includes/normal/UtfNormal.php

   1 <?php
   2 # Copyright (C) 2004 Brion Vibber <brion@pobox.com>
   3 # http://www.mediawiki.org/
   4 #
   5 # This program is free software; you can redistribute it and/or modify
   6 # it under the terms of the GNU General Public License as published by
   7 # the Free Software Foundation; either version 2 of the License, or
   8 # (at your option) any later version.
   9 #
  10 # This program is distributed in the hope that it will be useful,
  11 # but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13 # GNU General Public License for more details.
  14 #
  15 # You should have received a copy of the GNU General Public License along
  16 # with this program; if not, write to the Free Software Foundation, Inc.,
  17 # 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
  18 # http://www.gnu.org/copyleft/gpl.html
  19
  20 /**
  21  * @defgroup UtfNormal UtfNormal
  22  */
  23
  24 /** */
  25 require_once dirname(__FILE__).'/UtfNormalUtil.php';
  26
  27 global $utfCombiningClass, $utfCanonicalComp, $utfCanonicalDecomp;
  28 $utfCombiningClass = null;
  29 $utfCanonicalComp = null;
  30 $utfCanonicalDecomp = null;
  31
  32 # Load compatibility decompositions on demand if they are needed.
  33 global $utfCompatibilityDecomp;
  34 $utfCompatibilityDecomp = null;
  35
  36 /**
  37  * For using the ICU wrapper
  38  */
  39 define( 'UNORM_NONE', 1 );
  40 define( 'UNORM_NFD',  2 );
  41 define( 'UNORM_NFKD', 3 );
  42 define( 'UNORM_NFC',  4 );
  43 define( 'UNORM_DEFAULT', UNORM_NFC );
  44 define( 'UNORM_NFKC', 5 );
  45 define( 'UNORM_FCD',  6 );
  46
  47 define( 'NORMALIZE_ICU', function_exists( 'utf8_normalize' ) );
  48 define( 'NORMALIZE_INTL', function_exists( 'normalizer_normalize' ) );
  49
  50 /**
  51  * Unicode normalization routines for working with UTF-8 strings.
  52  * Currently assumes that input strings are valid UTF-8!
  53  *
  54  * Not as fast as I'd like, but should be usable for most purposes.
  55  * UtfNormal::toNFC() will bail early if given ASCII text or text
  56  * it can quickly deterimine is already normalized.
  57  *
  58  * All functions can be called static.
  59  *
  60  * See description of forms at http://www.unicode.org/reports/tr15/
  61  *
  62  * @ingroup UtfNormal
  63  */
  64 class UtfNormal {
  65         /**
  66          * The ultimate convenience function! Clean up invalid UTF-8 sequences,
  67          * and convert to normal form C, canonical composition.
  68          *
  69          * Fast return for pure ASCII strings; some lesser optimizations for
  70          * strings containing only known-good characters. Not as fast as toNFC().
  71          *
  72          * @param $string String: a UTF-8 string
  73          * @return string a clean, shiny, normalized UTF-8 string
  74          */
  75         static function cleanUp( $string ) {
  76                 if( NORMALIZE_ICU || NORMALIZE_INTL ) {
  77                         # We exclude a few chars that ICU would not.
  78                         $string = preg_replace(
  79                                 '/[\x00-\x08\x0b\x0c\x0e-\x1f]/',
  80                                 UTF8_REPLACEMENT,
  81                                 $string );
  82                         $string = str_replace( UTF8_FFFE, UTF8_REPLACEMENT, $string );
  83                         $string = str_replace( UTF8_FFFF, UTF8_REPLACEMENT, $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                         if ( NORMALIZE_ICU ) return rtrim( utf8_normalize( $string . "\x01", UNORM_NFC ), "\x01" );
  88                         if ( NORMALIZE_INTL ) return normalizer_normalize( $string, Normalizer::FORM_C );
  89                 } elseif( UtfNormal::quickIsNFCVerify( $string ) ) {
  90                         # Side effect -- $string has had UTF-8 errors cleaned up.
