includes/normal/UtfNormal.php

   1 <?php
   2 /**
   3  * Unicode normalization routines
   4  *
   5  * Copyright © 2004 Brion Vibber <brion@pobox.com>
   6  * http://www.mediawiki.org/
   7  *
   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.
  12  *
  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.
  17  *
  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
  22  *
  23  * @file
  24  * @ingroup UtfNormal
  25  */
  26
  27 /**
  28  * @defgroup UtfNormal UtfNormal
  29  */
  30
  31 require_once dirname(__FILE__).'/UtfNormalUtil.php';
  32
  33 /**
  34  * For using the ICU wrapper
  35  */
  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 );
  43
  44 define( 'NORMALIZE_ICU', function_exists( 'utf8_normalize' ) );
  45 define( 'NORMALIZE_INTL', function_exists( 'normalizer_normalize' ) );
  46
  47 /**
  48  * Unicode normalization routines for working with UTF-8 strings.
  49  * Currently assumes that input strings are valid UTF-8!
  50  *
  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.
  54  *
  55  * All functions can be called static.
  56  *
  57  * See description of forms at http://www.unicode.org/reports/tr15/
  58  *
  59  * @ingroup UtfNormal
  60  */
  61 class UtfNormal {
  62         static $utfCombiningClass = null;
  63         static $utfCanonicalComp = null;
  64         static $utfCanonicalDecomp = null;
  65
  66         # Load compatibility decompositions on demand if they are needed.
  67         static $utfCompatibilityDecomp = null;
  68
  69         static $utfCheckNFC;
  70
  71         /**
  72          * The ultimate convenience function! Clean up invalid UTF-8 sequences,
  73          * and convert to normal form C, canonical composition.
  74          *
  75          * Fast return for pure ASCII strings; some lesser optimizations for
  76          * strings containing only known-good characters. Not as fast as toNFC().
  77          *
  78          * @param $string String: a UTF-8 string
  79          * @return string a clean, shiny, normalized UTF-8 string
  80          */
  81         static function cleanUp( $string ) {
  82                 if( NORMALIZE_ICU ) {
  83                         $string = self::replaceForNativeNormalize( $string );
  84
  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.
  95
  96                                 if( UtfNormal::quickIsNFCVerify( $string ) ) {
  97                                         # if that's true, the string is actually already normal.
  98                                         return $string;
  99                                 } else {
 100                                         # Now we are valid but non-normal
 101                                         return normalizer_normalize( $string, Normalizer::FORM_C );
 102                                 }
 103                         } else {
 104                                 return $norm;
 105                         }
 106                 } elseif( UtfNormal::quickIsNFCVerify( $string ) ) {
 107                         # Side effect -- $string has had UTF-8 errors cleaned up.
 108                         return $string;
 109                 } else {
 110                         return UtfNormal::NFC( $string );
 111                 }
 112         }
 113
 114         /**
 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.
 118          *
 119          * @param $string String: a valid UTF-8 string. Input is not validated.
 120          * @return string a UTF-8 string in normal form C
 121          */
 122         static function toNFC( $string ) {
 123                 if( NORMALIZE_INTL )
 124                         return normalizer_normalize( $string, Normalizer::FORM_C );
 125                 elseif( NORMALIZE_ICU )
 126                         return utf8_normalize( $string, UNORM_NFC );
 127                 elseif( UtfNormal::quickIsNFC( $string ) )
 128                         return $string;
 129                 else
 130                         return UtfNormal::NFC( $string );
 131         }
 132
 133         /**
 134          * Convert a UTF-8 string to normal form D, canonical decomposition.
 135          * Fast return for pure ASCII strings.
 136          *
 137          * @param $string String: a valid UTF-8 string. Input is not validated.
 138          * @return string a UTF-8 string in normal form D
 139          */
 140         static function toNFD( $string ) {
 141                 if( NORMALIZE_INTL )
 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 );
 147                 else
 148                         return $string;
 149         }
 150
 151         /**
 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.
