2 * This file is part of DisOrder
3 * Copyright (C) 2007, 2009 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 3 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU 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, see <http://www.gnu.org/licenses/>.
18 /** @file lib/unicode.c
19 * @brief Unicode support functions
21 * Here by UTF-8 and UTF-8 we mean the encoding forms of those names (not the
22 * encoding schemes). The primary encoding form is UTF-32 but convenience
23 * wrappers using UTF-8 are provided for a number of functions.
25 * The idea is that all the strings that hit the database will be in a
26 * particular normalization form, and for the search and tags database
27 * in case-folded form, so they can be naively compared within the
30 * As the code stands this guarantee is not well met!
35 * - @ref utf32iterator
47 /** @defgroup utf32props Unicode Code Point Properties */
50 static const struct unidata
*utf32__unidata_hard(uint32_t c
);
52 /** @brief Find definition of code point @p c
54 * @return Pointer to @ref unidata structure for @p c
56 * @p c can be any 32-bit value, a sensible value will be returned regardless.
57 * The returned pointer is NOT guaranteed to be unique to @p c.
59 static inline const struct unidata
*utf32__unidata(uint32_t c
) {
60 /* The bottom half of the table contains almost everything of interest
61 * and we can just return the right thing straight away */
62 if(c
< UNICODE_BREAK_START
)
63 return &unidata
[c
/ UNICODE_MODULUS
][c
% UNICODE_MODULUS
];
65 return utf32__unidata_hard(c
);
68 /** @brief Find definition of code point @p c
70 * @return Pointer to @ref unidata structure for @p c
72 * @p c can be any 32-bit value, a sensible value will be returned regardless.
73 * The returned pointer is NOT guaranteed to be unique to @p c.
75 * Don't use this function (although it will work fine) - use utf32__unidata()
78 static const struct unidata
*utf32__unidata_hard(uint32_t c
) {
79 if(c
< UNICODE_BREAK_START
)
80 return &unidata
[c
/ UNICODE_MODULUS
][c
% UNICODE_MODULUS
];
81 /* Within the break everything is unassigned */
82 if(c
< UNICODE_BREAK_END
)
83 return utf32__unidata(0xFFFF); /* guaranteed to be Cn */
84 /* Planes 15 and 16 are (mostly) private use */
85 if((c
>= 0xF0000 && c
<= 0xFFFFD)
86 || (c
>= 0x100000 && c
<= 0x10FFFD))
87 return utf32__unidata(0xE000); /* first Co code point */
88 /* Everything else above the break top is unassigned */
89 if(c
>= UNICODE_BREAK_TOP
)
90 return utf32__unidata(0xFFFF); /* guaranteed to be Cn */
91 /* Currently the rest is language tags and variation selectors */
92 c
-= (UNICODE_BREAK_END
- UNICODE_BREAK_START
);
93 return &unidata
[c
/ UNICODE_MODULUS
][c
% UNICODE_MODULUS
];
96 /** @brief Return the combining class of @p c
98 * @return Combining class of @p c
100 * @p c can be any 32-bit value, a sensible value will be returned regardless.
102 static inline int utf32__combining_class(uint32_t c
) {
103 return utf32__unidata(c
)->ccc
;
106 /** @brief Return the combining class of @p c
107 * @param c Code point
108 * @return Combining class of @p c
110 * @p c can be any 32-bit value, a sensible value will be returned regardless.
112 int utf32_combining_class(uint32_t c
) {
113 return utf32__combining_class(c
);
116 /** @brief Return the General_Category value for @p c
117 * @param c Code point
118 * @return General_Category property value
120 * @p c can be any 32-bit value, a sensible value will be returned regardless.
122 static inline enum unicode_General_Category
utf32__general_category(uint32_t c
) {
123 return utf32__unidata(c
)->general_category
;
126 /** @brief Determine Grapheme_Break property
127 * @param c Code point
128 * @return Grapheme_Break property value of @p c
130 * @p c can be any 32-bit value, a sensible value will be returned regardless.
132 static inline enum unicode_Grapheme_Break
utf32__grapheme_break(uint32_t c
) {
133 return utf32__unidata(c
)->grapheme_break
;
136 /** @brief Determine Word_Break property
137 * @param c Code point
138 * @return Word_Break property value of @p c
140 * @p c can be any 32-bit value, a sensible value will be returned regardless.
142 static inline enum unicode_Word_Break
utf32__word_break(uint32_t c
) {
143 return utf32__unidata(c
)->word_break
;
146 /** @brief Determine Sentence_Break property
147 * @param c Code point
148 * @return Word_Break property value of @p c
150 * @p c can be any 32-bit value, a sensible value will be returned regardless.
152 static inline enum unicode_Sentence_Break
utf32__sentence_break(uint32_t c
) {
153 return utf32__unidata(c
)->sentence_break
;
156 /** @brief Return true if @p c is ignorable for boundary specifications
157 * @param wb Word break property value
158 * @return non-0 if @p wb is unicode_Word_Break_Extend or unicode_Word_Break_Format
160 static inline int utf32__boundary_ignorable(enum unicode_Word_Break wb
) {
161 return (wb
== unicode_Word_Break_Extend
162 || wb
== unicode_Word_Break_Format
);
165 /** @brief Return the canonical decomposition of @p c
166 * @param c Code point
167 * @return 0-terminated canonical decomposition, or 0
169 static inline const uint32_t *utf32__decomposition_canon(uint32_t c
) {
170 const struct unidata
*const data
= utf32__unidata(c
);
171 const uint32_t *const decomp
= data
->decomp
;
173 if(decomp
&& !(data
->flags
& unicode_compatibility_decomposition
))
179 /** @brief Return the compatibility decomposition of @p c
180 * @param c Code point
181 * @return 0-terminated decomposition, or 0
183 static inline const uint32_t *utf32__decomposition_compat(uint32_t c
) {
184 return utf32__unidata(c
)->decomp
;
188 /** @defgroup utftransform Functions that transform between different Unicode encoding forms */
191 /** @brief Convert UTF-32 to UTF-8
192 * @param s Source string
193 * @param ns Length of source string in code points
194 * @param ndp Where to store length of destination string (or NULL)
195 * @return Newly allocated destination string or NULL on error
197 * If the UTF-32 is not valid then NULL is returned. A UTF-32 code point is
199 * - it codes for a UTF-16 surrogate
200 * - it codes for a value outside the unicode code space
202 * The return value is always 0-terminated. The value returned via @p *ndp
203 * does not include the terminator.
