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 '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          */
 128         function cleanUp( $string ) {
 129                 if( NORMALIZE_ICU ) {
 130                         # We exclude a few chars that ICU would not.
 131                         $string = preg_replace(
 132                                 '/[\x00-\x08\x0b\x0c\x0e-\x1f]/',
 133                                 UTF8_REPLACEMENT,
 134                                 $string );
 135                         $string = str_replace( UTF8_FFFE, UTF8_REPLACEMENT, $string );
 136                         $string = str_replace( UTF8_FFFF, UTF8_REPLACEMENT, $string );
 137
 138                         # UnicodeString constructor fails if the string ends with a
 139                         # head byte. Add a junk char at the end, we'll strip it off.
 140                         return rtrim( utf8_normalize( $string . "\x01", UNORM_NFC ), "\x01" );
 141                 } elseif( UtfNormal::quickIsNFCVerify( $string ) ) {
 142                         # Side effect -- $string has had UTF-8 errors cleaned up.
 143                         return $string;
 144                 } else {
 145                         return UtfNormal::NFC( $string );
 146                 }
 147         }
 148
 149         /**
 150          * Convert a UTF-8 string to normal form C, canonical composition.
 151          * Fast return for pure ASCII strings; some lesser optimizations for
 152          * strings containing only known-good characters.
 153          *
 154          * @param string $string a valid UTF-8 string. Input is not validated.
 155          * @return string a UTF-8 string in normal form C
 156          */
 157         function toNFC( $string ) {
 158                 if( NORMALIZE_ICU )
 159                         return utf8_normalize( $string, UNORM_NFC );
 160                 elseif( UtfNormal::quickIsNFC( $string ) )
 161                         return $string;
 162                 else
 163                         return UtfNormal::NFC( $string );
 164         }
 165
 166         /**
 167          * Convert a UTF-8 string to normal form D, canonical decomposition.
 168          * Fast return for pure ASCII strings.
 169          *
 170          * @param string $string a valid UTF-8 string. Input is not validated.
 171          * @return string a UTF-8 string in normal form D
 172          */
 173         function toNFD( $string ) {
 174                 if( NORMALIZE_ICU )
 175                         return utf8_normalize( $string, UNORM_NFD );
 176                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 177                         return UtfNormal::NFD( $string );
 178                 else
 179                         return $string;
 180         }
 181
 182         /**
 183          * Convert a UTF-8 string to normal form KC, compatibility composition.
 184          * This may cause irreversible information loss, use judiciously.
 185          * Fast return for pure ASCII strings.
 186          *
 187          * @param string $string a valid UTF-8 string. Input is not validated.
 188          * @return string a UTF-8 string in normal form KC
 189          */
 190         function toNFKC( $string ) {
 191                 if( NORMALIZE_ICU )
 192                         return utf8_normalize( $string, UNORM_NFKC );
 193                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 194                         return UtfNormal::NFKC( $string );
 195                 else
 196                         return $string;
 197         }
 198
 199         /**
 200          * Convert a UTF-8 string to normal form KD, compatibility decomposition.
 201          * This may cause irreversible information loss, use judiciously.
 202          * Fast return for pure ASCII strings.
 203          *
 204          * @param string $string a valid UTF-8 string. Input is not validated.
 205          * @return string a UTF-8 string in normal form KD
 206          */
 207         function toNFKD( $string ) {
 208                 if( NORMALIZE_ICU )
 209                         return utf8_normalize( $string, UNORM_NFKD );
 210                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 211                         return UtfNormal::NFKD( $string );
 212                 else
 213                         return $string;
 214         }
 215
 216         /**
 217          * Load the basic composition data if necessary
 218          * @private
 219          */
 220         function loadData() {
 221                 # fixme : are $utfCanonicalComp, $utfCanonicalDecomp really used?
 222                 global $utfCombiningClass, $utfCanonicalComp, $utfCanonicalDecomp;
 223                 if( !isset( $utfCombiningClass ) ) {
 224                         require_once( 'UtfNormalData.inc' );
 225                 }
 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 valid UTF-8 string. Input is not validated.
