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
32 * For using the ICU wrapper
34 define( 'UNORM_NONE', 1 );
35 define( 'UNORM_NFD', 2 );
36 define( 'UNORM_NFKD', 3 );
37 define( 'UNORM_NFC', 4 );
38 define( 'UNORM_DEFAULT', UNORM_NFC
);
39 define( 'UNORM_NFKC', 5 );
40 define( 'UNORM_FCD', 6 );
42 define( 'NORMALIZE_ICU', function_exists( 'utf8_normalize' ) );
43 define( 'NORMALIZE_INTL', function_exists( 'normalizer_normalize' ) );
46 * Unicode normalization routines for working with UTF-8 strings.
47 * Currently assumes that input strings are valid UTF-8!
49 * Not as fast as I'd like, but should be usable for most purposes.
50 * UtfNormal::toNFC() will bail early if given ASCII text or text
51 * it can quickly deterimine is already normalized.
53 * All functions can be called static.
55 * See description of forms at http://www.unicode.org/reports/tr15/
60 static $utfCombiningClass = null;
61 static $utfCanonicalComp = null;
62 static $utfCanonicalDecomp = null;
64 # Load compatibility decompositions on demand if they are needed.
65 static $utfCompatibilityDecomp = null;
70 * The ultimate convenience function! Clean up invalid UTF-8 sequences,
71 * and convert to normal form C, canonical composition.
73 * Fast return for pure ASCII strings; some lesser optimizations for
74 * strings containing only known-good characters. Not as fast as toNFC().
76 * @param $string String: a UTF-8 string
77 * @return string a clean, shiny, normalized UTF-8 string
79 static function cleanUp( $string ) {
81 $string = self
::replaceForNativeNormalize( $string );
83 # UnicodeString constructor fails if the string ends with a
84 # head byte. Add a junk char at the end, we'll strip it off.
85 return rtrim( utf8_normalize( $string . "\x01", UNORM_NFC
), "\x01" );
86 } elseif( NORMALIZE_INTL
) {
87 $string = self
::replaceForNativeNormalize( $string );
88 $norm = normalizer_normalize( $string, Normalizer
::FORM_C
);
89 if( $norm === null ||
$norm === false ) {
90 # normalizer_normalize will either return false or null
91 # (depending on which doc you read) if invalid utf8 string.
92 # quickIsNFCVerify cleans up invalid sequences.
94 if( UtfNormal
::quickIsNFCVerify( $string ) ) {
95 # if that's true, the string is actually already normal.
98 # Now we are valid but non-normal
99 return normalizer_normalize( $string, Normalizer
::FORM_C
);
104 } elseif( UtfNormal
::quickIsNFCVerify( $string ) ) {
105 # Side effect -- $string has had UTF-8 errors cleaned up.
108 return UtfNormal
::NFC( $string );
113 * Convert a UTF-8 string to normal form C, canonical composition.
114 * Fast return for pure ASCII strings; some lesser optimizations for
115 * strings containing only known-good characters.
117 * @param $string String: a valid UTF-8 string. Input is not validated.
118 * @return string a UTF-8 string in normal form C
120 static function toNFC( $string ) {
122 return normalizer_normalize( $string, Normalizer
::FORM_C
);
123 elseif( NORMALIZE_ICU
)
124 return utf8_normalize( $string, UNORM_NFC
);
125 elseif( UtfNormal
::quickIsNFC( $string ) )
128 return UtfNormal
::NFC( $string );
132 * Convert a UTF-8 string to normal form D, canonical decomposition.
133 * Fast return for pure ASCII strings.
135 * @param $string String: a valid UTF-8 string. Input is not validated.
136 * @return string a UTF-8 string in normal form D
138 static function toNFD( $string ) {
140 return normalizer_normalize( $string, Normalizer
::FORM_D
);
141 elseif( NORMALIZE_ICU
)
142 return utf8_normalize( $string, UNORM_NFD
);
143 elseif( preg_match( '/[\x80-\xff]/', $string ) )
144 return UtfNormal
::NFD( $string );
150 * Convert a UTF-8 string to normal form KC, compatibility composition.
151 * This may cause irreversible information loss, use judiciously.
152 * Fast return for pure ASCII strings.
154 * @param $string String: a valid UTF-8 string. Input is not validated.
