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  * 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 dirname(__FILE__).'/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', "\xea\xb0\x80" /*codepointToUtf8( UNICODE_HANGUL_FIRST )*/ );
  70 define( 'UTF8_HANGUL_LAST', "\xed\x9e\xa3" /*codepointToUtf8( UNICODE_HANGUL_LAST )*/ );
  71
  72 define( 'UTF8_HANGUL_LBASE', "\xe1\x84\x80" /*codepointToUtf8( UNICODE_HANGUL_LBASE )*/ );
  73 define( 'UTF8_HANGUL_VBASE', "\xe1\x85\xa1" /*codepointToUtf8( UNICODE_HANGUL_VBASE )*/ );
  74 define( 'UTF8_HANGUL_TBASE', "\xe1\x86\xa7" /*codepointToUtf8( UNICODE_HANGUL_TBASE )*/ );
  75
  76 define( 'UTF8_HANGUL_LEND', "\xe1\x84\x92" /*codepointToUtf8( UNICODE_HANGUL_LEND )*/ );
  77 define( 'UTF8_HANGUL_VEND', "\xe1\x85\xb5" /*codepointToUtf8( UNICODE_HANGUL_VEND )*/ );
  78 define( 'UTF8_HANGUL_TEND', "\xe1\x87\x82" /*codepointToUtf8( UNICODE_HANGUL_TEND )*/ );
  79
  80 define( 'UTF8_SURROGATE_FIRST', "\xed\xa0\x80" /*codepointToUtf8( UNICODE_SURROGATE_FIRST )*/ );
  81 define( 'UTF8_SURROGATE_LAST', "\xed\xbf\xbf" /*codepointToUtf8( UNICODE_SURROGATE_LAST )*/ );
  82 define( 'UTF8_MAX', "\xf4\x8f\xbf\xbf" /*codepointToUtf8( UNICODE_MAX )*/ );
  83 define( 'UTF8_REPLACEMENT', "\xef\xbf\xbd" /*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', "\xef\xb7\x90" /*codepointToUtf8( 0xfdd0 )*/ );
  92 define( 'UTF8_FDEF', "\xef\xb7\xaf" /*codepointToUtf8( 0xfdef )*/ );
  93 define( 'UTF8_FFFE', "\xef\xbf\xbe" /*codepointToUtf8( 0xfffe )*/ );
  94 define( 'UTF8_FFFF', "\xef\xbf\xbf" /*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      * @static
 128          */
 129         static function cleanUp( $string ) {
 130                 if( NORMALIZE_ICU ) {
 131                         # We exclude a few chars that ICU would not.
 132                         $string = preg_replace(
 133                                 '/[\x00-\x08\x0b\x0c\x0e-\x1f]/',
 134                                 UTF8_REPLACEMENT,
 135                                 $string );
 136                         $string = str_replace( UTF8_FFFE, UTF8_REPLACEMENT, $string );
 137                         $string = str_replace( UTF8_FFFF, UTF8_REPLACEMENT, $string );
 138
 139                         # UnicodeString constructor fails if the string ends with a
 140                         # head byte. Add a junk char at the end, we'll strip it off.
 141                         return rtrim( utf8_normalize( $string . "\x01", UNORM_NFC ), "\x01" );
 142                 } elseif( UtfNormal::quickIsNFCVerify( $string ) ) {
 143                         # Side effect -- $string has had UTF-8 errors cleaned up.
 144                         return $string;
 145                 } else {
 146                         return UtfNormal::NFC( $string );
 147                 }
 148         }
 149
 150         /**
 151          * Convert a UTF-8 string to normal form C, canonical composition.
 152          * Fast return for pure ASCII strings; some lesser optimizations for
 153          * strings containing only known-good characters.
 154          *
 155          * @param string $string a valid UTF-8 string. Input is not validated.
 156          * @return string a UTF-8 string in normal form C
 157          * @static
 158          */
 159         static function toNFC( $string ) {
 160                 if( NORMALIZE_ICU )
 161                         return utf8_normalize( $string, UNORM_NFC );
 162                 elseif( UtfNormal::quickIsNFC( $string ) )
 163                         return $string;
 164                 else
 165                         return UtfNormal::NFC( $string );
 166         }
 167
 168         /**
 169          * Convert a UTF-8 string to normal form D, canonical decomposition.
 170          * Fast return for pure ASCII strings.
 171          *
 172          * @param string $string a valid UTF-8 string. Input is not validated.
