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.annotation.Nullable; 20import android.view.View; 21 22import static android.text.TextDirectionHeuristics.FIRSTSTRONG_LTR; 23 24import java.util.Locale; 25 26/** 27 * Utility class for formatting text for display in a potentially opposite-directionality context 28 * without garbling. The directionality of the context is set at formatter creation and the 29 * directionality of the text can be either estimated or passed in when known. 30 * 31 * <p>To support versions lower than {@link android.os.Build.VERSION_CODES#JELLY_BEAN_MR2}, 32 * you can use the support library's {@link android.support.v4.text.BidiFormatter} class. 33 * 34 * <p>These APIs provides the following functionality: 35 * <p> 36 * 1. Bidi Wrapping 37 * When text in one language is mixed into a document in another, opposite-directionality language, 38 * e.g. when an English business name is embedded in some Hebrew text, both the inserted string 39 * and the text surrounding it may be displayed incorrectly unless the inserted string is explicitly 40 * separated from the surrounding text in a "wrapper" that: 41 * <p> 42 * - Declares its directionality so that the string is displayed correctly. This can be done in 43 * Unicode bidi formatting codes by {@link #unicodeWrap} and similar methods. 44 * <p> 45 * - Isolates the string's directionality, so it does not unduly affect the surrounding content. 46 * Currently, this can only be done using invisible Unicode characters of the same direction as 47 * the context (LRM or RLM) in addition to the directionality declaration above, thus "resetting" 48 * the directionality to that of the context. The "reset" may need to be done at both ends of the 49 * string. Without "reset" after the string, the string will "stick" to a number or logically 50 * separate opposite-direction text that happens to follow it in-line (even if separated by 51 * neutral content like spaces and punctuation). Without "reset" before the string, the same can 52 * happen there, but only with more opposite-direction text, not a number. One approach is to 53 * "reset" the direction only after each string, on the theory that if the preceding opposite- 54 * direction text is itself bidi-wrapped, the "reset" after it will prevent the sticking. (Doing 55 * the "reset" only before each string definitely does not work because we do not want to require 56 * bidi-wrapping numbers, and a bidi-wrapped opposite-direction string could be followed by a 57 * number.) Still, the safest policy is to do the "reset" on both ends of each string, since RTL 58 * message translations often contain untranslated Latin-script brand names and technical terms, 59 * and one of these can be followed by a bidi-wrapped inserted value. On the other hand, when one 60 * has such a message, it is best to do the "reset" manually in the message translation itself, 61 * since the message's opposite-direction text could be followed by an inserted number, which we 62 * would not bidi-wrap anyway. Thus, "reset" only after the string is the current default. In an 63 * alternative to "reset", recent additions to the HTML, CSS, and Unicode standards allow the 64 * isolation to be part of the directionality declaration. This form of isolation is better than 65 * "reset" because it takes less space, does not require knowing the context directionality, has a 66 * gentler effect than "reset", and protects both ends of the string. However, we do not yet allow 67 * using it because required platforms do not yet support it. 68 * <p> 69 * Providing these wrapping services is the basic purpose of the bidi formatter. 70 * <p> 71 * 2. Directionality estimation 72 * How does one know whether a string about to be inserted into surrounding text has the same 73 * directionality? Well, in many cases, one knows that this must be the case when writing the code 74 * doing the insertion, e.g. when a localized message is inserted into a localized page. In such 75 * cases there is no need to involve the bidi formatter at all. In some other cases, it need not be 76 * the same as the context, but is either constant (e.g. urls are always LTR) or otherwise known. 77 * In the remaining cases, e.g. when the string is user-entered or comes from a database, the 78 * language of the string (and thus its directionality) is not known a priori, and must be 79 * estimated at run-time. The bidi formatter can do this automatically using the default 80 * first-strong estimation algorithm. It can also be configured to use a custom directionality 81 * estimation object. 