1 /* 2 ******************************************************************************* 3 * 4 * Copyright (C) 2009-2014, International Business Machines 5 * Corporation and others. All Rights Reserved. 6 * 7 ******************************************************************************* 8 * file name: normalizer2impl.h 9 * encoding: US-ASCII 10 * tab size: 8 (not used) 11 * indentation:4 12 * 13 * created on: 2009nov22 14 * created by: Markus W. Scherer 15 */ 16 17 #ifndef __NORMALIZER2IMPL_H__ 18 #define __NORMALIZER2IMPL_H__ 19 20 #include "unicode/utypes.h" 21 22 #if !UCONFIG_NO_NORMALIZATION 23 24 #include "unicode/normalizer2.h" 25 #include "unicode/unistr.h" 26 #include "unicode/unorm.h" 27 #include "unicode/utf16.h" 28 #include "mutex.h" 29 #include "uset_imp.h" 30 #include "utrie2.h" 31 32 U_NAMESPACE_BEGIN 33 34 struct CanonIterData; 35 36 class U_COMMON_API Hangul { 37 public: 38 /* Korean Hangul and Jamo constants */ 39 enum { 40 JAMO_L_BASE=0x1100, /* "lead" jamo */ 41 JAMO_L_END=0x1112, 42 JAMO_V_BASE=0x1161, /* "vowel" jamo */ 43 JAMO_V_END=0x1175, 44 JAMO_T_BASE=0x11a7, /* "trail" jamo */ 45 JAMO_T_END=0x11c2, 46 47 HANGUL_BASE=0xac00, 48 HANGUL_END=0xd7a3, 49 50 JAMO_L_COUNT=19, 51 JAMO_V_COUNT=21, 52 JAMO_T_COUNT=28, 53 54 JAMO_VT_COUNT=JAMO_V_COUNT*JAMO_T_COUNT, 55 56 HANGUL_COUNT=JAMO_L_COUNT*JAMO_V_COUNT*JAMO_T_COUNT, 57 HANGUL_LIMIT=HANGUL_BASE+HANGUL_COUNT 58 }; 59 60 static inline UBool isHangul(UChar32 c) { 61 return HANGUL_BASE<=c && c<HANGUL_LIMIT; 62 } 63 static inline UBool 64 isHangulWithoutJamoT(UChar c) { 65 c-=HANGUL_BASE; 66 return c<HANGUL_COUNT && c%JAMO_T_COUNT==0; 67 } 68 static inline UBool isJamoL(UChar32 c) { 69 return (uint32_t)(c-JAMO_L_BASE)<JAMO_L_COUNT; 70 } 71 static inline UBool isJamoV(UChar32 c) { 72 return (uint32_t)(c-JAMO_V_BASE)<JAMO_V_COUNT; 73 } 74 75 /** 76 * Decomposes c, which must be a Hangul syllable, into buffer 77 * and returns the length of the decomposition (2 or 3). 78 */ 79 static inline int32_t decompose(UChar32 c, UChar buffer[3]) { 80 c-=HANGUL_BASE; 81 UChar32 c2=c%JAMO_T_COUNT; 82 c/=JAMO_T_COUNT; 83 buffer[0]=(UChar)(JAMO_L_BASE+c/JAMO_V_COUNT); 84 buffer[1]=(UChar)(JAMO_V_BASE+c%JAMO_V_COUNT); 85 if(c2==0) { 86 return 2; 87 } else { 88 buffer[2]=(UChar)(JAMO_T_BASE+c2); 89 return 3; 90 } 91 } 92 93 /** 94 * Decomposes c, which must be a Hangul syllable, into buffer. 95 * This is the raw, not recursive, decomposition. Its length is always 2. 96 */ 97 static inline void getRawDecomposition(UChar32 c, UChar buffer[2]) { 98 UChar32 orig=c; 99 c-=HANGUL_BASE; 100 UChar32 c2=c%JAMO_T_COUNT; 101 if(c2==0) { 102 c/=JAMO_T_COUNT; 103 buffer[0]=(UChar)(JAMO_L_BASE+c/JAMO_V_COUNT); 104 buffer[1]=(UChar)(JAMO_V_BASE+c%JAMO_V_COUNT); 105 } else { 106 buffer[0]=orig-c2; // LV syllable 107 buffer[1]=(UChar)(JAMO_T_BASE+c2); 108 } 109 } 110 private: 111 Hangul(); // no instantiation 112 }; 113 114 class Normalizer2Impl; 115 116 class U_COMMON_API ReorderingBuffer : public UMemory { 117 public: 118 ReorderingBuffer(const Normalizer2Impl &ni, UnicodeString &dest) : 119 impl(ni), str(dest), 120 start(NULL), reorderStart(NULL), limit(NULL), 121 remainingCapacity(0), lastCC(0) {} 122 ~ReorderingBuffer() { 123 if(start!=NULL) { 124 str.releaseBuffer((int32_t)(limit-start)); 125 } 126 } 127 UBool init(int32_t destCapacity, UErrorCode &errorCode); 128 129 UBool isEmpty() const { return start==limit; } 130 int32_t length() const { return (int32_t)(limit-start); } 131 UChar *getStart() { return start; } 132 UChar *getLimit() { return limit; } 133 uint8_t getLastCC() const { return lastCC; } 134 135 UBool equals(const UChar *start, const UChar *limit) const; 136 137 // For Hangul composition, replacing the Leading consonant Jamo with the syllable. 