 173          * @return string a UTF-8 string in normal form D
 174          * @static
 175          */
 176         static function toNFD( $string ) {
 177                 if( NORMALIZE_ICU )
 178                         return utf8_normalize( $string, UNORM_NFD );
 179                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 180                         return UtfNormal::NFD( $string );
 181                 else
 182                         return $string;
 183         }
 184
 185         /**
 186          * Convert a UTF-8 string to normal form KC, compatibility composition.
 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 KC
 192          * @static
 193          */
 194         static function toNFKC( $string ) {
 195                 if( NORMALIZE_ICU )
 196                         return utf8_normalize( $string, UNORM_NFKC );
 197                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 198                         return UtfNormal::NFKC( $string );
 199                 else
 200                         return $string;
 201         }
 202
 203         /**
 204          * Convert a UTF-8 string to normal form KD, compatibility decomposition.
 205          * This may cause irreversible information loss, use judiciously.
 206          * Fast return for pure ASCII strings.
 207          *
 208          * @param string $string a valid UTF-8 string. Input is not validated.
 209          * @return string a UTF-8 string in normal form KD
 210          * @static
 211          */
 212         static function toNFKD( $string ) {
 213                 if( NORMALIZE_ICU )
 214                         return utf8_normalize( $string, UNORM_NFKD );
 215                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 216                         return UtfNormal::NFKD( $string );
 217                 else
 218                         return $string;
 219         }
 220
 221         /**
 222          * Load the basic composition data if necessary
 223          * @private
 224          * @static
 225          */
 226         static function loadData() {
 227                 # fixme : are $utfCanonicalComp, $utfCanonicalDecomp really used?
 228                 global $utfCombiningClass, $utfCanonicalComp, $utfCanonicalDecomp;
 229                 if( !isset( $utfCombiningClass ) ) {
 230                         require_once( 'UtfNormalData.inc' );
 231                 }
 232         }
 233
 234         /**
 235          * Returns true if the string is _definitely_ in NFC.
 236          * Returns false if not or uncertain.
 237          * @param string $string a valid UTF-8 string. Input is not validated.
 238          * @return bool
 239          * @static
 240          */
 241         static function quickIsNFC( $string ) {
 242                 # ASCII is always valid NFC!
 243                 # If it's pure ASCII, let it through.
 244                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 245
 246                 UtfNormal::loadData();
 247                 global $utfCheckNFC, $utfCombiningClass;
 248                 $len = strlen( $string );
 249                 for( $i = 0; $i < $len; $i++ ) {
 250                         $c = $string{$i};
 251                         $n = ord( $c );
 252                         if( $n < 0x80 ) {
 253                                 continue;
 254                         } elseif( $n >= 0xf0 ) {
 255                                 $c = substr( $string, $i, 4 );
 256                                 $i += 3;
 257                         } elseif( $n >= 0xe0 ) {
 258                                 $c = substr( $string, $i, 3 );
 259                                 $i += 2;
 260                         } elseif( $n >= 0xc0 ) {
 261                                 $c = substr( $string, $i, 2 );
 262                                 $i++;
 263                         }
 264                         if( isset( $utfCheckNFC[$c] ) ) {
 265                                 # If it's NO or MAYBE, bail and do the slow check.
 266                                 return false;
 267                         }
 268                         if( isset( $utfCombiningClass[$c] ) ) {
 269                                 # Combining character? We might have to do sorting, at least.
 270                                 return false;
 271                         }
 272                 }
 273                 return true;
 274         }
 275
 276         /**
 277          * Returns true if the string is _definitely_ in NFC.
 278          * Returns false if not or uncertain.
 279          * @param string $string a UTF-8 string, altered on output to be valid UTF-8 safe for XML.
 280          * @static
 281          */
 282         static function quickIsNFCVerify( &$string ) {
 283                 # Screen out some characters that eg won't be allowed in XML
 284                 $string = preg_replace( '/[\x00-\x08\x0b\x0c\x0e-\x1f]/', UTF8_REPLACEMENT, $string );
 285
 286                 # ASCII is always valid NFC!
 287                 # If we're only ever given plain ASCII, we can avoid the overhead
 288                 # of initializing the decomposition tables by skipping out early.
 289                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 290
 291                 static $checkit = null, $tailBytes = null, $utfCheckOrCombining = null;
 292                 if( !isset( $checkit ) ) {
 293                         # Load/build some scary lookup tables...
