1 // Copyright (C) 2016 and later: Unicode, Inc. and others.
2 // License & terms of use: http://www.unicode.org/copyright.html
4 *****************************************************************************
5 * Copyright (C) 1996-2015, International Business Machines Corporation and
6 * others. All Rights Reserved.
7 *****************************************************************************
10 #include "unicode/utypes.h"
12 #if !UCONFIG_NO_NORMALIZATION
14 #include "unicode/caniter.h"
15 #include "unicode/normalizer2.h"
16 #include "unicode/uchar.h"
17 #include "unicode/uniset.h"
18 #include "unicode/usetiter.h"
19 #include "unicode/ustring.h"
20 #include "unicode/utf16.h"
23 #include "normalizer2impl.h"
26 * This class allows one to iterate through all the strings that are canonically equivalent to a given
27 * string. For example, here are some sample results:
28 Results for: {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA}
29 1: \u0041\u030A\u0064\u0307\u0327
30 = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA}
31 2: \u0041\u030A\u0064\u0327\u0307
32 = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D}{COMBINING CEDILLA}{COMBINING DOT ABOVE}
33 3: \u0041\u030A\u1E0B\u0327
34 = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D WITH DOT ABOVE}{COMBINING CEDILLA}
35 4: \u0041\u030A\u1E11\u0307
36 = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D WITH CEDILLA}{COMBINING DOT ABOVE}
37 5: \u00C5\u0064\u0307\u0327
38 = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA}
39 6: \u00C5\u0064\u0327\u0307
40 = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D}{COMBINING CEDILLA}{COMBINING DOT ABOVE}
42 = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D WITH DOT ABOVE}{COMBINING CEDILLA}
44 = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D WITH CEDILLA}{COMBINING DOT ABOVE}
45 9: \u212B\u0064\u0307\u0327
46 = {ANGSTROM SIGN}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA}
47 10: \u212B\u0064\u0327\u0307
48 = {ANGSTROM SIGN}{LATIN SMALL LETTER D}{COMBINING CEDILLA}{COMBINING DOT ABOVE}
49 11: \u212B\u1E0B\u0327
50 = {ANGSTROM SIGN}{LATIN SMALL LETTER D WITH DOT ABOVE}{COMBINING CEDILLA}
51 12: \u212B\u1E11\u0307
52 = {ANGSTROM SIGN}{LATIN SMALL LETTER D WITH CEDILLA}{COMBINING DOT ABOVE}
53 *<br>Note: the code is intended for use with small strings, and is not suitable for larger ones,
54 * since it has not been optimized for that situation.
63 // TODO: add boilerplate methods.
65 UOBJECT_DEFINE_RTTI_IMPLEMENTATION(CanonicalIterator)
68 *@param source string to get results for
70 CanonicalIterator::CanonicalIterator(const UnicodeString &sourceStr, UErrorCode &status) :
76 nfd(*Normalizer2::getNFDInstance(status)),
77 nfcImpl(*Normalizer2Factory::getNFCImpl(status))
79 if(U_SUCCESS(status) && nfcImpl.ensureCanonIterData(status)) {
80 setSource(sourceStr, status);
84 CanonicalIterator::~CanonicalIterator() {
88 void CanonicalIterator::cleanPieces() {
91 for(i = 0; i < pieces_length; i++) {
92 if(pieces[i] != NULL) {
100 if(pieces_lengths != NULL) {
101 uprv_free(pieces_lengths);
102 pieces_lengths = NULL;
104 if(current != NULL) {
112 *@return gets the source: NOTE: it is the NFD form of source
114 UnicodeString CanonicalIterator::getSource() {
119 * Resets the iterator so that one can start again from the beginning.
121 void CanonicalIterator::reset() {
123 for (int i = 0; i < current_length; ++i) {
129 *@return the next string that is canonically equivalent. The value null is returned when
130 * the iteration is done.
132 UnicodeString CanonicalIterator::next() {
140 // delete old contents
143 // construct return value
145 for (i = 0; i < pieces_length; ++i) {
146 buffer.append(pieces[i][current[i]]);
148 //String result = buffer.toString(); // not needed
150 // find next value for next time
152 for (i = current_length - 1; ; --i) {
158 if (current[i] < pieces_lengths[i]) break; // got sequence
165 *@param set the source string to iterate against. This allows the same iterator to be used
166 * while changing the source string, saving object creation.
