| 1 | /* |
| 2 | * This file is part of DisOrder |
| 3 | * Copyright (C) 2007 Richard Kettlewell |
| 4 | * |
| 5 | * This program is free software; you can redistribute it and/or modify |
| 6 | * it under the terms of the GNU General Public License as published by |
| 7 | * the Free Software Foundation; either version 2 of the License, or |
| 8 | * (at your option) any later version. |
| 9 | * |
| 10 | * This program is distributed in the hope that it will be useful, but |
| 11 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | * General Public License for more details. |
| 14 | * |
| 15 | * You should have received a copy of the GNU General Public License |
| 16 | * along with this program; if not, write to the Free Software |
| 17 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 |
| 18 | * USA |
| 19 | */ |
| 20 | /** @file lib/unicode.c |
| 21 | * @brief Unicode support functions |
| 22 | * |
| 23 | * Here by UTF-8 and UTF-8 we mean the encoding forms of those names (not the |
| 24 | * encoding schemes). The primary encoding form is UTF-32 but convenience |
| 25 | * wrappers using UTF-8 are provided for a number of functions. |
| 26 | * |
| 27 | * The idea is that all the strings that hit the database will be in a |
| 28 | * particular normalization form, and for the search and tags database |
| 29 | * in case-folded form, so they can be naively compared within the |
| 30 | * database code. |
| 31 | * |
| 32 | * As the code stands this guarantee is not well met! |
| 33 | */ |
| 34 | |
| 35 | #include <config.h> |
| 36 | #include "types.h" |
| 37 | |
| 38 | #include <string.h> |
| 39 | #include <stdio.h> /* TODO */ |
| 40 | |
| 41 | #include "mem.h" |
| 42 | #include "vector.h" |
| 43 | #include "unicode.h" |
| 44 | #include "unidata.h" |
| 45 | |
| 46 | /** @defgroup utftransform Functions that transform between different Unicode encoding forms */ |
| 47 | /*@{*/ |
| 48 | |
| 49 | /** @brief Convert UTF-32 to UTF-8 |
| 50 | * @param s Source string |
| 51 | * @param ns Length of source string in code points |
| 52 | * @param ndp Where to store length of destination string (or NULL) |
| 53 | * @return Newly allocated destination string or NULL on error |
| 54 | * |
| 55 | * If the UTF-32 is not valid then NULL is returned. A UTF-32 code point is |
| 56 | * invalid if: |
| 57 | * - it codes for a UTF-16 surrogate |
| 58 | * - it codes for a value outside the unicode code space |
| 59 | * |
| 60 | * The return value is always 0-terminated. The value returned via @p *ndp |
| 61 | * does not include the terminator. |
| 62 | */ |
| 63 | char *utf32_to_utf8(const uint32_t *s, size_t ns, size_t *ndp) { |
| 64 | struct dynstr d; |
| 65 | uint32_t c; |
| 66 | |
| 67 | dynstr_init(&d); |
| 68 | while(ns > 0) { |
| 69 | c = *s++; |
| 70 | if(c < 0x80) |
| 71 | dynstr_append(&d, c); |
| 72 | else if(c < 0x0800) { |
| 73 | dynstr_append(&d, 0xC0 | (c >> 6)); |
| 74 | dynstr_append(&d, 0x80 | (c & 0x3F)); |
| 75 | } else if(c < 0x10000) { |
| 76 | if(c >= 0xD800 && c <= 0xDFFF) |
| 77 | goto error; |
| 78 | dynstr_append(&d, 0xE0 | (c >> 12)); |
| 79 | dynstr_append(&d, 0x80 | ((c >> 6) & 0x3F)); |
| 80 | dynstr_append(&d, 0x80 | (c & 0x3F)); |
| 81 | } else if(c < 0x110000) { |
| 82 | dynstr_append(&d, 0xF0 | (c >> 18)); |
| 83 | dynstr_append(&d, 0x80 | ((c >> 12) & 0x3F)); |
| 84 | dynstr_append(&d, 0x80 | ((c >> 6) & 0x3F)); |
| 85 | dynstr_append(&d, 0x80 | (c & 0x3F)); |
| 86 | } else |
| 87 | goto error; |
| 88 | --ns; |
| 89 | } |
| 90 | dynstr_terminate(&d); |
| 91 | if(ndp) |
| 92 | *ndp = d.nvec; |
| 93 | return d.vec; |
| 94 | error: |
| 95 | xfree(d.vec); |
| 96 | return 0; |
| 97 | } |
| 98 | |
| 99 | /** @brief Convert UTF-8 to UTF-32 |
| 100 | * @param s Source string |
| 101 | * @param ns Length of source string in code points |
| 102 | * @param ndp Where to store length of destination string (or NULL) |
| 103 | * @return Newly allocated destination string or NULL |
| 104 | * |
| 105 | * The return value is always 0-terminated. The value returned via @p *ndp |
| 106 | * does not include the terminator. |
| 107 | * |
| 108 | * If the UTF-8 is not valid then NULL is returned. A UTF-8 sequence |
| 109 | * for a code point is invalid if: |
| 110 | * - it is not the shortest possible sequence for the code point |
| 111 | * - it codes for a UTF-16 surrogate |
| 112 | * - it codes for a value outside the unicode code space |
| 113 | */ |
