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 # 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  18 # http://www.gnu.org/copyleft/gpl.html
  19
  20 /**
  21  * Unicode normalization routines for working with UTF-8 strings.
  22  * Currently assumes that input strings are valid UTF-8!
  23  *
  24  * Not as fast as I'd like, but should be usable for most purposes.
  25  * UtfNormal::toNFC() will bail early if given ASCII text or text
  26  * it can quickly deterimine is already normalized.
  27  *
  28  * All functions can be called static.
  29  *
  30  * See description of forms at http://www.unicode.org/reports/tr15/
  31  *
  32  * @package UtfNormal
  33  */
  34
  35 /** */
  36 require_once 'UtfNormalUtil.php';
  37
  38 global $utfCombiningClass, $utfCanonicalComp, $utfCanonicalDecomp;
  39 $utfCombiningClass = NULL;
  40 $utfCanonicalComp = NULL;
  41 $utfCanonicalDecomp = NULL;
  42
  43 # Load compatibility decompositions on demand if they are needed.
  44 global $utfCompatibilityDecomp;
  45 $utfCompatibilityDecomp = NULL;
  46
  47 define( 'UNICODE_HANGUL_FIRST', 0xac00 );
  48 define( 'UNICODE_HANGUL_LAST',  0xd7a3 );
  49
  50 define( 'UNICODE_HANGUL_LBASE', 0x1100 );
  51 define( 'UNICODE_HANGUL_VBASE', 0x1161 );
  52 define( 'UNICODE_HANGUL_TBASE', 0x11a7 );
  53
  54 define( 'UNICODE_HANGUL_LCOUNT', 19 );
  55 define( 'UNICODE_HANGUL_VCOUNT', 21 );
  56 define( 'UNICODE_HANGUL_TCOUNT', 28 );
  57 define( 'UNICODE_HANGUL_NCOUNT', UNICODE_HANGUL_VCOUNT * UNICODE_HANGUL_TCOUNT );
  58
  59 define( 'UNICODE_HANGUL_LEND', UNICODE_HANGUL_LBASE + UNICODE_HANGUL_LCOUNT - 1 );
  60 define( 'UNICODE_HANGUL_VEND', UNICODE_HANGUL_VBASE + UNICODE_HANGUL_VCOUNT - 1 );
  61 define( 'UNICODE_HANGUL_TEND', UNICODE_HANGUL_TBASE + UNICODE_HANGUL_TCOUNT - 1 );
  62
  63 define( 'UNICODE_SURROGATE_FIRST', 0xd800 );
  64 define( 'UNICODE_SURROGATE_LAST', 0xdfff );
  65 define( 'UNICODE_MAX', 0x10ffff );
  66 define( 'UNICODE_REPLACEMENT', 0xfffd );
  67
  68
  69 define( 'UTF8_HANGUL_FIRST', codepointToUtf8( UNICODE_HANGUL_FIRST ) );
  70 define( 'UTF8_HANGUL_LAST', codepointToUtf8( UNICODE_HANGUL_LAST ) );
  71
  72 define( 'UTF8_HANGUL_LBASE', codepointToUtf8( UNICODE_HANGUL_LBASE ) );
  73 define( 'UTF8_HANGUL_VBASE', codepointToUtf8( UNICODE_HANGUL_VBASE ) );
  74 define( 'UTF8_HANGUL_TBASE', codepointToUtf8( UNICODE_HANGUL_TBASE ) );
  75
  76 define( 'UTF8_HANGUL_LEND', codepointToUtf8( UNICODE_HANGUL_LEND ) );
  77 define( 'UTF8_HANGUL_VEND', codepointToUtf8( UNICODE_HANGUL_VEND ) );
  78 define( 'UTF8_HANGUL_TEND', codepointToUtf8( UNICODE_HANGUL_TEND ) );
  79
  80 define( 'UTF8_SURROGATE_FIRST', codepointToUtf8( UNICODE_SURROGATE_FIRST ) );
  81 define( 'UTF8_SURROGATE_LAST', codepointToUtf8( UNICODE_SURROGATE_LAST ) );
  82 define( 'UTF8_MAX', codepointToUtf8( UNICODE_MAX ) );
  83 define( 'UTF8_REPLACEMENT', codepointToUtf8( UNICODE_REPLACEMENT ) );
  84 #define( 'UTF8_REPLACEMENT', '!' );
  85
  86 define( 'UTF8_OVERLONG_A', "\xc1\xbf" );
  87 define( 'UTF8_OVERLONG_B', "\xe0\x9f\xbf" );
  88 define( 'UTF8_OVERLONG_C', "\xf0\x8f\xbf\xbf" );
  89
  90 # These two ranges are illegal
  91 define( 'UTF8_FDD0', codepointToUtf8( 0xfdd0 ) );
  92 define( 'UTF8_FDEF', codepointToUtf8( 0xfdef ) );
  93 define( 'UTF8_FFFE', codepointToUtf8( 0xfffe ) );
  94 define( 'UTF8_FFFF', codepointToUtf8( 0xffff ) );
  95
  96 define( 'UTF8_HEAD', false );
  97 define( 'UTF8_TAIL', true );
  98
  99
 100 /**
 101  * For using the ICU wrapper
 102  */
 103 define( 'UNORM_NONE', 1 );
 104 define( 'UNORM_NFD',  2 );
