1 // 2016 and later: Unicode, Inc. and others. 2 // License & terms of use: http://www.unicode.org/copyright.html 3 /* 4 ******************************************************************************* 5 * Copyright (C) 2012-2015, International Business Machines 6 * Corporation and others. All Rights Reserved. 7 ******************************************************************************* 8 * collationdatabuilder.cpp 9 * 10 * (replaced the former ucol_elm.cpp) 11 * 12 * created on: 2012apr01 13 * created by: Markus W. Scherer 14 */ 15 16 #include "unicode/utypes.h" 17 18 #if !UCONFIG_NO_COLLATION 19 20 #include "unicode/localpointer.h" 21 #include "unicode/uchar.h" 22 #include "unicode/ucharstrie.h" 23 #include "unicode/ucharstriebuilder.h" 24 #include "unicode/uniset.h" 25 #include "unicode/unistr.h" 26 #include "unicode/usetiter.h" 27 #include "unicode/utf16.h" 28 #include "cmemory.h" 29 #include "collation.h" 30 #include "collationdata.h" 31 #include "collationdatabuilder.h" 32 #include "collationfastlatinbuilder.h" 33 #include "collationiterator.h" 34 #include "normalizer2impl.h" 35 #include "utrie2.h" 36 #include "uvectr32.h" 37 #include "uvectr64.h" 38 #include "uvector.h" 39 40 U_NAMESPACE_BEGIN 41 42 CollationDataBuilder::CEModifier::~CEModifier() {} 43 44 /** 45 * Build-time context and CE32 for a code point. 46 * If a code point has contextual mappings, then the default (no-context) mapping 47 * and all conditional mappings are stored in a singly-linked list 48 * of ConditionalCE32, sorted by context strings. 49 * 50 * Context strings sort by prefix length, then by prefix, then by contraction suffix. 51 * Context strings must be unique and in ascending order. 52 */ 53 struct ConditionalCE32 : public UMemory { 54 ConditionalCE32() 55 : context(), 56 ce32(0), defaultCE32(Collation::NO_CE32), builtCE32(Collation::NO_CE32), 57 next(-1) {} 58 ConditionalCE32(const UnicodeString &ct, uint32_t ce) 59 : context(ct), 60 ce32(ce), defaultCE32(Collation::NO_CE32), builtCE32(Collation::NO_CE32), 61 next(-1) {} 62 63 inline UBool hasContext() const { return context.length() > 1; } 64 inline int32_t prefixLength() const { return context.charAt(0); } 65 66 /** 67 * "\0" for the first entry for any code point, with its default CE32. 68 * 69 * Otherwise one unit with the length of the prefix string, 70 * then the prefix string, then the contraction suffix. 71 */ 72 UnicodeString context; 73 /** 74 * CE32 for the code point and its context. 75 * Can be special (e.g., for an expansion) but not contextual (prefix or contraction tag). 76 */ 77 uint32_t ce32; 78 /** 79 * Default CE32 for all contexts with this same prefix. 80 * Initially NO_CE32. Set only while building runtime data structures, 81 * and only on one of the nodes of a sub-list with the same prefix. 82 */ 83 uint32_t defaultCE32; 84 /** 85 * CE32 for the built contexts. 86 * When fetching CEs from the builder, the contexts are built into their runtime form 87 * so that the normal collation implementation can process them. 88 * The result is cached in the list head. It is reset when the contexts are modified. 89 */ 90 uint32_t builtCE32; 91 /** 92 * Index of the next ConditionalCE32. 93 * Negative for the end of the list. 94 */ 95 int32_t next; 96 }; 97 98 U_CDECL_BEGIN 99 100 U_CAPI void U_CALLCONV 101 uprv_deleteConditionalCE32(void *obj) { 102 delete static_cast<ConditionalCE32 *>(obj); 103 } 104 105 U_CDECL_END 106 107 /** 108 * Build-time collation element and character iterator. 109 * Uses the runtime CollationIterator for fetching CEs for a string 110 * but reads from the builder's unfinished data structures. 111 * In particular, this class reads from the unfinished trie 112 * and has to avoid CollationIterator::nextCE() and redirect other 113 * calls to data->getCE32() and data->getCE32FromSupplementary(). 114 * 115 * We do this so that we need not implement the collation algorithm 116 * again for the builder and make it behave exactly like the runtime code. 117 * That would be more difficult to test and maintain than this indirection. 118 * 119 * Some CE32 tags (for example, the DIGIT_TAG) do not occur in the builder data, 120 * so the data accesses from those code paths need not be modified. 121 * 122 * This class iterates directly over whole code points 123 * so that the CollationIterator does not need the finished trie 124 * for handling the LEAD_SURROGATE_TAG. 125 */ 126 class DataBuilderCollationIterator : public CollationIterator { 127 public: 128 DataBuilderCollationIterator(CollationDataBuilder &b); 129 130 virtual ~DataBuilderCollationIterator(); 131 132 int32_t fetchCEs(const UnicodeString &str, int32_t start, int64_t ces[], int32_t cesLength); 133 134 virtual void resetToOffset(int32_t newOffset); 135 virtual int32_t getOffset() const; 136 137 virtual UChar32 nextCodePoint(UErrorCode &errorCode); 138 virtual UChar32 previousCodePoint(UErrorCode &errorCode); 139 140 protected: 141 virtual void forwardNumCodePoints(int32_t num, UErrorCode &errorCode); 142 virtual void backwardNumCodePoints(int32_t num, UErrorCode &errorCode); 143 144 virtual uint32_t getDataCE32(UChar32 c) const; 145 virtual uint32_t getCE32FromBuilderData(uint32_t ce32, UErrorCode &errorCode); 146 147 CollationDataBuilder &builder; 148 CollationData builderData; 149 uint32_t jamoCE32s[CollationData::JAMO_CE32S_LENGTH]; 150 const UnicodeString *s; 151 int32_t pos; 152 }; 153 154 DataBuilderCollationIterator::DataBuilderCollationIterator(CollationDataBuilder &b) 155 : CollationIterator(&builderData, /*numeric=*/ FALSE), 156 builder(b), builderData(b.nfcImpl), 157 s(NULL), pos(0) { 158 builderData.base = builder.base; 159 // Set all of the jamoCE32s[] to indirection CE32s. 160 for(int32_t j = 0; j < CollationData::JAMO_CE32S_LENGTH; ++j) { // Count across Jamo types. 161 UChar32 jamo = CollationDataBuilder::jamoCpFromIndex(j); 162 jamoCE32s[j] = Collation::makeCE32FromTagAndIndex(Collation::BUILDER_DATA_TAG, jamo) | 163 CollationDataBuilder::IS_BUILDER_JAMO_CE32; 164 } 165 builderData.jamoCE32s = jamoCE32s; 166 } 167 168 DataBuilderCollationIterator::~DataBuilderCollationIterator() {} 169 170 int32_t 171 DataBuilderCollationIterator::fetchCEs(const UnicodeString &str, int32_t start, 172 int64_t ces[], int32_t cesLength) { 173 // Set the pointers each time, in case they changed due to reallocation. 174 builderData.ce32s = reinterpret_cast<const uint32_t *>(builder.ce32s.getBuffer()); 175 builderData.ces = builder.ce64s.getBuffer(); 176 builderData.contexts = builder.contexts.getBuffer(); 177 // Modified copy of CollationIterator::nextCE() and CollationIterator::nextCEFromCE32(). 178 reset(); 179 s = &str; 180 pos = start; 181 UErrorCode errorCode = U_ZERO_ERROR; 182 while(U_SUCCESS(errorCode) && pos < s->length()) { 183 // No need to keep all CEs in the iterator buffer. 