1// © 2016 and later: Unicode, Inc. and others.
2// License & terms of use: http://www.unicode.org/copyright.html
3/*
4*******************************************************************************
5* Copyright (C) 2013-2014, International Business Machines
6* Corporation and others. All Rights Reserved.
7*******************************************************************************
8* collationbuilder.cpp
9*
10* (replaced the former ucol_bld.cpp)
11*
12* created on: 2013may06
13* created by: Markus W. Scherer
14*/
15
16#ifdef DEBUG_COLLATION_BUILDER
17#include <stdio.h>
18#endif
19
20#include "unicode/utypes.h"
21
22#if !UCONFIG_NO_COLLATION
23
24#include "unicode/caniter.h"
25#include "unicode/normalizer2.h"
26#include "unicode/tblcoll.h"
27#include "unicode/parseerr.h"
28#include "unicode/uchar.h"
29#include "unicode/ucol.h"
30#include "unicode/unistr.h"
31#include "unicode/usetiter.h"
32#include "unicode/utf16.h"
33#include "unicode/uversion.h"
34#include "cmemory.h"
35#include "collation.h"
36#include "collationbuilder.h"
37#include "collationdata.h"
38#include "collationdatabuilder.h"
39#include "collationfastlatin.h"
40#include "collationroot.h"
41#include "collationrootelements.h"
42#include "collationruleparser.h"
43#include "collationsettings.h"
44#include "collationtailoring.h"
45#include "collationweights.h"
46#include "normalizer2impl.h"
47#include "uassert.h"
48#include "ucol_imp.h"
49#include "utf16collationiterator.h"
50
51U_NAMESPACE_BEGIN
52
53namespace {
54
55class BundleImporter : public CollationRuleParser::Importer {
56public:
57 BundleImporter() {}
58 virtual ~BundleImporter();
59 virtual void getRules(
60 const char *localeID, const char *collationType,
61 UnicodeString &rules,
62 const char *&errorReason, UErrorCode &errorCode);
63};
64
65BundleImporter::~BundleImporter() {}
66
67void
68BundleImporter::getRules(
69 const char *localeID, const char *collationType,
70 UnicodeString &rules,
71 const char *& /*errorReason*/, UErrorCode &errorCode) {
72 CollationLoader::loadRules(localeID, collationType, rules, errorCode);
73}
74
75} // namespace
76
77// RuleBasedCollator implementation ---------------------------------------- ***
78
79// These methods are here, rather than in rulebasedcollator.cpp,
80// for modularization:
81// Most code using Collator does not need to build a Collator from rules.
82// By moving these constructors and helper methods to a separate file,
83// most code will not have a static dependency on the builder code.
84
85RuleBasedCollator::RuleBasedCollator()
86 : data(NULL),
87 settings(NULL),
88 tailoring(NULL),
89 cacheEntry(NULL),
90 validLocale(""),
91 explicitlySetAttributes(0),
92 actualLocaleIsSameAsValid(FALSE) {
93}
94
95RuleBasedCollator::RuleBasedCollator(const UnicodeString &rules, UErrorCode &errorCode)
96 : data(NULL),
97 settings(NULL),
98 tailoring(NULL),
99 cacheEntry(NULL),
100 validLocale(""),
101 explicitlySetAttributes(0),
102 actualLocaleIsSameAsValid(FALSE) {
103 internalBuildTailoring(rules, UCOL_DEFAULT, UCOL_DEFAULT, NULL, NULL, errorCode);
104}
105
106RuleBasedCollator::RuleBasedCollator(const UnicodeString &rules, ECollationStrength strength,
107 UErrorCode &errorCode)
108 : data(NULL),
109 settings(NULL),
110 tailoring(NULL),
111 cacheEntry(NULL),
112 validLocale(""),
113 explicitlySetAttributes(0),
114 actualLocaleIsSameAsValid(FALSE) {
115 internalBuildTailoring(rules, strength, UCOL_DEFAULT, NULL, NULL, errorCode);
116}
117
118RuleBasedCollator::RuleBasedCollator(const UnicodeString &rules,
119 UColAttributeValue decompositionMode,
120 UErrorCode &errorCode)
121 : data(NULL),
122 settings(NULL),
123 tailoring(NULL),
124 cacheEntry(NULL),
125 validLocale(""),
126 explicitlySetAttributes(0),
127 actualLocaleIsSameAsValid(FALSE) {
128 internalBuildTailoring(rules, UCOL_DEFAULT, decompositionMode, NULL, NULL, errorCode);
129}
130
131RuleBasedCollator::RuleBasedCollator(const UnicodeString &rules,
132 ECollationStrength strength,
133 UColAttributeValue decompositionMode,
134 UErrorCode &errorCode)
135 : data(NULL),
136 settings(NULL),
137 tailoring(NULL),
138 cacheEntry(NULL),
139 validLocale(""),
140 explicitlySetAttributes(0),
141 actualLocaleIsSameAsValid(FALSE) {
142 internalBuildTailoring(rules, strength, decompositionMode, NULL, NULL, errorCode);
143}
144
145RuleBasedCollator::RuleBasedCollator(const UnicodeString &rules,
146 UParseError &parseError, UnicodeString &reason,
147 UErrorCode &errorCode)
148 : data(NULL),
149 settings(NULL),
150 tailoring(NULL),
151 cacheEntry(NULL),
152 validLocale(""),
153 explicitlySetAttributes(0),
154 actualLocaleIsSameAsValid(FALSE) {
155 internalBuildTailoring(rules, UCOL_DEFAULT, UCOL_DEFAULT, &parseError, &reason, errorCode);
156}
157
158void
159RuleBasedCollator::internalBuildTailoring(const UnicodeString &rules,
160 int32_t strength,
161 UColAttributeValue decompositionMode,
162 UParseError *outParseError, UnicodeString *outReason,
163 UErrorCode &errorCode) {
164 const CollationTailoring *base = CollationRoot::getRoot(errorCode);
165 if(U_FAILURE(errorCode)) { return; }
166 if(outReason != NULL) { outReason->remove(); }
167 CollationBuilder builder(base, errorCode);
168 UVersionInfo noVersion = { 0, 0, 0, 0 };
169 BundleImporter importer;
170 LocalPointer<CollationTailoring> t(builder.parseAndBuild(rules, noVersion,
171 &importer,
172 outParseError, errorCode));
173 if(U_FAILURE(errorCode)) {
174 const char *reason = builder.getErrorReason();
175 if(reason != NULL && outReason != NULL) {
176 *outReason = UnicodeString(reason, -1, US_INV);
177 }
178 return;
179 }
180 t->actualLocale.setToBogus();
181 adoptTailoring(t.orphan(), errorCode);
182 // Set attributes after building the collator,
183 // to keep the default settings consistent with the rule string.
184 if(strength != UCOL_DEFAULT) {
185 setAttribute(UCOL_STRENGTH, (UColAttributeValue)strength, errorCode);
186 }
187 if(decompositionMode != UCOL_DEFAULT) {
188 setAttribute(UCOL_NORMALIZATION_MODE, decompositionMode, errorCode);
189 }
190}
191
192// CollationBuilder implementation ----------------------------------------- ***
193
194// Some compilers don't care if constants are defined in the .cpp file.
195// MS Visual C++ does not like it, but gcc requires it. clang does not care.
196#ifndef _MSC_VER
197const int32_t CollationBuilder::HAS_BEFORE2;
198const int32_t CollationBuilder::HAS_BEFORE3;
199#endif
200
201CollationBuilder::CollationBuilder(const CollationTailoring *b, UErrorCode &errorCode)
202 : nfd(*Normalizer2::getNFDInstance(errorCode)),
203 fcd(*Normalizer2Factory::getFCDInstance(errorCode)),
204 nfcImpl(*Normalizer2Factory::getNFCImpl(errorCode)),
205 base(b),
206 baseData(b->data),
207 rootElements(b->data->rootElements, b->data->rootElementsLength),
208 variableTop(0),
209 dataBuilder(new CollationDataBuilder(errorCode)), fastLatinEnabled(TRUE),
210 errorReason(NULL),
211 cesLength(0),
212 rootPrimaryIndexes(errorCode), nodes(errorCode) {
213 nfcImpl.ensureCanonIterData(errorCode);
214 if(U_FAILURE(errorCode)) {
215 errorReason = "CollationBuilder fields initialization failed";
216 return;
217 }
218 if(dataBuilder == NULL) {
219 errorCode = U_MEMORY_ALLOCATION_ERROR;
220 return;
221 }
222 dataBuilder->initForTailoring(baseData, errorCode);
223 if(U_FAILURE(errorCode)) {
224 errorReason = "CollationBuilder initialization failed";
225 }
226}
227
228CollationBuilder::~CollationBuilder() {
229 delete dataBuilder;
230}
231
232CollationTailoring *
233CollationBuilder::parseAndBuild(const UnicodeString &ruleString,
234 const UVersionInfo rulesVersion,
235 CollationRuleParser::Importer *importer,
236 UParseError *outParseError,
237 UErrorCode &errorCode) {
238 if(U_FAILURE(errorCode)) { return NULL; }
239 if(baseData->rootElements == NULL) {
240 errorCode = U_MISSING_RESOURCE_ERROR;
241 errorReason = "missing root elements data, tailoring not supported";
242 return NULL;
243 }
244 LocalPointer<CollationTailoring> tailoring(new CollationTailoring(base->settings));
245 if(tailoring.isNull() || tailoring->isBogus()) {
246 errorCode = U_MEMORY_ALLOCATION_ERROR;
247 return NULL;
248 }
249 CollationRuleParser parser(baseData, errorCode);
250 if(U_FAILURE(errorCode)) { return NULL; }
251 // Note: This always bases &[last variable] and &[first regular]
252 // on the root collator's maxVariable/variableTop.
253 // If we wanted this to change after [maxVariable x], then we would keep
254 // the tailoring.settings pointer here and read its variableTop when we need it.
