includes/normal/UtfNormal.php

   1 <?php
   2 /**
   3  * Unicode normalization routines
   4  *
   5  * Copyright © 2004 Brion Vibber <brion@pobox.com>
   6  * http://www.mediawiki.org/
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 2 of the License, or
  11  * (at your option) any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License along
  19  * with this program; if not, write to the Free Software Foundation, Inc.,
  20  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
  21  * http://www.gnu.org/copyleft/gpl.html
  22  *
  23  * @file
  24  * @ingroup UtfNormal
  25  */
  26
  27 /**
  28  * @defgroup UtfNormal UtfNormal
  29  */
  30
  31 require_once dirname(__FILE__).'/UtfNormalUtil.php';
  32
  33 /**
  34  * For using the ICU wrapper
  35  */
  36 define( 'UNORM_NONE', 1 );
  37 define( 'UNORM_NFD',  2 );
  38 define( 'UNORM_NFKD', 3 );
  39 define( 'UNORM_NFC',  4 );
  40 define( 'UNORM_DEFAULT', UNORM_NFC );
  41 define( 'UNORM_NFKC', 5 );
  42 define( 'UNORM_FCD',  6 );
  43
  44 define( 'NORMALIZE_ICU', function_exists( 'utf8_normalize' ) );
  45 define( 'NORMALIZE_INTL', function_exists( 'normalizer_normalize' ) );
  46
  47 /**
  48  * Unicode normalization routines for working with UTF-8 strings.
  49  * Currently assumes that input strings are valid UTF-8!
  50  *
  51  * Not as fast as I'd like, but should be usable for most purposes.
  52  * UtfNormal::toNFC() will bail early if given ASCII text or text
  53  * it can quickly deterimine is already normalized.
  54  *
  55  * All functions can be called static.
  56  *
  57  * See description of forms at http://www.unicode.org/reports/tr15/
  58  *
  59  * @ingroup UtfNormal
  60  */
  61 class UtfNormal {
  62         static $utfCombiningClass = null;
  63         static $utfCanonicalComp = null;
  64         static $utfCanonicalDecomp = null;
  65
  66         # Load compatibility decompositions on demand if they are needed.
  67         static $utfCompatibilityDecomp = null;
  68
  69         static $utfCheckNFC;
  70
  71         /**
  72          * The ultimate convenience function! Clean up invalid UTF-8 sequences,
  73          * and convert to normal form C, canonical composition.
  74          *
  75          * Fast return for pure ASCII strings; some lesser optimizations for
  76          * strings containing only known-good characters. Not as fast as toNFC().
  77          *
  78          * @param $string String: a UTF-8 string
  79          * @return string a clean, shiny, normalized UTF-8 string
  80          */
  81         static function cleanUp( $string ) {
  82                 if( NORMALIZE_ICU || NORMALIZE_INTL ) {
  83                         # We exclude a few chars that ICU would not.
  84                         $string = preg_replace(
  85                                 '/[\x00-\x08\x0b\x0c\x0e-\x1f]/',
  86                                 UTF8_REPLACEMENT,
  87                                 $string );
  88                         $string = str_replace( UTF8_FFFE, UTF8_REPLACEMENT, $string );
  89                         $string = str_replace( UTF8_FFFF, UTF8_REPLACEMENT, $string );
  90
  91                         # UnicodeString constructor fails if the string ends with a
  92                         # head byte. Add a junk char at the end, we'll strip it off.
  93                         if ( NORMALIZE_ICU ) return rtrim( utf8_normalize( $string . "\x01", UNORM_NFC ), "\x01" );
  94                         if ( NORMALIZE_INTL ) return normalizer_normalize( $string, Normalizer::FORM_C );
  95                 } elseif( UtfNormal::quickIsNFCVerify( $string ) ) {
  96                         # Side effect -- $string has had UTF-8 errors cleaned up.