205 char *utf32_to_utf8(const uint32_t *s
, size_t ns
, size_t *ndp
) {
213 dynstr_append(&d
, c
);
214 else if(c
< 0x0800) {
215 dynstr_append(&d
, 0xC0 | (c
>> 6));
216 dynstr_append(&d
, 0x80 | (c
& 0x3F));
217 } else if(c
< 0x10000) {
218 if(c
>= 0xD800 && c
<= 0xDFFF)
220 dynstr_append(&d
, 0xE0 | (c
>> 12));
221 dynstr_append(&d
, 0x80 | ((c
>> 6) & 0x3F));
222 dynstr_append(&d
, 0x80 | (c
& 0x3F));
223 } else if(c
< 0x110000) {
224 dynstr_append(&d
, 0xF0 | (c
>> 18));
225 dynstr_append(&d
, 0x80 | ((c
>> 12) & 0x3F));
226 dynstr_append(&d
, 0x80 | ((c
>> 6) & 0x3F));
227 dynstr_append(&d
, 0x80 | (c
& 0x3F));
232 dynstr_terminate(&d
);
241 /** @brief Convert UTF-8 to UTF-32
242 * @param s Source string
243 * @param ns Length of source string in code points
244 * @param ndp Where to store length of destination string (or NULL)
245 * @return Newly allocated destination string or NULL on error
247 * The return value is always 0-terminated. The value returned via @p *ndp
248 * does not include the terminator.
250 * If the UTF-8 is not valid then NULL is returned. A UTF-8 sequence
251 * for a code point is invalid if:
252 * - it is not the shortest possible sequence for the code point
253 * - it codes for a UTF-16 surrogate
254 * - it codes for a value outside the unicode code space
256 uint32_t *utf8_to_utf32(const char *s
, size_t ns
, size_t *ndp
) {
257 struct dynstr_ucs4 d
;
259 const uint8_t *ss
= (const uint8_t *)s
;
262 dynstr_ucs4_init(&d
);
264 const struct unicode_utf8_row
*const r
= &unicode_utf8_valid
[*ss
];
271 if(ss
[1] < r
->min2
|| ss
[1] > r
->max2
)
276 if(ss
[1] < r
->min2
|| ss
[1] > r
->max2
)
281 if(ss
[1] < r
->min2
|| ss
[1] > r
->max2
)
290 for(n
= 1; n
< r
->count
; ++n
) {
291 if(ss
[n
] < 0x80 || ss
[n
] > 0xBF)
293 c32
= (c32
<< 6) | (ss
[n
] & 0x3F);
295 dynstr_ucs4_append(&d
, c32
);
299 dynstr_ucs4_terminate(&d
);
308 /** @brief Test whether [s,s+ns) is valid UTF-8
309 * @param s Start of string
310 * @param ns Length of string
311 * @return non-0 if @p s is valid UTF-8, 0 if it is not valid
313 * This function is intended to be much faster than calling utf8_to_utf32() and
314 * throwing away the result.
316 int utf8_valid(const char *s
, size_t ns
) {
317 const uint8_t *ss
= (const uint8_t *)s
;
319 const struct unicode_utf8_row
*const r
= &unicode_utf8_valid
[*ss
];
325 if(ss
[1] < r
->min2
|| ss
[1] > r
->max2
)
329 if(ss
[1] < r
->min2
|| ss
[1] > r
->max2
)
331 if(ss
[2] < 0x80 || ss
[2] > 0xBF)
335 if(ss
[1] < r
->min2
|| ss
[1] > r
->max2
)
337 if(ss
[2] < 0x80 || ss
[2] > 0xBF)
339 if(ss
[3] < 0x80 || ss
[3] > 0xBF)
354 /** @defgroup utf32iterator UTF-32 string iterators */
357 struct utf32_iterator_data
{
358 /** @brief Start of string */
361 /** @brief Length of string */
364 /** @brief Current position */
367 /** @brief Last two non-ignorable characters or (uint32_t)-1
369 * last[1] is the non-Extend/Format character just before position @p n;
370 * last[0] is the one just before that.
372 * Exception 1: if there is no such non-Extend/Format character then an
373 * Extend/Format character is accepted instead.
375 * Exception 2: if there is no such character even taking that into account
376 * the value is (uint32_t)-1.
380 /** @brief Tailoring for Word_Break */
381 unicode_property_tailor
*word_break
;
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;
397 utf32_iterator_set(it
, n
);
400 /** @brief Create a new iterator pointing at the start of a string
401 * @param s Start of string
402 * @param ns Length of string
403 * @return New iterator
405 utf32_iterator
utf32_iterator_new(const uint32_t *s
, size_t ns
) {
406 utf32_iterator it
= xmalloc(sizeof *it
);
407 utf32__iterator_init(it
, s
, ns
, 0);
411 /** @brief Tailor this iterator's interpretation of the Word_Break property.
413 * @param pt Property tailor function or NULL
415 * After calling this the iterator will call @p pt to determine the Word_Break
416 * property of each code point. If it returns -1 the default value will be
417 * used otherwise the returned value will be used.
419 * @p pt can be NULL to revert to the default value of the property.
421 * It is safe to call this function at any time; the iterator's internal state
422 * will be reset to suit the new tailoring.
