| 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 | * collationkeys.cpp |
| 9 | * |
| 10 | * created on: 2012sep02 |
| 11 | * created by: Markus W. Scherer |
| 12 | */ |
| 13 | |
| 14 | #include "unicode/utypes.h" |
| 15 | |
| 16 | #if !UCONFIG_NO_COLLATION |
| 17 | |
| 18 | #include "unicode/bytestream.h" |
| 19 | #include "collation.h" |
| 20 | #include "collationiterator.h" |
| 21 | #include "collationkeys.h" |
| 22 | #include "collationsettings.h" |
| 23 | #include "uassert.h" |
| 24 | |
| 25 | U_NAMESPACE_BEGIN |
| 26 | |
| 27 | SortKeyByteSink::~SortKeyByteSink() {} |
| 28 | |
| 29 | void |
| 30 | SortKeyByteSink::Append(const char *bytes, int32_t n) { |
| 31 | if (n <= 0 || bytes == NULL) { |
| 32 | return; |
| 33 | } |
| 34 | if (ignore_ > 0) { |
| 35 | int32_t ignoreRest = ignore_ - n; |
| 36 | if (ignoreRest >= 0) { |
| 37 | ignore_ = ignoreRest; |
| 38 | return; |
| 39 | } else { |
| 40 | bytes += ignore_; |
| 41 | n = -ignoreRest; |
| 42 | ignore_ = 0; |
| 43 | } |
| 44 | } |
| 45 | int32_t length = appended_; |
| 46 | appended_ += n; |
| 47 | if ((buffer_ + length) == bytes) { |
| 48 | return; // the caller used GetAppendBuffer() and wrote the bytes already |
| 49 | } |
| 50 | int32_t available = capacity_ - length; |
| 51 | if (n <= available) { |
| 52 | uprv_memcpy(buffer_ + length, bytes, n); |
| 53 | } else { |
| 54 | AppendBeyondCapacity(bytes, n, length); |
| 55 | } |
| 56 | } |
| 57 | |
| 58 | char * |
| 59 | SortKeyByteSink::GetAppendBuffer(int32_t min_capacity, |
| 60 | int32_t desired_capacity_hint, |
| 61 | char *scratch, |
| 62 | int32_t scratch_capacity, |
| 63 | int32_t *result_capacity) { |
| 64 | if (min_capacity < 1 || scratch_capacity < min_capacity) { |
| 65 | *result_capacity = 0; |
| 66 | return NULL; |
| 67 | } |
| 68 | if (ignore_ > 0) { |
| 69 | // Do not write ignored bytes right at the end of the buffer. |
| 70 | *result_capacity = scratch_capacity; |
| 71 | return scratch; |
| 72 | } |
| 73 | int32_t available = capacity_ - appended_; |
| 74 | if (available >= min_capacity) { |
| 75 | *result_capacity = available; |
| 76 | return buffer_ + appended_; |
| 77 | } else if (Resize(desired_capacity_hint, appended_)) { |
| 78 | *result_capacity = capacity_ - appended_; |
| 79 | return buffer_ + appended_; |
| 80 | } else { |
| 81 | *result_capacity = scratch_capacity; |
| 82 | return scratch; |
| 83 | } |
| 84 | } |
| 85 | |
| 86 | namespace { |
| 87 | |
| 88 | /** |
| 89 | * uint8_t byte buffer, similar to CharString but simpler. |
| 90 | */ |
| 91 | class SortKeyLevel : public UMemory { |
| 92 | public: |
| 93 | SortKeyLevel() : len(0), ok(TRUE) {} |
| 94 | ~SortKeyLevel() {} |
| 95 | |
| 96 | /** @return FALSE if memory allocation failed */ |
| 97 | UBool isOk() const { return ok; } |
| 98 | UBool isEmpty() const { return len == 0; } |
| 99 | int32_t length() const { return len; } |
| 100 | const uint8_t *data() const { return buffer.getAlias(); } |
| 101 | uint8_t operator[](int32_t index) const { return buffer[index]; } |
| 102 | |
| 103 | uint8_t *data() { return buffer.getAlias(); } |
| 104 | |
| 105 | void appendByte(uint32_t b); |
| 106 | void appendWeight16(uint32_t w); |
| 107 | void appendWeight32(uint32_t w); |
