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.h
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: #ifndef __NORMALIZER2IMPL_H__
michael@0: #define __NORMALIZER2IMPL_H__
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/unistr.h"
michael@0: #include "unicode/unorm.h"
michael@0: #include "unicode/utf16.h"
michael@0: #include "mutex.h"
michael@0: #include "uset_imp.h"
michael@0: #include "utrie2.h"
michael@0: 
michael@0: U_NAMESPACE_BEGIN
michael@0: 
michael@0: struct CanonIterData;
michael@0: 
michael@0: class Hangul {
michael@0: public:
michael@0:     /* Korean Hangul and Jamo constants */
michael@0:     enum {
michael@0:         JAMO_L_BASE=0x1100,     /* "lead" jamo */
michael@0:         JAMO_V_BASE=0x1161,     /* "vowel" jamo */
michael@0:         JAMO_T_BASE=0x11a7,     /* "trail" jamo */
michael@0: 
michael@0:         HANGUL_BASE=0xac00,
michael@0: 
michael@0:         JAMO_L_COUNT=19,
michael@0:         JAMO_V_COUNT=21,
michael@0:         JAMO_T_COUNT=28,
michael@0: 
michael@0:         JAMO_VT_COUNT=JAMO_V_COUNT*JAMO_T_COUNT,
michael@0: 
michael@0:         HANGUL_COUNT=JAMO_L_COUNT*JAMO_V_COUNT*JAMO_T_COUNT,
michael@0:         HANGUL_LIMIT=HANGUL_BASE+HANGUL_COUNT
michael@0:     };
michael@0: 
michael@0:     static inline UBool isHangul(UChar32 c) {
michael@0:         return HANGUL_BASE<=c && c<HANGUL_LIMIT;
michael@0:     }
michael@0:     static inline UBool
michael@0:     isHangulWithoutJamoT(UChar c) {
michael@0:         c-=HANGUL_BASE;
michael@0:         return c<HANGUL_COUNT && c%JAMO_T_COUNT==0;
michael@0:     }
michael@0:     static inline UBool isJamoL(UChar32 c) {
michael@0:         return (uint32_t)(c-JAMO_L_BASE)<JAMO_L_COUNT;
michael@0:     }
michael@0:     static inline UBool isJamoV(UChar32 c) {
michael@0:         return (uint32_t)(c-JAMO_V_BASE)<JAMO_V_COUNT;
michael@0:     }
michael@0: 
michael@0:     /**
michael@0:      * Decomposes c, which must be a Hangul syllable, into buffer
michael@0:      * and returns the length of the decomposition (2 or 3).
michael@0:      */
michael@0:     static inline int32_t decompose(UChar32 c, UChar buffer[3]) {
michael@0:         c-=HANGUL_BASE;
michael@0:         UChar32 c2=c%JAMO_T_COUNT;
michael@0:         c/=JAMO_T_COUNT;
michael@0:         buffer[0]=(UChar)(JAMO_L_BASE+c/JAMO_V_COUNT);
michael@0:         buffer[1]=(UChar)(JAMO_V_BASE+c%JAMO_V_COUNT);
michael@0:         if(c2==0) {
michael@0:             return 2;
michael@0:         } else {
michael@0:             buffer[2]=(UChar)(JAMO_T_BASE+c2);
michael@0:             return 3;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /**
michael@0:      * Decomposes c, which must be a Hangul syllable, into buffer.
michael@0:      * This is the raw, not recursive, decomposition. Its length is always 2.
michael@0:      */
michael@0:     static inline void getRawDecomposition(UChar32 c, UChar buffer[2]) {
michael@0:         UChar32 orig=c;
michael@0:         c-=HANGUL_BASE;
michael@0:         UChar32 c2=c%JAMO_T_COUNT;
michael@0:         if(c2==0) {
michael@0:             c/=JAMO_T_COUNT;
michael@0:             buffer[0]=(UChar)(JAMO_L_BASE+c/JAMO_V_COUNT);
michael@0:             buffer[1]=(UChar)(JAMO_V_BASE+c%JAMO_V_COUNT);
michael@0:         } else {
michael@0:             buffer[0]=orig-c2;  // LV syllable
michael@0:             buffer[1]=(UChar)(JAMO_T_BASE+c2);
michael@0:         }
michael@0:     }
michael@0: private:
michael@0:     Hangul();  // no instantiation
michael@0: };
michael@0: 
michael@0: class Normalizer2Impl;
michael@0: 
michael@0: class ReorderingBuffer : public UMemory {
michael@0: public:
michael@0:     ReorderingBuffer(const Normalizer2Impl &ni, UnicodeString &dest) :
michael@0:         impl(ni), str(dest),
michael@0:         start(NULL), reorderStart(NULL), limit(NULL),
michael@0:         remainingCapacity(0), lastCC(0) {}
michael@0:     ~ReorderingBuffer() {
michael@0:         if(start!=NULL) {
michael@0:             str.releaseBuffer((int32_t)(limit-start));
michael@0:         }
michael@0:     }
michael@0:     UBool init(int32_t destCapacity, UErrorCode &errorCode);
michael@0: 
michael@0:     UBool isEmpty() const { return start==limit; }
michael@0:     int32_t length() const { return (int32_t)(limit-start); }
michael@0:     UChar *getStart() { return start; }
michael@0:     UChar *getLimit() { return limit; }
michael@0:     uint8_t getLastCC() const { return lastCC; }
michael@0: 
michael@0:     UBool equals(const UChar *start, const UChar *limit) const;
michael@0: 
michael@0:     // For Hangul composition, replacing the Leading consonant Jamo with the syllable.