  97                         return $string;
  98                 } else {
  99                         return UtfNormal::NFC( $string );
 100                 }
 101         }
 102
 103         /**
 104          * Convert a UTF-8 string to normal form C, canonical composition.
 105          * Fast return for pure ASCII strings; some lesser optimizations for
 106          * strings containing only known-good characters.
 107          *
 108          * @param $string String: a valid UTF-8 string. Input is not validated.
 109          * @return string a UTF-8 string in normal form C
 110          */
 111         static function toNFC( $string ) {
 112                 if( NORMALIZE_INTL )
 113                         return normalizer_normalize( $string, Normalizer::FORM_C );
 114                 elseif( NORMALIZE_ICU )
 115                         return utf8_normalize( $string, UNORM_NFC );
 116                 elseif( UtfNormal::quickIsNFC( $string ) )
 117                         return $string;
 118                 else
 119                         return UtfNormal::NFC( $string );
 120         }
 121
 122         /**
 123          * Convert a UTF-8 string to normal form D, canonical decomposition.
 124          * Fast return for pure ASCII strings.
 125          *
 126          * @param $string String: a valid UTF-8 string. Input is not validated.
 127          * @return string a UTF-8 string in normal form D
 128          */
 129         static function toNFD( $string ) {
 130                 if( NORMALIZE_INTL )
 131                         return normalizer_normalize( $string, Normalizer::FORM_D );
 132                 elseif( NORMALIZE_ICU )
 133                         return utf8_normalize( $string, UNORM_NFD );
 134                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 135                         return UtfNormal::NFD( $string );
 136                 else
 137                         return $string;
 138         }
 139
 140         /**
 141          * Convert a UTF-8 string to normal form KC, compatibility composition.
 142          * This may cause irreversible information loss, use judiciously.
 143          * Fast return for pure ASCII strings.
 144          *
 145          * @param $string String: a valid UTF-8 string. Input is not validated.
 146          * @return string a UTF-8 string in normal form KC
 147          */
 148         static function toNFKC( $string ) {
 149                 if( NORMALIZE_INTL )
 150                         return normalizer_normalize( $string, Normalizer::FORM_KC );
 151                 elseif( NORMALIZE_ICU )
 152                         return utf8_normalize( $string, UNORM_NFKC );
 153                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 154                         return UtfNormal::NFKC( $string );
 155                 else
 156                         return $string;
 157         }
 158
 159         /**
 160          * Convert a UTF-8 string to normal form KD, compatibility decomposition.
 161          * This may cause irreversible information loss, use judiciously.
 162          * Fast return for pure ASCII strings.
 163          *
 164          * @param $string String: a valid UTF-8 string. Input is not validated.
 165          * @return string a UTF-8 string in normal form KD
 166          */
 167         static function toNFKD( $string ) {
 168                 if( NORMALIZE_INTL )
 169                         return normalizer_normalize( $string, Normalizer::FORM_KD );
 170                 elseif( NORMALIZE_ICU )
 171                         return utf8_normalize( $string, UNORM_NFKD );
 172                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 173                         return UtfNormal::NFKD( $string );
 174                 else
 175                         return $string;
 176         }
 177
 178         /**
 179          * Load the basic composition data if necessary
 180          * @private
 181          */
 182         static function loadData() {
 183                 if( !isset( self::$utfCombiningClass ) ) {
 184                         require_once( dirname(__FILE__) . '/UtfNormalData.inc' );
 185                 }
 186         }
 187
 188         /**
 189          * Returns true if the string is _definitely_ in NFC.
 190          * Returns false if not or uncertain.
 191          * @param $string String: a valid UTF-8 string. Input is not validated.
 192          * @return bool
 193          */
 194         static function quickIsNFC( $string ) {
 195                 # ASCII is always valid NFC!
 196                 # If it's pure ASCII, let it through.
