1/* 2******************************************************************************* 3* 4* Copyright (C) 2001-2011, International Business Machines 5* Corporation and others. All Rights Reserved. 6* 7******************************************************************************* 8* file name: unormimp.h 9* encoding: US-ASCII 10* tab size: 8 (not used) 11* indentation:4 12* 13* created on: 2001may25 14* created by: Markus W. Scherer 15*/ 16 17#ifndef __UNORMIMP_H__ 18#define __UNORMIMP_H__ 19 20#include "unicode/utypes.h" 21 22#if !UCONFIG_NO_NORMALIZATION 23 24#include "udataswp.h" 25 26/* 27 * The 2001-2010 implementation of the normalization code loads its data from 28 * unorm.icu, which is generated with the gennorm tool. 29 * The format of that file is described at the end of this file. 30 */ 31 32/* norm32 value constants */ 33enum { 34 /* quick check flags 0..3 set mean "no" for their forms */ 35 _NORM_QC_NFC=0x11, /* no|maybe */ 36 _NORM_QC_NFKC=0x22, /* no|maybe */ 37 _NORM_QC_NFD=4, /* no */ 38 _NORM_QC_NFKD=8, /* no */ 39 40 _NORM_QC_ANY_NO=0xf, 41 42 /* quick check flags 4..5 mean "maybe" for their forms; test flags>=_NORM_QC_MAYBE */ 43 _NORM_QC_MAYBE=0x10, 44 _NORM_QC_ANY_MAYBE=0x30, 45 46 _NORM_QC_MASK=0x3f, 47 48 _NORM_COMBINES_FWD=0x40, 49 _NORM_COMBINES_BACK=0x80, 50 _NORM_COMBINES_ANY=0xc0, 51 52 _NORM_CC_SHIFT=8, /* UnicodeData.txt combining class in bits 15..8 */ 53 _NORM_CC_MASK=0xff00, 54 55 _NORM_EXTRA_SHIFT=16, /* 16 bits for the index to UChars and other extra data */ 56 _NORM_EXTRA_INDEX_TOP=0xfc00, /* start of surrogate specials after shift */ 57 58 _NORM_EXTRA_SURROGATE_MASK=0x3ff, 59 _NORM_EXTRA_SURROGATE_TOP=0x3f0, /* hangul etc. */ 60 61 _NORM_EXTRA_HANGUL=_NORM_EXTRA_SURROGATE_TOP, 62 _NORM_EXTRA_JAMO_L, 63 _NORM_EXTRA_JAMO_V, 64 _NORM_EXTRA_JAMO_T 65}; 66 67/* norm32 value constants using >16 bits */ 68#define _NORM_MIN_SPECIAL 0xfc000000 69#define _NORM_SURROGATES_TOP 0xfff00000 70#define _NORM_MIN_HANGUL 0xfff00000 71#define _NORM_MIN_JAMO_V 0xfff20000 72#define _NORM_JAMO_V_TOP 0xfff30000 73 74/* value constants for auxTrie */ 75enum { 76 _NORM_AUX_COMP_EX_SHIFT=10, 77 _NORM_AUX_UNSAFE_SHIFT=11, 78 _NORM_AUX_NFC_SKIPPABLE_F_SHIFT=12 79}; 80 81#define _NORM_AUX_MAX_FNC ((int32_t)1<<_NORM_AUX_COMP_EX_SHIFT) 82 83#define _NORM_AUX_FNC_MASK (uint32_t)(_NORM_AUX_MAX_FNC-1) 84#define _NORM_AUX_COMP_EX_MASK ((uint32_t)1<<_NORM_AUX_COMP_EX_SHIFT) 85#define _NORM_AUX_UNSAFE_MASK ((uint32_t)1<<_NORM_AUX_UNSAFE_SHIFT) 86#define _NORM_AUX_NFC_SKIP_F_MASK ((uint32_t)1<<_NORM_AUX_NFC_SKIPPABLE_F_SHIFT) 87 88/* canonStartSets[0..31] contains indexes for what is in the array */ 89enum { 90 _NORM_SET_INDEX_CANON_SETS_LENGTH, /* number of uint16_t in canonical starter sets */ 91 _NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH, /* number of uint16_t in the BMP search table (contains pairs) */ 92 _NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH,/* number of uint16_t in the supplementary search table (contains triplets) */ 93 94 /* from formatVersion 2.3: */ 95 _NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET, /* uint16_t offset from canonStartSets[0] to the 96 exclusion set for CJK compatibility characters */ 97 _NORM_SET_INDEX_NX_UNICODE32_OFFSET, /* uint16_t offset from canonStartSets[0] to the 98 exclusion set for Unicode 3.2 characters */ 99 _NORM_SET_INDEX_NX_RESERVED_OFFSET, /* uint16_t offset from canonStartSets[0] to the 100 end of the previous exclusion set */ 101 102 _NORM_SET_INDEX_TOP=32 /* changing this requires a new formatVersion */ 103}; 104 105/* more constants for canonical starter sets */ 106 107/* 14 bit indexes to canonical USerializedSets */ 108#define _NORM_MAX_CANON_SETS 0x4000 109 110/* single-code point BMP sets are encoded directly in the search table except if result=0x4000..0x7fff */ 111#define _NORM_CANON_SET_BMP_MASK 0xc000 112#define _NORM_CANON_SET_BMP_IS_INDEX 0x4000 113 114/* indexes[] value names */ 115enum { 116 _NORM_INDEX_TRIE_SIZE, /* number of bytes in normalization trie */ 117 _NORM_INDEX_UCHAR_COUNT, /* number of UChars in extra data */ 118 119 _NORM_INDEX_COMBINE_DATA_COUNT, /* number of uint16_t words