michael@0: /*
michael@0:  *****************************************************************************
michael@0:  * Copyright (C) 1996-2011, International Business Machines Corporation and  *
michael@0:  * others. All Rights Reserved.                                              *
michael@0:  *****************************************************************************
michael@0:  */
michael@0: 
michael@0: #include "unicode/utypes.h"
michael@0: 
michael@0: #if !UCONFIG_NO_NORMALIZATION
michael@0: 
michael@0: #include "unicode/caniter.h"
michael@0: #include "unicode/normalizer2.h"
michael@0: #include "unicode/uchar.h"
michael@0: #include "unicode/uniset.h"
michael@0: #include "unicode/usetiter.h"
michael@0: #include "unicode/ustring.h"
michael@0: #include "unicode/utf16.h"
michael@0: #include "cmemory.h"
michael@0: #include "hash.h"
michael@0: #include "normalizer2impl.h"
michael@0: 
michael@0: /**
michael@0:  * This class allows one to iterate through all the strings that are canonically equivalent to a given
michael@0:  * string. For example, here are some sample results:
michael@0: Results for: {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA}
michael@0: 1: \u0041\u030A\u0064\u0307\u0327
michael@0:  = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA}
michael@0: 2: \u0041\u030A\u0064\u0327\u0307
michael@0:  = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D}{COMBINING CEDILLA}{COMBINING DOT ABOVE}
michael@0: 3: \u0041\u030A\u1E0B\u0327
michael@0:  = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D WITH DOT ABOVE}{COMBINING CEDILLA}
michael@0: 4: \u0041\u030A\u1E11\u0307
michael@0:  = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D WITH CEDILLA}{COMBINING DOT ABOVE}
michael@0: 5: \u00C5\u0064\u0307\u0327
michael@0:  = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA}
michael@0: 6: \u00C5\u0064\u0327\u0307
michael@0:  = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D}{COMBINING CEDILLA}{COMBINING DOT ABOVE}
michael@0: 7: \u00C5\u1E0B\u0327
michael@0:  = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D WITH DOT ABOVE}{COMBINING CEDILLA}
michael@0: 8: \u00C5\u1E11\u0307
michael@0:  = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D WITH CEDILLA}{COMBINING DOT ABOVE}
michael@0: 9: \u212B\u0064\u0307\u0327
michael@0:  = {ANGSTROM SIGN}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA}
michael@0: 10: \u212B\u0064\u0327\u0307
michael@0:  = {ANGSTROM SIGN}{LATIN SMALL LETTER D}{COMBINING CEDILLA}{COMBINING DOT ABOVE}
michael@0: 11: \u212B\u1E0B\u0327
michael@0:  = {ANGSTROM SIGN}{LATIN SMALL LETTER D WITH DOT ABOVE}{COMBINING CEDILLA}
michael@0: 12: \u212B\u1E11\u0307
michael@0:  = {ANGSTROM SIGN}{LATIN SMALL LETTER D WITH CEDILLA}{COMBINING DOT ABOVE}
michael@0:  *<br>Note: the code is intended for use with small strings, and is not suitable for larger ones,
michael@0:  * since it has not been optimized for that situation.
michael@0:  *@author M. Davis
michael@0:  *@draft
michael@0:  */
michael@0: 
michael@0: // public
michael@0: 
michael@0: U_NAMESPACE_BEGIN
michael@0: 
michael@0: // TODO: add boilerplate methods.
