| 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 | UPRV_UNREACHABLE; // require ce32 == base->getFinalCE32(ce32) |
| 856 | } |
| 857 | return ce32; |
| 858 | } |
| 859 | |
| 860 | int32_t |
| 861 | CollationDataBuilder::copyContractionsFromBaseCE32(UnicodeString &context, UChar32 c, uint32_t ce32, |
| 862 | ConditionalCE32 *cond, UErrorCode &errorCode) { |
| 863 | if(U_FAILURE(errorCode)) { return 0; } |
| 864 | const UChar *p = base->contexts + Collation::indexFromCE32(ce32); |
| 865 | int32_t index; |
| 866 | if((ce32 & Collation::CONTRACT_SINGLE_CP_NO_MATCH) != 0) { |
| 867 | // No match on the single code point. |
| 868 | // We are underneath a prefix, and the default mapping is just |
| 869 | // a fallback to the mappings for a shorter prefix. |
| 870 | U_ASSERT(context.length() > 1); |
| 871 | index = -1; |
| 872 | } else { |
| 873 | ce32 = CollationData::readCE32(p); // Default if no suffix match. |
| 874 | U_ASSERT(!Collation::isContractionCE32(ce32)); |
| 875 | ce32 = copyFromBaseCE32(c, ce32, TRUE, errorCode); |
| 876 | cond->next = index = addConditionalCE32(context, ce32, errorCode); |
| 877 | if(U_FAILURE(errorCode)) { return 0; } |
| 878 | cond = getConditionalCE32(index); |
| 879 | } |
| 880 | |
| 881 | int32_t suffixStart = context.length(); |
| 882 | UCharsTrie::Iterator suffixes(p + 2, 0, errorCode); |
| 883 | while(suffixes.next(errorCode)) { |
| 884 | context.append(suffixes.getString()); |
| 885 | ce32 = copyFromBaseCE32(c, (uint32_t)suffixes.getValue(), TRUE, errorCode); |
| 886 | cond->next = index = addConditionalCE32(context, ce32, errorCode); |
| 887 | if(U_FAILURE(errorCode)) { return 0; } |
| 888 | // No need to update the unsafeBackwardSet because the tailoring set |
| 889 | // is already a copy of the base set. |
| 890 | cond = getConditionalCE32(index); |
| 891 | context.truncate(suffixStart); |
| 892 | } |
| 893 | U_ASSERT(index >= 0); |
| 894 | return index; |
| 895 | } |
| 896 | |
| 897 | class CopyHelper { |
| 898 | public: |
| 899 | CopyHelper(const CollationDataBuilder &s, CollationDataBuilder &d, |
| 900 | const CollationDataBuilder::CEModifier &m, UErrorCode &initialErrorCode) |
| 901 | : src(s), dest(d), modifier(m), |
| 902 | errorCode(initialErrorCode) {} |
| 903 | |
| 904 | UBool copyRangeCE32(UChar32 start, UChar32 end, uint32_t ce32) { |
| 905 | ce32 = copyCE32(ce32); |
| 906 | utrie2_setRange32(dest.trie, start, end, ce32, TRUE, &errorCode); |
| 907 | if(CollationDataBuilder::isBuilderContextCE32(ce32)) { |
| 908 | dest.contextChars.add(start, end); |
| 909 | } |
| 910 | return U_SUCCESS(errorCode); |
| 911 | } |
| 912 | |
| 913 | uint32_t copyCE32(uint32_t ce32) { |
| 914 | if(!Collation::isSpecialCE32(ce32)) { |
| 915 | int64_t ce = modifier.modifyCE32(ce32); |
| 916 | if(ce != Collation::NO_CE) { |
| 917 | ce32 = dest.encodeOneCE(ce, errorCode); |
| 918 | } |
| 919 | } else { |
| 920 | int32_t tag = Collation::tagFromCE32(ce32); |
| 921 | if(tag == Collation::EXPANSION32_TAG) { |
| 922 | const uint32_t *srcCE32s = reinterpret_cast<uint32_t *>(src.ce32s.getBuffer()); |
| 923 | srcCE32s += Collation::indexFromCE32(ce32); |
| 924 | int32_t length = Collation::lengthFromCE32(ce32); |
| 925 | // Inspect the source CE32s. Just copy them if none are modified. |
| 926 | // Otherwise copy to modifiedCEs, with modifications. |
| 927 | UBool isModified = FALSE; |
| 928 | for(int32_t i = 0; i < length; ++i) { |
| 929 | ce32 = srcCE32s[i]; |
| 930 | int64_t ce; |
| 931 | if(Collation::isSpecialCE32(ce32) || |
| 932 | (ce = modifier.modifyCE32(ce32)) == Collation::NO_CE) { |
| 933 | if(isModified) { |
| 934 | modifiedCEs[i] = Collation::ceFromCE32(ce32); |
| 935 | } |