82 */ 83public final class BidiFormatter { 84 85 /** 86 * The default text direction heuristic. 87 */ 88 private static TextDirectionHeuristic DEFAULT_TEXT_DIRECTION_HEURISTIC = FIRSTSTRONG_LTR; 89 90 /** 91 * Unicode "Left-To-Right Embedding" (LRE) character. 92 */ 93 private static final char LRE = '\u202A'; 94 95 /** 96 * Unicode "Right-To-Left Embedding" (RLE) character. 97 */ 98 private static final char RLE = '\u202B'; 99 100 /** 101 * Unicode "Pop Directional Formatting" (PDF) character. 102 */ 103 private static final char PDF = '\u202C'; 104 105 /** 106 * Unicode "Left-To-Right Mark" (LRM) character. 107 */ 108 private static final char LRM = '\u200E'; 109 110 /* 111 * Unicode "Right-To-Left Mark" (RLM) character. 112 */ 113 private static final char RLM = '\u200F'; 114 115 /* 116 * String representation of LRM 117 */ 118 private static final String LRM_STRING = Character.toString(LRM); 119 120 /* 121 * String representation of RLM 122 */ 123 private static final String RLM_STRING = Character.toString(RLM); 124 125 /** 126 * Empty string constant. 127 */ 128 private static final String EMPTY_STRING = ""; 129 130 /** 131 * A class for building a BidiFormatter with non-default options. 132 */ 133 public static final class Builder { 134 private boolean mIsRtlContext; 135 private int mFlags; 136 private TextDirectionHeuristic mTextDirectionHeuristic; 137 138 /** 139 * Constructor. 140 * 141 */ 142 public Builder() { 143 initialize(isRtlLocale(Locale.getDefault())); 144 } 145 146 /** 147 * Constructor. 148 * 149 * @param rtlContext Whether the context directionality is RTL. 150 */ 151 public Builder(boolean rtlContext) { 152 initialize(rtlContext); 153 } 154 155 /** 156 * Constructor. 157 * 158 * @param locale The context locale. 159 */ 160 public Builder(Locale locale) { 161 initialize(isRtlLocale(locale)); 162 } 163 164 /** 165 * Initializes the builder with the given context directionality and default options. 166 * 167 * @param isRtlContext Whether the context is RTL or not. 168 */ 169 private void initialize(boolean isRtlContext) { 170 mIsRtlContext = isRtlContext; 171 mTextDirectionHeuristic = DEFAULT_TEXT_DIRECTION_HEURISTIC; 172 mFlags = DEFAULT_FLAGS; 173 } 174 175 /** 176 * Specifies whether the BidiFormatter to be built should also "reset" directionality before 177 * a string being bidi-wrapped, not just after it. The default is true. 178 */ 179 public Builder stereoReset(boolean stereoReset) { 180 if (stereoReset) { 181 mFlags |= FLAG_STEREO_RESET; 182 } else { 183 mFlags &= ~FLAG_STEREO_RESET; 184 } 185 return this; 186 } 187 188 /** 189 * Specifies the default directionality estimation algorithm to be used by the BidiFormatter. 190 * By default, uses the first-strong heuristic. 191 * 192 * @param heuristic the {@code TextDirectionHeuristic} to use. 193 * @return the builder itself. 194 */ 195 public Builder setTextDirectionHeuristic(TextDirectionHeuristic heuristic) { 196 mTextDirectionHeuristic = heuristic; 197 return this; 198 } 199 200 /** 201 * @return A BidiFormatter with the specified options. 202 */ 203 public BidiFormatter build() { 204 if (mFlags == DEFAULT_FLAGS && 205 mTextDirectionHeuristic == DEFAULT_TEXT_DIRECTION_HEURISTIC) { 206 return BidiFormatter.getDefaultInstanceFromContext(mIsRtlContext); 207 } 208 return new BidiFormatter(mIsRtlContext, mFlags, mTextDirectionHeuristic); 209 } 210 } 211 212 // 213 private static final int FLAG_STEREO_RESET = 2; 214 private static final int DEFAULT_FLAGS = FLAG_STEREO_RESET; 215 216 private static final BidiFormatter DEFAULT_LTR_INSTANCE = new BidiFormatter( 217 false /* LTR context */, 218 DEFAULT_FLAGS, 219 DEFAULT_TEXT_DIRECTION_HEURISTIC); 220 221 private static final BidiFormatter DEFAULT_RTL_INSTANCE = new BidiFormatter( 222 true /* RTL context */, 223 DEFAULT_FLAGS, 224 DEFAULT_TEXT_DIRECTION_HEURISTIC); 225 226 private final boolean mIsRtlContext; 227 private final int mFlags; 228 private final TextDirectionHeuristic mDefaultTextDirectionHeuristic; 229 230 /** 231 * Factory for creating an instance of BidiFormatter for the default locale directionality. 232 * 233 * This does not create any new objects, and returns already existing static instances. 234 * 235 */ 236 public static BidiFormatter getInstance() { 237 return getDefaultInstanceFromContext(isRtlLocale(Locale.getDefault())); 238 } 239 240 /** 241 * Factory for creating an instance of BidiFormatter given the context directionality. 242 * 243 * This does not create any new objects, and returns already existing static instances. 