155 * @return string a UTF-8 string in normal form KC
157 static function toNFKC( $string ) {
159 return normalizer_normalize( $string, Normalizer
::FORM_KC
);
160 elseif( NORMALIZE_ICU
)
161 return utf8_normalize( $string, UNORM_NFKC
);
162 elseif( preg_match( '/[\x80-\xff]/', $string ) )
163 return UtfNormal
::NFKC( $string );
169 * Convert a UTF-8 string to normal form KD, compatibility decomposition.
170 * This may cause irreversible information loss, use judiciously.
171 * Fast return for pure ASCII strings.
173 * @param $string String: a valid UTF-8 string. Input is not validated.
174 * @return string a UTF-8 string in normal form KD
176 static function toNFKD( $string ) {
178 return normalizer_normalize( $string, Normalizer
::FORM_KD
);
179 elseif( NORMALIZE_ICU
)
180 return utf8_normalize( $string, UNORM_NFKD
);
181 elseif( preg_match( '/[\x80-\xff]/', $string ) )
182 return UtfNormal
::NFKD( $string );
188 * Load the basic composition data if necessary
191 static function loadData() {
192 if( !isset( self
::$utfCombiningClass ) ) {
193 require_once( dirname(__FILE__
) . '/UtfNormalData.inc' );
198 * Returns true if the string is _definitely_ in NFC.
199 * Returns false if not or uncertain.
200 * @param $string String: a valid UTF-8 string. Input is not validated.
203 static function quickIsNFC( $string ) {
204 # ASCII is always valid NFC!
205 # If it's pure ASCII, let it through.
206 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
208 UtfNormal
::loadData();
209 $len = strlen( $string );
210 for( $i = 0; $i < $len; $i++
) {
215 } elseif( $n >= 0xf0 ) {
216 $c = substr( $string, $i, 4 );
218 } elseif( $n >= 0xe0 ) {
219 $c = substr( $string, $i, 3 );
221 } elseif( $n >= 0xc0 ) {
222 $c = substr( $string, $i, 2 );
225 if( isset( self
::$utfCheckNFC[$c] ) ) {
226 # If it's NO or MAYBE, bail and do the slow check.
229 if( isset( self
::$utfCombiningClass[$c] ) ) {
230 # Combining character? We might have to do sorting, at least.
238 * Returns true if the string is _definitely_ in NFC.
239 * Returns false if not or uncertain.
240 * @param $string String: a UTF-8 string, altered on output to be valid UTF-8 safe for XML.
242 static function quickIsNFCVerify( &$string ) {
243 # Screen out some characters that eg won't be allowed in XML
244 $string = preg_replace( '/[\x00-\x08\x0b\x0c\x0e-\x1f]/', UTF8_REPLACEMENT
, $string );
246 # ASCII is always valid NFC!
247 # If we're only ever given plain ASCII, we can avoid the overhead
248 # of initializing the decomposition tables by skipping out early.
249 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
251 static $checkit = null, $tailBytes = null, $utfCheckOrCombining = null;
252 if( !isset( $checkit ) ) {
253 # Load/build some scary lookup tables...
254 UtfNormal
::loadData();
256 $utfCheckOrCombining = array_merge( self
::$utfCheckNFC, self
::$utfCombiningClass );
258 # Head bytes for sequences which we should do further validity checks
259 $checkit = array_flip( array_map( 'chr',
260 array( 0xc0, 0xc1, 0xe0, 0xed, 0xef,
261 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7,
262 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff ) ) );
264 # Each UTF-8 head byte is followed by a certain
265 # number of tail bytes.
266 $tailBytes = array();
267 for( $n = 0; $n < 256; $n++
) {
270 } elseif( $n < 0xe0 ) {
272 } elseif( $n < 0xf0 ) {
274 } elseif( $n < 0xf8 ) {
276 } elseif( $n < 0xfc ) {
278 } elseif( $n < 0xfe ) {
283 $tailBytes[chr($n)] = $remaining;
287 # Chop the text into pure-ASCII and non-ASCII areas;
288 # large ASCII parts can be handled much more quickly.
289 # Don't chop up Unicode areas for punctuation, though,
290 # that wastes energy.
293 '/([\x00-\x7f]+|[\x80-\xff][\x00-\x40\x5b-\x5f\x7b-\xff]*)/',
299 foreach( $matches[1] as $str ) {
300 $chunk = strlen( $str );
302 if( $str[0] < "\x80" ) {
303 # ASCII chunk: guaranteed to be valid UTF-8
304 # and in normal form C, so skip over it.
