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( NORMALIZE_ICU ) {
 130                         # We exclude a few chars that ICU would not.
 131                         $string = preg_replace(
 132                                 '/[\x00-\x08\x0b\x0c\x0e-\x1f]/',
 133                                 UTF8_REPLACEMENT,
 134                                 $string );
 135                         $str = str_replace( UTF8_FFFE, UTF8_REPLACEMENT, $string );
 136
 137                         # UnicodeString constructor fails if the string ends with a
 138                         # head byte. Add a junk char at the end, we'll strip it off.
 139                         return rtrim( utf8_normalize( $str . "\x01", UNORM_NFC ), "\x01" );
 140                 } elseif( UtfNormal::quickIsNFCVerify( $string ) ) {
 141                         # Side effect -- $string has had UTF-8 errors cleaned up.
 142                         return $string;
 143                 } else {
 144                         return UtfNormal::NFC( $string );
 145                 }
 146         }
 147
 148         /**
 149          * Convert a UTF-8 string to normal form C, canonical composition.
 150          * Fast return for pure ASCII strings; some lesser optimizations for
 151          * strings containing only known-good characters.
 152          *
 153          * @param string $string a valid UTF-8 string. Input is not validated.
 154          * @return string a UTF-8 string in normal form C
 155          */
 156         function toNFC( $string ) {
 157                 if( NORMALIZE_ICU )
 158                         return utf8_normalize( $string, UNORM_NFC );
 159                 elseif( UtfNormal::quickIsNFC( $string ) )
 160                         return $string;
 161                 else
 162                         return UtfNormal::NFC( $string );
 163         }
 164
 165         /**
 166          * Convert a UTF-8 string to normal form D, canonical decomposition.
 167          * Fast return for pure ASCII strings.
 168          *
 169          * @param string $string a valid UTF-8 string. Input is not validated.
 170          * @return string a UTF-8 string in normal form D
 171          */
 172         function toNFD( $string ) {
 173                 if( NORMALIZE_ICU )
 174                         return utf8_normalize( $string, UNORM_NFD );
 175                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 176                         return UtfNormal::NFD( $string );
 177                 else
 178                         return $string;
 179         }
 180
 181         /**
 182          * Convert a UTF-8 string to normal form KC, compatibility composition.
 183          * This may cause irreversible information loss, use judiciously.
 184          * Fast return for pure ASCII strings.
 185          *
 186          * @param string $string a valid UTF-8 string. Input is not validated.
 187          * @return string a UTF-8 string in normal form KC
 188          */
 189         function toNFKC( $string ) {
 190                 if( NORMALIZE_ICU )
 191                         return utf8_normalize( $string, UNORM_NFKC );
 192                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 193                         return UtfNormal::NFKC( $string );
 194                 else
 195                         return $string;
 196         }
 197
 198         /**
 199          * Convert a UTF-8 string to normal form KD, compatibility decomposition.
 200          * This may cause irreversible information loss, use judiciously.
 201          * Fast return for pure ASCII strings.
 202          *
 203          * @param string $string a valid UTF-8 string. Input is not validated.
 204          * @return string a UTF-8 string in normal form KD
 205          */
 206         function toNFKD( $string ) {
 207                 if( NORMALIZE_ICU )
 208                         return utf8_normalize( $string, UNORM_NFKD );
 209                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 210                         return UtfNormal::NFKD( $string );
 211                 else
 212                         return $string;
 213         }
 214
 215         /**
 216          * Load the basic composition data if necessary
 217          * @access private
 218          */
 219         function loadData() {
 220                 global $utfCombiningClass, $utfCanonicalComp, $utfCanonicalDecomp;
 221                 if( !isset( $utfCombiningClass ) ) {
 222                         require_once( 'UtfNormalData.inc' );
 223                 }
 224         }
 225
 226         /**
 227          * Returns true if the string is _definitely_ in NFC.
 228          * Returns false if not or uncertain.
 229          * @param string $string a valid UTF-8 string. Input is not validated.