for combining data */ 120 _NORM_INDEX_COMBINE_FWD_COUNT, /* number of code points that combine forward */ 121 _NORM_INDEX_COMBINE_BOTH_COUNT, /* number of code points that combine forward and backward */ 122 _NORM_INDEX_COMBINE_BACK_COUNT, /* number of code points that combine backward */ 123 124 _NORM_INDEX_MIN_NFC_NO_MAYBE, /* first code point with quick check NFC NO/MAYBE */ 125 _NORM_INDEX_MIN_NFKC_NO_MAYBE, /* first code point with quick check NFKC NO/MAYBE */ 126 _NORM_INDEX_MIN_NFD_NO_MAYBE, /* first code point with quick check NFD NO/MAYBE */ 127 _NORM_INDEX_MIN_NFKD_NO_MAYBE, /* first code point with quick check NFKD NO/MAYBE */ 128 129 _NORM_INDEX_FCD_TRIE_SIZE, /* number of bytes in FCD trie */ 130 131 _NORM_INDEX_AUX_TRIE_SIZE, /* number of bytes in the auxiliary trie */ 132 _NORM_INDEX_CANON_SET_COUNT, /* number of uint16_t in the array of serialized USet */ 133 134 _NORM_INDEX_TOP=32 /* changing this requires a new formatVersion */ 135}; 136 137enum { 138 /* FCD check: everything below this code point is known to have a 0 lead combining class */ 139 _NORM_MIN_WITH_LEAD_CC=0x300 140}; 141 142enum { 143 /** 144 * Bit 7 of the length byte for a decomposition string in extra data is 145 * a flag indicating whether the decomposition string is 146 * preceded by a 16-bit word with the leading and trailing cc 147 * of the decomposition (like for A-umlaut); 148 * if not, then both cc's are zero (like for compatibility ideographs). 149 */ 150 _NORM_DECOMP_FLAG_LENGTH_HAS_CC=0x80, 151 /** 152 * Bits 6..0 of the length byte contain the actual length. 153 */ 154 _NORM_DECOMP_LENGTH_MASK=0x7f 155}; 156 157/** Constants for options flags for normalization. */ 158enum { 159 /** Options bit 0, do not decompose Hangul syllables. */ 160 UNORM_NX_HANGUL=1, 161 /** Options bit 1, do not decompose CJK compatibility characters. */ 162 UNORM_NX_CJK_COMPAT=2 163}; 164 165/** 166 * Description of the format of unorm.icu version 2.3. 167 * 168 * Main change from version 1 to version 2: 169 * Use of new, common UTrie instead of normalization-specific tries. 170 * Change to version 2.1: add third/auxiliary trie with associated data. 171 * Change to version 2.2: add skippable (f) flag data (_NORM_AUX_NFC_SKIP_F_MASK). 172 * Change to version 2.3: add serialized sets for normalization exclusions 173 * stored inside canonStartSets[] 174 * 175 * For more details of how to use the data structures see the code 176 * in unorm.cpp (runtime normalization code) and 177 * in gennorm.c and gennorm/store.c (build-time data generation). 178 * 179 * For the serialized format of UTrie see utrie.c/UTrieHeader. 180 * 181 * - Overall partition 182 * 183 * unorm.dat customarily begins with a UDataInfo structure, see udata.h and .c. 184 * After that there are the following structures: 185 * 186 * int32_t indexes[_NORM_INDEX_TOP]; -- _NORM_INDEX_TOP=32, see enum in this file 187 * 188 * UTrie normTrie; -- size in bytes=indexes[_NORM_INDEX_TRIE_SIZE] 189 * 190 * uint16_t extraData[extraDataTop]; -- extraDataTop=indexes[_NORM_INDEX_UCHAR_COUNT] 191 * extraData[0] contains the number of units for 192 * FC_NFKC_Closure (formatVersion>=2.1) 193 * 194 * uint16_t combiningTable[combiningTableTop]; -- combiningTableTop=indexes[_NORM_INDEX_COMBINE_DATA_COUNT] 195 * combiningTableTop may include one 16-bit padding unit 196 * to make sure that fcdTrie is 32-bit-aligned 197 * 198 * UTrie fcdTrie; -- size in bytes=indexes[_NORM_INDEX_FCD_TRIE_SIZE] 199 * 200 * UTrie auxTrie; -- size in bytes=indexes[_NORM_INDEX_AUX_TRIE_SIZE] 201 * 202 * uint16_t canonStartSets[canonStartSetsTop] -- canonStartSetsTop=indexes[_NORM_INDEX_CANON_SET_COUNT] 203 * serialized USets and binary search tables, see below 204 * 205 * 206 * The indexes array contains lengths and sizes of the following arrays and structures 207 * as well as the following values: 208 * indexes[_NORM_INDEX_COMBINE_FWD_COUNT]=combineFwdTop 209 * -- one more than the highest combining index computed for forward-only-combining characters 210 * indexes[_NORM_INDEX_COMBINE_BOTH_COUNT]=combineBothTop-combineFwdTop 211 * -- number of combining indexes computed for both-ways-combining characters 212 * indexes[_NORM_INDEX_COMBINE_BACK_COUNT]=combineBackTop-combineBothTop 213 * -- number of combining indexes computed for backward-only-combining characters 214 * 215 * indexes[_NORM_INDEX_MIN_NF*_NO_MAYBE] (where *={ C, D, KC, KD }) 216 * -- first code point with a quick check NF* value of NO/MAYBE 217 * 218 * 219 * - Tries 220 * 221 * The main structures are two UTrie tables ("compact arrays"), 222 * each with one index array and one data array. 