184 clearCEs(); 185 UChar32 c = s->char32At(pos); 186 pos += U16_LENGTH(c); 187 uint32_t ce32 = utrie2_get32(builder.trie, c); 188 const CollationData *d; 189 if(ce32 == Collation::FALLBACK_CE32) { 190 d = builder.base; 191 ce32 = builder.base->getCE32(c); 192 } else { 193 d = &builderData; 194 } 195 appendCEsFromCE32(d, c, ce32, /*forward=*/ TRUE, errorCode); 196 U_ASSERT(U_SUCCESS(errorCode)); 197 for(int32_t i = 0; i < getCEsLength(); ++i) { 198 int64_t ce = getCE(i); 199 if(ce != 0) { 200 if(cesLength < Collation::MAX_EXPANSION_LENGTH) { 201 ces[cesLength] = ce; 202 } 203 ++cesLength; 204 } 205 } 206 } 207 return cesLength; 208 } 209 210 void 211 DataBuilderCollationIterator::resetToOffset(int32_t newOffset) { 212 reset(); 213 pos = newOffset; 214 } 215 216 int32_t 217 DataBuilderCollationIterator::getOffset() const { 218 return pos; 219 } 220 221 UChar32 222 DataBuilderCollationIterator::nextCodePoint(UErrorCode & /*errorCode*/) { 223 if(pos == s->length()) { 224 return U_SENTINEL; 225 } 226 UChar32 c = s->char32At(pos); 227 pos += U16_LENGTH(c); 228 return c; 229 } 230 231 UChar32 232 DataBuilderCollationIterator::previousCodePoint(UErrorCode & /*errorCode*/) { 233 if(pos == 0) { 234 return U_SENTINEL; 235 } 236 UChar32 c = s->char32At(pos - 1); 237 pos -= U16_LENGTH(c); 238 return c; 239 } 240 241 void 242 DataBuilderCollationIterator::forwardNumCodePoints(int32_t num, UErrorCode & /*errorCode*/) { 243 pos = s->moveIndex32(pos, num); 244 } 245 246 void 247 DataBuilderCollationIterator::backwardNumCodePoints(int32_t num, UErrorCode & /*errorCode*/) { 248 pos = s->moveIndex32(pos, -num); 249 } 250 251 uint32_t 252 DataBuilderCollationIterator::getDataCE32(UChar32 c) const { 253 return utrie2_get32(builder.trie, c); 254 } 255 256 uint32_t 257 DataBuilderCollationIterator::getCE32FromBuilderData(uint32_t ce32, UErrorCode &errorCode) { 258 U_ASSERT(Collation::hasCE32Tag(ce32, Collation::BUILDER_DATA_TAG)); 259 if((ce32 & CollationDataBuilder::IS_BUILDER_JAMO_CE32) != 0) { 260 UChar32 jamo = Collation::indexFromCE32(ce32); 261 return utrie2_get32(builder.trie, jamo); 262 } else { 263 ConditionalCE32 *cond = builder.getConditionalCE32ForCE32(ce32); 264 if(cond->builtCE32 == Collation::NO_CE32) { 265 // Build the context-sensitive mappings into their runtime form and cache the result. 266 cond->builtCE32 = builder.buildContext(cond, errorCode); 267 if(errorCode == U_BUFFER_OVERFLOW_ERROR) { 268 errorCode = U_ZERO_ERROR; 269 builder.clearContexts(); 270 cond->builtCE32 = builder.buildContext(cond, errorCode); 271 } 272 builderData.contexts = builder.contexts.getBuffer(); 273 } 274 return cond->builtCE32; 275 } 276 } 277 278 // ------------------------------------------------------------------------- *** 279 280 CollationDataBuilder::CollationDataBuilder(UErrorCode &errorCode) 281 : nfcImpl(*Normalizer2Factory::getNFCImpl(errorCode)), 282 base(NULL), baseSettings(NULL), 283 trie(NULL), 284 ce32s(errorCode), ce64s(errorCode), conditionalCE32s(errorCode), 285 modified(FALSE), 286 fastLatinEnabled(FALSE), fastLatinBuilder(NULL), 287 collIter(NULL) { 288 // Reserve the first CE32 for U+0000. 289 ce32s.addElement(0, errorCode); 290 conditionalCE32s.setDeleter(uprv_deleteConditionalCE32); 291 } 292 293 CollationDataBuilder::~CollationDataBuilder() { 294 utrie2_close(trie); 295 delete fastLatinBuilder; 296 delete collIter; 297 } 298 299 void 300 CollationDataBuilder::initForTailoring(const CollationData *b, UErrorCode &errorCode) { 301 if(U_FAILURE(errorCode)) { return; } 302 if(trie != NULL) { 303 errorCode = U_INVALID_STATE_ERROR; 304 return; 305 } 306 if(b == NULL) { 307 errorCode = U_ILLEGAL_ARGUMENT_ERROR; 308 return; 309 } 310 base = b; 311 312 // For a tailoring, the default is to fall back to the base. 313 trie = utrie2_open(Collation::FALLBACK_CE32, Collation::FFFD_CE32, &errorCode); 314 315 // Set the Latin-1 letters block so that it is allocated first in the data array, 316 // to try to improve locality of reference when sorting Latin-1 text. 317 // Do not use utrie2_setRange32() since that will not actually allocate blocks 318 // that are filled with the default value. 319 // ASCII (0..7F) is already preallocated anyway. 320 for(UChar32 c = 0xc0; c <= 0xff; ++c) { 321 utrie2_set32(trie, c, Collation::FALLBACK_CE32, &errorCode); 322 } 323 324 // Hangul syllables are not tailorable (except via tailoring Jamos). 325 // Always set the Hangul tag to help performance. 326 // Do this here, rather than in buildMappings(), 327 // so that we see the HANGUL_TAG in various assertions. 328 uint32_t hangulCE32 = Collation::makeCE32FromTagAndIndex(Collation::HANGUL_TAG, 0); 329 utrie2_setRange32(trie, Hangul::HANGUL_BASE, Hangul::HANGUL_END, hangulCE32, TRUE, &errorCode); 330 331 // Copy the set contents but don't copy/clone the set as a whole because 332 // that would copy the isFrozen state too. 333 unsafeBackwardSet.addAll(*b->unsafeBackwardSet); 334 335 if(U_FAILURE(errorCode)) { return; } 336 } 337 338 UBool 339 CollationDataBuilder::maybeSetPrimaryRange(UChar32 start, UChar32 end, 340 uint32_t primary, int32_t step, 341 UErrorCode &errorCode) { 342 if(U_FAILURE(errorCode)) { return FALSE; } 343 U_ASSERT(start <= end); 344 // TODO: Do we need to check what values are currently set for start..end? 345 // An offset range is worth it only if we can achieve an overlap between 346 // adjacent UTrie2 blocks of 32 code points each. 347 // An offset CE is also a little more expensive to look up and compute 348 // than a simple CE. 349 // If the range spans at least three UTrie2 block boundaries (> 64 code points), 350 // then we take it. 351 // If the range spans one or two block boundaries and there are 352 // at least 4 code points on either side, then we take it. 353 // (We could additionally require a minimum range length of, say, 16.) 354 int32_t blockDelta = (end >> 5) - (start >> 5); 355 if(2 <= step && step <= 0x7f && 356 (blockDelta >= 3 || 357 (blockDelta > 0 && (start & 0x1f) <= 0x1c && (end & 0x1f) >= 3))) { 358 int64_t dataCE = ((int64_t)primary << 32) | (start << 8) | step; 359 if(isCompressiblePrimary(primary)) { dataCE |= 0x80; } 360 int32_t index = addCE(dataCE, errorCode); 361 if(U_FAILURE(errorCode)) { return 0; } 362 if(index > Collation::MAX_INDEX) { 363 errorCode = U_BUFFER_OVERFLOW_ERROR; 364 return 0; 365 } 366 uint32_t offsetCE32 = Collation::makeCE32FromTagAndIndex(Collation::OFFSET_TAG, index); 367 utrie2_setRange32(trie, start, end, offsetCE32, TRUE, &errorCode); 368 modified = TRUE; 369 return TRUE; 370 } else { 371 return FALSE; 372 } 373 } 374 375 uint32_t 376 CollationDataBuilder::setPrimaryRangeAndReturnNext(UChar32 start, UChar32 end, 377 uint32_t primary, int32_t step, 378 UErrorCode &errorCode) { 379 if(U_FAILURE(errorCode)) { return 0; } 380 UBool isCompressible = isCompressiblePrimary(primary); 381 if(maybeSetPrimaryRange(start, end, primary, step, errorCode)) { 382 return Collation::incThreeBytePrimaryByOffset(primary, isCompressible, 