255 // See http://unicode.org/cldr/trac/ticket/6070
256 variableTop = base->settings->variableTop;
257 parser.setSink(this);
258 parser.setImporter(importer);
259 CollationSettings &ownedSettings = *SharedObject::copyOnWrite(tailoring->settings);
260 parser.parse(ruleString, ownedSettings, outParseError, errorCode);
261 errorReason = parser.getErrorReason();
262 if(U_FAILURE(errorCode)) { return NULL; }
263 if(dataBuilder->hasMappings()) {
264 makeTailoredCEs(errorCode);
265 closeOverComposites(errorCode);
266 finalizeCEs(errorCode);
267 // Copy all of ASCII, and Latin-1 letters, into each tailoring.
268 optimizeSet.add(0, 0x7f);
269 optimizeSet.add(0xc0, 0xff);
270 // Hangul is decomposed on the fly during collation,
271 // and the tailoring data is always built with HANGUL_TAG specials.
272 optimizeSet.remove(Hangul::HANGUL_BASE, Hangul::HANGUL_END);
273 dataBuilder->optimize(optimizeSet, errorCode);
274 tailoring->ensureOwnedData(errorCode);
275 if(U_FAILURE(errorCode)) { return NULL; }
276 if(fastLatinEnabled) { dataBuilder->enableFastLatin(); }
277 dataBuilder->build(*tailoring->ownedData, errorCode);
278 tailoring->builder = dataBuilder;
279 dataBuilder = NULL;
280 } else {
281 tailoring->data = baseData;
282 }
283 if(U_FAILURE(errorCode)) { return NULL; }
284 ownedSettings.fastLatinOptions = CollationFastLatin::getOptions(
285 tailoring->data, ownedSettings,
286 ownedSettings.fastLatinPrimaries, UPRV_LENGTHOF(ownedSettings.fastLatinPrimaries));
287 tailoring->rules = ruleString;
288 tailoring->rules.getTerminatedBuffer(); // ensure NUL-termination
289 tailoring->setVersion(base->version, rulesVersion);
290 return tailoring.orphan();
291}
292
293void
294CollationBuilder::addReset(int32_t strength, const UnicodeString &str,
295 const char *&parserErrorReason, UErrorCode &errorCode) {
296 if(U_FAILURE(errorCode)) { return; }
297 U_ASSERT(!str.isEmpty());
298 if(str.charAt(0) == CollationRuleParser::POS_LEAD) {
299 ces[0] = getSpecialResetPosition(str, parserErrorReason, errorCode);
300 cesLength = 1;
301 if(U_FAILURE(errorCode)) { return; }
302 U_ASSERT((ces[0] & Collation::CASE_AND_QUATERNARY_MASK) == 0);
303 } else {
304 // normal reset to a character or string
305 UnicodeString nfdString = nfd.normalize(str, errorCode);
306 if(U_FAILURE(errorCode)) {
307 parserErrorReason = "normalizing the reset position";
308 return;
309 }
310 cesLength = dataBuilder->getCEs(nfdString, ces, 0);
311 if(cesLength > Collation::MAX_EXPANSION_LENGTH) {
312 errorCode = U_ILLEGAL_ARGUMENT_ERROR;
313 parserErrorReason = "reset position maps to too many collation elements (more than 31)";
314 return;
315 }
316 }
317 if(strength == UCOL_IDENTICAL) { return; } // simple reset-at-position
318
319 // &[before strength]position
320 U_ASSERT(UCOL_PRIMARY <= strength && strength <= UCOL_TERTIARY);
321 int32_t index = findOrInsertNodeForCEs(strength, parserErrorReason, errorCode);
322 if(U_FAILURE(errorCode)) { return; }
323
324 int64_t node = nodes.elementAti(index);
325 // If the index is for a "weaker" node,
326 // then skip backwards over this and further "weaker" nodes.
327 while(strengthFromNode(node) > strength) {
328 index = previousIndexFromNode(node);
329 node = nodes.elementAti(index);
330 }
331
332 // Find or insert a node whose index we will put into a temporary CE.
333 if(strengthFromNode(node) == strength && isTailoredNode(node)) {
334 // Reset to just before this same-strength tailored node.
335 index = previousIndexFromNode(node);
336 } else if(strength == UCOL_PRIMARY) {
337 // root primary node (has no previous index)
338 uint32_t p = weight32FromNode(node);
339 if(p == 0) {
340 errorCode = U_UNSUPPORTED_ERROR;
341 parserErrorReason = "reset primary-before ignorable not possible";
342 return;
343 }
344 if(p <= rootElements.getFirstPrimary()) {
345 // There is no primary gap between ignorables and the space-first-primary.
346 errorCode = U_UNSUPPORTED_ERROR;
347 parserErrorReason = "reset primary-before first non-ignorable not supported";
348 return;
349 }
350 if(p == Collation::FIRST_TRAILING_PRIMARY) {
351 // We do not support tailoring to an unassigned-implicit CE.
352 errorCode = U_UNSUPPORTED_ERROR;
353 parserErrorReason = "reset primary-before [first trailing] not supported";
354 return;
355 }
356 p = rootElements.getPrimaryBefore(p, baseData->isCompressiblePrimary(p));
357 index = findOrInsertNodeForPrimary(p, errorCode);
358 // Go to the last node in this list:
359 // Tailor after the last node between adjacent root nodes.
360 for(;;) {
361 node = nodes.elementAti(index);
362 int32_t nextIndex = nextIndexFromNode(node);
363 if(nextIndex == 0) { break; }
364 index = nextIndex;
365 }
366 } else {
367 // &[before 2] or &[before 3]
368 index = findCommonNode(index, UCOL_SECONDARY);
369 if(strength >= UCOL_TERTIARY) {
370 index = findCommonNode(index, UCOL_TERTIARY);
371 }
372 // findCommonNode() stayed on the stronger node or moved to
373 // an explicit common-weight node of the reset-before strength.
374 node = nodes.elementAti(index);
375 if(strengthFromNode(node) == strength) {
376 // Found a same-strength node with an explicit weight.
377 uint32_t weight16 = weight16FromNode(node);
378 if(weight16 == 0) {
379 errorCode = U_UNSUPPORTED_ERROR;
380 if(strength == UCOL_SECONDARY) {
381 parserErrorReason = "reset secondary-before secondary ignorable not possible";
382 } else {
383 parserErrorReason = "reset tertiary-before completely ignorable not possible";
384 }
385 return;
386 }
387 U_ASSERT(weight16 > Collation::BEFORE_WEIGHT16);
388 // Reset to just before this node.
389 // Insert the preceding same-level explicit weight if it is not there already.
390 // Which explicit weight immediately precedes this one?
391 weight16 = getWeight16Before(index, node, strength);
392 // Does this preceding weight have a node?
393 uint32_t previousWeight16;
394 int32_t previousIndex = previousIndexFromNode(node);
395 for(int32_t i = previousIndex;; i = previousIndexFromNode(node)) {
396 node = nodes.elementAti(i);
397 int32_t previousStrength = strengthFromNode(node);
398 if(previousStrength < strength) {
399 U_ASSERT(weight16 >= Collation::COMMON_WEIGHT16 || i == previousIndex);
400 // Either the reset element has an above-common weight and
401 // the parent node provides the implied common weight,
402 // or the reset element has a weight<=common in the node
403 // right after the parent, and we need to insert the preceding weight.
404 previousWeight16 = Collation::COMMON_WEIGHT16;
405 break;
406 } else if(previousStrength == strength && !isTailoredNode(node)) {
407 previousWeight16 = weight16FromNode(node);
408 break;
409 }
410 // Skip weaker nodes and same-level tailored nodes.
411 }
412 if(previousWeight16 == weight16) {
413 // The preceding weight has a node,
414 // maybe with following weaker or tailored nodes.
415 // Reset to the last of them.
416 index = previousIndex;
417 } else {
418 // Insert a node with the preceding weight, reset to that.
419 node = nodeFromWeight16(weight16) | nodeFromStrength(strength);
420 index = insertNodeBetween(previousIndex, index, node, errorCode);
421 }
422 } else {
423 // Found a stronger node with implied strength-common weight.
424 uint32_t weight16 = getWeight16Before(index, node, strength);
425 index = findOrInsertWeakNode(index, weight16, strength, errorCode);
426 }
427 // Strength of the temporary CE = strength of its reset position.
428 // Code above raises an error if the before-strength is stronger.
429 strength = ceStrength(ces[cesLength - 1]);
430 }
431 if(U_FAILURE(errorCode)) {
432 parserErrorReason = "inserting reset position for &[before n]";
433 return;
434 }
435 ces[cesLength - 1] = tempCEFromIndexAndStrength(index, strength);
436}
437
438uint32_t
439CollationBuilder::getWeight16Before(int32_t index, int64_t node, int32_t level) {
440 U_ASSERT(strengthFromNode(node) < level || !isTailoredNode(node));
441 // Collect the root CE weights if this node is for a root CE.
442 // If it is not, then return the low non-primary boundary for a tailored CE.
443 uint32_t t;
444 if(strengthFromNode(node) == UCOL_TERTIARY) {
445 t = weight16FromNode(node);
446 } else {
447 t = Collation::COMMON_WEIGHT16; // Stronger node with implied common weight.
448 }
449 while(strengthFromNode(node) > UCOL_SECONDARY) {
450 index = previousIndexFromNode(node);
451 node = nodes.elementAti(index);
452 }
453 if(isTailoredNode(node)) {
454 return Collation::BEFORE_WEIGHT16;
455 }
456 uint32_t s;
457 if(strengthFromNode(node) == UCOL_SECONDARY) {
458 s = weight16FromNode(node);
459 } else {
460 s = Collation::COMMON_WEIGHT16; // Stronger node with implied common weight.
