1 // © 2016 and later: Unicode, Inc. and others.
2 // License & terms of use: http://www.unicode.org/copyright.html
3 /*
4 *******************************************************************************
5 * Copyright (C) 2012-2015, International Business Machines
6 * Corporation and others. All Rights Reserved.
7 *******************************************************************************
8 * collationdatabuilder.cpp
9 *
10 * (replaced the former ucol_elm.cpp)
11 *
12 * created on: 2012apr01
13 * created by: Markus W. Scherer
14 */
15
16 #include "unicode/utypes.h"
17
18 #if !UCONFIG_NO_COLLATION
19
20 #include "unicode/localpointer.h"
21 #include "unicode/uchar.h"
22 #include "unicode/ucharstrie.h"
23 #include "unicode/ucharstriebuilder.h"
24 #include "unicode/uniset.h"
25 #include "unicode/unistr.h"
26 #include "unicode/usetiter.h"
27 #include "unicode/utf16.h"
28 #include "cmemory.h"
29 #include "collation.h"
30 #include "collationdata.h"
31 #include "collationdatabuilder.h"
32 #include "collationfastlatinbuilder.h"
33 #include "collationiterator.h"
34 #include "normalizer2impl.h"
35 #include "utrie2.h"
36 #include "uvectr32.h"
37 #include "uvectr64.h"
38 #include "uvector.h"
39
40 U_NAMESPACE_BEGIN
41
~CEModifier()42 CollationDataBuilder::CEModifier::~CEModifier() {}
43
44 /**
45 * Build-time context and CE32 for a code point.
46 * If a code point has contextual mappings, then the default (no-context) mapping
47 * and all conditional mappings are stored in a singly-linked list
48 * of ConditionalCE32, sorted by context strings.
49 *
50 * Context strings sort by prefix length, then by prefix, then by contraction suffix.
51 * Context strings must be unique and in ascending order.
52 */
53 struct ConditionalCE32 : public UMemory {
ConditionalCE32ConditionalCE3254 ConditionalCE32()
55 : context(),
56 ce32(0), defaultCE32(Collation::NO_CE32), builtCE32(Collation::NO_CE32),
57 next(-1) {}
ConditionalCE32ConditionalCE3258 ConditionalCE32(const UnicodeString &ct, uint32_t ce)
59 : context(ct),
60 ce32(ce), defaultCE32(Collation::NO_CE32), builtCE32(Collation::NO_CE32),
61 next(-1) {}
62
hasContextConditionalCE3263 inline UBool hasContext() const { return context.length() > 1; }
prefixLengthConditionalCE3264 inline int32_t prefixLength() const { return context.charAt(0); }
65
66 /**
67 * "\0" for the first entry for any code point, with its default CE32.
68 *
69 * Otherwise one unit with the length of the prefix string,
70 * then the prefix string, then the contraction suffix.
71 */
72 UnicodeString context;
73 /**
74 * CE32 for the code point and its context.
75 * Can be special (e.g., for an expansion) but not contextual (prefix or contraction tag).
76 */
77 uint32_t ce32;
78 /**
79 * Default CE32 for all contexts with this same prefix.
80 * Initially NO_CE32. Set only while building runtime data structures,
81 * and only on one of the nodes of a sub-list with the same prefix.
82 */
83 uint32_t defaultCE32;
84 /**
85 * CE32 for the built contexts.
86 * When fetching CEs from the builder, the contexts are built into their runtime form
87 * so that the normal collation implementation can process them.
88 * The result is cached in the list head. It is reset when the contexts are modified.
89 * All of these builtCE32 are invalidated by clearContexts(),
90 * via incrementing the contextsEra.
91 */
92 uint32_t builtCE32;
93 /**
94 * The "era" of building intermediate contexts when the above builtCE32 was set.
95 * When the array of cached, temporary contexts overflows, then clearContexts()
96 * removes them all and invalidates the builtCE32 that used to point to built tries.
97 */
98 int32_t era = -1;
99 /**
100 * Index of the next ConditionalCE32.
101 * Negative for the end of the list.
102 */
103 int32_t next;
104 // Note: We could create a separate class for all of the contextual mappings for
105 // a code point, with the builtCE32, the era, and a list of the actual mappings.
106 // The class that represents one mapping would then not need to
107 // store those fields in each element.
108 };
109
110 U_CDECL_BEGIN
111
112 void U_CALLCONV
uprv_deleteConditionalCE32(void * obj)113 uprv_deleteConditionalCE32(void *obj) {
114 delete static_cast<ConditionalCE32 *>(obj);
115 }
116
117 U_CDECL_END
118
119 /**
120 * Build-time collation element and character iterator.
121 * Uses the runtime CollationIterator for fetching CEs for a string
122 * but reads from the builder's unfinished data structures.
123 * In particular, this class reads from the unfinished trie
124 * and has to avoid CollationIterator::nextCE() and redirect other
125 * calls to data->getCE32() and data->getCE32FromSupplementary().
126 *
127 * We do this so that we need not implement the collation algorithm
128 * again for the builder and make it behave exactly like the runtime code.
129 * That would be more difficult to test and maintain than this indirection.
130 *
131 * Some CE32 tags (for example, the DIGIT_TAG) do not occur in the builder data,
132 * so the data accesses from those code paths need not be modified.
133 *
134 * This class iterates directly over whole code points
135 * so that the CollationIterator does not need the finished trie
136 * for handling the LEAD_SURROGATE_TAG.
137 */
138 class DataBuilderCollationIterator : public CollationIterator {
139 public:
140 DataBuilderCollationIterator(CollationDataBuilder &b);
141
142 virtual ~DataBuilderCollationIterator();
143
144 int32_t fetchCEs(const UnicodeString &str, int32_t start, int64_t ces[], int32_t cesLength);
145
146 virtual void resetToOffset(int32_t newOffset) override;
147 virtual int32_t getOffset() const override;
148
149 virtual UChar32 nextCodePoint(UErrorCode &errorCode) override;
150 virtual UChar32 previousCodePoint(UErrorCode &errorCode) override;
151
152 protected:
153 virtual void forwardNumCodePoints(int32_t num, UErrorCode &errorCode) override;
154 virtual void backwardNumCodePoints(int32_t num, UErrorCode &errorCode) override;
155
156 virtual uint32_t getDataCE32(UChar32 c) const override;
157 virtual uint32_t getCE32FromBuilderData(uint32_t ce32, UErrorCode &errorCode) override;
158
159 CollationDataBuilder &builder;
160 CollationData builderData;
161 uint32_t jamoCE32s[CollationData::JAMO_CE32S_LENGTH];
162 const UnicodeString *s;
163 int32_t pos;
164 };
165
DataBuilderCollationIterator(CollationDataBuilder & b)166 DataBuilderCollationIterator::DataBuilderCollationIterator(CollationDataBuilder &b)
167 : CollationIterator(&builderData, /*numeric=*/ false),
168 builder(b), builderData(b.nfcImpl),
169 s(nullptr), pos(0) {
170 builderData.base = builder.base;
171 // Set all of the jamoCE32s[] to indirection CE32s.
172 for(int32_t j = 0; j < CollationData::JAMO_CE32S_LENGTH; ++j) { // Count across Jamo types.
173 UChar32 jamo = CollationDataBuilder::jamoCpFromIndex(j);
174 jamoCE32s[j] = Collation::makeCE32FromTagAndIndex(Collation::BUILDER_DATA_TAG, jamo) |
175 CollationDataBuilder::IS_BUILDER_JAMO_CE32;
176 }
177 builderData.jamoCE32s = jamoCE32s;
178 }
179
~DataBuilderCollationIterator()180 DataBuilderCollationIterator::~DataBuilderCollationIterator() {}
181
182 int32_t
fetchCEs(const UnicodeString & str,int32_t start,int64_t ces[],int32_t cesLength)183 DataBuilderCollationIterator::fetchCEs(const UnicodeString &str, int32_t start,
184 int64_t ces[], int32_t cesLength) {
185 // Set the pointers each time, in case they changed due to reallocation.
186 builderData.ce32s = reinterpret_cast<const uint32_t *>(builder.ce32s.getBuffer());
187 builderData.ces = builder.ce64s.getBuffer();
188 builderData.contexts = builder.contexts.getBuffer();
189 // Modified copy of CollationIterator::nextCE() and CollationIterator::nextCEFromCE32().
190 reset();
191 s = &str;
192 pos = start;
193 UErrorCode errorCode = U_ZERO_ERROR;
194 while(U_SUCCESS(errorCode) && pos < s->length()) {
195 // No need to keep all CEs in the iterator buffer.