 105 define( 'UNORM_NFKD', 3 );
 106 define( 'UNORM_NFC',  4 );
 107 define( 'UNORM_DEFAULT', UNORM_NFC );
 108 define( 'UNORM_NFKC', 5 );
 109 define( 'UNORM_FCD',  6 );
 110
 111 define( 'NORMALIZE_ICU', function_exists( 'utf8_normalize' ) );
 112
 113 /**
 114  *
 115  * @package MediaWiki
 116  */
 117 class UtfNormal {
 118         /**
 119          * The ultimate convenience function! Clean up invalid UTF-8 sequences,
 120          * and convert to normal form C, canonical composition.
 121          *
 122          * Fast return for pure ASCII strings; some lesser optimizations for
 123          * strings containing only known-good characters. Not as fast as toNFC().
 124          *
 125          * @param string $string a UTF-8 string
 126          * @return string a clean, shiny, normalized UTF-8 string
 127          */
 128         function cleanUp( $string ) {
 129                 if( UtfNormal::quickIsNFCVerify( $string ) )
 130                         return $string;
 131                 else
 132                         return UtfNormal::NFC( $string );
 133         }
 134
 135         /**
 136          * Convert a UTF-8 string to normal form C, canonical composition.
 137          * Fast return for pure ASCII strings; some lesser optimizations for
 138          * strings containing only known-good characters.
 139          *
 140          * @param string $string a valid UTF-8 string. Input is not validated.
 141          * @return string a UTF-8 string in normal form C
 142          */
 143         function toNFC( $string ) {
 144                 if( NORMALIZE_ICU )
 145                         return utf8_normalize( $string, UNORM_NFC );
 146                 elseif( UtfNormal::quickIsNFC( $string ) )
 147                         return $string;
 148                 else
 149                         return UtfNormal::NFC( $string );
 150         }
 151
 152         /**
 153          * Convert a UTF-8 string to normal form D, canonical decomposition.
 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 D
 158          */
 159         function toNFD( $string ) {
 160                 if( NORMALIZE_ICU )
 161                         return utf8_normalize( $string, UNORM_NFD );
 162                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 163                         return UtfNormal::NFD( $string );
 164                 else
 165                         return $string;
 166         }
 167
 168         /**
 169          * Convert a UTF-8 string to normal form KC, compatibility composition.
 170          * This may cause irreversible information loss, use judiciously.
 171          * Fast return for pure ASCII strings.
 172          *
 173          * @param string $string a valid UTF-8 string. Input is not validated.
 174          * @return string a UTF-8 string in normal form KC
 175          */
 176         function toNFKC( $string ) {
 177                 if( NORMALIZE_ICU )
 178                         return utf8_normalize( $string, UNORM_NFKC );
 179                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 180                         return UtfNormal::NFKC( $string );
 181                 else
 182                         return $string;
 183         }
 184
 185         /**
 186          * Convert a UTF-8 string to normal form KD, compatibility decomposition.
 187          * This may cause irreversible information loss, use judiciously.
 188          * Fast return for pure ASCII strings.
 189          *
 190          * @param string $string a valid UTF-8 string. Input is not validated.