244 * 245 * @param rtlContext Whether the context directionality is RTL. 246 */ 247 public static BidiFormatter getInstance(boolean rtlContext) { 248 return getDefaultInstanceFromContext(rtlContext); 249 } 250 251 /** 252 * Factory for creating an instance of BidiFormatter given the context locale. 253 * 254 * This does not create any new objects, and returns already existing static instances. 255 * 256 * @param locale The context locale. 257 */ 258 public static BidiFormatter getInstance(Locale locale) { 259 return getDefaultInstanceFromContext(isRtlLocale(locale)); 260 } 261 262 /** 263 * @param isRtlContext Whether the context directionality is RTL or not. 264 * @param flags The option flags. 265 * @param heuristic The default text direction heuristic. 266 */ 267 private BidiFormatter(boolean isRtlContext, int flags, TextDirectionHeuristic heuristic) { 268 mIsRtlContext = isRtlContext; 269 mFlags = flags; 270 mDefaultTextDirectionHeuristic = heuristic; 271 } 272 273 /** 274 * @return Whether the context directionality is RTL 275 */ 276 public boolean isRtlContext() { 277 return mIsRtlContext; 278 } 279 280 /** 281 * @return Whether directionality "reset" should also be done before a string being 282 * bidi-wrapped, not just after it. 283 */ 284 public boolean getStereoReset() { 285 return (mFlags & FLAG_STEREO_RESET) != 0; 286 } 287 288 /** 289 * Returns a Unicode bidi mark matching the context directionality (LRM or RLM) if either the 290 * overall or the exit directionality of a given string is opposite to the context directionality. 291 * Putting this after the string (including its directionality declaration wrapping) prevents it 292 * from "sticking" to other opposite-directionality text or a number appearing after it inline 293 * with only neutral content in between. Otherwise returns the empty string. While the exit 294 * directionality is determined by scanning the end of the string, the overall directionality is 295 * given explicitly by a heuristic to estimate the {@code str}'s directionality. 296 * 297 * @param str CharSequence after which the mark may need to appear. 298 * @param heuristic The text direction heuristic that will be used to estimate the {@code str}'s 299 * directionality. 300 * @return LRM for RTL text in LTR context; RLM for LTR text in RTL context; 301 * else, the empty string. 302 * 303 * @hide 304 */ 305 public String markAfter(CharSequence str, TextDirectionHeuristic heuristic) { 306 final boolean isRtl = heuristic.isRtl(str, 0, str.length()); 307 // getExitDir() is called only if needed (short-circuit). 308 if (!mIsRtlContext && (isRtl || getExitDir(str) == DIR_RTL)) { 309 return LRM_STRING; 310 } 311 if (mIsRtlContext && (!isRtl || getExitDir(str) == DIR_LTR)) { 312 return RLM_STRING; 313 } 314 return EMPTY_STRING; 315 } 316 317 /** 318 * Returns a Unicode bidi mark matching the context directionality (LRM or RLM) if either the 319 * overall or the entry directionality of a given string is opposite to the context 320 * directionality. Putting this before the string (including its directionality declaration 321 * wrapping) prevents it from "sticking" to other opposite-directionality text appearing before 322 * it inline with only neutral content in between. Otherwise returns the empty string. While the 323 * entry directionality is determined by scanning the beginning of the string, the overall 324 * directionality is given explicitly by a heuristic to estimate the {@code str}'s directionality. 325 * 326 * @param str CharSequence before which the mark may need to appear. 327 * @param heuristic The text direction heuristic that will be used to estimate the {@code str}'s 328 * directionality. 329 * @return LRM for RTL text in LTR context; RLM for LTR text in RTL context; 330 * else, the empty string. 331 * 332 * @hide 333 */ 334 public String markBefore(CharSequence str, TextDirectionHeuristic heuristic) { 335 final boolean isRtl = heuristic.isRtl(str, 0, str.length()); 336 // getEntryDir() is called only if needed (short-circuit). 337 if (!mIsRtlContext && (isRtl || getEntryDir(str) == DIR_RTL)) { 338 return LRM_STRING; 339 } 340 if (mIsRtlContext && (!isRtl || getEntryDir(str) == DIR_LTR)) { 341 return RLM_STRING; 342 } 343 return EMPTY_STRING; 344 } 345 346 /** 347 * Estimates the directionality of a string using the default text direction heuristic. 348 * 349 * @param str String whose directionality is to be estimated. 350 * @return true if {@code str}'s estimated overall directionality is RTL. Otherwise returns 351 * false. 