138 void setLastChar(UChar c) { 139 *(limit-1)=c; 140 } 141 142 UBool append(UChar32 c, uint8_t cc, UErrorCode &errorCode) { 143 return (c<=0xffff) ? 144 appendBMP((UChar)c, cc, errorCode) : 145 appendSupplementary(c, cc, errorCode); 146 } 147 // s must be in NFD, otherwise change the implementation. 148 UBool append(const UChar *s, int32_t length, 149 uint8_t leadCC, uint8_t trailCC, 150 UErrorCode &errorCode); 151 UBool appendBMP(UChar c, uint8_t cc, UErrorCode &errorCode) { 152 if(remainingCapacity==0 && !resize(1, errorCode)) { 153 return FALSE; 154 } 155 if(lastCC<=cc || cc==0) { 156 *limit++=c; 157 lastCC=cc; 158 if(cc<=1) { 159 reorderStart=limit; 160 } 161 } else { 162 insert(c, cc); 163 } 164 --remainingCapacity; 165 return TRUE; 166 } 167 UBool appendZeroCC(UChar32 c, UErrorCode &errorCode); 168 UBool appendZeroCC(const UChar *s, const UChar *sLimit, UErrorCode &errorCode); 169 void remove(); 170 void removeSuffix(int32_t suffixLength); 171 void setReorderingLimit(UChar *newLimit) { 172 remainingCapacity+=(int32_t)(limit-newLimit); 173 reorderStart=limit=newLimit; 174 lastCC=0; 175 } 176 void copyReorderableSuffixTo(UnicodeString &s) const { 177 s.setTo(reorderStart, (int32_t)(limit-reorderStart)); 178 } 179 private: 180 /* 181 * TODO: Revisit whether it makes sense to track reorderStart. 182 * It is set to after the last known character with cc<=1, 183 * which stops previousCC() before it reads that character and looks up its cc. 184 * previousCC() is normally only called from insert(). 185 * In other words, reorderStart speeds up the insertion of a combining mark 186 * into a multi-combining mark sequence where it does not belong at the end. 187 * This might not be worth the trouble. 188 * On the other hand, it's not a huge amount of trouble. 189 * 190 * We probably need it for UNORM_SIMPLE_APPEND. 191 */ 192 193 UBool appendSupplementary(UChar32 c, uint8_t cc, UErrorCode &errorCode); 194 void insert(UChar32 c, uint8_t cc); 195 static void writeCodePoint(UChar *p, UChar32 c) { 196 if(c<=0xffff) { 197 *p=(UChar)c; 198 } else { 199 p[0]=U16_LEAD(c); 200 p[1]=U16_TRAIL(c); 201 } 202 } 203 UBool resize(int32_t appendLength, UErrorCode &errorCode); 204 205 const Normalizer2Impl &impl; 206 UnicodeString &str; 207 UChar *start, *reorderStart, *limit; 208 int32_t remainingCapacity; 209 uint8_t lastCC; 210 211 // private backward iterator 212 void setIterator() { codePointStart=limit; } 213 void skipPrevious(); // Requires start<codePointStart. 214 uint8_t previousCC(); // Returns 0 if there is no previous character. 215 216 UChar *codePointStart, *codePointLimit; 217 }; 218 219 class U_COMMON_API Normalizer2Impl : public UObject { 220 public: 221 Normalizer2Impl() : normTrie(NULL), fCanonIterData(NULL) { 222 fCanonIterDataInitOnce.reset(); 223 } 224 virtual ~Normalizer2Impl(); 225 226 void init(const int32_t *inIndexes, const UTrie2 *inTrie, 227 const uint16_t *inExtraData, const uint8_t *inSmallFCD); 228 229 void addLcccChars(UnicodeSet &set) const; 230 void addPropertyStarts(const USetAdder *sa, UErrorCode &errorCode) const; 231 void addCanonIterPropertyStarts(const USetAdder *sa, UErrorCode &errorCode) const; 232 233 // low-level properties ------------------------------------------------ *** 234 235 const UTrie2 *getNormTrie() const { return normTrie; } 236 237 UBool ensureCanonIterData(UErrorCode &errorCode) const; 238 239 uint16_t getNorm16(UChar32 c) const { return UTRIE2_GET16(normTrie, c); } 240 241 UNormalizationCheckResult getCompQuickCheck(uint16_t norm16) const { 242 if(norm16<minNoNo || MIN_YES_YES_WITH_CC<=norm16) { 243 return UNORM_YES; 244 } else if(minMaybeYes<=norm16) { 245 return UNORM_MAYBE; 246 } else { 247 return UNORM_NO; 248 } 249 } 250 UBool isAlgorithmicNoNo(uint16_t