| 1 | /* |
| 2 | ******************************************************************************* |
| 3 | * |
| 4 | * Copyright (C) 2001-2004, 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 | #ifdef XP_CPLUSPLUS |
| 25 | #include "unicode/uniset.h" |
| 26 | #endif |
| 27 | |
| 28 | #include "unicode/uiter.h" |
| 29 | #include "unicode/unorm.h" |
| 30 | #include "unicode/uset.h" |
| 31 | #include "utrie.h" |
| 32 | #include "ustr_imp.h" |
| 33 | #include "udataswp.h" |
| 34 | |
| 35 | /* |
| 36 | * This new implementation of the normalization code loads its data from |
| 37 | * unorm.icu, which is generated with the gennorm tool. |
| 38 | * The format of that file is described at the end of this file. |
| 39 | */ |
| 40 | |
| 41 | /* norm32 value constants */ |
| 42 | enum { |
| 43 | /* quick check flags 0..3 set mean "no" for their forms */ |
| 44 | _NORM_QC_NFC=0x11, /* no|maybe */ |
| 45 | _NORM_QC_NFKC=0x22, /* no|maybe */ |
| 46 | _NORM_QC_NFD=4, /* no */ |
| 47 | _NORM_QC_NFKD=8, /* no */ |
| 48 | |
| 49 | _NORM_QC_ANY_NO=0xf, |
| 50 | |
| 51 | /* quick check flags 4..5 mean "maybe" for their forms; test flags>=_NORM_QC_MAYBE */ |
| 52 | _NORM_QC_MAYBE=0x10, |
| 53 | _NORM_QC_ANY_MAYBE=0x30, |
| 54 | |
| 55 | _NORM_QC_MASK=0x3f, |
| 56 | |
| 57 | _NORM_COMBINES_FWD=0x40, |
| 58 | _NORM_COMBINES_BACK=0x80, |
| 59 | _NORM_COMBINES_ANY=0xc0, |
| 60 | |
| 61 | _NORM_CC_SHIFT=8, /* UnicodeData.txt combining class in bits 15..8 */ |
| 62 | _NORM_CC_MASK=0xff00, |
| 63 | |
| 64 | _NORM_EXTRA_SHIFT=16, /* 16 bits for the index to UChars and other extra data */ |
| 65 | _NORM_EXTRA_INDEX_TOP=0xfc00, /* start of surrogate specials after shift */ |
| 66 | |
| 67 | _NORM_EXTRA_SURROGATE_MASK=0x3ff, |
| 68 | _NORM_EXTRA_SURROGATE_TOP=0x3f0, /* hangul etc. */ |
| 69 | |
| 70 | _NORM_EXTRA_HANGUL=_NORM_EXTRA_SURROGATE_TOP, |
| 71 | _NORM_EXTRA_JAMO_L, |
| 72 | _NORM_EXTRA_JAMO_V, |
| 73 | _NORM_EXTRA_JAMO_T |
| 74 | }; |
| 75 | |
| 76 | /* norm32 value constants using >16 bits */ |
| 77 | #define _NORM_MIN_SPECIAL 0xfc000000 |
| 78 | #define _NORM_SURROGATES_TOP 0xfff00000 |
| 79 | #define _NORM_MIN_HANGUL 0xfff00000 |
| 80 | #define _NORM_MIN_JAMO_V 0xfff20000 |
| 81 | #define _NORM_JAMO_V_TOP 0xfff30000 |
| 82 | |
| 83 | /* value constants for auxTrie */ |
| 84 | enum { |
| 85 | _NORM_AUX_COMP_EX_SHIFT=10, |
| 86 | _NORM_AUX_UNSAFE_SHIFT=11, |
| 87 | _NORM_AUX_NFC_SKIPPABLE_F_SHIFT=12 |
| 88 | }; |
| 89 | |
| 90 | #define _NORM_AUX_MAX_FNC ((int32_t)1<<_NORM_AUX_COMP_EX_SHIFT) |
| 91 | |
| 92 | #define _NORM_AUX_FNC_MASK (uint32_t)(_NORM_AUX_MAX_FNC-1) |
| 93 | #define _NORM_AUX_COMP_EX_MASK ((uint32_t)1<<_NORM_AUX_COMP_EX_SHIFT) |
| 94 | #define _NORM_AUX_UNSAFE_MASK ((uint32_t)1<<_NORM_AUX_UNSAFE_SHIFT) |
| 95 | #define _NORM_AUX_NFC_SKIP_F_MASK ((uint32_t)1<<_NORM_AUX_NFC_SKIPPABLE_F_SHIFT) |
| 96 | |
| 97 | /* canonStartSets[0..31] contains indexes for what is in the array */ |
| 98 | enum { |
| 99 | _NORM_SET_INDEX_CANON_SETS_LENGTH, /* number of uint16_t in canonical starter sets */ |
| 100 | _NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH, /* number of uint16_t in the BMP search table (contains pairs) */ |
| 101 | _NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH,/* number of uint16_t in the supplementary search table (contains triplets) */ |
| 102 | |
| 103 | /* from formatVersion 2.3: */ |
| 104 | _NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET, /* uint16_t offset from canonStartSets[0] to the |
| 105 | exclusion set for CJK compatibility characters */ |
| 106 | _NORM_SET_INDEX_NX_UNICODE32_OFFSET, /* uint16_t offset from canonStartSets[0] to the |
| 107 | exclusion set for Unicode 3.2 characters */ |
| 108 | _NORM_SET_INDEX_NX_RESERVED_OFFSET, /* uint16_t offset from canonStartSets[0] to the |
| 109 | end of the previous exclusion set */ |
| 110 | |
| 111 | _NORM_SET_INDEX_TOP=32 /* changing this requires a new formatVersion */ |
| 112 | }; |
| 113 | |
| 114 | /* more constants for canonical starter sets */ |
| 115 | |
| 116 | /* 14 bit indexes to canonical USerializedSets */ |
| 117 | #define _NORM_MAX_CANON_SETS 0x4000 |
| 118 | |
