michael@0: /* michael@0: ******************************************************************************* michael@0: * michael@0: * Copyright (C) 2001-2011, International Business Machines michael@0: * Corporation and others. All Rights Reserved. michael@0: * michael@0: ******************************************************************************* michael@0: * file name: unormimp.h michael@0: * encoding: US-ASCII michael@0: * tab size: 8 (not used) michael@0: * indentation:4 michael@0: * michael@0: * created on: 2001may25 michael@0: * created by: Markus W. Scherer michael@0: */ michael@0: michael@0: #ifndef __UNORMIMP_H__ michael@0: #define __UNORMIMP_H__ michael@0: michael@0: #include "unicode/utypes.h" michael@0: michael@0: #if !UCONFIG_NO_NORMALIZATION michael@0: michael@0: #include "udataswp.h" michael@0: michael@0: /* michael@0: * The 2001-2010 implementation of the normalization code loads its data from michael@0: * unorm.icu, which is generated with the gennorm tool. michael@0: * The format of that file is described at the end of this file. michael@0: */ michael@0: michael@0: /* norm32 value constants */ michael@0: enum { michael@0: /* quick check flags 0..3 set mean "no" for their forms */ michael@0: _NORM_QC_NFC=0x11, /* no|maybe */ michael@0: _NORM_QC_NFKC=0x22, /* no|maybe */ michael@0: _NORM_QC_NFD=4, /* no */ michael@0: _NORM_QC_NFKD=8, /* no */ michael@0: michael@0: _NORM_QC_ANY_NO=0xf, michael@0: michael@0: /* quick check flags 4..5 mean "maybe" for their forms; test flags>=_NORM_QC_MAYBE */ michael@0: _NORM_QC_MAYBE=0x10, michael@0: _NORM_QC_ANY_MAYBE=0x30, michael@0: michael@0: _NORM_QC_MASK=0x3f, michael@0: michael@0: _NORM_COMBINES_FWD=0x40, michael@0: _NORM_COMBINES_BACK=0x80, michael@0: _NORM_COMBINES_ANY=0xc0, michael@0: michael@0: _NORM_CC_SHIFT=8, /* UnicodeData.txt combining class in bits 15..8 */ michael@0: _NORM_CC_MASK=0xff00, michael@0: michael@0: _NORM_EXTRA_SHIFT=16, /* 16 bits for the index to UChars and other extra data */ michael@0: _NORM_EXTRA_INDEX_TOP=0xfc00, /* start of surrogate specials after shift */ michael@0: michael@0: _NORM_EXTRA_SURROGATE_MASK=0x3ff, michael@0: _NORM_EXTRA_SURROGATE_TOP=0x3f0, /* hangul etc. */ michael@0: michael@0: _NORM_EXTRA_HANGUL=_NORM_EXTRA_SURROGATE_TOP, michael@0: _NORM_EXTRA_JAMO_L, michael@0: _NORM_EXTRA_JAMO_V, michael@0: _NORM_EXTRA_JAMO_T michael@0: }; michael@0: michael@0: /* norm32 value constants using >16 bits */ michael@0: #define _NORM_MIN_SPECIAL 0xfc000000 michael@0: #define _NORM_SURROGATES_TOP 0xfff00000 michael@0: #define _NORM_MIN_HANGUL 0xfff00000 michael@0: #define _NORM_MIN_JAMO_V 0xfff20000 michael@0: #define _NORM_JAMO_V_TOP 0xfff30000 michael@0: michael@0: /* value constants for auxTrie */ michael@0: enum { michael@0: _NORM_AUX_COMP_EX_SHIFT=10, michael@0: _NORM_AUX_UNSAFE_SHIFT=11, michael@0: _NORM_AUX_NFC_SKIPPABLE_F_SHIFT=12 michael@0: }; michael@0: michael@0: #define _NORM_AUX_MAX_FNC ((int32_t)1<<_NORM_AUX_COMP_EX_SHIFT) michael@0: michael@0: #define _NORM_AUX_FNC_MASK (uint32_t)(_NORM_AUX_MAX_FNC-1) michael@0: #define _NORM_AUX_COMP_EX_MASK ((uint32_t)1<<_NORM_AUX_COMP_EX_SHIFT) michael@0: #define _NORM_AUX_UNSAFE_MASK ((uint32_t)1<<_NORM_AUX_UNSAFE_SHIFT) michael@0: #define _NORM_AUX_NFC_SKIP_F_MASK ((uint32_t)1<<_NORM_AUX_NFC_SKIPPABLE_F_SHIFT) michael@0: michael@0: /* canonStartSets[0..31] contains indexes for what is in the array */ michael@0: enum { michael@0: _NORM_SET_INDEX_CANON_SETS_LENGTH, /* number of uint16_t in canonical starter sets */ michael@0: _NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH, /* number of uint16_t in the BMP search table (contains pairs) */ michael@0: _NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH,/* number of uint16_t in the supplementary search table (contains triplets) */ michael@0: michael@0: /* from formatVersion 2.3: */ michael@0: _NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET, /* uint16_t offset from canonStartSets[0] to the michael@0: exclusion set for CJK compatibility characters */ michael@0: _NORM_SET_INDEX_NX_UNICODE32_OFFSET, /* uint16_t