461 }
462 while(strengthFromNode(node) > UCOL_PRIMARY) {
463 index = previousIndexFromNode(node);
464 node = nodes.elementAti(index);
465 }
466 if(isTailoredNode(node)) {
467 return Collation::BEFORE_WEIGHT16;
468 }
469 // [p, s, t] is a root CE. Return the preceding weight for the requested level.
470 uint32_t p = weight32FromNode(node);
471 uint32_t weight16;
472 if(level == UCOL_SECONDARY) {
473 weight16 = rootElements.getSecondaryBefore(p, s);
474 } else {
475 weight16 = rootElements.getTertiaryBefore(p, s, t);
476 U_ASSERT((weight16 & ~Collation::ONLY_TERTIARY_MASK) == 0);
477 }
478 return weight16;
479}
480
481int64_t
482CollationBuilder::getSpecialResetPosition(const UnicodeString &str,
483 const char *&parserErrorReason, UErrorCode &errorCode) {
484 U_ASSERT(str.length() == 2);
485 int64_t ce;
486 int32_t strength = UCOL_PRIMARY;
487 UBool isBoundary = FALSE;
488 UChar32 pos = str.charAt(1) - CollationRuleParser::POS_BASE;
489 U_ASSERT(0 <= pos && pos <= CollationRuleParser::LAST_TRAILING);
490 switch(pos) {
491 case CollationRuleParser::FIRST_TERTIARY_IGNORABLE:
492 // Quaternary CEs are not supported.
493 // Non-zero quaternary weights are possible only on tertiary or stronger CEs.
494 return 0;
495 case CollationRuleParser::LAST_TERTIARY_IGNORABLE:
496 return 0;
497 case CollationRuleParser::FIRST_SECONDARY_IGNORABLE: {
498 // Look for a tailored tertiary node after [0, 0, 0].
499 int32_t index = findOrInsertNodeForRootCE(0, UCOL_TERTIARY, errorCode);
500 if(U_FAILURE(errorCode)) { return 0; }
501 int64_t node = nodes.elementAti(index);
502 if((index = nextIndexFromNode(node)) != 0) {
503 node = nodes.elementAti(index);
504 U_ASSERT(strengthFromNode(node) <= UCOL_TERTIARY);
505 if(isTailoredNode(node) && strengthFromNode(node) == UCOL_TERTIARY) {
506 return tempCEFromIndexAndStrength(index, UCOL_TERTIARY);
507 }
508 }
509 return rootElements.getFirstTertiaryCE();
510 // No need to look for nodeHasAnyBefore() on a tertiary node.
511 }
512 case CollationRuleParser::LAST_SECONDARY_IGNORABLE:
513 ce = rootElements.getLastTertiaryCE();
514 strength = UCOL_TERTIARY;
515 break;
516 case CollationRuleParser::FIRST_PRIMARY_IGNORABLE: {
517 // Look for a tailored secondary node after [0, 0, *].
518 int32_t index = findOrInsertNodeForRootCE(0, UCOL_SECONDARY, errorCode);
519 if(U_FAILURE(errorCode)) { return 0; }
520 int64_t node = nodes.elementAti(index);
521 while((index = nextIndexFromNode(node)) != 0) {
522 node = nodes.elementAti(index);
523 strength = strengthFromNode(node);
524 if(strength < UCOL_SECONDARY) { break; }
525 if(strength == UCOL_SECONDARY) {
526 if(isTailoredNode(node)) {
527 if(nodeHasBefore3(node)) {
528 index = nextIndexFromNode(nodes.elementAti(nextIndexFromNode(node)));
529 U_ASSERT(isTailoredNode(nodes.elementAti(index)));
530 }
531 return tempCEFromIndexAndStrength(index, UCOL_SECONDARY);
532 } else {
533 break;
534 }
535 }
536 }
537 ce = rootElements.getFirstSecondaryCE();
538 strength = UCOL_SECONDARY;
539 break;
540 }
541 case CollationRuleParser::LAST_PRIMARY_IGNORABLE:
542 ce = rootElements.getLastSecondaryCE();
543 strength = UCOL_SECONDARY;
544 break;
545 case CollationRuleParser::FIRST_VARIABLE:
546 ce = rootElements.getFirstPrimaryCE();
547 isBoundary = TRUE; // FractionalUCA.txt: FDD1 00A0, SPACE first primary
548 break;
549 case CollationRuleParser::LAST_VARIABLE:
550 ce = rootElements.lastCEWithPrimaryBefore(variableTop + 1);
551 break;
552 case CollationRuleParser::FIRST_REGULAR:
553 ce = rootElements.firstCEWithPrimaryAtLeast(variableTop + 1);
554 isBoundary = TRUE; // FractionalUCA.txt: FDD1 263A, SYMBOL first primary
555 break;
556 case CollationRuleParser::LAST_REGULAR:
557 // Use the Hani-first-primary rather than the actual last "regular" CE before it,
558 // for backward compatibility with behavior before the introduction of
559 // script-first-primary CEs in the root collator.
560 ce = rootElements.firstCEWithPrimaryAtLeast(
561 baseData->getFirstPrimaryForGroup(USCRIPT_HAN));
562 break;
563 case CollationRuleParser::FIRST_IMPLICIT:
564 ce = baseData->getSingleCE(0x4e00, errorCode);
565 break;
566 case CollationRuleParser::LAST_IMPLICIT:
567 // We do not support tailoring to an unassigned-implicit CE.
568 errorCode = U_UNSUPPORTED_ERROR;
569 parserErrorReason = "reset to [last implicit] not supported";
570 return 0;
571 case CollationRuleParser::FIRST_TRAILING:
572 ce = Collation::makeCE(Collation::FIRST_TRAILING_PRIMARY);
573 isBoundary = TRUE; // trailing first primary (there is no mapping for it)
574 break;
575 case CollationRuleParser::LAST_TRAILING:
576 errorCode = U_ILLEGAL_ARGUMENT_ERROR;
577 parserErrorReason = "LDML forbids tailoring to U+FFFF";
578 return 0;
579 default:
580 UPRV_UNREACHABLE;
581 }
582
583 int32_t index = findOrInsertNodeForRootCE(ce, strength, errorCode);
584 if(U_FAILURE(errorCode)) { return 0; }
585 int64_t node = nodes.elementAti(index);
586 if((pos & 1) == 0) {
587 // even pos = [first xyz]
588 if(!nodeHasAnyBefore(node) && isBoundary) {
589 // A <group> first primary boundary is artificially added to FractionalUCA.txt.
590 // It is reachable via its special contraction, but is not normally used.
591 // Find the first character tailored after the boundary CE,
592 // or the first real root CE after it.
593 if((index = nextIndexFromNode(node)) != 0) {
594 // If there is a following node, then it must be tailored
595 // because there are no root CEs with a boundary primary
596 // and non-common secondary/tertiary weights.
597 node = nodes.elementAti(index);
598 U_ASSERT(isTailoredNode(node));
599 ce = tempCEFromIndexAndStrength(index, strength);
600 } else {
601 U_ASSERT(strength == UCOL_PRIMARY);
602 uint32_t p = (uint32_t)(ce >> 32);
603 int32_t pIndex = rootElements.findPrimary(p);
604 UBool isCompressible = baseData->isCompressiblePrimary(p);
605 p = rootElements.getPrimaryAfter(p, pIndex, isCompressible);
606 ce = Collation::makeCE(p);
607 index = findOrInsertNodeForRootCE(ce, UCOL_PRIMARY, errorCode);
608 if(U_FAILURE(errorCode)) { return 0; }
609 node = nodes.elementAti(index);
610 }
611 }
612 if(nodeHasAnyBefore(node)) {
613 // Get the first node that was tailored before this one at a weaker strength.
614 if(nodeHasBefore2(node)) {
615 index = nextIndexFromNode(nodes.elementAti(nextIndexFromNode(node)));
616 node = nodes.elementAti(index);
617 }
618 if(nodeHasBefore3(node)) {
619 index = nextIndexFromNode(nodes.elementAti(nextIndexFromNode(node)));
620 }
621 U_ASSERT(isTailoredNode(nodes.elementAti(index)));
622 ce = tempCEFromIndexAndStrength(index, strength);
623 }
624 } else {
625 // odd pos = [last xyz]
626 // Find the last node that was tailored after the [last xyz]
627 // at a strength no greater than the position's strength.
628 for(;;) {
629 int32_t nextIndex = nextIndexFromNode(node);
630 if(nextIndex == 0) { break; }
631 int64_t nextNode = nodes.elementAti(nextIndex);
632 if(strengthFromNode(nextNode) < strength) { break; }
633 index = nextIndex;
634 node = nextNode;
635 }
636 // Do not make a temporary CE for a root node.
637 // This last node might be the node for the root CE itself,
638 // or a node with a common secondary or tertiary weight.
639 if(isTailoredNode(node)) {
640 ce = tempCEFromIndexAndStrength(index, strength);
641 }
642 }
643 return ce;
644}
645
646void
647CollationBuilder::addRelation(int32_t strength, const UnicodeString &prefix,
648 const UnicodeString &str, const UnicodeString &extension,
649 const char *&parserErrorReason, UErrorCode &errorCode) {
650 if(U_FAILURE(errorCode)) { return; }
651 UnicodeString nfdPrefix;
652 if(!prefix.isEmpty()) {
653 nfd.normalize(prefix, nfdPrefix, errorCode);
654 if(U_FAILURE(errorCode)) {
655 parserErrorReason = "normalizing the relation prefix";
656 return;
657 }
658 }
659 UnicodeString nfdString = nfd.normalize(str, errorCode);
660 if(U_FAILURE(errorCode)) {
661 parserErrorReason = "normalizing the relation string";
662 return;
663 }
664
665 // The runtime code decomposes Hangul syllables on the fly,
666 // with recursive processing but without making the Jamo pieces visible for matching.
667 // It does not work with certain types of contextual mappings.
