1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/intl/icu/source/common/unormimp.h Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,486 @@
1.4 +/*
1.5 +*******************************************************************************
1.6 +*
1.7 +* Copyright (C) 2001-2011, International Business Machines
1.8 +* Corporation and others. All Rights Reserved.
1.9 +*
1.10 +*******************************************************************************
1.11 +* file name: unormimp.h
1.12 +* encoding: US-ASCII
1.13 +* tab size: 8 (not used)
1.14 +* indentation:4
1.15 +*
1.16 +* created on: 2001may25
1.17 +* created by: Markus W. Scherer
1.18 +*/
1.19 +
1.20 +#ifndef __UNORMIMP_H__
1.21 +#define __UNORMIMP_H__
1.22 +
1.23 +#include "unicode/utypes.h"
1.24 +
1.25 +#if !UCONFIG_NO_NORMALIZATION
1.26 +
1.27 +#include "udataswp.h"
1.28 +
1.29 +/*
1.30 + * The 2001-2010 implementation of the normalization code loads its data from
1.31 + * unorm.icu, which is generated with the gennorm tool.
1.32 + * The format of that file is described at the end of this file.
1.33 + */
1.34 +
1.35 +/* norm32 value constants */
1.36 +enum {
1.37 + /* quick check flags 0..3 set mean "no" for their forms */
1.38 + _NORM_QC_NFC=0x11, /* no|maybe */
1.39 + _NORM_QC_NFKC=0x22, /* no|maybe */
1.40 + _NORM_QC_NFD=4, /* no */
1.41 + _NORM_QC_NFKD=8, /* no */
1.42 +
1.43 + _NORM_QC_ANY_NO=0xf,
1.44 +
1.45 + /* quick check flags 4..5 mean "maybe" for their forms; test flags>=_NORM_QC_MAYBE */
1.46 + _NORM_QC_MAYBE=0x10,
1.47 + _NORM_QC_ANY_MAYBE=0x30,
1.48 +
1.49 + _NORM_QC_MASK=0x3f,
1.50 +
1.51 + _NORM_COMBINES_FWD=0x40,
1.52 + _NORM_COMBINES_BACK=0x80,
1.53 + _NORM_COMBINES_ANY=0xc0,
1.54 +
1.55 + _NORM_CC_SHIFT=8, /* UnicodeData.txt combining class in bits 15..8 */
1.56 + _NORM_CC_MASK=0xff00,
1.57 +
1.58 + _NORM_EXTRA_SHIFT=16, /* 16 bits for the index to UChars and other extra data */
1.59 + _NORM_EXTRA_INDEX_TOP=0xfc00, /* start of surrogate specials after shift */
1.60 +
1.61 + _NORM_EXTRA_SURROGATE_MASK=0x3ff,
1.62 + _NORM_EXTRA_SURROGATE_TOP=0x3f0, /* hangul etc. */
1.63 +
1.64 + _NORM_EXTRA_HANGUL=_NORM_EXTRA_SURROGATE_TOP,
1.65 + _NORM_EXTRA_JAMO_L,
1.66 + _NORM_EXTRA_JAMO_V,
1.67 + _NORM_EXTRA_JAMO_T
1.68 +};
1.69 +
1.70 +/* norm32 value constants using >16 bits */
1.71 +#define _NORM_MIN_SPECIAL 0xfc000000
1.72 +#define _NORM_SURROGATES_TOP 0xfff00000
1.73 +#define _NORM_MIN_HANGUL 0xfff00000
1.74 +#define _NORM_MIN_JAMO_V 0xfff20000
1.75 +#define _NORM_JAMO_V_TOP 0xfff30000
1.76 +
1.77 +/* value constants for auxTrie */
1.78 +enum {
1.79 + _NORM_AUX_COMP_EX_SHIFT=10,
1.80 + _NORM_AUX_UNSAFE_SHIFT=11,
1.81 + _NORM_AUX_NFC_SKIPPABLE_F_SHIFT=12
1.82 +};
1.83 +
1.84 +#define _NORM_AUX_MAX_FNC ((int32_t)1<<_NORM_AUX_COMP_EX_SHIFT)
1.85 +
1.86 +#define _NORM_AUX_FNC_MASK (uint32_t)(_NORM_AUX_MAX_FNC-1)
1.87 +#define _NORM_AUX_COMP_EX_MASK ((uint32_t)1<<_NORM_AUX_COMP_EX_SHIFT)
1.88 +#define _NORM_AUX_UNSAFE_MASK ((uint32_t)1<<_NORM_AUX_UNSAFE_SHIFT)
1.89 +#define _NORM_AUX_NFC_SKIP_F_MASK ((uint32_t)1<<_NORM_AUX_NFC_SKIPPABLE_F_SHIFT)
1.90 +
1.91 +/* canonStartSets[0..31] contains indexes for what is in the array */
1.92 +enum {
1.93 + _NORM_SET_INDEX_CANON_SETS_LENGTH, /* number of uint16_t in canonical starter sets */