| 108 | void appendReverseWeight16(uint32_t w); |
| 109 | |
| 110 | /** Appends all but the last byte to the sink. The last byte should be the 01 terminator. */ |
| 111 | void appendTo(ByteSink &sink) const { |
| 112 | U_ASSERT(len > 0 && buffer[len - 1] == 1); |
| 113 | sink.Append(reinterpret_cast<const char *>(buffer.getAlias()), len - 1); |
| 114 | } |
| 115 | |
| 116 | private: |
| 117 | MaybeStackArray<uint8_t, 40> buffer; |
| 118 | int32_t len; |
| 119 | UBool ok; |
| 120 | |
| 121 | UBool ensureCapacity(int32_t appendCapacity); |
| 122 | |
| 123 | SortKeyLevel(const SortKeyLevel &other); // forbid copying of this class |
| 124 | SortKeyLevel &operator=(const SortKeyLevel &other); // forbid copying of this class |
| 125 | }; |
| 126 | |
| 127 | void SortKeyLevel::appendByte(uint32_t b) { |
| 128 | if(len < buffer.getCapacity() || ensureCapacity(1)) { |
| 129 | buffer[len++] = (uint8_t)b; |
| 130 | } |
| 131 | } |
| 132 | |
| 133 | void |
| 134 | SortKeyLevel::appendWeight16(uint32_t w) { |
| 135 | U_ASSERT((w & 0xffff) != 0); |
| 136 | uint8_t b0 = (uint8_t)(w >> 8); |
| 137 | uint8_t b1 = (uint8_t)w; |
| 138 | int32_t appendLength = (b1 == 0) ? 1 : 2; |
| 139 | if((len + appendLength) <= buffer.getCapacity() || ensureCapacity(appendLength)) { |
| 140 | buffer[len++] = b0; |
| 141 | if(b1 != 0) { |
| 142 | buffer[len++] = b1; |
| 143 | } |
| 144 | } |
| 145 | } |
| 146 | |
| 147 | void |
| 148 | SortKeyLevel::appendWeight32(uint32_t w) { |
| 149 | U_ASSERT(w != 0); |
| 150 | uint8_t bytes[4] = { (uint8_t)(w >> 24), (uint8_t)(w >> 16), (uint8_t)(w >> 8), (uint8_t)w }; |
| 151 | int32_t appendLength = (bytes[1] == 0) ? 1 : (bytes[2] == 0) ? 2 : (bytes[3] == 0) ? 3 : 4; |
| 152 | if((len + appendLength) <= buffer.getCapacity() || ensureCapacity(appendLength)) { |
| 153 | buffer[len++] = bytes[0]; |
| 154 | if(bytes[1] != 0) { |
| 155 | buffer[len++] = bytes[1]; |
| 156 | if(bytes[2] != 0) { |
| 157 | buffer[len++] = bytes[2]; |
| 158 | if(bytes[3] != 0) { |
| 159 | buffer[len++] = bytes[3]; |
| 160 | } |
| 161 | } |
| 162 | } |
| 163 | } |
| 164 | } |
| 165 | |
| 166 | void |
| 167 | SortKeyLevel::appendReverseWeight16(uint32_t w) { |
| 168 | U_ASSERT((w & 0xffff) != 0); |
| 169 | uint8_t b0 = (uint8_t)(w >> 8); |
| 170 | uint8_t b1 = (uint8_t)w; |
| 171 | int32_t appendLength = (b1 == 0) ? 1 : 2; |
| 172 | if((len + appendLength) <= buffer.getCapacity() || ensureCapacity(appendLength)) { |
| 173 | if(b1 == 0) { |
| 174 | buffer[len++] = b0; |
| 175 | } else { |
| 176 | buffer[len] = b1; |
| 177 | buffer[len + 1] = b0; |
| 178 | len += 2; |
| 179 | } |
| 180 | } |
| 181 | } |
| 182 | |
| 183 | UBool SortKeyLevel::ensureCapacity(int32_t appendCapacity) { |
| 184 | if(!ok) { |
| 185 | return FALSE; |
| 186 | } |
| 187 | int32_t newCapacity = 2 * buffer.getCapacity(); |
| 188 | int32_t altCapacity = len + 2 * appendCapacity; |
| 189 | if (newCapacity < altCapacity) { |
| 190 | newCapacity = altCapacity; |
| 191 | } |
| 192 | if (newCapacity < 200) { |
| 193 | newCapacity = 200; |
| 194 | } |
| 195 | if(buffer.resize(newCapacity, len)==NULL) { |
| 196 | return ok = FALSE; |
| 197 | } |
| 198 | return TRUE; |
| 199 | } |
| 200 | |
| 201 | } // namespace |
| 202 | |
| 203 | CollationKeys::LevelCallback::~LevelCallback() {} |