 294                         UtfNormal::loadData();
 295                         global $utfCheckNFC, $utfCombiningClass;
 296
 297                         $utfCheckOrCombining = array_merge( $utfCheckNFC, $utfCombiningClass );
 298
 299                         # Head bytes for sequences which we should do further validity checks
 300                         $checkit = array_flip( array_map( 'chr',
 301                                         array( 0xc0, 0xc1, 0xe0, 0xed, 0xef,
 302                                                    0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7,
 303                                                    0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff ) ) );
 304
 305                         # Each UTF-8 head byte is followed by a certain
 306                         # number of tail bytes.
 307                         $tailBytes = array();
 308                         for( $n = 0; $n < 256; $n++ ) {
 309                                 if( $n < 0xc0 ) {
 310                                         $remaining = 0;
 311                                 } elseif( $n < 0xe0 ) {
 312                                         $remaining = 1;
 313                                 } elseif( $n < 0xf0 ) {
 314                                         $remaining = 2;
 315                                 } elseif( $n < 0xf8 ) {
 316                                         $remaining = 3;
 317                                 } elseif( $n < 0xfc ) {
 318                                         $remaining = 4;
 319                                 } elseif( $n < 0xfe ) {
 320                                         $remaining = 5;
 321                                 } else {
 322                                         $remaining = 0;
 323                                 }
 324                                 $tailBytes[chr($n)] = $remaining;
 325                         }
 326                 }
 327
 328                 # Chop the text into pure-ASCII and non-ASCII areas;
 329                 # large ASCII parts can be handled much more quickly.
 330                 # Don't chop up Unicode areas for punctuation, though,
 331                 # that wastes energy.
 332                 preg_match_all(
 333                         '/([\x00-\x7f]+|[\x80-\xff][\x00-\x40\x5b-\x5f\x7b-\xff]*)/',
 334                         $string, $matches );
 335
 336                 $looksNormal = true;
 337                 $base = 0;
 338                 $replace = array();
 339                 foreach( $matches[1] as $str ) {
 340                         $chunk = strlen( $str );
 341
 342                         if( $str{0} < "\x80" ) {
 343                                 # ASCII chunk: guaranteed to be valid UTF-8
 344                                 # and in normal form C, so skip over it.
 345                                 $base += $chunk;
 346                                 continue;
 347                         }
 348
 349                         # We'll have to examine the chunk byte by byte to ensure
 350                         # that it consists of valid UTF-8 sequences, and to see
 351                         # if any of them might not be normalized.
 352                         #
 353                         # Since PHP is not the fastest language on earth, some of
 354                         # this code is a little ugly with inner loop optimizations.
 355
 356                         $head = '';
 357                         $len = $chunk + 1; # Counting down is faster. I'm *so* sorry.
 358
 359                         for( $i = -1; --$len; ) {
 360                                 if( $remaining = $tailBytes[$c = $str{++$i}] ) {
 361                                         # UTF-8 head byte!
 362                                         $sequence = $head = $c;
 363                                         do {
 364                                                 # Look for the defined number of tail bytes...
 365                                                 if( --$len && ( $c = $str{++$i} ) >= "\x80" && $c < "\xc0" ) {
 366                                                         # Legal tail bytes are nice.
 367                                                         $sequence .= $c;
 368                                                 } else {
 369                                                         if( 0 == $len ) {
 370                                                                 # Premature end of string!
 371                                                                 # Drop a replacement character into output to
 372                                                                 # represent the invalid UTF-8 sequence.
 373                                                                 $replace[] = array( UTF8_REPLACEMENT,
 374                                                                                                         $base + $i + 1 - strlen( $sequence ),
 375                                                                                                         strlen( $sequence ) );
 376                                                                 break 2;
 377                                                         } else {
 378                                                                 # Illegal tail byte; abandon the sequence.
 379                                                                 $replace[] = array( UTF8_REPLACEMENT,
 380                                                                                                         $base + $i - strlen( $sequence ),
 381                                                                                                         strlen( $sequence ) );
 382                                                                 # Back up and reprocess this byte; it may itself
 383                                                                 # be a legal ASCII or UTF-8 sequence head.
 384                                                                 --$i;
 385                                                                 ++$len;
 386                                                                 continue 2;
 387                                                         }
 388                                                 }
 389                                         } while( --$remaining );
 390
 391                                         if( isset( $checkit[$head] ) ) {
 392                                                 # Do some more detailed validity checks, for
 393                                                 # invalid characters and illegal sequences.