  91                         return $string;
  92                 } else {
  93                         return UtfNormal::NFC( $string );
  94                 }
  95         }
  96
  97         /**
  98          * Convert a UTF-8 string to normal form C, canonical composition.
  99          * Fast return for pure ASCII strings; some lesser optimizations for
 100          * strings containing only known-good characters.
 101          *
 102          * @param $string String: a valid UTF-8 string. Input is not validated.
 103          * @return string a UTF-8 string in normal form C
 104          */
 105         static function toNFC( $string ) {
 106                 if( NORMALIZE_INTL )
 107                         return normalizer_normalize( $string, Normalizer::FORM_C );
 108                 elseif( NORMALIZE_ICU )
 109                         return utf8_normalize( $string, UNORM_NFC );
 110                 elseif( UtfNormal::quickIsNFC( $string ) )
 111                         return $string;
 112                 else
 113                         return UtfNormal::NFC( $string );
 114         }
 115
 116         /**
 117          * Convert a UTF-8 string to normal form D, canonical decomposition.
 118          * Fast return for pure ASCII strings.
 119          *
 120          * @param $string String: a valid UTF-8 string. Input is not validated.
 121          * @return string a UTF-8 string in normal form D
 122          */
 123         static function toNFD( $string ) {
 124                 if( NORMALIZE_INTL )
 125                         return normalizer_normalize( $string, Normalizer::FORM_D );
 126                 elseif( NORMALIZE_ICU )
 127                         return utf8_normalize( $string, UNORM_NFD );
 128                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 129                         return UtfNormal::NFD( $string );
 130                 else
 131                         return $string;
 132         }
 133
 134         /**
 135          * Convert a UTF-8 string to normal form KC, compatibility composition.
 136          * This may cause irreversible information loss, use judiciously.
 137          * Fast return for pure ASCII strings.
 138          *
 139          * @param $string String: a valid UTF-8 string. Input is not validated.
 140          * @return string a UTF-8 string in normal form KC
 141          */
 142         static function toNFKC( $string ) {
 143                 if( NORMALIZE_INTL )
 144                         return normalizer_normalize( $string, Normalizer::FORM_KC );
 145                 elseif( NORMALIZE_ICU )
 146                         return utf8_normalize( $string, UNORM_NFKC );
 147                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 148                         return UtfNormal::NFKC( $string );
 149                 else
 150                         return $string;
 151         }
 152
 153         /**
 154          * Convert a UTF-8 string to normal form KD, compatibility decomposition.
 155          * This may cause irreversible information loss, use judiciously.
 156          * Fast return for pure ASCII strings.
 157          *
 158          * @param $string String: a valid UTF-8 string. Input is not validated.
 159          * @return string a UTF-8 string in normal form KD
 160          */
 161         static function toNFKD( $string ) {
 162                 if( NORMALIZE_INTL )
 163                         return normalizer_normalize( $string, Normalizer::FORM_KD );
 164                 elseif( NORMALIZE_ICU )
 165                         return utf8_normalize( $string, UNORM_NFKD );
 166                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 167                         return UtfNormal::NFKD( $string );
 168                 else
 169                         return $string;
 170         }
 171
 172         /**
 173          * Load the basic composition data if necessary
 174          * @private
 175          */
 176         static function loadData() {
 177                 global $utfCombiningClass;
 178                 if( !isset( $utfCombiningClass ) ) {
 179                         require_once( dirname(__FILE__) . '/UtfNormalData.inc' );
 180                 }
 181         }
 182
 183         /**
 184          * Returns true if the string is _definitely_ in NFC.
 185          * Returns false if not or uncertain.
 186          * @param $string String: a valid UTF-8 string. Input is not validated.
 187          * @return bool
 188          */
 189         static function quickIsNFC( $string ) {
 190                 # ASCII is always valid NFC!
 191                 # If it's pure ASCII, let it through.