michael@0: 
michael@0: UOBJECT_DEFINE_RTTI_IMPLEMENTATION(CanonicalIterator)
michael@0: 
michael@0: /**
michael@0:  *@param source string to get results for
michael@0:  */
michael@0: CanonicalIterator::CanonicalIterator(const UnicodeString &sourceStr, UErrorCode &status) :
michael@0:     pieces(NULL),
michael@0:     pieces_length(0),
michael@0:     pieces_lengths(NULL),
michael@0:     current(NULL),
michael@0:     current_length(0),
michael@0:     nfd(*Normalizer2Factory::getNFDInstance(status)),
michael@0:     nfcImpl(*Normalizer2Factory::getNFCImpl(status))
michael@0: {
michael@0:     if(U_SUCCESS(status) && nfcImpl.ensureCanonIterData(status)) {
michael@0:       setSource(sourceStr, status);
michael@0:     }
michael@0: }
michael@0: 
michael@0: CanonicalIterator::~CanonicalIterator() {
michael@0:   cleanPieces();
michael@0: }
michael@0: 
michael@0: void CanonicalIterator::cleanPieces() {
michael@0:     int32_t i = 0;
michael@0:     if(pieces != NULL) {
michael@0:         for(i = 0; i < pieces_length; i++) {
michael@0:             if(pieces[i] != NULL) {
michael@0:                 delete[] pieces[i];
michael@0:             }
michael@0:         }
michael@0:         uprv_free(pieces);
michael@0:         pieces = NULL;
michael@0:         pieces_length = 0;
michael@0:     }
michael@0:     if(pieces_lengths != NULL) {
michael@0:         uprv_free(pieces_lengths);
michael@0:         pieces_lengths = NULL;
michael@0:     }
michael@0:     if(current != NULL) {
michael@0:         uprv_free(current);
michael@0:         current = NULL;
michael@0:         current_length = 0;
michael@0:     }
michael@0: }
michael@0: 
michael@0: /**
michael@0:  *@return gets the source: NOTE: it is the NFD form of source
michael@0:  */
michael@0: UnicodeString CanonicalIterator::getSource() {
michael@0:   return source;
michael@0: }
michael@0: 
michael@0: /**
michael@0:  * Resets the iterator so that one can start again from the beginning.
michael@0:  */
michael@0: void CanonicalIterator::reset() {
michael@0:     done = FALSE;
michael@0:     for (int i = 0; i < current_length; ++i) {
michael@0:         current[i] = 0;
michael@0:     }
michael@0: }
michael@0: 
michael@0: /**
michael@0:  *@return the next string that is canonically equivalent. The value null is returned when
michael@0:  * the iteration is done.
michael@0:  */
michael@0: UnicodeString CanonicalIterator::next() {
michael@0:     int32_t i = 0;
michael@0: 
michael@0:     if (done) {
michael@0:       buffer.setToBogus();
michael@0:       return buffer;
michael@0:     }
michael@0: 
michael@0:     // delete old contents
michael@0:     buffer.remove();
michael@0: 
michael@0:     // construct return value
michael@0: 
michael@0:     for (i = 0; i < pieces_length; ++i) {
michael@0:         buffer.append(pieces[i][current[i]]);
michael@0:     }
michael@0:     //String result = buffer.toString(); // not needed
michael@0: 
michael@0:     // find next value for next time
michael@0: 
michael@0:     for (i = current_length - 1; ; --i) {
michael@0:         if (i < 0) {
michael@0:             done = TRUE;
michael@0:             break;
michael@0:         }
michael@0:         current[i]++;
michael@0:         if (current[i] < pieces_lengths[i]) break; // got sequence
michael@0:         current[i] = 0;
michael@0:     }
michael@0:     return buffer;
michael@0: }
michael@0: 
michael@0: /**
michael@0:  *@param set the source string to iterate against. This allows the same iterator to be used
michael@0:  * while changing the source string, saving object creation.
