1 /* 2 ****************************************************************************** 3 * 4 * Copyright (C) 2003-2013, International Business Machines 5 * Corporation and others. All Rights Reserved. 6 * 7 ****************************************************************************** 8 * file name: ucnv_ext.h 9 * encoding: US-ASCII 10 * tab size: 8 (not used) 11 * indentation:4 12 * 13 * created on: 2003jun13 14 * created by: Markus W. Scherer 15 * 16 * Conversion extensions 17 */ 18 19 #ifndef __UCNV_EXT_H__ 20 #define __UCNV_EXT_H__ 21 22 #include "unicode/utypes.h" 23 24 #if !UCONFIG_NO_CONVERSION 25 26 #include "unicode/ucnv.h" 27 #include "ucnv_cnv.h" 28 29 /* 30 * See icuhtml/design/conversion/conversion_extensions.html 31 * 32 * Conversion extensions serve three purposes: 33 * 1. They support m:n mappings. 34 * 2. They support extension-only conversion files that are used together 35 * with the regular conversion data in base files. 36 * 3. They support mappings with more complicated meta data, 37 * for example "good one-way" mappings (|4). 38 * 39 * A base file may contain an extension table (explicitly requested or 40 * implicitly generated for m:n mappings), but its extension table is not 41 * used when an extension-only file is used. 42 * 43 * It is an error if a base file contains any regular (not extension) mapping 44 * from the same sequence as a mapping in the extension file 45 * because the base mapping would hide the extension mapping. 46 * 47 * 48 * Data for conversion extensions: 49 * 50 * One set of data structures per conversion direction (to/from Unicode). 51 * The data structures are sorted by input units to allow for binary search. 52 * Input sequences of more than one unit are handled like contraction tables 53 * in collation: 54 * The lookup value of a unit points to another table that is to be searched 55 * for the next unit, recursively. 56 * 57 * For conversion from Unicode, the initial code point is looked up in 58 * a 3-stage trie for speed, 59 * with an additional table of unique results to save space. 60 * 61 * Long output strings are stored in separate arrays, with length and index 62 * in the lookup tables. 63 * Output results also include a flag distinguishing roundtrip from 64 * (reverse) fallback mappings. 65 * 66 * Input Unicode strings must not begin or end with unpaired surrogates 67 * to avoid problems with matches on parts of surrogate pairs. 68 * 69 * Mappings from multiple characters (code points or codepage state 70 * table sequences) must be searched preferring the longest match. 71 * For this to work and be efficient, the variable-width table must contain 72 * all mappings that contain prefixes of the multiple characters. 73 * If an extension table is built on top of a base table in another file 74 * and a base table entry is a prefix of a multi-character mapping, then 75 * this is an error. 76 * 77 * 78 * Implementation note: 79 * 80 * Currently, the parser and several checks in the code limit the number 81 * of UChars or bytes in a mapping to 82 * UCNV_EXT_MAX_UCHARS and UCNV_EXT_MAX_BYTES, respectively, 83 * which are output value limits in the data structure. 84 * 85 * For input, this is not strictly necessary - it is a hard limit only for the 86 * buffers in UConverter that are used to store partial matches. 87 * 88 * Input sequences could otherwise be arbitrarily long if partial matches 89 * need not be stored (i.e., if a sequence does not span several buffers with too 90 * many units before the last buffer), although then results would differ 91 * depending on whether partial matches exceed the limits or not, 92 * which depends on the pattern of buffer sizes. 93 * 94 * 95 * Data structure: 96 * 97 * int32_t indexes[>=32]; 98 * 99 * Array of indexes and lengths etc. The length of the array is at least 32. 