424 void utf32_iterator_tailor_word_break(utf32_iterator it
,
425 unicode_property_tailor
*pt
) {
427 utf32_iterator_set(it
, it
->n
);
430 static inline enum unicode_Word_Break
utf32__iterator_word_break(utf32_iterator it
,
433 return utf32__word_break(c
);
435 const int t
= it
->word_break(c
);
438 return utf32__word_break(c
);
444 /** @brief Destroy an iterator
447 void utf32_iterator_destroy(utf32_iterator it
) {
451 /** @brief Find the current position of an interator
454 size_t utf32_iterator_where(utf32_iterator it
) {
458 /** @brief Set an iterator's absolute position
460 * @param n Absolute position
461 * @return 0 on success, non-0 on error
463 * It is an error to position the iterator outside the string (but acceptable
464 * to point it at the hypothetical post-final character). If an invalid value
465 * of @p n is specified then the iterator is not changed.
467 * This function works by backing up and then advancing to reconstruct the
468 * iterator's internal state for position @p n. The worst case will be O(n)
469 * time complexity (with a worse constant factor that utf32_iterator_advance())
470 * but the typical case is essentially constant-time.
472 int utf32_iterator_set(utf32_iterator it
, size_t n
) {
473 /* We can't just jump to position @p n; the @p last[] values will be wrong.
474 * What we need is to jump a bit behind @p n and then advance forward,
475 * updating @p last[] along the way. How far back? We need to cross two
476 * non-ignorable code points as we advance forwards, so we'd better pass two
477 * such characters on the way back (if such are available).
481 if(n
> it
->ns
) /* range check */
483 /* Walk backwards skipping ignorable code points */
486 && (utf32__boundary_ignorable(utf32__iterator_word_break(it
,
489 /* Either m=0 or s[m-1] is not ignorable */
492 /* s[m] is our first non-ignorable code; look for a second in the same
495 && (utf32__boundary_ignorable(utf32__iterator_word_break(it
,
498 /* Either m=0 or s[m-1] is not ignorable */
502 it
->last
[0] = it
->last
[1] = -1;
504 return utf32_iterator_advance(it
, n
- m
);
507 /** @brief Advance an iterator
509 * @param count Number of code points to advance by
510 * @return 0 on success, non-0 on error
512 * It is an error to advance an iterator beyond the hypothetical post-final
513 * character of the string. If an invalid value of @p n is specified then the
514 * iterator is not changed.
516 * This function has O(n) time complexity: it works by advancing naively
517 * forwards through the string.
519 int utf32_iterator_advance(utf32_iterator it
, size_t count
) {
520 if(count
<= it
->ns
- it
->n
) {
522 const uint32_t c
= it
->s
[it
->n
];
523 const enum unicode_Word_Break wb
= utf32__iterator_word_break(it
, c
);
524 if(it
->last
[1] == (uint32_t)-1
525 || !utf32__boundary_ignorable(wb
)) {
526 it
->last
[0] = it
->last
[1];
537 /** @brief Find the current code point
539 * @return Current code point or 0
541 * If the iterator points at the hypothetical post-final character of the
542 * string then 0 is returned. NB that this doesn't mean that there aren't any
543 * 0 code points inside the string!
545 uint32_t utf32_iterator_code(utf32_iterator it
) {
552 /** @brief Test for a grapheme boundary
554 * @return Non-0 if pointing just after a grapheme boundary, otherwise 0
556 * This function identifies default grapheme cluster boundaries as described in
557 * UAX #29 s3. It returns non-0 if @p it points at the code point just after a
558 * grapheme cluster boundary (including the hypothetical code point just after
559 * the end of the string).
561 int utf32_iterator_grapheme_boundary(utf32_iterator it
) {
562 uint32_t before
, after
;
563 enum unicode_Grapheme_Break gbbefore
, gbafter
;
565 if(it
->n
== 0 || it
->n
== it
->ns
)
567 /* Now we know that s[n-1] and s[n] are safe to inspect */
569 before
= it
->s
[it
->n
-1];
570 after
= it
->s
[it
->n
];
571 if(before
== 0x000D && after
== 0x000A)
573 gbbefore
= utf32__grapheme_break(before
);
574 gbafter
= utf32__grapheme_break(after
);
576 if(gbbefore
== unicode_Grapheme_Break_Control
581 if(gbafter
== unicode_Grapheme_Break_Control
586 if(gbbefore
== unicode_Grapheme_Break_L
587 && (gbafter
== unicode_Grapheme_Break_L
588 || gbafter
== unicode_Grapheme_Break_V
589 || gbafter
== unicode_Grapheme_Break_LV
590 || gbafter
== unicode_Grapheme_Break_LVT
))
593 if((gbbefore
== unicode_Grapheme_Break_LV
594 || gbbefore
== unicode_Grapheme_Break_V
)
595 && (gbafter
== unicode_Grapheme_Break_V
596 || gbafter
== unicode_Grapheme_Break_T
))
599 if((gbbefore
== unicode_Grapheme_Break_LVT
600 || gbbefore
== unicode_Grapheme_Break_T
)
601 && gbafter
== unicode_Grapheme_Break_T
)
604 if(gbafter
== unicode_Grapheme_Break_Extend
)
611 /** @brief Test for a word boundary
613 * @return Non-0 if pointing just after a word boundary, otherwise 0
615 * This function identifies default word boundaries as described in UAX #29 s4.
616 * It returns non-0 if @p it points at the code point just after a word
617 * boundary (including the hypothetical code point just after the end of the
618 * string) and 0 otherwise.