 192                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 193
 194                 UtfNormal::loadData();
 195                 global $utfCheckNFC, $utfCombiningClass;
 196                 $len = strlen( $string );
 197                 for( $i = 0; $i < $len; $i++ ) {
 198                         $c = $string{$i};
 199                         $n = ord( $c );
 200                         if( $n < 0x80 ) {
 201                                 continue;
 202                         } elseif( $n >= 0xf0 ) {
 203                                 $c = substr( $string, $i, 4 );
 204                                 $i += 3;
 205                         } elseif( $n >= 0xe0 ) {
 206                                 $c = substr( $string, $i, 3 );
 207                                 $i += 2;
 208                         } elseif( $n >= 0xc0 ) {
 209                                 $c = substr( $string, $i, 2 );
 210                                 $i++;
 211                         }
 212                         if( isset( $utfCheckNFC[$c] ) ) {
 213                                 # If it's NO or MAYBE, bail and do the slow check.
 214                                 return false;
 215                         }
 216                         if( isset( $utfCombiningClass[$c] ) ) {
 217                                 # Combining character? We might have to do sorting, at least.
 218                                 return false;
 219                         }
 220                 }
 221                 return true;
 222         }
 223
 224         /**
 225          * Returns true if the string is _definitely_ in NFC.
 226          * Returns false if not or uncertain.
 227          * @param $string String: a UTF-8 string, altered on output to be valid UTF-8 safe for XML.
 228          */
 229         static function quickIsNFCVerify( &$string ) {
 230                 # Screen out some characters that eg won't be allowed in XML
 231                 $string = preg_replace( '/[\x00-\x08\x0b\x0c\x0e-\x1f]/', UTF8_REPLACEMENT, $string );
 232
 233                 # ASCII is always valid NFC!
 234                 # If we're only ever given plain ASCII, we can avoid the overhead
 235                 # of initializing the decomposition tables by skipping out early.
 236                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 237
 238                 static $checkit = null, $tailBytes = null, $utfCheckOrCombining = null;
 239                 if( !isset( $checkit ) ) {
 240                         # Load/build some scary lookup tables...
 241                         UtfNormal::loadData();
 242                         global $utfCheckNFC, $utfCombiningClass;
 243
 244                         $utfCheckOrCombining = array_merge( $utfCheckNFC, $utfCombiningClass );
 245
 246                         # Head bytes for sequences which we should do further validity checks
 247                         $checkit = array_flip( array_map( 'chr',
 248                                         array( 0xc0, 0xc1, 0xe0, 0xed, 0xef,
 249                                                    0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7,
 250                                                    0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff ) ) );
 251
 252                         # Each UTF-8 head byte is followed by a certain
 253                         # number of tail bytes.
 254                         $tailBytes = array();
 255                         for( $n = 0; $n < 256; $n++ ) {
 256                                 if( $n < 0xc0 ) {
 257                                         $remaining = 0;
 258                                 } elseif( $n < 0xe0 ) {
 259                                         $remaining = 1;
 260                                 } elseif( $n < 0xf0 ) {
 261                                         $remaining = 2;
 262                                 } elseif( $n < 0xf8 ) {
 263                                         $remaining = 3;
 264                                 } elseif( $n < 0xfc ) {
 265                                         $remaining = 4;
 266                                 } elseif( $n < 0xfe ) {
 267                                         $remaining = 5;
 268                                 } else {
 269                                         $remaining = 0;
 270                                 }
 271                                 $tailBytes[chr($n)] = $remaining;
 272                         }
 273                 }
 274
 275                 # Chop the text into pure-ASCII and non-ASCII areas;
 276                 # large ASCII parts can be handled much more quickly.
 277                 # Don't chop up Unicode areas for punctuation, though,
 278                 # that wastes energy.
 279                 $matches = array();
 280                 preg_match_all(
 281                         '/([\x00-\x7f]+|[\x80-\xff][\x00-\x40\x5b-\x5f\x7b-\xff]*)/',
 282                         $string, $matches );
 283
 284                 $looksNormal = true;
 285                 $base = 0;
 286                 $replace = array();
 287                 foreach( $matches[1] as $str ) {
 288                         $chunk = strlen( $str );
 289
 290                         if( $str{0} < "\x80" ) {
 291                                 # ASCII chunk: guaranteed to be valid UTF-8
 292                                 # and in normal form C, so skip over it.
