/* ******************************************************************************* * * Copyright (C) 1998-2001, International Business Machines * Corporation and others. All Rights Reserved. * ******************************************************************************* * * Private implementation header for C collation * file name: ucol_imp.h * encoding: US-ASCII * tab size: 8 (not used) * indentation:4 * * created on: 2000dec11 * created by: Vladimir Weinstein * * Modification history * Date Name Comments * 02/16/2001 synwee Added UCOL_GETPREVCE for the use in ucoleitr * 02/27/2001 synwee Added getMaxExpansion data structure in UCollator * 03/02/2001 synwee Added UCOL_IMPLICIT_CE * 03/12/2001 synwee Added pointer start to collIterate. */ #ifndef UCOL_IMP_H #define UCOL_IMP_H #include "unicode/ucol.h" /*#include "ucmpe32.h"*/ #include "utrie.h" #include "unicode/ures.h" #include "unicode/udata.h" /* UDataInfo for UCA mapping table */ static const UDataInfo ucaDataInfo={ sizeof(UDataInfo), 0, U_IS_BIG_ENDIAN, U_CHARSET_FAMILY, sizeof(UChar), 0, {0x55, 0x43, 0x6f, 0x6c}, /* dataFormat="UCol" */ /* 03/26/2002 bumped up version since format has changed */ {2, 0, 0, 0}, /* formatVersion */ {3, 0, 0, 0} /* dataVersion = Unicode Version*/ }; /* UDataInfo for inverse UCA table */ static const UDataInfo invUcaDataInfo={ sizeof(UDataInfo), 0, U_IS_BIG_ENDIAN, U_CHARSET_FAMILY, sizeof(UChar), 0, {0x49, 0x6E, 0x76, 0x43}, /* dataFormat="InvC" */ /* 03/26/2002 bumped up version since format has changed */ {2, 0, 0, 0}, /* formatVersion */ {3, 0, 0, 0} /* dataVersion = Unicode Version*/ }; /* This is the size of the stack allocated buffer for sortkey generation and similar operations */ /* if it is too small, heap allocation will occur.*/ /* you can change this value if you need memory - it will affect the performance, though, since we're going to malloc */ #define UCOL_MAX_BUFFER 128 #define UCOL_PRIMARY_MAX_BUFFER 8*UCOL_MAX_BUFFER #define UCOL_SECONDARY_MAX_BUFFER UCOL_MAX_BUFFER #define UCOL_TERTIARY_MAX_BUFFER UCOL_MAX_BUFFER #define UCOL_CASE_MAX_BUFFER UCOL_MAX_BUFFER/4 #define UCOL_QUAD_MAX_BUFFER 2*UCOL_MAX_BUFFER #define UCOL_NORMALIZATION_GROWTH 2 #define UCOL_NORMALIZATION_MAX_BUFFER UCOL_MAX_BUFFER*UCOL_NORMALIZATION_GROWTH /* This writable buffer is used if we encounter Thai and need to reorder the string on the fly */ /* Sometimes we already have a writable buffer (like in case of normalized strings). */ /* you can change this value to any value >= 4 if you need memory - it will affect the performance, though, since we're going to malloc. Note 3 is the minimum value for Thai collation and 4 is the minimum number for special Jamo */ #define UCOL_WRITABLE_BUFFER_SIZE 256 /* This is the size of the buffer for expansion CE's */ /* In reality we should not have to deal with expm sequences longer then 16 */ /* you can change this value if you need memory */ /* WARNING THIS BUFFER DOES NOT HAVE MALLOC FALLBACK. If you make it too small, you'll get in trouble */ /* Reasonable small value is around 10, if you don't do Arabic or other funky collations that have long expansion sequence */ /* This is the longest expansion sequence we can handle without bombing out */ #define UCOL_EXPAND_CE_BUFFER_SIZE 512 /* synwee :TODO revert back 64*/ /* Unsafe UChar hash table table size. */ /* size is 32 bytes for 1 bit for each latin 1 char + some power of two for */ /* hashing the rest of the chars. Size in bytes */ #define UCOL_UNSAFECP_TABLE_SIZE 1056 /* mask value down to "some