| 119 | /* single-code point BMP sets are encoded directly in the search table except if result=0x4000..0x7fff */ |
| 120 | #define _NORM_CANON_SET_BMP_MASK 0xc000 |
| 121 | #define _NORM_CANON_SET_BMP_IS_INDEX 0x4000 |
| 122 | |
| 123 | /* indexes[] value names */ |
| 124 | enum { |
| 125 | _NORM_INDEX_TRIE_SIZE, /* number of bytes in normalization trie */ |
| 126 | _NORM_INDEX_UCHAR_COUNT, /* number of UChars in extra data */ |
| 127 | |
| 128 | _NORM_INDEX_COMBINE_DATA_COUNT, /* number of uint16_t words for combining data */ |
| 129 | _NORM_INDEX_COMBINE_FWD_COUNT, /* number of code points that combine forward */ |
| 130 | _NORM_INDEX_COMBINE_BOTH_COUNT, /* number of code points that combine forward and backward */ |
| 131 | _NORM_INDEX_COMBINE_BACK_COUNT, /* number of code points that combine backward */ |
| 132 | |
| 133 | _NORM_INDEX_MIN_NFC_NO_MAYBE, /* first code point with quick check NFC NO/MAYBE */ |
| 134 | _NORM_INDEX_MIN_NFKC_NO_MAYBE, /* first code point with quick check NFKC NO/MAYBE */ |
| 135 | _NORM_INDEX_MIN_NFD_NO_MAYBE, /* first code point with quick check NFD NO/MAYBE */ |
| 136 | _NORM_INDEX_MIN_NFKD_NO_MAYBE, /* first code point with quick check NFKD NO/MAYBE */ |
| 137 | |
| 138 | _NORM_INDEX_FCD_TRIE_SIZE, /* number of bytes in FCD trie */ |
| 139 | |
| 140 | _NORM_INDEX_AUX_TRIE_SIZE, /* number of bytes in the auxiliary trie */ |
| 141 | _NORM_INDEX_CANON_SET_COUNT, /* number of uint16_t in the array of serialized USet */ |
| 142 | |
| 143 | _NORM_INDEX_TOP=32 /* changing this requires a new formatVersion */ |
| 144 | }; |
| 145 | |
| 146 | enum { |
| 147 | /* FCD check: everything below this code point is known to have a 0 lead combining class */ |
| 148 | _NORM_MIN_WITH_LEAD_CC=0x300 |
| 149 | }; |
| 150 | |
| 151 | enum { |
| 152 | /** |
| 153 | * Bit 7 of the length byte for a decomposition string in extra data is |
| 154 | * a flag indicating whether the decomposition string is |
| 155 | * preceded by a 16-bit word with the leading and trailing cc |
| 156 | * of the decomposition (like for A-umlaut); |
| 157 | * if not, then both cc's are zero (like for compatibility ideographs). |
| 158 | */ |
| 159 | _NORM_DECOMP_FLAG_LENGTH_HAS_CC=0x80, |
| 160 | /** |
| 161 | * Bits 6..0 of the length byte contain the actual length. |
| 162 | */ |
| 163 | _NORM_DECOMP_LENGTH_MASK=0x7f |
| 164 | }; |
| 165 | |
| 166 | #endif /* #if !UCONFIG_NO_NORMALIZATION */ |
| 167 | |
| 168 | /* Korean Hangul and Jamo constants */ |
| 169 | enum { |
| 170 | JAMO_L_BASE=0x1100, /* "lead" jamo */ |
| 171 | JAMO_V_BASE=0x1161, /* "vowel" jamo */ |
| 172 | JAMO_T_BASE=0x11a7, /* "trail" jamo */ |
| 173 | |
| 174 | HANGUL_BASE=0xac00, |
| 175 | |
| 176 | JAMO_L_COUNT=19, |
| 177 | JAMO_V_COUNT=21, |
| 178 | JAMO_T_COUNT=28, |
| 179 | |
| 180 | HANGUL_COUNT=JAMO_L_COUNT*JAMO_V_COUNT*JAMO_T_COUNT |
| 181 | }; |
| 182 | |
| 183 | #if !UCONFIG_NO_NORMALIZATION |
| 184 | |
| 185 | /* Constants for options flags for normalization. @draft ICU 2.6 */ |
| 186 | enum { |
| 187 | /** Options bit 0, do not decompose Hangul syllables. @draft ICU 2.6 */ |
| 188 | UNORM_NX_HANGUL=1, |
| 189 | /** Options bit 1, do not decompose CJK compatibility characters. @draft ICU 2.6 */ |
| 190 | UNORM_NX_CJK_COMPAT=2, |
| 191 | /** |
| 192 | * Options bit 8, use buggy recomposition described in |
| 193 | * Unicode Public Review Issue #29 |
| 194 | * at http://www.unicode.org/review/resolved-pri.html#pri29 |
| 195 | * |
| 196 | * Used in IDNA implementation according to strict interpretation |
| 197 | * of IDNA definition based on Unicode 3.2 which predates PRI #29. |
| 198 | */ |
| 199 | UNORM_BEFORE_PRI_29=0x100 |
| 200 | }; |
| 201 | |
| 202 | /** |
| 203 | * Is the normalizer data loaded? |
| 204 | * This is used internally before other internal normalizer functions |
| 205 | * are called. |
| 206 | * It saves this check in each of many normalization calls that |
| 207 | * are made for, e.g., collation. |
| 208 | * |
| 209 | * @param pErrorCode as usual |
| 210 | * @return boolean value for whether the normalization data is loaded |
| 211 | * |
| 212 | * @internal |
| 213 | */ |
| 214 | U_CAPI UBool U_EXPORT2 |
| 215 | unorm_haveData(UErrorCode *pErrorCode); |
