Commit | Line | Data |
---|---|---|
e5a5a138 RK |
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 | |
35b651f0 RK |
24 | * encoding schemes). The primary encoding form is UTF-32 but convenience |
25 | * wrappers using UTF-8 are provided for a number of functions. | |
e5a5a138 RK |
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 | ||
092f426f RK |
46 | /** @defgroup utf32props Unicode Code Point Properties */ |
47 | /*@{*/ | |
48 | ||
49 | static const struct unidata *utf32__unidata_hard(uint32_t c); | |
50 | ||
51 | /** @brief Find definition of code point @p c | |
52 | * @param c Code point | |
53 | * @return Pointer to @ref unidata structure for @p c | |
54 | * | |
55 | * @p c can be any 32-bit value, a sensible value will be returned regardless. | |
56 | * The returned pointer is NOT guaranteed to be unique to @p c. | |
57 | */ | |
58 | static inline const struct unidata *utf32__unidata(uint32_t c) { | |
59 | /* The bottom half of the table contains almost everything of interest | |
60 | * and we can just return the right thing straight away */ | |
61 | if(c < UNICODE_BREAK_START) | |
62 | return &unidata[c / UNICODE_MODULUS][c % UNICODE_MODULUS]; | |
63 | else | |
64 | return utf32__unidata_hard(c); | |
65 | } | |
66 | ||
67 | /** @brief Find definition of code point @p c | |
68 | * @param c Code point | |
69 | * @return Pointer to @ref unidata structure for @p c | |
70 | * | |
71 | * @p c can be any 32-bit value, a sensible value will be returned regardless. | |
72 | * The returned pointer is NOT guaranteed to be unique to @p c. | |
73 | * | |
74 | * Don't use this function (although it will work fine) - use utf32__unidata() | |
75 | * instead. | |
76 | */ | |
77 | static const struct unidata *utf32__unidata_hard(uint32_t c) { | |
78 | if(c < UNICODE_BREAK_START) | |
79 | return &unidata[c / UNICODE_MODULUS][c % UNICODE_MODULUS]; | |
80 | /* Within the break everything is unassigned */ | |
81 | if(c < UNICODE_BREAK_END) | |
82 | return utf32__unidata(0xFFFF); /* guaranteed to be Cn */ | |
83 | /* Planes 15 and 16 are (mostly) private use */ | |
84 | if((c >= 0xF0000 && c <= 0xFFFFD) | |
85 | || (c >= 0x100000 && c <= 0x10FFFD)) | |
86 | return utf32__unidata(0xE000); /* first Co code point */ | |
87 | /* Everything else above the break top is unassigned */ | |
88 | if(c >= UNICODE_BREAK_TOP) | |
89 | return utf32__unidata(0xFFFF); /* guaranteed to be Cn */ | |
90 | /* Currently the rest is language tags and variation selectors */ | |
91 | c -= (UNICODE_BREAK_END - UNICODE_BREAK_START); | |
92 | return &unidata[c / UNICODE_MODULUS][c % UNICODE_MODULUS]; | |
93 | } | |
94 | ||
95 | /** @brief Return the combining class of @p c | |
96 | * @param c Code point | |
97 | * @return Combining class of @p c | |
98 | * | |
99 | * @p c can be any 32-bit value, a sensible value will be returned regardless. | |
100 | */ | |
101 | static inline int utf32__combining_class(uint32_t c) { | |
102 | return utf32__unidata(c)->ccc; | |
103 | } | |
104 | ||
105 | /** @brief Return the General_Category value for @p c | |
106 | * @param Code point | |
107 | * @return General_Category property value | |
108 | * | |
109 | * @p c can be any 32-bit value, a sensible value will be returned regardless. | |
110 | */ | |
111 | static inline enum unicode_General_Category utf32__general_category(uint32_t c) { | |
112 | return utf32__unidata(c)->general_category; | |
113 | } | |
114 | ||
115 | /** @brief Determine Grapheme_Break property | |
116 | * @param c Code point | |
117 | * @return Grapheme_Break property value of @p c | |
118 | * | |
119 | * @p c can be any 32-bit value, a sensible value will be returned regardless. | |
120 | */ | |
121 | static inline enum unicode_Grapheme_Break utf32__grapheme_break(uint32_t c) { | |
122 | return utf32__unidata(c)->grapheme_break; | |
123 | } | |
124 | ||
125 | /** @brief Determine Word_Break property | |
126 | * @param c Code point | |
127 | * @return Word_Break property value of @p c | |
128 | * | |
129 | * @p c can be any 32-bit value, a sensible value will be returned regardless. | |
130 | */ | |
131 | static inline enum unicode_Word_Break utf32__word_break(uint32_t c) { | |
132 | return utf32__unidata(c)->word_break; | |
133 | } | |
134 | ||
135 | /** @brief Determine Sentence_Break property | |
136 | * @param c Code point | |
137 | * @return Word_Break property value of @p c | |
138 | * | |
139 | * @p c can be any 32-bit value, a sensible value will be returned regardless. | |
140 | */ | |
141 | static inline enum unicode_Sentence_Break utf32__sentence_break(uint32_t c) { | |
142 | return utf32__unidata(c)->sentence_break; | |
143 | } | |
144 | ||
145 | /** @brief Return true if @p c is ignorable for boundary specifications | |
146 | * @param wb Word break property value | |
147 | * @return non-0 if @p wb is unicode_Word_Break_Extend or unicode_Word_Break_Format | |
148 | */ | |
149 | static inline int utf32__boundary_ignorable(enum unicode_Word_Break wb) { | |
150 | return (wb == unicode_Word_Break_Extend | |
151 | || wb == unicode_Word_Break_Format); | |
152 | } | |
153 | ||
f98fcddb RK |
154 | /** @brief Return the canonical decomposition of @p c |
155 | * @param c Code point | |
156 | * @return 0-terminated canonical decomposition, or 0 | |
157 | */ | |
158 | static inline const uint32_t *utf32__decomposition_canon(uint32_t c) { | |
159 | const struct unidata *const data = utf32__unidata(c); | |
160 | const uint32_t *const decomp = data->decomp; | |
161 | ||
162 | if(decomp && !(data->flags & unicode_compatibility_decomposition)) | |
163 | return decomp; | |
164 | else | |
165 | return 0; | |
166 | } | |
167 | ||
168 | /** @brief Return the compatibility decomposition of @p c | |
169 | * @param c Code point | |
170 | * @return 0-terminated decomposition, or 0 | |
171 | */ | |
172 | static inline const uint32_t *utf32__decomposition_compat(uint32_t c) { | |
173 | return utf32__unidata(c)->decomp; | |
174 | } | |
175 | ||
092f426f | 176 | /*@}*/ |
e5a5a138 RK |
177 | /** @defgroup utftransform Functions that transform between different Unicode encoding forms */ |
178 | /*@{*/ | |
179 | ||
180 | /** @brief Convert UTF-32 to UTF-8 | |
181 | * @param s Source string | |
182 | * @param ns Length of source string in code points | |
183 | * @param ndp Where to store length of destination string (or NULL) | |
184 | * @return Newly allocated destination string or NULL on error | |
185 | * | |
56fd389c RK |
186 | * If the UTF-32 is not valid then NULL is returned. A UTF-32 code point is |
187 | * invalid if: | |
e5a5a138 RK |
188 | * - it codes for a UTF-16 surrogate |