196 clearCEs();
197 UChar32 c = s->char32At(pos);
198 pos += U16_LENGTH(c);
199 uint32_t ce32 = utrie2_get32(builder.trie, c);
200 const CollationData *d;
201 if(ce32 == Collation::FALLBACK_CE32) {
202 d = builder.base;
203 ce32 = builder.base->getCE32(c);
204 } else {
205 d = &builderData;
206 }
207 appendCEsFromCE32(d, c, ce32, /*forward=*/ true, errorCode);
208 U_ASSERT(U_SUCCESS(errorCode));
209 for(int32_t i = 0; i < getCEsLength(); ++i) {
210 int64_t ce = getCE(i);
211 if(ce != 0) {
212 if(cesLength < Collation::MAX_EXPANSION_LENGTH) {
213 ces[cesLength] = ce;
214 }
215 ++cesLength;
216 }
217 }
218 }
219 return cesLength;
220 }
221
222 void
resetToOffset(int32_t newOffset)223 DataBuilderCollationIterator::resetToOffset(int32_t newOffset) {
224 reset();
225 pos = newOffset;
226 }
227
228 int32_t
getOffset() const229 DataBuilderCollationIterator::getOffset() const {
230 return pos;
231 }
232
233 UChar32
nextCodePoint(UErrorCode &)234 DataBuilderCollationIterator::nextCodePoint(UErrorCode & /*errorCode*/) {
235 if(pos == s->length()) {
236 return U_SENTINEL;
237 }
238 UChar32 c = s->char32At(pos);
239 pos += U16_LENGTH(c);
240 return c;
241 }
242
243 UChar32
previousCodePoint(UErrorCode &)244 DataBuilderCollationIterator::previousCodePoint(UErrorCode & /*errorCode*/) {
245 if(pos == 0) {
246 return U_SENTINEL;
247 }
248 UChar32 c = s->char32At(pos - 1);
249 pos -= U16_LENGTH(c);
250 return c;
251 }
252
253 void
forwardNumCodePoints(int32_t num,UErrorCode &)254 DataBuilderCollationIterator::forwardNumCodePoints(int32_t num, UErrorCode & /*errorCode*/) {
255 pos = s->moveIndex32(pos, num);
256 }
257
258 void
backwardNumCodePoints(int32_t num,UErrorCode &)259 DataBuilderCollationIterator::backwardNumCodePoints(int32_t num, UErrorCode & /*errorCode*/) {
260 pos = s->moveIndex32(pos, -num);
261 }
262
263 uint32_t
getDataCE32(UChar32 c) const264 DataBuilderCollationIterator::getDataCE32(UChar32 c) const {
265 return utrie2_get32(builder.trie, c);
266 }
267
268 uint32_t
getCE32FromBuilderData(uint32_t ce32,UErrorCode & errorCode)269 DataBuilderCollationIterator::getCE32FromBuilderData(uint32_t ce32, UErrorCode &errorCode) {
270 if (U_FAILURE(errorCode)) { return 0; }
271 U_ASSERT(Collation::hasCE32Tag(ce32, Collation::BUILDER_DATA_TAG));
272 if((ce32 & CollationDataBuilder::IS_BUILDER_JAMO_CE32) != 0) {
273 UChar32 jamo = Collation::indexFromCE32(ce32);
274 return utrie2_get32(builder.trie, jamo);
275 } else {
276 ConditionalCE32 *cond = builder.getConditionalCE32ForCE32(ce32);
277 if (cond == nullptr) {
278 errorCode = U_INTERNAL_PROGRAM_ERROR;
279 // TODO: ICU-21531 figure out why this happens.
280 return 0;
281 }
282 if(cond->builtCE32 == Collation::NO_CE32 || cond->era != builder.contextsEra) {
283 // Build the context-sensitive mappings into their runtime form and cache the result.
284 cond->builtCE32 = builder.buildContext(cond, errorCode);
285 if(errorCode == U_BUFFER_OVERFLOW_ERROR) {
286 errorCode = U_ZERO_ERROR;
287 builder.clearContexts();
288 cond->builtCE32 = builder.buildContext(cond, errorCode);
289 }
290 cond->era = builder.contextsEra;
291 builderData.contexts = builder.contexts.getBuffer();
292 }
293 return cond->builtCE32;
294 }
295 }
296
297 // ------------------------------------------------------------------------- ***
298
CollationDataBuilder(UBool icu4xMode,UErrorCode & errorCode)299 CollationDataBuilder::CollationDataBuilder(UBool icu4xMode, UErrorCode &errorCode)
300 : nfcImpl(*Normalizer2Factory::getNFCImpl(errorCode)),
301 base(nullptr), baseSettings(nullptr),
302 trie(nullptr),
303 ce32s(errorCode), ce64s(errorCode), conditionalCE32s(errorCode),
304 modified(false),
305 icu4xMode(icu4xMode),
306 fastLatinEnabled(false), fastLatinBuilder(nullptr),
307 collIter(nullptr) {
308 // Reserve the first CE32 for U+0000.
309 if (!icu4xMode) {
310 ce32s.addElement(0, errorCode);
311 }
312 conditionalCE32s.setDeleter(uprv_deleteConditionalCE32);
313 }
314
~CollationDataBuilder()315 CollationDataBuilder::~CollationDataBuilder() {
316 utrie2_close(trie);
317 delete fastLatinBuilder;
318 delete collIter;
319 }
320
321 void
initForTailoring(const CollationData * b,UErrorCode & errorCode)322 CollationDataBuilder::initForTailoring(const CollationData *b, UErrorCode &errorCode) {
323 if(U_FAILURE(errorCode)) { return; }
324 if(trie != nullptr) {
325 errorCode = U_INVALID_STATE_ERROR;
326 return;
327 }
328 if(b == nullptr) {
329 errorCode = U_ILLEGAL_ARGUMENT_ERROR;
330 return;
331 }
332 base = b;
333
334 // For a tailoring, the default is to fall back to the base.
335 // For ICU4X, use the same value for fallback as for the default
336 // to avoid having to have different blocks for the two.
337 trie = utrie2_open(Collation::FALLBACK_CE32, icu4xMode ? Collation::FALLBACK_CE32 : Collation::FFFD_CE32, &errorCode);
338
339 if (!icu4xMode) {
340 // Set the Latin-1 letters block so that it is allocated first in the data array,
341 // to try to improve locality of reference when sorting Latin-1 text.
342 // Do not use utrie2_setRange32() since that will not actually allocate blocks
343 // that are filled with the default value.
344 // ASCII (0..7F) is already preallocated anyway.
345 for(UChar32 c = 0xc0; c <= 0xff; ++c) {
346 utrie2_set32(trie, c, Collation::FALLBACK_CE32, &errorCode);
347 }
348
349 // Hangul syllables are not tailorable (except via tailoring Jamos).
350 // Always set the Hangul tag to help performance.
351 // Do this here, rather than in buildMappings(),
352 // so that we see the HANGUL_TAG in various assertions.
353 uint32_t hangulCE32 = Collation::makeCE32FromTagAndIndex(Collation::HANGUL_TAG, 0);
354 utrie2_setRange32(trie, Hangul::HANGUL_BASE, Hangul::HANGUL_END, hangulCE32, true, &errorCode);
355
356 // Copy the set contents but don't copy/clone the set as a whole because
357 // that would copy the isFrozen state too.
358 unsafeBackwardSet.addAll(*b->unsafeBackwardSet);
359 }
360
361 if(U_FAILURE(errorCode)) { return; }
362 }
363
364 UBool
maybeSetPrimaryRange(UChar32 start,UChar32 end,uint32_t primary,int32_t step,UErrorCode & errorCode)365 CollationDataBuilder::maybeSetPrimaryRange(UChar32 start, UChar32 end,
366 uint32_t primary, int32_t step,
367 UErrorCode &errorCode) {
368 if(U_FAILURE(errorCode)) { return false; }
369 U_ASSERT(start <= end);
370 // TODO: Do we need to check what values are currently set for start..end?
371 // An offset range is worth it only if we can achieve an overlap between
372 // adjacent UTrie2 blocks of 32 code points each.
373 // An offset CE is also a little more expensive to look up and compute
374 // than a simple CE.
375 // If the range spans at least three UTrie2 block boundaries (> 64 code points),
376 // then we take it.
377 // If the range spans one or two block boundaries and there are
378 // at least 4 code points on either side, then we take it.
379 // (We could additionally require a minimum range length of, say, 16.)
380 int32_t blockDelta = (end >> 5) - (start >> 5);
381 if(2 <= step && step <= 0x7f &&
382 (blockDelta >= 3 ||
383 (blockDelta > 0 && (start & 0x1f) <= 0x1c && (end & 0x1f) >= 3))) {
384 int64_t dataCE = ((int64_t)primary << 32) | (start << 8) | step;
385 if(isCompressiblePrimary(primary)) { dataCE |= 0x80; }
386 int32_t index = addCE(dataCE, errorCode);
387 if(U_FAILURE(errorCode)) { return 0; }
388 if(index > Collation::MAX_INDEX) {
389 errorCode = U_BUFFER_OVERFLOW_ERROR;
390 return 0;
391 }
392 uint32_t offsetCE32 = Collation::makeCE32FromTagAndIndex(Collation::OFFSET_TAG, index);
393 utrie2_setRange32(trie, start, end, offsetCE32, true, &errorCode);
394 modified = true;
395 return true;
396 } else {
397 return false;
398 }
399 }
400
401 uint32_t
setPrimaryRangeAndReturnNext(UChar32 start,UChar32 end,uint32_t primary,int32_t step,UErrorCode & errorCode)402 CollationDataBuilder::setPrimaryRangeAndReturnNext(UChar32 start, UChar32 end,
403 uint32_t primary, int32_t step,
404 UErrorCode &errorCode) {
405 if(U_FAILURE(errorCode)) { return 0; }
406 UBool isCompressible = isCompressiblePrimary(primary);
407 if(maybeSetPrimaryRange(start, end, primary, step, errorCode)) {
408 return Collation::incThreeBytePrimaryByOffset(primary, isCompressible,
409 (end - start + 1) * step);
410 } else {
411 // Short range: Set individual CE32s.
412 for(;;) {
413 utrie2_set32(trie, start, Collation::makeLongPrimaryCE32(primary), &errorCode);
414 ++start;
415 primary = Collation::incThreeBytePrimaryByOffset(primary, isCompressible, step);
416 if(start > end) { return primary; }
417 }
418 modified = true;
419 }
420 }
421
422 uint32_t
getCE32FromOffsetCE32(UBool fromBase,UChar32 c,uint32_t ce32) const423 CollationDataBuilder::getCE32FromOffsetCE32(UBool fromBase, UChar32 c, uint32_t ce32) const {
424 int32_t i = Collation::indexFromCE32(ce32);
425 int64_t dataCE = fromBase ? base->ces[i] : ce64s.elementAti(i);
426 uint32_t p = Collation::getThreeBytePrimaryForOffsetData(c, dataCE);
427 return Collation::makeLongPrimaryCE32(p);
428 }
429
430 UBool
isCompressibleLeadByte(uint32_t b) const431 CollationDataBuilder::isCompressibleLeadByte(uint32_t b) const {
432 return base->isCompressibleLeadByte(b);
433 }
434
435 UBool
isAssigned(UChar32 c) const436 CollationDataBuilder::isAssigned(UChar32 c) const {
437 return Collation::isAssignedCE32(utrie2_get32(trie, c));
438 }
439
440 uint32_t
getLongPrimaryIfSingleCE(UChar32 c) const441 CollationDataBuilder::getLongPrimaryIfSingleCE(UChar32 c) const {
442 uint32_t ce32 = utrie2_get32(trie, c);
443 if(Collation::isLongPrimaryCE32(ce32)) {
444 return Collation::primaryFromLongPrimaryCE32(ce32);
445 } else {
446 return 0;
447 }
448 }
449
450 int64_t
getSingleCE(UChar32 c,UErrorCode & errorCode) const451 CollationDataBuilder::getSingleCE(UChar32 c, UErrorCode &errorCode) const {
452 if(U_FAILURE(errorCode)) { return 0; }
453 // Keep parallel with CollationData::getSingleCE().