power of two"-1 */ /* number of bits, not num of bytes. */ #define UCOL_UNSAFECP_TABLE_MASK 0x1fff /* flags bits for collIterate.flags */ /* */ /* NORM - set for incremental normalize of source string */ #define UCOL_ITER_NORM 1 #define UCOL_ITER_HASLEN 2 /* UCOL_ITER_INNORMBUF - set if the "pos" is in */ /* the writable side buffer, handling */ /* incrementally normalized characters. */ #define UCOL_ITER_INNORMBUF 4 /* UCOL_ITER_ALLOCATED - set if this iterator has */ /* malloced storage to expand a buffer. */ #define UCOL_ITER_ALLOCATED 8 /* UCOL_HIRAGANA_Q - note if the codepoint was hiragana */ #define UCOL_HIRAGANA_Q 16 /* UCOL_WAS_HIRAGANA - set to TRUE if there was a Hiragana */ /* otherwise set to false */ #define UCOL_WAS_HIRAGANA 32 #define NFC_ZERO_CC_BLOCK_LIMIT_ 0x300 struct collIterate { UChar *string; /* Original string */ /* UChar *start; Pointer to the start of the source string. Either points to string or to writableBuffer */ UChar *endp; /* string end ptr. Is undefined for null terminated strings */ UChar *pos; /* This is position in the string. Can be to original or writable buf */ uint32_t *toReturn; /* This is the CE from CEs buffer that should be returned */ uint32_t *CEpos; /* This is the position to which we have stored processed CEs */ UChar *writableBuffer; uint32_t writableBufSize; UChar *fcdPosition; /* Position in the original string to continue FCD check from. */ const UCollator *coll; uint8_t flags; uint8_t origFlags; uint32_t CEs[UCOL_EXPAND_CE_BUFFER_SIZE]; /* This is where we store CEs */ UChar stackWritableBuffer[UCOL_WRITABLE_BUFFER_SIZE]; /* A writable buffer. */ }; /* struct used internally in getSpecial*CE. data similar to collIterate. */ struct collIterateState { UChar *pos; /* This is position in the string. Can be to original or writable buf */ UChar *fcdPosition; /* Position in the original string to continue FCD check from. */ UChar *bufferaddress; /* address of the normalization buffer */ uint32_t buffersize; uint8_t flags; uint8_t origFlags; }; U_CAPI void U_EXPORT2 uprv_init_collIterate(const UCollator *collator, const UChar *sourceString, int32_t sourceLen, collIterate *s); struct UCollationElements { /** * Struct wrapper for source data */ collIterate iteratordata_; /** * Indicates if this data has been reset. */ UBool reset_; /** * Indicates if the data should be deleted. */ UBool isWritable; }; #define UCOL_LEVELTERMINATOR 1 /* mask off anything but primary order */ #define UCOL_PRIMARYORDERMASK 0xffff0000 /* mask off anything but secondary order */ #define UCOL_SECONDARYORDERMASK 0x0000ff00 /* mask off anything but tertiary order */ #define UCOL_TERTIARYORDERMASK 0x000000ff /* primary order shift */ #define UCOL_PRIMARYORDERSHIFT 16 /* secondary order shift */ #define UCOL_SECONDARYORDERSHIFT 8 #define UCOL_BYTE_SIZE_MASK 0xFF #define UCOL_CASE_BYTE_START 0x80 #define UCOL_CASE_SHIFT_START 7 #define UCOL_IGNORABLE 0 /* get weights from a CE */ #define UCOL_PRIMARYORDER(order) (((order) & UCOL_PRIMARYORDERMASK)>> UCOL_PRIMARYORDERSHIFT) #define UCOL_SECONDARYORDER(order) (((order) & UCOL_SECONDARYORDERMASK)>> UCOL_SECONDARYORDERSHIFT) #define UCOL_TERTIARYORDER(order) ((order) & UCOL_TERTIARYORDERMASK) /** * Determine if a character is a Thai vowel (which sorts after * its base consonant). */ #define UCOL_ISTHAIPREVOWEL(ch) ((((uint32_t)(ch) - 0xe40) <= (0xe44 - 0xe40)) || \ (((uint32_t)(ch) - 0xec0) <= (0xec4 - 0xec0))) /** * Determine if a character is a Thai base consonant */ #define UCOL_ISTHAIBASECONSONANT(ch) ((uint32_t)(ch) - 