 293                                 $base += $chunk;
 294                                 continue;
 295                         }
 296
 297                         # We'll have to examine the chunk byte by byte to ensure
 298                         # that it consists of valid UTF-8 sequences, and to see
 299                         # if any of them might not be normalized.
 300                         #
 301                         # Since PHP is not the fastest language on earth, some of
 302                         # this code is a little ugly with inner loop optimizations.
 303
 304                         $head = '';
 305                         $len = $chunk + 1; # Counting down is faster. I'm *so* sorry.
 306
 307                         for( $i = -1; --$len; ) {
 308                                 if( $remaining = $tailBytes[$c = $str{++$i}] ) {
 309                                         # UTF-8 head byte!
 310                                         $sequence = $head = $c;
 311                                         do {
 312                                                 # Look for the defined number of tail bytes...
 313                                                 if( --$len && ( $c = $str{++$i} ) >= "\x80" && $c < "\xc0" ) {
 314                                                         # Legal tail bytes are nice.
 315                                                         $sequence .= $c;
 316                                                 } else {
 317                                                         if( 0 == $len ) {
 318                                                                 # Premature end of string!
 319                                                                 # Drop a replacement character into output to
 320                                                                 # represent the invalid UTF-8 sequence.
 321                                                                 $replace[] = array( UTF8_REPLACEMENT,
 322                                                                                                         $base + $i + 1 - strlen( $sequence ),
 323                                                                                                         strlen( $sequence ) );
 324                                                                 break 2;
 325                                                         } else {
 326                                                                 # Illegal tail byte; abandon the sequence.
 327                                                                 $replace[] = array( UTF8_REPLACEMENT,
 328                                                                                                         $base + $i - strlen( $sequence ),
 329                                                                                                         strlen( $sequence ) );
 330                                                                 # Back up and reprocess this byte; it may itself
 331                                                                 # be a legal ASCII or UTF-8 sequence head.
 332                                                                 --$i;
 333                                                                 ++$len;
 334                                                                 continue 2;
 335                                                         }
 336                                                 }
 337                                         } while( --$remaining );
 338
 339                                         if( isset( $checkit[$head] ) ) {
 340                                                 # Do some more detailed validity checks, for
 341                                                 # invalid characters and illegal sequences.
 342                                                 if( $head == "\xed" ) {
 343                                                         # 0xed is relatively frequent in Korean, which
 344                                                         # abuts the surrogate area, so we're doing
 345                                                         # this check separately to speed things up.
 346
 347                                                         if( $sequence >= UTF8_SURROGATE_FIRST ) {
 348                                                                 # Surrogates are legal only in UTF-16 code.
 349                                                                 # They are totally forbidden here in UTF-8
 350                                                                 # utopia.
 351                                                                 $replace[] = array( UTF8_REPLACEMENT,
 352                                                                              $base + $i + 1 - strlen( $sequence ),
 353                                                                              strlen( $sequence ) );
 354                                                                 $head = '';
 355                                                                 continue;
 356                                                         }
 357                                                 } else {
 358                                                         # Slower, but rarer checks...
 359                                                         $n = ord( $head );
 360                                                         if(
 361                                                                 # "Overlong sequences" are those that are syntactically
 362                                                                 # correct but use more UTF-8 bytes than are necessary to
 363                                                                 # encode a character. Naïve string comparisons can be
 364                                                                 # tricked into failing to see a match for an ASCII
 365                                                                 # character, for instance, which can be a security hole
 366                                                                 # if blacklist checks are being used.
 367                                                                ($n  < 0xc2 && $sequence <= UTF8_OVERLONG_A)
 368                                                                 || ($n == 0xe0 && $sequence <= UTF8_OVERLONG_B)
 369                                                                 || ($n == 0xf0 && $sequence <= UTF8_OVERLONG_C)
 370
 371                                                                 # U+FFFE and U+FFFF are explicitly forbidden in Unicode.