454 UBool fromBase = false;
455 uint32_t ce32 = utrie2_get32(trie, c);
456 if(ce32 == Collation::FALLBACK_CE32) {
457 fromBase = true;
458 ce32 = base->getCE32(c);
459 }
460 while(Collation::isSpecialCE32(ce32)) {
461 switch(Collation::tagFromCE32(ce32)) {
462 case Collation::LATIN_EXPANSION_TAG:
463 case Collation::BUILDER_DATA_TAG:
464 case Collation::PREFIX_TAG:
465 case Collation::CONTRACTION_TAG:
466 case Collation::HANGUL_TAG:
467 case Collation::LEAD_SURROGATE_TAG:
468 errorCode = U_UNSUPPORTED_ERROR;
469 return 0;
470 case Collation::FALLBACK_TAG:
471 case Collation::RESERVED_TAG_3:
472 errorCode = U_INTERNAL_PROGRAM_ERROR;
473 return 0;
474 case Collation::LONG_PRIMARY_TAG:
475 return Collation::ceFromLongPrimaryCE32(ce32);
476 case Collation::LONG_SECONDARY_TAG:
477 return Collation::ceFromLongSecondaryCE32(ce32);
478 case Collation::EXPANSION32_TAG:
479 if(Collation::lengthFromCE32(ce32) == 1) {
480 int32_t i = Collation::indexFromCE32(ce32);
481 ce32 = fromBase ? base->ce32s[i] : ce32s.elementAti(i);
482 break;
483 } else {
484 errorCode = U_UNSUPPORTED_ERROR;
485 return 0;
486 }
487 case Collation::EXPANSION_TAG: {
488 if(Collation::lengthFromCE32(ce32) == 1) {
489 int32_t i = Collation::indexFromCE32(ce32);
490 return fromBase ? base->ces[i] : ce64s.elementAti(i);
491 } else {
492 errorCode = U_UNSUPPORTED_ERROR;
493 return 0;
494 }
495 }
496 case Collation::DIGIT_TAG:
497 // Fetch the non-numeric-collation CE32 and continue.
498 ce32 = ce32s.elementAti(Collation::indexFromCE32(ce32));
499 break;
500 case Collation::U0000_TAG:
501 U_ASSERT(c == 0);
502 // Fetch the normal ce32 for U+0000 and continue.
503 ce32 = fromBase ? base->ce32s[0] : ce32s.elementAti(0);
504 break;
505 case Collation::OFFSET_TAG:
506 ce32 = getCE32FromOffsetCE32(fromBase, c, ce32);
507 break;
508 case Collation::IMPLICIT_TAG:
509 return Collation::unassignedCEFromCodePoint(c);
510 }
511 }
512 return Collation::ceFromSimpleCE32(ce32);
513 }
514
515 int32_t
addCE(int64_t ce,UErrorCode & errorCode)516 CollationDataBuilder::addCE(int64_t ce, UErrorCode &errorCode) {
517 int32_t length = ce64s.size();
518 for(int32_t i = 0; i < length; ++i) {
519 if(ce == ce64s.elementAti(i)) { return i; }
520 }
521 ce64s.addElement(ce, errorCode);
522 return length;
523 }
524
525 int32_t
addCE32(uint32_t ce32,UErrorCode & errorCode)526 CollationDataBuilder::addCE32(uint32_t ce32, UErrorCode &errorCode) {
527 int32_t length = ce32s.size();
528 for(int32_t i = 0; i < length; ++i) {
529 if(ce32 == (uint32_t)ce32s.elementAti(i)) { return i; }
530 }
531 ce32s.addElement((int32_t)ce32, errorCode);
532 return length;
533 }
534
535 int32_t
addConditionalCE32(const UnicodeString & context,uint32_t ce32,UErrorCode & errorCode)536 CollationDataBuilder::addConditionalCE32(const UnicodeString &context, uint32_t ce32,
537 UErrorCode &errorCode) {
538 if(U_FAILURE(errorCode)) { return -1; }
539 U_ASSERT(!context.isEmpty());
540 int32_t index = conditionalCE32s.size();
541 if(index > Collation::MAX_INDEX) {
542 errorCode = U_BUFFER_OVERFLOW_ERROR;
543 return -1;
544 }
545 LocalPointer<ConditionalCE32> cond(new ConditionalCE32(context, ce32), errorCode);
546 conditionalCE32s.adoptElement(cond.orphan(), errorCode);
547 if(U_FAILURE(errorCode)) {
548 return -1;
549 }
550 return index;
551 }
552
553 void
add(const UnicodeString & prefix,const UnicodeString & s,const int64_t ces[],int32_t cesLength,UErrorCode & errorCode)554 CollationDataBuilder::add(const UnicodeString &prefix, const UnicodeString &s,
555 const int64_t ces[], int32_t cesLength,
556 UErrorCode &errorCode) {
557 uint32_t ce32 = encodeCEs(ces, cesLength, errorCode);
558 addCE32(prefix, s, ce32, errorCode);
559 }
560
561 void
addCE32(const UnicodeString & prefix,const UnicodeString & s,uint32_t ce32,UErrorCode & errorCode)562 CollationDataBuilder::addCE32(const UnicodeString &prefix, const UnicodeString &s,
563 uint32_t ce32, UErrorCode &errorCode) {
564 if(U_FAILURE(errorCode)) { return; }
565 if(s.isEmpty()) {
566 errorCode = U_ILLEGAL_ARGUMENT_ERROR;
567 return;
568 }
569 if(trie == nullptr || utrie2_isFrozen(trie)) {
570 errorCode = U_INVALID_STATE_ERROR;
571 return;
572 }
573 UChar32 c = s.char32At(0);
574 int32_t cLength = U16_LENGTH(c);
575 uint32_t oldCE32 = utrie2_get32(trie, c);
576 UBool hasContext = !prefix.isEmpty() || s.length() > cLength;
577
578 if (icu4xMode) {
579 if (base && c >= 0x1100 && c < 0x1200) {
580 // Omit jamo tailorings.
581 // TODO(https://github.com/unicode-org/icu4x/issues/1941).
582 }
583 const Normalizer2* nfdNormalizer = Normalizer2::getNFDInstance(errorCode);
584 UnicodeString sInNfd;
585 nfdNormalizer->normalize(s, sInNfd, errorCode);
586 if (s != sInNfd) {
587 // s is not in NFD, so it cannot match in ICU4X, since ICU4X only
588 // does NFD lookups.
589 // Now check that we're only rejecting known cases.
590 if (s.length() == 2) {
591 char16_t second = s.charAt(1);
592 if (second == 0x0F73 || second == 0x0F75 || second == 0x0F81) {
593 // Second is a special decomposing Tibetan vowel sign.
594 // These also get added in the decomposed form, so ignoring
595 // this instance is OK.
596 return;
597 }
598 if (c == 0xFDD1 && second == 0xAC00) {
599 // This strange contraction exists in the root and
600 // doesn't have a decomposed counterpart there.
601 // This won't match in ICU4X anyway and is very strange:
602 // Unassigned Arabic presentation form contracting with
603 // the very first Hangul syllable. Let's ignore this
604 // explicitly.
605 return;
606 }
607 }
608 // Unknown case worth investigating if ever found.
609 errorCode = U_UNSUPPORTED_ERROR;
610 return;
611 }
612
613 if (!prefix.isEmpty()) {
614 UnicodeString prefixInNfd;
615 nfdNormalizer->normalize(prefix, prefixInNfd, errorCode);
616 if (prefix != prefixInNfd) {
617 errorCode = U_UNSUPPORTED_ERROR;
618 return;
619 }
620
621 int32_t count = prefix.countChar32();
622 if (count > 2) {
623 // Prefix too long for ICU4X.
624 errorCode = U_UNSUPPORTED_ERROR;
625 return;
626 }
627 UChar32 utf32[4];
628 int32_t len = prefix.toUTF32(utf32, 4, errorCode);
629 if (len != count) {
630 errorCode = U_INVALID_STATE_ERROR;
631 return;
632 }
633 UChar32 c = utf32[0];
634 if (u_getCombiningClass(c)) {
635 // Prefix must start with as starter for ICU4X.
636 errorCode = U_UNSUPPORTED_ERROR;
637 return;
638 }
639 // XXX: Korean searchjl has jamo in prefix, so commenting out this
640 // check for now. ICU4X currently ignores non-root jamo tables anyway.
641 // searchjl was added in
642 // https://unicode-org.atlassian.net/browse/CLDR-3560
643 // Contractions were changed to prefixes in
644 // https://unicode-org.atlassian.net/browse/CLDR-6546
645 //
646 // if ((c >= 0x1100 && c < 0x1200) || (c >= 0xAC00 && c < 0xD7A4)) {
647 // errorCode = U_UNSUPPORTED_ERROR;
648 // return;
649 // }
650 if ((len > 1) && !(utf32[1] == 0x3099 || utf32[1] == 0x309A)) {
651 // Second character in prefix, if present, must be a kana voicing mark for ICU4X.
652 errorCode = U_UNSUPPORTED_ERROR;
653 return;
654 }
655 }
656
657 if (s.length() > cLength) {
658 // Check that there's no modern Hangul in contractions.
659 for (int32_t i = 0; i < s.length(); ++i) {
660 char16_t c = s.charAt(i);
661 if ((c >= 0x1100 && c < 0x1100 + 19) || (c >= 0x1161 && c < 0x1161 + 21) || (c >= 0x11A7 && c < 0x11A7 + 28) || (c >= 0xAC00 && c < 0xD7A4)) {
662 errorCode = U_UNSUPPORTED_ERROR;
663 return;
664 }
665 }
666 }
667 }
668
669 if(oldCE32 == Collation::FALLBACK_CE32) {
670 // First tailoring for c.