309 # We'll have to examine the chunk byte by byte to ensure
310 # that it consists of valid UTF-8 sequences, and to see
311 # if any of them might not be normalized.
313 # Since PHP is not the fastest language on earth, some of
314 # this code is a little ugly with inner loop optimizations.
317 $len = $chunk +
1; # Counting down is faster. I'm *so* sorry.
319 for( $i = -1; --$len; ) {
320 $remaining = $tailBytes[$c = $str[++
$i]];
323 $sequence = $head = $c;
325 # Look for the defined number of tail bytes...
326 if( --$len && ( $c = $str[++
$i] ) >= "\x80" && $c < "\xc0" ) {
327 # Legal tail bytes are nice.
331 # Premature end of string!
332 # Drop a replacement character into output to
333 # represent the invalid UTF-8 sequence.
334 $replace[] = array( UTF8_REPLACEMENT
,
335 $base +
$i +
1 - strlen( $sequence ),
336 strlen( $sequence ) );
339 # Illegal tail byte; abandon the sequence.
340 $replace[] = array( UTF8_REPLACEMENT
,
341 $base +
$i - strlen( $sequence ),
342 strlen( $sequence ) );
343 # Back up and reprocess this byte; it may itself
344 # be a legal ASCII or UTF-8 sequence head.
350 } while( --$remaining );
352 if( isset( $checkit[$head] ) ) {
353 # Do some more detailed validity checks, for
354 # invalid characters and illegal sequences.
355 if( $head == "\xed" ) {
356 # 0xed is relatively frequent in Korean, which
357 # abuts the surrogate area, so we're doing
358 # this check separately to speed things up.
360 if( $sequence >= UTF8_SURROGATE_FIRST
) {
361 # Surrogates are legal only in UTF-16 code.
362 # They are totally forbidden here in UTF-8
364 $replace[] = array( UTF8_REPLACEMENT
,
365 $base +
$i +
1 - strlen( $sequence ),
366 strlen( $sequence ) );
371 # Slower, but rarer checks...
374 # "Overlong sequences" are those that are syntactically
375 # correct but use more UTF-8 bytes than are necessary to
376 # encode a character. Naïve string comparisons can be
377 # tricked into failing to see a match for an ASCII
378 # character, for instance, which can be a security hole
379 # if blacklist checks are being used.
380 ($n < 0xc2 && $sequence <= UTF8_OVERLONG_A
)
381 ||
($n == 0xe0 && $sequence <= UTF8_OVERLONG_B
)
382 ||
($n == 0xf0 && $sequence <= UTF8_OVERLONG_C
)
384 # U+FFFE and U+FFFF are explicitly forbidden in Unicode.
386 ($sequence == UTF8_FFFE
)
387 ||
($sequence == UTF8_FFFF
) )
389 # Unicode has been limited to 21 bits; longer
390 # sequences are not allowed.
391 ||
($n >= 0xf0 && $sequence > UTF8_MAX
) ) {
393 $replace[] = array( UTF8_REPLACEMENT
,
394 $base +
$i +
1 - strlen( $sequence ),
395 strlen( $sequence ) );
402 if( isset( $utfCheckOrCombining[$sequence] ) ) {
403 # If it's NO or MAYBE, we'll have to rip
404 # the string apart and put it back together.
405 # That's going to be mighty slow.
406 $looksNormal = false;
409 # The sequence is legal!
411 } elseif( $c < "\x80" ) {
414 } elseif( $c < "\xc0" ) {
418 $replace[] = array( UTF8_REPLACEMENT
, $base +
$i, 1 );
420 # Don't add if we're continuing a broken sequence;
421 # we already put a replacement character when we looked
422 # at the broken sequence.
423 $replace[] = array( '', $base +
$i, 1 );
426 # Miscellaneous freaks.
427 $replace[] = array( UTF8_REPLACEMENT
, $base +
$i, 1 );
433 if( count( $replace ) ) {
434 # There were illegal UTF-8 sequences we need to fix up.