168 void CanonicalIterator::setSource(const UnicodeString &newSource, UErrorCode &status) {
169 int32_t list_length = 0;
173 UnicodeString *list = NULL;
175 nfd.normalize(newSource, source, status);
176 if(U_FAILURE(status)) {
183 // catch degenerate case
184 if (newSource.length() == 0) {
185 pieces = (UnicodeString **)uprv_malloc(sizeof(UnicodeString *));
186 pieces_lengths = (int32_t*)uprv_malloc(1 * sizeof(int32_t));
188 current = (int32_t*)uprv_malloc(1 * sizeof(int32_t));
190 if (pieces == NULL || pieces_lengths == NULL || current == NULL) {
191 status = U_MEMORY_ALLOCATION_ERROR;
192 goto CleanPartialInitialization;
195 pieces[0] = new UnicodeString[1];
196 pieces_lengths[0] = 1;
197 if (pieces[0] == 0) {
198 status = U_MEMORY_ALLOCATION_ERROR;
199 goto CleanPartialInitialization;
205 list = new UnicodeString[source.length()];
207 status = U_MEMORY_ALLOCATION_ERROR;
208 goto CleanPartialInitialization;
211 // i should initialy be the number of code units at the
212 // start of the string
213 i = U16_LENGTH(source.char32At(0));
216 // This code iterates through the source string and
217 // extracts segments that end up on a codepoint that
218 // doesn't start any decompositions. (Analysis is done
219 // on the NFD form - see above).
220 for (; i < source.length(); i += U16_LENGTH(cp)) {
221 cp = source.char32At(i);
222 if (nfcImpl.isCanonSegmentStarter(cp)) {
223 source.extract(start, i-start, list[list_length++]); // add up to i
227 source.extract(start, i-start, list[list_length++]); // add last one
230 // allocate the arrays, and find the strings that are CE to each segment
231 pieces = (UnicodeString **)uprv_malloc(list_length * sizeof(UnicodeString *));
232 pieces_length = list_length;
233 pieces_lengths = (int32_t*)uprv_malloc(list_length * sizeof(int32_t));
234 current = (int32_t*)uprv_malloc(list_length * sizeof(int32_t));
235 current_length = list_length;
236 if (pieces == NULL || pieces_lengths == NULL || current == NULL) {
237 status = U_MEMORY_ALLOCATION_ERROR;
238 goto CleanPartialInitialization;
241 for (i = 0; i < current_length; i++) {
244 // for each segment, get all the combinations that can produce
245 // it after NFD normalization
246 for (i = 0; i < pieces_length; ++i) {
247 //if (PROGRESS) printf("SEGMENT\n");
248 pieces[i] = getEquivalents(list[i], pieces_lengths[i], status);
253 // Common section to cleanup all local variables and reset object variables.
254 CleanPartialInitialization:
262 * Dumb recursive implementation of permutation.
264 * @param source the string to find permutations for
265 * @return the results in a set.
267 void U_EXPORT2 CanonicalIterator::permute(UnicodeString &source, UBool skipZeros, Hashtable *result, UErrorCode &status) {
268 if(U_FAILURE(status)) {
271 //if (PROGRESS) printf("Permute: %s\n", UToS(Tr(source)));
275 // if zero or one character, just return a set with it
276 // we check for length < 2 to keep from counting code points all the time
277 if (source.length() <= 2 && source.countChar32() <= 1) {
278 UnicodeString *toPut = new UnicodeString(source);
281 status = U_MEMORY_ALLOCATION_ERROR;
284 result->put(source, toPut, status);
288 // otherwise iterate through the string, and recursively permute all the other characters
290 Hashtable subpermute(status);
291 if(U_FAILURE(status)) {
294 subpermute.setValueDeleter(uprv_deleteUObject);
296 for (i = 0; i < source.length(); i += U16_LENGTH(cp)) {
297 cp = source.char32At(i);
298 const UHashElement *ne = NULL;
299 int32_t el = UHASH_FIRST;
300 UnicodeString subPermuteString = source;
303 // if the character is canonical combining class zero,
305 if (skipZeros && i != 0 && u_getCombiningClass(cp) == 0) {
306 //System.out.println("Skipping " + Utility.hex(UTF16.valueOf(source, i)));
310 subpermute.removeAll();
312 // see what the permutations of the characters before and after this one are
313 //Hashtable *subpermute = permute(source.substring(0,i) + source.substring(i + UTF16.getCharCount(cp)));
314 permute(subPermuteString.replace(i, U16_LENGTH(cp), NULL, 0), skipZeros, &subpermute, status);
315 /* Test for buffer overflows */
316 if(U_FAILURE(status)) {
319 // The upper replace is destructive. The question is do we have to make a copy, or we don't care about the contents
320 // of source at this point.