671 // If c has contextual base mappings or if we add a contextual mapping,
672 // then copy the base mappings.
673 // Otherwise we just override the base mapping.
674 uint32_t baseCE32 = base->getFinalCE32(base->getCE32(c));
675 if(hasContext || Collation::ce32HasContext(baseCE32)) {
676 oldCE32 = copyFromBaseCE32(c, baseCE32, true, errorCode);
677 utrie2_set32(trie, c, oldCE32, &errorCode);
678 if(U_FAILURE(errorCode)) { return; }
679 }
680 }
681 if(!hasContext) {
682 // No prefix, no contraction.
683 if(!isBuilderContextCE32(oldCE32)) {
684 utrie2_set32(trie, c, ce32, &errorCode);
685 } else {
686 ConditionalCE32 *cond = getConditionalCE32ForCE32(oldCE32);
687 cond->builtCE32 = Collation::NO_CE32;
688 cond->ce32 = ce32;
689 }
690 } else {
691 ConditionalCE32 *cond;
692 if(!isBuilderContextCE32(oldCE32)) {
693 // Replace the simple oldCE32 with a builder context CE32
694 // pointing to a new ConditionalCE32 list head.
695 int32_t index = addConditionalCE32(UnicodeString((char16_t)0), oldCE32, errorCode);
696 if(U_FAILURE(errorCode)) { return; }
697 uint32_t contextCE32 = makeBuilderContextCE32(index);
698 utrie2_set32(trie, c, contextCE32, &errorCode);
699 contextChars.add(c);
700 cond = getConditionalCE32(index);
701 } else {
702 cond = getConditionalCE32ForCE32(oldCE32);
703 cond->builtCE32 = Collation::NO_CE32;
704 }
705 UnicodeString suffix(s, cLength);
706 UnicodeString context((char16_t)prefix.length());
707 context.append(prefix).append(suffix);
708 unsafeBackwardSet.addAll(suffix);
709 for(;;) {
710 // invariant: context > cond->context
711 int32_t next = cond->next;
712 if(next < 0) {
713 // Append a new ConditionalCE32 after cond.
714 int32_t index = addConditionalCE32(context, ce32, errorCode);
715 if(U_FAILURE(errorCode)) { return; }
716 cond->next = index;
717 break;
718 }
719 ConditionalCE32 *nextCond = getConditionalCE32(next);
720 int8_t cmp = context.compare(nextCond->context);
721 if(cmp < 0) {
722 // Insert a new ConditionalCE32 between cond and nextCond.
723 int32_t index = addConditionalCE32(context, ce32, errorCode);
724 if(U_FAILURE(errorCode)) { return; }
725 cond->next = index;
726 getConditionalCE32(index)->next = next;
727 break;
728 } else if(cmp == 0) {
729 // Same context as before, overwrite its ce32.
730 nextCond->ce32 = ce32;
731 break;
732 }
733 cond = nextCond;
734 }
735 }
736 modified = true;
737 }
738
739 uint32_t
encodeOneCEAsCE32(int64_t ce)740 CollationDataBuilder::encodeOneCEAsCE32(int64_t ce) {
741 uint32_t p = (uint32_t)(ce >> 32);
742 uint32_t lower32 = (uint32_t)ce;
743 uint32_t t = (uint32_t)(ce & 0xffff);
744 U_ASSERT((t & 0xc000) != 0xc000); // Impossible case bits 11 mark special CE32s.
745 if((ce & INT64_C(0xffff00ff00ff)) == 0) {
746 // normal form ppppsstt
747 return p | (lower32 >> 16) | (t >> 8);
748 } else if((ce & INT64_C(0xffffffffff)) == Collation::COMMON_SEC_AND_TER_CE) {
749 // long-primary form ppppppC1
750 return Collation::makeLongPrimaryCE32(p);
751 } else if(p == 0 && (t & 0xff) == 0) {
752 // long-secondary form ssssttC2
753 return Collation::makeLongSecondaryCE32(lower32);
754 }
755 return Collation::NO_CE32;
756 }
757
758 uint32_t
encodeOneCE(int64_t ce,UErrorCode & errorCode)759 CollationDataBuilder::encodeOneCE(int64_t ce, UErrorCode &errorCode) {
760 // Try to encode one CE as one CE32.
761 uint32_t ce32 = encodeOneCEAsCE32(ce);
762 if(ce32 != Collation::NO_CE32) { return ce32; }
763 int32_t index = addCE(ce, errorCode);
764 if(U_FAILURE(errorCode)) { return 0; }
765 if(index > Collation::MAX_INDEX) {
766 errorCode = U_BUFFER_OVERFLOW_ERROR;
767 return 0;
768 }
769 return Collation::makeCE32FromTagIndexAndLength(Collation::EXPANSION_TAG, index, 1);
770 }
771
772 uint32_t
encodeCEs(const int64_t ces[],int32_t cesLength,UErrorCode & errorCode)773 CollationDataBuilder::encodeCEs(const int64_t ces[], int32_t cesLength,
774 UErrorCode &errorCode) {
775 if(U_FAILURE(errorCode)) { return 0; }
776 if(cesLength < 0 || cesLength > Collation::MAX_EXPANSION_LENGTH) {
777 errorCode = U_ILLEGAL_ARGUMENT_ERROR;
778 return 0;
779 }
780 if(trie == nullptr || utrie2_isFrozen(trie)) {
781 errorCode = U_INVALID_STATE_ERROR;
782 return 0;
783 }
784 if(cesLength == 0) {
785 // Convenience: We cannot map to nothing, but we can map to a completely ignorable CE.
786 // Do this here so that callers need not do it.
787 return encodeOneCEAsCE32(0);
788 } else if(cesLength == 1) {
789 return encodeOneCE(ces[0], errorCode);
790 } else if(cesLength == 2 && !icu4xMode) {
791 // Try to encode two CEs as one CE32.
792 // Turn this off for ICU4X, because without the canonical closure
793 // these are so rare that it doesn't make sense to spend a branch
794 // on checking this tag when using the data.
795 int64_t ce0 = ces[0];
796 int64_t ce1 = ces[1];
797 uint32_t p0 = (uint32_t)(ce0 >> 32);
798 if((ce0 & INT64_C(0xffffffffff00ff)) == Collation::COMMON_SECONDARY_CE &&
799 (ce1 & INT64_C(0xffffffff00ffffff)) == Collation::COMMON_TERTIARY_CE &&
800 p0 != 0) {
801 // Latin mini expansion
802 return
803 p0 |
804 (((uint32_t)ce0 & 0xff00u) << 8) |
805 (uint32_t)(ce1 >> 16) |
806 Collation::SPECIAL_CE32_LOW_BYTE |
807 Collation::LATIN_EXPANSION_TAG;
808 }
809 }
810 // Try to encode two or more CEs as CE32s.
811 int32_t newCE32s[Collation::MAX_EXPANSION_LENGTH];
812 for(int32_t i = 0;; ++i) {
813 if(i == cesLength) {
814 return encodeExpansion32(newCE32s, cesLength, errorCode);
815 }
816 uint32_t ce32 = encodeOneCEAsCE32(ces[i]);
817 if(ce32 == Collation::NO_CE32) { break; }
818 newCE32s[i] = (int32_t)ce32;
819 }
820 return encodeExpansion(ces, cesLength, errorCode);
821 }
822
823 uint32_t
encodeExpansion(const int64_t ces[],int32_t length,UErrorCode & errorCode)824 CollationDataBuilder::encodeExpansion(const int64_t ces[], int32_t length, UErrorCode &errorCode) {
825 if(U_FAILURE(errorCode)) { return 0; }
826 // See if this sequence of CEs has already been stored.
827 int64_t first = ces[0];
828 int32_t ce64sMax = ce64s.size() - length;
829 for(int32_t i = 0; i <= ce64sMax; ++i) {
830 if(first == ce64s.elementAti(i)) {
831 if(i > Collation::MAX_INDEX) {
832 errorCode = U_BUFFER_OVERFLOW_ERROR;
833 return 0;
834 }
835 for(int32_t j = 1;; ++j) {
836 if(j == length) {
837 return Collation::makeCE32FromTagIndexAndLength(
838 Collation::EXPANSION_TAG, i, length);
839 }
840 if(ce64s.elementAti(i + j) != ces[j]) { break; }
841 }
842 }
843 }
844 // Store the new sequence.
845 int32_t i = ce64s.size();
846 if(i > Collation::MAX_INDEX) {
847 errorCode = U_BUFFER_OVERFLOW_ERROR;
848 return 0;
849 }
850 for(int32_t j = 0; j < length; ++j) {
851 ce64s.addElement(ces[j], errorCode);
852 }
853 return Collation::makeCE32FromTagIndexAndLength(Collation::EXPANSION_TAG, i, length);
854 }
855
856 uint32_t
encodeExpansion32(const int32_t newCE32s[],int32_t length,UErrorCode & errorCode)857 CollationDataBuilder::encodeExpansion32(const int32_t newCE32s[], int32_t length,
858 UErrorCode &errorCode) {
859 if(U_FAILURE(errorCode)) { return 0; }
860 // See if this sequence of CE32s has already been stored.
861 int32_t first = newCE32s[0];
862 int32_t ce32sMax = ce32s.size() - length;
863 for(int32_t i = 0; i <= ce32sMax; ++i) {
864 if(first == ce32s.elementAti(i)) {
865 if(i > Collation::MAX_INDEX) {
866 errorCode = U_BUFFER_OVERFLOW_ERROR;
867 return 0;
868 }
869 for(int32_t j = 1;; ++j) {
870 if(j == length) {
871 return Collation::makeCE32FromTagIndexAndLength(
872 Collation::EXPANSION32_TAG, i, length);
873 }
874 if(ce32s.elementAti(i + j) != newCE32s[j]) { break; }
875 }
876 }
877 }
878 // Store the new sequence.
