michael@0: /*
michael@0: *******************************************************************************
michael@0: *
michael@0: *   Copyright (C) 2009-2013, International Business Machines
michael@0: *   Corporation and others.  All Rights Reserved.
michael@0: *
michael@0: *******************************************************************************
michael@0: *   file name:  normalizer2impl.cpp
michael@0: *   encoding:   US-ASCII
michael@0: *   tab size:   8 (not used)
michael@0: *   indentation:4
michael@0: *
michael@0: *   created on: 2009nov22
michael@0: *   created by: Markus W. Scherer
michael@0: */
michael@0: 
michael@0: #include "unicode/utypes.h"
michael@0: 
michael@0: #if !UCONFIG_NO_NORMALIZATION
michael@0: 
michael@0: #include "unicode/normalizer2.h"
michael@0: #include "unicode/udata.h"
michael@0: #include "unicode/ustring.h"
michael@0: #include "unicode/utf16.h"
michael@0: #include "cmemory.h"
michael@0: #include "mutex.h"
michael@0: #include "normalizer2impl.h"
michael@0: #include "putilimp.h"
michael@0: #include "uassert.h"
michael@0: #include "uset_imp.h"
michael@0: #include "utrie2.h"
michael@0: #include "uvector.h"
michael@0: 
michael@0: U_NAMESPACE_BEGIN
michael@0: 
michael@0: // ReorderingBuffer -------------------------------------------------------- ***
michael@0: 
michael@0: UBool ReorderingBuffer::init(int32_t destCapacity, UErrorCode &errorCode) {
michael@0:     int32_t length=str.length();
michael@0:     start=str.getBuffer(destCapacity);
michael@0:     if(start==NULL) {
michael@0:         // getBuffer() already did str.setToBogus()
michael@0:         errorCode=U_MEMORY_ALLOCATION_ERROR;
michael@0:         return FALSE;
michael@0:     }
michael@0:     limit=start+length;
michael@0:     remainingCapacity=str.getCapacity()-length;
michael@0:     reorderStart=start;
michael@0:     if(start==limit) {
michael@0:         lastCC=0;
michael@0:     } else {
michael@0:         setIterator();
michael@0:         lastCC=previousCC();
michael@0:         // Set reorderStart after the last code point with cc<=1 if there is one.
michael@0:         if(lastCC>1) {
michael@0:             while(previousCC()>1) {}
michael@0:         }
michael@0:         reorderStart=codePointLimit;
michael@0:     }
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: UBool ReorderingBuffer::equals(const UChar *otherStart, const UChar *otherLimit) const {
michael@0:     int32_t length=(int32_t)(limit-start);
michael@0:     return
michael@0:         length==(int32_t)(otherLimit-otherStart) &&
michael@0:         0==u_memcmp(start, otherStart, length);
michael@0: }
michael@0: 
michael@0: UBool ReorderingBuffer::appendSupplementary(UChar32 c, uint8_t cc, UErrorCode &errorCode) {
michael@0:     if(remainingCapacity<2 && !resize(2, errorCode)) {
michael@0:         return FALSE;
michael@0:     }
michael@0:     if(lastCC<=cc || cc==0) {
michael@0:         limit[0]=U16_LEAD(c);
michael@0:         limit[1]=U16_TRAIL(c);
michael@0:         limit+=2;
michael@0:         lastCC=cc;
michael@0:         if(cc<=1) {
michael@0:             reorderStart=limit;
michael@0:         }
michael@0:     } else {
michael@0:         insert(c, cc);
michael@0:     }
michael@0:     remainingCapacity-=2;
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: UBool ReorderingBuffer::append(const UChar *s, int32_t length,
michael@0:                                uint8_t leadCC, uint8_t trailCC,
michael@0:                                UErrorCode &errorCode) {
michael@0:     if(length==0) {
michael@0:         return TRUE;
michael@0:     }
michael@0:     if(remainingCapacity<length && !resize(length, errorCode)) {
michael@0:         return FALSE;
michael@0:     }
michael@0:     remainingCapacity-=length;
michael@0:     if(lastCC<=leadCC || leadCC==0) {
michael@0:         if(trailCC<=1) {
michael@0:             reorderStart=limit+length;
michael@0:         } else if(leadCC<=1) {
michael@0:             reorderStart=limit+1;  // Ok if not a code point boundary.
michael@0:         }
michael@0:         const UChar *sLimit=s+length;
michael@0:         do { *limit++=*s++; } while(s!=sLimit);
michael@0:         lastCC=trailCC;
michael@0:     } else {
michael@0:         int32_t i=0;
michael@0:         UChar32 c;
michael@0:         U16_NEXT(s, i, length, c);
michael@0:         insert(c, leadCC);  // insert first code point
michael@0:         while(i<length) {
michael@0:             U16_NEXT(s, i, length, c);
michael@0:             if(i<length) {
michael@0:                 // s must be in NFD, otherwise we need to use getCC().
michael@0:                 leadCC=Normalizer2Impl::getCCFromYesOrMaybe(impl.getNorm16(c));
michael@0:             } else {
michael@0:                 leadCC=trailCC;
michael@0:             }
michael@0:             append(c, leadCC, errorCode);
michael@0:         }
michael@0:     }
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: UBool ReorderingBuffer::appendZeroCC(UChar32 c, UErrorCode &errorCode) {
michael@0:     int32_t cpLength=U16_LENGTH(c);
michael@0:     if(remainingCapacity<cpLength && !resize(cpLength, errorCode)) {
michael@0:         return FALSE;
michael@0:     }
michael@0:     remainingCapacity-=cpLength;
michael@0:     if(cpLength==1) {
michael@0:         *limit++=(UChar)c;
michael@0:     } else {
michael@0:         limit[0]=U16_LEAD(c);
michael@0:         limit[1]=U16_TRAIL(c);
michael@0:         limit+=2;
michael@0:     }
michael@0:     lastCC=0;
michael@0:     reorderStart=limit;
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: UBool ReorderingBuffer::appendZeroCC(const UChar *s, const UChar *sLimit, UErrorCode &errorCode) {
michael@0:     if(s==sLimit) {
michael@0:         return TRUE;
michael@0:     }
michael@0:     int32_t length=(int32_t)(sLimit-s);
michael@0:     if(remainingCapacity<length && !resize(length, errorCode)) {
michael@0:         return FALSE;
michael@0:     }
michael@0:     u_memcpy(limit, s, length);
michael@0:     limit+=length;
michael@0:     remainingCapacity-=length;
michael@0:     lastCC=0;
michael@0:     reorderStart=limit;
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: void ReorderingBuffer::remove() {
michael@0:     reorderStart=limit=start;
michael@0:     remainingCapacity=str.getCapacity();
michael@0:     lastCC=0;
michael@0: }
michael@0: 
michael@0: void ReorderingBuffer::removeSuffix(int32_t suffixLength) {
michael@0:     if(suffixLength<(limit-start)) {
michael@0:         limit-=suffixLength;
michael@0:         remainingCapacity+=suffixLength;
michael@0:     } else {
michael@0:         limit=start;
michael@0:         remainingCapacity=str.getCapacity();
michael@0:     }
michael@0:     lastCC=0;
michael@0:     reorderStart=limit;
michael@0: }
michael@0: 
michael@0: UBool ReorderingBuffer::resize(int32_t appendLength, UErrorCode &errorCode) {
michael@0:     int32_t reorderStartIndex=(int32_t)(reorderStart-start);
michael@0:     int32_t length=(int32_t)(limit-start);
michael@0:     str.releaseBuffer(length);
michael@0:     int32_t newCapacity=length+appendLength;
michael@0:     int32_t doubleCapacity=2*str.getCapacity();
michael@0:     if(newCapacity<doubleCapacity) {
michael@0:         newCapacity=doubleCapacity;
michael@0:     }
michael@0:     if(newCapacity<256) {
michael@0:         newCapacity=256;
michael@0:     }
michael@0:     start=str.getBuffer(newCapacity);
michael@0:     if(start==NULL) {
michael@0:         // getBuffer() already did str.setToBogus()
michael@0:         errorCode=U_MEMORY_ALLOCATION_ERROR;
michael@0:         return FALSE;
michael@0:     }
michael@0:     reorderStart=start+reorderStartIndex;
michael@0:     limit=start+length;
michael@0:     remainingCapacity=str.getCapacity()-length;
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: void ReorderingBuffer::skipPrevious() {
michael@0:     codePointLimit=codePointStart;
michael@0:     UChar c=*--codePointStart;
michael@0:     if(U16_IS_TRAIL(c) && start<codePointStart && U16_IS_LEAD(*(codePointStart-1))) {
michael@0:         --codePointStart;
michael@0:     }
michael@0: }
michael@0: 
michael@0: uint8_t ReorderingBuffer::previousCC() {
michael@0:     codePointLimit=codePointStart;
michael@0:     if(reorderStart>=codePointStart) {
michael@0:         return 0;
michael@0:     }
michael@0:     UChar32 c=*--codePointStart;
michael@0:     if(c<Normalizer2Impl::MIN_CCC_LCCC_CP) {
michael@0:         return 0;
michael@0:     }
michael@0: 
michael@0:     UChar c2;
michael@0:     if(U16_IS_TRAIL(c) && start<codePointStart && U16_IS_LEAD(c2=*(codePointStart-1))) {
michael@0:         --codePointStart;
michael@0:         c=U16_GET_SUPPLEMENTARY(c2, c);
michael@0:     }
michael@0:     return Normalizer2Impl::getCCFromYesOrMaybe(impl.getNorm16(c));
michael@0: }
michael@0: 
michael@0: // Inserts c somewhere before the last character.
michael@0: // Requires 0<cc<lastCC which implies reorderStart<limit.
michael@0: void ReorderingBuffer::insert(UChar32 c, uint8_t cc) {
michael@0:     for(setIterator(), skipPrevious(); previousCC()>cc;) {}
michael@0:     // insert c at codePointLimit, after the character with prevCC<=cc
michael@0:     UChar *q=limit;
michael@0:     UChar *r=limit+=U16_LENGTH(c);
michael@0:     do {
michael@0:         *--r=*--q;
michael@0:     } while(codePointLimit!=q);
michael@0:     writeCodePoint(q, c);
michael@0:     if(cc<=1) {
michael@0:         reorderStart=r;
michael@0:     }
michael@0: }
michael@0: 
michael@0: // Normalizer2Impl --------------------------------------------------------- ***
michael@0: 
michael@0: struct CanonIterData : public UMemory {
michael@0:     CanonIterData(UErrorCode &errorCode);
michael@0:     ~CanonIterData();
michael@0:     void addToStartSet(UChar32 origin, UChar32 decompLead, UErrorCode &errorCode);
michael@0:     UTrie2 *trie;
michael@0:     UVector canonStartSets;  // contains UnicodeSet *
michael@0: };
michael@0: 
michael@0: Normalizer2Impl::~Normalizer2Impl() {
michael@0:     udata_close(memory);
michael@0:     utrie2_close(normTrie);
michael@0:     delete fCanonIterData;
michael@0: }
michael@0: 
michael@0: UBool U_CALLCONV
michael@0: Normalizer2Impl::isAcceptable(void *context,
michael@0:                               const char * /* type */, const char * /*name*/,
michael@0:                               const UDataInfo *pInfo) {
michael@0:     if(
michael@0:         pInfo->size>=20 &&
michael@0:         pInfo->isBigEndian==U_IS_BIG_ENDIAN &&
michael@0:         pInfo->charsetFamily==U_CHARSET_FAMILY &&
michael@0:         pInfo->dataFormat[0]==0x4e &&    /* dataFormat="Nrm2" */
michael@0:         pInfo->dataFormat[1]==0x72 &&
michael@0:         pInfo->dataFormat[2]==0x6d &&
michael@0:         pInfo->dataFormat[3]==0x32 &&
michael@0:         pInfo->formatVersion[0]==2
michael@0:     ) {
michael@0:         Normalizer2Impl *me=(Normalizer2Impl *)context;
michael@0:         uprv_memcpy(me->dataVersion, pInfo->dataVersion, 4);
michael@0:         return TRUE;
michael@0:     } else {
michael@0:         return FALSE;
michael@0:     }
michael@0: }
michael@0: 
michael@0: void
michael@0: Normalizer2Impl::load(const char *packageName, const char *name, UErrorCode &errorCode) {
michael@0:     if(U_FAILURE(errorCode)) {
michael@0:         return;
michael@0:     }
michael@0:     memory=udata_openChoice(packageName, "nrm", name, isAcceptable, this, &errorCode);
michael@0:     if(U_FAILURE(errorCode)) {
michael@0:         return;
michael@0:     }
michael@0:     const uint8_t *inBytes=(const uint8_t *)udata_getMemory(memory);
michael@0:     const int32_t *inIndexes=(const int32_t *)inBytes;
michael@0:     int32_t indexesLength=inIndexes[IX_NORM_TRIE_OFFSET]/4;
michael@0:     if(indexesLength<=IX_MIN_MAYBE_YES) {
michael@0:         errorCode=U_INVALID_FORMAT_ERROR;  // Not enough indexes.