383 (end - start + 1) * step); 384 } else { 385 // Short range: Set individual CE32s. 386 for(;;) { 387 utrie2_set32(trie, start, Collation::makeLongPrimaryCE32(primary), &errorCode); 388 ++start; 389 primary = Collation::incThreeBytePrimaryByOffset(primary, isCompressible, step); 390 if(start > end) { return primary; } 391 } 392 modified = TRUE; 393 } 394 } 395 396 uint32_t 397 CollationDataBuilder::getCE32FromOffsetCE32(UBool fromBase, UChar32 c, uint32_t ce32) const { 398 int32_t i = Collation::indexFromCE32(ce32); 399 int64_t dataCE = fromBase ? base->ces[i] : ce64s.elementAti(i); 400 uint32_t p = Collation::getThreeBytePrimaryForOffsetData(c, dataCE); 401 return Collation::makeLongPrimaryCE32(p); 402 } 403 404 UBool 405 CollationDataBuilder::isCompressibleLeadByte(uint32_t b) const { 406 return base->isCompressibleLeadByte(b); 407 } 408 409 UBool 410 CollationDataBuilder::isAssigned(UChar32 c) const { 411 return Collation::isAssignedCE32(utrie2_get32(trie, c)); 412 } 413 414 uint32_t 415 CollationDataBuilder::getLongPrimaryIfSingleCE(UChar32 c) const { 416 uint32_t ce32 = utrie2_get32(trie, c); 417 if(Collation::isLongPrimaryCE32(ce32)) { 418 return Collation::primaryFromLongPrimaryCE32(ce32); 419 } else { 420 return 0; 421 } 422 } 423 424 int64_t 425 CollationDataBuilder::getSingleCE(UChar32 c, UErrorCode &errorCode) const { 426 if(U_FAILURE(errorCode)) { return 0; } 427 // Keep parallel with CollationData::getSingleCE(). 428 UBool fromBase = FALSE; 429 uint32_t ce32 = utrie2_get32(trie, c); 430 if(ce32 == Collation::FALLBACK_CE32) { 431 fromBase = TRUE; 432 ce32 = base->getCE32(c); 433 } 434 while(Collation::isSpecialCE32(ce32)) { 435 switch(Collation::tagFromCE32(ce32)) { 436 case Collation::LATIN_EXPANSION_TAG: 437 case Collation::BUILDER_DATA_TAG: 438 case Collation::PREFIX_TAG: 439 case Collation::CONTRACTION_TAG: 440 case Collation::HANGUL_TAG: 441 case Collation::LEAD_SURROGATE_TAG: 442 errorCode = U_UNSUPPORTED_ERROR; 443 return 0; 444 case Collation::FALLBACK_TAG: 445 case Collation::RESERVED_TAG_3: 446 errorCode = U_INTERNAL_PROGRAM_ERROR; 447 return 0; 448 case Collation::LONG_PRIMARY_TAG: 449 return Collation::ceFromLongPrimaryCE32(ce32); 450 case Collation::LONG_SECONDARY_TAG: 451 return Collation::ceFromLongSecondaryCE32(ce32); 452 case Collation::EXPANSION32_TAG: 453 if(Collation::lengthFromCE32(ce32) == 1) { 454 int32_t i = Collation::indexFromCE32(ce32); 455 ce32 = fromBase ? base->ce32s[i] : ce32s.elementAti(i); 456 break; 457 } else { 458 errorCode = U_UNSUPPORTED_ERROR; 459 return 0; 460 } 461 case Collation::EXPANSION_TAG: { 462 if(Collation::lengthFromCE32(ce32) == 1) { 463 int32_t i = Collation::indexFromCE32(ce32); 464 return fromBase ? base->ces[i] : ce64s.elementAti(i); 465 } else { 466 errorCode = U_UNSUPPORTED_ERROR; 467 return 0; 468 } 469 } 470 case Collation::DIGIT_TAG: 471 // Fetch the non-numeric-collation CE32 and continue. 472 ce32 = ce32s.elementAti(Collation::indexFromCE32(ce32)); 473 break; 474 case Collation::U0000_TAG: 475 U_ASSERT(c == 0); 476 // Fetch the normal ce32 for U+0000 and continue. 477 ce32 = fromBase ? base->ce32s[0] : ce32s.elementAti(0); 478 break; 479 case Collation::OFFSET_TAG: 480 ce32 = getCE32FromOffsetCE32(fromBase, c, ce32); 481 break; 482 case Collation::IMPLICIT_TAG: 483 return Collation::unassignedCEFromCodePoint(c); 484 } 485 } 486 return Collation::ceFromSimpleCE32(ce32); 487 } 488 489 int32_t 490 CollationDataBuilder::addCE(int64_t ce, UErrorCode &errorCode) { 491 int32_t length = ce64s.size(); 492 for(int32_t i = 0; i < length; ++i) { 493 if(ce == ce64s.elementAti(i)) { return i; } 494 } 495 ce64s.addElement(ce, errorCode); 496 return length; 497 } 498 499 int32_t 500 CollationDataBuilder::addCE32(uint32_t ce32, UErrorCode &errorCode) { 501 int32_t length = ce32s.size(); 502 for(int32_t i = 0; i < length; ++i) { 503 if(ce32 == (uint32_t)ce32s.elementAti(i)) { return i; } 504 } 505 ce32s.addElement((int32_t)ce32, errorCode); 506 return length; 507 } 508 509 int32_t 510 CollationDataBuilder::addConditionalCE32(const UnicodeString &context, uint32_t ce32, 511 UErrorCode &errorCode) { 512 if(U_FAILURE(errorCode)) { return -1; } 513 U_ASSERT(!context.isEmpty()); 514 int32_t index = conditionalCE32s.size(); 515 if(index > Collation::MAX_INDEX) { 516 errorCode = U_BUFFER_OVERFLOW_ERROR; 517 return -1; 518 } 519 ConditionalCE32 *cond = new ConditionalCE32(context, ce32); 520 if(cond == NULL) { 521 errorCode = U_MEMORY_ALLOCATION_ERROR; 522 return -1; 523 } 524 conditionalCE32s.addElement(cond, errorCode); 525 return index; 526 } 527 528 void 529 CollationDataBuilder::add(const UnicodeString &prefix, const UnicodeString &s, 530 const int64_t ces[], int32_t cesLength, 531 UErrorCode &errorCode) { 532 uint32_t ce32 = encodeCEs(ces, cesLength, errorCode); 533 addCE32(prefix, s, ce32, errorCode); 534 } 535 536 void 537 CollationDataBuilder::addCE32(const UnicodeString &prefix, const UnicodeString &s, 538 uint32_t ce32, UErrorCode &errorCode) { 539 if(U_FAILURE(errorCode)) { return; } 540 if(s.isEmpty()) { 541 errorCode = U_ILLEGAL_ARGUMENT_ERROR; 542 return; 543 } 544 if(trie == NULL || utrie2_isFrozen(trie)) { 545 errorCode = U_INVALID_STATE_ERROR; 546 return; 547 } 548 UChar32 c = s.char32At(0); 549 int32_t cLength = U16_LENGTH(c); 550 uint32_t oldCE32 = utrie2_get32(trie, c); 551 UBool hasContext = !prefix.isEmpty() || s.length() > cLength; 552 if(oldCE32 == Collation::FALLBACK_CE32) { 553 // First tailoring for c. 554 // If c has contextual base mappings or if we add a contextual mapping, 555 // then copy the base mappings. 556 // Otherwise we just override the base mapping. 557 uint32_t baseCE32 = base->getFinalCE32(base->getCE32(c)); 558 if(hasContext || Collation::ce32HasContext(baseCE32)) { 559 oldCE32 = copyFromBaseCE32(c, baseCE32, TRUE, errorCode); 560 utrie2_set32(trie, c, oldCE32, &errorCode); 561 if(U_FAILURE(errorCode)) { return; } 562 } 563 } 564 if(!hasContext) { 565 // No prefix, no contraction. 566 if(!isBuilderContextCE32(oldCE32)) { 567 utrie2_set32(trie, c, ce32, &errorCode); 568 } else { 569 ConditionalCE32 *cond = getConditionalCE32ForCE32(oldCE32); 570 cond->builtCE32 = Collation::NO_CE32; 571 cond->ce32 = ce32; 572 } 573 } else { 574 ConditionalCE32 *cond; 575 if(!isBuilderContextCE32(oldCE32)) { 576 // Replace the simple oldCE32 with a builder context CE32 577 // pointing to a new ConditionalCE32 list head. 578 int32_t index = addConditionalCE32(UnicodeString((UChar)0), oldCE32, errorCode); 579 if(U_FAILURE(errorCode)) { return; } 580 uint32_t contextCE32 = makeBuilderContextCE32(index); 581 utrie2_set32(trie, c, contextCE32, &errorCode); 582 contextChars.add(c); 583 cond = getConditionalCE32(index); 584 } else { 585 cond = getConditionalCE32ForCE32(oldCE32); 586 cond->builtCE32 = Collation::NO_CE32; 587 } 588 UnicodeString suffix(s, cLength); 589 UnicodeString context((UChar)prefix.length()); 590 context.append(prefix).append(suffix); 591 unsafeBackwardSet.addAll(suffix); 592 for(;;) { 593 // invariant: context > cond->context 594 int32_t next = cond->next; 595 if(next < 0) { 596 // Append a new ConditionalCE32 after cond. 597 int32_t index = addConditionalCE32(context, ce32, errorCode); 598 if(U_FAILURE(errorCode)) { return; } 599 cond->next = index; 600 break; 601 } 602 ConditionalCE32 *nextCond = getConditionalCE32(next); 603 int8_t cmp = context.compare(nextCond->context); 604 if(cmp < 0) { 605 // Insert a new ConditionalCE32 between cond and nextCond. 