0xe01) <= (0xe2e - 0xe01) #define UCOL_ISJAMO(ch) ((((uint32_t)(ch) - 0x1100) <= (0x1112 - 0x1100)) || \ (((uint32_t)(ch) - 0x1161) <= (0x1175 - 0x1161)) || \ (((uint32_t)(ch) - 0x11A8) <= (0x11C2 - 0x11A8))) #if 0 /* initializes collIterate structure */ /* made as macro to speed up things */ #define init_collIterate(collator, sourceString, sourceLen, s) { \ (s)->start = (s)->string = (s)->pos = (UChar *)(sourceString); \ (s)->endp = (sourceLen) == -1 ? NULL :(UChar *)(sourceString)+(sourceLen); \ (s)->CEpos = (s)->toReturn = (s)->CEs; \ (s)->isThai = TRUE; \ (s)->writableBuffer = (s)->stackWritableBuffer; \ (s)->writableBufSize = UCOL_WRITABLE_BUFFER_SIZE; \ (s)->coll = (collator); \ (s)->fcdPosition = 0; \ (s)->flags = 0; \ if(((collator)->normalizationMode == UCOL_ON)) (s)->flags |= UCOL_ITER_NORM; \ } #endif /* * Macro to get the maximum size of an expansion ending with the argument ce. * Used in the Boyer Moore algorithm. * Note for tailoring, the UCA maxexpansion table has been merged. * Hence we only have to search the tailored collator only. * @param coll const UCollator pointer * @param order last collation element of the expansion sequence * @param result size of the longest expansion with argument collation element * as the last element */ #define UCOL_GETMAXEXPANSION(coll, order, result) { \ const uint32_t *start; \ const uint32_t *limit; \ const uint32_t *mid; \ start = (coll)->endExpansionCE; \ limit = (coll)->lastEndExpansionCE; \ while (start < limit - 1) { \ mid = start + ((limit - start) >> 1); \ if ((order) <= *mid) { \ limit = mid; \ } \ else { \ start = mid; \ } \ } \ if (*start == order) { \ result = *((coll)->expansionCESize + (start - (coll)->endExpansionCE)); \ } \ else if (*limit == order) { \ result = *(coll->expansionCESize + (limit - coll->endExpansionCE)); \ } \ else if ((order & 0xFFFF) == 0x00C0) { \ result = 2; \ } \ else { \ result = 1; \ } \ } U_CFUNC uint32_t ucol_prv_getSpecialCE(const UCollator *coll, UChar ch, uint32_t CE, collIterate *source, UErrorCode *status); U_CFUNC uint32_t ucol_prv_getSpecialPrevCE(const UCollator *coll, UChar ch, uint32_t CE, collIterate *source, UErrorCode *status); U_CAPI uint32_t U_EXPORT2 ucol_getNextCE(const UCollator *coll, collIterate *collationSource, UErrorCode *status); U_CAPI uint32_t U_EXPORT2 ucol_getPrevCE(const UCollator *coll, collIterate *collationSource, UErrorCode *status); /* function used by C++ getCollationKey to prevent restarting the calculation */ U_CFUNC int32_t ucol_getSortKeyWithAllocation(const UCollator *coll, const UChar *source, int32_t sourceLength, uint8_t **pResult, UErrorCode *pErrorCode); /* get some memory */ void *ucol_getABuffer(const UCollator *coll, uint32_t size); /* worker function for generating sortkeys */ U_CFUNC int32_t U_CALLCONV ucol_calcSortKey(const UCollator *coll, const UChar *source, int32_t sourceLength, uint8_t **result, uint32_t resultLength, UBool allocatePrimary, UErrorCode *status); U_CFUNC int32_t U_CALLCONV ucol_calcSortKeySimpleTertiary(const UCollator *coll, const UChar *source, int32_t sourceLength, uint8_t **result, uint32_t resultLength, UBool allocatePrimary, UErrorCode *status); U_CFUNC int32_t ucol_getSortKeySize(const UCollator *coll, collIterate *s, int32_t currentSize, UColAttributeValue strength, int32_t len); /** * Makes a copy of the Collator's rule data. The format is * that of .col files. * * @param length returns the length of the data, in bytes. * @param status the error status * @return memory, owned