 197                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 198
 199                 UtfNormal::loadData();
 200                 $len = strlen( $string );
 201                 for( $i = 0; $i < $len; $i++ ) {
 202                         $c = $string{$i};
 203                         $n = ord( $c );
 204                         if( $n < 0x80 ) {
 205                                 continue;
 206                         } elseif( $n >= 0xf0 ) {
 207                                 $c = substr( $string, $i, 4 );
 208                                 $i += 3;
 209                         } elseif( $n >= 0xe0 ) {
 210                                 $c = substr( $string, $i, 3 );
 211                                 $i += 2;
 212                         } elseif( $n >= 0xc0 ) {
 213                                 $c = substr( $string, $i, 2 );
 214                                 $i++;
 215                         }
 216                         if( isset( self::$utfCheckNFC[$c] ) ) {
 217                                 # If it's NO or MAYBE, bail and do the slow check.
 218                                 return false;
 219                         }
 220                         if( isset( self::$utfCombiningClass[$c] ) ) {
 221                                 # Combining character? We might have to do sorting, at least.
 222                                 return false;
 223                         }
 224                 }
 225                 return true;
 226         }
 227
 228         /**
 229          * Returns true if the string is _definitely_ in NFC.
 230          * Returns false if not or uncertain.
 231          * @param $string String: a UTF-8 string, altered on output to be valid UTF-8 safe for XML.
 232          */
 233         static function quickIsNFCVerify( &$string ) {
 234                 # Screen out some characters that eg won't be allowed in XML
 235                 $string = preg_replace( '/[\x00-\x08\x0b\x0c\x0e-\x1f]/', UTF8_REPLACEMENT, $string );
 236
 237                 # ASCII is always valid NFC!
 238                 # If we're only ever given plain ASCII, we can avoid the overhead
 239                 # of initializing the decomposition tables by skipping out early.
 240                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 241
 242                 static $checkit = null, $tailBytes = null, $utfCheckOrCombining = null;
 243                 if( !isset( $checkit ) ) {
 244                         # Load/build some scary lookup tables...
 245                         UtfNormal::loadData();
 246
 247                         $utfCheckOrCombining = array_merge( self::$utfCheckNFC, self::$utfCombiningClass );
 248
 249                         # Head bytes for sequences which we should do further validity checks
 250                         $checkit = array_flip( array_map( 'chr',
 251                                         array( 0xc0, 0xc1, 0xe0, 0xed, 0xef,
 252                                                    0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7,
 253                                                    0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff ) ) );
 254
 255                         # Each UTF-8 head byte is followed by a certain
 256                         # number of tail bytes.
 257                         $tailBytes = array();
 258                         for( $n = 0; $n < 256; $n++ ) {
 259                                 if( $n < 0xc0 ) {
 260                                         $remaining = 0;
 261                                 } elseif( $n < 0xe0 ) {
 262                                         $remaining = 1;
 263                                 } elseif( $n < 0xf0 ) {
 264                                         $remaining = 2;
 265                                 } elseif( $n < 0xf8 ) {
 266                                         $remaining = 3;
 267                                 } elseif( $n < 0xfc ) {
 268                                         $remaining = 4;
 269                                 } elseif( $n < 0xfe ) {
 270                                         $remaining = 5;
 271                                 } else {
 272                                         $remaining = 0;
 273                                 }
 274                                 $tailBytes[chr($n)] = $remaining;
 275                         }
 276                 }
 277
 278                 # Chop the text into pure-ASCII and non-ASCII areas;
 279                 # large ASCII parts can be handled much more quickly.
 280                 # Don't chop up Unicode areas for punctuation, though,
 281                 # that wastes energy.
 282                 $matches = array();
 283                 preg_match_all(
 284                         '/([\x00-\x7f]+|[\x80-\xff][\x00-\x40\x5b-\x5f\x7b-\xff]*)/',
 285                         $string, $matches );
 286
 287                 $looksNormal = true;
 288                 $base = 0;
 289                 $replace = array();
 290                 foreach( $matches[1] as $str ) {
 291                         $chunk = strlen( $str );
 292
 293                         if( $str{0} < "\x80" ) {
 294                                 # ASCII chunk: guaranteed to be valid UTF-8
 295                                 # and in normal form C, so skip over it.