 230          * @return bool
 231          */
 232         function quickIsNFC( $string ) {
 233                 # ASCII is always valid NFC!
 234                 # If it's pure ASCII, let it through.
 235                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 236
 237                 UtfNormal::loadData();
 238                 global $utfCheckNFC, $utfCombiningClass;
 239                 $len = strlen( $string );
 240                 for( $i = 0; $i < $len; $i++ ) {
 241                         $c = $string{$i};
 242                         $n = ord( $c );
 243                         if( $n < 0x80 ) {
 244                                 continue;
 245                         } elseif( $n >= 0xf0 ) {
 246                                 $c = substr( $string, $i, 4 );
 247                                 $i += 3;
 248                         } elseif( $n >= 0xe0 ) {
 249                                 $c = substr( $string, $i, 3 );
 250                                 $i += 2;
 251                         } elseif( $n >= 0xc0 ) {
 252                                 $c = substr( $string, $i, 2 );
 253                                 $i++;
 254                         }
 255                         if( isset( $utfCheckNFC[$c] ) ) {
 256                                 # If it's NO or MAYBE, bail and do the slow check.
 257                                 return false;
 258                         }
 259                         if( isset( $utfCombiningClass[$c] ) ) {
 260                                 # Combining character? We might have to do sorting, at least.
 261                                 return false;
 262                         }
 263                 }
 264                 return true;
 265         }
 266
 267         /**
 268          * Returns true if the string is _definitely_ in NFC.
 269          * Returns false if not or uncertain.
 270          * @param string $string a UTF-8 string, altered on output to be valid UTF-8 safe for XML.
 271          */
 272         function quickIsNFCVerify( &$string ) {
 273                 # Screen out some characters that eg won't be allowed in XML
 274                 $string = preg_replace( '/[\x00-\x08\x0b\x0c\x0e-\x1f]/', UTF8_REPLACEMENT, $string );
 275
 276                 # ASCII is always valid NFC!
 277                 # If we're only ever given plain ASCII, we can avoid the overhead
 278                 # of initializing the decomposition tables by skipping out early.
 279                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 280
 281                 static $checkit = null, $tailBytes = null, $utfCheckOrCombining = null;
 282                 if( !isset( $checkit ) ) {
 283                         # Load/build some scary lookup tables...
 284                         UtfNormal::loadData();
 285                         global $utfCheckNFC, $utfCombiningClass;
 286
 287                         $utfCheckOrCombining = array_merge( $utfCheckNFC, $utfCombiningClass );
 288
 289                         # Head bytes for sequences which we should do further validity checks
 290                         $checkit = array_flip( array_map( 'chr',
 291                                         array( 0xc0, 0xc1, 0xe0, 0xed, 0xef,
 292                                                    0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7,
 293                                                    0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff ) ) );
 294
 295                         # Each UTF-8 head byte is followed by a certain
 296                         # number of tail bytes.
 297                         $tailBytes = array();
 298                         for( $n = 0; $n < 256; $n++ ) {
 299                                 if( $n < 0xc0 ) {
 300                                         $remaining = 0;
 301                                 } elseif( $n < 0xe0 ) {
 302                                         $remaining = 1;
 303                                 } elseif( $n < 0xf0 ) {
 304                                         $remaining = 2;
 305                                 } elseif( $n < 0xf8 ) {
 306                                         $remaining = 3;
 307                                 } elseif( $n < 0xfc ) {
 308                                         $remaining = 4;
 309                                 } elseif( $n < 0xfe ) {
 310                                         $remaining = 5;
 311                                 } else {
 312                                         $remaining = 0;
 313                                 }
 314                                 $tailBytes[chr($n)] = $remaining;
 315                         }
 316                 }
 317
 318                 # Chop the text into pure-ASCII and non-ASCII areas;
 319                 # large ASCII parts can be handled much more quickly.
 320                 # Don't chop up Unicode areas for punctuation, though,
 321                 # that wastes energy.
