2 * This file is part of DisOrder
3 * Copyright (C) 2007 Richard Kettlewell
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.
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.
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
20 /** @file lib/unicode.c
21 * @brief Unicode support functions
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.
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
32 * As the code stands this guarantee is not well met!
39 #include <stdio.h> /* TODO */
46 /** @defgroup utf32props Unicode Code Point Properties */
49 static const struct unidata *utf32__unidata_hard(uint32_t c);
51 /** @brief Find definition of code point @p c
53 * @return Pointer to @ref unidata structure for @p c
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.
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];
64 return utf32__unidata_hard(c);
67 /** @brief Find definition of code point @p c
69 * @return Pointer to @ref unidata structure for @p c
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.
74 * Don't use this function (although it will work fine) - use utf32__unidata()
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];
95 /** @brief Return the combining class of @p c
97 * @return Combining class of @p c
99 * @p c can be any 32-bit value, a sensible value will be returned regardless.
101 static inline int utf32__combining_class(uint32_t c) {
102 return utf32__unidata(c)->ccc;
105 /** @brief Return the General_Category value for @p c
107 * @return General_Category property value
109 * @p c can be any 32-bit value, a sensible value will be returned regardless.
111 static inline enum unicode_General_Category utf32__general_category(uint32_t c) {
112 return utf32__unidata(c)->general_category;
115 /** @brief Determine Grapheme_Break property
116 * @param c Code point
117 * @return Grapheme_Break property value of @p c
119 * @p c can be any 32-bit value, a sensible value will be returned regardless.
121 static inline enum unicode_Grapheme_Break utf32__grapheme_break(uint32_t c) {
122 return utf32__unidata(c)->grapheme_break;
125 /** @brief Determine Word_Break property
126 * @param c Code point
127 * @return Word_Break property value of @p c
129 * @p c can be any 32-bit value, a sensible value will be returned regardless.
131 static inline enum unicode_Word_Break utf32__word_break(uint32_t c) {
132 return utf32__unidata(c)->word_break;
135 /** @brief Determine Sentence_Break property
136 * @param c Code point
137 * @return Word_Break property value of @p c
139 * @p c can be any 32-bit value, a sensible value will be returned regardless.
141 static inline enum unicode_Sentence_Break utf32__sentence_break(uint32_t c) {
142 return utf32__unidata(c)->sentence_break;
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
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);
154 /** @brief Return the canonical decomposition of @p c
155 * @param c Code point
156 * @return 0-terminated canonical decomposition, or 0
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;
162 if(decomp && !(data->flags & unicode_compatibility_decomposition))
168 /** @brief Return the compatibility decomposition of @p c
169 * @param c Code point
170 * @return 0-terminated decomposition, or 0
172 static inline const uint32_t *utf32__decomposition_compat(uint32_t c) {
173 return utf32__unidata(c)->decomp;
177 /** @defgroup utftransform Functions that transform between different Unicode encoding forms */
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
186 * If the UTF-32 is not valid then NULL is returned. A UTF-32 code point is
188 * - it codes for a UTF-16 surrogate
189 * - it codes for a value outside the unicode code space
191 * The return value is always 0-terminated. The value returned via @p *ndp
192 * does not include the terminator.
194 char *utf32_to_utf8(const uint32_t *s, size_t ns, size_t *ndp) {
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) {
207 if(c >= 0xD800 && c <= 0xDFFF)
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));
221 dynstr_terminate(&d);
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)
234 * @return Newly allocated destination string or NULL on error
236 * The return value is always 0-terminated. The value returned via @p *ndp
237 * does not include the terminator.
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
245 uint32_t *utf8_to_utf32(const char *s, size_t ns, size_t *ndp) {
246 struct dynstr_ucs4 d;
248 const uint8_t *ss = (const uint8_t *)s;
251 dynstr_ucs4_init(&d);
253 const struct unicode_utf8_row *const r = &unicode_utf8_valid[*ss];
260 if(ss[1] < r->min2 || ss[1] > r->max2)
265 if(ss[1] < r->min2 || ss[1] > r->max2)
270 if(ss[1] < r->min2 || ss[1] > r->max2)
279 for(n = 1; n < r->count; ++n) {
280 if(ss[n] < 0x80 || ss[n] > 0xBF)
282 c32 = (c32 << 6) | (ss[n] & 0x3F);
284 dynstr_ucs4_append(&d, c32);
288 dynstr_ucs4_terminate(&d);
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
302 * This function is intended to be much faster than calling utf8_to_utf32() and
303 * throwing away the result.