 232          * @return bool
 233          */
 234         function quickIsNFC( $string ) {
 235                 # ASCII is always valid NFC!
 236                 # If it's pure ASCII, let it through.
 237                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 238
 239                 UtfNormal::loadData();
 240                 global $utfCheckNFC, $utfCombiningClass;
 241                 $len = strlen( $string );
 242                 for( $i = 0; $i < $len; $i++ ) {
 243                         $c = $string{$i};
 244                         $n = ord( $c );
 245                         if( $n < 0x80 ) {
 246                                 continue;
 247                         } elseif( $n >= 0xf0 ) {
 248                                 $c = substr( $string, $i, 4 );
 249                                 $i += 3;
 250                         } elseif( $n >= 0xe0 ) {
 251                                 $c = substr( $string, $i, 3 );
 252                                 $i += 2;
 253                         } elseif( $n >= 0xc0 ) {
 254                                 $c = substr( $string, $i, 2 );
 255                                 $i++;
 256                         }
 257                         if( isset( $utfCheckNFC[$c] ) ) {
 258                                 # If it's NO or MAYBE, bail and do the slow check.
 259                                 return false;
 260                         }
 261                         if( isset( $utfCombiningClass[$c] ) ) {
 262                                 # Combining character? We might have to do sorting, at least.
 263                                 return false;
 264                         }
 265                 }
 266                 return true;
 267         }
 268
 269         /**
 270          * Returns true if the string is _definitely_ in NFC.
 271          * Returns false if not or uncertain.
 272          * @param string $string a UTF-8 string, altered on output to be valid UTF-8 safe for XML.
 273          */
 274         function quickIsNFCVerify( &$string ) {
 275                 # Screen out some characters that eg won't be allowed in XML
 276                 $string = preg_replace( '/[\x00-\x08\x0b\x0c\x0e-\x1f]/', UTF8_REPLACEMENT, $string );
 277
 278                 # ASCII is always valid NFC!
 279                 # If we're only ever given plain ASCII, we can avoid the overhead
 280                 # of initializing the decomposition tables by skipping out early.
 281                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 282
 283                 static $checkit = null, $tailBytes = null, $utfCheckOrCombining = null;
 284                 if( !isset( $checkit ) ) {
 285                         # Load/build some scary lookup tables...
 286                         UtfNormal::loadData();
 287                         global $utfCheckNFC, $utfCombiningClass;
 288
 289                         $utfCheckOrCombining = array_merge( $utfCheckNFC, $utfCombiningClass );
 290
 291                         # Head bytes for sequences which we should do further validity checks
 292                         $checkit = array_flip( array_map( 'chr',
 293                                         array( 0xc0, 0xc1, 0xe0, 0xed, 0xef,
 294                                                    0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7,
 295                                                    0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff ) ) );
 296
 297                         # Each UTF-8 head byte is followed by a certain
 298                         # number of tail bytes.
 299                         $tailBytes = array();
 300                         for( $n = 0; $n < 256; $n++ ) {
 301                                 if( $n < 0xc0 ) {
 302                                         $remaining = 0;
 303                                 } elseif( $n < 0xe0 ) {
 304                                         $remaining = 1;
 305                                 } elseif( $n < 0xf0 ) {
 306                                         $remaining = 2;
 307                                 } elseif( $n < 0xf8 ) {
 308                                         $remaining = 3;
 309                                 } elseif( $n < 0xfc ) {
 310                                         $remaining = 4;
 311                                 } elseif( $n < 0xfe ) {
 312                                         $remaining = 5;
 313                                 } else {
 314                                         $remaining = 0;
 315                                 }
 316                                 $tailBytes[chr($n)] = $remaining;
 317                         }
 318                 }
 319
 320                 # Chop the text into pure-ASCII and non-ASCII areas;
 321                 # large ASCII parts can be handled much more quickly.
 322                 # Don't chop up Unicode areas for punctuation, though,
 323                 # that wastes energy.
