| 1 | // © 2016 and later: Unicode, Inc. and others. |
| 2 | // License & terms of use: http://www.unicode.org/copyright.html |
| 3 | /* |
| 4 | ****************************************************************************** |
| 5 | * Copyright (C) 2001-2016, International Business Machines |
| 6 | * Corporation and others. All Rights Reserved. |
| 7 | ****************************************************************************** |
| 8 | * |
| 9 | * File ucoleitr.cpp |
| 10 | * |
| 11 | * Modification History: |
| 12 | * |
| 13 | * Date Name Description |
| 14 | * 02/15/2001 synwee Modified all methods to process its own function |
| 15 | * instead of calling the equivalent c++ api (coleitr.h) |
| 16 | * 2012-2014 markus Rewritten in C++ again. |
| 17 | ******************************************************************************/ |
| 18 | |
| 19 | #include "unicode/utypes.h" |
| 20 | |
| 21 | #if !UCONFIG_NO_COLLATION |
| 22 | |
| 23 | #include "unicode/coleitr.h" |
| 24 | #include "unicode/tblcoll.h" |
| 25 | #include "unicode/ucoleitr.h" |
| 26 | #include "unicode/ustring.h" |
| 27 | #include "unicode/sortkey.h" |
| 28 | #include "unicode/uobject.h" |
| 29 | #include "cmemory.h" |
| 30 | #include "usrchimp.h" |
| 31 | |
| 32 | U_NAMESPACE_USE |
| 33 | |
| 34 | #define BUFFER_LENGTH 100 |
| 35 | |
| 36 | #define DEFAULT_BUFFER_SIZE 16 |
| 37 | #define BUFFER_GROW 8 |
| 38 | |
| 39 | #define ARRAY_COPY(dst, src, count) uprv_memcpy((void *) (dst), (void *) (src), (size_t)(count) * sizeof (src)[0]) |
| 40 | |
| 41 | #define NEW_ARRAY(type, count) (type *) uprv_malloc((size_t)(count) * sizeof(type)) |
| 42 | |
| 43 | #define DELETE_ARRAY(array) uprv_free((void *) (array)) |
| 44 | |
| 45 | struct RCEI |
| 46 | { |
| 47 | uint32_t ce; |
| 48 | int32_t low; |
| 49 | int32_t high; |
| 50 | }; |
| 51 | |
| 52 | U_NAMESPACE_BEGIN |
| 53 | |
| 54 | struct RCEBuffer |
| 55 | { |
| 56 | RCEI defaultBuffer[DEFAULT_BUFFER_SIZE]; |
| 57 | RCEI *buffer; |
| 58 | int32_t bufferIndex; |
| 59 | int32_t bufferSize; |
| 60 | |
| 61 | RCEBuffer(); |
| 62 | ~RCEBuffer(); |
| 63 | |
| 64 | UBool isEmpty() const; |
| 65 | void put(uint32_t ce, int32_t ixLow, int32_t ixHigh, UErrorCode &errorCode); |
| 66 | const RCEI *get(); |
| 67 | }; |
| 68 | |
| 69 | RCEBuffer::RCEBuffer() |
| 70 | { |
| 71 | buffer = defaultBuffer; |
| 72 | bufferIndex = 0; |
| 73 | bufferSize = UPRV_LENGTHOF(defaultBuffer); |
| 74 | } |
| 75 | |
| 76 | RCEBuffer::~RCEBuffer() |
| 77 | { |
| 78 | if (buffer != defaultBuffer) { |
| 79 | DELETE_ARRAY(buffer); |
| 80 | } |
| 81 | } |
| 82 | |
| 83 | UBool RCEBuffer::isEmpty() const |
| 84 | { |
| 85 | return bufferIndex <= 0; |
| 86 | } |
| 87 | |
| 88 | void RCEBuffer::put(uint32_t ce, int32_t ixLow, int32_t ixHigh, UErrorCode &errorCode) |
| 89 | { |
| 90 | if (U_FAILURE(errorCode)) { |
| 91 | return; |
| 92 | } |
| 93 | if (bufferIndex >= bufferSize) { |
| 94 | RCEI *newBuffer = NEW_ARRAY(RCEI, bufferSize + BUFFER_GROW); |
| 95 | if (newBuffer == NULL) { |
| 96 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
| 97 | return; |
| 98 | } |
| 99 | |
| 100 | ARRAY_COPY(newBuffer, buffer, bufferSize); |
| 101 | |