437 foreach( $replace as $rep ) {
438 list( $replacement, $start, $length ) = $rep;
439 if( $last < $start ) {
440 $out .= substr( $string, $last, $start - $last );
442 $out .= $replacement;
443 $last = $start +
$length;
445 if( $last < strlen( $string ) ) {
446 $out .= substr( $string, $last );
453 # These take a string and run the normalization on them, without
454 # checking for validity or any optimization etc. Input must be
457 * @param $string string
461 static function NFC( $string ) {
462 return UtfNormal
::fastCompose( UtfNormal
::NFD( $string ) );
466 * @param $string string
470 static function NFD( $string ) {
471 UtfNormal
::loadData();
473 return UtfNormal
::fastCombiningSort(
474 UtfNormal
::fastDecompose( $string, self
::$utfCanonicalDecomp ) );
478 * @param $string string
482 static function NFKC( $string ) {
483 return UtfNormal
::fastCompose( UtfNormal
::NFKD( $string ) );
487 * @param $string string
491 static function NFKD( $string ) {
492 if( !isset( self
::$utfCompatibilityDecomp ) ) {
493 require_once( 'UtfNormalDataK.inc' );
495 return self
::fastCombiningSort(
496 self
::fastDecompose( $string, self
::$utfCompatibilityDecomp ) );
501 * Perform decomposition of a UTF-8 string into either D or KD form
502 * (depending on which decomposition map is passed to us).
503 * Input is assumed to be *valid* UTF-8. Invalid code will break.
505 * @param $string String: valid UTF-8 string
506 * @param $map Array: hash of expanded decomposition map
507 * @return string a UTF-8 string decomposed, not yet normalized (needs sorting)
509 static function fastDecompose( $string, $map ) {
510 UtfNormal
::loadData();
511 $len = strlen( $string );
513 for( $i = 0; $i < $len; $i++
) {
517 # ASCII chars never decompose
521 } elseif( $n >= 0xf0 ) {
522 $c = substr( $string, $i, 4 );
524 } elseif( $n >= 0xe0 ) {
525 $c = substr( $string, $i, 3 );
527 } elseif( $n >= 0xc0 ) {
528 $c = substr( $string, $i, 2 );
531 if( isset( $map[$c] ) ) {
535 if( $c >= UTF8_HANGUL_FIRST
&& $c <= UTF8_HANGUL_LAST
) {
536 # Decompose a hangul syllable into jamo;
537 # hardcoded for three-byte UTF-8 sequence.
538 # A lookup table would be slightly faster,
539 # but adds a lot of memory & disk needs.
541 $index = ( (ord( $c[0] ) & 0x0f) << 12
542 |
(ord( $c[1] ) & 0x3f) << 6
543 |
(ord( $c[2] ) & 0x3f) )
544 - UNICODE_HANGUL_FIRST
;
545 $l = intval( $index / UNICODE_HANGUL_NCOUNT
);
546 $v = intval( ($index % UNICODE_HANGUL_NCOUNT
) / UNICODE_HANGUL_TCOUNT
);
547 $t = $index % UNICODE_HANGUL_TCOUNT
;
548 $out .= "\xe1\x84" . chr( 0x80 +
$l ) . "\xe1\x85" . chr( 0xa1 +
$v );
550 $out .= "\xe1\x87" . chr( 0x80 +
$t - 25 );
552 $out .= "\xe1\x86" . chr( 0xa7 +
$t );
563 * Sorts combining characters into canonical order. This is the
564 * final step in creating decomposed normal forms D and KD.
566 * @param $string String: a valid, decomposed UTF-8 string. Input is not validated.
567 * @return string a UTF-8 string with combining characters sorted in canonical order
569 static function fastCombiningSort( $string ) {
570 UtfNormal
::loadData();
571 $len = strlen( $string );
573 $combiners = array();
575 for( $i = 0; $i < $len; $i++
) {
580 $c = substr( $string, $i, 4 );
582 } elseif( $n >= 0xe0 ) {
583 $c = substr( $string, $i, 3 );
585 } elseif( $n >= 0xc0 ) {
586 $c = substr( $string, $i, 2 );
589 if( isset( self
::$utfCombiningClass[$c] ) ) {
590 $lastClass = self
::$utfCombiningClass[$c];
591 if( isset( $combiners[$lastClass] ) ) {
592 $combiners[$lastClass] .= $c;
594 $combiners[$lastClass] = $c;
601 $out .= implode( '', $combiners );
602 $combiners = array();
609 $out .= implode( '', $combiners );
615 * Produces canonically composed sequences, i.e. normal form C or KC.
618 * @param $string String: a valid UTF-8 string in sorted normal form D or KD. Input is not validated.