 394                                                 if( $head == "\xed" ) {
 395                                                         # 0xed is relatively frequent in Korean, which
 396                                                         # abuts the surrogate area, so we're doing
 397                                                         # this check separately to speed things up.
 398
 399                                                         if( $sequence >= UTF8_SURROGATE_FIRST ) {
 400                                                                 # Surrogates are legal only in UTF-16 code.
 401                                                                 # They are totally forbidden here in UTF-8
 402                                                                 # utopia.
 403                                                                 $replace[] = array( UTF8_REPLACEMENT,
 404                                                                              $base + $i + 1 - strlen( $sequence ),
 405                                                                              strlen( $sequence ) );
 406                                                                 $head = '';
 407                                                                 continue;
 408                                                         }
 409                                                 } else {
 410                                                         # Slower, but rarer checks...
 411                                                         $n = ord( $head );
 412                                                         if(
 413                                                                 # "Overlong sequences" are those that are syntactically
 414                                                                 # correct but use more UTF-8 bytes than are necessary to
 415                                                                 # encode a character. Naïve string comparisons can be
 416                                                                 # tricked into failing to see a match for an ASCII
 417                                                                 # character, for instance, which can be a security hole
 418                                                                 # if blacklist checks are being used.
 419                                                                ($n  < 0xc2 && $sequence <= UTF8_OVERLONG_A)
 420                                                                 || ($n == 0xe0 && $sequence <= UTF8_OVERLONG_B)
 421                                                                 || ($n == 0xf0 && $sequence <= UTF8_OVERLONG_C)
 422
 423                                                                 # U+FFFE and U+FFFF are explicitly forbidden in Unicode.
 424                                                                 || ($n == 0xef &&
 425                                                                            ($sequence == UTF8_FFFE)
 426                                                                         || ($sequence == UTF8_FFFF) )
 427
 428                                                                 # Unicode has been limited to 21 bits; longer
 429                                                                 # sequences are not allowed.
 430                                                                 || ($n >= 0xf0 && $sequence > UTF8_MAX) ) {
 431
 432                                                                 $replace[] = array( UTF8_REPLACEMENT,
 433                                                                                     $base + $i + 1 - strlen( $sequence ),
 434                                                                                     strlen( $sequence ) );
 435                                                                 $head = '';
 436                                                                 continue;
 437                                                         }
 438                                                 }
 439                                         }
 440
 441                                         if( isset( $utfCheckOrCombining[$sequence] ) ) {
 442                                                 # If it's NO or MAYBE, we'll have to rip
 443                                                 # the string apart and put it back together.
 444                                                 # That's going to be mighty slow.
 445                                                 $looksNormal = false;
 446                                         }
 447
 448                                         # The sequence is legal!
 449                                         $head = '';
 450                                 } elseif( $c < "\x80" ) {
 451                                         # ASCII byte.
 452                                         $head = '';
 453                                 } elseif( $c < "\xc0" ) {
 454                                         # Illegal tail bytes
 455                                         if( $head == '' ) {
 456                                                 # Out of the blue!
 457                                                 $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 458                                         } else {
 459                                                 # Don't add if we're continuing a broken sequence;
 460                                                 # we already put a replacement character when we looked
 461                                                 # at the broken sequence.
 462                                                 $replace[] = array( '', $base + $i, 1 );
 463                                         }
 464                                 } else {
 465                                         # Miscellaneous freaks.
 466                                         $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 467                                         $head = '';
 468                                 }
 469                         }
 470                         $base += $chunk;
 471                 }
 472                 if( count( $replace ) ) {
 473                         # There were illegal UTF-8 sequences we need to fix up.
 474                         $out = '';
 475                         $last = 0;
 476                         foreach( $replace as $rep ) {
 477                                 list( $replacement, $start, $length ) = $rep;
 478                                 if( $last < $start ) {
 479                                         $out .= substr( $string, $last, $start - $last );
 480                                 }
 481                                 $out .= $replacement;
 482                                 $last = $start + $length;
 483                         }
 484                         if( $last < strlen( $string ) ) {
 485                                 $out .= substr( $string, $last );
 486                         }
 487                         $string = $out;
 488                 }
 489                 return $looksNormal;
 490         }
 491
 492         # These take a string and run the normalization on them, without
 493         # checking for validity or any optimization etc. Input must be
 494         # VALID UTF-8!