606 int32_t index = addConditionalCE32(context, ce32, errorCode); 607 if(U_FAILURE(errorCode)) { return; } 608 cond->next = index; 609 getConditionalCE32(index)->next = next; 610 break; 611 } else if(cmp == 0) { 612 // Same context as before, overwrite its ce32. 613 nextCond->ce32 = ce32; 614 break; 615 } 616 cond = nextCond; 617 } 618 } 619 modified = TRUE; 620 } 621 622 uint32_t 623 CollationDataBuilder::encodeOneCEAsCE32(int64_t ce) { 624 uint32_t p = (uint32_t)(ce >> 32); 625 uint32_t lower32 = (uint32_t)ce; 626 uint32_t t = (uint32_t)(ce & 0xffff); 627 U_ASSERT((t & 0xc000) != 0xc000); // Impossible case bits 11 mark special CE32s. 628 if((ce & INT64_C(0xffff00ff00ff)) == 0) { 629 // normal form ppppsstt 630 return p | (lower32 >> 16) | (t >> 8); 631 } else if((ce & INT64_C(0xffffffffff)) == Collation::COMMON_SEC_AND_TER_CE) { 632 // long-primary form ppppppC1 633 return Collation::makeLongPrimaryCE32(p); 634 } else if(p == 0 && (t & 0xff) == 0) { 635 // long-secondary form ssssttC2 636 return Collation::makeLongSecondaryCE32(lower32); 637 } 638 return Collation::NO_CE32; 639 } 640 641 uint32_t 642 CollationDataBuilder::encodeOneCE(int64_t ce, UErrorCode &errorCode) { 643 // Try to encode one CE as one CE32. 644 uint32_t ce32 = encodeOneCEAsCE32(ce); 645 if(ce32 != Collation::NO_CE32) { return ce32; } 646 int32_t index = addCE(ce, errorCode); 647 if(U_FAILURE(errorCode)) { return 0; } 648 if(index > Collation::MAX_INDEX) { 649 errorCode = U_BUFFER_OVERFLOW_ERROR; 650 return 0; 651 } 652 return Collation::makeCE32FromTagIndexAndLength(Collation::EXPANSION_TAG, index, 1); 653 } 654 655 uint32_t 656 CollationDataBuilder::encodeCEs(const int64_t ces[], int32_t cesLength, 657 UErrorCode &errorCode) { 658 if(U_FAILURE(errorCode)) { return 0; } 659 if(cesLength < 0 || cesLength > Collation::MAX_EXPANSION_LENGTH) { 660 errorCode = U_ILLEGAL_ARGUMENT_ERROR; 661 return 0; 662 } 663 if(trie == NULL || utrie2_isFrozen(trie)) { 664 errorCode = U_INVALID_STATE_ERROR; 665 return 0; 666 } 667 if(cesLength == 0) { 668 // Convenience: We cannot map to nothing, but we can map to a completely ignorable CE. 669 // Do this here so that callers need not do it. 670 return encodeOneCEAsCE32(0); 671 } else if(cesLength == 1) { 672 return encodeOneCE(ces[0], errorCode); 673 } else if(cesLength == 2) { 674 // Try to encode two CEs as one CE32. 675 int64_t ce0 = ces[0]; 676 int64_t ce1 = ces[1]; 677 uint32_t p0 = (uint32_t)(ce0 >> 32); 678 if((ce0 & INT64_C(0xffffffffff00ff)) == Collation::COMMON_SECONDARY_CE && 679 (ce1 & INT64_C(0xffffffff00ffffff)) == Collation::COMMON_TERTIARY_CE && 680 p0 != 0) { 681 // Latin mini expansion 682 return 683 p0 | 684 (((uint32_t)ce0 & 0xff00u) << 8) | 685 (uint32_t)(ce1 >> 16) | 686 Collation::SPECIAL_CE32_LOW_BYTE | 687 Collation::LATIN_EXPANSION_TAG; 688 } 689 } 690 // Try to encode two or more CEs as CE32s. 691 int32_t newCE32s[Collation::MAX_EXPANSION_LENGTH]; 692 for(int32_t i = 0;; ++i) { 693 if(i == cesLength) { 694 return encodeExpansion32(newCE32s, cesLength, errorCode); 695 } 696 uint32_t ce32 = encodeOneCEAsCE32(ces[i]); 697 if(ce32 == Collation::NO_CE32) { break; } 698 newCE32s[i] = (int32_t)ce32; 699 } 700 return encodeExpansion(ces, cesLength, errorCode); 701 } 702 703 uint32_t 704 CollationDataBuilder::encodeExpansion(const int64_t ces[], int32_t length, UErrorCode &errorCode) { 705 if(U_FAILURE(errorCode)) { return 0; } 706 // See if this sequence of CEs has already been stored. 707 int64_t first = ces[0]; 708 int32_t ce64sMax = ce64s.size() - length; 709 for(int32_t i = 0; i <= ce64sMax; ++i) { 710 if(first == ce64s.elementAti(i)) { 711 if(i > Collation::MAX_INDEX) { 712 errorCode = U_BUFFER_OVERFLOW_ERROR; 713 return 0; 714 } 715 for(int32_t j = 1;; ++j) { 716 if(j == length) { 717 return Collation::makeCE32FromTagIndexAndLength( 718 Collation::EXPANSION_TAG, i, length); 719 } 720 if(ce64s.elementAti(i + j) != ces[j]) { break; } 721 } 722 } 723 } 724 // Store the new sequence. 725 int32_t i = ce64s.size(); 726 if(i > Collation::MAX_INDEX) { 727 errorCode = U_BUFFER_OVERFLOW_ERROR; 728 return 0; 729 } 730 for(int32_t j = 0; j < length; ++j) { 731 ce64s.addElement(ces[j], errorCode); 732 } 733 return Collation::makeCE32FromTagIndexAndLength(Collation::EXPANSION_TAG, i, length); 734 } 735 736 uint32_t 737 CollationDataBuilder::encodeExpansion32(const int32_t newCE32s[], int32_t length, 738 UErrorCode &errorCode) { 739 if(U_FAILURE(errorCode)) { return 0; } 740 // See if this sequence of CE32s has already been stored. 741 int32_t first = newCE32s[0]; 742 int32_t ce32sMax = ce32s.size() - length; 743 for(int32_t i = 0; i <= ce32sMax; ++i) { 744 if(first == ce32s.elementAti(i)) { 745 if(i > Collation::MAX_INDEX) { 746 errorCode = U_BUFFER_OVERFLOW_ERROR; 747 return 0; 748 } 749 for(int32_t j = 1;; ++j) { 750 if(j == length) { 751 return Collation::makeCE32FromTagIndexAndLength( 752 Collation::EXPANSION32_TAG, i, length); 753 } 754 if(ce32s.elementAti(i + j) != newCE32s[j]) { break; } 755 } 756 } 757 } 758 // Store the new sequence. 759 int32_t i = ce32s.size(); 760 if(i > Collation::MAX_INDEX) { 761 errorCode = U_BUFFER_OVERFLOW_ERROR; 762 return 0; 763 } 764 for(int32_t j = 0; j < length; ++j) { 765 ce32s.addElement(newCE32s[j], errorCode); 766 } 767 return Collation::makeCE32FromTagIndexAndLength(Collation::EXPANSION32_TAG, i, length); 768 } 769 770 uint32_t 771 CollationDataBuilder::copyFromBaseCE32(UChar32 c, uint32_t ce32, UBool withContext, 772 UErrorCode &errorCode) { 773 if(U_FAILURE(errorCode)) { return 0; } 774 if(!Collation::isSpecialCE32(ce32)) { return ce32; } 775 switch(Collation::tagFromCE32(ce32)) { 776 case Collation::LONG_PRIMARY_TAG: 777 case Collation::LONG_SECONDARY_TAG: 778 case Collation::LATIN_EXPANSION_TAG: 779 // copy as is 780 break; 781 case Collation::EXPANSION32_TAG: { 782 const uint32_t *baseCE32s = base->ce32s + Collation::indexFromCE32(ce32); 783 int32_t length = Collation::lengthFromCE32(ce32); 784 ce32 = encodeExpansion32( 785 reinterpret_cast<const int32_t *>(baseCE32s), length, errorCode); 786 break; 787 } 788 case Collation::EXPANSION_TAG: { 789 const int64_t *baseCEs = base->ces + Collation::indexFromCE32(ce32); 790 int32_t length = Collation::lengthFromCE32(ce32); 791 ce32 = encodeExpansion(baseCEs, length, errorCode); 792 break; 793 } 794 case Collation::PREFIX_TAG: { 795 // Flatten prefixes and nested suffixes (contractions) 796 // into a linear list of ConditionalCE32. 