 322                 preg_match_all(
 323                         '/([\x00-\x7f]+|[\x80-\xff][\x00-\x40\x5b-\x5f\x7b-\xff]*)/',
 324                         $string, $matches );
 325
 326                 $looksNormal = true;
 327                 $base = 0;
 328                 $replace = array();
 329                 foreach( $matches[1] as $str ) {
 330                         $chunk = strlen( $str );
 331
 332                         if( $str{0} < "\x80" ) {
 333                                 # ASCII chunk: guaranteed to be valid UTF-8
 334                                 # and in normal form C, so skip over it.
 335                                 $base += $chunk;
 336                                 continue;
 337                         }
 338
 339                         # We'll have to examine the chunk byte by byte to ensure
 340                         # that it consists of valid UTF-8 sequences, and to see
 341                         # if any of them might not be normalized.
 342                         #
 343                         # Since PHP is not the fastest language on earth, some of
 344                         # this code is a little ugly with inner loop optimizations.
 345
 346                         $head = '';
 347                         $len = $chunk + 1; # Counting down is faster. I'm *so* sorry.
 348
 349                         for( $i = -1; --$len; ) {
 350                                 if( $remaining = $tailBytes[$c = $str{++$i}] ) {
 351                                         # UTF-8 head byte!
 352                                         $sequence = $head = $c;
 353                                         do {
 354                                                 # Look for the defined number of tail bytes...
 355                                                 if( --$len && ( $c = $str{++$i} ) >= "\x80" && $c < "\xc0" ) {
 356                                                         # Legal tail bytes are nice.
 357                                                         $sequence .= $c;
 358                                                 } else {
 359                                                         if( 0 == $len ) {
 360                                                                 # Premature end of string!
 361                                                                 # Drop a replacement character into output to
 362                                                                 # represent the invalid UTF-8 sequence.
 363                                                                 $replace[] = array( UTF8_REPLACEMENT,
 364                                                                                                         $base + $i + 1 - strlen( $sequence ),
 365                                                                                                         strlen( $sequence ) );
 366                                                                 break 2;
 367                                                         } else {
 368                                                                 # Illegal tail byte; abandon the sequence.
 369                                                                 $replace[] = array( UTF8_REPLACEMENT,
 370                                                                                                         $base + $i - strlen( $sequence ),
 371                                                                                                         strlen( $sequence ) );
 372                                                                 # Back up and reprocess this byte; it may itself
 373                                                                 # be a legal ASCII or UTF-8 sequence head.
 374                                                                 --$i;
 375                                                                 ++$len;
 376                                                                 continue 2;
 377                                                         }
 378                                                 }
 379                                         } while( --$remaining );
 380
 381                                         if( isset( $checkit[$head] ) ) {
 382                                                 # Do some more detailed validity checks, for
 383                                                 # invalid characters and illegal sequences.
 384                                                 if( $head == "\xed" ) {
 385                                                         # 0xed is relatively frequent in Korean, which
 386                                                         # abuts the surrogate area, so we're doing
 387                                                         # this check separately to speed things up.
 388
 389                                                         if( $sequence >= UTF8_SURROGATE_FIRST ) {
 390                                                                 # Surrogates are legal only in UTF-16 code.
 391                                                                 # They are totally forbidden here in UTF-8
 392                                                                 # utopia.
 393                                                                 $replace[] = array( UTF8_REPLACEMENT,
 394                                                                              $base + $i + 1 - strlen( $sequence ),
 395                                                                              strlen( $sequence ) );
 396                                                                 $head = '';
 397                                                                 continue;
 398                                                         }
 399                                                 } else {
 400                                                         # Slower, but rarer checks...
 401                                                         $n = ord( $head );
 402                                                         if(
 403                                                                 # "Overlong sequences" are those that are syntactically
 404                                                                 # correct but use more UTF-8 bytes than are necessary to
 405                                                                 # encode a character. Naïve string comparisons can be
 406                                                                 # tricked into failing to see a match for an ASCII
 407                                                                 # character, for instance, which can be a security hole
 408                                                                 # if blacklist checks are being used.