| 114 | uint32_t *utf8_to_utf32(const char *s, size_t ns, size_t *ndp) { |
| 115 | struct dynstr_ucs4 d; |
| 116 | uint32_t c32, c; |
| 117 | const uint8_t *ss = (const uint8_t *)s; |
| 118 | |
| 119 | dynstr_ucs4_init(&d); |
| 120 | while(ns > 0) { |
| 121 | c = *ss++; |
| 122 | --ns; |
| 123 | /* Acceptable UTF-8 is that which codes for Unicode Scalar Values |
| 124 | * (Unicode 5.0.0 s3.9 D76) |
| 125 | * |
| 126 | * 0xxxxxxx |
| 127 | * 7 data bits gives 0x00 - 0x7F and all are acceptable |
| 128 | * |
| 129 | * 110xxxxx 10xxxxxx |
| 130 | * 11 data bits gives 0x0000 - 0x07FF but only 0x0080 - 0x07FF acceptable |
| 131 | * |
| 132 | * 1110xxxx 10xxxxxx 10xxxxxx |
| 133 | * 16 data bits gives 0x0000 - 0xFFFF but only 0x0800 - 0xFFFF acceptable |
| 134 | * (and UTF-16 surrogates are not acceptable) |
| 135 | * |
| 136 | * 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx |
| 137 | * 21 data bits gives 0x00000000 - 0x001FFFFF |
| 138 | * but only 0x00010000 - 0x0010FFFF are acceptable |
| 139 | * |
| 140 | * It is NOT always the case that the data bits in the first byte are |
| 141 | * always non-0 for the acceptable values, so we do a separate check after |
| 142 | * decoding. |
| 143 | */ |
| 144 | if(c < 0x80) |
| 145 | c32 = c; |
| 146 | else if(c <= 0xDF) { |
| 147 | if(ns < 1) goto error; |
| 148 | c32 = c & 0x1F; |
| 149 | c = *ss++; |
| 150 | if((c & 0xC0) != 0x80) goto error; |
| 151 | c32 = (c32 << 6) | (c & 0x3F); |
| 152 | if(c32 < 0x80) goto error; |
| 153 | } else if(c <= 0xEF) { |
| 154 | if(ns < 2) goto error; |
| 155 | c32 = c & 0x0F; |
| 156 | c = *ss++; |
| 157 | if((c & 0xC0) != 0x80) goto error; |
| 158 | c32 = (c32 << 6) | (c & 0x3F); |
| 159 | c = *ss++; |
| 160 | if((c & 0xC0) != 0x80) goto error; |
| 161 | c32 = (c32 << 6) | (c & 0x3F); |
| 162 | if(c32 < 0x0800 || (c32 >= 0xD800 && c32 <= 0xDFFF)) goto error; |
| 163 | } else if(c <= 0xF7) { |
| 164 | if(ns < 3) goto error; |
| 165 | c32 = c & 0x07; |
| 166 | c = *ss++; |
| 167 | if((c & 0xC0) != 0x80) goto error; |
| 168 | c32 = (c32 << 6) | (c & 0x3F); |
| 169 | c = *ss++; |
| 170 | if((c & 0xC0) != 0x80) goto error; |
| 171 | c32 = (c32 << 6) | (c & 0x3F); |
| 172 | c = *ss++; |
| 173 | if((c & 0xC0) != 0x80) goto error; |
| 174 | c32 = (c32 << 6) | (c & 0x3F); |
| 175 | if(c32 < 0x00010000 || c32 > 0x0010FFFF) goto error; |
| 176 | } else |
| 177 | goto error; |
| 178 | dynstr_ucs4_append(&d, c32); |
| 179 | } |
| 180 | dynstr_ucs4_terminate(&d); |
| 181 | if(ndp) |
| 182 | *ndp = d.nvec; |
| 183 | return d.vec; |
| 184 | error: |
| 185 | xfree(d.vec); |
| 186 | return 0; |
| 187 | } |
| 188 | |
| 189 | /*@}*/ |
| 190 | /** @defgroup utf32 Functions that operate on UTF-32 strings */ |
| 191 | /*@{*/ |
| 192 | |
| 193 | /** @brief Return the length of a 0-terminated UTF-32 string |
| 194 | * @param s Pointer to 0-terminated string |
| 195 | * @return Length of string in code points (excluding terminator) |
| 196 | * |
| 197 | * Unlike the conversion functions no validity checking is done on the string. |
| 198 | */ |
| 199 | size_t utf32_len(const uint32_t *s) { |
| 200 | const uint32_t *t = s; |
| 201 | |
| 202 | while(*t) |
| 203 | ++t; |
| 204 | return (size_t)(t - s); |
| 205 | } |
| 206 | |
| 207 | /** @brief Return the @ref unidata structure for code point @p c |
| 208 | * |
| 209 | * @p c can be any 32-bit value, a sensible value will be returned regardless. |
| 210 | */ |
| 211 | static const struct unidata *utf32__unidata(uint32_t c) { |
| 212 | if(c < UNICODE_NCHARS) |
| 213 | return &unidata[c / UNICODE_MODULUS][c % UNICODE_MODULUS]; |
| 214 | else if((c >= 0xF0000 && c <= 0xFFFFD) |
| 215 | || (c >= 0x100000 && c <= 0x10FFFD)) |
| 216 | return utf32__unidata(0xE000); /* Co */ |
| 217 | else |
| 218 | return utf32__unidata(0xFFFF); /* Cn */ |
| 219 | } |
| 220 | |
| 221 | /** @brief Return the combining class of @p c |
| 222 | * @param c Code point |
| 223 | * @return Combining class of @p c |
| 224 | */ |
| 225 | static inline int utf32__combining_class(uint32_t c) { |
| 226 | return utf32__unidata(c)->ccc; |
| 227 | } |
| 228 | |
| 229 | /** @brief Stably sort [s,s+ns) into descending order of combining class |
| 230 | * @param s Start of array |
| 231 | * @param ns Number of elements, must be at least 1 |
| 232 | * @param buffer Buffer of at least @p ns elements |
| 233 | */ |
| 234 | static void utf32__sort_ccc(uint32_t *s, size_t ns, uint32_t *buffer) { |
| 235 | uint32_t *a, *b, *bp; |
| 236 | size_t na, nb; |