352 */ 353 public boolean isRtl(String str) { 354 return isRtl((CharSequence) str); 355 } 356 357 /** 358 * @hide 359 */ 360 public boolean isRtl(CharSequence str) { 361 return mDefaultTextDirectionHeuristic.isRtl(str, 0, str.length()); 362 } 363 364 /** 365 * Formats a string of given directionality for use in plain-text output of the context 366 * directionality, so an opposite-directionality string is neither garbled nor garbles its 367 * surroundings. This makes use of Unicode bidi formatting characters. 368 * <p> 369 * The algorithm: In case the given directionality doesn't match the context directionality, wraps 370 * the string with Unicode bidi formatting characters: RLE+{@code str}+PDF for RTL text, or 371 * LRE+{@code str}+PDF for LTR text. 372 * <p> 373 * If {@code isolate}, directionally isolates the string so that it does not garble its 374 * surroundings. Currently, this is done by "resetting" the directionality after the string by 375 * appending a trailing Unicode bidi mark matching the context directionality (LRM or RLM) when 376 * either the overall directionality or the exit directionality of the string is opposite to 377 * that of the context. Unless the formatter was built using 378 * {@link Builder#stereoReset(boolean)} with a {@code false} argument, also prepends a Unicode 379 * bidi mark matching the context directionality when either the overall directionality or the 380 * entry directionality of the string is opposite to that of the context. Note that as opposed 381 * to the overall directionality, the entry and exit directionalities are determined from the 382 * string itself. 383 * <p> 384 * Does *not* do HTML-escaping. 385 * 386 * @param str The input string. 387 * @param heuristic The algorithm to be used to estimate the string's overall direction. 388 * See {@link TextDirectionHeuristics} for pre-defined heuristics. 389 * @param isolate Whether to directionally isolate the string to prevent it from garbling the 390 * content around it 391 * @return Input string after applying the above processing. {@code null} if {@code str} is 392 * {@code null}. 393 */ 394 public @Nullable String unicodeWrap(@Nullable String str, TextDirectionHeuristic heuristic, 395 boolean isolate) { 396 if (str == null) return null; 397 return unicodeWrap((CharSequence) str, heuristic, isolate).toString(); 398 } 399 400 /** 401 * @hide 402 */ 403 public @Nullable CharSequence unicodeWrap(@Nullable CharSequence str, 404 TextDirectionHeuristic heuristic, boolean isolate) { 405 if (str == null) return null; 406 final boolean isRtl = heuristic.isRtl(str, 0, str.length()); 407 SpannableStringBuilder result = new SpannableStringBuilder(); 408 if (getStereoReset() && isolate) { 409 result.append(markBefore(str, 410 isRtl ? TextDirectionHeuristics.RTL : TextDirectionHeuristics.LTR)); 411 } 412 if (isRtl != mIsRtlContext) { 413 result.append(isRtl ? RLE : LRE); 414 result.append(str); 415 result.append(PDF); 416 } else { 417 result.append(str); 418 } 419 if (isolate) { 420 result.append(markAfter(str, 421 isRtl ? TextDirectionHeuristics.RTL : TextDirectionHeuristics.LTR)); 422 } 423 return result; 424 } 425 426 /** 427 * Operates like {@link #unicodeWrap(String, TextDirectionHeuristic, boolean)}, but assumes 428 * {@code isolate} is true. 429 * 430 * @param str The input string. 431 * @param heuristic The algorithm to be used to estimate the string's overall direction. 432 * See {@link TextDirectionHeuristics} for pre-defined heuristics. 433 * @return Input string after applying the above processing. 434 */ 435 public String unicodeWrap(String str, TextDirectionHeuristic heuristic) { 436 return unicodeWrap(str, heuristic, true /* isolate */); 437 } 438 439 /** 440 * @hide 441 */ 442 public CharSequence unicodeWrap(CharSequence str, TextDirectionHeuristic heuristic) { 443 return unicodeWrap(str, heuristic, true /* isolate */); 444 } 445 446 447 /** 448 * Operates like {@link #unicodeWrap(String, TextDirectionHeuristic, boolean)}, but uses the 449 * formatter's default direction estimation algorithm. 450 * 451 * @param str The input string. 452 * @param isolate Whether to directionally isolate the string to prevent it from garbling the 453 * content around it 454 * @return Input string after applying the above processing. 455 */ 456 public String unicodeWrap(String str, boolean isolate) { 457 return unicodeWrap(str, mDefaultTextDirectionHeuristic, isolate); 458 } 459 460 /** 461 * @hide 462 */ 463 public CharSequence unicodeWrap(CharSequence str, boolean isolate) { 464 return unicodeWrap(str, mDefaultTextDirectionHeuristic, isolate); 465 } 466 467 /** 468 * Operates like {@link #unicodeWrap(String, TextDirectionHeuristic, boolean)}, but uses the 469 * formatter's default direction estimation algorithm and assumes {@code isolate} is true. 470 * 471 * @param str The input string. 