 409                                                                ($n  < 0xc2 && $sequence <= UTF8_OVERLONG_A)
 410                                                                 || ($n == 0xe0 && $sequence <= UTF8_OVERLONG_B)
 411                                                                 || ($n == 0xf0 && $sequence <= UTF8_OVERLONG_C)
 412
 413                                                                 # U+FFFE and U+FFFF are explicitly forbidden in Unicode.
 414                                                                 || ($n == 0xef &&
 415                                                                            ($sequence == UTF8_FFFE)
 416                                                                         || ($sequence == UTF8_FFFF) )
 417
 418                                                                 # Unicode has been limited to 21 bits; longer
 419                                                                 # sequences are not allowed.
 420                                                                 || ($n >= 0xf0 && $sequence > UTF8_MAX) ) {
 421
 422                                                                 $replace[] = array( UTF8_REPLACEMENT,
 423                                                                                     $base + $i + 1 - strlen( $sequence ),
 424                                                                                     strlen( $sequence ) );
 425                                                                 $head = '';
 426                                                                 continue;
 427                                                         }
 428                                                 }
 429                                         }
 430
 431                                         if( isset( $utfCheckOrCombining[$sequence] ) ) {
 432                                                 # If it's NO or MAYBE, we'll have to rip
 433                                                 # the string apart and put it back together.
 434                                                 # That's going to be mighty slow.
 435                                                 $looksNormal = false;
 436                                         }
 437
 438                                         # The sequence is legal!
 439                                         $head = '';
 440                                 } elseif( $c < "\x80" ) {
 441                                         # ASCII byte.
 442                                         $head = '';
 443                                 } elseif( $c < "\xc0" ) {
 444                                         # Illegal tail bytes
 445                                         if( $head == '' ) {
 446                                                 # Out of the blue!
 447                                                 $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 448                                         } else {
 449                                                 # Don't add if we're continuing a broken sequence;
 450                                                 # we already put a replacement character when we looked
 451                                                 # at the broken sequence.
 452                                                 $replace[] = array( '', $base + $i, 1 );
 453                                         }
 454                                 } else {
 455                                         # Miscellaneous freaks.
 456                                         $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 457                                         $head = '';
 458                                 }
 459                         }
 460                         $base += $chunk;
 461                 }
 462                 if( count( $replace ) ) {
 463                         # There were illegal UTF-8 sequences we need to fix up.
 464                         $out = '';
 465                         $last = 0;
 466                         foreach( $replace as $rep ) {
 467                                 list( $replacement, $start, $length ) = $rep;
 468                                 if( $last < $start ) {
 469                                         $out .= substr( $string, $last, $start - $last );
 470                                 }
 471                                 $out .= $replacement;
 472                                 $last = $start + $length;
 473                         }
 474                         if( $last < strlen( $string ) ) {
 475                                 $out .= substr( $string, $last );
 476                         }
 477                         $string = $out;
 478                 }
 479                 return $looksNormal;
 480         }
 481
 482         # These take a string and run the normalization on them, without
 483         # checking for validity or any optimization etc. Input must be
 484         # VALID UTF-8!
 485         /**
 486          * @param string $string
 487          * @return string
 488          * @access private
 489          */
 490         function NFC( $string ) {
 491                 return UtfNormal::fastCompose( UtfNormal::NFD( $string ) );
 492         }
 493
 494         /**
 495          * @param string $string
 496          * @return string
 497          * @access private
 498          */
 499         function NFD( $string ) {
 500                 UtfNormal::loadData();
 501                 global $utfCanonicalDecomp;
 502                 return UtfNormal::fastCombiningSort(
 503                         UtfNormal::fastDecompose( $string, $utfCanonicalDecomp ) );
 504         }
 505
 506         /**
 507          * @param string $string
 508          * @return string
 509          * @access private
 510          */
 511         function NFKC( $string ) {
 512                 return UtfNormal::fastCompose( UtfNormal::NFKD( $string ) );
 513         }
 514
 515         /**
 516          * @param string $string
 517          * @return string
 518          * @access private
 519          */
 520         function NFKD( $string ) {
 521                 global $utfCompatibilityDecomp;
 522                 if( !isset( $utfCompatibilityDecomp ) ) {
 523                         require_once( 'UtfNormalDataK.inc' );
 524                 }
 525                 return UtfNormal::fastCombiningSort(
 526                         UtfNormal::fastDecompose( $string, $utfCompatibilityDecomp ) );
 527         }
 528
 529
 530         /**
 531          * Perform decomposition of a UTF-8 string into either D or KD form
 532          * (depending on which decomposition map is passed to us).
