3 * Unicode normalization routines
5 * Copyright © 2004 Brion Vibber <brion@pobox.com>
6 * http://www.mediawiki.org/
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License along
19 * with this program; if not, write to the Free Software Foundation, Inc.,
20 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
21 * http://www.gnu.org/copyleft/gpl.html
28 * @defgroup UtfNormal UtfNormal
31 require_once dirname(__FILE__
).'/UtfNormalUtil.php';
34 * For using the ICU wrapper
36 define( 'UNORM_NONE', 1 );
37 define( 'UNORM_NFD', 2 );
38 define( 'UNORM_NFKD', 3 );
39 define( 'UNORM_NFC', 4 );
40 define( 'UNORM_DEFAULT', UNORM_NFC
);
41 define( 'UNORM_NFKC', 5 );
42 define( 'UNORM_FCD', 6 );
44 define( 'NORMALIZE_ICU', function_exists( 'utf8_normalize' ) );
45 define( 'NORMALIZE_INTL', function_exists( 'normalizer_normalize' ) );
48 * Unicode normalization routines for working with UTF-8 strings.
49 * Currently assumes that input strings are valid UTF-8!
51 * Not as fast as I'd like, but should be usable for most purposes.
52 * UtfNormal::toNFC() will bail early if given ASCII text or text
53 * it can quickly deterimine is already normalized.
55 * All functions can be called static.
57 * See description of forms at http://www.unicode.org/reports/tr15/
62 static $utfCombiningClass = null;
63 static $utfCanonicalComp = null;
64 static $utfCanonicalDecomp = null;
66 # Load compatibility decompositions on demand if they are needed.
67 static $utfCompatibilityDecomp = null;
72 * The ultimate convenience function! Clean up invalid UTF-8 sequences,
73 * and convert to normal form C, canonical composition.
75 * Fast return for pure ASCII strings; some lesser optimizations for
76 * strings containing only known-good characters. Not as fast as toNFC().
78 * @param $string String: a UTF-8 string
79 * @return string a clean, shiny, normalized UTF-8 string
81 static function cleanUp( $string ) {
83 $string = self
::replaceForNativeNormalize( $string );
85 # UnicodeString constructor fails if the string ends with a
86 # head byte. Add a junk char at the end, we'll strip it off.
87 return rtrim( utf8_normalize( $string . "\x01", UNORM_NFC
), "\x01" );
88 } elseif( NORMALIZE_INTL
) {
89 $string = self
::replaceForNativeNormalize( $string );
90 $norm = normalizer_normalize( $string, Normalizer
::FORM_C
);
91 if( $norm === null ||
$norm === false ) {
92 # normalizer_normalize will either return false or null
93 # (depending on which doc you read) if invalid utf8 string.
94 # quickIsNFCVerify cleans up invalid sequences.
96 if( UtfNormal
::quickIsNFCVerify( $string ) ) {
97 # if that's true, the string is actually already normal.
100 # Now we are valid but non-normal
101 return normalizer_normalize( $string, Normalizer
::FORM_C
);
106 } elseif( UtfNormal
::quickIsNFCVerify( $string ) ) {
107 # Side effect -- $string has had UTF-8 errors cleaned up.
110 return UtfNormal
::NFC( $string );
115 * Convert a UTF-8 string to normal form C, canonical composition.
116 * Fast return for pure ASCII strings; some lesser optimizations for
117 * strings containing only known-good characters.
119 * @param $string String: a valid UTF-8 string. Input is not validated.
120 * @return string a UTF-8 string in normal form C
122 static function toNFC( $string ) {
124 return normalizer_normalize( $string, Normalizer
::FORM_C
);
125 elseif( NORMALIZE_ICU
)
126 return utf8_normalize( $string, UNORM_NFC
);
127 elseif( UtfNormal
::quickIsNFC( $string ) )
130 return UtfNormal
::NFC( $string );
134 * Convert a UTF-8 string to normal form D, canonical decomposition.
135 * Fast return for pure ASCII strings.
137 * @param $string String: a valid UTF-8 string. Input is not validated.
138 * @return string a UTF-8 string in normal form D
140 static function toNFD( $string ) {
142 return normalizer_normalize( $string, Normalizer
::FORM_D
);
143 elseif( NORMALIZE_ICU
)
144 return utf8_normalize( $string, UNORM_NFD
);
145 elseif( preg_match( '/[\x80-\xff]/', $string ) )
146 return UtfNormal
::NFD( $string );
152 * Convert a UTF-8 string to normal form KC, compatibility composition.