472 * @return Input string after applying the above processing. 473 */ 474 public String unicodeWrap(String str) { 475 return unicodeWrap(str, mDefaultTextDirectionHeuristic, true /* isolate */); 476 } 477 478 /** 479 * @hide 480 */ 481 public CharSequence unicodeWrap(CharSequence str) { 482 return unicodeWrap(str, mDefaultTextDirectionHeuristic, true /* isolate */); 483 } 484 485 private static BidiFormatter getDefaultInstanceFromContext(boolean isRtlContext) { 486 return isRtlContext ? DEFAULT_RTL_INSTANCE : DEFAULT_LTR_INSTANCE; 487 } 488 489 /** 490 * Helper method to return true if the Locale directionality is RTL. 491 * 492 * @param locale The Locale whose directionality will be checked to be RTL or LTR 493 * @return true if the {@code locale} directionality is RTL. False otherwise. 494 */ 495 private static boolean isRtlLocale(Locale locale) { 496 return (TextUtils.getLayoutDirectionFromLocale(locale) == View.LAYOUT_DIRECTION_RTL); 497 } 498 499 /** 500 * Enum for directionality type. 501 */ 502 private static final int DIR_LTR = -1; 503 private static final int DIR_UNKNOWN = 0; 504 private static final int DIR_RTL = +1; 505 506 /** 507 * Returns the directionality of the last character with strong directionality in the string, or 508 * DIR_UNKNOWN if none was encountered. For efficiency, actually scans backwards from the end of 509 * the string. Treats a non-BN character between an LRE/RLE/LRO/RLO and its matching PDF as a 510 * strong character, LTR after LRE/LRO, and RTL after RLE/RLO. The results are undefined for a 511 * string containing unbalanced LRE/RLE/LRO/RLO/PDF characters. The intended use is to check 512 * whether a logically separate item that starts with a number or a character of the string's 513 * exit directionality and follows this string inline (not counting any neutral characters in 514 * between) would "stick" to it in an opposite-directionality context, thus being displayed in 515 * an incorrect position. An LRM or RLM character (the one of the context's directionality) 516 * between the two will prevent such sticking. 517 * 518 * @param str the string to check. 519 */ 520 private static int getExitDir(CharSequence str) { 521 return new DirectionalityEstimator(str, false /* isHtml */).getExitDir(); 522 } 523 524 /** 525 * Returns the directionality of the first character with strong directionality in the string, 526 * or DIR_UNKNOWN if none was encountered. Treats a non-BN character between an 527 * LRE/RLE/LRO/RLO and its matching PDF as a strong character, LTR after LRE/LRO, and RTL after 528 * RLE/RLO. The results are undefined for a string containing unbalanced LRE/RLE/LRO/RLO/PDF 529 * characters. The intended use is to check whether a logically separate item that ends with a 530 * character of the string's entry directionality and precedes the string inline (not counting 531 * any neutral characters in between) would "stick" to it in an opposite-directionality context, 532 * thus being displayed in an incorrect position. An LRM or RLM character (the one of the 533 * context's directionality) between the two will prevent such sticking. 534 * 535 * @param str the string to check. 536 */ 537 private static int getEntryDir(CharSequence str) { 538 return new DirectionalityEstimator(str, false /* isHtml */).getEntryDir(); 539 } 540 541 /** 542 * An object that estimates the directionality of a given string by various methods. 543 * 544 */ 545 private static class DirectionalityEstimator { 546 547 // Internal static variables and constants. 548 549 /** 550 * Size of the bidi character class cache. The results of the Character.getDirectionality() 551 * calls on the lowest DIR_TYPE_CACHE_SIZE codepoints are kept in an array for speed. 552 * The 0x700 value is designed to leave all the European and Near Eastern languages in the 553 * cache. It can be reduced to 0x180, restricting the cache to the Western European 554 * languages. 555 */ 556 private static final int DIR_TYPE_CACHE_SIZE = 0x700; 557 558 /** 559 * The bidi character class cache. 560 */ 561 private static final byte DIR_TYPE_CACHE[]; 562 563 static { 564 DIR_TYPE_CACHE = new byte[DIR_TYPE_CACHE_SIZE]; 565 for (int i = 0; i < DIR_TYPE_CACHE_SIZE; i++) { 566 DIR_TYPE_CACHE[i] = Character.getDirectionality(i); 567 } 568 } 569 570 // Internal instance variables. 571 572 /** 573 * The text to be scanned. 