michael@0:     void setLastChar(UChar c) {
michael@0:         *(limit-1)=c;
michael@0:     }
michael@0: 
michael@0:     UBool append(UChar32 c, uint8_t cc, UErrorCode &errorCode) {
michael@0:         return (c<=0xffff) ?
michael@0:             appendBMP((UChar)c, cc, errorCode) :
michael@0:             appendSupplementary(c, cc, errorCode);
michael@0:     }
michael@0:     // s must be in NFD, otherwise change the implementation.
michael@0:     UBool append(const UChar *s, int32_t length,
michael@0:                  uint8_t leadCC, uint8_t trailCC,
michael@0:                  UErrorCode &errorCode);
michael@0:     UBool appendBMP(UChar c, uint8_t cc, UErrorCode &errorCode) {
michael@0:         if(remainingCapacity==0 && !resize(1, errorCode)) {
michael@0:             return FALSE;
michael@0:         }
michael@0:         if(lastCC<=cc || cc==0) {
michael@0:             *limit++=c;
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;
michael@0:         return TRUE;
michael@0:     }
michael@0:     UBool appendZeroCC(UChar32 c, UErrorCode &errorCode);
michael@0:     UBool appendZeroCC(const UChar *s, const UChar *sLimit, UErrorCode &errorCode);
michael@0:     void remove();
michael@0:     void removeSuffix(int32_t suffixLength);
michael@0:     void setReorderingLimit(UChar *newLimit) {
michael@0:         remainingCapacity+=(int32_t)(limit-newLimit);
michael@0:         reorderStart=limit=newLimit;
michael@0:         lastCC=0;
michael@0:     }
michael@0:     void copyReorderableSuffixTo(UnicodeString &s) const {
michael@0:         s.setTo(reorderStart, (int32_t)(limit-reorderStart));
michael@0:     }
michael@0: private:
michael@0:     /*
michael@0:      * TODO: Revisit whether it makes sense to track reorderStart.
michael@0:      * It is set to after the last known character with cc<=1,
michael@0:      * which stops previousCC() before it reads that character and looks up its cc.
michael@0:      * previousCC() is normally only called from insert().
michael@0:      * In other words, reorderStart speeds up the insertion of a combining mark
michael@0:      * into a multi-combining mark sequence where it does not belong at the end.
michael@0:      * This might not be worth the trouble.
michael@0:      * On the other hand, it's not a huge amount of trouble.
michael@0:      *
michael@0:      * We probably need it for UNORM_SIMPLE_APPEND.
michael@0:      */
michael@0: 
michael@0:     UBool appendSupplementary(UChar32 c, uint8_t cc, UErrorCode &errorCode);
michael@0:     void insert(UChar32 c, uint8_t cc);
michael@0:     static void writeCodePoint(UChar *p, UChar32 c) {
michael@0:         if(c<=0xffff) {
michael@0:             *p=(UChar)c;
michael@0:         } else {
michael@0:             p[0]=U16_LEAD(c);
michael@0:             p[1]=U16_TRAIL(c);
michael@0:         }
michael@0:     }
michael@0:     UBool resize(int32_t appendLength, UErrorCode &errorCode);
michael@0: 
michael@0:     const Normalizer2Impl &impl;
michael@0:     UnicodeString &str;
michael@0:     UChar *start, *reorderStart, *limit;
michael@0:     int32_t remainingCapacity;
michael@0:     uint8_t lastCC;
michael@0: 
michael@0:     // private backward iterator
michael@0:     void setIterator() { codePointStart=limit; }
michael@0:     void skipPrevious();  // Requires start<codePointStart.
michael@0:     uint8_t previousCC();  // Returns 0 if there is no previous character.