| 102 | if (buffer != defaultBuffer) { |
| 103 | DELETE_ARRAY(buffer); |
| 104 | } |
| 105 | |
| 106 | buffer = newBuffer; |
| 107 | bufferSize += BUFFER_GROW; |
| 108 | } |
| 109 | |
| 110 | buffer[bufferIndex].ce = ce; |
| 111 | buffer[bufferIndex].low = ixLow; |
| 112 | buffer[bufferIndex].high = ixHigh; |
| 113 | |
| 114 | bufferIndex += 1; |
| 115 | } |
| 116 | |
| 117 | const RCEI *RCEBuffer::get() |
| 118 | { |
| 119 | if (bufferIndex > 0) { |
| 120 | return &buffer[--bufferIndex]; |
| 121 | } |
| 122 | |
| 123 | return NULL; |
| 124 | } |
| 125 | |
| 126 | PCEBuffer::PCEBuffer() |
| 127 | { |
| 128 | buffer = defaultBuffer; |
| 129 | bufferIndex = 0; |
| 130 | bufferSize = UPRV_LENGTHOF(defaultBuffer); |
| 131 | } |
| 132 | |
| 133 | PCEBuffer::~PCEBuffer() |
| 134 | { |
| 135 | if (buffer != defaultBuffer) { |
| 136 | DELETE_ARRAY(buffer); |
| 137 | } |
| 138 | } |
| 139 | |
| 140 | void PCEBuffer::reset() |
| 141 | { |
| 142 | bufferIndex = 0; |
| 143 | } |
| 144 | |
| 145 | UBool PCEBuffer::isEmpty() const |
| 146 | { |
| 147 | return bufferIndex <= 0; |
| 148 | } |
| 149 | |
| 150 | void PCEBuffer::put(uint64_t ce, int32_t ixLow, int32_t ixHigh, UErrorCode &errorCode) |
| 151 | { |
| 152 | if (U_FAILURE(errorCode)) { |
| 153 | return; |
| 154 | } |
| 155 | if (bufferIndex >= bufferSize) { |
| 156 | PCEI *newBuffer = NEW_ARRAY(PCEI, bufferSize + BUFFER_GROW); |
| 157 | if (newBuffer == NULL) { |
| 158 | errorCode = U_MEMORY_ALLOCATION_ERROR; |
| 159 | return; |
| 160 | } |
| 161 | |
| 162 | ARRAY_COPY(newBuffer, buffer, bufferSize); |
| 163 | |
| 164 | if (buffer != defaultBuffer) { |
| 165 | DELETE_ARRAY(buffer); |
| 166 | } |
| 167 | |
| 168 | buffer = newBuffer; |
| 169 | bufferSize += BUFFER_GROW; |
| 170 | } |
| 171 | |
| 172 | buffer[bufferIndex].ce = ce; |
| 173 | buffer[bufferIndex].low = ixLow; |
| 174 | buffer[bufferIndex].high = ixHigh; |
| 175 | |
| 176 | bufferIndex += 1; |
| 177 | } |
| 178 | |
| 179 | const PCEI *PCEBuffer::get() |
| 180 | { |
| 181 | if (bufferIndex > 0) { |
| 182 | return &buffer[--bufferIndex]; |
| 183 | } |
| 184 | |
| 185 | return NULL; |
| 186 | } |
| 187 | |
| 188 | UCollationPCE::UCollationPCE(UCollationElements *elems) { init(elems); } |
| 189 | |
| 190 | UCollationPCE::UCollationPCE(CollationElementIterator *iter) { init(iter); } |
| 191 | |
| 192 | void UCollationPCE::init(UCollationElements *elems) { |
| 193 | init(CollationElementIterator::fromUCollationElements(elems)); |
| 194 | } |
| 195 | |
| 196 | void UCollationPCE::init(CollationElementIterator *iter) |
| 197 | { |
| 198 | cei = iter; |
| 199 | init(*iter->rbc_); |
| 200 | } |
| 201 | |
| 202 | void UCollationPCE::init(const Collator &coll) |
| 203 | { |
| 204 | UErrorCode status = U_ZERO_ERROR; |
| 205 | |
| 206 | strength = coll.getAttribute(UCOL_STRENGTH, status); |
| 207 | toShift = coll.getAttribute(UCOL_ALTERNATE_HANDLING, status) == UCOL_SHIFTED; |
| 208 | isShifted = FALSE; |
| 209 | variableTop = coll.getVariableTop(status); |
| 210 | } |
| 211 | |
| 212 | UCollationPCE::~UCollationPCE() |
| 213 | { |
| 214 | // nothing to do |
| 215 | } |
| 216 | |
| 217 | uint64_t UCollationPCE::processCE(uint32_t ce) |