 155          *
 156          * @param $string String: a valid UTF-8 string. Input is not validated.
 157          * @return string a UTF-8 string in normal form KC
 158          */
 159         static function toNFKC( $string ) {
 160                 if( NORMALIZE_INTL )
 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 );
 166                 else
 167                         return $string;
 168         }
 169
 170         /**
 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.
 174          *
 175          * @param $string String: a valid UTF-8 string. Input is not validated.
 176          * @return string a UTF-8 string in normal form KD
 177          */
 178         static function toNFKD( $string ) {
 179                 if( NORMALIZE_INTL )
 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 );
 185                 else
 186                         return $string;
 187         }
 188
 189         /**
 190          * Load the basic composition data if necessary
 191          * @private
 192          */
 193         static function loadData() {
 194                 if( !isset( self::$utfCombiningClass ) ) {
 195                         require_once( dirname(__FILE__) . '/UtfNormalData.inc' );
 196                 }
 197         }
 198
 199         /**
 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.
 203          * @return bool
 204          */
 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;
 209
 210                 UtfNormal::loadData();
 211                 $len = strlen( $string );
 212                 for( $i = 0; $i < $len; $i++ ) {
 213                         $c = $string{$i};
 214                         $n = ord( $c );
 215                         if( $n < 0x80 ) {
 216                                 continue;
 217                         } elseif( $n >= 0xf0 ) {
 218                                 $c = substr( $string, $i, 4 );
 219                                 $i += 3;
 220                         } elseif( $n >= 0xe0 ) {
 221                                 $c = substr( $string, $i, 3 );
 222                                 $i += 2;
 223                         } elseif( $n >= 0xc0 ) {
 224                                 $c = substr( $string, $i, 2 );
 225                                 $i++;
 226                         }
 227                         if( isset( self::$utfCheckNFC[$c] ) ) {
 228                                 # If it's NO or MAYBE, bail and do the slow check.
 229                                 return false;
 230                         }
 231                         if( isset( self::$utfCombiningClass[$c] ) ) {
 232                                 # Combining character? We might have to do sorting, at least.
 233                                 return false;
 234                         }
 235                 }
 236                 return true;
 237         }
 238
 239         /**
 240          * Returns true if the string is _definitely_ in NFC.
 241          * Returns false if not or uncertain.
 242          * @param $string String: a UTF-8 string, altered on output to be valid UTF-8 safe for XML.
 243          */
 244         static function quickIsNFCVerify( &$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 );
 247
 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;
 252
 253                 static $checkit = null, $tailBytes = null, $utfCheckOrCombining = null;
 254                 if( !isset( $checkit ) ) {
 255                         # Load/build some scary lookup tables...
 256                         UtfNormal::loadData();
 257
 258                         $utfCheckOrCombining = array_merge( self::$utfCheckNFC, self::$utfCombiningClass );
 259
 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 ) ) );
 265
 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++ ) {
 270                                 if( $n < 0xc0 ) {
 271                                         $remaining = 0;
 272                                 } elseif( $n < 0xe0 ) {
 273                                         $remaining = 1;
 274                                 } elseif( $n < 0xf0 ) {
 275                                         $remaining = 2;
 276                                 } elseif( $n < 0xf8 ) {
 277                                         $remaining = 3;
 278                                 } elseif( $n < 0xfc ) {
 279                                         $remaining = 4;
 280                                 } elseif( $n < 0xfe ) {
 281                                         $remaining = 5;
 282                                 } else {
 283                                         $remaining = 0;
 284                                 }
 285                                 $tailBytes[chr($n)] = $remaining;
 286                         }
 287                 }
 288
 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.
 293                 $matches = array();
 294                 preg_match_all(
 295                         '/([\x00-\x7f]+|[\x80-\xff][\x00-\x40\x5b-\x5f\x7b-\xff]*)/',
 296                         $string, $matches );
 297
 298                 $looksNormal = true;
 299                 $base = 0;
 300                 $replace = array();
 301                 foreach( $matches[1] as $str ) {
 302                         $chunk = strlen( $str );
 303
 304                         if( $str{0} < "\x80" ) {
 305                                 # ASCII chunk: guaranteed to be valid UTF-8
 306                                 # and in normal form C, so skip over it.