michael@0:  */
michael@0: void CanonicalIterator::setSource(const UnicodeString &newSource, UErrorCode &status) {
michael@0:     int32_t list_length = 0;
michael@0:     UChar32 cp = 0;
michael@0:     int32_t start = 0;
michael@0:     int32_t i = 0;
michael@0:     UnicodeString *list = NULL;
michael@0: 
michael@0:     nfd.normalize(newSource, source, status);
michael@0:     if(U_FAILURE(status)) {
michael@0:       return;
michael@0:     }
michael@0:     done = FALSE;
michael@0: 
michael@0:     cleanPieces();
michael@0: 
michael@0:     // catch degenerate case
michael@0:     if (newSource.length() == 0) {
michael@0:         pieces = (UnicodeString **)uprv_malloc(sizeof(UnicodeString *));
michael@0:         pieces_lengths = (int32_t*)uprv_malloc(1 * sizeof(int32_t));
michael@0:         pieces_length = 1;
michael@0:         current = (int32_t*)uprv_malloc(1 * sizeof(int32_t));
michael@0:         current_length = 1;
michael@0:         if (pieces == NULL || pieces_lengths == NULL || current == NULL) {
michael@0:             status = U_MEMORY_ALLOCATION_ERROR;
michael@0:             goto CleanPartialInitialization;
michael@0:         }
michael@0:         current[0] = 0;
michael@0:         pieces[0] = new UnicodeString[1];
michael@0:         pieces_lengths[0] = 1;
michael@0:         if (pieces[0] == 0) {
michael@0:             status = U_MEMORY_ALLOCATION_ERROR;
michael@0:             goto CleanPartialInitialization;
michael@0:         }
michael@0:         return;
michael@0:     }
michael@0: 
michael@0: 
michael@0:     list = new UnicodeString[source.length()];
michael@0:     if (list == 0) {
michael@0:         status = U_MEMORY_ALLOCATION_ERROR;
michael@0:         goto CleanPartialInitialization;
michael@0:     }
michael@0: 
michael@0:     // i should initialy be the number of code units at the 
michael@0:     // start of the string
michael@0:     i = U16_LENGTH(source.char32At(0));
michael@0:     //int32_t i = 1;
michael@0:     // find the segments
michael@0:     // This code iterates through the source string and 
michael@0:     // extracts segments that end up on a codepoint that
michael@0:     // doesn't start any decompositions. (Analysis is done
michael@0:     // on the NFD form - see above).
michael@0:     for (; i < source.length(); i += U16_LENGTH(cp)) {
michael@0:         cp = source.char32At(i);
michael@0:         if (nfcImpl.isCanonSegmentStarter(cp)) {
michael@0:             source.extract(start, i-start, list[list_length++]); // add up to i
michael@0:             start = i;
michael@0:         }
michael@0:     }
michael@0:     source.extract(start, i-start, list[list_length++]); // add last one
michael@0: 
michael@0: 
michael@0:     // allocate the arrays, and find the strings that are CE to each segment
michael@0:     pieces = (UnicodeString **)uprv_malloc(list_length * sizeof(UnicodeString *));
michael@0:     pieces_length = list_length;
michael@0:     pieces_lengths = (int32_t*)uprv_malloc(list_length * sizeof(int32_t));
michael@0:     current = (int32_t*)uprv_malloc(list_length * sizeof(int32_t));
michael@0:     current_length = list_length;
michael@0:     if (pieces == NULL || pieces_lengths == NULL || current == NULL) {
michael@0:         status = U_MEMORY_ALLOCATION_ERROR;
michael@0:         goto CleanPartialInitialization;
michael@0:     }
michael@0: 
michael@0:     for (i = 0; i < current_length; i++) {
michael@0:         current[i] = 0;
michael@0:     }
michael@0:     // for each segment, get all the combinations that can produce 
michael@0:     // it after NFD normalization
michael@0:     for (i = 0; i < pieces_length; ++i) {
michael@0:         //if (PROGRESS) printf("SEGMENT\n");
michael@0:         pieces[i] = getEquivalents(list[i], pieces_lengths[i], status);
michael@0:     }
michael@0: 
michael@0:     delete[] list;
michael@0:     return;
michael@0: // Common section to cleanup all local variables and reset object variables.