michael@0:         return;
michael@0:     }
michael@0: 
michael@0:     minDecompNoCP=inIndexes[IX_MIN_DECOMP_NO_CP];
michael@0:     minCompNoMaybeCP=inIndexes[IX_MIN_COMP_NO_MAYBE_CP];
michael@0: 
michael@0:     minYesNo=inIndexes[IX_MIN_YES_NO];
michael@0:     minYesNoMappingsOnly=inIndexes[IX_MIN_YES_NO_MAPPINGS_ONLY];
michael@0:     minNoNo=inIndexes[IX_MIN_NO_NO];
michael@0:     limitNoNo=inIndexes[IX_LIMIT_NO_NO];
michael@0:     minMaybeYes=inIndexes[IX_MIN_MAYBE_YES];
michael@0: 
michael@0:     int32_t offset=inIndexes[IX_NORM_TRIE_OFFSET];
michael@0:     int32_t nextOffset=inIndexes[IX_EXTRA_DATA_OFFSET];
michael@0:     normTrie=utrie2_openFromSerialized(UTRIE2_16_VALUE_BITS,
michael@0:                                        inBytes+offset, nextOffset-offset, NULL,
michael@0:                                        &errorCode);
michael@0:     if(U_FAILURE(errorCode)) {
michael@0:         return;
michael@0:     }
michael@0: 
michael@0:     offset=nextOffset;
michael@0:     nextOffset=inIndexes[IX_SMALL_FCD_OFFSET];
michael@0:     maybeYesCompositions=(const uint16_t *)(inBytes+offset);
michael@0:     extraData=maybeYesCompositions+(MIN_NORMAL_MAYBE_YES-minMaybeYes);
michael@0: 
michael@0:     // smallFCD: new in formatVersion 2
michael@0:     offset=nextOffset;
michael@0:     smallFCD=inBytes+offset;
michael@0: 
michael@0:     // Build tccc180[].
michael@0:     // gennorm2 enforces lccc=0 for c<MIN_CCC_LCCC_CP=U+0300.
michael@0:     uint8_t bits=0;
michael@0:     for(UChar c=0; c<0x180; bits>>=1) {
michael@0:         if((c&0xff)==0) {
michael@0:             bits=smallFCD[c>>8];  // one byte per 0x100 code points
michael@0:         }
michael@0:         if(bits&1) {
michael@0:             for(int i=0; i<0x20; ++i, ++c) {
michael@0:                 tccc180[c]=(uint8_t)getFCD16FromNormData(c);
michael@0:             }
michael@0:         } else {
michael@0:             uprv_memset(tccc180+c, 0, 0x20);
michael@0:             c+=0x20;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: uint8_t Normalizer2Impl::getTrailCCFromCompYesAndZeroCC(const UChar *cpStart, const UChar *cpLimit) const {
michael@0:     UChar32 c;
michael@0:     if(cpStart==(cpLimit-1)) {
michael@0:         c=*cpStart;
michael@0:     } else {
michael@0:         c=U16_GET_SUPPLEMENTARY(cpStart[0], cpStart[1]);
michael@0:     }
michael@0:     uint16_t prevNorm16=getNorm16(c);
michael@0:     if(prevNorm16<=minYesNo) {
michael@0:         return 0;  // yesYes and Hangul LV/LVT have ccc=tccc=0
michael@0:     } else {
michael@0:         return (uint8_t)(*getMapping(prevNorm16)>>8);  // tccc from yesNo
michael@0:     }
michael@0: }
michael@0: 
michael@0: U_CDECL_BEGIN
michael@0: 
michael@0: static UBool U_CALLCONV
michael@0: enumPropertyStartsRange(const void *context, UChar32 start, UChar32 /*end*/, uint32_t /*value*/) {
michael@0:     /* add the start code point to the USet */
michael@0:     const USetAdder *sa=(const USetAdder *)context;
michael@0:     sa->add(sa->set, start);
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: static uint32_t U_CALLCONV
michael@0: segmentStarterMapper(const void * /*context*/, uint32_t value) {
michael@0:     return value&CANON_NOT_SEGMENT_STARTER;
michael@0: }
michael@0: 
michael@0: U_CDECL_END
michael@0: 
michael@0: void
michael@0: Normalizer2Impl::addPropertyStarts(const USetAdder *sa, UErrorCode & /*errorCode*/) const {
michael@0:     /* add the start code point of each same-value range of each trie */
michael@0:     utrie2_enum(normTrie, NULL, enumPropertyStartsRange, sa);
michael@0: 
michael@0:     /* add Hangul LV syllables and LV+1 because of skippables */
michael@0:     for(UChar c=Hangul::HANGUL_BASE; c<Hangul::HANGUL_LIMIT; c+=Hangul::JAMO_T_COUNT) {
michael@0:         sa->add(sa->set, c);
michael@0:         sa->add(sa->set, c+1);
michael@0:     }
michael@0:     sa->add(sa->set, Hangul::HANGUL_LIMIT); /* add Hangul+1 to continue with other properties */
michael@0: }
michael@0: 
michael@0: void
michael@0: Normalizer2Impl::addCanonIterPropertyStarts(const USetAdder *sa, UErrorCode &errorCode) const {
michael@0:     /* add the start code point of each same-value range of the canonical iterator data trie */
michael@0:     if(ensureCanonIterData(errorCode)) {
michael@0:         // currently only used for the SEGMENT_STARTER property
michael@0:         utrie2_enum(fCanonIterData->trie, segmentStarterMapper, enumPropertyStartsRange, sa);
michael@0:     }
michael@0: }
michael@0: 
michael@0: const UChar *
michael@0: Normalizer2Impl::copyLowPrefixFromNulTerminated(const UChar *src,
michael@0:                                                 UChar32 minNeedDataCP,
michael@0:                                                 ReorderingBuffer *buffer,
michael@0:                                                 UErrorCode &errorCode) const {
michael@0:     // Make some effort to support NUL-terminated strings reasonably.
michael@0:     // Take the part of the fast quick check loop that does not look up
michael@0:     // data and check the first part of the string.
michael@0:     // After this prefix, determine the string length to simplify the rest
michael@0:     // of the code.
michael@0:     const UChar *prevSrc=src;
michael@0:     UChar c;
michael@0:     while((c=*src++)<minNeedDataCP && c!=0) {}
michael@0:     // Back out the last character for full processing.
michael@0:     // Copy this prefix.
michael@0:     if(--src!=prevSrc) {
michael@0:         if(buffer!=NULL) {
michael@0:             buffer->appendZeroCC(prevSrc, src, errorCode);
michael@0:         }
michael@0:     }
michael@0:     return src;
michael@0: }
michael@0: 
michael@0: // Dual functionality:
michael@0: // buffer!=NULL: normalize
michael@0: // buffer==NULL: isNormalized/spanQuickCheckYes
michael@0: const UChar *
michael@0: Normalizer2Impl::decompose(const UChar *src, const UChar *limit,
michael@0:                            ReorderingBuffer *buffer,
michael@0:                            UErrorCode &errorCode) const {
michael@0:     UChar32 minNoCP=minDecompNoCP;
michael@0:     if(limit==NULL) {
michael@0:         src=copyLowPrefixFromNulTerminated(src, minNoCP, buffer, errorCode);
michael@0:         if(U_FAILURE(errorCode)) {
michael@0:             return src;
michael@0:         }
michael@0:         limit=u_strchr(src, 0);
michael@0:     }
michael@0: 
michael@0:     const UChar *prevSrc;
michael@0:     UChar32 c=0;
michael@0:     uint16_t norm16=0;
michael@0: 
michael@0:     // only for quick check
michael@0:     const UChar *prevBoundary=src;
michael@0:     uint8_t prevCC=0;
michael@0: 
michael@0:     for(;;) {
michael@0:         // count code units below the minimum or with irrelevant data for the quick check
michael@0:         for(prevSrc=src; src!=limit;) {
michael@0:             if( (c=*src)<minNoCP ||
michael@0:                 isMostDecompYesAndZeroCC(norm16=UTRIE2_GET16_FROM_U16_SINGLE_LEAD(normTrie, c))
michael@0:             ) {
michael@0:                 ++src;
michael@0:             } else if(!U16_IS_SURROGATE(c)) {
michael@0:                 break;
michael@0:             } else {
michael@0:                 UChar c2;
michael@0:                 if(U16_IS_SURROGATE_LEAD(c)) {
michael@0:                     if((src+1)!=limit && U16_IS_TRAIL(c2=src[1])) {
michael@0:                         c=U16_GET_SUPPLEMENTARY(c, c2);
michael@0:                     }
michael@0:                 } else /* trail surrogate */ {
michael@0:                     if(prevSrc<src && U16_IS_LEAD(c2=*(src-1))) {
michael@0:                         --src;
michael@0:                         c=U16_GET_SUPPLEMENTARY(c2, c);
michael@0:                     }
michael@0:                 }
michael@0:                 if(isMostDecompYesAndZeroCC(norm16=getNorm16(c))) {
michael@0:                     src+=U16_LENGTH(c);
michael@0:                 } else {
michael@0:                     break;
michael@0:                 }
michael@0:             }
michael@0:         }
michael@0:         // copy these code units all at once
michael@0:         if(src!=prevSrc) {
michael@0:             if(buffer!=NULL) {
michael@0:                 if(!buffer->appendZeroCC(prevSrc, src, errorCode)) {
michael@0:                     break;
michael@0:                 }
michael@0:             } else {
michael@0:                 prevCC=0;
michael@0:                 prevBoundary=src;
michael@0:             }
michael@0:         }
michael@0:         if(src==limit) {
michael@0:             break;
michael@0:         }
michael@0: 
michael@0:         // Check one above-minimum, relevant code point.
michael@0:         src+=U16_LENGTH(c);
michael@0:         if(buffer!=NULL) {
michael@0:             if(!decompose(c, norm16, *buffer, errorCode)) {
michael@0:                 break;
michael@0:             }
michael@0:         } else {
michael@0:             if(isDecompYes(norm16)) {
michael@0:                 uint8_t cc=getCCFromYesOrMaybe(norm16);
michael@0:                 if(prevCC<=cc || cc==0) {
michael@0:                     prevCC=cc;
michael@0:                     if(cc<=1) {
michael@0:                         prevBoundary=src;
michael@0:                     }
michael@0:                     continue;
michael@0:                 }
michael@0:             }
michael@0:             return prevBoundary;  // "no" or cc out of order
michael@0:         }
michael@0:     }
michael@0:     return src;
michael@0: }
michael@0: 
michael@0: // Decompose a short piece of text which is likely to contain characters that
michael@0: // fail the quick check loop and/or where the quick check loop's overhead
michael@0: // is unlikely to be amortized.
michael@0: // Called by the compose() and makeFCD() implementations.
michael@0: UBool Normalizer2Impl::decomposeShort(const UChar *src, const UChar *limit,
michael@0:                                       ReorderingBuffer &buffer,
michael@0:                                       UErrorCode &errorCode) const {
michael@0:     while(src<limit) {
michael@0:         UChar32 c;
michael@0:         uint16_t norm16;
michael@0:         UTRIE2_U16_NEXT16(normTrie, src, limit, c, norm16);
michael@0:         if(!decompose(c, norm16, buffer, errorCode)) {
michael@0:             return FALSE;
michael@0:         }
michael@0:     }
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: UBool Normalizer2Impl::decompose(UChar32 c, uint16_t norm16,
michael@0:                                  ReorderingBuffer &buffer,
michael@0:                                  UErrorCode &errorCode) const {
michael@0:     // Only loops for 1:1 algorithmic mappings.