| 204 | |
| 205 | UBool |
| 206 | CollationKeys::LevelCallback::needToWrite(Collation::Level /*level*/) { return TRUE; } |
| 207 | |
| 208 | /** |
| 209 | * Map from collation strength (UColAttributeValue) |
| 210 | * to a mask of Collation::Level bits up to that strength, |
| 211 | * excluding the CASE_LEVEL which is independent of the strength, |
| 212 | * and excluding IDENTICAL_LEVEL which this function does not write. |
| 213 | */ |
| 214 | static const uint32_t levelMasks[UCOL_STRENGTH_LIMIT] = { |
| 215 | 2, // UCOL_PRIMARY -> PRIMARY_LEVEL |
| 216 | 6, // UCOL_SECONDARY -> up to SECONDARY_LEVEL |
| 217 | 0x16, // UCOL_TERTIARY -> up to TERTIARY_LEVEL |
| 218 | 0x36, // UCOL_QUATERNARY -> up to QUATERNARY_LEVEL |
| 219 | 0, 0, 0, 0, |
| 220 | 0, 0, 0, 0, |
| 221 | 0, 0, 0, |
| 222 | 0x36 // UCOL_IDENTICAL -> up to QUATERNARY_LEVEL |
| 223 | }; |
| 224 | |
| 225 | void |
| 226 | CollationKeys::writeSortKeyUpToQuaternary(CollationIterator &iter, |
| 227 | const UBool *compressibleBytes, |
| 228 | const CollationSettings &settings, |
| 229 | SortKeyByteSink &sink, |
| 230 | Collation::Level minLevel, LevelCallback &callback, |
| 231 | UBool preflight, UErrorCode &errorCode) { |
| 232 | if(U_FAILURE(errorCode)) { return; } |
| 233 | |
| 234 | int32_t options = settings.options; |
| 235 | // Set of levels to process and write. |
| 236 | uint32_t levels = levelMasks[CollationSettings::getStrength(options)]; |
| 237 | if((options & CollationSettings::CASE_LEVEL) != 0) { |
| 238 | levels |= Collation::CASE_LEVEL_FLAG; |
| 239 | } |
| 240 | // Minus the levels below minLevel. |
| 241 | levels &= ~(((uint32_t)1 << minLevel) - 1); |
| 242 | if(levels == 0) { return; } |
| 243 | |
| 244 | uint32_t variableTop; |
| 245 | if((options & CollationSettings::ALTERNATE_MASK) == 0) { |
| 246 | variableTop = 0; |
| 247 | } else { |
| 248 | // +1 so that we can use "<" and primary ignorables test out early. |
| 249 | variableTop = settings.variableTop + 1; |
| 250 | } |
| 251 | |
| 252 | uint32_t tertiaryMask = CollationSettings::getTertiaryMask(options); |
| 253 | |
| 254 | SortKeyLevel cases; |
| 255 | SortKeyLevel secondaries; |
| 256 | SortKeyLevel tertiaries; |
| 257 | SortKeyLevel quaternaries; |
| 258 | |
| 259 | uint32_t prevReorderedPrimary = 0; // 0==no compression |
| 260 | int32_t commonCases = 0; |
| 261 | int32_t commonSecondaries = 0; |
| 262 | int32_t commonTertiaries = 0; |
| 263 | int32_t commonQuaternaries = 0; |
| 264 | |
| 265 | uint32_t prevSecondary = 0; |
| 266 | int32_t secSegmentStart = 0; |
| 267 | |
| 268 | for(;;) { |
| 269 | // No need to keep all CEs in the buffer when we write a sort key. |
| 270 | iter.clearCEsIfNoneRemaining(); |
| 271 | int64_t ce = iter.nextCE(errorCode); |
| 272 | uint32_t p = (uint32_t)(ce >> 32); |
| 273 | if(p < variableTop && p > Collation::MERGE_SEPARATOR_PRIMARY) { |
| 274 | // Variable CE, shift it to quaternary level. |
| 275 | // Ignore all following primary ignorables, and shift further variable CEs. |
| 276 | if(commonQuaternaries != 0) { |
| 277 | --commonQuaternaries; |
| 278 | while(commonQuaternaries >= QUAT_COMMON_MAX_COUNT) { |
| 279 | quaternaries.appendByte(QUAT_COMMON_MIDDLE); |
| 280 | commonQuaternaries -= QUAT_COMMON_MAX_COUNT; |