620 int utf32_iterator_word_boundary(utf32_iterator it
) {
621 enum unicode_Word_Break twobefore
, before
, after
, twoafter
;
625 if(it
->n
== 0 || it
->n
== it
->ns
)
628 if(it
->s
[it
->n
-1] == 0x000D && it
->s
[it
->n
] == 0x000A)
631 /* (!Sep) x (Extend|Format) as in UAX #29 s6.2 */
632 if(utf32__sentence_break(it
->s
[it
->n
-1]) != unicode_Sentence_Break_Sep
633 && utf32__boundary_ignorable(utf32__iterator_word_break(it
, it
->s
[it
->n
])))
635 /* Gather the property values we'll need for the rest of the test taking the
636 * s6.2 changes into account */
637 /* First we look at the code points after the proposed boundary */
638 nn
= it
->n
; /* <it->ns */
639 after
= utf32__iterator_word_break(it
, it
->s
[nn
++]);
640 if(!utf32__boundary_ignorable(after
)) {
641 /* X (Extend|Format)* -> X */
643 && utf32__boundary_ignorable(utf32__iterator_word_break(it
,
647 /* It's possible now that nn=ns */
649 twoafter
= utf32__iterator_word_break(it
, it
->s
[nn
]);
651 twoafter
= unicode_Word_Break_Other
;
653 /* We've already recorded the non-ignorable code points before the proposed
655 before
= utf32__iterator_word_break(it
, it
->last
[1]);
656 twobefore
= utf32__iterator_word_break(it
, it
->last
[0]);
659 if(before
== unicode_Word_Break_ALetter
660 && after
== unicode_Word_Break_ALetter
)
663 if(before
== unicode_Word_Break_ALetter
664 && after
== unicode_Word_Break_MidLetter
665 && twoafter
== unicode_Word_Break_ALetter
)
668 if(twobefore
== unicode_Word_Break_ALetter
669 && before
== unicode_Word_Break_MidLetter
670 && after
== unicode_Word_Break_ALetter
)
673 if(before
== unicode_Word_Break_Numeric
674 && after
== unicode_Word_Break_Numeric
)
677 if(before
== unicode_Word_Break_ALetter
678 && after
== unicode_Word_Break_Numeric
)
681 if(before
== unicode_Word_Break_Numeric
682 && after
== unicode_Word_Break_ALetter
)
685 if(twobefore
== unicode_Word_Break_Numeric
686 && before
== unicode_Word_Break_MidNum
687 && after
== unicode_Word_Break_Numeric
)
690 if(before
== unicode_Word_Break_Numeric
691 && after
== unicode_Word_Break_MidNum
692 && twoafter
== unicode_Word_Break_Numeric
)
695 if(before
== unicode_Word_Break_Katakana
696 && after
== unicode_Word_Break_Katakana
)
699 if((before
== unicode_Word_Break_ALetter
700 || before
== unicode_Word_Break_Numeric
701 || before
== unicode_Word_Break_Katakana
702 || before
== unicode_Word_Break_ExtendNumLet
)
703 && after
== unicode_Word_Break_ExtendNumLet
)
706 if(before
== unicode_Word_Break_ExtendNumLet
707 && (after
== unicode_Word_Break_ALetter
708 || after
== unicode_Word_Break_Numeric
709 || after
== unicode_Word_Break_Katakana
))
716 /** @defgroup utf32 Functions that operate on UTF-32 strings */
719 /** @brief Return the length of a 0-terminated UTF-32 string
720 * @param s Pointer to 0-terminated string
721 * @return Length of string in code points (excluding terminator)
723 * Unlike the conversion functions no validity checking is done on the string.
725 size_t utf32_len(const uint32_t *s
) {
726 const uint32_t *t
= s
;
730 return (size_t)(t
- s
);
733 /** @brief Stably sort [s,s+ns) into descending order of combining class
734 * @param s Start of array
735 * @param ns Number of elements, must be at least 1
736 * @param buffer Buffer of at least @p ns elements
738 static void utf32__sort_ccc(uint32_t *s
, size_t ns
, uint32_t *buffer
) {
739 uint32_t *a
, *b
, *bp
;
743 case 1: /* 1-element array is always sorted */
745 case 2: /* 2-element arrays are trivial to sort */
746 if(utf32__combining_class(s
[0]) > utf32__combining_class(s
[1])) {
753 /* Partition the array */
758 /* Sort the two halves of the array */
759 utf32__sort_ccc(a
, na
, buffer
);
760 utf32__sort_ccc(b
, nb
, buffer
);
761 /* Merge them back into one, via the buffer */
763 while(na
> 0 && nb
> 0) {
764 /* We want ascending order of combining class (hence <)
765 * and we want stability within combining classes (hence <=)
767 if(utf32__combining_class(*a
) <= utf32__combining_class(*b
)) {
783 memcpy(s
, buffer
, ns
* sizeof(uint32_t));
788 /** @brief Put combining characters into canonical order
789 * @param s Pointer to UTF-32 string
790 * @param ns Length of @p s
791 * @return 0 on success, non-0 on error
793 * @p s is modified in-place. See Unicode 5.0 s3.11 for details of the
796 * Currently we only support a maximum of 1024 combining characters after each
797 * base character. If this limit is exceeded then a non-0 value is returned.