offset from canonStartSets[0] to the michael@0: exclusion set for Unicode 3.2 characters */ michael@0: _NORM_SET_INDEX_NX_RESERVED_OFFSET, /* uint16_t offset from canonStartSets[0] to the michael@0: end of the previous exclusion set */ michael@0: michael@0: _NORM_SET_INDEX_TOP=32 /* changing this requires a new formatVersion */ michael@0: }; michael@0: michael@0: /* more constants for canonical starter sets */ michael@0: michael@0: /* 14 bit indexes to canonical USerializedSets */ michael@0: #define _NORM_MAX_CANON_SETS 0x4000 michael@0: michael@0: /* single-code point BMP sets are encoded directly in the search table except if result=0x4000..0x7fff */ michael@0: #define _NORM_CANON_SET_BMP_MASK 0xc000 michael@0: #define _NORM_CANON_SET_BMP_IS_INDEX 0x4000 michael@0: michael@0: /* indexes[] value names */ michael@0: enum { michael@0: _NORM_INDEX_TRIE_SIZE, /* number of bytes in normalization trie */ michael@0: _NORM_INDEX_UCHAR_COUNT, /* number of UChars in extra data */ michael@0: michael@0: _NORM_INDEX_COMBINE_DATA_COUNT, /* number of uint16_t words for combining data */ michael@0: _NORM_INDEX_COMBINE_FWD_COUNT, /* number of code points that combine forward */ michael@0: _NORM_INDEX_COMBINE_BOTH_COUNT, /* number of code points that combine forward and backward */ michael@0: _NORM_INDEX_COMBINE_BACK_COUNT, /* number of code points that combine backward */ michael@0: michael@0: _NORM_INDEX_MIN_NFC_NO_MAYBE, /* first code point with quick check NFC NO/MAYBE */ michael@0: _NORM_INDEX_MIN_NFKC_NO_MAYBE, /* first code point with quick check NFKC NO/MAYBE */ michael@0: _NORM_INDEX_MIN_NFD_NO_MAYBE, /* first code point with quick check NFD NO/MAYBE */ michael@0: _NORM_INDEX_MIN_NFKD_NO_MAYBE, /* first code point with quick check NFKD NO/MAYBE */ michael@0: michael@0: _NORM_INDEX_FCD_TRIE_SIZE, /* number of bytes in FCD trie */ michael@0: michael@0: _NORM_INDEX_AUX_TRIE_SIZE, /* number of bytes in the auxiliary trie */ michael@0: _NORM_INDEX_CANON_SET_COUNT, /* number of uint16_t in the array of serialized USet */ michael@0: michael@0: _NORM_INDEX_TOP=32 /* changing this requires a new formatVersion */ michael@0: }; michael@0: michael@0: enum { michael@0: /* FCD check: everything below this code point is known to have a 0 lead combining class */ michael@0: _NORM_MIN_WITH_LEAD_CC=0x300 michael@0: }; michael@0: michael@0: enum { michael@0: /** michael@0: * Bit 7 of the length byte for a decomposition string in extra data is michael@0: * a flag indicating whether the decomposition string is michael@0: * preceded by a 16-bit word with the leading and trailing cc michael@0: * of the decomposition (like for A-umlaut); michael@0: * if not, then both cc's are zero (like for compatibility ideographs). michael@0: */ michael@0: _NORM_DECOMP_FLAG_LENGTH_HAS_CC=0x80, michael@0: /** michael@0: * Bits 6..0 of the length byte contain the actual length. michael@0: */ michael@0: _NORM_DECOMP_LENGTH_MASK=0x7f michael@0: }; michael@0: michael@0: /** Constants for options flags for normalization. */ michael@0: enum { michael@0: /** Options bit 0, do not decompose Hangul syllables. */ michael@0: UNORM_NX_HANGUL=1, michael@0: /** Options bit 1, do not decompose CJK compatibility characters. */ michael@0: UNORM_NX_CJK_COMPAT=2 michael@0: }; michael@0: michael@0: /** michael@0: * Description of the format of unorm.icu version 2.3. michael@0: * michael@0: * Main change from version 1 to version 2: michael@0: * Use of new, common UTrie instead of normalization-specific tries. michael@0: * Change to version 2.1: add third/auxiliary trie with associated data. michael@0: * Change to version 2.2: add skippable (f) flag data (_NORM_AUX_NFC_SKIP_F_MASK). michael@0: * Change to version 2.3: add serialized sets for normalization exclusions michael@0: * stored inside canonStartSets[] michael@0: * michael@0: * For more details of how to use the data structures see the code michael@0: * in unorm.cpp (runtime normalization code) and michael@0: * in gennorm.c and gennorm/store.c (build-time data generation). michael@0: * michael@0: * For the serialized format of UTrie see utrie.c/UTrieHeader. michael@0: * michael@0: * - Overall partition michael@0: * michael@0: * unorm.dat customarily begins with a UDataInfo structure, see udata.h and .c. michael@0: * After that there are the following structures: michael@0: * michael@0: * int32_t indexes[_NORM_INDEX_TOP]; -- _NORM_INDEX_TOP=32, see enum in this file michael@0: * michael@0: * UTrie normTrie; -- size in bytes=indexes[_NORM_INDEX_TRIE_SIZE] michael@0: * michael@0: * uint16_t extraData[extraDataTop]; -- extraDataTop=indexes[_NORM_INDEX_UCHAR_COUNT] michael@0: * extraData[0] contains the number of units for michael@0: * FC_NFKC_Closure (formatVersion>=2.1) michael@0: * michael@0: * uint16_t combiningTable[combiningTableTop]; -- combiningTableTop=indexes[_NORM_INDEX_COMBINE_DATA_COUNT] michael@0: * combiningTableTop may include one 16-bit padding unit michael@0: * to make sure that fcdTrie is 32-bit-aligned michael@0: * michael@0: * UTrie fcdTrie; -- size in bytes=indexes[_NORM_INDEX_FCD_TRIE_SIZE] michael@0: * michael@0: * UTrie auxTrie; -- size in bytes=indexes[_NORM_INDEX_AUX_TRIE_SIZE] michael@0: * michael@0: * uint16_t canonStartSets[canonStartSetsTop] -- canonStartSetsTop=indexes[_NORM_INDEX_CANON_SET_COUNT] michael@0: * serialized USets and binary search tables, see below michael@0: * michael@0: * michael@0: * The indexes array contains lengths and sizes of the following arrays and structures michael@0: * as well as the following values: michael@0: * indexes[_NORM_INDEX_COMBINE_FWD_COUNT]=combineFwdTop michael@0: * -- one more than the highest combining index computed for forward-only-combining characters michael@0: * indexes[_NORM_INDEX_COMBINE_BOTH_COUNT]=combineBothTop-combineFwdTop michael@0: * -- number of combining indexes computed for both-ways-combining characters michael@0: * indexes[_NORM_INDEX_COMBINE_BACK_COUNT]=combineBackTop-combineBothTop michael@0: * -- number of combining indexes computed for backward-only-combining characters michael@0: * michael@0: * indexes[_NORM_INDEX_MIN_NF*_NO_MAYBE] (where *={ C, D, KC, KD }) michael@0: * -- first code point with a quick check NF* value of NO/MAYBE michael@0: * michael@0: * michael@0: * - Tries michael@0: * michael@0: * The main structures are two UTrie tables ("compact arrays"), michael@0: * each with one index array and one data array. michael@0: * See utrie.h and utrie.c. michael@0: * michael@0: * michael@0: * - Tries in unorm.dat michael@0: * michael@0: * The first trie (normTrie above) michael@0: * provides data for the NF* quick checks and normalization. michael@0: * The second trie (fcdTrie above) provides data just for FCD checks. michael@0: * michael@0: * michael@0: * - norm32 data words from the first trie michael@0: * michael@0: * The norm32Table contains one 32-bit word "norm32" per code point. michael@0: * It contains the following bit fields: michael@0: * 31..16 extra data index, _NORM_EXTRA_SHIFT is used to shift this field down michael@0: * if this index is <_NORM_EXTRA_INDEX_TOP then it is an index into michael@0: * extraData[] where variable-length normalization data for this michael@0: * code point is found michael@0: * if this index is <_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP michael@0: * then this is a norm32 for a leading surrogate, and the index michael@0: * value is used together with the following trailing surrogate michael@0: * code unit in the second trie access michael@0: * if this index is >=_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP michael@0: * then this is a norm32 for a "special" character, michael@0: * i.e., the character is a Hangul syllable or a Jamo michael@0: * see _NORM_EXTRA_HANGUL etc. michael@0: * generally, instead of extracting this index from the norm32 and michael@0: * comparing it with the above constants, michael@0: * the normalization code compares the entire norm32 value michael@0: * with _NORM_MIN_SPECIAL, _NORM_SURROGATES_TOP, _NORM_MIN_HANGUL etc. michael@0: * michael@0: * 15..8 combining class (cc) according