| 936 | } else { |
| 937 | if(!isModified) { |
| 938 | for(int32_t j = 0; j < i; ++j) { |
| 939 | modifiedCEs[j] = Collation::ceFromCE32(srcCE32s[j]); |
| 940 | } |
| 941 | isModified = TRUE; |
| 942 | } |
| 943 | modifiedCEs[i] = ce; |
| 944 | } |
| 945 | } |
| 946 | if(isModified) { |
| 947 | ce32 = dest.encodeCEs(modifiedCEs, length, errorCode); |
| 948 | } else { |
| 949 | ce32 = dest.encodeExpansion32( |
| 950 | reinterpret_cast<const int32_t *>(srcCE32s), length, errorCode); |
| 951 | } |
| 952 | } else if(tag == Collation::EXPANSION_TAG) { |
| 953 | const int64_t *srcCEs = src.ce64s.getBuffer(); |
| 954 | srcCEs += Collation::indexFromCE32(ce32); |
| 955 | int32_t length = Collation::lengthFromCE32(ce32); |
| 956 | // Inspect the source CEs. Just copy them if none are modified. |
| 957 | // Otherwise copy to modifiedCEs, with modifications. |
| 958 | UBool isModified = FALSE; |
| 959 | for(int32_t i = 0; i < length; ++i) { |
| 960 | int64_t srcCE = srcCEs[i]; |
| 961 | int64_t ce = modifier.modifyCE(srcCE); |
| 962 | if(ce == Collation::NO_CE) { |
| 963 | if(isModified) { |
| 964 | modifiedCEs[i] = srcCE; |
| 965 | } |
| 966 | } else { |
| 967 | if(!isModified) { |
| 968 | for(int32_t j = 0; j < i; ++j) { |
| 969 | modifiedCEs[j] = srcCEs[j]; |
| 970 | } |
| 971 | isModified = TRUE; |
| 972 | } |
| 973 | modifiedCEs[i] = ce; |
| 974 | } |
| 975 | } |
| 976 | if(isModified) { |
| 977 | ce32 = dest.encodeCEs(modifiedCEs, length, errorCode); |
| 978 | } else { |
| 979 | ce32 = dest.encodeExpansion(srcCEs, length, errorCode); |
| 980 | } |
| 981 | } else if(tag == Collation::BUILDER_DATA_TAG) { |
| 982 | // Copy the list of ConditionalCE32. |
| 983 | ConditionalCE32 *cond = src.getConditionalCE32ForCE32(ce32); |
| 984 | U_ASSERT(!cond->hasContext()); |
| 985 | int32_t destIndex = dest.addConditionalCE32( |
| 986 | cond->context, copyCE32(cond->ce32), errorCode); |
| 987 | ce32 = CollationDataBuilder::makeBuilderContextCE32(destIndex); |
| 988 | while(cond->next >= 0) { |
| 989 | cond = src.getConditionalCE32(cond->next); |
| 990 | ConditionalCE32 *prevDestCond = dest.getConditionalCE32(destIndex); |
| 991 | destIndex = dest.addConditionalCE32( |
| 992 | cond->context, copyCE32(cond->ce32), errorCode); |
| 993 | int32_t suffixStart = cond->prefixLength() + 1; |
| 994 | dest.unsafeBackwardSet.addAll(cond->context.tempSubString(suffixStart)); |
| 995 | prevDestCond->next = destIndex; |
| 996 | } |
| 997 | } else { |
| 998 | // Just copy long CEs and Latin mini expansions (and other expected values) as is, |
| 999 | // assuming that the modifier would not modify them. |
| 1000 | U_ASSERT(tag == Collation::LONG_PRIMARY_TAG || |
| 1001 | tag == Collation::LONG_SECONDARY_TAG || |
| 1002 | tag == Collation::LATIN_EXPANSION_TAG || |
| 1003 | tag == Collation::HANGUL_TAG); |
| 1004 | } |
| 1005 | } |
| 1006 | return ce32; |
| 1007 | } |
| 1008 | |
| 1009 | const CollationDataBuilder &src; |
| 1010 | CollationDataBuilder &dest; |
| 1011 | const CollationDataBuilder::CEModifier &modifier; |
| 1012 | int64_t modifiedCEs[Collation::MAX_EXPANSION_LENGTH]; |
| 1013 | UErrorCode errorCode; |
| 1014 | }; |
| 1015 | |
| 1016 | U_CDECL_BEGIN |
| 1017 | |
| 1018 | static UBool U_CALLCONV |
| 1019 | enumRangeForCopy(const void *context, UChar32 start, UChar32 end, uint32_t value) { |
| 1020 | return |
| 1021 | value == Collation::UNASSIGNED_CE32 || value == Collation::FALLBACK_CE32 || |
| 1022 | ((CopyHelper *)context)->copyRangeCE32(start, end, value); |
| 1023 | } |
| 1024 | |
| 1025 | U_CDECL_END |
| 1026 | |
| 1027 | void |
| 1028 | CollationDataBuilder::copyFrom(const CollationDataBuilder &src, const CEModifier &modifier, |