by the caller, of size 'length' bytes. * @internal INTERNAL USE ONLY */ U_CAPI uint8_t* U_EXPORT2 ucol_cloneRuleData(const UCollator *coll, int32_t *length, UErrorCode *status); #define UCOL_SPECIAL_FLAG 0xF0000000 #define UCOL_TAG_SHIFT 24 #define UCOL_TAG_MASK 0x0F000000 #define INIT_EXP_TABLE_SIZE 1024 #define UCOL_NOT_FOUND 0xF0000000 #define UCOL_EXPANSION 0xF1000000 #define UCOL_CONTRACTION 0xF2000000 #define UCOL_THAI 0xF3000000 #define UCOL_UNMARKED 0x03 #define UCOL_NEW_TERTIARYORDERMASK 0x0000003f /* Bit mask for primary collation strength. */ #define UCOL_PRIMARYMASK 0xFFFF0000 /* Bit mask for secondary collation strength. */ #define UCOL_SECONDARYMASK 0x0000FF00 /* Bit mask for tertiary collation strength. */ #define UCOL_TERTIARYMASK 0x000000FF /** * Internal. * This indicates the last element in a UCollationElements has been consumed. * Compare with the UCOL_NULLORDER, UCOL_NULLORDER is returned if error occurs. */ #define UCOL_NO_MORE_CES 0x00010101 #define UCOL_NO_MORE_CES_PRIMARY 0x00010000 #define UCOL_NO_MORE_CES_SECONDARY 0x00000100 #define UCOL_NO_MORE_CES_TERTIARY 0x00000001 #define isSpecial(CE) ((((CE)&UCOL_SPECIAL_FLAG)>>28)==0xF) #define UCOL_UPPER_CASE 0x80 #define UCOL_MIXED_CASE 0x40 #define UCOL_LOWER_CASE 0x00 #define UCOL_CONTINUATION_MARKER 0xC0 #define UCOL_REMOVE_CONTINUATION 0xFFFFFF3F #define isContinuation(CE) (((CE) & UCOL_CONTINUATION_MARKER) == UCOL_CONTINUATION_MARKER) #define isFlagged(CE) (((CE) & 0x80) == 0x80) #define isLongPrimary(CE) (((CE) & 0xC0) == 0xC0) #define getCETag(CE) (((CE)&UCOL_TAG_MASK)>>UCOL_TAG_SHIFT) #define isContraction(CE) (isSpecial((CE)) && (getCETag((CE)) == CONTRACTION_TAG)) #define isPrefix(CE) (isSpecial((CE)) && (getCETag((CE)) == SPEC_PROC_TAG)) #define constructContractCE(tag, CE) (UCOL_SPECIAL_FLAG | ((tag)<>4) #define getExpansionCount(CE) ((CE)&0xF) #define isCEIgnorable(CE) (((CE) & 0xFFFFFFBF) == 0) /* StringSearch internal use */ #define inNormBuf(coleiter) ((coleiter)->iteratordata_.flags & UCOL_ITER_INNORMBUF) #define isFCDPointerNull(coleiter) ((coleiter)->iteratordata_.fcdPosition == NULL) #define hasExpansion(coleiter) ((coleiter)->iteratordata_.CEpos != (coleiter)->iteratordata_.CEs) #define getExpansionPrefix(coleiter) ((coleiter)->iteratordata_.toReturn - (coleiter)->iteratordata_.CEs) #define setExpansionPrefix(coleiter, offset) ((coleiter)->iteratordata_.CEs + offset) #define getExpansionSuffix(coleiter) ((coleiter)->iteratordata_.CEpos - (coleiter)->iteratordata_.toReturn) #define setExpansionSuffix(coleiter, offset) ((coleiter)->iteratordata_.toReturn = (coleiter)->iteratordata_.CEpos - leftoverces) #define UCA_DATA_TYPE "icu" #define UCA_DATA_NAME "ucadata" #define INVC_DATA_TYPE "icu" #define INVC_DATA_NAME "invuca" /* This is an enum that lists magic special byte values from the fractional UCA */ /* TODO: all the #defines that refer to special byte values from the UCA should be changed to point here */ enum { UCOL_BYTE_ZERO = 0x00, UCOL_BYTE_LEVEL_SEPARATOR = 0x01, UCOL_BYTE_SORTKEY_GLUE = 0x02, UCOL_BYTE_SHIFT_PREFIX = 0x03, UCOL_BYTE_UNSHIFTED_MIN = UCOL_BYTE_SHIFT_PREFIX, UCOL_BYTE_FIRST_TAILORED = 0x04, UCOL_BYTE_COMMON = 0x05, UCOL_BYTE_FIRST_UCA = UCOL_BYTE_COMMON, UCOL_BYTE_LAST_LATIN_PRIMARY = 0x4C, UCOL_BYTE_FIRST_NON_LATIN_PRIMARY = 0x4D, UCOL_BYTE_UNSHIFTED_MAX = 0xFF }; #if 0 #define UCOL_RESET_TOP_VALUE 0x9F000303 #define UCOL_FIRST_PRIMARY_IGNORABLE 0x00008705 #define UCOL_LAST_PRIMARY_IGNORABLE 0x0000DD05 #define UCOL_LAST_PRIMARY_IGNORABLE_CONT 