 324                 preg_match_all(
 325                         '/([\x00-\x7f]+|[\x80-\xff][\x00-\x40\x5b-\x5f\x7b-\xff]*)/',
 326                         $string, $matches );
 327
 328                 $looksNormal = true;
 329                 $base = 0;
 330                 $replace = array();
 331                 foreach( $matches[1] as $str ) {
 332                         $chunk = strlen( $str );
 333
 334                         if( $str{0} < "\x80" ) {
 335                                 # ASCII chunk: guaranteed to be valid UTF-8
 336                                 # and in normal form C, so skip over it.
 337                                 $base += $chunk;
 338                                 continue;
 339                         }
 340
 341                         # We'll have to examine the chunk byte by byte to ensure
 342                         # that it consists of valid UTF-8 sequences, and to see
 343                         # if any of them might not be normalized.
 344                         #
 345                         # Since PHP is not the fastest language on earth, some of
 346                         # this code is a little ugly with inner loop optimizations.
 347
 348                         $head = '';
 349                         $len = $chunk + 1; # Counting down is faster. I'm *so* sorry.
 350
 351                         for( $i = -1; --$len; ) {
 352                                 if( $remaining = $tailBytes[$c = $str{++$i}] ) {
 353                                         # UTF-8 head byte!
 354                                         $sequence = $head = $c;
 355                                         do {
 356                                                 # Look for the defined number of tail bytes...
 357                                                 if( --$len && ( $c = $str{++$i} ) >= "\x80" && $c < "\xc0" ) {
 358                                                         # Legal tail bytes are nice.
 359                                                         $sequence .= $c;
 360                                                 } else {
 361                                                         if( 0 == $len ) {
 362                                                                 # Premature end of string!
 363                                                                 # Drop a replacement character into output to
 364                                                                 # represent the invalid UTF-8 sequence.
 365                                                                 $replace[] = array( UTF8_REPLACEMENT,
 366                                                                                                         $base + $i + 1 - strlen( $sequence ),
 367                                                                                                         strlen( $sequence ) );
 368                                                                 break 2;
 369                                                         } else {
 370                                                                 # Illegal tail byte; abandon the sequence.
 371                                                                 $replace[] = array( UTF8_REPLACEMENT,
 372                                                                                                         $base + $i - strlen( $sequence ),
 373                                                                                                         strlen( $sequence ) );
 374                                                                 # Back up and reprocess this byte; it may itself
 375                                                                 # be a legal ASCII or UTF-8 sequence head.
 376                                                                 --$i;
 377                                                                 ++$len;
 378                                                                 continue 2;
 379                                                         }
 380                                                 }
 381                                         } while( --$remaining );
 382
 383                                         if( isset( $checkit[$head] ) ) {
 384                                                 # Do some more detailed validity checks, for
 385                                                 # invalid characters and illegal sequences.
 386                                                 if( $head == "\xed" ) {
 387                                                         # 0xed is relatively frequent in Korean, which
 388                                                         # abuts the surrogate area, so we're doing
 389                                                         # this check separately to speed things up.
 390
 391                                                         if( $sequence >= UTF8_SURROGATE_FIRST ) {
 392                                                                 # Surrogates are legal only in UTF-16 code.
 393                                                                 # They are totally forbidden here in UTF-8
 394                                                                 # utopia.
 395                                                                 $replace[] = array( UTF8_REPLACEMENT,
 396                                                                              $base + $i + 1 - strlen( $sequence ),
 397                                                                              strlen( $sequence ) );
 398                                                                 $head = '';
 399                                                                 continue;
 400                                                         }
 401                                                 } else {
 402                                                         # Slower, but rarer checks...
 403                                                         $n = ord( $head );
 404                                                         if(
 405                                                                 # "Overlong sequences" are those that are syntactically
 406                                                                 # correct but use more UTF-8 bytes than are necessary to
 407                                                                 # encode a character. Naïve string comparisons can be
 408                                                                 # tricked into failing to see a match for an ASCII
 409                                                                 # character, for instance, which can be a security hole
 410                                                                 # if blacklist checks are being used.