| 216 | |
| 217 | /** |
| 218 | * Internal API for normalizing. |
| 219 | * Does not check for bad input. |
| 220 | * @internal |
| 221 | */ |
| 222 | U_CAPI int32_t U_EXPORT2 |
| 223 | unorm_internalNormalize(UChar *dest, int32_t destCapacity, |
| 224 | const UChar *src, int32_t srcLength, |
| 225 | UNormalizationMode mode, int32_t options, |
| 226 | UErrorCode *pErrorCode); |
| 227 | |
| 228 | #ifdef XP_CPLUSPLUS |
| 229 | |
| 230 | /** |
| 231 | * Internal API for normalizing. |
| 232 | * Does not check for bad input. |
| 233 | * Requires _haveData() to be true. |
| 234 | * @internal |
| 235 | */ |
| 236 | U_CFUNC int32_t |
| 237 | unorm_internalNormalizeWithNX(UChar *dest, int32_t destCapacity, |
| 238 | const UChar *src, int32_t srcLength, |
| 239 | UNormalizationMode mode, int32_t options, const UnicodeSet *nx, |
| 240 | UErrorCode *pErrorCode); |
| 241 | |
| 242 | #endif |
| 243 | |
| 244 | /** |
| 245 | * internal API, used by normlzr.cpp |
| 246 | * @internal |
| 247 | */ |
| 248 | U_CAPI int32_t U_EXPORT2 |
| 249 | unorm_decompose(UChar *dest, int32_t destCapacity, |
| 250 | const UChar *src, int32_t srcLength, |
| 251 | UBool compat, int32_t options, |
| 252 | UErrorCode *pErrorCode); |
| 253 | |
| 254 | /** |
| 255 | * internal API, used by normlzr.cpp |
| 256 | * @internal |
| 257 | */ |
| 258 | U_CAPI int32_t U_EXPORT2 |
| 259 | unorm_compose(UChar *dest, int32_t destCapacity, |
| 260 | const UChar *src, int32_t srcLength, |
| 261 | UBool compat, int32_t options, |
| 262 | UErrorCode *pErrorCode); |
| 263 | |
| 264 | #ifdef XP_CPLUSPLUS |
| 265 | |
| 266 | /** |
| 267 | * internal API, used by unormcmp.cpp |
| 268 | * @internal |
| 269 | */ |
| 270 | U_CFUNC UNormalizationCheckResult |
| 271 | unorm_internalQuickCheck(const UChar *src, |
| 272 | int32_t srcLength, |
| 273 | UNormalizationMode mode, |
| 274 | UBool allowMaybe, |
| 275 | const UnicodeSet *nx, |
| 276 | UErrorCode *pErrorCode); |
| 277 | |
| 278 | #endif |
| 279 | |
| 280 | #endif /* #if !UCONFIG_NO_NORMALIZATION */ |
| 281 | |
| 282 | /** |
| 283 | * Internal option for unorm_cmpEquivFold() for decomposing. |
| 284 | * If not set, just do strcasecmp(). |
| 285 | * @internal |
| 286 | */ |
| 287 | #define _COMPARE_EQUIV 0x80000 |
| 288 | |
| 289 | #ifndef U_COMPARE_IGNORE_CASE |
| 290 | /* see also unorm.h */ |
| 291 | /** |
| 292 | * Option bit for unorm_compare: |
| 293 | * Perform case-insensitive comparison. |
| 294 | * @draft ICU 2.2 |
| 295 | */ |
| 296 | #define U_COMPARE_IGNORE_CASE 0x10000 |
| 297 | #endif |
| 298 | |
| 299 | /** |
| 300 | * Internal option for unorm_cmpEquivFold() for strncmp style. |
| 301 | * If set, checks for both string length and terminating NUL. |
| 302 | * @internal |
| 303 | */ |
| 304 | #define _STRNCMP_STYLE 0x1000 |
| 305 | |
| 306 | #if !UCONFIG_NO_NORMALIZATION |
| 307 | |
| 308 | /** |
| 309 | * Internal API to get the 16-bit FCD value (lccc + tccc) for c, |
| 310 | * for u_getIntPropertyValue(). |
| 311 | * @internal |
| 312 | */ |
| 313 | U_CAPI uint16_t U_EXPORT2 |
| 314 | unorm_getFCD16FromCodePoint(UChar32 c); |
| 315 | |
| 316 | /** |
| 317 | * Internal API, used by collation code. |
| 318 | * Get access to the internal FCD trie table to be able to perform |
| 319 | * incremental, per-code unit, FCD checks in collation. |
| 320 | * One pointer is sufficient because the trie index values are offset |
| 321 | * by the index size, so that the same pointer is used to access the trie data. |
| 322 | * @internal |
| 323 | */ |
| 324 | U_CAPI const uint16_t * U_EXPORT2 |
| 325 | unorm_getFCDTrie(UErrorCode *pErrorCode); |
| 326 | |
| 327 | #ifdef XP_CPLUSPLUS |
| 328 | |
| 329 | U_NAMESPACE_BEGIN |
| 330 | /** |
| 331 | * Internal API, used by collation code. |
| 332 | * Get the FCD value for a code unit, with |
| 333 | * bits 15..8 lead combining class |
| 334 | * bits 7..0 trail combining class |
| 335 | * |
| 336 | * If c is a lead surrogate and the value is not 0, |
| 337 | * then instead of combining classes the value |
| 338 | * is used in unorm_getFCD16FromSurrogatePair() to get the real value |
| 339 | * of the supplementary code point. |
| 340 | * |
| 341 | * @internal |
| 342 | */ |