norm16) const { return limitNoNo<=norm16 && norm16<minMaybeYes; } 251 UBool isCompNo(uint16_t norm16) const { return minNoNo<=norm16 && norm16<minMaybeYes; } 252 UBool isDecompYes(uint16_t norm16) const { return norm16<minYesNo || minMaybeYes<=norm16; } 253 254 uint8_t getCC(uint16_t norm16) const { 255 if(norm16>=MIN_NORMAL_MAYBE_YES) { 256 return (uint8_t)norm16; 257 } 258 if(norm16<minNoNo || limitNoNo<=norm16) { 259 return 0; 260 } 261 return getCCFromNoNo(norm16); 262 } 263 static uint8_t getCCFromYesOrMaybe(uint16_t norm16) { 264 return norm16>=MIN_NORMAL_MAYBE_YES ? (uint8_t)norm16 : 0; 265 } 266 267 /** 268 * Returns the FCD data for code point c. 269 * @param c A Unicode code point. 270 * @return The lccc(c) in bits 15..8 and tccc(c) in bits 7..0. 271 */ 272 uint16_t getFCD16(UChar32 c) const { 273 if(c<0) { 274 return 0; 275 } else if(c<0x180) { 276 return tccc180[c]; 277 } else if(c<=0xffff) { 278 if(!singleLeadMightHaveNonZeroFCD16(c)) { return 0; } 279 } 280 return getFCD16FromNormData(c); 281 } 282 /** 283 * Returns the FCD data for the next code point (post-increment). 284 * Might skip only a lead surrogate rather than the whole surrogate pair if none of 285 * the supplementary code points associated with the lead surrogate have non-zero FCD data. 286 * @param s A valid pointer into a string. Requires s!=limit. 287 * @param limit The end of the string, or NULL. 288 * @return The lccc(c) in bits 15..8 and tccc(c) in bits 7..0. 289 */ 290 uint16_t nextFCD16(const UChar *&s, const UChar *limit) const { 291 UChar32 c=*s++; 292 if(c<0x180) { 293 return tccc180[c]; 294 } else if(!singleLeadMightHaveNonZeroFCD16(c)) { 295 return 0; 296 } 297 UChar c2; 298 if(U16_IS_LEAD(c) && s!=limit && U16_IS_TRAIL(c2=*s)) { 299 c=U16_GET_SUPPLEMENTARY(c, c2); 300 ++s; 301 } 302 return getFCD16FromNormData(c); 303 } 304 /** 305 * Returns the FCD data for the previous code point (pre-decrement). 306 * @param start The start of the string. 307 * @param s A valid pointer into a string. Requires start<s. 308 * @return The lccc(c) in bits 15..8 and tccc(c) in bits 7..0. 309 */ 310 uint16_t previousFCD16(const UChar *start, const UChar *&s) const { 311 UChar32 c=*--s; 312 if(c<0x180) { 313 return tccc180[c]; 314 } 315 if(!U16_IS_TRAIL(c)) { 316 if(!singleLeadMightHaveNonZeroFCD16(c)) { 317 return 0; 318 } 319 } else { 320 UChar c2; 321 if(start<s && U16_IS_LEAD(c2=*(s-1))) { 322 c=U16_GET_SUPPLEMENTARY(c2, c); 323 --s; 324 } 325 } 326 return getFCD16FromNormData(c); 327 } 328 329 /** Returns the FCD data for U+0000<=c<U+0180. */ 330 uint16_t getFCD16FromBelow180(UChar32 c) const { return tccc180[c]; } 331 /** Returns TRUE if the single-or-lead code unit c might have non-zero FCD data. */ 332 UBool singleLeadMightHaveNonZeroFCD16(UChar32 lead) const { 333 // 0<=lead<=0xffff 334 uint8_t bits=smallFCD[lead>>8]; 335 if(bits==0) { return false; } 336 return (UBool)((bits>>((lead>>5)&7))&1); 337 } 338 /** Returns the FCD value from the regular normalization data. */ 339 uint16_t getFCD16FromNormData(UChar32 c) const; 340 341 void makeCanonIterDataFromNorm16(UChar32 start, UChar32 end, uint16_t norm16, 342 CanonIterData &newData, UErrorCode &errorCode) const; 343 344 /** 345 * Gets the decomposition for one code point. 346 * @param c code point 347 * @param buffer out-only buffer for algorithmic decompositions 348 * @param length out-only, takes the length of the decomposition, if any 349 * @return pointer to the decomposition, or NULL if none 350 */ 351 const UChar *getDecomposition(UChar32 c, UChar buffer[4], int32_t &length) const; 352 353 /** 354 * Gets the raw decomposition for one code point. 