includes/Collation.php

   1 <?php
   2 /**
   3  * Database row sorting.
   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  * @file
  21  */
  22
  23 abstract class Collation {
  24         private static $instance;
  25
  26         /**
  27          * @return Collation
  28          */
  29         static function singleton() {
  30                 if ( !self::$instance ) {
  31                         global $wgCategoryCollation;
  32                         self::$instance = self::factory( $wgCategoryCollation );
  33                 }
  34                 return self::$instance;
  35         }
  36
  37         /**
  38          * @throws MWException
  39          * @param string $collationName
  40          * @return Collation
  41          */
  42         static function factory( $collationName ) {
  43                 switch ( $collationName ) {
  44                         case 'uppercase':
  45                                 return new UppercaseCollation;
  46                         case 'identity':
  47                                 return new IdentityCollation;
  48                         case 'uca-default':
  49                                 return new IcuCollation( 'root' );
  50                         case 'xx-uca-ckb':
  51                                 return new CollationCkb;
  52                         case 'xx-uca-et':
  53                                 return new CollationEt;
  54                         default:
  55                                 $match = array();
  56                                 if ( preg_match( '/^uca-([a-z@=-]+)$/', $collationName, $match ) ) {
  57                                         return new IcuCollation( $match[1] );
  58                                 }
  59
  60                                 # Provide a mechanism for extensions to hook in.
  61                                 $collationObject = null;
  62                                 Hooks::run( 'Collation::factory', array( $collationName, &$collationObject ) );
  63
  64                                 if ( $collationObject instanceof Collation ) {
  65                                         return $collationObject;
  66                                 }
  67
  68                                 // If all else fails...
  69                                 throw new MWException( __METHOD__ . ": unknown collation type \"$collationName\"" );
  70                 }
  71         }
  72
  73         /**
  74          * Given a string, convert it to a (hopefully short) key that can be used
  75          * for efficient sorting.  A binary sort according to the sortkeys
  76          * corresponds to a logical sort of the corresponding strings.  Current
  77          * code expects that a line feed character should sort before all others, but
  78          * has no other particular expectations (and that one can be changed if
  79          * necessary).
  80          *
  81          * @param string $string UTF-8 string
  82          * @return string Binary sortkey
  83          */
  84         abstract function getSortKey( $string );
  85
  86         /**
  87          * Given a string, return the logical "first letter" to be used for
  88          * grouping on category pages and so on.  This has to be coordinated
  89          * carefully with convertToSortkey(), or else the sorted list might jump
  90          * back and forth between the same "initial letters" or other pathological
  91          * behavior.  For instance, if you just return the first character, but "a"
  92          * sorts the same as "A" based on getSortKey(), then you might get a
  93          * list like
  94          *
  95          * == A ==
  96          * * [[Aardvark]]
  97          *
  98          * == a ==
  99          * * [[antelope]]
 100          *
 101          * == A ==
 102          * * [[Ape]]
 103          *
 104          * etc., assuming for the sake of argument that $wgCapitalLinks is false.
 105          *
 106          * @param string $string UTF-8 string
 107          * @return string UTF-8 string corresponding to the first letter of input
 108          */
 109         abstract function getFirstLetter( $string );
 110 }
 111
 112 class UppercaseCollation extends Collation {
 113         private $lang;
 114
 115         function __construct() {
 116                 // Get a language object so that we can use the generic UTF-8 uppercase
 117                 // function there
 118                 $this->lang = Language::factory( 'en' );
 119         }
 120
 121         function getSortKey( $string ) {
 122                 return $this->lang->uc( $string );
 123         }
 124
 125         function getFirstLetter( $string ) {
 126                 if ( $string[0] == "\0" ) {
 127                         $string = substr( $string, 1 );
 128                 }
 129                 return $this->lang->ucfirst( $this->lang->firstChar( $string ) );
 130         }
 131 }
 132
 133 /**
 134  * Collation class that's essentially a no-op.
 135  *
 136  * Does sorting based on binary value of the string.