| 343 | inline uint16_t |
| 344 | unorm_getFCD16(const uint16_t *fcdTrieIndex, UChar c) { |
| 345 | return |
| 346 | fcdTrieIndex[ |
| 347 | (fcdTrieIndex[ |
| 348 | c>>UTRIE_SHIFT |
| 349 | ]<<UTRIE_INDEX_SHIFT)+ |
| 350 | (c&UTRIE_MASK) |
| 351 | ]; |
| 352 | } |
| 353 | |
| 354 | /** |
| 355 | * Internal API, used by collation code. |
| 356 | * Get the FCD value for a supplementary code point, with |
| 357 | * bits 15..8 lead combining class |
| 358 | * bits 7..0 trail combining class |
| 359 | * |
| 360 | * @param fcd16 The FCD value for the lead surrogate, not 0. |
| 361 | * @param c2 The trail surrogate code unit. |
| 362 | * |
| 363 | * @internal |
| 364 | */ |
| 365 | inline uint16_t |
| 366 | unorm_getFCD16FromSurrogatePair(const uint16_t *fcdTrieIndex, uint16_t fcd16, UChar c2) { |
| 367 | return |
| 368 | fcdTrieIndex[ |
| 369 | (fcdTrieIndex[ |
| 370 | (int32_t)fcd16+((c2&0x3ff)>>UTRIE_SHIFT) |
| 371 | ]<<UTRIE_INDEX_SHIFT)+ |
| 372 | (c2&UTRIE_MASK) |
| 373 | ]; |
| 374 | } |
| 375 | |
| 376 | U_NAMESPACE_END |
| 377 | |
| 378 | #endif |
| 379 | |
| 380 | /** |
| 381 | * internal API, used by StringPrep |
| 382 | * @internal |
| 383 | */ |
| 384 | U_CAPI void U_EXPORT2 |
| 385 | unorm_getUnicodeVersion(UVersionInfo *versionInfo, UErrorCode *pErrorCode); |
| 386 | |
| 387 | /** |
| 388 | * Get the canonical decomposition for one code point. |
| 389 | * Requires unorm_haveData() and buffer!=NULL and pLength!=NULL. |
| 390 | * @param c code point |
| 391 | * @param buffer out-only buffer for algorithmic decompositions of Hangul |
| 392 | * @param length out-only, takes the length of the decomposition, if any |
| 393 | * @return pointer to decomposition, or 0 if none |
| 394 | * @internal |
| 395 | */ |
| 396 | U_CFUNC const UChar * |
| 397 | unorm_getCanonicalDecomposition(UChar32 c, UChar buffer[4], int32_t *pLength); |
| 398 | |
| 399 | /** |
| 400 | * internal API, used by the canonical iterator |
| 401 | * TODO Consider using signature similar to unorm_getCanonicalDecomposition() |
| 402 | * for more efficiency |
| 403 | * @internal |
| 404 | */ |
| 405 | U_CAPI int32_t U_EXPORT2 |
| 406 | unorm_getDecomposition(UChar32 c, UBool compat, |
| 407 | UChar *dest, int32_t destCapacity); |
| 408 | |
| 409 | /** |
| 410 | * internal API, used by uprops.cpp |
| 411 | * @internal |
| 412 | */ |
| 413 | U_CAPI UBool U_EXPORT2 |
| 414 | unorm_internalIsFullCompositionExclusion(UChar32 c); |
| 415 | |
| 416 | /** |
| 417 | * Internal API, used by enumeration of canonically equivalent strings |
| 418 | * @internal |
| 419 | */ |
| 420 | U_CAPI UBool U_EXPORT2 |
| 421 | unorm_isCanonSafeStart(UChar32 c); |
| 422 | |
| 423 | /** |
| 424 | * Internal API, used by enumeration of canonically equivalent strings |
| 425 | * @internal |
| 426 | */ |
| 427 | U_CAPI UBool U_EXPORT2 |
| 428 | unorm_getCanonStartSet(UChar32 c, USerializedSet *fillSet); |
| 429 | |
| 430 | /** |
| 431 | * Is c an NF<mode>-skippable code point? See unormimp.h. |
| 432 | * @internal |
| 433 | */ |
| 434 | U_CAPI UBool U_EXPORT2 |
| 435 | unorm_isNFSkippable(UChar32 c, UNormalizationMode mode); |
| 436 | |
| 437 | #ifdef XP_CPLUSPLUS |
| 438 | |
| 439 | /** |
| 440 | * Get normalization exclusion set for the options. |
| 441 | * Requires unorm_haveData(). |
| 442 | * @internal |
| 443 | */ |
| 444 | U_CFUNC const UnicodeSet * |
| 445 | unorm_getNX(int32_t options, UErrorCode *pErrorCode); |
| 446 | |
| 447 | #endif |
| 448 | |
| 449 | /** |
| 450 | * Enumerate each normalization data trie and add the |
| 451 | * start of each range of same properties to the set. |
| 452 | * @internal |
| 453 | */ |
| 454 | U_CAPI void U_EXPORT2 |
| 455 | unorm_addPropertyStarts(USetAdder *sa, UErrorCode *pErrorCode); |
| 456 | |
| 457 | /** |
| 458 | * Swap unorm.icu. See udataswp.h. |
| 459 | * @internal |
| 460 | */ |
| 461 | U_CAPI int32_t U_EXPORT2 |
| 462 | unorm_swap(const UDataSwapper *ds, |
| 463 | const void *inData, int32_t length, void *outData, |
| 464 | UErrorCode *pErrorCode); |
| 465 | |
| 466 | /** |
| 467 | * Get the NF*_QC property for a code point, for u_getIntPropertyValue(). |
| 468 | * @internal |
| 469 | */ |
