includes/normal/UtfNormal.php

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