668 int32_t nfdLength = nfdString.length();
669 if(nfdLength >= 2) {
670 UChar c = nfdString.charAt(0);
671 if(Hangul::isJamoL(c) || Hangul::isJamoV(c)) {
672 // While handling a Hangul syllable, contractions starting with Jamo L or V
673 // would not see the following Jamo of that syllable.
674 errorCode = U_UNSUPPORTED_ERROR;
675 parserErrorReason = "contractions starting with conjoining Jamo L or V not supported";
676 return;
677 }
678 c = nfdString.charAt(nfdLength - 1);
679 if(Hangul::isJamoL(c) ||
680 (Hangul::isJamoV(c) && Hangul::isJamoL(nfdString.charAt(nfdLength - 2)))) {
681 // A contraction ending with Jamo L or L+V would require
682 // generating Hangul syllables in addTailComposites() (588 for a Jamo L),
683 // or decomposing a following Hangul syllable on the fly, during contraction matching.
684 errorCode = U_UNSUPPORTED_ERROR;
685 parserErrorReason = "contractions ending with conjoining Jamo L or L+V not supported";
686 return;
687 }
688 // A Hangul syllable completely inside a contraction is ok.
689 }
690 // Note: If there is a prefix, then the parser checked that
691 // both the prefix and the string beging with NFC boundaries (not Jamo V or T).
692 // Therefore: prefix.isEmpty() || !isJamoVOrT(nfdString.charAt(0))
693 // (While handling a Hangul syllable, prefixes on Jamo V or T
694 // would not see the previous Jamo of that syllable.)
695
696 if(strength != UCOL_IDENTICAL) {
697 // Find the node index after which we insert the new tailored node.
698 int32_t index = findOrInsertNodeForCEs(strength, parserErrorReason, errorCode);
699 U_ASSERT(cesLength > 0);
700 int64_t ce = ces[cesLength - 1];
701 if(strength == UCOL_PRIMARY && !isTempCE(ce) && (uint32_t)(ce >> 32) == 0) {
702 // There is no primary gap between ignorables and the space-first-primary.
703 errorCode = U_UNSUPPORTED_ERROR;
704 parserErrorReason = "tailoring primary after ignorables not supported";
705 return;
706 }
707 if(strength == UCOL_QUATERNARY && ce == 0) {
708 // The CE data structure does not support non-zero quaternary weights
709 // on tertiary ignorables.
710 errorCode = U_UNSUPPORTED_ERROR;
711 parserErrorReason = "tailoring quaternary after tertiary ignorables not supported";
712 return;
713 }
714 // Insert the new tailored node.
715 index = insertTailoredNodeAfter(index, strength, errorCode);
716 if(U_FAILURE(errorCode)) {
717 parserErrorReason = "modifying collation elements";
718 return;
719 }
720 // Strength of the temporary CE:
721 // The new relation may yield a stronger CE but not a weaker one.
722 int32_t tempStrength = ceStrength(ce);
723 if(strength < tempStrength) { tempStrength = strength; }
724 ces[cesLength - 1] = tempCEFromIndexAndStrength(index, tempStrength);
725 }
726
727 setCaseBits(nfdString, parserErrorReason, errorCode);
728 if(U_FAILURE(errorCode)) { return; }
729
730 int32_t cesLengthBeforeExtension = cesLength;
731 if(!extension.isEmpty()) {
732 UnicodeString nfdExtension = nfd.normalize(extension, errorCode);
733 if(U_FAILURE(errorCode)) {
734 parserErrorReason = "normalizing the relation extension";
735 return;
736 }
737 cesLength = dataBuilder->getCEs(nfdExtension, ces, cesLength);
738 if(cesLength > Collation::MAX_EXPANSION_LENGTH) {
739 errorCode = U_ILLEGAL_ARGUMENT_ERROR;
740 parserErrorReason =
741 "extension string adds too many collation elements (more than 31 total)";
742 return;
743 }
744 }
745 uint32_t ce32 = Collation::UNASSIGNED_CE32;
746 if((prefix != nfdPrefix || str != nfdString) &&
747 !ignorePrefix(prefix, errorCode) && !ignoreString(str, errorCode)) {
748 // Map from the original input to the CEs.
749 // We do this in case the canonical closure is incomplete,
750 // so that it is possible to explicitly provide the missing mappings.
751 ce32 = addIfDifferent(prefix, str, ces, cesLength, ce32, errorCode);
752 }
753 addWithClosure(nfdPrefix, nfdString, ces, cesLength, ce32, errorCode);
754 if(U_FAILURE(errorCode)) {
755 parserErrorReason = "writing collation elements";
756 return;
757 }
758 cesLength = cesLengthBeforeExtension;
759}
760
761int32_t
762CollationBuilder::findOrInsertNodeForCEs(int32_t strength, const char *&parserErrorReason,
763 UErrorCode &errorCode) {
764 if(U_FAILURE(errorCode)) { return 0; }
765 U_ASSERT(UCOL_PRIMARY <= strength && strength <= UCOL_QUATERNARY);
766
767 // Find the last CE that is at least as "strong" as the requested difference.
768 // Note: Stronger is smaller (UCOL_PRIMARY=0).
769 int64_t ce;
770 for(;; --cesLength) {
771 if(cesLength == 0) {
772 ce = ces[0] = 0;
773 cesLength = 1;
774 break;
775 } else {
776 ce = ces[cesLength - 1];
777 }
778 if(ceStrength(ce) <= strength) { break; }
779 }
780
781 if(isTempCE(ce)) {
782 // No need to findCommonNode() here for lower levels
783 // because insertTailoredNodeAfter() will do that anyway.
784 return indexFromTempCE(ce);
785 }
786
787 // root CE
788 if((uint8_t)(ce >> 56) == Collation::UNASSIGNED_IMPLICIT_BYTE) {
789 errorCode = U_UNSUPPORTED_ERROR;
790 parserErrorReason = "tailoring relative to an unassigned code point not supported";
791 return 0;
792 }
793 return findOrInsertNodeForRootCE(ce, strength, errorCode);
794}
795
796int32_t
797CollationBuilder::findOrInsertNodeForRootCE(int64_t ce, int32_t strength, UErrorCode &errorCode) {
798 if(U_FAILURE(errorCode)) { return 0; }
799 U_ASSERT((uint8_t)(ce >> 56) != Collation::UNASSIGNED_IMPLICIT_BYTE);
800
801 // Find or insert the node for each of the root CE's weights,
802 // down to the requested level/strength.
803 // Root CEs must have common=zero quaternary weights (for which we never insert any nodes).
804 U_ASSERT((ce & 0xc0) == 0);
805 int32_t index = findOrInsertNodeForPrimary((uint32_t)(ce >> 32), errorCode);
806 if(strength >= UCOL_SECONDARY) {
807 uint32_t lower32 = (uint32_t)ce;
808 index = findOrInsertWeakNode(index, lower32 >> 16, UCOL_SECONDARY, errorCode);
809 if(strength >= UCOL_TERTIARY) {
810 index = findOrInsertWeakNode(index, lower32 & Collation::ONLY_TERTIARY_MASK,
811 UCOL_TERTIARY, errorCode);
812 }
813 }
814 return index;
815}
816
817namespace {
818
819/**
820 * Like Java Collections.binarySearch(List, key, Comparator).
821 *
822 * @return the index>=0 where the item was found,
823 * or the index<0 for inserting the string at ~index in sorted order
824 * (index into rootPrimaryIndexes)
825 */
826int32_t
827binarySearchForRootPrimaryNode(const int32_t *rootPrimaryIndexes, int32_t length,
828 const int64_t *nodes, uint32_t p) {
829 if(length == 0) { return ~0; }
830 int32_t start = 0;
831 int32_t limit = length;
832 for (;;) {
833 int32_t i = (start + limit) / 2;
834 int64_t node = nodes[rootPrimaryIndexes[i]];
835 uint32_t nodePrimary = (uint32_t)(node >> 32); // weight32FromNode(node)
836 if (p == nodePrimary) {
837 return i;
838 } else if (p < nodePrimary) {
839 if (i == start) {
840 return ~start; // insert s before i
841 }
842 limit = i;
843 } else {
844 if (i == start) {
845 return ~(start + 1); // insert s after i
846 }
847 start = i;
848 }
849 }
850}
851
852} // namespace
853
854int32_t
855CollationBuilder::findOrInsertNodeForPrimary(uint32_t p, UErrorCode &errorCode) {
856 if(U_FAILURE(errorCode)) { return 0; }
857
858 int32_t rootIndex = binarySearchForRootPrimaryNode(
859 rootPrimaryIndexes.getBuffer(), rootPrimaryIndexes.size(), nodes.getBuffer(), p);
860 if(rootIndex >= 0) {
861 return rootPrimaryIndexes.elementAti(rootIndex);
862 } else {
863 // Start a new list of nodes with this primary.
864 int32_t index = nodes.size();
865 nodes.addElement(nodeFromWeight32(p), errorCode);
866 rootPrimaryIndexes.insertElementAt(index, ~rootIndex, errorCode);
867 return index;
868 }
869}
870
871int32_t
872CollationBuilder::findOrInsertWeakNode(int32_t index, uint32_t weight16, int32_t level, UErrorCode &errorCode) {
873 if(U_FAILURE(errorCode)) { return 0; }
874 U_ASSERT(0 <= index && index < nodes.size());
875 U_ASSERT(UCOL_SECONDARY <= level && level <= UCOL_TERTIARY);
876
877 if(weight16 == Collation::COMMON_WEIGHT16) {
878 return findCommonNode(index, level);
879 }
880
881 // If this will be the first below-common weight for the parent node,
882 // then we will also need to insert a common weight after it.