189 | * - it codes for a value outside the unicode code space | |
190 | * | |
56fd389c RK |
191 | * The return value is always 0-terminated. The value returned via @p *ndp |
192 | * does not include the terminator. | |
e5a5a138 RK |
193 | */ |
194 | char *utf32_to_utf8(const uint32_t *s, size_t ns, size_t *ndp) { | |
195 | struct dynstr d; | |
196 | uint32_t c; | |
197 | ||
198 | dynstr_init(&d); | |
199 | while(ns > 0) { | |
200 | c = *s++; | |
201 | if(c < 0x80) | |
202 | dynstr_append(&d, c); | |
203 | else if(c < 0x0800) { | |
204 | dynstr_append(&d, 0xC0 | (c >> 6)); | |
205 | dynstr_append(&d, 0x80 | (c & 0x3F)); | |
206 | } else if(c < 0x10000) { | |
56fd389c | 207 | if(c >= 0xD800 && c <= 0xDFFF) |
e5a5a138 RK |
208 | goto error; |
209 | dynstr_append(&d, 0xE0 | (c >> 12)); | |
210 | dynstr_append(&d, 0x80 | ((c >> 6) & 0x3F)); | |
211 | dynstr_append(&d, 0x80 | (c & 0x3F)); | |
212 | } else if(c < 0x110000) { | |
213 | dynstr_append(&d, 0xF0 | (c >> 18)); | |
214 | dynstr_append(&d, 0x80 | ((c >> 12) & 0x3F)); | |
215 | dynstr_append(&d, 0x80 | ((c >> 6) & 0x3F)); | |
216 | dynstr_append(&d, 0x80 | (c & 0x3F)); | |
217 | } else | |
218 | goto error; | |
219 | --ns; | |
220 | } | |
221 | dynstr_terminate(&d); | |
222 | if(ndp) | |
223 | *ndp = d.nvec; | |
224 | return d.vec; | |
225 | error: | |
226 | xfree(d.vec); | |
227 | return 0; | |
228 | } | |
229 | ||
230 | /** @brief Convert UTF-8 to UTF-32 | |
231 | * @param s Source string | |
232 | * @param ns Length of source string in code points | |
233 | * @param ndp Where to store length of destination string (or NULL) | |
f98fcddb | 234 | * @return Newly allocated destination string or NULL on error |
e5a5a138 | 235 | * |
56fd389c RK |
236 | * The return value is always 0-terminated. The value returned via @p *ndp |
237 | * does not include the terminator. | |
e5a5a138 RK |
238 | * |
239 | * If the UTF-8 is not valid then NULL is returned. A UTF-8 sequence | |
240 | * for a code point is invalid if: | |
241 | * - it is not the shortest possible sequence for the code point | |
242 | * - it codes for a UTF-16 surrogate | |
243 | * - it codes for a value outside the unicode code space | |
244 | */ | |
245 | uint32_t *utf8_to_utf32(const char *s, size_t ns, size_t *ndp) { | |
246 | struct dynstr_ucs4 d; | |
32b158f2 | 247 | uint32_t c32; |
e5a5a138 | 248 | const uint8_t *ss = (const uint8_t *)s; |
32b158f2 | 249 | int n; |
e5a5a138 RK |
250 | |
251 | dynstr_ucs4_init(&d); | |
252 | while(ns > 0) { | |
32b158f2 RK |
253 | const struct unicode_utf8_row *const r = &unicode_utf8_valid[*ss]; |
254 | if(r->count <= ns) { | |
255 | switch(r->count) { | |
256 | case 1: | |
257 | c32 = *ss; | |
258 | break; | |
259 | case 2: | |
260 | if(ss[1] < r->min2 || ss[1] > r->max2) | |
261 | goto error; | |
262 | c32 = *ss & 0x1F; | |
263 | break; | |
264 | case 3: | |
265 | if(ss[1] < r->min2 || ss[1] > r->max2) | |
266 | goto error; | |
267 | c32 = *ss & 0x0F; | |
268 | break; | |
269 | case 4: | |
270 | if(ss[1] < r->min2 || ss[1] > r->max2) | |
271 | goto error; | |
272 | c32 = *ss & 0x07; | |
273 | break; | |
274 | default: | |
275 | goto error; | |
276 | } | |
e5a5a138 RK |
277 | } else |
278 | goto error; | |
32b158f2 RK |
279 | for(n = 1; n < r->count; ++n) { |
280 | if(ss[n] < 0x80 || ss[n] > 0xBF) | |
281 | goto error; | |
282 | c32 = (c32 << 6) | (ss[n] & 0x3F); | |
283 | } | |
e5a5a138 | 284 | dynstr_ucs4_append(&d, c32); |
32b158f2 RK |
285 | ss += r->count; |
286 | ns -= r->count; | |
e5a5a138 RK |
287 | } |
288 | dynstr_ucs4_terminate(&d); | |
289 | if(ndp) | |
290 | *ndp = d.nvec; | |
291 | return d.vec; | |
292 | error: | |
293 | xfree(d.vec); | |
294 | return 0; | |
295 | } | |
296 | ||
18cda350 RK |
297 | /** @brief Test whether [s,s+ns) is valid UTF-8 |
298 | * @param s Start of string | |
299 | * @param ns Length of string | |
300 | * @return non-0 if @p s is valid UTF-8, 0 if it is not valid | |
301 | * | |
302 | * This function is intended to be much faster than calling utf8_to_utf32() and | |
303 | * throwing away the result. | |
304 | */ | |
305 | int utf8_valid(const char *s, size_t ns) { | |
306 | const uint8_t *ss = (const uint8_t *)s; | |
307 | while(ns > 0) { | |
308 | const struct unicode_utf8_row *const r = &unicode_utf8_valid[*ss]; | |
309 | if(r->count <= ns) { | |
310 | switch(r->count) { | |
311 | case 1: | |
312 | break; | |
313 | case 2: | |
314 | if(ss[1] < r->min2 || ss[1] > r->max2) | |
315 | return 0; | |
316 | break; | |
317 | case 3: | |
318 | if(ss[1] < r->min2 || ss[1] > r->max2) | |
319 | return 0; | |
320 | if(ss[2] < 0x80 || ss[2] > 0xBF) | |
321 | return 0; | |
322 | break; | |
323 | case 4: | |
324 | if(ss[1] < r->min2 || ss[1] > r->max2) | |
325 | return 0; | |
326 | if(ss[2] < 0x80 || ss[2] > 0xBF) | |
327 | return 0; | |
328 | if(ss[3] < 0x80 || ss[3] > 0xBF) | |
329 | return 0; | |
330 | break; | |
331 | default: | |
332 | return 0; | |
333 | } | |
334 | } else | |
335 | return 0; | |
336 | ss += r->count; | |
337 | ns -= r->count; | |
338 | } | |
339 | return 1; | |
340 | } | |
341 | ||
092f426f RK |
342 | /*@}*/ |
343 | /** @defgroup utf32iterator UTF-32 string iterators */ | |
344 | /*@{*/ | |
345 | ||
346 | struct utf32_iterator_data { | |
347 | /** @brief Start of string */ | |
348 | const uint32_t *s; | |
349 | ||
350 | /** @brief Length of string */ | |
351 | size_t ns; | |
352 | ||
353 | /** @brief Current position */ | |
354 | size_t n; | |
355 | ||
356 | /** @brief Last two non-ignorable characters or (uint32_t)-1 | |
357 | * | |
358 | * last[1] is the non-Extend/Format character just before position @p n; | |
359 | * last[0] is the one just before that. | |
360 | * | |
361 | * Exception 1: if there is no such non-Extend/Format character then an | |
362 | * Extend/Format character is accepted instead. | |
363 | * | |
364 | * Exception 2: if there is no such character even taking that into account | |
365 | * the value is (uint32_t)-1. | |
366 | */ | |
367 | uint32_t last[2]; | |
368 | }; | |
369 | ||
370 | /** @brief Create a new iterator pointing at the start of a string | |
371 | * @param s Start of string | |
372 | * @param ns Length of string | |
373 | * @return New iterator | |
374 | */ | |
375 | utf32_iterator utf32_iterator_new(const uint32_t *s, size_t ns) { | |
376 | utf32_iterator it = xmalloc(sizeof *it); | |
377 | it->s = s; | |
378 | it->ns = ns; | |
379 | it->n = 0; | |
380 | it->last[0] = it->last[1] = -1; | |
381 | return it; | |
382 | } | |
383 | ||
384 | /** @brief Initialize an internal private iterator | |
385 | * @param it Iterator | |
386 | * @param s Start of string | |
387 | * @param ns Length of string | |