| 1029 | UErrorCode &errorCode) { |
| 1030 | if(U_FAILURE(errorCode)) { return; } |
| 1031 | if(trie == NULL || utrie2_isFrozen(trie)) { |
| 1032 | errorCode = U_INVALID_STATE_ERROR; |
| 1033 | return; |
| 1034 | } |
| 1035 | CopyHelper helper(src, *this, modifier, errorCode); |
| 1036 | utrie2_enum(src.trie, NULL, enumRangeForCopy, &helper); |
| 1037 | errorCode = helper.errorCode; |
| 1038 | // Update the contextChars and the unsafeBackwardSet while copying, |
| 1039 | // in case a character had conditional mappings in the source builder |
| 1040 | // and they were removed later. |
| 1041 | modified |= src.modified; |
| 1042 | } |
| 1043 | |
| 1044 | void |
| 1045 | CollationDataBuilder::optimize(const UnicodeSet &set, UErrorCode &errorCode) { |
| 1046 | if(U_FAILURE(errorCode) || set.isEmpty()) { return; } |
| 1047 | UnicodeSetIterator iter(set); |
| 1048 | while(iter.next() && !iter.isString()) { |
| 1049 | UChar32 c = iter.getCodepoint(); |
| 1050 | uint32_t ce32 = utrie2_get32(trie, c); |
| 1051 | if(ce32 == Collation::FALLBACK_CE32) { |
| 1052 | ce32 = base->getFinalCE32(base->getCE32(c)); |
| 1053 | ce32 = copyFromBaseCE32(c, ce32, TRUE, errorCode); |
| 1054 | utrie2_set32(trie, c, ce32, &errorCode); |
| 1055 | } |
| 1056 | } |
| 1057 | modified = TRUE; |
| 1058 | } |
| 1059 | |
| 1060 | void |
| 1061 | CollationDataBuilder::suppressContractions(const UnicodeSet &set, UErrorCode &errorCode) { |
| 1062 | if(U_FAILURE(errorCode) || set.isEmpty()) { return; } |
| 1063 | UnicodeSetIterator iter(set); |
| 1064 | while(iter.next() && !iter.isString()) { |
| 1065 | UChar32 c = iter.getCodepoint(); |
| 1066 | uint32_t ce32 = utrie2_get32(trie, c); |
| 1067 | if(ce32 == Collation::FALLBACK_CE32) { |
| 1068 | ce32 = base->getFinalCE32(base->getCE32(c)); |
| 1069 | if(Collation::ce32HasContext(ce32)) { |
| 1070 | ce32 = copyFromBaseCE32(c, ce32, FALSE /* without context */, errorCode); |
| 1071 | utrie2_set32(trie, c, ce32, &errorCode); |
| 1072 | } |
| 1073 | } else if(isBuilderContextCE32(ce32)) { |
| 1074 | ce32 = getConditionalCE32ForCE32(ce32)->ce32; |
| 1075 | // Simply abandon the list of ConditionalCE32. |
| 1076 | // The caller will copy this builder in the end, |
| 1077 | // eliminating unreachable data. |
| 1078 | utrie2_set32(trie, c, ce32, &errorCode); |
| 1079 | contextChars.remove(c); |
| 1080 | } |
| 1081 | } |
| 1082 | modified = TRUE; |
| 1083 | } |
| 1084 | |
| 1085 | UBool |
| 1086 | CollationDataBuilder::getJamoCE32s(uint32_t jamoCE32s[], UErrorCode &errorCode) { |
| 1087 | if(U_FAILURE(errorCode)) { return FALSE; } |
| 1088 | UBool anyJamoAssigned = base == NULL; // always set jamoCE32s in the base data |
| 1089 | UBool needToCopyFromBase = FALSE; |
| 1090 | for(int32_t j = 0; j < CollationData::JAMO_CE32S_LENGTH; ++j) { // Count across Jamo types. |
| 1091 | UChar32 jamo = jamoCpFromIndex(j); |
| 1092 | UBool fromBase = FALSE; |
| 1093 | uint32_t ce32 = utrie2_get32(trie, jamo); |
| 1094 | anyJamoAssigned |= Collation::isAssignedCE32(ce32); |
| 1095 | // TODO: Try to prevent [optimize [Jamo]] from counting as anyJamoAssigned. |
| 1096 | // (As of CLDR 24 [2013] the Korean tailoring does not optimize conjoining Jamo.) |
| 1097 | if(ce32 == Collation::FALLBACK_CE32) { |
| 1098 | fromBase = TRUE; |
| 1099 | ce32 = base->getCE32(jamo); |
| 1100 | } |
| 1101 | if(Collation::isSpecialCE32(ce32)) { |
| 1102 | switch(Collation::tagFromCE32(ce32)) { |
| 1103 | case Collation::LONG_PRIMARY_TAG: |
| 1104 | case Collation::LONG_SECONDARY_TAG: |
| 1105 | case Collation::LATIN_EXPANSION_TAG: |
| 1106 | // Copy the ce32 as-is. |
| 1107 | break; |
| 1108 | case Collation::EXPANSION32_TAG: |