799 static int utf32__canonical_ordering(uint32_t *s
, size_t ns
) {
801 uint32_t buffer
[1024];
803 /* The ordering amounts to a stable sort of each contiguous group of
804 * characters with non-0 combining class. */
806 /* Skip non-combining characters */
807 if(utf32__combining_class(*s
) == 0) {
812 /* We must now have at least one combining character; see how many
814 for(nc
= 1; nc
< ns
&& utf32__combining_class(s
[nc
]) != 0; ++nc
)
819 utf32__sort_ccc(s
, nc
, buffer
);
826 /* Magic numbers from UAX #15 s16 */
834 #define NCount (VCount * TCount)
835 #define SCount (LCount * NCount)
837 /** @brief Guts of the decomposition lookup functions */
838 #define utf32__decompose_one_generic(WHICH) do { \
839 const uint32_t *dc = utf32__decomposition_##WHICH(c); \
841 /* Found a canonical decomposition in the table */ \
843 utf32__decompose_one_##WHICH(d, *dc++); \
844 } else if(c >= SBase && c < SBase + SCount) { \
845 /* Mechanically decomposable Hangul syllable (UAX #15 s16) */ \
846 const uint32_t SIndex = c - SBase; \
847 const uint32_t L = LBase + SIndex / NCount; \
848 const uint32_t V = VBase + (SIndex % NCount) / TCount; \
849 const uint32_t T = TBase + SIndex % TCount; \
850 dynstr_ucs4_append(d, L); \
851 dynstr_ucs4_append(d, V); \
853 dynstr_ucs4_append(d, T); \
855 /* Equal to own canonical decomposition */ \
856 dynstr_ucs4_append(d, c); \
859 /** @brief Recursively compute the canonical decomposition of @p c
860 * @param d Dynamic string to store decomposition in
861 * @param c Code point to decompose (must be a valid!)
862 * @return 0 on success, non-0 on error
864 static void utf32__decompose_one_canon(struct dynstr_ucs4
*d
, uint32_t c
) {
865 utf32__decompose_one_generic(canon
);
868 /** @brief Recursively compute the compatibility decomposition of @p c
869 * @param d Dynamic string to store decomposition in
870 * @param c Code point to decompose (must be a valid!)
871 * @return 0 on success, non-0 on error
873 static void utf32__decompose_one_compat(struct dynstr_ucs4
*d
, uint32_t c
) {
874 utf32__decompose_one_generic(compat
);
877 /** @brief Magic utf32__compositions() return value for Hangul Choseong */
878 static const uint32_t utf32__hangul_L
[1];
880 /** @brief Return the list of compositions that @p c starts
881 * @param c Starter code point
882 * @return Composition list or NULL
884 * For Hangul leading (Choseong) jamo we return the special value
885 * utf32__hangul_L. These code points are not listed as the targets of
886 * canonical decompositions (make-unidata checks) so there is no confusion with
887 * real decompositions here.
889 static const uint32_t *utf32__compositions(uint32_t c
) {
890 const uint32_t *compositions
= utf32__unidata(c
)->composed
;
894 /* Special-casing for Hangul */
895 switch(utf32__grapheme_break(c
)) {
898 case unicode_Grapheme_Break_L
:
899 return utf32__hangul_L
;
903 /** @brief Composition step
904 * @param s Start of string
905 * @param ns Length of string
906 * @return New length of string
908 * This is called from utf32__decompose_generic() to compose the result string
911 static size_t utf32__compose(uint32_t *s
, size_t ns
) {
912 const uint32_t *compositions
;
913 uint32_t *start
= s
, *t
= s
, *tt
, cc
;
916 uint32_t starter
= *s
++;
917 int block_starters
= 0;
919 /* We don't attempt to compose the following things:
920 * - final characters whatever kind they are
921 * - non-starter characters
922 * - starters that don't take part in a canonical decomposition mapping
925 || utf32__combining_class(starter
)
926 || !(compositions
= utf32__compositions(starter
))) {
930 if(compositions
!= utf32__hangul_L
) {
931 /* Where we'll put the eventual starter */
934 /* See if we can find composition of starter+*s */
935 const uint32_t cchar
= *s
, *cp
= compositions
;
936 while((cc
= *cp
++)) {
937 const uint32_t *decomp
= utf32__decomposition_canon(cc
);
938 /* We know decomp[0] == starter */
939 if(decomp
[1] == cchar
)
943 /* Found a composition: cc decomposes to starter,*s */
945 compositions
= utf32__compositions(starter
);
949 /* No composition found. */
950 const int class = utf32__combining_class(*s
);
952 /* Transfer the uncomposable combining character to the output */
955 /* All the combining characters of the same class of the
956 * uncomposable character are blocked by it, but there may be
957 * others of higher class later. We eat the uncomposable and
958 * blocked characters and go back round the loop for that higher
960 while(ns
> 0 && utf32__combining_class(*s
) == class) {
964 /* Block any subsequent starters */
967 /* The uncombinable character is itself a starter, so we don't
968 * transfer it to the output but instead go back round the main
973 /* Keep going while there are still characters and the starter takes
974 * part in some composition */
975 } while(ns
> 0 && compositions
976 && (!block_starters
|| utf32__combining_class(*s
)));
977 /* Store any remaining combining characters */
978 while(ns
> 0 && utf32__combining_class(*s
)) {
982 /* Store the resulting starter */
985 /* Special-casing for Hangul
987 * If there are combining characters between the L and the V then they
988 * will block the V and so no composition happens. Similarly combining
989 * characters between V and T will block the T and so we only get as far
992 if(utf32__grapheme_break(*s
) == unicode_Grapheme_Break_V
) {
993 const uint32_t V
= *s
++;
994 const uint32_t LIndex
= starter
- LBase
;
995 const uint32_t VIndex
= V
- VBase
;
999 && utf32__grapheme_break(*s
) == unicode_Grapheme_Break_T
) {
1000 /* We have an L V T sequence */
1001 const uint32_t T
= *s
++;
1007 /* Compose to LVT or LV as appropriate */
1008 starter
= (LIndex
* VCount
+ VIndex
) * TCount
+ TIndex
+ SBase
;
1009 } /* else we only have L or LV and no V or T */
1011 /* There could be some combining characters that belong to the V or T.