100 * The actual length is stored in indexes[0] to be forward compatible. 101 * 102 * Each index to another array is the number of bytes from indexes[]. 103 * Each length of an array is the number of array base units in that array. 104 * 105 * Some of the structures may not be present, in which case their indexes 106 * and lengths are 0. 107 * 108 * Usage of indexes[i]: 109 * [0] length of indexes[] 110 * 111 * // to Unicode table 112 * [1] index of toUTable[] (array of uint32_t) 113 * [2] length of toUTable[] 114 * [3] index of toUUChars[] (array of UChar) 115 * [4] length of toUUChars[] 116 * 117 * // from Unicode table, not for the initial code point 118 * [5] index of fromUTableUChars[] (array of UChar) 119 * [6] index of fromUTableValues[] (array of uint32_t) 120 * [7] length of fromUTableUChars[] and fromUTableValues[] 121 * [8] index of fromUBytes[] (array of char) 122 * [9] length of fromUBytes[] 123 * 124 * // from Unicode trie for initial-code point lookup 125 * [10] index of fromUStage12[] (combined array of uint16_t for stages 1 & 2) 126 * [11] length of stage 1 portion of fromUStage12[] 127 * [12] length of fromUStage12[] 128 * [13] index of fromUStage3[] (array of uint16_t indexes into fromUStage3b[]) 129 * [14] length of fromUStage3[] 130 * [15] index of fromUStage3b[] (array of uint32_t like fromUTableValues[]) 131 * [16] length of fromUStage3b[] 132 * 133 * [17] Bit field containing numbers of bytes: 134 * 31..24 reserved, 0 135 * 23..16 maximum input bytes 136 * 15.. 8 maximum output bytes 137 * 7.. 0 maximum bytes per UChar 138 * 139 * [18] Bit field containing numbers of UChars: 140 * 31..24 reserved, 0 141 * 23..16 maximum input UChars 142 * 15.. 8 maximum output UChars 143 * 7.. 0 maximum UChars per byte 144 * 145 * [19] Bit field containing flags: 146 * (extension table unicodeMask) 147 * 1 UCNV_HAS_SURROGATES flag for the extension table 148 * 0 UCNV_HAS_SUPPLEMENTARY flag for the extension table 149 * 150 * [20]..[30] reserved, 0 151 * [31] number of bytes for the entire extension structure 152 * [>31] reserved; there are indexes[0] indexes 153 * 154 * 155 * uint32_t toUTable[]; 156 * 157 * Array of byte/value pairs for lookups for toUnicode conversion. 158 * The array is partitioned into sections like collation contraction tables. 159 * Each section contains one word with the number of following words and 160 * a default value for when the lookup in this section yields no match. 161 * 162 * A section is sorted in ascending order of input bytes, 163 * allowing for fast linear or binary searches. 164 * The builder may store entries for a contiguous range of byte values 165 * (compare difference between the first and last one with count), 166 * which then allows for direct array access. 167 * The builder should always do this for the initial table section. 168 * 169 * Entries may have 0 values, see below. 170 * No two entries in a section have the same byte values. 171 * 172 * Each uint32_t contains an input byte value in bits 31..24 and the 173 * corresponding lookup value in bits 23..0. 174 * Interpret the value as follows: 175 * if(value==0) { 176 * no match, see below 177 * } else if(value<0x1f0000) { 178 * partial match - use value as index to the next toUTable section 179 * and match the next unit; (value indexes toUTable[value]) 180 * } else { 181 * if(bit 23 set) { 182 * roundtrip; 183 * } else { 184 * fallback; 185 * } 186 * unset value bit 23; 187 * if(value<=0x2fffff) { 188 * (value-0x1f0000) is a code point; (BMP: value<=0x1fffff) 189 * } else { 190 * bits 17..0 (value&0x3ffff) is an index to 191 * the result UChars in toUUChars[]; (0 indexes toUUChars[0]) 192 * length of the result=((value>>18)-12); (length=0..19) 193 * } 194 * } 195 * 196 * The first word in a section contains the number of following words in the 197 * input byte position (bits 31..24, number=1..0xff). 