 191          * @return string a UTF-8 string in normal form KD
 192          */
 193         function toNFKD( $string ) {
 194                 if( NORMALIZE_ICU )
 195                         return utf8_normalize( $string, UNORM_NFKD );
 196                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 197                         return UtfNormal::NFKD( $string );
 198                 else
 199                         return $string;
 200         }
 201
 202         /**
 203          * Load the basic composition data if necessary
 204          * @access private
 205          */
 206         function loadData() {
 207                 global $utfCombiningClass, $utfCanonicalComp, $utfCanonicalDecomp;
 208                 if( !isset( $utfCombiningClass ) ) {
 209                         require_once( 'UtfNormalData.inc' );
 210                 }
 211         }
 212
 213         /**
 214          * Returns true if the string is _definitely_ in NFC.
 215          * Returns false if not or uncertain.
 216          * @param string $string a valid UTF-8 string. Input is not validated.
 217          * @return bool
 218          */
 219         function quickIsNFC( $string ) {
 220                 # ASCII is always valid NFC!
 221                 # If it's pure ASCII, let it through.
 222                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 223
 224                 UtfNormal::loadData();
 225                 global $utfCheckNFC, $utfCombiningClass;
 226                 $len = strlen( $string );
 227                 for( $i = 0; $i < $len; $i++ ) {
 228                         $c = $string{$i};
 229                         $n = ord( $c );
 230                         if( $n < 0x80 ) {
 231                                 continue;
 232                         } elseif( $n >= 0xf0 ) {
 233                                 $c = substr( $string, $i, 4 );
 234                                 $i += 3;
 235                         } elseif( $n >= 0xe0 ) {
 236                                 $c = substr( $string, $i, 3 );
 237                                 $i += 2;
 238                         } elseif( $n >= 0xc0 ) {
 239                                 $c = substr( $string, $i, 2 );
 240                                 $i++;
 241                         }
 242                         if( isset( $utfCheckNFC[$c] ) ) {
 243                                 # If it's NO or MAYBE, bail and do the slow check.
 244                                 return false;
 245                         }
 246                         if( isset( $utfCombiningClass[$c] ) ) {
 247                                 # Combining character? We might have to do sorting, at least.
 248                                 return false;
 249                         }
 250                 }
 251                 return true;
 252         }
 253
 254         /**
 255          * Returns true if the string is _definitely_ in NFC.
 256          * Returns false if not or uncertain.
 257          * @param string $string a UTF-8 string, altered on output to be valid UTF-8 safe for XML.
 258          */
 259         function quickIsNFCVerify( &$string ) {
 260                 # Screen out some characters that eg won't be allowed in XML
 261                 preg_replace( '/[\x00-\x08\x0b\x0c\x0e-\x1f]/', UTF8_REPLACEMENT, $string );
 262
 263                 # ASCII is always valid NFC!
 264                 # If we're only ever given plain ASCII, we can avoid the overhead
 265                 # of initializing the decomposition tables by skipping out early.
 266                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 267
 268                 UtfNormal::loadData();
 269                 global $utfCheckNFC, $utfCombiningClass;
 270
 271                 static $checkit = null, $tailBytes = null;
 272                 if( !isset( $checkit ) ) {
 273                         # Head bytes for sequences which we should do further validity checks
 274                         $checkit = array_flip( array_map( 'chr',
 275                                         array( 0xc0, 0xc1, 0xe0, 0xed, 0xef,
 276                                                    0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7,
 277                                                    0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff ) ) );
 278
 279                         $tailBytes = array();
 280                         for( $n = 0; $n < 256; $n++ ) {
 281                                 if( $n < 0xc0 ) {
 282                                         $remaining = 0;
 283                                 } elseif( $n < 0xe0 ) {
 284                                         $remaining = 1;
 285                                 } elseif( $n < 0xf0 ) {
 286                                         $remaining = 2;
 287                                 } elseif( $n < 0xf8 ) {
 288                                         $remaining = 3;
 289                                 } elseif( $n < 0xfc ) {
 290                                         $remaining = 4;
 291                                 } elseif( $n < 0xfe ) {
 292                                         $remaining = 5;
 293                                 } else {
 294                                         $remaining = 0;
 295                                 }
 296                                 $tailBytes[chr($n)] = $remaining;
 297                         }
 298                 }
 299
 300                 # Chop the text into pure-ASCII and non-ASCII areas;
 301                 # large ASCII parts can be handled much more quickly.
 302                 # Don't chop up Unicode areas for punctuation, though,
 303                 # that wastes energy.