 533          * Input is assumed to be *valid* UTF-8. Invalid code will break.
 534          * @access private
 535          * @param string $string Valid UTF-8 string
 536          * @param array $map hash of expanded decomposition map
 537          * @return string a UTF-8 string decomposed, not yet normalized (needs sorting)
 538          */
 539         function fastDecompose( $string, &$map ) {
 540                 UtfNormal::loadData();
 541                 $len = strlen( $string );
 542                 $out = '';
 543                 for( $i = 0; $i < $len; $i++ ) {
 544                         $c = $string{$i};
 545                         $n = ord( $c );
 546                         if( $n < 0x80 ) {
 547                                 # ASCII chars never decompose
 548                                 # THEY ARE IMMORTAL
 549                                 $out .= $c;
 550                                 continue;
 551                         } elseif( $n >= 0xf0 ) {
 552                                 $c = substr( $string, $i, 4 );
 553                                 $i += 3;
 554                         } elseif( $n >= 0xe0 ) {
 555                                 $c = substr( $string, $i, 3 );
 556                                 $i += 2;
 557                         } elseif( $n >= 0xc0 ) {
 558                                 $c = substr( $string, $i, 2 );
 559                                 $i++;
 560                         }
 561                         if( isset( $map[$c] ) ) {
 562                                 $out .= $map[$c];
 563                                 continue;
 564                         } else {
 565                                 if( $c >= UTF8_HANGUL_FIRST && $c <= UTF8_HANGUL_LAST ) {
 566                                         # Decompose a hangul syllable into jamo;
 567                                         # hardcoded for three-byte UTF-8 sequence.
 568                                         # A lookup table would be slightly faster,
 569                                         # but adds a lot of memory & disk needs.
 570                                         #
 571                                         $index = ( (ord( $c{0} ) & 0x0f) << 12
 572                                                  | (ord( $c{1} ) & 0x3f) <<  6
 573                                                  | (ord( $c{2} ) & 0x3f) )
 574                                                - UNICODE_HANGUL_FIRST;
 575                                         $l = IntVal( $index / UNICODE_HANGUL_NCOUNT );
 576                                         $v = IntVal( ($index % UNICODE_HANGUL_NCOUNT) / UNICODE_HANGUL_TCOUNT);
 577                                         $t = $index % UNICODE_HANGUL_TCOUNT;
 578                                         $out .= "\xe1\x84" . chr( 0x80 + $l ) . "\xe1\x85" . chr( 0xa1 + $v );
 579                                         if( $t >= 25 ) {
 580                                                 $out .= "\xe1\x87" . chr( 0x80 + $t - 25 );
 581                                         } elseif( $t ) {
 582                                                 $out .= "\xe1\x86" . chr( 0xa7 + $t );
 583                                         }
 584                                         continue;
 585                                 }
 586                         }
 587                         $out .= $c;
 588                 }
 589                 return $out;
 590         }
 591
 592         /**
 593          * Sorts combining characters into canonical order. This is the
 594          * final step in creating decomposed normal forms D and KD.
 595          * @access private
 596          * @param string $string a valid, decomposed UTF-8 string. Input is not validated.