 307                                 $base += $chunk;
 308                                 continue;
 309                         }
 310
 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.
 314                         #
 315                         # Since PHP is not the fastest language on earth, some of
 316                         # this code is a little ugly with inner loop optimizations.
 317
 318                         $head = '';
 319                         $len = $chunk + 1; # Counting down is faster. I'm *so* sorry.
 320
 321                         for( $i = -1; --$len; ) {
 322                                 $remaining = $tailBytes[$c = $str{++$i}];
 323                                 if( $remaining ) {
 324                                         # UTF-8 head byte!
 325                                         $sequence = $head = $c;
 326                                         do {
 327                                                 # Look for the defined number of tail bytes...
 328                                                 if( --$len && ( $c = $str{++$i} ) >= "\x80" && $c < "\xc0" ) {
 329                                                         # Legal tail bytes are nice.
 330                                                         $sequence .= $c;
 331                                                 } else {
 332                                                         if( 0 == $len ) {
 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 ) );
 339                                                                 break 2;
 340                                                         } else {
 341                                                                 # Illegal tail byte; abandon the sequence.
 342                                                                 $replace[] = array( UTF8_REPLACEMENT,
 343                                                                                                         $base + $i - strlen( $sequence ),
 344                                                                                                         strlen( $sequence ) );
 345                                                                 # Back up and reprocess this byte; it may itself
 346                                                                 # be a legal ASCII or UTF-8 sequence head.
 347                                                                 --$i;
 348                                                                 ++$len;
 349                                                                 continue 2;
 350                                                         }
 351                                                 }
 352                                         } while( --$remaining );
 353
 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.
 361
 362                                                         if( $sequence >= UTF8_SURROGATE_FIRST ) {
 363                                                                 # Surrogates are legal only in UTF-16 code.
 364                                                                 # They are totally forbidden here in UTF-8
 365                                                                 # utopia.
 366                                                                 $replace[] = array( UTF8_REPLACEMENT,
 367                                                                              $base + $i + 1 - strlen( $sequence ),
 368                                                                              strlen( $sequence ) );
 369                                                                 $head = '';
 370                                                                 continue;
 371                                                         }
 372                                                 } else {
 373                                                         # Slower, but rarer checks...
 374                                                         $n = ord( $head );
 375                                                         if(
 376                                                                 # "Overlong sequences" are those that are syntactically
 377                                                                 # correct but use more UTF-8 bytes than are necessary to
 378                                                                 # encode a character. Naïve string comparisons can be
 379                                                                 # tricked into failing to see a match for an ASCII
 380                                                                 # character, for instance, which can be a security hole
 381                                                                 # if blacklist checks are being used.
 382                                                                ($n  < 0xc2 && $sequence <= UTF8_OVERLONG_A)
 383                                                                 || ($n == 0xe0 && $sequence <= UTF8_OVERLONG_B)
 384                                                                 || ($n == 0xf0 && $sequence <= UTF8_OVERLONG_C)
 385
 386                                                                 # U+FFFE and U+FFFF are explicitly forbidden in Unicode.
 387                                                                 || ($n == 0xef &&
 388                                                                            ($sequence == UTF8_FFFE)
 389                                                                         || ($sequence == UTF8_FFFF) )
 390
 391                                                                 # Unicode has been limited to 21 bits; longer
 392                                                                 # sequences are not allowed.
 393                                                                 || ($n >= 0xf0 && $sequence > UTF8_MAX) ) {
 394
 395                                                                 $replace[] = array( UTF8_REPLACEMENT,
 396                                                                                     $base + $i + 1 - strlen( $sequence ),
 397                                                                                     strlen( $sequence ) );
 398                                                                 $head = '';
 399                                                                 continue;
 400                                                         }
 401                                                 }
 402                                         }
 403
 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;
 409                                         }
 410
 411                                         # The sequence is legal!