 495         /**
 496          * @param string $string
 497          * @return string
 498          * @private
 499          * @static
 500          */
 501         static function NFC( $string ) {
 502                 return UtfNormal::fastCompose( UtfNormal::NFD( $string ) );
 503         }
 504
 505         /**
 506          * @param string $string
 507          * @return string
 508          * @private
 509          * @static
 510          */
 511         static function NFD( $string ) {
 512                 UtfNormal::loadData();
 513                 global $utfCanonicalDecomp;
 514                 return UtfNormal::fastCombiningSort(
 515                         UtfNormal::fastDecompose( $string, $utfCanonicalDecomp ) );
 516         }
 517
 518         /**
 519          * @param string $string
 520          * @return string
 521          * @private
 522          * @static
 523          */
 524         static function NFKC( $string ) {
 525                 return UtfNormal::fastCompose( UtfNormal::NFKD( $string ) );
 526         }
 527
 528         /**
 529          * @param string $string
 530          * @return string
 531          * @private
 532          * @static
 533          */
 534         static function NFKD( $string ) {
 535                 global $utfCompatibilityDecomp;
 536                 if( !isset( $utfCompatibilityDecomp ) ) {
 537                         require_once( 'UtfNormalDataK.inc' );
 538                 }
 539                 return UtfNormal::fastCombiningSort(
 540                         UtfNormal::fastDecompose( $string, $utfCompatibilityDecomp ) );
 541         }
 542
 543
 544         /**
 545          * Perform decomposition of a UTF-8 string into either D or KD form
 546          * (depending on which decomposition map is passed to us).
 547          * Input is assumed to be *valid* UTF-8. Invalid code will break.
 548          * @private
 549          * @param string $string Valid UTF-8 string
 550          * @param array $map hash of expanded decomposition map
 551          * @return string a UTF-8 string decomposed, not yet normalized (needs sorting)
 552          * @static
 553          */
 554         static function fastDecompose( $string, &$map ) {
 555                 UtfNormal::loadData();
 556                 $len = strlen( $string );
 557                 $out = '';
 558                 for( $i = 0; $i < $len; $i++ ) {
 559                         $c = $string{$i};
 560                         $n = ord( $c );
 561                         if( $n < 0x80 ) {
 562                                 # ASCII chars never decompose
 563                                 # THEY ARE IMMORTAL
 564                                 $out .= $c;
 565                                 continue;
 566                         } elseif( $n >= 0xf0 ) {
 567                                 $c = substr( $string, $i, 4 );
 568                                 $i += 3;
 569                         } elseif( $n >= 0xe0 ) {
 570                                 $c = substr( $string, $i, 3 );
 571                                 $i += 2;
 572                         } elseif( $n >= 0xc0 ) {
 573                                 $c = substr( $string, $i, 2 );
 574                                 $i++;
 575                         }
 576                         if( isset( $map[$c] ) ) {
 577                                 $out .= $map[$c];
 578                                 continue;
 579                         } else {
 580                                 if( $c >= UTF8_HANGUL_FIRST && $c <= UTF8_HANGUL_LAST ) {
 581                                         # Decompose a hangul syllable into jamo;
 582                                         # hardcoded for three-byte UTF-8 sequence.
 583                                         # A lookup table would be slightly faster,
 584                                         # but adds a lot of memory & disk needs.
 585                                         #
 586                                         $index = ( (ord( $c{0} ) & 0x0f) << 12
 587                                                  | (ord( $c{1} ) & 0x3f) <<  6
 588                                                  | (ord( $c{2} ) & 0x3f) )
 589                                                - UNICODE_HANGUL_FIRST;
 590                                         $l = intval( $index / UNICODE_HANGUL_NCOUNT );
 591                                         $v = intval( ($index % UNICODE_HANGUL_NCOUNT) / UNICODE_HANGUL_TCOUNT);
 592                                         $t = $index % UNICODE_HANGUL_TCOUNT;
 593                                         $out .= "\xe1\x84" . chr( 0x80 + $l ) . "\xe1\x85" . chr( 0xa1 + $v );
 594                                         if( $t >= 25 ) {
 595                                                 $out .= "\xe1\x87" . chr( 0x80 + $t - 25 );
 596                                         } elseif( $t ) {
 597                                                 $out .= "\xe1\x86" . chr( 0xa7 + $t );
 598                                         }
 599                                         continue;
 600                                 }
 601                         }
 602                         $out .= $c;
 603                 }
 604                 return $out;
 605         }
 606
 607         /**
 608          * Sorts combining characters into canonical order. This is the
 609          * final step in creating decomposed normal forms D and KD.