michael@0:     for(;;) {
michael@0:         // get the decomposition and the lead and trail cc's
michael@0:         if(isDecompYes(norm16)) {
michael@0:             // c does not decompose
michael@0:             return buffer.append(c, getCCFromYesOrMaybe(norm16), errorCode);
michael@0:         } else if(isHangul(norm16)) {
michael@0:             // Hangul syllable: decompose algorithmically
michael@0:             UChar jamos[3];
michael@0:             return buffer.appendZeroCC(jamos, jamos+Hangul::decompose(c, jamos), errorCode);
michael@0:         } else if(isDecompNoAlgorithmic(norm16)) {
michael@0:             c=mapAlgorithmic(c, norm16);
michael@0:             norm16=getNorm16(c);
michael@0:         } else {
michael@0:             // c decomposes, get everything from the variable-length extra data
michael@0:             const uint16_t *mapping=getMapping(norm16);
michael@0:             uint16_t firstUnit=*mapping;
michael@0:             int32_t length=firstUnit&MAPPING_LENGTH_MASK;
michael@0:             uint8_t leadCC, trailCC;
michael@0:             trailCC=(uint8_t)(firstUnit>>8);
michael@0:             if(firstUnit&MAPPING_HAS_CCC_LCCC_WORD) {
michael@0:                 leadCC=(uint8_t)(*(mapping-1)>>8);
michael@0:             } else {
michael@0:                 leadCC=0;
michael@0:             }
michael@0:             return buffer.append((const UChar *)mapping+1, length, leadCC, trailCC, errorCode);
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: const UChar *
michael@0: Normalizer2Impl::getDecomposition(UChar32 c, UChar buffer[4], int32_t &length) const {
michael@0:     const UChar *decomp=NULL;
michael@0:     uint16_t norm16;
michael@0:     for(;;) {
michael@0:         if(c<minDecompNoCP || isDecompYes(norm16=getNorm16(c))) {
michael@0:             // c does not decompose
michael@0:             return decomp;
michael@0:         } else if(isHangul(norm16)) {
michael@0:             // Hangul syllable: decompose algorithmically
michael@0:             length=Hangul::decompose(c, buffer);
michael@0:             return buffer;
michael@0:         } else if(isDecompNoAlgorithmic(norm16)) {
michael@0:             c=mapAlgorithmic(c, norm16);
michael@0:             decomp=buffer;
michael@0:             length=0;
michael@0:             U16_APPEND_UNSAFE(buffer, length, c);
michael@0:         } else {
michael@0:             // c decomposes, get everything from the variable-length extra data
michael@0:             const uint16_t *mapping=getMapping(norm16);
michael@0:             length=*mapping&MAPPING_LENGTH_MASK;
michael@0:             return (const UChar *)mapping+1;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: // The capacity of the buffer must be 30=MAPPING_LENGTH_MASK-1
michael@0: // so that a raw mapping fits that consists of one unit ("rm0")
michael@0: // plus all but the first two code units of the normal mapping.
michael@0: // The maximum length of a normal mapping is 31=MAPPING_LENGTH_MASK.
michael@0: const UChar *
michael@0: Normalizer2Impl::getRawDecomposition(UChar32 c, UChar buffer[30], int32_t &length) const {
michael@0:     // We do not loop in this method because an algorithmic mapping itself
michael@0:     // becomes a final result rather than having to be decomposed recursively.
michael@0:     uint16_t norm16;
michael@0:     if(c<minDecompNoCP || isDecompYes(norm16=getNorm16(c))) {
michael@0:         // c does not decompose
michael@0:         return NULL;
michael@0:     } else if(isHangul(norm16)) {
michael@0:         // Hangul syllable: decompose algorithmically
michael@0:         Hangul::getRawDecomposition(c, buffer);
michael@0:         length=2;
michael@0:         return buffer;
michael@0:     } else if(isDecompNoAlgorithmic(norm16)) {
michael@0:         c=mapAlgorithmic(c, norm16);
michael@0:         length=0;
michael@0:         U16_APPEND_UNSAFE(buffer, length, c);
michael@0:         return buffer;
michael@0:     } else {
michael@0:         // c decomposes, get everything from the variable-length extra data
michael@0:         const uint16_t *mapping=getMapping(norm16);
michael@0:         uint16_t firstUnit=*mapping;
michael@0:         int32_t mLength=firstUnit&MAPPING_LENGTH_MASK;  // length of normal mapping
michael@0:         if(firstUnit&MAPPING_HAS_RAW_MAPPING) {
michael@0:             // Read the raw mapping from before the firstUnit and before the optional ccc/lccc word.
michael@0:             // Bit 7=MAPPING_HAS_CCC_LCCC_WORD
michael@0:             const uint16_t *rawMapping=mapping-((firstUnit>>7)&1)-1;
michael@0:             uint16_t rm0=*rawMapping;
michael@0:             if(rm0<=MAPPING_LENGTH_MASK) {
michael@0:                 length=rm0;
michael@0:                 return (const UChar *)rawMapping-rm0;
michael@0:             } else {
michael@0:                 // Copy the normal mapping and replace its first two code units with rm0.
michael@0:                 buffer[0]=(UChar)rm0;
michael@0:                 u_memcpy(buffer+1, (const UChar *)mapping+1+2, mLength-2);
michael@0:                 length=mLength-1;
michael@0:                 return buffer;
michael@0:             }
michael@0:         } else {
michael@0:             length=mLength;
michael@0:             return (const UChar *)mapping+1;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: void Normalizer2Impl::decomposeAndAppend(const UChar *src, const UChar *limit,
michael@0:                                          UBool doDecompose,
michael@0:                                          UnicodeString &safeMiddle,
michael@0:                                          ReorderingBuffer &buffer,
michael@0:                                          UErrorCode &errorCode) const {
michael@0:     buffer.copyReorderableSuffixTo(safeMiddle);
michael@0:     if(doDecompose) {
michael@0:         decompose(src, limit, &buffer, errorCode);
michael@0:         return;
michael@0:     }
michael@0:     // Just merge the strings at the boundary.
michael@0:     ForwardUTrie2StringIterator iter(normTrie, src, limit);
michael@0:     uint8_t firstCC, prevCC, cc;
michael@0:     firstCC=prevCC=cc=getCC(iter.next16());
michael@0:     while(cc!=0) {
michael@0:         prevCC=cc;
michael@0:         cc=getCC(iter.next16());
michael@0:     };
michael@0:     if(limit==NULL) {  // appendZeroCC() needs limit!=NULL
michael@0:         limit=u_strchr(iter.codePointStart, 0);
michael@0:     }
michael@0: 
michael@0:     if (buffer.append(src, (int32_t)(iter.codePointStart-src), firstCC, prevCC, errorCode)) {
michael@0:         buffer.appendZeroCC(iter.codePointStart, limit, errorCode);
michael@0:     }
michael@0: }
michael@0: 
michael@0: // Note: hasDecompBoundary() could be implemented as aliases to
michael@0: // hasFCDBoundaryBefore() and hasFCDBoundaryAfter()
michael@0: // at the cost of building the FCD trie for a decomposition normalizer.
michael@0: UBool Normalizer2Impl::hasDecompBoundary(UChar32 c, UBool before) const {
michael@0:     for(;;) {
michael@0:         if(c<minDecompNoCP) {
michael@0:             return TRUE;
michael@0:         }
michael@0:         uint16_t norm16=getNorm16(c);
michael@0:         if(isHangul(norm16) || isDecompYesAndZeroCC(norm16)) {
michael@0:             return TRUE;
michael@0:         } else if(norm16>MIN_NORMAL_MAYBE_YES) {
michael@0:             return FALSE;  // ccc!=0
michael@0:         } else if(isDecompNoAlgorithmic(norm16)) {
michael@0:             c=mapAlgorithmic(c, norm16);
michael@0:         } else {
michael@0:             // c decomposes, get everything from the variable-length extra data
michael@0:             const uint16_t *mapping=getMapping(norm16);
michael@0:             uint16_t firstUnit=*mapping;
michael@0:             if((firstUnit&MAPPING_LENGTH_MASK)==0) {
michael@0:                 return FALSE;
michael@0:             }
michael@0:             if(!before) {
michael@0:                 // decomp after-boundary: same as hasFCDBoundaryAfter(),
michael@0:                 // fcd16<=1 || trailCC==0
michael@0:                 if(firstUnit>0x1ff) {
michael@0:                     return FALSE;  // trailCC>1
michael@0:                 }
michael@0:                 if(firstUnit<=0xff) {
michael@0:                     return TRUE;  // trailCC==0
michael@0:                 }
michael@0:                 // if(trailCC==1) test leadCC==0, same as checking for before-boundary
michael@0:             }
michael@0:             // TRUE if leadCC==0 (hasFCDBoundaryBefore())
michael@0:             return (firstUnit&MAPPING_HAS_CCC_LCCC_WORD)==0 || (*(mapping-1)&0xff00)==0;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Finds the recomposition result for
michael@0:  * a forward-combining "lead" character,
michael@0:  * specified with a pointer to its compositions list,
michael@0:  * and a backward-combining "trail" character.
michael@0:  *
michael@0:  * If the lead and trail characters combine, then this function returns
michael@0:  * the following "compositeAndFwd" value:
michael@0:  * Bits 21..1  composite character
michael@0:  * Bit      0  set if the composite is a forward-combining starter
michael@0:  * otherwise it returns -1.
michael@0:  *
michael@0:  * The compositions list has (trail, compositeAndFwd) pair entries,
michael@0:  * encoded as either pairs or triples of 16-bit units.
michael@0:  * The last entry has the high bit of its first unit set.
michael@0:  *
michael@0:  * The list is sorted by ascending trail characters (there are no duplicates).
michael@0:  * A linear search is used.
michael@0:  *
michael@0:  * See normalizer2impl.h for a more detailed description
michael@0:  * of the compositions list format.
michael@0:  */
michael@0: int32_t Normalizer2Impl::combine(const uint16_t *list, UChar32 trail) {
michael@0:     uint16_t key1, firstUnit;
michael@0:     if(trail<COMP_1_TRAIL_LIMIT) {
michael@0:         // trail character is 0..33FF
michael@0:         // result entry may have 2 or 3 units
michael@0:         key1=(uint16_t)(trail<<1);
michael@0:         while(key1>(firstUnit=*list)) {
michael@0:             list+=2+(firstUnit&COMP_1_TRIPLE);
michael@0:         }
michael@0:         if(key1==(firstUnit&COMP_1_TRAIL_MASK)) {
michael@0:             if(firstUnit&COMP_1_TRIPLE) {
michael@0:                 return ((int32_t)list[1]<<16)|list[2];
michael@0:             } else {
michael@0:                 return list[1];
michael@0:             }
michael@0:         }
michael@0:     } else {
michael@0:         // trail character is 3400..10FFFF
michael@0:         // result entry has 3 units
michael@0:         key1=(uint16_t)(COMP_1_TRAIL_LIMIT+
michael@0:                         (((trail>>COMP_1_TRAIL_SHIFT))&
michael@0:                           ~COMP_1_TRIPLE));
michael@0:         uint16_t key2=(uint16_t)(trail<<COMP_2_TRAIL_SHIFT);
michael@0:         uint16_t secondUnit;
michael@0:         for(;;) {
michael@0:             if(key1>(firstUnit=*list)) {
michael@0:                 list+=2+(firstUnit&COMP_1_TRIPLE);
michael@0:             } else if(key1==(firstUnit&COMP_1_TRAIL_MASK)) {
michael@0:                 if(key2>(secondUnit=list[1])) {
michael@0:                     if(firstUnit&COMP_1_LAST_TUPLE) {
michael@0:                         break;
michael@0:                     } else {
michael@0:                         list+=3;
michael@0:                     }
michael@0:                 } else if(key2==(secondUnit&COMP_2_TRAIL_MASK)) {
michael@0:                     return ((int32_t)(secondUnit&~COMP_2_TRAIL_MASK)<<16)|list[2];
michael@0:                 } else {
michael@0:                     break;
michael@0:                 }
michael@0:             } else {
michael@0:                 break;
michael@0:             }
michael@0:         }
michael@0:     }
michael@0:     return -1;
michael@0: }
michael@0: 
michael@0: /**
michael@0:   * @param list some character's compositions list
michael@0:   * @param set recursively receives the composites from these compositions
michael@0:   */
michael@0: void Normalizer2Impl::addComposites(const uint16_t *list, UnicodeSet &set) const {
michael@0:     uint16_t firstUnit;
michael@0:     int32_t compositeAndFwd;
michael@0:     do {
michael@0:         firstUnit=*list;
michael@0:         if((firstUnit&COMP_1_TRIPLE)==0) {
michael@0:             compositeAndFwd=list[1];
michael@0:             list+=2;
michael@0:         } else {
michael@0:             compositeAndFwd=(((int32_t)list[1]&~COMP_2_TRAIL_MASK)<<16)|list[2];
michael@0:             list+=3;
michael@0:         }
michael@0:         UChar32 composite=compositeAndFwd>>1;
michael@0:         if((compositeAndFwd&1)!=0) {
michael@0:             addComposites(getCompositionsListForComposite(getNorm16(composite)), set);
michael@0:         }
michael@0:         set.add(composite);
michael@0:     } while((firstUnit&COMP_1_LAST_TUPLE)==0);
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Recomposes the buffer text starting at recomposeStartIndex
michael@0:  * (which is in NFD - decomposed and canonically ordered),
michael@0:  * and truncates the buffer contents.