 412                                         $head = '';
 413                                 } elseif( $c < "\x80" ) {
 414                                         # ASCII byte.
 415                                         $head = '';
 416                                 } elseif( $c < "\xc0" ) {
 417                                         # Illegal tail bytes
 418                                         if( $head == '' ) {
 419                                                 # Out of the blue!
 420                                                 $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 421                                         } else {
 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 );
 426                                         }
 427                                 } else {
 428                                         # Miscellaneous freaks.
 429                                         $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 430                                         $head = '';
 431                                 }
 432                         }
 433                         $base += $chunk;
 434                 }
 435                 if( count( $replace ) ) {
 436                         # There were illegal UTF-8 sequences we need to fix up.
 437                         $out = '';
 438                         $last = 0;
 439                         foreach( $replace as $rep ) {
 440                                 list( $replacement, $start, $length ) = $rep;
 441                                 if( $last < $start ) {
 442                                         $out .= substr( $string, $last, $start - $last );
 443                                 }
 444                                 $out .= $replacement;
 445                                 $last = $start + $length;
 446                         }
 447                         if( $last < strlen( $string ) ) {
 448                                 $out .= substr( $string, $last );
 449                         }
 450                         $string = $out;
 451                 }
 452                 return $looksNormal;
 453         }
 454
 455         # These take a string and run the normalization on them, without
 456         # checking for validity or any optimization etc. Input must be
 457         # VALID UTF-8!
 458         /**
 459          * @param $string string
 460          * @return string
 461          * @private
 462          */
 463         static function NFC( $string ) {
 464                 return UtfNormal::fastCompose( UtfNormal::NFD( $string ) );
 465         }
 466
 467         /**
 468          * @param $string string
 469          * @return string
 470          * @private
 471          */
 472         static function NFD( $string ) {
 473                 UtfNormal::loadData();
 474
 475                 return UtfNormal::fastCombiningSort(
 476                         UtfNormal::fastDecompose( $string, self::$utfCanonicalDecomp ) );
 477         }
 478
 479         /**
 480          * @param $string string
 481          * @return string
 482          * @private
 483          */
 484         static function NFKC( $string ) {
 485                 return UtfNormal::fastCompose( UtfNormal::NFKD( $string ) );
 486         }
 487
 488         /**
 489          * @param $string string
 490          * @return string
 491          * @private
 492          */
 493         static function NFKD( $string ) {
 494                 if( !isset( self::$utfCompatibilityDecomp ) ) {
 495                         require_once( 'UtfNormalDataK.inc' );
 496                 }
 497                 return self::fastCombiningSort(
 498                         self::fastDecompose( $string, self::$utfCompatibilityDecomp ) );
 499         }
 500
 501
 502         /**
 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.
 506          * @private
 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)
 510          */
 511         static function fastDecompose( $string, $map ) {
 512                 UtfNormal::loadData();
 513                 $len = strlen( $string );
 514                 $out = '';
 515                 for( $i = 0; $i < $len; $i++ ) {
 516                         $c = $string{$i};
 517                         $n = ord( $c );
 518                         if( $n < 0x80 ) {
 519                                 # ASCII chars never decompose
 520                                 # THEY ARE IMMORTAL
 521                                 $out .= $c;
 522                                 continue;
 523                         } elseif( $n >= 0xf0 ) {
 524                                 $c = substr( $string, $i, 4 );
 525                                 $i += 3;
 526                         } elseif( $n >= 0xe0 ) {
 527                                 $c = substr( $string, $i, 3 );
 528                                 $i += 2;
 529                         } elseif( $n >= 0xc0 ) {
 530                                 $c = substr( $string, $i, 2 );
 531                                 $i++;
 532                         }
 533                         if( isset( $map[$c] ) ) {
 534                                 $out .= $map[$c];
 535                                 continue;
 536                         } else {
 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.