388 | * @param n Absolute position | |
389 | */ | |
390 | static void utf32__iterator_init(utf32_iterator it, | |
391 | const uint32_t *s, size_t ns, size_t n) { | |
392 | it->s = s; | |
393 | it->ns = ns; | |
394 | it->n = 0; | |
395 | it->last[0] = it->last[1] = -1; | |
b21a155c | 396 | utf32_iterator_set(it, n); |
092f426f RK |
397 | } |
398 | ||
399 | /** @brief Destroy an iterator | |
400 | * @param it Iterator | |
401 | */ | |
402 | void utf32_iterator_destroy(utf32_iterator it) { | |
403 | xfree(it); | |
404 | } | |
405 | ||
406 | /** @brief Find the current position of an interator | |
407 | * @param it Iterator | |
408 | */ | |
409 | size_t utf32_iterator_where(utf32_iterator it) { | |
410 | return it->n; | |
411 | } | |
412 | ||
413 | /** @brief Set an iterator's absolute position | |
414 | * @param it Iterator | |
415 | * @param n Absolute position | |
416 | * @return 0 on success, non-0 on error | |
417 | * | |
418 | * It is an error to position the iterator outside the string (but acceptable | |
419 | * to point it at the hypothetical post-final character). If an invalid value | |
420 | * of @p n is specified then the iterator is not changed. | |
f98fcddb RK |
421 | * |
422 | * This function works by backing up and then advancing to reconstruct the | |
423 | * iterator's internal state for position @p n. The worst case will be O(n) | |
424 | * time complexity (with a worse constant factor that utf32_iterator_advance()) | |
425 | * but the typical case is essentially constant-time. | |
092f426f RK |
426 | */ |
427 | int utf32_iterator_set(utf32_iterator it, size_t n) { | |
5617aaff RK |
428 | /* We can't just jump to position @p n; the @p last[] values will be wrong. |
429 | * What we need is to jump a bit behind @p n and then advance forward, | |
430 | * updating @p last[] along the way. How far back? We need to cross two | |
431 | * non-ignorable code points as we advance forwards, so we'd better pass two | |
432 | * such characters on the way back (if such are available). | |
433 | */ | |
b21a155c | 434 | size_t m; |
5617aaff RK |
435 | |
436 | if(n > it->ns) /* range check */ | |
092f426f | 437 | return -1; |
b21a155c RK |
438 | /* Walk backwards skipping ignorable code points */ |
439 | m = n; | |
440 | while(m > 0 && (utf32__boundary_ignorable(utf32__word_break(it->s[m-1])))) | |
441 | --m; | |
442 | /* Either m=0 or s[m-1] is not ignorable */ | |
443 | if(m > 0) { | |
444 | --m; | |
445 | /* s[m] is our first non-ignorable code; look for a second in the same | |
446 | way **/ | |
447 | while(m > 0 && (utf32__boundary_ignorable(utf32__word_break(it->s[m-1])))) | |
5617aaff | 448 | --m; |
b21a155c RK |
449 | /* Either m=0 or s[m-1] is not ignorable */ |
450 | if(m > 0) | |
451 | --m; | |
452 | } | |
453 | it->last[0] = it->last[1] = -1; | |
5617aaff RK |
454 | it->n = m; |
455 | return utf32_iterator_advance(it, n - m); | |
092f426f RK |
456 | } |
457 | ||
458 | /** @brief Advance an iterator | |
459 | * @param it Iterator | |
460 | * @param count Number of code points to advance by | |
461 | * @return 0 on success, non-0 on error | |
462 | * | |
463 | * It is an error to advance an iterator beyond the hypothetical post-final | |
464 | * character of the string. If an invalid value of @p n is specified then the | |
465 | * iterator is not changed. | |
466 | * | |
467 | * This function has O(n) time complexity: it works by advancing naively | |
468 | * forwards through the string. | |
469 | */ | |
470 | int utf32_iterator_advance(utf32_iterator it, size_t count) { | |
471 | if(count <= it->ns - it->n) { | |
472 | while(count > 0) { | |
473 | const uint32_t c = it->s[it->n]; | |
474 | const enum unicode_Word_Break wb = utf32__word_break(c); | |
475 | if(it->last[1] == (uint32_t)-1 | |
476 | || !utf32__boundary_ignorable(wb)) { | |
477 | it->last[0] = it->last[1]; | |
478 | it->last[1] = c; | |
479 | } | |
480 | ++it->n; | |
481 | --count; | |
482 | } | |
483 | return 0; | |
484 | } else | |
485 | return -1; | |
486 | } | |
487 | ||
488 | /** @brief Find the current code point | |
489 | * @param it Iterator | |
490 | * @return Current code point or 0 | |
491 | * | |
492 | * If the iterator points at the hypothetical post-final character of the | |
493 | * string then 0 is returned. NB that this doesn't mean that there aren't any | |
494 | * 0 code points inside the string! | |
495 | */ | |
496 | uint32_t utf32_iterator_code(utf32_iterator it) { | |
497 | if(it->n < it->ns) | |
498 | return it->s[it->n]; | |
499 | else | |
500 | return 0; | |
501 | } | |
502 | ||
503 | /** @brief Test for a grapheme boundary | |
504 | * @param it Iterator | |
505 | * @return Non-0 if pointing just after a grapheme boundary, otherwise 0 | |
f98fcddb RK |
506 | * |
507 | * This function identifies default grapheme cluster boundaries as described in | |
508 | * UAX #29 s3. It returns non-0 if @p it points at the code point just after a | |
509 | * grapheme cluster boundary (including the hypothetical code point just after | |
510 | * the end of the string). | |
092f426f RK |
511 | */ |
512 | int utf32_iterator_grapheme_boundary(utf32_iterator it) { | |
513 | uint32_t before, after; | |
514 | enum unicode_Grapheme_Break gbbefore, gbafter; | |
515 | /* GB1 and GB2 */ | |
516 | if(it->n == 0 || it->n == it->ns) | |
517 | return 1; | |
518 | /* Now we know that s[n-1] and s[n] are safe to inspect */ | |
519 | /* GB3 */ | |
520 | before = it->s[it->n-1]; | |
521 | after = it->s[it->n]; | |
522 | if(before == 0x000D && after == 0x000A) | |
523 | return 0; | |
524 | gbbefore = utf32__grapheme_break(before); | |
525 | gbafter = utf32__grapheme_break(after); | |
526 | /* GB4 */ | |
527 | if(gbbefore == unicode_Grapheme_Break_Control | |
528 | || before == 0x000D | |
529 | || before == 0x000A) | |
530 | return 1; | |
531 | /* GB5 */ | |
532 | if(gbafter == unicode_Grapheme_Break_Control | |
533 | || after == 0x000D | |
534 | || after == 0x000A) | |
535 | return 1; | |
536 | /* GB6 */ | |
537 | if(gbbefore == unicode_Grapheme_Break_L | |
538 | && (gbafter == unicode_Grapheme_Break_L | |
539 | || gbafter == unicode_Grapheme_Break_V | |
540 | || gbafter == unicode_Grapheme_Break_LV | |
541 | || gbafter == unicode_Grapheme_Break_LVT)) | |
542 | return 0; | |
543 | /* GB7 */ | |
544 | if((gbbefore == unicode_Grapheme_Break_LV | |
545 | || gbbefore == unicode_Grapheme_Break_V) | |
546 | && (gbafter == unicode_Grapheme_Break_V | |
547 | || gbafter == unicode_Grapheme_Break_T)) | |
548 | return 0; | |
549 | /* GB8 */ | |
550 | if((gbbefore == unicode_Grapheme_Break_LVT | |
551 | || gbbefore == unicode_Grapheme_Break_T) | |