198 * The value of the initial word is used when the current byte is not found 199 * in this section. 200 * If the value is not 0, then it represents a result as above. 201 * If the value is 0, then the search has to return a shorter match with an 202 * earlier default value as the result, or result in "unmappable" even for the 203 * initial bytes. 204 * If the value is 0 for the initial toUTable entry, then the initial byte 205 * does not start any mapping input. 206 * 207 * 208 * UChar toUUChars[]; 209 * 210 * Contains toUnicode mapping results, stored as sequences of UChars. 211 * Indexes and lengths stored in the toUTable[]. 212 * 213 * 214 * UChar fromUTableUChars[]; 215 * uint32_t fromUTableValues[]; 216 * 217 * The fromUTable is split into two arrays, but works otherwise much like 218 * the toUTable. The array is partitioned into sections like collation 219 * contraction tables and toUTable. 220 * A row in the table consists of same-index entries in fromUTableUChars[] 221 * and fromUTableValues[]. 222 * 223 * Interpret a value as follows: 224 * if(value==0) { 225 * no match, see below 226 * } else if(value<=0xffffff) { (bits 31..24 are 0) 227 * partial match - use value as index to the next fromUTable section 228 * and match the next unit; (value indexes fromUTable[value]) 229 * } else { 230 * if(value==0x80000001) { 231 * return no mapping, but request for <subchar1>; 232 * } 233 * if(bit 31 set) { 234 * roundtrip (|0); 235 * } else if(bit 30 set) { 236 * "good one-way" mapping (|4); -- new in ICU4C 51, _MBCSHeader.version 5.4/4.4 237 * } else { 238 * normal fallback (|1); 239 * } 240 * // bit 29 reserved, 0 241 * length=(value>>24)&0x1f; (bits 28..24) 242 * if(length==1..3) { 243 * bits 23..0 contain 1..3 bytes, padded with 00s on the left; 244 * } else { 245 * bits 23..0 (value&0xffffff) is an index to 246 * the result bytes in fromUBytes[]; (0 indexes fromUBytes[0]) 247 * } 248 * } 249 * 250 * The first pair in a section contains the number of following pairs in the 251 * UChar position (16 bits, number=1..0xffff). 252 * The value of the initial pair is used when the current UChar is not found 253 * in this section. 254 * If the value is not 0, then it represents a result as above. 255 * If the value is 0, then the search has to return a shorter match with an 256 * earlier default value as the result, or result in "unmappable" even for the 257 * initial UChars. 258 * 259 * If the from Unicode trie is present, then the from Unicode search tables 260 * are not used for initial code points. 261 * In this case, the first entries (index 0) in the tables are not used 262 * (reserved, set to 0) because a value of 0 is used in trie results 263 * to indicate no mapping. 264 * 265 * 266 * uint16_t fromUStage12[]; 267 * 268 * Stages 1 & 2 of a trie that maps an initial code point. 269 * Indexes in stage 1 are all offset by the length of stage 1 so that the 270 * same array pointer can be used for both stages. 271 * If (c>>10)>=(length of stage 1) then c does not start any mapping. 272 * Same bit distribution as for regular conversion tries. 273 * 274 * 275 * uint16_t fromUStage3[]; 276 * uint32_t fromUStage3b[]; 277 * 278 * Stage 3 of the trie. The first array simply contains indexes to the second, 279 * which contains words in the same format as fromUTableValues[]. 280 * Use a stage 3 granularity of 4, which allows for 256k stage 3 entries, 281 * and 16-bit entries in stage 3 allow for 64k stage 3b entries. 282 * The stage 3 granularity means that the stage 2 entry needs to be left-shifted. 283 * 284 * Two arrays are used because it is expected that more than half of the stage 3 285 * entries will be zero. The 16-bit index stage 3 array saves space even 286 * considering storing a total of 6 bytes per non-zero entry in both arrays 287 * together. 