 411                                                                ($n  < 0xc2 && $sequence <= UTF8_OVERLONG_A)
 412                                                                 || ($n == 0xe0 && $sequence <= UTF8_OVERLONG_B)
 413                                                                 || ($n == 0xf0 && $sequence <= UTF8_OVERLONG_C)
 414
 415                                                                 # U+FFFE and U+FFFF are explicitly forbidden in Unicode.
 416                                                                 || ($n == 0xef &&
 417                                                                            ($sequence == UTF8_FFFE)
 418                                                                         || ($sequence == UTF8_FFFF) )
 419
 420                                                                 # Unicode has been limited to 21 bits; longer
 421                                                                 # sequences are not allowed.
 422                                                                 || ($n >= 0xf0 && $sequence > UTF8_MAX) ) {
 423
 424                                                                 $replace[] = array( UTF8_REPLACEMENT,
 425                                                                                     $base + $i + 1 - strlen( $sequence ),
 426                                                                                     strlen( $sequence ) );
 427                                                                 $head = '';
 428                                                                 continue;
 429                                                         }
 430                                                 }
 431                                         }
 432
 433                                         if( isset( $utfCheckOrCombining[$sequence] ) ) {
 434                                                 # If it's NO or MAYBE, we'll have to rip
 435                                                 # the string apart and put it back together.
 436                                                 # That's going to be mighty slow.
 437                                                 $looksNormal = false;
 438                                         }
 439
 440                                         # The sequence is legal!
 441                                         $head = '';
 442                                 } elseif( $c < "\x80" ) {
 443                                         # ASCII byte.
 444                                         $head = '';
 445                                 } elseif( $c < "\xc0" ) {
 446                                         # Illegal tail bytes
 447                                         if( $head == '' ) {
 448                                                 # Out of the blue!
 449                                                 $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 450                                         } else {
 451                                                 # Don't add if we're continuing a broken sequence;
 452                                                 # we already put a replacement character when we looked
 453                                                 # at the broken sequence.
 454                                                 $replace[] = array( '', $base + $i, 1 );
 455                                         }
 456                                 } else {
 457                                         # Miscellaneous freaks.
 458                                         $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 459                                         $head = '';
 460                                 }
 461                         }
 462                         $base += $chunk;
 463                 }
 464                 if( count( $replace ) ) {
 465                         # There were illegal UTF-8 sequences we need to fix up.
 466                         $out = '';
 467                         $last = 0;
 468                         foreach( $replace as $rep ) {
 469                                 list( $replacement, $start, $length ) = $rep;
 470                                 if( $last < $start ) {
 471                                         $out .= substr( $string, $last, $start - $last );
 472                                 }
 473                                 $out .= $replacement;
 474                                 $last = $start + $length;
 475                         }
 476                         if( $last < strlen( $string ) ) {
 477                                 $out .= substr( $string, $last );
 478                         }
 479                         $string = $out;
 480                 }
 481                 return $looksNormal;
 482         }
 483
 484         # These take a string and run the normalization on them, without
 485         # checking for validity or any optimization etc. Input must be
 486         # VALID UTF-8!
 487         /**
 488          * @param string $string
 489          * @return string
 490          * @private
 491          */
 492         function NFC( $string ) {
 493                 return UtfNormal::fastCompose( UtfNormal::NFD( $string ) );
 494         }
 495
 496         /**
 497          * @param string $string
 498          * @return string
 499          * @private
 500          */
 501         function NFD( $string ) {
 502                 UtfNormal::loadData();
 503                 global $utfCanonicalDecomp;
 504                 return UtfNormal::fastCombiningSort(
 505                         UtfNormal::fastDecompose( $string, $utfCanonicalDecomp ) );
 506         }
 507
 508         /**
 509          * @param string $string
 510          * @return string
 511          * @private
 512          */
 513         function NFKC( $string ) {
 514                 return UtfNormal::fastCompose( UtfNormal::NFKD( $string ) );
 515         }
 516
 517         /**
 518          * @param string $string
 519          * @return string
 520          * @private
 521          */
 522         function NFKD( $string ) {
 523                 global $utfCompatibilityDecomp;
 524                 if( !isset( $utfCompatibilityDecomp ) ) {
 525                         require_once( 'UtfNormalDataK.inc' );
 526                 }
 527                 return UtfNormal::fastCombiningSort(
 528                         UtfNormal::fastDecompose( $string, $utfCompatibilityDecomp ) );
 529         }
 530
 531
 532         /**
 533          * Perform decomposition of a UTF-8 string into either D or KD form
 534          * (depending on which decomposition map is passed to us).