1.94 + _NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH, /* number of uint16_t in the BMP search table (contains pairs) */
1.95 + _NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH,/* number of uint16_t in the supplementary search table (contains triplets) */
1.96 +
1.97 + /* from formatVersion 2.3: */
1.98 + _NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET, /* uint16_t offset from canonStartSets[0] to the
1.99 + exclusion set for CJK compatibility characters */
1.100 + _NORM_SET_INDEX_NX_UNICODE32_OFFSET, /* uint16_t offset from canonStartSets[0] to the
1.101 + exclusion set for Unicode 3.2 characters */
1.102 + _NORM_SET_INDEX_NX_RESERVED_OFFSET, /* uint16_t offset from canonStartSets[0] to the
1.103 + end of the previous exclusion set */
1.104 +
1.105 + _NORM_SET_INDEX_TOP=32 /* changing this requires a new formatVersion */
1.106 +};
1.107 +
1.108 +/* more constants for canonical starter sets */
1.109 +
1.110 +/* 14 bit indexes to canonical USerializedSets */
1.111 +#define _NORM_MAX_CANON_SETS 0x4000
1.112 +
1.113 +/* single-code point BMP sets are encoded directly in the search table except if result=0x4000..0x7fff */
1.114 +#define _NORM_CANON_SET_BMP_MASK 0xc000
1.115 +#define _NORM_CANON_SET_BMP_IS_INDEX 0x4000
1.116 +
1.117 +/* indexes[] value names */
1.118 +enum {
1.119 + _NORM_INDEX_TRIE_SIZE, /* number of bytes in normalization trie */
1.120 + _NORM_INDEX_UCHAR_COUNT, /* number of UChars in extra data */
1.121 +
1.122 + _NORM_INDEX_COMBINE_DATA_COUNT, /* number of uint16_t words for combining data */
1.123 + _NORM_INDEX_COMBINE_FWD_COUNT, /* number of code points that combine forward */
1.124 + _NORM_INDEX_COMBINE_BOTH_COUNT, /* number of code points that combine forward and backward */
1.125 + _NORM_INDEX_COMBINE_BACK_COUNT, /* number of code points that combine backward */
1.126 +
1.127 + _NORM_INDEX_MIN_NFC_NO_MAYBE, /* first code point with quick check NFC NO/MAYBE */
1.128 + _NORM_INDEX_MIN_NFKC_NO_MAYBE, /* first code point with quick check NFKC NO/MAYBE */
1.129 + _NORM_INDEX_MIN_NFD_NO_MAYBE, /* first code point with quick check NFD NO/MAYBE */
1.130 + _NORM_INDEX_MIN_NFKD_NO_MAYBE, /* first code point with quick check NFKD NO/MAYBE */
1.131 +
1.132 + _NORM_INDEX_FCD_TRIE_SIZE, /* number of bytes in FCD trie */
1.133 +
1.134 + _NORM_INDEX_AUX_TRIE_SIZE, /* number of bytes in the auxiliary trie */
1.135 + _NORM_INDEX_CANON_SET_COUNT, /* number of uint16_t in the array of serialized USet */
1.136 +
1.137 + _NORM_INDEX_TOP=32 /* changing this requires a new formatVersion */
1.138 +};
1.139 +
1.140 +enum {
1.141 + /* FCD check: everything below this code point is known to have a 0 lead combining class */
1.142 + _NORM_MIN_WITH_LEAD_CC=0x300
1.143 +};
1.144 +
1.145 +enum {
1.146 + /**
1.147 + * Bit 7 of the length byte for a decomposition string in extra data is
1.148 + * a flag indicating whether the decomposition string is
1.149 + * preceded by a 16-bit word with the leading and trailing cc
1.150 + * of the decomposition (like for A-umlaut);
1.151 + * if not, then both cc's are zero (like for compatibility ideographs).
1.152 + */
1.153 + _NORM_DECOMP_FLAG_LENGTH_HAS_CC=0x80,
1.154 + /**
1.155 + * Bits 6..0 of the length byte contain the actual length.