288 * Using a stage 3 granularity of >1 diminishes the compactability in that stage 289 * but provides a larger effective addressing space in stage 2. 290 * All but the final result stage use 16-bit entries to save space. 291 * 292 * fromUStage3b[] contains a zero for "no mapping" at its index 0, 293 * and may contain UCNV_EXT_FROM_U_SUBCHAR1 at index 1 for "<subchar1> SUB mapping" 294 * (i.e., "no mapping" with preference for <subchar1> rather than <subchar>), 295 * and all other items are unique non-zero results. 296 * 297 * The default value of a fromUTableValues[] section that is referenced 298 * _directly_ from a fromUStage3b[] item may also be UCNV_EXT_FROM_U_SUBCHAR1, 299 * but this value must not occur anywhere else in fromUTableValues[] 300 * because "no mapping" is always a property of a single code point, 301 * never of multiple. 302 * 303 * 304 * char fromUBytes[]; 305 * 306 * Contains fromUnicode mapping results, stored as sequences of chars. 307 * Indexes and lengths stored in the fromUTableValues[]. 308 */ 309 enum { 310 UCNV_EXT_INDEXES_LENGTH, /* 0 */ 311 312 UCNV_EXT_TO_U_INDEX, /* 1 */ 313 UCNV_EXT_TO_U_LENGTH, 314 UCNV_EXT_TO_U_UCHARS_INDEX, 315 UCNV_EXT_TO_U_UCHARS_LENGTH, 316 317 UCNV_EXT_FROM_U_UCHARS_INDEX, /* 5 */ 318 UCNV_EXT_FROM_U_VALUES_INDEX, 319 UCNV_EXT_FROM_U_LENGTH, 320 UCNV_EXT_FROM_U_BYTES_INDEX, 321 UCNV_EXT_FROM_U_BYTES_LENGTH, 322 323 UCNV_EXT_FROM_U_STAGE_12_INDEX, /* 10 */ 324 UCNV_EXT_FROM_U_STAGE_1_LENGTH, 325 UCNV_EXT_FROM_U_STAGE_12_LENGTH, 326 UCNV_EXT_FROM_U_STAGE_3_INDEX, 327 UCNV_EXT_FROM_U_STAGE_3_LENGTH, 328 UCNV_EXT_FROM_U_STAGE_3B_INDEX, 329 UCNV_EXT_FROM_U_STAGE_3B_LENGTH, 330 331 UCNV_EXT_COUNT_BYTES, /* 17 */ 332 UCNV_EXT_COUNT_UCHARS, 333 UCNV_EXT_FLAGS, 334 335 UCNV_EXT_RESERVED_INDEX, /* 20, moves with additional indexes */ 336 337 UCNV_EXT_SIZE=31, 338 UCNV_EXT_INDEXES_MIN_LENGTH=32 339 }; 340 341 /* get the pointer to an extension array from indexes[index] */ 342 #define UCNV_EXT_ARRAY(indexes, index, itemType) \ 343 ((const itemType *)((const char *)(indexes)+(indexes)[index])) 344 345 #define UCNV_GET_MAX_BYTES_PER_UCHAR(indexes) \ 346 ((indexes)[UCNV_EXT_COUNT_BYTES]&0xff) 347 348 /* internal API ------------------------------------------------------------- */ 349 350 U_CFUNC UBool 351 ucnv_extInitialMatchToU(UConverter *cnv, const int32_t *cx, 352 int32_t firstLength, 353 const char **src, const char *srcLimit, 354 UChar **target, const UChar *targetLimit, 355 int32_t **offsets, int32_t srcIndex, 356 UBool flush, 357 UErrorCode *pErrorCode); 358 359 U_CFUNC UChar32 360 ucnv_extSimpleMatchToU(const int32_t *cx, 361 const char *source, int32_t length, 362 UBool useFallback); 363 364 U_CFUNC void 365 ucnv_extContinueMatchToU(UConverter *cnv, 366 UConverterToUnicodeArgs *pArgs, int32_t srcIndex, 367 UErrorCode *pErrorCode); 368 369 370 U_CFUNC UBool 371 ucnv_extInitialMatchFromU(UConverter *cnv, const int32_t *cx, 372 UChar32 cp, 373 const UChar **src, const UChar *srcLimit, 374 char **target, const char *targetLimit, 375 int32_t **offsets, int32_t srcIndex, 376 UBool flush, 377 UErrorCode *pErrorCode); 378 379 U_CFUNC int32_t 380 ucnv_extSimpleMatchFromU(const int32_t *cx, 381 UChar32 cp, uint32_t *pValue, 382 UBool useFallback); 383 384 U_CFUNC void 385 ucnv_extContinueMatchFromU(UConverter *cnv, 386 UConverterFromUnicodeArgs *pArgs, int32_t srcIndex, 387 UErrorCode *pErrorCode); 388 389 /* 390 * Add code points and strings to the set according to the extension mappings. 391 * Limitation on the UConverterSetFilter: 392 * The filters currently assume that they are used with 1:1 mappings. 393 * They only apply to single input code points, and then they pass through 394 * only mappings with single-charset-code results. 