797 const UChar *p = base->contexts + Collation::indexFromCE32(ce32); 798 ce32 = CollationData::readCE32(p); // Default if no prefix match. 799 if(!withContext) { 800 return copyFromBaseCE32(c, ce32, FALSE, errorCode); 801 } 802 ConditionalCE32 head; 803 UnicodeString context((UChar)0); 804 int32_t index; 805 if(Collation::isContractionCE32(ce32)) { 806 index = copyContractionsFromBaseCE32(context, c, ce32, &head, errorCode); 807 } else { 808 ce32 = copyFromBaseCE32(c, ce32, TRUE, errorCode); 809 head.next = index = addConditionalCE32(context, ce32, errorCode); 810 } 811 if(U_FAILURE(errorCode)) { return 0; } 812 ConditionalCE32 *cond = getConditionalCE32(index); // the last ConditionalCE32 so far 813 UCharsTrie::Iterator prefixes(p + 2, 0, errorCode); 814 while(prefixes.next(errorCode)) { 815 context = prefixes.getString(); 816 context.reverse(); 817 context.insert(0, (UChar)context.length()); 818 ce32 = (uint32_t)prefixes.getValue(); 819 if(Collation::isContractionCE32(ce32)) { 820 index = copyContractionsFromBaseCE32(context, c, ce32, cond, errorCode); 821 } else { 822 ce32 = copyFromBaseCE32(c, ce32, TRUE, errorCode); 823 cond->next = index = addConditionalCE32(context, ce32, errorCode); 824 } 825 if(U_FAILURE(errorCode)) { return 0; } 826 cond = getConditionalCE32(index); 827 } 828 ce32 = makeBuilderContextCE32(head.next); 829 contextChars.add(c); 830 break; 831 } 832 case Collation::CONTRACTION_TAG: { 833 if(!withContext) { 834 const UChar *p = base->contexts + Collation::indexFromCE32(ce32); 835 ce32 = CollationData::readCE32(p); // Default if no suffix match. 836 return copyFromBaseCE32(c, ce32, FALSE, errorCode); 837 } 838 ConditionalCE32 head; 839 UnicodeString context((UChar)0); 840 copyContractionsFromBaseCE32(context, c, ce32, &head, errorCode); 841 ce32 = makeBuilderContextCE32(head.next); 842 contextChars.add(c); 843 break; 844 } 845 case Collation::HANGUL_TAG: 846 errorCode = U_UNSUPPORTED_ERROR; // We forbid tailoring of Hangul syllables. 847 break; 848 case Collation::OFFSET_TAG: 849 ce32 = getCE32FromOffsetCE32(TRUE, c, ce32); 850 break; 851 case Collation::IMPLICIT_TAG: 852 ce32 = encodeOneCE(Collation::unassignedCEFromCodePoint(c), errorCode); 853 break; 854 default: 855 U_ASSERT(FALSE); // require ce32 == base->getFinalCE32(ce32) 856 break; 857 } 858 return ce32; 859 } 860 861 int32_t 862 CollationDataBuilder::copyContractionsFromBaseCE32(UnicodeString &context, UChar32 c, uint32_t ce32, 863 ConditionalCE32 *cond, UErrorCode &errorCode) { 864 if(U_FAILURE(errorCode)) { return 0; } 865 const UChar *p = base->contexts + Collation::indexFromCE32(ce32); 866 int32_t index; 867 if((ce32 & Collation::CONTRACT_SINGLE_CP_NO_MATCH) != 0) { 868 // No match on the single code point. 869 // We are underneath a prefix, and the default mapping is just 870 // a fallback to the mappings for a shorter prefix. 871 U_ASSERT(context.length() > 1); 872 index = -1; 873 } else { 874 ce32 = CollationData::readCE32(p); // Default if no suffix match. 875 U_ASSERT(!Collation::isContractionCE32(ce32)); 876 ce32 = copyFromBaseCE32(c, ce32, TRUE, errorCode); 877 cond->next = index = addConditionalCE32(context, ce32, errorCode); 878 if(U_FAILURE(errorCode)) { return 0; } 879 cond = getConditionalCE32(index); 880 } 881 882 int32_t suffixStart = context.length(); 883 UCharsTrie::Iterator suffixes(p + 2, 0, errorCode); 884 while(suffixes.next(errorCode)) { 885 context.append(suffixes.getString()); 886 ce32 = copyFromBaseCE32(c, (uint32_t)suffixes.getValue(), TRUE, errorCode); 887 cond->next = index = addConditionalCE32(context, ce32, errorCode); 888 if(U_FAILURE(errorCode)) { return 0; } 889 // No need to update the unsafeBackwardSet because the tailoring set 890 // is already a copy of the base set. 891 cond = getConditionalCE32(index); 892 context.truncate(suffixStart); 893 } 894 U_ASSERT(index >= 0); 895 return index; 896 } 897 898 class CopyHelper { 899 public: 900 CopyHelper(const CollationDataBuilder &s, CollationDataBuilder &d, 901 const CollationDataBuilder::CEModifier &m, UErrorCode &initialErrorCode) 902 : src(s), dest(d), modifier(m), 903 errorCode(initialErrorCode) {} 904 905 UBool copyRangeCE32(UChar32 start, UChar32 end, uint32_t ce32) { 906 ce32 = copyCE32(ce32); 907 utrie2_setRange32(dest.trie, start, end, ce32, TRUE, &errorCode); 908 if(CollationDataBuilder::isBuilderContextCE32(ce32)) { 909 dest.contextChars.add(start, end); 910 } 911 return U_SUCCESS(errorCode); 912 } 913 914 uint32_t copyCE32(uint32_t ce32) { 915 if(!Collation::isSpecialCE32(ce32)) { 916 int64_t ce = modifier.modifyCE32(ce32); 917 if(ce != Collation::NO_CE) { 918 ce32 = dest.encodeOneCE(ce, errorCode); 919 } 920 } else { 921 int32_t tag = Collation::tagFromCE32(ce32); 922 if(tag == Collation::EXPANSION32_TAG) { 923 const uint32_t *srcCE32s = reinterpret_cast<uint32_t *>(src.ce32s.getBuffer()); 924 srcCE32s += Collation::indexFromCE32(ce32); 925 int32_t length = Collation::lengthFromCE32(ce32); 926 // Inspect the source CE32s. Just copy them if none are modified. 927 // Otherwise copy to modifiedCEs, with modifications. 928 UBool isModified = FALSE; 929 for(int32_t i = 0; i < length; ++i) { 930 ce32 = srcCE32s[i]; 931 int64_t ce; 932 if(Collation::isSpecialCE32(ce32) || 933 (ce = modifier.modifyCE32(ce32)) == Collation::NO_CE) { 934 if(isModified) { 935 modifiedCEs[i] = Collation::ceFromCE32(ce32); 936 } 937 } else { 938 if(!isModified) { 939 for(int32_t j = 0; j < i; ++j) { 940 modifiedCEs[j] = Collation::ceFromCE32(srcCE32s[j]); 941 } 942 isModified = TRUE; 943 } 944 modifiedCEs[i] = ce; 945 } 946 } 947 if(isModified) { 948 ce32 = dest.encodeCEs(modifiedCEs, length, errorCode); 949 } else { 950 ce32 = dest.encodeExpansion32( 951 reinterpret_cast<const int32_t *>(srcCE32s), length, errorCode); 952 } 953 } else if(tag == Collation::EXPANSION_TAG) { 954 const int64_t *srcCEs = src.ce64s.getBuffer(); 955 srcCEs += Collation::indexFromCE32(ce32); 956 int32_t length = Collation::lengthFromCE32(ce32); 957 // Inspect the source CEs. Just copy them if none are modified. 958 // Otherwise copy to modifiedCEs, with modifications. 959 UBool isModified = FALSE; 960 for(int32_t i = 0; i < length; ++i) { 961 int64_t srcCE = srcCEs[i]; 962 int64_t ce = modifier.modifyCE(srcCE); 963 if(ce == Collation::NO_CE) { 964 if(isModified) { 965 modifiedCEs[i] = srcCE; 966 } 967 } else { 968 if(!isModified) { 969 for(int32_t j = 0; j < i; ++j) { 970 modifiedCEs[j] = srcCEs[j]; 971 } 972 isModified = TRUE; 973 } 974 modifiedCEs[i] = ce; 975 } 976 } 977 if(isModified) { 978 ce32 = dest.encodeCEs(modifiedCEs, length, errorCode); 979 } else { 980 ce32 = dest.encodeExpansion(srcCEs, length, errorCode); 981 } 982 } else if(tag == Collation::BUILDER_DATA_TAG) { 983 // Copy the list of ConditionalCE32. 