diff options
author | Pierre Schmitz <pierre@archlinux.de> | 2008-12-15 18:02:47 +0100 |
---|---|---|
committer | Pierre Schmitz <pierre@archlinux.de> | 2008-12-15 18:02:47 +0100 |
commit | 396b28f3d881f5debd888ba9bb9b47c2d478a76f (patch) | |
tree | 10d6e1a721ee4ef69def34a57f02d7eb3fc9e31e /includes/Parser.php | |
parent | 0be4d3ccf6c4fe98a72704f9463ecdea2ee5e615 (diff) |
update to Mediawiki 1.13.3; some cleanups
Diffstat (limited to 'includes/Parser.php')
-rw-r--r-- | includes/Parser.php | 4913 |
1 files changed, 0 insertions, 4913 deletions
diff --git a/includes/Parser.php b/includes/Parser.php deleted file mode 100644 index 41eabe4f..00000000 --- a/includes/Parser.php +++ /dev/null @@ -1,4913 +0,0 @@ -<?php - -/** - * - * File for Parser and related classes - * - * @addtogroup Parser - */ - - -/** - * PHP Parser - Processes wiki markup (which uses a more user-friendly - * syntax, such as "[[link]]" for making links), and provides a one-way - * transformation of that wiki markup it into XHTML output / markup - * (which in turn the browser understands, and can display). - * - * <pre> - * There are five main entry points into the Parser class: - * parse() - * produces HTML output - * preSaveTransform(). - * produces altered wiki markup. - * preprocess() - * removes HTML comments and expands templates - * cleanSig() - * Cleans a signature before saving it to preferences - * extractSections() - * Extracts sections from an article for section editing - * - * Globals used: - * objects: $wgLang, $wgContLang - * - * NOT $wgArticle, $wgUser or $wgTitle. Keep them away! - * - * settings: - * $wgUseTex*, $wgUseDynamicDates*, $wgInterwikiMagic*, - * $wgNamespacesWithSubpages, $wgAllowExternalImages*, - * $wgLocaltimezone, $wgAllowSpecialInclusion*, - * $wgMaxArticleSize* - * - * * only within ParserOptions - * </pre> - * - * @addtogroup Parser - */ -class Parser -{ - /** - * Update this version number when the ParserOutput format - * changes in an incompatible way, so the parser cache - * can automatically discard old data. - */ - const VERSION = '1.6.4'; - - # Flags for Parser::setFunctionHook - # Also available as global constants from Defines.php - const SFH_NO_HASH = 1; - const SFH_OBJECT_ARGS = 2; - - # Constants needed for external link processing - # Everything except bracket, space, or control characters - const EXT_LINK_URL_CLASS = '[^][<>"\\x00-\\x20\\x7F]'; - const EXT_IMAGE_REGEX = '/^(http:\/\/|https:\/\/)([^][<>"\\x00-\\x20\\x7F]+) - \\/([A-Za-z0-9_.,~%\\-+&;#*?!=()@\\x80-\\xFF]+)\\.((?i)gif|png|jpg|jpeg)$/Sx'; - - // State constants for the definition list colon extraction - const COLON_STATE_TEXT = 0; - const COLON_STATE_TAG = 1; - const COLON_STATE_TAGSTART = 2; - const COLON_STATE_CLOSETAG = 3; - const COLON_STATE_TAGSLASH = 4; - const COLON_STATE_COMMENT = 5; - const COLON_STATE_COMMENTDASH = 6; - const COLON_STATE_COMMENTDASHDASH = 7; - - // Flags for preprocessToDom - const PTD_FOR_INCLUSION = 1; - - // Allowed values for $this->mOutputType - // Parameter to startExternalParse(). - const OT_HTML = 1; - const OT_WIKI = 2; - const OT_PREPROCESS = 3; - const OT_MSG = 3; - - /**#@+ - * @private - */ - # Persistent: - var $mTagHooks, $mTransparentTagHooks, $mFunctionHooks, $mFunctionSynonyms, $mVariables, - $mImageParams, $mImageParamsMagicArray, $mStripList, $mMarkerSuffix, $mMarkerIndex, - $mExtLinkBracketedRegex, $mPreprocessor, $mDefaultStripList, $mVarCache, $mConf; - - - # Cleared with clearState(): - var $mOutput, $mAutonumber, $mDTopen, $mStripState; - var $mIncludeCount, $mArgStack, $mLastSection, $mInPre; - var $mInterwikiLinkHolders, $mLinkHolders; - var $mIncludeSizes, $mPPNodeCount, $mDefaultSort; - var $mTplExpandCache; // empty-frame expansion cache - var $mTplRedirCache, $mTplDomCache, $mHeadings; - - # Temporary - # These are variables reset at least once per parse regardless of $clearState - var $mOptions, // ParserOptions object - $mTitle, // Title context, used for self-link rendering and similar things - $mOutputType, // Output type, one of the OT_xxx constants - $ot, // Shortcut alias, see setOutputType() - $mRevisionId, // ID to display in {{REVISIONID}} tags - $mRevisionTimestamp, // The timestamp of the specified revision ID - $mRevIdForTs; // The revision ID which was used to fetch the timestamp - - /**#@-*/ - - /** - * Constructor - * - * @public - */ - function __construct( $conf = array() ) { - $this->mConf = $conf; - $this->mTagHooks = array(); - $this->mTransparentTagHooks = array(); - $this->mFunctionHooks = array(); - $this->mFunctionSynonyms = array( 0 => array(), 1 => array() ); - $this->mDefaultStripList = $this->mStripList = array( 'nowiki', 'gallery' ); - $this->mMarkerSuffix = "-QINU\x7f"; - $this->mExtLinkBracketedRegex = '/\[(\b(' . wfUrlProtocols() . ')'. - '[^][<>"\\x00-\\x20\\x7F]+) *([^\]\\x0a\\x0d]*?)\]/S'; - $this->mVarCache = array(); - if ( isset( $conf['preprocessorClass'] ) ) { - $this->mPreprocessorClass = $conf['preprocessorClass']; - } else { - $this->mPreprocessorClass = 'Preprocessor_DOM'; - } - $this->mMarkerIndex = 0; - $this->mFirstCall = true; - } - - /** - * Do various kinds of initialisation on the first call of the parser - */ - function firstCallInit() { - if ( !$this->mFirstCall ) { - return; - } - $this->mFirstCall = false; - - wfProfileIn( __METHOD__ ); - global $wgAllowDisplayTitle, $wgAllowSlowParserFunctions; - - $this->setHook( 'pre', array( $this, 'renderPreTag' ) ); - - # Syntax for arguments (see self::setFunctionHook): - # "name for lookup in localized magic words array", - # function callback, - # optional SFH_NO_HASH to omit the hash from calls (e.g. {{int:...} - # instead of {{#int:...}}) - $this->setFunctionHook( 'int', array( 'CoreParserFunctions', 'intFunction' ), SFH_NO_HASH ); - $this->setFunctionHook( 'ns', array( 'CoreParserFunctions', 'ns' ), SFH_NO_HASH ); - $this->setFunctionHook( 'urlencode', array( 'CoreParserFunctions', 'urlencode' ), SFH_NO_HASH ); - $this->setFunctionHook( 'lcfirst', array( 'CoreParserFunctions', 'lcfirst' ), SFH_NO_HASH ); - $this->setFunctionHook( 'ucfirst', array( 'CoreParserFunctions', 'ucfirst' ), SFH_NO_HASH ); - $this->setFunctionHook( 'lc', array( 'CoreParserFunctions', 'lc' ), SFH_NO_HASH ); - $this->setFunctionHook( 'uc', array( 'CoreParserFunctions', 'uc' ), SFH_NO_HASH ); - $this->setFunctionHook( 'localurl', array( 'CoreParserFunctions', 'localurl' ), SFH_NO_HASH ); - $this->setFunctionHook( 'localurle', array( 'CoreParserFunctions', 'localurle' ), SFH_NO_HASH ); - $this->setFunctionHook( 'fullurl', array( 'CoreParserFunctions', 'fullurl' ), SFH_NO_HASH ); - $this->setFunctionHook( 'fullurle', array( 'CoreParserFunctions', 'fullurle' ), SFH_NO_HASH ); - $this->setFunctionHook( 'formatnum', array( 'CoreParserFunctions', 'formatnum' ), SFH_NO_HASH ); - $this->setFunctionHook( 'grammar', array( 'CoreParserFunctions', 'grammar' ), SFH_NO_HASH ); - $this->setFunctionHook( 'plural', array( 'CoreParserFunctions', 'plural' ), SFH_NO_HASH ); - $this->setFunctionHook( 'numberofpages', array( 'CoreParserFunctions', 'numberofpages' ), SFH_NO_HASH ); - $this->setFunctionHook( 'numberofusers', array( 'CoreParserFunctions', 'numberofusers' ), SFH_NO_HASH ); - $this->setFunctionHook( 'numberofarticles', array( 'CoreParserFunctions', 'numberofarticles' ), SFH_NO_HASH ); - $this->setFunctionHook( 'numberoffiles', array( 'CoreParserFunctions', 'numberoffiles' ), SFH_NO_HASH ); - $this->setFunctionHook( 'numberofadmins', array( 'CoreParserFunctions', 'numberofadmins' ), SFH_NO_HASH ); - $this->setFunctionHook( 'numberofedits', array( 'CoreParserFunctions', 'numberofedits' ), SFH_NO_HASH ); - $this->setFunctionHook( 'language', array( 'CoreParserFunctions', 'language' ), SFH_NO_HASH ); - $this->setFunctionHook( 'padleft', array( 'CoreParserFunctions', 'padleft' ), SFH_NO_HASH ); - $this->setFunctionHook( 'padright', array( 'CoreParserFunctions', 'padright' ), SFH_NO_HASH ); - $this->setFunctionHook( 'anchorencode', array( 'CoreParserFunctions', 'anchorencode' ), SFH_NO_HASH ); - $this->setFunctionHook( 'special', array( 'CoreParserFunctions', 'special' ) ); - $this->setFunctionHook( 'defaultsort', array( 'CoreParserFunctions', 'defaultsort' ), SFH_NO_HASH ); - $this->setFunctionHook( 'filepath', array( 'CoreParserFunctions', 'filepath' ), SFH_NO_HASH ); - $this->setFunctionHook( 'tag', array( 'CoreParserFunctions', 'tagObj' ), SFH_OBJECT_ARGS ); - - if ( $wgAllowDisplayTitle ) { - $this->setFunctionHook( 'displaytitle', array( 'CoreParserFunctions', 'displaytitle' ), SFH_NO_HASH ); - } - if ( $wgAllowSlowParserFunctions ) { - $this->setFunctionHook( 'pagesinnamespace', array( 'CoreParserFunctions', 'pagesinnamespace' ), SFH_NO_HASH ); - } - - $this->initialiseVariables(); - - wfRunHooks( 'ParserFirstCallInit', array( &$this ) ); - wfProfileOut( __METHOD__ ); - } - - /** - * Clear Parser state - * - * @private - */ - function clearState() { - wfProfileIn( __METHOD__ ); - if ( $this->mFirstCall ) { - $this->firstCallInit(); - } - $this->mOutput = new ParserOutput; - $this->mAutonumber = 0; - $this->mLastSection = ''; - $this->mDTopen = false; - $this->mIncludeCount = array(); - $this->mStripState = new StripState; - $this->mArgStack = false; - $this->mInPre = false; - $this->mInterwikiLinkHolders = array( - 'texts' => array(), - 'titles' => array() - ); - $this->mLinkHolders = array( - 'namespaces' => array(), - 'dbkeys' => array(), - 'queries' => array(), - 'texts' => array(), - 'titles' => array() - ); - $this->mRevisionTimestamp = $this->mRevisionId = null; - - /** - * Prefix for temporary replacement strings for the multipass parser. - * \x07 should never appear in input as it's disallowed in XML. - * Using it at the front also gives us a little extra robustness - * since it shouldn't match when butted up against identifier-like - * string constructs. - * - * Must not consist of all title characters, or else it will change - * the behaviour of <nowiki> in a link. - */ - #$this->mUniqPrefix = "\x07UNIQ" . Parser::getRandomString(); - # Changed to \x7f to allow XML double-parsing -- TS - $this->mUniqPrefix = "\x7fUNIQ" . Parser::getRandomString(); - - - # Clear these on every parse, bug 4549 - $this->mTplExpandCache = $this->mTplRedirCache = $this->mTplDomCache = array(); - - $this->mShowToc = true; - $this->mForceTocPosition = false; - $this->mIncludeSizes = array( - 'post-expand' => 0, - 'arg' => 0, - ); - $this->mPPNodeCount = 0; - $this->mDefaultSort = false; - $this->mHeadings = array(); - - # Fix cloning - if ( isset( $this->mPreprocessor ) && $this->mPreprocessor->parser !== $this ) { - $this->mPreprocessor = null; - } - - wfRunHooks( 'ParserClearState', array( &$this ) ); - wfProfileOut( __METHOD__ ); - } - - function setOutputType( $ot ) { - $this->mOutputType = $ot; - // Shortcut alias - $this->ot = array( - 'html' => $ot == self::OT_HTML, - 'wiki' => $ot == self::OT_WIKI, - 'pre' => $ot == self::OT_PREPROCESS, - ); - } - - /** - * Set the context title - */ - function setTitle( $t ) { - if ( !$t || $t instanceof FakeTitle ) { - $t = Title::newFromText( 'NO TITLE' ); - } - if ( strval( $t->getFragment() ) !== '' ) { - # Strip the fragment to avoid various odd effects - $this->mTitle = clone $t; - $this->mTitle->setFragment( '' ); - } else { - $this->mTitle = $t; - } - } - - /** - * Accessor for mUniqPrefix. - * - * @public - */ - function uniqPrefix() { - if( !isset( $this->mUniqPrefix ) ) { - // @fixme this is probably *horribly wrong* - // LanguageConverter seems to want $wgParser's uniqPrefix, however - // if this is called for a parser cache hit, the parser may not - // have ever been initialized in the first place. - // Not really sure what the heck is supposed to be going on here. - return ''; - //throw new MWException( "Accessing uninitialized mUniqPrefix" ); - } - return $this->mUniqPrefix; - } - - /** - * Convert wikitext to HTML - * Do not call this function recursively. - * - * @param string $text Text we want to parse - * @param Title &$title A title object - * @param array $options - * @param boolean $linestart - * @param boolean $clearState - * @param int $revid number to pass in {{REVISIONID}} - * @return ParserOutput a ParserOutput - */ - public function parse( $text, &$title, $options, $linestart = true, $clearState = true, $revid = null ) { - /** - * First pass--just handle <nowiki> sections, pass the rest off - * to internalParse() which does all the real work. - */ - - global $wgUseTidy, $wgAlwaysUseTidy, $wgContLang; - $fname = 'Parser::parse-' . wfGetCaller(); - wfProfileIn( __METHOD__ ); - wfProfileIn( $fname ); - - if ( $clearState ) { - $this->clearState(); - } - - $this->mOptions = $options; - $this->setTitle( $title ); - $oldRevisionId = $this->mRevisionId; - $oldRevisionTimestamp = $this->mRevisionTimestamp; - if( $revid !== null ) { - $this->mRevisionId = $revid; - $this->mRevisionTimestamp = null; - } - $this->setOutputType( self::OT_HTML ); - wfRunHooks( 'ParserBeforeStrip', array( &$this, &$text, &$this->mStripState ) ); - # No more strip! - wfRunHooks( 'ParserAfterStrip', array( &$this, &$text, &$this->mStripState ) ); - $text = $this->internalParse( $text ); - $text = $this->mStripState->unstripGeneral( $text ); - - # Clean up special characters, only run once, next-to-last before doBlockLevels - $fixtags = array( - # french spaces, last one Guillemet-left - # only if there is something before the space - '/(.) (?=\\?|:|;|!|%|\\302\\273)/' => '\\1 \\2', - # french spaces, Guillemet-right - '/(\\302\\253) /' => '\\1 ', - ); - $text = preg_replace( array_keys($fixtags), array_values($fixtags), $text ); - - # only once and last - $text = $this->doBlockLevels( $text, $linestart ); - - $this->replaceLinkHolders( $text ); - - # the position of the parserConvert() call should not be changed. it - # assumes that the links are all replaced and the only thing left - # is the <nowiki> mark. - # Side-effects: this calls $this->mOutput->setTitleText() - $text = $wgContLang->parserConvert( $text, $this ); - - $text = $this->mStripState->unstripNoWiki( $text ); - - wfRunHooks( 'ParserBeforeTidy', array( &$this, &$text ) ); - -//!JF Move to its own function - - $uniq_prefix = $this->mUniqPrefix; - $matches = array(); - $elements = array_keys( $this->mTransparentTagHooks ); - $text = Parser::extractTagsAndParams( $elements, $text, $matches, $uniq_prefix ); - - foreach( $matches as $marker => $data ) { - list( $element, $content, $params, $tag ) = $data; - $tagName = strtolower( $element ); - if( isset( $this->mTransparentTagHooks[$tagName] ) ) { - $output = call_user_func_array( $this->mTransparentTagHooks[$tagName], - array( $content, $params, $this ) ); - } else { - $output = $tag; - } - $this->mStripState->general->setPair( $marker, $output ); - } - $text = $this->mStripState->unstripGeneral( $text ); - - $text = Sanitizer::normalizeCharReferences( $text ); - - if (($wgUseTidy and $this->mOptions->mTidy) or $wgAlwaysUseTidy) { - $text = Parser::tidy($text); - } else { - # attempt to sanitize at least some nesting problems - # (bug #2702 and quite a few others) - $tidyregs = array( - # ''Something [http://www.cool.com cool''] --> - # <i>Something</i><a href="http://www.cool.com"..><i>cool></i></a> - '/(<([bi])>)(<([bi])>)?([^<]*)(<\/?a[^<]*>)([^<]*)(<\/\\4>)?(<\/\\2>)/' => - '\\1\\3\\5\\8\\9\\6\\1\\3\\7\\8\\9', - # fix up an anchor inside another anchor, only - # at least for a single single nested link (bug 3695) - '/(<a[^>]+>)([^<]*)(<a[^>]+>[^<]*)<\/a>(.*)<\/a>/' => - '\\1\\2</a>\\3</a>\\1\\4</a>', - # fix div inside inline elements- doBlockLevels won't wrap a line which - # contains a div, so fix it up here; replace - # div with escaped text - '/(<([aib]) [^>]+>)([^<]*)(<div([^>]*)>)(.*)(<\/div>)([^<]*)(<\/\\2>)/' => - '\\1\\3<div\\5>\\6</div>\\8\\9', - # remove empty italic or bold tag pairs, some - # introduced by rules above - '/<([bi])><\/\\1>/' => '', - ); - - $text = preg_replace( - array_keys( $tidyregs ), - array_values( $tidyregs ), - $text ); - } - - wfRunHooks( 'ParserAfterTidy', array( &$this, &$text ) ); - - # Information on include size limits, for the benefit of users who try to skirt them - if ( $this->mOptions->getEnableLimitReport() ) { - $max = $this->mOptions->getMaxIncludeSize(); - $limitReport = - "NewPP limit report\n" . - "Preprocessor node count: {$this->mPPNodeCount}/{$this->mOptions->mMaxPPNodeCount}\n" . - "Post-expand include size: {$this->mIncludeSizes['post-expand']}/$max bytes\n" . - "Template argument size: {$this->mIncludeSizes['arg']}/$max bytes\n"; - wfRunHooks( 'ParserLimitReport', array( $this, &$limitReport ) ); - $text .= "\n<!-- \n$limitReport-->\n"; - } - $this->mOutput->setText( $text ); - $this->mRevisionId = $oldRevisionId; - $this->mRevisionTimestamp = $oldRevisionTimestamp; - wfProfileOut( $fname ); - wfProfileOut( __METHOD__ ); - - return $this->mOutput; - } - - /** - * Recursive parser entry point that can be called from an extension tag - * hook. - */ - function recursiveTagParse( $text ) { - wfProfileIn( __METHOD__ ); - wfRunHooks( 'ParserBeforeStrip', array( &$this, &$text, &$this->mStripState ) ); - wfRunHooks( 'ParserAfterStrip', array( &$this, &$text, &$this->mStripState ) ); - $text = $this->internalParse( $text ); - wfProfileOut( __METHOD__ ); - return $text; - } - - /** - * Expand templates and variables in the text, producing valid, static wikitext. - * Also removes comments. - */ - function preprocess( $text, $title, $options, $revid = null ) { - wfProfileIn( __METHOD__ ); - $this->clearState(); - $this->setOutputType( self::OT_PREPROCESS ); - $this->mOptions = $options; - $this->setTitle( $title ); - if( $revid !== null ) { - $this->mRevisionId = $revid; - } - wfRunHooks( 'ParserBeforeStrip', array( &$this, &$text, &$this->mStripState ) ); - wfRunHooks( 'ParserAfterStrip', array( &$this, &$text, &$this->mStripState ) ); - $text = $this->replaceVariables( $text ); - $text = $this->mStripState->unstripBoth( $text ); - wfProfileOut( __METHOD__ ); - return $text; - } - - /** - * Get a random string - * - * @private - * @static - */ - function getRandomString() { - return dechex(mt_rand(0, 0x7fffffff)) . dechex(mt_rand(0, 0x7fffffff)); - } - - function &getTitle() { return $this->mTitle; } - function getOptions() { return $this->mOptions; } - - function getFunctionLang() { - global $wgLang, $wgContLang; - return $this->mOptions->getInterfaceMessage() ? $wgLang : $wgContLang; - } - - /** - * Get a preprocessor object - */ - function getPreprocessor() { - if ( !isset( $this->mPreprocessor ) ) { - $class = $this->mPreprocessorClass; - $this->mPreprocessor = new $class( $this ); - } - return $this->mPreprocessor; - } - - /** - * Replaces all occurrences of HTML-style comments and the given tags - * in the text with a random marker and returns the next text. The output - * parameter $matches will be an associative array filled with data in - * the form: - * 'UNIQ-xxxxx' => array( - * 'element', - * 'tag content', - * array( 'param' => 'x' ), - * '<element param="x">tag content</element>' ) ) - * - * @param $elements list of element names. Comments are always extracted. - * @param $text Source text string. - * @param $uniq_prefix - * - * @public - * @static - */ - function extractTagsAndParams($elements, $text, &$matches, $uniq_prefix = ''){ - static $n = 1; - $stripped = ''; - $matches = array(); - - $taglist = implode( '|', $elements ); - $start = "/<($taglist)(\\s+[^>]*?