395 * For example, the Shift-JIS filter only works for 2-byte results and tests 396 * that those 2 bytes are in the JIS X 0208 range of Shift-JIS. 397 */ 398 U_CFUNC void 399 ucnv_extGetUnicodeSet(const UConverterSharedData *sharedData, 400 const USetAdder *sa, 401 UConverterUnicodeSet which, 402 UConverterSetFilter filter, 403 UErrorCode *pErrorCode); 404 405 /* toUnicode helpers -------------------------------------------------------- */ 406 407 #define UCNV_EXT_TO_U_BYTE_SHIFT 24 408 #define UCNV_EXT_TO_U_VALUE_MASK 0xffffff 409 #define UCNV_EXT_TO_U_MIN_CODE_POINT 0x1f0000 410 #define UCNV_EXT_TO_U_MAX_CODE_POINT 0x2fffff 411 #define UCNV_EXT_TO_U_ROUNDTRIP_FLAG ((uint32_t)1<<23) 412 #define UCNV_EXT_TO_U_INDEX_MASK 0x3ffff 413 #define UCNV_EXT_TO_U_LENGTH_SHIFT 18 414 #define UCNV_EXT_TO_U_LENGTH_OFFSET 12 415 416 /* maximum number of indexed UChars */ 417 #define UCNV_EXT_MAX_UCHARS 19 418 419 #define UCNV_EXT_TO_U_MAKE_WORD(byte, value) (((uint32_t)(byte)<<UCNV_EXT_TO_U_BYTE_SHIFT)|(value)) 420 421 #define UCNV_EXT_TO_U_GET_BYTE(word) ((word)>>UCNV_EXT_TO_U_BYTE_SHIFT) 422 #define UCNV_EXT_TO_U_GET_VALUE(word) ((word)&UCNV_EXT_TO_U_VALUE_MASK) 423 424 #define UCNV_EXT_TO_U_IS_PARTIAL(value) ((value)<UCNV_EXT_TO_U_MIN_CODE_POINT) 425 #define UCNV_EXT_TO_U_GET_PARTIAL_INDEX(value) (value) 426 427 #define UCNV_EXT_TO_U_IS_ROUNDTRIP(value) (((value)&UCNV_EXT_TO_U_ROUNDTRIP_FLAG)!=0) 428 #define UCNV_EXT_TO_U_MASK_ROUNDTRIP(value) ((value)&~UCNV_EXT_TO_U_ROUNDTRIP_FLAG) 429 430 /* use after masking off the roundtrip flag */ 431 #define UCNV_EXT_TO_U_IS_CODE_POINT(value) ((value)<=UCNV_EXT_TO_U_MAX_CODE_POINT) 432 #define UCNV_EXT_TO_U_GET_CODE_POINT(value) ((value)-UCNV_EXT_TO_U_MIN_CODE_POINT) 433 434 #define UCNV_EXT_TO_U_GET_INDEX(value) ((value)&UCNV_EXT_TO_U_INDEX_MASK) 435 #define UCNV_EXT_TO_U_GET_LENGTH(value) (((value)>>UCNV_EXT_TO_U_LENGTH_SHIFT)-UCNV_EXT_TO_U_LENGTH_OFFSET) 436 437 /* fromUnicode helpers ------------------------------------------------------ */ 438 439 /* most trie constants are shared with ucnvmbcs.h */ 440 441 /* see similar utrie.h UTRIE_INDEX_SHIFT and UTRIE_DATA_GRANULARITY */ 442 #define UCNV_EXT_STAGE_2_LEFT_SHIFT 2 443 #define UCNV_EXT_STAGE_3_GRANULARITY 4 444 445 /* trie access, returns the stage 3 value=index to stage 3b; s1Index=c>>10 */ 446 #define UCNV_EXT_FROM_U(stage12, stage3, s1Index, c) \ 447 (stage3)[ ((int32_t)(stage12)[ (stage12)[s1Index] +(((c)>>4)&0x3f) ]<<UCNV_EXT_STAGE_2_LEFT_SHIFT) +((c)&0xf) ] 448 449 #define UCNV_EXT_FROM_U_LENGTH_SHIFT 24 450 #define UCNV_EXT_FROM_U_ROUNDTRIP_FLAG ((uint32_t)1<<31) 451 #define UCNV_EXT_FROM_U_GOOD_ONE_WAY_FLAG 0x40000000 452 #define UCNV_EXT_FROM_U_STATUS_MASK 0xc0000000 453 #define UCNV_EXT_FROM_U_RESERVED_MASK 0x20000000 454 #define UCNV_EXT_FROM_U_DATA_MASK 0xffffff 455 456 /* special value for "no mapping" to <subchar1> (impossible roundtrip to 0 bytes, value 01) */ 457 #define UCNV_EXT_FROM_U_SUBCHAR1 0x80000001 458 459 /* at most 3 bytes in the lower part of the value */ 460 #define UCNV_EXT_FROM_U_MAX_DIRECT_LENGTH 3 461 462 /* maximum number of indexed bytes */ 463 #define UCNV_EXT_MAX_BYTES 0x1f 464 465 #define UCNV_EXT_FROM_U_IS_PARTIAL(value) (((value)>>UCNV_EXT_FROM_U_LENGTH_SHIFT)==0) 466 #define UCNV_EXT_FROM_U_GET_PARTIAL_INDEX(value) (value) 467 468 #define UCNV_EXT_FROM_U_IS_ROUNDTRIP(value) (((value)&UCNV_EXT_FROM_U_ROUNDTRIP_FLAG)!=0) 469 #define UCNV_EXT_FROM_U_MASK_ROUNDTRIP(value) ((value)&~UCNV_EXT_FROM_U_ROUNDTRIP_FLAG) 470 471 /* get length; masks away all other bits */ 472 #define UCNV_EXT_FROM_U_GET_LENGTH(value) (int32_t)(((value)>>UCNV_EXT_FROM_U_LENGTH_SHIFT)&UCNV_EXT_MAX_BYTES) 473 474 /* get bytes or bytes index */ 475 #define UCNV_EXT_FROM_U_GET_DATA(value) ((value)&UCNV_EXT_FROM_U_DATA_MASK) 476 477 #endif 478 479 #endif 480