| 237 | |
| 238 | switch(ns) { |
| 239 | case 1: /* 1-element array is always sorted */ |
| 240 | return; |
| 241 | case 2: /* 2-element arrays are trivial to sort */ |
| 242 | if(utf32__combining_class(s[0]) > utf32__combining_class(s[1])) { |
| 243 | uint32_t tmp = s[0]; |
| 244 | s[0] = s[1]; |
| 245 | s[1] = tmp; |
| 246 | } |
| 247 | return; |
| 248 | default: |
| 249 | /* Partition the array */ |
| 250 | na = ns / 2; |
| 251 | nb = ns - na; |
| 252 | a = s; |
| 253 | b = s + na; |
| 254 | /* Sort the two halves of the array */ |
| 255 | utf32__sort_ccc(a, na, buffer); |
| 256 | utf32__sort_ccc(b, nb, buffer); |
| 257 | /* Merge them back into one, via the buffer */ |
| 258 | bp = buffer; |
| 259 | while(na > 0 && nb > 0) { |
| 260 | /* We want descending order of combining class (hence <) |
| 261 | * and we want stability within combining classes (hence <=) |
| 262 | */ |
| 263 | if(utf32__combining_class(*a) <= utf32__combining_class(*b)) { |
| 264 | *bp++ = *a++; |
| 265 | --na; |
| 266 | } else { |
| 267 | *bp++ = *b++; |
| 268 | --nb; |
| 269 | } |
| 270 | } |
| 271 | while(na > 0) { |
| 272 | *bp++ = *a++; |
| 273 | --na; |
| 274 | } |
| 275 | while(nb > 0) { |
| 276 | *bp++ = *b++; |
| 277 | --nb; |
| 278 | } |
| 279 | memcpy(s, buffer, ns * sizeof(uint32_t)); |
| 280 | return; |
| 281 | } |
| 282 | } |
| 283 | |
| 284 | /** @brief Put combining characters into canonical order |
| 285 | * @param s Pointer to UTF-32 string |
| 286 | * @param ns Length of @p s |
| 287 | * @return 0 on success, -1 on error |
| 288 | * |
| 289 | * @p s is modified in-place. See Unicode 5.0 s3.11 for details of the |
| 290 | * ordering. |
| 291 | * |
| 292 | * Currently we only support a maximum of 1024 combining characters after each |
| 293 | * base character. If this limit is exceeded then -1 is returned. |
| 294 | */ |
| 295 | static int utf32__canonical_ordering(uint32_t *s, size_t ns) { |
| 296 | size_t nc; |
| 297 | uint32_t buffer[1024]; |
| 298 | |
| 299 | /* The ordering amounts to a stable sort of each contiguous group of |
| 300 | * characters with non-0 combining class. */ |
| 301 | while(ns > 0) { |
| 302 | /* Skip non-combining characters */ |
| 303 | if(utf32__combining_class(*s) == 0) { |
| 304 | ++s; |
| 305 | --ns; |
| 306 | continue; |
| 307 | } |
| 308 | /* We must now have at least one combining character; see how many |
| 309 | * there are */ |
| 310 | for(nc = 1; nc < ns && utf32__combining_class(s[nc]) != 0; ++nc) |
| 311 | ; |
| 312 | if(nc > 1024) |
| 313 | return -1; |
| 314 | /* Sort the array */ |
| 315 | utf32__sort_ccc(s, nc, buffer); |
| 316 | s += nc; |
| 317 | ns -= nc; |
| 318 | } |
| 319 | return 0; |
| 320 | } |
| 321 | |
| 322 | /* Magic numbers from UAX #15 s16 */ |
| 323 | #define SBase 0xAC00 |
| 324 | #define LBase 0x1100 |
| 325 | #define VBase 0x1161 |
| 326 | #define TBase 0x11A7 |
| 327 | #define LCount 19 |
| 328 | #define VCount 21 |
| 329 | #define TCount 28 |
| 330 | #define NCount (VCount * TCount) |
| 331 | #define SCount (LCount * NCount) |
| 332 | |
| 333 | /** @brief Guts of the decomposition lookup functions */ |
| 334 | #define utf32__decompose_one_generic(WHICH) do { \ |
| 335 | const uint32_t *dc = utf32__unidata(c)->WHICH; \ |
| 336 | if(dc) { \ |
| 337 | /* Found a canonical decomposition in the table */ \ |
| 338 | while(*dc) \ |
| 339 | utf32__decompose_one_##WHICH(d, *dc++); \ |
| 340 | } else if(c >= SBase && c < SBase + SCount) { \ |
| 341 | /* Mechanically decomposable Hangul syllable (UAX #15 s16) */ \ |
| 342 | const uint32_t SIndex = c - SBase; \ |
| 343 | const uint32_t L = LBase + SIndex / NCount; \ |
| 344 | const uint32_t V = VBase + (SIndex % NCount) / TCount; \ |
| 345 | const uint32_t T = TBase + SIndex % TCount; \ |
| 346 | dynstr_ucs4_append(d, L); \ |
| 347 | dynstr_ucs4_append(d, V); \ |
| 348 | if(T != TBase) \ |
| 349 | dynstr_ucs4_append(d, T); \ |
| 350 | } else \ |
| 351 | /* Equal to own canonical decomposition */ \ |
| 352 | dynstr_ucs4_append(d, c); \ |
| 353 | } while(0) |
| 354 | |
| 355 | /** @brief Recursively compute the canonical decomposition of @p c |
| 356 | * @param d Dynamic string to store decomposition in |
| 357 | * @param c Code point to decompose (must be a valid!) |
| 358 | * @return 0 on success, -1 on error |
| 359 | */ |
| 360 | static void utf32__decompose_one_canon(struct dynstr_ucs4 *d, uint32_t c) { |
| 361 | utf32__decompose_one_generic(canon); |
| 362 | } |
| 363 | |
