 } 631 632 // We have found a non-empty embedding. 633 if (embeddingLevelDir != DIR_UNKNOWN) { 634 // We know the directionality of the non-empty embedding. 635 return embeddingLevelDir; 636 } 637 638 // We do not remember the directionality of the non-empty embedding we found. So, we go 639 // backwards to find the start of the non-empty embedding and get its directionality. 640 while (charIndex > 0) { 641 switch (dirTypeBackward()) { 642 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_EMBEDDING: 643 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_OVERRIDE: 644 if (firstNonEmptyEmbeddingLevel == embeddingLevel) { 645 return DIR_LTR; 646 } 647 --embeddingLevel; 648 break; 649 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_EMBEDDING: 650 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_OVERRIDE: 651 if (firstNonEmptyEmbeddingLevel == embeddingLevel) { 652 return DIR_RTL; 653 } 654 --embeddingLevel; 655 break; 656 case Character.DIRECTIONALITY_POP_DIRECTIONAL_FORMAT: 657 ++embeddingLevel; 658 break; 659 } 660 } 661 // We should never get here. 662 return DIR_UNKNOWN; 663 } 664 665 /** 666 * Returns the directionality of the last character with strong directionality in the 667 * string, or DIR_UNKNOWN if none was encountered. For efficiency, actually scans backwards 668 * from the end of the string. Treats a non-BN character between an LRE/RLE/LRO/RLO and its 669 * matching PDF as a strong character, LTR after LRE/LRO, and RTL after RLE/RLO. The results 670 * are undefined for a string containing unbalanced LRE/RLE/LRO/RLO/PDF characters. 671 */ 672 int getExitDir() { 673 // The reason for this method name, as opposed to getLastStrongDir(), is that "last 674 // strong" sounds like the exact opposite of "first strong", which is a commonly used 675 // description of Unicode's estimation algorithm (getUnicodeDir() above), but the two 676 // must treat formatting characters quite differently. Thus, we are staying away from 677 // both "first" and "last" in these method names to avoid confusion. 678 charIndex = length; 679 int embeddingLevel = 0; 680 int lastNonEmptyEmbeddingLevel = 0; 681 while (charIndex > 0) { 682 switch (dirTypeBackward()) { 683 case Character.DIRECTIONALITY_LEFT_TO_RIGHT: 684 if (embeddingLevel == 0) { 685 return DIR_LTR; 686 } 687 if (lastNonEmptyEmbeddingLevel == 0) { 688 lastNonEmptyEmbeddingLevel = embeddingLevel; 689 } 690 break; 691 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_EMBEDDING: 692 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_OVERRIDE: 693 if (lastNonEmptyEmbeddingLevel == embeddingLevel) { 694 return DIR_LTR; 695 } 696 --embeddingLevel; 697 break; 698 case Character.DIRECTIONALITY_RIGHT_TO_LEFT: 699 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_ARABIC: 700 if (embeddingLevel == 0) { 701 return DIR_RTL; 702 } 703 if (lastNonEmptyEmbeddingLevel == 0) { 704 lastNonEmptyEmbeddingLevel = embeddingLevel; 705 } 706 break; 707 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_EMBEDDING: 708 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_OVERRIDE: 709 if (lastNonEmptyEmbeddingLevel == embeddingLevel) { 710 return DIR_RTL; 711 } 712 --embeddingLevel; 713 break; 714 case Character.DIRECTIONALITY_POP_DIRECTIONAL_FORMAT: 715 ++embeddingLevel; 716 break; 717 case Character.DIRECTIONALITY_BOUNDARY_NEUTRAL: 718 break; 719 default: 720 if (lastNonEmptyEmbeddingLevel == 0) { 721 lastNonEmptyEmbeddingLevel = embeddingLevel; 722 } 723 break; 724 } 725 } 726 return DIR_UNKNOWN; 727 } 728 729 // Internal methods 730 731 /** 732 * Gets the bidi character class, i.e. Character.getDirectionality(), of a given char, using 733 * a cache for speed. Not designed for supplementary codepoints, whose results we do not 734 * cache. 735 */ 736 private static byte getCachedDirectionality(char c) { 737 return c < DIR_TYPE_CACHE_SIZE ? DIR_TYPE_CACHE[c] : Character.getDirectionality(c); 738 } 739 740 /** 741 * Returns the Character.DIRECTIONALITY_... value of the next codepoint and advances 742 * charIndex. If isHtml, and the codepoint is '<' or '&', advances through the tag/entity, 743 * and returns Character.DIRECTIONALITY_WHITESPACE. For an entity, it would be best to 744 * figure out the actual character, and return its dirtype, but treating it as whitespace is 745 * good enough for our purposes. 746 * 747 * @throws java.lang.IndexOutOfBoundsException if called when charIndex >= length or < 0. 748 */ 749 byte dirTypeForward() { 750 lastChar = text.charAt(charIndex); 751 if (Character.isHighSurrogate(lastChar)) { 752 int codePoint = Character.codePointAt(text, charIndex); 753 charIndex += Character.charCount(codePoint); 754 return Character.getDirectionality(codePoint); 755 } 756 charIndex++; 757 byte dirType = getCachedDirectionality(lastChar); 758 if (isHtml) { 759 // Process tags and entities. 