1.156 + */
1.157 + _NORM_DECOMP_LENGTH_MASK=0x7f
1.158 +};
1.159 +
1.160 +/** Constants for options flags for normalization. */
1.161 +enum {
1.162 + /** Options bit 0, do not decompose Hangul syllables. */
1.163 + UNORM_NX_HANGUL=1,
1.164 + /** Options bit 1, do not decompose CJK compatibility characters. */
1.165 + UNORM_NX_CJK_COMPAT=2
1.166 +};
1.167 +
1.168 +/**
1.169 + * Description of the format of unorm.icu version 2.3.
1.170 + *
1.171 + * Main change from version 1 to version 2:
1.172 + * Use of new, common UTrie instead of normalization-specific tries.
1.173 + * Change to version 2.1: add third/auxiliary trie with associated data.
1.174 + * Change to version 2.2: add skippable (f) flag data (_NORM_AUX_NFC_SKIP_F_MASK).
1.175 + * Change to version 2.3: add serialized sets for normalization exclusions
1.176 + * stored inside canonStartSets[]
1.177 + *
1.178 + * For more details of how to use the data structures see the code
1.179 + * in unorm.cpp (runtime normalization code) and
1.180 + * in gennorm.c and gennorm/store.c (build-time data generation).
1.181 + *
1.182 + * For the serialized format of UTrie see utrie.c/UTrieHeader.
1.183 + *
1.184 + * - Overall partition
1.185 + *
1.186 + * unorm.dat customarily begins with a UDataInfo structure, see udata.h and .c.
1.187 + * After that there are the following structures:
1.188 + *
1.189 + * int32_t indexes[_NORM_INDEX_TOP]; -- _NORM_INDEX_TOP=32, see enum in this file
1.190 + *
1.191 + * UTrie normTrie; -- size in bytes=indexes[_NORM_INDEX_TRIE_SIZE]
1.192 + *
1.193 + * uint16_t extraData[extraDataTop]; -- extraDataTop=indexes[_NORM_INDEX_UCHAR_COUNT]
1.194 + * extraData[0] contains the number of units for
1.195 + * FC_NFKC_Closure (formatVersion>=2.1)
1.196 + *
1.197 + * uint16_t combiningTable[combiningTableTop]; -- combiningTableTop=indexes[_NORM_INDEX_COMBINE_DATA_COUNT]
1.198 + * combiningTableTop may include one 16-bit padding unit
1.199 + * to make sure that fcdTrie is 32-bit-aligned
1.200 + *
1.201 + * UTrie fcdTrie; -- size in bytes=indexes[_NORM_INDEX_FCD_TRIE_SIZE]
1.202 + *
1.203 + * UTrie auxTrie; -- size in bytes=indexes[_NORM_INDEX_AUX_TRIE_SIZE]
1.204 + *
1.205 + * uint16_t canonStartSets[canonStartSetsTop] -- canonStartSetsTop=indexes[_NORM_INDEX_CANON_SET_COUNT]
1.206 + * serialized USets and binary search tables, see below
1.207 + *
1.208 + *
1.209 + * The indexes array contains lengths and sizes of the following arrays and structures
1.210 + * as well as the following values:
1.211 + * indexes[_NORM_INDEX_COMBINE_FWD_COUNT]=combineFwdTop
1.212 + * -- one more than the highest combining index computed for forward-only-combining characters
1.213 + * indexes[_NORM_INDEX_COMBINE_BOTH_COUNT]=combineBothTop-combineFwdTop
1.214 + * -- number of combining indexes computed for both-ways-combining characters
1.215 + * indexes[_NORM_INDEX_COMBINE_BACK_COUNT]=combineBackTop-combineBothTop
1.216 + * -- number of combining indexes computed for backward-only-combining characters
1.217 + *
1.218 + * indexes[_NORM_INDEX_MIN_NF*_NO_MAYBE] (where *={ C, D, KC, KD })
1.219 + * -- first code point with a quick check NF* value of NO/MAYBE
1.220 + *
1.221 + *
1.222 + * - Tries
1.223 + *
1.224 + * The main structures are two UTrie tables ("compact arrays"),
1.225 + * each with one index array and one data array.
1.226 + * See utrie.h and utrie.c.
1.227 + *
1.228 + *
1.229 + * - Tries in unorm.dat
1.230 + *
1.231 + * The first trie (normTrie above)
1.232 + * provides data for the NF* quick checks and normalization.