michael@0: CleanPartialInitialization:
michael@0:     if (list != NULL) {
michael@0:         delete[] list;
michael@0:     }
michael@0:     cleanPieces();
michael@0: }
michael@0: 
michael@0: /**
michael@0:  * Dumb recursive implementation of permutation.
michael@0:  * TODO: optimize
michael@0:  * @param source the string to find permutations for
michael@0:  * @return the results in a set.
michael@0:  */
michael@0: void U_EXPORT2 CanonicalIterator::permute(UnicodeString &source, UBool skipZeros, Hashtable *result, UErrorCode &status) {
michael@0:     if(U_FAILURE(status)) {
michael@0:         return;
michael@0:     }
michael@0:     //if (PROGRESS) printf("Permute: %s\n", UToS(Tr(source)));
michael@0:     int32_t i = 0;
michael@0: 
michael@0:     // optimization:
michael@0:     // if zero or one character, just return a set with it
michael@0:     // we check for length < 2 to keep from counting code points all the time
michael@0:     if (source.length() <= 2 && source.countChar32() <= 1) {
michael@0:         UnicodeString *toPut = new UnicodeString(source);
michael@0:         /* test for NULL */
michael@0:         if (toPut == 0) {
michael@0:             status = U_MEMORY_ALLOCATION_ERROR;
michael@0:             return;
michael@0:         }
michael@0:         result->put(source, toPut, status);
michael@0:         return;
michael@0:     }
michael@0: 
michael@0:     // otherwise iterate through the string, and recursively permute all the other characters
michael@0:     UChar32 cp;
michael@0:     Hashtable subpermute(status);
michael@0:     if(U_FAILURE(status)) {
michael@0:         return;
michael@0:     }
michael@0:     subpermute.setValueDeleter(uprv_deleteUObject);
michael@0: 
michael@0:     for (i = 0; i < source.length(); i += U16_LENGTH(cp)) {
michael@0:         cp = source.char32At(i);
michael@0:         const UHashElement *ne = NULL;
michael@0:         int32_t el = -1;
michael@0:         UnicodeString subPermuteString = source;
michael@0: 
michael@0:         // optimization:
michael@0:         // if the character is canonical combining class zero,
michael@0:         // don't permute it
michael@0:         if (skipZeros && i != 0 && u_getCombiningClass(cp) == 0) {
michael@0:             //System.out.println("Skipping " + Utility.hex(UTF16.valueOf(source, i)));
michael@0:             continue;
michael@0:         }
michael@0: 
michael@0:         subpermute.removeAll();
michael@0: 
michael@0:         // see what the permutations of the characters before and after this one are
michael@0:         //Hashtable *subpermute = permute(source.substring(0,i) + source.substring(i + UTF16.getCharCount(cp)));
michael@0:         permute(subPermuteString.replace(i, U16_LENGTH(cp), NULL, 0), skipZeros, &subpermute, status);
michael@0:         /* Test for buffer overflows */
michael@0:         if(U_FAILURE(status)) {
michael@0:             return;
michael@0:         }
michael@0:         // The upper replace is destructive. The question is do we have to make a copy, or we don't care about the contents 
michael@0:         // of source at this point.
michael@0: 
michael@0:         // prefix this character to all of them
michael@0:         ne = subpermute.nextElement(el);
michael@0:         while (ne != NULL) {
michael@0:             UnicodeString *permRes = (UnicodeString *)(ne->value.pointer);
michael@0:             UnicodeString *chStr = new UnicodeString(cp);
michael@0:             //test for  NULL
michael@0:             if (chStr == NULL) {
michael@0:                 status = U_MEMORY_ALLOCATION_ERROR;
michael@0:                 return;
michael@0:             }
michael@0:             chStr->append(*permRes); //*((UnicodeString *)(ne->value.pointer));
michael@0:             //if (PROGRESS) printf("  Piece: %s\n", UToS(*chStr));
michael@0:             result->put(*chStr, chStr, status);
michael@0:             ne = subpermute.nextElement(el);
michael@0:         }
michael@0:     }
michael@0:     //return result;
michael@0: }
michael@0: 
michael@0: // privates
michael@0: 
michael@0: // we have a segment, in NFD. Find all the strings that are canonically equivalent to it.