619 * @return string a UTF-8 string with canonical precomposed characters used where possible
621 static function fastCompose( $string ) {
622 UtfNormal
::loadData();
623 $len = strlen( $string );
629 $x1 = ord(substr(UTF8_HANGUL_VBASE
,0,1));
630 $x2 = ord(substr(UTF8_HANGUL_TEND
,0,1));
631 for( $i = 0; $i < $len; $i++
) {
635 # No combining characters here...
642 } elseif( $n >= 0xf0 ) {
643 $c = substr( $string, $i, 4 );
645 } elseif( $n >= 0xe0 ) {
646 $c = substr( $string, $i, 3 );
648 } elseif( $n >= 0xc0 ) {
649 $c = substr( $string, $i, 2 );
652 $pair = $startChar . $c;
654 if( isset( self
::$utfCombiningClass[$c] ) ) {
655 # A combining char; see what we can do with it
656 $class = self
::$utfCombiningClass[$c];
657 if( !empty( $startChar ) &&
658 $lastClass < $class &&
660 isset( self
::$utfCanonicalComp[$pair] ) ) {
661 $startChar = self
::$utfCanonicalComp[$pair];
672 if( $lastClass == 0 ) {
673 if( isset( self
::$utfCanonicalComp[$pair] ) ) {
674 $startChar = self
::$utfCanonicalComp[$pair];
678 if( $n >= $x1 && $n <= $x2 ) {
679 # WARNING: Hangul code is painfully slow.
680 # I apologize for this ugly, ugly code; however
681 # performance is even more teh suck if we call
682 # out to nice clean functions. Lookup tables are
683 # marginally faster, but require a lot of space.
685 if( $c >= UTF8_HANGUL_VBASE
&&
686 $c <= UTF8_HANGUL_VEND
&&
687 $startChar >= UTF8_HANGUL_LBASE
&&
688 $startChar <= UTF8_HANGUL_LEND
) {
690 #$lIndex = utf8ToCodepoint( $startChar ) - UNICODE_HANGUL_LBASE;
691 #$vIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_VBASE;
692 $lIndex = ord( $startChar[2] ) - 0x80;
693 $vIndex = ord( $c[2] ) - 0xa1;
695 $hangulPoint = UNICODE_HANGUL_FIRST +
696 UNICODE_HANGUL_TCOUNT
*
697 (UNICODE_HANGUL_VCOUNT
* $lIndex +
$vIndex);
699 # Hardcode the limited-range UTF-8 conversion:
700 $startChar = chr( $hangulPoint >> 12 & 0x0f |
0xe0 ) .
701 chr( $hangulPoint >> 6 & 0x3f |
0x80 ) .
702 chr( $hangulPoint & 0x3f |
0x80 );
705 } elseif( $c >= UTF8_HANGUL_TBASE
&&
706 $c <= UTF8_HANGUL_TEND
&&
707 $startChar >= UTF8_HANGUL_FIRST
&&
708 $startChar <= UTF8_HANGUL_LAST
&&
710 # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
711 $tIndex = ord( $c[2] ) - 0xa7;
712 if( $tIndex < 0 ) $tIndex = ord( $c[2] ) - 0x80 +
(0x11c0 - 0x11a7);
714 # Increment the code point by $tIndex, without
715 # the function overhead of decoding and recoding UTF-8
717 $tail = ord( $startChar[2] ) +
$tIndex;
720 $mid = ord( $startChar[1] ) +
1;
722 $startChar[0] = chr( ord( $startChar[0] ) +
1 );
725 $startChar[1] = chr( $mid );
727 $startChar[2] = chr( $tail );
729 # If there's another jamo char after this, *don't* try to merge it.
742 $out .= $startChar . $combining;
747 * This is just used for the benchmark, comparing how long it takes to
748 * interate through a string without really doing anything of substance.
749 * @param $string string
752 static function placebo( $string ) {
753 $len = strlen( $string );
755 for( $i = 0; $i < $len; $i++
) {
761 * Function to replace some characters that we don't want
762 * but most of the native normalize functions keep.
764 * @param $string String The string
765 * @return String String with the character codes replaced.
767 private static function replaceForNativeNormalize( $string ) {
768 $string = preg_replace(
769 '/[\x00-\x08\x0b\x0c\x0e-\x1f]/',
772 $string = str_replace( UTF8_FFFE
, UTF8_REPLACEMENT
, $string );
773 $string = str_replace( UTF8_FFFF
, UTF8_REPLACEMENT
, $string );