| 218 | { |
| 219 | uint64_t primary = 0, secondary = 0, tertiary = 0, quaternary = 0; |
| 220 | |
| 221 | // This is clean, but somewhat slow... |
| 222 | // We could apply the mask to ce and then |
| 223 | // just get all three orders... |
| 224 | switch(strength) { |
| 225 | default: |
| 226 | tertiary = ucol_tertiaryOrder(ce); |
| 227 | U_FALLTHROUGH; |
| 228 | |
| 229 | case UCOL_SECONDARY: |
| 230 | secondary = ucol_secondaryOrder(ce); |
| 231 | U_FALLTHROUGH; |
| 232 | |
| 233 | case UCOL_PRIMARY: |
| 234 | primary = ucol_primaryOrder(ce); |
| 235 | } |
| 236 | |
| 237 | // **** This should probably handle continuations too. **** |
| 238 | // **** That means that we need 24 bits for the primary **** |
| 239 | // **** instead of the 16 that we're currently using. **** |
| 240 | // **** So we can lay out the 64 bits as: 24.12.12.16. **** |
| 241 | // **** Another complication with continuations is that **** |
| 242 | // **** the *second* CE is marked as a continuation, so **** |
| 243 | // **** we always have to peek ahead to know how long **** |
| 244 | // **** the primary is... **** |
| 245 | if ((toShift && variableTop > ce && primary != 0) |
| 246 | || (isShifted && primary == 0)) { |
| 247 | |
| 248 | if (primary == 0) { |
| 249 | return UCOL_IGNORABLE; |
| 250 | } |
| 251 | |
| 252 | if (strength >= UCOL_QUATERNARY) { |
| 253 | quaternary = primary; |
| 254 | } |
| 255 | |
| 256 | primary = secondary = tertiary = 0; |
| 257 | isShifted = TRUE; |
| 258 | } else { |
| 259 | if (strength >= UCOL_QUATERNARY) { |
| 260 | quaternary = 0xFFFF; |
| 261 | } |
| 262 | |
| 263 | isShifted = FALSE; |
| 264 | } |
| 265 | |
| 266 | return primary << 48 | secondary << 32 | tertiary << 16 | quaternary; |
| 267 | } |
| 268 | |
| 269 | U_NAMESPACE_END |
| 270 | |
| 271 | /* public methods ---------------------------------------------------- */ |
| 272 | |
| 273 | U_CAPI UCollationElements* U_EXPORT2 |
| 274 | ucol_openElements(const UCollator *coll, |
| 275 | const UChar *text, |
| 276 | int32_t textLength, |
| 277 | UErrorCode *status) |
| 278 | { |
| 279 | if (U_FAILURE(*status)) { |
| 280 | return NULL; |
| 281 | } |
| 282 | if (coll == NULL || (text == NULL && textLength != 0)) { |
| 283 | *status = U_ILLEGAL_ARGUMENT_ERROR; |
| 284 | return NULL; |
| 285 | } |
| 286 | const RuleBasedCollator *rbc = RuleBasedCollator::rbcFromUCollator(coll); |
| 287 | if (rbc == NULL) { |
| 288 | *status = U_UNSUPPORTED_ERROR; // coll is a Collator but not a RuleBasedCollator |
| 289 | return NULL; |
| 290 | } |
| 291 | |
| 292 | UnicodeString s((UBool)(textLength < 0), text, textLength); |
| 293 | CollationElementIterator *cei = rbc->createCollationElementIterator(s); |
| 294 | if (cei == NULL) { |
| 295 | *status = U_MEMORY_ALLOCATION_ERROR; |
| 296 | return NULL; |
| 297 | } |
| 298 | |
| 299 | return cei->toUCollationElements(); |
| 300 | } |
| 301 | |
| 302 | |
| 303 | U_CAPI void U_EXPORT2 |
| 304 | ucol_closeElements(UCollationElements *elems) |
| 305 | { |
| 306 | delete CollationElementIterator::fromUCollationElements(elems); |
| 307 | } |
| 308 | |
| 309 | U_CAPI void U_EXPORT2 |
| 310 | ucol_reset(UCollationElements *elems) |
| 311 | { |