michael@0:  *
michael@0:  * Note that recomposition never lengthens the text:
michael@0:  * Any character consists of either one or two code units;
michael@0:  * a composition may contain at most one more code unit than the original starter,
michael@0:  * while the combining mark that is removed has at least one code unit.
michael@0:  */
michael@0: void Normalizer2Impl::recompose(ReorderingBuffer &buffer, int32_t recomposeStartIndex,
michael@0:                                 UBool onlyContiguous) const {
michael@0:     UChar *p=buffer.getStart()+recomposeStartIndex;
michael@0:     UChar *limit=buffer.getLimit();
michael@0:     if(p==limit) {
michael@0:         return;
michael@0:     }
michael@0: 
michael@0:     UChar *starter, *pRemove, *q, *r;
michael@0:     const uint16_t *compositionsList;
michael@0:     UChar32 c, compositeAndFwd;
michael@0:     uint16_t norm16;
michael@0:     uint8_t cc, prevCC;
michael@0:     UBool starterIsSupplementary;
michael@0: 
michael@0:     // Some of the following variables are not used until we have a forward-combining starter
michael@0:     // and are only initialized now to avoid compiler warnings.
michael@0:     compositionsList=NULL;  // used as indicator for whether we have a forward-combining starter
michael@0:     starter=NULL;
michael@0:     starterIsSupplementary=FALSE;
michael@0:     prevCC=0;
michael@0: 
michael@0:     for(;;) {
michael@0:         UTRIE2_U16_NEXT16(normTrie, p, limit, c, norm16);
michael@0:         cc=getCCFromYesOrMaybe(norm16);
michael@0:         if( // this character combines backward and
michael@0:             isMaybe(norm16) &&
michael@0:             // we have seen a starter that combines forward and
michael@0:             compositionsList!=NULL &&
michael@0:             // the backward-combining character is not blocked
michael@0:             (prevCC<cc || prevCC==0)
michael@0:         ) {
michael@0:             if(isJamoVT(norm16)) {
michael@0:                 // c is a Jamo V/T, see if we can compose it with the previous character.
michael@0:                 if(c<Hangul::JAMO_T_BASE) {
michael@0:                     // c is a Jamo Vowel, compose with previous Jamo L and following Jamo T.
michael@0:                     UChar prev=(UChar)(*starter-Hangul::JAMO_L_BASE);
michael@0:                     if(prev<Hangul::JAMO_L_COUNT) {
michael@0:                         pRemove=p-1;
michael@0:                         UChar syllable=(UChar)
michael@0:                             (Hangul::HANGUL_BASE+
michael@0:                              (prev*Hangul::JAMO_V_COUNT+(c-Hangul::JAMO_V_BASE))*
michael@0:                              Hangul::JAMO_T_COUNT);
michael@0:                         UChar t;
michael@0:                         if(p!=limit && (t=(UChar)(*p-Hangul::JAMO_T_BASE))<Hangul::JAMO_T_COUNT) {
michael@0:                             ++p;
michael@0:                             syllable+=t;  // The next character was a Jamo T.
michael@0:                         }
michael@0:                         *starter=syllable;
michael@0:                         // remove the Jamo V/T
michael@0:                         q=pRemove;
michael@0:                         r=p;
michael@0:                         while(r<limit) {
michael@0:                             *q++=*r++;
michael@0:                         }
michael@0:                         limit=q;
michael@0:                         p=pRemove;
michael@0:                     }
michael@0:                 }
michael@0:                 /*
michael@0:                  * No "else" for Jamo T:
michael@0:                  * Since the input is in NFD, there are no Hangul LV syllables that
michael@0:                  * a Jamo T could combine with.
michael@0:                  * All Jamo Ts are combined above when handling Jamo Vs.
michael@0:                  */
michael@0:                 if(p==limit) {
michael@0:                     break;
michael@0:                 }
michael@0:                 compositionsList=NULL;
michael@0:                 continue;
michael@0:             } else if((compositeAndFwd=combine(compositionsList, c))>=0) {
michael@0:                 // The starter and the combining mark (c) do combine.
michael@0:                 UChar32 composite=compositeAndFwd>>1;
michael@0: 
michael@0:                 // Replace the starter with the composite, remove the combining mark.
michael@0:                 pRemove=p-U16_LENGTH(c);  // pRemove & p: start & limit of the combining mark
michael@0:                 if(starterIsSupplementary) {
michael@0:                     if(U_IS_SUPPLEMENTARY(composite)) {
michael@0:                         // both are supplementary
michael@0:                         starter[0]=U16_LEAD(composite);
michael@0:                         starter[1]=U16_TRAIL(composite);
michael@0:                     } else {
michael@0:                         *starter=(UChar)composite;
michael@0:                         // The composite is shorter than the starter,
michael@0:                         // move the intermediate characters forward one.
michael@0:                         starterIsSupplementary=FALSE;
michael@0:                         q=starter+1;
michael@0:                         r=q+1;
michael@0:                         while(r<pRemove) {
michael@0:                             *q++=*r++;
michael@0:                         }
michael@0:                         --pRemove;
michael@0:                     }
michael@0:                 } else if(U_IS_SUPPLEMENTARY(composite)) {
michael@0:                     // The composite is longer than the starter,
michael@0:                     // move the intermediate characters back one.
michael@0:                     starterIsSupplementary=TRUE;
michael@0:                     ++starter;  // temporarily increment for the loop boundary
michael@0:                     q=pRemove;
michael@0:                     r=++pRemove;
michael@0:                     while(starter<q) {
michael@0:                         *--r=*--q;
michael@0:                     }
michael@0:                     *starter=U16_TRAIL(composite);
michael@0:                     *--starter=U16_LEAD(composite);  // undo the temporary increment
michael@0:                 } else {
michael@0:                     // both are on the BMP
michael@0:                     *starter=(UChar)composite;
michael@0:                 }
michael@0: 
michael@0:                 /* remove the combining mark by moving the following text over it */
michael@0:                 if(pRemove<p) {
michael@0:                     q=pRemove;
michael@0:                     r=p;
michael@0:                     while(r<limit) {
michael@0:                         *q++=*r++;
michael@0:                     }
michael@0:                     limit=q;
michael@0:                     p=pRemove;
michael@0:                 }
michael@0:                 // Keep prevCC because we removed the combining mark.
michael@0: 
michael@0:                 if(p==limit) {
michael@0:                     break;
michael@0:                 }
michael@0:                 // Is the composite a starter that combines forward?
michael@0:                 if(compositeAndFwd&1) {
michael@0:                     compositionsList=
michael@0:                         getCompositionsListForComposite(getNorm16(composite));
michael@0:                 } else {
michael@0:                     compositionsList=NULL;
michael@0:                 }
michael@0: 
michael@0:                 // We combined; continue with looking for compositions.
michael@0:                 continue;
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         // no combination this time
michael@0:         prevCC=cc;
michael@0:         if(p==limit) {
michael@0:             break;
michael@0:         }
michael@0: 
michael@0:         // If c did not combine, then check if it is a starter.
michael@0:         if(cc==0) {
michael@0:             // Found a new starter.
michael@0:             if((compositionsList=getCompositionsListForDecompYes(norm16))!=NULL) {
michael@0:                 // It may combine with something, prepare for it.
michael@0:                 if(U_IS_BMP(c)) {
michael@0:                     starterIsSupplementary=FALSE;
michael@0:                     starter=p-1;
michael@0:                 } else {
michael@0:                     starterIsSupplementary=TRUE;
michael@0:                     starter=p-2;
michael@0:                 }
michael@0:             }
michael@0:         } else if(onlyContiguous) {
michael@0:             // FCC: no discontiguous compositions; any intervening character blocks.
michael@0:             compositionsList=NULL;
michael@0:         }
michael@0:     }
michael@0:     buffer.setReorderingLimit(limit);
michael@0: }
michael@0: 
michael@0: UChar32
michael@0: Normalizer2Impl::composePair(UChar32 a, UChar32 b) const {
michael@0:     uint16_t norm16=getNorm16(a);  // maps an out-of-range 'a' to inert norm16=0
michael@0:     const uint16_t *list;
michael@0:     if(isInert(norm16)) {
michael@0:         return U_SENTINEL;
michael@0:     } else if(norm16<minYesNoMappingsOnly) {
michael@0:         if(isJamoL(norm16)) {
michael@0:             b-=Hangul::JAMO_V_BASE;
michael@0:             if(0<=b && b<Hangul::JAMO_V_COUNT) {
michael@0:                 return
michael@0:                     (Hangul::HANGUL_BASE+
michael@0:                      ((a-Hangul::JAMO_L_BASE)*Hangul::JAMO_V_COUNT+b)*
michael@0:                      Hangul::JAMO_T_COUNT);
michael@0:             } else {
michael@0:                 return U_SENTINEL;
michael@0:             }
michael@0:         } else if(isHangul(norm16)) {
michael@0:             b-=Hangul::JAMO_T_BASE;
michael@0:             if(Hangul::isHangulWithoutJamoT(a) && 0<b && b<Hangul::JAMO_T_COUNT) {  // not b==0!
michael@0:                 return a+b;
michael@0:             } else {
michael@0:                 return U_SENTINEL;
michael@0:             }
michael@0:         } else {
michael@0:             // 'a' has a compositions list in extraData
michael@0:             list=extraData+norm16;
michael@0:             if(norm16>minYesNo) {  // composite 'a' has both mapping & compositions list
michael@0:                 list+=  // mapping pointer
michael@0:                     1+  // +1 to skip the first unit with the mapping lenth
michael@0:                     (*list&MAPPING_LENGTH_MASK);  // + mapping length
michael@0:             }
michael@0:         }
michael@0:     } else if(norm16<minMaybeYes || MIN_NORMAL_MAYBE_YES<=norm16) {
michael@0:         return U_SENTINEL;
michael@0:     } else {
michael@0:         list=maybeYesCompositions+norm16-minMaybeYes;
michael@0:     }
michael@0:     if(b<0 || 0x10ffff<b) {  // combine(list, b) requires a valid code point b
michael@0:         return U_SENTINEL;
michael@0:     }
michael@0: #if U_SIGNED_RIGHT_SHIFT_IS_ARITHMETIC
michael@0:     return combine(list, b)>>1;
michael@0: #else
michael@0:     int32_t compositeAndFwd=combine(list, b);
michael@0:     return compositeAndFwd>=0 ? compositeAndFwd>>1 : U_SENTINEL;
michael@0: #endif
michael@0: }
michael@0: 
michael@0: // Very similar to composeQuickCheck(): Make the same changes in both places if relevant.
michael@0: // doCompose: normalize
michael@0: // !doCompose: isNormalized (buffer must be empty and initialized)
michael@0: UBool
michael@0: Normalizer2Impl::compose(const UChar *src, const UChar *limit,
michael@0:                          UBool onlyContiguous,
michael@0:                          UBool doCompose,
michael@0:                          ReorderingBuffer &buffer,
michael@0:                          UErrorCode &errorCode) const {
michael@0:     /*
michael@0:      * prevBoundary points to the last character before the current one
michael@0:      * that has a composition boundary before it with ccc==0 and quick check "yes".
michael@0:      * Keeping track of prevBoundary saves us looking for a composition boundary
michael@0:      * when we find a "no" or "maybe".
michael@0:      *
michael@0:      * When we back out from prevSrc back to prevBoundary,
michael@0:      * then we also remove those same characters (which had been simply copied
michael@0:      * or canonically-order-inserted) from the ReorderingBuffer.
michael@0:      * Therefore, at all times, the [prevBoundary..prevSrc[ source units
michael@0:      * must correspond 1:1 to destination units at the end of the destination buffer.
michael@0:      */
michael@0:     const UChar *prevBoundary=src;
michael@0:     UChar32 minNoMaybeCP=minCompNoMaybeCP;
michael@0:     if(limit==NULL) {
michael@0:         src=copyLowPrefixFromNulTerminated(src, minNoMaybeCP,
michael@0:                                            doCompose ? &buffer : NULL,
michael@0:                                            errorCode);
michael@0:         if(U_FAILURE(errorCode)) {
michael@0:             return FALSE;
michael@0:         }
michael@0:         if(prevBoundary<src) {
michael@0:             // Set prevBoundary to the last character in the prefix.