michael@0: 
michael@0:     UChar *codePointStart, *codePointLimit;
michael@0: };
michael@0: 
michael@0: class U_COMMON_API Normalizer2Impl : public UMemory {
michael@0: public:
michael@0:     Normalizer2Impl() : memory(NULL), normTrie(NULL), fCanonIterData(NULL) {
michael@0:         fCanonIterDataInitOnce.reset();
michael@0:     }
michael@0:     ~Normalizer2Impl();
michael@0: 
michael@0:     void load(const char *packageName, const char *name, UErrorCode &errorCode);
michael@0: 
michael@0:     void addPropertyStarts(const USetAdder *sa, UErrorCode &errorCode) const;
michael@0:     void addCanonIterPropertyStarts(const USetAdder *sa, UErrorCode &errorCode) const;
michael@0: 
michael@0:     // low-level properties ------------------------------------------------ ***
michael@0: 
michael@0:     const UTrie2 *getNormTrie() const { return normTrie; }
michael@0: 
michael@0:     UBool ensureCanonIterData(UErrorCode &errorCode) const;
michael@0: 
michael@0:     uint16_t getNorm16(UChar32 c) const { return UTRIE2_GET16(normTrie, c); }
michael@0: 
michael@0:     UNormalizationCheckResult getCompQuickCheck(uint16_t norm16) const {
michael@0:         if(norm16<minNoNo || MIN_YES_YES_WITH_CC<=norm16) {
michael@0:             return UNORM_YES;
michael@0:         } else if(minMaybeYes<=norm16) {
michael@0:             return UNORM_MAYBE;
michael@0:         } else {
michael@0:             return UNORM_NO;
michael@0:         }
michael@0:     }
michael@0:     UBool isCompNo(uint16_t norm16) const { return minNoNo<=norm16 && norm16<minMaybeYes; }
michael@0:     UBool isDecompYes(uint16_t norm16) const { return norm16<minYesNo || minMaybeYes<=norm16; }
michael@0: 
michael@0:     uint8_t getCC(uint16_t norm16) const {
michael@0:         if(norm16>=MIN_NORMAL_MAYBE_YES) {
michael@0:             return (uint8_t)norm16;
michael@0:         }
michael@0:         if(norm16<minNoNo || limitNoNo<=norm16) {
michael@0:             return 0;
michael@0:         }
michael@0:         return getCCFromNoNo(norm16);
michael@0:     }
michael@0:     static uint8_t getCCFromYesOrMaybe(uint16_t norm16) {
michael@0:         return norm16>=MIN_NORMAL_MAYBE_YES ? (uint8_t)norm16 : 0;
michael@0:     }
michael@0: 
michael@0:     /**
michael@0:      * Returns the FCD data for code point c.
michael@0:      * @param c A Unicode code point.
michael@0:      * @return The lccc(c) in bits 15..8 and tccc(c) in bits 7..0.
michael@0:      */
michael@0:     uint16_t getFCD16(UChar32 c) const {
michael@0:         if(c<0) {
michael@0:             return 0;
michael@0:         } else if(c<0x180) {
michael@0:             return tccc180[c];
michael@0:         } else if(c<=0xffff) {
michael@0:             if(!singleLeadMightHaveNonZeroFCD16(c)) { return 0; }
michael@0:         }
michael@0:         return getFCD16FromNormData(c);
michael@0:     }
michael@0:     /**
michael@0:      * Returns the FCD data for the next code point (post-increment).
michael@0:      * Might skip only a lead surrogate rather than the whole surrogate pair if none of
michael@0:      * the supplementary code points associated with the lead surrogate have non-zero FCD data.
michael@0:      * @param s A valid pointer into a string. Requires s!=limit.
michael@0:      * @param limit The end of the string, or NULL.
michael@0:      * @return The lccc(c) in bits 15..8 and tccc(c) in bits 7..0.
michael@0:      */
michael@0:     uint16_t nextFCD16(const UChar *&s, const UChar *limit) const {
michael@0:         UChar32 c=*s++;
michael@0:         if(c<0x180) {
michael@0:             return tccc180[c];
michael@0:         } else if(!singleLeadMightHaveNonZeroFCD16(c)) {
michael@0:             return 0;
michael@0:         }
michael@0:         UChar c2;
michael@0:         if(U16_IS_LEAD(c) && s!=limit && U16_IS_TRAIL(c2=*s)) {
michael@0:             c=U16_GET_SUPPLEMENTARY(c, c2);
michael@0:             ++s;
michael@0:         }
michael@0:         return getFCD16FromNormData(c);
michael@0:     }
michael@0:     /**
michael@0:      * Returns the FCD data for the previous code point (pre-decrement).
michael@0:      * @param start The start of the string.
michael@0:      * @param s A valid pointer into a string. Requires start<s.
michael@0:      * @return The lccc(c) in bits 15..8 and tccc(c) in bits 7..0.
michael@0:      */
michael@0:     uint16_t previousFCD16(const UChar *start, const UChar *&s) const {
michael@0:         UChar32 c=*--s;
michael@0:         if(c<0x180) {
michael@0:             return tccc180[c];
michael@0:         }
michael@0:         if(!U16_IS_TRAIL(c)) {
michael@0:             if(!singleLeadMightHaveNonZeroFCD16(c)) {
michael@0:                 return 0;
michael@0:             }
michael@0:         } else {
michael@0:             UChar c2;
michael@0:             if(start<s && U16_IS_LEAD(c2=*(s-1))) {
michael@0:                 c=U16_GET_SUPPLEMENTARY(c2, c);
michael@0:                 --s;
michael@0:             }
michael@0:         }
michael@0:         return getFCD16FromNormData(c);
michael@0:     }
michael@0: 
michael@0:     /** Returns the FCD data for U+0000<=c<U+0180. */
michael@0:     uint16_t getFCD16FromBelow180(UChar32 c) const { return tccc180[c]; }
michael@0:     /** Returns TRUE if the single-or-lead code unit c might have non-zero FCD data. */
michael@0:     UBool singleLeadMightHaveNonZeroFCD16(UChar32 lead) const {
michael@0:         // 0<=lead<=0xffff
michael@0:         uint8_t bits=smallFCD[lead>>8];
michael@0:         if(bits==0) { return false; }
michael@0:         return (UBool)((bits>>((lead>>5)&7))&1);
michael@0:     }
michael@0:     /** Returns the FCD value from the regular normalization data. */
michael@0:     uint16_t getFCD16FromNormData(UChar32 c) const;
michael@0: 
michael@0:     void makeCanonIterDataFromNorm16(UChar32 start, UChar32 end, uint16_t norm16,
michael@0:                                      CanonIterData &newData, UErrorCode &errorCode) const;
michael@0: 
michael@0:     /**
michael@0:      * Gets the decomposition for one code point.