| 281 | } |
| 282 | // Shifted primary weights are lower than the common weight. |
| 283 | quaternaries.appendByte(QUAT_COMMON_LOW + commonQuaternaries); |
| 284 | commonQuaternaries = 0; |
| 285 | } |
| 286 | do { |
| 287 | if((levels & Collation::QUATERNARY_LEVEL_FLAG) != 0) { |
| 288 | if(settings.hasReordering()) { |
| 289 | p = settings.reorder(p); |
| 290 | } |
| 291 | if((p >> 24) >= QUAT_SHIFTED_LIMIT_BYTE) { |
| 292 | // Prevent shifted primary lead bytes from |
| 293 | // overlapping with the common compression range. |
| 294 | quaternaries.appendByte(QUAT_SHIFTED_LIMIT_BYTE); |
| 295 | } |
| 296 | quaternaries.appendWeight32(p); |
| 297 | } |
| 298 | do { |
| 299 | ce = iter.nextCE(errorCode); |
| 300 | p = (uint32_t)(ce >> 32); |
| 301 | } while(p == 0); |
| 302 | } while(p < variableTop && p > Collation::MERGE_SEPARATOR_PRIMARY); |
| 303 | } |
| 304 | // ce could be primary ignorable, or NO_CE, or the merge separator, |
| 305 | // or a regular primary CE, but it is not variable. |
| 306 | // If ce==NO_CE, then write nothing for the primary level but |
| 307 | // terminate compression on all levels and then exit the loop. |
| 308 | if(p > Collation::NO_CE_PRIMARY && (levels & Collation::PRIMARY_LEVEL_FLAG) != 0) { |
| 309 | // Test the un-reordered primary for compressibility. |
| 310 | UBool isCompressible = compressibleBytes[p >> 24]; |
| 311 | if(settings.hasReordering()) { |
| 312 | p = settings.reorder(p); |
| 313 | } |
| 314 | uint32_t p1 = p >> 24; |
| 315 | if(!isCompressible || p1 != (prevReorderedPrimary >> 24)) { |
| 316 | if(prevReorderedPrimary != 0) { |
| 317 | if(p < prevReorderedPrimary) { |
| 318 | // No primary compression terminator |
| 319 | // at the end of the level or merged segment. |
| 320 | if(p1 > Collation::MERGE_SEPARATOR_BYTE) { |
| 321 | sink.Append(Collation::PRIMARY_COMPRESSION_LOW_BYTE); |
| 322 | } |
| 323 | } else { |
| 324 | sink.Append(Collation::PRIMARY_COMPRESSION_HIGH_BYTE); |
| 325 | } |
| 326 | } |
| 327 | sink.Append(p1); |
| 328 | if(isCompressible) { |
| 329 | prevReorderedPrimary = p; |
| 330 | } else { |
| 331 | prevReorderedPrimary = 0; |
| 332 | } |
| 333 | } |
| 334 | char p2 = (char)(p >> 16); |
| 335 | if(p2 != 0) { |
| 336 | char buffer[3] = { p2, (char)(p >> 8), (char)p }; |
| 337 | sink.Append(buffer, (buffer[1] == 0) ? 1 : (buffer[2] == 0) ? 2 : 3); |
| 338 | } |
| 339 | // Optimization for internalNextSortKeyPart(): |
| 340 | // When the primary level overflows we can stop because we need not |
| 341 | // calculate (preflight) the whole sort key length. |
| 342 | if(!preflight && sink.Overflowed()) { |
| 343 | if(U_SUCCESS(errorCode) && !sink.IsOk()) { |
| 344 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
| 345 | } |
| 346 | return; |
| 347 | } |
| 348 | } |
| 349 | |
| 350 | uint32_t lower32 = (uint32_t)ce; |
| 351 | if(lower32 == 0) { continue; } // completely ignorable, no secondary/case/tertiary/quaternary |
| 352 | |
| 353 | if((levels & Collation::SECONDARY_LEVEL_FLAG) != 0) { |
| 354 | uint32_t s = lower32 >> 16; |
| 355 | if(s == 0) { |
| 356 | // secondary ignorable |
| 357 | } else if(s == Collation::COMMON_WEIGHT16 && |
| 358 | ((options & CollationSettings::BACKWARD_SECONDARY) == 0 || |