to UnicodeData.txt michael@0: * michael@0: * 7..6 _NORM_COMBINES_ANY flags, used in composition to see if a character michael@0: * combines with any following or preceding character(s) michael@0: * at all michael@0: * 7 _NORM_COMBINES_BACK michael@0: * 6 _NORM_COMBINES_FWD michael@0: * michael@0: * 5..0 quick check flags, set for "no" or "maybe", with separate flags for michael@0: * each normalization form michael@0: * the higher bits are "maybe" flags; for NF*D there are no such flags michael@0: * the lower bits are "no" flags for all forms, in the same order michael@0: * as the "maybe" flags, michael@0: * which is (MSB to LSB): NFKD NFD NFKC NFC michael@0: * 5..4 _NORM_QC_ANY_MAYBE michael@0: * 3..0 _NORM_QC_ANY_NO michael@0: * see further related constants michael@0: * michael@0: * michael@0: * - Extra data per code point michael@0: * michael@0: * "Extra data" is referenced by the index in norm32. michael@0: * It is variable-length data. It is only present, and only those parts michael@0: * of it are, as needed for a given character. michael@0: * The norm32 extra data index is added to the beginning of extraData[] michael@0: * to get to a vector of 16-bit words with data at the following offsets: michael@0: * michael@0: * [-1] Combining index for composition. michael@0: * Stored only if norm32&_NORM_COMBINES_ANY . michael@0: * [0] Lengths of the canonical and compatibility decomposition strings. michael@0: * Stored only if there are decompositions, i.e., michael@0: * if norm32&(_NORM_QC_NFD|_NORM_QC_NFKD) michael@0: * High byte: length of NFKD, or 0 if none michael@0: * Low byte: length of NFD, or 0 if none michael@0: * Each length byte also has another flag: michael@0: * Bit 7 of a length byte is set if there are non-zero michael@0: * combining classes (cc's) associated with the respective michael@0: * decomposition. If this flag is set, then the decomposition michael@0: * is preceded by a 16-bit word that contains the michael@0: * leading and trailing cc's. michael@0: * Bits 6..0 of a length byte are the length of the michael@0: * decomposition string, not counting the cc word. michael@0: * [1..n] NFD michael@0: * [n+1..] NFKD michael@0: * michael@0: * Each of the two decompositions consists of up to two parts: michael@0: * - The 16-bit words with the leading and trailing cc's. michael@0: * This is only stored if bit 7 of the corresponding length byte michael@0: * is set. In this case, at least one of the cc's is not zero. michael@0: * High byte: leading cc==cc of the first code point in the decomposition string michael@0: * Low byte: trailing cc==cc of the last code point in the decomposition string michael@0: * - The decomposition string in UTF-16, with length code units. michael@0: * michael@0: * michael@0: * - Combining indexes and combiningTable[] michael@0: * michael@0: * Combining indexes are stored at the [-1] offset of the extra data michael@0: * if the character combines forward or backward with any other characters. michael@0: * They are used for (re)composition in NF*C. michael@0: * Values of combining indexes are arranged according to whether a character michael@0: * combines forward, backward, or both ways: michael@0: * forward-only < both ways < backward-only michael@0: * michael@0: * The index values for forward-only and both-ways combining characters michael@0: * are indexes into the combiningTable[]. michael@0: * The index values for backward-only combining characters are simply michael@0: * incremented from the preceding index values to be unique. michael@0: * michael@0: * In the combiningTable[], a variable-length list michael@0: * of variable-length (back-index, code point) pair entries is stored michael@0: * for each forward-combining character. michael@0: * michael@0: * These back-indexes are the combining indexes of both-ways or backward-only michael@0: * combining characters that the forward-combining character combines with. michael@0: * michael@0: * Each list is sorted in ascending order of back-indexes. michael@0: * Each list is terminated with the