574 */ 575 private final CharSequence text; 576 577 /** 578 * Whether the text to be scanned is to be treated as HTML, i.e. skipping over tags and 579 * entities when looking for the next / preceding dir type. 580 */ 581 private final boolean isHtml; 582 583 /** 584 * The length of the text in chars. 585 */ 586 private final int length; 587 588 /** 589 * The current position in the text. 590 */ 591 private int charIndex; 592 593 /** 594 * The char encountered by the last dirTypeForward or dirTypeBackward call. If it 595 * encountered a supplementary codepoint, this contains a char that is not a valid 596 * codepoint. This is ok, because this member is only used to detect some well-known ASCII 597 * syntax, e.g. "http://" and the beginning of an HTML tag or entity. 598 */ 599 private char lastChar; 600 601 /** 602 * Constructor. 603 * 604 * @param text The string to scan. 605 * @param isHtml Whether the text to be scanned is to be treated as HTML, i.e. skipping over 606 * tags and entities. 607 */ 608 DirectionalityEstimator(CharSequence text, boolean isHtml) { 609 this.text = text; 610 this.isHtml = isHtml; 611 length = text.length(); 612 } 613 614 /** 615 * Returns the directionality of the first character with strong directionality in the 616 * string, or DIR_UNKNOWN if none was encountered. Treats a non-BN character between an 617 * LRE/RLE/LRO/RLO and its matching PDF as a strong character, LTR after LRE/LRO, and RTL 618 * after RLE/RLO. The results are undefined for a string containing unbalanced 619 * LRE/RLE/LRO/RLO/PDF characters. 620 */ 621 int getEntryDir() { 622 // The reason for this method name, as opposed to getFirstStrongDir(), is that 623 // "first strong" is a commonly used description of Unicode's estimation algorithm, 624 // but the two must treat formatting characters quite differently. Thus, we are staying 625 // away from both "first" and "last" in these method names to avoid confusion. 626 charIndex = 0; 627 int embeddingLevel = 0; 628 int embeddingLevelDir = DIR_UNKNOWN; 629 int firstNonEmptyEmbeddingLevel = 0; 630 while (charIndex < length && firstNonEmptyEmbeddingLevel == 0) { 631 switch (dirTypeForward()) { 632 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_EMBEDDING: 633 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_OVERRIDE: 634 ++embeddingLevel; 635 embeddingLevelDir = DIR_LTR; 636 break; 637 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_EMBEDDING: 638 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_OVERRIDE: 639 ++embeddingLevel; 640 embeddingLevelDir = DIR_RTL; 641 break; 642 case Character.DIRECTIONALITY_POP_DIRECTIONAL_FORMAT: 643 --embeddingLevel; 644 // To restore embeddingLevelDir to its previous value, we would need a 645 // stack, which we want to avoid. Thus, at this point we do not know the 646 // current embedding's directionality. 647 embeddingLevelDir = DIR_UNKNOWN; 648 break; 649 case Character.DIRECTIONALITY_BOUNDARY_NEUTRAL: 650 break; 651 case Character.DIRECTIONALITY_LEFT_TO_RIGHT: 652 if (embeddingLevel == 0) { 653 return DIR_LTR; 654 } 655 firstNonEmptyEmbeddingLevel = embeddingLevel; 656 break; 657 case Character.DIRECTIONALITY_RIGHT_TO_LEFT: 658 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_ARABIC: 659 if (embeddingLevel == 0) { 660 return DIR_RTL; 661 } 662 firstNonEmptyEmbeddingLevel = embeddingLevel; 663 break; 664 default: 665 firstNonEmptyEmbeddingLevel = embeddingLevel; 666 break; 667 } 668 } 669 670 // We have either found a non-empty embedding or scanned the entire string finding 671 // neither a non-empty embedding nor a strong character outside of an embedding. 672 if (firstNonEmptyEmbeddingLevel == 0) { 673 // We have not found a non-empty embedding. Thus, the string contains neither a 674 // non-empty embedding nor a strong character outside of an embedding. 675 return DIR_UNKNOWN; 676 } 677 678 // We have found a non-empty embedding. 679 if (embeddingLevelDir != DIR_UNKNOWN) { 680 // We know the directionality of the non-empty embedding. 681 return embeddingLevelDir; 682 } 683 684 // We do not remember the directionality of the non-empty embedding we found. So, we go 685 // backwards to find the start of the non-empty embedding and get its directionality. 