3 * Unicode normalization routines
5 * Copyright © 2004 Brion Vibber <brion@pobox.com>
6 * https://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 define( 'NORMALIZE_ICU', function_exists( 'utf8_normalize' ) );
32 define( 'NORMALIZE_INTL', function_exists( 'normalizer_normalize' ) );
35 * Unicode normalization routines for working with UTF-8 strings.
36 * Currently assumes that input strings are valid UTF-8!
38 * Not as fast as I'd like, but should be usable for most purposes.
39 * UtfNormal::toNFC() will bail early if given ASCII text or text
40 * it can quickly determine is already normalized.
42 * All functions can be called static.
44 * See description of forms at http://www.unicode.org/reports/tr15/
50 * For using the ICU wrapper
58 const UNORM_DEFAULT
= self
::UNORM_NFC
;
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 ) {
80 if ( NORMALIZE_ICU
) {
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", self
::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 ) {
121 if ( NORMALIZE_INTL
)
122 return normalizer_normalize( $string, Normalizer
::FORM_C
);
123 elseif ( NORMALIZE_ICU
)
124 return utf8_normalize( $string, self
::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 ) {
139 if ( NORMALIZE_INTL
)
140 return normalizer_normalize( $string, Normalizer
::FORM_D
);
141 elseif ( NORMALIZE_ICU
)
142 return utf8_normalize( $string, self
::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 ) {
158 if ( NORMALIZE_INTL
)
159 return normalizer_normalize( $string, Normalizer
::FORM_KC
);
160 elseif ( NORMALIZE_ICU
)
161 return utf8_normalize( $string, self
::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 ) {
177 if ( NORMALIZE_INTL
)
178 return normalizer_normalize( $string, Normalizer
::FORM_KD
);
179 elseif ( NORMALIZE_ICU
)
180 return utf8_normalize( $string, self
::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 __DIR__
. '/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.
239 * Returns true if the string is _definitely_ in NFC.
240 * Returns false if not or uncertain.
241 * @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
)
396 $replace[] = array( UTF8_REPLACEMENT
,
397 $base +
$i +
1 - strlen( $sequence ),
398 strlen( $sequence ) );
405 if ( isset( $utfCheckOrCombining[$sequence] ) ) {
406 # If it's NO or MAYBE, we'll have to rip
407 # the string apart and put it back together.
408 # That's going to be mighty slow.
409 $looksNormal = false;
412 # The sequence is legal!
414 } elseif ( $c < "\x80" ) {
417 } elseif ( $c < "\xc0" ) {
421 $replace[] = array( UTF8_REPLACEMENT
, $base +
$i, 1 );
423 # Don't add if we're continuing a broken sequence;
424 # we already put a replacement character when we looked
425 # at the broken sequence.
426 $replace[] = array( '', $base +
$i, 1 );
429 # Miscellaneous freaks.
430 $replace[] = array( UTF8_REPLACEMENT
, $base +
$i, 1 );
436 if ( count( $replace ) ) {
437 # There were illegal UTF-8 sequences we need to fix up.
440 foreach ( $replace as $rep ) {
441 list( $replacement, $start, $length ) = $rep;
442 if ( $last < $start ) {
443 $out .= substr( $string, $last, $start - $last );
445 $out .= $replacement;
446 $last = $start +
$length;
448 if ( $last < strlen( $string ) ) {
449 $out .= substr( $string, $last );
457 # These take a string and run the normalization on them, without
458 # checking for validity or any optimization etc. Input must be
461 * @param $string string
465 static function NFC( $string ) {
466 return UtfNormal
::fastCompose( UtfNormal
::NFD( $string ) );
470 * @param $string string
474 static function NFD( $string ) {
475 UtfNormal
::loadData();
477 return UtfNormal
::fastCombiningSort(
478 UtfNormal
::fastDecompose( $string, self
::$utfCanonicalDecomp ) );
482 * @param $string string
486 static function NFKC( $string ) {
487 return UtfNormal
::fastCompose( UtfNormal
::NFKD( $string ) );
491 * @param $string string
495 static function NFKD( $string ) {
496 if ( !isset( self
::$utfCompatibilityDecomp ) ) {
497 require_once 'UtfNormalDataK.inc';
500 return self
::fastCombiningSort(
501 self
::fastDecompose( $string, self
::$utfCompatibilityDecomp ) );
505 * Perform decomposition of a UTF-8 string into either D or KD form
506 * (depending on which decomposition map is passed to us).