305 int utf8_valid(const char *s, size_t ns) {
306 const uint8_t *ss = (const uint8_t *)s;
308 const struct unicode_utf8_row *const r = &unicode_utf8_valid[*ss];
314 if(ss[1] < r->min2 || ss[1] > r->max2)
318 if(ss[1] < r->min2 || ss[1] > r->max2)
320 if(ss[2] < 0x80 || ss[2] > 0xBF)
324 if(ss[1] < r->min2 || ss[1] > r->max2)
326 if(ss[2] < 0x80 || ss[2] > 0xBF)
328 if(ss[3] < 0x80 || ss[3] > 0xBF)
343 /** @defgroup utf32iterator UTF-32 string iterators */
346 struct utf32_iterator_data {
347 /** @brief Start of string */
350 /** @brief Length of string */
353 /** @brief Current position */
356 /** @brief Last two non-ignorable characters or (uint32_t)-1
358 * last[1] is the non-Extend/Format character just before position @p n;
359 * last[0] is the one just before that.
361 * Exception 1: if there is no such non-Extend/Format character then an
362 * Extend/Format character is accepted instead.
364 * Exception 2: if there is no such character even taking that into account
365 * the value is (uint32_t)-1.
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
375 utf32_iterator utf32_iterator_new(const uint32_t *s, size_t ns) {
376 utf32_iterator it = xmalloc(sizeof *it);
380 it->last[0] = it->last[1] = -1;
384 /** @brief Initialize an internal private iterator
386 * @param s Start of string
387 * @param ns Length of string
388 * @param n Absolute position
390 static void utf32__iterator_init(utf32_iterator it,
391 const uint32_t *s, size_t ns, size_t n) {
395 it->last[0] = it->last[1] = -1;
396 utf32_iterator_set(it, n);
399 /** @brief Destroy an iterator
402 void utf32_iterator_destroy(utf32_iterator it) {
406 /** @brief Find the current position of an interator
409 size_t utf32_iterator_where(utf32_iterator it) {
413 /** @brief Set an iterator's absolute position
415 * @param n Absolute position
416 * @return 0 on success, non-0 on error
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.
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.
427 int utf32_iterator_set(utf32_iterator it, size_t n) {
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).
436 if(n > it->ns) /* range check */
438 /* Walk backwards skipping ignorable code points */
440 while(m > 0 && (utf32__boundary_ignorable(utf32__word_break(it->s[m-1]))))
442 /* Either m=0 or s[m-1] is not ignorable */
445 /* s[m] is our first non-ignorable code; look for a second in the same
447 while(m > 0 && (utf32__boundary_ignorable(utf32__word_break(it->s[m-1]))))
449 /* Either m=0 or s[m-1] is not ignorable */
453 it->last[0] = it->last[1] = -1;
455 return utf32_iterator_advance(it, n - m);
458 /** @brief Advance an iterator
460 * @param count Number of code points to advance by
461 * @return 0 on success, non-0 on error
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.
467 * This function has O(n) time complexity: it works by advancing naively
468 * forwards through the string.
470 int utf32_iterator_advance(utf32_iterator it, size_t count) {
471 if(count <= it->ns - it->n) {
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];
488 /** @brief Find the current code point
490 * @return Current code point or 0
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!
496 uint32_t utf32_iterator_code(utf32_iterator it) {
503 /** @brief Test for a grapheme boundary
505 * @return Non-0 if pointing just after a grapheme boundary, otherwise 0
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).
512 int utf32_iterator_grapheme_boundary(utf32_iterator it) {
513 uint32_t before, after;
514 enum unicode_Grapheme_Break gbbefore, gbafter;
516 if(it->n == 0 || it->n == it->ns)
518 /* Now we know that s[n-1] and s[n] are safe to inspect */
520 before = it->s[it->n-1];
521 after = it->s[it->n];
522 if(before == 0x000D && after == 0x000A)
524 gbbefore = utf32__grapheme_break(before);
525 gbafter = utf32__grapheme_break(after);
527 if(gbbefore == unicode_Grapheme_Break_Control
532 if(gbafter == unicode_Grapheme_Break_Control
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))
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))
550 if((gbbefore == unicode_Grapheme_Break_LVT
551 || gbbefore == unicode_Grapheme_Break_T)
552 && gbafter == unicode_Grapheme_Break_T)
555 if(gbafter == unicode_Grapheme_Break_Extend)
562 /** @brief Test for a word boundary
564 * @return Non-0 if pointing just after a word boundary, otherwise 0
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.