michael@0: UnicodeString* CanonicalIterator::getEquivalents(const UnicodeString &segment, int32_t &result_len, UErrorCode &status) {
michael@0:     Hashtable result(status);
michael@0:     Hashtable permutations(status);
michael@0:     Hashtable basic(status);
michael@0:     if (U_FAILURE(status)) {
michael@0:         return 0;
michael@0:     }
michael@0:     result.setValueDeleter(uprv_deleteUObject);
michael@0:     permutations.setValueDeleter(uprv_deleteUObject);
michael@0:     basic.setValueDeleter(uprv_deleteUObject);
michael@0: 
michael@0:     UChar USeg[256];
michael@0:     int32_t segLen = segment.extract(USeg, 256, status);
michael@0:     getEquivalents2(&basic, USeg, segLen, status);
michael@0: 
michael@0:     // now get all the permutations
michael@0:     // add only the ones that are canonically equivalent
michael@0:     // TODO: optimize by not permuting any class zero.
michael@0: 
michael@0:     const UHashElement *ne = NULL;
michael@0:     int32_t el = -1;
michael@0:     //Iterator it = basic.iterator();
michael@0:     ne = basic.nextElement(el);
michael@0:     //while (it.hasNext())
michael@0:     while (ne != NULL) {
michael@0:         //String item = (String) it.next();
michael@0:         UnicodeString item = *((UnicodeString *)(ne->value.pointer));
michael@0: 
michael@0:         permutations.removeAll();
michael@0:         permute(item, CANITER_SKIP_ZEROES, &permutations, status);
michael@0:         const UHashElement *ne2 = NULL;
michael@0:         int32_t el2 = -1;
michael@0:         //Iterator it2 = permutations.iterator();
michael@0:         ne2 = permutations.nextElement(el2);
michael@0:         //while (it2.hasNext())
michael@0:         while (ne2 != NULL) {
michael@0:             //String possible = (String) it2.next();
michael@0:             //UnicodeString *possible = new UnicodeString(*((UnicodeString *)(ne2->value.pointer)));
michael@0:             UnicodeString possible(*((UnicodeString *)(ne2->value.pointer)));
michael@0:             UnicodeString attempt;
michael@0:             nfd.normalize(possible, attempt, status);
michael@0: 
michael@0:             // TODO: check if operator == is semanticaly the same as attempt.equals(segment)
michael@0:             if (attempt==segment) {
michael@0:                 //if (PROGRESS) printf("Adding Permutation: %s\n", UToS(Tr(*possible)));
michael@0:                 // TODO: use the hashtable just to catch duplicates - store strings directly (somehow).