michael@0:      * @param c code point
michael@0:      * @param buffer out-only buffer for algorithmic decompositions
michael@0:      * @param length out-only, takes the length of the decomposition, if any
michael@0:      * @return pointer to the decomposition, or NULL if none
michael@0:      */
michael@0:     const UChar *getDecomposition(UChar32 c, UChar buffer[4], int32_t &length) const;
michael@0: 
michael@0:     /**
michael@0:      * Gets the raw decomposition for one code point.
michael@0:      * @param c code point
michael@0:      * @param buffer out-only buffer for algorithmic decompositions
michael@0:      * @param length out-only, takes the length of the decomposition, if any
michael@0:      * @return pointer to the decomposition, or NULL if none
michael@0:      */
michael@0:     const UChar *getRawDecomposition(UChar32 c, UChar buffer[30], int32_t &length) const;
michael@0: 
michael@0:     UChar32 composePair(UChar32 a, UChar32 b) const;
michael@0: 
michael@0:     UBool isCanonSegmentStarter(UChar32 c) const;
michael@0:     UBool getCanonStartSet(UChar32 c, UnicodeSet &set) const;
michael@0: 
michael@0:     enum {
michael@0:         MIN_CCC_LCCC_CP=0x300
michael@0:     };
michael@0: 
michael@0:     enum {
michael@0:         MIN_YES_YES_WITH_CC=0xff01,
michael@0:         JAMO_VT=0xff00,
michael@0:         MIN_NORMAL_MAYBE_YES=0xfe00,
michael@0:         JAMO_L=1,
michael@0:         MAX_DELTA=0x40
michael@0:     };
michael@0: 
michael@0:     enum {
michael@0:         // Byte offsets from the start of the data, after the generic header.
michael@0:         IX_NORM_TRIE_OFFSET,
michael@0:         IX_EXTRA_DATA_OFFSET,
michael@0:         IX_SMALL_FCD_OFFSET,
michael@0:         IX_RESERVED3_OFFSET,
michael@0:         IX_RESERVED4_OFFSET,
michael@0:         IX_RESERVED5_OFFSET,
michael@0:         IX_RESERVED6_OFFSET,
michael@0:         IX_TOTAL_SIZE,
michael@0: 
michael@0:         // Code point thresholds for quick check codes.
michael@0:         IX_MIN_DECOMP_NO_CP,
michael@0:         IX_MIN_COMP_NO_MAYBE_CP,
michael@0: 
michael@0:         // Norm16 value thresholds for quick check combinations and types of extra data.
michael@0:         IX_MIN_YES_NO,  // Mappings & compositions in [minYesNo..minYesNoMappingsOnly[.
michael@0:         IX_MIN_NO_NO,
michael@0:         IX_LIMIT_NO_NO,
michael@0:         IX_MIN_MAYBE_YES,
michael@0: 
michael@0:         IX_MIN_YES_NO_MAPPINGS_ONLY,  // Mappings only in [minYesNoMappingsOnly..minNoNo[.
michael@0: 
michael@0:         IX_RESERVED15,
michael@0:         IX_COUNT
michael@0:     };
michael@0: 
michael@0:     enum {
michael@0:         MAPPING_HAS_CCC_LCCC_WORD=0x80,
michael@0:         MAPPING_HAS_RAW_MAPPING=0x40,
michael@0:         MAPPING_NO_COMP_BOUNDARY_AFTER=0x20,
michael@0:         MAPPING_LENGTH_MASK=0x1f
michael@0:     };
michael@0: 
michael@0:     enum {
michael@0:         COMP_1_LAST_TUPLE=0x8000,
michael@0:         COMP_1_TRIPLE=1,
michael@0:         COMP_1_TRAIL_LIMIT=0x3400,
michael@0:         COMP_1_TRAIL_MASK=0x7ffe,
michael@0:         COMP_1_TRAIL_SHIFT=9,  // 10-1 for the "triple" bit
michael@0:         COMP_2_TRAIL_SHIFT=6,
michael@0:         COMP_2_TRAIL_MASK=0xffc0
michael@0:     };
michael@0: 
michael@0:     // higher-level functionality ------------------------------------------ ***
michael@0: 
michael@0:     const UChar *decompose(const UChar *src, const UChar *limit,
michael@0:                            ReorderingBuffer *buffer, UErrorCode &errorCode) const;
michael@0:     void 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:     UBool 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:     const UChar *composeQuickCheck(const UChar *src, const UChar *limit,
michael@0:                                    UBool onlyContiguous,
michael@0:                                    UNormalizationCheckResult *pQCResult) const;
michael@0:     void 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:     const UChar *makeFCD(const UChar *src, const UChar *limit,
michael@0:                          ReorderingBuffer *buffer, UErrorCode &errorCode) const;