552 | && gbafter == unicode_Grapheme_Break_T) | |
553 | return 0; | |
554 | /* GB9 */ | |
555 | if(gbafter == unicode_Grapheme_Break_Extend) | |
556 | return 0; | |
557 | /* GB10 */ | |
558 | return 1; | |
559 | ||
560 | } | |
561 | ||
562 | /** @brief Test for a word boundary | |
563 | * @param it Iterator | |
564 | * @return Non-0 if pointing just after a word boundary, otherwise 0 | |
f98fcddb RK |
565 | * |
566 | * This function identifies default word boundaries as described in UAX #29 s4. | |
567 | * It returns non-0 if @p it points at the code point just after a word | |
568 | * boundary (including the hypothetical code point just after the end of the | |
569 | * string) and 0 otherwise. | |
092f426f RK |
570 | */ |
571 | int utf32_iterator_word_boundary(utf32_iterator it) { | |
572 | enum unicode_Word_Break twobefore, before, after, twoafter; | |
573 | size_t nn; | |
574 | ||
575 | /* WB1 and WB2 */ | |
576 | if(it->n == 0 || it->n == it->ns) | |
577 | return 1; | |
578 | /* WB3 */ | |
579 | if(it->s[it->n-1] == 0x000D && it->s[it->n] == 0x000A) | |
580 | return 0; | |
581 | /* WB4 */ | |
582 | /* (!Sep) x (Extend|Format) as in UAX #29 s6.2 */ | |
583 | if(utf32__sentence_break(it->s[it->n-1]) != unicode_Sentence_Break_Sep | |
584 | && utf32__boundary_ignorable(utf32__word_break(it->s[it->n]))) | |
585 | return 0; | |
586 | /* Gather the property values we'll need for the rest of the test taking the | |
587 | * s6.2 changes into account */ | |
588 | /* First we look at the code points after the proposed boundary */ | |
589 | nn = it->n; /* <it->ns */ | |
590 | after = utf32__word_break(it->s[nn++]); | |
591 | if(!utf32__boundary_ignorable(after)) { | |
592 | /* X (Extend|Format)* -> X */ | |
593 | while(nn < it->ns | |
594 | && utf32__boundary_ignorable(utf32__word_break(it->s[nn]))) | |
595 | ++nn; | |
596 | } | |
597 | /* It's possible now that nn=ns */ | |
598 | if(nn < it->ns) | |
599 | twoafter = utf32__word_break(it->s[nn]); | |
600 | else | |
601 | twoafter = unicode_Word_Break_Other; | |
602 | ||
603 | /* We've already recorded the non-ignorable code points before the proposed | |
604 | * boundary */ | |
605 | before = utf32__word_break(it->last[1]); | |
606 | twobefore = utf32__word_break(it->last[0]); | |
607 | ||
608 | /* WB5 */ | |
609 | if(before == unicode_Word_Break_ALetter | |
610 | && after == unicode_Word_Break_ALetter) | |
611 | return 0; | |
612 | /* WB6 */ | |
613 | if(before == unicode_Word_Break_ALetter | |
614 | && after == unicode_Word_Break_MidLetter | |
615 | && twoafter == unicode_Word_Break_ALetter) | |
616 | return 0; | |
617 | /* WB7 */ | |
618 | if(twobefore == unicode_Word_Break_ALetter | |
619 | && before == unicode_Word_Break_MidLetter | |
620 | && after == unicode_Word_Break_ALetter) | |
621 | return 0; | |
622 | /* WB8 */ | |
623 | if(before == unicode_Word_Break_Numeric | |
624 | && after == unicode_Word_Break_Numeric) | |
625 | return 0; | |
626 | /* WB9 */ | |
627 | if(before == unicode_Word_Break_ALetter | |
628 | && after == unicode_Word_Break_Numeric) | |
629 | return 0; | |
630 | /* WB10 */ | |
631 | if(before == unicode_Word_Break_Numeric | |
632 | && after == unicode_Word_Break_ALetter) | |
633 | return 0; | |
634 | /* WB11 */ | |
635 | if(twobefore == unicode_Word_Break_Numeric | |
636 | && before == unicode_Word_Break_MidNum | |
637 | && after == unicode_Word_Break_Numeric) | |
638 | return 0; | |
639 | /* WB12 */ | |
640 | if(before == unicode_Word_Break_Numeric | |
641 | && after == unicode_Word_Break_MidNum | |
642 | && twoafter == unicode_Word_Break_Numeric) | |
643 | return 0; | |
644 | /* WB13 */ | |
645 | if(before == unicode_Word_Break_Katakana | |
646 | && after == unicode_Word_Break_Katakana) | |
647 | return 0; | |
648 | /* WB13a */ | |
649 | if((before == unicode_Word_Break_ALetter | |
650 | || before == unicode_Word_Break_Numeric | |
651 | || before == unicode_Word_Break_Katakana | |
652 | || before == unicode_Word_Break_ExtendNumLet) | |
653 | && after == unicode_Word_Break_ExtendNumLet) | |
654 | return 0; | |
655 | /* WB13b */ | |
656 | if(before == unicode_Word_Break_ExtendNumLet | |
657 | && (after == unicode_Word_Break_ALetter | |
658 | || after == unicode_Word_Break_Numeric | |
659 | || after == unicode_Word_Break_Katakana)) | |
660 | return 0; | |
661 | /* WB14 */ | |
662 | return 1; | |
663 | } | |
664 | ||
e5a5a138 RK |
665 | /*@}*/ |
666 | /** @defgroup utf32 Functions that operate on UTF-32 strings */ | |
667 | /*@{*/ | |
668 | ||
669 | /** @brief Return the length of a 0-terminated UTF-32 string | |
670 | * @param s Pointer to 0-terminated string | |
671 | * @return Length of string in code points (excluding terminator) | |
672 | * | |
56fd389c | 673 | * Unlike the conversion functions no validity checking is done on the string. |
e5a5a138 RK |
674 | */ |
675 | size_t utf32_len(const uint32_t *s) { | |
676 | const uint32_t *t = s; | |
677 | ||
678 | while(*t) | |
679 | ++t; | |
680 | return (size_t)(t - s); | |
681 | } | |
682 | ||
e5a5a138 RK |
683 | /** @brief Stably sort [s,s+ns) into descending order of combining class |
684 | * @param s Start of array | |
685 | * @param ns Number of elements, must be at least 1 | |
686 | * @param buffer Buffer of at least @p ns elements | |
687 | */ | |
688 | static void utf32__sort_ccc(uint32_t *s, size_t ns, uint32_t *buffer) { | |
689 | uint32_t *a, *b, *bp; | |
690 | size_t na, nb; | |
691 | ||
692 | switch(ns) { | |
693 | case 1: /* 1-element array is always sorted */ | |
694 | return; | |
695 | case 2: /* 2-element arrays are trivial to sort */ | |
696 | if(utf32__combining_class(s[0]) > utf32__combining_class(s[1])) { | |
697 | uint32_t tmp = s[0]; | |
698 | s[0] = s[1]; | |
699 | s[1] = tmp; | |
700 | } | |
701 | return; | |
702 | default: | |
703 | /* Partition the array */ | |
704 | na = ns / 2; | |
705 | nb = ns - na; | |
706 | a = s; | |
707 | b = s + na; | |
708 | /* Sort the two halves of the array */ | |
709 | utf32__sort_ccc(a, na, buffer); | |
710 | utf32__sort_ccc(b, nb, buffer); | |
711 | /* Merge them back into one, via the buffer */ | |
712 | bp = buffer; | |
713 | while(na > 0 && nb > 0) { | |
16506c9d | 714 | /* We want ascending order of combining class (hence <) |
e5a5a138 RK |
715 | * and we want stability within combining classes (hence <=) |
716 | */ | |
717 | if(utf32__combining_class(*a) <= utf32__combining_class(*b)) { | |
718 | *bp++ = *a++; | |
719 | --na; | |
720 | } else { | |
721 | *bp++ = *b++; | |
722 | --nb; | |
723 | } | |
724 | } | |
725 | while(na > 0) { | |
726 | *bp++ = *a++; | |
727 | --na; | |
728 | } | |
729 | while(nb > 0) { | |
730 | *bp++ = *b++; | |