355 * @param c code point 356 * @param buffer out-only buffer for algorithmic decompositions 357 * @param length out-only, takes the length of the decomposition, if any 358 * @return pointer to the decomposition, or NULL if none 359 */ 360 const UChar *getRawDecomposition(UChar32 c, UChar buffer[30], int32_t &length) const; 361 362 UChar32 composePair(UChar32 a, UChar32 b) const; 363 364 UBool isCanonSegmentStarter(UChar32 c) const; 365 UBool getCanonStartSet(UChar32 c, UnicodeSet &set) const; 366 367 enum { 368 MIN_CCC_LCCC_CP=0x300 369 }; 370 371 enum { 372 MIN_YES_YES_WITH_CC=0xff01, 373 JAMO_VT=0xff00, 374 MIN_NORMAL_MAYBE_YES=0xfe00, 375 JAMO_L=1, 376 MAX_DELTA=0x40 377 }; 378 379 enum { 380 // Byte offsets from the start of the data, after the generic header. 381 IX_NORM_TRIE_OFFSET, 382 IX_EXTRA_DATA_OFFSET, 383 IX_SMALL_FCD_OFFSET, 384 IX_RESERVED3_OFFSET, 385 IX_RESERVED4_OFFSET, 386 IX_RESERVED5_OFFSET, 387 IX_RESERVED6_OFFSET, 388 IX_TOTAL_SIZE, 389 390 // Code point thresholds for quick check codes. 391 IX_MIN_DECOMP_NO_CP, 392 IX_MIN_COMP_NO_MAYBE_CP, 393 394 // Norm16 value thresholds for quick check combinations and types of extra data. 395 IX_MIN_YES_NO, // Mappings & compositions in [minYesNo..minYesNoMappingsOnly[. 396 IX_MIN_NO_NO, 397 IX_LIMIT_NO_NO, 398 IX_MIN_MAYBE_YES, 399 400 IX_MIN_YES_NO_MAPPINGS_ONLY, // Mappings only in [minYesNoMappingsOnly..minNoNo[. 401 402 IX_RESERVED15, 403 IX_COUNT 404 }; 405 406 enum { 407 MAPPING_HAS_CCC_LCCC_WORD=0x80, 408 MAPPING_HAS_RAW_MAPPING=0x40, 409 MAPPING_NO_COMP_BOUNDARY_AFTER=0x20, 410 MAPPING_LENGTH_MASK=0x1f 411 }; 412 413 enum { 414 COMP_1_LAST_TUPLE=0x8000, 415 COMP_1_TRIPLE=1, 416 COMP_1_TRAIL_LIMIT=0x3400, 417 COMP_1_TRAIL_MASK=0x7ffe, 418 COMP_1_TRAIL_SHIFT=9, // 10-1 for the "triple" bit 419 COMP_2_TRAIL_SHIFT=6, 420 COMP_2_TRAIL_MASK=0xffc0 421 }; 422 423 // higher-level functionality ------------------------------------------ *** 424 425 // NFD without an NFD Normalizer2 instance. 426 UnicodeString &decompose(const UnicodeString &src, UnicodeString &dest, 427 UErrorCode &errorCode) const; 428 /** 429 * Decomposes [src, limit[ and writes the result to dest. 430 * limit can be NULL if src is NUL-terminated. 431 * destLengthEstimate is the initial dest buffer capacity and can be -1. 432 */ 433 void decompose(const UChar *src, const UChar *limit, 434 UnicodeString &dest, int32_t destLengthEstimate, 435 UErrorCode &errorCode) const; 436 437 const UChar *decompose(const UChar *src, const UChar *limit, 438 ReorderingBuffer *buffer, UErrorCode &errorCode) const; 439 void decomposeAndAppend(const UChar *src, const UChar *limit, 440 UBool doDecompose, 441 UnicodeString &safeMiddle, 442 ReorderingBuffer &buffer, 443 UErrorCode &errorCode) const; 444 UBool compose(const UChar *src, const UChar *limit, 445 UBool onlyContiguous, 446 UBool doCompose, 447 ReorderingBuffer &buffer, 448 UErrorCode &errorCode) const; 449 const UChar *composeQuickCheck(const UChar *src, const UChar *limit, 450 UBool onlyContiguous, 451 UNormalizationCheckResult *pQCResult) const; 452 void composeAndAppend(const UChar *src, const UChar *limit, 453 UBool doCompose, 454 UBool onlyContiguous, 455 UnicodeString &safeMiddle, 456 ReorderingBuffer &buffer, 457 UErrorCode &errorCode) const; 458 const UChar *makeFCD(const UChar *src, const UChar *limit, 459 ReorderingBuffer *buffer, UErrorCode &errorCode) const; 460 void makeFCDAndAppend(const UChar *src, const UChar *limit, 461 UBool doMakeFCD, 462 UnicodeString &safeMiddle, 463 ReorderingBuffer &buffer, 464 UErrorCode &errorCode) const; 465 466 UBool hasDecompBoundary(UChar32 c, UBool before) const; 467 UBool isDecompInert(UChar32 c) const { return isDecompYesAndZeroCC(getNorm16(c)); } 468 469 UBool hasCompBoundaryBefore(UChar32 c) const { 470 return c<minCompNoMaybeCP || hasCompBoundaryBefore(c, getNorm16(c)); 471 } 472 UBool