 542                                         #
 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 );
 551                                         if( $t >= 25 ) {
 552                                                 $out .= "\xe1\x87" . chr( 0x80 + $t - 25 );
 553                                         } elseif( $t ) {
 554                                                 $out .= "\xe1\x86" . chr( 0xa7 + $t );
 555                                         }
 556                                         continue;
 557                                 }
 558                         }
 559                         $out .= $c;
 560                 }
 561                 return $out;
 562         }
 563
 564         /**
 565          * Sorts combining characters into canonical order. This is the
 566          * final step in creating decomposed normal forms D and KD.
 567          * @private
 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
 570          */
 571         static function fastCombiningSort( $string ) {
 572                 UtfNormal::loadData();
 573                 $len = strlen( $string );
 574                 $out = '';
 575                 $combiners = array();
 576                 $lastClass = -1;
 577                 for( $i = 0; $i < $len; $i++ ) {
 578                         $c = $string{$i};
 579                         $n = ord( $c );
 580                         if( $n >= 0x80 ) {
 581                                 if( $n >= 0xf0 ) {
 582                                         $c = substr( $string, $i, 4 );
 583                                         $i += 3;
 584                                 } elseif( $n >= 0xe0 ) {
 585                                         $c = substr( $string, $i, 3 );
 586                                         $i += 2;
 587                                 } elseif( $n >= 0xc0 ) {
 588                                         $c = substr( $string, $i, 2 );
 589                                         $i++;
 590                                 }
 591                                 if( isset( self::$utfCombiningClass[$c] ) ) {
 592                                         $lastClass = self::$utfCombiningClass[$c];
 593                                         if( isset( $combiners[$lastClass] ) ) {
 594                                                 $combiners[$lastClass] .= $c;
 595                                         } else {
 596                                                 $combiners[$lastClass] = $c;
 597                                         }
 598                                         continue;
 599                                 }
 600                         }
 601                         if( $lastClass ) {
 602                                 ksort( $combiners );
 603                                 $out .= implode( '', $combiners );
 604                                 $combiners = array();
 605                         }
 606                         $out .= $c;
 607                         $lastClass = 0;
 608                 }
 609                 if( $lastClass ) {
 610                         ksort( $combiners );
 611                         $out .= implode( '', $combiners );
 612                 }
 613                 return $out;
 614         }
 615
 616         /**
 617          * Produces canonically composed sequences, i.e. normal form C or KC.
 618          *
 619          * @private
 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
 622          */
 623         static function fastCompose( $string ) {
 624                 UtfNormal::loadData();
 625                 $len = strlen( $string );
 626                 $out = '';
 627                 $lastClass = -1;
 628                 $lastHangul = 0;
 629                 $startChar = '';
 630                 $combining = '';
 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++ ) {
 634                         $c = $string{$i};
 635                         $n = ord( $c );
 636                         if( $n < 0x80 ) {
 637                                 # No combining characters here...
 638                                 $out .= $startChar;
 639                                 $out .= $combining;
 640                                 $startChar = $c;
 641                                 $combining = '';
 642                                 $lastClass = 0;
 643                                 continue;
 644                         } elseif( $n >= 0xf0 ) {
 645                                 $c = substr( $string, $i, 4 );
 646                                 $i += 3;
 647                         } elseif( $n >= 0xe0 ) {
 648                                 $c = substr( $string, $i, 3 );
 649                                 $i += 2;
 650                         } elseif( $n >= 0xc0 ) {
 651                                 $c = substr( $string, $i, 2 );
 652                                 $i++;
 653                         }
 654                         $pair = $startChar . $c;
 655                         if( $n > 0x80 ) {
 656                                 if( isset( self::$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 &&
 661                                                 $class > 0 &&
 662                                                 isset( self::$utfCanonicalComp[$pair] ) ) {
 663                                                 $startChar = self::$utfCanonicalComp[$pair];
 664                                                 $class = 0;
 665                                         } else {
 666                                                 $combining .= $c;
 667                                         }
 668                                         $lastClass = $class;
 669                                         $lastHangul = 0;
 670                                         continue;
 671                                 }
 672                         }
 673                         # New start char
 674                         if( $lastClass == 0 ) {
 675                                 if( isset( self::$utfCanonicalComp[$pair] ) ) {
 676                                         $startChar = self::$utfCanonicalComp[$pair];
 677                                         $lastHangul = 0;
 678                                         continue;
 679                                 }
 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.