last back-index having bit 15 set. michael@0: * michael@0: * Each pair (back-index, code point) takes up either 2 or 3 michael@0: * 16-bit words. michael@0: * The first word of a list entry is the back-index, with its bit 15 set if michael@0: * this is the last pair in the list. michael@0: * michael@0: * The second word contains flags in bits 15..13 that determine michael@0: * if there is a third word and how the combined character is encoded: michael@0: * 15 set if there is a third word in this list entry michael@0: * 14 set if the result is a supplementary character michael@0: * 13 set if the result itself combines forward michael@0: * michael@0: * According to these bits 15..14 of the second word, michael@0: * the result character is encoded as follows: michael@0: * 00 or 01 The result is <=0x1fff and stored in bits 12..0 of michael@0: * the second word. michael@0: * 10 The result is 0x2000..0xffff and stored in the third word. michael@0: * Bits 12..0 of the second word are not used. michael@0: * 11 The result is a supplementary character. michael@0: * Bits 9..0 of the leading surrogate are in bits 9..0 of michael@0: * the second word. michael@0: * Add 0xd800 to these bits to get the complete surrogate. michael@0: * Bits 12..10 of the second word are not used. michael@0: * The trailing surrogate is stored in the third word. michael@0: * michael@0: * michael@0: * - FCD trie michael@0: * michael@0: * The FCD trie is very simple. michael@0: * It is a folded trie with 16-bit data words. michael@0: * In each word, the high byte contains the leading cc of the character, michael@0: * and the low byte contains the trailing cc of the character. michael@0: * These cc's are the cc's of the first and last code points in the michael@0: * canonical decomposition of the character. michael@0: * michael@0: * Since all 16 bits are used for cc's, lead surrogates must be tested michael@0: * by checking the code unit instead of the trie data. michael@0: * This is done only if the 16-bit data word is not zero. michael@0: * If the code unit is a leading surrogate and the data word is not zero, michael@0: * then instead of cc's it contains the offset for the second trie lookup. michael@0: * michael@0: * michael@0: * - Auxiliary trie and data michael@0: * michael@0: * The auxiliary 16-bit trie contains data for additional properties. michael@0: * Bits michael@0: * 15..13 reserved michael@0: * 12 not NFC_Skippable (f) (formatVersion>=2.2) michael@0: * 11 flag: not a safe starter for canonical closure michael@0: * 10 composition exclusion michael@0: * 9.. 0 index into extraData[] to FC_NFKC_Closure string michael@0: * (not for lead surrogate), michael@0: * or lead surrogate offset (for lead surrogate, if 9..0 not zero) michael@0: * michael@0: * - FC_NFKC_Closure strings in extraData[] michael@0: * michael@0: * Strings are either stored as a single code unit or as the length michael@0: * followed by that many units. michael@0: * const UChar *s=extraData+(index from auxTrie data bits 9..0); michael@0: * int32_t length; michael@0: * if(*s<0xff00) { michael@0: * // s points to the single-unit string michael@0: * length=1; michael@0: * } else { michael@0: * length=*s&0xff; michael@0: * ++s; michael@0: * } michael@0: * michael@0: * Conditions for "NF* Skippable" from Mark Davis' com.ibm.text.UCD.NFSkippable: michael@0: * (used in NormalizerTransliterator) michael@0: * michael@0: * A skippable character is michael@0: * a) unassigned, or ALL of the following: michael@0: * b) of combining class 0. michael@0: * c) not decomposed by this normalization form. michael@0: * AND if NFC or NFKC, michael@0: * d) can never compose with a previous character. michael@0: * e) can never compose with a following character. michael@0: * f) can never change if another character is added. michael@0: * Example: a-breve might satisfy all but f, but if you michael@0: * add an ogonek it changes to a-ogonek + breve michael@0: * michael@0: * a)..e) must be tested from norm32. michael@0: * Since f) is more complicated, the (not-)NFC_Skippable flag (f) is built michael@0: * into the auxiliary trie. michael@0: * The same bit is used for NFC and NFKC; (c) differs for them. michael@0: * As usual, we build the "not skippable" flags so that unassigned michael@0: * code points get a 0 bit. michael@0: * This bit is only valid after (a)..