|\\s*?)(\/?>)|<(!--)/i"; - - while ( '' != $text ) { - $p = preg_split( $start, $text, 2, PREG_SPLIT_DELIM_CAPTURE ); - $stripped .= $p[0]; - if( count( $p ) < 5 ) { - break; - } - if( count( $p ) > 5 ) { - // comment - $element = $p[4]; - $attributes = ''; - $close = ''; - $inside = $p[5]; - } else { - // tag - $element = $p[1]; - $attributes = $p[2]; - $close = $p[3]; - $inside = $p[4]; - } - - $marker = "$uniq_prefix-$element-" . sprintf('%08X', $n++) . $this->mMarkerSuffix; - $stripped .= $marker; - - if ( $close === '/>' ) { - // Empty element tag, <tag /> - $content = null; - $text = $inside; - $tail = null; - } else { - if( $element == '!--' ) { - $end = '/(-->)/'; - } else { - $end = "/(<\\/$element\\s*>)/i"; - } - $q = preg_split( $end, $inside, 2, PREG_SPLIT_DELIM_CAPTURE ); - $content = $q[0]; - if( count( $q ) < 3 ) { - # No end tag -- let it run out to the end of the text. - $tail = ''; - $text = ''; - } else { - $tail = $q[1]; - $text = $q[2]; - } - } - - $matches[$marker] = array( $element, - $content, - Sanitizer::decodeTagAttributes( $attributes ), - "<$element$attributes$close$content$tail" ); - } - return $stripped; - } - - /** - * Get a list of strippable XML-like elements - */ - function getStripList() { - global $wgRawHtml; - $elements = $this->mStripList; - if( $wgRawHtml ) { - $elements[] = 'html'; - } - if( $this->mOptions->getUseTeX() ) { - $elements[] = 'math'; - } - return $elements; - } - - /** - * @deprecated use replaceVariables - */ - function strip( $text, $state, $stripcomments = false , $dontstrip = array () ) { - return $text; - } - - /** - * Restores pre, math, and other extensions removed by strip() - * - * always call unstripNoWiki() after this one - * @private - * @deprecated use $this->mStripState->unstrip() - */ - function unstrip( $text, $state ) { - return $state->unstripGeneral( $text ); - } - - /** - * Always call this after unstrip() to preserve the order - * - * @private - * @deprecated use $this->mStripState->unstrip() - */ - function unstripNoWiki( $text, $state ) { - return $state->unstripNoWiki( $text ); - } - - /** - * @deprecated use $this->mStripState->unstripBoth() - */ - function unstripForHTML( $text ) { - return $this->mStripState->unstripBoth( $text ); - } - - /** - * Add an item to the strip state - * Returns the unique tag which must be inserted into the stripped text - * The tag will be replaced with the original text in unstrip() - * - * @private - */ - function insertStripItem( $text ) { - $rnd = "{$this->mUniqPrefix}-item-{$this->mMarkerIndex}-{$this->mMarkerSuffix}"; - $this->mMarkerIndex++; - $this->mStripState->general->setPair( $rnd, $text ); - return $rnd; - } - - /** - * Interface with html tidy, used if $wgUseTidy = true. - * If tidy isn't able to correct the markup, the original will be - * returned in all its glory with a warning comment appended. - * - * Either the external tidy program or the in-process tidy extension - * will be used depending on availability. Override the default - * $wgTidyInternal setting to disable the internal if it's not working. - * - * @param string $text Hideous HTML input - * @return string Corrected HTML output - * @public - * @static - */ - function tidy( $text ) { - global $wgTidyInternal; - $wrappedtext = '<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"'. -' "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html>'. -'<head><title>test</title></head><body>'.$text.'</body></html>'; - if( $wgTidyInternal ) { - $correctedtext = Parser::internalTidy( $wrappedtext ); - } else { - $correctedtext = Parser::externalTidy( $wrappedtext ); - } - if( is_null( $correctedtext ) ) { - wfDebug( "Tidy error detected!\n" ); - return $text . "\n<!-- Tidy found serious XHTML errors -->\n"; - } - return $correctedtext; - } - - /** - * Spawn an external HTML tidy process and get corrected markup back from it. - * - * @private - * @static - */ - function externalTidy( $text ) { - global $wgTidyConf, $wgTidyBin, $wgTidyOpts; - $fname = 'Parser::externalTidy'; - wfProfileIn( $fname ); - - $cleansource = ''; - $opts = ' -utf8'; - - $descriptorspec = array( - 0 => array('pipe', 'r'), - 1 => array('pipe', 'w'), - 2 => array('file', wfGetNull(), 'a') - ); - $pipes = array(); - $process = proc_open("$wgTidyBin -config $wgTidyConf $wgTidyOpts$opts", $descriptorspec, $pipes); - if (is_resource($process)) { - // Theoretically, this style of communication could cause a deadlock - // here. If the stdout buffer fills up, then writes to stdin could - // block. This doesn't appear to happen with tidy, because tidy only - // writes to stdout after it's finished reading from stdin. Search - // for tidyParseStdin and tidySaveStdout in console/tidy.c - fwrite($pipes[0], $text); - fclose($pipes[0]); - while (!feof($pipes[1])) { - $cleansource .= fgets($pipes[1], 1024); - } - fclose($pipes[1]); - proc_close($process); - } - - wfProfileOut( $fname ); - - if( $cleansource == '' && $text != '') { - // Some kind of error happened, so we couldn't get the corrected text. - // Just give up; we'll use the source text and append a warning. - return null; - } else { - return $cleansource; - } - } - - /** - * Use the HTML tidy PECL extension to use the tidy library in-process, - * saving the overhead of spawning a new process. - * - * 'pear install tidy' should be able to compile the extension module. - * - * @private - * @static - */ - function internalTidy( $text ) { - global $wgTidyConf, $IP, $wgDebugTidy; - $fname = 'Parser::internalTidy'; - wfProfileIn( $fname ); - - $tidy = new tidy; - $tidy->parseString( $text, $wgTidyConf, 'utf8' ); - $tidy->cleanRepair(); - if( $tidy->getStatus() == 2 ) { - // 2 is magic number for fatal error - // http://www.php.net/manual/en/function.tidy-get-status.php - $cleansource = null; - } else { - $cleansource = tidy_get_output( $tidy ); - } - if ( $wgDebugTidy && $tidy->getStatus() > 0 ) { - $cleansource .= "<!--\nTidy reports:\n" . - str_replace( '-->', '-->', $tidy->errorBuffer ) . - "\n-->"; - } - - wfProfileOut( $fname ); - return $cleansource; - } - - /** - * parse the wiki syntax used to render tables - * - * @private - */ - function doTableStuff ( $text ) { - $fname = 'Parser::doTableStuff'; - wfProfileIn( $fname ); - - $lines = explode ( "\n" , $text ); - $td_history = array (); // Is currently a td tag open? - $last_tag_history = array (); // Save history of last lag activated (td, th or caption) - $tr_history = array (); // Is currently a tr tag open? - $tr_attributes = array (); // history of tr attributes - $has_opened_tr = array(); // Did this table open a <tr> element? - $indent_level = 0; // indent level of the table - foreach ( $lines as $key => $line ) - { - $line = trim ( $line ); - - if( $line == '' ) { // empty line, go to next line - continue; - } - $first_character = $line{0}; - $matches = array(); - - if ( preg_match( '/^(:*)\{\|(.*)$/' , $line , $matches ) ) { - // First check if we are starting a new table - $indent_level = strlen( $matches[1] ); - - $attributes = $this->mStripState->unstripBoth( $matches[2] ); - $attributes = Sanitizer::fixTagAttributes ( $attributes , 'table' ); - - $lines[$key] = str_repeat( '<dl><dd>' , $indent_level ) . "<table{$attributes}>"; - array_push ( $td_history , false ); - array_push ( $last_tag_history , '' ); - array_push ( $tr_history , false ); - array_push ( $tr_attributes , '' ); - array_push ( $has_opened_tr , false ); - } else if ( count ( $td_history ) == 0 ) { - // Don't do any of the following - continue; - } else if ( substr ( $line , 0 , 2 ) == '|}' ) { - // We are ending a table - $line = '</table>' . substr ( $line , 2 ); - $last_tag = array_pop ( $last_tag_history ); - - if ( !array_pop ( $has_opened_tr ) ) { - $line = "<tr><td></td></tr>{$line}"; - } - - if ( array_pop ( $tr_history ) ) { - $line = "</tr>{$line}"; - } - - if ( array_pop ( $td_history ) ) { - $line = "</{$last_tag}>{$line}"; - } - array_pop ( $tr_attributes ); - $lines[$key] = $line . str_repeat( '</dd></dl>' , $indent_level ); - } else if ( substr ( $line , 0 , 2 ) == '|-' ) { - // Now we have a table row - $line = preg_replace( '#^\|-+#', '', $line ); - - // Whats after the tag is now only attributes - $attributes = $this->mStripState->unstripBoth( $line ); - $attributes = Sanitizer::fixTagAttributes ( $attributes , 'tr' ); - array_pop ( $tr_attributes ); - array_push ( $tr_attributes , $attributes ); - - $line = ''; - $last_tag = array_pop ( $last_tag_history ); - array_pop ( $has_opened_tr ); - array_push ( $has_opened_tr , true ); - - if ( array_pop ( $tr_history ) ) { - $line = '</tr>'; - } - - if ( array_pop ( $td_history ) ) { - $line = "</{$last_tag}>{$line}"; - } - - $lines[$key] = $line; - array_push ( $tr_history , false ); - array_push ( $td_history , false ); - array_push ( $last_tag_history , '' ); - } - else if ( $first_character == '|' || $first_character == '!' || substr ( $line , 0 , 2 ) == '|+' ) { - // This might be cell elements, td, th or captions - if ( substr ( $line , 0 , 2 ) == '|+' ) { - $first_character = '+'; - $line = substr ( $line , 1 ); - } - - $line = substr ( $line , 1 ); - - if ( $first_character == '!' ) { - $line = str_replace ( '!!' , '||' , $line ); - } - - // Split up multiple cells on the same line. - // FIXME : This can result in improper nesting of tags processed - // by earlier parser steps, but should avoid splitting up eg - // attribute values containing literal "||". - $cells = StringUtils::explodeMarkup( '||' , $line ); - - $lines[$key] = ''; - - // Loop through each table cell - foreach ( $cells as $cell ) - { - $previous = ''; - if ( $first_character != '+' ) - { - $tr_after = array_pop ( $tr_attributes ); - if ( !array_pop ( $tr_history ) ) { - $previous = "<tr{$tr_after}>\n"; - } - array_push ( $tr_history , true ); - array_push ( $tr_attributes , '' ); - array_pop ( $has_opened_tr ); - array_push ( $has_opened_tr , true ); - } - - $last_tag = array_pop ( $last_tag_history ); - - if ( array_pop ( $td_history ) ) { - $previous = "</{$last_tag}>{$previous}"; - } - - if ( $first_character == '|' ) { - $last_tag = 'td'; - } else if ( $first_character == '!' ) { - $last_tag = 'th'; - } else if ( $first_character == '+' ) { - $last_tag = 'caption'; - } else { - $last_tag = ''; - } - - array_push ( $last_tag_history , $last_tag ); - - // A cell could contain both parameters and data - $cell_data = explode ( '|' , $cell , 2 ); - - // Bug 553: Note that a '|' inside an invalid link should not - // be mistaken as delimiting cell parameters - if ( strpos( $cell_data[0], '[[' ) !== false ) { - $cell = "{$previous}<{$last_tag}>{$cell}"; - } else if ( count ( $cell_data ) == 1 ) - $cell = "{$previous}<{$last_tag}>{$cell_data[0]}"; - else { - $attributes = $this->mStripState->unstripBoth( $cell_data[0] ); - $attributes = Sanitizer::fixTagAttributes( $attributes , $last_tag ); - $cell = "{$previous}<{$last_tag}{$attributes}>{$cell_data[1]}"; - } - - $lines[$key] .= $cell; - array_push ( $td_history , true ); - } - } - } - - // Closing open td, tr && table - while ( count ( $td_history ) > 0 ) - { - if ( array_pop ( $td_history ) ) { - $lines[] = '</td>' ; - } - if ( array_pop ( $tr_history ) ) { - $lines[] = '</tr>' ; - } - if ( !array_pop ( $has_opened_tr ) ) { - $lines[] = "<tr><td></td></tr>" ; - } - - $lines[] = '</table>' ; - } - - $output = implode ( "\n" , $lines ) ; - - // special case: don't return empty table - if( $output == "<table>\n<tr><td></td></tr>\n</table>" ) { - $output = ''; - } - - wfProfileOut( $fname ); - - return $output; - } - - /** - * Helper function for parse() that transforms wiki markup into - * HTML. Only called for $mOutputType == self::OT_HTML. - * - * @private - */ - function internalParse( $text ) { - $isMain = true; - $fname = 'Parser::internalParse'; - wfProfileIn( $fname ); - - # Hook to suspend the parser in this state - if ( !wfRunHooks( 'ParserBeforeInternalParse', array( &$this, &$text, &$this->mStripState ) ) ) { - wfProfileOut( $fname ); - return $text ; - } - - $text = $this->replaceVariables( $text ); - $text = Sanitizer::removeHTMLtags( $text, array( &$this, 'attributeStripCallback' ), false, array_keys( $this->mTransparentTagHooks ) ); - wfRunHooks( 'InternalParseBeforeLinks', array( &$this, &$text, &$this->mStripState ) ); - - // Tables need to come after variable replacement for things to work - // properly; putting them before other transformations should keep - // exciting things like link expansions from showing up in surprising - // places. - $text = $this->doTableStuff( $text ); - - $text = preg_replace( '/(^|\n)-----*/', '\\1<hr />', $text ); - - $text = $this->stripToc( $text ); - $this->stripNoGallery( $text ); - $text = $this->doHeadings( $text ); - if($this->mOptions->getUseDynamicDates()) { - $df =& DateFormatter::getInstance(); - $text = $df->reformat( $this->mOptions->getDateFormat(), $text ); - } - $text = $this->doAllQuotes( $text ); - $text = $this->replaceInternalLinks( $text ); - $text = $this->replaceExternalLinks( $text ); - - # replaceInternalLinks may sometimes leave behind - # absolute URLs, which have to be masked to hide them from replaceExternalLinks - $text = str_replace($this->mUniqPrefix."NOPARSE", "", $text); - - $text = $this->doMagicLinks( $text ); - $text = $this->formatHeadings( $text, $isMain ); - - wfProfileOut( $fname ); - return $text; - } - - /** - * Replace special strings like "ISBN xxx" and "RFC xxx" with - * magic external links. - * - * @private - */ - function doMagicLinks( $text ) { - wfProfileIn( __METHOD__ ); - $text = preg_replace_callback( - '!(?: # Start cases - <a.*?</a> | # Skip link text - <.*?> | # Skip stuff inside HTML elements - (?:RFC|PMID)\s+([0-9]+) | # RFC or PMID, capture number as m[1] - ISBN\s+(\b # ISBN, capture number as m[2] - (?: 97[89] [\ \-]? )? # optional 13-digit ISBN prefix - (?: [0-9] [\ \-]? ){9} # 9 digits with opt. delimiters - [0-9Xx] # check digit - \b) - )!x', array( &$this, 'magicLinkCallback' ), $text ); - wfProfileOut( __METHOD__ ); - return $text; - } - - function magicLinkCallback( $m ) { - if ( substr( $m[0], 0, 1 ) == '<' ) { - # Skip HTML element - return $m[0]; - } elseif ( substr( $m[0], 0, 4 ) == 'ISBN' ) { - $isbn = $m[2]; - $num = strtr( $isbn, array( - '-' => '', - ' ' => '', - 'x' => 'X', - )); - $titleObj = SpecialPage::getTitleFor( 'Booksources' ); - $text = '<a href="' . - $titleObj->escapeLocalUrl( "isbn=$num" ) . - "\" class=\"internal\">ISBN $isbn</a>"; - } else { - if ( substr( $m[0], 0, 3 ) == 'RFC' ) { - $keyword = 'RFC'; - $urlmsg = 'rfcurl'; - $id = $m[1]; - } elseif ( substr( $m[0], 0, 4 ) == 'PMID' ) { - $keyword = 'PMID'; - $urlmsg = 'pubmedurl'; - $id = $m[1]; - } else { - throw new MWException( __METHOD__.': unrecognised match type "' . - substr($m[0], 0, 20 ) . '"' ); - } - - $url = wfMsg( $urlmsg, $id); - $sk = $this->mOptions->getSkin(); - $la = $sk->getExternalLinkAttributes( $url, $keyword.$id ); - $text = "<a href=\"{$url}\"{$la}>{$keyword} {$id}</a>"; - } - return $text; - } - - /** - * Parse headers and return html - * - * @private - */ - function doHeadings( $text ) { - $fname = 'Parser::doHeadings'; - wfProfileIn( $fname ); - for ( $i = 6; $i >= 1; --$i ) { - $h = str_repeat( '=', $i ); - $text = preg_replace( "/^$h(.+)$h\\s*$/m", - "<h$i>\\1</h$i>", $text ); - } - wfProfileOut( $fname ); - return $text; - } - - /** - * Replace single quotes with HTML markup - * @private - * @return string the altered text - */ - function doAllQuotes( $text ) { - $fname = 'Parser::doAllQuotes'; - wfProfileIn( $fname ); - $outtext = ''; - $lines = explode( "\n", $text ); - foreach ( $lines as $line ) { - $outtext .= $this->doQuotes ( $line ) . "\n"; - } - $outtext = substr($outtext, 0,-1); - wfProfileOut( $fname ); - return $outtext; - } - - /** - * Helper function for doAllQuotes() - */ - public function doQuotes( $text ) { - $arr = preg_split( "/(''+)/", $text, -1, PREG_SPLIT_DELIM_CAPTURE ); - if ( count( $arr ) == 1 ) - return $text; - else - { - # First, do some preliminary work. This may shift some apostrophes from - # being mark-up to being text. It also counts the number of occurrences - # of bold and italics mark-ups. - $i = 0; - $numbold = 0; - $numitalics = 0; - foreach ( $arr as $r ) - { - if ( ( $i % 2 ) == 1 ) - { - # If there are ever four apostrophes, assume the first is supposed to - # be text, and the remaining three constitute mark-up for bold text. - if ( strlen( $arr[$i] ) == 4 ) - { - $arr[$i-1] .= "'"; - $arr[$i] = "'''"; - } - # If there are more than 5 apostrophes in a row, assume they're all - # text except for the last 5. - else if ( strlen( $arr[$i] ) > 5 ) - { - $arr[$i-1] .= str_repeat( "'", strlen( $arr[$i] ) - 5 ); - $arr[$i] = "'''''"; - } - # Count the number of occurrences of bold and italics mark-ups. - # We are not counting sequences of five apostrophes. - if ( strlen( $arr[$i] ) == 2 ) { $numitalics++; } - else if ( strlen( $arr[$i] ) == 3 ) { $numbold++; } - else if ( strlen( $arr[$i] ) == 5 ) { $numitalics++; $numbold++; } - } - $i++; - } - - # If there is an odd number of both bold and italics, it is likely - # that one of the bold ones was meant to be an apostrophe followed - # by italics. Which one we cannot know for certain, but it is more - # likely to be one that has a single-letter word before it. - if ( ( $numbold % 2 == 1 ) && ( $numitalics % 2 == 1 ) ) - { - $i = 0; - $firstsingleletterword = -1; - $firstmultiletterword = -1; - $firstspace = -1; - foreach ( $arr as $r ) - { - if ( ( $i % 2 == 1 ) and ( strlen( $r ) == 3 ) ) - { - $x1 = substr ($arr[$i-1], -1); - $x2 = substr ($arr[$i-1], -2, 1); - if ($x1 == ' ') { - if ($firstspace == -1) $firstspace = $i; - } else if ($x2 == ' ') { - if ($firstsingleletterword == -1) $firstsingleletterword = $i; - } else { - if ($firstmultiletterword == -1) $firstmultiletterword = $i; - } - } - $i++; - } - - # If there is a single-letter word, use it! - if ($firstsingleletterword > -1) - { - $arr [ $firstsingleletterword ] = "''"; - $arr [ $firstsingleletterword-1 ] .= "'"; - } - # If not, but there's a multi-letter word, use that one. - else if ($firstmultiletterword > -1) - { - $arr [ $firstmultiletterword ] = "''"; - $arr [ $firstmultiletterword-1 ] .= "'"; - } - # ... otherwise use the first one that has neither. - # (notice that it is possible for all three to be -1 if, for example, - # there is only one pentuple-apostrophe in the line) - else if ($firstspace > -1) - { - $arr [ $firstspace ] = "''"; - $arr [ $firstspace-1 ] .= "'"; - } - } - - # Now let's actually convert our apostrophic mush to HTML! - $output = ''; - $buffer = ''; - $state = ''; - $i = 0; - foreach ($arr as $r) - { - if (($i % 2) == 0) - { - if ($state == 'both') - $buffer .