0x000051C0 #define UCOL_FIRST_SECONDARY_IGNORABLE 0x00000000 #define UCOL_LAST_SECONDARY_IGNORABLE 0x00000500 #define UCOL_FIRST_TERTIARY_IGNORABLE 0x00000000 #define UCOL_LAST_TERTIARY_IGNORABLE 0x00000000 #define UCOL_FIRST_VARIABLE 0x05070505 #define UCOL_LAST_VARIABLE 0x179B0505 #define UCOL_FIRST_NON_VARIABLE 0x1A200505 #define UCOL_LAST_NON_VARIABLE 0x7B41058F #define UCOL_NEXT_TOP_VALUE 0xE8960303 #define UCOL_NEXT_FIRST_PRIMARY_IGNORABLE 0x00008905 #define UCOL_NEXT_LAST_PRIMARY_IGNORABLE 0x03000303 #define UCOL_NEXT_FIRST_SECONDARY_IGNORABLE 0x00008705 #define UCOL_NEXT_LAST_SECONDARY_IGNORABLE 0x00000500 #define UCOL_NEXT_FIRST_TERTIARY_IGNORABLE 0x00000000 #define UCOL_NEXT_LAST_TERTIARY_IGNORABLE 0x00000000 #define UCOL_NEXT_FIRST_VARIABLE 0x05090505 #define UCOL_NEXT_LAST_VARIABLE 0x1A200505 #define PRIMARY_IMPLICIT_MIN 0xE8000000 #define PRIMARY_IMPLICIT_MAX 0xF0000000 #endif /* These constants can be changed - sortkey size is affected by them */ #define UCOL_PROPORTION2 0.5 #define UCOL_PROPORTION3 0.667 /* These values come from the UCA */ #define UCOL_COMMON_BOT2 UCOL_BYTE_COMMON #define UCOL_COMMON_TOP2 0x86u #define UCOL_TOTAL2 (UCOL_COMMON_TOP2-UCOL_COMMON_BOT2-1) #define UCOL_FLAG_BIT_MASK_CASE_SW_OFF 0x80 #define UCOL_FLAG_BIT_MASK_CASE_SW_ON 0x40 #define UCOL_COMMON_TOP3_CASE_SW_OFF 0x85 #define UCOL_COMMON_TOP3_CASE_SW_LOWER 0x45 #define UCOL_COMMON_TOP3_CASE_SW_UPPER 0xC5 /* These values come from the UCA */ #define UCOL_COMMON_BOT3 0x05 #define UCOL_COMMON_BOTTOM3_CASE_SW_UPPER 0x86; #define UCOL_COMMON_BOTTOM3_CASE_SW_LOWER UCOL_COMMON_BOT3; #define UCOL_TOP_COUNT2 (UCOL_PROPORTION2*UCOL_TOTAL2) #define UCOL_BOT_COUNT2 (UCOL_TOTAL2-UCOL_TOP_COUNT2) #define UCOL_COMMON2 UCOL_COMMON_BOT2 #define UCOL_COMMON3_UPPERFIRST 0xC5 #define UCOL_COMMON3_NORMAL UCOL_COMMON_BOT3 #define UCOL_COMMON4 0xFF /* constants for case level/case first handling */ /* used to instantiate UCollators fields in ucol_updateInternalState */ #define UCOL_CASE_SWITCH 0xC0 #define UCOL_NO_CASE_SWITCH 0x00 #define UCOL_REMOVE_CASE 0x3F #define UCOL_KEEP_CASE 0xFF #define UCOL_CASE_BIT_MASK 0xC0 #define UCOL_TERT_CASE_MASK 0xFF typedef enum { NOT_FOUND_TAG = 0, EXPANSION_TAG = 1, /* This code point results in an expansion */ CONTRACTION_TAG = 2, /* Start of a contraction */ THAI_TAG = 3, /* Thai character - do the reordering */ CHARSET_TAG = 4, /* Charset processing, not yet implemented */ SURROGATE_TAG = 5, /* Lead surrogate that is tailored and doesn't start a contraction */ HANGUL_SYLLABLE_TAG = 6, /* AC00-D7AF*/ LEAD_SURROGATE_TAG = 7, /* D800-DBFF*/ TRAIL_SURROGATE_TAG = 8, /* DC00-DFFF*/ CJK_IMPLICIT_TAG = 9, /* 0x3400-0x4DB5, 0x4E00-0x9FA5, 0xF900-0xFA2D*/ IMPLICIT_TAG = 10, SPEC_PROC_TAG = 11, /* ICU 2.1 */ LONG_PRIMARY_TAG = 12, /* This is a three byte primary with starting secondaries and tertiaries */ /* It fits in a single 32 bit CE and is used instead of expansion to save */ /* space without affecting the performance (hopefully) */ CE_TAGS_COUNT } UColCETags; /* ***************************************************************************************** * set to zero * NON_CHARACTER FDD0 - FDEF, FFFE, FFFF, 1FFFE, 1FFFF, 2FFFE, 2FFFF,...e.g. **FFFE, **FFFF ****************************************************************************************** */ typedef struct { uint32_t variableTopValue; UColAttributeValue frenchCollation; UColAttributeValue alternateHandling; /* attribute for handling variable elements*/ UColAttributeValue caseFirst; /* who goes first, lower