984 ConditionalCE32 *cond = src.getConditionalCE32ForCE32(ce32); 985 U_ASSERT(!cond->hasContext()); 986 int32_t destIndex = dest.addConditionalCE32( 987 cond->context, copyCE32(cond->ce32), errorCode); 988 ce32 = CollationDataBuilder::makeBuilderContextCE32(destIndex); 989 while(cond->next >= 0) { 990 cond = src.getConditionalCE32(cond->next); 991 ConditionalCE32 *prevDestCond = dest.getConditionalCE32(destIndex); 992 destIndex = dest.addConditionalCE32( 993 cond->context, copyCE32(cond->ce32), errorCode); 994 int32_t suffixStart = cond->prefixLength() + 1; 995 dest.unsafeBackwardSet.addAll(cond->context.tempSubString(suffixStart)); 996 prevDestCond->next = destIndex; 997 } 998 } else { 999 // Just copy long CEs and Latin mini expansions (and other expected values) as is, 1000 // assuming that the modifier would not modify them. 1001 U_ASSERT(tag == Collation::LONG_PRIMARY_TAG || 1002 tag == Collation::LONG_SECONDARY_TAG || 1003 tag == Collation::LATIN_EXPANSION_TAG || 1004 tag == Collation::HANGUL_TAG); 1005 } 1006 } 1007 return ce32; 1008 } 1009 1010 const CollationDataBuilder &src; 1011 CollationDataBuilder &dest; 1012 const CollationDataBuilder::CEModifier &modifier; 1013 int64_t modifiedCEs[Collation::MAX_EXPANSION_LENGTH]; 1014 UErrorCode errorCode; 1015 }; 1016 1017 U_CDECL_BEGIN 1018 1019 static UBool U_CALLCONV 1020 enumRangeForCopy(const void *context, UChar32 start, UChar32 end, uint32_t value) { 1021 return 1022 value == Collation::UNASSIGNED_CE32 || value == Collation::FALLBACK_CE32 || 1023 ((CopyHelper *)context)->copyRangeCE32(start, end, value); 1024 } 1025 1026 U_CDECL_END 1027 1028 void 1029 CollationDataBuilder::copyFrom(const CollationDataBuilder &src, const CEModifier &modifier, 1030 UErrorCode &errorCode) { 1031 if(U_FAILURE(errorCode)) { return; } 1032 if(trie == NULL || utrie2_isFrozen(trie)) { 1033 errorCode = U_INVALID_STATE_ERROR; 1034 return; 1035 } 1036 CopyHelper helper(src, *this, modifier, errorCode); 1037 utrie2_enum(src.trie, NULL, enumRangeForCopy, &helper); 1038 errorCode = helper.errorCode; 1039 // Update the contextChars and the unsafeBackwardSet while copying, 1040 // in case a character had conditional mappings in the source builder 1041 // and they were removed later. 1042 modified |= src.modified; 1043 } 1044 1045 void 1046 CollationDataBuilder::optimize(const UnicodeSet &set, UErrorCode &errorCode) { 1047 if(U_FAILURE(errorCode) || set.isEmpty()) { return; } 1048 UnicodeSetIterator iter(set); 1049 while(iter.next() && !iter.isString()) { 1050 UChar32 c = iter.getCodepoint(); 1051 uint32_t ce32 = utrie2_get32(trie, c); 1052 if(ce32 == Collation::FALLBACK_CE32) { 1053 ce32 = base->getFinalCE32(base->getCE32(c)); 1054 ce32 = copyFromBaseCE32(c, ce32, TRUE, errorCode); 1055 utrie2_set32(trie, c, ce32, &errorCode); 1056 } 1057 } 1058 modified = TRUE; 1059 } 1060 1061 void 1062 CollationDataBuilder::suppressContractions(const UnicodeSet &set, UErrorCode &errorCode) { 1063 if(U_FAILURE(errorCode) || set.isEmpty()) { return; } 1064 UnicodeSetIterator iter(set); 1065 while(iter.next() && !iter.isString()) { 1066 UChar32 c = iter.getCodepoint(); 1067 uint32_t ce32 = utrie2_get32(trie, c); 1068 if(ce32 == Collation::FALLBACK_CE32) { 1069 ce32 = base->getFinalCE32(base->getCE32(c)); 1070 if(Collation::ce32HasContext(ce32)) { 1071 ce32 = copyFromBaseCE32(c, ce32, FALSE /* without context */, errorCode); 1072 utrie2_set32(trie, c, ce32, &errorCode); 1073 } 1074 } else if(isBuilderContextCE32(ce32)) { 1075 ce32 = getConditionalCE32ForCE32(ce32)->ce32; 1076 // Simply abandon the list of ConditionalCE32. 1077 // The caller will copy this builder in the end, 1078 // eliminating unreachable data. 1079 utrie2_set32(trie, c, ce32, &errorCode); 1080 contextChars.remove(c); 1081 } 1082 } 1083 modified = TRUE; 1084 } 1085 1086 UBool 1087 CollationDataBuilder::getJamoCE32s(uint32_t jamoCE32s[], UErrorCode &errorCode) { 1088 if(U_FAILURE(errorCode)) { return FALSE; } 1089 UBool anyJamoAssigned = base == NULL; // always set jamoCE32s in the base data 1090 UBool needToCopyFromBase = FALSE; 1091 for(int32_t j = 0; j < CollationData::JAMO_CE32S_LENGTH; ++j) { // Count across Jamo types. 1092 UChar32 jamo = jamoCpFromIndex(j); 1093 UBool fromBase = FALSE; 1094 uint32_t ce32 = utrie2_get32(trie, jamo); 1095 anyJamoAssigned |= Collation::isAssignedCE32(ce32); 1096 // TODO: Try to prevent [optimize [Jamo]] from counting as anyJamoAssigned. 1097 // (As of CLDR 24 [2013] the Korean tailoring does not optimize conjoining Jamo.) 1098 if(ce32 == Collation::FALLBACK_CE32) { 1099 fromBase = TRUE; 1100 ce32 = base->getCE32(jamo); 1101 } 1102 if(Collation::isSpecialCE32(ce32)) { 1103 switch(Collation::tagFromCE32(ce32)) { 1104 case Collation::LONG_PRIMARY_TAG: 1105 case Collation::LONG_SECONDARY_TAG: 1106 case Collation::LATIN_EXPANSION_TAG: 1107 // Copy the ce32 as-is. 1108 break; 1109 case Collation::EXPANSION32_TAG: 1110 case Collation::EXPANSION_TAG: 1111 case Collation::PREFIX_TAG: 1112 case Collation::CONTRACTION_TAG: 1113 if(fromBase) { 1114 // Defer copying until we know if anyJamoAssigned. 1115 ce32 = Collation::FALLBACK_CE32; 1116 needToCopyFromBase = TRUE; 1117 } 1118 break; 1119 case Collation::IMPLICIT_TAG: 1120 // An unassigned Jamo should only occur in tests with incomplete bases. 1121 U_ASSERT(fromBase); 1122 ce32 = Collation::FALLBACK_CE32; 1123 needToCopyFromBase = TRUE; 1124 break; 1125 case Collation::OFFSET_TAG: 1126 ce32 = getCE32FromOffsetCE32(fromBase, jamo, ce32); 1127 break; 1128 case Collation::FALLBACK_TAG: 1129 case Collation::RESERVED_TAG_3: 1130 case Collation::BUILDER_DATA_TAG: 1131 case Collation::DIGIT_TAG: 1132 case Collation::U0000_TAG: 1133 case Collation::HANGUL_TAG: 1134 case Collation::LEAD_SURROGATE_TAG: 1135 errorCode = U_INTERNAL_PROGRAM_ERROR; 1136 return FALSE; 1137 } 1138 } 1139 jamoCE32s[j] = ce32; 1140 } 1141 if(anyJamoAssigned && needToCopyFromBase) { 1142 for(int32_t j = 0; j < CollationData::JAMO_CE32S_LENGTH; ++j) { 1143 if(jamoCE32s[j] == Collation::FALLBACK_CE32) { 1144 UChar32 jamo = jamoCpFromIndex(j); 1145 jamoCE32s[j] = copyFromBaseCE32(jamo, base->getCE32(jamo), 1146 /*withContext=*/ TRUE, errorCode); 1147 } 1148 } 1149 } 1150 return anyJamoAssigned && U_SUCCESS(errorCode); 1151 } 1152 1153 void 1154 CollationDataBuilder::setDigitTags(UErrorCode &errorCode) { 1155 UnicodeSet digits(UNICODE_STRING_SIMPLE("[:Nd:]"), errorCode); 1156 if(U_FAILURE(errorCode)) { return; } 1157 UnicodeSetIterator iter(digits); 1158 while(iter.next()) { 1159 U_ASSERT(!iter.isString()); 1160 UChar32 c = iter.getCodepoint(); 1161 uint32_t ce32 = utrie2_get32(trie, c); 1162 if(ce32 != Collation::FALLBACK_CE32 && ce32 != Collation::UNASSIGNED_CE32) { 1163 int32_t index = addCE32(ce32, errorCode); 1164 if(U_FAILURE(errorCode)) { return; } 1165 if(index > Collation::MAX_INDEX) { 1166 errorCode = U_BUFFER_OVERFLOW_ERROR; 1167 return; 