= $r; - else - $output .= $r; - } - else - { - if (strlen ($r) == 2) - { - if ($state == 'i') - { $output .= '</i>'; $state = ''; } - else if ($state == 'bi') - { $output .= '</i>'; $state = 'b'; } - else if ($state == 'ib') - { $output .= '</b></i><b>'; $state = 'b'; } - else if ($state == 'both') - { $output .= '<b><i>'.$buffer.'</i>'; $state = 'b'; } - else # $state can be 'b' or '' - { $output .= '<i>'; $state .= 'i'; } - } - else if (strlen ($r) == 3) - { - if ($state == 'b') - { $output .= '</b>'; $state = ''; } - else if ($state == 'bi') - { $output .= '</i></b><i>'; $state = 'i'; } - else if ($state == 'ib') - { $output .= '</b>'; $state = 'i'; } - else if ($state == 'both') - { $output .= '<i><b>'.$buffer.'</b>'; $state = 'i'; } - else # $state can be 'i' or '' - { $output .= '<b>'; $state .= 'b'; } - } - else if (strlen ($r) == 5) - { - if ($state == 'b') - { $output .= '</b><i>'; $state = 'i'; } - else if ($state == 'i') - { $output .= '</i><b>'; $state = 'b'; } - else if ($state == 'bi') - { $output .= '</i></b>'; $state = ''; } - else if ($state == 'ib') - { $output .= '</b></i>'; $state = ''; } - else if ($state == 'both') - { $output .= '<i><b>'.$buffer.'</b></i>'; $state = ''; } - else # ($state == '') - { $buffer = ''; $state = 'both'; } - } - } - $i++; - } - # Now close all remaining tags. Notice that the order is important. - if ($state == 'b' || $state == 'ib') - $output .= '</b>'; - if ($state == 'i' || $state == 'bi' || $state == 'ib') - $output .= '</i>'; - if ($state == 'bi') - $output .= '</b>'; - # There might be lonely ''''', so make sure we have a buffer - if ($state == 'both' && $buffer) - $output .= '<b><i>'.$buffer.'</i></b>'; - return $output; - } - } - - /** - * Replace external links - * - * Note: this is all very hackish and the order of execution matters a lot. - * Make sure to run maintenance/parserTests.php if you change this code. - * - * @private - */ - function replaceExternalLinks( $text ) { - global $wgContLang; - $fname = 'Parser::replaceExternalLinks'; - wfProfileIn( $fname ); - - $sk = $this->mOptions->getSkin(); - - $bits = preg_split( $this->mExtLinkBracketedRegex, $text, -1, PREG_SPLIT_DELIM_CAPTURE ); - - $s = $this->replaceFreeExternalLinks( array_shift( $bits ) ); - - $i = 0; - while ( $i<count( $bits ) ) { - $url = $bits[$i++]; - $protocol = $bits[$i++]; - $text = $bits[$i++]; - $trail = $bits[$i++]; - - # The characters '<' and '>' (which were escaped by - # removeHTMLtags()) should not be included in - # URLs, per RFC 2396. - $m2 = array(); - if (preg_match('/&(lt|gt);/', $url, $m2, PREG_OFFSET_CAPTURE)) { - $text = substr($url, $m2[0][1]) . ' ' . $text; - $url = substr($url, 0, $m2[0][1]); - } - - # If the link text is an image URL, replace it with an <img> tag - # This happened by accident in the original parser, but some people used it extensively - $img = $this->maybeMakeExternalImage( $text ); - if ( $img !== false ) { - $text = $img; - } - - $dtrail = ''; - - # Set linktype for CSS - if URL==text, link is essentially free - $linktype = ($text == $url) ? 'free' : 'text'; - - # No link text, e.g. [http://domain.tld/some.link] - if ( $text == '' ) { - # Autonumber if allowed. See bug #5918 - if ( strpos( wfUrlProtocols(), substr($protocol, 0, strpos($protocol, ':')) ) !== false ) { - $text = '[' . ++$this->mAutonumber . ']'; - $linktype = 'autonumber'; - } else { - # Otherwise just use the URL - $text = htmlspecialchars( $url ); - $linktype = 'free'; - } - } else { - # Have link text, e.g. [http://domain.tld/some.link text]s - # Check for trail - list( $dtrail, $trail ) = Linker::splitTrail( $trail ); - } - - $text = $wgContLang->markNoConversion($text); - - $url = Sanitizer::cleanUrl( $url ); - - # Process the trail (i.e. everything after this link up until start of the next link), - # replacing any non-bracketed links - $trail = $this->replaceFreeExternalLinks( $trail ); - - # Use the encoded URL - # This means that users can paste URLs directly into the text - # Funny characters like ö aren't valid in URLs anyway - # This was changed in August 2004 - $s .= $sk->makeExternalLink( $url, $text, false, $linktype, $this->mTitle->getNamespace() ) . $dtrail . $trail; - - # Register link in the output object. - # Replace unnecessary URL escape codes with the referenced character - # This prevents spammers from hiding links from the filters - $pasteurized = Parser::replaceUnusualEscapes( $url ); - $this->mOutput->addExternalLink( $pasteurized ); - } - - wfProfileOut( $fname ); - return $s; - } - - /** - * Replace anything that looks like a URL with a link - * @private - */ - function replaceFreeExternalLinks( $text ) { - global $wgContLang; - $fname = 'Parser::replaceFreeExternalLinks'; - wfProfileIn( $fname ); - - $bits = preg_split( '/(\b(?:' . wfUrlProtocols() . '))/S', $text, -1, PREG_SPLIT_DELIM_CAPTURE ); - $s = array_shift( $bits ); - $i = 0; - - $sk = $this->mOptions->getSkin(); - - while ( $i < count( $bits ) ){ - $protocol = $bits[$i++]; - $remainder = $bits[$i++]; - - $m = array(); - if ( preg_match( '/^('.self::EXT_LINK_URL_CLASS.'+)(.*)$/s', $remainder, $m ) ) { - # Found some characters after the protocol that look promising - $url = $protocol . $m[1]; - $trail = $m[2]; - - # special case: handle urls as url args: - # http://www.example.com/foo?=http://www.example.com/bar - if(strlen($trail) == 0 && - isset($bits[$i]) && - preg_match('/^'. wfUrlProtocols() . '$/S', $bits[$i]) && - preg_match( '/^('.self::EXT_LINK_URL_CLASS.'+)(.*)$/s', $bits[$i + 1], $m )) - { - # add protocol, arg - $url .= $bits[$i] . $m[1]; # protocol, url as arg to previous link - $i += 2; - $trail = $m[2]; - } - - # The characters '<' and '>' (which were escaped by - # removeHTMLtags()) should not be included in - # URLs, per RFC 2396. - $m2 = array(); - if (preg_match('/&(lt|gt);/', $url, $m2, PREG_OFFSET_CAPTURE)) { - $trail = substr($url, $m2[0][1]) . $trail; - $url = substr($url, 0, $m2[0][1]); - } - - # Move trailing punctuation to $trail - $sep = ',;\.:!?'; - # If there is no left bracket, then consider right brackets fair game too - if ( strpos( $url, '(' ) === false ) { - $sep .= ')'; - } - - $numSepChars = strspn( strrev( $url ), $sep ); - if ( $numSepChars ) { - $trail = substr( $url, -$numSepChars ) . $trail; - $url = substr( $url, 0, -$numSepChars ); - } - - $url = Sanitizer::cleanUrl( $url ); - - # Is this an external image? - $text = $this->maybeMakeExternalImage( $url ); - if ( $text === false ) { - # Not an image, make a link - $text = $sk->makeExternalLink( $url, $wgContLang->markNoConversion($url), true, 'free', $this->mTitle->getNamespace() ); - # Register it in the output object... - # Replace unnecessary URL escape codes with their equivalent characters - $pasteurized = Parser::replaceUnusualEscapes( $url ); - $this->mOutput->addExternalLink( $pasteurized ); - } - $s .= $text . $trail; - } else { - $s .= $protocol . $remainder; - } - } - wfProfileOut( $fname ); - return $s; - } - - /** - * Replace unusual URL escape codes with their equivalent characters - * @param string - * @return string - * @static - * @todo This can merge genuinely required bits in the path or query string, - * breaking legit URLs. A proper fix would treat the various parts of - * the URL differently; as a workaround, just use the output for - * statistical records, not for actual linking/output. - */ - static function replaceUnusualEscapes( $url ) { - return preg_replace_callback( '/%[0-9A-Fa-f]{2}/', - array( 'Parser', 'replaceUnusualEscapesCallback' ), $url ); - } - - /** - * Callback function used in replaceUnusualEscapes(). - * Replaces unusual URL escape codes with their equivalent character - * @static - * @private - */ - private static function replaceUnusualEscapesCallback( $matches ) { - $char = urldecode( $matches[0] ); - $ord = ord( $char ); - // Is it an unsafe or HTTP reserved character according to RFC 1738? - if ( $ord > 32 && $ord < 127 && strpos( '<>"#{}|\^~[]`;/?', $char ) === false ) { - // No, shouldn't be escaped - return $char; - } else { - // Yes, leave it escaped - return $matches[0]; - } - } - - /** - * make an image if it's allowed, either through the global - * option or through the exception - * @private - */ - function maybeMakeExternalImage( $url ) { - $sk = $this->mOptions->getSkin(); - $imagesfrom = $this->mOptions->getAllowExternalImagesFrom(); - $imagesexception = !empty($imagesfrom); - $text = false; - if ( $this->mOptions->getAllowExternalImages() - || ( $imagesexception && strpos( $url, $imagesfrom ) === 0 ) ) { - if ( preg_match( self::EXT_IMAGE_REGEX, $url ) ) { - # Image found - $text = $sk->makeExternalImage( htmlspecialchars( $url ) ); - } - } - return $text; - } - - /** - * Process [[ ]] wikilinks - * - * @private - */ - function replaceInternalLinks( $s ) { - global $wgContLang; - static $fname = 'Parser::replaceInternalLinks' ; - - wfProfileIn( $fname ); - - wfProfileIn( $fname.'-setup' ); - static $tc = FALSE; - # the % is needed to support urlencoded titles as well - if ( !$tc ) { $tc = Title::legalChars() . '#%'; } - - $sk = $this->mOptions->getSkin(); - - #split the entire text string on occurences of [[ - $a = explode( '[[', ' ' . $s ); - #get the first element (all text up to first [[), and remove the space we added - $s = array_shift( $a ); - $s = substr( $s, 1 ); - - # Match a link having the form [[namespace:link|alternate]]trail - static $e1 = FALSE; - if ( !$e1 ) { $e1 = "/^([{$tc}]+)(?:\\|(.+?))?]](.*)\$/sD"; } - # Match cases where there is no "]]", which might still be images - static $e1_img = FALSE; - if ( !$e1_img ) { $e1_img = "/^([{$tc}]+)\\|(.*)\$/sD"; } - - $useLinkPrefixExtension = $wgContLang->linkPrefixExtension(); - $e2 = null; - if ( $useLinkPrefixExtension ) { - # Match the end of a line for a word that's not followed by whitespace, - # e.g. in the case of 'The Arab al[[Razi]]', 'al' will be matched - $e2 = wfMsgForContent( 'linkprefix' ); - } - - if( is_null( $this->mTitle ) ) { - throw new MWException( __METHOD__.": \$this->mTitle is null\n" ); - } - $nottalk = !$this->mTitle->isTalkPage(); - - if ( $useLinkPrefixExtension ) { - $m = array(); - if ( preg_match( $e2, $s, $m ) ) { - $first_prefix = $m[2]; - } else { - $first_prefix = false; - } - } else { - $prefix = ''; - } - - if($wgContLang->hasVariants()) { - $selflink = $wgContLang->convertLinkToAllVariants($this->mTitle->getPrefixedText()); - } else { - $selflink = array($this->mTitle->getPrefixedText()); - } - $useSubpages = $this->areSubpagesAllowed(); - wfProfileOut( $fname.'-setup' ); - - # Loop for each link - for ($k = 0; isset( $a[$k] ); $k++) { - $line = $a[$k]; - if ( $useLinkPrefixExtension ) { - wfProfileIn( $fname.'-prefixhandling' ); - if ( preg_match( $e2, $s, $m ) ) { - $prefix = $m[2]; - $s = $m[1]; - } else { - $prefix=''; - } - # first link - if($first_prefix) { - $prefix = $first_prefix; - $first_prefix = false; - } - wfProfileOut( $fname.'-prefixhandling' ); - } - - $might_be_img = false; - - wfProfileIn( "$fname-e1" ); - if ( preg_match( $e1, $line, $m ) ) { # page with normal text or alt - $text = $m[2]; - # If we get a ] at the beginning of $m[3] that means we have a link that's something like: - # [[Image:Foo.jpg|[http://example.com desc]]] <- having three ] in a row fucks up, - # the real problem is with the $e1 regex - # See bug 1300. - # - # Still some problems for cases where the ] is meant to be outside punctuation, - # and no image is in sight. See bug 2095. - # - if( $text !== '' && - substr( $m[3], 0, 1 ) === ']' && - strpos($text, '[') !== false - ) - { - $text .= ']'; # so that replaceExternalLinks($text) works later - $m[3] = substr( $m[3], 1 ); - } - # fix up urlencoded title texts - if( strpos( $m[1], '%' ) !== false ) { - # Should anchors '#' also be rejected? - $m[1] = str_replace( array('<', '>'), array('<', '>'), urldecode($m[1]) ); - } - $trail = $m[3]; - } elseif( preg_match($e1_img, $line, $m) ) { # Invalid, but might be an image with a link in its caption - $might_be_img = true; - $text = $m[2]; - if ( strpos( $m[1], '%' ) !== false ) { - $m[1] = urldecode($m[1]); - } - $trail = ""; - } else { # Invalid form; output directly - $s .= $prefix . '[[' . $line ; - wfProfileOut( "$fname-e1" ); - continue; - } - wfProfileOut( "$fname-e1" ); - wfProfileIn( "$fname-misc" ); - - # Don't allow internal links to pages containing - # PROTO: where PROTO is a valid URL protocol; these - # should be external links. - if (preg_match('/^\b(?:' . wfUrlProtocols() . ')/', $m[1])) { - $s .= $prefix . '[[' . $line ; - continue; - } - - # Make subpage if necessary - if( $useSubpages ) { - $link = $this->maybeDoSubpageLink( $m[1], $text ); - } else { - $link = $m[1]; - } - - $noforce = (substr($m[1], 0, 1) != ':'); - if (!$noforce) { - # Strip off leading ':' - $link = substr($link, 1); - } - - wfProfileOut( "$fname-misc" ); - wfProfileIn( "$fname-title" ); - $nt = Title::newFromText( $this->mStripState->unstripNoWiki($link) ); - if( !$nt ) { - $s .= $prefix . '[[' . $line; - wfProfileOut( "$fname-title" ); - continue; - } - - $ns = $nt->getNamespace(); - $iw = $nt->getInterWiki(); - wfProfileOut( "$fname-title" ); - - if ($might_be_img) { # if this is actually an invalid link - wfProfileIn( "$fname-might_be_img" ); - if ($ns == NS_IMAGE && $noforce) { #but might be an image - $found = false; - while (isset ($a[$k+1]) ) { - #look at the next 'line' to see if we can close it there - $spliced = array_splice( $a, $k + 1, 1 ); - $next_line = array_shift( $spliced ); - $m = explode( ']]', $next_line, 3 ); - if ( count( $m ) == 3 ) { - # the first ]] closes the inner link, the second the image - $found = true; - $text .= "[[{$m[0]}]]{$m[1]}"; - $trail = $m[2]; - break; - } elseif ( count( $m ) == 2 ) { - #if there's exactly one ]] that's fine, we'll keep looking - $text .= "[[{$m[0]}]]{$m[1]}"; - } else { - #if $next_line is invalid too, we need look no further - $text .= '[[' . $next_line; - break; - } - } - if ( !$found ) { - # we couldn't find the end of this imageLink, so output it raw - #but don't ignore what might be perfectly normal links in the text we've examined - $text = $this->replaceInternalLinks($text); - $s .= "{$prefix}[[$link|$text"; - # note: no $trail, because without an end, there *is* no trail - wfProfileOut( "$fname-might_be_img" ); - continue; - } - } else { #it's not an image, so output it raw - $s .= "{$prefix}[[$link|$text"; - # note: no $trail, because without an end, there *is* no trail - wfProfileOut( "$fname-might_be_img" ); - continue; - } - wfProfileOut( "$fname-might_be_img" ); - } - - $wasblank = ( '' == $text ); - if( $wasblank ) $text = $link; - - # Link not escaped by : , create the various objects - if( $noforce ) { - - # Interwikis - wfProfileIn( "$fname-interwiki" ); - if( $iw && $this->mOptions->getInterwikiMagic() && $nottalk && $wgContLang->getLanguageName( $iw ) ) { - $this->mOutput->addLanguageLink( $nt->getFullText() ); - $s = rtrim($s . $prefix); - $s .= trim($trail, "\n") == '' ? '': $prefix . $trail; - wfProfileOut( "$fname-interwiki" ); - continue; - } - wfProfileOut( "$fname-interwiki" ); - - if ( $ns == NS_IMAGE ) { - wfProfileIn( "$fname-image" ); - if ( !wfIsBadImage( $nt->getDBkey(), $this->mTitle ) ) { - # recursively parse links inside the image caption - # actually, this will parse them in any other parameters, too, - # but it might be hard to fix that, and it doesn't matter ATM - $text = $this->replaceExternalLinks($text); - $text = $this->replaceInternalLinks($text); - - # cloak any absolute URLs inside the image markup, so replaceExternalLinks() won't touch them - $s .= $prefix . $this->armorLinks( $this->makeImage( $nt, $text ) ) . $trail; - $this->mOutput->addImage( $nt->getDBkey() ); - - wfProfileOut( "$fname-image" ); - continue; - } else { - # We still need to record the image's presence on the page - $this->mOutput->addImage( $nt->getDBkey() ); - } - wfProfileOut( "$fname-image" ); - - } - - if ( $ns == NS_CATEGORY ) { - wfProfileIn( "$fname-category" ); - $s = rtrim($s . "\n"); # bug 87 - - if ( $wasblank ) { - $sortkey = $this->getDefaultSort(); - } else { - $sortkey = $text; - } - $sortkey = Sanitizer::decodeCharReferences( $sortkey ); - $sortkey = str_replace( "\n", '', $sortkey ); - $sortkey = $wgContLang->convertCategoryKey( $sortkey ); - $this->mOutput->addCategory( $nt->getDBkey(), $sortkey ); - - /** - * Strip the whitespace Category links produce, see bug 87 - * @todo We might want to use trim($tmp, "\n") here. - */ - $s .= trim($prefix . $trail, "\n") == '' ? '': $prefix . $trail; - - wfProfileOut( "$fname-category" ); - continue; - } - } - - # Self-link checking - if( $nt->getFragment() === '' ) { - if( in_array( $nt->getPrefixedText(), $selflink, true ) ) { - $s .= $prefix . $sk->makeSelfLinkObj( $nt, $text, '', $trail ); - continue; - } - } - - # Special and Media are pseudo-namespaces; no pages actually exist in them - if( $ns == NS_MEDIA ) { - $link = $sk->makeMediaLinkObj( $nt, $text ); - # Cloak with NOPARSE to avoid replacement in replaceExternalLinks - $s .= $prefix . $this->armorLinks( $link ) . $trail; - $this->mOutput->addImage( $nt->getDBkey() ); - continue; - } elseif( $ns == NS_SPECIAL ) { - if( SpecialPage::exists( $nt->getDBkey() ) ) { - $s .= $this->makeKnownLinkHolder( $nt, $text, '', $trail, $prefix ); - } else { - $s .= $this->makeLinkHolder( $nt, $text, '', $trail, $prefix ); - } - continue; - } elseif( $ns == NS_IMAGE ) { - $img = wfFindFile( $nt ); - if( $img ) { - // Force a blue link if the file exists; may be a remote - // upload on the shared repository, and we want to see its - // auto-generated page. - $s .= $this->makeKnownLinkHolder( $nt, $text, '', $trail, $prefix ); - $this->mOutput->addLink( $nt ); - continue; - } - } - $s .= $this->makeLinkHolder( $nt, $text, '', $trail, $prefix ); - } - wfProfileOut( $fname ); - return $s; - } - - /** - * Make a link placeholder. The text returned can be later resolved to a real link with - * replaceLinkHolders(). This is done for two reasons: firstly to avoid further - * parsing of interwiki links, and secondly to allow all existence checks and - * article length checks (for stub links) to be bundled into a single query. - * - */ - function makeLinkHolder( &$nt, $text = '', $query = '', $trail = '', $prefix = '' ) { - wfProfileIn( __METHOD__ ); - if ( ! is_object($nt) ) { - # Fail gracefully - $retVal = "<!-- ERROR -->{$prefix}{$text}{$trail}"; - } else { - # Separate the link trail from the rest of the link - list( $inside, $trail ) = Linker::splitTrail( $trail ); - - if ( $nt->isExternal() ) { - $nr = array_push( $this->mInterwikiLinkHolders['texts'], $prefix.$text.$inside ); - $this->mInterwikiLinkHolders['titles'][] = $nt; - $retVal = '<!--IWLINK '. ($nr-1) ."-->{$trail}"; - } else { - $nr = array_push( $this->mLinkHolders['namespaces'], $nt->getNamespace() ); - $this->mLinkHolders['dbkeys'][] = $nt->getDBkey(); - $this->mLinkHolders['queries'][] = $query; - $this->mLinkHolders['texts'][] = $prefix.