 304                 preg_match_all(
 305                         '/([\x00-\x7f]+|[\x80-\xff][\x00-\x40\x5b-\x5f\x7b-\xff]*)/',
 306                         $string, $matches );
 307
 308                 ob_start();
 309                 $looksNormal = true;
 310                 foreach( $matches[1] as $str ) {
 311                         if( $str{0} < "\x80" ) {
 312                                 # ASCII chunk: guaranteed to be valid UTF-8
 313                                 # and in normal form C, so output it quick.
 314                                 echo $str;
 315                                 continue;
 316                         }
 317
 318                         # We'll have to examine the chunk byte by byte to ensure
 319                         # that it consists of valid UTF-8 sequences, and to see
 320                         # if any of them might not be normalized.
 321                         #
 322                         # Since PHP is not the fastest language on earth, some of
 323                         # this code is a little ugly with inner loop optimizations.
 324
 325                         $len = strlen( $str );
 326                         $tail = false;
 327                         $head = '';
 328                         for( $i = 0; $i < $len; $i++ ) {
 329                                 if( $tail ) {
 330                                         if( ( $c = $str{$i} ) >= "\x80" && $c < "\xc0" ) {
 331                                                 $sequence .= $c;
 332                                                 if( --$remaining ) {
 333                                                         # Keep adding bytes...
 334                                                         continue;
 335                                                 }
 336
 337                                                 # We have come to the end of the sequence...
 338                                                 if( isset( $checkit[$head] ) ) {
 339                                                         # Do some more detailed validity checks, for
 340                                                         # invalid characters and illegal sequences.
 341                                                         if( $head == "\xed" ) {
 342                                                                 # 0xed is relatively frequent in Korean, which
 343                                                                 # abuts the surrogate area, so we're doing
 344                                                                 # this check separately.
 345                                                                 if( $sequence >= UTF8_SURROGATE_FIRST ) {
 346                                                                         echo UTF8_REPLACEMENT;
 347                                                                         $tail = false;
 348                                                                         continue;
 349                                                                 }
 350                                                         } else {
 351                                                                 $n = ord( $head );
 352                                                                 if(    ($n  < 0xc2 && $sequence <= UTF8_OVERLONG_A)
 353                                                                         || ($n == 0xe0 && $sequence <= UTF8_OVERLONG_B)
 354                                                                         || ($n == 0xef &&
 355                                                                                 ($sequence >= UTF8_FDD0 && $sequence <= UTF8_FDEF)
 356                                                                                 || ($sequence == UTF8_FFFE)
 357                                                                                 || ($sequence == UTF8_FFFF) )
 358                                                                         || ($n == 0xf0 && $sequence <= UTF8_OVERLONG_C)
 359                                                                         || ($n >= 0xf0 && $sequence > UTF8_MAX) ) {
 360                                                                         echo UTF8_REPLACEMENT;
 361                                                                         $tail = false;
 362                                                                         continue;
 363                                                                 }
 364                                                         }
 365                                                 }
 366
 367                                                 if( isset( $utfCheckNFC[$sequence] ) ||
 368                                                         isset( $utfCombiningClass[$sequence] ) ) {
 369                                                         # If it's NO or MAYBE, we'll have to rip
 370                                                         # the string apart and put it back together.
 371                                                         # That's going to be mighty slow.
 372                                                         $looksNormal = false;
 373                                                 }
 374
 375                                                 # The sequence is legal!
 376                                                 echo $sequence;
 377                                                 $tail = false;
 378                                                 $head = '';
 379                                                 continue;
 380                                         }
 381                                         # Not a valid tail byte! DIscard the char we've been building.
 382                                         #printf ("Invalid '%x' in tail with %d remaining bytes\n", $n, $remaining );
 383                                         $tail = false;
 384                                         echo UTF8_REPLACEMENT;
 385                                 }
 386                                 if( $remaining = $tailBytes[$c = $str{$i}] ) {
 387                                         $tail = true;
 388                                         $sequence = $head = $c;
 389                                 } elseif( $c < "\x80" ) {
 390                                         echo $c;
 391                                 } elseif( $c < "\xc0" ) {
 392                                         # illegal tail bytes or head byte of overlong sequence
 393                                         if( $head == '' ) {
 394                                                 # Don't add if we're continuing a too-long sequence
 395                                                 echo UTF8_REPLACEMENT;
 396                                         }
 397                                 } else {
 398                                         echo UTF8_REPLACEMENT;
 399                                 }
 400                         }
 401                         if( $tail ) {
 402                                 echo UTF8_REPLACEMENT;
 403                         }
 404                 }
 405                 $string = ob_get_contents();
 406                 ob_end_clean();
 407                 return $looksNormal;
 408         }
 409
 410         # These take a string and run the normalization on them, without
 411         # checking for validity or any optimization etc. Input must be
 412         # VALID UTF-8!