| 1109 | case Collation::EXPANSION_TAG: |
| 1110 | case Collation::PREFIX_TAG: |
| 1111 | case Collation::CONTRACTION_TAG: |
| 1112 | if(fromBase) { |
| 1113 | // Defer copying until we know if anyJamoAssigned. |
| 1114 | ce32 = Collation::FALLBACK_CE32; |
| 1115 | needToCopyFromBase = TRUE; |
| 1116 | } |
| 1117 | break; |
| 1118 | case Collation::IMPLICIT_TAG: |
| 1119 | // An unassigned Jamo should only occur in tests with incomplete bases. |
| 1120 | U_ASSERT(fromBase); |
| 1121 | ce32 = Collation::FALLBACK_CE32; |
| 1122 | needToCopyFromBase = TRUE; |
| 1123 | break; |
| 1124 | case Collation::OFFSET_TAG: |
| 1125 | ce32 = getCE32FromOffsetCE32(fromBase, jamo, ce32); |
| 1126 | break; |
| 1127 | case Collation::FALLBACK_TAG: |
| 1128 | case Collation::RESERVED_TAG_3: |
| 1129 | case Collation::BUILDER_DATA_TAG: |
| 1130 | case Collation::DIGIT_TAG: |
| 1131 | case Collation::U0000_TAG: |
| 1132 | case Collation::HANGUL_TAG: |
| 1133 | case Collation::LEAD_SURROGATE_TAG: |
| 1134 | errorCode = U_INTERNAL_PROGRAM_ERROR; |
| 1135 | return FALSE; |
| 1136 | } |
| 1137 | } |
| 1138 | jamoCE32s[j] = ce32; |
| 1139 | } |
| 1140 | if(anyJamoAssigned && needToCopyFromBase) { |
| 1141 | for(int32_t j = 0; j < CollationData::JAMO_CE32S_LENGTH; ++j) { |
| 1142 | if(jamoCE32s[j] == Collation::FALLBACK_CE32) { |
| 1143 | UChar32 jamo = jamoCpFromIndex(j); |
| 1144 | jamoCE32s[j] = copyFromBaseCE32(jamo, base->getCE32(jamo), |
| 1145 | /*withContext=*/ TRUE, errorCode); |
| 1146 | } |
| 1147 | } |
| 1148 | } |
| 1149 | return anyJamoAssigned && U_SUCCESS(errorCode); |
| 1150 | } |
| 1151 | |
| 1152 | void |
| 1153 | CollationDataBuilder::setDigitTags(UErrorCode &errorCode) { |
| 1154 | UnicodeSet digits(UNICODE_STRING_SIMPLE("[:Nd:]" ), errorCode); |
| 1155 | if(U_FAILURE(errorCode)) { return; } |
| 1156 | UnicodeSetIterator iter(digits); |
| 1157 | while(iter.next()) { |
| 1158 | U_ASSERT(!iter.isString()); |
| 1159 | UChar32 c = iter.getCodepoint(); |
| 1160 | uint32_t ce32 = utrie2_get32(trie, c); |
| 1161 | if(ce32 != Collation::FALLBACK_CE32 && ce32 != Collation::UNASSIGNED_CE32) { |
| 1162 | int32_t index = addCE32(ce32, errorCode); |
| 1163 | if(U_FAILURE(errorCode)) { return; } |
| 1164 | if(index > Collation::MAX_INDEX) { |
| 1165 | errorCode = U_BUFFER_OVERFLOW_ERROR; |
| 1166 | return; |
| 1167 | } |
| 1168 | ce32 = Collation::makeCE32FromTagIndexAndLength( |
| 1169 | Collation::DIGIT_TAG, index, u_charDigitValue(c)); |
| 1170 | utrie2_set32(trie, c, ce32, &errorCode); |
| 1171 | } |
| 1172 | } |
| 1173 | } |
| 1174 | |
| 1175 | U_CDECL_BEGIN |
| 1176 | |
| 1177 | static UBool U_CALLCONV |
| 1178 | enumRangeLeadValue(const void *context, UChar32 /*start*/, UChar32 /*end*/, uint32_t value) { |
| 1179 | int32_t *pValue = (int32_t *)context; |
| 1180 | if(value == Collation::UNASSIGNED_CE32) { |
| 1181 | value = Collation::LEAD_ALL_UNASSIGNED; |
| 1182 | } else if(value == Collation::FALLBACK_CE32) { |
| 1183 | value = Collation::LEAD_ALL_FALLBACK; |
| 1184 | } else { |
| 1185 | *pValue = Collation::LEAD_MIXED; |
| 1186 | return FALSE; |
| 1187 | } |
| 1188 | if(*pValue < 0) { |
| 1189 | *pValue = (int32_t)value; |
| 1190 | } else if(*pValue != (int32_t)value) { |
| 1191 | *pValue = Collation::LEAD_MIXED; |
| 1192 | return FALSE; |
| 1193 | } |
| 1194 | return TRUE; |
| 1195 | } |
| 1196 | |
| 1197 | U_CDECL_END |
| 1198 | |
| 1199 | void |
| 1200 | CollationDataBuilder::setLeadSurrogates(UErrorCode &errorCode) { |
| 1201 | for(UChar lead = 0xd800; lead < 0xdc00; ++lead) { |
| 1202 | int32_t value = -1; |
| 1203 | utrie2_enumForLeadSurrogate(trie, lead, NULL, enumRangeLeadValue, &value); |