| 470 | U_CAPI UNormalizationCheckResult U_EXPORT2 |
| 471 | unorm_getQuickCheck(UChar32 c, UNormalizationMode mode); |
| 472 | |
| 473 | /** |
| 474 | * Description of the format of unorm.icu version 2.3. |
| 475 | * |
| 476 | * Main change from version 1 to version 2: |
| 477 | * Use of new, common UTrie instead of normalization-specific tries. |
| 478 | * Change to version 2.1: add third/auxiliary trie with associated data. |
| 479 | * Change to version 2.2: add skippable (f) flag data (_NORM_AUX_NFC_SKIP_F_MASK). |
| 480 | * Change to version 2.3: add serialized sets for normalization exclusions |
| 481 | * stored inside canonStartSets[] |
| 482 | * |
| 483 | * For more details of how to use the data structures see the code |
| 484 | * in unorm.cpp (runtime normalization code) and |
| 485 | * in gennorm.c and gennorm/store.c (build-time data generation). |
| 486 | * |
| 487 | * For the serialized format of UTrie see utrie.c/UTrieHeader. |
| 488 | * |
| 489 | * - Overall partition |
| 490 | * |
| 491 | * unorm.dat customarily begins with a UDataInfo structure, see udata.h and .c. |
| 492 | * After that there are the following structures: |
| 493 | * |
| 494 | * int32_t indexes[_NORM_INDEX_TOP]; -- _NORM_INDEX_TOP=32, see enum in this file |
| 495 | * |
| 496 | * UTrie normTrie; -- size in bytes=indexes[_NORM_INDEX_TRIE_SIZE] |
| 497 | * |
| 498 | * uint16_t extraData[extraDataTop]; -- extraDataTop=indexes[_NORM_INDEX_UCHAR_COUNT] |
| 499 | * extraData[0] contains the number of units for |
| 500 | * FC_NFKC_Closure (formatVersion>=2.1) |
| 501 | * |
| 502 | * uint16_t combiningTable[combiningTableTop]; -- combiningTableTop=indexes[_NORM_INDEX_COMBINE_DATA_COUNT] |
| 503 | * combiningTableTop may include one 16-bit padding unit |
| 504 | * to make sure that fcdTrie is 32-bit-aligned |
| 505 | * |
| 506 | * UTrie fcdTrie; -- size in bytes=indexes[_NORM_INDEX_FCD_TRIE_SIZE] |
| 507 | * |
| 508 | * UTrie auxTrie; -- size in bytes=indexes[_NORM_INDEX_AUX_TRIE_SIZE] |
| 509 | * |
| 510 | * uint16_t canonStartSets[canonStartSetsTop] -- canonStartSetsTop=indexes[_NORM_INDEX_CANON_SET_COUNT] |
| 511 | * serialized USets and binary search tables, see below |
| 512 | * |
| 513 | * |
| 514 | * The indexes array contains lengths and sizes of the following arrays and structures |
| 515 | * as well as the following values: |
| 516 | * indexes[_NORM_INDEX_COMBINE_FWD_COUNT]=combineFwdTop |
| 517 | * -- one more than the highest combining index computed for forward-only-combining characters |
| 518 | * indexes[_NORM_INDEX_COMBINE_BOTH_COUNT]=combineBothTop-combineFwdTop |
| 519 | * -- number of combining indexes computed for both-ways-combining characters |
| 520 | * indexes[_NORM_INDEX_COMBINE_BACK_COUNT]=combineBackTop-combineBothTop |
| 521 | * -- number of combining indexes computed for backward-only-combining characters |
| 522 | * |
| 523 | * indexes[_NORM_INDEX_MIN_NF*_NO_MAYBE] (where *={ C, D, KC, KD }) |
| 524 | * -- first code point with a quick check NF* value of NO/MAYBE |
| 525 | * |
| 526 | * |
| 527 | * - Tries |
| 528 | * |
| 529 | * The main structures are two UTrie tables ("compact arrays"), |
| 530 | * each with one index array and one data array. |
| 531 | * See utrie.h and utrie.c. |
| 532 | * |
| 533 | * |
| 534 | * - Tries in unorm.dat |
| 535 | * |
| 536 | * The first trie (normTrie above) |
| 537 | * provides data for the NF* quick checks and normalization. |
| 538 | * The second trie (fcdTrie above) provides data just for FCD checks. |
| 539 | * |
| 540 | * |
| 541 | * - norm32 data words from the first trie |
| 542 | * |
| 543 | * The norm32Table contains one 32-bit word "norm32" per code point. |
| 544 | * It contains the following bit fields: |
| 545 | * 31..16 extra data index, _NORM_EXTRA_SHIFT is used to shift this field down |
| 546 | * if this index is <_NORM_EXTRA_INDEX_TOP then it is an index into |
| 547 | * extraData[] where variable-length normalization data for this |
| 548 | * code point is found |
| 549 | * if this index is <_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP |
| 550 | * then this is a norm32 for a leading surrogate, and the index |
| 551 | * value is used together with the following trailing surrogate |