153 * This may cause irreversible information loss, use judiciously.
154 * Fast return for pure ASCII strings.
156 * @param $string String: a valid UTF-8 string. Input is not validated.
157 * @return string a UTF-8 string in normal form KC
159 static function toNFKC( $string ) {
161 return normalizer_normalize( $string, Normalizer
::FORM_KC
);
162 elseif( NORMALIZE_ICU
)
163 return utf8_normalize( $string, UNORM_NFKC
);
164 elseif( preg_match( '/[\x80-\xff]/', $string ) )
165 return UtfNormal
::NFKC( $string );
171 * Convert a UTF-8 string to normal form KD, compatibility decomposition.
172 * This may cause irreversible information loss, use judiciously.
173 * Fast return for pure ASCII strings.
175 * @param $string String: a valid UTF-8 string. Input is not validated.
176 * @return string a UTF-8 string in normal form KD
178 static function toNFKD( $string ) {
180 return normalizer_normalize( $string, Normalizer
::FORM_KD
);
181 elseif( NORMALIZE_ICU
)
182 return utf8_normalize( $string, UNORM_NFKD
);
183 elseif( preg_match( '/[\x80-\xff]/', $string ) )
184 return UtfNormal
::NFKD( $string );
190 * Load the basic composition data if necessary
193 static function loadData() {
194 if( !isset( self
::$utfCombiningClass ) ) {
195 require_once( dirname(__FILE__
) . '/UtfNormalData.inc' );
200 * Returns true if the string is _definitely_ in NFC.
201 * Returns false if not or uncertain.
202 * @param $string String: a valid UTF-8 string. Input is not validated.
205 static function quickIsNFC( $string ) {
206 # ASCII is always valid NFC!
207 # If it's pure ASCII, let it through.
208 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
210 UtfNormal
::loadData();
211 $len = strlen( $string );
212 for( $i = 0; $i < $len; $i++
) {
217 } elseif( $n >= 0xf0 ) {
218 $c = substr( $string, $i, 4 );
220 } elseif( $n >= 0xe0 ) {
221 $c = substr( $string, $i, 3 );
223 } elseif( $n >= 0xc0 ) {
224 $c = substr( $string, $i, 2 );
227 if( isset( self
::$utfCheckNFC[$c] ) ) {
228 # If it's NO or MAYBE, bail and do the slow check.
231 if( isset( self
::$utfCombiningClass[$c] ) ) {
232 # Combining character? We might have to do sorting, at least.
240 * Returns true if the string is _definitely_ in NFC.
241 * Returns false if not or uncertain.
242 * @param $string String: a UTF-8 string, altered on output to be valid UTF-8 safe for XML.
244 static function quickIsNFCVerify( &$string ) {
245 # Screen out some characters that eg won't be allowed in XML
246 $string = preg_replace( '/[\x00-\x08\x0b\x0c\x0e-\x1f]/', UTF8_REPLACEMENT
, $string );
248 # ASCII is always valid NFC!
249 # If we're only ever given plain ASCII, we can avoid the overhead
250 # of initializing the decomposition tables by skipping out early.
251 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
253 static $checkit = null, $tailBytes = null, $utfCheckOrCombining = null;
254 if( !isset( $checkit ) ) {
255 # Load/build some scary lookup tables...
256 UtfNormal
::loadData();
258 $utfCheckOrCombining = array_merge( self
::$utfCheckNFC, self
::$utfCombiningClass );
260 # Head bytes for sequences which we should do further validity checks
261 $checkit = array_flip( array_map( 'chr',
262 array( 0xc0, 0xc1, 0xe0, 0xed, 0xef,
263 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7,
264 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff ) ) );
266 # Each UTF-8 head byte is followed by a certain
267 # number of tail bytes.
268 $tailBytes = array();
269 for( $n = 0; $n < 256; $n++
) {
272 } elseif( $n < 0xe0 ) {
274 } elseif( $n < 0xf0 ) {
276 } elseif( $n < 0xf8 ) {
278 } elseif( $n < 0xfc ) {
280 } elseif( $n < 0xfe ) {
285 $tailBytes[chr($n)] = $remaining;
289 # Chop the text into pure-ASCII and non-ASCII areas;
290 # large ASCII parts can be handled much more quickly.
291 # Don't chop up Unicode areas for punctuation, though,
292 # that wastes energy.