 137  * Like how things were pre 1.17.
 138  */
 139 class IdentityCollation extends Collation {
 140
 141         function getSortKey( $string ) {
 142                 return $string;
 143         }
 144
 145         function getFirstLetter( $string ) {
 146                 global $wgContLang;
 147                 // Copied from UppercaseCollation.
 148                 // I'm kind of unclear on when this could happen...
 149                 if ( $string[0] == "\0" ) {
 150                         $string = substr( $string, 1 );
 151                 }
 152                 return $wgContLang->firstChar( $string );
 153         }
 154 }
 155
 156 class IcuCollation extends Collation {
 157         const FIRST_LETTER_VERSION = 2;
 158
 159         /** @var Collator */
 160         private $primaryCollator;
 161
 162         /** @var Collator */
 163         private $mainCollator;
 164
 165         /** @var string */
 166         private $locale;
 167
 168         /** @var Language */
 169         protected $digitTransformLanguage;
 170
 171         /** @var array */
 172         private $firstLetterData;
 173
 174         /**
 175          * Unified CJK blocks.
 176          *
 177          * The same definition of a CJK block must be used for both Collation and
 178          * generateCollationData.php. These blocks are omitted from the first
 179          * letter data, as an optimisation measure and because the default UCA table
 180          * is pretty useless for sorting Chinese text anyway. Japanese and Korean
 181          * blocks are not included here, because they are smaller and more useful.
 182          */
 183         private static $cjkBlocks = array(
 184                 array( 0x2E80, 0x2EFF ), // CJK Radicals Supplement
 185                 array( 0x2F00, 0x2FDF ), // Kangxi Radicals
 186                 array( 0x2FF0, 0x2FFF ), // Ideographic Description Characters
 187                 array( 0x3000, 0x303F ), // CJK Symbols and Punctuation
 188                 array( 0x31C0, 0x31EF ), // CJK Strokes
 189                 array( 0x3200, 0x32FF ), // Enclosed CJK Letters and Months
 190                 array( 0x3300, 0x33FF ), // CJK Compatibility
 191                 array( 0x3400, 0x4DBF ), // CJK Unified Ideographs Extension A
 192                 array( 0x4E00, 0x9FFF ), // CJK Unified Ideographs
 193                 array( 0xF900, 0xFAFF ), // CJK Compatibility Ideographs
 194                 array( 0xFE30, 0xFE4F ), // CJK Compatibility Forms
 195                 array( 0x20000, 0x2A6DF ), // CJK Unified Ideographs Extension B
 196                 array( 0x2A700, 0x2B73F ), // CJK Unified Ideographs Extension C
 197                 array( 0x2B740, 0x2B81F ), // CJK Unified Ideographs Extension D
 198                 array( 0x2F800, 0x2FA1F ), // CJK Compatibility Ideographs Supplement
 199         );
 200
 201         /**
 202          * Additional characters (or character groups) to be considered separate
 203          * letters for given languages, or to be removed from the list of such
 204          * letters (denoted by keys starting with '-').
 205          *
 206          * These are additions to (or subtractions from) the data stored in the
 207          * first-letters-root.ser file (which among others includes full basic latin,
 208          * cyrillic and greek alphabets).
 209          *
 210          * "Separate letter" is a letter that would have a separate heading/section
 211          * for it in a dictionary or a phone book in this language. This data isn't
 212          * used for sorting (the ICU library handles that), only for deciding which
 213          * characters (or character groups) to use as headings.
 214          *
 215          * Initially generated based on the primary level of Unicode collation
 216          * tailorings available at http://developer.mimer.com/charts/tailorings.htm ,
 217          * later modified.
 218          *
 219          * Empty arrays are intended; this signifies that the data for the language is
 220          * available and that there are, in fact, no additional letters to consider.