 296                                 $base += $chunk;
 297                                 continue;
 298                         }
 299
 300                         # We'll have to examine the chunk byte by byte to ensure
 301                         # that it consists of valid UTF-8 sequences, and to see
 302                         # if any of them might not be normalized.
 303                         #
 304                         # Since PHP is not the fastest language on earth, some of
 305                         # this code is a little ugly with inner loop optimizations.
 306
 307                         $head = '';
 308                         $len = $chunk + 1; # Counting down is faster. I'm *so* sorry.
 309
 310                         for( $i = -1; --$len; ) {
 311                                 $remaining = $tailBytes[$c = $str{++$i}]
 312                                 if( $remaining ) {
 313                                         # UTF-8 head byte!
 314                                         $sequence = $head = $c;
 315                                         do {
 316                                                 # Look for the defined number of tail bytes...
 317                                                 if( --$len && ( $c = $str{++$i} ) >= "\x80" && $c < "\xc0" ) {
 318                                                         # Legal tail bytes are nice.
 319                                                         $sequence .= $c;
 320                                                 } else {
 321                                                         if( 0 == $len ) {
 322                                                                 # Premature end of string!
 323                                                                 # Drop a replacement character into output to
 324                                                                 # represent the invalid UTF-8 sequence.
 325                                                                 $replace[] = array( UTF8_REPLACEMENT,
 326                                                                                                         $base + $i + 1 - strlen( $sequence ),
 327                                                                                                         strlen( $sequence ) );
 328                                                                 break 2;
 329                                                         } else {
 330                                                                 # Illegal tail byte; abandon the sequence.
 331                                                                 $replace[] = array( UTF8_REPLACEMENT,
 332                                                                                                         $base + $i - strlen( $sequence ),
 333                                                                                                         strlen( $sequence ) );
 334                                                                 # Back up and reprocess this byte; it may itself
 335                                                                 # be a legal ASCII or UTF-8 sequence head.
 336                                                                 --$i;
 337                                                                 ++$len;
 338                                                                 continue 2;
 339                                                         }
 340                                                 }
 341                                         } while( --$remaining );
 342
 343                                         if( isset( $checkit[$head] ) ) {
 344                                                 # Do some more detailed validity checks, for
 345                                                 # invalid characters and illegal sequences.
 346                                                 if( $head == "\xed" ) {
 347                                                         # 0xed is relatively frequent in Korean, which
 348                                                         # abuts the surrogate area, so we're doing
 349                                                         # this check separately to speed things up.
 350
 351                                                         if( $sequence >= UTF8_SURROGATE_FIRST ) {
 352                                                                 # Surrogates are legal only in UTF-16 code.
 353                                                                 # They are totally forbidden here in UTF-8
 354                                                                 # utopia.
 355                                                                 $replace[] = array( UTF8_REPLACEMENT,
 356                                                                              $base + $i + 1 - strlen( $sequence ),
 357                                                                              strlen( $sequence ) );
 358                                                                 $head = '';
 359                                                                 continue;
 360                                                         }
 361                                                 } else {
 362                                                         # Slower, but rarer checks...
 363                                                         $n = ord( $head );
 364                                                         if(
 365                                                                 # "Overlong sequences" are those that are syntactically
 366                                                                 # correct but use more UTF-8 bytes than are necessary to
 367                                                                 # encode a character. Naïve string comparisons can be
 368                                                                 # tricked into failing to see a match for an ASCII
 369                                                                 # character, for instance, which can be a security hole
 370                                                                 # if blacklist checks are being used.
 371                                                                ($n  < 0xc2 && $sequence <= UTF8_OVERLONG_A)
 372                                                                 || ($n == 0xe0 && $sequence <= UTF8_OVERLONG_B)
 373                                                                 || ($n == 0xf0 && $sequence <= UTF8_OVERLONG_C)
 374
 375                                                                 # U+FFFE and U+FFFF are explicitly forbidden in Unicode.