| 359 | p != Collation::MERGE_SEPARATOR_PRIMARY)) { |
| 360 | // s is a common secondary weight, and |
| 361 | // backwards-secondary is off or the ce is not the merge separator. |
| 362 | ++commonSecondaries; |
| 363 | } else if((options & CollationSettings::BACKWARD_SECONDARY) == 0) { |
| 364 | if(commonSecondaries != 0) { |
| 365 | --commonSecondaries; |
| 366 | while(commonSecondaries >= SEC_COMMON_MAX_COUNT) { |
| 367 | secondaries.appendByte(SEC_COMMON_MIDDLE); |
| 368 | commonSecondaries -= SEC_COMMON_MAX_COUNT; |
| 369 | } |
| 370 | uint32_t b; |
| 371 | if(s < Collation::COMMON_WEIGHT16) { |
| 372 | b = SEC_COMMON_LOW + commonSecondaries; |
| 373 | } else { |
| 374 | b = SEC_COMMON_HIGH - commonSecondaries; |
| 375 | } |
| 376 | secondaries.appendByte(b); |
| 377 | commonSecondaries = 0; |
| 378 | } |
| 379 | secondaries.appendWeight16(s); |
| 380 | } else { |
| 381 | if(commonSecondaries != 0) { |
| 382 | --commonSecondaries; |
| 383 | // Append reverse weights. The level will be re-reversed later. |
| 384 | int32_t remainder = commonSecondaries % SEC_COMMON_MAX_COUNT; |
| 385 | uint32_t b; |
| 386 | if(prevSecondary < Collation::COMMON_WEIGHT16) { |
| 387 | b = SEC_COMMON_LOW + remainder; |
| 388 | } else { |
| 389 | b = SEC_COMMON_HIGH - remainder; |
| 390 | } |
| 391 | secondaries.appendByte(b); |
| 392 | commonSecondaries -= remainder; |
| 393 | // commonSecondaries is now a multiple of SEC_COMMON_MAX_COUNT. |
| 394 | while(commonSecondaries > 0) { // same as >= SEC_COMMON_MAX_COUNT |
| 395 | secondaries.appendByte(SEC_COMMON_MIDDLE); |
| 396 | commonSecondaries -= SEC_COMMON_MAX_COUNT; |
| 397 | } |
| 398 | // commonSecondaries == 0 |
| 399 | } |
| 400 | if(0 < p && p <= Collation::MERGE_SEPARATOR_PRIMARY) { |
| 401 | // The backwards secondary level compares secondary weights backwards |
| 402 | // within segments separated by the merge separator (U+FFFE). |
| 403 | uint8_t *secs = secondaries.data(); |
| 404 | int32_t last = secondaries.length() - 1; |
| 405 | if(secSegmentStart < last) { |
| 406 | uint8_t *q = secs + secSegmentStart; |
| 407 | uint8_t *r = secs + last; |
| 408 | do { |
| 409 | uint8_t b = *q; |
| 410 | *q++ = *r; |
| 411 | *r-- = b; |
| 412 | } while(q < r); |
| 413 | } |
| 414 | secondaries.appendByte(p == Collation::NO_CE_PRIMARY ? |
| 415 | Collation::LEVEL_SEPARATOR_BYTE : Collation::MERGE_SEPARATOR_BYTE); |
| 416 | prevSecondary = 0; |
| 417 | secSegmentStart = secondaries.length(); |
| 418 | } else { |
| 419 | secondaries.appendReverseWeight16(s); |
| 420 | prevSecondary = s; |
| 421 | } |
| 422 | } |
| 423 | } |
| 424 | |
| 425 | if((levels & Collation::CASE_LEVEL_FLAG) != 0) { |
| 426 | if((CollationSettings::getStrength(options) == UCOL_PRIMARY) ? |
| 427 | p == 0 : lower32 <= 0xffff) { |
| 428 | // Primary+caseLevel: Ignore case level weights of primary ignorables. |
| 429 | // Otherwise: Ignore case level weights of secondary ignorables. |
| 430 | // For details see the comments in the CollationCompare class. |
| 431 | } else { |
| 432 | uint32_t c = (lower32 >> 8) & 0xff; // case bits & tertiary lead byte |
| 433 | U_ASSERT((c & 0xc0) != 0xc0); |
| 434 | if((c & 0xc0) == 0 && c > Collation::LEVEL_SEPARATOR_BYTE) { |
| 435 | ++commonCases; |