 535          * Input is assumed to be *valid* UTF-8. Invalid code will break.
 536          * @private
 537          * @param string $string Valid UTF-8 string
 538          * @param array $map hash of expanded decomposition map
 539          * @return string a UTF-8 string decomposed, not yet normalized (needs sorting)
 540          */
 541         function fastDecompose( $string, &$map ) {
 542                 UtfNormal::loadData();
 543                 $len = strlen( $string );
 544                 $out = '';
 545                 for( $i = 0; $i < $len; $i++ ) {
 546                         $c = $string{$i};
 547                         $n = ord( $c );
 548                         if( $n < 0x80 ) {
 549                                 # ASCII chars never decompose
 550                                 # THEY ARE IMMORTAL
 551                                 $out .= $c;
 552                                 continue;
 553                         } elseif( $n >= 0xf0 ) {
 554                                 $c = substr( $string, $i, 4 );
 555                                 $i += 3;
 556                         } elseif( $n >= 0xe0 ) {
 557                                 $c = substr( $string, $i, 3 );
 558                                 $i += 2;
 559                         } elseif( $n >= 0xc0 ) {
 560                                 $c = substr( $string, $i, 2 );
 561                                 $i++;
 562                         }
 563                         if( isset( $map[$c] ) ) {
 564                                 $out .= $map[$c];
 565                                 continue;
 566                         } else {
 567                                 if( $c >= UTF8_HANGUL_FIRST && $c <= UTF8_HANGUL_LAST ) {
 568                                         # Decompose a hangul syllable into jamo;
 569                                         # hardcoded for three-byte UTF-8 sequence.
 570                                         # A lookup table would be slightly faster,
 571                                         # but adds a lot of memory & disk needs.
 572                                         #
 573                                         $index = ( (ord( $c{0} ) & 0x0f) << 12
 574                                                  | (ord( $c{1} ) & 0x3f) <<  6
 575                                                  | (ord( $c{2} ) & 0x3f) )
 576                                                - UNICODE_HANGUL_FIRST;
 577                                         $l = intval( $index / UNICODE_HANGUL_NCOUNT );
 578                                         $v = intval( ($index % UNICODE_HANGUL_NCOUNT) / UNICODE_HANGUL_TCOUNT);
 579                                         $t = $index % UNICODE_HANGUL_TCOUNT;
 580                                         $out .= "\xe1\x84" . chr( 0x80 + $l ) . "\xe1\x85" . chr( 0xa1 + $v );
 581                                         if( $t >= 25 ) {
 582                                                 $out .= "\xe1\x87" . chr( 0x80 + $t - 25 );
 583                                         } elseif( $t ) {
 584                                                 $out .= "\xe1\x86" . chr( 0xa7 + $t );
 585                                         }
 586                                         continue;
 587                                 }
 588                         }
 589                         $out .= $c;
 590                 }
 591                 return $out;
 592         }
 593
 594         /**
 595          * Sorts combining characters into canonical order. This is the
 596          * final step in creating decomposed normal forms D and KD.
 597          * @private
 598          * @param string $string a valid, decomposed UTF-8 string. Input is not validated.