case or uppercase */ UColAttributeValue caseLevel; /* do we have an extra case level */ UColAttributeValue normalizationMode; /* attribute for normalization */ UColAttributeValue strength; /* attribute for strength */ UColAttributeValue hiraganaQ; /* attribuge for special Hiragana */ } UColOptionSet; typedef struct { uint32_t UCA_FIRST_TERTIARY_IGNORABLE[2]; /*0x00000000*/ uint32_t UCA_LAST_TERTIARY_IGNORABLE[2]; /*0x00000000*/ uint32_t UCA_FIRST_PRIMARY_IGNORABLE[2]; /*0x00008705*/ uint32_t UCA_FIRST_SECONDARY_IGNORABLE[2]; /*0x00000000*/ uint32_t UCA_LAST_SECONDARY_IGNORABLE[2]; /*0x00000500*/ uint32_t UCA_LAST_PRIMARY_IGNORABLE[2]; /*0x0000DD05*/ uint32_t UCA_FIRST_VARIABLE[2]; /*0x05070505*/ uint32_t UCA_LAST_VARIABLE[2]; /*0x13CF0505*/ uint32_t UCA_FIRST_NON_VARIABLE[2]; /*0x16200505*/ uint32_t UCA_LAST_NON_VARIABLE[2]; /*0x767C0505*/ uint32_t UCA_RESET_TOP_VALUE[2]; /*0x9F000303*/ uint32_t UCA_FIRST_IMPLICIT[2]; uint32_t UCA_LAST_IMPLICIT[2]; uint32_t UCA_FIRST_TRAILING[2]; uint32_t UCA_LAST_TRAILING[2]; #if 0 uint32_t UCA_NEXT_TOP_VALUE[2]; /*0xE8960303*/ uint32_t UCA_NEXT_FIRST_PRIMARY_IGNORABLE; /*0x00008905*/ uint32_t UCA_NEXT_LAST_PRIMARY_IGNORABLE; /*0x03000303*/ uint32_t UCA_NEXT_FIRST_SECONDARY_IGNORABLE; /*0x00008705*/ uint32_t UCA_NEXT_LAST_SECONDARY_IGNORABLE; /*0x00000500*/ uint32_t UCA_NEXT_FIRST_TERTIARY_IGNORABLE; /*0x00000000*/ uint32_t UCA_NEXT_LAST_TERTIARY_IGNORABLE; /*0x00000000*/ uint32_t UCA_NEXT_FIRST_VARIABLE; /*0x05090505*/ uint32_t UCA_NEXT_LAST_VARIABLE; /*0x16200505*/ #endif uint32_t UCA_PRIMARY_TOP_MIN; uint32_t UCA_PRIMARY_IMPLICIT_MIN; /*0xE8000000*/ uint32_t UCA_PRIMARY_IMPLICIT_MAX; /*0xF0000000*/ uint32_t UCA_PRIMARY_TRAILING_MIN; /*0xE8000000*/ uint32_t UCA_PRIMARY_TRAILING_MAX; /*0xF0000000*/ uint32_t UCA_PRIMARY_SPECIAL_MIN; /*0xE8000000*/ uint32_t UCA_PRIMARY_SPECIAL_MAX; /*0xF0000000*/ } UCAConstants; typedef struct { int32_t size; /* all the offsets are in bytes */ /* to get the address add to the header address and cast properly */ uint32_t options; /* these are the default options for the collator */ uint32_t UCAConsts; /* structure which holds values for indirect positioning and implicit ranges */ uint32_t contractionUCACombos; /* this one is needed only for UCA, to copy the appropriate contractions */ uint32_t unusedReserved1; /* reserved for future use */ uint32_t mappingPosition; /* const uint8_t *mappingPosition; */ uint32_t expansion; /* uint32_t *expansion; */ uint32_t contractionIndex; /* UChar *contractionIndex; */ uint32_t contractionCEs; /* uint32_t *contractionCEs; */ uint32_t contractionSize; /* needed for various closures */ /*int32_t latinOneMapping;*/ /* this is now handled in the trie itself *//* fast track to latin1 chars */ uint32_t endExpansionCE; /* array of last collation element in expansion */ uint32_t expansionCESize; /* array of maximum expansion size corresponding to the expansion collation elements with last element in endExpansionCE*/ int32_t endExpansionCECount; /* size of endExpansionCE */ uint32_t unsafeCP; /* hash table of unsafe code points */ uint32_t contrEndCP; /* hash table of final code points */ /* in contractions. */ int32_t CEcount; UBool jamoSpecial; /* is jamoSpecial */ uint8_t padding[3]; /* for guaranteed alignment */ UVersionInfo version; char charsetName[32]; /* for charset CEs */ uint8_t reserved[64]; /* for future use */ } UCATableHeader; #define U_UNKNOWN_STATE 0 #define U_COLLATOR_STATE 0x01 #define U_STATE_LIMIT 0x02 /* This is the first structure in a state */ /* it should be machine independent */ typedef struct { /* this structure is supposed to be readable on all the platforms.