michael@0:                 result.put(possible, new UnicodeString(possible), status); //add(possible);
michael@0:             } else {
michael@0:                 //if (PROGRESS) printf("-Skipping Permutation: %s\n", UToS(Tr(*possible)));
michael@0:             }
michael@0: 
michael@0:             ne2 = permutations.nextElement(el2);
michael@0:         }
michael@0:         ne = basic.nextElement(el);
michael@0:     }
michael@0: 
michael@0:     /* Test for buffer overflows */
michael@0:     if(U_FAILURE(status)) {
michael@0:         return 0;
michael@0:     }
michael@0:     // convert into a String[] to clean up storage
michael@0:     //String[] finalResult = new String[result.size()];
michael@0:     UnicodeString *finalResult = NULL;
michael@0:     int32_t resultCount;
michael@0:     if((resultCount = result.count())) {
michael@0:         finalResult = new UnicodeString[resultCount];
michael@0:         if (finalResult == 0) {
michael@0:             status = U_MEMORY_ALLOCATION_ERROR;
michael@0:             return NULL;
michael@0:         }
michael@0:     }
michael@0:     else {
michael@0:         status = U_ILLEGAL_ARGUMENT_ERROR;
michael@0:         return NULL;
michael@0:     }
michael@0:     //result.toArray(finalResult);
michael@0:     result_len = 0;
michael@0:     el = -1;
michael@0:     ne = result.nextElement(el);
michael@0:     while(ne != NULL) {
michael@0:         finalResult[result_len++] = *((UnicodeString *)(ne->value.pointer));
michael@0:         ne = result.nextElement(el);
michael@0:     }
michael@0: 
michael@0: 
michael@0:     return finalResult;
michael@0: }
michael@0: 
michael@0: Hashtable *CanonicalIterator::getEquivalents2(Hashtable *fillinResult, const UChar *segment, int32_t segLen, UErrorCode &status) {
michael@0: 
michael@0:     if (U_FAILURE(status)) {
michael@0:         return NULL;
michael@0:     }
michael@0: 
michael@0:     //if (PROGRESS) printf("Adding: %s\n", UToS(Tr(segment)));
michael@0: 
michael@0:     UnicodeString toPut(segment, segLen);
michael@0: 
michael@0:     fillinResult->put(toPut, new UnicodeString(toPut), status);
michael@0: 
michael@0:     UnicodeSet starts;
michael@0: 
michael@0:     // cycle through all the characters
michael@0:     UChar32 cp;
michael@0:     for (int32_t i = 0; i < segLen; i += U16_LENGTH(cp)) {
michael@0:         // see if any character is at the start of some decomposition
michael@0:         U16_GET(segment, 0, i, segLen, cp);
michael@0:         if (!nfcImpl.getCanonStartSet(cp, starts)) {
michael@0:             continue;
michael@0:         }
michael@0:         // if so, see which decompositions match
michael@0:         UnicodeSetIterator iter(starts);
michael@0:         while (iter.next()) {
michael@0:             UChar32 cp2 = iter.getCodepoint();
michael@0:             Hashtable remainder(status);
michael@0:             remainder.setValueDeleter(uprv_deleteUObject);
michael@0:             if (extract(&remainder, cp2, segment, segLen, i, status) == NULL) {
michael@0:                 continue;
michael@0:             }
michael@0: 
michael@0:             // there were some matches, so add all the possibilities to the set.
michael@0:             UnicodeString prefix(segment, i);
michael@0:             prefix += cp2;
michael@0: 
michael@0:             int32_t el = -1;
michael@0:             const UHashElement *ne = remainder.nextElement(el);
michael@0:             while (ne != NULL) {
michael@0:                 UnicodeString item = *((UnicodeString *)(ne->value.pointer));
michael@0:                 UnicodeString *toAdd = new UnicodeString(prefix);
michael@0:                 /* test for NULL */
michael@0:                 if (toAdd == 0) {
michael@0:                     status = U_MEMORY_ALLOCATION_ERROR;
michael@0:                     return NULL;
michael@0:                 }
michael@0:                 *toAdd += item;
michael@0:                 fillinResult->put(*toAdd, toAdd, status);
michael@0: 
michael@0:                 //if (PROGRESS) printf("Adding: %s\n", UToS(Tr(*toAdd)));
michael@0: 
michael@0:                 ne = remainder.nextElement(el);
michael@0:             }
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /* Test for buffer overflows */
michael@0:     if(U_FAILURE(status)) {
michael@0:         return NULL;
michael@0:     }
michael@0:     return fillinResult;
michael@0: }
michael@0: 
michael@0: /**
michael@0:  * See if the decomposition of cp2 is at segment starting at segmentPos 
michael@0:  * (with canonical rearrangment!)