731 | --nb; | |
732 | } | |
733 | memcpy(s, buffer, ns * sizeof(uint32_t)); | |
734 | return; | |
735 | } | |
736 | } | |
737 | ||
738 | /** @brief Put combining characters into canonical order | |
739 | * @param s Pointer to UTF-32 string | |
740 | * @param ns Length of @p s | |
f98fcddb | 741 | * @return 0 on success, non-0 on error |
e5a5a138 | 742 | * |
56fd389c RK |
743 | * @p s is modified in-place. See Unicode 5.0 s3.11 for details of the |
744 | * ordering. | |
e5a5a138 | 745 | * |
56fd389c | 746 | * Currently we only support a maximum of 1024 combining characters after each |
f98fcddb | 747 | * base character. If this limit is exceeded then a non-0 value is returned. |
e5a5a138 RK |
748 | */ |
749 | static int utf32__canonical_ordering(uint32_t *s, size_t ns) { | |
750 | size_t nc; | |
751 | uint32_t buffer[1024]; | |
752 | ||
753 | /* The ordering amounts to a stable sort of each contiguous group of | |
754 | * characters with non-0 combining class. */ | |
755 | while(ns > 0) { | |
756 | /* Skip non-combining characters */ | |
757 | if(utf32__combining_class(*s) == 0) { | |
758 | ++s; | |
759 | --ns; | |
760 | continue; | |
761 | } | |
762 | /* We must now have at least one combining character; see how many | |
763 | * there are */ | |
764 | for(nc = 1; nc < ns && utf32__combining_class(s[nc]) != 0; ++nc) | |
765 | ; | |
766 | if(nc > 1024) | |
767 | return -1; | |
768 | /* Sort the array */ | |
769 | utf32__sort_ccc(s, nc, buffer); | |
770 | s += nc; | |
771 | ns -= nc; | |
772 | } | |
773 | return 0; | |
774 | } | |
775 | ||
776 | /* Magic numbers from UAX #15 s16 */ | |
777 | #define SBase 0xAC00 | |
778 | #define LBase 0x1100 | |
779 | #define VBase 0x1161 | |
780 | #define TBase 0x11A7 | |
781 | #define LCount 19 | |
782 | #define VCount 21 | |
783 | #define TCount 28 | |
784 | #define NCount (VCount * TCount) | |
785 | #define SCount (LCount * NCount) | |
786 | ||
787 | /** @brief Guts of the decomposition lookup functions */ | |
788 | #define utf32__decompose_one_generic(WHICH) do { \ | |
f98fcddb | 789 | const uint32_t *dc = utf32__decomposition_##WHICH(c); \ |
e5a5a138 RK |
790 | if(dc) { \ |
791 | /* Found a canonical decomposition in the table */ \ | |
792 | while(*dc) \ | |
793 | utf32__decompose_one_##WHICH(d, *dc++); \ | |
794 | } else if(c >= SBase && c < SBase + SCount) { \ | |
795 | /* Mechanically decomposable Hangul syllable (UAX #15 s16) */ \ | |
796 | const uint32_t SIndex = c - SBase; \ | |
797 | const uint32_t L = LBase + SIndex / NCount; \ | |
798 | const uint32_t V = VBase + (SIndex % NCount) / TCount; \ | |
799 | const uint32_t T = TBase + SIndex % TCount; \ | |
800 | dynstr_ucs4_append(d, L); \ | |
801 | dynstr_ucs4_append(d, V); \ | |
802 | if(T != TBase) \ | |
803 | dynstr_ucs4_append(d, T); \ | |
804 | } else \ | |
805 | /* Equal to own canonical decomposition */ \ | |
806 | dynstr_ucs4_append(d, c); \ | |
807 | } while(0) | |
808 | ||
809 | /** @brief Recursively compute the canonical decomposition of @p c | |
810 | * @param d Dynamic string to store decomposition in | |
811 | * @param c Code point to decompose (must be a valid!) | |
f98fcddb | 812 | * @return 0 on success, non-0 on error |
e5a5a138 RK |
813 | */ |
814 | static void utf32__decompose_one_canon(struct dynstr_ucs4 *d, uint32_t c) { | |
815 | utf32__decompose_one_generic(canon); | |
816 | } | |
817 | ||
818 | /** @brief Recursively compute the compatibility decomposition of @p c | |
819 | * @param d Dynamic string to store decomposition in | |
820 | * @param c Code point to decompose (must be a valid!) | |
f98fcddb | 821 | * @return 0 on success, non-0 on error |
e5a5a138 RK |
822 | */ |
823 | static void utf32__decompose_one_compat(struct dynstr_ucs4 *d, uint32_t c) { | |
824 | utf32__decompose_one_generic(compat); | |
825 | } | |
826 | ||
16506c9d RK |
827 | /** @brief Magic utf32__compositions() return value for Hangul Choseong */ |
828 | static const uint32_t utf32__hangul_L[1]; | |
829 | ||
830 | /** @brief Return the list of compositions that @p c starts | |
831 | * @param c Starter code point | |
832 | * @return Composition list or NULL | |
833 | * | |
834 | * For Hangul leading (Choseong) jamo we return the special value | |
835 | * utf32__hangul_L. These code points are not listed as the targets of | |
836 | * canonical decompositions (make-unidata checks) so there is no confusion with | |
837 | * real decompositions here. | |
838 | */ | |
839 | static const uint32_t *utf32__compositions(uint32_t c) { | |
840 | const uint32_t *compositions = utf32__unidata(c)->composed; | |
841 | ||
842 | if(compositions) | |
843 | return compositions; | |
844 | /* Special-casing for Hangul */ | |
845 | switch(utf32__grapheme_break(c)) { | |
846 | default: | |
847 | return 0; | |
848 | case unicode_Grapheme_Break_L: | |
849 | return utf32__hangul_L; | |
850 | } | |
851 | } | |
852 | ||
853 | /** @brief Composition step | |
854 | * @param s Start of string | |
855 | * @param ns Length of string | |
856 | * @return New length of string | |
857 | * | |
858 | * This is called from utf32__decompose_generic() to compose the result string | |
859 | * in place. | |
860 | */ | |
861 | static size_t utf32__compose(uint32_t *s, size_t ns) { | |
862 | const uint32_t *compositions; | |
863 | uint32_t *start = s, *t = s, *tt, cc; | |
864 | ||
865 | while(ns > 0) { | |
866 | uint32_t starter = *s++; | |
867 | int block_starters = 0; | |
868 | --ns; | |
869 | /* We don't attempt to compose the following things: | |
870 | * - final characters whatever kind they are | |
871 | * - non-starter characters | |
872 | * - starters that don't take part in a canonical decomposition mapping | |
873 | */ | |
874 | if(ns == 0 | |
875 | || utf32__combining_class(starter) | |
876 | || !(compositions = utf32__compositions(starter))) { | |
877 | *t++ = starter; | |
878 | continue; | |
879 | } | |
880 | if(compositions != utf32__hangul_L) { | |
881 | /* Where we'll put the eventual starter */ | |
882 | tt = t++; | |
883 | do { | |
884 | /* See if we can find composition of starter+*s */ | |
885 | const uint32_t cchar = *s, *cp = compositions; | |
886 | while((cc = *cp++)) { | |
887 | const uint32_t *decomp = utf32__decomposition_canon(cc); | |
888 | /* We know decomp[0] == starter */ | |
889 | if(decomp[1] == cchar) | |
890 | break; | |
891 | } | |
892 | if(cc) { | |
893 | /* Found a composition: cc decomposes to starter,*s */ | |
894 | starter = cc; | |
895 | compositions = utf32__compositions(starter); | |
896 | ++s; | |
897 | --ns; | |
898 | } else { | |
899 | /* No composition found. */ | |
900 | const int class = utf32__combining_class(*s); | |