571 int utf32_iterator_word_boundary(utf32_iterator it) {
572 enum unicode_Word_Break twobefore, before, after, twoafter;
576 if(it->n == 0 || it->n == it->ns)
579 if(it->s[it->n-1] == 0x000D && it->s[it->n] == 0x000A)
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])))
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 */
594 && utf32__boundary_ignorable(utf32__word_break(it->s[nn])))
597 /* It's possible now that nn=ns */
599 twoafter = utf32__word_break(it->s[nn]);
601 twoafter = unicode_Word_Break_Other;
603 /* We've already recorded the non-ignorable code points before the proposed
605 before = utf32__word_break(it->last[1]);
606 twobefore = utf32__word_break(it->last[0]);
609 if(before == unicode_Word_Break_ALetter
610 && after == unicode_Word_Break_ALetter)
613 if(before == unicode_Word_Break_ALetter
614 && after == unicode_Word_Break_MidLetter
615 && twoafter == unicode_Word_Break_ALetter)
618 if(twobefore == unicode_Word_Break_ALetter
619 && before == unicode_Word_Break_MidLetter
620 && after == unicode_Word_Break_ALetter)
623 if(before == unicode_Word_Break_Numeric
624 && after == unicode_Word_Break_Numeric)
627 if(before == unicode_Word_Break_ALetter
628 && after == unicode_Word_Break_Numeric)
631 if(before == unicode_Word_Break_Numeric
632 && after == unicode_Word_Break_ALetter)
635 if(twobefore == unicode_Word_Break_Numeric
636 && before == unicode_Word_Break_MidNum
637 && after == unicode_Word_Break_Numeric)
640 if(before == unicode_Word_Break_Numeric
641 && after == unicode_Word_Break_MidNum
642 && twoafter == unicode_Word_Break_Numeric)
645 if(before == unicode_Word_Break_Katakana
646 && after == unicode_Word_Break_Katakana)
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)
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))
666 /** @defgroup utf32 Functions that operate on UTF-32 strings */
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)
673 * Unlike the conversion functions no validity checking is done on the string.
675 size_t utf32_len(const uint32_t *s) {
676 const uint32_t *t = s;
680 return (size_t)(t - s);
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
688 static void utf32__sort_ccc(uint32_t *s, size_t ns, uint32_t *buffer) {
689 uint32_t *a, *b, *bp;
693 case 1: /* 1-element array is always sorted */
695 case 2: /* 2-element arrays are trivial to sort */
696 if(utf32__combining_class(s[0]) > utf32__combining_class(s[1])) {
703 /* Partition the array */
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 */
713 while(na > 0 && nb > 0) {
714 /* We want ascending order of combining class (hence <)
715 * and we want stability within combining classes (hence <=)
717 if(utf32__combining_class(*a) <= utf32__combining_class(*b)) {
733 memcpy(s, buffer, ns * sizeof(uint32_t));
738 /** @brief Put combining characters into canonical order
739 * @param s Pointer to UTF-32 string
740 * @param ns Length of @p s
741 * @return 0 on success, non-0 on error
743 * @p s is modified in-place. See Unicode 5.0 s3.11 for details of the
746 * Currently we only support a maximum of 1024 combining characters after each
747 * base character. If this limit is exceeded then a non-0 value is returned.
749 static int utf32__canonical_ordering(uint32_t *s, size_t ns) {
751 uint32_t buffer[1024];
753 /* The ordering amounts to a stable sort of each contiguous group of
754 * characters with non-0 combining class. */
756 /* Skip non-combining characters */
757 if(utf32__combining_class(*s) == 0) {
762 /* We must now have at least one combining character; see how many
764 for(nc = 1; nc < ns && utf32__combining_class(s[nc]) != 0; ++nc)
769 utf32__sort_ccc(s, nc, buffer);
776 /* Magic numbers from UAX #15 s16 */
784 #define NCount (VCount * TCount)
785 #define SCount (LCount * NCount)
787 /** @brief Guts of the decomposition lookup functions */
788 #define utf32__decompose_one_generic(WHICH) do { \
789 const uint32_t *dc = utf32__decomposition_##WHICH(c); \
791 /* Found a canonical decomposition in the table */ \
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); \
803 dynstr_ucs4_append(d, T); \
805 /* Equal to own canonical decomposition */ \
806 dynstr_ucs4_append(d, c); \
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!)
812 * @return 0 on success, non-0 on error
814 static void utf32__decompose_one_canon(struct dynstr_ucs4 *d, uint32_t c) {
815 utf32__decompose_one_generic(canon);
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!)