1168 } 1169 ce32 = Collation::makeCE32FromTagIndexAndLength( 1170 Collation::DIGIT_TAG, index, u_charDigitValue(c)); 1171 utrie2_set32(trie, c, ce32, &errorCode); 1172 } 1173 } 1174 } 1175 1176 U_CDECL_BEGIN 1177 1178 static UBool U_CALLCONV 1179 enumRangeLeadValue(const void *context, UChar32 /*start*/, UChar32 /*end*/, uint32_t value) { 1180 int32_t *pValue = (int32_t *)context; 1181 if(value == Collation::UNASSIGNED_CE32) { 1182 value = Collation::LEAD_ALL_UNASSIGNED; 1183 } else if(value == Collation::FALLBACK_CE32) { 1184 value = Collation::LEAD_ALL_FALLBACK; 1185 } else { 1186 *pValue = Collation::LEAD_MIXED; 1187 return FALSE; 1188 } 1189 if(*pValue < 0) { 1190 *pValue = (int32_t)value; 1191 } else if(*pValue != (int32_t)value) { 1192 *pValue = Collation::LEAD_MIXED; 1193 return FALSE; 1194 } 1195 return TRUE; 1196 } 1197 1198 U_CDECL_END 1199 1200 void 1201 CollationDataBuilder::setLeadSurrogates(UErrorCode &errorCode) { 1202 for(UChar lead = 0xd800; lead < 0xdc00; ++lead) { 1203 int32_t value = -1; 1204 utrie2_enumForLeadSurrogate(trie, lead, NULL, enumRangeLeadValue, &value); 1205 utrie2_set32ForLeadSurrogateCodeUnit( 1206 trie, lead, 1207 Collation::makeCE32FromTagAndIndex(Collation::LEAD_SURROGATE_TAG, 0) | (uint32_t)value, 1208 &errorCode); 1209 } 1210 } 1211 1212 void 1213 CollationDataBuilder::build(CollationData &data, UErrorCode &errorCode) { 1214 buildMappings(data, errorCode); 1215 if(base != NULL) { 1216 data.numericPrimary = base->numericPrimary; 1217 data.compressibleBytes = base->compressibleBytes; 1218 data.numScripts = base->numScripts; 1219 data.scriptsIndex = base->scriptsIndex; 1220 data.scriptStarts = base->scriptStarts; 1221 data.scriptStartsLength = base->scriptStartsLength; 1222 } 1223 buildFastLatinTable(data, errorCode); 1224 } 1225 1226 void 1227 CollationDataBuilder::buildMappings(CollationData &data, UErrorCode &errorCode) { 1228 if(U_FAILURE(errorCode)) { return; } 1229 if(trie == NULL || utrie2_isFrozen(trie)) { 1230 errorCode = U_INVALID_STATE_ERROR; 1231 return; 1232 } 1233 1234 buildContexts(errorCode); 1235 1236 uint32_t jamoCE32s[CollationData::JAMO_CE32S_LENGTH]; 1237 int32_t jamoIndex = -1; 1238 if(getJamoCE32s(jamoCE32s, errorCode)) { 1239 jamoIndex = ce32s.size(); 1240 for(int32_t i = 0; i < CollationData::JAMO_CE32S_LENGTH; ++i) { 1241 ce32s.addElement((int32_t)jamoCE32s[i], errorCode); 1242 } 1243 // Small optimization: Use a bit in the Hangul ce32 1244 // to indicate that none of the Jamo CE32s are isSpecialCE32() 1245 // (as it should be in the root collator). 1246 // It allows CollationIterator to avoid recursive function calls and per-Jamo tests. 1247 // In order to still have good trie compression and keep this code simple, 1248 // we only set this flag if a whole block of 588 Hangul syllables starting with 1249 // a common leading consonant (Jamo L) has this property. 1250 UBool isAnyJamoVTSpecial = FALSE; 1251 for(int32_t i = Hangul::JAMO_L_COUNT; i < CollationData::JAMO_CE32S_LENGTH; ++i) { 1252 if(Collation::isSpecialCE32(jamoCE32s[i])) { 1253 isAnyJamoVTSpecial = TRUE; 1254 break; 1255 } 1256 } 1257 uint32_t hangulCE32 = Collation::makeCE32FromTagAndIndex(Collation::HANGUL_TAG, 0); 1258 UChar32 c = Hangul::HANGUL_BASE; 1259 for(int32_t i = 0; i < Hangul::JAMO_L_COUNT; ++i) { // iterate over the Jamo L 1260 uint32_t ce32 = hangulCE32; 1261 if(!isAnyJamoVTSpecial && !Collation::isSpecialCE32(jamoCE32s[i])) { 1262 ce32 |= Collation::HANGUL_NO_SPECIAL_JAMO; 1263 } 1264 UChar32 limit = c + Hangul::JAMO_VT_COUNT; 1265 utrie2_setRange32(trie, c, limit - 1, ce32, TRUE, &errorCode); 1266 c = limit; 1267 } 1268 } else { 1269 // Copy the Hangul CE32s from the base in blocks per Jamo L, 1270 // assuming that HANGUL_NO_SPECIAL_JAMO is set or not set for whole blocks. 1271 for(UChar32 c = Hangul::HANGUL_BASE; c < Hangul::HANGUL_LIMIT;) { 1272 uint32_t ce32 = base->getCE32(c); 1273 U_ASSERT(Collation::hasCE32Tag(ce32, Collation::HANGUL_TAG)); 1274 UChar32 limit = c + Hangul::JAMO_VT_COUNT; 1275 utrie2_setRange32(trie, c, limit - 1, ce32, TRUE, &errorCode); 1276 c = limit; 1277 } 1278 } 1279 1280 setDigitTags(errorCode); 1281 setLeadSurrogates(errorCode); 1282 1283 // For U+0000, move its normal ce32 into CE32s[0] and set U0000_TAG. 1284 ce32s.setElementAt((int32_t)utrie2_get32(trie, 0), 0); 1285 utrie2_set32(trie, 0, Collation::makeCE32FromTagAndIndex(Collation::U0000_TAG, 0), &errorCode); 1286 1287 utrie2_freeze(trie, UTRIE2_32_VALUE_BITS, &errorCode); 1288 if(U_FAILURE(errorCode)) { return; } 1289 1290 // Mark each lead surrogate as "unsafe" 1291 // if any of its 1024 associated supplementary code points is "unsafe". 1292 UChar32 c = 0x10000; 1293 for(UChar lead = 0xd800; lead < 0xdc00; ++lead, c += 0x400) { 1294 if(unsafeBackwardSet.containsSome(c, c + 0x3ff)) { 1295 unsafeBackwardSet.add(lead); 1296 } 1297 } 1298 unsafeBackwardSet.freeze(); 1299 1300 data.trie = trie; 1301 data.ce32s = reinterpret_cast<const uint32_t *>(ce32s.getBuffer()); 1302 data.ces = ce64s.getBuffer(); 1303 data.contexts = contexts.getBuffer(); 1304 1305 data.ce32sLength = ce32s.size(); 1306 data.cesLength = ce64s.size(); 1307 data.contextsLength = contexts.length(); 1308 1309 data.base = base; 1310 if(jamoIndex >= 0) { 1311 data.jamoCE32s = data.ce32s + jamoIndex; 1312 } else { 1313 data.jamoCE32s = base->jamoCE32s; 1314 } 1315 data.unsafeBackwardSet = &unsafeBackwardSet; 1316 } 1317 1318 void 1319 CollationDataBuilder::clearContexts() { 1320 contexts.remove(); 1321 UnicodeSetIterator iter(contextChars); 1322 while(iter.next()) { 1323 U_ASSERT(!iter.isString()); 1324 uint32_t ce32 = utrie2_get32(trie, iter.getCodepoint()); 1325 U_ASSERT(isBuilderContextCE32(ce32)); 1326 getConditionalCE32ForCE32(ce32)->builtCE32 = Collation::NO_CE32; 1327 } 1328 } 1329 1330 void 1331 CollationDataBuilder::buildContexts(UErrorCode &errorCode) { 1332 if(U_FAILURE(errorCode)) { return; } 1333 // Ignore abandoned lists and the cached builtCE32, 1334 // and build all contexts from scratch. 1335 contexts.remove(); 1336 UnicodeSetIterator iter(contextChars); 1337 while(U_SUCCESS(errorCode) && iter.next()) { 1338 U_ASSERT(!iter.isString()); 1339 UChar32 c = iter.getCodepoint(); 1340 uint32_t ce32 = utrie2_get32(trie, c); 1341 if(!isBuilderContextCE32(ce32)) { 1342 // Impossible: No context data for c in contextChars. 1343 errorCode = U_INTERNAL_PROGRAM_ERROR; 1344 return; 1345 } 1346 ConditionalCE32 *cond = getConditionalCE32ForCE32(ce32); 1347 ce32 = buildContext(cond, errorCode); 1348 utrie2_set32(trie, c, ce32, &errorCode); 1349 } 1350 } 1351 1352 uint32_t 1353 CollationDataBuilder::buildContext(ConditionalCE32 *head, UErrorCode &errorCode) { 1354 if(U_FAILURE(errorCode)) { return 0; } 1355 // The list head must have no context. 1356 U_ASSERT(!head->hasContext()); 1357 // The list head must be followed by one or more nodes that all do have context. 