hasCompBoundaryAfter(UChar32 c, UBool onlyContiguous, UBool testInert) const; 473 474 UBool hasFCDBoundaryBefore(UChar32 c) const { return c<MIN_CCC_LCCC_CP || getFCD16(c)<=0xff; } 475 UBool hasFCDBoundaryAfter(UChar32 c) const { 476 uint16_t fcd16=getFCD16(c); 477 return fcd16<=1 || (fcd16&0xff)==0; 478 } 479 UBool isFCDInert(UChar32 c) const { return getFCD16(c)<=1; } 480 private: 481 UBool isMaybe(uint16_t norm16) const { return minMaybeYes<=norm16 && norm16<=JAMO_VT; } 482 UBool isMaybeOrNonZeroCC(uint16_t norm16) const { return norm16>=minMaybeYes; } 483 static UBool isInert(uint16_t norm16) { return norm16==0; } 484 static UBool isJamoL(uint16_t norm16) { return norm16==1; } 485 static UBool isJamoVT(uint16_t norm16) { return norm16==JAMO_VT; } 486 UBool isHangul(uint16_t norm16) const { return norm16==minYesNo; } 487 UBool isCompYesAndZeroCC(uint16_t norm16) const { return norm16<minNoNo; } 488 // UBool isCompYes(uint16_t norm16) const { 489 // return norm16>=MIN_YES_YES_WITH_CC || norm16<minNoNo; 490 // } 491 // UBool isCompYesOrMaybe(uint16_t norm16) const { 492 // return norm16<minNoNo || minMaybeYes<=norm16; 493 // } 494 // UBool hasZeroCCFromDecompYes(uint16_t norm16) const { 495 // return norm16<=MIN_NORMAL_MAYBE_YES || norm16==JAMO_VT; 496 // } 497 UBool isDecompYesAndZeroCC(uint16_t norm16) const { 498 return norm16<minYesNo || 499 norm16==JAMO_VT || 500 (minMaybeYes<=norm16 && norm16<=MIN_NORMAL_MAYBE_YES); 501 } 502 /** 503 * A little faster and simpler than isDecompYesAndZeroCC() but does not include 504 * the MaybeYes which combine-forward and have ccc=0. 505 * (Standard Unicode 5.2 normalization does not have such characters.) 506 */ 507 UBool isMostDecompYesAndZeroCC(uint16_t norm16) const { 508 return norm16<minYesNo || norm16==MIN_NORMAL_MAYBE_YES || norm16==JAMO_VT; 509 } 510 UBool isDecompNoAlgorithmic(uint16_t norm16) const { return norm16>=limitNoNo; } 511 512 // For use with isCompYes(). 513 // Perhaps the compiler can combine the two tests for MIN_YES_YES_WITH_CC. 514 // static uint8_t getCCFromYes(uint16_t norm16) { 515 // return norm16>=MIN_YES_YES_WITH_CC ? (uint8_t)norm16 : 0; 516 // } 517 uint8_t getCCFromNoNo(uint16_t norm16) const { 518 const uint16_t *mapping=getMapping(norm16); 519 if(*mapping&MAPPING_HAS_CCC_LCCC_WORD) { 520 return (uint8_t)*(mapping-1); 521 } else { 522 return 0; 523 } 524 } 525 // requires that the [cpStart..cpLimit[ character passes isCompYesAndZeroCC() 526 uint8_t getTrailCCFromCompYesAndZeroCC(const UChar *cpStart, const UChar *cpLimit) const; 527 528 // Requires algorithmic-NoNo. 529 UChar32 mapAlgorithmic(UChar32 c, uint16_t norm16) const { 530 return c+norm16-(minMaybeYes-MAX_DELTA-1); 531 } 532 533 // Requires minYesNo<norm16<limitNoNo. 534 const uint16_t *getMapping(uint16_t norm16) const { return extraData+norm16; } 535 const uint16_t *getCompositionsListForDecompYes(uint16_t norm16) const { 536 if(norm16==0 || MIN_NORMAL_MAYBE_YES<=norm16) { 537 return NULL; 538 } else if(norm16<minMaybeYes) { 539 return extraData+norm16; // for yesYes; if Jamo L: harmless empty list 540 } else { 541 return maybeYesCompositions+norm16-minMaybeYes; 542 } 543 } 544 const uint16_t *getCompositionsListForComposite(uint16_t norm16) const { 545 const uint16_t *list=extraData+norm16; // composite has both mapping & compositions list 546 return list+ // mapping pointer 547 1+ // +1 to skip the first unit with the mapping lenth 548 (*list&MAPPING_LENGTH_MASK); // + mapping length 549 } 550 /** 551 * @param c code point must have compositions 552 * @return compositions list pointer 553 */ 554 const uint16_t *getCompositionsList(uint16_t norm16) const { 555 return isDecompYes(norm16) ? 