$text.$inside; - $this->mLinkHolders['titles'][] = $nt; - - $retVal = '<!--LINK '. ($nr-1) ."-->{$trail}"; - } - } - wfProfileOut( __METHOD__ ); - return $retVal; - } - - /** - * Render a forced-blue link inline; protect against double expansion of - * URLs if we're in a mode that prepends full URL prefixes to internal links. - * Since this little disaster has to split off the trail text to avoid - * breaking URLs in the following text without breaking trails on the - * wiki links, it's been made into a horrible function. - * - * @param Title $nt - * @param string $text - * @param string $query - * @param string $trail - * @param string $prefix - * @return string HTML-wikitext mix oh yuck - */ - function makeKnownLinkHolder( $nt, $text = '', $query = '', $trail = '', $prefix = '' ) { - list( $inside, $trail ) = Linker::splitTrail( $trail ); - $sk = $this->mOptions->getSkin(); - $link = $sk->makeKnownLinkObj( $nt, $text, $query, $inside, $prefix ); - return $this->armorLinks( $link ) . $trail; - } - - /** - * Insert a NOPARSE hacky thing into any inline links in a chunk that's - * going to go through further parsing steps before inline URL expansion. - * - * In particular this is important when using action=render, which causes - * full URLs to be included. - * - * Oh man I hate our multi-layer parser! - * - * @param string more-or-less HTML - * @return string less-or-more HTML with NOPARSE bits - */ - function armorLinks( $text ) { - return preg_replace( '/\b(' . wfUrlProtocols() . ')/', - "{$this->mUniqPrefix}NOPARSE$1", $text ); - } - - /** - * Return true if subpage links should be expanded on this page. - * @return bool - */ - function areSubpagesAllowed() { - # Some namespaces don't allow subpages - global $wgNamespacesWithSubpages; - return !empty($wgNamespacesWithSubpages[$this->mTitle->getNamespace()]); - } - - /** - * Handle link to subpage if necessary - * @param string $target the source of the link - * @param string &$text the link text, modified as necessary - * @return string the full name of the link - * @private - */ - function maybeDoSubpageLink($target, &$text) { - # Valid link forms: - # Foobar -- normal - # :Foobar -- override special treatment of prefix (images, language links) - # /Foobar -- convert to CurrentPage/Foobar - # /Foobar/ -- convert to CurrentPage/Foobar, strip the initial / from text - # ../ -- convert to CurrentPage, from CurrentPage/CurrentSubPage - # ../Foobar -- convert to CurrentPage/Foobar, from CurrentPage/CurrentSubPage - - $fname = 'Parser::maybeDoSubpageLink'; - wfProfileIn( $fname ); - $ret = $target; # default return value is no change - - # Some namespaces don't allow subpages, - # so only perform processing if subpages are allowed - if( $this->areSubpagesAllowed() ) { - $hash = strpos( $target, '#' ); - if( $hash !== false ) { - $suffix = substr( $target, $hash ); - $target = substr( $target, 0, $hash ); - } else { - $suffix = ''; - } - # bug 7425 - $target = trim( $target ); - # Look at the first character - if( $target != '' && $target{0} == '/' ) { - # / at end means we don't want the slash to be shown - $m = array(); - $trailingSlashes = preg_match_all( '%(/+)$%', $target, $m ); - if( $trailingSlashes ) { - $noslash = $target = substr( $target, 1, -strlen($m[0][0]) ); - } else { - $noslash = substr( $target, 1 ); - } - - $ret = $this->mTitle->getPrefixedText(). '/' . trim($noslash) . $suffix; - if( '' === $text ) { - $text = $target . $suffix; - } # this might be changed for ugliness reasons - } else { - # check for .. subpage backlinks - $dotdotcount = 0; - $nodotdot = $target; - while( strncmp( $nodotdot, "../", 3 ) == 0 ) { - ++$dotdotcount; - $nodotdot = substr( $nodotdot, 3 ); - } - if($dotdotcount > 0) { - $exploded = explode( '/', $this->mTitle->GetPrefixedText() ); - if( count( $exploded ) > $dotdotcount ) { # not allowed to go below top level page - $ret = implode( '/', array_slice( $exploded, 0, -$dotdotcount ) ); - # / at the end means don't show full path - if( substr( $nodotdot, -1, 1 ) == '/' ) { - $nodotdot = substr( $nodotdot, 0, -1 ); - if( '' === $text ) { - $text = $nodotdot . $suffix; - } - } - $nodotdot = trim( $nodotdot ); - if( $nodotdot != '' ) { - $ret .= '/' . $nodotdot; - } - $ret .= $suffix; - } - } - } - } - - wfProfileOut( $fname ); - return $ret; - } - - /**#@+ - * Used by doBlockLevels() - * @private - */ - /* private */ function closeParagraph() { - $result = ''; - if ( '' != $this->mLastSection ) { - $result = '</' . $this->mLastSection . ">\n"; - } - $this->mInPre = false; - $this->mLastSection = ''; - return $result; - } - # getCommon() returns the length of the longest common substring - # of both arguments, starting at the beginning of both. - # - /* private */ function getCommon( $st1, $st2 ) { - $fl = strlen( $st1 ); - $shorter = strlen( $st2 ); - if ( $fl < $shorter ) { $shorter = $fl; } - - for ( $i = 0; $i < $shorter; ++$i ) { - if ( $st1{$i} != $st2{$i} ) { break; } - } - return $i; - } - # These next three functions open, continue, and close the list - # element appropriate to the prefix character passed into them. - # - /* private */ function openList( $char ) { - $result = $this->closeParagraph(); - - if ( '*' == $char ) { $result .= '<ul><li>'; } - else if ( '#' == $char ) { $result .= '<ol><li>'; } - else if ( ':' == $char ) { $result .= '<dl><dd>'; } - else if ( ';' == $char ) { - $result .= '<dl><dt>'; - $this->mDTopen = true; - } - else { $result = '<!-- ERR 1 -->'; } - - return $result; - } - - /* private */ function nextItem( $char ) { - if ( '*' == $char || '#' == $char ) { return '</li><li>'; } - else if ( ':' == $char || ';' == $char ) { - $close = '</dd>'; - if ( $this->mDTopen ) { $close = '</dt>'; } - if ( ';' == $char ) { - $this->mDTopen = true; - return $close . '<dt>'; - } else { - $this->mDTopen = false; - return $close . '<dd>'; - } - } - return '<!-- ERR 2 -->'; - } - - /* private */ function closeList( $char ) { - if ( '*' == $char ) { $text = '</li></ul>'; } - else if ( '#' == $char ) { $text = '</li></ol>'; } - else if ( ':' == $char ) { - if ( $this->mDTopen ) { - $this->mDTopen = false; - $text = '</dt></dl>'; - } else { - $text = '</dd></dl>'; - } - } - else { return '<!-- ERR 3 -->'; } - return $text."\n"; - } - /**#@-*/ - - /** - * Make lists from lines starting with ':', '*', '#', etc. - * - * @private - * @return string the lists rendered as HTML - */ - function doBlockLevels( $text, $linestart ) { - $fname = 'Parser::doBlockLevels'; - wfProfileIn( $fname ); - - # Parsing through the text line by line. The main thing - # happening here is handling of block-level elements p, pre, - # and making lists from lines starting with * # : etc. - # - $textLines = explode( "\n", $text ); - - $lastPrefix = $output = ''; - $this->mDTopen = $inBlockElem = false; - $prefixLength = 0; - $paragraphStack = false; - - if ( !$linestart ) { - $output .= array_shift( $textLines ); - } - foreach ( $textLines as $oLine ) { - $lastPrefixLength = strlen( $lastPrefix ); - $preCloseMatch = preg_match('/<\\/pre/i', $oLine ); - $preOpenMatch = preg_match('/<pre/i', $oLine ); - if ( !$this->mInPre ) { - # Multiple prefixes may abut each other for nested lists. - $prefixLength = strspn( $oLine, '*#:;' ); - $pref = substr( $oLine, 0, $prefixLength ); - - # eh? - $pref2 = str_replace( ';', ':', $pref ); - $t = substr( $oLine, $prefixLength ); - $this->mInPre = !empty($preOpenMatch); - } else { - # Don't interpret any other prefixes in preformatted text - $prefixLength = 0; - $pref = $pref2 = ''; - $t = $oLine; - } - - # List generation - if( $prefixLength && 0 == strcmp( $lastPrefix, $pref2 ) ) { - # Same as the last item, so no need to deal with nesting or opening stuff - $output .= $this->nextItem( substr( $pref, -1 ) ); - $paragraphStack = false; - - if ( substr( $pref, -1 ) == ';') { - # The one nasty exception: definition lists work like this: - # ; title : definition text - # So we check for : in the remainder text to split up the - # title and definition, without b0rking links. - $term = $t2 = ''; - if ($this->findColonNoLinks($t, $term, $t2) !== false) { - $t = $t2; - $output .= $term . $this->nextItem( ':' ); - } - } - } elseif( $prefixLength || $lastPrefixLength ) { - # Either open or close a level... - $commonPrefixLength = $this->getCommon( $pref, $lastPrefix ); - $paragraphStack = false; - - while( $commonPrefixLength < $lastPrefixLength ) { - $output .= $this->closeList( $lastPrefix{$lastPrefixLength-1} ); - --$lastPrefixLength; - } - if ( $prefixLength <= $commonPrefixLength && $commonPrefixLength > 0 ) { - $output .= $this->nextItem( $pref{$commonPrefixLength-1} ); - } - while ( $prefixLength > $commonPrefixLength ) { - $char = substr( $pref, $commonPrefixLength, 1 ); - $output .= $this->openList( $char ); - - if ( ';' == $char ) { - # FIXME: This is dupe of code above - if ($this->findColonNoLinks($t, $term, $t2) !== false) { - $t = $t2; - $output .= $term . $this->nextItem( ':' ); - } - } - ++$commonPrefixLength; - } - $lastPrefix = $pref2; - } - if( 0 == $prefixLength ) { - wfProfileIn( "$fname-paragraph" ); - # No prefix (not in list)--go to paragraph mode - // XXX: use a stack for nestable elements like span, table and div - $openmatch = preg_match('/(?:<table|<blockquote|<h1|<h2|<h3|<h4|<h5|<h6|<pre|<tr|<p|<ul|<ol|<li|<\\/tr|<\\/td|<\\/th)/iS', $t ); - $closematch = preg_match( - '/(?:<\\/table|<\\/blockquote|<\\/h1|<\\/h2|<\\/h3|<\\/h4|<\\/h5|<\\/h6|'. - '<td|<th|<\\/?div|<hr|<\\/pre|<\\/p|'.$this->mUniqPrefix.'-pre|<\\/li|<\\/ul|<\\/ol|<\\/?center)/iS', $t ); - if ( $openmatch or $closematch ) { - $paragraphStack = false; - #Â TODO bug 5718: paragraph closed - $output .= $this->closeParagraph(); - if ( $preOpenMatch and !$preCloseMatch ) { - $this->mInPre = true; - } - if ( $closematch ) { - $inBlockElem = false; - } else { - $inBlockElem = true; - } - } else if ( !$inBlockElem && !$this->mInPre ) { - if ( ' ' == $t{0} and ( $this->mLastSection == 'pre' or trim($t) != '' ) ) { - // pre - if ($this->mLastSection != 'pre') { - $paragraphStack = false; - $output .= $this->closeParagraph().'<pre>'; - $this->mLastSection = 'pre'; - } - $t = substr( $t, 1 ); - } else { - // paragraph - if ( '' == trim($t) ) { - if ( $paragraphStack ) { - $output .= $paragraphStack.'<br />'; - $paragraphStack = false; - $this->mLastSection = 'p'; - } else { - if ($this->mLastSection != 'p' ) { - $output .= $this->closeParagraph(); - $this->mLastSection = ''; - $paragraphStack = '<p>'; - } else { - $paragraphStack = '</p><p>'; - } - } - } else { - if ( $paragraphStack ) { - $output .= $paragraphStack; - $paragraphStack = false; - $this->mLastSection = 'p'; - } else if ($this->mLastSection != 'p') { - $output .= $this->closeParagraph().'<p>'; - $this->mLastSection = 'p'; - } - } - } - } - wfProfileOut( "$fname-paragraph" ); - } - // somewhere above we forget to get out of pre block (bug 785) - if($preCloseMatch && $this->mInPre) { - $this->mInPre = false; - } - if ($paragraphStack === false) { - $output .= $t."\n"; - } - } - while ( $prefixLength ) { - $output .= $this->closeList( $pref2{$prefixLength-1} ); - --$prefixLength; - } - if ( '' != $this->mLastSection ) { - $output .= '</' . $this->mLastSection . '>'; - $this->mLastSection = ''; - } - - wfProfileOut( $fname ); - return $output; - } - - /** - * Split up a string on ':', ignoring any occurences inside tags - * to prevent illegal overlapping. - * @param string $str the string to split - * @param string &$before set to everything before the ':' - * @param string &$after set to everything after the ':' - * return string the position of the ':', or false if none found - */ - function findColonNoLinks($str, &$before, &$after) { - $fname = 'Parser::findColonNoLinks'; - wfProfileIn( $fname ); - - $pos = strpos( $str, ':' ); - if( $pos === false ) { - // Nothing to find! - wfProfileOut( $fname ); - return false; - } - - $lt = strpos( $str, '<' ); - if( $lt === false || $lt > $pos ) { - // Easy; no tag nesting to worry about - $before = substr( $str, 0, $pos ); - $after = substr( $str, $pos+1 ); - wfProfileOut( $fname ); - return $pos; - } - - // Ugly state machine to walk through avoiding tags. - $state = self::COLON_STATE_TEXT; - $stack = 0; - $len = strlen( $str ); - for( $i = 0; $i < $len; $i++ ) { - $c = $str{$i}; - - switch( $state ) { - // (Using the number is a performance hack for common cases) - case 0: // self::COLON_STATE_TEXT: - switch( $c ) { - case "<": - // Could be either a <start> tag or an </end> tag - $state = self::COLON_STATE_TAGSTART; - break; - case ":": - if( $stack == 0 ) { - // We found it! - $before = substr( $str, 0, $i ); - $after = substr( $str, $i + 1 ); - wfProfileOut( $fname ); - return $i; - } - // Embedded in a tag; don't break it. - break; - default: - // Skip ahead looking for something interesting - $colon = strpos( $str, ':', $i ); - if( $colon === false ) { - // Nothing else interesting - wfProfileOut( $fname ); - return false; - } - $lt = strpos( $str, '<', $i ); - if( $stack === 0 ) { - if( $lt === false || $colon < $lt ) { - // We found it! - $before = substr( $str, 0, $colon ); - $after = substr( $str, $colon + 1 ); - wfProfileOut( $fname ); - return $i; - } - } - if( $lt === false ) { - // Nothing else interesting to find; abort! - // We're nested, but there's no close tags left. Abort! - break 2; - } - // Skip ahead to next tag start - $i = $lt; - $state = self::COLON_STATE_TAGSTART; - } - break; - case 1: // self::COLON_STATE_TAG: - // In a <tag> - switch( $c ) { - case ">": - $stack++; - $state = self::COLON_STATE_TEXT; - break; - case "/": - // Slash may be followed by >? - $state = self::COLON_STATE_TAGSLASH; - break; - default: - // ignore - } - break; - case 2: // self::COLON_STATE_TAGSTART: - switch( $c ) { - case "/": - $state = self::COLON_STATE_CLOSETAG; - break; - case "!": - $state = self::COLON_STATE_COMMENT; - break; - case ">": - // Illegal early close? This shouldn't happen D: - $state = self::COLON_STATE_TEXT; - break; - default: - $state = self::COLON_STATE_TAG; - } - break; - case 3: // self::COLON_STATE_CLOSETAG: - // In a </tag> - if( $c == ">" ) { - $stack--; - if( $stack < 0 ) { - wfDebug( "Invalid input in $fname; too many close tags\n" ); - wfProfileOut( $fname ); - return false; - } - $state = self::COLON_STATE_TEXT; - } - break; - case self::COLON_STATE_TAGSLASH: - if( $c == ">" ) { - // Yes, a self-closed tag <blah/> - $state = self::COLON_STATE_TEXT; - } else { - // Probably we're jumping the gun, and this is an attribute - $state = self::COLON_STATE_TAG; - } - break; - case 5: // self::COLON_STATE_COMMENT: - if( $c == "-" ) { - $state = self::COLON_STATE_COMMENTDASH; - } - break; - case self::COLON_STATE_COMMENTDASH: - if( $c == "-" ) { - $state = self::COLON_STATE_COMMENTDASHDASH; - } else { - $state = self::COLON_STATE_COMMENT; - } - break; - case self::COLON_STATE_COMMENTDASHDASH: - if( $c == ">" ) { - $state = self::COLON_STATE_TEXT; - } else { - $state = self::COLON_STATE_COMMENT; - } - break; - default: - throw new MWException( "State machine error in $fname" ); - } - } - if( $stack > 0 ) { - wfDebug( "Invalid input in $fname; not enough close tags (stack $stack, state $state)\n" ); - return false; - } - wfProfileOut( $fname ); - return false; - } - - /** - * Return value of a magic variable (like PAGENAME) - * - * @private - */ - function getVariableValue( $index ) { - global $wgContLang, $wgSitename, $wgServer, $wgServerName, $wgScriptPath; - - /** - * Some of these require message or data lookups and can be - * expensive to check many times. - */ - if ( wfRunHooks( 'ParserGetVariableValueVarCache', array( &$this, &$this->mVarCache ) ) ) { - if ( isset( $this->mVarCache[$index] ) ) { - return $this->mVarCache[$index]; - } - } - - $ts = wfTimestamp( TS_UNIX, $this->mOptions->getTimestamp() ); - wfRunHooks( 'ParserGetVariableValueTs', array( &$this, &$ts ) ); - - # Use the time zone - global $wgLocaltimezone; - if ( isset( $wgLocaltimezone ) ) { - $oldtz = getenv( 'TZ' ); - putenv( 'TZ='.$wgLocaltimezone ); - } - - wfSuppressWarnings(); // E_STRICT system time bitching - $localTimestamp = date( 'YmdHis', $ts ); - $localMonth = date( 'm', $ts ); - $localMonthName = date( 'n', $ts ); - $localDay = date( 'j', $ts ); - $localDay2 = date( 'd', $ts ); - $localDayOfWeek = date( 'w', $ts ); - $localWeek = date( 'W', $ts ); - $localYear = date( 'Y', $ts ); - $localHour = date( 'H', $ts ); - if ( isset( $wgLocaltimezone ) ) { - putenv( 'TZ='.$oldtz ); - } - wfRestoreWarnings(); - - switch ( $index ) { - case 'currentmonth': - return $this->mVarCache[$index] = $wgContLang->formatNum( gmdate( 'm', $ts ) ); - case 'currentmonthname': - return $this->mVarCache[$index] = $wgContLang->getMonthName( gmdate( 'n', $ts ) ); - case 'currentmonthnamegen': - return $this->mVarCache[$index] = $wgContLang->getMonthNameGen( gmdate( 'n', $ts ) ); - case 'currentmonthabbrev': - return $this->mVarCache[$index] = $wgContLang->getMonthAbbreviation( gmdate( 'n', $ts ) ); - case 'currentday': - return $this->mVarCache[$index] = $wgContLang->formatNum( gmdate( 'j', $ts ) ); - case 'currentday2': - return $this->mVarCache[$index] = $wgContLang->formatNum( gmdate( 'd', $ts ) ); - case 'localmonth': - return $this->mVarCache[$index] = $wgContLang->formatNum( $localMonth ); - case 'localmonthname': - return $this->mVarCache[$index] = $wgContLang->getMonthName( $localMonthName ); - case 'localmonthnamegen': - return $this->mVarCache[$index] = $wgContLang->getMonthNameGen( $localMonthName ); - case 'localmonthabbrev': - return $this->mVarCache[$index] = $wgContLang->getMonthAbbreviation( $localMonthName ); - case 'localday': - return $this->mVarCache[$index] = $wgContLang->formatNum( $localDay ); - case 'localday2': - return $this->mVarCache[$index] = $wgContLang->formatNum( $localDay2 ); - case 'pagename': - return wfEscapeWikiText( $this->mTitle->getText() ); - case 'pagenamee': - return $this->mTitle->getPartialURL(); - case 'fullpagename': - return wfEscapeWikiText( $this->mTitle->getPrefixedText() ); - case 'fullpagenamee': - return $this->mTitle->getPrefixedURL(); - case 'subpagename': - return wfEscapeWikiText( $this->mTitle->getSubpageText() ); - case 'subpagenamee': - return $this->mTitle->getSubpageUrlForm(); - case 'basepagename': - return wfEscapeWikiText( $this->mTitle->getBaseText() ); - case 'basepagenamee': - return wfUrlEncode( str_replace( ' ', '_', $this->mTitle->getBaseText() ) ); - case 'talkpagename': - if( $this->mTitle->canTalk() ) { - $talkPage = $this->mTitle->getTalkPage(); - return wfEscapeWikiText( $talkPage->getPrefixedText() ); - } else { - return ''; - } - case 'talkpagenamee': - if( $this->mTitle->canTalk() ) { - $talkPage = $this->mTitle->getTalkPage(); - return $talkPage->getPrefixedUrl(); - } else { - return ''; - } - case 'subjectpagename': - $subjPage = $this->mTitle->getSubjectPage(); - return wfEscapeWikiText( $subjPage->getPrefixedText() ); - case 'subjectpagenamee': - $subjPage = $this->mTitle->getSubjectPage(); - return $subjPage->getPrefixedUrl(); - case 'revisionid': - // Let the edit saving system know we should parse the page - // *after* a revision ID has been assigned. - $this->mOutput->setFlag( 'vary-revision' ); - wfDebug( __METHOD__ . ": {{REVISIONID}} used, setting vary-revision...