1012 * These will be treated as non-starter characters at the top of the loop
1013 * and thuss transferred to the output. */
1019 /** @brief Guts of the composition and decomposition functions
1020 * @param WHICH @c canon or @c compat to choose decomposition
1021 * @param COMPOSE @c 0 or @c 1 to compose
1023 #define utf32__decompose_generic(WHICH, COMPOSE) do { \
1024 struct dynstr_ucs4 d; \
1027 dynstr_ucs4_init(&d); \
1030 if((c >= 0xD800 && c <= 0xDFFF) || c > 0x10FFFF) \
1032 utf32__decompose_one_##WHICH(&d, c); \
1035 if(utf32__canonical_ordering(d.vec, d.nvec)) \
1038 d.nvec = utf32__compose(d.vec, d.nvec); \
1039 dynstr_ucs4_terminate(&d); \
1048 /** @brief Canonically decompose @p [s,s+ns)
1049 * @param s Pointer to string
1050 * @param ns Length of string
1051 * @param ndp Where to store length of result
1052 * @return Pointer to result string, or NULL on error
1054 * Computes NFD (Normalization Form D) of the string at @p s. This implies
1055 * performing all canonical decompositions and then normalizing the order of
1056 * combining characters.
1058 * Returns NULL if the string is not valid for either of the following reasons:
1059 * - it codes for a UTF-16 surrogate
1060 * - it codes for a value outside the unicode code space
1063 * - utf32_decompose_compat()
1064 * - utf32_compose_canon()
1066 uint32_t *utf32_decompose_canon(const uint32_t *s
, size_t ns
, size_t *ndp
) {
1067 utf32__decompose_generic(canon
, 0);
1070 /** @brief Compatibility decompose @p [s,s+ns)
1071 * @param s Pointer to string
1072 * @param ns Length of string
1073 * @param ndp Where to store length of result
1074 * @return Pointer to result string, or NULL on error
1076 * Computes NFKD (Normalization Form KD) of the string at @p s. This implies
1077 * performing all canonical and compatibility decompositions and then
1078 * normalizing the order of combining characters.
1080 * Returns NULL if the string is not valid for either of the following reasons:
1081 * - it codes for a UTF-16 surrogate
1082 * - it codes for a value outside the unicode code space
1085 * - utf32_decompose_canon()
1086 * - utf32_compose_compat()
1088 uint32_t *utf32_decompose_compat(const uint32_t *s
, size_t ns
, size_t *ndp
) {
1089 utf32__decompose_generic(compat
, 0);
1092 /** @brief Canonically compose @p [s,s+ns)
1093 * @param s Pointer to string
1094 * @param ns Length of string
1095 * @param ndp Where to store length of result
1096 * @return Pointer to result string, or NULL on error
1098 * Computes NFC (Normalization Form C) of the string at @p s. This implies
1099 * performing all canonical decompositions, normalizing the order of combining
1100 * characters and then composing all unblocked primary compositables.
1102 * Returns NULL if the string is not valid for either of the following reasons:
1103 * - it codes for a UTF-16 surrogate
1104 * - it codes for a value outside the unicode code space
1107 * - utf32_compose_compat()
1108 * - utf32_decompose_canon()
1110 uint32_t *utf32_compose_canon(const uint32_t *s
, size_t ns
, size_t *ndp
) {
1111 utf32__decompose_generic(canon
, 1);
1114 /** @brief Compatibility compose @p [s,s+ns)
1115 * @param s Pointer to string
1116 * @param ns Length of string
1117 * @param ndp Where to store length of result
1118 * @return Pointer to result string, or NULL on error
1120 * Computes NFKC (Normalization Form KC) of the string at @p s. This implies
1121 * performing all canonical and compatibility decompositions, normalizing the
1122 * order of combining characters and then composing all unblocked primary
1125 * Returns NULL if the string is not valid for either of the following reasons:
1126 * - it codes for a UTF-16 surrogate
1127 * - it codes for a value outside the unicode code space
1130 * - utf32_compose_canon()
1131 * - utf32_decompose_compat()
1133 uint32_t *utf32_compose_compat(const uint32_t *s
, size_t ns
, size_t *ndp
) {
1134 utf32__decompose_generic(compat
, 1);
1137 /** @brief Single-character case-fold and decompose operation */
1138 #define utf32__casefold_one(WHICH) do { \
1139 const uint32_t *cf = utf32__unidata(c)->casefold; \
1141 /* Found a case-fold mapping in the table */ \
1143 utf32__decompose_one_##WHICH(&d, *cf++); \
1145 utf32__decompose_one_##WHICH(&d, c); \
1148 /** @brief Case-fold @p [s,s+ns)
1149 * @param s Pointer to string
1150 * @param ns Length of string
1151 * @param ndp Where to store length of result
1152 * @return Pointer to result string, or NULL on error
1154 * Case-fold the string at @p s according to full default case-folding rules
1155 * (s3.13) for caseless matching. The result will be in NFD.
1157 * Returns NULL if the string is not valid for either of the following reasons:
1158 * - it codes for a UTF-16 surrogate
1159 * - it codes for a value outside the unicode code space
1161 uint32_t *utf32_casefold_canon(const uint32_t *s
, size_t ns
, size_t *ndp
) {
1162 struct dynstr_ucs4 d
;
1167 /* If the canonical decomposition of the string includes any combining
1168 * character that case-folds to a non-combining character then we must
1169 * normalize before we fold. In Unicode 5.0.0 this means 0345 COMBINING
1170 * GREEK YPOGEGRAMMENI in its decomposition and the various characters that
1171 * canonically decompose to it. */
1172 for(n
= 0; n
< ns
; ++n
)
1173 if(utf32__unidata(s
[n
])->flags
& unicode_normalize_before_casefold
)
1176 /* We need a preliminary decomposition */
1177 if(!(ss
= utf32_decompose_canon(s
, ns
, &ns
)))
1181 dynstr_ucs4_init(&d
);
1184 if((c
>= 0xD800 && c
<= 0xDFFF) || c
> 0x10FFFF)
1186 utf32__casefold_one(canon
);
1189 if(utf32__canonical_ordering(d
.vec
, d
.nvec
))
1191 dynstr_ucs4_terminate(&d
);
1201 /** @brief Compatibility case-fold @p [s,s+ns)
1202 * @param s Pointer to string
1203 * @param ns Length of string
1204 * @param ndp Where to store length of result
1205 * @return Pointer to result string, or NULL on error
1207 * Case-fold the string at @p s according to full default case-folding rules
1208 * (s3.13) for compatibility caseless matching. The result will be in NFKD.