901 | if(class) { | |
902 | /* Transfer the uncomposable combining character to the output */ | |
903 | *t++ = *s++; | |
904 | --ns; | |
905 | /* All the combining characters of the same class of the | |
906 | * uncomposable character are blocked by it, but there may be | |
907 | * others of higher class later. We eat the uncomposable and | |
908 | * blocked characters and go back round the loop for that higher | |
909 | * class. */ | |
910 | while(ns > 0 && utf32__combining_class(*s) == class) { | |
911 | *t++ = *s++; | |
912 | --ns; | |
913 | } | |
914 | /* Block any subsequent starters */ | |
915 | block_starters = 1; | |
916 | } else { | |
917 | /* The uncombinable character is itself a starter, so we don't | |
918 | * transfer it to the output but instead go back round the main | |
919 | * loop. */ | |
920 | break; | |
921 | } | |
922 | } | |
923 | /* Keep going while there are still characters and the starter takes | |
924 | * part in some composition */ | |
925 | } while(ns > 0 && compositions | |
926 | && (!block_starters || utf32__combining_class(*s))); | |
927 | /* Store any remaining combining characters */ | |
928 | while(ns > 0 && utf32__combining_class(*s)) { | |
929 | *t++ = *s++; | |
930 | --ns; | |
931 | } | |
932 | /* Store the resulting starter */ | |
933 | *tt = starter; | |
934 | } else { | |
935 | /* Special-casing for Hangul | |
936 | * | |
937 | * If there are combining characters between the L and the V then they | |
938 | * will block the V and so no composition happens. Similarly combining | |
939 | * characters between V and T will block the T and so we only get as far | |
940 | * as LV. | |
941 | */ | |
942 | if(utf32__grapheme_break(*s) == unicode_Grapheme_Break_V) { | |
943 | const uint32_t V = *s++; | |
944 | const uint32_t LIndex = starter - LBase; | |
945 | const uint32_t VIndex = V - VBase; | |
946 | uint32_t TIndex; | |
947 | --ns; | |
948 | if(ns > 0 | |
949 | && utf32__grapheme_break(*s) == unicode_Grapheme_Break_T) { | |
950 | /* We have an L V T sequence */ | |
951 | const uint32_t T = *s++; | |
952 | TIndex = T - TBase; | |
953 | --ns; | |
954 | } else | |
955 | /* It's just L V */ | |
956 | TIndex = 0; | |
957 | /* Compose to LVT or LV as appropriate */ | |
958 | starter = (LIndex * VCount + VIndex) * TCount + TIndex + SBase; | |
959 | } /* else we only have L or LV and no V or T */ | |
960 | *t++ = starter; | |
961 | /* There could be some combining characters that belong to the V or T. | |
962 | * These will be treated as non-starter characters at the top of the loop | |
963 | * and thuss transferred to the output. */ | |
964 | } | |
965 | } | |
966 | return t - start; | |
967 | } | |
968 | ||
969 | /** @brief Guts of the composition and decomposition functions | |
970 | * @param WHICH @c canon or @c compat to choose decomposition | |
971 | * @param COMPOSE @c 0 or @c 1 to compose | |
972 | */ | |
973 | #define utf32__decompose_generic(WHICH, COMPOSE) do { \ | |
e5a5a138 RK |
974 | struct dynstr_ucs4 d; \ |
975 | uint32_t c; \ | |
976 | \ | |
977 | dynstr_ucs4_init(&d); \ | |
978 | while(ns) { \ | |
979 | c = *s++; \ | |
56fd389c | 980 | if((c >= 0xD800 && c <= 0xDFFF) || c > 0x10FFFF) \ |
e5a5a138 RK |
981 | goto error; \ |
982 | utf32__decompose_one_##WHICH(&d, c); \ | |
983 | --ns; \ | |
984 | } \ | |
985 | if(utf32__canonical_ordering(d.vec, d.nvec)) \ | |
986 | goto error; \ | |
16506c9d RK |
987 | if(COMPOSE) \ |
988 | d.nvec = utf32__compose(d.vec, d.nvec); \ | |
e5a5a138 RK |
989 | dynstr_ucs4_terminate(&d); \ |
990 | if(ndp) \ | |
991 | *ndp = d.nvec; \ | |
992 | return d.vec; \ | |
993 | error: \ | |
994 | xfree(d.vec); \ | |
995 | return 0; \ | |
996 | } while(0) | |
997 | ||
998 | /** @brief Canonically decompose @p [s,s+ns) | |
999 | * @param s Pointer to string | |
1000 | * @param ns Length of string | |
1001 | * @param ndp Where to store length of result | |
f98fcddb | 1002 | * @return Pointer to result string, or NULL on error |
e5a5a138 | 1003 | * |
16506c9d RK |
1004 | * Computes NFD (Normalization Form D) of the string at @p s. This implies |
1005 | * performing all canonical decompositions and then normalizing the order of | |
1006 | * combining characters. | |
e5a5a138 | 1007 | * |
56fd389c | 1008 | * Returns NULL if the string is not valid for either of the following reasons: |
e5a5a138 RK |
1009 | * - it codes for a UTF-16 surrogate |
1010 | * - it codes for a value outside the unicode code space | |
16506c9d RK |
1011 | * |
1012 | * See also: | |
1013 | * - utf32_decompose_compat() | |
1014 | * - utf32_compose_canon() | |
e5a5a138 RK |
1015 | */ |
1016 | uint32_t *utf32_decompose_canon(const uint32_t *s, size_t ns, size_t *ndp) { | |
16506c9d | 1017 | utf32__decompose_generic(canon, 0); |
e5a5a138 RK |
1018 | } |
1019 | ||
1020 | /** @brief Compatibility decompose @p [s,s+ns) | |
1021 | * @param s Pointer to string | |
1022 | * @param ns Length of string | |
1023 | * @param ndp Where to store length of result | |
f98fcddb | 1024 | * @return Pointer to result string, or NULL on error |
e5a5a138 | 1025 | * |
16506c9d RK |
1026 | * Computes NFKD (Normalization Form KD) of the string at @p s. This implies |
1027 | * performing all canonical and compatibility decompositions and then | |
1028 | * normalizing the order of combining characters. | |
e5a5a138 | 1029 | * |
56fd389c | 1030 | * Returns NULL if the string is not valid for either of the following reasons: |
e5a5a138 RK |
1031 | * - it codes for a UTF-16 surrogate |
1032 | * - it codes for a value outside the unicode code space | |
16506c9d RK |
1033 | * |
1034 | * See also: | |
1035 | * - utf32_decompose_canon() | |
1036 | * - utf32_compose_compat() | |
e5a5a138 RK |
1037 | */ |
1038 | uint32_t *utf32_decompose_compat(const uint32_t *s, size_t ns, size_t *ndp) { | |
16506c9d RK |
1039 | utf32__decompose_generic(compat, 0); |
1040 | } | |
1041 | ||
1042 | /** @brief Canonically compose @p [s,s+ns) | |
1043 | * @param s Pointer to string | |
1044 | * @param ns Length of string | |
1045 | * @param ndp Where to store length of result | |
1046 | * @return Pointer to result string, or NULL on error | |
1047 | * | |
1048 | * Computes NFC (Normalization Form C) of the string at @p s. This implies | |
1049 | * performing all canonical decompositions, normalizing the order of combining | |
1050 | * characters and then composing all unblocked primary compositables. | |
1051 | * | |
1052 | * Returns NULL if the string is not valid for either of the following reasons: | |
1053 | * - it codes for a UTF-16 surrogate | |
1054 | * - it codes for a value outside the unicode code space | |