(e) test FALSE; test NFD_NO before (f) as well. michael@0: * Test Hangul LV syllables entirely in code. michael@0: * michael@0: * michael@0: * - structure inside canonStartSets[] michael@0: * michael@0: * This array maps from code points c to sets of code points (USerializedSet). michael@0: * The result sets are the code points whose canonical decompositions start michael@0: * with c. michael@0: * michael@0: * canonStartSets[] contains the following sub-arrays: michael@0: * michael@0: * indexes[_NORM_SET_INDEX_TOP] michael@0: * - contains lengths of sub-arrays etc. michael@0: * michael@0: * startSets[indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH]-_NORM_SET_INDEX_TOP] michael@0: * - contains serialized sets (USerializedSet) of canonical starters for michael@0: * enumerating canonically equivalent strings michael@0: * indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH] includes _NORM_SET_INDEX_TOP michael@0: * for details about the structure see uset.c michael@0: * michael@0: * bmpTable[indexes[_NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH]] michael@0: * - a sorted search table for BMP code points whose results are michael@0: * either indexes to USerializedSets or single code points for michael@0: * single-code point sets; michael@0: * each entry is a pair of { code point, result } with result=(binary) yy xxxxxx xxxxxxxx michael@0: * if yy==01 then there is a USerializedSet at canonStartSets+x michael@0: * else build a USerializedSet with result as the single code point michael@0: * michael@0: * suppTable[indexes[_NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH]] michael@0: * - a sorted search table for supplementary code points whose results are michael@0: * either indexes to USerializedSets or single code points for michael@0: * single-code point sets; michael@0: * each entry is a triplet of { high16(cp), low16(cp), result } michael@0: * each code point's high-word may contain extra data in bits 15..5: michael@0: * if the high word has bit 15 set, then build a set with a single code point michael@0: * which is (((high16(cp)&0x1f00)<<8)|result; michael@0: * else there is a USerializedSet at canonStartSets+result michael@0: * michael@0: * FormatVersion 2.3 adds 2 serialized sets for normalization exclusions. michael@0: * They are stored in the data file so that the runtime normalization code need michael@0: * not depend on other properties and their data and implementation files. michael@0: * The _NORM_SET_INDEX_NX_..._OFFSET offsets in the canonStartSets index table michael@0: * give the location for each set. michael@0: * There is no set stored for UNORM_NX_HANGUL because it's trivial to create michael@0: * without using properties. michael@0: * michael@0: * Set contents: michael@0: * michael@0: * _NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET (for UNORM_NX_CJK_COMPAT) michael@0: * [[:Ideographic:]&[:NFD_QC=No:]] michael@0: * =[CJK Ideographs]&[has canonical decomposition] michael@0: * michael@0: * _NORM_SET_INDEX_NX_UNICODE32_OFFSET (for UNORM_UNICODE_3_2) michael@0: * [:^Age=3.2:] michael@0: * =set with all code points that were not designated by the specified Unicode version michael@0: * michael@0: * _NORM_SET_INDEX_NX_RESERVED_OFFSET michael@0: * This is an offset that points to where the next, future set would start. michael@0: * Currently it indicates where the previous set ends, and thus its length. michael@0: * The name for this enum constant may in the future be applied to different michael@0: * index slots. In order to get the limit of a set, use its index slot and michael@0: * the immediately following one regardless of that one's enum name. michael@0: */ michael@0: michael@0: #endif /* #if !UCONFIG_NO_NORMALIZATION */ michael@0: michael@0: #endif