michael@0:     void 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: 
michael@0:     UBool hasDecompBoundary(UChar32 c, UBool before) const;
michael@0:     UBool isDecompInert(UChar32 c) const { return isDecompYesAndZeroCC(getNorm16(c)); }
michael@0: 
michael@0:     UBool hasCompBoundaryBefore(UChar32 c) const {
michael@0:         return c<minCompNoMaybeCP || hasCompBoundaryBefore(c, getNorm16(c));
michael@0:     }
michael@0:     UBool hasCompBoundaryAfter(UChar32 c, UBool onlyContiguous, UBool testInert) const;
michael@0: 
michael@0:     UBool hasFCDBoundaryBefore(UChar32 c) const { return c<MIN_CCC_LCCC_CP || getFCD16(c)<=0xff; }
michael@0:     UBool hasFCDBoundaryAfter(UChar32 c) const {
michael@0:         uint16_t fcd16=getFCD16(c);
michael@0:         return fcd16<=1 || (fcd16&0xff)==0;
michael@0:     }
michael@0:     UBool isFCDInert(UChar32 c) const { return getFCD16(c)<=1; }
michael@0: private:
michael@0:     static UBool U_CALLCONV
michael@0:     isAcceptable(void *context, const char *type, const char *name, const UDataInfo *pInfo);
michael@0: 
michael@0:     UBool isMaybe(uint16_t norm16) const { return minMaybeYes<=norm16 && norm16<=JAMO_VT; }
michael@0:     UBool isMaybeOrNonZeroCC(uint16_t norm16) const { return norm16>=minMaybeYes; }
michael@0:     static UBool isInert(uint16_t norm16) { return norm16==0; }
michael@0:     static UBool isJamoL(uint16_t norm16) { return norm16==1; }
michael@0:     static UBool isJamoVT(uint16_t norm16) { return norm16==JAMO_VT; }
michael@0:     UBool isHangul(uint16_t norm16) const { return norm16==minYesNo; }
michael@0:     UBool isCompYesAndZeroCC(uint16_t norm16) const { return norm16<minNoNo; }
michael@0:     // UBool isCompYes(uint16_t norm16) const {
michael@0:     //     return norm16>=MIN_YES_YES_WITH_CC || norm16<minNoNo;
michael@0:     // }
michael@0:     // UBool isCompYesOrMaybe(uint16_t norm16) const {
michael@0:     //     return norm16<minNoNo || minMaybeYes<=norm16;
michael@0:     // }
michael@0:     // UBool hasZeroCCFromDecompYes(uint16_t norm16) const {
michael@0:     //     return norm16<=MIN_NORMAL_MAYBE_YES || norm16==JAMO_VT;
michael@0:     // }
michael@0:     UBool isDecompYesAndZeroCC(uint16_t norm16) const {
michael@0:         return norm16<minYesNo ||
michael@0:                norm16==JAMO_VT ||
michael@0:                (minMaybeYes<=norm16 && norm16<=MIN_NORMAL_MAYBE_YES);
michael@0:     }
michael@0:     /**
michael@0:      * A little faster and simpler than isDecompYesAndZeroCC() but does not include
michael@0:      * the MaybeYes which combine-forward and have ccc=0.
michael@0:      * (Standard Unicode 5.2 normalization does not have such characters.)
michael@0:      */
michael@0:     UBool isMostDecompYesAndZeroCC(uint16_t norm16) const {
michael@0:         return norm16<minYesNo || norm16==MIN_NORMAL_MAYBE_YES || norm16==JAMO_VT;
michael@0:     }
michael@0:     UBool isDecompNoAlgorithmic(uint16_t norm16) const { return norm16>=limitNoNo; }
michael@0: 
michael@0:     // For use with isCompYes().
michael@0:     // Perhaps the compiler can combine the two tests for MIN_YES_YES_WITH_CC.
michael@0:     // static uint8_t getCCFromYes(uint16_t norm16) {
michael@0:     //     return norm16>=MIN_YES_YES_WITH_CC ? (uint8_t)norm16 : 0;
michael@0:     // }
michael@0:     uint8_t getCCFromNoNo(uint16_t norm16) const {
michael@0:         const uint16_t *mapping=getMapping(norm16);
michael@0:         if(*mapping&MAPPING_HAS_CCC_LCCC_WORD) {
michael@0:             return (uint8_t)*(mapping-1);
michael@0:         } else {
michael@0:             return 0;
michael@0:         }
michael@0:     }
michael@0:     // requires that the [cpStart..cpLimit[ character passes isCompYesAndZeroCC()
michael@0:     uint8_t getTrailCCFromCompYesAndZeroCC(const UChar *cpStart, const UChar *cpLimit) const;
michael@0: 
michael@0:     // Requires algorithmic-NoNo.
michael@0:     UChar32 mapAlgorithmic(UChar32 c, uint16_t norm16) const {
michael@0:         return c+norm16-(minMaybeYes-MAX_DELTA-1);
michael@0:     }
michael@0: 
michael@0:     // Requires minYesNo<norm16<limitNoNo.