| 436 | } else { |
| 437 | if((options & CollationSettings::UPPER_FIRST) == 0) { |
| 438 | // lowerFirst: Compress common weights to nibbles 1..7..13, mixed=14, upper=15. |
| 439 | // If there are only common (=lowest) weights in the whole level, |
| 440 | // then we need not write anything. |
| 441 | // Level length differences are handled already on the next-higher level. |
| 442 | if(commonCases != 0 && |
| 443 | (c > Collation::LEVEL_SEPARATOR_BYTE || !cases.isEmpty())) { |
| 444 | --commonCases; |
| 445 | while(commonCases >= CASE_LOWER_FIRST_COMMON_MAX_COUNT) { |
| 446 | cases.appendByte(CASE_LOWER_FIRST_COMMON_MIDDLE << 4); |
| 447 | commonCases -= CASE_LOWER_FIRST_COMMON_MAX_COUNT; |
| 448 | } |
| 449 | uint32_t b; |
| 450 | if(c <= Collation::LEVEL_SEPARATOR_BYTE) { |
| 451 | b = CASE_LOWER_FIRST_COMMON_LOW + commonCases; |
| 452 | } else { |
| 453 | b = CASE_LOWER_FIRST_COMMON_HIGH - commonCases; |
| 454 | } |
| 455 | cases.appendByte(b << 4); |
| 456 | commonCases = 0; |
| 457 | } |
| 458 | if(c > Collation::LEVEL_SEPARATOR_BYTE) { |
| 459 | c = (CASE_LOWER_FIRST_COMMON_HIGH + (c >> 6)) << 4; // 14 or 15 |
| 460 | } |
| 461 | } else { |
| 462 | // upperFirst: Compress common weights to nibbles 3..15, mixed=2, upper=1. |
| 463 | // The compressed common case weights only go up from the "low" value |
| 464 | // because with upperFirst the common weight is the highest one. |
| 465 | if(commonCases != 0) { |
| 466 | --commonCases; |
| 467 | while(commonCases >= CASE_UPPER_FIRST_COMMON_MAX_COUNT) { |
| 468 | cases.appendByte(CASE_UPPER_FIRST_COMMON_LOW << 4); |
| 469 | commonCases -= CASE_UPPER_FIRST_COMMON_MAX_COUNT; |
| 470 | } |
| 471 | cases.appendByte((CASE_UPPER_FIRST_COMMON_LOW + commonCases) << 4); |
| 472 | commonCases = 0; |
| 473 | } |
| 474 | if(c > Collation::LEVEL_SEPARATOR_BYTE) { |
| 475 | c = (CASE_UPPER_FIRST_COMMON_LOW - (c >> 6)) << 4; // 2 or 1 |
| 476 | } |
| 477 | } |
| 478 | // c is a separator byte 01, |
| 479 | // or a left-shifted nibble 0x10, 0x20, ... 0xf0. |
| 480 | cases.appendByte(c); |
| 481 | } |
| 482 | } |
| 483 | } |
| 484 | |
| 485 | if((levels & Collation::TERTIARY_LEVEL_FLAG) != 0) { |
| 486 | uint32_t t = lower32 & tertiaryMask; |
| 487 | U_ASSERT((lower32 & 0xc000) != 0xc000); |
| 488 | if(t == Collation::COMMON_WEIGHT16) { |
| 489 | ++commonTertiaries; |
| 490 | } else if((tertiaryMask & 0x8000) == 0) { |
| 491 | // Tertiary weights without case bits. |
| 492 | // Move lead bytes 06..3F to C6..FF for a large common-weight range. |
| 493 | if(commonTertiaries != 0) { |
| 494 | --commonTertiaries; |
| 495 | while(commonTertiaries >= TER_ONLY_COMMON_MAX_COUNT) { |
| 496 | tertiaries.appendByte(TER_ONLY_COMMON_MIDDLE); |
| 497 | commonTertiaries -= TER_ONLY_COMMON_MAX_COUNT; |
| 498 | } |
| 499 | uint32_t b; |
| 500 | if(t < Collation::COMMON_WEIGHT16) { |
| 501 | b = TER_ONLY_COMMON_LOW + commonTertiaries; |
| 502 | } else { |
| 503 | b = TER_ONLY_COMMON_HIGH - commonTertiaries; |
| 504 | } |
| 505 | tertiaries.appendByte(b); |
| 506 | commonTertiaries = 0; |
| 507 | } |
| 508 | if(t > Collation::COMMON_WEIGHT16) { t += 0xc000; } |
| 509 | tertiaries.appendWeight16(t); |
| 510 | } else if((options & CollationSettings::UPPER_FIRST) == 0) { |