| 364 | /** @brief Recursively compute the compatibility decomposition of @p c |
| 365 | * @param d Dynamic string to store decomposition in |
| 366 | * @param c Code point to decompose (must be a valid!) |
| 367 | * @return 0 on success, -1 on error |
| 368 | */ |
| 369 | static void utf32__decompose_one_compat(struct dynstr_ucs4 *d, uint32_t c) { |
| 370 | utf32__decompose_one_generic(compat); |
| 371 | } |
| 372 | |
| 373 | /** @brief Guts of the decomposition functions */ |
| 374 | #define utf32__decompose_generic(WHICH) do { \ |
| 375 | struct dynstr_ucs4 d; \ |
| 376 | uint32_t c; \ |
| 377 | \ |
| 378 | dynstr_ucs4_init(&d); \ |
| 379 | while(ns) { \ |
| 380 | c = *s++; \ |
| 381 | if((c >= 0xD800 && c <= 0xDFFF) || c > 0x10FFFF) \ |
| 382 | goto error; \ |
| 383 | utf32__decompose_one_##WHICH(&d, c); \ |
| 384 | --ns; \ |
| 385 | } \ |
| 386 | if(utf32__canonical_ordering(d.vec, d.nvec)) \ |
| 387 | goto error; \ |
| 388 | dynstr_ucs4_terminate(&d); \ |
| 389 | if(ndp) \ |
| 390 | *ndp = d.nvec; \ |
| 391 | return d.vec; \ |
| 392 | error: \ |
| 393 | xfree(d.vec); \ |
| 394 | return 0; \ |
| 395 | } while(0) |
| 396 | |
| 397 | /** @brief Canonically decompose @p [s,s+ns) |
| 398 | * @param s Pointer to string |
| 399 | * @param ns Length of string |
| 400 | * @param ndp Where to store length of result |
| 401 | * @return Pointer to result string, or NULL |
| 402 | * |
| 403 | * Computes the canonical decomposition of a string and stably sorts combining |
| 404 | * characters into canonical order. The result is in Normalization Form D and |
| 405 | * (at the time of writing!) passes the NFD tests defined in Unicode 5.0's |
| 406 | * NormalizationTest.txt. |
| 407 | * |
| 408 | * Returns NULL if the string is not valid for either of the following reasons: |
| 409 | * - it codes for a UTF-16 surrogate |
| 410 | * - it codes for a value outside the unicode code space |
| 411 | */ |
| 412 | uint32_t *utf32_decompose_canon(const uint32_t *s, size_t ns, size_t *ndp) { |
| 413 | utf32__decompose_generic(canon); |
| 414 | } |
| 415 | |
| 416 | /** @brief Compatibility decompose @p [s,s+ns) |
| 417 | * @param s Pointer to string |
| 418 | * @param ns Length of string |
| 419 | * @param ndp Where to store length of result |
| 420 | * @return Pointer to result string, or NULL |
| 421 | * |
| 422 | * Computes the compatibility decomposition of a string and stably sorts |
| 423 | * combining characters into canonical order. The result is in Normalization |
| 424 | * Form KD and (at the time of writing!) passes the NFKD tests defined in |
| 425 | * Unicode 5.0's NormalizationTest.txt. |
| 426 | * |
| 427 | * Returns NULL if the string is not valid for either of the following reasons: |
| 428 | * - it codes for a UTF-16 surrogate |
| 429 | * - it codes for a value outside the unicode code space |
| 430 | */ |
| 431 | uint32_t *utf32_decompose_compat(const uint32_t *s, size_t ns, size_t *ndp) { |
| 432 | utf32__decompose_generic(compat); |
| 433 | } |
| 434 | |
| 435 | /** @brief Single-character case-fold and decompose operation */ |
| 436 | #define utf32__casefold_one(WHICH) do { \ |
| 437 | const uint32_t *cf = utf32__unidata(c)->casefold; \ |
| 438 | if(cf) { \ |
| 439 | /* Found a case-fold mapping in the table */ \ |
| 440 | while(*cf) \ |
| 441 | utf32__decompose_one_##WHICH(&d, *cf++); \ |
| 442 | } else \ |
| 443 | utf32__decompose_one_##WHICH(&d, c); \ |
| 444 | } while(0) |
| 445 | |
| 446 | /** @brief Case-fold @p [s,s+ns) |
| 447 | * @param s Pointer to string |
| 448 | * @param ns Length of string |
| 449 | * @param ndp Where to store length of result |
| 450 | * @return Pointer to result string, or NULL |
| 451 | * |
| 452 | * Case-fold the string at @p s according to full default case-folding rules |
| 453 | * (s3.13) for caseless matching. The result will be in NFD. |
| 454 | * |
| 455 | * Returns NULL if the string is not valid for either of the following reasons: |
| 456 | * - it codes for a UTF-16 surrogate |
| 457 | * - it codes for a value outside the unicode code space |
| 458 | */ |
| 459 | uint32_t *utf32_casefold_canon(const uint32_t *s, size_t ns, size_t *ndp) { |
| 460 | struct dynstr_ucs4 d; |
| 461 | uint32_t c; |
| 462 | size_t n; |
| 463 | uint32_t *ss = 0; |
| 464 | |
| 465 | /* If the canonical decomposition of the string includes any combining |
| 466 | * character that case-folds to a non-combining character then we must |
| 467 | * normalize before we fold. In Unicode 5.0.0 this means 0345 COMBINING |