507 * Input is assumed to be *valid* UTF-8. Invalid code will break.
509 * @param string $string valid UTF-8 string
510 * @param array $map hash of expanded decomposition map
511 * @return string a UTF-8 string decomposed, not yet normalized (needs sorting)
513 static function fastDecompose( $string, $map ) {
514 UtfNormal
::loadData();
515 $len = strlen( $string );
517 for ( $i = 0; $i < $len; $i++
) {
521 # ASCII chars never decompose
525 } elseif ( $n >= 0xf0 ) {
526 $c = substr( $string, $i, 4 );
528 } elseif ( $n >= 0xe0 ) {
529 $c = substr( $string, $i, 3 );
531 } elseif ( $n >= 0xc0 ) {
532 $c = substr( $string, $i, 2 );
535 if ( isset( $map[$c] ) ) {
539 if ( $c >= UTF8_HANGUL_FIRST
&& $c <= UTF8_HANGUL_LAST
) {
540 # Decompose a hangul syllable into jamo;
541 # hardcoded for three-byte UTF-8 sequence.
542 # A lookup table would be slightly faster,
543 # but adds a lot of memory & disk needs.
545 $index = ( ( ord( $c[0] ) & 0x0f ) << 12
546 |
( ord( $c[1] ) & 0x3f ) << 6
547 |
( ord( $c[2] ) & 0x3f ) )
548 - UNICODE_HANGUL_FIRST
;
549 $l = intval( $index / UNICODE_HANGUL_NCOUNT
);
550 $v = intval( ( $index % UNICODE_HANGUL_NCOUNT
) / UNICODE_HANGUL_TCOUNT
);
551 $t = $index % UNICODE_HANGUL_TCOUNT
;
552 $out .= "\xe1\x84" . chr( 0x80 +
$l ) . "\xe1\x85" . chr( 0xa1 +
$v );
554 $out .= "\xe1\x87" . chr( 0x80 +
$t - 25 );
556 $out .= "\xe1\x86" . chr( 0xa7 +
$t );
568 * Sorts combining characters into canonical order. This is the
569 * final step in creating decomposed normal forms D and KD.
571 * @param string $string a valid, decomposed UTF-8 string. Input is not validated.
572 * @return string a UTF-8 string with combining characters sorted in canonical order
574 static function fastCombiningSort( $string ) {
575 UtfNormal
::loadData();
576 $len = strlen( $string );
578 $combiners = array();
580 for ( $i = 0; $i < $len; $i++
) {
585 $c = substr( $string, $i, 4 );
587 } elseif ( $n >= 0xe0 ) {
588 $c = substr( $string, $i, 3 );
590 } elseif ( $n >= 0xc0 ) {
591 $c = substr( $string, $i, 2 );
594 if ( isset( self
::$utfCombiningClass[$c] ) ) {
595 $lastClass = self
::$utfCombiningClass[$c];
596 if ( isset( $combiners[$lastClass] ) ) {
597 $combiners[$lastClass] .= $c;
599 $combiners[$lastClass] = $c;
606 $out .= implode( '', $combiners );
607 $combiners = array();
614 $out .= implode( '', $combiners );
621 * Produces canonically composed sequences, i.e. normal form C or KC.
624 * @param string $string a valid UTF-8 string in sorted normal form D or KD. Input is not validated.
625 * @return string a UTF-8 string with canonical precomposed characters used where possible
627 static function fastCompose( $string ) {
628 UtfNormal
::loadData();
629 $len = strlen( $string );
635 $x1 = ord( substr( UTF8_HANGUL_VBASE
, 0, 1 ) );
636 $x2 = ord( substr( UTF8_HANGUL_TEND
, 0, 1 ) );
637 for ( $i = 0; $i < $len; $i++
) {
641 # No combining characters here...