760 if (lastChar == '<') { 761 dirType = skipTagForward(); 762 } else if (lastChar == '&') { 763 dirType = skipEntityForward(); 764 } 765 } 766 return dirType; 767 } 768 769 /** 770 * Returns the Character.DIRECTIONALITY_... value of the preceding codepoint and advances 771 * charIndex backwards. If isHtml, and the codepoint is the end of a complete HTML tag or 772 * entity, advances over the whole tag/entity and returns 773 * Character.DIRECTIONALITY_WHITESPACE. For an entity, it would be best to figure out the 774 * actual character, and return its dirtype, but treating it as whitespace is good enough 775 * for our purposes. 776 * 777 * @throws java.lang.IndexOutOfBoundsException if called when charIndex > length or <= 0. 778 */ 779 byte dirTypeBackward() { 780 lastChar = text.charAt(charIndex - 1); 781 if (Character.isLowSurrogate(lastChar)) { 782 int codePoint = Character.codePointBefore(text, charIndex); 783 charIndex -= Character.charCount(codePoint); 784 return Character.getDirectionality(codePoint); 785 } 786 charIndex--; 787 byte dirType = getCachedDirectionality(lastChar); 788 if (isHtml) { 789 // Process tags and entities. 790 if (lastChar == '>') { 791 dirType = skipTagBackward(); 792 } else if (lastChar == ';') { 793 dirType = skipEntityBackward(); 794 } 795 } 796 return dirType; 797 } 798 799 /** 800 * Advances charIndex forward through an HTML tag (after the opening < has already been 801 * read) and returns Character.DIRECTIONALITY_WHITESPACE. If there is no matching >, 802 * does not change charIndex and returns Character.DIRECTIONALITY_OTHER_NEUTRALS (for the 803 * < that hadn't been part of a tag after all). 804 */ 805 private byte skipTagForward() { 806 int initialCharIndex = charIndex; 807 while (charIndex < length) { 808 lastChar = text.charAt(charIndex++); 809 if (lastChar == '>') { 810 // The end of the tag. 811 return Character.DIRECTIONALITY_WHITESPACE; 812 } 813 if (lastChar == '"' || lastChar == '\'') { 814 // Skip over a quoted attribute value inside the tag. 815 char quote = lastChar; 816 while (charIndex < length && (lastChar = text.charAt(charIndex++)) != quote) {} 817 } 818 } 819 // The original '<' wasn't the start of a tag after all. 820 charIndex = initialCharIndex; 821 lastChar = '<'; 822 return Character.DIRECTIONALITY_OTHER_NEUTRALS; 823 } 824 825 /** 826 * Advances charIndex backward through an HTML tag (after the closing > has already been 827 * read) and returns Character.DIRECTIONALITY_WHITESPACE. If there is no matching <, does 828 * not change charIndex and returns Character.DIRECTIONALITY_OTHER_NEUTRALS (for the > 829 * that hadn't been part of a tag after all). Nevertheless, the running time for calling 830 * skipTagBackward() in a loop remains linear in the size of the text, even for a text like 831 * ">>>>", because skipTagBackward() also stops looking for a matching < 832 * when it encounters another >. 833 */ 834 private byte skipTagBackward() { 835 int initialCharIndex = charIndex; 836 while (charIndex > 0) { 837 lastChar = text.charAt(--charIndex); 838 if (lastChar == '<') { 839 // The start of the tag. 840 return Character.DIRECTIONALITY_WHITESPACE; 841 } 842 if (lastChar == '>') { 843 break; 844 } 845 if (lastChar == '"' || lastChar == '\'') { 846 // Skip over a quoted attribute value inside the tag. 847 char quote = lastChar; 848 while (charIndex > 0 && (lastChar = text.charAt(--charIndex)) != quote) {} 849 } 850 } 851 // The original '>' wasn't the end of a tag after all. 852 charIndex = initialCharIndex; 853 lastChar = '>'; 854 return Character.DIRECTIONALITY_OTHER_NEUTRALS; 855 } 856 857 /** 858 * Advances charIndex forward through an HTML character entity tag (after the opening 859 * & has already been read) and returns Character.DIRECTIONALITY_WHITESPACE. It would be 860 * best to figure out the actual character and return its dirtype, but this is good enough. 861 */ 862 private byte skipEntityForward() { 863 while (charIndex < length && (lastChar = text.charAt(charIndex++)) != ';') {} 864 return Character.DIRECTIONALITY_WHITESPACE; 865 } 866 867 /** 868 * Advances charIndex backward through an HTML character entity tag (after the closing ; 869 * has already been read) and returns Character.DIRECTIONALITY_WHITESPACE. It would be best 870 * to figure out the actual character and return its dirtype, but this is good enough. 871 * If there is no matching &, does not change charIndex and returns 872 * Character.DIRECTIONALITY_OTHER_NEUTRALS (for the ';' that did not start an entity after 873 * all). Nevertheless, the running time for calling skipEntityBackward() in a loop remains 874 * linear in the size of the text, even for a text like ";;;;;;;", because skipTagBackward() 875 * also stops looking for a matching & when it encounters another ;. 876 */ 877 private byte skipEntityBackward() { 878 int initialCharIndex = charIndex; 879 while (charIndex > 0) { 880 lastChar = text.charAt(--charIndex); 881 if (lastChar == '&') { 882 return Character.DIRECTIONALITY_WHITESPACE; 883 } 884 if (lastChar == ';') { 885 break; 886 } 887 } 888 charIndex = initialCharIndex; 889 lastChar = ';'; 890 return Character.DIRECTIONALITY_OTHER_NEUTRALS; 891 } 892 } 893 }