*/ /* first 2 fields hold the size of the structure in a platform independent way */ uint8_t sizeLo; uint8_t sizeHi; /* identifying the writing platform */ uint8_t isBigEndian; /* see U_CHARSET_FAMILY values in utypes.h */ uint8_t charsetFamily; /* version of ICU this state structure comes from */ uint8_t icuVersion[4]; /* What is the data following this state */ uint8_t type; /* more stuff to come, keep it on 16 byte boundary */ uint8_t reserved[7]; } UStateStruct; /* This structure follows UStatusStruct */ /* and contains data specific for the collators */ /* Endianess needs to be decided before accessing this structure */ /* However, it's size IS endianess independent */ typedef struct { /* size of this structure */ uint8_t sizeLo; uint8_t sizeHi; /* This state is followed by the frozen tailoring */ uint8_t containsTailoring; /* This state is followed by the frozen UCA */ uint8_t containsUCA; /* Version info - the same one */ uint8_t versionInfo[4]; /* for charset CEs */ uint8_t charsetName[32]; /* this is the resolved locale name*/ uint8_t locale[32]; /* Attributes. Open ended */ /* all the following will be moved to uint32_t because of portability */ /* variable top value */ uint32_t variableTopValue; /* attribute for handling variable elements*/ uint32_t /*UColAttributeValue*/ alternateHandling; /* how to handle secondary weights */ uint32_t /*UColAttributeValue*/ frenchCollation; /* who goes first, lower case or uppercase */ uint32_t /*UColAttributeValue*/ caseFirst; /* do we have an extra case level */ uint32_t /*UColAttributeValue*/ caseLevel; /* attribute for normalization */ uint32_t /*UColAttributeValue*/ normalizationMode; /* attribute for strength */ uint32_t /*UColAttributeValue*/ strength; /* to be immediately 16 byte aligned */ uint8_t reserved[12]; } UColStateStruct; #define UCOL_INV_SIZEMASK 0xFFF00000 #define UCOL_INV_OFFSETMASK 0x000FFFFF #define UCOL_INV_SHIFTVALUE 20 U_CDECL_BEGIN typedef struct { uint32_t byteSize; uint32_t tableSize; uint32_t contsSize; uint32_t table; uint32_t conts; } InverseTableHeader; typedef int32_t U_CALLCONV SortKeyGenerator(const UCollator *coll, const UChar *source, int32_t sourceLength, uint8_t **result, uint32_t resultLength, UBool allocatePrimary, UErrorCode *status); struct UCollator { UColOptionSet *options; SortKeyGenerator *sortKeyGen; char* requestedLocale; UResourceBundle *rb; UResourceBundle *binary; const UCATableHeader *image; /*CompactEIntArray *mapping;*/ UTrie *mapping; const uint32_t *latinOneMapping; const uint32_t *expansion; const UChar *contractionIndex; const uint32_t *contractionCEs; const uint8_t *scriptOrder; const uint32_t *endExpansionCE; /* array of last ces in an expansion ce. corresponds to expansionCESize */ const uint32_t *lastEndExpansionCE;/* pointer to the last element in endExpansionCE */ const uint8_t *expansionCESize; /* array of the maximum size of a expansion ce with the last ce corresponding to endExpansionCE, terminated with a null */ const uint8_t *unsafeCP; /* unsafe code points hashtable */ const uint8_t *contrEndCP; /* Contraction ending chars hash table */ UChar minUnsafeCP; /* Smallest unsafe Code Point. */ UChar minContrEndCP; /* Smallest code point at end of a contraction */ const UChar *rules; int32_t rulesLength; UErrorCode errorCode; /* internal error code */ uint32_t variableTopValue; UColAttributeValue frenchCollation; UColAttributeValue