 221          */
 222         private static $tailoringFirstLetters = array(
 223                 // Verified by native speakers
 224                 'be' => array( "Ё" ),
 225                 'be-tarask' => array( "Ё" ),
 226                 'cy' => array( "Ch", "Dd", "Ff", "Ng", "Ll", "Ph", "Rh", "Th" ),
 227                 'en' => array(),
 228                 'fa' => array( "آ", "ء", "ه" ),
 229                 'fi' => array( "Å", "Ä", "Ö" ),
 230                 'fr' => array(),
 231                 'hu' => array( "Cs", "Dz", "Dzs", "Gy", "Ly", "Ny", "Ö", "Sz", "Ty", "Ü", "Zs" ),
 232                 'is' => array( "Á", "Ð", "É", "Í", "Ó", "Ú", "Ý", "Þ", "Æ", "Ö", "Å" ),
 233                 'it' => array(),
 234                 'lv' => array( "Č", "Ģ", "Ķ", "Ļ", "Ņ", "Š", "Ž" ),
 235                 'pl' => array( "Ą", "Ć", "Ę", "Ł", "Ń", "Ó", "Ś", "Ź", "Ż" ),
 236                 'pt' => array(),
 237                 'ru' => array(),
 238                 'sv' => array( "Å", "Ä", "Ö" ),
 239                 'sv@collation=standard' => array( "Å", "Ä", "Ö" ),
 240                 'uk' => array( "Ґ", "Ь" ),
 241                 'vi' => array( "Ă", "Â", "Đ", "Ê", "Ô", "Ơ", "Ư" ),
 242                 // Not verified, but likely correct
 243                 'af' => array(),
 244                 'ast' => array( "Ch", "Ll", "Ñ" ),
 245                 'az' => array( "Ç", "Ə", "Ğ", "İ", "Ö", "Ş", "Ü" ),
 246                 'bg' => array(),
 247                 'br' => array( "Ch", "C'h" ),
 248                 'bs' => array( "Č", "Ć", "Dž", "Đ", "Lj", "Nj", "Š", "Ž" ),
 249                 'ca' => array(),
 250                 'co' => array(),
 251                 'cs' => array( "Č", "Ch", "Ř", "Š", "Ž" ),
 252                 'da' => array( "Æ", "Ø", "Å" ),
 253                 'de' => array(),
 254                 'dsb' => array( "Č", "Ć", "Dź", "Ě", "Ch", "Ł", "Ń", "Ŕ", "Š", "Ś", "Ž", "Ź" ),
 255                 'el' => array(),
 256                 'eo' => array( "Ĉ", "Ĝ", "Ĥ", "Ĵ", "Ŝ", "Ŭ" ),
 257                 'es' => array( "Ñ" ),
 258                 'et' => array( "Š", "Ž", "Õ", "Ä", "Ö", "Ü", "W" ), // added W for CollationEt (xx-uca-et)
 259                 'eu' => array( "Ñ" ),
 260                 'fo' => array( "Á", "Ð", "Í", "Ó", "Ú", "Ý", "Æ", "Ø", "Å" ),
 261                 'fur' => array( "À", "Á", "Â", "È", "Ì", "Ò", "Ù" ),
 262                 'fy' => array(),
 263                 'ga' => array(),
 264                 'gd' => array(),
 265                 'gl' => array( "Ch", "Ll", "Ñ" ),
 266                 'hr' => array( "Č", "Ć", "Dž", "Đ", "Lj", "Nj", "Š", "Ž" ),
 267                 'hsb' => array( "Č", "Dź", "Ě", "Ch", "Ł", "Ń", "Ř", "Š", "Ć", "Ž" ),
 268                 'kk' => array( "Ү", "І" ),
 269                 'kl' => array( "Æ", "Ø", "Å" ),
 270                 'ku' => array( "Ç", "Ê", "Î", "Ş", "Û" ),
 271                 'ky' => array( "Ё" ),
 272                 'la' => array(),
 273                 'lb' => array(),
 274                 'lt' => array( "Č", "Š", "Ž" ),
 275                 'mk' => array(),
 276                 'mo' => array( "Ă", "Â", "Î", "Ş", "Ţ" ),
 277                 'mt' => array( "Ċ", "Ġ", "Għ", "Ħ", "Ż" ),
 278                 'nl' => array(),
 279                 'no' => array( "Æ", "Ø", "Å" ),
 280                 'oc' => array(),
 281                 'rm' => array(),
 282                 'ro' => array( "Ă", "Â", "Î", "Ş", "Ţ" ),
 283                 'rup' => array( "Ă", "Â", "Î", "Ľ", "Ń", "Ş", "Ţ" ),
 284                 'sco' => array(),
 285                 'sk' => array( "Ä", "Č", "Ch", "Ô", "Š", "Ž" ),
 286                 'sl' => array( "Č", "Š", "Ž" ),
 287                 'smn' => array( "Á", "Č", "Đ", "Ŋ", "Š", "Ŧ", "Ž", "Æ", "Ø", "Å", "Ä", "Ö" ),
 288                 'sq' => array( "Ç", "Dh", "Ë", "Gj", "Ll", "Nj", "Rr", "Sh", "Th", "Xh", "Zh" ),
 289                 'sr' => array(),
 290                 'tk' => array( "Ç", "Ä", "Ž", "Ň", "Ö", "Ş", "Ü", "Ý" ),