 597          * @return string a UTF-8 string with combining characters sorted in canonical order
 598          */
 599         function fastCombiningSort( $string ) {
 600                 UtfNormal::loadData();
 601                 global $utfCombiningClass;
 602                 $len = strlen( $string );
 603                 $out = '';
 604                 $combiners = array();
 605                 $lastClass = -1;
 606                 for( $i = 0; $i < $len; $i++ ) {
 607                         $c = $string{$i};
 608                         $n = ord( $c );
 609                         if( $n >= 0x80 ) {
 610                                 if( $n >= 0xf0 ) {
 611                                         $c = substr( $string, $i, 4 );
 612                                         $i += 3;
 613                                 } elseif( $n >= 0xe0 ) {
 614                                         $c = substr( $string, $i, 3 );
 615                                         $i += 2;
 616                                 } elseif( $n >= 0xc0 ) {
 617                                         $c = substr( $string, $i, 2 );
 618                                         $i++;
 619                                 }
 620                                 if( isset( $utfCombiningClass[$c] ) ) {
 621                                         $lastClass = $utfCombiningClass[$c];
 622                                         @$combiners[$lastClass] .= $c;
 623                                         continue;
 624                                 }
 625                         }
 626                         if( $lastClass ) {
 627                                 ksort( $combiners );
 628                                 $out .= implode( '', $combiners );
 629                                 $combiners = array();
 630                         }
 631                         $out .= $c;
 632                         $lastClass = 0;
 633                 }
 634                 if( $lastClass ) {
 635                         ksort( $combiners );
 636                         $out .= implode( '', $combiners );
 637                 }
 638                 return $out;
 639         }
 640
 641         /**
 642          * Produces canonically composed sequences, i.e. normal form C or KC.
 643          *
 644          * @access private
 645          * @param string $string a valid UTF-8 string in sorted normal form D or KD. Input is not validated.
 646          * @return string a UTF-8 string with canonical precomposed characters used where possible
 647          */
 648         function fastCompose( $string ) {
 649                 UtfNormal::loadData();
 650                 global $utfCanonicalComp, $utfCombiningClass;
 651                 $len = strlen( $string );
 652                 $out = '';
 653                 $lastClass = -1;
 654                 $startChar = '';
 655                 $combining = '';
 656                 $x1 = ord(substr(UTF8_HANGUL_VBASE,0,1));
 657                 $x2 = ord(substr(UTF8_HANGUL_TEND,0,1));
 658                 for( $i = 0; $i < $len; $i++ ) {
 659                         $c = $string{$i};
 660                         $n = ord( $c );
 661                         if( $n < 0x80 ) {
 662                                 # No combining characters here...
 663                                 $out .= $startChar;
 664                                 $out .= $combining;
 665                                 $startChar = $c;
 666                                 $combining = '';
 667                                 $lastClass = 0;
 668                                 continue;
 669                         } elseif( $n >= 0xf0 ) {
 670                                 $c = substr( $string, $i, 4 );
 671                                 $i += 3;
 672                         } elseif( $n >= 0xe0 ) {
 673                                 $c = substr( $string, $i, 3 );
 674                                 $i += 2;
 675                         } elseif( $n >= 0xc0 ) {
 676                                 $c = substr( $string, $i, 2 );
 677                                 $i++;
 678                         }
 679                         $pair = $startChar . $c;
 680                         if( $n > 0x80 ) {
 681                                 if( isset( $utfCombiningClass[$c] ) ) {
 682                                         # A combining char; see what we can do with it
 683                                         $class = $utfCombiningClass[$c];
 684                                         if( !empty( $startChar ) &&
 685                                                 $lastClass < $class &&
 686                                                 $class > 0 &&
 687                                                 isset( $utfCanonicalComp[$pair] ) ) {
 688                                                 $startChar = $utfCanonicalComp[$pair];
 689                                                 $class = 0;
 690                                         } else {
 691                                                 $combining .= $c;
 692                                         }
 693                                         $lastClass = $class;
 694                                         continue;
 695                                 }
 696                         }
 697                         # New start char
 698                         if( $lastClass == 0 ) {
 699                                 if( isset( $utfCanonicalComp[$pair] ) ) {
 700                                         $startChar = $utfCanonicalComp[$pair];
 701                                         continue;
 702                                 }
 703                                 if( $n >= $x1 && $n <= $x2 ) {
 704                                         # WARNING: Hangul code is painfully slow.