686 while (charIndex > 0) { 687 switch (dirTypeBackward()) { 688 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_EMBEDDING: 689 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_OVERRIDE: 690 if (firstNonEmptyEmbeddingLevel == embeddingLevel) { 691 return DIR_LTR; 692 } 693 --embeddingLevel; 694 break; 695 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_EMBEDDING: 696 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_OVERRIDE: 697 if (firstNonEmptyEmbeddingLevel == embeddingLevel) { 698 return DIR_RTL; 699 } 700 --embeddingLevel; 701 break; 702 case Character.DIRECTIONALITY_POP_DIRECTIONAL_FORMAT: 703 ++embeddingLevel; 704 break; 705 } 706 } 707 // We should never get here. 708 return DIR_UNKNOWN; 709 } 710 711 /** 712 * Returns the directionality of the last character with strong directionality in the 713 * string, or DIR_UNKNOWN if none was encountered. For efficiency, actually scans backwards 714 * from the end of the string. Treats a non-BN character between an LRE/RLE/LRO/RLO and its 715 * matching PDF as a strong character, LTR after LRE/LRO, and RTL after RLE/RLO. The results 716 * are undefined for a string containing unbalanced LRE/RLE/LRO/RLO/PDF characters. 717 */ 718 int getExitDir() { 719 // The reason for this method name, as opposed to getLastStrongDir(), is that "last 720 // strong" sounds like the exact opposite of "first strong", which is a commonly used 721 // description of Unicode's estimation algorithm (getUnicodeDir() above), but the two 722 // must treat formatting characters quite differently. Thus, we are staying away from 723 // both "first" and "last" in these method names to avoid confusion. 724 charIndex = length; 725 int embeddingLevel = 0; 726 int lastNonEmptyEmbeddingLevel = 0; 727 while (charIndex > 0) { 728 switch (dirTypeBackward()) { 729 case Character.DIRECTIONALITY_LEFT_TO_RIGHT: 730 if (embeddingLevel == 0) { 731 return DIR_LTR; 732 } 733 if (lastNonEmptyEmbeddingLevel == 0) { 734 lastNonEmptyEmbeddingLevel = embeddingLevel; 735 } 736 break; 737 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_EMBEDDING: 738 case Character.DIRECTIONALITY_LEFT_TO_RIGHT_OVERRIDE: 739 if (lastNonEmptyEmbeddingLevel == embeddingLevel) { 740 return DIR_LTR; 741 } 742 --embeddingLevel; 743 break; 744 case Character.DIRECTIONALITY_RIGHT_TO_LEFT: 745 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_ARABIC: 746 if (embeddingLevel == 0) { 747 return DIR_RTL; 748 } 749 if (lastNonEmptyEmbeddingLevel == 0) { 750 lastNonEmptyEmbeddingLevel = embeddingLevel; 751 } 752 break; 753 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_EMBEDDING: 754 case Character.DIRECTIONALITY_RIGHT_TO_LEFT_OVERRIDE: 755 if (lastNonEmptyEmbeddingLevel == embeddingLevel) { 756 return DIR_RTL; 757 } 758 --embeddingLevel; 759 break; 760 case Character.DIRECTIONALITY_POP_DIRECTIONAL_FORMAT: 761 ++embeddingLevel; 762 break; 763 case Character.DIRECTIONALITY_BOUNDARY_NEUTRAL: 764 break; 765 default: 766 if (lastNonEmptyEmbeddingLevel == 0) { 767 lastNonEmptyEmbeddingLevel = embeddingLevel; 768 } 769 break; 770 } 771 } 772 return DIR_UNKNOWN; 773 } 774 775 // Internal methods 776 777 /** 778 * Gets the bidi character class, i.e. Character.getDirectionality(), of a given char, using 779 * a cache for speed. Not designed for supplementary codepoints, whose results we do not 780 * cache. 781 */ 782 private static byte getCachedDirectionality(char c) { 783 return c < DIR_TYPE_CACHE_SIZE ? DIR_TYPE_CACHE[c] : Character.getDirectionality(c); 784 } 785 786 /** 787 * Returns the Character.DIRECTIONALITY_... value of the next codepoint and advances 788 * charIndex. If isHtml, and the codepoint is '<' or '&', advances through the tag/entity, 789 * and returns Character.DIRECTIONALITY_WHITESPACE. For an entity, it would be best to 790 * figure out the actual character, and return its dirtype, but treating it as whitespace is 791 * good enough for our purposes. 792 * 793 * @throws java.lang.IndexOutOfBoundsException if called when charIndex >= length or < 0. 794 */ 795 byte dirTypeForward() { 796 lastChar = text.charAt(charIndex); 797 if (Character.isHighSurrogate(lastChar)) { 798 int codePoint = Character.codePointAt(text, charIndex); 799 charIndex += Character.charCount(codePoint); 800 return Character.getDirectionality(codePoint); 801 } 802 charIndex++; 803 byte dirType = getCachedDirectionality(lastChar); 804 if (isHtml) { 805 // Process tags and entities. 806 if (lastChar == '<') { 807 dirType = skipTagForward(); 808 } else if (lastChar == '&') { 809 dirType = skipEntityForward(); 810 } 811 } 812 return dirType; 813 } 814 815 /** 816 * Returns the Character.DIRECTIONALITY_... value of the preceding codepoint and advances 817 * charIndex backwards. If isHtml, and the codepoint is the end of a complete HTML tag or 818 * entity, advances over the whole tag/entity and returns 819 * Character.DIRECTIONALITY_WHITESPACE. For an entity, it would be best to figure out the 820 * actual character, and return its dirtype, but treating it as whitespace is good enough 821 * for our purposes. 