556 getCompositionsListForDecompYes(norm16) : 557 getCompositionsListForComposite(norm16); 558 } 559 560 const UChar *copyLowPrefixFromNulTerminated(const UChar *src, 561 UChar32 minNeedDataCP, 562 ReorderingBuffer *buffer, 563 UErrorCode &errorCode) const; 564 UBool decomposeShort(const UChar *src, const UChar *limit, 565 ReorderingBuffer &buffer, UErrorCode &errorCode) const; 566 UBool decompose(UChar32 c, uint16_t norm16, 567 ReorderingBuffer &buffer, UErrorCode &errorCode) const; 568 569 static int32_t combine(const uint16_t *list, UChar32 trail); 570 void addComposites(const uint16_t *list, UnicodeSet &set) const; 571 void recompose(ReorderingBuffer &buffer, int32_t recomposeStartIndex, 572 UBool onlyContiguous) const; 573 574 UBool hasCompBoundaryBefore(UChar32 c, uint16_t norm16) const; 575 const UChar *findPreviousCompBoundary(const UChar *start, const UChar *p) const; 576 const UChar *findNextCompBoundary(const UChar *p, const UChar *limit) const; 577 578 const UChar *findPreviousFCDBoundary(const UChar *start, const UChar *p) const; 579 const UChar *findNextFCDBoundary(const UChar *p, const UChar *limit) const; 580 581 int32_t getCanonValue(UChar32 c) const; 582 const UnicodeSet &getCanonStartSet(int32_t n) const; 583 584 // UVersionInfo dataVersion; 585 586 // Code point thresholds for quick check codes. 587 UChar32 minDecompNoCP; 588 UChar32 minCompNoMaybeCP; 589 590 // Norm16 value thresholds for quick check combinations and types of extra data. 591 uint16_t minYesNo; 592 uint16_t minYesNoMappingsOnly; 593 uint16_t minNoNo; 594 uint16_t limitNoNo; 595 uint16_t minMaybeYes; 596 597 const UTrie2 *normTrie; 598 const uint16_t *maybeYesCompositions; 599 const uint16_t *extraData; // mappings and/or compositions for yesYes, yesNo & noNo characters 600 const uint8_t *smallFCD; // [0x100] one bit per 32 BMP code points, set if any FCD!=0 601 uint8_t tccc180[0x180]; // tccc values for U+0000..U+017F 602 603 public: // CanonIterData is public to allow access from C callback functions. 604 UInitOnce fCanonIterDataInitOnce; 605 CanonIterData *fCanonIterData; 606 }; 607 608 // bits in canonIterData 609 #define CANON_NOT_SEGMENT_STARTER 0x80000000 610 #define CANON_HAS_COMPOSITIONS 0x40000000 611 #define CANON_HAS_SET 0x200000 612 #define CANON_VALUE_MASK 0x1fffff 613 614 /** 615 * ICU-internal shortcut for quick access to standard Unicode normalization. 616 */ 617 class U_COMMON_API Normalizer2Factory { 618 public: 619 static const Normalizer2 *getFCDInstance(UErrorCode &errorCode); 620 static const Normalizer2 *getFCCInstance(UErrorCode &errorCode); 621 static const Normalizer2 *getNoopInstance(UErrorCode &errorCode); 622 623 static const Normalizer2 *getInstance(UNormalizationMode mode, UErrorCode &errorCode); 624 625 static const Normalizer2Impl *getNFCImpl(UErrorCode &errorCode); 626 static const Normalizer2Impl *getNFKCImpl(UErrorCode &errorCode); 627 static const Normalizer2Impl *getNFKC_CFImpl(UErrorCode &errorCode); 628 629 // Get the Impl instance of the Normalizer2. 630 // Must be used only when it is known that norm2 is a Normalizer2WithImpl instance. 631 static const Normalizer2Impl *getImpl(const Normalizer2 *norm2); 632 private: 633 Normalizer2Factory(); // No instantiation. 634 }; 635 636 U_NAMESPACE_END 637 638 U_CAPI int32_t U_EXPORT2 639 unorm2_swap(const UDataSwapper *ds, 640 const void *inData, int32_t length, void *outData, 641 UErrorCode *pErrorCode); 642 643 /** 644 * Get the NF*_QC property for a code point, for u_getIntPropertyValue(). 645 * @internal 646 */ 647 U_CFUNC UNormalizationCheckResult 648 unorm_getQuickCheck(UChar32 c, UNormalizationMode mode); 649 650 /** 651 * Gets the 16-bit FCD value (lead & trail CCs) for a code point, for u_getIntPropertyValue(). 652 * @internal 653 */ 654 U_CFUNC uint16_t 655 unorm_getFCD16(UChar32 c); 656 657 /** 658 * Format of Normalizer2 .nrm data files. 