 705                                         # I apologize for this ugly, ugly code; however
 706                                         # performance is even more teh suck if we call
 707                                         # out to nice clean functions. Lookup tables are
 708                                         # marginally faster, but require a lot of space.
 709                                         #
 710                                         if( $c >= UTF8_HANGUL_VBASE &&
 711                                                 $c <= UTF8_HANGUL_VEND &&
 712                                                 $startChar >= UTF8_HANGUL_LBASE &&
 713                                                 $startChar <= UTF8_HANGUL_LEND ) {
 714                                                 #
 715                                                 #$lIndex = utf8ToCodepoint( $startChar ) - UNICODE_HANGUL_LBASE;
 716                                                 #$vIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_VBASE;
 717                                                 $lIndex = ord( $startChar{2} ) - 0x80;
 718                                                 $vIndex = ord( $c{2}         ) - 0xa1;
 719
 720                                                 $hangulPoint = UNICODE_HANGUL_FIRST +
 721                                                         UNICODE_HANGUL_TCOUNT *
 722                                                         (UNICODE_HANGUL_VCOUNT * $lIndex + $vIndex);
 723
 724                                                 # Hardcode the limited-range UTF-8 conversion:
 725                                                 $startChar = chr( $hangulPoint >> 12 & 0x0f | 0xe0 ) .
 726                                                                          chr( $hangulPoint >>  6 & 0x3f | 0x80 ) .
 727                                                                          chr( $hangulPoint       & 0x3f | 0x80 );
 728                                                 continue;
 729                                         } elseif( $c >= UTF8_HANGUL_TBASE &&
 730                                                           $c <= UTF8_HANGUL_TEND &&
 731                                                           $startChar >= UTF8_HANGUL_FIRST &&
 732                                                           $startChar <= UTF8_HANGUL_LAST ) {
 733                                                 # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
 734                                                 $tIndex = ord( $c{2} ) - 0xa7;
 735                                                 if( $tIndex < 0 ) $tIndex = ord( $c{2} ) - 0x80 + (0x11c0 - 0x11a7);
 736
 737                                                 # Increment the code point by $tIndex, without
 738                                                 # the function overhead of decoding and recoding UTF-8
 739                                                 #
 740                                                 $tail = ord( $startChar{2} ) + $tIndex;
 741                                                 if( $tail > 0xbf ) {
 742                                                         $tail -= 0x40;
 743                                                         $mid = ord( $startChar{1} ) + 1;
 744                                                         if( $mid > 0xbf ) {
 745                                                                 $startChar{0} = chr( ord( $startChar{0} ) + 1 );
 746                                                                 $mid -= 0x40;
 747                                                         }
 748                                                         $startChar{1} = chr( $mid );
 749                                                 }
 750                                                 $startChar{2} = chr( $tail );
 751                                                 continue;
 752                                         }
 753                                 }
 754                         }
 755                         $out .= $startChar;
 756                         $out .= $combining;
 757                         $startChar = $c;
 758                         $combining = '';
 759                         $lastClass = 0;
 760                 }
 761                 $out .= $startChar . $combining;
 762                 return $out;
 763         }
 764
 765         /**
 766          * This is just used for the benchmark, comparing how long it takes to
 767          * interate through a string without really doing anything of substance.
 768          * @param string $string
 769          * @return string
 770          */
 771         function placebo( $string ) {
 772                 $len = strlen( $string );
 773                 $out = '';
 774                 for( $i = 0; $i < $len; $i++ ) {
 775                         $out .= $string{$i};
 776                 }
 777                 return $out;
 778         }
 779 }
 780
 781 ?>