322 // prefix this character to all of them
323 ne = subpermute.nextElement(el);
325 UnicodeString *permRes = (UnicodeString *)(ne->value.pointer);
326 UnicodeString *chStr = new UnicodeString(cp);
329 status = U_MEMORY_ALLOCATION_ERROR;
332 chStr->append(*permRes); //*((UnicodeString *)(ne->value.pointer));
333 //if (PROGRESS) printf(" Piece: %s\n", UToS(*chStr));
334 result->put(*chStr, chStr, status);
335 ne = subpermute.nextElement(el);
343 // we have a segment, in NFD. Find all the strings that are canonically equivalent to it.
344 UnicodeString* CanonicalIterator::getEquivalents(const UnicodeString &segment, int32_t &result_len, UErrorCode &status) {
345 Hashtable result(status);
346 Hashtable permutations(status);
347 Hashtable basic(status);
348 if (U_FAILURE(status)) {
351 result.setValueDeleter(uprv_deleteUObject);
352 permutations.setValueDeleter(uprv_deleteUObject);
353 basic.setValueDeleter(uprv_deleteUObject);
356 int32_t segLen = segment.extract(USeg, 256, status);
357 getEquivalents2(&basic, USeg, segLen, status);
359 // now get all the permutations
360 // add only the ones that are canonically equivalent
361 // TODO: optimize by not permuting any class zero.
363 const UHashElement *ne = NULL;
364 int32_t el = UHASH_FIRST;
365 //Iterator it = basic.iterator();
366 ne = basic.nextElement(el);
367 //while (it.hasNext())
369 //String item = (String) it.next();
370 UnicodeString item = *((UnicodeString *)(ne->value.pointer));
372 permutations.removeAll();
373 permute(item, CANITER_SKIP_ZEROES, &permutations, status);
374 const UHashElement *ne2 = NULL;
375 int32_t el2 = UHASH_FIRST;
376 //Iterator it2 = permutations.iterator();
377 ne2 = permutations.nextElement(el2);
378 //while (it2.hasNext())
379 while (ne2 != NULL) {
380 //String possible = (String) it2.next();
381 //UnicodeString *possible = new UnicodeString(*((UnicodeString *)(ne2->value.pointer)));
382 UnicodeString possible(*((UnicodeString *)(ne2->value.pointer)));
383 UnicodeString attempt;
384 nfd.normalize(possible, attempt, status);
386 // TODO: check if operator == is semanticaly the same as attempt.equals(segment)
387 if (attempt==segment) {
388 //if (PROGRESS) printf("Adding Permutation: %s\n", UToS(Tr(*possible)));
389 // TODO: use the hashtable just to catch duplicates - store strings directly (somehow).
390 result.put(possible, new UnicodeString(possible), status); //add(possible);
392 //if (PROGRESS) printf("-Skipping Permutation: %s\n", UToS(Tr(*possible)));
395 ne2 = permutations.nextElement(el2);
397 ne = basic.nextElement(el);
400 /* Test for buffer overflows */
401 if(U_FAILURE(status)) {
404 // convert into a String[] to clean up storage
405 //String[] finalResult = new String[result.size()];
406 UnicodeString *finalResult = NULL;
408 if((resultCount = result.count())) {
409 finalResult = new UnicodeString[resultCount];
410 if (finalResult == 0) {
411 status = U_MEMORY_ALLOCATION_ERROR;
416 status = U_ILLEGAL_ARGUMENT_ERROR;
419 //result.toArray(finalResult);
422 ne = result.nextElement(el);
424 finalResult[result_len++] = *((UnicodeString *)(ne->value.pointer));
425 ne = result.nextElement(el);
432 Hashtable *CanonicalIterator::getEquivalents2(Hashtable *fillinResult, const UChar *segment, int32_t segLen, UErrorCode &status) {
434 if (U_FAILURE(status)) {
438 //if (PROGRESS) printf("Adding: %s\n", UToS(Tr(segment)));
440 UnicodeString toPut(segment, segLen);
442 fillinResult->put(toPut, new UnicodeString(toPut), status);
446 // cycle through all the characters
448 for (int32_t i = 0; i < segLen; i += U16_LENGTH(cp)) {
449 // see if any character is at the start of some decomposition
450 U16_GET(segment, 0, i, segLen, cp);
451 if (!nfcImpl.getCanonStartSet(cp, starts)) {
454 // if so, see which decompositions match
455 UnicodeSetIterator iter(starts);
456 while (iter.next()) {
457 UChar32 cp2 = iter.getCodepoint();
458 Hashtable remainder(status);
459 remainder.setValueDeleter(uprv_deleteUObject);
460 if (extract(&remainder, cp2, segment, segLen, i, status) == NULL) {
464 // there were some matches, so add all the possibilities to the set.