michael@0:     const uint16_t *getMapping(uint16_t norm16) const { return extraData+norm16; }
michael@0:     const uint16_t *getCompositionsListForDecompYes(uint16_t norm16) const {
michael@0:         if(norm16==0 || MIN_NORMAL_MAYBE_YES<=norm16) {
michael@0:             return NULL;
michael@0:         } else if(norm16<minMaybeYes) {
michael@0:             return extraData+norm16;  // for yesYes; if Jamo L: harmless empty list
michael@0:         } else {
michael@0:             return maybeYesCompositions+norm16-minMaybeYes;
michael@0:         }
michael@0:     }
michael@0:     const uint16_t *getCompositionsListForComposite(uint16_t norm16) const {
michael@0:         const uint16_t *list=extraData+norm16;  // composite has both mapping & compositions list
michael@0:         return 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:      * @param c code point must have compositions
michael@0:      * @return compositions list pointer
michael@0:      */
michael@0:     const uint16_t *getCompositionsList(uint16_t norm16) const {
michael@0:         return isDecompYes(norm16) ?
michael@0:                 getCompositionsListForDecompYes(norm16) :
michael@0:                 getCompositionsListForComposite(norm16);
michael@0:     }
michael@0: 
michael@0:     const UChar *copyLowPrefixFromNulTerminated(const UChar *src,
michael@0:                                                 UChar32 minNeedDataCP,
michael@0:                                                 ReorderingBuffer *buffer,
michael@0:                                                 UErrorCode &errorCode) const;
michael@0:     UBool decomposeShort(const UChar *src, const UChar *limit,
michael@0:                          ReorderingBuffer &buffer, UErrorCode &errorCode) const;
michael@0:     UBool decompose(UChar32 c, uint16_t norm16,
michael@0:                     ReorderingBuffer &buffer, UErrorCode &errorCode) const;
michael@0: 
michael@0:     static int32_t combine(const uint16_t *list, UChar32 trail);
michael@0:     void addComposites(const uint16_t *list, UnicodeSet &set) const;
michael@0:     void recompose(ReorderingBuffer &buffer, int32_t recomposeStartIndex,
michael@0:                    UBool onlyContiguous) const;
michael@0: 
michael@0:     UBool hasCompBoundaryBefore(UChar32 c, uint16_t norm16) const;
michael@0:     const UChar *findPreviousCompBoundary(const UChar *start, const UChar *p) const;
michael@0:     const UChar *findNextCompBoundary(const UChar *p, const UChar *limit) const;
michael@0: 
michael@0:     const UChar *findPreviousFCDBoundary(const UChar *start, const UChar *p) const;
michael@0:     const UChar *findNextFCDBoundary(const UChar *p, const UChar *limit) const;
michael@0: 
michael@0:     int32_t getCanonValue(UChar32 c) const;
michael@0:     const UnicodeSet &getCanonStartSet(int32_t n) const;
michael@0: 
michael@0:     UDataMemory *memory;
michael@0:     UVersionInfo dataVersion;
michael@0: 
michael@0:     // Code point thresholds for quick check codes.
michael@0:     UChar32 minDecompNoCP;
michael@0:     UChar32 minCompNoMaybeCP;
michael@0: 
michael@0:     // Norm16 value thresholds for quick check combinations and types of extra data.
michael@0:     uint16_t minYesNo;
michael@0:     uint16_t minYesNoMappingsOnly;
michael@0:     uint16_t minNoNo;
michael@0:     uint16_t limitNoNo;
michael@0:     uint16_t minMaybeYes;
michael@0: 
michael@0:     UTrie2 *normTrie;
michael@0:     const uint16_t *maybeYesCompositions;
michael@0:     const uint16_t *extraData;  // mappings and/or compositions for yesYes, yesNo & noNo characters
michael@0:     const uint8_t *smallFCD;  // [0x100] one bit per 32 BMP code points, set if any FCD!=0
michael@0:     uint8_t tccc180[0x180];  // tccc values for U+0000..U+017F
michael@0: 
michael@0:   public:           // CanonIterData is public to allow access from C callback functions.
michael@0:     UInitOnce       fCanonIterDataInitOnce;
michael@0:     CanonIterData  *fCanonIterData;
michael@0: };
michael@0: 
michael@0: // bits in canonIterData
michael@0: #define CANON_NOT_SEGMENT_STARTER 0x80000000
michael@0: #define CANON_HAS_COMPOSITIONS 0x40000000
michael@0: #define CANON_HAS_SET 0x200000
michael@0: #define CANON_VALUE_MASK 0x1fffff
michael@0: 
michael@0: /**
michael@0:  * ICU-internal shortcut for quick access to standard Unicode normalization.
michael@0:  */
michael@0: class U_COMMON_API Normalizer2Factory {
michael@0: public:
michael@0:     static const Normalizer2 *getNFCInstance(UErrorCode &errorCode);
michael@0:     static const Normalizer2 *getNFDInstance(UErrorCode &errorCode);
michael@0:     static const Normalizer2 *getFCDInstance(UErrorCode &errorCode);
michael@0:     static const Normalizer2 *getFCCInstance(UErrorCode &errorCode);
michael@0:     static const Normalizer2 *getNFKCInstance(UErrorCode &errorCode);
michael@0:     static const Normalizer2 *getNFKDInstance(UErrorCode &errorCode);
michael@0:     static const Normalizer2 *getNFKC_CFInstance(UErrorCode &errorCode);
michael@0:     static const Normalizer2 *getNoopInstance(UErrorCode &errorCode);
michael@0: 
michael@0:     static const Normalizer2 *getInstance(UNormalizationMode mode, UErrorCode &errorCode);
michael@0: 
michael@0:     static const Normalizer2Impl *getNFCImpl(UErrorCode &errorCode);
michael@0:     static const Normalizer2Impl *getNFKCImpl(UErrorCode &errorCode);
michael@0:     static const Normalizer2Impl *getNFKC_CFImpl(UErrorCode &errorCode);
michael@0: 
michael@0:     // Get the Impl instance of the Normalizer2.