\n" ); - return $this->mRevisionId; - case 'revisionday': - // Let the edit saving system know we should parse the page - // *after* a revision ID has been assigned. This is for null edits. - $this->mOutput->setFlag( 'vary-revision' ); - wfDebug( __METHOD__ . ": {{REVISIONDAY}} used, setting vary-revision...\n" ); - return intval( substr( $this->getRevisionTimestamp(), 6, 2 ) ); - case 'revisionday2': - // Let the edit saving system know we should parse the page - // *after* a revision ID has been assigned. This is for null edits. - $this->mOutput->setFlag( 'vary-revision' ); - wfDebug( __METHOD__ . ": {{REVISIONDAY2}} used, setting vary-revision...\n" ); - return substr( $this->getRevisionTimestamp(), 6, 2 ); - case 'revisionmonth': - // Let the edit saving system know we should parse the page - // *after* a revision ID has been assigned. This is for null edits. - $this->mOutput->setFlag( 'vary-revision' ); - wfDebug( __METHOD__ . ": {{REVISIONMONTH}} used, setting vary-revision...\n" ); - return intval( substr( $this->getRevisionTimestamp(), 4, 2 ) ); - case 'revisionyear': - // Let the edit saving system know we should parse the page - // *after* a revision ID has been assigned. This is for null edits. - $this->mOutput->setFlag( 'vary-revision' ); - wfDebug( __METHOD__ . ": {{REVISIONYEAR}} used, setting vary-revision...\n" ); - return substr( $this->getRevisionTimestamp(), 0, 4 ); - case 'revisiontimestamp': - // Let the edit saving system know we should parse the page - // *after* a revision ID has been assigned. This is for null edits. - $this->mOutput->setFlag( 'vary-revision' ); - wfDebug( __METHOD__ . ": {{REVISIONTIMESTAMP}} used, setting vary-revision...\n" ); - return $this->getRevisionTimestamp(); - case 'namespace': - return str_replace('_',' ',$wgContLang->getNsText( $this->mTitle->getNamespace() ) ); - case 'namespacee': - return wfUrlencode( $wgContLang->getNsText( $this->mTitle->getNamespace() ) ); - case 'talkspace': - return $this->mTitle->canTalk() ? str_replace('_',' ',$this->mTitle->getTalkNsText()) : ''; - case 'talkspacee': - return $this->mTitle->canTalk() ? wfUrlencode( $this->mTitle->getTalkNsText() ) : ''; - case 'subjectspace': - return $this->mTitle->getSubjectNsText(); - case 'subjectspacee': - return( wfUrlencode( $this->mTitle->getSubjectNsText() ) ); - case 'currentdayname': - return $this->mVarCache[$index] = $wgContLang->getWeekdayName( gmdate( 'w', $ts ) + 1 ); - case 'currentyear': - return $this->mVarCache[$index] = $wgContLang->formatNum( gmdate( 'Y', $ts ), true ); - case 'currenttime': - return $this->mVarCache[$index] = $wgContLang->time( wfTimestamp( TS_MW, $ts ), false, false ); - case 'currenthour': - return $this->mVarCache[$index] = $wgContLang->formatNum( gmdate( 'H', $ts ), true ); - case 'currentweek': - // @bug 4594 PHP5 has it zero padded, PHP4 does not, cast to - // int to remove the padding - return $this->mVarCache[$index] = $wgContLang->formatNum( (int)gmdate( 'W', $ts ) ); - case 'currentdow': - return $this->mVarCache[$index] = $wgContLang->formatNum( gmdate( 'w', $ts ) ); - case 'localdayname': - return $this->mVarCache[$index] = $wgContLang->getWeekdayName( $localDayOfWeek + 1 ); - case 'localyear': - return $this->mVarCache[$index] = $wgContLang->formatNum( $localYear, true ); - case 'localtime': - return $this->mVarCache[$index] = $wgContLang->time( $localTimestamp, false, false ); - case 'localhour': - return $this->mVarCache[$index] = $wgContLang->formatNum( $localHour, true ); - case 'localweek': - // @bug 4594 PHP5 has it zero padded, PHP4 does not, cast to - // int to remove the padding - return $this->mVarCache[$index] = $wgContLang->formatNum( (int)$localWeek ); - case 'localdow': - return $this->mVarCache[$index] = $wgContLang->formatNum( $localDayOfWeek ); - case 'numberofarticles': - return $this->mVarCache[$index] = $wgContLang->formatNum( SiteStats::articles() ); - case 'numberoffiles': - return $this->mVarCache[$index] = $wgContLang->formatNum( SiteStats::images() ); - case 'numberofusers': - return $this->mVarCache[$index] = $wgContLang->formatNum( SiteStats::users() ); - case 'numberofpages': - return $this->mVarCache[$index] = $wgContLang->formatNum( SiteStats::pages() ); - case 'numberofadmins': - return $this->mVarCache[$index] = $wgContLang->formatNum( SiteStats::admins() ); - case 'numberofedits': - return $this->mVarCache[$index] = $wgContLang->formatNum( SiteStats::edits() ); - case 'currenttimestamp': - return $this->mVarCache[$index] = wfTimestamp( TS_MW, $ts ); - case 'localtimestamp': - return $this->mVarCache[$index] = $localTimestamp; - case 'currentversion': - return $this->mVarCache[$index] = SpecialVersion::getVersion(); - case 'sitename': - return $wgSitename; - case 'server': - return $wgServer; - case 'servername': - return $wgServerName; - case 'scriptpath': - return $wgScriptPath; - case 'directionmark': - return $wgContLang->getDirMark(); - case 'contentlanguage': - global $wgContLanguageCode; - return $wgContLanguageCode; - default: - $ret = null; - if ( wfRunHooks( 'ParserGetVariableValueSwitch', array( &$this, &$this->mVarCache, &$index, &$ret ) ) ) - return $ret; - else - return null; - } - } - - /** - * initialise the magic variables (like CURRENTMONTHNAME) - * - * @private - */ - function initialiseVariables() { - $fname = 'Parser::initialiseVariables'; - wfProfileIn( $fname ); - $variableIDs = MagicWord::getVariableIDs(); - - $this->mVariables = new MagicWordArray( $variableIDs ); - wfProfileOut( $fname ); - } - - /** - * Preprocess some wikitext and return the document tree. - * This is the ghost of replace_variables(). - * - * @param string $text The text to parse - * @param integer flags Bitwise combination of: - * self::PTD_FOR_INCLUSION Handle <noinclude>/<includeonly> as if the text is being - * included. Default is to assume a direct page view. - * - * The generated DOM tree must depend only on the input text and the flags. - * The DOM tree must be the same in OT_HTML and OT_WIKI mode, to avoid a regression of bug 4899. - * - * Any flag added to the $flags parameter here, or any other parameter liable to cause a - * change in the DOM tree for a given text, must be passed through the section identifier - * in the section edit link and thus back to extractSections(). - * - * The output of this function is currently only cached in process memory, but a persistent - * cache may be implemented at a later date which takes further advantage of these strict - * dependency requirements. - * - * @private - */ - function preprocessToDom ( $text, $flags = 0 ) { - $dom = $this->getPreprocessor()->preprocessToObj( $text, $flags ); - return $dom; - } - - /* - * Return a three-element array: leading whitespace, string contents, trailing whitespace - */ - public static function splitWhitespace( $s ) { - $ltrimmed = ltrim( $s ); - $w1 = substr( $s, 0, strlen( $s ) - strlen( $ltrimmed ) ); - $trimmed = rtrim( $ltrimmed ); - $diff = strlen( $ltrimmed ) - strlen( $trimmed ); - if ( $diff > 0 ) { - $w2 = substr( $ltrimmed, -$diff ); - } else { - $w2 = ''; - } - return array( $w1, $trimmed, $w2 ); - } - - /** - * Replace magic variables, templates, and template arguments - * with the appropriate text. Templates are substituted recursively, - * taking care to avoid infinite loops. - * - * Note that the substitution depends on value of $mOutputType: - * self::OT_WIKI: only {{subst:}} templates - * self::OT_PREPROCESS: templates but not extension tags - * self::OT_HTML: all templates and extension tags - * - * @param string $tex The text to transform - * @param PPFrame $frame Object describing the arguments passed to the template - * @param bool $argsOnly Only do argument (triple-brace) expansion, not double-brace expansion - * @private - */ - function replaceVariables( $text, $frame = false, $argsOnly = false ) { - # Prevent too big inclusions - if( strlen( $text ) > $this->mOptions->getMaxIncludeSize() ) { - return $text; - } - - $fname = __METHOD__; - wfProfileIn( $fname ); - - if ( $frame === false ) { - $frame = $this->getPreprocessor()->newFrame(); - } elseif ( !( $frame instanceof PPFrame ) ) { - throw new MWException( __METHOD__ . ' called using the old argument format' ); - } - - $dom = $this->preprocessToDom( $text ); - $flags = $argsOnly ? PPFrame::NO_TEMPLATES : 0; - $text = $frame->expand( $dom, $flags ); - - wfProfileOut( $fname ); - return $text; - } - - /// Clean up argument array - refactored in 1.9 so parserfunctions can use it, too. - static function createAssocArgs( $args ) { - $assocArgs = array(); - $index = 1; - foreach( $args as $arg ) { - $eqpos = strpos( $arg, '=' ); - if ( $eqpos === false ) { - $assocArgs[$index++] = $arg; - } else { - $name = trim( substr( $arg, 0, $eqpos ) ); - $value = trim( substr( $arg, $eqpos+1 ) ); - if ( $value === false ) { - $value = ''; - } - if ( $name !== false ) { - $assocArgs[$name] = $value; - } - } - } - - return $assocArgs; - } - - /** - * Return the text of a template, after recursively - * replacing any variables or templates within the template. - * - * @param array $piece The parts of the template - * $piece['title']: the title, i.e. the part before the | - * $piece['parts']: the parameter array - * $piece['lineStart']: whether the brace was at the start of a line - * @param PPFrame The current frame, contains template arguments - * @return string the text of the template - * @private - */ - function braceSubstitution( $piece, $frame ) { - global $wgContLang, $wgLang, $wgAllowDisplayTitle, $wgNonincludableNamespaces; - $fname = __METHOD__; - wfProfileIn( $fname ); - wfProfileIn( __METHOD__.'-setup' ); - - # Flags - $found = false; # $text has been filled - $nowiki = false; # wiki markup in $text should be escaped - $isHTML = false; # $text is HTML, armour it against wikitext transformation - $forceRawInterwiki = false; # Force interwiki transclusion to be done in raw mode not rendered - $isChildObj = false; # $text is a DOM node needing expansion in a child frame - $isLocalObj = false; # $text is a DOM node needing expansion in the current frame - - # Title object, where $text came from - $title = NULL; - - # $part1 is the bit before the first |, and must contain only title characters. - # Various prefixes will be stripped from it later. - $titleWithSpaces = $frame->expand( $piece['title'] ); - $part1 = trim( $titleWithSpaces ); - $titleText = false; - - # Original title text preserved for various purposes - $originalTitle = $part1; - - # $args is a list of argument nodes, starting from index 0, not including $part1 - $args = (null == $piece['parts']) ? array() : $piece['parts']; - wfProfileOut( __METHOD__.'-setup' ); - - # SUBST - wfProfileIn( __METHOD__.'-modifiers' ); - if ( !$found ) { - $mwSubst =& MagicWord::get( 'subst' ); - if ( $mwSubst->matchStartAndRemove( $part1 ) xor $this->ot['wiki'] ) { - # One of two possibilities is true: - # 1) Found SUBST but not in the PST phase - # 2) Didn't find SUBST and in the PST phase - # In either case, return without further processing - $text = $frame->virtualBracketedImplode( '{{', '|', '}}', $titleWithSpaces, $args ); - $isLocalObj = true; - $found = true; - } - } - - # Variables - if ( !$found && $args->getLength() == 0 ) { - $id = $this->mVariables->matchStartToEnd( $part1 ); - if ( $id !== false ) { - $text = $this->getVariableValue( $id ); - if (MagicWord::getCacheTTL($id)>-1) - $this->mOutput->mContainsOldMagic = true; - $found = true; - } - } - - # MSG, MSGNW and RAW - if ( !$found ) { - # Check for MSGNW: - $mwMsgnw =& MagicWord::get( 'msgnw' ); - if ( $mwMsgnw->matchStartAndRemove( $part1 ) ) { - $nowiki = true; - } else { - # Remove obsolete MSG: - $mwMsg =& MagicWord::get( 'msg' ); - $mwMsg->matchStartAndRemove( $part1 ); - } - - # Check for RAW: - $mwRaw =& MagicWord::get( 'raw' ); - if ( $mwRaw->matchStartAndRemove( $part1 ) ) { - $forceRawInterwiki = true; - } - } - wfProfileOut( __METHOD__.'-modifiers' ); - - # Parser functions - if ( !$found ) { - wfProfileIn( __METHOD__ . '-pfunc' ); - - $colonPos = strpos( $part1, ':' ); - if ( $colonPos !== false ) { - # Case sensitive functions - $function = substr( $part1, 0, $colonPos ); - if ( isset( $this->mFunctionSynonyms[1][$function] ) ) { - $function = $this->mFunctionSynonyms[1][$function]; - } else { - # Case insensitive functions - $function = strtolower( $function ); - if ( isset( $this->mFunctionSynonyms[0][$function] ) ) { - $function = $this->mFunctionSynonyms[0][$function]; - } else { - $function = false; - } - } - if ( $function ) { - list( $callback, $flags ) = $this->mFunctionHooks[$function]; - $initialArgs = array( &$this ); - $funcArgs = array( trim( substr( $part1, $colonPos + 1 ) ) ); - if ( $flags & SFH_OBJECT_ARGS ) { - # Add a frame parameter, and pass the arguments as an array - $allArgs = $initialArgs; - $allArgs[] = $frame; - for ( $i = 0; $i < $args->getLength(); $i++ ) { - $funcArgs[] = $args->item( $i ); - } - $allArgs[] = $funcArgs; - } else { - # Convert arguments to plain text - for ( $i = 0; $i < $args->getLength(); $i++ ) { - $funcArgs[] = trim( $frame->expand( $args->item( $i ) ) ); - } - $allArgs = array_merge( $initialArgs, $funcArgs ); - } - - # Workaround for PHP bug 35229 and similar - if ( !is_callable( $callback ) ) { - throw new MWException( "Tag hook for $name is not callable\n" ); - } - $result = call_user_func_array( $callback, $allArgs ); - $found = true; - - if ( is_array( $result ) ) { - if ( isset( $result[0] ) ) { - $text = $result[0]; - unset( $result[0] ); - } - - // Extract flags into the local scope - // This allows callers to set flags such as nowiki, found, etc. - extract( $result ); - } else { - $text = $result; - } - } - } - wfProfileOut( __METHOD__ . '-pfunc' ); - } - - # Finish mangling title and then check for loops. - # Set $title to a Title object and $titleText to the PDBK - if ( !$found ) { - $ns = NS_TEMPLATE; - # Split the title into page and subpage - $subpage = ''; - $part1 = $this->maybeDoSubpageLink( $part1, $subpage ); - if ($subpage !== '') { - $ns = $this->mTitle->getNamespace(); - } - $title = Title::newFromText( $part1, $ns ); - if ( $title ) { - $titleText = $title->getPrefixedText(); - # Check for language variants if the template is not found - if($wgContLang->hasVariants() && $title->getArticleID() == 0){ - $wgContLang->findVariantLink($part1, $title); - } - # Do infinite loop check - if ( !$frame->loopCheck( $title ) ) { - $found = true; - $text = "<span class=\"error\">Template loop detected: [[$titleText]]</span>"; - wfDebug( __METHOD__.": template loop broken at '$titleText'\n" ); - } - # Do recursion depth check - $limit = $this->mOptions->getMaxTemplateDepth(); - if ( $frame->depth >= $limit ) { - $found = true; - $text = "<span class=\"error\">Template recursion depth limit exceeded ($limit)</span>"; - } - } - } - - # Load from database - if ( !$found && $title ) { - wfProfileIn( __METHOD__ . '-loadtpl' ); - if ( !$title->isExternal() ) { - if ( $title->getNamespace() == NS_SPECIAL && $this->mOptions->getAllowSpecialInclusion() && $this->ot['html'] ) { - $text = SpecialPage::capturePath( $title ); - if ( is_string( $text ) ) { - $found = true; - $isHTML = true; - $this->disableCache(); - } - } else if ( $wgNonincludableNamespaces && in_array( $title->getNamespace(), $wgNonincludableNamespaces ) ) { - $found = false; //access denied - wfDebug( "$fname: template inclusion denied for " . $title->getPrefixedDBkey() ); - } else { - list( $text, $title ) = $this->getTemplateDom( $title ); - if ( $text !== false ) { - $found = true; - $isChildObj = true; - } - } - - # If the title is valid but undisplayable, make a link to it - if ( !$found && ( $this->ot['html'] || $this->ot['pre'] ) ) { - $text = "[[:$titleText]]"; - $found = true; - } - } elseif ( $title->isTrans() ) { - // Interwiki transclusion - if ( $this->ot['html'] && !$forceRawInterwiki ) { - $text = $this->interwikiTransclude( $title, 'render' ); - $isHTML = true; - } else { - $text = $this->interwikiTransclude( $title, 'raw' ); - // Preprocess it like a template - $text = $this->preprocessToDom( $text, self::PTD_FOR_INCLUSION ); - $isChildObj = true; - } - $found = true; - } - wfProfileOut( __METHOD__ . '-loadtpl' ); - } - - # If we haven't found text to substitute by now, we're done - # Recover the source wikitext and return it - if ( !$found ) { - $text = $frame->virtualBracketedImplode( '{{', '|', '}}', $titleWithSpaces, $args ); - wfProfileOut( $fname ); - return array( 'object' => $text ); - } - - # Expand DOM-style return values in a child frame - if ( $isChildObj ) { - # Clean up argument array - $newFrame = $frame->newChild( $args, $title ); - - if ( $nowiki ) { - $text = $newFrame->expand( $text, PPFrame::RECOVER_ORIG ); - } elseif ( $titleText !== false && $newFrame->isEmpty() ) { - # Expansion is eligible for the empty-frame cache - if ( isset( $this->mTplExpandCache[$titleText] ) ) { - $text = $this->mTplExpandCache[$titleText]; - } else { - $text = $newFrame->expand( $text ); - $this->mTplExpandCache[$titleText] = $text; - } - } else { - # Uncached expansion - $text = $newFrame->expand( $text ); - } - } - if ( $isLocalObj && $nowiki ) { - $text = $frame->expand( $text, PPFrame::RECOVER_ORIG ); - $isLocalObj = false; - } - - # Replace raw HTML by a placeholder - # Add a blank line preceding, to prevent it from mucking up - # immediately preceding headings - if ( $isHTML ) { - $text = "\n\n" . $this->insertStripItem( $text ); - } - # Escape nowiki-style return values - elseif ( $nowiki && ( $this->ot['html'] || $this->ot['pre'] ) ) { - $text = wfEscapeWikiText( $text ); - } - # Bug 529: if the template begins with a table or block-level - # element, it should be treated as beginning a new line. - # This behaviour is somewhat controversial. - elseif ( is_string( $text ) && !$piece['lineStart'] && preg_match('/^(?:{\\||:|;|#|\*)/', $text)) /*}*/{ - $text = "\n" . $text; - } - - if ( is_string( $text ) && !$this->incrementIncludeSize( 'post-expand', strlen( $text ) ) ) { - # Error, oversize inclusion - $text = "[[$originalTitle]]" . - $this->insertStripItem( '<!