michael@0:             prevBoundary=src-1;
michael@0:         }
michael@0:         limit=u_strchr(src, 0);
michael@0:     }
michael@0: 
michael@0:     const UChar *prevSrc;
michael@0:     UChar32 c=0;
michael@0:     uint16_t norm16=0;
michael@0: 
michael@0:     // only for isNormalized
michael@0:     uint8_t prevCC=0;
michael@0: 
michael@0:     for(;;) {
michael@0:         // count code units below the minimum or with irrelevant data for the quick check
michael@0:         for(prevSrc=src; src!=limit;) {
michael@0:             if( (c=*src)<minNoMaybeCP ||
michael@0:                 isCompYesAndZeroCC(norm16=UTRIE2_GET16_FROM_U16_SINGLE_LEAD(normTrie, c))
michael@0:             ) {
michael@0:                 ++src;
michael@0:             } else if(!U16_IS_SURROGATE(c)) {
michael@0:                 break;
michael@0:             } else {
michael@0:                 UChar c2;
michael@0:                 if(U16_IS_SURROGATE_LEAD(c)) {
michael@0:                     if((src+1)!=limit && U16_IS_TRAIL(c2=src[1])) {
michael@0:                         c=U16_GET_SUPPLEMENTARY(c, c2);
michael@0:                     }
michael@0:                 } else /* trail surrogate */ {
michael@0:                     if(prevSrc<src && U16_IS_LEAD(c2=*(src-1))) {
michael@0:                         --src;
michael@0:                         c=U16_GET_SUPPLEMENTARY(c2, c);
michael@0:                     }
michael@0:                 }
michael@0:                 if(isCompYesAndZeroCC(norm16=getNorm16(c))) {
michael@0:                     src+=U16_LENGTH(c);
michael@0:                 } else {
michael@0:                     break;
michael@0:                 }
michael@0:             }
michael@0:         }
michael@0:         // copy these code units all at once
michael@0:         if(src!=prevSrc) {
michael@0:             if(doCompose) {
michael@0:                 if(!buffer.appendZeroCC(prevSrc, src, errorCode)) {
michael@0:                     break;
michael@0:                 }
michael@0:             } else {
michael@0:                 prevCC=0;
michael@0:             }
michael@0:             if(src==limit) {
michael@0:                 break;
michael@0:             }
michael@0:             // Set prevBoundary to the last character in the quick check loop.
michael@0:             prevBoundary=src-1;
michael@0:             if( U16_IS_TRAIL(*prevBoundary) && prevSrc<prevBoundary &&
michael@0:                 U16_IS_LEAD(*(prevBoundary-1))
michael@0:             ) {
michael@0:                 --prevBoundary;
michael@0:             }
michael@0:             // The start of the current character (c).
michael@0:             prevSrc=src;
michael@0:         } else if(src==limit) {
michael@0:             break;
michael@0:         }
michael@0: 
michael@0:         src+=U16_LENGTH(c);
michael@0:         /*
michael@0:          * isCompYesAndZeroCC(norm16) is false, that is, norm16>=minNoNo.
michael@0:          * c is either a "noNo" (has a mapping) or a "maybeYes" (combines backward)
michael@0:          * or has ccc!=0.
michael@0:          * Check for Jamo V/T, then for regular characters.
michael@0:          * c is not a Hangul syllable or Jamo L because those have "yes" properties.
michael@0:          */
michael@0:         if(isJamoVT(norm16) && prevBoundary!=prevSrc) {
michael@0:             UChar prev=*(prevSrc-1);
michael@0:             UBool needToDecompose=FALSE;
michael@0:             if(c<Hangul::JAMO_T_BASE) {
michael@0:                 // c is a Jamo Vowel, compose with previous Jamo L and following Jamo T.
michael@0:                 prev=(UChar)(prev-Hangul::JAMO_L_BASE);
michael@0:                 if(prev<Hangul::JAMO_L_COUNT) {
michael@0:                     if(!doCompose) {
michael@0:                         return FALSE;
michael@0:                     }
michael@0:                     UChar syllable=(UChar)
michael@0:                         (Hangul::HANGUL_BASE+
michael@0:                          (prev*Hangul::JAMO_V_COUNT+(c-Hangul::JAMO_V_BASE))*
michael@0:                          Hangul::JAMO_T_COUNT);
michael@0:                     UChar t;
michael@0:                     if(src!=limit && (t=(UChar)(*src-Hangul::JAMO_T_BASE))<Hangul::JAMO_T_COUNT) {
michael@0:                         ++src;
michael@0:                         syllable+=t;  // The next character was a Jamo T.
michael@0:                         prevBoundary=src;
michael@0:                         buffer.setLastChar(syllable);
michael@0:                         continue;
michael@0:                     }
michael@0:                     // If we see L+V+x where x!=T then we drop to the slow path,
michael@0:                     // decompose and recompose.
michael@0:                     // This is to deal with NFKC finding normal L and V but a
michael@0:                     // compatibility variant of a T. We need to either fully compose that
michael@0:                     // combination here (which would complicate the code and may not work
michael@0:                     // with strange custom data) or use the slow path -- or else our replacing
michael@0:                     // two input characters (L+V) with one output character (LV syllable)
michael@0:                     // would violate the invariant that [prevBoundary..prevSrc[ has the same
michael@0:                     // length as what we appended to the buffer since prevBoundary.
michael@0:                     needToDecompose=TRUE;
michael@0:                 }
michael@0:             } else if(Hangul::isHangulWithoutJamoT(prev)) {
michael@0:                 // c is a Jamo Trailing consonant,
michael@0:                 // compose with previous Hangul LV that does not contain a Jamo T.
michael@0:                 if(!doCompose) {
michael@0:                     return FALSE;
michael@0:                 }
michael@0:                 buffer.setLastChar((UChar)(prev+c-Hangul::JAMO_T_BASE));
michael@0:                 prevBoundary=src;
michael@0:                 continue;
michael@0:             }
michael@0:             if(!needToDecompose) {
michael@0:                 // The Jamo V/T did not compose into a Hangul syllable.
michael@0:                 if(doCompose) {
michael@0:                     if(!buffer.appendBMP((UChar)c, 0, errorCode)) {
michael@0:                         break;
michael@0:                     }
michael@0:                 } else {
michael@0:                     prevCC=0;
michael@0:                 }
michael@0:                 continue;
michael@0:             }
michael@0:         }
michael@0:         /*
michael@0:          * Source buffer pointers:
michael@0:          *
michael@0:          *  all done      quick check   current char  not yet
michael@0:          *                "yes" but     (c)           processed
michael@0:          *                may combine
michael@0:          *                forward
michael@0:          * [-------------[-------------[-------------[-------------[
michael@0:          * |             |             |             |             |
michael@0:          * orig. src     prevBoundary  prevSrc       src           limit
michael@0:          *
michael@0:          *
michael@0:          * Destination buffer pointers inside the ReorderingBuffer:
michael@0:          *
michael@0:          *  all done      might take    not filled yet
michael@0:          *                characters for
michael@0:          *                reordering
michael@0:          * [-------------[-------------[-------------[
michael@0:          * |             |             |             |
michael@0:          * start         reorderStart  limit         |
michael@0:          *                             +remainingCap.+
michael@0:          */
michael@0:         if(norm16>=MIN_YES_YES_WITH_CC) {
michael@0:             uint8_t cc=(uint8_t)norm16;  // cc!=0
michael@0:             if( onlyContiguous &&  // FCC
michael@0:                 (doCompose ? buffer.getLastCC() : prevCC)==0 &&
michael@0:                 prevBoundary<prevSrc &&
michael@0:                 // buffer.getLastCC()==0 && prevBoundary<prevSrc tell us that
michael@0:                 // [prevBoundary..prevSrc[ (which is exactly one character under these conditions)
michael@0:                 // passed the quick check "yes && ccc==0" test.
michael@0:                 // Check whether the last character was a "yesYes" or a "yesNo".
michael@0:                 // If a "yesNo", then we get its trailing ccc from its
michael@0:                 // mapping and check for canonical order.
michael@0:                 // All other cases are ok.
michael@0:                 getTrailCCFromCompYesAndZeroCC(prevBoundary, prevSrc)>cc
michael@0:             ) {
michael@0:                 // Fails FCD test, need to decompose and contiguously recompose.
michael@0:                 if(!doCompose) {
michael@0:                     return FALSE;
michael@0:                 }
michael@0:             } else if(doCompose) {
michael@0:                 if(!buffer.append(c, cc, errorCode)) {
michael@0:                     break;
michael@0:                 }
michael@0:                 continue;
michael@0:             } else if(prevCC<=cc) {
michael@0:                 prevCC=cc;
michael@0:                 continue;
michael@0:             } else {
michael@0:                 return FALSE;
michael@0:             }
michael@0:         } else if(!doCompose && !isMaybeOrNonZeroCC(norm16)) {
michael@0:             return FALSE;
michael@0:         }
michael@0: 
michael@0:         /*
michael@0:          * Find appropriate boundaries around this character,
michael@0:          * decompose the source text from between the boundaries,
michael@0:          * and recompose it.
michael@0:          *
michael@0:          * We may need to remove the last few characters from the ReorderingBuffer
michael@0:          * to account for source text that was copied or appended
michael@0:          * but needs to take part in the recomposition.
michael@0:          */
michael@0: 
michael@0:         /*
michael@0:          * Find the last composition boundary in [prevBoundary..src[.
michael@0:          * It is either the decomposition of the current character (at prevSrc),
michael@0:          * or prevBoundary.
michael@0:          */
michael@0:         if(hasCompBoundaryBefore(c, norm16)) {
michael@0:             prevBoundary=prevSrc;
michael@0:         } else if(doCompose) {
michael@0:             buffer.removeSuffix((int32_t)(prevSrc-prevBoundary));
michael@0:         }
michael@0: 
michael@0:         // Find the next composition boundary in [src..limit[ -
michael@0:         // modifies src to point to the next starter.
michael@0:         src=(UChar *)findNextCompBoundary(src, limit);
michael@0: 
michael@0:         // Decompose [prevBoundary..src[ into the buffer and then recompose that part of it.
michael@0:         int32_t recomposeStartIndex=buffer.length();
michael@0:         if(!decomposeShort(prevBoundary, src, buffer, errorCode)) {
michael@0:             break;
michael@0:         }
michael@0:         recompose(buffer, recomposeStartIndex, onlyContiguous);
michael@0:         if(!doCompose) {
michael@0:             if(!buffer.equals(prevBoundary, src)) {
michael@0:                 return FALSE;
michael@0:             }
michael@0:             buffer.remove();
michael@0:             prevCC=0;
michael@0:         }
michael@0: 
michael@0:         // Move to the next starter. We never need to look back before this point again.
michael@0:         prevBoundary=src;
michael@0:     }
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: // Very similar to compose(): Make the same changes in both places if relevant.
michael@0: // pQCResult==NULL: spanQuickCheckYes
michael@0: // pQCResult!=NULL: quickCheck (*pQCResult must be UNORM_YES)
michael@0: const UChar *
michael@0: Normalizer2Impl::composeQuickCheck(const UChar *src, const UChar *limit,
michael@0:                                    UBool onlyContiguous,
michael@0:                                    UNormalizationCheckResult *pQCResult) const {
michael@0:     /*
michael@0:      * prevBoundary points to the last character before the current one
michael@0:      * that has a composition boundary before it with ccc==0 and quick check "yes".
michael@0:      */
michael@0:     const UChar *prevBoundary=src;
michael@0:     UChar32 minNoMaybeCP=minCompNoMaybeCP;
michael@0:     if(limit==NULL) {
michael@0:         UErrorCode errorCode=U_ZERO_ERROR;
michael@0:         src=copyLowPrefixFromNulTerminated(src, minNoMaybeCP, NULL, errorCode);
michael@0:         if(prevBoundary<src) {
michael@0:             // Set prevBoundary to the last character in the prefix.