1055 | * | |
1056 | * See also: | |
1057 | * - utf32_compose_compat() | |
1058 | * - utf32_decompose_canon() | |
1059 | */ | |
1060 | uint32_t *utf32_compose_canon(const uint32_t *s, size_t ns, size_t *ndp) { | |
1061 | utf32__decompose_generic(canon, 1); | |
1062 | } | |
1063 | ||
1064 | /** @brief Compatibility compose @p [s,s+ns) | |
1065 | * @param s Pointer to string | |
1066 | * @param ns Length of string | |
1067 | * @param ndp Where to store length of result | |
1068 | * @return Pointer to result string, or NULL on error | |
1069 | * | |
1070 | * Computes NFKC (Normalization Form KC) of the string at @p s. This implies | |
1071 | * performing all canonical and compatibility decompositions, normalizing the | |
1072 | * order of combining characters and then composing all unblocked primary | |
1073 | * compositables. | |
1074 | * | |
1075 | * Returns NULL if the string is not valid for either of the following reasons: | |
1076 | * - it codes for a UTF-16 surrogate | |
1077 | * - it codes for a value outside the unicode code space | |
1078 | * | |
1079 | * See also: | |
1080 | * - utf32_compose_canon() | |
1081 | * - utf32_decompose_compat() | |
1082 | */ | |
1083 | uint32_t *utf32_compose_compat(const uint32_t *s, size_t ns, size_t *ndp) { | |
1084 | utf32__decompose_generic(compat, 1); | |
e5a5a138 RK |
1085 | } |
1086 | ||
56fd389c RK |
1087 | /** @brief Single-character case-fold and decompose operation */ |
1088 | #define utf32__casefold_one(WHICH) do { \ | |
bcf9ed7f | 1089 | const uint32_t *cf = utf32__unidata(c)->casefold; \ |
56fd389c RK |
1090 | if(cf) { \ |
1091 | /* Found a case-fold mapping in the table */ \ | |
1092 | while(*cf) \ | |
1093 | utf32__decompose_one_##WHICH(&d, *cf++); \ | |
1094 | } else \ | |
1095 | utf32__decompose_one_##WHICH(&d, c); \ | |
1096 | } while(0) | |
e5a5a138 RK |
1097 | |
1098 | /** @brief Case-fold @p [s,s+ns) | |
1099 | * @param s Pointer to string | |
1100 | * @param ns Length of string | |
1101 | * @param ndp Where to store length of result | |
f98fcddb | 1102 | * @return Pointer to result string, or NULL on error |
e5a5a138 RK |
1103 | * |
1104 | * Case-fold the string at @p s according to full default case-folding rules | |
56fd389c | 1105 | * (s3.13) for caseless matching. The result will be in NFD. |
e5a5a138 | 1106 | * |
56fd389c | 1107 | * Returns NULL if the string is not valid for either of the following reasons: |
e5a5a138 RK |
1108 | * - it codes for a UTF-16 surrogate |
1109 | * - it codes for a value outside the unicode code space | |
1110 | */ | |
1111 | uint32_t *utf32_casefold_canon(const uint32_t *s, size_t ns, size_t *ndp) { | |
1112 | struct dynstr_ucs4 d; | |
1113 | uint32_t c; | |
1114 | size_t n; | |
1115 | uint32_t *ss = 0; | |
1116 | ||
1117 | /* If the canonical decomposition of the string includes any combining | |
1118 | * character that case-folds to a non-combining character then we must | |
1119 | * normalize before we fold. In Unicode 5.0.0 this means 0345 COMBINING | |
1120 | * GREEK YPOGEGRAMMENI in its decomposition and the various characters that | |
1121 | * canonically decompose to it. */ | |
bcf9ed7f RK |
1122 | for(n = 0; n < ns; ++n) |
1123 | if(utf32__unidata(s[n])->flags & unicode_normalize_before_casefold) | |
e5a5a138 | 1124 | break; |
e5a5a138 RK |
1125 | if(n < ns) { |
1126 | /* We need a preliminary decomposition */ | |
1127 | if(!(ss = utf32_decompose_canon(s, ns, &ns))) | |
1128 | return 0; | |
1129 | s = ss; | |
1130 | } | |
1131 | dynstr_ucs4_init(&d); | |
1132 | while(ns) { | |
1133 | c = *s++; | |
56fd389c | 1134 | if((c >= 0xD800 && c <= 0xDFFF) || c > 0x10FFFF) |
e5a5a138 | 1135 | goto error; |
56fd389c | 1136 | utf32__casefold_one(canon); |
e5a5a138 RK |
1137 | --ns; |
1138 | } | |
1139 | if(utf32__canonical_ordering(d.vec, d.nvec)) | |
1140 | goto error; | |
1141 | dynstr_ucs4_terminate(&d); | |
1142 | if(ndp) | |
1143 | *ndp = d.nvec; | |
1144 | return d.vec; | |
1145 | error: | |
1146 | xfree(d.vec); | |
1147 | xfree(ss); | |
1148 | return 0; | |
1149 | } | |
1150 | ||
f98fcddb | 1151 | /** @brief Compatibility case-fold @p [s,s+ns) |
56fd389c RK |
1152 | * @param s Pointer to string |
1153 | * @param ns Length of string | |
1154 | * @param ndp Where to store length of result | |
f98fcddb | 1155 | * @return Pointer to result string, or NULL on error |
56fd389c RK |
1156 | * |
1157 | * Case-fold the string at @p s according to full default case-folding rules | |
1158 | * (s3.13) for compatibility caseless matching. The result will be in NFKD. | |
1159 | * | |
1160 | * Returns NULL if the string is not valid for either of the following reasons: | |
1161 | * - it codes for a UTF-16 surrogate | |
1162 | * - it codes for a value outside the unicode code space | |
1163 | */ | |
1164 | uint32_t *utf32_casefold_compat(const uint32_t *s, size_t ns, size_t *ndp) { | |
1165 | struct dynstr_ucs4 d; | |
1166 | uint32_t c; | |
1167 | size_t n; | |
1168 | uint32_t *ss = 0; | |
1169 | ||
bcf9ed7f RK |
1170 | for(n = 0; n < ns; ++n) |
1171 | if(utf32__unidata(s[n])->flags & unicode_normalize_before_casefold) | |
56fd389c | 1172 | break; |
56fd389c RK |
1173 | if(n < ns) { |
1174 | /* We need a preliminary _canonical_ decomposition */ | |
1175 | if(!(ss = utf32_decompose_canon(s, ns, &ns))) | |
1176 | return 0; | |
1177 | s = ss; | |
1178 | } | |
1179 | /* This computes NFKD(toCaseFold(s)) */ | |
1180 | #define compat_casefold_middle() do { \ | |
1181 | dynstr_ucs4_init(&d); \ | |
1182 | while(ns) { \ | |
1183 | c = *s++; \ | |
1184 | if((c >= 0xD800 && c <= 0xDFFF) || c > 0x10FFFF) \ | |
1185 | goto error; \ | |
1186 | utf32__casefold_one(compat); \ | |
1187 | --ns; \ | |
1188 | } \ | |
1189 | if(utf32__canonical_ordering(d.vec, d.nvec)) \ | |
1190 | goto error; \ | |
1191 | } while(0) | |
1192 | /* Do the inner (NFKD o toCaseFold) */ | |
1193 | compat_casefold_middle(); | |
1194 | /* We can do away with the NFD'd copy of the input now */ | |
1195 | xfree(ss); | |
1196 | s = ss = d.vec; | |
1197 | ns = d.nvec; | |
1198 | /* Do the outer (NFKD o toCaseFold) */ | |
1199 | compat_casefold_middle(); | |
1200 | /* That's all */ | |
1201 | dynstr_ucs4_terminate(&d); | |
1202 | if(ndp) | |
1203 | *ndp = d.nvec; | |
1204 | return d.vec; | |
1205 | error: | |
1206 | xfree(d.vec); | |
1207 | xfree(ss); | |
1208 | return 0; | |
1209 | } | |
1210 | ||
e5a5a138 RK |
1211 | /** @brief Order a pair of UTF-32 strings |
1212 | * @param a First 0-terminated string | |
1213 | * @param b Second 0-terminated string | |
1214 | * @return -1, 0 or 1 for a less than, equal to or greater than b | |
1215 | * | |
1216 | * "Comparable to strcmp() at its best." | |