| 312 | CollationElementIterator::fromUCollationElements(elems)->reset(); |
| 313 | } |
| 314 | |
| 315 | U_CAPI int32_t U_EXPORT2 |
| 316 | ucol_next(UCollationElements *elems, |
| 317 | UErrorCode *status) |
| 318 | { |
| 319 | if (U_FAILURE(*status)) { |
| 320 | return UCOL_NULLORDER; |
| 321 | } |
| 322 | |
| 323 | return CollationElementIterator::fromUCollationElements(elems)->next(*status); |
| 324 | } |
| 325 | |
| 326 | U_NAMESPACE_BEGIN |
| 327 | |
| 328 | int64_t |
| 329 | UCollationPCE::nextProcessed( |
| 330 | int32_t *ixLow, |
| 331 | int32_t *ixHigh, |
| 332 | UErrorCode *status) |
| 333 | { |
| 334 | int64_t result = UCOL_IGNORABLE; |
| 335 | uint32_t low = 0, high = 0; |
| 336 | |
| 337 | if (U_FAILURE(*status)) { |
| 338 | return UCOL_PROCESSED_NULLORDER; |
| 339 | } |
| 340 | |
| 341 | pceBuffer.reset(); |
| 342 | |
| 343 | do { |
| 344 | low = cei->getOffset(); |
| 345 | int32_t ce = cei->next(*status); |
| 346 | high = cei->getOffset(); |
| 347 | |
| 348 | if (ce == UCOL_NULLORDER) { |
| 349 | result = UCOL_PROCESSED_NULLORDER; |
| 350 | break; |
| 351 | } |
| 352 | |
| 353 | result = processCE((uint32_t)ce); |
| 354 | } while (result == UCOL_IGNORABLE); |
| 355 | |
| 356 | if (ixLow != NULL) { |
| 357 | *ixLow = low; |
| 358 | } |
| 359 | |
| 360 | if (ixHigh != NULL) { |
| 361 | *ixHigh = high; |
| 362 | } |
| 363 | |
| 364 | return result; |
| 365 | } |
| 366 | |
| 367 | U_NAMESPACE_END |
| 368 | |
| 369 | U_CAPI int32_t U_EXPORT2 |
| 370 | ucol_previous(UCollationElements *elems, |
| 371 | UErrorCode *status) |
| 372 | { |
| 373 | if(U_FAILURE(*status)) { |
| 374 | return UCOL_NULLORDER; |
| 375 | } |
| 376 | return CollationElementIterator::fromUCollationElements(elems)->previous(*status); |
| 377 | } |
| 378 | |
| 379 | U_NAMESPACE_BEGIN |
| 380 | |
| 381 | int64_t |
| 382 | UCollationPCE::previousProcessed( |
| 383 | int32_t *ixLow, |
| 384 | int32_t *ixHigh, |
| 385 | UErrorCode *status) |
| 386 | { |
| 387 | int64_t result = UCOL_IGNORABLE; |
| 388 | int32_t low = 0, high = 0; |
| 389 | |
| 390 | if (U_FAILURE(*status)) { |
| 391 | return UCOL_PROCESSED_NULLORDER; |
| 392 | } |
| 393 | |
| 394 | // pceBuffer.reset(); |
| 395 | |
| 396 | while (pceBuffer.isEmpty()) { |
| 397 | // buffer raw CEs up to non-ignorable primary |
| 398 | RCEBuffer rceb; |
| 399 | int32_t ce; |
| 400 | |
| 401 | // **** do we need to reset rceb, or will it always be empty at this point **** |
| 402 | do { |
| 403 | high = cei->getOffset(); |
| 404 | ce = cei->previous(*status); |
| 405 | low = cei->getOffset(); |
| 406 | |
| 407 | if (ce == UCOL_NULLORDER) { |
| 408 | if (!rceb.isEmpty()) { |
| 409 | break; |
| 410 | } |
| 411 | |
| 412 | goto finish; |
| 413 | } |
| 414 | |
| 415 | rceb.put((uint32_t)ce, low, high, *status); |
| 416 | } while (U_SUCCESS(*status) && ((ce & UCOL_PRIMARYORDERMASK) == 0 || isContinuation(ce))); |
| 417 | |
| 418 | // process the raw CEs |
| 419 | while (U_SUCCESS(*status) && !rceb.isEmpty()) { |
| 420 | const RCEI *rcei = rceb.get(); |
| 421 | |
| 422 | result = processCE(rcei->ce); |
| 423 | |
| 424 | if (result != UCOL_IGNORABLE) { |
| 425 | pceBuffer.put(result, rcei->low, rcei->high, *status); |