1210 * Returns NULL if the string is not valid for either of the following reasons:
1211 * - it codes for a UTF-16 surrogate
1212 * - it codes for a value outside the unicode code space
1214 uint32_t *utf32_casefold_compat(const uint32_t *s
, size_t ns
, size_t *ndp
) {
1215 struct dynstr_ucs4 d
;
1220 for(n
= 0; n
< ns
; ++n
)
1221 if(utf32__unidata(s
[n
])->flags
& unicode_normalize_before_casefold
)
1224 /* We need a preliminary _canonical_ decomposition */
1225 if(!(ss
= utf32_decompose_canon(s
, ns
, &ns
)))
1229 /* This computes NFKD(toCaseFold(s)) */
1230 #define compat_casefold_middle() do { \
1231 dynstr_ucs4_init(&d); \
1234 if((c >= 0xD800 && c <= 0xDFFF) || c > 0x10FFFF) \
1236 utf32__casefold_one(compat); \
1239 if(utf32__canonical_ordering(d.vec, d.nvec)) \
1242 /* Do the inner (NFKD o toCaseFold) */
1243 compat_casefold_middle();
1244 /* We can do away with the NFD'd copy of the input now */
1248 /* Do the outer (NFKD o toCaseFold) */
1249 compat_casefold_middle();
1251 dynstr_ucs4_terminate(&d
);
1261 /** @brief Order a pair of UTF-32 strings
1262 * @param a First 0-terminated string
1263 * @param b Second 0-terminated string
1264 * @return -1, 0 or 1 for a less than, equal to or greater than b
1266 * "Comparable to strcmp() at its best."
1268 int utf32_cmp(const uint32_t *a
, const uint32_t *b
) {
1269 while(*a
&& *b
&& *a
== *b
) {
1273 return *a
< *b ?
-1 : (*a
> *b ?
1 : 0);
1276 /** @brief Identify a grapheme cluster boundary
1277 * @param s Start of string (must be NFD)
1278 * @param ns Length of string
1279 * @param n Index within string (in [0,ns].)
1280 * @return 1 at a grapheme cluster boundary, 0 otherwise
1282 * This function identifies default grapheme cluster boundaries as described in
1283 * UAX #29 s3. It returns non-0 if @p n points at the code point just after a
1284 * grapheme cluster boundary (including the hypothetical code point just after
1285 * the end of the string).
1287 * This function uses utf32_iterator_set() internally; see that function for
1288 * remarks on performance.
1290 int utf32_is_grapheme_boundary(const uint32_t *s
, size_t ns
, size_t n
) {
1291 struct utf32_iterator_data it
[1];
1293 utf32__iterator_init(it
, s
, ns
, n
);
1294 return utf32_iterator_grapheme_boundary(it
);
1297 /** @brief Identify a word boundary
1298 * @param s Start of string (must be NFD)
1299 * @param ns Length of string
1300 * @param n Index within string (in [0,ns].)
1301 * @return 1 at a word boundary, 0 otherwise
1303 * This function identifies default word boundaries as described in UAX #29 s4.
1304 * It returns non-0 if @p n points at the code point just after a word boundary
1305 * (including the hypothetical code point just after the end of the string).
1307 * This function uses utf32_iterator_set() internally; see that function for
1308 * remarks on performance.
1310 int utf32_is_word_boundary(const uint32_t *s
, size_t ns
, size_t n
) {
1311 struct utf32_iterator_data it
[1];
1313 utf32__iterator_init(it
, s
, ns
, n
);
1314 return utf32_iterator_word_boundary(it
);
1317 /** @brief Split [s,ns) into multiple words
1318 * @param s Pointer to start of string
1319 * @param ns Length of string
1320 * @param nwp Where to store word count, or NULL
1321 * @param wbreak Word_Break property tailor, or NULL
1322 * @return Pointer to array of pointers to words
1324 * The returned array is terminated by a NULL pointer and individual
1325 * strings are 0-terminated.
1327 uint32_t **utf32_word_split(const uint32_t *s
, size_t ns
, size_t *nwp
,
1328 unicode_property_tailor
*wbreak
) {
1329 struct utf32_iterator_data it
[1];
1330 size_t b1
= 0, b2
= 0 ,i
;
1332 struct vector32 v32
[1];
1336 utf32__iterator_init(it
, s
, ns
, 0);
1337 it
->word_break
= wbreak
;
1338 /* Work our way through the string stopping at each word break. */
1340 if(utf32_iterator_word_boundary(it
)) {
1341 /* We've found a new boundary */
1344 /*fprintf(stderr, "[%zu, %zu) is a candidate word\n", b1, b2);*/
1345 /* Inspect the characters between the boundary and form an opinion as to
1346 * whether they are a word or not */
1348 for(i
= b1
; i
< b2
; ++i
) {
1349 switch(utf32__iterator_word_break(it
, it
->s
[i
])) {
1350 case unicode_Word_Break_ALetter
:
1351 case unicode_Word_Break_Numeric
:
1352 case unicode_Word_Break_Katakana
:
1359 /* If it's a word add it to the list of results */
1361 const size_t len
= b2
- b1
;
1362 w
= xcalloc_noptr(len
+ 1, sizeof(uint32_t));
1363 memcpy(w
, it
->s
+ b1
, len
* sizeof (uint32_t));
1365 vector32_append(v32
, w
);
1368 } while(!utf32_iterator_advance(it
, 1));
1369 vector32_terminate(v32
);
1376 /** @defgroup utf8 Functions that operate on UTF-8 strings */
1379 /** @brief Wrapper to transform a UTF-8 string using the UTF-32 function */
1380 #define utf8__transform(FN) do { \
1381 uint32_t *to32 = 0, *decomp32 = 0; \
1382 size_t nto32, ndecomp32; \
1383 char *decomp8 = 0; \
1385 if(!(to32 = utf8_to_utf32(s, ns, &nto32))) goto error; \
1386 if(!(decomp32 = FN(to32, nto32, &ndecomp32))) goto error; \
1387 decomp8 = utf32_to_utf8(decomp32, ndecomp32, ndp); \
1394 /** @brief Canonically decompose @p [s,s+ns)
1395 * @param s Pointer to string
1396 * @param ns Length of string
1397 * @param ndp Where to store length of result
1398 * @return Pointer to result string, or NULL on error
1400 * Computes NFD (Normalization Form D) of the string at @p s. This implies
1401 * performing all canonical decompositions and then normalizing the order of
1402 * combining characters.