 376                                                                 || ($n == 0xef &&
 377                                                                            ($sequence == UTF8_FFFE)
 378                                                                         || ($sequence == UTF8_FFFF) )
 379
 380                                                                 # Unicode has been limited to 21 bits; longer
 381                                                                 # sequences are not allowed.
 382                                                                 || ($n >= 0xf0 && $sequence > UTF8_MAX) ) {
 383
 384                                                                 $replace[] = array( UTF8_REPLACEMENT,
 385                                                                                     $base + $i + 1 - strlen( $sequence ),
 386                                                                                     strlen( $sequence ) );
 387                                                                 $head = '';
 388                                                                 continue;
 389                                                         }
 390                                                 }
 391                                         }
 392
 393                                         if( isset( $utfCheckOrCombining[$sequence] ) ) {
 394                                                 # If it's NO or MAYBE, we'll have to rip
 395                                                 # the string apart and put it back together.
 396                                                 # That's going to be mighty slow.
 397                                                 $looksNormal = false;
 398                                         }
 399
 400                                         # The sequence is legal!
 401                                         $head = '';
 402                                 } elseif( $c < "\x80" ) {
 403                                         # ASCII byte.
 404                                         $head = '';
 405                                 } elseif( $c < "\xc0" ) {
 406                                         # Illegal tail bytes
 407                                         if( $head == '' ) {
 408                                                 # Out of the blue!
 409                                                 $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 410                                         } else {
 411                                                 # Don't add if we're continuing a broken sequence;
 412                                                 # we already put a replacement character when we looked
 413                                                 # at the broken sequence.
 414                                                 $replace[] = array( '', $base + $i, 1 );
 415                                         }
 416                                 } else {
 417                                         # Miscellaneous freaks.
 418                                         $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 419                                         $head = '';
 420                                 }
 421                         }
 422                         $base += $chunk;
 423                 }
 424                 if( count( $replace ) ) {
 425                         # There were illegal UTF-8 sequences we need to fix up.
 426                         $out = '';
 427                         $last = 0;
 428                         foreach( $replace as $rep ) {
 429                                 list( $replacement, $start, $length ) = $rep;
 430                                 if( $last < $start ) {
 431                                         $out .= substr( $string, $last, $start - $last );
 432                                 }
 433                                 $out .= $replacement;
 434                                 $last = $start + $length;
 435                         }
 436                         if( $last < strlen( $string ) ) {
 437                                 $out .= substr( $string, $last );
 438                         }
 439                         $string = $out;
 440                 }
 441                 return $looksNormal;
 442         }
 443
 444         # These take a string and run the normalization on them, without
 445         # checking for validity or any optimization etc. Input must be
 446         # VALID UTF-8!
 447         /**
 448          * @param $string string
 449          * @return string
 450          * @private
 451          */
 452         static function NFC( $string ) {
 453                 return UtfNormal::fastCompose( UtfNormal::NFD( $string ) );
 454         }
 455
 456         /**
 457          * @param $string string
 458          * @return string
 459          * @private
 460          */
 461         static function NFD( $string ) {
 462                 UtfNormal::loadData();
 463
 464                 return UtfNormal::fastCombiningSort(
 465                         UtfNormal::fastDecompose( $string, self::$utfCanonicalDecomp ) );
 466         }
 467
 468         /**
 469          * @param $string string
 470          * @return string
 471          * @private
 472          */
 473         static function NFKC( $string ) {
 474                 return UtfNormal::fastCompose( UtfNormal::NFKD( $string ) );
 475         }
 476
 477         /**
 478          * @param $string string
 479          * @return string
 480          * @private
 481          */
 482         static function NFKD( $string ) {
 483                 if( !isset( self::$utfCompatibilityDecomp ) ) {
 484                         require_once( 'UtfNormalDataK.inc' );
 485                 }
 486                 return self::fastCombiningSort(
 487                         self::fastDecompose( $string, self::$utfCompatibilityDecomp ) );
 488         }
 489
 490
 491         /**
 492          * Perform decomposition of a UTF-8 string into either D or KD form
 493          * (depending on which decomposition map is passed to us).