295 '/([\x00-\x7f]+|[\x80-\xff][\x00-\x40\x5b-\x5f\x7b-\xff]*)/',
301 foreach( $matches[1] as $str ) {
302 $chunk = strlen( $str );
304 if( $str[0] < "\x80" ) {
305 # ASCII chunk: guaranteed to be valid UTF-8
306 # and in normal form C, so skip over it.
311 # We'll have to examine the chunk byte by byte to ensure
312 # that it consists of valid UTF-8 sequences, and to see
313 # if any of them might not be normalized.
315 # Since PHP is not the fastest language on earth, some of
316 # this code is a little ugly with inner loop optimizations.
319 $len = $chunk +
1; # Counting down is faster. I'm *so* sorry.
321 for( $i = -1; --$len; ) {
322 $remaining = $tailBytes[$c = $str[++
$i]];
325 $sequence = $head = $c;
327 # Look for the defined number of tail bytes...
328 if( --$len && ( $c = $str[++
$i] ) >= "\x80" && $c < "\xc0" ) {
329 # Legal tail bytes are nice.
333 # Premature end of string!
334 # Drop a replacement character into output to
335 # represent the invalid UTF-8 sequence.
336 $replace[] = array( UTF8_REPLACEMENT
,
337 $base +
$i +
1 - strlen( $sequence ),
338 strlen( $sequence ) );
341 # Illegal tail byte; abandon the sequence.
342 $replace[] = array( UTF8_REPLACEMENT
,
343 $base +
$i - strlen( $sequence ),
344 strlen( $sequence ) );
345 # Back up and reprocess this byte; it may itself
346 # be a legal ASCII or UTF-8 sequence head.
352 } while( --$remaining );
354 if( isset( $checkit[$head] ) ) {
355 # Do some more detailed validity checks, for
356 # invalid characters and illegal sequences.
357 if( $head == "\xed" ) {
358 # 0xed is relatively frequent in Korean, which
359 # abuts the surrogate area, so we're doing
360 # this check separately to speed things up.
362 if( $sequence >= UTF8_SURROGATE_FIRST
) {
363 # Surrogates are legal only in UTF-16 code.
364 # They are totally forbidden here in UTF-8
366 $replace[] = array( UTF8_REPLACEMENT
,
367 $base +
$i +
1 - strlen( $sequence ),
368 strlen( $sequence ) );
373 # Slower, but rarer checks...
376 # "Overlong sequences" are those that are syntactically
377 # correct but use more UTF-8 bytes than are necessary to
378 # encode a character. Naïve string comparisons can be
379 # tricked into failing to see a match for an ASCII
380 # character, for instance, which can be a security hole
381 # if blacklist checks are being used.
382 ($n < 0xc2 && $sequence <= UTF8_OVERLONG_A
)
383 ||
($n == 0xe0 && $sequence <= UTF8_OVERLONG_B
)
384 ||
($n == 0xf0 && $sequence <= UTF8_OVERLONG_C
)
386 # U+FFFE and U+FFFF are explicitly forbidden in Unicode.
388 ($sequence == UTF8_FFFE
)
389 ||
($sequence == UTF8_FFFF
) )
391 # Unicode has been limited to 21 bits; longer
392 # sequences are not allowed.
393 ||
($n >= 0xf0 && $sequence > UTF8_MAX
) ) {
395 $replace[] = array( UTF8_REPLACEMENT
,
396 $base +
$i +
1 - strlen( $sequence ),
397 strlen( $sequence ) );
404 if( isset( $utfCheckOrCombining[$sequence] ) ) {
405 # If it's NO or MAYBE, we'll have to rip
406 # the string apart and put it back together.
407 # That's going to be mighty slow.
408 $looksNormal = false;
411 # The sequence is legal!
413 } elseif( $c < "\x80" ) {
416 } elseif( $c < "\xc0" ) {
420 $replace[] = array( UTF8_REPLACEMENT
, $base +
$i, 1 );
422 # Don't add if we're continuing a broken sequence;
423 # we already put a replacement character when we looked
424 # at the broken sequence.
425 $replace[] = array( '', $base +
$i, 1 );
428 # Miscellaneous freaks.
429 $replace[] = array( UTF8_REPLACEMENT
, $base +
$i, 1 );
435 if( count( $replace ) ) {
436 # There were illegal UTF-8 sequences we need to fix up.