 372                                                                 || ($n == 0xef &&
 373                                                                            ($sequence == UTF8_FFFE)
 374                                                                         || ($sequence == UTF8_FFFF) )
 375
 376                                                                 # Unicode has been limited to 21 bits; longer
 377                                                                 # sequences are not allowed.
 378                                                                 || ($n >= 0xf0 && $sequence > UTF8_MAX) ) {
 379
 380                                                                 $replace[] = array( UTF8_REPLACEMENT,
 381                                                                                     $base + $i + 1 - strlen( $sequence ),
 382                                                                                     strlen( $sequence ) );
 383                                                                 $head = '';
 384                                                                 continue;
 385                                                         }
 386                                                 }
 387                                         }
 388
 389                                         if( isset( $utfCheckOrCombining[$sequence] ) ) {
 390                                                 # If it's NO or MAYBE, we'll have to rip
 391                                                 # the string apart and put it back together.
 392                                                 # That's going to be mighty slow.
 393                                                 $looksNormal = false;
 394                                         }
 395
 396                                         # The sequence is legal!
 397                                         $head = '';
 398                                 } elseif( $c < "\x80" ) {
 399                                         # ASCII byte.
 400                                         $head = '';
 401                                 } elseif( $c < "\xc0" ) {
 402                                         # Illegal tail bytes
 403                                         if( $head == '' ) {
 404                                                 # Out of the blue!
 405                                                 $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 406                                         } else {
 407                                                 # Don't add if we're continuing a broken sequence;
 408                                                 # we already put a replacement character when we looked
 409                                                 # at the broken sequence.
 410                                                 $replace[] = array( '', $base + $i, 1 );
 411                                         }
 412                                 } else {
 413                                         # Miscellaneous freaks.
 414                                         $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 415                                         $head = '';
 416                                 }
 417                         }
 418                         $base += $chunk;
 419                 }
 420                 if( count( $replace ) ) {
 421                         # There were illegal UTF-8 sequences we need to fix up.
 422                         $out = '';
 423                         $last = 0;
 424                         foreach( $replace as $rep ) {
 425                                 list( $replacement, $start, $length ) = $rep;
 426                                 if( $last < $start ) {
 427                                         $out .= substr( $string, $last, $start - $last );
 428                                 }
 429                                 $out .= $replacement;
 430                                 $last = $start + $length;
 431                         }
 432                         if( $last < strlen( $string ) ) {
 433                                 $out .= substr( $string, $last );
 434                         }
 435                         $string = $out;
 436                 }
 437                 return $looksNormal;
 438         }
 439
 440         # These take a string and run the normalization on them, without
 441         # checking for validity or any optimization etc. Input must be
 442         # VALID UTF-8!
 443         /**
 444          * @param $string string
 445          * @return string
 446          * @private
 447          */
 448         static function NFC( $string ) {
 449                 return UtfNormal::fastCompose( UtfNormal::NFD( $string ) );
 450         }
 451
 452         /**
 453          * @param $string string
 454          * @return string
 455          * @private
 456          */
 457         static function NFD( $string ) {
 458                 UtfNormal::loadData();
 459                 global $utfCanonicalDecomp;
 460                 return UtfNormal::fastCombiningSort(
 461                         UtfNormal::fastDecompose( $string, $utfCanonicalDecomp ) );
 462         }
 463
 464         /**
 465          * @param $string string
 466          * @return string
 467          * @private
 468          */
 469         static function NFKC( $string ) {
 470                 return UtfNormal::fastCompose( UtfNormal::NFKD( $string ) );
 471         }
 472
 473         /**
 474          * @param $string string
 475          * @return string
 476          * @private
 477          */
 478         static function NFKD( $string ) {
 479                 global $utfCompatibilityDecomp;
 480                 if( !isset( $utfCompatibilityDecomp ) ) {
 481                         require_once( 'UtfNormalDataK.inc' );
 482                 }
 483                 return UtfNormal::fastCombiningSort(
 484                         UtfNormal::fastDecompose( $string, $utfCompatibilityDecomp ) );
 485         }
 486
 487
 488         /**
 489          * Perform decomposition of a UTF-8 string into either D or KD form
 490          * (depending on which decomposition map is passed to us).