| 468 | * GREEK YPOGEGRAMMENI in its decomposition and the various characters that |
| 469 | * canonically decompose to it. */ |
| 470 | for(n = 0; n < ns; ++n) |
| 471 | if(utf32__unidata(s[n])->flags & unicode_normalize_before_casefold) |
| 472 | break; |
| 473 | if(n < ns) { |
| 474 | /* We need a preliminary decomposition */ |
| 475 | if(!(ss = utf32_decompose_canon(s, ns, &ns))) |
| 476 | return 0; |
| 477 | s = ss; |
| 478 | } |
| 479 | dynstr_ucs4_init(&d); |
| 480 | while(ns) { |
| 481 | c = *s++; |
| 482 | if((c >= 0xD800 && c <= 0xDFFF) || c > 0x10FFFF) |
| 483 | goto error; |
| 484 | utf32__casefold_one(canon); |
| 485 | --ns; |
| 486 | } |
| 487 | if(utf32__canonical_ordering(d.vec, d.nvec)) |
| 488 | goto error; |
| 489 | dynstr_ucs4_terminate(&d); |
| 490 | if(ndp) |
| 491 | *ndp = d.nvec; |
| 492 | return d.vec; |
| 493 | error: |
| 494 | xfree(d.vec); |
| 495 | xfree(ss); |
| 496 | return 0; |
| 497 | } |
| 498 | |
| 499 | /** @brief Compatibilit case-fold @p [s,s+ns) |
| 500 | * @param s Pointer to string |
| 501 | * @param ns Length of string |
| 502 | * @param ndp Where to store length of result |
| 503 | * @return Pointer to result string, or NULL |
| 504 | * |
| 505 | * Case-fold the string at @p s according to full default case-folding rules |
| 506 | * (s3.13) for compatibility caseless matching. The result will be in NFKD. |
| 507 | * |
| 508 | * Returns NULL if the string is not valid for either of the following reasons: |
| 509 | * - it codes for a UTF-16 surrogate |
| 510 | * - it codes for a value outside the unicode code space |
| 511 | */ |
| 512 | uint32_t *utf32_casefold_compat(const uint32_t *s, size_t ns, size_t *ndp) { |
| 513 | struct dynstr_ucs4 d; |
| 514 | uint32_t c; |
| 515 | size_t n; |
| 516 | uint32_t *ss = 0; |
| 517 | |
| 518 | for(n = 0; n < ns; ++n) |
| 519 | if(utf32__unidata(s[n])->flags & unicode_normalize_before_casefold) |
| 520 | break; |
| 521 | if(n < ns) { |
| 522 | /* We need a preliminary _canonical_ decomposition */ |
| 523 | if(!(ss = utf32_decompose_canon(s, ns, &ns))) |
| 524 | return 0; |
| 525 | s = ss; |
| 526 | } |
| 527 | /* This computes NFKD(toCaseFold(s)) */ |
| 528 | #define compat_casefold_middle() do { \ |
| 529 | dynstr_ucs4_init(&d); \ |
| 530 | while(ns) { \ |
| 531 | c = *s++; \ |
| 532 | if((c >= 0xD800 && c <= 0xDFFF) || c > 0x10FFFF) \ |
| 533 | goto error; \ |
| 534 | utf32__casefold_one(compat); \ |
| 535 | --ns; \ |
| 536 | } \ |
| 537 | if(utf32__canonical_ordering(d.vec, d.nvec)) \ |
| 538 | goto error; \ |
| 539 | } while(0) |
| 540 | /* Do the inner (NFKD o toCaseFold) */ |
| 541 | compat_casefold_middle(); |
| 542 | /* We can do away with the NFD'd copy of the input now */ |
| 543 | xfree(ss); |
| 544 | s = ss = d.vec; |
| 545 | ns = d.nvec; |
| 546 | /* Do the outer (NFKD o toCaseFold) */ |
| 547 | compat_casefold_middle(); |
| 548 | /* That's all */ |
| 549 | dynstr_ucs4_terminate(&d); |
| 550 | if(ndp) |
| 551 | *ndp = d.nvec; |
| 552 | return d.vec; |
| 553 | error: |
| 554 | xfree(d.vec); |
| 555 | xfree(ss); |
| 556 | return 0; |
| 557 | } |
| 558 | |
| 559 | /** @brief Order a pair of UTF-32 strings |
| 560 | * @param a First 0-terminated string |
| 561 | * @param b Second 0-terminated string |
| 562 | * @return -1, 0 or 1 for a less than, equal to or greater than b |
| 563 | * |
| 564 | * "Comparable to strcmp() at its best." |
| 565 | */ |
| 566 | int utf32_cmp(const uint32_t *a, const uint32_t *b) { |
| 567 | while(*a && *b && *a == *b) { |
| 568 | ++a; |
| 569 | ++b; |
| 570 | } |
| 571 | return *a < *b ? -1 : (*a > *b ? 1 : 0); |
| 572 | } |
| 573 | |
| 574 | /** @brief Return the General_Category value for @p c |
| 575 | * @param Code point |
| 576 | * @return General_Category property value |
| 577 | */ |
| 578 | static inline enum unicode_General_Category utf32__general_category(uint32_t c) { |
| 579 | return utf32__unidata(c)->general_category; |
| 580 | } |
| 581 | |
| 582 | /** @brief Check Grapheme_Cluster_Break property |
| 583 | * @param c Code point |
| 584 | * @return 0 if it is as described, 1 otherwise |
| 585 | */ |
| 586 | static int utf32__is_control_or_cr_or_lf(uint32_t c) { |
| 587 | switch(utf32__general_category(c)) { |
| 588 | default: |
| 589 | return 0; |
| 590 | case unicode_General_Category_Zl: |
| 591 | case unicode_General_Category_Zp: |
| 592 | case unicode_General_Category_Cc: |
| 593 | return 1; |