1.233 + * The second trie (fcdTrie above) provides data just for FCD checks.
1.234 + *
1.235 + *
1.236 + * - norm32 data words from the first trie
1.237 + *
1.238 + * The norm32Table contains one 32-bit word "norm32" per code point.
1.239 + * It contains the following bit fields:
1.240 + * 31..16 extra data index, _NORM_EXTRA_SHIFT is used to shift this field down
1.241 + * if this index is <_NORM_EXTRA_INDEX_TOP then it is an index into
1.242 + * extraData[] where variable-length normalization data for this
1.243 + * code point is found
1.244 + * if this index is <_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP
1.245 + * then this is a norm32 for a leading surrogate, and the index
1.246 + * value is used together with the following trailing surrogate
1.247 + * code unit in the second trie access
1.248 + * if this index is >=_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP
1.249 + * then this is a norm32 for a "special" character,
1.250 + * i.e., the character is a Hangul syllable or a Jamo
1.251 + * see _NORM_EXTRA_HANGUL etc.
1.252 + * generally, instead of extracting this index from the norm32 and
1.253 + * comparing it with the above constants,
1.254 + * the normalization code compares the entire norm32 value
1.255 + * with _NORM_MIN_SPECIAL, _NORM_SURROGATES_TOP, _NORM_MIN_HANGUL etc.
1.256 + *
1.257 + * 15..8 combining class (cc) according to UnicodeData.txt
1.258 + *
1.259 + * 7..6 _NORM_COMBINES_ANY flags, used in composition to see if a character
1.260 + * combines with any following or preceding character(s)
1.261 + * at all
1.262 + * 7 _NORM_COMBINES_BACK
1.263 + * 6 _NORM_COMBINES_FWD
1.264 + *
1.265 + * 5..0 quick check flags, set for "no" or "maybe", with separate flags for
1.266 + * each normalization form
1.267 + * the higher bits are "maybe" flags; for NF*D there are no such flags
1.268 + * the lower bits are "no" flags for all forms, in the same order
1.269 + * as the "maybe" flags,
1.270 + * which is (MSB to LSB): NFKD NFD NFKC NFC
1.271 + * 5..4 _NORM_QC_ANY_MAYBE
1.272 + * 3..0 _NORM_QC_ANY_NO
1.273 + * see further related constants
1.274 + *
1.275 + *
1.276 + * - Extra data per code point
1.277 + *
1.278 + * "Extra data" is referenced by the index in norm32.
1.279 + * It is variable-length data. It is only present, and only those parts
1.280 + * of it are, as needed for a given character.
1.281 + * The norm32 extra data index is added to the beginning of extraData[]
1.282 + * to get to a vector of 16-bit words with data at the following offsets:
1.283 + *
1.284 + * [-1] Combining index for composition.
1.285 + * Stored only if norm32&_NORM_COMBINES_ANY .
1.286 + * [0] Lengths of the canonical and compatibility decomposition strings.
1.287 + * Stored only if there are decompositions, i.e.,
1.288 + * if norm32&(_NORM_QC_NFD|_NORM_QC_NFKD)
1.289 + * High byte: length of NFKD, or 0 if none
1.290 + * Low byte: length of NFD, or 0 if none
1.291 + * Each length byte also has another flag:
1.292 + * Bit 7 of a length byte is set if there are non-zero
1.293 + * combining classes (cc's) associated with the respective
1.294 + * decomposition. If this flag is set, then the decomposition
1.295 + * is preceded by a 16-bit word that contains the
1.296 + * leading and trailing cc's.
1.297 + * Bits 6..0 of a length byte are the length of the
1.298 + * decomposition string, not counting the cc word.
1.299 + * [1..n] NFD
1.300 + * [n+1..] NFKD
1.301 + *
1.302 + * Each of the two decompositions consists of up to two parts:
1.303 + * - The 16-bit words with the leading and trailing cc's.
1.304 + * This is only stored if bit 7 of the corresponding length byte
1.305 + * is set. In this case, at least one of the cc's is not zero.
1.306 + * High byte: leading cc==cc of the first code point in the decomposition string
1.307 + * Low byte: trailing cc==cc of the last code point in the decomposition string
1.308 + * - The decomposition string in UTF-16, with length code units.
1.309 + *
1.310 + *
1.311 + * - Combining indexes and combiningTable[]
1.312 + *
1.313 + * Combining indexes are stored at the [-1] offset of the extra data
1.314 + * if the character combines forward or backward with any other characters.