 491          * Input is assumed to be *valid* UTF-8. Invalid code will break.
 492          * @private
 493          * @param $string String: valid UTF-8 string
 494          * @param $map Array: hash of expanded decomposition map
 495          * @return string a UTF-8 string decomposed, not yet normalized (needs sorting)
 496          */
 497         static function fastDecompose( $string, $map ) {
 498                 UtfNormal::loadData();
 499                 $len = strlen( $string );
 500                 $out = '';
 501                 for( $i = 0; $i < $len; $i++ ) {
 502                         $c = $string{$i};
 503                         $n = ord( $c );
 504                         if( $n < 0x80 ) {
 505                                 # ASCII chars never decompose
 506                                 # THEY ARE IMMORTAL
 507                                 $out .= $c;
 508                                 continue;
 509                         } elseif( $n >= 0xf0 ) {
 510                                 $c = substr( $string, $i, 4 );
 511                                 $i += 3;
 512                         } elseif( $n >= 0xe0 ) {
 513                                 $c = substr( $string, $i, 3 );
 514                                 $i += 2;
 515                         } elseif( $n >= 0xc0 ) {
 516                                 $c = substr( $string, $i, 2 );
 517                                 $i++;
 518                         }
 519                         if( isset( $map[$c] ) ) {
 520                                 $out .= $map[$c];
 521                                 continue;
 522                         } else {
 523                                 if( $c >= UTF8_HANGUL_FIRST && $c <= UTF8_HANGUL_LAST ) {
 524                                         # Decompose a hangul syllable into jamo;
 525                                         # hardcoded for three-byte UTF-8 sequence.
 526                                         # A lookup table would be slightly faster,
 527                                         # but adds a lot of memory & disk needs.
 528                                         #
 529                                         $index = ( (ord( $c{0} ) & 0x0f) << 12
 530                                                  | (ord( $c{1} ) & 0x3f) <<  6
 531                                                  | (ord( $c{2} ) & 0x3f) )
 532                                                - UNICODE_HANGUL_FIRST;
 533                                         $l = intval( $index / UNICODE_HANGUL_NCOUNT );
 534                                         $v = intval( ($index % UNICODE_HANGUL_NCOUNT) / UNICODE_HANGUL_TCOUNT);
 535                                         $t = $index % UNICODE_HANGUL_TCOUNT;
 536                                         $out .= "\xe1\x84" . chr( 0x80 + $l ) . "\xe1\x85" . chr( 0xa1 + $v );
 537                                         if( $t >= 25 ) {
 538                                                 $out .= "\xe1\x87" . chr( 0x80 + $t - 25 );
 539                                         } elseif( $t ) {
 540                                                 $out .= "\xe1\x86" . chr( 0xa7 + $t );
 541                                         }
 542                                         continue;
 543                                 }
 544                         }
 545                         $out .= $c;
 546                 }
 547                 return $out;
 548         }
 549
 550         /**
 551          * Sorts combining characters into canonical order. This is the
 552          * final step in creating decomposed normal forms D and KD.
 553          * @private
 554          * @param $string String: a valid, decomposed UTF-8 string. Input is not validated.