 291                 'tl' => array( "Ñ", "Ng" ),
 292                 'tr' => array( "Ç", "Ğ", "İ", "Ö", "Ş", "Ü" ),
 293                 'tt' => array( "Ә", "Ө", "Ү", "Җ", "Ң", "Һ" ),
 294                 'uz' => array( "Ch", "G'", "Ng", "O'", "Sh" ),
 295         );
 296
 297         const RECORD_LENGTH = 14;
 298
 299         function __construct( $locale ) {
 300                 if ( !extension_loaded( 'intl' ) ) {
 301                         throw new MWException( 'An ICU collation was requested, ' .
 302                                 'but the intl extension is not available.' );
 303                 }
 304
 305                 $this->locale = $locale;
 306                 // Drop everything after the '@' in locale's name
 307                 $localeParts = explode( '@', $locale );
 308                 $this->digitTransformLanguage = Language::factory( $locale === 'root' ? 'en' : $localeParts[0] );
 309
 310                 $this->mainCollator = Collator::create( $locale );
 311                 if ( !$this->mainCollator ) {
 312                         throw new MWException( "Invalid ICU locale specified for collation: $locale" );
 313                 }
 314
 315                 $this->primaryCollator = Collator::create( $locale );
 316                 $this->primaryCollator->setStrength( Collator::PRIMARY );
 317         }
 318
 319         function getSortKey( $string ) {
 320                 // intl extension produces non null-terminated
 321                 // strings. Appending '' fixes it so that it doesn't generate
 322                 // a warning on each access in debug php.
 323                 MediaWiki\suppressWarnings();
 324                 $key = $this->mainCollator->getSortKey( $string ) . '';
 325                 MediaWiki\restoreWarnings();
 326                 return $key;
 327         }
 328
 329         function getPrimarySortKey( $string ) {
 330                 MediaWiki\suppressWarnings();
 331                 $key = $this->primaryCollator->getSortKey( $string ) . '';
 332                 MediaWiki\restoreWarnings();
 333                 return $key;
 334         }
 335
 336         function getFirstLetter( $string ) {
 337                 $string = strval( $string );
 338                 if ( $string === '' ) {
 339                         return '';
 340                 }
 341
 342                 // Check for CJK
 343                 $firstChar = mb_substr( $string, 0, 1, 'UTF-8' );
 344                 if ( ord( $firstChar ) > 0x7f && self::isCjk( UtfNormal\Utils::utf8ToCodepoint( $firstChar ) ) ) {
 345                         return $firstChar;
 346                 }
 347
 348                 $sortKey = $this->getPrimarySortKey( $string );
 349
 350                 // Do a binary search to find the correct letter to sort under
 351                 $min = ArrayUtils::findLowerBound(
 352                         array( $this, 'getSortKeyByLetterIndex' ),
 353                         $this->getFirstLetterCount(),
 354                         'strcmp',
 355                         $sortKey );
 356
 357                 if ( $min === false ) {
 358                         // Before the first letter
 359                         return '';
 360                 }
 361                 return $this->getLetterByIndex( $min );
 362         }
 363
 364         function getFirstLetterData() {
 365                 if ( $this->firstLetterData !== null ) {
 366                         return $this->firstLetterData;
 367                 }
 368
 369                 $cache = wfGetCache( CACHE_ANYTHING );
 370                 $cacheKey = wfMemcKey(
 371                         'first-letters',
 372                         $this->locale,
 373                         $this->digitTransformLanguage->getCode(),
 374                         self::getICUVersion()
 375                 );
 376                 $cacheEntry = $cache->get( $cacheKey );
 377
 378                 if ( $cacheEntry && isset( $cacheEntry['version'] )