223 * See utrie.h and utrie.c. 224 * 225 * 226 * - Tries in unorm.dat 227 * 228 * The first trie (normTrie above) 229 * provides data for the NF* quick checks and normalization. 230 * The second trie (fcdTrie above) provides data just for FCD checks. 231 * 232 * 233 * - norm32 data words from the first trie 234 * 235 * The norm32Table contains one 32-bit word "norm32" per code point. 236 * It contains the following bit fields: 237 * 31..16 extra data index, _NORM_EXTRA_SHIFT is used to shift this field down 238 * if this index is <_NORM_EXTRA_INDEX_TOP then it is an index into 239 * extraData[] where variable-length normalization data for this 240 * code point is found 241 * if this index is <_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP 242 * then this is a norm32 for a leading surrogate, and the index 243 * value is used together with the following trailing surrogate 244 * code unit in the second trie access 245 * if this index is >=_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP 246 * then this is a norm32 for a "special" character, 247 * i.e., the character is a Hangul syllable or a Jamo 248 * see _NORM_EXTRA_HANGUL etc. 249 * generally, instead of extracting this index from the norm32 and 250 * comparing it with the above constants, 251 * the normalization code compares the entire norm32 value 252 * with _NORM_MIN_SPECIAL, _NORM_SURROGATES_TOP, _NORM_MIN_HANGUL etc. 253 * 254 * 15..8 combining class (cc) according to UnicodeData.txt 255 * 256 * 7..6 _NORM_COMBINES_ANY flags, used in composition to see if a character 257 * combines with any following or preceding character(s) 258 * at all 259 * 7 _NORM_COMBINES_BACK 260 * 6 _NORM_COMBINES_FWD 261 * 262 * 5..0 quick check flags, set for "no" or "maybe", with separate flags for 263 * each normalization form 264 * the higher bits are "maybe" flags; for NF*D there are no such flags 265 * the lower bits are "no" flags for all forms, in the same order 266 * as the "maybe" flags, 267 * which is (MSB to LSB): NFKD NFD NFKC NFC 268 * 5..4 _NORM_QC_ANY_MAYBE 269 * 3..0 _NORM_QC_ANY_NO 270 * see further related constants 271 * 272 * 273 * - Extra data per code point 274 * 275 * "Extra data" is referenced by the index in norm32. 276 * It is variable-length data. It is only present, and only those parts 277 * of it are, as needed for a given character. 278 * The norm32 extra data index is added to the beginning of extraData[] 279 * to get to a vector of 16-bit words with data at the following offsets: 280 * 281 * [-1] Combining index for composition. 282 * Stored only if norm32&_NORM_COMBINES_ANY . 283 * [0] Lengths of the canonical and compatibility decomposition strings. 284 * Stored only if there are decompositions, i.e., 285 * if norm32&(_NORM_QC_NFD|_NORM_QC_NFKD) 286 * High byte: length of NFKD, or 0 if none 287 * Low byte: length of NFD, or 0 if none 288 * Each length byte also has another flag: 289 * Bit 7 of a length byte is set if there are non-zero 290 * combining classes (cc's) associated with the respective 291 * decomposition. If this flag is set, then the decomposition 292 * is preceded by a 16-bit word that contains the 293 * leading and trailing cc's. 294 * Bits 6..0 of a length byte are the length of the 295 * decomposition string, not counting the cc word. 296 * [1..n] NFD 297 * [n+1..] NFKD 298 * 299 * Each of the two decompositions consists of up to two parts: 300 * - The 16-bit words with the leading and trailing cc's. 301 * This is only stored if bit 7 of the corresponding length byte 302 * is set. In this case, at least one of the cc's is not zero. 303 * High byte: leading cc==cc of the first code point in the decomposition string 304 * Low byte: trailing cc==cc of the last code point in the decomposition string 305 * - The decomposition string in UTF-16, with length code units. 