| 426 | } |
| 427 | } |
| 428 | if (U_FAILURE(*status)) { |
| 429 | return UCOL_PROCESSED_NULLORDER; |
| 430 | } |
| 431 | } |
| 432 | |
| 433 | finish: |
| 434 | if (pceBuffer.isEmpty()) { |
| 435 | // **** Is -1 the right value for ixLow, ixHigh? **** |
| 436 | if (ixLow != NULL) { |
| 437 | *ixLow = -1; |
| 438 | } |
| 439 | |
| 440 | if (ixHigh != NULL) { |
| 441 | *ixHigh = -1 |
| 442 | ; |
| 443 | } |
| 444 | return UCOL_PROCESSED_NULLORDER; |
| 445 | } |
| 446 | |
| 447 | const PCEI *pcei = pceBuffer.get(); |
| 448 | |
| 449 | if (ixLow != NULL) { |
| 450 | *ixLow = pcei->low; |
| 451 | } |
| 452 | |
| 453 | if (ixHigh != NULL) { |
| 454 | *ixHigh = pcei->high; |
| 455 | } |
| 456 | |
| 457 | return pcei->ce; |
| 458 | } |
| 459 | |
| 460 | U_NAMESPACE_END |
| 461 | |
| 462 | U_CAPI int32_t U_EXPORT2 |
| 463 | ucol_getMaxExpansion(const UCollationElements *elems, |
| 464 | int32_t order) |
| 465 | { |
| 466 | return CollationElementIterator::fromUCollationElements(elems)->getMaxExpansion(order); |
| 467 | |
| 468 | // TODO: The old code masked the order according to strength and then did a binary search. |
| 469 | // However this was probably at least partially broken because of the following comment. |
| 470 | // Still, it might have found a match when this version may not. |
| 471 | |
| 472 | // FIXME: with a masked search, there might be more than one hit, |
| 473 | // so we need to look forward and backward from the match to find all |
| 474 | // of the hits... |
| 475 | } |
| 476 | |
| 477 | U_CAPI void U_EXPORT2 |
| 478 | ucol_setText( UCollationElements *elems, |
| 479 | const UChar *text, |
| 480 | int32_t textLength, |
| 481 | UErrorCode *status) |
| 482 | { |
| 483 | if (U_FAILURE(*status)) { |
| 484 | return; |
| 485 | } |
| 486 | |
| 487 | if ((text == NULL && textLength != 0)) { |
| 488 | *status = U_ILLEGAL_ARGUMENT_ERROR; |
| 489 | return; |
| 490 | } |
| 491 | UnicodeString s((UBool)(textLength < 0), text, textLength); |
| 492 | return CollationElementIterator::fromUCollationElements(elems)->setText(s, *status); |
| 493 | } |
| 494 | |
| 495 | U_CAPI int32_t U_EXPORT2 |
| 496 | ucol_getOffset(const UCollationElements *elems) |
| 497 | { |
| 498 | return CollationElementIterator::fromUCollationElements(elems)->getOffset(); |
| 499 | } |
| 500 | |
| 501 | U_CAPI void U_EXPORT2 |
| 502 | ucol_setOffset(UCollationElements *elems, |
| 503 | int32_t offset, |
| 504 | UErrorCode *status) |
| 505 | { |
| 506 | if (U_FAILURE(*status)) { |
| 507 | return; |
| 508 | } |
| 509 | |
| 510 | CollationElementIterator::fromUCollationElements(elems)->setOffset(offset, *status); |
| 511 | } |
| 512 | |
| 513 | U_CAPI int32_t U_EXPORT2 |
| 514 | ucol_primaryOrder (int32_t order) |
| 515 | { |
| 516 | return (order >> 16) & 0xffff; |
| 517 | } |
| 518 | |
| 519 | U_CAPI int32_t U_EXPORT2 |
| 520 | ucol_secondaryOrder (int32_t order) |
| 521 | { |
| 522 | return (order >> 8) & 0xff; |
| 523 | } |
| 524 | |
| 525 | U_CAPI int32_t U_EXPORT2 |
| 526 | ucol_tertiaryOrder (int32_t order) |
| 527 | { |
| 528 | return order & 0xff; |
| 529 | } |
| 530 | |
| 531 | #endif /* #if !UCONFIG_NO_COLLATION */ |
| 532 | |