| 1204 | utrie2_set32ForLeadSurrogateCodeUnit( |
| 1205 | trie, lead, |
| 1206 | Collation::makeCE32FromTagAndIndex(Collation::LEAD_SURROGATE_TAG, 0) | (uint32_t)value, |
| 1207 | &errorCode); |
| 1208 | } |
| 1209 | } |
| 1210 | |
| 1211 | void |
| 1212 | CollationDataBuilder::build(CollationData &data, UErrorCode &errorCode) { |
| 1213 | buildMappings(data, errorCode); |
| 1214 | if(base != NULL) { |
| 1215 | data.numericPrimary = base->numericPrimary; |
| 1216 | data.compressibleBytes = base->compressibleBytes; |
| 1217 | data.numScripts = base->numScripts; |
| 1218 | data.scriptsIndex = base->scriptsIndex; |
| 1219 | data.scriptStarts = base->scriptStarts; |
| 1220 | data.scriptStartsLength = base->scriptStartsLength; |
| 1221 | } |
| 1222 | buildFastLatinTable(data, errorCode); |
| 1223 | } |
| 1224 | |
| 1225 | void |
| 1226 | CollationDataBuilder::buildMappings(CollationData &data, UErrorCode &errorCode) { |
| 1227 | if(U_FAILURE(errorCode)) { return; } |
| 1228 | if(trie == NULL || utrie2_isFrozen(trie)) { |
| 1229 | errorCode = U_INVALID_STATE_ERROR; |
| 1230 | return; |
| 1231 | } |
| 1232 | |
| 1233 | buildContexts(errorCode); |
| 1234 | |
| 1235 | uint32_t jamoCE32s[CollationData::JAMO_CE32S_LENGTH]; |
| 1236 | int32_t jamoIndex = -1; |
| 1237 | if(getJamoCE32s(jamoCE32s, errorCode)) { |
| 1238 | jamoIndex = ce32s.size(); |
| 1239 | for(int32_t i = 0; i < CollationData::JAMO_CE32S_LENGTH; ++i) { |
| 1240 | ce32s.addElement((int32_t)jamoCE32s[i], errorCode); |
| 1241 | } |
| 1242 | // Small optimization: Use a bit in the Hangul ce32 |
| 1243 | // to indicate that none of the Jamo CE32s are isSpecialCE32() |
| 1244 | // (as it should be in the root collator). |
| 1245 | // It allows CollationIterator to avoid recursive function calls and per-Jamo tests. |
| 1246 | // In order to still have good trie compression and keep this code simple, |
| 1247 | // we only set this flag if a whole block of 588 Hangul syllables starting with |
| 1248 | // a common leading consonant (Jamo L) has this property. |
| 1249 | UBool isAnyJamoVTSpecial = FALSE; |
| 1250 | for(int32_t i = Hangul::JAMO_L_COUNT; i < CollationData::JAMO_CE32S_LENGTH; ++i) { |
| 1251 | if(Collation::isSpecialCE32(jamoCE32s[i])) { |
| 1252 | isAnyJamoVTSpecial = TRUE; |
| 1253 | break; |
| 1254 | } |
| 1255 | } |
| 1256 | uint32_t hangulCE32 = Collation::makeCE32FromTagAndIndex(Collation::HANGUL_TAG, 0); |
| 1257 | UChar32 c = Hangul::HANGUL_BASE; |
| 1258 | for(int32_t i = 0; i < Hangul::JAMO_L_COUNT; ++i) { // iterate over the Jamo L |
| 1259 | uint32_t ce32 = hangulCE32; |
| 1260 | if(!isAnyJamoVTSpecial && !Collation::isSpecialCE32(jamoCE32s[i])) { |
| 1261 | ce32 |= Collation::HANGUL_NO_SPECIAL_JAMO; |
| 1262 | } |
| 1263 | UChar32 limit = c + Hangul::JAMO_VT_COUNT; |
| 1264 | utrie2_setRange32(trie, c, limit - 1, ce32, TRUE, &errorCode); |
| 1265 | c = limit; |
| 1266 | } |
| 1267 | } else { |
| 1268 | // Copy the Hangul CE32s from the base in blocks per Jamo L, |
| 1269 | // assuming that HANGUL_NO_SPECIAL_JAMO is set or not set for whole blocks. |
| 1270 | for(UChar32 c = Hangul::HANGUL_BASE; c < Hangul::HANGUL_LIMIT;) { |
| 1271 | uint32_t ce32 = base->getCE32(c); |
| 1272 | U_ASSERT(Collation::hasCE32Tag(ce32, Collation::HANGUL_TAG)); |
| 1273 | UChar32 limit = c + Hangul::JAMO_VT_COUNT; |
| 1274 | utrie2_setRange32(trie, c, limit - 1, ce32, TRUE, &errorCode); |
| 1275 | c = limit; |
| 1276 | } |
| 1277 | } |
| 1278 | |
| 1279 | setDigitTags(errorCode); |
| 1280 | setLeadSurrogates(errorCode); |
| 1281 | |
| 1282 | // For U+0000, move its normal ce32 into CE32s[0] and set U0000_TAG. |
| 1283 | ce32s.setElementAt((int32_t)utrie2_get32(trie, 0), 0); |