| 594 | case unicode_General_Category_Cf: |
| 595 | if(c == 0x200C || c == 0x200D) |
| 596 | return 0; |
| 597 | return 1; |
| 598 | } |
| 599 | } |
| 600 | |
| 601 | #define Hangul_Syllable_Type_NA 0 |
| 602 | #define Hangul_Syllable_Type_L 0x1100 |
| 603 | #define Hangul_Syllable_Type_V 0x1160 |
| 604 | #define Hangul_Syllable_Type_T 0x11A8 |
| 605 | #define Hangul_Syllable_Type_LV 0xAC00 |
| 606 | #define Hangul_Syllable_Type_LVT 0xAC01 |
| 607 | |
| 608 | /** @brief Determine Hangul_Syllable_Type of @p c |
| 609 | * @param c Code point |
| 610 | * @return Equivalance class of @p c, or Hangul_Syllable_Type_NA |
| 611 | * |
| 612 | * If this is a Hangul character then a representative member of its |
| 613 | * equivalence class is returned. Otherwise Hangul_Syllable_Type_NA is |
| 614 | * returned. |
| 615 | */ |
| 616 | static uint32_t utf32__hangul_syllable_type(uint32_t c) { |
| 617 | /* Dispose of the bulk of the non-Hangul code points first */ |
| 618 | if(c < 0x1100) return Hangul_Syllable_Type_NA; |
| 619 | if(c > 0x1200 && c < 0xAC00) return Hangul_Syllable_Type_NA; |
| 620 | if(c >= 0xD800) return Hangul_Syllable_Type_NA; |
| 621 | /* Now we pick out the assigned Hangul code points */ |
| 622 | if((c >= 0x1100 && c <= 0x1159) || c == 0x115F) return Hangul_Syllable_Type_L; |
| 623 | if(c >= 0x1160 && c <= 0x11A2) return Hangul_Syllable_Type_V; |
| 624 | if(c >= 0x11A8 && c <= 0x11F9) return Hangul_Syllable_Type_T; |
| 625 | if(c >= 0xAC00 && c <= 0xD7A3) { |
| 626 | if(c % 28 == 16) |
| 627 | return Hangul_Syllable_Type_LV; |
| 628 | else |
| 629 | return Hangul_Syllable_Type_LVT; |
| 630 | } |
| 631 | return Hangul_Syllable_Type_NA; |
| 632 | } |
| 633 | |
| 634 | /** @brief Determine Word_Break property |
| 635 | * @param c Code point |
| 636 | * @return Word_Break property value of @p c |
| 637 | */ |
| 638 | static enum unicode_Word_Break utf32__word_break(uint32_t c) { |
| 639 | if(c < 0xAC00 || c > 0xD7A3) |
| 640 | return utf32__unidata(c)->word_break; |
| 641 | else |
| 642 | return unicode_Word_Break_ALetter; |
| 643 | } |
| 644 | |
| 645 | /** @brief Identify a grapheme cluster boundary |
| 646 | * @param s Start of string (must be NFD) |
| 647 | * @param ns Length of string |
| 648 | * @param n Index within string (in [0,ns].) |
| 649 | * @return 1 at a grapheme cluster boundary, 0 otherwise |
| 650 | * |
| 651 | * This function identifies default grapheme cluster boundaries as described in |
| 652 | * UAX #29 s3. It returns 1 if @p n points at the code point just after a |
| 653 | * grapheme cluster boundary (including the hypothetical code point just after |
| 654 | * the end of the string). |
| 655 | */ |
| 656 | int utf32_is_gcb(const uint32_t *s, size_t ns, size_t n) { |
| 657 | uint32_t before, after; |
| 658 | uint32_t hbefore, hafter; |
| 659 | /* GB1 and GB2 */ |
| 660 | if(n == 0 || n == ns) |
| 661 | return 1; |
| 662 | /* Now we know that s[n-1] and s[n] are safe to inspect */ |
| 663 | /* GB3 */ |
| 664 | before = s[n-1]; |
| 665 | after = s[n]; |
| 666 | if(before == 0x000D && after == 0x000A) |
| 667 | return 0; |
| 668 | /* GB4 and GB5 */ |
| 669 | if(utf32__is_control_or_cr_or_lf(before) |
| 670 | || utf32__is_control_or_cr_or_lf(after)) |
| 671 | return 1; |
| 672 | hbefore = utf32__hangul_syllable_type(before); |
| 673 | hafter = utf32__hangul_syllable_type(after); |
| 674 | /* GB6 */ |
| 675 | if(hbefore == Hangul_Syllable_Type_L |
| 676 | && (hafter == Hangul_Syllable_Type_L |
| 677 | || hafter == Hangul_Syllable_Type_V |
| 678 | || hafter == Hangul_Syllable_Type_LV |
| 679 | || hafter == Hangul_Syllable_Type_LVT)) |
| 680 | return 0; |
| 681 | /* GB7 */ |
| 682 | if((hbefore == Hangul_Syllable_Type_LV |
| 683 | || hbefore == Hangul_Syllable_Type_V) |
| 684 | && (hafter == Hangul_Syllable_Type_V |
| 685 | || hafter == Hangul_Syllable_Type_T)) |
| 686 | return 0; |
| 687 | /* GB8 */ |
| 688 | if((hbefore == Hangul_Syllable_Type_LVT |
| 689 | || hbefore == Hangul_Syllable_Type_T) |
| 690 | && hafter == Hangul_Syllable_Type_T) |
| 691 | return 0; |
| 692 | /* GB9 */ |
| 693 | if(utf32__word_break(after) == unicode_Word_Break_Extend) |
| 694 | return 0; |
| 695 | /* GB10 */ |
| 696 | return 1; |
| 697 | } |
| 698 | |
| 699 | /** @brief Return true if @p c is ignorable for boundary specifications */ |
| 700 | static inline int utf32__boundary_ignorable(enum unicode_Word_Break wb) { |
| 701 | return (wb == unicode_Word_Break_Extend |
| 702 | || wb == unicode_Word_Break_Format); |