michael@0:     // Must be used only when it is known that norm2 is a Normalizer2WithImpl instance.
michael@0:     static const Normalizer2Impl *getImpl(const Normalizer2 *norm2);
michael@0: private:
michael@0:     Normalizer2Factory();  // No instantiation.
michael@0: };
michael@0: 
michael@0: U_NAMESPACE_END
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: 
michael@0: /**
michael@0:  * Get the NF*_QC property for a code point, for u_getIntPropertyValue().
michael@0:  * @internal
michael@0:  */
michael@0: U_CFUNC UNormalizationCheckResult
michael@0: unorm_getQuickCheck(UChar32 c, UNormalizationMode mode);
michael@0: 
michael@0: /**
michael@0:  * Gets the 16-bit FCD value (lead & trail CCs) for a code point, for u_getIntPropertyValue().
michael@0:  * @internal
michael@0:  */
michael@0: U_CFUNC uint16_t
michael@0: unorm_getFCD16(UChar32 c);
michael@0: 
michael@0: /**
michael@0:  * Format of Normalizer2 .nrm data files.
michael@0:  * Format version 2.0.
michael@0:  *
michael@0:  * Normalizer2 .nrm data files provide data for the Unicode Normalization algorithms.
michael@0:  * ICU ships with data files for standard Unicode Normalization Forms
michael@0:  * NFC and NFD (nfc.nrm), NFKC and NFKD (nfkc.nrm) and NFKC_Casefold (nfkc_cf.nrm).
michael@0:  * Custom (application-specific) data can be built into additional .nrm files
michael@0:  * with the gennorm2 build tool.
michael@0:  *
michael@0:  * Normalizer2.getInstance() causes a .nrm file to be loaded, unless it has been
michael@0:  * cached already. Internally, Normalizer2Impl.load() reads the .nrm file.
michael@0:  *
michael@0:  * A .nrm file begins with a standard ICU data file header
michael@0:  * (DataHeader, see ucmndata.h and unicode/udata.h).
michael@0:  * The UDataInfo.dataVersion field usually contains the Unicode version
michael@0:  * for which the data was generated.
michael@0:  *
michael@0:  * After the header, the file contains the following parts.
michael@0:  * Constants are defined as enum values of the Normalizer2Impl class.
michael@0:  *
michael@0:  * Many details of the data structures are described in the design doc
michael@0:  * which is at http://site.icu-project.org/design/normalization/custom
michael@0:  *
michael@0:  * int32_t indexes[indexesLength]; -- indexesLength=indexes[IX_NORM_TRIE_OFFSET]/4;
michael@0:  *
michael@0:  *      The first eight indexes are byte offsets in ascending order.
michael@0:  *      Each byte offset marks the start of the next part in the data file,
michael@0:  *      and the end of the previous one.
michael@0:  *      When two consecutive byte offsets are the same, then the corresponding part is empty.
michael@0:  *      Byte offsets are offsets from after the header,
michael@0:  *      that is, from the beginning of the indexes[].
michael@0:  *      Each part starts at an offset with proper alignment for its data.
michael@0:  *      If necessary, the previous part may include padding bytes to achieve this alignment.
michael@0:  *
michael@0:  *      minDecompNoCP=indexes[IX_MIN_DECOMP_NO_CP] is the lowest code point
michael@0:  *      with a decomposition mapping, that is, with NF*D_QC=No.
michael@0:  *      minCompNoMaybeCP=indexes[IX_MIN_COMP_NO_MAYBE_CP] is the lowest code point
michael@0:  *      with NF*C_QC=No (has a one-way mapping) or Maybe (combines backward).
michael@0:  *
michael@0:  *      The next five indexes are thresholds of 16-bit trie values for ranges of
michael@0:  *      values indicating multiple normalization properties.
michael@0:  *          minYesNo=indexes[IX_MIN_YES_NO];
michael@0:  *          minNoNo=indexes[IX_MIN_NO_NO];
michael@0:  *          limitNoNo=indexes[IX_LIMIT_NO_NO];
michael@0:  *          minMaybeYes=indexes[IX_MIN_MAYBE_YES];
michael@0:  *          minYesNoMappingsOnly=indexes[IX_MIN_YES_NO_MAPPINGS_ONLY];
michael@0:  *      See the normTrie description below and the design doc for details.
michael@0:  *
michael@0:  * UTrie2 normTrie; -- see utrie2_impl.h and utrie2.h
michael@0:  *
michael@0:  *      The trie holds the main normalization data. Each code point is mapped to a 16-bit value.
michael@0:  *      Rather than using independent bits in the value (which would require more than 16 bits),
michael@0:  *      information is extracted primarily via range checks.
michael@0:  *      For example, a 16-bit value norm16 in the range minYesNo<=norm16<minNoNo
michael@0:  *      means that the character has NF*C_QC=Yes and NF*D_QC=No properties,
michael@0:  *      which means it has a two-way (round-trip) decomposition mapping.