465 UnicodeString prefix(segment, i);
468 int32_t el = UHASH_FIRST;
469 const UHashElement *ne = remainder.nextElement(el);
471 UnicodeString item = *((UnicodeString *)(ne->value.pointer));
472 UnicodeString *toAdd = new UnicodeString(prefix);
475 status = U_MEMORY_ALLOCATION_ERROR;
479 fillinResult->put(*toAdd, toAdd, status);
481 //if (PROGRESS) printf("Adding: %s\n", UToS(Tr(*toAdd)));
483 ne = remainder.nextElement(el);
488 /* Test for buffer overflows */
489 if(U_FAILURE(status)) {
496 * See if the decomposition of cp2 is at segment starting at segmentPos
497 * (with canonical rearrangment!)
498 * If so, take the remainder, and return the equivalents
500 Hashtable *CanonicalIterator::extract(Hashtable *fillinResult, UChar32 comp, const UChar *segment, int32_t segLen, int32_t segmentPos, UErrorCode &status) {
501 //Hashtable *CanonicalIterator::extract(UChar32 comp, const UnicodeString &segment, int32_t segLen, int32_t segmentPos, UErrorCode &status) {
502 //if (PROGRESS) printf(" extract: %s, ", UToS(Tr(UnicodeString(comp))));
503 //if (PROGRESS) printf("%s, %i\n", UToS(Tr(segment)), segmentPos);
505 if (U_FAILURE(status)) {
509 UnicodeString temp(comp);
510 int32_t inputLen=temp.length();
511 UnicodeString decompString;
512 nfd.normalize(temp, decompString, status);
513 if (U_FAILURE(status)) {
516 if (decompString.isBogus()) {
517 status = U_MEMORY_ALLOCATION_ERROR;
520 const UChar *decomp=decompString.getBuffer();
521 int32_t decompLen=decompString.length();
523 // See if it matches the start of segment (at segmentPos)
526 int32_t decompPos = 0;
528 U16_NEXT(decomp, decompPos, decompLen, decompCp);
530 int32_t i = segmentPos;
532 U16_NEXT(segment, i, segLen, cp);
534 if (cp == decompCp) { // if equal, eat another cp from decomp
536 //if (PROGRESS) printf(" matches: %s\n", UToS(Tr(UnicodeString(cp))));
538 if (decompPos == decompLen) { // done, have all decomp characters!
539 temp.append(segment+i, segLen-i);
543 U16_NEXT(decomp, decompPos, decompLen, decompCp);
545 //if (PROGRESS) printf(" buffer: %s\n", UToS(Tr(UnicodeString(cp))));
547 // brute force approach
551 // since we know that the classes are monotonically increasing, after zero
553 // we can do an optimization
554 // there are only a few cases that work: zero, less, same, greater
555 // if both classes are the same, we fail
556 // if the decomp class < the segment class, we fail
558 segClass = getClass(cp);
559 if (decompClass <= segClass) return null;
564 return NULL; // we failed, characters left over
566 //if (PROGRESS) printf("Matches\n");
568 if (inputLen == temp.length()) {
569 fillinResult->put(UnicodeString(), new UnicodeString(), status);
570 return fillinResult; // succeed, but no remainder
573 // brute force approach
574 // check to make sure result is canonically equivalent
576 nfd.normalize(temp, trial, status);
577 if(U_FAILURE(status) || trial.compare(segment+segmentPos, segLen - segmentPos) != 0) {
581 return getEquivalents2(fillinResult, temp.getBuffer()+inputLen, temp.length()-inputLen, status);
586 #endif /* #if !UCONFIG_NO_NORMALIZATION */