| 511 | // Tertiary weights with caseFirst=lowerFirst. |
| 512 | // Move lead bytes 06..BF to 46..FF for the common-weight range. |
| 513 | if(commonTertiaries != 0) { |
| 514 | --commonTertiaries; |
| 515 | while(commonTertiaries >= TER_LOWER_FIRST_COMMON_MAX_COUNT) { |
| 516 | tertiaries.appendByte(TER_LOWER_FIRST_COMMON_MIDDLE); |
| 517 | commonTertiaries -= TER_LOWER_FIRST_COMMON_MAX_COUNT; |
| 518 | } |
| 519 | uint32_t b; |
| 520 | if(t < Collation::COMMON_WEIGHT16) { |
| 521 | b = TER_LOWER_FIRST_COMMON_LOW + commonTertiaries; |
| 522 | } else { |
| 523 | b = TER_LOWER_FIRST_COMMON_HIGH - commonTertiaries; |
| 524 | } |
| 525 | tertiaries.appendByte(b); |
| 526 | commonTertiaries = 0; |
| 527 | } |
| 528 | if(t > Collation::COMMON_WEIGHT16) { t += 0x4000; } |
| 529 | tertiaries.appendWeight16(t); |
| 530 | } else { |
| 531 | // Tertiary weights with caseFirst=upperFirst. |
| 532 | // Do not change the artificial uppercase weight of a tertiary CE (0.0.ut), |
| 533 | // to keep tertiary CEs well-formed. |
| 534 | // Their case+tertiary weights must be greater than those of |
| 535 | // primary and secondary CEs. |
| 536 | // |
| 537 | // Separator 01 -> 01 (unchanged) |
| 538 | // Lowercase 02..04 -> 82..84 (includes uncased) |
| 539 | // Common weight 05 -> 85..C5 (common-weight compression range) |
| 540 | // Lowercase 06..3F -> C6..FF |
| 541 | // Mixed case 42..7F -> 42..7F |
| 542 | // Uppercase 82..BF -> 02..3F |
| 543 | // Tertiary CE 86..BF -> C6..FF |
| 544 | if(t <= Collation::NO_CE_WEIGHT16) { |
| 545 | // Keep separators unchanged. |
| 546 | } else if(lower32 > 0xffff) { |
| 547 | // Invert case bits of primary & secondary CEs. |
| 548 | t ^= 0xc000; |
| 549 | if(t < (TER_UPPER_FIRST_COMMON_HIGH << 8)) { |
| 550 | t -= 0x4000; |
| 551 | } |
| 552 | } else { |
| 553 | // Keep uppercase bits of tertiary CEs. |
| 554 | U_ASSERT(0x8600 <= t && t <= 0xbfff); |
| 555 | t += 0x4000; |
| 556 | } |
| 557 | if(commonTertiaries != 0) { |
| 558 | --commonTertiaries; |
| 559 | while(commonTertiaries >= TER_UPPER_FIRST_COMMON_MAX_COUNT) { |
| 560 | tertiaries.appendByte(TER_UPPER_FIRST_COMMON_MIDDLE); |
| 561 | commonTertiaries -= TER_UPPER_FIRST_COMMON_MAX_COUNT; |
| 562 | } |
| 563 | uint32_t b; |
| 564 | if(t < (TER_UPPER_FIRST_COMMON_LOW << 8)) { |
| 565 | b = TER_UPPER_FIRST_COMMON_LOW + commonTertiaries; |
| 566 | } else { |
| 567 | b = TER_UPPER_FIRST_COMMON_HIGH - commonTertiaries; |
| 568 | } |
| 569 | tertiaries.appendByte(b); |
| 570 | commonTertiaries = 0; |
| 571 | } |
| 572 | tertiaries.appendWeight16(t); |
| 573 | } |
| 574 | } |
| 575 | |
| 576 | if((levels & Collation::QUATERNARY_LEVEL_FLAG) != 0) { |
| 577 | uint32_t q = lower32 & 0xffff; |
| 578 | if((q & 0xc0) == 0 && q > Collation::NO_CE_WEIGHT16) { |
| 579 | ++commonQuaternaries; |
| 580 | } else if(q == Collation::NO_CE_WEIGHT16 && |
| 581 | (options & CollationSettings::ALTERNATE_MASK) == 0 && |
| 582 | quaternaries.isEmpty()) { |
| 583 | // If alternate=non-ignorable and there are only common quaternary weights, |
| 584 | // then we need not write anything. |
| 585 | // The only weights greater than the merge separator and less than the common weight |