1358 U_ASSERT(head->next >= 0); 1359 UCharsTrieBuilder prefixBuilder(errorCode); 1360 UCharsTrieBuilder contractionBuilder(errorCode); 1361 for(ConditionalCE32 *cond = head;; cond = getConditionalCE32(cond->next)) { 1362 // After the list head, the prefix or suffix can be empty, but not both. 1363 U_ASSERT(cond == head || cond->hasContext()); 1364 int32_t prefixLength = cond->prefixLength(); 1365 UnicodeString prefix(cond->context, 0, prefixLength + 1); 1366 // Collect all contraction suffixes for one prefix. 1367 ConditionalCE32 *firstCond = cond; 1368 ConditionalCE32 *lastCond = cond; 1369 while(cond->next >= 0 && 1370 (cond = getConditionalCE32(cond->next))->context.startsWith(prefix)) { 1371 lastCond = cond; 1372 } 1373 uint32_t ce32; 1374 int32_t suffixStart = prefixLength + 1; // == prefix.length() 1375 if(lastCond->context.length() == suffixStart) { 1376 // One prefix without contraction suffix. 1377 U_ASSERT(firstCond == lastCond); 1378 ce32 = lastCond->ce32; 1379 cond = lastCond; 1380 } else { 1381 // Build the contractions trie. 1382 contractionBuilder.clear(); 1383 // Entry for an empty suffix, to be stored before the trie. 1384 uint32_t emptySuffixCE32 = 0; 1385 uint32_t flags = 0; 1386 if(firstCond->context.length() == suffixStart) { 1387 // There is a mapping for the prefix and the single character c. (p|c) 1388 // If no other suffix matches, then we return this value. 1389 emptySuffixCE32 = firstCond->ce32; 1390 cond = getConditionalCE32(firstCond->next); 1391 } else { 1392 // There is no mapping for the prefix and just the single character. 1393 // (There is no p|c, only p|cd, p|ce etc.) 1394 flags |= Collation::CONTRACT_SINGLE_CP_NO_MATCH; 1395 // When the prefix matches but none of the prefix-specific suffixes, 1396 // then we fall back to the mappings with the next-longest prefix, 1397 // and ultimately to mappings with no prefix. 1398 // Each fallback might be another set of contractions. 1399 // For example, if there are mappings for ch, p|cd, p|ce, but not for p|c, 1400 // then in text "pch" we find the ch contraction. 1401 for(cond = head;; cond = getConditionalCE32(cond->next)) { 1402 int32_t length = cond->prefixLength(); 1403 if(length == prefixLength) { break; } 1404 if(cond->defaultCE32 != Collation::NO_CE32 && 1405 (length==0 || prefix.endsWith(cond->context, 1, length))) { 1406 emptySuffixCE32 = cond->defaultCE32; 1407 } 1408 } 1409 cond = firstCond; 1410 } 1411 // Optimization: Set a flag when 1412 // the first character of every contraction suffix has lccc!=0. 1413 // Short-circuits contraction matching when a normal letter follows. 1414 flags |= Collation::CONTRACT_NEXT_CCC; 1415 // Add all of the non-empty suffixes into the contraction trie. 1416 for(;;) { 1417 UnicodeString suffix(cond->context, suffixStart); 1418 uint16_t fcd16 = nfcImpl.getFCD16(suffix.char32At(0)); 1419 if(fcd16 <= 0xff) { 1420 flags &= ~Collation::CONTRACT_NEXT_CCC; 1421 } 1422 fcd16 = nfcImpl.getFCD16(suffix.char32At(suffix.length() - 1)); 1423 if(fcd16 > 0xff) { 1424 // The last suffix character has lccc!=0, allowing for discontiguous contractions. 1425 flags |= Collation::CONTRACT_TRAILING_CCC; 1426 } 1427 contractionBuilder.add(suffix, (int32_t)cond->ce32, errorCode); 1428 if(cond == lastCond) { break; } 1429 cond = getConditionalCE32(cond->next); 1430 } 1431 int32_t index = addContextTrie(emptySuffixCE32, contractionBuilder, errorCode); 1432 if(U_FAILURE(errorCode)) { return 0; } 1433 if(index > Collation::MAX_INDEX) { 1434 errorCode = U_BUFFER_OVERFLOW_ERROR; 1435 return 0; 1436 } 1437 ce32 = Collation::makeCE32FromTagAndIndex(Collation::CONTRACTION_TAG, index) | flags; 1438 } 1439 U_ASSERT(cond == lastCond); 1440 firstCond->defaultCE32 = ce32; 1441 if(prefixLength == 0) { 1442 if(cond->next < 0) { 1443 // No non-empty prefixes, only contractions. 1444 return ce32; 1445 } 1446 } else { 1447 prefix.remove(0, 1); // Remove the length unit. 1448 prefix.reverse(); 1449 prefixBuilder.add(prefix, (int32_t)ce32, errorCode); 1450 if(cond->next < 0) { break; } 1451 } 1452 } 1453 U_ASSERT(head->defaultCE32 != Collation::NO_CE32); 1454 int32_t index = addContextTrie(head->defaultCE32, prefixBuilder, errorCode); 1455 if(U_FAILURE(errorCode)) { return 0; } 1456 if(index > Collation::MAX_INDEX) { 1457 errorCode = U_BUFFER_OVERFLOW_ERROR; 1458 return 0; 1459 } 1460 return Collation::makeCE32FromTagAndIndex(Collation::PREFIX_TAG, index); 1461 } 1462 1463 int32_t 1464 CollationDataBuilder::addContextTrie(uint32_t defaultCE32, UCharsTrieBuilder &trieBuilder, 1465 UErrorCode &errorCode) { 1466 UnicodeString context; 1467 context.append((UChar)(defaultCE32 >> 16)).append((UChar)defaultCE32); 1468 UnicodeString trieString; 1469 context.append(trieBuilder.buildUnicodeString(USTRINGTRIE_BUILD_SMALL, trieString, errorCode)); 1470 if(U_FAILURE(errorCode)) { return -1; } 1471 int32_t index = contexts.indexOf(context); 1472 if(index < 0) { 1473 index = contexts.length(); 1474 contexts.append(context); 1475 } 1476 return index; 1477 } 1478 1479 void 1480 CollationDataBuilder::buildFastLatinTable(CollationData &data, UErrorCode &errorCode) { 1481 if(U_FAILURE(errorCode) || !fastLatinEnabled) { return; } 1482 1483 delete fastLatinBuilder; 1484 fastLatinBuilder = new CollationFastLatinBuilder(errorCode); 1485 if(fastLatinBuilder == NULL) { 1486 errorCode = U_MEMORY_ALLOCATION_ERROR; 1487 return; 1488 } 1489 if(fastLatinBuilder->forData(data, errorCode)) { 1490 const uint16_t *table = fastLatinBuilder->getTable(); 1491 int32_t length = fastLatinBuilder->lengthOfTable(); 1492 if(base != NULL && length == base->fastLatinTableLength && 1493 uprv_memcmp(table, base->fastLatinTable, length * 2) == 0) { 1494 // Same fast Latin table as in the base, use that one instead. 1495 delete fastLatinBuilder; 1496 fastLatinBuilder = NULL; 1497 table = base->fastLatinTable; 1498 } 1499 data.fastLatinTable = table; 1500 data.fastLatinTableLength = length; 1501 } else { 1502 delete fastLatinBuilder; 1503 fastLatinBuilder = NULL; 1504 } 1505 } 1506 1507 int32_t 1508 CollationDataBuilder::getCEs(const UnicodeString &s, int64_t ces[], int32_t cesLength) { 1509 return getCEs(s, 0, ces, cesLength); 1510 } 1511 1512 int32_t 1513 CollationDataBuilder::getCEs(const UnicodeString &prefix, const UnicodeString &s, 1514 int64_t ces[], int32_t cesLength) { 1515 int32_t prefixLength = prefix.length(); 1516 if(prefixLength == 0) { 1517 return getCEs(s, 0, ces, cesLength); 1518 } else { 1519 return getCEs(prefix + s, prefixLength, ces, cesLength); 1520 } 1521 } 1522 1523 int32_t 1524 CollationDataBuilder::getCEs(const UnicodeString &s, int32_t start, 1525 int64_t ces[], int32_t cesLength) { 1526 if(collIter == NULL) { 1527 collIter = new DataBuilderCollationIterator(*this); 1528 if(collIter == NULL) { return 0; } 1529 } 1530 return collIter->fetchCEs(s, start, ces, cesLength); 1531 } 1532 1533 U_NAMESPACE_END 1534 1535 #endif // !UCONFIG_NO_COLLATION 1536