883 int64_t node = nodes.elementAti(index);
884 U_ASSERT(strengthFromNode(node) < level); // parent node is stronger
885 if(weight16 != 0 && weight16 < Collation::COMMON_WEIGHT16) {
886 int32_t hasThisLevelBefore = level == UCOL_SECONDARY ? HAS_BEFORE2 : HAS_BEFORE3;
887 if((node & hasThisLevelBefore) == 0) {
888 // The parent node has an implied level-common weight.
889 int64_t commonNode =
890 nodeFromWeight16(Collation::COMMON_WEIGHT16) | nodeFromStrength(level);
891 if(level == UCOL_SECONDARY) {
892 // Move the HAS_BEFORE3 flag from the parent node
893 // to the new secondary common node.
894 commonNode |= node & HAS_BEFORE3;
895 node &= ~(int64_t)HAS_BEFORE3;
896 }
897 nodes.setElementAt(node | hasThisLevelBefore, index);
898 // Insert below-common-weight node.
899 int32_t nextIndex = nextIndexFromNode(node);
900 node = nodeFromWeight16(weight16) | nodeFromStrength(level);
901 index = insertNodeBetween(index, nextIndex, node, errorCode);
902 // Insert common-weight node.
903 insertNodeBetween(index, nextIndex, commonNode, errorCode);
904 // Return index of below-common-weight node.
905 return index;
906 }
907 }
908
909 // Find the root CE's weight for this level.
910 // Postpone insertion if not found:
911 // Insert the new root node before the next stronger node,
912 // or before the next root node with the same strength and a larger weight.
913 int32_t nextIndex;
914 while((nextIndex = nextIndexFromNode(node)) != 0) {
915 node = nodes.elementAti(nextIndex);
916 int32_t nextStrength = strengthFromNode(node);
917 if(nextStrength <= level) {
918 // Insert before a stronger node.
919 if(nextStrength < level) { break; }
920 // nextStrength == level
921 if(!isTailoredNode(node)) {
922 uint32_t nextWeight16 = weight16FromNode(node);
923 if(nextWeight16 == weight16) {
924 // Found the node for the root CE up to this level.
925 return nextIndex;
926 }
927 // Insert before a node with a larger same-strength weight.
928 if(nextWeight16 > weight16) { break; }
929 }
930 }
931 // Skip the next node.
932 index = nextIndex;
933 }
934 node = nodeFromWeight16(weight16) | nodeFromStrength(level);
935 return insertNodeBetween(index, nextIndex, node, errorCode);
936}
937
938int32_t
939CollationBuilder::insertTailoredNodeAfter(int32_t index, int32_t strength, UErrorCode &errorCode) {
940 if(U_FAILURE(errorCode)) { return 0; }
941 U_ASSERT(0 <= index && index < nodes.size());
942 if(strength >= UCOL_SECONDARY) {
943 index = findCommonNode(index, UCOL_SECONDARY);
944 if(strength >= UCOL_TERTIARY) {
945 index = findCommonNode(index, UCOL_TERTIARY);
946 }
947 }
948 // Postpone insertion:
949 // Insert the new node before the next one with a strength at least as strong.
950 int64_t node = nodes.elementAti(index);
951 int32_t nextIndex;
952 while((nextIndex = nextIndexFromNode(node)) != 0) {
953 node = nodes.elementAti(nextIndex);
954 if(strengthFromNode(node) <= strength) { break; }
955 // Skip the next node which has a weaker (larger) strength than the new one.
956 index = nextIndex;
957 }
958 node = IS_TAILORED | nodeFromStrength(strength);
959 return insertNodeBetween(index, nextIndex, node, errorCode);
960}
961
962int32_t
963CollationBuilder::insertNodeBetween(int32_t index, int32_t nextIndex, int64_t node,
964 UErrorCode &errorCode) {
965 if(U_FAILURE(errorCode)) { return 0; }
966 U_ASSERT(previousIndexFromNode(node) == 0);
967 U_ASSERT(nextIndexFromNode(node) == 0);
968 U_ASSERT(nextIndexFromNode(nodes.elementAti(index)) == nextIndex);
969 // Append the new node and link it to the existing nodes.
970 int32_t newIndex = nodes.size();
971 node |= nodeFromPreviousIndex(index) | nodeFromNextIndex(nextIndex);
972 nodes.addElement(node, errorCode);
973 if(U_FAILURE(errorCode)) { return 0; }
974 // nodes[index].nextIndex = newIndex
975 node = nodes.elementAti(index);
976 nodes.setElementAt(changeNodeNextIndex(node, newIndex), index);
977 // nodes[nextIndex].previousIndex = newIndex
978 if(nextIndex != 0) {
979 node = nodes.elementAti(nextIndex);
980 nodes.setElementAt(changeNodePreviousIndex(node, newIndex), nextIndex);
981 }
982 return newIndex;
983}
984
985int32_t
986CollationBuilder::findCommonNode(int32_t index, int32_t strength) const {
987 U_ASSERT(UCOL_SECONDARY <= strength && strength <= UCOL_TERTIARY);
988 int64_t node = nodes.elementAti(index);
989 if(strengthFromNode(node) >= strength) {
990 // The current node is no stronger.
991 return index;
992 }
993 if(strength == UCOL_SECONDARY ? !nodeHasBefore2(node) : !nodeHasBefore3(node)) {
994 // The current node implies the strength-common weight.
995 return index;
996 }
997 index = nextIndexFromNode(node);
998 node = nodes.elementAti(index);
999 U_ASSERT(!isTailoredNode(node) && strengthFromNode(node) == strength &&
1000 weight16FromNode(node) < Collation::COMMON_WEIGHT16);
1001 // Skip to the explicit common node.
1002 do {
1003 index = nextIndexFromNode(node);
1004 node = nodes.elementAti(index);
1005 U_ASSERT(strengthFromNode(node) >= strength);
1006 } while(isTailoredNode(node) || strengthFromNode(node) > strength ||
1007 weight16FromNode(node) < Collation::COMMON_WEIGHT16);
1008 U_ASSERT(weight16FromNode(node) == Collation::COMMON_WEIGHT16);
1009 return index;
1010}
1011
1012void
1013CollationBuilder::setCaseBits(const UnicodeString &nfdString,
1014 const char *&parserErrorReason, UErrorCode &errorCode) {
1015 if(U_FAILURE(errorCode)) { return; }
1016 int32_t numTailoredPrimaries = 0;
1017 for(int32_t i = 0; i < cesLength; ++i) {
1018 if(ceStrength(ces[i]) == UCOL_PRIMARY) { ++numTailoredPrimaries; }
1019 }
1020 // We should not be able to get too many case bits because
1021 // cesLength<=31==MAX_EXPANSION_LENGTH.
1022 // 31 pairs of case bits fit into an int64_t without setting its sign bit.
1023 U_ASSERT(numTailoredPrimaries <= 31);
1024
1025 int64_t cases = 0;
1026 if(numTailoredPrimaries > 0) {
1027 const UChar *s = nfdString.getBuffer();
1028 UTF16CollationIterator baseCEs(baseData, FALSE, s, s, s + nfdString.length());
1029 int32_t baseCEsLength = baseCEs.fetchCEs(errorCode) - 1;
1030 if(U_FAILURE(errorCode)) {
1031 parserErrorReason = "fetching root CEs for tailored string";
1032 return;
1033 }
1034 U_ASSERT(baseCEsLength >= 0 && baseCEs.getCE(baseCEsLength) == Collation::NO_CE);
1035
1036 uint32_t lastCase = 0;
1037 int32_t numBasePrimaries = 0;
1038 for(int32_t i = 0; i < baseCEsLength; ++i) {
1039 int64_t ce = baseCEs.getCE(i);
1040 if((ce >> 32) != 0) {
1041 ++numBasePrimaries;
1042 uint32_t c = ((uint32_t)ce >> 14) & 3;
1043 U_ASSERT(c == 0 || c == 2); // lowercase or uppercase, no mixed case in any base CE
1044 if(numBasePrimaries < numTailoredPrimaries) {
1045 cases |= (int64_t)c << ((numBasePrimaries - 1) * 2);
1046 } else if(numBasePrimaries == numTailoredPrimaries) {
1047 lastCase = c;
1048 } else if(c != lastCase) {
1049 // There are more base primary CEs than tailored primaries.
1050 // Set mixed case if the case bits of the remainder differ.
1051 lastCase = 1;
1052 // Nothing more can change.
1053 break;
1054 }
1055 }
1056 }
1057 if(numBasePrimaries >= numTailoredPrimaries) {
1058 cases |= (int64_t)lastCase << ((numTailoredPrimaries - 1) * 2);
1059 }
1060 }
1061
1062 for(int32_t i = 0; i < cesLength; ++i) {
1063 int64_t ce = ces[i] & INT64_C(0xffffffffffff3fff); // clear old case bits
1064 int32_t strength = ceStrength(ce);
1065 if(strength == UCOL_PRIMARY) {
1066 ce |= (cases & 3) << 14;
1067 cases >>= 2;
1068 } else if(strength == UCOL_TERTIARY) {
1069 // Tertiary CEs must have uppercase bits.
1070 // See the LDML spec, and comments in class CollationCompare.
1071 ce |= 0x8000;
1072 }
1073 // Tertiary ignorable CEs must have 0 case bits.
1074 // We set 0 case bits for secondary CEs too
1075 // since currently only U+0345 is cased and maps to a secondary CE,
1076 // and it is lowercase. Other secondaries are uncased.
1077 // See [[:Cased:]&[:uca1=:]] where uca1 queries the root primary weight.