| 552 | * code unit in the second trie access |
| 553 | * if this index is >=_NORM_EXTRA_INDEX_TOP+_NORM_EXTRA_SURROGATE_TOP |
| 554 | * then this is a norm32 for a "special" character, |
| 555 | * i.e., the character is a Hangul syllable or a Jamo |
| 556 | * see _NORM_EXTRA_HANGUL etc. |
| 557 | * generally, instead of extracting this index from the norm32 and |
| 558 | * comparing it with the above constants, |
| 559 | * the normalization code compares the entire norm32 value |
| 560 | * with _NORM_MIN_SPECIAL, _NORM_SURROGATES_TOP, _NORM_MIN_HANGUL etc. |
| 561 | * |
| 562 | * 15..8 combining class (cc) according to UnicodeData.txt |
| 563 | * |
| 564 | * 7..6 _NORM_COMBINES_ANY flags, used in composition to see if a character |
| 565 | * combines with any following or preceding character(s) |
| 566 | * at all |
| 567 | * 7 _NORM_COMBINES_BACK |
| 568 | * 6 _NORM_COMBINES_FWD |
| 569 | * |
| 570 | * 5..0 quick check flags, set for "no" or "maybe", with separate flags for |
| 571 | * each normalization form |
| 572 | * the higher bits are "maybe" flags; for NF*D there are no such flags |
| 573 | * the lower bits are "no" flags for all forms, in the same order |
| 574 | * as the "maybe" flags, |
| 575 | * which is (MSB to LSB): NFKD NFD NFKC NFC |
| 576 | * 5..4 _NORM_QC_ANY_MAYBE |
| 577 | * 3..0 _NORM_QC_ANY_NO |
| 578 | * see further related constants |
| 579 | * |
| 580 | * |
| 581 | * - Extra data per code point |
| 582 | * |
| 583 | * "Extra data" is referenced by the index in norm32. |
| 584 | * It is variable-length data. It is only present, and only those parts |
| 585 | * of it are, as needed for a given character. |
| 586 | * The norm32 extra data index is added to the beginning of extraData[] |
| 587 | * to get to a vector of 16-bit words with data at the following offsets: |
| 588 | * |
| 589 | * [-1] Combining index for composition. |
| 590 | * Stored only if norm32&_NORM_COMBINES_ANY . |
| 591 | * [0] Lengths of the canonical and compatibility decomposition strings. |
| 592 | * Stored only if there are decompositions, i.e., |
| 593 | * if norm32&(_NORM_QC_NFD|_NORM_QC_NFKD) |
| 594 | * High byte: length of NFKD, or 0 if none |
| 595 | * Low byte: length of NFD, or 0 if none |
| 596 | * Each length byte also has another flag: |
| 597 | * Bit 7 of a length byte is set if there are non-zero |
| 598 | * combining classes (cc's) associated with the respective |
| 599 | * decomposition. If this flag is set, then the decomposition |
| 600 | * is preceded by a 16-bit word that contains the |
| 601 | * leading and trailing cc's. |
| 602 | * Bits 6..0 of a length byte are the length of the |
| 603 | * decomposition string, not counting the cc word. |
| 604 | * [1..n] NFD |
| 605 | * [n+1..] NFKD |
| 606 | * |
| 607 | * Each of the two decompositions consists of up to two parts: |
| 608 | * - The 16-bit words with the leading and trailing cc's. |
| 609 | * This is only stored if bit 7 of the corresponding length byte |
| 610 | * is set. In this case, at least one of the cc's is not zero. |
| 611 | * High byte: leading cc==cc of the first code point in the decomposition string |
| 612 | * Low byte: trailing cc==cc of the last code point in the decomposition string |
| 613 | * - The decomposition string in UTF-16, with length code units. |
| 614 | * |
| 615 | * |
| 616 | * - Combining indexes and combiningTable[] |
| 617 | * |
| 618 | * Combining indexes are stored at the [-1] offset of the extra data |
| 619 | * if the character combines forward or backward with any other characters. |
| 620 | * They are used for (re)composition in NF*C. |
| 621 | * Values of combining indexes are arranged according to whether a character |
| 622 | * combines forward, backward, or both ways: |
| 623 | * forward-only < both ways < backward-only |
| 624 | * |
| 625 | * The index values for forward-only and both-ways combining characters |
| 626 | * are indexes into the combiningTable[]. |
| 627 | * The index values for backward-only combining characters are simply |
| 628 | * incremented from the preceding index values to be unique. |
| 629 | * |
| 630 | * In the combiningTable[], a variable-length list |