| 1284 | utrie2_set32(trie, 0, Collation::makeCE32FromTagAndIndex(Collation::U0000_TAG, 0), &errorCode); |
| 1285 | |
| 1286 | utrie2_freeze(trie, UTRIE2_32_VALUE_BITS, &errorCode); |
| 1287 | if(U_FAILURE(errorCode)) { return; } |
| 1288 | |
| 1289 | // Mark each lead surrogate as "unsafe" |
| 1290 | // if any of its 1024 associated supplementary code points is "unsafe". |
| 1291 | UChar32 c = 0x10000; |
| 1292 | for(UChar lead = 0xd800; lead < 0xdc00; ++lead, c += 0x400) { |
| 1293 | if(unsafeBackwardSet.containsSome(c, c + 0x3ff)) { |
| 1294 | unsafeBackwardSet.add(lead); |
| 1295 | } |
| 1296 | } |
| 1297 | unsafeBackwardSet.freeze(); |
| 1298 | |
| 1299 | data.trie = trie; |
| 1300 | data.ce32s = reinterpret_cast<const uint32_t *>(ce32s.getBuffer()); |
| 1301 | data.ces = ce64s.getBuffer(); |
| 1302 | data.contexts = contexts.getBuffer(); |
| 1303 | |
| 1304 | data.ce32sLength = ce32s.size(); |
| 1305 | data.cesLength = ce64s.size(); |
| 1306 | data.contextsLength = contexts.length(); |
| 1307 | |
| 1308 | data.base = base; |
| 1309 | if(jamoIndex >= 0) { |
| 1310 | data.jamoCE32s = data.ce32s + jamoIndex; |
| 1311 | } else { |
| 1312 | data.jamoCE32s = base->jamoCE32s; |
| 1313 | } |
| 1314 | data.unsafeBackwardSet = &unsafeBackwardSet; |
| 1315 | } |
| 1316 | |
| 1317 | void |
| 1318 | CollationDataBuilder::clearContexts() { |
| 1319 | contexts.remove(); |
| 1320 | UnicodeSetIterator iter(contextChars); |
| 1321 | while(iter.next()) { |
| 1322 | U_ASSERT(!iter.isString()); |
| 1323 | uint32_t ce32 = utrie2_get32(trie, iter.getCodepoint()); |
| 1324 | U_ASSERT(isBuilderContextCE32(ce32)); |
| 1325 | getConditionalCE32ForCE32(ce32)->builtCE32 = Collation::NO_CE32; |
| 1326 | } |
| 1327 | } |
| 1328 | |
| 1329 | void |
| 1330 | CollationDataBuilder::buildContexts(UErrorCode &errorCode) { |
| 1331 | if(U_FAILURE(errorCode)) { return; } |
| 1332 | // Ignore abandoned lists and the cached builtCE32, |
| 1333 | // and build all contexts from scratch. |
| 1334 | contexts.remove(); |
| 1335 | UnicodeSetIterator iter(contextChars); |
| 1336 | while(U_SUCCESS(errorCode) && iter.next()) { |
| 1337 | U_ASSERT(!iter.isString()); |
| 1338 | UChar32 c = iter.getCodepoint(); |
| 1339 | uint32_t ce32 = utrie2_get32(trie, c); |
| 1340 | if(!isBuilderContextCE32(ce32)) { |
| 1341 | // Impossible: No context data for c in contextChars. |
| 1342 | errorCode = U_INTERNAL_PROGRAM_ERROR; |
| 1343 | return; |
| 1344 | } |
| 1345 | ConditionalCE32 *cond = getConditionalCE32ForCE32(ce32); |
| 1346 | ce32 = buildContext(cond, errorCode); |
| 1347 | utrie2_set32(trie, c, ce32, &errorCode); |
| 1348 | } |
| 1349 | } |
| 1350 | |
| 1351 | uint32_t |
| 1352 | CollationDataBuilder::buildContext(ConditionalCE32 *head, UErrorCode &errorCode) { |
| 1353 | if(U_FAILURE(errorCode)) { return 0; } |
| 1354 | // The list head must have no context. |
| 1355 | U_ASSERT(!head->hasContext()); |
| 1356 | // The list head must be followed by one or more nodes that all do have context. |
| 1357 | U_ASSERT(head->next >= 0); |
| 1358 | UCharsTrieBuilder prefixBuilder(errorCode); |
| 1359 | UCharsTrieBuilder contractionBuilder(errorCode); |
| 1360 | for(ConditionalCE32 *cond = head;; cond = getConditionalCE32(cond->next)) { |
| 1361 | // After the list head, the prefix or suffix can be empty, but not both. |
| 1362 | U_ASSERT(cond == head || cond->hasContext()); |
| 1363 | int32_t prefixLength = cond->prefixLength(); |
| 1364 | UnicodeString prefix(cond->context, 0, prefixLength + 1); |
| 1365 | // Collect all contraction suffixes for one prefix. |
| 1366 | ConditionalCE32 *firstCond = cond; |
| 1367 | ConditionalCE32 *lastCond = cond; |