879 int32_t i = ce32s.size();
880 if(i > Collation::MAX_INDEX) {
881 errorCode = U_BUFFER_OVERFLOW_ERROR;
882 return 0;
883 }
884 for(int32_t j = 0; j < length; ++j) {
885 ce32s.addElement(newCE32s[j], errorCode);
886 }
887 return Collation::makeCE32FromTagIndexAndLength(Collation::EXPANSION32_TAG, i, length);
888 }
889
890 uint32_t
copyFromBaseCE32(UChar32 c,uint32_t ce32,UBool withContext,UErrorCode & errorCode)891 CollationDataBuilder::copyFromBaseCE32(UChar32 c, uint32_t ce32, UBool withContext,
892 UErrorCode &errorCode) {
893 if(U_FAILURE(errorCode)) { return 0; }
894 if(!Collation::isSpecialCE32(ce32)) { return ce32; }
895 switch(Collation::tagFromCE32(ce32)) {
896 case Collation::LONG_PRIMARY_TAG:
897 case Collation::LONG_SECONDARY_TAG:
898 case Collation::LATIN_EXPANSION_TAG:
899 // copy as is
900 break;
901 case Collation::EXPANSION32_TAG: {
902 const uint32_t *baseCE32s = base->ce32s + Collation::indexFromCE32(ce32);
903 int32_t length = Collation::lengthFromCE32(ce32);
904 ce32 = encodeExpansion32(
905 reinterpret_cast<const int32_t *>(baseCE32s), length, errorCode);
906 break;
907 }
908 case Collation::EXPANSION_TAG: {
909 const int64_t *baseCEs = base->ces + Collation::indexFromCE32(ce32);
910 int32_t length = Collation::lengthFromCE32(ce32);
911 ce32 = encodeExpansion(baseCEs, length, errorCode);
912 break;
913 }
914 case Collation::PREFIX_TAG: {
915 // Flatten prefixes and nested suffixes (contractions)
916 // into a linear list of ConditionalCE32.
917 const char16_t *p = base->contexts + Collation::indexFromCE32(ce32);
918 ce32 = CollationData::readCE32(p); // Default if no prefix match.
919 if(!withContext) {
920 return copyFromBaseCE32(c, ce32, false, errorCode);
921 }
922 ConditionalCE32 head;
923 UnicodeString context((char16_t)0);
924 int32_t index;
925 if(Collation::isContractionCE32(ce32)) {
926 index = copyContractionsFromBaseCE32(context, c, ce32, &head, errorCode);
927 } else {
928 ce32 = copyFromBaseCE32(c, ce32, true, errorCode);
929 head.next = index = addConditionalCE32(context, ce32, errorCode);
930 }
931 if(U_FAILURE(errorCode)) { return 0; }
932 ConditionalCE32 *cond = getConditionalCE32(index); // the last ConditionalCE32 so far
933 UCharsTrie::Iterator prefixes(p + 2, 0, errorCode);
934 while(prefixes.next(errorCode)) {
935 context = prefixes.getString();
936 context.reverse();
937 context.insert(0, (char16_t)context.length());
938 ce32 = (uint32_t)prefixes.getValue();
939 if(Collation::isContractionCE32(ce32)) {
940 index = copyContractionsFromBaseCE32(context, c, ce32, cond, errorCode);
941 } else {
942 ce32 = copyFromBaseCE32(c, ce32, true, errorCode);
943 cond->next = index = addConditionalCE32(context, ce32, errorCode);
944 }
945 if(U_FAILURE(errorCode)) { return 0; }
946 cond = getConditionalCE32(index);
947 }
948 ce32 = makeBuilderContextCE32(head.next);
949 contextChars.add(c);
950 break;
951 }
952 case Collation::CONTRACTION_TAG: {
953 if(!withContext) {
954 const char16_t *p = base->contexts + Collation::indexFromCE32(ce32);
955 ce32 = CollationData::readCE32(p); // Default if no suffix match.
956 return copyFromBaseCE32(c, ce32, false, errorCode);
957 }
958 ConditionalCE32 head;
959 UnicodeString context((char16_t)0);
960 copyContractionsFromBaseCE32(context, c, ce32, &head, errorCode);
961 ce32 = makeBuilderContextCE32(head.next);
962 contextChars.add(c);
963 break;
964 }
965 case Collation::HANGUL_TAG:
966 errorCode = U_UNSUPPORTED_ERROR; // We forbid tailoring of Hangul syllables.
967 break;
968 case Collation::OFFSET_TAG:
969 ce32 = getCE32FromOffsetCE32(true, c, ce32);
970 break;
971 case Collation::IMPLICIT_TAG:
972 ce32 = encodeOneCE(Collation::unassignedCEFromCodePoint(c), errorCode);
973 break;
974 default:
975 UPRV_UNREACHABLE_EXIT; // require ce32 == base->getFinalCE32(ce32)
976 }
977 return ce32;
978 }
979
980 int32_t
copyContractionsFromBaseCE32(UnicodeString & context,UChar32 c,uint32_t ce32,ConditionalCE32 * cond,UErrorCode & errorCode)981 CollationDataBuilder::copyContractionsFromBaseCE32(UnicodeString &context, UChar32 c, uint32_t ce32,
982 ConditionalCE32 *cond, UErrorCode &errorCode) {
983 if(U_FAILURE(errorCode)) { return 0; }
984 const char16_t *p = base->contexts + Collation::indexFromCE32(ce32);
985 int32_t index;
986 if((ce32 & Collation::CONTRACT_SINGLE_CP_NO_MATCH) != 0) {
987 // No match on the single code point.
988 // We are underneath a prefix, and the default mapping is just
989 // a fallback to the mappings for a shorter prefix.
990 U_ASSERT(context.length() > 1);
991 index = -1;
992 } else {
993 ce32 = CollationData::readCE32(p); // Default if no suffix match.
994 U_ASSERT(!Collation::isContractionCE32(ce32));
995 ce32 = copyFromBaseCE32(c, ce32, true, errorCode);
996 cond->next = index = addConditionalCE32(context, ce32, errorCode);
997 if(U_FAILURE(errorCode)) { return 0; }
998 cond = getConditionalCE32(index);
999 }
1000
1001 int32_t suffixStart = context.length();
1002 UCharsTrie::Iterator suffixes(p + 2, 0, errorCode);
1003 while(suffixes.next(errorCode)) {
1004 context.append(suffixes.getString());
1005 ce32 = copyFromBaseCE32(c, (uint32_t)suffixes.getValue(), true, errorCode);
1006 cond->next = index = addConditionalCE32(context, ce32, errorCode);
1007 if(U_FAILURE(errorCode)) { return 0; }
1008 // No need to update the unsafeBackwardSet because the tailoring set
1009 // is already a copy of the base set.
1010 cond = getConditionalCE32(index);
1011 context.truncate(suffixStart);
1012 }
1013 U_ASSERT(index >= 0);
1014 return index;
1015 }
1016
1017 class CopyHelper {
1018 public:
CopyHelper(const CollationDataBuilder & s,CollationDataBuilder & d,const CollationDataBuilder::CEModifier & m,UErrorCode & initialErrorCode)1019 CopyHelper(const CollationDataBuilder &s, CollationDataBuilder &d,
1020 const CollationDataBuilder::CEModifier &m, UErrorCode &initialErrorCode)
1021 : src(s), dest(d), modifier(m),
1022 errorCode(initialErrorCode) {}
1023
copyRangeCE32(UChar32 start,UChar32 end,uint32_t ce32)1024 UBool copyRangeCE32(UChar32 start, UChar32 end, uint32_t ce32) {
1025 ce32 = copyCE32(ce32);
1026 utrie2_setRange32(dest.trie, start, end, ce32, true, &errorCode);
1027 if(CollationDataBuilder::isBuilderContextCE32(ce32)) {
1028 dest.contextChars.add(start, end);
1029 }
1030 return U_SUCCESS(errorCode);
1031 }
1032
copyCE32(uint32_t ce32)1033 uint32_t copyCE32(uint32_t ce32) {
1034 if(!Collation::isSpecialCE32(ce32)) {
1035 int64_t ce = modifier.modifyCE32(ce32);
1036 if(ce != Collation::NO_CE) {
1037 ce32 = dest.encodeOneCE(ce, errorCode);
1038 }
1039 } else {
1040 int32_t tag = Collation::tagFromCE32(ce32);
1041 if(tag == Collation::EXPANSION32_TAG) {
1042 const uint32_t *srcCE32s = reinterpret_cast<uint32_t *>(src.ce32s.getBuffer());
1043 srcCE32s += Collation::indexFromCE32(ce32);
1044 int32_t length = Collation::lengthFromCE32(ce32);
1045 // Inspect the source CE32s. Just copy them if none are modified.
1046 // Otherwise copy to modifiedCEs, with modifications.
1047 UBool isModified = false;
1048 for(int32_t i = 0; i < length; ++i) {
1049 ce32 = srcCE32s[i];
1050 int64_t ce;
1051 if(Collation::isSpecialCE32(ce32) ||
1052 (ce = modifier.modifyCE32(ce32)) == Collation::NO_CE) {
1053 if(isModified) {
1054 modifiedCEs[i] = Collation::ceFromCE32(ce32);
1055 }
1056 } else {
1057 if(!isModified) {
1058 for(int32_t j = 0; j < i; ++j) {
1059 modifiedCEs[j] = Collation::ceFromCE32(srcCE32s[j]);
1060 }
1061 isModified = true;
1062 }
1063 modifiedCEs[i] = ce;
1064 }
1065 }
1066 if(isModified) {
1067 ce32 = dest.encodeCEs(modifiedCEs, length, errorCode);
1068 } else {
1069 ce32 = dest.encodeExpansion32(
1070 reinterpret_cast<const int32_t *>(srcCE32s), length, errorCode);
1071 }
1072 } else if(tag == Collation::EXPANSION_TAG) {
1073 const int64_t *srcCEs = src.ce64s.getBuffer();
1074 srcCEs += Collation::indexFromCE32(ce32);
1075 int32_t length = Collation::lengthFromCE32(ce32);
1076 // Inspect the source CEs. Just copy them if none are modified.
1077 // Otherwise copy to modifiedCEs, with modifications.