-- WARNING: template omitted, post-expand include size too large -->' ); - } - - if ( $isLocalObj ) { - $ret = array( 'object' => $text ); - } else { - $ret = array( 'text' => $text ); - } - - wfProfileOut( $fname ); - return $ret; - } - - /** - * Get the semi-parsed DOM representation of a template with a given title, - * and its redirect destination title. Cached. - */ - function getTemplateDom( $title ) { - $cacheTitle = $title; - $titleText = $title->getPrefixedDBkey(); - - if ( isset( $this->mTplRedirCache[$titleText] ) ) { - list( $ns, $dbk ) = $this->mTplRedirCache[$titleText]; - $title = Title::makeTitle( $ns, $dbk ); - $titleText = $title->getPrefixedDBkey(); - } - if ( isset( $this->mTplDomCache[$titleText] ) ) { - return array( $this->mTplDomCache[$titleText], $title ); - } - - // Cache miss, go to the database - list( $text, $title ) = $this->fetchTemplateAndTitle( $title ); - - if ( $text === false ) { - $this->mTplDomCache[$titleText] = false; - return array( false, $title ); - } - - $dom = $this->preprocessToDom( $text, self::PTD_FOR_INCLUSION ); - $this->mTplDomCache[ $titleText ] = $dom; - - if (! $title->equals($cacheTitle)) { - $this->mTplRedirCache[$cacheTitle->getPrefixedDBkey()] = - array( $title->getNamespace(),$cdb = $title->getDBkey() ); - } - - return array( $dom, $title ); - } - - /** - * Fetch the unparsed text of a template and register a reference to it. - */ - function fetchTemplateAndTitle( $title ) { - $templateCb = $this->mOptions->getTemplateCallback(); - $stuff = call_user_func( $templateCb, $title ); - $text = $stuff['text']; - $finalTitle = isset( $stuff['finalTitle'] ) ? $stuff['finalTitle'] : $title; - if ( isset( $stuff['deps'] ) ) { - foreach ( $stuff['deps'] as $dep ) { - $this->mOutput->addTemplate( $dep['title'], $dep['page_id'], $dep['rev_id'] ); - } - } - return array($text,$finalTitle); - } - - function fetchTemplate( $title ) { - $rv = $this->fetchTemplateAndTitle($title); - return $rv[0]; - } - - /** - * Static function to get a template - * Can be overridden via ParserOptions::setTemplateCallback(). - */ - static function statelessFetchTemplate( $title ) { - $text = $skip = false; - $finalTitle = $title; - $deps = array(); - - // Loop to fetch the article, with up to 1 redirect - for ( $i = 0; $i < 2 && is_object( $title ); $i++ ) { - # Give extensions a chance to select the revision instead - $id = false; // Assume current - wfRunHooks( 'BeforeParserFetchTemplateAndtitle', array( false, &$title, &$skip, &$id ) ); - - if( $skip ) { - $text = false; - $deps[] = array( - 'title' => $title, - 'page_id' => $title->getArticleID(), - 'rev_id' => null ); - break; - } - $rev = $id ? Revision::newFromId( $id ) : Revision::newFromTitle( $title ); - $rev_id = $rev ? $rev->getId() : 0; - - $deps[] = array( - 'title' => $title, - 'page_id' => $title->getArticleID(), - 'rev_id' => $rev_id ); - - if( $rev ) { - $text = $rev->getText(); - } elseif( $title->getNamespace() == NS_MEDIAWIKI ) { - global $wgLang; - $message = $wgLang->lcfirst( $title->getText() ); - $text = wfMsgForContentNoTrans( $message ); - if( wfEmptyMsg( $message, $text ) ) { - $text = false; - break; - } - } else { - break; - } - if ( $text === false ) { - break; - } - // Redirect? - $finalTitle = $title; - $title = Title::newFromRedirect( $text ); - } - return array( - 'text' => $text, - 'finalTitle' => $finalTitle, - 'deps' => $deps ); - } - - /** - * Transclude an interwiki link. - */ - function interwikiTransclude( $title, $action ) { - global $wgEnableScaryTranscluding; - - if (!$wgEnableScaryTranscluding) - return wfMsg('scarytranscludedisabled'); - - $url = $title->getFullUrl( "action=$action" ); - - if (strlen($url) > 255) - return wfMsg('scarytranscludetoolong'); - return $this->fetchScaryTemplateMaybeFromCache($url); - } - - function fetchScaryTemplateMaybeFromCache($url) { - global $wgTranscludeCacheExpiry; - $dbr = wfGetDB(DB_SLAVE); - $obj = $dbr->selectRow('transcache', array('tc_time', 'tc_contents'), - array('tc_url' => $url)); - if ($obj) { - $time = $obj->tc_time; - $text = $obj->tc_contents; - if ($time && time() < $time + $wgTranscludeCacheExpiry ) { - return $text; - } - } - - $text = Http::get($url); - if (!$text) - return wfMsg('scarytranscludefailed', $url); - - $dbw = wfGetDB(DB_MASTER); - $dbw->replace('transcache', array('tc_url'), array( - 'tc_url' => $url, - 'tc_time' => time(), - 'tc_contents' => $text)); - return $text; - } - - - /** - * Triple brace replacement -- used for template arguments - * @private - */ - function argSubstitution( $piece, $frame ) { - wfProfileIn( __METHOD__ ); - - $error = false; - $parts = $piece['parts']; - $nameWithSpaces = $frame->expand( $piece['title'] ); - $argName = trim( $nameWithSpaces ); - $object = false; - $text = $frame->getArgument( $argName ); - if ( $text === false && $parts->getLength() > 0 - && ( - $this->ot['html'] - || $this->ot['pre'] - || ( $this->ot['wiki'] && $frame->isTemplate() ) - ) - ) { - # No match in frame, use the supplied default - $object = $parts->item( 0 )->getChildren(); - } - if ( !$this->incrementIncludeSize( 'arg', strlen( $text ) ) ) { - $error = '<!-- WARNING: argument omitted, expansion size too large -->'; - } - - if ( $text === false && $object === false ) { - # No match anywhere - $object = $frame->virtualBracketedImplode( '{{{', '|', '}}}', $nameWithSpaces, $parts ); - } - if ( $error !== false ) { - $text .= $error; - } - if ( $object !== false ) { - $ret = array( 'object' => $object ); - } else { - $ret = array( 'text' => $text ); - } - - wfProfileOut( __METHOD__ ); - return $ret; - } - - /** - * Return the text to be used for a given extension tag. - * This is the ghost of strip(). - * - * @param array $params Associative array of parameters: - * name PPNode for the tag name - * attr PPNode for unparsed text where tag attributes are thought to be - * attributes Optional associative array of parsed attributes - * inner Contents of extension element - * noClose Original text did not have a close tag - * @param PPFrame $frame - */ - function extensionSubstitution( $params, $frame ) { - global $wgRawHtml, $wgContLang; - - $name = $frame->expand( $params['name'] ); - $attrText = !isset( $params['attr'] ) ? null : $frame->expand( $params['attr'] ); - $content = !isset( $params['inner'] ) ? null : $frame->expand( $params['inner'] ); - - $marker = "{$this->mUniqPrefix}-$name-" . sprintf('%08X', $this->mMarkerIndex++) . $this->mMarkerSuffix; - - if ( $this->ot['html'] ) { - $name = strtolower( $name ); - - $attributes = Sanitizer::decodeTagAttributes( $attrText ); - if ( isset( $params['attributes'] ) ) { - $attributes = $attributes + $params['attributes']; - } - switch ( $name ) { - case 'html': - if( $wgRawHtml ) { - $output = $content; - break; - } else { - throw new MWException( '<html> extension tag encountered unexpectedly' ); - } - case 'nowiki': - $output = Xml::escapeTagsOnly( $content ); - break; - case 'math': - $output = $wgContLang->armourMath( - MathRenderer::renderMath( $content, $attributes ) ); - break; - case 'gallery': - $output = $this->renderImageGallery( $content, $attributes ); - break; - default: - if( isset( $this->mTagHooks[$name] ) ) { - # Workaround for PHP bug 35229 and similar - if ( !is_callable( $this->mTagHooks[$name] ) ) { - throw new MWException( "Tag hook for $name is not callable\n" ); - } - $output = call_user_func_array( $this->mTagHooks[$name], - array( $content, $attributes, $this ) ); - } else { - throw new MWException( "Invalid call hook $name" ); - } - } - } else { - if ( is_null( $attrText ) ) { - $attrText = ''; - } - if ( isset( $params['attributes'] ) ) { - foreach ( $params['attributes'] as $attrName => $attrValue ) { - $attrText .= ' ' . htmlspecialchars( $attrName ) . '="' . - htmlspecialchars( $attrValue ) . '"'; - } - } - if ( $content === null ) { - $output = "<$name$attrText/>"; - } else { - $close = is_null( $params['close'] ) ? '' : $frame->expand( $params['close'] ); - $output = "<$name$attrText>$content$close"; - } - } - - if ( $name == 'html' || $name == 'nowiki' ) { - $this->mStripState->nowiki->setPair( $marker, $output ); - } else { - $this->mStripState->general->setPair( $marker, $output ); - } - return $marker; - } - - /** - * Increment an include size counter - * - * @param string $type The type of expansion - * @param integer $size The size of the text - * @return boolean False if this inclusion would take it over the maximum, true otherwise - */ - function incrementIncludeSize( $type, $size ) { - if ( $this->mIncludeSizes[$type] + $size > $this->mOptions->getMaxIncludeSize( $type ) ) { - return false; - } else { - $this->mIncludeSizes[$type] += $size; - return true; - } - } - - /** - * Detect __NOGALLERY__ magic word and set a placeholder - */ - function stripNoGallery( &$text ) { - # if the string __NOGALLERY__ (not case-sensitive) occurs in the HTML, - # do not add TOC - $mw = MagicWord::get( 'nogallery' ); - $this->mOutput->mNoGallery = $mw->matchAndRemove( $text ) ; - } - - /** - * Find the first __TOC__ magic word and set a <!--MWTOC--> - * placeholder that will then be replaced by the real TOC in - * ->formatHeadings, this works because at this points real - * comments will have already been discarded by the sanitizer. - * - * Any additional __TOC__ magic words left over will be discarded - * as there can only be one TOC on the page. - */ - function stripToc( $text ) { - # if the string __NOTOC__ (not case-sensitive) occurs in the HTML, - # do not add TOC - $mw = MagicWord::get( 'notoc' ); - if( $mw->matchAndRemove( $text ) ) { - $this->mShowToc = false; - } - - $mw = MagicWord::get( 'toc' ); - if( $mw->match( $text ) ) { - $this->mShowToc = true; - $this->mForceTocPosition = true; - - // Set a placeholder. At the end we'll fill it in with the TOC. - $text = $mw->replace( '<!--MWTOC-->', $text, 1 ); - - // Only keep the first one. - $text = $mw->replace( '', $text ); - } - return $text; - } - - /** - * This function accomplishes several tasks: - * 1) Auto-number headings if that option is enabled - * 2) Add an [edit] link to sections for users who have enabled the option and can edit the page - * 3) Add a Table of contents on the top for users who have enabled the option - * 4) Auto-anchor headings - * - * It loops through all headlines, collects the necessary data, then splits up the - * string and re-inserts the newly formatted headlines. - * - * @param string $text - * @param boolean $isMain - * @private - */ - function formatHeadings( $text, $isMain=true ) { - global $wgMaxTocLevel, $wgContLang; - - $doNumberHeadings = $this->mOptions->getNumberHeadings(); - if( !$this->mTitle->quickUserCan( 'edit' ) ) { - $showEditLink = 0; - } else { - $showEditLink = $this->mOptions->getEditSection(); - } - - # Inhibit editsection links if requested in the page - $esw =& MagicWord::get( 'noeditsection' ); - if( $esw->matchAndRemove( $text ) ) { - $showEditLink = 0; - } - - # Get all headlines for numbering them and adding funky stuff like [edit] - # links - this is for later, but we need the number of headlines right now - $matches = array(); - $numMatches = preg_match_all( '/<H(?P<level>[1-6])(?P<attrib>.*?'.'>)(?P<header>.*?)<\/H[1-6] *>/i', $text, $matches ); - - # if there are fewer than 4 headlines in the article, do not show TOC - # unless it's been explicitly enabled. - $enoughToc = $this->mShowToc && - (($numMatches >= 4) || $this->mForceTocPosition); - - # Allow user to stipulate that a page should have a "new section" - # link added via __NEWSECTIONLINK__ - $mw =& MagicWord::get( 'newsectionlink' ); - if( $mw->matchAndRemove( $text ) ) - $this->mOutput->setNewSection( true ); - - # if the string __FORCETOC__ (not case-sensitive) occurs in the HTML, - # override above conditions and always show TOC above first header - $mw =& MagicWord::get( 'forcetoc' ); - if ($mw->matchAndRemove( $text ) ) { - $this->mShowToc = true; - $enoughToc = true; - } - - # We need this to perform operations on the HTML - $sk = $this->mOptions->getSkin(); - - # headline counter - $headlineCount = 0; - $numVisible = 0; - - # Ugh .. the TOC should have neat indentation levels which can be - # passed to the skin functions. These are determined here - $toc = ''; - $full = ''; - $head = array(); - $sublevelCount = array(); - $levelCount = array(); - $toclevel = 0; - $level = 0; - $prevlevel = 0; - $toclevel = 0; - $prevtoclevel = 0; - $markerRegex = "{$this->mUniqPrefix}-h-(\d+)-{$this->mMarkerSuffix}"; - $baseTitleText = $this->mTitle->getPrefixedDBkey(); - $tocraw = array(); - - foreach( $matches[3] as $headline ) { - $isTemplate = false; - $titleText = false; - $sectionIndex = false; - $numbering = ''; - $markerMatches = array(); - if (preg_match("/^$markerRegex/", $headline, $markerMatches)) { - $serial = $markerMatches[1]; - list( $titleText, $sectionIndex ) = $this->mHeadings[$serial]; - $isTemplate = ($titleText != $baseTitleText); - $headline = preg_replace("/^$markerRegex/", "", $headline); - } - - if( $toclevel ) { - $prevlevel = $level; - $prevtoclevel = $toclevel; - } - $level = $matches[1][$headlineCount]; - - if( $doNumberHeadings || $enoughToc ) { - - if ( $level > $prevlevel ) { - # Increase TOC level - $toclevel++; - $sublevelCount[$toclevel] = 0; - if( $toclevel<$wgMaxTocLevel ) { - $prevtoclevel = $toclevel; - $toc .= $sk->tocIndent(); - $numVisible++; - } - } - elseif ( $level < $prevlevel && $toclevel > 1 ) { - # Decrease TOC level, find level to jump to - - if ( $toclevel == 2 && $level <= $levelCount[1] ) { - # Can only go down to level 1 - $toclevel = 1; - } else { - for ($i = $toclevel; $i > 0; $i--) { - if ( $levelCount[$i] == $level ) { - # Found last matching level - $toclevel = $i; - break; - } - elseif ( $levelCount[$i] < $level ) { - # Found first matching level below current level - $toclevel = $i + 1; - break; - } - } - } - if( $toclevel<$wgMaxTocLevel ) { - if($prevtoclevel < $wgMaxTocLevel) { - # Unindent only if the previous toc level was shown :p - $toc .= $sk->tocUnindent( $prevtoclevel - $toclevel ); - } else { - $toc .= $sk->tocLineEnd(); - } - } - } - else { - # No change in level, end TOC line - if( $toclevel<$wgMaxTocLevel ) { - $toc .= $sk->tocLineEnd(); - } - } - - $levelCount[$toclevel] = $level; - - # count number of headlines for each level - @$sublevelCount[$toclevel]++; - $dot = 0; - for( $i = 1; $i <= $toclevel; $i++ ) { - if( !empty( $sublevelCount[$i] ) ) { - if( $dot ) { - $numbering .= '.'; - } - $numbering .= $wgContLang->formatNum( $sublevelCount[$i] ); - $dot = 1; - } - } - } - - # The safe header is a version of the header text safe to use for links - # Avoid insertion of weird stuff like <math> by expanding the relevant sections - $safeHeadline = $this->mStripState->unstripBoth( $headline ); - - # Remove link placeholders by the link text. - # <!--LINK number--> - # turns into - # link text with suffix - $safeHeadline = preg_replace( '/<!--LINK ([0-9]*)-->/e', - "\$this->mLinkHolders['texts'][\$1]", - $safeHeadline ); - $safeHeadline = preg_replace( '/<!--IWLINK ([0-9]*)-->/e', - "\$this->mInterwikiLinkHolders['texts'][\$1]", - $safeHeadline ); - - # Strip out HTML (other than plain <sup> and <sub>: bug 8393) - $tocline = preg_replace( - array( '#<(?!/?(sup|sub)).*?'.'>#', '#<(/?(sup|sub)).*?'.'>#' ), - array( '', '<$1>'), - $safeHeadline - ); - $tocline = trim( $tocline ); - - # For the anchor, strip out HTML-y stuff period - $safeHeadline = preg_replace( '/<.*?'.'>/', '', $safeHeadline ); - $safeHeadline = trim( $safeHeadline ); - - # Save headline for section edit hint before it's escaped - $headlineHint = $safeHeadline; - $safeHeadline = Sanitizer::escapeId( $safeHeadline ); - $refers[$headlineCount] = $safeHeadline; - - # count how many in assoc. array so we can track dupes in anchors - isset( $refers[$safeHeadline] ) ? $refers[$safeHeadline]++ : $refers[$safeHeadline] = 1; - $refcount[$headlineCount] = $refers[$safeHeadline]; - - # Don't number the heading if it is the only one (looks silly) - if( $doNumberHeadings && count( $matches[3] ) > 1) { - # the two are different if the line contains a link - $headline=$numbering . ' ' . $headline; - } - - # Create the anchor for linking from the TOC to the section - $anchor = $safeHeadline; - if($refcount[$headlineCount] > 1 ) { - $anchor .= '_' . $refcount[$headlineCount]; - } - if( $enoughToc && ( !isset($wgMaxTocLevel) || $toclevel<$wgMaxTocLevel ) ) { - $toc .= $sk->tocLine($anchor, $tocline, $numbering, $toclevel); - $tocraw[] = array( 'toclevel' => $toclevel, 'level' => $level, 'line' => $tocline, 'number' => $numbering ); - } - # give headline the correct <h#> tag - if( $showEditLink && $sectionIndex !== false ) { - if( $isTemplate ) { - # Put a T flag in the section identifier, to indicate to extractSections() - # that sections inside <includeonly> should be counted. - $editlink = $sk->editSectionLinkForOther($titleText, "T-$sectionIndex"); - } else { - $editlink = $sk->editSectionLink($this->mTitle, $sectionIndex, $headlineHint); - } - } else { - $editlink = ''; - } - $head[$headlineCount] = $sk->makeHeadline( $level, $matches['attrib'][$headlineCount], $anchor, $headline, $editlink ); - - $headlineCount++; - } - - $this->mOutput->setSections( $tocraw ); - - # Never ever show TOC if no headers - if( $numVisible < 1 ) { - $enoughToc = false; - } - - if( $enoughToc ) { - if( $prevtoclevel > 0 && $prevtoclevel < $wgMaxTocLevel ) { - $toc .= $sk->tocUnindent( $prevtoclevel - 1 ); - } - $toc = $sk->tocList( $toc ); - } - - # split up and insert constructed headlines - - $blocks = preg_split( '/<H[1-6].*?' . '>.*?<\/H[1-6]>/i', $text ); - $i = 0; - - foreach( $blocks as $block ) { - if( $showEditLink && $headlineCount > 0 && $i == 0 && $block != "\n" ) { - # This is the [edit] link that appears for the top block of text when - # section editing is enabled - - # Disabled because it broke block formatting - # For example, a bullet point in the top line - # $full .= $sk->editSectionLink(0); - } - $full .= $block; - if( $enoughToc && !$i && $isMain && !$this->mForceTocPosition ) { - # Top anchor now in skin - $full = $full.$toc; - } - - if( !empty( $head[$i] ) ) { - $full .= $head[$i]; - } - $i++; - } - if( $this->mForceTocPosition ) { - return str_replace( '<!--MWTOC-->', $toc, $full ); - } else { - return $full; - } - } - - /** - * Transform wiki markup when saving a page by doing \r\n -> \n - * conversion, substitting signatures, {{subst:}} templates, etc. - * - * @param string $text the text to transform - * @param Title &$title the Title object for the current article - * @param User &$user the User object describing the current user - * @param ParserOptions $options parsing options - * @param bool $clearState whether to clear the parser state first - * @return string the altered wiki markup - * @public - */ - function preSaveTransform( $text, &$title, $user, $options, $clearState = true ) { - $this->mOptions = $options; - $this->setTitle( $title ); - $this->setOutputType( self::OT_WIKI ); - - if ( $clearState ) { - $this->clearState(); - } - - $pairs = array( - "\r\n" => "\n", - ); - $text = str_replace( array_keys( $pairs ), array_values( $pairs ), $text ); - $text = $this->pstPass2( $text, $user ); - $text = $this->mStripState->unstripBoth( $text ); - return $text; - } - - /** - * Pre-save transform helper function - * @private - */ - function pstPass2( $text, $user ) { - global $wgContLang, $wgLocaltimezone; - - /* Note: This is the timestamp saved as hardcoded wikitext to - * the database, we use $wgContLang here in order to give - * everyone the same signature and use the default one rather - * than the one selected in each user's preferences. - * - * (see also bug 12815) - */ - $ts = $this->mOptions->getTimestamp(); - $tz = 'UTC'; - if ( isset( $wgLocaltimezone ) ) { - $unixts = wfTimestamp( TS_UNIX, $ts ); - $oldtz = getenv( 'TZ' ); - putenv( 'TZ='.