alternateHandling; /* attribute for handling variable elements*/ UColAttributeValue caseFirst; /* who goes first, lower case or uppercase */ UColAttributeValue caseLevel; /* do we have an extra case level */ UColAttributeValue normalizationMode; /* attribute for normalization */ UColAttributeValue strength; /* attribute for strength */ UColAttributeValue hiraganaQ; /* attribute for Hiragana */ UBool variableTopValueisDefault; UBool frenchCollationisDefault; UBool alternateHandlingisDefault; /* attribute for handling variable elements*/ UBool caseFirstisDefault; /* who goes first, lower case or uppercase */ UBool caseLevelisDefault; /* do we have an extra case level */ UBool normalizationModeisDefault; /* attribute for normalization */ UBool strengthisDefault; /* attribute for strength */ UBool hiraganaQisDefault; /* attribute for Hiragana */ UBool hasRealData; /* some collators have only options, like French, no rules */ /* to speed up things, we use the UCA image, but we don't want it */ /* to run around */ UBool freeOnClose; UBool freeOptionsOnClose; UBool freeRulesOnClose; int32_t tertiaryAddition; /* when switching case, we need to add or subtract different values */ uint8_t caseSwitch; uint8_t tertiaryCommon; uint8_t tertiaryMask; uint8_t tertiaryTop; /* Upper range when compressing */ uint8_t tertiaryBottom; /* Upper range when compressing */ uint8_t tertiaryTopCount; uint8_t tertiaryBottomCount; UDataInfo dataInfo; /* Data info of UCA table */ }; U_CDECL_END /* various internal functions */ U_CFUNC void ucol_initUCA(UErrorCode *status); U_CFUNC UCollator* ucol_initCollator(const UCATableHeader *image, UCollator *fillIn, UErrorCode *status); U_CFUNC void ucol_setOptionsFromHeader(UCollator* result, UColOptionSet * opts, UErrorCode *status); #if 0 U_CFUNC void ucol_putOptionsToHeader(UCollator* result, UColOptionSet * opts, UErrorCode *status); #endif U_CFUNC void ucol_updateInternalState(UCollator *coll); U_CAPI uint32_t U_EXPORT2 ucol_getFirstCE(const UCollator *coll, UChar u, UErrorCode *status); U_CAPI char* U_EXPORT2 ucol_sortKeyToString(const UCollator *coll, const uint8_t *sortkey, char *buffer, uint32_t *len); U_CAPI UBool U_EXPORT2 ucol_isTailored(const UCollator *coll, const UChar u, UErrorCode *status); U_CAPI const InverseTableHeader* U_EXPORT2 ucol_initInverseUCA(UErrorCode *status); U_CAPI int32_t U_EXPORT2 ucol_inv_getNextCE(uint32_t CE, uint32_t contCE, uint32_t *nextCE, uint32_t *nextContCE, uint32_t strength); U_CAPI int32_t U_EXPORT2 ucol_inv_getPrevCE(uint32_t CE, uint32_t contCE, uint32_t *prevCE, uint32_t *prevContCE, uint32_t strength); U_CAPI void U_EXPORT2 uprv_uca_initImplicitConstants(uint32_t baseByte); #ifdef XP_CPLUSPLUS /* * Test whether a character is potentially "unsafe" for use as a collation * starting point. Unsafe chars are those with combining class != 0 plus * those that are the 2nd thru nth character in a contraction sequence. * * Function is in header file because it's used in both collation and string search, * and needs to be inline for performance. */ static inline UBool ucol_unsafeCP(UChar c, const UCollator *coll) { int32_t hash; uint8_t htbyte; if (c < coll->minUnsafeCP) { return FALSE; } hash = c; if (hash >= UCOL_UNSAFECP_TABLE_SIZE*8) { if(UTF_IS_LEAD(c) || UTF_IS_TRAIL(c)) { /* Trail surrogate */ /* These are always considered unsafe. */ return TRUE; } hash = (hash & UCOL_UNSAFECP_TABLE_MASK) + 256; } htbyte = coll->unsafeCP[hash>>3]; return (((htbyte >> (hash & 7)) & 1) == 1); } #endif /* XP_CPLUSPLUS */ #endif