306 * 307 * 308 * - Combining indexes and combiningTable[] 309 * 310 * Combining indexes are stored at the [-1] offset of the extra data 311 * if the character combines forward or backward with any other characters. 312 * They are used for (re)composition in NF*C. 313 * Values of combining indexes are arranged according to whether a character 314 * combines forward, backward, or both ways: 315 * forward-only < both ways < backward-only 316 * 317 * The index values for forward-only and both-ways combining characters 318 * are indexes into the combiningTable[]. 319 * The index values for backward-only combining characters are simply 320 * incremented from the preceding index values to be unique. 321 * 322 * In the combiningTable[], a variable-length list 323 * of variable-length (back-index, code point) pair entries is stored 324 * for each forward-combining character. 325 * 326 * These back-indexes are the combining indexes of both-ways or backward-only 327 * combining characters that the forward-combining character combines with. 328 * 329 * Each list is sorted in ascending order of back-indexes. 330 * Each list is terminated with the last back-index having bit 15 set. 331 * 332 * Each pair (back-index, code point) takes up either 2 or 3 333 * 16-bit words. 334 * The first word of a list entry is the back-index, with its bit 15 set if 335 * this is the last pair in the list. 336 * 337 * The second word contains flags in bits 15..13 that determine 338 * if there is a third word and how the combined character is encoded: 339 * 15 set if there is a third word in this list entry 340 * 14 set if the result is a supplementary character 341 * 13 set if the result itself combines forward 342 * 343 * According to these bits 15..14 of the second word, 344 * the result character is encoded as follows: 345 * 00 or 01 The result is <=0x1fff and stored in bits 12..0 of 346 * the second word. 347 * 10 The result is 0x2000..0xffff and stored in the third word. 348 * Bits 12..0 of the second word are not used. 349 * 11 The result is a supplementary character. 350 * Bits 9..0 of the leading surrogate are in bits 9..0 of 351 * the second word. 352 * Add 0xd800 to these bits to get the complete surrogate. 353 * Bits 12..10 of the second word are not used. 354 * The trailing surrogate is stored in the third word. 355 * 356 * 357 * - FCD trie 358 * 359 * The FCD trie is very simple. 360 * It is a folded trie with 16-bit data words. 361 * In each word, the high byte contains the leading cc of the character, 362 * and the low byte contains the trailing cc of the character. 363 * These cc's are the cc's of the first and last code points in the 364 * canonical decomposition of the character. 365 * 366 * Since all 16 bits are used for cc's, lead surrogates must be tested 367 * by checking the code unit instead of the trie data. 368 * This is done only if the 16-bit data word is not zero. 369 * If the code unit is a leading surrogate and the data word is not zero, 370 * then instead of cc's it contains the offset for the second trie lookup. 371 * 372 * 373 * - Auxiliary trie and data 374 * 375 * The auxiliary 16-bit trie contains data for additional properties. 376 * Bits 377 * 15..13 reserved 378 * 12 not NFC_Skippable (f) (formatVersion>=2.2) 379 * 11 flag: not a safe starter for canonical closure 380 * 10 composition exclusion 381 * 9.. 0 index into extraData[] to FC_NFKC_Closure string 382 * (not for lead surrogate), 383 * or lead surrogate offset (for lead surrogate, if 9..0 not zero) 384 * 385 * - FC_NFKC_Closure strings in extraData[] 386 * 387 * Strings are either stored as a single code unit or as the length 388 * followed by that many units. 389 * const UChar *s=extraData+(index from auxTrie data bits 9..0); 390 * int32_t length; 391 * if(*s<0xff00) { 392 * // s points to the single-unit string 393 * length=1; 394 * } else { 395 * length=*s&0xff; 396 * ++s; 397 * } 398 * 399 * Conditions for "NF* Skippable" from Mark Davis' com.ibm.text.UCD.NFSkippable: 400 * (used in NormalizerTransliterator) 401 * 402 * A skippable character is 403 * a) unassigned, or ALL of the following: 404 * b) of combining class 0. 