 555          * @return string a UTF-8 string with combining characters sorted in canonical order
 556          */
 557         static function fastCombiningSort( $string ) {
 558                 UtfNormal::loadData();
 559                 global $utfCombiningClass;
 560                 $len = strlen( $string );
 561                 $out = '';
 562                 $combiners = array();
 563                 $lastClass = -1;
 564                 for( $i = 0; $i < $len; $i++ ) {
 565                         $c = $string{$i};
 566                         $n = ord( $c );
 567                         if( $n >= 0x80 ) {
 568                                 if( $n >= 0xf0 ) {
 569                                         $c = substr( $string, $i, 4 );
 570                                         $i += 3;
 571                                 } elseif( $n >= 0xe0 ) {
 572                                         $c = substr( $string, $i, 3 );
 573                                         $i += 2;
 574                                 } elseif( $n >= 0xc0 ) {
 575                                         $c = substr( $string, $i, 2 );
 576                                         $i++;
 577                                 }
 578                                 if( isset( $utfCombiningClass[$c] ) ) {
 579                                         $lastClass = $utfCombiningClass[$c];
 580                                         if( isset( $combiners[$lastClass] ) ) {
 581                                                 $combiners[$lastClass] .= $c;
 582                                         } else {
 583                                                 $combiners[$lastClass] = $c;
 584                                         }
 585                                         continue;
 586                                 }
 587                         }
 588                         if( $lastClass ) {
 589                                 ksort( $combiners );
 590                                 $out .= implode( '', $combiners );
 591                                 $combiners = array();
 592                         }
 593                         $out .= $c;
 594                         $lastClass = 0;
 595                 }
 596                 if( $lastClass ) {
 597                         ksort( $combiners );
 598                         $out .= implode( '', $combiners );
 599                 }
 600                 return $out;
 601         }
 602
 603         /**
 604          * Produces canonically composed sequences, i.e. normal form C or KC.
 605          *
 606          * @private
 607          * @param $string String: a valid UTF-8 string in sorted normal form D or KD. Input is not validated.
 608          * @return string a UTF-8 string with canonical precomposed characters used where possible
 609          */
 610         static function fastCompose( $string ) {
 611                 UtfNormal::loadData();
 612                 global $utfCanonicalComp, $utfCombiningClass;
 613                 $len = strlen( $string );
 614                 $out = '';
 615                 $lastClass = -1;
 616                 $lastHangul = 0;
 617                 $startChar = '';
 618                 $combining = '';
 619                 $x1 = ord(substr(UTF8_HANGUL_VBASE,0,1));
 620                 $x2 = ord(substr(UTF8_HANGUL_TEND,0,1));
 621                 for( $i = 0; $i < $len; $i++ ) {
 622                         $c = $string{$i};
 623                         $n = ord( $c );
 624                         if( $n < 0x80 ) {
 625                                 # No combining characters here...
 626                                 $out .= $startChar;
 627                                 $out .= $combining;
 628                                 $startChar = $c;
 629                                 $combining = '';
 630                                 $lastClass = 0;
 631                                 continue;
 632                         } elseif( $n >= 0xf0 ) {
 633                                 $c = substr( $string, $i, 4 );
 634                                 $i += 3;
 635                         } elseif( $n >= 0xe0 ) {
 636                                 $c = substr( $string, $i, 3 );
 637                                 $i += 2;
 638                         } elseif( $n >= 0xc0 ) {
 639                                 $c = substr( $string, $i, 2 );
 640                                 $i++;
 641                         }
 642                         $pair = $startChar . $c;
 643                         if( $n > 0x80 ) {
 644                                 if( isset( $utfCombiningClass[$c] ) ) {
 645                                         # A combining char; see what we can do with it
 646                                         $class = $utfCombiningClass[$c];
 647                                         if( !empty( $startChar ) &&
 648                                                 $lastClass < $class &&
 649                                                 $class > 0 &&
 650                                                 isset( $utfCanonicalComp[$pair] ) ) {
 651                                                 $startChar = $utfCanonicalComp[$pair];
 652                                                 $class = 0;
 653                                         } else {
 654                                                 $combining .= $c;
 655                                         }
 656                                         $lastClass = $class;
 657                                         $lastHangul = 0;
 658                                         continue;
 659                                 }
 660                         }
 661                         # New start char
 662                         if( $lastClass == 0 ) {
 663                                 if( isset( $utfCanonicalComp[$pair] ) ) {
 664                                         $startChar = $utfCanonicalComp[$pair];
 665                                         $lastHangul = 0;
 666                                         continue;
 667                                 }
 668                                 if( $n >= $x1 && $n <= $x2 ) {
 669                                         # WARNING: Hangul code is painfully slow.