1078 UBool isModified = false;
1079 for(int32_t i = 0; i < length; ++i) {
1080 int64_t srcCE = srcCEs[i];
1081 int64_t ce = modifier.modifyCE(srcCE);
1082 if(ce == Collation::NO_CE) {
1083 if(isModified) {
1084 modifiedCEs[i] = srcCE;
1085 }
1086 } else {
1087 if(!isModified) {
1088 for(int32_t j = 0; j < i; ++j) {
1089 modifiedCEs[j] = srcCEs[j];
1090 }
1091 isModified = true;
1092 }
1093 modifiedCEs[i] = ce;
1094 }
1095 }
1096 if(isModified) {
1097 ce32 = dest.encodeCEs(modifiedCEs, length, errorCode);
1098 } else {
1099 ce32 = dest.encodeExpansion(srcCEs, length, errorCode);
1100 }
1101 } else if(tag == Collation::BUILDER_DATA_TAG) {
1102 // Copy the list of ConditionalCE32.
1103 ConditionalCE32 *cond = src.getConditionalCE32ForCE32(ce32);
1104 U_ASSERT(!cond->hasContext());
1105 int32_t destIndex = dest.addConditionalCE32(
1106 cond->context, copyCE32(cond->ce32), errorCode);
1107 ce32 = CollationDataBuilder::makeBuilderContextCE32(destIndex);
1108 while(cond->next >= 0) {
1109 cond = src.getConditionalCE32(cond->next);
1110 ConditionalCE32 *prevDestCond = dest.getConditionalCE32(destIndex);
1111 destIndex = dest.addConditionalCE32(
1112 cond->context, copyCE32(cond->ce32), errorCode);
1113 int32_t suffixStart = cond->prefixLength() + 1;
1114 dest.unsafeBackwardSet.addAll(cond->context.tempSubString(suffixStart));
1115 prevDestCond->next = destIndex;
1116 }
1117 } else {
1118 // Just copy long CEs and Latin mini expansions (and other expected values) as is,
1119 // assuming that the modifier would not modify them.
1120 U_ASSERT(tag == Collation::LONG_PRIMARY_TAG ||
1121 tag == Collation::LONG_SECONDARY_TAG ||
1122 tag == Collation::LATIN_EXPANSION_TAG ||
1123 tag == Collation::HANGUL_TAG);
1124 }
1125 }
1126 return ce32;
1127 }
1128
1129 const CollationDataBuilder &src;
1130 CollationDataBuilder &dest;
1131 const CollationDataBuilder::CEModifier &modifier;
1132 int64_t modifiedCEs[Collation::MAX_EXPANSION_LENGTH];
1133 UErrorCode errorCode;
1134 };
1135
1136 U_CDECL_BEGIN
1137
1138 static UBool U_CALLCONV
enumRangeForCopy(const void * context,UChar32 start,UChar32 end,uint32_t value)1139 enumRangeForCopy(const void *context, UChar32 start, UChar32 end, uint32_t value) {
1140 return
1141 value == Collation::UNASSIGNED_CE32 || value == Collation::FALLBACK_CE32 ||
1142 ((CopyHelper *)context)->copyRangeCE32(start, end, value);
1143 }
1144
1145 U_CDECL_END
1146
1147 void
copyFrom(const CollationDataBuilder & src,const CEModifier & modifier,UErrorCode & errorCode)1148 CollationDataBuilder::copyFrom(const CollationDataBuilder &src, const CEModifier &modifier,
1149 UErrorCode &errorCode) {
1150 if(U_FAILURE(errorCode)) { return; }
1151 if(trie == nullptr || utrie2_isFrozen(trie)) {
1152 errorCode = U_INVALID_STATE_ERROR;
1153 return;
1154 }
1155 CopyHelper helper(src, *this, modifier, errorCode);
1156 utrie2_enum(src.trie, nullptr, enumRangeForCopy, &helper);
1157 errorCode = helper.errorCode;
1158 // Update the contextChars and the unsafeBackwardSet while copying,
1159 // in case a character had conditional mappings in the source builder
1160 // and they were removed later.
1161 modified |= src.modified;
1162 }
1163
1164 void
optimize(const UnicodeSet & set,UErrorCode & errorCode)1165 CollationDataBuilder::optimize(const UnicodeSet &set, UErrorCode &errorCode) {
1166 if(U_FAILURE(errorCode) || set.isEmpty()) { return; }
1167 UnicodeSetIterator iter(set);
1168 while(iter.next() && !iter.isString()) {
1169 UChar32 c = iter.getCodepoint();
1170 uint32_t ce32 = utrie2_get32(trie, c);
1171 if(ce32 == Collation::FALLBACK_CE32) {
1172 ce32 = base->getFinalCE32(base->getCE32(c));
1173 ce32 = copyFromBaseCE32(c, ce32, true, errorCode);
1174 utrie2_set32(trie, c, ce32, &errorCode);
1175 }
1176 }
1177 modified = true;
1178 }
1179
1180 void
suppressContractions(const UnicodeSet & set,UErrorCode & errorCode)1181 CollationDataBuilder::suppressContractions(const UnicodeSet &set, UErrorCode &errorCode) {
1182 if(U_FAILURE(errorCode) || set.isEmpty()) { return; }
1183 UnicodeSetIterator iter(set);
1184 while(iter.next() && !iter.isString()) {
1185 UChar32 c = iter.getCodepoint();
1186 uint32_t ce32 = utrie2_get32(trie, c);
1187 if(ce32 == Collation::FALLBACK_CE32) {
1188 ce32 = base->getFinalCE32(base->getCE32(c));
1189 if(Collation::ce32HasContext(ce32)) {
1190 ce32 = copyFromBaseCE32(c, ce32, false /* without context */, errorCode);
1191 utrie2_set32(trie, c, ce32, &errorCode);
1192 }
1193 } else if(isBuilderContextCE32(ce32)) {
1194 ce32 = getConditionalCE32ForCE32(ce32)->ce32;
1195 // Simply abandon the list of ConditionalCE32.
1196 // The caller will copy this builder in the end,
1197 // eliminating unreachable data.
1198 utrie2_set32(trie, c, ce32, &errorCode);
1199 contextChars.remove(c);
1200 }
1201 }
1202 modified = true;
1203 }
1204
1205 UBool
getJamoCE32s(uint32_t jamoCE32s[],UErrorCode & errorCode)1206 CollationDataBuilder::getJamoCE32s(uint32_t jamoCE32s[], UErrorCode &errorCode) {
1207 if(U_FAILURE(errorCode)) { return false; }
1208 UBool anyJamoAssigned = base == nullptr; // always set jamoCE32s in the base data
1209 UBool needToCopyFromBase = false;
1210 for(int32_t j = 0; j < CollationData::JAMO_CE32S_LENGTH; ++j) { // Count across Jamo types.
1211 UChar32 jamo = jamoCpFromIndex(j);
1212 UBool fromBase = false;
1213 uint32_t ce32 = utrie2_get32(trie, jamo);
1214 anyJamoAssigned |= Collation::isAssignedCE32(ce32);
1215 // TODO: Try to prevent [optimize [Jamo]] from counting as anyJamoAssigned.
1216 // (As of CLDR 24 [2013] the Korean tailoring does not optimize conjoining Jamo.)
1217 if(ce32 == Collation::FALLBACK_CE32) {
1218 fromBase = true;
1219 ce32 = base->getCE32(jamo);
1220 }
1221 if(Collation::isSpecialCE32(ce32)) {
1222 switch(Collation::tagFromCE32(ce32)) {
1223 case Collation::LONG_PRIMARY_TAG:
1224 case Collation::LONG_SECONDARY_TAG:
1225 case Collation::LATIN_EXPANSION_TAG:
1226 // Copy the ce32 as-is.
1227 break;
1228 case Collation::EXPANSION32_TAG:
1229 case Collation::EXPANSION_TAG:
1230 case Collation::PREFIX_TAG:
1231 case Collation::CONTRACTION_TAG:
1232 if(fromBase) {
1233 // Defer copying until we know if anyJamoAssigned.
1234 ce32 = Collation::FALLBACK_CE32;
1235 needToCopyFromBase = true;
1236 }
1237 break;
1238 case Collation::IMPLICIT_TAG:
1239 // An unassigned Jamo should only occur in tests with incomplete bases.