822 * 823 * @throws java.lang.IndexOutOfBoundsException if called when charIndex > length or <= 0. 824 */ 825 byte dirTypeBackward() { 826 lastChar = text.charAt(charIndex - 1); 827 if (Character.isLowSurrogate(lastChar)) { 828 int codePoint = Character.codePointBefore(text, charIndex); 829 charIndex -= Character.charCount(codePoint); 830 return Character.getDirectionality(codePoint); 831 } 832 charIndex--; 833 byte dirType = getCachedDirectionality(lastChar); 834 if (isHtml) { 835 // Process tags and entities. 836 if (lastChar == '>') { 837 dirType = skipTagBackward(); 838 } else if (lastChar == ';') { 839 dirType = skipEntityBackward(); 840 } 841 } 842 return dirType; 843 } 844 845 /** 846 * Advances charIndex forward through an HTML tag (after the opening < has already been 847 * read) and returns Character.DIRECTIONALITY_WHITESPACE. If there is no matching >, 848 * does not change charIndex and returns Character.DIRECTIONALITY_OTHER_NEUTRALS (for the 849 * < that hadn't been part of a tag after all). 850 */ 851 private byte skipTagForward() { 852 int initialCharIndex = charIndex; 853 while (charIndex < length) { 854 lastChar = text.charAt(charIndex++); 855 if (lastChar == '>') { 856 // The end of the tag. 857 return Character.DIRECTIONALITY_WHITESPACE; 858 } 859 if (lastChar == '"' || lastChar == '\'') { 860 // Skip over a quoted attribute value inside the tag. 861 char quote = lastChar; 862 while (charIndex < length && (lastChar = text.charAt(charIndex++)) != quote) {} 863 } 864 } 865 // The original '<' wasn't the start of a tag after all. 866 charIndex = initialCharIndex; 867 lastChar = '<'; 868 return Character.DIRECTIONALITY_OTHER_NEUTRALS; 869 } 870 871 /** 872 * Advances charIndex backward through an HTML tag (after the closing > has already been 873 * read) and returns Character.DIRECTIONALITY_WHITESPACE. If there is no matching <, does 874 * not change charIndex and returns Character.DIRECTIONALITY_OTHER_NEUTRALS (for the > 875 * that hadn't been part of a tag after all). Nevertheless, the running time for calling 876 * skipTagBackward() in a loop remains linear in the size of the text, even for a text like 877 * ">>>>", because skipTagBackward() also stops looking for a matching < 878 * when it encounters another >. 879 */ 880 private byte skipTagBackward() { 881 int initialCharIndex = charIndex; 882 while (charIndex > 0) { 883 lastChar = text.charAt(--charIndex); 884 if (lastChar == '<') { 885 // The start of the tag. 886 return Character.DIRECTIONALITY_WHITESPACE; 887 } 888 if (lastChar == '>') { 889 break; 890 } 891 if (lastChar == '"' || lastChar == '\'') { 892 // Skip over a quoted attribute value inside the tag. 893 char quote = lastChar; 894 while (charIndex > 0 && (lastChar = text.charAt(--charIndex)) != quote) {} 895 } 896 } 897 // The original '>' wasn't the end of a tag after all. 898 charIndex = initialCharIndex; 899 lastChar = '>'; 900 return Character.DIRECTIONALITY_OTHER_NEUTRALS; 901 } 902 903 /** 904 * Advances charIndex forward through an HTML character entity tag (after the opening 905 * & has already been read) and returns Character.DIRECTIONALITY_WHITESPACE. It would be 906 * best to figure out the actual character and return its dirtype, but this is good enough. 907 */ 908 private byte skipEntityForward() { 909 while (charIndex < length && (lastChar = text.charAt(charIndex++)) != ';') {} 910 return Character.DIRECTIONALITY_WHITESPACE; 911 } 912 913 /** 914 * Advances charIndex backward through an HTML character entity tag (after the closing ; 915 * has already been read) and returns Character.DIRECTIONALITY_WHITESPACE. It would be best 916 * to figure out the actual character and return its dirtype, but this is good enough. 917 * If there is no matching &, does not change charIndex and returns 918 * Character.DIRECTIONALITY_OTHER_NEUTRALS (for the ';' that did not start an entity after 919 * all). Nevertheless, the running time for calling skipEntityBackward() in a loop remains 920 * linear in the size of the text, even for a text like ";;;;;;;", because skipTagBackward() 921 * also stops looking for a matching & when it encounters another ;. 922 */ 923 private byte skipEntityBackward() { 924 int initialCharIndex = charIndex; 925 while (charIndex > 0) { 926 lastChar = text.charAt(--charIndex); 927 if (lastChar == '&') { 928 return Character.DIRECTIONALITY_WHITESPACE; 929 } 930 if (lastChar == ';') { 931 break; 932 } 933 } 934 charIndex = initialCharIndex; 935 lastChar = ';'; 936 return Character.DIRECTIONALITY_OTHER_NEUTRALS; 937 } 938 } 939} 940