 610          * @private
 611          * @param string $string a valid, decomposed UTF-8 string. Input is not validated.
 612          * @return string a UTF-8 string with combining characters sorted in canonical order
 613          * @static
 614          */
 615         static function fastCombiningSort( $string ) {
 616                 UtfNormal::loadData();
 617                 global $utfCombiningClass;
 618                 $len = strlen( $string );
 619                 $out = '';
 620                 $combiners = array();
 621                 $lastClass = -1;
 622                 for( $i = 0; $i < $len; $i++ ) {
 623                         $c = $string{$i};
 624                         $n = ord( $c );
 625                         if( $n >= 0x80 ) {
 626                                 if( $n >= 0xf0 ) {
 627                                         $c = substr( $string, $i, 4 );
 628                                         $i += 3;
 629                                 } elseif( $n >= 0xe0 ) {
 630                                         $c = substr( $string, $i, 3 );
 631                                         $i += 2;
 632                                 } elseif( $n >= 0xc0 ) {
 633                                         $c = substr( $string, $i, 2 );
 634                                         $i++;
 635                                 }
 636                                 if( isset( $utfCombiningClass[$c] ) ) {
 637                                         $lastClass = $utfCombiningClass[$c];
 638                                         @$combiners[$lastClass] .= $c;
 639                                         continue;
 640                                 }
 641                         }
 642                         if( $lastClass ) {
 643                                 ksort( $combiners );
 644                                 $out .= implode( '', $combiners );
 645                                 $combiners = array();
 646                         }
 647                         $out .= $c;
 648                         $lastClass = 0;
 649                 }
 650                 if( $lastClass ) {
 651                         ksort( $combiners );
 652                         $out .= implode( '', $combiners );
 653                 }
 654                 return $out;
 655         }
 656
 657         /**
 658          * Produces canonically composed sequences, i.e. normal form C or KC.
 659          *
 660          * @private
 661          * @param string $string a valid UTF-8 string in sorted normal form D or KD. Input is not validated.
 662          * @return string a UTF-8 string with canonical precomposed characters used where possible
 663          * @static
 664          */
 665         static function fastCompose( $string ) {
 666                 UtfNormal::loadData();
 667                 global $utfCanonicalComp, $utfCombiningClass;
 668                 $len = strlen( $string );
 669                 $out = '';
 670                 $lastClass = -1;
 671                 $lastHangul = 0;
 672                 $startChar = '';
 673                 $combining = '';
 674                 $x1 = ord(substr(UTF8_HANGUL_VBASE,0,1));
 675                 $x2 = ord(substr(UTF8_HANGUL_TEND,0,1));
 676                 for( $i = 0; $i < $len; $i++ ) {
 677                         $c = $string{$i};
 678                         $n = ord( $c );
 679                         if( $n < 0x80 ) {
 680                                 # No combining characters here...
 681                                 $out .= $startChar;
 682                                 $out .= $combining;
 683                                 $startChar = $c;
 684                                 $combining = '';
 685                                 $lastClass = 0;
 686                                 continue;
 687                         } elseif( $n >= 0xf0 ) {
 688                                 $c = substr( $string, $i, 4 );
 689                                 $i += 3;
 690                         } elseif( $n >= 0xe0 ) {
 691                                 $c = substr( $string, $i, 3 );
 692                                 $i += 2;
 693                         } elseif( $n >= 0xc0 ) {
 694                                 $c = substr( $string, $i, 2 );
 695                                 $i++;
 696                         }
 697                         $pair = $startChar . $c;
 698                         if( $n > 0x80 ) {
 699                                 if( isset( $utfCombiningClass[$c] ) ) {
 700                                         # A combining char; see what we can do with it
 701                                         $class = $utfCombiningClass[$c];
 702                                         if( !empty( $startChar ) &&
 703                                                 $lastClass < $class &&
 704                                                 $class > 0 &&
 705                                                 isset( $utfCanonicalComp[$pair] ) ) {
 706                                                 $startChar = $utfCanonicalComp[$pair];
 707                                                 $class = 0;
 708                                         } else {
 709                                                 $combining .= $c;
 710                                         }
 711                                         $lastClass = $class;
 712                                         $lastHangul = 0;
 713                                         continue;
 714                                 }
 715                         }
 716                         # New start char
 717                         if( $lastClass == 0 ) {
 718                                 if( isset( $utfCanonicalComp[$pair] ) ) {
 719                                         $startChar = $utfCanonicalComp[$pair];
 720                                         $lastHangul = 0;
 721                                         continue;
 722                                 }
 723                                 if( $n >= $x1 && $n <= $x2 ) {
 724                                         # WARNING: Hangul code is painfully slow.