1240 U_ASSERT(fromBase);
1241 ce32 = Collation::FALLBACK_CE32;
1242 needToCopyFromBase = true;
1243 break;
1244 case Collation::OFFSET_TAG:
1245 ce32 = getCE32FromOffsetCE32(fromBase, jamo, ce32);
1246 break;
1247 case Collation::FALLBACK_TAG:
1248 case Collation::RESERVED_TAG_3:
1249 case Collation::BUILDER_DATA_TAG:
1250 case Collation::DIGIT_TAG:
1251 case Collation::U0000_TAG:
1252 case Collation::HANGUL_TAG:
1253 case Collation::LEAD_SURROGATE_TAG:
1254 errorCode = U_INTERNAL_PROGRAM_ERROR;
1255 return false;
1256 }
1257 }
1258 jamoCE32s[j] = ce32;
1259 }
1260 if(anyJamoAssigned && needToCopyFromBase) {
1261 for(int32_t j = 0; j < CollationData::JAMO_CE32S_LENGTH; ++j) {
1262 if(jamoCE32s[j] == Collation::FALLBACK_CE32) {
1263 UChar32 jamo = jamoCpFromIndex(j);
1264 jamoCE32s[j] = copyFromBaseCE32(jamo, base->getCE32(jamo),
1265 /*withContext=*/ true, errorCode);
1266 }
1267 }
1268 }
1269 return anyJamoAssigned && U_SUCCESS(errorCode);
1270 }
1271
1272 void
setDigitTags(UErrorCode & errorCode)1273 CollationDataBuilder::setDigitTags(UErrorCode &errorCode) {
1274 UnicodeSet digits(UNICODE_STRING_SIMPLE("[:Nd:]"), errorCode);
1275 if(U_FAILURE(errorCode)) { return; }
1276 UnicodeSetIterator iter(digits);
1277 while(iter.next()) {
1278 U_ASSERT(!iter.isString());
1279 UChar32 c = iter.getCodepoint();
1280 uint32_t ce32 = utrie2_get32(trie, c);
1281 if(ce32 != Collation::FALLBACK_CE32 && ce32 != Collation::UNASSIGNED_CE32) {
1282 int32_t index = addCE32(ce32, errorCode);
1283 if(U_FAILURE(errorCode)) { return; }
1284 if(index > Collation::MAX_INDEX) {
1285 errorCode = U_BUFFER_OVERFLOW_ERROR;
1286 return;
1287 }
1288 ce32 = Collation::makeCE32FromTagIndexAndLength(
1289 Collation::DIGIT_TAG, index, u_charDigitValue(c));
1290 utrie2_set32(trie, c, ce32, &errorCode);
1291 }
1292 }
1293 }
1294
1295 U_CDECL_BEGIN
1296
1297 static UBool U_CALLCONV
enumRangeLeadValue(const void * context,UChar32,UChar32,uint32_t value)1298 enumRangeLeadValue(const void *context, UChar32 /*start*/, UChar32 /*end*/, uint32_t value) {
1299 int32_t *pValue = (int32_t *)context;
1300 if(value == Collation::UNASSIGNED_CE32) {
1301 value = Collation::LEAD_ALL_UNASSIGNED;
1302 } else if(value == Collation::FALLBACK_CE32) {
1303 value = Collation::LEAD_ALL_FALLBACK;
1304 } else {
1305 *pValue = Collation::LEAD_MIXED;
1306 return false;
1307 }
1308 if(*pValue < 0) {
1309 *pValue = (int32_t)value;
1310 } else if(*pValue != (int32_t)value) {
1311 *pValue = Collation::LEAD_MIXED;
1312 return false;
1313 }
1314 return true;
1315 }
1316
1317 U_CDECL_END
1318
1319 void
setLeadSurrogates(UErrorCode & errorCode)1320 CollationDataBuilder::setLeadSurrogates(UErrorCode &errorCode) {
1321 for(char16_t lead = 0xd800; lead < 0xdc00; ++lead) {
1322 int32_t value = -1;
1323 utrie2_enumForLeadSurrogate(trie, lead, nullptr, enumRangeLeadValue, &value);
1324 utrie2_set32ForLeadSurrogateCodeUnit(
1325 trie, lead,
1326 Collation::makeCE32FromTagAndIndex(Collation::LEAD_SURROGATE_TAG, 0) | (uint32_t)value,
1327 &errorCode);
1328 }
1329 }
1330
1331 void
build(CollationData & data,UErrorCode & errorCode)1332 CollationDataBuilder::build(CollationData &data, UErrorCode &errorCode) {
1333 buildMappings(data, errorCode);
1334 if(base != nullptr) {
1335 data.numericPrimary = base->numericPrimary;
1336 data.compressibleBytes = base->compressibleBytes;
1337 data.numScripts = base->numScripts;
1338 data.scriptsIndex = base->scriptsIndex;
1339 data.scriptStarts = base->scriptStarts;
1340 data.scriptStartsLength = base->scriptStartsLength;
1341 }
1342 buildFastLatinTable(data, errorCode);
1343 }
1344
1345 void
buildMappings(CollationData & data,UErrorCode & errorCode)1346 CollationDataBuilder::buildMappings(CollationData &data, UErrorCode &errorCode) {
1347 if(U_FAILURE(errorCode)) { return; }
1348 if(trie == nullptr || utrie2_isFrozen(trie)) {
1349 errorCode = U_INVALID_STATE_ERROR;
1350 return;
1351 }
1352
1353 buildContexts(errorCode);
1354
1355 uint32_t jamoCE32s[CollationData::JAMO_CE32S_LENGTH];
1356 int32_t jamoIndex = -1;
1357 if(getJamoCE32s(jamoCE32s, errorCode)) {
1358 jamoIndex = ce32s.size();
1359 for(int32_t i = 0; i < CollationData::JAMO_CE32S_LENGTH; ++i) {
1360 ce32s.addElement((int32_t)jamoCE32s[i], errorCode);
1361 }
1362 // Small optimization: Use a bit in the Hangul ce32
1363 // to indicate that none of the Jamo CE32s are isSpecialCE32()
1364 // (as it should be in the root collator).
1365 // It allows CollationIterator to avoid recursive function calls and per-Jamo tests.
1366 // In order to still have good trie compression and keep this code simple,
1367 // we only set this flag if a whole block of 588 Hangul syllables starting with
1368 // a common leading consonant (Jamo L) has this property.
1369 UBool isAnyJamoVTSpecial = false;
1370 for(int32_t i = Hangul::JAMO_L_COUNT; i < CollationData::JAMO_CE32S_LENGTH; ++i) {
1371 if(Collation::isSpecialCE32(jamoCE32s[i])) {
1372 isAnyJamoVTSpecial = true;
1373 break;
1374 }
1375 }
1376 uint32_t hangulCE32 = Collation::makeCE32FromTagAndIndex(Collation::HANGUL_TAG, 0);
1377 UChar32 c = Hangul::HANGUL_BASE;
1378 for(int32_t i = 0; i < Hangul::JAMO_L_COUNT; ++i) { // iterate over the Jamo L
1379 uint32_t ce32 = hangulCE32;
1380 if(!isAnyJamoVTSpecial && !Collation::isSpecialCE32(jamoCE32s[i])) {
1381 ce32 |= Collation::HANGUL_NO_SPECIAL_JAMO;
1382 }
1383 UChar32 limit = c + Hangul::JAMO_VT_COUNT;
1384 utrie2_setRange32(trie, c, limit - 1, ce32, true, &errorCode);
1385 c = limit;
1386 }
1387 } else {
1388 // Copy the Hangul CE32s from the base in blocks per Jamo L,
1389 // assuming that HANGUL_NO_SPECIAL_JAMO is set or not set for whole blocks.
1390 for(UChar32 c = Hangul::HANGUL_BASE; c < Hangul::HANGUL_LIMIT;) {
1391 uint32_t ce32 = base->getCE32(c);
1392 U_ASSERT(Collation::hasCE32Tag(ce32, Collation::HANGUL_TAG));
1393 UChar32 limit = c + Hangul::JAMO_VT_COUNT;
1394 utrie2_setRange32(trie, c, limit - 1, ce32, true, &errorCode);
1395 c = limit;
1396 }
1397 }
1398
1399 setDigitTags(errorCode);
1400 setLeadSurrogates(errorCode);
1401
1402 if (!icu4xMode) {
1403 // For U+0000, move its normal ce32 into CE32s[0] and set U0000_TAG.
1404 ce32s.setElementAt((int32_t)utrie2_get32(trie, 0), 0);
1405 utrie2_set32(trie, 0, Collation::makeCE32FromTagAndIndex(Collation::U0000_TAG, 0), &errorCode);
1406 }
1407
1408 utrie2_freeze(trie, UTRIE2_32_VALUE_BITS, &errorCode);
1409 if(U_FAILURE(errorCode)) { return; }
1410
1411 // Mark each lead surrogate as "unsafe"
1412 // if any of its 1024 associated supplementary code points is "unsafe".
1413 UChar32 c = 0x10000;
1414 for(char16_t lead = 0xd800; lead < 0xdc00; ++lead, c += 0x400) {
1415 if(unsafeBackwardSet.containsSome(c, c + 0x3ff)) {
1416 unsafeBackwardSet.add(lead);
1417 }
1418 }
1419 unsafeBackwardSet.freeze();
1420
1421 data.trie = trie;
1422 data.ce32s = reinterpret_cast<const uint32_t *>(ce32s.getBuffer());
1423 data.ces = ce64s.getBuffer();
1424 data.contexts = contexts.getBuffer();
1425
1426 data.ce32sLength = ce32s.size();
1427 data.cesLength = ce64s.size();
1428 data.contextsLength = contexts.length();
1429
1430 data.base = base;
1431 if(jamoIndex >= 0) {
1432 data.jamoCE32s = data.ce32s + jamoIndex;
1433 } else {
1434 data.jamoCE32s = base->jamoCE32s;
1435 }
1436 data.unsafeBackwardSet = &unsafeBackwardSet;
1437 }
1438
1439 void
clearContexts()1440 CollationDataBuilder::clearContexts() {
1441 contexts.remove();
1442 // Incrementing the contexts build "era" invalidates all of the builtCE32
1443 // from before this clearContexts() call.
1444 // Simpler than finding and resetting all of those fields.
1445 ++contextsEra;
1446 }
1447
1448 void
buildContexts(UErrorCode & errorCode)1449 CollationDataBuilder::buildContexts(UErrorCode &errorCode) {
1450 if(U_FAILURE(errorCode)) { return; }
1451 // Ignore abandoned lists and the cached builtCE32,
1452 // and build all contexts from scratch.
1453 clearContexts();
1454 UnicodeSetIterator iter(contextChars);
1455 while(U_SUCCESS(errorCode) && iter.next()) {
1456 U_ASSERT(!iter.isString());
1457 UChar32 c = iter.getCodepoint();
1458 uint32_t ce32 = utrie2_get32(trie, c);
1459 if(!isBuilderContextCE32(ce32)) {
1460 // Impossible: No context data for c in contextChars.
1461 errorCode = U_INTERNAL_PROGRAM_ERROR;
1462 return;
1463 }
1464 ConditionalCE32 *cond = getConditionalCE32ForCE32(ce32);
1465 ce32 = buildContext(cond, errorCode);
1466 utrie2_set32(trie, c, ce32, &errorCode);
1467 }
1468 }
1469
1470 uint32_t
buildContext(ConditionalCE32 * head,UErrorCode & errorCode)1471 CollationDataBuilder::buildContext(ConditionalCE32 *head, UErrorCode &errorCode) {
1472 if(U_FAILURE(errorCode)) { return 0; }
1473 // The list head must have no context.
1474 U_ASSERT(!head->hasContext());
1475 // The list head must be followed by one or more nodes that all do have context.
1476 U_ASSERT(head->next >= 0);
1477 UCharsTrieBuilder prefixBuilder(errorCode);
1478 UCharsTrieBuilder contractionBuilder(errorCode);
1479 // This outer loop goes from each prefix to the next.
1480 // For each prefix it finds the one or more same-prefix entries (firstCond..lastCond).