| 703 | } |
| 704 | |
| 705 | /** @brief Identify a word boundary |
| 706 | * @param s Start of string (must be NFD) |
| 707 | * @param ns Length of string |
| 708 | * @param n Index within string (in [0,ns].) |
| 709 | * @return 1 at a word boundary, 0 otherwise |
| 710 | * |
| 711 | * This function identifies default word boundaries as described in UAX #29 s4. |
| 712 | * It returns 1 if @p n points at the code point just after a word boundary |
| 713 | * (including the hypothetical code point just after the end of the string). |
| 714 | */ |
| 715 | int utf32_is_word_boundary(const uint32_t *s, size_t ns, size_t n) { |
| 716 | enum unicode_Word_Break twobefore, before, after, twoafter; |
| 717 | size_t nn; |
| 718 | |
| 719 | /* WB1 and WB2 */ |
| 720 | if(n == 0 || n == ns) |
| 721 | return 1; |
| 722 | /* WB3 */ |
| 723 | if(s[n-1] == 0x000D && s[n] == 0x000A) |
| 724 | return 0; |
| 725 | /* WB4 */ |
| 726 | /* (!Sep) x (Extend|Format) as in UAX #29 s6.2 */ |
| 727 | switch(s[n-1]) { /* bit of a bodge */ |
| 728 | case 0x000A: |
| 729 | case 0x000D: |
| 730 | case 0x0085: |
| 731 | case 0x2028: |
| 732 | case 0x2029: |
| 733 | break; |
| 734 | default: |
| 735 | if(utf32__boundary_ignorable(utf32__word_break(s[n]))) |
| 736 | return 0; |
| 737 | break; |
| 738 | } |
| 739 | /* Gather the property values we'll need for the rest of the test taking the |
| 740 | * s6.2 changes into account */ |
| 741 | /* First we look at the code points after the proposed boundary */ |
| 742 | nn = n; /* <ns */ |
| 743 | after = utf32__word_break(s[nn++]); |
| 744 | if(!utf32__boundary_ignorable(after)) { |
| 745 | /* X (Extend|Format)* -> X */ |
| 746 | while(nn < ns && utf32__boundary_ignorable(utf32__word_break(s[nn]))) |
| 747 | ++nn; |
| 748 | } |
| 749 | /* It's possible now that nn=ns */ |
| 750 | if(nn < ns) |
| 751 | twoafter = utf32__word_break(s[nn]); |
| 752 | else |
| 753 | twoafter = unicode_Word_Break_Other; |
| 754 | |
| 755 | /* Next we look at the code points before the proposed boundary. This is a |
| 756 | * bit fiddlier. */ |
| 757 | nn = n; |
| 758 | while(nn > 0 && utf32__boundary_ignorable(utf32__word_break(s[nn - 1]))) |
| 759 | --nn; |
| 760 | if(nn == 0) { |
| 761 | /* s[nn] must be ignorable */ |
| 762 | before = utf32__word_break(s[nn]); |
| 763 | twobefore = unicode_Word_Break_Other; |
| 764 | } else { |
| 765 | /* s[nn] is ignorable or after the proposed boundary; but s[nn-1] is not |
| 766 | * ignorable. */ |
| 767 | before = utf32__word_break(s[nn - 1]); |
| 768 | --nn; |
| 769 | /* Repeat the exercise */ |
| 770 | while(nn > 0 && utf32__boundary_ignorable(utf32__word_break(s[nn - 1]))) |
| 771 | --nn; |
| 772 | if(nn == 0) |
| 773 | twobefore = utf32__word_break(s[nn]); |
| 774 | else |
| 775 | twobefore = utf32__word_break(s[nn - 1]); |
| 776 | } |
| 777 | |
| 778 | /* WB5 */ |
| 779 | if(before == unicode_Word_Break_ALetter |
| 780 | && after == unicode_Word_Break_ALetter) |
| 781 | return 0; |
| 782 | /* WB6 */ |
| 783 | if(before == unicode_Word_Break_ALetter |
| 784 | && after == unicode_Word_Break_MidLetter |
| 785 | && twoafter == unicode_Word_Break_ALetter) |
| 786 | return 0; |
| 787 | /* WB7 */ |
| 788 | if(twobefore == unicode_Word_Break_ALetter |
| 789 | && before == unicode_Word_Break_MidLetter |
| 790 | && after == unicode_Word_Break_ALetter) |
| 791 | return 0; |
| 792 | /* WB8 */ |
| 793 | if(before == unicode_Word_Break_Numeric |
| 794 | && after == unicode_Word_Break_Numeric) |
| 795 | return 0; |
| 796 | /* WB9 */ |
| 797 | if(before == unicode_Word_Break_ALetter |
| 798 | && after == unicode_Word_Break_Numeric) |
| 799 | return 0; |
| 800 | /* WB10 */ |
| 801 | if(before == unicode_Word_Break_Numeric |
| 802 | && after == unicode_Word_Break_ALetter) |
| 803 | return 0; |
| 804 | /* WB11 */ |
| 805 | if(twobefore == unicode_Word_Break_Numeric |
| 806 | && before == unicode_Word_Break_MidNum |
| 807 | && after == unicode_Word_Break_Numeric) |
| 808 | return 0; |
| 809 | /* WB12 */ |
| 810 | if(before == unicode_Word_Break_Numeric |
| 811 | && after == unicode_Word_Break_MidNum |
| 812 | && twoafter == unicode_Word_Break_Numeric) |
| 813 | return 0; |
| 814 | /* WB13 */ |
| 815 | if(before == unicode_Word_Break_Katakana |
| 816 | && after == unicode_Word_Break_Katakana) |
| 817 | return 0; |
| 818 | /* WB13a */ |
| 819 | if((before == unicode_Word_Break_ALetter |
| 820 | || before == unicode_Word_Break_Numeric |