 379                         && $cacheEntry['version'] == self::FIRST_LETTER_VERSION
 380                 ) {
 381                         $this->firstLetterData = $cacheEntry;
 382                         return $this->firstLetterData;
 383                 }
 384
 385                 // Generate data from serialized data file
 386
 387                 if ( isset( self::$tailoringFirstLetters[$this->locale] ) ) {
 388                         $letters = wfGetPrecompiledData( "first-letters-root.ser" );
 389                         // Append additional characters
 390                         $letters = array_merge( $letters, self::$tailoringFirstLetters[$this->locale] );
 391                         // Remove unnecessary ones, if any
 392                         if ( isset( self::$tailoringFirstLetters['-' . $this->locale] ) ) {
 393                                 $letters = array_diff( $letters, self::$tailoringFirstLetters['-' . $this->locale] );
 394                         }
 395                         // Apply digit transforms
 396                         $digits = array( '0', '1', '2', '3', '4', '5', '6', '7', '8', '9' );
 397                         $letters = array_diff( $letters, $digits );
 398                         foreach ( $digits as $digit ) {
 399                                 $letters[] = $this->digitTransformLanguage->formatNum( $digit, true );
 400                         }
 401                 } else {
 402                         $letters = wfGetPrecompiledData( "first-letters-{$this->locale}.ser" );
 403                         if ( $letters === false ) {
 404                                 throw new MWException( "MediaWiki does not support ICU locale " .
 405                                         "\"{$this->locale}\"" );
 406                         }
 407                 }
 408
 409                 /* Sort the letters.
 410                  *
 411                  * It's impossible to have the precompiled data file properly sorted,
 412                  * because the sort order changes depending on ICU version. If the
 413                  * array is not properly sorted, the binary search will return random
 414                  * results.
 415                  *
 416                  * We also take this opportunity to remove primary collisions.
 417                  */
 418                 $letterMap = array();
 419                 foreach ( $letters as $letter ) {
 420                         $key = $this->getPrimarySortKey( $letter );
 421                         if ( isset( $letterMap[$key] ) ) {
 422                                 // Primary collision
 423                                 // Keep whichever one sorts first in the main collator
 424                                 if ( $this->mainCollator->compare( $letter, $letterMap[$key] ) < 0 ) {
 425                                         $letterMap[$key] = $letter;
 426                                 }
 427                         } else {
 428                                 $letterMap[$key] = $letter;
 429                         }
 430                 }
 431                 ksort( $letterMap, SORT_STRING );
 432
 433                 /* Remove duplicate prefixes. Basically if something has a sortkey
 434                  * which is a prefix of some other sortkey, then it is an
 435                  * expansion and probably should not be considered a section
 436                  * header.
 437                  *
 438                  * For example 'þ' is sometimes sorted as if it is the letters
 439                  * 'th'. Other times it is its own primary element. Another
 440                  * example is '₨'. Sometimes its a currency symbol. Sometimes it
 441                  * is an 'R' followed by an 's'.
 442                  *
 443                  * Additionally an expanded element should always sort directly
 444                  * after its first element due to they way sortkeys work.