 599          * @return string a UTF-8 string with combining characters sorted in canonical order
 600          */
 601         function fastCombiningSort( $string ) {
 602                 UtfNormal::loadData();
 603                 global $utfCombiningClass;
 604                 $len = strlen( $string );
 605                 $out = '';
 606                 $combiners = array();
 607                 $lastClass = -1;
 608                 for( $i = 0; $i < $len; $i++ ) {
 609                         $c = $string{$i};
 610                         $n = ord( $c );
 611                         if( $n >= 0x80 ) {
 612                                 if( $n >= 0xf0 ) {
 613                                         $c = substr( $string, $i, 4 );
 614                                         $i += 3;
 615                                 } elseif( $n >= 0xe0 ) {
 616                                         $c = substr( $string, $i, 3 );
 617                                         $i += 2;
 618                                 } elseif( $n >= 0xc0 ) {
 619                                         $c = substr( $string, $i, 2 );
 620                                         $i++;
 621                                 }
 622                                 if( isset( $utfCombiningClass[$c] ) ) {
 623                                         $lastClass = $utfCombiningClass[$c];
 624                                         @$combiners[$lastClass] .= $c;
 625                                         continue;
 626                                 }
 627                         }
 628                         if( $lastClass ) {
 629                                 ksort( $combiners );
 630                                 $out .= implode( '', $combiners );
 631                                 $combiners = array();
 632                         }
 633                         $out .= $c;
 634                         $lastClass = 0;
 635                 }
 636                 if( $lastClass ) {
 637                         ksort( $combiners );
 638                         $out .= implode( '', $combiners );
 639                 }
 640                 return $out;
 641         }
 642
 643         /**
 644          * Produces canonically composed sequences, i.e. normal form C or KC.
 645          *
 646          * @private
 647          * @param string $string a valid UTF-8 string in sorted normal form D or KD. Input is not validated.
 648          * @return string a UTF-8 string with canonical precomposed characters used where possible
 649          */
 650         function fastCompose( $string ) {
 651                 UtfNormal::loadData();
 652                 global $utfCanonicalComp, $utfCombiningClass;
 653                 $len = strlen( $string );
 654                 $out = '';
 655                 $lastClass = -1;
 656                 $lastHangul = 0;
 657                 $startChar = '';
 658                 $combining = '';
 659                 $x1 = ord(substr(UTF8_HANGUL_VBASE,0,1));
 660                 $x2 = ord(substr(UTF8_HANGUL_TEND,0,1));
 661                 for( $i = 0; $i < $len; $i++ ) {
 662                         $c = $string{$i};
 663                         $n = ord( $c );
 664                         if( $n < 0x80 ) {
 665                                 # No combining characters here...
 666                                 $out .= $startChar;
 667                                 $out .= $combining;
 668                                 $startChar = $c;
 669                                 $combining = '';
 670                                 $lastClass = 0;
 671                                 continue;
 672                         } elseif( $n >= 0xf0 ) {
 673                                 $c = substr( $string, $i, 4 );
 674                                 $i += 3;
 675                         } elseif( $n >= 0xe0 ) {
 676                                 $c = substr( $string, $i, 3 );
 677                                 $i += 2;
 678                         } elseif( $n >= 0xc0 ) {
 679                                 $c = substr( $string, $i, 2 );
 680                                 $i++;
 681                         }
 682                         $pair = $startChar . $c;
 683                         if( $n > 0x80 ) {
 684                                 if( isset( $utfCombiningClass[$c] ) ) {
 685                                         # A combining char; see what we can do with it
 686                                         $class = $utfCombiningClass[$c];
 687                                         if( !empty( $startChar ) &&
 688                                                 $lastClass < $class &&
 689                                                 $class > 0 &&
 690                                                 isset( $utfCanonicalComp[$pair] ) ) {
 691                                                 $startChar = $utfCanonicalComp[$pair];
 692                                                 $class = 0;
 693                                         } else {
 694                                                 $combining .= $c;
 695                                         }
 696                                         $lastClass = $class;
 697                                         $lastHangul = 0;
 698                                         continue;
 699                                 }
 700                         }
 701                         # New start char
 702                         if( $lastClass == 0 ) {
 703                                 if( isset( $utfCanonicalComp[$pair] ) ) {
 704                                         $startChar = $utfCanonicalComp[$pair];
 705                                         $lastHangul = 0;
 706                                         continue;
 707                                 }
 708                                 if( $n >= $x1 && $n <= $x2 ) {
 709                                         # WARNING: Hangul code is painfully slow.