1217 | */ | |
1218 | int utf32_cmp(const uint32_t *a, const uint32_t *b) { | |
1219 | while(*a && *b && *a == *b) { | |
1220 | ++a; | |
1221 | ++b; | |
1222 | } | |
1223 | return *a < *b ? -1 : (*a > *b ? 1 : 0); | |
1224 | } | |
1225 | ||
35b651f0 RK |
1226 | /** @brief Identify a grapheme cluster boundary |
1227 | * @param s Start of string (must be NFD) | |
1228 | * @param ns Length of string | |
1229 | * @param n Index within string (in [0,ns].) | |
1230 | * @return 1 at a grapheme cluster boundary, 0 otherwise | |
1231 | * | |
1232 | * This function identifies default grapheme cluster boundaries as described in | |
f98fcddb | 1233 | * UAX #29 s3. It returns non-0 if @p n points at the code point just after a |
35b651f0 RK |
1234 | * grapheme cluster boundary (including the hypothetical code point just after |
1235 | * the end of the string). | |
f98fcddb RK |
1236 | * |
1237 | * This function uses utf32_iterator_set() internally; see that function for | |
1238 | * remarks on performance. | |
35b651f0 | 1239 | */ |
1625e11a | 1240 | int utf32_is_grapheme_boundary(const uint32_t *s, size_t ns, size_t n) { |
092f426f | 1241 | struct utf32_iterator_data it[1]; |
35b651f0 | 1242 | |
092f426f RK |
1243 | utf32__iterator_init(it, s, ns, n); |
1244 | return utf32_iterator_grapheme_boundary(it); | |
0b7052da RK |
1245 | } |
1246 | ||
1247 | /** @brief Identify a word boundary | |
1248 | * @param s Start of string (must be NFD) | |
1249 | * @param ns Length of string | |
1250 | * @param n Index within string (in [0,ns].) | |
1251 | * @return 1 at a word boundary, 0 otherwise | |
1252 | * | |
1253 | * This function identifies default word boundaries as described in UAX #29 s4. | |
f98fcddb | 1254 | * It returns non-0 if @p n points at the code point just after a word boundary |
0b7052da | 1255 | * (including the hypothetical code point just after the end of the string). |
f98fcddb RK |
1256 | * |
1257 | * This function uses utf32_iterator_set() internally; see that function for | |
1258 | * remarks on performance. | |
0b7052da RK |
1259 | */ |
1260 | int utf32_is_word_boundary(const uint32_t *s, size_t ns, size_t n) { | |
092f426f | 1261 | struct utf32_iterator_data it[1]; |
0b7052da | 1262 | |
092f426f RK |
1263 | utf32__iterator_init(it, s, ns, n); |
1264 | return utf32_iterator_word_boundary(it); | |
0b7052da RK |
1265 | } |
1266 | ||
e5a5a138 | 1267 | /*@}*/ |
349b7b74 | 1268 | /** @defgroup utf8 Functions that operate on UTF-8 strings */ |
e5a5a138 RK |
1269 | /*@{*/ |
1270 | ||
1271 | /** @brief Wrapper to transform a UTF-8 string using the UTF-32 function */ | |
1272 | #define utf8__transform(FN) do { \ | |
1273 | uint32_t *to32 = 0, *decomp32 = 0; \ | |
1274 | size_t nto32, ndecomp32; \ | |
1275 | char *decomp8 = 0; \ | |
1276 | \ | |
1277 | if(!(to32 = utf8_to_utf32(s, ns, &nto32))) goto error; \ | |
1278 | if(!(decomp32 = FN(to32, nto32, &ndecomp32))) goto error; \ | |
1279 | decomp8 = utf32_to_utf8(decomp32, ndecomp32, ndp); \ | |
1280 | error: \ | |
1281 | xfree(to32); \ | |
1282 | xfree(decomp32); \ | |
1283 | return decomp8; \ | |
1284 | } while(0) | |
1285 | ||
1286 | /** @brief Canonically decompose @p [s,s+ns) | |
1287 | * @param s Pointer to string | |
1288 | * @param ns Length of string | |
1289 | * @param ndp Where to store length of result | |
f98fcddb | 1290 | * @return Pointer to result string, or NULL on error |
e5a5a138 RK |
1291 | * |
1292 | * Computes the canonical decomposition of a string and stably sorts combining | |
1293 | * characters into canonical order. The result is in Normalization Form D and | |
1294 | * (at the time of writing!) passes the NFD tests defined in Unicode 5.0's | |
1295 | * NormalizationTest.txt. | |
1296 | * | |
1297 | * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why | |
1298 | * this might be. | |
1299 | * | |
1300 | * See also utf32_decompose_canon(). | |
1301 | */ | |
1302 | char *utf8_decompose_canon(const char *s, size_t ns, size_t *ndp) { | |
1303 | utf8__transform(utf32_decompose_canon); | |
1304 | } | |
1305 | ||
1306 | /** @brief Compatibility decompose @p [s,s+ns) | |
1307 | * @param s Pointer to string | |
1308 | * @param ns Length of string | |
1309 | * @param ndp Where to store length of result | |
f98fcddb | 1310 | * @return Pointer to result string, or NULL on error |
e5a5a138 RK |
1311 | * |
1312 | * Computes the compatibility decomposition of a string and stably sorts | |
1313 | * combining characters into canonical order. The result is in Normalization | |
1314 | * Form KD and (at the time of writing!) passes the NFKD tests defined in | |
1315 | * Unicode 5.0's NormalizationTest.txt. | |
1316 | * | |
1317 | * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why | |
1318 | * this might be. | |
1319 | * | |
1320 | * See also utf32_decompose_compat(). | |
1321 | */ | |
1322 | char *utf8_decompose_compat(const char *s, size_t ns, size_t *ndp) { | |
1323 | utf8__transform(utf32_decompose_compat); | |
1324 | } | |
1325 | ||
1326 | /** @brief Case-fold @p [s,s+ns) | |
1327 | * @param s Pointer to string | |
1328 | * @param ns Length of string | |
1329 | * @param ndp Where to store length of result | |
f98fcddb | 1330 | * @return Pointer to result string, or NULL on error |
e5a5a138 RK |
1331 | * |
1332 | * Case-fold the string at @p s according to full default case-folding rules | |
1333 | * (s3.13). The result will be in NFD. | |
1334 | * | |
1335 | * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why | |
1336 | * this might be. | |
1337 | */ | |
1338 | char *utf8_casefold_canon(const char *s, size_t ns, size_t *ndp) { | |
1339 | utf8__transform(utf32_casefold_canon); | |
1340 | } | |
1341 | ||
1342 | /** @brief Compatibility case-fold @p [s,s+ns) | |
1343 | * @param s Pointer to string | |
1344 | * @param ns Length of string | |
1345 | * @param ndp Where to store length of result | |
f98fcddb | 1346 | * @return Pointer to result string, or NULL on error |
e5a5a138 RK |
1347 | * |
1348 | * Case-fold the string at @p s according to full default case-folding rules | |
1349 | * (s3.13). The result will be in NFKD. | |
1350 | * | |
1351 | * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why | |
1352 | * this might be. | |
1353 | */ | |
e5a5a138 RK |
1354 | char *utf8_casefold_compat(const char *s, size_t ns, size_t *ndp) { |
1355 | utf8__transform(utf32_casefold_compat); | |
1356 | } | |
e5a5a138 RK |
1357 | |
1358 | /*@}*/ | |
1359 | ||
1360 | /* | |
1361 | Local Variables: | |
1362 | c-basic-offset:2 | |
1363 | comment-column:40 | |
1364 | fill-column:79 | |
1365 | indent-tabs-mode:nil | |
1366 | End: | |
1367 | */ |