 494          * Input is assumed to be *valid* UTF-8. Invalid code will break.
 495          * @private
 496          * @param $string String: valid UTF-8 string
 497          * @param $map Array: hash of expanded decomposition map
 498          * @return string a UTF-8 string decomposed, not yet normalized (needs sorting)
 499          */
 500         static function fastDecompose( $string, $map ) {
 501                 UtfNormal::loadData();
 502                 $len = strlen( $string );
 503                 $out = '';
 504                 for( $i = 0; $i < $len; $i++ ) {
 505                         $c = $string{$i};
 506                         $n = ord( $c );
 507                         if( $n < 0x80 ) {
 508                                 # ASCII chars never decompose
 509                                 # THEY ARE IMMORTAL
 510                                 $out .= $c;
 511                                 continue;
 512                         } elseif( $n >= 0xf0 ) {
 513                                 $c = substr( $string, $i, 4 );
 514                                 $i += 3;
 515                         } elseif( $n >= 0xe0 ) {
 516                                 $c = substr( $string, $i, 3 );
 517                                 $i += 2;
 518                         } elseif( $n >= 0xc0 ) {
 519                                 $c = substr( $string, $i, 2 );
 520                                 $i++;
 521                         }
 522                         if( isset( $map[$c] ) ) {
 523                                 $out .= $map[$c];
 524                                 continue;
 525                         } else {
 526                                 if( $c >= UTF8_HANGUL_FIRST && $c <= UTF8_HANGUL_LAST ) {
 527                                         # Decompose a hangul syllable into jamo;
 528                                         # hardcoded for three-byte UTF-8 sequence.
 529                                         # A lookup table would be slightly faster,
 530                                         # but adds a lot of memory & disk needs.
 531                                         #
 532                                         $index = ( (ord( $c{0} ) & 0x0f) << 12
 533                                                  | (ord( $c{1} ) & 0x3f) <<  6
 534                                                  | (ord( $c{2} ) & 0x3f) )
 535                                                - UNICODE_HANGUL_FIRST;
 536                                         $l = intval( $index / UNICODE_HANGUL_NCOUNT );
 537                                         $v = intval( ($index % UNICODE_HANGUL_NCOUNT) / UNICODE_HANGUL_TCOUNT);
 538                                         $t = $index % UNICODE_HANGUL_TCOUNT;
 539                                         $out .= "\xe1\x84" . chr( 0x80 + $l ) . "\xe1\x85" . chr( 0xa1 + $v );
 540                                         if( $t >= 25 ) {
 541                                                 $out .= "\xe1\x87" . chr( 0x80 + $t - 25 );
 542                                         } elseif( $t ) {
 543                                                 $out .= "\xe1\x86" . chr( 0xa7 + $t );
 544                                         }
 545                                         continue;
 546                                 }
 547                         }
 548                         $out .= $c;
 549                 }
 550                 return $out;
 551         }
 552
 553         /**
 554          * Sorts combining characters into canonical order. This is the
 555          * final step in creating decomposed normal forms D and KD.
 556          * @private
 557          * @param $string String: a valid, decomposed UTF-8 string. Input is not validated.