439 foreach( $replace as $rep ) {
440 list( $replacement, $start, $length ) = $rep;
441 if( $last < $start ) {
442 $out .= substr( $string, $last, $start - $last );
444 $out .= $replacement;
445 $last = $start +
$length;
447 if( $last < strlen( $string ) ) {
448 $out .= substr( $string, $last );
455 # These take a string and run the normalization on them, without
456 # checking for validity or any optimization etc. Input must be
459 * @param $string string
463 static function NFC( $string ) {
464 return UtfNormal
::fastCompose( UtfNormal
::NFD( $string ) );
468 * @param $string string
472 static function NFD( $string ) {
473 UtfNormal
::loadData();
475 return UtfNormal
::fastCombiningSort(
476 UtfNormal
::fastDecompose( $string, self
::$utfCanonicalDecomp ) );
480 * @param $string string
484 static function NFKC( $string ) {
485 return UtfNormal
::fastCompose( UtfNormal
::NFKD( $string ) );
489 * @param $string string
493 static function NFKD( $string ) {
494 if( !isset( self
::$utfCompatibilityDecomp ) ) {
495 require_once( 'UtfNormalDataK.inc' );
497 return self
::fastCombiningSort(
498 self
::fastDecompose( $string, self
::$utfCompatibilityDecomp ) );
503 * Perform decomposition of a UTF-8 string into either D or KD form
504 * (depending on which decomposition map is passed to us).
505 * Input is assumed to be *valid* UTF-8. Invalid code will break.
507 * @param $string String: valid UTF-8 string
508 * @param $map Array: hash of expanded decomposition map
509 * @return string a UTF-8 string decomposed, not yet normalized (needs sorting)
511 static function fastDecompose( $string, $map ) {
512 UtfNormal
::loadData();
513 $len = strlen( $string );
515 for( $i = 0; $i < $len; $i++
) {
519 # ASCII chars never decompose
523 } elseif( $n >= 0xf0 ) {
524 $c = substr( $string, $i, 4 );
526 } elseif( $n >= 0xe0 ) {
527 $c = substr( $string, $i, 3 );
529 } elseif( $n >= 0xc0 ) {
530 $c = substr( $string, $i, 2 );
533 if( isset( $map[$c] ) ) {
537 if( $c >= UTF8_HANGUL_FIRST
&& $c <= UTF8_HANGUL_LAST
) {
538 # Decompose a hangul syllable into jamo;
539 # hardcoded for three-byte UTF-8 sequence.
540 # A lookup table would be slightly faster,
541 # but adds a lot of memory & disk needs.
543 $index = ( (ord( $c[0] ) & 0x0f) << 12
544 |
(ord( $c[1] ) & 0x3f) << 6
545 |
(ord( $c[2] ) & 0x3f) )
546 - UNICODE_HANGUL_FIRST
;
547 $l = intval( $index / UNICODE_HANGUL_NCOUNT
);
548 $v = intval( ($index % UNICODE_HANGUL_NCOUNT
) / UNICODE_HANGUL_TCOUNT
);
549 $t = $index % UNICODE_HANGUL_TCOUNT
;
550 $out .= "\xe1\x84" . chr( 0x80 +
$l ) . "\xe1\x85" . chr( 0xa1 +
$v );
552 $out .= "\xe1\x87" . chr( 0x80 +
$t - 25 );
554 $out .= "\xe1\x86" . chr( 0xa7 +
$t );
565 * Sorts combining characters into canonical order. This is the
566 * final step in creating decomposed normal forms D and KD.
568 * @param $string String: a valid, decomposed UTF-8 string. Input is not validated.
569 * @return string a UTF-8 string with combining characters sorted in canonical order
571 static function fastCombiningSort( $string ) {
572 UtfNormal
::loadData();
573 $len = strlen( $string );
575 $combiners = array();
577 for( $i = 0; $i < $len; $i++
) {
582 $c = substr( $string, $i, 4 );
584 } elseif( $n >= 0xe0 ) {
585 $c = substr( $string, $i, 3 );
587 } elseif( $n >= 0xc0 ) {
588 $c = substr( $string, $i, 2 );
591 if( isset( self
::$utfCombiningClass[$c] ) ) {
592 $lastClass = self
::$utfCombiningClass[$c];
593 if( isset( $combiners[$lastClass] ) ) {
594 $combiners[$lastClass] .= $c;
596 $combiners[$lastClass] = $c;
603 $out .= implode( '', $combiners );
604 $combiners = array();
611 $out .= implode( '', $combiners );
617 * Produces canonically composed sequences, i.e. normal form C or KC.
620 * @param $string String: a valid UTF-8 string in sorted normal form D or KD. Input is not validated.