1.315 + * They are used for (re)composition in NF*C.
1.316 + * Values of combining indexes are arranged according to whether a character
1.317 + * combines forward, backward, or both ways:
1.318 + * forward-only < both ways < backward-only
1.319 + *
1.320 + * The index values for forward-only and both-ways combining characters
1.321 + * are indexes into the combiningTable[].
1.322 + * The index values for backward-only combining characters are simply
1.323 + * incremented from the preceding index values to be unique.
1.324 + *
1.325 + * In the combiningTable[], a variable-length list
1.326 + * of variable-length (back-index, code point) pair entries is stored
1.327 + * for each forward-combining character.
1.328 + *
1.329 + * These back-indexes are the combining indexes of both-ways or backward-only
1.330 + * combining characters that the forward-combining character combines with.
1.331 + *
1.332 + * Each list is sorted in ascending order of back-indexes.
1.333 + * Each list is terminated with the last back-index having bit 15 set.
1.334 + *
1.335 + * Each pair (back-index, code point) takes up either 2 or 3
1.336 + * 16-bit words.
1.337 + * The first word of a list entry is the back-index, with its bit 15 set if
1.338 + * this is the last pair in the list.
1.339 + *
1.340 + * The second word contains flags in bits 15..13 that determine
1.341 + * if there is a third word and how the combined character is encoded:
1.342 + * 15 set if there is a third word in this list entry
1.343 + * 14 set if the result is a supplementary character
1.344 + * 13 set if the result itself combines forward
1.345 + *
1.346 + * According to these bits 15..14 of the second word,
1.347 + * the result character is encoded as follows:
1.348 + * 00 or 01 The result is <=0x1fff and stored in bits 12..0 of
1.349 + * the second word.
1.350 + * 10 The result is 0x2000..0xffff and stored in the third word.
1.351 + * Bits 12..0 of the second word are not used.
1.352 + * 11 The result is a supplementary character.
1.353 + * Bits 9..0 of the leading surrogate are in bits 9..0 of
1.354 + * the second word.
1.355 + * Add 0xd800 to these bits to get the complete surrogate.
1.356 + * Bits 12..10 of the second word are not used.
1.357 + * The trailing surrogate is stored in the third word.
1.358 + *
1.359 + *
1.360 + * - FCD trie
1.361 + *
1.362 + * The FCD trie is very simple.
1.363 + * It is a folded trie with 16-bit data words.
1.364 + * In each word, the high byte contains the leading cc of the character,
1.365 + * and the low byte contains the trailing cc of the character.
1.366 + * These cc's are the cc's of the first and last code points in the
1.367 + * canonical decomposition of the character.
1.368 + *
1.369 + * Since all 16 bits are used for cc's, lead surrogates must be tested
1.370 + * by checking the code unit instead of the trie data.
1.371 + * This is done only if the 16-bit data word is not zero.
1.372 + * If the code unit is a leading surrogate and the data word is not zero,
1.373 + * then instead of cc's it contains the offset for the second trie lookup.
1.374 + *
1.375 + *
1.376 + * - Auxiliary trie and data
1.377 + *
1.378 + * The auxiliary 16-bit trie contains data for additional properties.
1.379 + * Bits
1.380 + * 15..13 reserved
1.381 + * 12 not NFC_Skippable (f) (formatVersion>=2.2)
1.382 + * 11 flag: not a safe starter for canonical closure
1.383 + * 10 composition exclusion
1.384 + * 9.. 0 index into extraData[] to FC_NFKC_Closure string
1.385 + * (not for lead surrogate),
1.386 + * or lead surrogate offset (for lead surrogate, if 9..0 not zero)
1.387 + *
1.388 + * - FC_NFKC_Closure strings in extraData[]
1.389 + *
1.390 + * Strings are either stored as a single code unit or as the length
1.391 + * followed by that many units.
1.392 + * const UChar *s=extraData+(index from auxTrie data bits 9..0);
1.393 + * int32_t length;
1.394 + * if(*s<0xff00) {
1.395 + * // s points to the single-unit string
1.396 + * length=1;
1.397 + * } else {
1.398 + * length=*s&0xff;
1.399 + * ++s;
1.400 + * }
1.401 + *
1.402 + * Conditions for "NF* Skippable" from Mark Davis' com.ibm.text.UCD.NFSkippable:
1.403 + * (used in NormalizerTransliterator)
1.404 + *
1.405 + * A skippable character is
1.406 + * a) unassigned, or ALL of the following:
1.407 + * b) of combining class 0.