michael@0:  *      Values in the range 2<=norm16<limitNoNo are also directly indexes into the extraData
michael@0:  *      pointing to mappings, compositions lists, or both.
michael@0:  *      Value norm16==0 means that the character is normalization-inert, that is,
michael@0:  *      it does not have a mapping, does not participate in composition, has a zero
michael@0:  *      canonical combining class, and forms a boundary where text before it and after it
michael@0:  *      can be normalized independently.
michael@0:  *      For details about how multiple properties are encoded in 16-bit values
michael@0:  *      see the design doc.
michael@0:  *      Note that the encoding cannot express all combinations of the properties involved;
michael@0:  *      it only supports those combinations that are allowed by
michael@0:  *      the Unicode Normalization algorithms. Details are in the design doc as well.
michael@0:  *      The gennorm2 tool only builds .nrm files for data that conforms to the limitations.
michael@0:  *
michael@0:  *      The trie has a value for each lead surrogate code unit representing the "worst case"
michael@0:  *      properties of the 1024 supplementary characters whose UTF-16 form starts with
michael@0:  *      the lead surrogate. If all of the 1024 supplementary characters are normalization-inert,
michael@0:  *      then their lead surrogate code unit has the trie value 0.
michael@0:  *      When the lead surrogate unit's value exceeds the quick check minimum during processing,
michael@0:  *      the properties for the full supplementary code point need to be looked up.
michael@0:  *
michael@0:  * uint16_t maybeYesCompositions[MIN_NORMAL_MAYBE_YES-minMaybeYes];
michael@0:  * uint16_t extraData[];
michael@0:  *
michael@0:  *      There is only one byte offset for the end of these two arrays.
michael@0:  *      The split between them is given by the constant and variable mentioned above.
michael@0:  *
michael@0:  *      The maybeYesCompositions array contains compositions lists for characters that
michael@0:  *      combine both forward (as starters in composition pairs)
michael@0:  *      and backward (as trailing characters in composition pairs).
michael@0:  *      Such characters do not occur in Unicode 5.2 but are allowed by
michael@0:  *      the Unicode Normalization algorithms.
michael@0:  *      If there are no such characters, then minMaybeYes==MIN_NORMAL_MAYBE_YES
michael@0:  *      and the maybeYesCompositions array is empty.
michael@0:  *      If there are such characters, then minMaybeYes is subtracted from their norm16 values
michael@0:  *      to get the index into this array.
michael@0:  *
michael@0:  *      The extraData array contains compositions lists for "YesYes" characters,
michael@0:  *      followed by mappings and optional compositions lists for "YesNo" characters,
michael@0:  *      followed by only mappings for "NoNo" characters.
michael@0:  *      (Referring to pairs of NFC/NFD quick check values.)
michael@0:  *      The norm16 values of those characters are directly indexes into the extraData array.
michael@0:  *
michael@0:  *      The data structures for compositions lists and mappings are described in the design doc.
michael@0:  *
michael@0:  * uint8_t smallFCD[0x100]; -- new in format version 2
michael@0:  *
michael@0:  *      This is a bit set to help speed up FCD value lookups in the absence of a full
michael@0:  *      UTrie2 or other large data structure with the full FCD value mapping.
michael@0:  *
michael@0:  *      Each smallFCD bit is set if any of the corresponding 32 BMP code points
michael@0:  *      has a non-zero FCD value (lccc!=0 or tccc!=0).
michael@0:  *      Bit 0 of smallFCD[0] is for U+0000..U+001F. Bit 7 of smallFCD[0xff] is for U+FFE0..U+FFFF.
michael@0:  *      A bit for 32 lead surrogates is set if any of the 32k corresponding
michael@0:  *      _supplementary_ code points has a non-zero FCD value.
michael@0:  *
michael@0:  *      This bit set is most useful for the large blocks of CJK characters with FCD=0.
michael@0:  *
michael@0:  * Changes from format version 1 to format version 2 ---------------------------
michael@0:  *
michael@0:  * - Addition of data for raw (not recursively decomposed) mappings.
michael@0:  *   + The MAPPING_NO_COMP_BOUNDARY_AFTER bit in the extraData is now also set when
michael@0:  *     the mapping is to an empty string or when the character combines-forward.
michael@0:  *     This subsumes the one actual use of the MAPPING_PLUS_COMPOSITION_LIST bit which
michael@0:  *     is then repurposed for the MAPPING_HAS_RAW_MAPPING bit.
michael@0:  *   + For details see the design doc.
michael@0:  * - Addition of indexes[IX_MIN_YES_NO_MAPPINGS_ONLY] and separation of the yesNo extraData into
michael@0:  *   distinct ranges (combines-forward vs. not)
michael@0:  *   so that a range check can be used to find out if there is a compositions list.
michael@0:  *   This is fully equivalent with formatVersion 1's MAPPING_PLUS_COMPOSITION_LIST flag.
michael@0:  *   It is needed for the new (in ICU 49) composePair(), not for other normalization.
michael@0:  * - Addition of the smallFCD[] bit set.
michael@0:  */
michael@0: 
michael@0: #endif  /* !UCONFIG_NO_NORMALIZATION */
michael@0: #endif  /* __NORMALIZER2IMPL_H__ */