659 * Format version 2.0. 660 * 661 * Normalizer2 .nrm data files provide data for the Unicode Normalization algorithms. 662 * ICU ships with data files for standard Unicode Normalization Forms 663 * NFC and NFD (nfc.nrm), NFKC and NFKD (nfkc.nrm) and NFKC_Casefold (nfkc_cf.nrm). 664 * Custom (application-specific) data can be built into additional .nrm files 665 * with the gennorm2 build tool. 666 * 667 * Normalizer2.getInstance() causes a .nrm file to be loaded, unless it has been 668 * cached already. Internally, Normalizer2Impl.load() reads the .nrm file. 669 * 670 * A .nrm file begins with a standard ICU data file header 671 * (DataHeader, see ucmndata.h and unicode/udata.h). 672 * The UDataInfo.dataVersion field usually contains the Unicode version 673 * for which the data was generated. 674 * 675 * After the header, the file contains the following parts. 676 * Constants are defined as enum values of the Normalizer2Impl class. 677 * 678 * Many details of the data structures are described in the design doc 679 * which is at http://site.icu-project.org/design/normalization/custom 680 * 681 * int32_t indexes[indexesLength]; -- indexesLength=indexes[IX_NORM_TRIE_OFFSET]/4; 682 * 683 * The first eight indexes are byte offsets in ascending order. 684 * Each byte offset marks the start of the next part in the data file, 685 * and the end of the previous one. 686 * When two consecutive byte offsets are the same, then the corresponding part is empty. 687 * Byte offsets are offsets from after the header, 688 * that is, from the beginning of the indexes[]. 689 * Each part starts at an offset with proper alignment for its data. 690 * If necessary, the previous part may include padding bytes to achieve this alignment. 691 * 692 * minDecompNoCP=indexes[IX_MIN_DECOMP_NO_CP] is the lowest code point 693 * with a decomposition mapping, that is, with NF*D_QC=No. 694 * minCompNoMaybeCP=indexes[IX_MIN_COMP_NO_MAYBE_CP] is the lowest code point 695 * with NF*C_QC=No (has a one-way mapping) or Maybe (combines backward). 696 * 697 * The next five indexes are thresholds of 16-bit trie values for ranges of 698 * values indicating multiple normalization properties. 699 * minYesNo=indexes[IX_MIN_YES_NO]; 700 * minNoNo=indexes[IX_MIN_NO_NO]; 701 * limitNoNo=indexes[IX_LIMIT_NO_NO]; 702 * minMaybeYes=indexes[IX_MIN_MAYBE_YES]; 703 * minYesNoMappingsOnly=indexes[IX_MIN_YES_NO_MAPPINGS_ONLY]; 704 * See the normTrie description below and the design doc for details. 705 * 706 * UTrie2 normTrie; -- see utrie2_impl.h and utrie2.h 707 * 708 * The trie holds the main normalization data. Each code point is mapped to a 16-bit value. 709 * Rather than using independent bits in the value (which would require more than 16 bits), 710 * information is extracted primarily via range checks. 711 * For example, a 16-bit value norm16 in the range minYesNo<=norm16<minNoNo 712 * means that the character has NF*C_QC=Yes and NF*D_QC=No properties, 713 * which means it has a two-way (round-trip) decomposition mapping. 714 * Values in the range 2<=norm16<limitNoNo are also directly indexes into the extraData 715 * pointing to mappings, compositions lists, or both. 716 * Value norm16==0 means that the character is normalization-inert, that is, 717 * it does not have a mapping, does not participate in composition, has a zero 718 * canonical combining class, and forms a boundary where text before it and after it 719 * can be normalized independently. 720 * For details about how multiple properties are encoded in 16-bit values 721 * see the design doc. 722 * Note that the encoding cannot express all combinations of the properties involved; 723 * it only supports those combinations that are allowed by 724 * the Unicode Normalization algorithms. Details are in the design doc as well. 725 * The gennorm2 tool only builds .nrm files for data that conforms to the limitations. 