michael@0:             prevBoundary=src-1;
michael@0:         }
michael@0:         limit=u_strchr(src, 0);
michael@0:     }
michael@0: 
michael@0:     const UChar *prevSrc;
michael@0:     UChar32 c=0;
michael@0:     uint16_t norm16=0;
michael@0:     uint8_t prevCC=0;
michael@0: 
michael@0:     for(;;) {
michael@0:         // count code units below the minimum or with irrelevant data for the quick check
michael@0:         for(prevSrc=src;;) {
michael@0:             if(src==limit) {
michael@0:                 return src;
michael@0:             }
michael@0:             if( (c=*src)<minNoMaybeCP ||
michael@0:                 isCompYesAndZeroCC(norm16=UTRIE2_GET16_FROM_U16_SINGLE_LEAD(normTrie, c))
michael@0:             ) {
michael@0:                 ++src;
michael@0:             } else if(!U16_IS_SURROGATE(c)) {
michael@0:                 break;
michael@0:             } else {
michael@0:                 UChar c2;
michael@0:                 if(U16_IS_SURROGATE_LEAD(c)) {
michael@0:                     if((src+1)!=limit && U16_IS_TRAIL(c2=src[1])) {
michael@0:                         c=U16_GET_SUPPLEMENTARY(c, c2);
michael@0:                     }
michael@0:                 } else /* trail surrogate */ {
michael@0:                     if(prevSrc<src && U16_IS_LEAD(c2=*(src-1))) {
michael@0:                         --src;
michael@0:                         c=U16_GET_SUPPLEMENTARY(c2, c);
michael@0:                     }
michael@0:                 }
michael@0:                 if(isCompYesAndZeroCC(norm16=getNorm16(c))) {
michael@0:                     src+=U16_LENGTH(c);
michael@0:                 } else {
michael@0:                     break;
michael@0:                 }
michael@0:             }
michael@0:         }
michael@0:         if(src!=prevSrc) {
michael@0:             // Set prevBoundary to the last character in the quick check loop.
michael@0:             prevBoundary=src-1;
michael@0:             if( U16_IS_TRAIL(*prevBoundary) && prevSrc<prevBoundary &&
michael@0:                 U16_IS_LEAD(*(prevBoundary-1))
michael@0:             ) {
michael@0:                 --prevBoundary;
michael@0:             }
michael@0:             prevCC=0;
michael@0:             // The start of the current character (c).
michael@0:             prevSrc=src;
michael@0:         }
michael@0: 
michael@0:         src+=U16_LENGTH(c);
michael@0:         /*
michael@0:          * isCompYesAndZeroCC(norm16) is false, that is, norm16>=minNoNo.
michael@0:          * c is either a "noNo" (has a mapping) or a "maybeYes" (combines backward)
michael@0:          * or has ccc!=0.
michael@0:          */
michael@0:         if(isMaybeOrNonZeroCC(norm16)) {
michael@0:             uint8_t cc=getCCFromYesOrMaybe(norm16);
michael@0:             if( onlyContiguous &&  // FCC
michael@0:                 cc!=0 &&
michael@0:                 prevCC==0 &&
michael@0:                 prevBoundary<prevSrc &&
michael@0:                 // prevCC==0 && prevBoundary<prevSrc tell us that
michael@0:                 // [prevBoundary..prevSrc[ (which is exactly one character under these conditions)
michael@0:                 // passed the quick check "yes && ccc==0" test.
michael@0:                 // Check whether the last character was a "yesYes" or a "yesNo".
michael@0:                 // If a "yesNo", then we get its trailing ccc from its
michael@0:                 // mapping and check for canonical order.
michael@0:                 // All other cases are ok.
michael@0:                 getTrailCCFromCompYesAndZeroCC(prevBoundary, prevSrc)>cc
michael@0:             ) {
michael@0:                 // Fails FCD test.
michael@0:             } else if(prevCC<=cc || cc==0) {
michael@0:                 prevCC=cc;
michael@0:                 if(norm16<MIN_YES_YES_WITH_CC) {
michael@0:                     if(pQCResult!=NULL) {
michael@0:                         *pQCResult=UNORM_MAYBE;
michael@0:                     } else {
michael@0:                         return prevBoundary;
michael@0:                     }
michael@0:                 }
michael@0:                 continue;
michael@0:             }
michael@0:         }
michael@0:         if(pQCResult!=NULL) {
michael@0:             *pQCResult=UNORM_NO;
michael@0:         }
michael@0:         return prevBoundary;
michael@0:     }
michael@0: }
michael@0: 
michael@0: void Normalizer2Impl::composeAndAppend(const UChar *src, const UChar *limit,
michael@0:                                        UBool doCompose,
michael@0:                                        UBool onlyContiguous,
michael@0:                                        UnicodeString &safeMiddle,
michael@0:                                        ReorderingBuffer &buffer,
michael@0:                                        UErrorCode &errorCode) const {
michael@0:     if(!buffer.isEmpty()) {
michael@0:         const UChar *firstStarterInSrc=findNextCompBoundary(src, limit);
michael@0:         if(src!=firstStarterInSrc) {
michael@0:             const UChar *lastStarterInDest=findPreviousCompBoundary(buffer.getStart(),
michael@0:                                                                     buffer.getLimit());
michael@0:             int32_t destSuffixLength=(int32_t)(buffer.getLimit()-lastStarterInDest);
michael@0:             UnicodeString middle(lastStarterInDest, destSuffixLength);
michael@0:             buffer.removeSuffix(destSuffixLength);
michael@0:             safeMiddle=middle;
michael@0:             middle.append(src, (int32_t)(firstStarterInSrc-src));
michael@0:             const UChar *middleStart=middle.getBuffer();
michael@0:             compose(middleStart, middleStart+middle.length(), onlyContiguous,
michael@0:                     TRUE, buffer, errorCode);
michael@0:             if(U_FAILURE(errorCode)) {
michael@0:                 return;
michael@0:             }
michael@0:             src=firstStarterInSrc;
michael@0:         }
michael@0:     }
michael@0:     if(doCompose) {
michael@0:         compose(src, limit, onlyContiguous, TRUE, buffer, errorCode);
michael@0:     } else {
michael@0:         if(limit==NULL) {  // appendZeroCC() needs limit!=NULL
michael@0:             limit=u_strchr(src, 0);
michael@0:         }
michael@0:         buffer.appendZeroCC(src, limit, errorCode);
michael@0:     }
michael@0: }
michael@0: 
michael@0: /**
michael@0:  * Does c have a composition boundary before it?
michael@0:  * True if its decomposition begins with a character that has
michael@0:  * ccc=0 && NFC_QC=Yes (isCompYesAndZeroCC()).
michael@0:  * As a shortcut, this is true if c itself has ccc=0 && NFC_QC=Yes
michael@0:  * (isCompYesAndZeroCC()) so we need not decompose.
michael@0:  */
michael@0: UBool Normalizer2Impl::hasCompBoundaryBefore(UChar32 c, uint16_t norm16) const {
michael@0:     for(;;) {
michael@0:         if(isCompYesAndZeroCC(norm16)) {
michael@0:             return TRUE;
michael@0:         } else if(isMaybeOrNonZeroCC(norm16)) {
michael@0:             return FALSE;
michael@0:         } else if(isDecompNoAlgorithmic(norm16)) {
michael@0:             c=mapAlgorithmic(c, norm16);
michael@0:             norm16=getNorm16(c);
michael@0:         } else {
michael@0:             // c decomposes, get everything from the variable-length extra data
michael@0:             const uint16_t *mapping=getMapping(norm16);
michael@0:             uint16_t firstUnit=*mapping;
michael@0:             if((firstUnit&MAPPING_LENGTH_MASK)==0) {
michael@0:                 return FALSE;
michael@0:             }
michael@0:             if((firstUnit&MAPPING_HAS_CCC_LCCC_WORD) && (*(mapping-1)&0xff00)) {
michael@0:                 return FALSE;  // non-zero leadCC
michael@0:             }
michael@0:             int32_t i=1;  // skip over the firstUnit
michael@0:             UChar32 c;
michael@0:             U16_NEXT_UNSAFE(mapping, i, c);
michael@0:             return isCompYesAndZeroCC(getNorm16(c));
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: UBool Normalizer2Impl::hasCompBoundaryAfter(UChar32 c, UBool onlyContiguous, UBool testInert) const {
michael@0:     for(;;) {
michael@0:         uint16_t norm16=getNorm16(c);
michael@0:         if(isInert(norm16)) {
michael@0:             return TRUE;
michael@0:         } else if(norm16<=minYesNo) {
michael@0:             // Hangul: norm16==minYesNo
michael@0:             // Hangul LVT has a boundary after it.
michael@0:             // Hangul LV and non-inert yesYes characters combine forward.
michael@0:             return isHangul(norm16) && !Hangul::isHangulWithoutJamoT((UChar)c);
michael@0:         } else if(norm16>= (testInert ? minNoNo : minMaybeYes)) {
michael@0:             return FALSE;
michael@0:         } else if(isDecompNoAlgorithmic(norm16)) {
michael@0:             c=mapAlgorithmic(c, norm16);
michael@0:         } else {
michael@0:             // c decomposes, get everything from the variable-length extra data.
michael@0:             // If testInert, then c must be a yesNo character which has lccc=0,
michael@0:             // otherwise it could be a noNo.
michael@0:             const uint16_t *mapping=getMapping(norm16);
michael@0:             uint16_t firstUnit=*mapping;
michael@0:             // TRUE if
michael@0:             //   not MAPPING_NO_COMP_BOUNDARY_AFTER
michael@0:             //     (which is set if
michael@0:             //       c is not deleted, and
michael@0:             //       it and its decomposition do not combine forward, and it has a starter)
michael@0:             //   and if FCC then trailCC<=1
michael@0:             return
michael@0:                 (firstUnit&MAPPING_NO_COMP_BOUNDARY_AFTER)==0 &&
michael@0:                 (!onlyContiguous || firstUnit<=0x1ff);
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: const UChar *Normalizer2Impl::findPreviousCompBoundary(const UChar *start, const UChar *p) const {
michael@0:     BackwardUTrie2StringIterator iter(normTrie, start, p);
michael@0:     uint16_t norm16;
michael@0:     do {
michael@0:         norm16=iter.previous16();
michael@0:     } while(!hasCompBoundaryBefore(iter.codePoint, norm16));
michael@0:     // We could also test hasCompBoundaryAfter() and return iter.codePointLimit,
michael@0:     // but that's probably not worth the extra cost.
michael@0:     return iter.codePointStart;
michael@0: }
michael@0: 
michael@0: const UChar *Normalizer2Impl::findNextCompBoundary(const UChar *p, const UChar *limit) const {
michael@0:     ForwardUTrie2StringIterator iter(normTrie, p, limit);
michael@0:     uint16_t norm16;
michael@0:     do {
michael@0:         norm16=iter.next16();
michael@0:     } while(!hasCompBoundaryBefore(iter.codePoint, norm16));
michael@0:     return iter.codePointStart;
michael@0: }
michael@0: 
michael@0: // Note: normalizer2impl.cpp r30982 (2011-nov-27)
michael@0: // still had getFCDTrie() which built and cached an FCD trie.
michael@0: // That provided faster access to FCD data than getFCD16FromNormData()
michael@0: // but required synchronization and consumed some 10kB of heap memory
michael@0: // in any process that uses FCD (e.g., via collation).
michael@0: // tccc180[] and smallFCD[] are intended to help with any loss of performance,
michael@0: // at least for Latin & CJK.
michael@0: 
michael@0: // Gets the FCD value from the regular normalization data.
michael@0: uint16_t Normalizer2Impl::getFCD16FromNormData(UChar32 c) const {
michael@0:     // Only loops for 1:1 algorithmic mappings.
michael@0:     for(;;) {
michael@0:         uint16_t norm16=getNorm16(c);
michael@0:         if(norm16<=minYesNo) {
michael@0:             // no decomposition or Hangul syllable, all zeros
michael@0:             return 0;
michael@0:         } else if(norm16>=MIN_NORMAL_MAYBE_YES) {
michael@0:             // combining mark
michael@0:             norm16&=0xff;
michael@0:             return norm16|(norm16<<8);
michael@0:         } else if(norm16>=minMaybeYes) {
michael@0:             return 0;
michael@0:         } else if(isDecompNoAlgorithmic(norm16)) {
michael@0:             c=mapAlgorithmic(c, norm16);
michael@0:         } else {
michael@0:             // c decomposes, get everything from the variable-length extra data
michael@0:             const uint16_t *mapping=getMapping(norm16);
michael@0:             uint16_t firstUnit=*mapping;
michael@0:             if((firstUnit&MAPPING_LENGTH_MASK)==0) {
michael@0:                 // A character that is deleted (maps to an empty string) must
michael@0:                 // get the worst-case lccc and tccc values because arbitrary
michael@0:                 // characters on both sides will become adjacent.
michael@0:                 return 0x1ff;
michael@0:             } else {
michael@0:                 norm16=firstUnit>>8;  // tccc
michael@0:                 if(firstUnit&MAPPING_HAS_CCC_LCCC_WORD) {
michael@0:                     norm16|=*(mapping-1)&0xff00;  // lccc
michael@0:                 }
michael@0:                 return norm16;
michael@0:             }
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: // Dual functionality:
michael@0: // buffer!=NULL: normalize
michael@0: // buffer==NULL: isNormalized/quickCheck/spanQuickCheckYes
michael@0: const UChar *
michael@0: Normalizer2Impl::makeFCD(const UChar *src, const UChar *limit,
michael@0:                          ReorderingBuffer *buffer,
michael@0:                          UErrorCode &errorCode) const {
michael@0:     // Tracks the last FCD-safe boundary, before lccc=0 or after properly-ordered tccc<=1.