821 * @return 0 on success, non-0 on error
823 static void utf32__decompose_one_compat(struct dynstr_ucs4 *d, uint32_t c) {
824 utf32__decompose_one_generic(compat);
827 /** @brief Magic utf32__compositions() return value for Hangul Choseong */
828 static const uint32_t utf32__hangul_L[1];
830 /** @brief Return the list of compositions that @p c starts
831 * @param c Starter code point
832 * @return Composition list or NULL
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.
839 static const uint32_t *utf32__compositions(uint32_t c) {
840 const uint32_t *compositions = utf32__unidata(c)->composed;
844 /* Special-casing for Hangul */
845 switch(utf32__grapheme_break(c)) {
848 case unicode_Grapheme_Break_L:
849 return utf32__hangul_L;
853 /** @brief Composition step
854 * @param s Start of string
855 * @param ns Length of string
856 * @return New length of string
858 * This is called from utf32__decompose_generic() to compose the result string
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;
866 uint32_t starter = *s++;
867 int block_starters = 0;
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
875 || utf32__combining_class(starter)
876 || !(compositions = utf32__compositions(starter))) {
880 if(compositions != utf32__hangul_L) {
881 /* Where we'll put the eventual starter */
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)
893 /* Found a composition: cc decomposes to starter,*s */
895 compositions = utf32__compositions(starter);
899 /* No composition found. */
900 const int class = utf32__combining_class(*s);
902 /* Transfer the uncomposable combining character to the output */
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
910 while(ns > 0 && utf32__combining_class(*s) == class) {
914 /* Block any subsequent starters */
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
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)) {
932 /* Store the resulting starter */
935 /* Special-casing for Hangul
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
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;
949 && utf32__grapheme_break(*s) == unicode_Grapheme_Break_T) {
950 /* We have an L V T sequence */
951 const uint32_t T = *s++;
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 */
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. */
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
973 #define utf32__decompose_generic(WHICH, COMPOSE) do { \
974 struct dynstr_ucs4 d; \
977 dynstr_ucs4_init(&d); \
980 if((c >= 0xD800 && c <= 0xDFFF) || c > 0x10FFFF) \
982 utf32__decompose_one_##WHICH(&d, c); \
985 if(utf32__canonical_ordering(d.vec, d.nvec)) \
988 d.nvec = utf32__compose(d.vec, d.nvec); \
989 dynstr_ucs4_terminate(&d); \
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
1002 * @return Pointer to result string, or NULL on error
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.
1008 * Returns NULL if the string is not valid for either of the following reasons:
1009 * - it codes for a UTF-16 surrogate
1010 * - it codes for a value outside the unicode code space
1013 * - utf32_decompose_compat()
1014 * - utf32_compose_canon()
1016 uint32_t *utf32_decompose_canon(const uint32_t *s, size_t ns, size_t *ndp) {
1017 utf32__decompose_generic(canon, 0);
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
1024 * @return Pointer to result string, or NULL on error
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.
1030 * Returns NULL if the string is not valid for either of the following reasons:
1031 * - it codes for a UTF-16 surrogate
1032 * - it codes for a value outside the unicode code space
1035 * - utf32_decompose_canon()
1036 * - utf32_compose_compat()
1038 uint32_t *utf32_decompose_compat(const uint32_t *s, size_t ns, size_t *ndp) {
1039 utf32__decompose_generic(compat, 0);
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
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.
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
1057 * - utf32_compose_compat()
1058 * - utf32_decompose_canon()
1060 uint32_t *utf32_compose_canon(const uint32_t *s, size_t ns, size_t *ndp) {
1061 utf32__decompose_generic(canon, 1);
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
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
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
1080 * - utf32_compose_canon()
1081 * - utf32_decompose_compat()
1083 uint32_t *utf32_compose_compat(const uint32_t *s, size_t ns, size_t *ndp) {
1084 utf32__decompose_generic(compat, 1);
1087 /** @brief Single-character case-fold and decompose operation */
1088 #define utf32__casefold_one(WHICH) do { \
1089 const uint32_t *cf = utf32__unidata(c)->casefold; \
1091 /* Found a case-fold mapping in the table */ \
1093 utf32__decompose_one_##WHICH(&d, *cf++); \
1095 utf32__decompose_one_##WHICH(&d, c); \
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
1102 * @return Pointer to result string, or NULL on error
1104 * Case-fold the string at @p s according to full default case-folding rules
1105 * (s3.13) for caseless matching. The result will be in NFD.