| 1368 | while(cond->next >= 0 && |
| 1369 | (cond = getConditionalCE32(cond->next))->context.startsWith(prefix)) { |
| 1370 | lastCond = cond; |
| 1371 | } |
| 1372 | uint32_t ce32; |
| 1373 | int32_t suffixStart = prefixLength + 1; // == prefix.length() |
| 1374 | if(lastCond->context.length() == suffixStart) { |
| 1375 | // One prefix without contraction suffix. |
| 1376 | U_ASSERT(firstCond == lastCond); |
| 1377 | ce32 = lastCond->ce32; |
| 1378 | cond = lastCond; |
| 1379 | } else { |
| 1380 | // Build the contractions trie. |
| 1381 | contractionBuilder.clear(); |
| 1382 | // Entry for an empty suffix, to be stored before the trie. |
| 1383 | uint32_t emptySuffixCE32 = 0; |
| 1384 | uint32_t flags = 0; |
| 1385 | if(firstCond->context.length() == suffixStart) { |
| 1386 | // There is a mapping for the prefix and the single character c. (p|c) |
| 1387 | // If no other suffix matches, then we return this value. |
| 1388 | emptySuffixCE32 = firstCond->ce32; |
| 1389 | cond = getConditionalCE32(firstCond->next); |
| 1390 | } else { |
| 1391 | // There is no mapping for the prefix and just the single character. |
| 1392 | // (There is no p|c, only p|cd, p|ce etc.) |
| 1393 | flags |= Collation::CONTRACT_SINGLE_CP_NO_MATCH; |
| 1394 | // When the prefix matches but none of the prefix-specific suffixes, |
| 1395 | // then we fall back to the mappings with the next-longest prefix, |
| 1396 | // and ultimately to mappings with no prefix. |
| 1397 | // Each fallback might be another set of contractions. |
| 1398 | // For example, if there are mappings for ch, p|cd, p|ce, but not for p|c, |
| 1399 | // then in text "pch" we find the ch contraction. |
| 1400 | for(cond = head;; cond = getConditionalCE32(cond->next)) { |
| 1401 | int32_t length = cond->prefixLength(); |
| 1402 | if(length == prefixLength) { break; } |
| 1403 | if(cond->defaultCE32 != Collation::NO_CE32 && |
| 1404 | (length==0 || prefix.endsWith(cond->context, 1, length))) { |
| 1405 | emptySuffixCE32 = cond->defaultCE32; |
| 1406 | } |
| 1407 | } |
| 1408 | cond = firstCond; |
| 1409 | } |
| 1410 | // Optimization: Set a flag when |
| 1411 | // the first character of every contraction suffix has lccc!=0. |
| 1412 | // Short-circuits contraction matching when a normal letter follows. |
| 1413 | flags |= Collation::CONTRACT_NEXT_CCC; |
| 1414 | // Add all of the non-empty suffixes into the contraction trie. |
| 1415 | for(;;) { |
| 1416 | UnicodeString suffix(cond->context, suffixStart); |
| 1417 | uint16_t fcd16 = nfcImpl.getFCD16(suffix.char32At(0)); |
| 1418 | if(fcd16 <= 0xff) { |
| 1419 | flags &= ~Collation::CONTRACT_NEXT_CCC; |
| 1420 | } |
| 1421 | fcd16 = nfcImpl.getFCD16(suffix.char32At(suffix.length() - 1)); |
| 1422 | if(fcd16 > 0xff) { |
| 1423 | // The last suffix character has lccc!=0, allowing for discontiguous contractions. |
| 1424 | flags |= Collation::CONTRACT_TRAILING_CCC; |
| 1425 | } |
| 1426 | contractionBuilder.add(suffix, (int32_t)cond->ce32, errorCode); |
| 1427 | if(cond == lastCond) { break; } |
| 1428 | cond = getConditionalCE32(cond->next); |
| 1429 | } |
| 1430 | int32_t index = addContextTrie(emptySuffixCE32, contractionBuilder, errorCode); |
| 1431 | if(U_FAILURE(errorCode)) { return 0; } |
| 1432 | if(index > Collation::MAX_INDEX) { |
| 1433 | errorCode = U_BUFFER_OVERFLOW_ERROR; |
| 1434 | return 0; |
| 1435 | } |
| 1436 | ce32 = Collation::makeCE32FromTagAndIndex(Collation::CONTRACTION_TAG, index) | flags; |
| 1437 | } |
| 1438 | U_ASSERT(cond == lastCond); |
| 1439 | firstCond->defaultCE32 = ce32; |
| 1440 | if(prefixLength == 0) { |
| 1441 | if(cond->next < 0) { |
| 1442 | // No non-empty prefixes, only contractions. |