michael@0:     // Similar to the prevBoundary in the compose() implementation.
michael@0:     const UChar *prevBoundary=src;
michael@0:     int32_t prevFCD16=0;
michael@0:     if(limit==NULL) {
michael@0:         src=copyLowPrefixFromNulTerminated(src, MIN_CCC_LCCC_CP, buffer, errorCode);
michael@0:         if(U_FAILURE(errorCode)) {
michael@0:             return src;
michael@0:         }
michael@0:         if(prevBoundary<src) {
michael@0:             prevBoundary=src;
michael@0:             // We know that the previous character's lccc==0.
michael@0:             // Fetching the fcd16 value was deferred for this below-U+0300 code point.
michael@0:             prevFCD16=getFCD16(*(src-1));
michael@0:             if(prevFCD16>1) {
michael@0:                 --prevBoundary;
michael@0:             }
michael@0:         }
michael@0:         limit=u_strchr(src, 0);
michael@0:     }
michael@0: 
michael@0:     // Note: In this function we use buffer->appendZeroCC() because we track
michael@0:     // the lead and trail combining classes here, rather than leaving it to
michael@0:     // the ReorderingBuffer.
michael@0:     // The exception is the call to decomposeShort() which uses the buffer
michael@0:     // in the normal way.
michael@0: 
michael@0:     const UChar *prevSrc;
michael@0:     UChar32 c=0;
michael@0:     uint16_t fcd16=0;
michael@0: 
michael@0:     for(;;) {
michael@0:         // count code units with lccc==0
michael@0:         for(prevSrc=src; src!=limit;) {
michael@0:             if((c=*src)<MIN_CCC_LCCC_CP) {
michael@0:                 prevFCD16=~c;
michael@0:                 ++src;
michael@0:             } else if(!singleLeadMightHaveNonZeroFCD16(c)) {
michael@0:                 prevFCD16=0;
michael@0:                 ++src;
michael@0:             } else {
michael@0:                 if(U16_IS_SURROGATE(c)) {
michael@0:                     UChar c2;
michael@0:                     if(U16_IS_SURROGATE_LEAD(c)) {
michael@0:                         if((src+1)!=limit && U16_IS_TRAIL(c2=src[1])) {
michael@0:                             c=U16_GET_SUPPLEMENTARY(c, c2);
michael@0:                         }
michael@0:                     } else /* trail surrogate */ {
michael@0:                         if(prevSrc<src && U16_IS_LEAD(c2=*(src-1))) {
michael@0:                             --src;
michael@0:                             c=U16_GET_SUPPLEMENTARY(c2, c);
michael@0:                         }
michael@0:                     }
michael@0:                 }
michael@0:                 if((fcd16=getFCD16FromNormData(c))<=0xff) {
michael@0:                     prevFCD16=fcd16;
michael@0:                     src+=U16_LENGTH(c);
michael@0:                 } else {
michael@0:                     break;
michael@0:                 }
michael@0:             }
michael@0:         }
michael@0:         // copy these code units all at once
michael@0:         if(src!=prevSrc) {
michael@0:             if(buffer!=NULL && !buffer->appendZeroCC(prevSrc, src, errorCode)) {
michael@0:                 break;
michael@0:             }
michael@0:             if(src==limit) {
michael@0:                 break;
michael@0:             }
michael@0:             prevBoundary=src;
michael@0:             // We know that the previous character's lccc==0.
michael@0:             if(prevFCD16<0) {
michael@0:                 // Fetching the fcd16 value was deferred for this below-U+0300 code point.
michael@0:                 UChar32 prev=~prevFCD16;
michael@0:                 prevFCD16= prev<0x180 ? tccc180[prev] : getFCD16FromNormData(prev);
michael@0:                 if(prevFCD16>1) {
michael@0:                     --prevBoundary;
michael@0:                 }
michael@0:             } else {
michael@0:                 const UChar *p=src-1;
michael@0:                 if(U16_IS_TRAIL(*p) && prevSrc<p && U16_IS_LEAD(*(p-1))) {
michael@0:                     --p;
michael@0:                     // Need to fetch the previous character's FCD value because
michael@0:                     // prevFCD16 was just for the trail surrogate code point.
michael@0:                     prevFCD16=getFCD16FromNormData(U16_GET_SUPPLEMENTARY(p[0], p[1]));
michael@0:                     // Still known to have lccc==0 because its lead surrogate unit had lccc==0.
michael@0:                 }
michael@0:                 if(prevFCD16>1) {
michael@0:                     prevBoundary=p;
michael@0:                 }
michael@0:             }
michael@0:             // The start of the current character (c).
michael@0:             prevSrc=src;
michael@0:         } else if(src==limit) {
michael@0:             break;
michael@0:         }
michael@0: 
michael@0:         src+=U16_LENGTH(c);
michael@0:         // The current character (c) at [prevSrc..src[ has a non-zero lead combining class.
michael@0:         // Check for proper order, and decompose locally if necessary.
michael@0:         if((prevFCD16&0xff)<=(fcd16>>8)) {
michael@0:             // proper order: prev tccc <= current lccc
michael@0:             if((fcd16&0xff)<=1) {
michael@0:                 prevBoundary=src;
michael@0:             }
michael@0:             if(buffer!=NULL && !buffer->appendZeroCC(c, errorCode)) {
michael@0:                 break;
michael@0:             }
michael@0:             prevFCD16=fcd16;
michael@0:             continue;
michael@0:         } else if(buffer==NULL) {
michael@0:             return prevBoundary;  // quick check "no"
michael@0:         } else {
michael@0:             /*
michael@0:              * Back out the part of the source that we copied or appended
michael@0:              * already but is now going to be decomposed.
michael@0:              * prevSrc is set to after what was copied/appended.
michael@0:              */
michael@0:             buffer->removeSuffix((int32_t)(prevSrc-prevBoundary));
michael@0:             /*
michael@0:              * Find the part of the source that needs to be decomposed,
michael@0:              * up to the next safe boundary.
michael@0:              */
michael@0:             src=findNextFCDBoundary(src, limit);
michael@0:             /*
michael@0:              * The source text does not fulfill the conditions for FCD.
michael@0:              * Decompose and reorder a limited piece of the text.
michael@0:              */
michael@0:             if(!decomposeShort(prevBoundary, src, *buffer, errorCode)) {
michael@0:                 break;
michael@0:             }
michael@0:             prevBoundary=src;
michael@0:             prevFCD16=0;
michael@0:         }
michael@0:     }
michael@0:     return src;
michael@0: }
michael@0: 
michael@0: void Normalizer2Impl::makeFCDAndAppend(const UChar *src, const UChar *limit,
michael@0:                                        UBool doMakeFCD,
michael@0:                                        UnicodeString &safeMiddle,
michael@0:                                        ReorderingBuffer &buffer,
michael@0:                                        UErrorCode &errorCode) const {
michael@0:     if(!buffer.isEmpty()) {
michael@0:         const UChar *firstBoundaryInSrc=findNextFCDBoundary(src, limit);
michael@0:         if(src!=firstBoundaryInSrc) {
michael@0:             const UChar *lastBoundaryInDest=findPreviousFCDBoundary(buffer.getStart(),
michael@0:                                                                     buffer.getLimit());
michael@0:             int32_t destSuffixLength=(int32_t)(buffer.getLimit()-lastBoundaryInDest);
michael@0:             UnicodeString middle(lastBoundaryInDest, destSuffixLength);
michael@0:             buffer.removeSuffix(destSuffixLength);
michael@0:             safeMiddle=middle;
michael@0:             middle.append(src, (int32_t)(firstBoundaryInSrc-src));
michael@0:             const UChar *middleStart=middle.getBuffer();
michael@0:             makeFCD(middleStart, middleStart+middle.length(), &buffer, errorCode);
michael@0:             if(U_FAILURE(errorCode)) {
michael@0:                 return;
michael@0:             }
michael@0:             src=firstBoundaryInSrc;
michael@0:         }
michael@0:     }
michael@0:     if(doMakeFCD) {
michael@0:         makeFCD(src, limit, &buffer, errorCode);
michael@0:     } else {
michael@0:         if(limit==NULL) {  // appendZeroCC() needs limit!=NULL
michael@0:             limit=u_strchr(src, 0);
michael@0:         }
michael@0:         buffer.appendZeroCC(src, limit, errorCode);
michael@0:     }
michael@0: }
michael@0: 
michael@0: const UChar *Normalizer2Impl::findPreviousFCDBoundary(const UChar *start, const UChar *p) const {
michael@0:     while(start<p && previousFCD16(start, p)>0xff) {}
michael@0:     return p;
michael@0: }
michael@0: 
michael@0: const UChar *Normalizer2Impl::findNextFCDBoundary(const UChar *p, const UChar *limit) const {
michael@0:     while(p<limit) {
michael@0:         const UChar *codePointStart=p;
michael@0:         if(nextFCD16(p, limit)<=0xff) {
michael@0:             return codePointStart;
michael@0:         }
michael@0:     }
michael@0:     return p;
michael@0: }
michael@0: 
michael@0: // CanonicalIterator data -------------------------------------------------- ***
michael@0: 
michael@0: CanonIterData::CanonIterData(UErrorCode &errorCode) :
michael@0:         trie(utrie2_open(0, 0, &errorCode)),
michael@0:         canonStartSets(uprv_deleteUObject, NULL, errorCode) {}
michael@0: 
michael@0: CanonIterData::~CanonIterData() {
michael@0:     utrie2_close(trie);
michael@0: }
michael@0: 
michael@0: void CanonIterData::addToStartSet(UChar32 origin, UChar32 decompLead, UErrorCode &errorCode) {
michael@0:     uint32_t canonValue=utrie2_get32(trie, decompLead);
michael@0:     if((canonValue&(CANON_HAS_SET|CANON_VALUE_MASK))==0 && origin!=0) {
michael@0:         // origin is the first character whose decomposition starts with
michael@0:         // the character for which we are setting the value.
michael@0:         utrie2_set32(trie, decompLead, canonValue|origin, &errorCode);
michael@0:     } else {
michael@0:         // origin is not the first character, or it is U+0000.
michael@0:         UnicodeSet *set;
michael@0:         if((canonValue&CANON_HAS_SET)==0) {
michael@0:             set=new UnicodeSet;
michael@0:             if(set==NULL) {
michael@0:                 errorCode=U_MEMORY_ALLOCATION_ERROR;
michael@0:                 return;
michael@0:             }
michael@0:             UChar32 firstOrigin=(UChar32)(canonValue&CANON_VALUE_MASK);
michael@0:             canonValue=(canonValue&~CANON_VALUE_MASK)|CANON_HAS_SET|(uint32_t)canonStartSets.size();
michael@0:             utrie2_set32(trie, decompLead, canonValue, &errorCode);
michael@0:             canonStartSets.addElement(set, errorCode);
michael@0:             if(firstOrigin!=0) {
michael@0:                 set->add(firstOrigin);
michael@0:             }
michael@0:         } else {
michael@0:             set=(UnicodeSet *)canonStartSets[(int32_t)(canonValue&CANON_VALUE_MASK)];
michael@0:         }
michael@0:         set->add(origin);
michael@0:     }
michael@0: }
michael@0: 
michael@0: U_CDECL_BEGIN
michael@0: 
michael@0: // Call Normalizer2Impl::makeCanonIterDataFromNorm16() for a range of same-norm16 characters.
michael@0: //     context: the Normalizer2Impl
michael@0: static UBool U_CALLCONV
michael@0: enumCIDRangeHandler(const void *context, UChar32 start, UChar32 end, uint32_t value) {
michael@0:     UErrorCode errorCode = U_ZERO_ERROR;
michael@0:     if (value != 0) {
michael@0:         Normalizer2Impl *impl = (Normalizer2Impl *)context;
michael@0:         impl->makeCanonIterDataFromNorm16(
michael@0:             start, end, (uint16_t)value, *impl->fCanonIterData, errorCode);
michael@0:     }
michael@0:     return U_SUCCESS(errorCode);
michael@0: }
michael@0: 
michael@0: 
michael@0: 
michael@0: // UInitOnce instantiation function for CanonIterData
michael@0: 
michael@0: static void U_CALLCONV 
michael@0: initCanonIterData(Normalizer2Impl *impl, UErrorCode &errorCode) {
michael@0:     U_ASSERT(impl->fCanonIterData == NULL);
michael@0:     impl->fCanonIterData = new CanonIterData(errorCode);
michael@0:     if (impl->fCanonIterData == NULL) {
michael@0:         errorCode=U_MEMORY_ALLOCATION_ERROR;
michael@0:     }
michael@0:     if (U_SUCCESS(errorCode)) {
michael@0:         utrie2_enum(impl->getNormTrie(), NULL, enumCIDRangeHandler, impl);
michael@0:         utrie2_freeze(impl->fCanonIterData->trie, UTRIE2_32_VALUE_BITS, &errorCode);
michael@0:     }
michael@0:     if (U_FAILURE(errorCode)) {
michael@0:         delete impl->fCanonIterData;
michael@0:         impl->fCanonIterData = NULL;
michael@0:     }
michael@0: }
michael@0: 
michael@0: U_CDECL_END
michael@0: 
michael@0: void Normalizer2Impl::makeCanonIterDataFromNorm16(UChar32 start, UChar32 end, uint16_t norm16,
michael@0:                                                   CanonIterData &newData,
michael@0:                                                   UErrorCode &errorCode) const {
michael@0:     if(norm16==0 || (minYesNo<=norm16 && norm16<minNoNo)) {
michael@0:         // Inert, or 2-way mapping (including Hangul syllable).