$wgLocaltimezone ); - $ts = date( 'YmdHis', $unixts ); - $tz = date( 'T', $unixts ); # might vary on DST changeover! - putenv( 'TZ='.$oldtz ); - } - $d = $wgContLang->timeanddate( $ts, false, false ) . " ($tz)"; - - # Variable replacement - # Because mOutputType is OT_WIKI, this will only process {{subst:xxx}} type tags - $text = $this->replaceVariables( $text ); - - # Signatures - $sigText = $this->getUserSig( $user ); - $text = strtr( $text, array( - '~~~~~' => $d, - '~~~~' => "$sigText $d", - '~~~' => $sigText - ) ); - - # Context links: [[|name]] and [[name (context)|]] - # - global $wgLegalTitleChars; - $tc = "[$wgLegalTitleChars]"; - $nc = '[ _0-9A-Za-z\x80-\xff]'; # Namespaces can use non-ascii! - - $p1 = "/\[\[(:?$nc+:|:|)($tc+?)( \\($tc+\\))\\|]]/"; # [[ns:page (context)|]] - $p3 = "/\[\[(:?$nc+:|:|)($tc+?)( \\($tc+\\)|)(, $tc+|)\\|]]/"; # [[ns:page (context), context|]] - $p2 = "/\[\[\\|($tc+)]]/"; # [[|page]] - - # try $p1 first, to turn "[[A, B (C)|]]" into "[[A, B (C)|A, B]]" - $text = preg_replace( $p1, '[[\\1\\2\\3|\\2]]', $text ); - $text = preg_replace( $p3, '[[\\1\\2\\3\\4|\\2]]', $text ); - - $t = $this->mTitle->getText(); - $m = array(); - if ( preg_match( "/^($nc+:|)$tc+?( \\($tc+\\))$/", $t, $m ) ) { - $text = preg_replace( $p2, "[[$m[1]\\1$m[2]|\\1]]", $text ); - } elseif ( preg_match( "/^($nc+:|)$tc+?(, $tc+|)$/", $t, $m ) && '' != "$m[1]$m[2]" ) { - $text = preg_replace( $p2, "[[$m[1]\\1$m[2]|\\1]]", $text ); - } else { - # if there's no context, don't bother duplicating the title - $text = preg_replace( $p2, '[[\\1]]', $text ); - } - - # Trim trailing whitespace - $text = rtrim( $text ); - - return $text; - } - - /** - * Fetch the user's signature text, if any, and normalize to - * validated, ready-to-insert wikitext. - * - * @param User $user - * @return string - * @private - */ - function getUserSig( &$user ) { - global $wgMaxSigChars; - - $username = $user->getName(); - $nickname = $user->getOption( 'nickname' ); - $nickname = $nickname === '' ? $username : $nickname; - - if( mb_strlen( $nickname ) > $wgMaxSigChars ) { - $nickname = $username; - wfDebug( __METHOD__ . ": $username has overlong signature.\n" ); - } elseif( $user->getBoolOption( 'fancysig' ) !== false ) { - # Sig. might contain markup; validate this - if( $this->validateSig( $nickname ) !== false ) { - # Validated; clean up (if needed) and return it - return $this->cleanSig( $nickname, true ); - } else { - # Failed to validate; fall back to the default - $nickname = $username; - wfDebug( "Parser::getUserSig: $username has bad XML tags in signature.\n" ); - } - } - - // Make sure nickname doesnt get a sig in a sig - $nickname = $this->cleanSigInSig( $nickname ); - - # If we're still here, make it a link to the user page - $userText = wfEscapeWikiText( $username ); - $nickText = wfEscapeWikiText( $nickname ); - if ( $user->isAnon() ) { - return wfMsgExt( 'signature-anon', array( 'content', 'parsemag' ), $userText, $nickText ); - } else { - return wfMsgExt( 'signature', array( 'content', 'parsemag' ), $userText, $nickText ); - } - } - - /** - * Check that the user's signature contains no bad XML - * - * @param string $text - * @return mixed An expanded string, or false if invalid. - */ - function validateSig( $text ) { - return( wfIsWellFormedXmlFragment( $text ) ? $text : false ); - } - - /** - * Clean up signature text - * - * 1) Strip ~~~, ~~~~ and ~~~~~ out of signatures @see cleanSigInSig - * 2) Substitute all transclusions - * - * @param string $text - * @param $parsing Whether we're cleaning (preferences save) or parsing - * @return string Signature text - */ - function cleanSig( $text, $parsing = false ) { - if ( !$parsing ) { - global $wgTitle; - $this->clearState(); - $this->setTitle( $wgTitle ); - $this->mOptions = new ParserOptions; - $this->setOutputType = self::OT_PREPROCESS; - } - - # FIXME: regex doesn't respect extension tags or nowiki - # => Move this logic to braceSubstitution() - $substWord = MagicWord::get( 'subst' ); - $substRegex = '/\{\{(?!(?:' . $substWord->getBaseRegex() . '))/x' . $substWord->getRegexCase(); - $substText = '{{' . $substWord->getSynonym( 0 ); - - $text = preg_replace( $substRegex, $substText, $text ); - $text = $this->cleanSigInSig( $text ); - $dom = $this->preprocessToDom( $text ); - $frame = $this->getPreprocessor()->newFrame(); - $text = $frame->expand( $dom ); - - if ( !$parsing ) { - $text = $this->mStripState->unstripBoth( $text ); - } - - return $text; - } - - /** - * Strip ~~~, ~~~~ and ~~~~~ out of signatures - * @param string $text - * @return string Signature text with /~{3,5}/ removed - */ - function cleanSigInSig( $text ) { - $text = preg_replace( '/~{3,5}/', '', $text ); - return $text; - } - - /** - * Set up some variables which are usually set up in parse() - * so that an external function can call some class members with confidence - * @public - */ - function startExternalParse( &$title, $options, $outputType, $clearState = true ) { - $this->setTitle( $title ); - $this->mOptions = $options; - $this->setOutputType( $outputType ); - if ( $clearState ) { - $this->clearState(); - } - } - - /** - * Wrapper for preprocess() - * - * @param string $text the text to preprocess - * @param ParserOptions $options options - * @return string - * @public - */ - function transformMsg( $text, $options ) { - global $wgTitle; - static $executing = false; - - $fname = "Parser::transformMsg"; - - # Guard against infinite recursion - if ( $executing ) { - return $text; - } - $executing = true; - - wfProfileIn($fname); - $text = $this->preprocess( $text, $wgTitle, $options ); - - $executing = false; - wfProfileOut($fname); - return $text; - } - - /** - * Create an HTML-style tag, e.g. <yourtag>special text</yourtag> - * The callback should have the following form: - * function myParserHook( $text, $params, &$parser ) { ... } - * - * Transform and return $text. Use $parser for any required context, e.g. use - * $parser->getTitle() and $parser->getOptions() not $wgTitle or $wgOut->mParserOptions - * - * @public - * - * @param mixed $tag The tag to use, e.g. 'hook' for <hook> - * @param mixed $callback The callback function (and object) to use for the tag - * - * @return The old value of the mTagHooks array associated with the hook - */ - function setHook( $tag, $callback ) { - $tag = strtolower( $tag ); - $oldVal = isset( $this->mTagHooks[$tag] ) ? $this->mTagHooks[$tag] : null; - $this->mTagHooks[$tag] = $callback; - $this->mStripList[] = $tag; - - return $oldVal; - } - - function setTransparentTagHook( $tag, $callback ) { - $tag = strtolower( $tag ); - $oldVal = isset( $this->mTransparentTagHooks[$tag] ) ? $this->mTransparentTagHooks[$tag] : null; - $this->mTransparentTagHooks[$tag] = $callback; - - return $oldVal; - } - - /** - * Remove all tag hooks - */ - function clearTagHooks() { - $this->mTagHooks = array(); - $this->mStripList = $this->mDefaultStripList; - } - - /** - * Create a function, e.g. {{sum:1|2|3}} - * The callback function should have the form: - * function myParserFunction( &$parser, $arg1, $arg2, $arg3 ) { ... } - * - * The callback may either return the text result of the function, or an array with the text - * in element 0, and a number of flags in the other elements. The names of the flags are - * specified in the keys. Valid flags are: - * found The text returned is valid, stop processing the template. This - * is on by default. - * nowiki Wiki markup in the return value should be escaped - * isHTML The returned text is HTML, armour it against wikitext transformation - * - * @public - * - * @param string $id The magic word ID - * @param mixed $callback The callback function (and object) to use - * @param integer $flags a combination of the following flags: - * SFH_NO_HASH No leading hash, i.e. {{plural:...}} instead of {{#if:...}} - * - * @return The old callback function for this name, if any - */ - function setFunctionHook( $id, $callback, $flags = 0 ) { - $oldVal = isset( $this->mFunctionHooks[$id] ) ? $this->mFunctionHooks[$id][0] : null; - $this->mFunctionHooks[$id] = array( $callback, $flags ); - - # Add to function cache - $mw = MagicWord::get( $id ); - if( !$mw ) - throw new MWException( 'Parser::setFunctionHook() expecting a magic word identifier.' ); - - $synonyms = $mw->getSynonyms(); - $sensitive = intval( $mw->isCaseSensitive() ); - - foreach ( $synonyms as $syn ) { - # Case - if ( !$sensitive ) { - $syn = strtolower( $syn ); - } - # Add leading hash - if ( !( $flags & SFH_NO_HASH ) ) { - $syn = '#' . $syn; - } - # Remove trailing colon - if ( substr( $syn, -1, 1 ) == ':' ) { - $syn = substr( $syn, 0, -1 ); - } - $this->mFunctionSynonyms[$sensitive][$syn] = $id; - } - return $oldVal; - } - - /** - * Get all registered function hook identifiers - * - * @return array - */ - function getFunctionHooks() { - return array_keys( $this->mFunctionHooks ); - } - - /** - * Replace <!--LINK--> link placeholders with actual links, in the buffer - * Placeholders created in Skin::makeLinkObj() - * Returns an array of link CSS classes, indexed by PDBK. - * $options is a bit field, RLH_FOR_UPDATE to select for update - */ - function replaceLinkHolders( &$text, $options = 0 ) { - global $wgUser; - global $wgContLang; - - $fname = 'Parser::replaceLinkHolders'; - wfProfileIn( $fname ); - - $pdbks = array(); - $colours = array(); - $linkcolour_ids = array(); - $sk = $this->mOptions->getSkin(); - $linkCache =& LinkCache::singleton(); - - if ( !empty( $this->mLinkHolders['namespaces'] ) ) { - wfProfileIn( $fname.'-check' ); - $dbr = wfGetDB( DB_SLAVE ); - $page = $dbr->tableName( 'page' ); - $threshold = $wgUser->getOption('stubthreshold'); - - # Sort by namespace - asort( $this->mLinkHolders['namespaces'] ); - - # Generate query - $query = false; - $current = null; - foreach ( $this->mLinkHolders['namespaces'] as $key => $ns ) { - # Make title object - $title = $this->mLinkHolders['titles'][$key]; - - # Skip invalid entries. - # Result will be ugly, but prevents crash. - if ( is_null( $title ) ) { - continue; - } - $pdbk = $pdbks[$key] = $title->getPrefixedDBkey(); - - # Check if it's a static known link, e.g. interwiki - if ( $title->isAlwaysKnown() ) { - $colours[$pdbk] = ''; - } elseif ( ( $id = $linkCache->getGoodLinkID( $pdbk ) ) != 0 ) { - $colours[$pdbk] = ''; - $this->mOutput->addLink( $title, $id ); - } elseif ( $linkCache->isBadLink( $pdbk ) ) { - $colours[$pdbk] = 'new'; - } elseif ( $title->getNamespace() == NS_SPECIAL && !SpecialPage::exists( $pdbk ) ) { - $colours[$pdbk] = 'new'; - } else { - # Not in the link cache, add it to the query - if ( !isset( $current ) ) { - $current = $ns; - $query = "SELECT page_id, page_namespace, page_title, page_is_redirect"; - if ( $threshold > 0 ) { - $query .= ', page_len'; - } - $query .= " FROM $page WHERE (page_namespace=$ns AND page_title IN("; - } elseif ( $current != $ns ) { - $current = $ns; - $query .= ")) OR (page_namespace=$ns AND page_title IN("; - } else { - $query .= ', '; - } - - $query .= $dbr->addQuotes( $this->mLinkHolders['dbkeys'][$key] ); - } - } - if ( $query ) { - $query .= '))'; - if ( $options & RLH_FOR_UPDATE ) { - $query .= ' FOR UPDATE'; - } - - $res = $dbr->query( $query, $fname ); - - # Fetch data and form into an associative array - # non-existent = broken - while ( $s = $dbr->fetchObject($res) ) { - $title = Title::makeTitle( $s->page_namespace, $s->page_title ); - $pdbk = $title->getPrefixedDBkey(); - $linkCache->addGoodLinkObj( $s->page_id, $title ); - $this->mOutput->addLink( $title, $s->page_id ); - $colours[$pdbk] = $sk->getLinkColour( $s, $threshold ); - //add id to the extension todolist - $linkcolour_ids[$s->page_id] = $pdbk; - } - //pass an array of page_ids to an extension - wfRunHooks( 'GetLinkColours', array( $linkcolour_ids, &$colours ) ); - } - wfProfileOut( $fname.'-check' ); - - # Do a second query for different language variants of links and categories - if($wgContLang->hasVariants()){ - $linkBatch = new LinkBatch(); - $variantMap = array(); // maps $pdbkey_Variant => $keys (of link holders) - $categoryMap = array(); // maps $category_variant => $category (dbkeys) - $varCategories = array(); // category replacements oldDBkey => newDBkey - - $categories = $this->mOutput->getCategoryLinks(); - - // Add variants of links to link batch - foreach ( $this->mLinkHolders['namespaces'] as $key => $ns ) { - $title = $this->mLinkHolders['titles'][$key]; - if ( is_null( $title ) ) - continue; - - $pdbk = $title->getPrefixedDBkey(); - $titleText = $title->getText(); - - // generate all variants of the link title text - $allTextVariants = $wgContLang->convertLinkToAllVariants($titleText); - - // if link was not found (in first query), add all variants to query - if ( !isset($colours[$pdbk]) ){ - foreach($allTextVariants as $textVariant){ - if($textVariant != $titleText){ - $variantTitle = Title::makeTitle( $ns, $textVariant ); - if(is_null($variantTitle)) continue; - $linkBatch->addObj( $variantTitle ); - $variantMap[$variantTitle->getPrefixedDBkey()][] = $key; - } - } - } - } - - // process categories, check if a category exists in some variant - foreach( $categories as $category ){ - $variants = $wgContLang->convertLinkToAllVariants($category); - foreach($variants as $variant){ - if($variant != $category){ - $variantTitle = Title::newFromDBkey( Title::makeName(NS_CATEGORY,$variant) ); - if(is_null($variantTitle)) continue; - $linkBatch->addObj( $variantTitle ); - $categoryMap[$variant] = $category; - } - } - } - - - if(!$linkBatch->isEmpty()){ - // construct query - $titleClause = $linkBatch->constructSet('page', $dbr); - - $variantQuery = "SELECT page_id, page_namespace, page_title, page_is_redirect"; - if ( $threshold > 0 ) { - $variantQuery .= ', page_len'; - } - - $variantQuery .= " FROM $page WHERE $titleClause"; - if ( $options & RLH_FOR_UPDATE ) { - $variantQuery .= ' FOR UPDATE'; - } - - $varRes = $dbr->query( $variantQuery, $fname ); - - // for each found variants, figure out link holders and replace - while ( $s = $dbr->fetchObject($varRes) ) { - - $variantTitle = Title::makeTitle( $s->page_namespace, $s->page_title ); - $varPdbk = $variantTitle->getPrefixedDBkey(); - $vardbk = $variantTitle->getDBkey(); - - $holderKeys = array(); - if(isset($variantMap[$varPdbk])){ - $holderKeys = $variantMap[$varPdbk]; - $linkCache->addGoodLinkObj( $s->page_id, $variantTitle ); - $this->mOutput->addLink( $variantTitle, $s->page_id ); - } - - // loop over link holders - foreach($holderKeys as $key){ - $title = $this->mLinkHolders['titles'][$key]; - if ( is_null( $title ) ) continue; - - $pdbk = $title->getPrefixedDBkey(); - - if(!isset($colours[$pdbk])){ - // found link in some of the variants, replace the link holder data - $this->mLinkHolders['titles'][$key] = $variantTitle; - $this->mLinkHolders['dbkeys'][$key] = $variantTitle->getDBkey(); - - // set pdbk and colour - $pdbks[$key] = $varPdbk; - $colours[$varPdbk] = $sk->getLinkColour( $s, $threshold ); - $linkcolour_ids[$s->page_id] = $pdbk; - } - wfRunHooks( 'GetLinkColours', array( $linkcolour_ids, &$colours ) ); - } - - // check if the object is a variant of a category - if(isset($categoryMap[$vardbk])){ - $oldkey = $categoryMap[$vardbk]; - if($oldkey != $vardbk) - $varCategories[$oldkey]=$vardbk; - } - } - - // rebuild the categories in original order (if there are replacements) - if(count($varCategories)>0){ - $newCats = array(); - $originalCats = $this->mOutput->getCategories(); - foreach($originalCats as $cat => $sortkey){ - // make the replacement - if( array_key_exists($cat,$varCategories) ) - $newCats[$varCategories[$cat]] = $sortkey; - else $newCats[$cat] = $sortkey; - } - $this->mOutput->setCategoryLinks($newCats); - } - } - } - - # Construct search and replace arrays - wfProfileIn( $fname.'-construct' ); - $replacePairs = array(); - foreach ( $this->mLinkHolders['namespaces'] as $key => $ns ) { - $pdbk = $pdbks[$key]; - $searchkey = "<!--LINK $key-->"; - $title = $this->mLinkHolders['titles'][$key]; - if ( !isset( $colours[$pdbk] ) || $colours[$pdbk] == 'new' ) { - $linkCache->addBadLinkObj( $title ); - $colours[$pdbk] = 'new'; - $this->mOutput->addLink( $title, 0 ); - $replacePairs[$searchkey] = $sk->makeBrokenLinkObj( $title, - $this->mLinkHolders['texts'][$key], - $this->mLinkHolders['queries'][$key] ); - } else { - $replacePairs[$searchkey] = $sk->makeColouredLinkObj( $title, $colours[$pdbk], - $this->mLinkHolders['texts'][$key], - $this->mLinkHolders['queries'][$key] ); - } - } - $replacer = new HashtableReplacer( $replacePairs, 1 ); - wfProfileOut( $fname.'-construct' ); - - # Do the thing - wfProfileIn( $fname.'-replace' ); - $text = preg_replace_callback( - '/(<!--LINK .*?-->)/', - $replacer->cb(), - $text); - - wfProfileOut( $fname.'-replace' ); - } - - # Now process interwiki link holders - # This is quite a bit simpler than internal links - if ( !empty( $this->mInterwikiLinkHolders['texts'] ) ) { - wfProfileIn( $fname.'-interwiki' ); - # Make interwiki link HTML - $replacePairs = array(); - foreach( $this->mInterwikiLinkHolders['texts'] as $key => $link ) { - $title = $this->mInterwikiLinkHolders['titles'][$key]; - $replacePairs[$key] = $sk->makeLinkObj( $title, $link ); - } - $replacer = new HashtableReplacer( $replacePairs, 1 ); - - $text = preg_replace_callback( - '/<!--IWLINK (.*?)-->/', - $replacer->cb(), - $text ); - wfProfileOut( $fname.'-interwiki' ); - } - - wfProfileOut( $fname ); - return $colours; - } - - /** - * Replace <!--LINK--> link placeholders with plain text of links - * (not HTML-formatted). - * @param string $text - * @return string - */ - function replaceLinkHoldersText( $text ) { - $fname = 'Parser::replaceLinkHoldersText'; - wfProfileIn( $fname ); - - $text = preg_replace_callback( - '/<!--(LINK|IWLINK) (.*?)