621 * @return string a UTF-8 string with canonical precomposed characters used where possible
623 static function fastCompose( $string ) {
624 UtfNormal
::loadData();
625 $len = strlen( $string );
631 $x1 = ord(substr(UTF8_HANGUL_VBASE
,0,1));
632 $x2 = ord(substr(UTF8_HANGUL_TEND
,0,1));
633 for( $i = 0; $i < $len; $i++
) {
637 # No combining characters here...
644 } elseif( $n >= 0xf0 ) {
645 $c = substr( $string, $i, 4 );
647 } elseif( $n >= 0xe0 ) {
648 $c = substr( $string, $i, 3 );
650 } elseif( $n >= 0xc0 ) {
651 $c = substr( $string, $i, 2 );
654 $pair = $startChar . $c;
656 if( isset( self
::$utfCombiningClass[$c] ) ) {
657 # A combining char; see what we can do with it
658 $class = self
::$utfCombiningClass[$c];
659 if( !empty( $startChar ) &&
660 $lastClass < $class &&
662 isset( self
::$utfCanonicalComp[$pair] ) ) {
663 $startChar = self
::$utfCanonicalComp[$pair];
674 if( $lastClass == 0 ) {
675 if( isset( self
::$utfCanonicalComp[$pair] ) ) {
676 $startChar = self
::$utfCanonicalComp[$pair];
680 if( $n >= $x1 && $n <= $x2 ) {
681 # WARNING: Hangul code is painfully slow.
682 # I apologize for this ugly, ugly code; however
683 # performance is even more teh suck if we call
684 # out to nice clean functions. Lookup tables are
685 # marginally faster, but require a lot of space.
687 if( $c >= UTF8_HANGUL_VBASE
&&
688 $c <= UTF8_HANGUL_VEND
&&
689 $startChar >= UTF8_HANGUL_LBASE
&&
690 $startChar <= UTF8_HANGUL_LEND
) {
692 #$lIndex = utf8ToCodepoint( $startChar ) - UNICODE_HANGUL_LBASE;
693 #$vIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_VBASE;
694 $lIndex = ord( $startChar[2] ) - 0x80;
695 $vIndex = ord( $c[2] ) - 0xa1;
697 $hangulPoint = UNICODE_HANGUL_FIRST +
698 UNICODE_HANGUL_TCOUNT
*
699 (UNICODE_HANGUL_VCOUNT
* $lIndex +
$vIndex);
701 # Hardcode the limited-range UTF-8 conversion:
702 $startChar = chr( $hangulPoint >> 12 & 0x0f |
0xe0 ) .
703 chr( $hangulPoint >> 6 & 0x3f |
0x80 ) .
704 chr( $hangulPoint & 0x3f |
0x80 );
707 } elseif( $c >= UTF8_HANGUL_TBASE
&&
708 $c <= UTF8_HANGUL_TEND
&&
709 $startChar >= UTF8_HANGUL_FIRST
&&
710 $startChar <= UTF8_HANGUL_LAST
&&
712 # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
713 $tIndex = ord( $c[2] ) - 0xa7;
714 if( $tIndex < 0 ) $tIndex = ord( $c[2] ) - 0x80 +
(0x11c0 - 0x11a7);
716 # Increment the code point by $tIndex, without
717 # the function overhead of decoding and recoding UTF-8
719 $tail = ord( $startChar[2] ) +
$tIndex;
722 $mid = ord( $startChar[1] ) +
1;
724 $startChar[0] = chr( ord( $startChar[0] ) +
1 );
727 $startChar[1] = chr( $mid );
729 $startChar[2] = chr( $tail );
731 # If there's another jamo char after this, *don't* try to merge it.
744 $out .= $startChar . $combining;
749 * This is just used for the benchmark, comparing how long it takes to
750 * interate through a string without really doing anything of substance.
751 * @param $string string
754 static function placebo( $string ) {
755 $len = strlen( $string );
757 for( $i = 0; $i < $len; $i++
) {
763 * Function to replace some characters that we don't want
764 * but most of the native normalize functions keep.
766 * @param $string String The string
767 * @return String String with the character codes replaced.
769 private static function replaceForNativeNormalize( $string ) {
770 $string = preg_replace(
771 '/[\x00-\x08\x0b\x0c\x0e-\x1f]/',
774 $string = str_replace( UTF8_FFFE
, UTF8_REPLACEMENT
, $string );
775 $string = str_replace( UTF8_FFFF
, UTF8_REPLACEMENT
, $string );