 725                                         # I apologize for this ugly, ugly code; however
 726                                         # performance is even more teh suck if we call
 727                                         # out to nice clean functions. Lookup tables are
 728                                         # marginally faster, but require a lot of space.
 729                                         #
 730                                         if( $c >= UTF8_HANGUL_VBASE &&
 731                                                 $c <= UTF8_HANGUL_VEND &&
 732                                                 $startChar >= UTF8_HANGUL_LBASE &&
 733                                                 $startChar <= UTF8_HANGUL_LEND ) {
 734                                                 #
 735                                                 #$lIndex = utf8ToCodepoint( $startChar ) - UNICODE_HANGUL_LBASE;
 736                                                 #$vIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_VBASE;
 737                                                 $lIndex = ord( $startChar{2} ) - 0x80;
 738                                                 $vIndex = ord( $c{2}         ) - 0xa1;
 739
 740                                                 $hangulPoint = UNICODE_HANGUL_FIRST +
 741                                                         UNICODE_HANGUL_TCOUNT *
 742                                                         (UNICODE_HANGUL_VCOUNT * $lIndex + $vIndex);
 743
 744                                                 # Hardcode the limited-range UTF-8 conversion:
 745                                                 $startChar = chr( $hangulPoint >> 12 & 0x0f | 0xe0 ) .
 746                                                                          chr( $hangulPoint >>  6 & 0x3f | 0x80 ) .
 747                                                                          chr( $hangulPoint       & 0x3f | 0x80 );
 748                                                 $lastHangul = 0;
 749                                                 continue;
 750                                         } elseif( $c >= UTF8_HANGUL_TBASE &&
 751                                                           $c <= UTF8_HANGUL_TEND &&
 752                                                           $startChar >= UTF8_HANGUL_FIRST &&
 753                                                           $startChar <= UTF8_HANGUL_LAST &&
 754                                                           !$lastHangul ) {
 755                                                 # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
 756                                                 $tIndex = ord( $c{2} ) - 0xa7;
 757                                                 if( $tIndex < 0 ) $tIndex = ord( $c{2} ) - 0x80 + (0x11c0 - 0x11a7);
 758
 759                                                 # Increment the code point by $tIndex, without
 760                                                 # the function overhead of decoding and recoding UTF-8
 761                                                 #
 762                                                 $tail = ord( $startChar{2} ) + $tIndex;
 763                                                 if( $tail > 0xbf ) {
 764                                                         $tail -= 0x40;
 765                                                         $mid = ord( $startChar{1} ) + 1;
 766                                                         if( $mid > 0xbf ) {
 767                                                                 $startChar{0} = chr( ord( $startChar{0} ) + 1 );
 768                                                                 $mid -= 0x40;
 769                                                         }
 770                                                         $startChar{1} = chr( $mid );
 771                                                 }
 772                                                 $startChar{2} = chr( $tail );
 773
 774                                                 # If there's another jamo char after this, *don't* try to merge it.
 775                                                 $lastHangul = 1;
 776                                                 continue;
 777                                         }
 778                                 }
 779                         }
 780                         $out .= $startChar;
 781                         $out .= $combining;
 782                         $startChar = $c;
 783                         $combining = '';
 784                         $lastClass = 0;
 785                         $lastHangul = 0;
 786                 }
 787                 $out .= $startChar . $combining;
 788                 return $out;
 789         }
 790
 791         /**
 792          * This is just used for the benchmark, comparing how long it takes to
 793          * interate through a string without really doing anything of substance.
 794          * @param string $string
 795          * @return string
 796          * @static
 797          */
 798         static function placebo( $string ) {
 799                 $len = strlen( $string );
 800                 $out = '';
 801                 for( $i = 0; $i < $len; $i++ ) {
 802                         $out .= $string{$i};
 803                 }
 804                 return $out;
 805         }
 806 }
 807
 808 ?>