 445                  *
 446                  * UCA sortkey elements are of variable length but no collation
 447                  * element should be a prefix of some other element, so I think
 448                  * this is safe. See:
 449                  * - https://ssl.icu-project.org/repos/icu/icuhtml/trunk/design/collation/ICU_collation_design.htm
 450                  * - http://site.icu-project.org/design/collation/uca-weight-allocation
 451                  *
 452                  * Additionally, there is something called primary compression to
 453                  * worry about. Basically, if you have two primary elements that
 454                  * are more than one byte and both start with the same byte then
 455                  * the first byte is dropped on the second primary. Additionally
 456                  * either \x03 or \xFF may be added to mean that the next primary
 457                  * does not start with the first byte of the first primary.
 458                  *
 459                  * This shouldn't matter much, as the first primary is not
 460                  * changed, and that is what we are comparing against.
 461                  *
 462                  * tl;dr: This makes some assumptions about how icu implements
 463                  * collations. It seems incredibly unlikely these assumptions
 464                  * will change, but nonetheless they are assumptions.
 465                  */
 466
 467                 $prev = false;
 468                 $duplicatePrefixes = array();
 469                 foreach ( $letterMap as $key => $value ) {
 470                         // Remove terminator byte. Otherwise the prefix
 471                         // comparison will get hung up on that.
 472                         $trimmedKey = rtrim( $key, "\0" );
 473                         if ( $prev === false || $prev === '' ) {
 474                                 $prev = $trimmedKey;
 475                                 // We don't yet have a collation element
 476                                 // to compare against, so continue.
 477                                 continue;
 478                         }
 479
 480                         // Due to the fact the array is sorted, we only have
 481                         // to compare with the element directly previous
 482                         // to the current element (skipping expansions).
 483                         // An element "X" will always sort directly
 484                         // before "XZ" (Unless we have "XY", but we
 485                         // do not update $prev in that case).
 486                         if ( substr( $trimmedKey, 0, strlen( $prev ) ) === $prev ) {
 487                                 $duplicatePrefixes[] = $key;
 488                                 // If this is an expansion, we don't want to
 489                                 // compare the next element to this element,
 490                                 // but to what is currently $prev
 491                                 continue;
 492                         }
 493                         $prev = $trimmedKey;
 494                 }
 495                 foreach ( $duplicatePrefixes as $badKey ) {
 496                         wfDebug( "Removing '{$letterMap[$badKey]}' from first letters.\n" );
 497                         unset( $letterMap[$badKey] );
 498                         // This code assumes that unsetting does not change sort order.
 499                 }
 500                 $data = array(
 501                         'chars' => array_values( $letterMap ),
 502                         'keys' => array_keys( $letterMap ),
 503                         'version' => self::FIRST_LETTER_VERSION,
 504                 );
 505
 506                 // Reduce memory usage before caching
 507                 unset( $letterMap );
 508
 509                 // Save to cache
 510                 $this->firstLetterData = $data;
 511                 $cache->set( $cacheKey, $data, $cache::TTL_WEEK );
 512                 return $data;
 513         }
 514
 515         function getLetterByIndex( $index ) {
 516                 if ( $this->firstLetterData === null ) {
 517                         $this->getFirstLetterData();
 518                 }
 519                 return $this->firstLetterData['chars'][$index];
 520         }
 521
 522         function getSortKeyByLetterIndex( $index ) {
 523                 if ( $this->firstLetterData === null ) {
 524                         $this->getFirstLetterData();
 525                 }
 526                 return $this->firstLetterData['keys'][$index];
 527         }
 528
 529         function getFirstLetterCount() {
 530                 if ( $this->firstLetterData === null ) {
 531                         $this->getFirstLetterData();
 532                 }
 533                 return count( $this->firstLetterData['chars'] );
 534         }
 535
 536         static function isCjk( $codepoint ) {
 537                 foreach ( self::$cjkBlocks as $block ) {
 538                         if ( $codepoint >= $block[0] && $codepoint <= $block[1] ) {
 539                                 return true;
 540                         }
 541                 }
 542                 return false;
 543         }
 544
 545         /**
 546          * Return the version of ICU library used by PHP's intl extension,
 547          * or false when the extension is not installed of the version
 548          * can't be determined.