| 631 | * of variable-length (back-index, code point) pair entries is stored |
| 632 | * for each forward-combining character. |
| 633 | * |
| 634 | * These back-indexes are the combining indexes of both-ways or backward-only |
| 635 | * combining characters that the forward-combining character combines with. |
| 636 | * |
| 637 | * Each list is sorted in ascending order of back-indexes. |
| 638 | * Each list is terminated with the last back-index having bit 15 set. |
| 639 | * |
| 640 | * Each pair (back-index, code point) takes up either 2 or 3 |
| 641 | * 16-bit words. |
| 642 | * The first word of a list entry is the back-index, with its bit 15 set if |
| 643 | * this is the last pair in the list. |
| 644 | * |
| 645 | * The second word contains flags in bits 15..13 that determine |
| 646 | * if there is a third word and how the combined character is encoded: |
| 647 | * 15 set if there is a third word in this list entry |
| 648 | * 14 set if the result is a supplementary character |
| 649 | * 13 set if the result itself combines forward |
| 650 | * |
| 651 | * According to these bits 15..14 of the second word, |
| 652 | * the result character is encoded as follows: |
| 653 | * 00 or 01 The result is <=0x1fff and stored in bits 12..0 of |
| 654 | * the second word. |
| 655 | * 10 The result is 0x2000..0xffff and stored in the third word. |
| 656 | * Bits 12..0 of the second word are not used. |
| 657 | * 11 The result is a supplementary character. |
| 658 | * Bits 9..0 of the leading surrogate are in bits 9..0 of |
| 659 | * the second word. |
| 660 | * Add 0xd800 to these bits to get the complete surrogate. |
| 661 | * Bits 12..10 of the second word are not used. |
| 662 | * The trailing surrogate is stored in the third word. |
| 663 | * |
| 664 | * |
| 665 | * - FCD trie |
| 666 | * |
| 667 | * The FCD trie is very simple. |
| 668 | * It is a folded trie with 16-bit data words. |
| 669 | * In each word, the high byte contains the leading cc of the character, |
| 670 | * and the low byte contains the trailing cc of the character. |
| 671 | * These cc's are the cc's of the first and last code points in the |
| 672 | * canonical decomposition of the character. |
| 673 | * |
| 674 | * Since all 16 bits are used for cc's, lead surrogates must be tested |
| 675 | * by checking the code unit instead of the trie data. |
| 676 | * This is done only if the 16-bit data word is not zero. |
| 677 | * If the code unit is a leading surrogate and the data word is not zero, |
| 678 | * then instead of cc's it contains the offset for the second trie lookup. |
| 679 | * |
| 680 | * |
| 681 | * - Auxiliary trie and data |
| 682 | * |
| 683 | * The auxiliary 16-bit trie contains data for additional properties. |
| 684 | * Bits |
| 685 | * 15..13 reserved |
| 686 | * 12 not NFC_Skippable (f) (formatVersion>=2.2) |
| 687 | * 11 flag: not a safe starter for canonical closure |
| 688 | * 10 composition exclusion |
| 689 | * 9.. 0 index into extraData[] to FC_NFKC_Closure string |
| 690 | * (not for lead surrogate), |
| 691 | * or lead surrogate offset (for lead surrogate, if 9..0 not zero) |
| 692 | * |
| 693 | * - FC_NFKC_Closure strings in extraData[] |
| 694 | * |
| 695 | * Strings are either stored as a single code unit or as the length |
| 696 | * followed by that many units. |
| 697 | * const UChar *s=extraData+(index from auxTrie data bits 9..0); |
| 698 | * int32_t length; |
| 699 | * if(*s<0xff00) { |
| 700 | * // s points to the single-unit string |
| 701 | * length=1; |
| 702 | * } else { |
| 703 | * length=*s&0xff; |
| 704 | * ++s; |
| 705 | * } |
| 706 | * |
| 707 | * Conditions for "NF* Skippable" from Mark Davis' com.ibm.text.UCD.NFSkippable: |
| 708 | * (used in NormalizerTransliterator) |
| 709 | * |
| 710 | * A skippable character is |
| 711 | * a) unassigned, or ALL of the following: |
| 712 | * b) of combining class 0. |
| 713 | * c) not decomposed by this normalization form. |
| 714 | * AND if NFC or NFKC, |
| 715 | * d) can never compose with a previous character. |
| 716 | * e) can never compose with a following character. |
| 717 | * f) can never change if another character is added. |