| 586 | // are shifted primary weights, which are not generated for alternate=non-ignorable. |
| 587 | // There are also exactly as many quaternary weights as tertiary weights, |
| 588 | // so level length differences are handled already on tertiary level. |
| 589 | // Any above-common quaternary weight will compare greater regardless. |
| 590 | quaternaries.appendByte(Collation::LEVEL_SEPARATOR_BYTE); |
| 591 | } else { |
| 592 | if(q == Collation::NO_CE_WEIGHT16) { |
| 593 | q = Collation::LEVEL_SEPARATOR_BYTE; |
| 594 | } else { |
| 595 | q = 0xfc + ((q >> 6) & 3); |
| 596 | } |
| 597 | if(commonQuaternaries != 0) { |
| 598 | --commonQuaternaries; |
| 599 | while(commonQuaternaries >= QUAT_COMMON_MAX_COUNT) { |
| 600 | quaternaries.appendByte(QUAT_COMMON_MIDDLE); |
| 601 | commonQuaternaries -= QUAT_COMMON_MAX_COUNT; |
| 602 | } |
| 603 | uint32_t b; |
| 604 | if(q < QUAT_COMMON_LOW) { |
| 605 | b = QUAT_COMMON_LOW + commonQuaternaries; |
| 606 | } else { |
| 607 | b = QUAT_COMMON_HIGH - commonQuaternaries; |
| 608 | } |
| 609 | quaternaries.appendByte(b); |
| 610 | commonQuaternaries = 0; |
| 611 | } |
| 612 | quaternaries.appendByte(q); |
| 613 | } |
| 614 | } |
| 615 | |
| 616 | if((lower32 >> 24) == Collation::LEVEL_SEPARATOR_BYTE) { break; } // ce == NO_CE |
| 617 | } |
| 618 | |
| 619 | if(U_FAILURE(errorCode)) { return; } |
| 620 | |
| 621 | // Append the beyond-primary levels. |
| 622 | UBool ok = TRUE; |
| 623 | if((levels & Collation::SECONDARY_LEVEL_FLAG) != 0) { |
| 624 | if(!callback.needToWrite(Collation::SECONDARY_LEVEL)) { return; } |
| 625 | ok &= secondaries.isOk(); |
| 626 | sink.Append(Collation::LEVEL_SEPARATOR_BYTE); |
| 627 | secondaries.appendTo(sink); |
| 628 | } |
| 629 | |
| 630 | if((levels & Collation::CASE_LEVEL_FLAG) != 0) { |
| 631 | if(!callback.needToWrite(Collation::CASE_LEVEL)) { return; } |
| 632 | ok &= cases.isOk(); |
| 633 | sink.Append(Collation::LEVEL_SEPARATOR_BYTE); |
| 634 | // Write pairs of nibbles as bytes, except separator bytes as themselves. |
| 635 | int32_t length = cases.length() - 1; // Ignore the trailing NO_CE. |
| 636 | uint8_t b = 0; |
| 637 | for(int32_t i = 0; i < length; ++i) { |
| 638 | uint8_t c = (uint8_t)cases[i]; |
| 639 | U_ASSERT((c & 0xf) == 0 && c != 0); |
| 640 | if(b == 0) { |
| 641 | b = c; |
| 642 | } else { |
| 643 | sink.Append(b | (c >> 4)); |
| 644 | b = 0; |
| 645 | } |
| 646 | } |
| 647 | if(b != 0) { |
| 648 | sink.Append(b); |
| 649 | } |
| 650 | } |
| 651 | |
| 652 | if((levels & Collation::TERTIARY_LEVEL_FLAG) != 0) { |
| 653 | if(!callback.needToWrite(Collation::TERTIARY_LEVEL)) { return; } |
| 654 | ok &= tertiaries.isOk(); |
| 655 | sink.Append(Collation::LEVEL_SEPARATOR_BYTE); |
| 656 | tertiaries.appendTo(sink); |
| 657 | } |
| 658 | |
| 659 | if((levels & Collation::QUATERNARY_LEVEL_FLAG) != 0) { |
| 660 | if(!callback.needToWrite(Collation::QUATERNARY_LEVEL)) { return; } |
| 661 | ok &= quaternaries.isOk(); |
| 662 | sink.Append(Collation::LEVEL_SEPARATOR_BYTE); |
| 663 | quaternaries.appendTo(sink); |
| 664 | } |
| 665 | |
| 666 | if(!ok || !sink.IsOk()) { |
| 667 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
| 668 | } |
| 669 | } |
| 670 | |
| 671 | U_NAMESPACE_END |
| 672 | |
| 673 | #endif // !UCONFIG_NO_COLLATION |
| 674 | |