 558          * @return string a UTF-8 string with combining characters sorted in canonical order
 559          */
 560         static function fastCombiningSort( $string ) {
 561                 UtfNormal::loadData();
 562                 $len = strlen( $string );
 563                 $out = '';
 564                 $combiners = array();
 565                 $lastClass = -1;
 566                 for( $i = 0; $i < $len; $i++ ) {
 567                         $c = $string{$i};
 568                         $n = ord( $c );
 569                         if( $n >= 0x80 ) {
 570                                 if( $n >= 0xf0 ) {
 571                                         $c = substr( $string, $i, 4 );
 572                                         $i += 3;
 573                                 } elseif( $n >= 0xe0 ) {
 574                                         $c = substr( $string, $i, 3 );
 575                                         $i += 2;
 576                                 } elseif( $n >= 0xc0 ) {
 577                                         $c = substr( $string, $i, 2 );
 578                                         $i++;
 579                                 }
 580                                 if( isset( self::$utfCombiningClass[$c] ) ) {
 581                                         $lastClass = self::$utfCombiningClass[$c];
 582                                         if( isset( $combiners[$lastClass] ) ) {
 583                                                 $combiners[$lastClass] .= $c;
 584                                         } else {
 585                                                 $combiners[$lastClass] = $c;
 586                                         }
 587                                         continue;
 588                                 }
 589                         }
 590                         if( $lastClass ) {
 591                                 ksort( $combiners );
 592                                 $out .= implode( '', $combiners );
 593                                 $combiners = array();
 594                         }
 595                         $out .= $c;
 596                         $lastClass = 0;
 597                 }
 598                 if( $lastClass ) {
 599                         ksort( $combiners );
 600                         $out .= implode( '', $combiners );
 601                 }
 602                 return $out;
 603         }
 604
 605         /**
 606          * Produces canonically composed sequences, i.e. normal form C or KC.
 607          *
 608          * @private
 609          * @param $string String: a valid UTF-8 string in sorted normal form D or KD. Input is not validated.
 610          * @return string a UTF-8 string with canonical precomposed characters used where possible
 611          */
 612         static function fastCompose( $string ) {
 613                 UtfNormal::loadData();
 614                 $len = strlen( $string );
 615                 $out = '';
 616                 $lastClass = -1;
 617                 $lastHangul = 0;
 618                 $startChar = '';
 619                 $combining = '';
 620                 $x1 = ord(substr(UTF8_HANGUL_VBASE,0,1));
 621                 $x2 = ord(substr(UTF8_HANGUL_TEND,0,1));
 622                 for( $i = 0; $i < $len; $i++ ) {
 623                         $c = $string{$i};
 624                         $n = ord( $c );
 625                         if( $n < 0x80 ) {
 626                                 # No combining characters here...
 627                                 $out .= $startChar;
 628                                 $out .= $combining;
 629                                 $startChar = $c;
 630                                 $combining = '';
 631                                 $lastClass = 0;
 632                                 continue;
 633                         } elseif( $n >= 0xf0 ) {
 634                                 $c = substr( $string, $i, 4 );
 635                                 $i += 3;
 636                         } elseif( $n >= 0xe0 ) {
 637                                 $c = substr( $string, $i, 3 );
 638                                 $i += 2;
 639                         } elseif( $n >= 0xc0 ) {
 640                                 $c = substr( $string, $i, 2 );
 641                                 $i++;
 642                         }
 643                         $pair = $startChar . $c;
 644                         if( $n > 0x80 ) {
 645                                 if( isset( self::$utfCombiningClass[$c] ) ) {
 646                                         # A combining char; see what we can do with it
 647                                         $class = self::$utfCombiningClass[$c];
 648                                         if( !empty( $startChar ) &&
 649                                                 $lastClass < $class &&
 650                                                 $class > 0 &&
 651                                                 isset( self::$utfCanonicalComp[$pair] ) ) {
 652                                                 $startChar = self::$utfCanonicalComp[$pair];
 653                                                 $class = 0;
 654                                         } else {
 655                                                 $combining .= $c;
 656                                         }
 657                                         $lastClass = $class;
 658                                         $lastHangul = 0;
 659                                         continue;
 660                                 }
 661                         }
 662                         # New start char
 663                         if( $lastClass == 0 ) {
 664                                 if( isset( self::$utfCanonicalComp[$pair] ) ) {
 665                                         $startChar = self::$utfCanonicalComp[$pair];
 666                                         $lastHangul = 0;
 667                                         continue;
 668                                 }
 669                                 if( $n >= $x1 && $n <= $x2 ) {
 670                                         # WARNING: Hangul code is painfully slow.