 413         /**
 414          * @param string $string
 415          * @return string
 416          * @access private
 417          */
 418         function NFC( $string ) {
 419                 return UtfNormal::fastCompose( UtfNormal::NFD( $string ) );
 420         }
 421
 422         /**
 423          * @param string $string
 424          * @return string
 425          * @access private
 426          */
 427         function NFD( $string ) {
 428                 UtfNormal::loadData();
 429                 global $utfCanonicalDecomp;
 430                 return UtfNormal::fastCombiningSort(
 431                         UtfNormal::fastDecompose( $string, $utfCanonicalDecomp ) );
 432         }
 433
 434         /**
 435          * @param string $string
 436          * @return string
 437          * @access private
 438          */
 439         function NFKC( $string ) {
 440                 return UtfNormal::fastCompose( UtfNormal::NFKD( $string ) );
 441         }
 442
 443         /**
 444          * @param string $string
 445          * @return string
 446          * @access private
 447          */
 448         function NFKD( $string ) {
 449                 global $utfCompatibilityDecomp;
 450                 if( !isset( $utfCompatibilityDecomp ) ) {
 451                         require_once( 'UtfNormalDataK.inc' );
 452                 }
 453                 return UtfNormal::fastCombiningSort(
 454                         UtfNormal::fastDecompose( $string, $utfCompatibilityDecomp ) );
 455         }
 456
 457
 458         /**
 459          * Perform decomposition of a UTF-8 string into either D or KD form
 460          * (depending on which decomposition map is passed to us).
 461          * Input is assumed to be *valid* UTF-8. Invalid code will break.
 462          * @access private
 463          * @param string $string Valid UTF-8 string
 464          * @param array $map hash of expanded decomposition map
 465          * @return string a UTF-8 string decomposed, not yet normalized (needs sorting)
 466          */
 467         function fastDecompose( $string, &$map ) {
 468                 UtfNormal::loadData();
 469                 $len = strlen( $string );
 470                 $out = '';
 471                 for( $i = 0; $i < $len; $i++ ) {
 472                         $c = $string{$i};
 473                         $n = ord( $c );
 474                         if( $n < 0x80 ) {
 475                                 # ASCII chars never decompose
 476                                 # THEY ARE IMMORTAL
 477                                 $out .= $c;
 478                                 continue;
 479                         } elseif( $n >= 0xf0 ) {
 480                                 $c = substr( $string, $i, 4 );
 481                                 $i += 3;
 482                         } elseif( $n >= 0xe0 ) {
 483                                 $c = substr( $string, $i, 3 );
 484                                 $i += 2;
 485                         } elseif( $n >= 0xc0 ) {
 486                                 $c = substr( $string, $i, 2 );
 487                                 $i++;
 488                         }
 489                         if( isset( $map[$c] ) ) {
 490                                 $out .= $map[$c];
 491                                 continue;
 492                         } else {
 493                                 if( $c >= UTF8_HANGUL_FIRST && $c <= UTF8_HANGUL_LAST ) {
 494                                         # Decompose a hangul syllable into jamo;
 495                                         # hardcoded for three-byte UTF-8 sequence.
 496                                         # A lookup table would be slightly faster,
 497                                         # but adds a lot of memory & disk needs.
 498                                         #
 499                                         $index = ( (ord( $c{0} ) & 0x0f) << 12
 500                                                  | (ord( $c{1} ) & 0x3f) <<  6
 501                                                  | (ord( $c{2} ) & 0x3f) )
 502                                                - UNICODE_HANGUL_FIRST;
 503                                         $l = IntVal( $index / UNICODE_HANGUL_NCOUNT );
 504                                         $v = IntVal( ($index % UNICODE_HANGUL_NCOUNT) / UNICODE_HANGUL_TCOUNT);
 505                                         $t = $index % UNICODE_HANGUL_TCOUNT;
 506                                         $out .= "\xe1\x84" . chr( 0x80 + $l ) . "\xe1\x85" . chr( 0xa1 + $v );
 507                                         if( $t >= 25 ) {
 508                                                 $out .= "\xe1\x87" . chr( 0x80 + $t - 25 );
 509                                         } elseif( $t ) {
 510                                                 $out .= "\xe1\x86" . chr( 0xa7 + $t );
 511                                         }
 512                                         continue;
 513                                 }
 514                         }
 515                         $out .= $c;
 516                 }
 517                 return $out;
 518         }
 519
 520         /**
 521          * Sorts combining characters into canonical order. This is the
 522          * final step in creating decomposed normal forms D and KD.