 549          *
 550          * The constant INTL_ICU_VERSION this function refers to isn't really
 551          * documented. It is available since PHP 5.3.7 (see PHP bug 54561).
 552          * This function will return false on older PHPs.
 553          *
 554          * @since 1.21
 555          * @return string|bool
 556          */
 557         static function getICUVersion() {
 558                 return defined( 'INTL_ICU_VERSION' ) ? INTL_ICU_VERSION : false;
 559         }
 560
 561         /**
 562          * Return the version of Unicode appropriate for the version of ICU library
 563          * currently in use, or false when it can't be determined.
 564          *
 565          * @since 1.21
 566          * @return string|bool
 567          */
 568         static function getUnicodeVersionForICU() {
 569                 $icuVersion = IcuCollation::getICUVersion();
 570                 if ( !$icuVersion ) {
 571                         return false;
 572                 }
 573
 574                 $versionPrefix = substr( $icuVersion, 0, 3 );
 575                 // Source: http://site.icu-project.org/download
 576                 $map = array(
 577                         '50.' => '6.2',
 578                         '49.' => '6.1',
 579                         '4.8' => '6.0',
 580                         '4.6' => '6.0',
 581                         '4.4' => '5.2',
 582                         '4.2' => '5.1',
 583                         '4.0' => '5.1',
 584                         '3.8' => '5.0',
 585                         '3.6' => '5.0',
 586                         '3.4' => '4.1',
 587                 );
 588
 589                 if ( isset( $map[$versionPrefix] ) ) {
 590                         return $map[$versionPrefix];
 591                 } else {
 592                         return false;
 593                 }
 594         }
 595 }
 596
 597 /**
 598  * Workaround for the lack of support of Sorani Kurdish / Central Kurdish language ('ckb') in ICU.
 599  *
 600  * Uses the same collation rules as Persian / Farsi ('fa'), but different characters for digits.
 601  */
 602 class CollationCkb extends IcuCollation {
 603         function __construct() {
 604                 // This will set $locale and collators, which affect the actual sorting order
 605                 parent::__construct( 'fa' );
 606                 // Override the 'fa' language set by parent constructor, which affects #getFirstLetterData()
 607                 $this->digitTransformLanguage = Language::factory( 'ckb' );
 608         }
 609 }
 610
 611 /**
 612  * Workaround for incorrect collation of Estonian language ('et') in ICU (bug 54168).
 613  *
 614  * 'W' and 'V' should not be considered the same letter for the purposes of collation in modern
 615  * Estonian. We work around this by replacing 'W' and 'w' with 'ᴡ' U+1D21 'LATIN LETTER SMALL
 616  * CAPITAL W' for sortkey generation, which is collated like 'W' and is not tailored to have the
 617  * same primary weight as 'V' in Estonian.
 618  */
 619 class CollationEt extends IcuCollation {
 620         function __construct() {
 621                 parent::__construct( 'et' );
 622         }
 623
 624         private static function mangle( $string ) {
 625                 return str_replace(
 626                         array( 'w', 'W' ),
 627                         'ᴡ', // U+1D21 'LATIN LETTER SMALL CAPITAL W'
 628                         $string
 629                 );
 630         }
 631
 632         private static function unmangle( $string ) {
 633                 // Casing data is lost…
 634                 return str_replace(
 635                         'ᴡ', // U+1D21 'LATIN LETTER SMALL CAPITAL W'
 636                         'W',
 637                         $string
 638                 );
 639         }
 640
 641         function getSortKey( $string ) {
 642                 return parent::getSortKey( self::mangle( $string ) );
 643         }
 644
 645         function getFirstLetter( $string ) {
 646                 return self::unmangle( parent::getFirstLetter( self::mangle( $string ) ) );
 647         }
 648 }