michael@0:  * If so, take the remainder, and return the equivalents 
michael@0:  */
michael@0: Hashtable *CanonicalIterator::extract(Hashtable *fillinResult, UChar32 comp, const UChar *segment, int32_t segLen, int32_t segmentPos, UErrorCode &status) {
michael@0: //Hashtable *CanonicalIterator::extract(UChar32 comp, const UnicodeString &segment, int32_t segLen, int32_t segmentPos, UErrorCode &status) {
michael@0:     //if (PROGRESS) printf(" extract: %s, ", UToS(Tr(UnicodeString(comp))));
michael@0:     //if (PROGRESS) printf("%s, %i\n", UToS(Tr(segment)), segmentPos);
michael@0: 
michael@0:     if (U_FAILURE(status)) {
michael@0:         return NULL;
michael@0:     }
michael@0: 
michael@0:     UnicodeString temp(comp);
michael@0:     int32_t inputLen=temp.length();
michael@0:     UnicodeString decompString;
michael@0:     nfd.normalize(temp, decompString, status);
michael@0:     const UChar *decomp=decompString.getBuffer();
michael@0:     int32_t decompLen=decompString.length();
michael@0: 
michael@0:     // See if it matches the start of segment (at segmentPos)
michael@0:     UBool ok = FALSE;
michael@0:     UChar32 cp;
michael@0:     int32_t decompPos = 0;
michael@0:     UChar32 decompCp;
michael@0:     U16_NEXT(decomp, decompPos, decompLen, decompCp);
michael@0: 
michael@0:     int32_t i = segmentPos;
michael@0:     while(i < segLen) {
michael@0:         U16_NEXT(segment, i, segLen, cp);
michael@0: 
michael@0:         if (cp == decompCp) { // if equal, eat another cp from decomp
michael@0: 
michael@0:             //if (PROGRESS) printf("  matches: %s\n", UToS(Tr(UnicodeString(cp))));
michael@0: 
michael@0:             if (decompPos == decompLen) { // done, have all decomp characters!
michael@0:                 temp.append(segment+i, segLen-i);
michael@0:                 ok = TRUE;
michael@0:                 break;
michael@0:             }
michael@0:             U16_NEXT(decomp, decompPos, decompLen, decompCp);
michael@0:         } else {
michael@0:             //if (PROGRESS) printf("  buffer: %s\n", UToS(Tr(UnicodeString(cp))));
michael@0: 
michael@0:             // brute force approach
michael@0:             temp.append(cp);
michael@0: 
michael@0:             /* TODO: optimize
michael@0:             // since we know that the classes are monotonically increasing, after zero
michael@0:             // e.g. 0 5 7 9 0 3
michael@0:             // we can do an optimization
michael@0:             // there are only a few cases that work: zero, less, same, greater
michael@0:             // if both classes are the same, we fail
michael@0:             // if the decomp class < the segment class, we fail
michael@0: 
michael@0:             segClass = getClass(cp);
michael@0:             if (decompClass <= segClass) return null;
michael@0:             */
michael@0:         }
michael@0:     }
michael@0:     if (!ok)
michael@0:         return NULL; // we failed, characters left over
michael@0: 
michael@0:     //if (PROGRESS) printf("Matches\n");
michael@0: 
michael@0:     if (inputLen == temp.length()) {
michael@0:         fillinResult->put(UnicodeString(), new UnicodeString(), status);
michael@0:         return fillinResult; // succeed, but no remainder
michael@0:     }
michael@0: 
michael@0:     // brute force approach
michael@0:     // check to make sure result is canonically equivalent
michael@0:     UnicodeString trial;
michael@0:     nfd.normalize(temp, trial, status);
michael@0:     if(U_FAILURE(status) || trial.compare(segment+segmentPos, segLen - segmentPos) != 0) {
michael@0:         return NULL;
michael@0:     }
michael@0: 
michael@0:     return getEquivalents2(fillinResult, temp.getBuffer()+inputLen, temp.length()-inputLen, status);
michael@0: }
michael@0: 
michael@0: U_NAMESPACE_END
michael@0: 
michael@0: #endif /* #if !UCONFIG_NO_NORMALIZATION */