 523          * @access private
 524          * @param string $string a valid, decomposed UTF-8 string. Input is not validated.
 525          * @return string a UTF-8 string with combining characters sorted in canonical order
 526          */
 527         function fastCombiningSort( $string ) {
 528                 UtfNormal::loadData();
 529                 global $utfCombiningClass;
 530                 $len = strlen( $string );
 531                 $out = '';
 532                 $combiners = array();
 533                 $lastClass = -1;
 534                 for( $i = 0; $i < $len; $i++ ) {
 535                         $c = $string{$i};
 536                         $n = ord( $c );
 537                         if( $n >= 0x80 ) {
 538                                 if( $n >= 0xf0 ) {
 539                                         $c = substr( $string, $i, 4 );
 540                                         $i += 3;
 541                                 } elseif( $n >= 0xe0 ) {
 542                                         $c = substr( $string, $i, 3 );
 543                                         $i += 2;
 544                                 } elseif( $n >= 0xc0 ) {
 545                                         $c = substr( $string, $i, 2 );
 546                                         $i++;
 547                                 }
 548                                 if( isset( $utfCombiningClass[$c] ) ) {
 549                                         $lastClass = $utfCombiningClass[$c];
 550                                         @$combiners[$lastClass] .= $c;
 551                                         continue;
 552                                 }
 553                         }
 554                         if( $lastClass ) {
 555                                 ksort( $combiners );
 556                                 $out .= implode( '', $combiners );
 557                                 $combiners = array();
 558                         }
 559                         $out .= $c;
 560                         $lastClass = 0;
 561                 }
 562                 if( $lastClass ) {
 563                         ksort( $combiners );
 564                         $out .= implode( '', $combiners );
 565                 }
 566                 return $out;
 567         }
 568
 569         /**
 570          * Produces canonically composed sequences, i.e. normal form C or KC.
 571          *
 572          * @access private
 573          * @param string $string a valid UTF-8 string in sorted normal form D or KD. Input is not validated.
 574          * @return string a UTF-8 string with canonical precomposed characters used where possible
 575          */
 576         function fastCompose( $string ) {
 577                 UtfNormal::loadData();
 578                 global $utfCanonicalComp, $utfCombiningClass;
 579                 $len = strlen( $string );
 580                 $out = '';
 581                 $lastClass = -1;
 582                 $startChar = '';
 583                 $combining = '';
 584                 $x1 = ord(substr(UTF8_HANGUL_VBASE,0,1));
 585                 $x2 = ord(substr(UTF8_HANGUL_TEND,0,1));
 586                 for( $i = 0; $i < $len; $i++ ) {
 587                         $c = $string{$i};
 588                         $n = ord( $c );
 589                         if( $n < 0x80 ) {
 590                                 # No combining characters here...
 591                                 $out .= $startChar;
 592                                 $out .= $combining;
 593                                 $startChar = $c;
 594                                 $combining = '';
 595                                 $lastClass = 0;
 596                                 continue;
 597                         } elseif( $n >= 0xf0 ) {
 598                                 $c = substr( $string, $i, 4 );
 599                                 $i += 3;
 600                         } elseif( $n >= 0xe0 ) {
 601                                 $c = substr( $string, $i, 3 );
 602                                 $i += 2;
 603                         } elseif( $n >= 0xc0 ) {
 604                                 $c = substr( $string, $i, 2 );
 605                                 $i++;
 606                         }
 607                         $pair = $startChar . $c;
 608                         if( $n > 0x80 ) {
 609                                 if( isset( $utfCombiningClass[$c] ) ) {
 610                                         # A combining char; see what we can do with it
 611                                         $class = $utfCombiningClass[$c];
 612                                         if( !empty( $startChar ) &&
 613                                                 $lastClass < $class &&
 614                                                 $class > 0 &&
 615                                                 isset( $utfCanonicalComp[$pair] ) ) {
 616                                                 $startChar = $utfCanonicalComp[$pair];
 617                                                 $class = 0;
 618                                         } else {
 619                                                 $combining .= $c;
 620                                         }
 621                                         $lastClass = $class;
 622                                         continue;
 623                                 }
 624                         }
 625                         # New start char
 626                         if( $lastClass == 0 ) {
 627                                 if( isset( $utfCanonicalComp[$pair] ) ) {
 628                                         $startChar = $utfCanonicalComp[$pair];
 629                                         continue;
 630                                 }
 631                                 if( $n >= $x1 && $n <= $x2 ) {
 632                                         # WARNING: Hangul code is painfully slow.