-->/', - array( &$this, 'replaceLinkHoldersTextCallback' ), - $text ); - - wfProfileOut( $fname ); - return $text; - } - - /** - * @param array $matches - * @return string - * @private - */ - function replaceLinkHoldersTextCallback( $matches ) { - $type = $matches[1]; - $key = $matches[2]; - if( $type == 'LINK' ) { - if( isset( $this->mLinkHolders['texts'][$key] ) ) { - return $this->mLinkHolders['texts'][$key]; - } - } elseif( $type == 'IWLINK' ) { - if( isset( $this->mInterwikiLinkHolders['texts'][$key] ) ) { - return $this->mInterwikiLinkHolders['texts'][$key]; - } - } - return $matches[0]; - } - - /** - * Tag hook handler for 'pre'. - */ - function renderPreTag( $text, $attribs ) { - // Backwards-compatibility hack - $content = StringUtils::delimiterReplace( '<nowiki>', '</nowiki>', '$1', $text, 'i' ); - - $attribs = Sanitizer::validateTagAttributes( $attribs, 'pre' ); - return wfOpenElement( 'pre', $attribs ) . - Xml::escapeTagsOnly( $content ) . - '</pre>'; - } - - /** - * Renders an image gallery from a text with one line per image. - * text labels may be given by using |-style alternative text. E.g. - * Image:one.jpg|The number "1" - * Image:tree.jpg|A tree - * given as text will return the HTML of a gallery with two images, - * labeled 'The number "1"' and - * 'A tree'. - */ - function renderImageGallery( $text, $params ) { - $ig = new ImageGallery(); - $ig->setContextTitle( $this->mTitle ); - $ig->setShowBytes( false ); - $ig->setShowFilename( false ); - $ig->setParser( $this ); - $ig->setHideBadImages(); - $ig->setAttributes( Sanitizer::validateTagAttributes( $params, 'table' ) ); - $ig->useSkin( $this->mOptions->getSkin() ); - $ig->mRevisionId = $this->mRevisionId; - - if( isset( $params['caption'] ) ) { - $caption = $params['caption']; - $caption = htmlspecialchars( $caption ); - $caption = $this->replaceInternalLinks( $caption ); - $ig->setCaptionHtml( $caption ); - } - if( isset( $params['perrow'] ) ) { - $ig->setPerRow( $params['perrow'] ); - } - if( isset( $params['widths'] ) ) { - $ig->setWidths( $params['widths'] ); - } - if( isset( $params['heights'] ) ) { - $ig->setHeights( $params['heights'] ); - } - - wfRunHooks( 'BeforeParserrenderImageGallery', array( &$this, &$ig ) ); - - $lines = explode( "\n", $text ); - foreach ( $lines as $line ) { - # match lines like these: - # Image:someimage.jpg|This is some image - $matches = array(); - preg_match( "/^([^|]+)(\\|(.*))?$/", $line, $matches ); - # Skip empty lines - if ( count( $matches ) == 0 ) { - continue; - } - $tp = Title::newFromText( $matches[1] ); - $nt =& $tp; - if( is_null( $nt ) ) { - # Bogus title. Ignore these so we don't bomb out later. - continue; - } - if ( isset( $matches[3] ) ) { - $label = $matches[3]; - } else { - $label = ''; - } - - $html = $this->recursiveTagParse( trim( $label ) ); - - $ig->add( $nt, $html ); - - # Only add real images (bug #5586) - if ( $nt->getNamespace() == NS_IMAGE ) { - $this->mOutput->addImage( $nt->getDBkey() ); - } - } - return $ig->toHTML(); - } - - function getImageParams( $handler ) { - if ( $handler ) { - $handlerClass = get_class( $handler ); - } else { - $handlerClass = ''; - } - if ( !isset( $this->mImageParams[$handlerClass] ) ) { - // Initialise static lists - static $internalParamNames = array( - 'horizAlign' => array( 'left', 'right', 'center', 'none' ), - 'vertAlign' => array( 'baseline', 'sub', 'super', 'top', 'text-top', 'middle', - 'bottom', 'text-bottom' ), - 'frame' => array( 'thumbnail', 'manualthumb', 'framed', 'frameless', - 'upright', 'border' ), - ); - static $internalParamMap; - if ( !$internalParamMap ) { - $internalParamMap = array(); - foreach ( $internalParamNames as $type => $names ) { - foreach ( $names as $name ) { - $magicName = str_replace( '-', '_', "img_$name" ); - $internalParamMap[$magicName] = array( $type, $name ); - } - } - } - - // Add handler params - $paramMap = $internalParamMap; - if ( $handler ) { - $handlerParamMap = $handler->getParamMap(); - foreach ( $handlerParamMap as $magic => $paramName ) { - $paramMap[$magic] = array( 'handler', $paramName ); - } - } - $this->mImageParams[$handlerClass] = $paramMap; - $this->mImageParamsMagicArray[$handlerClass] = new MagicWordArray( array_keys( $paramMap ) ); - } - return array( $this->mImageParams[$handlerClass], $this->mImageParamsMagicArray[$handlerClass] ); - } - - /** - * Parse image options text and use it to make an image - */ - function makeImage( $title, $options ) { - # @TODO: let the MediaHandler specify its transform parameters - # - # Check if the options text is of the form "options|alt text" - # Options are: - # * thumbnail make a thumbnail with enlarge-icon and caption, alignment depends on lang - # * left no resizing, just left align. label is used for alt= only - # * right same, but right aligned - # * none same, but not aligned - # * ___px scale to ___ pixels width, no aligning. e.g. use in taxobox - # * center center the image - # * framed Keep original image size, no magnify-button. - # * frameless like 'thumb' but without a frame. Keeps user preferences for width - # * upright reduce width for upright images, rounded to full __0 px - # * border draw a 1px border around the image - # vertical-align values (no % or length right now): - # * baseline - # * sub - # * super - # * top - # * text-top - # * middle - # * bottom - # * text-bottom - - $parts = array_map( 'trim', explode( '|', $options) ); - $sk = $this->mOptions->getSkin(); - - # Give extensions a chance to select the file revision for us - $skip = $time = false; - wfRunHooks( 'BeforeParserMakeImageLinkObj', array( &$this, &$title, &$skip, &$time ) ); - - if ( $skip ) { - return $sk->makeLinkObj( $title ); - } - - # Get parameter map - $file = wfFindFile( $title, $time ); - $handler = $file ? $file->getHandler() : false; - - list( $paramMap, $mwArray ) = $this->getImageParams( $handler ); - - # Process the input parameters - $caption = ''; - $params = array( 'frame' => array(), 'handler' => array(), - 'horizAlign' => array(), 'vertAlign' => array() ); - foreach( $parts as $part ) { - list( $magicName, $value ) = $mwArray->matchVariableStartToEnd( $part ); - if ( isset( $paramMap[$magicName] ) ) { - list( $type, $paramName ) = $paramMap[$magicName]; - $params[$type][$paramName] = $value; - - // Special case; width and height come in one variable together - if( $type == 'handler' && $paramName == 'width' ) { - $m = array(); - if ( preg_match( '/^([0-9]*)x([0-9]*)$/', $value, $m ) ) { - $params[$type]['width'] = intval( $m[1] ); - $params[$type]['height'] = intval( $m[2] ); - } else { - $params[$type]['width'] = intval( $value ); - } - } - } else { - $caption = $part; - } - } - - # Process alignment parameters - if ( $params['horizAlign'] ) { - $params['frame']['align'] = key( $params['horizAlign'] ); - } - if ( $params['vertAlign'] ) { - $params['frame']['valign'] = key( $params['vertAlign'] ); - } - - # Validate the handler parameters - if ( $handler ) { - foreach ( $params['handler'] as $name => $value ) { - if ( !$handler->validateParam( $name, $value ) ) { - unset( $params['handler'][$name] ); - } - } - } - - # Strip bad stuff out of the alt text - $alt = $this->replaceLinkHoldersText( $caption ); - - # make sure there are no placeholders in thumbnail attributes - # that are later expanded to html- so expand them now and - # remove the tags - $alt = $this->mStripState->unstripBoth( $alt ); - $alt = Sanitizer::stripAllTags( $alt ); - - $params['frame']['alt'] = $alt; - $params['frame']['caption'] = $caption; - - # Linker does the rest - $ret = $sk->makeImageLink2( $title, $file, $params['frame'], $params['handler'] ); - - # Give the handler a chance to modify the parser object - if ( $handler ) { - $handler->parserTransformHook( $this, $file ); - } - - return $ret; - } - - /** - * Set a flag in the output object indicating that the content is dynamic and - * shouldn't be cached. - */ - function disableCache() { - wfDebug( "Parser output marked as uncacheable.\n" ); - $this->mOutput->mCacheTime = -1; - } - - /**#@+ - * Callback from the Sanitizer for expanding items found in HTML attribute - * values, so they can be safely tested and escaped. - * @param string $text - * @param PPFrame $frame - * @return string - * @private - */ - function attributeStripCallback( &$text, $frame = false ) { - $text = $this->replaceVariables( $text, $frame ); - $text = $this->mStripState->unstripBoth( $text ); - return $text; - } - - /**#@-*/ - - /**#@+ - * Accessor/mutator - */ - function Title( $x = NULL ) { return wfSetVar( $this->mTitle, $x ); } - function Options( $x = NULL ) { return wfSetVar( $this->mOptions, $x ); } - function OutputType( $x = NULL ) { return wfSetVar( $this->mOutputType, $x ); } - /**#@-*/ - - /**#@+ - * Accessor - */ - function getTags() { return array_merge( array_keys($this->mTransparentTagHooks), array_keys( $this->mTagHooks ) ); } - /**#@-*/ - - - /** - * Break wikitext input into sections, and either pull or replace - * some particular section's text. - * - * External callers should use the getSection and replaceSection methods. - * - * @param string $text Page wikitext - * @param string $section A section identifier string of the form: - * <flag1> - <flag2> - ... - <section number> - * - * Currently the only recognised flag is "T", which means the target section number - * was derived during a template inclusion parse, in other words this is a template - * section edit link. If no flags are given, it was an ordinary section edit link. - * This flag is required to avoid a section numbering mismatch when a section is - * enclosed by <includeonly> (bug 6563). - * - * The section number 0 pulls the text before the first heading; other numbers will - * pull the given section along with its lower-level subsections. If the section is - * not found, $mode=get will return $newtext, and $mode=replace will return $text. - * - * @param string $mode One of "get" or "replace" - * @param string $newText Replacement text for section data. - * @return string for "get", the extracted section text. - * for "replace", the whole page with the section replaced. - */ - private function extractSections( $text, $section, $mode, $newText='' ) { - global $wgTitle; - $this->clearState(); - $this->setTitle( $wgTitle ); // not generally used but removes an ugly failure mode - $this->mOptions = new ParserOptions; - $this->setOutputType( self::OT_WIKI ); - $outText = ''; - $frame = $this->getPreprocessor()->newFrame(); - - // Process section extraction flags - $flags = 0; - $sectionParts = explode( '-', $section ); - $sectionIndex = array_pop( $sectionParts ); - foreach ( $sectionParts as $part ) { - if ( $part == 'T' ) { - $flags |= self::PTD_FOR_INCLUSION; - } - } - // Preprocess the text - $root = $this->preprocessToDom( $text, $flags ); - - // <h> nodes indicate section breaks - // They can only occur at the top level, so we can find them by iterating the root's children - $node = $root->getFirstChild(); - - // Find the target section - if ( $sectionIndex == 0 ) { - // Section zero doesn't nest, level=big - $targetLevel = 1000; - } else { - while ( $node ) { - if ( $node->getName() == 'h' ) { - $bits = $node->splitHeading(); - if ( $bits['i'] == $sectionIndex ) { - $targetLevel = $bits['level']; - break; - } - } - if ( $mode == 'replace' ) { - $outText .= $frame->expand( $node, PPFrame::RECOVER_ORIG ); - } - $node = $node->getNextSibling(); - } - } - - if ( !$node ) { - // Not found - if ( $mode == 'get' ) { - return $newText; - } else { - return $text; - } - } - - // Find the end of the section, including nested sections - do { - if ( $node->getName() == 'h' ) { - $bits = $node->splitHeading(); - $curLevel = $bits['level']; - if ( $bits['i'] != $sectionIndex && $curLevel <= $targetLevel ) { - break; - } - } - if ( $mode == 'get' ) { - $outText .= $frame->expand( $node, PPFrame::RECOVER_ORIG ); - } - $node = $node->getNextSibling(); - } while ( $node ); - - // Write out the remainder (in replace mode only) - if ( $mode == 'replace' ) { - // Output the replacement text - // Add two newlines on -- trailing whitespace in $newText is conventionally - // stripped by the editor, so we need both newlines to restore the paragraph gap - $outText .= $newText . "\n\n"; - while ( $node ) { - $outText .= $frame->expand( $node, PPFrame::RECOVER_ORIG ); - $node = $node->getNextSibling(); - } - } - - if ( is_string( $outText ) ) { - // Re-insert stripped tags - $outText = trim( $this->mStripState->unstripBoth( $outText ) ); - } - - return $outText; - } - - /** - * This function returns the text of a section, specified by a number ($section). - * A section is text under a heading like == Heading == or \<h1\>Heading\</h1\>, or - * the first section before any such heading (section 0). - * - * If a section contains subsections, these are also returned. - * - * @param string $text text to look in - * @param string $section section identifier - * @param string $deftext default to return if section is not found - * @return string text of the requested section - */ - public function getSection( $text, $section, $deftext='' ) { - return $this->extractSections( $text, $section, "get", $deftext ); - } - - public function replaceSection( $oldtext, $section, $text ) { - return $this->extractSections( $oldtext, $section, "replace", $text ); - } - - /** - * Get the timestamp associated with the current revision, adjusted for - * the default server-local timestamp - */ - function getRevisionTimestamp() { - if ( is_null( $this->mRevisionTimestamp ) ) { - wfProfileIn( __METHOD__ ); - global $wgContLang; - $dbr = wfGetDB( DB_SLAVE ); - $timestamp = $dbr->selectField( 'revision', 'rev_timestamp', - array( 'rev_id' => $this->mRevisionId ), __METHOD__ ); - - // Normalize timestamp to internal MW format for timezone processing. - // This has the added side-effect of replacing a null value with - // the current time, which gives us more sensible behavior for - // previews. - $timestamp = wfTimestamp( TS_MW, $timestamp ); - - // The cryptic '' timezone parameter tells to use the site-default - // timezone offset instead of the user settings. - // - // Since this value will be saved into the parser cache, served - // to other users, and potentially even used inside links and such, - // it needs to be consistent for all visitors. - $this->mRevisionTimestamp = $wgContLang->userAdjust( $timestamp, '' ); - - wfProfileOut( __METHOD__ ); - } - return $this->mRevisionTimestamp; - } - - /** - * Mutator for $mDefaultSort - * - * @param $sort New value - */ - public function setDefaultSort( $sort ) { - $this->mDefaultSort = $sort; - } - - /** - * Accessor for $mDefaultSort - * Will use the title/prefixed title if none is set - * - * @return string - */ - public function getDefaultSort() { - if( $this->mDefaultSort !== false ) { - return $this->mDefaultSort; - } else { - return $this->mTitle->getNamespace() == NS_CATEGORY - ? $this->mTitle->getText() - : $this->mTitle->getPrefixedText(); - } - } - - /** - * Try to guess the section anchor name based on a wikitext fragment - * presumably extracted from a heading, for example "Header" from - * "== Header ==". - */ - public function guessSectionNameFromWikiText( $text ) { - # Strip out wikitext links(they break the anchor) - $text = $this->stripSectionName( $text ); - $headline = Sanitizer::decodeCharReferences( $text ); - # strip out HTML - $headline = StringUtils::delimiterReplace( '<', '>', '', $headline ); - $headline = trim( $headline ); - $sectionanchor = '#' . urlencode( str_replace( ' ', '_', $headline ) ); - $replacearray = array( - '%3A' => ':', - '%' => '.' - ); - return str_replace( - array_keys( $replacearray ), - array_values( $replacearray ), - $sectionanchor ); - } - - /** - * Strips a text string of wikitext for use in a section anchor - * - * Accepts a text string and then removes all wikitext from the - * string and leaves only the resultant text (i.e. the result of - * [[User:WikiSysop|Sysop]] would be "Sysop" and the result of - * [[User:WikiSysop]] would be "User:WikiSysop") - this is intended - * to create valid section anchors by mimicing the output of the - * parser when headings are parsed. - * - * @param $text string Text string to be stripped of wikitext - * for use in a Section anchor - * @return Filtered text string - */ - public function stripSectionName( $text ) { - # Strip internal link markup - $text = preg_replace('/\[\[:?([^[|]+)\|([^[]+)\]\]/','$2',$text); - $text = preg_replace('/\[\[:?([^[]+)\|?\]\]/','$1',$text); - - # Strip external link markup (FIXME: Not Tolerant to blank link text - # I.E. [http://www.mediawiki.org] will render as [1] or something depending - # on how many empty links there are on the page - need to figure that out. - $text = preg_replace('/\[(?:' . wfUrlProtocols() . ')([^ ]+?) ([^[]+)\]/','$2',$text); - - # Parse wikitext quotes (italics & bold) - $text = $this->doQuotes($text); - - # Strip HTML tags - $text = StringUtils::delimiterReplace( '<', '>', '', $text ); - return $text; - } - - function srvus( $text ) { - return $this->testSrvus( $text, $this->mOutputType ); - } - - /** - * strip/replaceVariables/unstrip for preprocessor regression testing - */ - function testSrvus( $text, $title, $options, $outputType = self::OT_HTML ) { - $this->clearState(); - if ( ! ( $title instanceof Title ) ) { - $title = Title::newFromText( $title ); - } - $this->mTitle = $title; - $this->mOptions = $options; - $this->setOutputType( $outputType ); - $text = $this->replaceVariables( $text ); - $text = $this->mStripState->unstripBoth( $text ); - $text = Sanitizer::removeHTMLtags( $text ); - return $text; - } - - function testPst( $text, $title, $options ) { - global $wgUser; - if ( ! ( $title instanceof Title ) ) { - $title = Title::newFromText( $title ); - } - return $this->preSaveTransform( $text, $title, $wgUser, $options ); - } - - function testPreprocess( $text, $title, $options ) { - if ( ! ( $title instanceof Title ) ) { - $title = Title::newFromText( $title ); - } - return $this->testSrvus( $text, $title, $options, self::OT_PREPROCESS ); - } - - function markerSkipCallback( $s, $callback ) { - $i = 0; - $out = ''; - while ( $i < strlen( $s ) ) { - $markerStart = strpos( $s, $this->mUniqPrefix, $i ); - if ( $markerStart === false ) { - $out .= call_user_func( $callback, substr( $s, $i ) ); - break; - } else { - $out .= call_user_func( $callback, substr( $s, $i, $markerStart - $i ) ); - $markerEnd = strpos( $s, $this->mMarkerSuffix, $markerStart ); - if ( $markerEnd === false ) { - $out .= substr( $s, $markerStart ); - break; - } else { - $markerEnd += strlen( $this->mMarkerSuffix ); - $out .= substr( $s, $markerStart, $markerEnd - $markerStart ); - $i = $markerEnd; - } - } - } - return $out; - } -} - -/** - * @todo document, briefly. - * @addtogroup Parser - */ -class StripState { - var $general, $nowiki; - - function __construct() { - $this->general = new ReplacementArray; - $this->nowiki = new ReplacementArray; - } - - function unstripGeneral( $text ) { - wfProfileIn( __METHOD__ ); - do { - $oldText = $text; - $text = $this->general->replace( $text ); - } while ( $text != $oldText ); - wfProfileOut( __METHOD__ ); - return $text; - } - - function unstripNoWiki( $text ) { - wfProfileIn( __METHOD__ ); - do { - $oldText = $text; - $text = $this->nowiki->replace( $text ); - } while ( $text != $oldText ); - wfProfileOut( __METHOD__ ); - return $text; - } - - function unstripBoth( $text ) { - wfProfileIn( __METHOD__ ); - do { - $oldText = $text; - $text = $this->general->replace( $text ); - $text = $this->nowiki->replace( $text ); - } while ( $text != $oldText ); - wfProfileOut( __METHOD__ ); - return $text; - } -} - -/** - * @todo document, briefly. - * @addtogroup Parser - */ -class OnlyIncludeReplacer { - var $output = ''; - - function replace( $matches ) { - if ( substr( $matches[1], -1 ) == "\n" ) { - $this->output .= substr( $matches[1], 0, -1 ); - } else { - $this->output .= $matches[1]; - } - } -} - |