| 821 | || before == unicode_Word_Break_Katakana |
| 822 | || before == unicode_Word_Break_ExtendNumLet) |
| 823 | && after == unicode_Word_Break_ExtendNumLet) |
| 824 | return 0; |
| 825 | /* WB13b */ |
| 826 | if(before == unicode_Word_Break_ExtendNumLet |
| 827 | && (after == unicode_Word_Break_ALetter |
| 828 | || after == unicode_Word_Break_Numeric |
| 829 | || after == unicode_Word_Break_Katakana)) |
| 830 | return 0; |
| 831 | /* WB14 */ |
| 832 | return 1; |
| 833 | } |
| 834 | |
| 835 | /*@}*/ |
| 836 | /** @defgroup utf8 Functions that operate on UTF-8 strings */ |
| 837 | /*@{*/ |
| 838 | |
| 839 | /** @brief Wrapper to transform a UTF-8 string using the UTF-32 function */ |
| 840 | #define utf8__transform(FN) do { \ |
| 841 | uint32_t *to32 = 0, *decomp32 = 0; \ |
| 842 | size_t nto32, ndecomp32; \ |
| 843 | char *decomp8 = 0; \ |
| 844 | \ |
| 845 | if(!(to32 = utf8_to_utf32(s, ns, &nto32))) goto error; \ |
| 846 | if(!(decomp32 = FN(to32, nto32, &ndecomp32))) goto error; \ |
| 847 | decomp8 = utf32_to_utf8(decomp32, ndecomp32, ndp); \ |
| 848 | error: \ |
| 849 | xfree(to32); \ |
| 850 | xfree(decomp32); \ |
| 851 | return decomp8; \ |
| 852 | } while(0) |
| 853 | |
| 854 | /** @brief Canonically decompose @p [s,s+ns) |
| 855 | * @param s Pointer to string |
| 856 | * @param ns Length of string |
| 857 | * @param ndp Where to store length of result |
| 858 | * @return Pointer to result string, or NULL |
| 859 | * |
| 860 | * Computes the canonical decomposition of a string and stably sorts combining |
| 861 | * characters into canonical order. The result is in Normalization Form D and |
| 862 | * (at the time of writing!) passes the NFD tests defined in Unicode 5.0's |
| 863 | * NormalizationTest.txt. |
| 864 | * |
| 865 | * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why |
| 866 | * this might be. |
| 867 | * |
| 868 | * See also utf32_decompose_canon(). |
| 869 | */ |
| 870 | char *utf8_decompose_canon(const char *s, size_t ns, size_t *ndp) { |
| 871 | utf8__transform(utf32_decompose_canon); |
| 872 | } |
| 873 | |
| 874 | /** @brief Compatibility decompose @p [s,s+ns) |
| 875 | * @param s Pointer to string |
| 876 | * @param ns Length of string |
| 877 | * @param ndp Where to store length of result |
| 878 | * @return Pointer to result string, or NULL |
| 879 | * |
| 880 | * Computes the compatibility decomposition of a string and stably sorts |
| 881 | * combining characters into canonical order. The result is in Normalization |
| 882 | * Form KD and (at the time of writing!) passes the NFKD tests defined in |
| 883 | * Unicode 5.0's NormalizationTest.txt. |
| 884 | * |
| 885 | * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why |
| 886 | * this might be. |
| 887 | * |
| 888 | * See also utf32_decompose_compat(). |
| 889 | */ |
| 890 | char *utf8_decompose_compat(const char *s, size_t ns, size_t *ndp) { |
| 891 | utf8__transform(utf32_decompose_compat); |
| 892 | } |
| 893 | |
| 894 | /** @brief Case-fold @p [s,s+ns) |
| 895 | * @param s Pointer to string |
| 896 | * @param ns Length of string |
| 897 | * @param ndp Where to store length of result |
| 898 | * @return Pointer to result string, or NULL |
| 899 | * |
| 900 | * Case-fold the string at @p s according to full default case-folding rules |
| 901 | * (s3.13). The result will be in NFD. |
| 902 | * |
| 903 | * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why |
| 904 | * this might be. |
| 905 | */ |
| 906 | char *utf8_casefold_canon(const char *s, size_t ns, size_t *ndp) { |
| 907 | utf8__transform(utf32_casefold_canon); |
| 908 | } |
| 909 | |
| 910 | /** @brief Compatibility case-fold @p [s,s+ns) |
| 911 | * @param s Pointer to string |
| 912 | * @param ns Length of string |
| 913 | * @param ndp Where to store length of result |
| 914 | * @return Pointer to result string, or NULL |
| 915 | * |
| 916 | * Case-fold the string at @p s according to full default case-folding rules |
| 917 | * (s3.13). The result will be in NFKD. |
| 918 | * |
| 919 | * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why |
| 920 | * this might be. |
| 921 | */ |
| 922 | char *utf8_casefold_compat(const char *s, size_t ns, size_t *ndp) { |
| 923 | utf8__transform(utf32_casefold_compat); |
| 924 | } |
| 925 | |
| 926 | /*@}*/ |
| 927 | |
| 928 | /* |
| 929 | Local Variables: |
| 930 | c-basic-offset:2 |
| 931 | comment-column:40 |
| 932 | fill-column:79 |
| 933 | indent-tabs-mode:nil |
| 934 | End: |
| 935 | */ |