 670                                         # I apologize for this ugly, ugly code; however
 671                                         # performance is even more teh suck if we call
 672                                         # out to nice clean functions. Lookup tables are
 673                                         # marginally faster, but require a lot of space.
 674                                         #
 675                                         if( $c >= UTF8_HANGUL_VBASE &&
 676                                                 $c <= UTF8_HANGUL_VEND &&
 677                                                 $startChar >= UTF8_HANGUL_LBASE &&
 678                                                 $startChar <= UTF8_HANGUL_LEND ) {
 679                                                 #
 680                                                 #$lIndex = utf8ToCodepoint( $startChar ) - UNICODE_HANGUL_LBASE;
 681                                                 #$vIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_VBASE;
 682                                                 $lIndex = ord( $startChar{2} ) - 0x80;
 683                                                 $vIndex = ord( $c{2}         ) - 0xa1;
 684
 685                                                 $hangulPoint = UNICODE_HANGUL_FIRST +
 686                                                         UNICODE_HANGUL_TCOUNT *
 687                                                         (UNICODE_HANGUL_VCOUNT * $lIndex + $vIndex);
 688
 689                                                 # Hardcode the limited-range UTF-8 conversion:
 690                                                 $startChar = chr( $hangulPoint >> 12 & 0x0f | 0xe0 ) .
 691                                                                          chr( $hangulPoint >>  6 & 0x3f | 0x80 ) .
 692                                                                          chr( $hangulPoint       & 0x3f | 0x80 );
 693                                                 $lastHangul = 0;
 694                                                 continue;
 695                                         } elseif( $c >= UTF8_HANGUL_TBASE &&
 696                                                           $c <= UTF8_HANGUL_TEND &&
 697                                                           $startChar >= UTF8_HANGUL_FIRST &&
 698                                                           $startChar <= UTF8_HANGUL_LAST &&
 699                                                           !$lastHangul ) {
 700                                                 # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
 701                                                 $tIndex = ord( $c{2} ) - 0xa7;
 702                                                 if( $tIndex < 0 ) $tIndex = ord( $c{2} ) - 0x80 + (0x11c0 - 0x11a7);
 703
 704                                                 # Increment the code point by $tIndex, without
 705                                                 # the function overhead of decoding and recoding UTF-8
 706                                                 #
 707                                                 $tail = ord( $startChar{2} ) + $tIndex;
 708                                                 if( $tail > 0xbf ) {
 709                                                         $tail -= 0x40;
 710                                                         $mid = ord( $startChar{1} ) + 1;
 711                                                         if( $mid > 0xbf ) {
 712                                                                 $startChar{0} = chr( ord( $startChar{0} ) + 1 );
 713                                                                 $mid -= 0x40;
 714                                                         }
 715                                                         $startChar{1} = chr( $mid );
 716                                                 }
 717                                                 $startChar{2} = chr( $tail );
 718
 719                                                 # If there's another jamo char after this, *don't* try to merge it.
 720                                                 $lastHangul = 1;
 721                                                 continue;
 722                                         }
 723                                 }
 724                         }
 725                         $out .= $startChar;
 726                         $out .= $combining;
 727                         $startChar = $c;
 728                         $combining = '';
 729                         $lastClass = 0;
 730                         $lastHangul = 0;
 731                 }
 732                 $out .= $startChar . $combining;
 733                 return $out;
 734         }
 735
 736         /**
 737          * This is just used for the benchmark, comparing how long it takes to
 738          * interate through a string without really doing anything of substance.
 739          * @param $string string
 740          * @return string
 741          */
 742         static function placebo( $string ) {
 743                 $len = strlen( $string );
 744                 $out = '';
 745                 for( $i = 0; $i < $len; $i++ ) {
 746                         $out .= $string{$i};
 747                 }
 748                 return $out;
 749         }
 750 }