648 } elseif ( $n >= 0xf0 ) {
649 $c = substr( $string, $i, 4 );
651 } elseif ( $n >= 0xe0 ) {
652 $c = substr( $string, $i, 3 );
654 } elseif ( $n >= 0xc0 ) {
655 $c = substr( $string, $i, 2 );
658 $pair = $startChar . $c;
660 if ( isset( self
::$utfCombiningClass[$c] ) ) {
661 # A combining char; see what we can do with it
662 $class = self
::$utfCombiningClass[$c];
663 if ( !empty( $startChar ) &&
664 $lastClass < $class &&
666 isset( self
::$utfCanonicalComp[$pair] )
668 $startChar = self
::$utfCanonicalComp[$pair];
679 if ( $lastClass == 0 ) {
680 if ( isset( self
::$utfCanonicalComp[$pair] ) ) {
681 $startChar = self
::$utfCanonicalComp[$pair];
685 if ( $n >= $x1 && $n <= $x2 ) {
686 # WARNING: Hangul code is painfully slow.
687 # I apologize for this ugly, ugly code; however
688 # performance is even more teh suck if we call
689 # out to nice clean functions. Lookup tables are
690 # marginally faster, but require a lot of space.
692 if ( $c >= UTF8_HANGUL_VBASE
&&
693 $c <= UTF8_HANGUL_VEND
&&
694 $startChar >= UTF8_HANGUL_LBASE
&&
695 $startChar <= UTF8_HANGUL_LEND
698 #$lIndex = utf8ToCodepoint( $startChar ) - UNICODE_HANGUL_LBASE;
699 #$vIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_VBASE;
700 $lIndex = ord( $startChar[2] ) - 0x80;
701 $vIndex = ord( $c[2] ) - 0xa1;
703 $hangulPoint = UNICODE_HANGUL_FIRST +
704 UNICODE_HANGUL_TCOUNT
*
705 ( UNICODE_HANGUL_VCOUNT
* $lIndex +
$vIndex );
707 # Hardcode the limited-range UTF-8 conversion:
708 $startChar = chr( $hangulPoint >> 12 & 0x0f |
0xe0 ) .
709 chr( $hangulPoint >> 6 & 0x3f |
0x80 ) .
710 chr( $hangulPoint & 0x3f |
0x80 );
713 } elseif ( $c >= UTF8_HANGUL_TBASE
&&
714 $c <= UTF8_HANGUL_TEND
&&
715 $startChar >= UTF8_HANGUL_FIRST
&&
716 $startChar <= UTF8_HANGUL_LAST
&&
719 # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
720 $tIndex = ord( $c[2] ) - 0xa7;
721 if ( $tIndex < 0 ) $tIndex = ord( $c[2] ) - 0x80 +
( 0x11c0 - 0x11a7 );
723 # Increment the code point by $tIndex, without
724 # the function overhead of decoding and recoding UTF-8
726 $tail = ord( $startChar[2] ) +
$tIndex;
727 if ( $tail > 0xbf ) {
729 $mid = ord( $startChar[1] ) +
1;
731 $startChar[0] = chr( ord( $startChar[0] ) +
1 );
734 $startChar[1] = chr( $mid );
736 $startChar[2] = chr( $tail );
738 # If there's another jamo char after this, *don't* try to merge it.
751 $out .= $startChar . $combining;
757 * This is just used for the benchmark, comparing how long it takes to
758 * interate through a string without really doing anything of substance.
759 * @param $string string
762 static function placebo( $string ) {
763 $len = strlen( $string );
765 for ( $i = 0; $i < $len; $i++
) {
773 * Function to replace some characters that we don't want
774 * but most of the native normalize functions keep.
776 * @param string $string The string
777 * @return String String with the character codes replaced.
779 private static function replaceForNativeNormalize( $string ) {
780 $string = preg_replace(
781 '/[\x00-\x08\x0b\x0c\x0e-\x1f]/',
784 $string = str_replace( UTF8_FFFE
, UTF8_REPLACEMENT
, $string );
785 $string = str_replace( UTF8_FFFF
, UTF8_REPLACEMENT
, $string );