1404 * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
1408 * - utf32_decompose_canon().
1409 * - utf8_decompose_compat()
1410 * - utf8_compose_canon()
1412 char *utf8_decompose_canon(const char *s
, size_t ns
, size_t *ndp
) {
1413 utf8__transform(utf32_decompose_canon
);
1416 /** @brief Compatibility decompose @p [s,s+ns)
1417 * @param s Pointer to string
1418 * @param ns Length of string
1419 * @param ndp Where to store length of result
1420 * @return Pointer to result string, or NULL on error
1422 * Computes NFKD (Normalization Form KD) of the string at @p s. This implies
1423 * performing all canonical and compatibility decompositions and then
1424 * normalizing the order of combining characters.
1426 * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
1430 * - utf32_decompose_compat().
1431 * - utf8_decompose_canon()
1432 * - utf8_compose_compat()
1434 char *utf8_decompose_compat(const char *s
, size_t ns
, size_t *ndp
) {
1435 utf8__transform(utf32_decompose_compat
);
1438 /** @brief Canonically compose @p [s,s+ns)
1439 * @param s Pointer to string
1440 * @param ns Length of string
1441 * @param ndp Where to store length of result
1442 * @return Pointer to result string, or NULL on error
1444 * Computes NFC (Normalization Form C) of the string at @p s. This implies
1445 * performing all canonical decompositions, normalizing the order of combining
1446 * characters and then composing all unblocked primary compositables.
1448 * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
1452 * - utf32_compose_canon()
1453 * - utf8_compose_compat()
1454 * - utf8_decompose_canon()
1456 char *utf8_compose_canon(const char *s
, size_t ns
, size_t *ndp
) {
1457 utf8__transform(utf32_compose_canon
);
1460 /** @brief Compatibility compose @p [s,s+ns)
1461 * @param s Pointer to string
1462 * @param ns Length of string
1463 * @param ndp Where to store length of result
1464 * @return Pointer to result string, or NULL on error
1466 * Computes NFKC (Normalization Form KC) of the string at @p s. This implies
1467 * performing all canonical and compatibility decompositions, normalizing the
1468 * order of combining characters and then composing all unblocked primary
1471 * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
1475 * - utf32_compose_compat()
1476 * - utf8_compose_canon()
1477 * - utf8_decompose_compat()
1479 char *utf8_compose_compat(const char *s
, size_t ns
, size_t *ndp
) {
1480 utf8__transform(utf32_compose_compat
);
1483 /** @brief Case-fold @p [s,s+ns)
1484 * @param s Pointer to string
1485 * @param ns Length of string
1486 * @param ndp Where to store length of result
1487 * @return Pointer to result string, or NULL on error
1489 * Case-fold the string at @p s according to full default case-folding rules
1490 * (s3.13). The result will be in NFD.
1492 * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
1495 char *utf8_casefold_canon(const char *s
, size_t ns
, size_t *ndp
) {
1496 utf8__transform(utf32_casefold_canon
);
1499 /** @brief Compatibility case-fold @p [s,s+ns)
1500 * @param s Pointer to string
1501 * @param ns Length of string
1502 * @param ndp Where to store length of result
1503 * @return Pointer to result string, or NULL on error
1505 * Case-fold the string at @p s according to full default case-folding rules
1506 * (s3.13). The result will be in NFKD.
1508 * Returns NULL if the string is not valid; see utf8_to_utf32() for reasons why
1511 char *utf8_casefold_compat(const char *s
, size_t ns
, size_t *ndp
) {
1512 utf8__transform(utf32_casefold_compat
);
1515 /** @brief Split [s,ns) into multiple words
1516 * @param s Pointer to start of string
1517 * @param ns Length of string
1518 * @param nwp Where to store word count, or NULL
1519 * @param wbreak Word_Break property tailor, or NULL
1520 * @return Pointer to array of pointers to words
1522 * The returned array is terminated by a NULL pointer and individual
1523 * strings are 0-terminated.
1525 char **utf8_word_split(const char *s
, size_t ns
, size_t *nwp
,
1526 unicode_property_tailor
*wbreak
) {
1527 uint32_t *to32
= 0, **v32
= 0;
1528 size_t nto32
, nv
, n
;
1529 char **v8
= 0, **ret
= 0;
1531 if(!(to32
= utf8_to_utf32(s
, ns
, &nto32
))) goto error
;
1532 if(!(v32
= utf32_word_split(to32
, nto32
, &nv
, wbreak
))) goto error
;
1533 v8
= xcalloc(sizeof (char *), nv
+ 1);
1534 for(n
= 0; n
< nv
; ++n
)
1535 if(!(v8
[n
] = utf32_to_utf8(v32
[n
], utf32_len(v32
[n
]), 0)))
1539 v8
= 0; /* don't free */
1542 for(n
= 0; n
< nv
; ++n
)
1547 for(n
= 0; n
< nv
; ++n
)
1563 indent-tabs-mode:nil