1078 ces[i] = ce;
1079 }
1080}
1081
1082void
1083CollationBuilder::suppressContractions(const UnicodeSet &set, const char *&parserErrorReason,
1084 UErrorCode &errorCode) {
1085 if(U_FAILURE(errorCode)) { return; }
1086 dataBuilder->suppressContractions(set, errorCode);
1087 if(U_FAILURE(errorCode)) {
1088 parserErrorReason = "application of [suppressContractions [set]] failed";
1089 }
1090}
1091
1092void
1093CollationBuilder::optimize(const UnicodeSet &set, const char *& /* parserErrorReason */,
1094 UErrorCode &errorCode) {
1095 if(U_FAILURE(errorCode)) { return; }
1096 optimizeSet.addAll(set);
1097}
1098
1099uint32_t
1100CollationBuilder::addWithClosure(const UnicodeString &nfdPrefix, const UnicodeString &nfdString,
1101 const int64_t newCEs[], int32_t newCEsLength, uint32_t ce32,
1102 UErrorCode &errorCode) {
1103 // Map from the NFD input to the CEs.
1104 ce32 = addIfDifferent(nfdPrefix, nfdString, newCEs, newCEsLength, ce32, errorCode);
1105 ce32 = addOnlyClosure(nfdPrefix, nfdString, newCEs, newCEsLength, ce32, errorCode);
1106 addTailComposites(nfdPrefix, nfdString, errorCode);
1107 return ce32;
1108}
1109
1110uint32_t
1111CollationBuilder::addOnlyClosure(const UnicodeString &nfdPrefix, const UnicodeString &nfdString,
1112 const int64_t newCEs[], int32_t newCEsLength, uint32_t ce32,
1113 UErrorCode &errorCode) {
1114 if(U_FAILURE(errorCode)) { return ce32; }
1115
1116 // Map from canonically equivalent input to the CEs. (But not from the all-NFD input.)
1117 if(nfdPrefix.isEmpty()) {
1118 CanonicalIterator stringIter(nfdString, errorCode);
1119 if(U_FAILURE(errorCode)) { return ce32; }
1120 UnicodeString prefix;
1121 for(;;) {
1122 UnicodeString str = stringIter.next();
1123 if(str.isBogus()) { break; }
1124 if(ignoreString(str, errorCode) || str == nfdString) { continue; }
1125 ce32 = addIfDifferent(prefix, str, newCEs, newCEsLength, ce32, errorCode);
1126 if(U_FAILURE(errorCode)) { return ce32; }
1127 }
1128 } else {
1129 CanonicalIterator prefixIter(nfdPrefix, errorCode);
1130 CanonicalIterator stringIter(nfdString, errorCode);
1131 if(U_FAILURE(errorCode)) { return ce32; }
1132 for(;;) {
1133 UnicodeString prefix = prefixIter.next();
1134 if(prefix.isBogus()) { break; }
1135 if(ignorePrefix(prefix, errorCode)) { continue; }
1136 UBool samePrefix = prefix == nfdPrefix;
1137 for(;;) {
1138 UnicodeString str = stringIter.next();
1139 if(str.isBogus()) { break; }
1140 if(ignoreString(str, errorCode) || (samePrefix && str == nfdString)) { continue; }
1141 ce32 = addIfDifferent(prefix, str, newCEs, newCEsLength, ce32, errorCode);
1142 if(U_FAILURE(errorCode)) { return ce32; }
1143 }
1144 stringIter.reset();
1145 }
1146 }
1147 return ce32;
1148}
1149
1150void
1151CollationBuilder::addTailComposites(const UnicodeString &nfdPrefix, const UnicodeString &nfdString,
1152 UErrorCode &errorCode) {
1153 if(U_FAILURE(errorCode)) { return; }
1154
1155 // Look for the last starter in the NFD string.
1156 UChar32 lastStarter;
1157 int32_t indexAfterLastStarter = nfdString.length();
1158 for(;;) {
1159 if(indexAfterLastStarter == 0) { return; } // no starter at all
1160 lastStarter = nfdString.char32At(indexAfterLastStarter - 1);
1161 if(nfd.getCombiningClass(lastStarter) == 0) { break; }
1162 indexAfterLastStarter -= U16_LENGTH(lastStarter);
1163 }
1164 // No closure to Hangul syllables since we decompose them on the fly.
1165 if(Hangul::isJamoL(lastStarter)) { return; }
1166
1167 // Are there any composites whose decomposition starts with the lastStarter?
1168 // Note: Normalizer2Impl does not currently return start sets for NFC_QC=Maybe characters.
1169 // We might find some more equivalent mappings here if it did.
1170 UnicodeSet composites;
1171 if(!nfcImpl.getCanonStartSet(lastStarter, composites)) { return; }
1172
1173 UnicodeString decomp;
1174 UnicodeString newNFDString, newString;
1175 int64_t newCEs[Collation::MAX_EXPANSION_LENGTH];
1176 UnicodeSetIterator iter(composites);
1177 while(iter.next()) {
1178 U_ASSERT(!iter.isString());
1179 UChar32 composite = iter.getCodepoint();
1180 nfd.getDecomposition(composite, decomp);
1181 if(!mergeCompositeIntoString(nfdString, indexAfterLastStarter, composite, decomp,
1182 newNFDString, newString, errorCode)) {
1183 continue;
1184 }
1185 int32_t newCEsLength = dataBuilder->getCEs(nfdPrefix, newNFDString, newCEs, 0);
1186 if(newCEsLength > Collation::MAX_EXPANSION_LENGTH) {
1187 // Ignore mappings that we cannot store.
1188 continue;
1189 }
1190 // Note: It is possible that the newCEs do not make use of the mapping
1191 // for which we are adding the tail composites, in which case we might be adding
1192 // unnecessary mappings.
1193 // For example, when we add tail composites for ae^ (^=combining circumflex),
1194 // UCA discontiguous-contraction matching does not find any matches
1195 // for ae_^ (_=any combining diacritic below) *unless* there is also
1196 // a contraction mapping for ae.
1197 // Thus, if there is no ae contraction, then the ae^ mapping is ignored
1198 // while fetching the newCEs for ae_^.
1199 // TODO: Try to detect this effectively.
1200 // (Alternatively, print a warning when prefix contractions are missing.)
1201
1202 // We do not need an explicit mapping for the NFD strings.
1203 // It is fine if the NFD input collates like this via a sequence of mappings.
1204 // It also saves a little bit of space, and may reduce the set of characters with contractions.
1205 uint32_t ce32 = addIfDifferent(nfdPrefix, newString,
1206 newCEs, newCEsLength, Collation::UNASSIGNED_CE32, errorCode);
1207 if(ce32 != Collation::UNASSIGNED_CE32) {
1208 // was different, was added
1209 addOnlyClosure(nfdPrefix, newNFDString, newCEs, newCEsLength, ce32, errorCode);
1210 }
1211 }
1212}
1213
1214UBool
1215CollationBuilder::mergeCompositeIntoString(const UnicodeString &nfdString,
1216 int32_t indexAfterLastStarter,
1217 UChar32 composite, const UnicodeString &decomp,
1218 UnicodeString &newNFDString, UnicodeString &newString,
1219 UErrorCode &errorCode) const {
1220 if(U_FAILURE(errorCode)) { return FALSE; }
1221 U_ASSERT(nfdString.char32At(indexAfterLastStarter - 1) == decomp.char32At(0));
1222 int32_t lastStarterLength = decomp.moveIndex32(0, 1);
1223 if(lastStarterLength == decomp.length()) {
1224 // Singleton decompositions should be found by addWithClosure()
1225 // and the CanonicalIterator, so we can ignore them here.
1226 return FALSE;
1227 }
1228 if(nfdString.compare(indexAfterLastStarter, 0x7fffffff,
1229 decomp, lastStarterLength, 0x7fffffff) == 0) {
1230 // same strings, nothing new to be found here
1231 return FALSE;
1232 }
1233
1234 // Make new FCD strings that combine a composite, or its decomposition,
1235 // into the nfdString's last starter and the combining marks following it.
1236 // Make an NFD version, and a version with the composite.
1237 newNFDString.setTo(nfdString, 0, indexAfterLastStarter);
1238 newString.setTo(nfdString, 0, indexAfterLastStarter - lastStarterLength).append(composite);
1239
1240 // The following is related to discontiguous contraction matching,
1241 // but builds only FCD strings (or else returns FALSE).
1242 int32_t sourceIndex = indexAfterLastStarter;
1243 int32_t decompIndex = lastStarterLength;
1244 // Small optimization: We keep the source character across loop iterations
1245 // because we do not always consume it,
1246 // and then need not fetch it again nor look up its combining class again.
1247 UChar32 sourceChar = U_SENTINEL;
1248 // The cc variables need to be declared before the loop so that at the end
1249 // they are set to the last combining classes seen.
1250 uint8_t sourceCC = 0;
1251 uint8_t decompCC = 0;
1252 for(;;) {
1253 if(sourceChar < 0) {
1254 if(sourceIndex >= nfdString.length()) { break; }
1255 sourceChar = nfdString.char32At(sourceIndex);
1256 sourceCC = nfd.getCombiningClass(sourceChar);
1257 U_ASSERT(sourceCC != 0);
1258 }
1259 // We consume a decomposition character in each iteration.
1260 if(decompIndex >= decomp.length()) { break; }
1261 UChar32 decompChar = decomp.char32At(decompIndex);
1262 decompCC = nfd.getCombiningClass(decompChar);
1263 // Compare the two characters and their combining classes.
1264 if(decompCC == 0) {
1265 // Unable to merge because the source contains a non-zero combining mark
1266 // but the composite's decomposition contains another starter.
1267 // The strings would not be equivalent.
1268 return FALSE;
1269 } else if(sourceCC < decompCC) {
1270 // Composite + sourceChar would not be FCD.
1271 return FALSE;
1272 } else if(decompCC < sourceCC) {
1273 newNFDString.append(decompChar);
1274 decompIndex += U16_LENGTH(decompChar);
1275 } else if(decompChar != sourceChar) {
1276 // Blocked because same combining class.
1277 return FALSE;
1278 } else { // match: decompChar == sourceChar
1279 newNFDString.append(decompChar);
1280 decompIndex += U16_LENGTH(decompChar);
1281 sourceIndex += U16_LENGTH(decompChar);
1282 sourceChar = U_SENTINEL;
1283 }
1284 }
1285 // We are at the end of at least one of the two inputs.