1481 // If there are multiple suffixes for the same prefix,
1482 // then an inner loop builds a contraction trie for them.
1483 for(ConditionalCE32 *cond = head;; cond = getConditionalCE32(cond->next)) {
1484 if(U_FAILURE(errorCode)) { return 0; } // early out for memory allocation errors
1485 // After the list head, the prefix or suffix can be empty, but not both.
1486 U_ASSERT(cond == head || cond->hasContext());
1487 int32_t prefixLength = cond->prefixLength();
1488 UnicodeString prefix(cond->context, 0, prefixLength + 1);
1489 // Collect all contraction suffixes for one prefix.
1490 ConditionalCE32 *firstCond = cond;
1491 ConditionalCE32 *lastCond;
1492 do {
1493 lastCond = cond;
1494 // Clear the defaultCE32 fields as we go.
1495 // They are left over from building a previous version of this list of contexts.
1496 //
1497 // One of the code paths below may copy a preceding defaultCE32
1498 // into its emptySuffixCE32.
1499 // If a new suffix has been inserted before what used to be
1500 // the firstCond for its prefix, then that previous firstCond could still
1501 // contain an outdated defaultCE32 from an earlier buildContext() and
1502 // result in an incorrect emptySuffixCE32.
1503 // So we reset all defaultCE32 before reading and setting new values.
1504 cond->defaultCE32 = Collation::NO_CE32;
1505 } while(cond->next >= 0 &&
1506 (cond = getConditionalCE32(cond->next))->context.startsWith(prefix));
1507 uint32_t ce32;
1508 int32_t suffixStart = prefixLength + 1; // == prefix.length()
1509 if(lastCond->context.length() == suffixStart) {
1510 // One prefix without contraction suffix.
1511 U_ASSERT(firstCond == lastCond);
1512 ce32 = lastCond->ce32;
1513 cond = lastCond;
1514 } else {
1515 // Build the contractions trie.
1516 contractionBuilder.clear();
1517 // Entry for an empty suffix, to be stored before the trie.
1518 uint32_t emptySuffixCE32 = 0;
1519 uint32_t flags = 0;
1520 if(firstCond->context.length() == suffixStart) {
1521 // There is a mapping for the prefix and the single character c. (p|c)
1522 // If no other suffix matches, then we return this value.
1523 emptySuffixCE32 = firstCond->ce32;
1524 cond = getConditionalCE32(firstCond->next);
1525 } else {
1526 // There is no mapping for the prefix and just the single character.
1527 // (There is no p|c, only p|cd, p|ce etc.)
1528 flags |= Collation::CONTRACT_SINGLE_CP_NO_MATCH;
1529 // When the prefix matches but none of the prefix-specific suffixes,
1530 // then we fall back to the mappings with the next-longest prefix,
1531 // and ultimately to mappings with no prefix.
1532 // Each fallback might be another set of contractions.
1533 // For example, if there are mappings for ch, p|cd, p|ce, but not for p|c,
1534 // then in text "pch" we find the ch contraction.
1535 for(cond = head;; cond = getConditionalCE32(cond->next)) {
1536 int32_t length = cond->prefixLength();
1537 if(length == prefixLength) { break; }
1538 if(cond->defaultCE32 != Collation::NO_CE32 &&
1539 (length==0 || prefix.endsWith(cond->context, 1, length))) {
1540 emptySuffixCE32 = cond->defaultCE32;
1541 }
1542 }
1543 cond = firstCond;
1544 }
1545 // Optimization: Set a flag when
1546 // the first character of every contraction suffix has lccc!=0.
1547 // Short-circuits contraction matching when a normal letter follows.
1548 flags |= Collation::CONTRACT_NEXT_CCC;
1549 // Add all of the non-empty suffixes into the contraction trie.
1550 for(;;) {
1551 UnicodeString suffix(cond->context, suffixStart);
1552 uint16_t fcd16 = nfcImpl.getFCD16(suffix.char32At(0));
1553 if(fcd16 <= 0xff) {
1554 flags &= ~Collation::CONTRACT_NEXT_CCC;
1555 }
1556 fcd16 = nfcImpl.getFCD16(suffix.char32At(suffix.length() - 1));
1557 if(fcd16 > 0xff) {
1558 // The last suffix character has lccc!=0, allowing for discontiguous contractions.
1559 flags |= Collation::CONTRACT_TRAILING_CCC;
1560 }
1561 if (icu4xMode && (flags & Collation::CONTRACT_HAS_STARTER) == 0) {
1562 for (int32_t i = 0; i < suffix.length();) {
1563 UChar32 c = suffix.char32At(i);
1564 if (!u_getCombiningClass(c)) {
1565 flags |= Collation::CONTRACT_HAS_STARTER;
1566 break;
1567 }
1568 if (c > 0xFFFF) {
1569 i += 2;
1570 } else {
1571 ++i;
1572 }
1573 }
1574 }
1575 contractionBuilder.add(suffix, (int32_t)cond->ce32, errorCode);
1576 if(cond == lastCond) { break; }
1577 cond = getConditionalCE32(cond->next);
1578 }
1579 int32_t index = addContextTrie(emptySuffixCE32, contractionBuilder, errorCode);
1580 if(U_FAILURE(errorCode)) { return 0; }
1581 if(index > Collation::MAX_INDEX) {
1582 errorCode = U_BUFFER_OVERFLOW_ERROR;
1583 return 0;
1584 }
1585 ce32 = Collation::makeCE32FromTagAndIndex(Collation::CONTRACTION_TAG, index) | flags;
1586 }
1587 U_ASSERT(cond == lastCond);
1588 firstCond->defaultCE32 = ce32;
1589 if(prefixLength == 0) {
1590 if(cond->next < 0) {
1591 // No non-empty prefixes, only contractions.
1592 return ce32;
1593 }
1594 } else {
1595 prefix.remove(0, 1); // Remove the length unit.
1596 prefix.reverse();
1597 prefixBuilder.add(prefix, (int32_t)ce32, errorCode);
1598 if(cond->next < 0) { break; }
1599 }
1600 }
1601 U_ASSERT(head->defaultCE32 != Collation::NO_CE32);
1602 int32_t index = addContextTrie(head->defaultCE32, prefixBuilder, errorCode);
1603 if(U_FAILURE(errorCode)) { return 0; }
1604 if(index > Collation::MAX_INDEX) {
1605 errorCode = U_BUFFER_OVERFLOW_ERROR;
1606 return 0;
1607 }
1608 return Collation::makeCE32FromTagAndIndex(Collation::PREFIX_TAG, index);
1609 }
1610
1611 int32_t
addContextTrie(uint32_t defaultCE32,UCharsTrieBuilder & trieBuilder,UErrorCode & errorCode)1612 CollationDataBuilder::addContextTrie(uint32_t defaultCE32, UCharsTrieBuilder &trieBuilder,
1613 UErrorCode &errorCode) {
1614 UnicodeString context;
1615 context.append((char16_t)(defaultCE32 >> 16)).append((char16_t)defaultCE32);
1616 UnicodeString trieString;
1617 context.append(trieBuilder.buildUnicodeString(USTRINGTRIE_BUILD_SMALL, trieString, errorCode));
1618 if(U_FAILURE(errorCode)) { return -1; }
1619 int32_t index = contexts.indexOf(context);
1620 if(index < 0) {
1621 index = contexts.length();
1622 contexts.append(context);
1623 }
1624 return index;
1625 }
1626
1627 void
buildFastLatinTable(CollationData & data,UErrorCode & errorCode)1628 CollationDataBuilder::buildFastLatinTable(CollationData &data, UErrorCode &errorCode) {
1629 if(U_FAILURE(errorCode) || !fastLatinEnabled) { return; }
1630
1631 delete fastLatinBuilder;
1632 fastLatinBuilder = new CollationFastLatinBuilder(errorCode);
1633 if(fastLatinBuilder == nullptr) {
1634 errorCode = U_MEMORY_ALLOCATION_ERROR;
1635 return;
1636 }
1637 if(fastLatinBuilder->forData(data, errorCode)) {
1638 const uint16_t *table = fastLatinBuilder->getTable();
1639 int32_t length = fastLatinBuilder->lengthOfTable();
1640 if(base != nullptr && length == base->fastLatinTableLength &&
1641 uprv_memcmp(table, base->fastLatinTable, length * 2) == 0) {
1642 // Same fast Latin table as in the base, use that one instead.
1643 delete fastLatinBuilder;
1644 fastLatinBuilder = nullptr;
1645 table = base->fastLatinTable;
1646 }
1647 data.fastLatinTable = table;
1648 data.fastLatinTableLength = length;
1649 } else {
1650 delete fastLatinBuilder;
1651 fastLatinBuilder = nullptr;
1652 }
1653 }
1654
1655 int32_t
getCEs(const UnicodeString & s,int64_t ces[],int32_t cesLength)1656 CollationDataBuilder::getCEs(const UnicodeString &s, int64_t ces[], int32_t cesLength) {
1657 return getCEs(s, 0, ces, cesLength);
1658 }
1659
1660 int32_t
getCEs(const UnicodeString & prefix,const UnicodeString & s,int64_t ces[],int32_t cesLength)1661 CollationDataBuilder::getCEs(const UnicodeString &prefix, const UnicodeString &s,
1662 int64_t ces[], int32_t cesLength) {
1663 int32_t prefixLength = prefix.length();
1664 if(prefixLength == 0) {
1665 return getCEs(s, 0, ces, cesLength);
1666 } else {
1667 return getCEs(prefix + s, prefixLength, ces, cesLength);
1668 }
1669 }
1670
1671 int32_t
getCEs(const UnicodeString & s,int32_t start,int64_t ces[],int32_t cesLength)1672 CollationDataBuilder::getCEs(const UnicodeString &s, int32_t start,
1673 int64_t ces[], int32_t cesLength) {
1674 if(collIter == nullptr) {
1675 collIter = new DataBuilderCollationIterator(*this);
1676 if(collIter == nullptr) { return 0; }
1677 }
1678 return collIter->fetchCEs(s, start, ces, cesLength);
1679 }
1680
1681 U_NAMESPACE_END
1682
1683 #endif // !UCONFIG_NO_COLLATION
1684