1.408 + * c) not decomposed by this normalization form.
1.409 + * AND if NFC or NFKC,
1.410 + * d) can never compose with a previous character.
1.411 + * e) can never compose with a following character.
1.412 + * f) can never change if another character is added.
1.413 + * Example: a-breve might satisfy all but f, but if you
1.414 + * add an ogonek it changes to a-ogonek + breve
1.415 + *
1.416 + * a)..e) must be tested from norm32.
1.417 + * Since f) is more complicated, the (not-)NFC_Skippable flag (f) is built
1.418 + * into the auxiliary trie.
1.419 + * The same bit is used for NFC and NFKC; (c) differs for them.
1.420 + * As usual, we build the "not skippable" flags so that unassigned
1.421 + * code points get a 0 bit.
1.422 + * This bit is only valid after (a)..(e) test FALSE; test NFD_NO before (f) as well.
1.423 + * Test Hangul LV syllables entirely in code.
1.424 + *
1.425 + *
1.426 + * - structure inside canonStartSets[]
1.427 + *
1.428 + * This array maps from code points c to sets of code points (USerializedSet).
1.429 + * The result sets are the code points whose canonical decompositions start
1.430 + * with c.
1.431 + *
1.432 + * canonStartSets[] contains the following sub-arrays:
1.433 + *
1.434 + * indexes[_NORM_SET_INDEX_TOP]
1.435 + * - contains lengths of sub-arrays etc.
1.436 + *
1.437 + * startSets[indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH]-_NORM_SET_INDEX_TOP]
1.438 + * - contains serialized sets (USerializedSet) of canonical starters for
1.439 + * enumerating canonically equivalent strings
1.440 + * indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH] includes _NORM_SET_INDEX_TOP
1.441 + * for details about the structure see uset.c
1.442 + *
1.443 + * bmpTable[indexes[_NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH]]
1.444 + * - a sorted search table for BMP code points whose results are
1.445 + * either indexes to USerializedSets or single code points for
1.446 + * single-code point sets;
1.447 + * each entry is a pair of { code point, result } with result=(binary) yy xxxxxx xxxxxxxx
1.448 + * if yy==01 then there is a USerializedSet at canonStartSets+x
1.449 + * else build a USerializedSet with result as the single code point
1.450 + *
1.451 + * suppTable[indexes[_NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH]]
1.452 + * - a sorted search table for supplementary code points whose results are
1.453 + * either indexes to USerializedSets or single code points for
1.454 + * single-code point sets;
1.455 + * each entry is a triplet of { high16(cp), low16(cp), result }
1.456 + * each code point's high-word may contain extra data in bits 15..5:
1.457 + * if the high word has bit 15 set, then build a set with a single code point
1.458 + * which is (((high16(cp)&0x1f00)<<8)|result;
1.459 + * else there is a USerializedSet at canonStartSets+result
1.460 + *
1.461 + * FormatVersion 2.3 adds 2 serialized sets for normalization exclusions.
1.462 + * They are stored in the data file so that the runtime normalization code need
1.463 + * not depend on other properties and their data and implementation files.
1.464 + * The _NORM_SET_INDEX_NX_..._OFFSET offsets in the canonStartSets index table
1.465 + * give the location for each set.
1.466 + * There is no set stored for UNORM_NX_HANGUL because it's trivial to create
1.467 + * without using properties.
1.468 + *
1.469 + * Set contents:
1.470 + *
1.471 + * _NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET (for UNORM_NX_CJK_COMPAT)
1.472 + * [[:Ideographic:]&[:NFD_QC=No:]]
1.473 + * =[CJK Ideographs]&[has canonical decomposition]
1.474 + *
1.475 + * _NORM_SET_INDEX_NX_UNICODE32_OFFSET (for UNORM_UNICODE_3_2)
1.476 + * [:^Age=3.2:]
1.477 + * =set with all code points that were not designated by the specified Unicode version
1.478 + *
1.479 + * _NORM_SET_INDEX_NX_RESERVED_OFFSET
1.480 + * This is an offset that points to where the next, future set would start.
1.481 + * Currently it indicates where the previous set ends, and thus its length.
1.482 + * The name for this enum constant may in the future be applied to different
1.483 + * index slots. In order to get the limit of a set, use its index slot and
1.484 + * the immediately following one regardless of that one's enum name.
1.485 + */
1.486 +
1.487 +#endif /* #if !UCONFIG_NO_NORMALIZATION */
1.488 +
1.489 +#endif