1286 if(sourceChar >= 0) { // more characters from nfdString but not from decomp
1287 if(sourceCC < decompCC) {
1288 // Appending the next source character to the composite would not be FCD.
1289 return FALSE;
1290 }
1291 newNFDString.append(nfdString, sourceIndex, 0x7fffffff);
1292 newString.append(nfdString, sourceIndex, 0x7fffffff);
1293 } else if(decompIndex < decomp.length()) { // more characters from decomp, not from nfdString
1294 newNFDString.append(decomp, decompIndex, 0x7fffffff);
1295 }
1296 U_ASSERT(nfd.isNormalized(newNFDString, errorCode));
1297 U_ASSERT(fcd.isNormalized(newString, errorCode));
1298 U_ASSERT(nfd.normalize(newString, errorCode) == newNFDString); // canonically equivalent
1299 return TRUE;
1300}
1301
1302UBool
1303CollationBuilder::ignorePrefix(const UnicodeString &s, UErrorCode &errorCode) const {
1304 // Do not map non-FCD prefixes.
1305 return !isFCD(s, errorCode);
1306}
1307
1308UBool
1309CollationBuilder::ignoreString(const UnicodeString &s, UErrorCode &errorCode) const {
1310 // Do not map non-FCD strings.
1311 // Do not map strings that start with Hangul syllables: We decompose those on the fly.
1312 return !isFCD(s, errorCode) || Hangul::isHangul(s.charAt(0));
1313}
1314
1315UBool
1316CollationBuilder::isFCD(const UnicodeString &s, UErrorCode &errorCode) const {
1317 return U_SUCCESS(errorCode) && fcd.isNormalized(s, errorCode);
1318}
1319
1320void
1321CollationBuilder::closeOverComposites(UErrorCode &errorCode) {
1322 UnicodeSet composites(UNICODE_STRING_SIMPLE("[:NFD_QC=N:]"), errorCode); // Java: static final
1323 if(U_FAILURE(errorCode)) { return; }
1324 // Hangul is decomposed on the fly during collation.
1325 composites.remove(Hangul::HANGUL_BASE, Hangul::HANGUL_END);
1326 UnicodeString prefix; // empty
1327 UnicodeString nfdString;
1328 UnicodeSetIterator iter(composites);
1329 while(iter.next()) {
1330 U_ASSERT(!iter.isString());
1331 nfd.getDecomposition(iter.getCodepoint(), nfdString);
1332 cesLength = dataBuilder->getCEs(nfdString, ces, 0);
1333 if(cesLength > Collation::MAX_EXPANSION_LENGTH) {
1334 // Too many CEs from the decomposition (unusual), ignore this composite.
1335 // We could add a capacity parameter to getCEs() and reallocate if necessary.
1336 // However, this can only really happen in contrived cases.
1337 continue;
1338 }
1339 const UnicodeString &composite(iter.getString());
1340 addIfDifferent(prefix, composite, ces, cesLength, Collation::UNASSIGNED_CE32, errorCode);
1341 }
1342}
1343
1344uint32_t
1345CollationBuilder::addIfDifferent(const UnicodeString &prefix, const UnicodeString &str,
1346 const int64_t newCEs[], int32_t newCEsLength, uint32_t ce32,
1347 UErrorCode &errorCode) {
1348 if(U_FAILURE(errorCode)) { return ce32; }
1349 int64_t oldCEs[Collation::MAX_EXPANSION_LENGTH];
1350 int32_t oldCEsLength = dataBuilder->getCEs(prefix, str, oldCEs, 0);
1351 if(!sameCEs(newCEs, newCEsLength, oldCEs, oldCEsLength)) {
1352 if(ce32 == Collation::UNASSIGNED_CE32) {
1353 ce32 = dataBuilder->encodeCEs(newCEs, newCEsLength, errorCode);
1354 }
1355 dataBuilder->addCE32(prefix, str, ce32, errorCode);
1356 }
1357 return ce32;
1358}
1359
1360UBool
1361CollationBuilder::sameCEs(const int64_t ces1[], int32_t ces1Length,
1362 const int64_t ces2[], int32_t ces2Length) {
1363 if(ces1Length != ces2Length) {
1364 return FALSE;
1365 }
1366 U_ASSERT(ces1Length <= Collation::MAX_EXPANSION_LENGTH);
1367 for(int32_t i = 0; i < ces1Length; ++i) {
1368 if(ces1[i] != ces2[i]) { return FALSE; }
1369 }
1370 return TRUE;
1371}
1372
1373#ifdef DEBUG_COLLATION_BUILDER
1374
1375uint32_t
1376alignWeightRight(uint32_t w) {
1377 if(w != 0) {
1378 while((w & 0xff) == 0) { w >>= 8; }
1379 }
1380 return w;
1381}
1382
1383#endif
1384
1385void
1386CollationBuilder::makeTailoredCEs(UErrorCode &errorCode) {
1387 if(U_FAILURE(errorCode)) { return; }
1388
1389 CollationWeights primaries, secondaries, tertiaries;
1390 int64_t *nodesArray = nodes.getBuffer();
1391#ifdef DEBUG_COLLATION_BUILDER
1392 puts("\nCollationBuilder::makeTailoredCEs()");
1393#endif
1394
1395 for(int32_t rpi = 0; rpi < rootPrimaryIndexes.size(); ++rpi) {
1396 int32_t i = rootPrimaryIndexes.elementAti(rpi);
1397 int64_t node = nodesArray[i];
1398 uint32_t p = weight32FromNode(node);
1399 uint32_t s = p == 0 ? 0 : Collation::COMMON_WEIGHT16;
1400 uint32_t t = s;
1401 uint32_t q = 0;
1402 UBool pIsTailored = FALSE;
1403 UBool sIsTailored = FALSE;
1404 UBool tIsTailored = FALSE;
1405#ifdef DEBUG_COLLATION_BUILDER
1406 printf("\nprimary %lx\n", (long)alignWeightRight(p));
1407#endif
1408 int32_t pIndex = p == 0 ? 0 : rootElements.findPrimary(p);
1409 int32_t nextIndex = nextIndexFromNode(node);
1410 while(nextIndex != 0) {
1411 i = nextIndex;
1412 node = nodesArray[i];
1413 nextIndex = nextIndexFromNode(node);
1414 int32_t strength = strengthFromNode(node);
1415 if(strength == UCOL_QUATERNARY) {
1416 U_ASSERT(isTailoredNode(node));
1417#ifdef DEBUG_COLLATION_BUILDER
1418 printf(" quat+ ");
1419#endif
1420 if(q == 3) {
1421 errorCode = U_BUFFER_OVERFLOW_ERROR;
1422 errorReason = "quaternary tailoring gap too small";
1423 return;
1424 }
1425 ++q;
1426 } else {
1427 if(strength == UCOL_TERTIARY) {
1428 if(isTailoredNode(node)) {
1429#ifdef DEBUG_COLLATION_BUILDER
1430 printf(" ter+ ");
1431#endif
1432 if(!tIsTailored) {
1433 // First tailored tertiary node for [p, s].
1434 int32_t tCount = countTailoredNodes(nodesArray, nextIndex,
1435 UCOL_TERTIARY) + 1;
1436 uint32_t tLimit;
1437 if(t == 0) {
1438 // Gap at the beginning of the tertiary CE range.
1439 t = rootElements.getTertiaryBoundary() - 0x100;
1440 tLimit = rootElements.getFirstTertiaryCE() & Collation::ONLY_TERTIARY_MASK;
1441 } else if(!pIsTailored && !sIsTailored) {
1442 // p and s are root weights.
1443 tLimit = rootElements.getTertiaryAfter(pIndex, s, t);
1444 } else if(t == Collation::BEFORE_WEIGHT16) {
1445 tLimit = Collation::COMMON_WEIGHT16;
1446 } else {
1447 // [p, s] is tailored.
1448 U_ASSERT(t == Collation::COMMON_WEIGHT16);
1449 tLimit = rootElements.getTertiaryBoundary();
1450 }
1451 U_ASSERT(tLimit == 0x4000 || (tLimit & ~Collation::ONLY_TERTIARY_MASK) == 0);
1452 tertiaries.initForTertiary();
1453 if(!tertiaries.allocWeights(t, tLimit, tCount)) {
1454 errorCode = U_BUFFER_OVERFLOW_ERROR;
1455 errorReason = "tertiary tailoring gap too small";
1456 return;
1457 }
1458 tIsTailored = TRUE;
1459 }
1460 t = tertiaries.nextWeight();
1461 U_ASSERT(t != 0xffffffff);
1462 } else {
1463 t = weight16FromNode(node);
1464 tIsTailored = FALSE;
1465#ifdef DEBUG_COLLATION_BUILDER
1466 printf(" ter %lx\n", (long)alignWeightRight(t));
1467#endif
1468 }
1469 } else {
1470 if(strength == UCOL_SECONDARY) {
1471 if(isTailoredNode(node)) {
1472#ifdef DEBUG_COLLATION_BUILDER
1473 printf(" sec+ ");
1474#endif
1475 if(!sIsTailored) {
1476 // First tailored secondary node for p.
1477 int32_t sCount = countTailoredNodes(nodesArray, nextIndex,
1478 UCOL_SECONDARY) + 1;
1479 uint32_t sLimit;
1480 if(s == 0) {
1481 // Gap at the beginning of the secondary CE range.
1482 s = rootElements.getSecondaryBoundary() - 0x100;
1483 sLimit = rootElements.getFirstSecondaryCE() >> 16;
1484 } else if(!pIsTailored) {
1485 // p is a root primary.
1486 sLimit = rootElements.getSecondaryAfter(pIndex, s);
1487 } else if(s == Collation::BEFORE_WEIGHT16) {
1488 sLimit = Collation::COMMON_WEIGHT16;
1489 } else {
1490 // p is a tailored primary.