| 1443 | return ce32; |
| 1444 | } |
| 1445 | } else { |
| 1446 | prefix.remove(0, 1); // Remove the length unit. |
| 1447 | prefix.reverse(); |
| 1448 | prefixBuilder.add(prefix, (int32_t)ce32, errorCode); |
| 1449 | if(cond->next < 0) { break; } |
| 1450 | } |
| 1451 | } |
| 1452 | U_ASSERT(head->defaultCE32 != Collation::NO_CE32); |
| 1453 | int32_t index = addContextTrie(head->defaultCE32, prefixBuilder, errorCode); |
| 1454 | if(U_FAILURE(errorCode)) { return 0; } |
| 1455 | if(index > Collation::MAX_INDEX) { |
| 1456 | errorCode = U_BUFFER_OVERFLOW_ERROR; |
| 1457 | return 0; |
| 1458 | } |
| 1459 | return Collation::makeCE32FromTagAndIndex(Collation::PREFIX_TAG, index); |
| 1460 | } |
| 1461 | |
| 1462 | int32_t |
| 1463 | CollationDataBuilder::addContextTrie(uint32_t defaultCE32, UCharsTrieBuilder &trieBuilder, |
| 1464 | UErrorCode &errorCode) { |
| 1465 | UnicodeString context; |
| 1466 | context.append((UChar)(defaultCE32 >> 16)).append((UChar)defaultCE32); |
| 1467 | UnicodeString trieString; |
| 1468 | context.append(trieBuilder.buildUnicodeString(USTRINGTRIE_BUILD_SMALL, trieString, errorCode)); |
| 1469 | if(U_FAILURE(errorCode)) { return -1; } |
| 1470 | int32_t index = contexts.indexOf(context); |
| 1471 | if(index < 0) { |
| 1472 | index = contexts.length(); |
| 1473 | contexts.append(context); |
| 1474 | } |
| 1475 | return index; |
| 1476 | } |
| 1477 | |
| 1478 | void |
| 1479 | CollationDataBuilder::buildFastLatinTable(CollationData &data, UErrorCode &errorCode) { |
| 1480 | if(U_FAILURE(errorCode) || !fastLatinEnabled) { return; } |
| 1481 | |
| 1482 | delete fastLatinBuilder; |
| 1483 | fastLatinBuilder = new CollationFastLatinBuilder(errorCode); |
| 1484 | if(fastLatinBuilder == NULL) { |
| 1485 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
| 1486 | return; |
| 1487 | } |
| 1488 | if(fastLatinBuilder->forData(data, errorCode)) { |
| 1489 | const uint16_t *table = fastLatinBuilder->getTable(); |
| 1490 | int32_t length = fastLatinBuilder->lengthOfTable(); |
| 1491 | if(base != NULL && length == base->fastLatinTableLength && |
| 1492 | uprv_memcmp(table, base->fastLatinTable, length * 2) == 0) { |
| 1493 | // Same fast Latin table as in the base, use that one instead. |
| 1494 | delete fastLatinBuilder; |
| 1495 | fastLatinBuilder = NULL; |
| 1496 | table = base->fastLatinTable; |
| 1497 | } |
| 1498 | data.fastLatinTable = table; |
| 1499 | data.fastLatinTableLength = length; |
| 1500 | } else { |
| 1501 | delete fastLatinBuilder; |
| 1502 | fastLatinBuilder = NULL; |
| 1503 | } |
| 1504 | } |
| 1505 | |
| 1506 | int32_t |
| 1507 | CollationDataBuilder::getCEs(const UnicodeString &s, int64_t ces[], int32_t cesLength) { |
| 1508 | return getCEs(s, 0, ces, cesLength); |
| 1509 | } |
| 1510 | |
| 1511 | int32_t |
| 1512 | CollationDataBuilder::getCEs(const UnicodeString &prefix, const UnicodeString &s, |
| 1513 | int64_t ces[], int32_t cesLength) { |
| 1514 | int32_t prefixLength = prefix.length(); |
| 1515 | if(prefixLength == 0) { |
| 1516 | return getCEs(s, 0, ces, cesLength); |
| 1517 | } else { |
| 1518 | return getCEs(prefix + s, prefixLength, ces, cesLength); |
| 1519 | } |
| 1520 | } |
| 1521 | |
| 1522 | int32_t |
| 1523 | CollationDataBuilder::getCEs(const UnicodeString &s, int32_t start, |
| 1524 | int64_t ces[], int32_t cesLength) { |
| 1525 | if(collIter == NULL) { |
| 1526 | collIter = new DataBuilderCollationIterator(*this); |
| 1527 | if(collIter == NULL) { return 0; } |
| 1528 | } |
| 1529 | return collIter->fetchCEs(s, start, ces, cesLength); |
| 1530 | } |
| 1531 | |
| 1532 | U_NAMESPACE_END |
| 1533 | |
| 1534 | #endif // !UCONFIG_NO_COLLATION |
| 1535 | |