 671                                         # I apologize for this ugly, ugly code; however
 672                                         # performance is even more teh suck if we call
 673                                         # out to nice clean functions. Lookup tables are
 674                                         # marginally faster, but require a lot of space.
 675                                         #
 676                                         if( $c >= UTF8_HANGUL_VBASE &&
 677                                                 $c <= UTF8_HANGUL_VEND &&
 678                                                 $startChar >= UTF8_HANGUL_LBASE &&
 679                                                 $startChar <= UTF8_HANGUL_LEND ) {
 680                                                 #
 681                                                 #$lIndex = utf8ToCodepoint( $startChar ) - UNICODE_HANGUL_LBASE;
 682                                                 #$vIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_VBASE;
 683                                                 $lIndex = ord( $startChar{2} ) - 0x80;
 684                                                 $vIndex = ord( $c{2}         ) - 0xa1;
 685
 686                                                 $hangulPoint = UNICODE_HANGUL_FIRST +
 687                                                         UNICODE_HANGUL_TCOUNT *
 688                                                         (UNICODE_HANGUL_VCOUNT * $lIndex + $vIndex);
 689
 690                                                 # Hardcode the limited-range UTF-8 conversion:
 691                                                 $startChar = chr( $hangulPoint >> 12 & 0x0f | 0xe0 ) .
 692                                                                          chr( $hangulPoint >>  6 & 0x3f | 0x80 ) .
 693                                                                          chr( $hangulPoint       & 0x3f | 0x80 );
 694                                                 $lastHangul = 0;
 695                                                 continue;
 696                                         } elseif( $c >= UTF8_HANGUL_TBASE &&
 697                                                           $c <= UTF8_HANGUL_TEND &&
 698                                                           $startChar >= UTF8_HANGUL_FIRST &&
 699                                                           $startChar <= UTF8_HANGUL_LAST &&
 700                                                           !$lastHangul ) {
 701                                                 # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
 702                                                 $tIndex = ord( $c{2} ) - 0xa7;
 703                                                 if( $tIndex < 0 ) $tIndex = ord( $c{2} ) - 0x80 + (0x11c0 - 0x11a7);
 704
 705                                                 # Increment the code point by $tIndex, without
 706                                                 # the function overhead of decoding and recoding UTF-8
 707                                                 #
 708                                                 $tail = ord( $startChar{2} ) + $tIndex;
 709                                                 if( $tail > 0xbf ) {
 710                                                         $tail -= 0x40;
 711                                                         $mid = ord( $startChar{1} ) + 1;
 712                                                         if( $mid > 0xbf ) {
 713                                                                 $startChar{0} = chr( ord( $startChar{0} ) + 1 );
 714                                                                 $mid -= 0x40;
 715                                                         }
 716                                                         $startChar{1} = chr( $mid );
 717                                                 }
 718                                                 $startChar{2} = chr( $tail );
 719
 720                                                 # If there's another jamo char after this, *don't* try to merge it.
 721                                                 $lastHangul = 1;
 722                                                 continue;
 723                                         }
 724                                 }
 725                         }
 726                         $out .= $startChar;
 727                         $out .= $combining;
 728                         $startChar = $c;
 729                         $combining = '';
 730                         $lastClass = 0;
 731                         $lastHangul = 0;
 732                 }
 733                 $out .= $startChar . $combining;
 734                 return $out;
 735         }
 736
 737         /**
 738          * This is just used for the benchmark, comparing how long it takes to
 739          * interate through a string without really doing anything of substance.
 740          * @param $string string
 741          * @return string
 742          */
 743         static function placebo( $string ) {
 744                 $len = strlen( $string );
 745                 $out = '';
 746                 for( $i = 0; $i < $len; $i++ ) {
 747                         $out .= $string{$i};
 748                 }
 749                 return $out;
 750         }
 751 }