726 * 727 * The trie has a value for each lead surrogate code unit representing the "worst case" 728 * properties of the 1024 supplementary characters whose UTF-16 form starts with 729 * the lead surrogate. If all of the 1024 supplementary characters are normalization-inert, 730 * then their lead surrogate code unit has the trie value 0. 731 * When the lead surrogate unit's value exceeds the quick check minimum during processing, 732 * the properties for the full supplementary code point need to be looked up. 733 * 734 * uint16_t maybeYesCompositions[MIN_NORMAL_MAYBE_YES-minMaybeYes]; 735 * uint16_t extraData[]; 736 * 737 * There is only one byte offset for the end of these two arrays. 738 * The split between them is given by the constant and variable mentioned above. 739 * 740 * The maybeYesCompositions array contains compositions lists for characters that 741 * combine both forward (as starters in composition pairs) 742 * and backward (as trailing characters in composition pairs). 743 * Such characters do not occur in Unicode 5.2 but are allowed by 744 * the Unicode Normalization algorithms. 745 * If there are no such characters, then minMaybeYes==MIN_NORMAL_MAYBE_YES 746 * and the maybeYesCompositions array is empty. 747 * If there are such characters, then minMaybeYes is subtracted from their norm16 values 748 * to get the index into this array. 749 * 750 * The extraData array contains compositions lists for "YesYes" characters, 751 * followed by mappings and optional compositions lists for "YesNo" characters, 752 * followed by only mappings for "NoNo" characters. 753 * (Referring to pairs of NFC/NFD quick check values.) 754 * The norm16 values of those characters are directly indexes into the extraData array. 755 * 756 * The data structures for compositions lists and mappings are described in the design doc. 757 * 758 * uint8_t smallFCD[0x100]; -- new in format version 2 759 * 760 * This is a bit set to help speed up FCD value lookups in the absence of a full 761 * UTrie2 or other large data structure with the full FCD value mapping. 762 * 763 * Each smallFCD bit is set if any of the corresponding 32 BMP code points 764 * has a non-zero FCD value (lccc!=0 or tccc!=0). 765 * Bit 0 of smallFCD[0] is for U+0000..U+001F. Bit 7 of smallFCD[0xff] is for U+FFE0..U+FFFF. 766 * A bit for 32 lead surrogates is set if any of the 32k corresponding 767 * _supplementary_ code points has a non-zero FCD value. 768 * 769 * This bit set is most useful for the large blocks of CJK characters with FCD=0. 770 * 771 * Changes from format version 1 to format version 2 --------------------------- 772 * 773 * - Addition of data for raw (not recursively decomposed) mappings. 774 * + The MAPPING_NO_COMP_BOUNDARY_AFTER bit in the extraData is now also set when 775 * the mapping is to an empty string or when the character combines-forward. 776 * This subsumes the one actual use of the MAPPING_PLUS_COMPOSITION_LIST bit which 777 * is then repurposed for the MAPPING_HAS_RAW_MAPPING bit. 778 * + For details see the design doc. 779 * - Addition of indexes[IX_MIN_YES_NO_MAPPINGS_ONLY] and separation of the yesNo extraData into 780 * distinct ranges (combines-forward vs. not) 781 * so that a range check can be used to find out if there is a compositions list. 782 * This is fully equivalent with formatVersion 1's MAPPING_PLUS_COMPOSITION_LIST flag. 783 * It is needed for the new (in ICU 49) composePair(), not for other normalization. 784 * - Addition of the smallFCD[] bit set. 785 */ 786 787 #endif /* !UCONFIG_NO_NORMALIZATION */ 788 #endif /* __NORMALIZER2IMPL_H__ */ 789