1491 U_ASSERT(s == Collation::COMMON_WEIGHT16);
1492 sLimit = rootElements.getSecondaryBoundary();
1493 }
1494 if(s == Collation::COMMON_WEIGHT16) {
1495 // Do not tailor into the getSortKey() range of
1496 // compressed common secondaries.
1497 s = rootElements.getLastCommonSecondary();
1498 }
1499 secondaries.initForSecondary();
1500 if(!secondaries.allocWeights(s, sLimit, sCount)) {
1501 errorCode = U_BUFFER_OVERFLOW_ERROR;
1502 errorReason = "secondary tailoring gap too small";
1503#ifdef DEBUG_COLLATION_BUILDER
1504 printf("!secondaries.allocWeights(%lx, %lx, sCount=%ld)\n",
1505 (long)alignWeightRight(s), (long)alignWeightRight(sLimit),
1506 (long)alignWeightRight(sCount));
1507#endif
1508 return;
1509 }
1510 sIsTailored = TRUE;
1511 }
1512 s = secondaries.nextWeight();
1513 U_ASSERT(s != 0xffffffff);
1514 } else {
1515 s = weight16FromNode(node);
1516 sIsTailored = FALSE;
1517#ifdef DEBUG_COLLATION_BUILDER
1518 printf(" sec %lx\n", (long)alignWeightRight(s));
1519#endif
1520 }
1521 } else /* UCOL_PRIMARY */ {
1522 U_ASSERT(isTailoredNode(node));
1523#ifdef DEBUG_COLLATION_BUILDER
1524 printf("pri+ ");
1525#endif
1526 if(!pIsTailored) {
1527 // First tailored primary node in this list.
1528 int32_t pCount = countTailoredNodes(nodesArray, nextIndex,
1529 UCOL_PRIMARY) + 1;
1530 UBool isCompressible = baseData->isCompressiblePrimary(p);
1531 uint32_t pLimit =
1532 rootElements.getPrimaryAfter(p, pIndex, isCompressible);
1533 primaries.initForPrimary(isCompressible);
1534 if(!primaries.allocWeights(p, pLimit, pCount)) {
1535 errorCode = U_BUFFER_OVERFLOW_ERROR; // TODO: introduce a more specific UErrorCode?
1536 errorReason = "primary tailoring gap too small";
1537 return;
1538 }
1539 pIsTailored = TRUE;
1540 }
1541 p = primaries.nextWeight();
1542 U_ASSERT(p != 0xffffffff);
1543 s = Collation::COMMON_WEIGHT16;
1544 sIsTailored = FALSE;
1545 }
1546 t = s == 0 ? 0 : Collation::COMMON_WEIGHT16;
1547 tIsTailored = FALSE;
1548 }
1549 q = 0;
1550 }
1551 if(isTailoredNode(node)) {
1552 nodesArray[i] = Collation::makeCE(p, s, t, q);
1553#ifdef DEBUG_COLLATION_BUILDER
1554 printf("%016llx\n", (long long)nodesArray[i]);
1555#endif
1556 }
1557 }
1558 }
1559}
1560
1561int32_t
1562CollationBuilder::countTailoredNodes(const int64_t *nodesArray, int32_t i, int32_t strength) {
1563 int32_t count = 0;
1564 for(;;) {
1565 if(i == 0) { break; }
1566 int64_t node = nodesArray[i];
1567 if(strengthFromNode(node) < strength) { break; }
1568 if(strengthFromNode(node) == strength) {
1569 if(isTailoredNode(node)) {
1570 ++count;
1571 } else {
1572 break;
1573 }
1574 }
1575 i = nextIndexFromNode(node);
1576 }
1577 return count;
1578}
1579
1580class CEFinalizer : public CollationDataBuilder::CEModifier {
1581public:
1582 CEFinalizer(const int64_t *ces) : finalCEs(ces) {}
1583 virtual ~CEFinalizer();
1584 virtual int64_t modifyCE32(uint32_t ce32) const {
1585 U_ASSERT(!Collation::isSpecialCE32(ce32));
1586 if(CollationBuilder::isTempCE32(ce32)) {
1587 // retain case bits
1588 return finalCEs[CollationBuilder::indexFromTempCE32(ce32)] | ((ce32 & 0xc0) << 8);
1589 } else {
1590 return Collation::NO_CE;
1591 }
1592 }
1593 virtual int64_t modifyCE(int64_t ce) const {
1594 if(CollationBuilder::isTempCE(ce)) {
1595 // retain case bits
1596 return finalCEs[CollationBuilder::indexFromTempCE(ce)] | (ce & 0xc000);
1597 } else {
1598 return Collation::NO_CE;
1599 }
1600 }
1601
1602private:
1603 const int64_t *finalCEs;
1604};
1605
1606CEFinalizer::~CEFinalizer() {}
1607
1608void
1609CollationBuilder::finalizeCEs(UErrorCode &errorCode) {
1610 if(U_FAILURE(errorCode)) { return; }
1611 LocalPointer<CollationDataBuilder> newBuilder(new CollationDataBuilder(errorCode), errorCode);
1612 if(U_FAILURE(errorCode)) {
1613 return;
1614 }
1615 newBuilder->initForTailoring(baseData, errorCode);
1616 CEFinalizer finalizer(nodes.getBuffer());
1617 newBuilder->copyFrom(*dataBuilder, finalizer, errorCode);
1618 if(U_FAILURE(errorCode)) { return; }
1619 delete dataBuilder;
1620 dataBuilder = newBuilder.orphan();
1621}
1622
1623int32_t
1624CollationBuilder::ceStrength(int64_t ce) {
1625 return
1626 isTempCE(ce) ? strengthFromTempCE(ce) :
1627 (ce & INT64_C(0xff00000000000000)) != 0 ? UCOL_PRIMARY :
1628 ((uint32_t)ce & 0xff000000) != 0 ? UCOL_SECONDARY :
1629 ce != 0 ? UCOL_TERTIARY :
1630 UCOL_IDENTICAL;
1631}
1632
1633U_NAMESPACE_END
1634
1635U_NAMESPACE_USE
1636
1637U_CAPI UCollator * U_EXPORT2
1638ucol_openRules(const UChar *rules, int32_t rulesLength,
1639 UColAttributeValue normalizationMode, UCollationStrength strength,
1640 UParseError *parseError, UErrorCode *pErrorCode) {
1641 if(U_FAILURE(*pErrorCode)) { return NULL; }
1642 if(rules == NULL && rulesLength != 0) {
1643 *pErrorCode = U_ILLEGAL_ARGUMENT_ERROR;
1644 return NULL;
1645 }
1646 RuleBasedCollator *coll = new RuleBasedCollator();
1647 if(coll == NULL) {
1648 *pErrorCode = U_MEMORY_ALLOCATION_ERROR;
1649 return NULL;
1650 }
1651 UnicodeString r((UBool)(rulesLength < 0), rules, rulesLength);
1652 coll->internalBuildTailoring(r, strength, normalizationMode, parseError, NULL, *pErrorCode);
1653 if(U_FAILURE(*pErrorCode)) {
1654 delete coll;
1655 return NULL;
1656 }
1657 return coll->toUCollator();
1658}
1659
1660static const int32_t internalBufferSize = 512;
1661
1662// The @internal ucol_getUnsafeSet() was moved here from ucol_sit.cpp
1663// because it calls UnicodeSet "builder" code that depends on all Unicode properties,
1664// and the rest of the collation "runtime" code only depends on normalization.
1665// This function is not related to the collation builder,
1666// but it did not seem worth moving it into its own .cpp file,
1667// nor rewriting it to use lower-level UnicodeSet and Normalizer2Impl methods.
1668U_CAPI int32_t U_EXPORT2
1669ucol_getUnsafeSet( const UCollator *coll,
1670 USet *unsafe,
1671 UErrorCode *status)
1672{
1673 UChar buffer[internalBufferSize];
1674 int32_t len = 0;
1675
1676 uset_clear(unsafe);
1677
1678 // cccpattern = "[[:^tccc=0:][:^lccc=0:]]", unfortunately variant
1679 static const UChar cccpattern[25] = { 0x5b, 0x5b, 0x3a, 0x5e, 0x74, 0x63, 0x63, 0x63, 0x3d, 0x30, 0x3a, 0x5d,
1680 0x5b, 0x3a, 0x5e, 0x6c, 0x63, 0x63, 0x63, 0x3d, 0x30, 0x3a, 0x5d, 0x5d, 0x00 };
1681
1682 // add chars that fail the fcd check
1683 uset_applyPattern(unsafe, cccpattern, 24, USET_IGNORE_SPACE, status);
1684
1685 // add lead/trail surrogates
1686 // (trail surrogates should need to be unsafe only if the caller tests for UTF-16 code *units*,
1687 // not when testing code *points*)
1688 uset_addRange(unsafe, 0xd800, 0xdfff);
1689
1690 USet *contractions = uset_open(0,0);
1691
1692 int32_t i = 0, j = 0;
1693 ucol_getContractionsAndExpansions(coll, contractions, NULL, FALSE, status);
1694 int32_t contsSize = uset_size(contractions);
1695 UChar32 c = 0;
1696 // Contraction set consists only of strings
1697 // to get unsafe code points, we need to
1698 // break the strings apart and add them to the unsafe set
1699 for(i = 0; i < contsSize; i++) {
1700 len = uset_getItem(contractions, i, NULL, NULL, buffer, internalBufferSize, status);
1701 if(len > 0) {
1702 j = 0;
1703 while(j < len) {
1704 U16_NEXT(buffer, j, len, c);
1705 if(j < len) {
1706 uset_add(unsafe, c);
1707 }
1708 }
1709 }
1710 }
1711
1712 uset_close(contractions);
1713
1714 return uset_size(unsafe);
1715}
1716
1717#endif // !UCONFIG_NO_COLLATION
1718