 633                                         # I apologize for this ugly, ugly code; however
 634                                         # performance is even more teh suck if we call
 635                                         # out to nice clean functions. Lookup tables are
 636                                         # marginally faster, but require a lot of space.
 637                                         #
 638                                         if( $c >= UTF8_HANGUL_VBASE &&
 639                                                 $c <= UTF8_HANGUL_VEND &&
 640                                                 $startChar >= UTF8_HANGUL_LBASE &&
 641                                                 $startChar <= UTF8_HANGUL_LEND ) {
 642                                                 #
 643                                                 #$lIndex = utf8ToCodepoint( $startChar ) - UNICODE_HANGUL_LBASE;
 644                                                 #$vIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_VBASE;
 645                                                 $lIndex = ord( $startChar{2} ) - 0x80;
 646                                                 $vIndex = ord( $c{2}         ) - 0xa1;
 647
 648                                                 $hangulPoint = UNICODE_HANGUL_FIRST +
 649                                                         UNICODE_HANGUL_TCOUNT *
 650                                                         (UNICODE_HANGUL_VCOUNT * $lIndex + $vIndex);
 651
 652                                                 # Hardcode the limited-range UTF-8 conversion:
 653                                                 $startChar = chr( $hangulPoint >> 12 & 0x0f | 0xe0 ) .
 654                                                                          chr( $hangulPoint >>  6 & 0x3f | 0x80 ) .
 655                                                                          chr( $hangulPoint       & 0x3f | 0x80 );
 656                                                 continue;
 657                                         } elseif( $c >= UTF8_HANGUL_TBASE &&
 658                                                           $c <= UTF8_HANGUL_TEND &&
 659                                                           $startChar >= UTF8_HANGUL_FIRST &&
 660                                                           $startChar <= UTF8_HANGUL_LAST ) {
 661                                                 # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
 662                                                 $tIndex = ord( $c{2} ) - 0xa7;
 663                                                 if( $tIndex < 0 ) $tIndex = ord( $c{2} ) - 0x80 + (0x11c0 - 0x11a7);
 664
 665                                                 # Increment the code point by $tIndex, without
 666                                                 # the function overhead of decoding and recoding UTF-8
 667                                                 #
 668                                                 $tail = ord( $startChar{2} ) + $tIndex;
 669                                                 if( $tail > 0xbf ) {
 670                                                         $tail -= 0x40;
 671                                                         $mid = ord( $startChar{1} ) + 1;
 672                                                         if( $mid > 0xbf ) {
 673                                                                 $startChar{0} = chr( ord( $startChar{0} ) + 1 );
 674                                                                 $mid -= 0x40;
 675                                                         }
 676                                                         $startChar{1} = chr( $mid );
 677                                                 }
 678                                                 $startChar{2} = chr( $tail );
 679                                                 continue;
 680                                         }
 681                                 }
 682                         }
 683                         $out .= $startChar;
 684                         $out .= $combining;
 685                         $startChar = $c;
 686                         $combining = '';
 687                         $lastClass = 0;
 688                 }
 689                 $out .= $startChar . $combining;
 690                 return $out;
 691         }
 692
 693         /**
 694          * This is just used for the benchmark, comparing how long it takes to
 695          * interate through a string without really doing anything of substance.
 696          * @param string $string
 697          * @return string
 698          */
 699         function placebo( $string ) {
 700                 $len = strlen( $string );
 701                 $out = '';
 702                 for( $i = 0; $i < $len; $i++ ) {
 703                         $out .= $string{$i};
 704                 }
 705                 return $out;
 706         }
 707 }
 708
 709 ?>