| 718 | * Example: a-breve might satisfy all but f, but if you |
| 719 | * add an ogonek it changes to a-ogonek + breve |
| 720 | * |
| 721 | * a)..e) must be tested from norm32. |
| 722 | * Since f) is more complicated, the (not-)NFC_Skippable flag (f) is built |
| 723 | * into the auxiliary trie. |
| 724 | * The same bit is used for NFC and NFKC; (c) differs for them. |
| 725 | * As usual, we build the "not skippable" flags so that unassigned |
| 726 | * code points get a 0 bit. |
| 727 | * This bit is only valid after (a)..(e) test FALSE; test NFD_NO before (f) as well. |
| 728 | * Test Hangul LV syllables entirely in code. |
| 729 | * |
| 730 | * |
| 731 | * - structure inside canonStartSets[] |
| 732 | * |
| 733 | * This array maps from code points c to sets of code points (USerializedSet). |
| 734 | * The result sets are the code points whose canonical decompositions start |
| 735 | * with c. |
| 736 | * |
| 737 | * canonStartSets[] contains the following sub-arrays: |
| 738 | * |
| 739 | * indexes[_NORM_SET_INDEX_TOP] |
| 740 | * - contains lengths of sub-arrays etc. |
| 741 | * |
| 742 | * startSets[indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH]-_NORM_SET_INDEX_TOP] |
| 743 | * - contains serialized sets (USerializedSet) of canonical starters for |
| 744 | * enumerating canonically equivalent strings |
| 745 | * indexes[_NORM_SET_INDEX_CANON_SETS_LENGTH] includes _NORM_SET_INDEX_TOP |
| 746 | * for details about the structure see uset.c |
| 747 | * |
| 748 | * bmpTable[indexes[_NORM_SET_INDEX_CANON_BMP_TABLE_LENGTH]] |
| 749 | * - a sorted search table for BMP code points whose results are |
| 750 | * either indexes to USerializedSets or single code points for |
| 751 | * single-code point sets; |
| 752 | * each entry is a pair of { code point, result } with result=(binary) yy xxxxxx xxxxxxxx |
| 753 | * if yy==01 then there is a USerializedSet at canonStartSets+x |
| 754 | * else build a USerializedSet with result as the single code point |
| 755 | * |
| 756 | * suppTable[indexes[_NORM_SET_INDEX_CANON_SUPP_TABLE_LENGTH]] |
| 757 | * - a sorted search table for supplementary code points whose results are |
| 758 | * either indexes to USerializedSets or single code points for |
| 759 | * single-code point sets; |
| 760 | * each entry is a triplet of { high16(cp), low16(cp), result } |
| 761 | * each code point's high-word may contain extra data in bits 15..5: |
| 762 | * if the high word has bit 15 set, then build a set with a single code point |
| 763 | * which is (((high16(cp)&0x1f00)<<8)|result; |
| 764 | * else there is a USerializedSet at canonStartSets+result |
| 765 | * |
| 766 | * FormatVersion 2.3 adds 2 serialized sets for normalization exclusions. |
| 767 | * They are stored in the data file so that the runtime normalization code need |
| 768 | * not depend on other properties and their data and implementation files. |
| 769 | * The _NORM_SET_INDEX_NX_..._OFFSET offsets in the canonStartSets index table |
| 770 | * give the location for each set. |
| 771 | * There is no set stored for UNORM_NX_HANGUL because it's trivial to create |
| 772 | * without using properties. |
| 773 | * |
| 774 | * Set contents: |
| 775 | * |
| 776 | * _NORM_SET_INDEX_NX_CJK_COMPAT_OFFSET (for UNORM_NX_CJK_COMPAT) |
| 777 | * [[:Ideographic:]&[:NFD_QC=No:]] |
| 778 | * =[CJK Ideographs]&[has canonical decomposition] |
| 779 | * |
| 780 | * _NORM_SET_INDEX_NX_UNICODE32_OFFSET (for UNORM_UNICODE_3_2) |
| 781 | * [:^Age=3.2:] |
| 782 | * =set with all code points that were not designated by the specified Unicode version |
| 783 | * |
| 784 | * _NORM_SET_INDEX_NX_RESERVED_OFFSET |
| 785 | * This is an offset that points to where the next, future set would start. |
| 786 | * Currently it indicates where the previous set ends, and thus its length. |
| 787 | * The name for this enum constant may in the future be applied to different |
| 788 | * index slots. In order to get the limit of a set, use its index slot and |
| 789 | * the immediately following one regardless of that one's enum name. |
| 790 | */ |
| 791 | |
| 792 | #endif /* #if !UCONFIG_NO_NORMALIZATION */ |
| 793 | |
| 794 | #endif |