 710                                         # I apologize for this ugly, ugly code; however
 711                                         # performance is even more teh suck if we call
 712                                         # out to nice clean functions. Lookup tables are
 713                                         # marginally faster, but require a lot of space.
 714                                         #
 715                                         if( $c >= UTF8_HANGUL_VBASE &&
 716                                                 $c <= UTF8_HANGUL_VEND &&
 717                                                 $startChar >= UTF8_HANGUL_LBASE &&
 718                                                 $startChar <= UTF8_HANGUL_LEND ) {
 719                                                 #
 720                                                 #$lIndex = utf8ToCodepoint( $startChar ) - UNICODE_HANGUL_LBASE;
 721                                                 #$vIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_VBASE;
 722                                                 $lIndex = ord( $startChar{2} ) - 0x80;
 723                                                 $vIndex = ord( $c{2}         ) - 0xa1;
 724
 725                                                 $hangulPoint = UNICODE_HANGUL_FIRST +
 726                                                         UNICODE_HANGUL_TCOUNT *
 727                                                         (UNICODE_HANGUL_VCOUNT * $lIndex + $vIndex);
 728
 729                                                 # Hardcode the limited-range UTF-8 conversion:
 730                                                 $startChar = chr( $hangulPoint >> 12 & 0x0f | 0xe0 ) .
 731                                                                          chr( $hangulPoint >>  6 & 0x3f | 0x80 ) .
 732                                                                          chr( $hangulPoint       & 0x3f | 0x80 );
 733                                                 $lastHangul = 0;
 734                                                 continue;
 735                                         } elseif( $c >= UTF8_HANGUL_TBASE &&
 736                                                           $c <= UTF8_HANGUL_TEND &&
 737                                                           $startChar >= UTF8_HANGUL_FIRST &&
 738                                                           $startChar <= UTF8_HANGUL_LAST &&
 739                                                           !$lastHangul ) {
 740                                                 # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
 741                                                 $tIndex = ord( $c{2} ) - 0xa7;
 742                                                 if( $tIndex < 0 ) $tIndex = ord( $c{2} ) - 0x80 + (0x11c0 - 0x11a7);
 743
 744                                                 # Increment the code point by $tIndex, without
 745                                                 # the function overhead of decoding and recoding UTF-8
 746                                                 #
 747                                                 $tail = ord( $startChar{2} ) + $tIndex;
 748                                                 if( $tail > 0xbf ) {
 749                                                         $tail -= 0x40;
 750                                                         $mid = ord( $startChar{1} ) + 1;
 751                                                         if( $mid > 0xbf ) {
 752                                                                 $startChar{0} = chr( ord( $startChar{0} ) + 1 );
 753                                                                 $mid -= 0x40;
 754                                                         }
 755                                                         $startChar{1} = chr( $mid );
 756                                                 }
 757                                                 $startChar{2} = chr( $tail );
 758
 759                                                 # If there's another jamo char after this, *don't* try to merge it.
 760                                                 $lastHangul = 1;
 761                                                 continue;
 762                                         }
 763                                 }
 764                         }
 765                         $out .= $startChar;
 766                         $out .= $combining;
 767                         $startChar = $c;
 768                         $combining = '';
 769                         $lastClass = 0;
 770                         $lastHangul = 0;
 771                 }
 772                 $out .= $startChar . $combining;
 773                 return $out;
 774         }
 775
 776         /**
 777          * This is just used for the benchmark, comparing how long it takes to
 778          * interate through a string without really doing anything of substance.
 779          * @param string $string
 780          * @return string
 781          */
 782         function placebo( $string ) {
 783                 $len = strlen( $string );
 784                 $out = '';
 785                 for( $i = 0; $i < $len; $i++ ) {
 786                         $out .= $string{$i};
 787                 }
 788                 return $out;
 789         }
 790 }
 791
 792 ?>