michael@0:         // We do not write a canonStartSet for any yesNo character.
michael@0:         // Composites from 2-way mappings are added at runtime from the
michael@0:         // starter's compositions list, and the other characters in
michael@0:         // 2-way mappings get CANON_NOT_SEGMENT_STARTER set because they are
michael@0:         // "maybe" characters.
michael@0:         return;
michael@0:     }
michael@0:     for(UChar32 c=start; c<=end; ++c) {
michael@0:         uint32_t oldValue=utrie2_get32(newData.trie, c);
michael@0:         uint32_t newValue=oldValue;
michael@0:         if(norm16>=minMaybeYes) {
michael@0:             // not a segment starter if it occurs in a decomposition or has cc!=0
michael@0:             newValue|=CANON_NOT_SEGMENT_STARTER;
michael@0:             if(norm16<MIN_NORMAL_MAYBE_YES) {
michael@0:                 newValue|=CANON_HAS_COMPOSITIONS;
michael@0:             }
michael@0:         } else if(norm16<minYesNo) {
michael@0:             newValue|=CANON_HAS_COMPOSITIONS;
michael@0:         } else {
michael@0:             // c has a one-way decomposition
michael@0:             UChar32 c2=c;
michael@0:             uint16_t norm16_2=norm16;
michael@0:             while(limitNoNo<=norm16_2 && norm16_2<minMaybeYes) {
michael@0:                 c2=mapAlgorithmic(c2, norm16_2);
michael@0:                 norm16_2=getNorm16(c2);
michael@0:             }
michael@0:             if(minYesNo<=norm16_2 && norm16_2<limitNoNo) {
michael@0:                 // c decomposes, get everything from the variable-length extra data
michael@0:                 const uint16_t *mapping=getMapping(norm16_2);
michael@0:                 uint16_t firstUnit=*mapping;
michael@0:                 int32_t length=firstUnit&MAPPING_LENGTH_MASK;
michael@0:                 if((firstUnit&MAPPING_HAS_CCC_LCCC_WORD)!=0) {
michael@0:                     if(c==c2 && (*(mapping-1)&0xff)!=0) {
michael@0:                         newValue|=CANON_NOT_SEGMENT_STARTER;  // original c has cc!=0
michael@0:                     }
michael@0:                 }
michael@0:                 // Skip empty mappings (no characters in the decomposition).
michael@0:                 if(length!=0) {
michael@0:                     ++mapping;  // skip over the firstUnit
michael@0:                     // add c to first code point's start set
michael@0:                     int32_t i=0;
michael@0:                     U16_NEXT_UNSAFE(mapping, i, c2);
michael@0:                     newData.addToStartSet(c, c2, errorCode);
michael@0:                     // Set CANON_NOT_SEGMENT_STARTER for each remaining code point of a
michael@0:                     // one-way mapping. A 2-way mapping is possible here after
michael@0:                     // intermediate algorithmic mapping.
michael@0:                     if(norm16_2>=minNoNo) {
michael@0:                         while(i<length) {
michael@0:                             U16_NEXT_UNSAFE(mapping, i, c2);
michael@0:                             uint32_t c2Value=utrie2_get32(newData.trie, c2);
michael@0:                             if((c2Value&CANON_NOT_SEGMENT_STARTER)==0) {
michael@0:                                 utrie2_set32(newData.trie, c2, c2Value|CANON_NOT_SEGMENT_STARTER,
michael@0:                                              &errorCode);
michael@0:                             }
michael@0:                         }
michael@0:                     }
michael@0:                 }
michael@0:             } else {
michael@0:                 // c decomposed to c2 algorithmically; c has cc==0
michael@0:                 newData.addToStartSet(c, c2, errorCode);
michael@0:             }
michael@0:         }
michael@0:         if(newValue!=oldValue) {
michael@0:             utrie2_set32(newData.trie, c, newValue, &errorCode);
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: UBool Normalizer2Impl::ensureCanonIterData(UErrorCode &errorCode) const {
michael@0:     // Logically const: Synchronized instantiation.
michael@0:     Normalizer2Impl *me=const_cast<Normalizer2Impl *>(this);
michael@0:     umtx_initOnce(me->fCanonIterDataInitOnce, &initCanonIterData, me, errorCode);
michael@0:     return U_SUCCESS(errorCode);
michael@0: }
michael@0: 
michael@0: int32_t Normalizer2Impl::getCanonValue(UChar32 c) const {
michael@0:     return (int32_t)utrie2_get32(fCanonIterData->trie, c);
michael@0: }
michael@0: 
michael@0: const UnicodeSet &Normalizer2Impl::getCanonStartSet(int32_t n) const {
michael@0:     return *(const UnicodeSet *)fCanonIterData->canonStartSets[n];
michael@0: }
michael@0: 
michael@0: UBool Normalizer2Impl::isCanonSegmentStarter(UChar32 c) const {
michael@0:     return getCanonValue(c)>=0;
michael@0: }
michael@0: 
michael@0: UBool Normalizer2Impl::getCanonStartSet(UChar32 c, UnicodeSet &set) const {
michael@0:     int32_t canonValue=getCanonValue(c)&~CANON_NOT_SEGMENT_STARTER;
michael@0:     if(canonValue==0) {
michael@0:         return FALSE;
michael@0:     }
michael@0:     set.clear();
michael@0:     int32_t value=canonValue&CANON_VALUE_MASK;
michael@0:     if((canonValue&CANON_HAS_SET)!=0) {
michael@0:         set.addAll(getCanonStartSet(value));
michael@0:     } else if(value!=0) {
michael@0:         set.add(value);
michael@0:     }
michael@0:     if((canonValue&CANON_HAS_COMPOSITIONS)!=0) {
michael@0:         uint16_t norm16=getNorm16(c);
michael@0:         if(norm16==JAMO_L) {
michael@0:             UChar32 syllable=
michael@0:                 (UChar32)(Hangul::HANGUL_BASE+(c-Hangul::JAMO_L_BASE)*Hangul::JAMO_VT_COUNT);
michael@0:             set.add(syllable, syllable+Hangul::JAMO_VT_COUNT-1);
michael@0:         } else {
michael@0:             addComposites(getCompositionsList(norm16), set);
michael@0:         }
michael@0:     }
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: U_NAMESPACE_END
michael@0: 
michael@0: // Normalizer2 data swapping ----------------------------------------------- ***
michael@0: 
michael@0: U_NAMESPACE_USE
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: unorm2_swap(const UDataSwapper *ds,
michael@0:             const void *inData, int32_t length, void *outData,
michael@0:             UErrorCode *pErrorCode) {
michael@0:     const UDataInfo *pInfo;
michael@0:     int32_t headerSize;
michael@0: 
michael@0:     const uint8_t *inBytes;
michael@0:     uint8_t *outBytes;
michael@0: 
michael@0:     const int32_t *inIndexes;
michael@0:     int32_t indexes[Normalizer2Impl::IX_MIN_MAYBE_YES+1];
michael@0: 
michael@0:     int32_t i, offset, nextOffset, size;
michael@0: 
michael@0:     /* udata_swapDataHeader checks the arguments */
michael@0:     headerSize=udata_swapDataHeader(ds, inData, length, outData, pErrorCode);
michael@0:     if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
michael@0:         return 0;
michael@0:     }
michael@0: 
michael@0:     /* check data format and format version */
michael@0:     pInfo=(const UDataInfo *)((const char *)inData+4);
michael@0:     if(!(
michael@0:         pInfo->dataFormat[0]==0x4e &&   /* dataFormat="Nrm2" */
michael@0:         pInfo->dataFormat[1]==0x72 &&
michael@0:         pInfo->dataFormat[2]==0x6d &&
michael@0:         pInfo->dataFormat[3]==0x32 &&
michael@0:         (pInfo->formatVersion[0]==1 || pInfo->formatVersion[0]==2)
michael@0:     )) {
michael@0:         udata_printError(ds, "unorm2_swap(): data format %02x.%02x.%02x.%02x (format version %02x) is not recognized as Normalizer2 data\n",
michael@0:                          pInfo->dataFormat[0], pInfo->dataFormat[1],
michael@0:                          pInfo->dataFormat[2], pInfo->dataFormat[3],
michael@0:                          pInfo->formatVersion[0]);
michael@0:         *pErrorCode=U_UNSUPPORTED_ERROR;
michael@0:         return 0;
michael@0:     }
michael@0: 
michael@0:     inBytes=(const uint8_t *)inData+headerSize;
michael@0:     outBytes=(uint8_t *)outData+headerSize;
michael@0: 
michael@0:     inIndexes=(const int32_t *)inBytes;
michael@0: 
michael@0:     if(length>=0) {
michael@0:         length-=headerSize;
michael@0:         if(length<(int32_t)sizeof(indexes)) {
michael@0:             udata_printError(ds, "unorm2_swap(): too few bytes (%d after header) for Normalizer2 data\n",
michael@0:                              length);
michael@0:             *pErrorCode=U_INDEX_OUTOFBOUNDS_ERROR;
michael@0:             return 0;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /* read the first few indexes */
michael@0:     for(i=0; i<=Normalizer2Impl::IX_MIN_MAYBE_YES; ++i) {
michael@0:         indexes[i]=udata_readInt32(ds, inIndexes[i]);
michael@0:     }
michael@0: 
michael@0:     /* get the total length of the data */
michael@0:     size=indexes[Normalizer2Impl::IX_TOTAL_SIZE];
michael@0: 
michael@0:     if(length>=0) {
michael@0:         if(length<size) {
michael@0:             udata_printError(ds, "unorm2_swap(): too few bytes (%d after header) for all of Normalizer2 data\n",
michael@0:                              length);
michael@0:             *pErrorCode=U_INDEX_OUTOFBOUNDS_ERROR;
michael@0:             return 0;
michael@0:         }
michael@0: 
michael@0:         /* copy the data for inaccessible bytes */
michael@0:         if(inBytes!=outBytes) {
michael@0:             uprv_memcpy(outBytes, inBytes, size);
michael@0:         }
michael@0: 
michael@0:         offset=0;
michael@0: 
michael@0:         /* swap the int32_t indexes[] */
michael@0:         nextOffset=indexes[Normalizer2Impl::IX_NORM_TRIE_OFFSET];
michael@0:         ds->swapArray32(ds, inBytes, nextOffset-offset, outBytes, pErrorCode);
michael@0:         offset=nextOffset;
michael@0: 
michael@0:         /* swap the UTrie2 */
michael@0:         nextOffset=indexes[Normalizer2Impl::IX_EXTRA_DATA_OFFSET];
michael@0:         utrie2_swap(ds, inBytes+offset, nextOffset-offset, outBytes+offset, pErrorCode);
michael@0:         offset=nextOffset;
michael@0: 
michael@0:         /* swap the uint16_t extraData[] */
michael@0:         nextOffset=indexes[Normalizer2Impl::IX_SMALL_FCD_OFFSET];
michael@0:         ds->swapArray16(ds, inBytes+offset, nextOffset-offset, outBytes+offset, pErrorCode);
michael@0:         offset=nextOffset;
michael@0: 
michael@0:         /* no need to swap the uint8_t smallFCD[] (new in formatVersion 2) */
michael@0:         nextOffset=indexes[Normalizer2Impl::IX_SMALL_FCD_OFFSET+1];
michael@0:         offset=nextOffset;
michael@0: 
michael@0:         U_ASSERT(offset==size);
michael@0:     }
michael@0: 
michael@0:     return headerSize+size;
michael@0: }
michael@0: 
michael@0: #endif  // !UCONFIG_NO_NORMALIZATION