405 * c) not decomposed by this normalization form. 406 * AND if NFC or NFKC, 407 * d) can never compose with a previous character. 408 * e) can never compose with a following character. 409 * f) can never change if another character is added. 410 * Example: a-breve might satisfy all but f, but if you 411 * add an ogonek it changes to a-ogonek + breve 412 * 413 * a)..e) must be tested from norm32. 414 * Since f) is more complicated, the (not-)NFC_Skippable flag (f) is built 415 * into the auxiliary trie. 416 * The same bit is used for NFC and NFKC; (c) differs for them. 417 * As usual, we build the "not skippable" flags so that unassigned 418 * code points get a 0 bit. 419 * This bit is only valid after (a)..(e) test FALSE; test NFD_NO before (f) as well. 420 * Test Hangul LV syllables entirely in code. 421 * 422 * 423 * - structure inside canonStartSets[] 424 * 425 * This array maps from code points c to sets of code points (USerializedSet). 426 * The result sets are the code points whose canonical decompositions start 427 * with c. 428 * 429 * canonStartSets[] contains the following sub-arrays: 430 * 431 * indexes[_NORM_SET_INDEX_TOP] 432 * - contains lengths of sub-arrays etc. 433 * 434 * startSets[indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH]-_NORM_SET_INDEX_TOP] 435 * - contains serialized sets (USerializedSet) of canonical starters for 436 * enumerating canonically equivalent strings 437 * indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH] includes _NORM_SET_INDEX_TOP 438 * for details about the structure see uset.c 439 * 440 * bmpTable[indexes[_NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH]] 441 * - a sorted search table for BMP code points whose results are 442 * either indexes to USerializedSets or single code points for 443 * single-code point sets; 444 * each entry is a pair of { code point, result } with result=(binary) yy xxxxxx xxxxxxxx 445 * if yy==01 then there is a USerializedSet at canonStartSets+x 446 * else build a USerializedSet with result as the single code point 447 * 448 * suppTable[indexes[_NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH]] 449 * - a sorted search table for supplementary code points whose results are 450 * either indexes to USerializedSets or single code points for 451 * single-code point sets; 452 * each entry is a triplet of { high16(cp), low16(cp), result } 453 * each code point's high-word may contain extra data in bits 15..5: 454 * if the high word has bit 15 set, then build a set with a single code point 455 * which is (((high16(cp)&0x1f00)<<8)|result; 456 * else there is a USerializedSet at canonStartSets+result 457 * 458 * FormatVersion 2.3 adds 2 serialized sets for normalization exclusions. 459 * They are stored in the data file so that the runtime normalization code need 460 * not depend on other properties and their data and implementation files. 461 * The _NORM_SET_INDEX_NX_..._OFFSET offsets in the canonStartSets index table 462 * give the location for each set. 463 * There is no set stored for UNORM_NX_HANGUL because it's trivial to create 464 * without using properties. 465 * 466 * Set contents: 467 * 468 * _NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET (for UNORM_NX_CJK_COMPAT) 469 * [[:Ideographic:]&[:NFD_QC=No:]] 470 * =[CJK Ideographs]&[has canonical decomposition] 471 * 472 * _NORM_SET_INDEX_NX_UNICODE32_OFFSET (for UNORM_UNICODE_3_2) 473 * [:^Age=3.2:] 474 * =set with all code points that were not designated by the specified Unicode version 475 * 476 * _NORM_SET_INDEX_NX_RESERVED_OFFSET 477 * This is an offset that points to where the next, future set would start. 478 * Currently it indicates where the previous set ends, and thus its length. 479 * The name for this enum constant may in the future be applied to different 480 * index slots. In order to get the limit of a set, use its index slot and 481 * the immediately following one regardless of that one's enum name. 482 */ 483 484#endif /* #if !UCONFIG_NO_NORMALIZATION */ 485 486#endif 487