1107 * Returns NULL if the string is not valid for either of the following reasons:
1108 * - it codes for a UTF-16 surrogate
1109 * - it codes for a value outside the unicode code space
1111 uint32_t *utf32_casefold_canon(const uint32_t *s, size_t ns, size_t *ndp) {
1112 struct dynstr_ucs4 d;
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. */
1122 for(n = 0; n < ns; ++n)
1123 if(utf32__unidata(s[n])->flags & unicode_normalize_before_casefold)
1126 /* We need a preliminary decomposition */
1127 if(!(ss = utf32_decompose_canon(s, ns, &ns)))
1131 dynstr_ucs4_init(&d);
1134 if((c >= 0xD800 && c <= 0xDFFF) || c > 0x10FFFF)
1136 utf32__casefold_one(canon);
1139 if(utf32__canonical_ordering(d.vec, d.nvec))
1141 dynstr_ucs4_terminate(&d);
1151 /** @brief Compatibility case-fold @p [s,s+ns)
1152 * @param s Pointer to string
1153 * @param ns Length of string
1154 * @param ndp Where to store length of result
1155 * @return Pointer to result string, or NULL on error
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.
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
1164 uint32_t *utf32_casefold_compat(const uint32_t *s, size_t ns, size_t *ndp) {
1165 struct dynstr_ucs4 d;
1170 for(n = 0; n < ns; ++n)
1171 if(utf32__unidata(s[n])->flags & unicode_normalize_before_casefold)
1174 /* We need a preliminary _canonical_ decomposition */
1175 if(!(ss = utf32_decompose_canon(s, ns, &ns)))
1179 /* This computes NFKD(toCaseFold(s)) */
1180 #define compat_casefold_middle() do { \
1181 dynstr_ucs4_init(&d); \
1184 if((c >= 0xD800 && c <= 0xDFFF) || c > 0x10FFFF) \
1186 utf32__casefold_one(compat); \
1189 if(utf32__canonical_ordering(d.vec, d.nvec)) \
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 */
1198 /* Do the outer (NFKD o toCaseFold) */
1199 compat_casefold_middle();
1201 dynstr_ucs4_terminate(&d);
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
1216 * "Comparable to strcmp() at its best."
1218 int utf32_cmp(const uint32_t *a, const uint32_t *b) {
1219 while(*a && *b && *a == *b) {
1223 return *a < *b ? -1 : (*a > *b ? 1 : 0);
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
1232 * This function identifies default grapheme cluster boundaries as described in
1233 * UAX #29 s3. It returns non-0 if @p n points at the code point just after a
1234 * grapheme cluster boundary (including the hypothetical code point just after
1235 * the end of the string).
1237 * This function uses utf32_iterator_set() internally; see that function for
1238 * remarks on performance.
1240 int utf32_is_grapheme_boundary(const uint32_t *s, size_t ns, size_t n) {
1241 struct utf32_iterator_data it[1];
1243 utf32__iterator_init(it, s, ns, n);
1244 return utf32_iterator_grapheme_boundary(it);
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
1253 * This function identifies default word boundaries as described in UAX #29 s4.
1254 * It returns non-0 if @p n points at the code point just after a word boundary
1255 * (including the hypothetical code point just after the end of the string).
1257 * This function uses utf32_iterator_set() internally; see that function for
1258 * remarks on performance.
1260 int utf32_is_word_boundary(const uint32_t *s, size_t ns, size_t n) {
1261 struct utf32_iterator_data it[1];
1263 utf32__iterator_init(it, s, ns, n);
1264 return utf32_iterator_word_boundary(it);
1268 /** @defgroup utf8 Functions that operate on UTF-8 strings */
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; \
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); \
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
1290 * @return Pointer to result string, or NULL on error
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.
1297 * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
1300 * See also utf32_decompose_canon().
1302 char *utf8_decompose_canon(const char *s, size_t ns, size_t *ndp) {
1303 utf8__transform(utf32_decompose_canon);
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
1310 * @return Pointer to result string, or NULL on error
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.
1317 * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
1320 * See also utf32_decompose_compat().
1322 char *utf8_decompose_compat(const char *s, size_t ns, size_t *ndp) {
1323 utf8__transform(utf32_decompose_compat);
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
1330 * @return Pointer to result string, or NULL on error
1332 * Case-fold the string at @p s according to full default case-folding rules
1333 * (s3.13). The result will be in NFD.
1335 * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
1338 char *utf8_casefold_canon(const char *s, size_t ns, size_t *ndp) {
1339 utf8__transform(utf32_casefold_canon);
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
1346 * @return Pointer to result string, or NULL on error
1348 * Case-fold the string at @p s according to full default case-folding rules
1349 * (s3.13). The result will be in NFKD.
1351 * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
1354 char *utf8_casefold_compat(const char *s, size_t ns, size_t *ndp) {
1355 utf8__transform(utf32_casefold_compat);
1365 indent-tabs-mode:nil