 686                                         #
 687                                         if( $c >= UTF8_HANGUL_VBASE &&
 688                                                 $c <= UTF8_HANGUL_VEND &&
 689                                                 $startChar >= UTF8_HANGUL_LBASE &&
 690                                                 $startChar <= UTF8_HANGUL_LEND ) {
 691                                                 #
 692                                                 #$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;
 696
 697                                                 $hangulPoint = UNICODE_HANGUL_FIRST +
 698                                                         UNICODE_HANGUL_TCOUNT *
 699                                                         (UNICODE_HANGUL_VCOUNT * $lIndex + $vIndex);
 700
 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 );
 705                                                 $lastHangul = 0;
 706                                                 continue;
 707                                         } elseif( $c >= UTF8_HANGUL_TBASE &&
 708                                                           $c <= UTF8_HANGUL_TEND &&
 709                                                           $startChar >= UTF8_HANGUL_FIRST &&
 710                                                           $startChar <= UTF8_HANGUL_LAST &&
 711                                                           !$lastHangul ) {
 712                                                 # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
 713                                                 $tIndex = ord( $c{2} ) - 0xa7;
 714                                                 if( $tIndex < 0 ) $tIndex = ord( $c{2} ) - 0x80 + (0x11c0 - 0x11a7);
 715
 716                                                 # Increment the code point by $tIndex, without
 717                                                 # the function overhead of decoding and recoding UTF-8
 718                                                 #
 719                                                 $tail = ord( $startChar{2} ) + $tIndex;
 720                                                 if( $tail > 0xbf ) {
 721                                                         $tail -= 0x40;
 722                                                         $mid = ord( $startChar{1} ) + 1;
 723                                                         if( $mid > 0xbf ) {
 724                                                                 $startChar{0} = chr( ord( $startChar{0} ) + 1 );
 725                                                                 $mid -= 0x40;
 726                                                         }
 727                                                         $startChar{1} = chr( $mid );
 728                                                 }
 729                                                 $startChar{2} = chr( $tail );
 730
 731                                                 # If there's another jamo char after this, *don't* try to merge it.
 732                                                 $lastHangul = 1;
 733                                                 continue;
 734                                         }
 735                                 }
 736                         }
 737                         $out .= $startChar;
 738                         $out .= $combining;
 739                         $startChar = $c;
 740                         $combining = '';
 741                         $lastClass = 0;
 742                         $lastHangul = 0;
 743                 }
 744                 $out .= $startChar . $combining;
 745                 return $out;
 746         }
 747
 748         /**
 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
 752          * @return string
 753          */
 754         static function placebo( $string ) {
 755                 $len = strlen( $string );
 756                 $out = '';
 757                 for( $i = 0; $i < $len; $i++ ) {
 758                         $out .= $string{$i};
 759                 }
 760                 return $out;
 761         }
 762         /**
 763          * Function to replace some characters that we don't want
 764          * but most of the native normalize functions keep.
 765          *
 766          * @param $string String The string
 767          * @return String String with the character codes replaced.
 768          */
 769         private static function replaceForNativeNormalize( $string ) {
 770                 $string = preg_replace(
 771                         '/[\x00-\x08\x0b\x0c\x0e-\x1f]/',
 772                         UTF8_REPLACEMENT,
 773                         $string );
 774                 $string = str_replace( UTF8_FFFE, UTF8_REPLACEMENT, $string );
 775                 $string = str_replace( UTF8_FFFF, UTF8_REPLACEMENT, $string );
 776                 return $string;
 777         }
 778 }