]> scripts.mit.edu Git - autoinstallsdev/mediawiki.git/blobdiff - includes/search/SearchMySQL.php
MediaWiki 1.16.0
[autoinstallsdev/mediawiki.git] / includes / search / SearchMySQL.php
diff --git a/includes/search/SearchMySQL.php b/includes/search/SearchMySQL.php
new file mode 100644 (file)
index 0000000..0c238be
--- /dev/null
@@ -0,0 +1,412 @@
+<?php
+# Copyright (C) 2004 Brion Vibber <brion@pobox.com>
+# http://www.mediawiki.org/
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License along
+# with this program; if not, write to the Free Software Foundation, Inc.,
+# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+# http://www.gnu.org/copyleft/gpl.html
+
+/**
+ * @file
+ * @ingroup Search
+ */
+
+/**
+ * Search engine hook for MySQL 4+
+ * @ingroup Search
+ */
+class SearchMySQL extends SearchEngine {
+       var $strictMatching = true;
+       static $mMinSearchLength;
+
+       /** @todo document */
+       function __construct( $db ) {
+               $this->db = $db;
+       }
+
+       /** 
+        * Parse the user's query and transform it into an SQL fragment which will 
+        * become part of a WHERE clause
+        */
+       function parseQuery( $filteredText, $fulltext ) {
+               global $wgContLang;
+               $lc = SearchEngine::legalSearchChars(); // Minus format chars
+               $searchon = '';
+               $this->searchTerms = array();
+
+               # FIXME: This doesn't handle parenthetical expressions.
+               $m = array();
+               if( preg_match_all( '/([-+<>~]?)(([' . $lc . ']+)(\*?)|"[^"]*")/',
+                         $filteredText, $m, PREG_SET_ORDER ) ) {
+                       foreach( $m as $bits ) {
+                               @list( /* all */, $modifier, $term, $nonQuoted, $wildcard ) = $bits;
+                               
+                               if( $nonQuoted != '' ) {
+                                       $term = $nonQuoted;
+                                       $quote = '';
+                               } else {
+                                       $term = str_replace( '"', '', $term );
+                                       $quote = '"';
+                               }
+                       
+                               if( $searchon !== '' ) $searchon .= ' ';
+                               if( $this->strictMatching && ($modifier == '') ) {
+                                       // If we leave this out, boolean op defaults to OR which is rarely helpful.
+                                       $modifier = '+';
+                               }
+                               
+                               // Some languages such as Serbian store the input form in the search index,
+                               // so we may need to search for matches in multiple writing system variants.
+                               $convertedVariants = $wgContLang->autoConvertToAllVariants( $term );
+                               if( is_array( $convertedVariants ) ) {
+                                       $variants = array_unique( array_values( $convertedVariants ) );
+                               } else {
+                                       $variants = array( $term );
+                               }
+                               
+                               // The low-level search index does some processing on input to work
+                               // around problems with minimum lengths and encoding in MySQL's
+                               // fulltext engine.
+                               // For Chinese this also inserts spaces between adjacent Han characters.
+                               $strippedVariants = array_map(
+                                       array( $wgContLang, 'normalizeForSearch' ),
+                                       $variants );
+                               
+                               // Some languages such as Chinese force all variants to a canonical
+                               // form when stripping to the low-level search index, so to be sure
+                               // let's check our variants list for unique items after stripping.
+                               $strippedVariants = array_unique( $strippedVariants );
+                               
+                               $searchon .= $modifier;
+                               if( count( $strippedVariants) > 1 )
+                                       $searchon .= '(';
+                               foreach( $strippedVariants as $stripped ) {
+                                       $stripped = $this->normalizeText( $stripped );
+                                       if( $nonQuoted && strpos( $stripped, ' ' ) !== false ) {
+                                               // Hack for Chinese: we need to toss in quotes for
+                                               // multiple-character phrases since normalizeForSearch()
+                                               // added spaces between them to make word breaks.
+                                               $stripped = '"' . trim( $stripped ) . '"';
+                                       }
+                                       $searchon .= "$quote$stripped$quote$wildcard ";
+                               }
+                               if( count( $strippedVariants) > 1 )
+                                       $searchon .= ')';
+                               
+                               // Match individual terms or quoted phrase in result highlighting...
+                               // Note that variants will be introduced in a later stage for highlighting!
+                               $regexp = $this->regexTerm( $term, $wildcard );
+                               $this->searchTerms[] = $regexp;
+                       }
+                       wfDebug( __METHOD__ . ": Would search with '$searchon'\n" );
+                       wfDebug( __METHOD__ . ': Match with /' . implode( '|', $this->searchTerms ) . "/\n" );
+               } else {
+                       wfDebug( __METHOD__ . ": Can't understand search query '{$filteredText}'\n" );
+               }
+
+               $searchon = $this->db->strencode( $searchon );
+               $field = $this->getIndexField( $fulltext );
+               return " MATCH($field) AGAINST('$searchon' IN BOOLEAN MODE) ";
+       }
+       
+       function regexTerm( $string, $wildcard ) {
+               global $wgContLang;
+               
+               $regex = preg_quote( $string, '/' );
+               if( $wgContLang->hasWordBreaks() ) {
+                       if( $wildcard ) {
+                               // Don't cut off the final bit!
+                               $regex = "\b$regex";
+                       } else {
+                               $regex = "\b$regex\b";
+                       }
+               } else {
+                       // For Chinese, words may legitimately abut other words in the text literal.
+                       // Don't add \b boundary checks... note this could cause false positives
+                       // for latin chars.
+               }
+               return $regex;
+       }
+
+       public static function legalSearchChars() {
+               return "\"*" . parent::legalSearchChars();
+       }
+
+       /**
+        * Perform a full text search query and return a result set.
+        *
+        * @param $term String: raw search term
+        * @return MySQLSearchResultSet
+        */
+       function searchText( $term ) {
+               return $this->searchInternal( $term, true );
+       }
+
+       /**
+        * Perform a title-only search query and return a result set.
+        *
+        * @param $term String: raw search term
+        * @return MySQLSearchResultSet
+        */
+       function searchTitle( $term ) {
+               return $this->searchInternal( $term, false );
+       }
+       
+       protected function searchInternal( $term, $fulltext ) {
+               global $wgCountTotalSearchHits;
+               
+               $filteredTerm = $this->filter( $term );
+               $resultSet = $this->db->query( $this->getQuery( $filteredTerm, $fulltext ) );
+               
+               $total = null;
+               if( $wgCountTotalSearchHits ) {
+                       $totalResult = $this->db->query( $this->getCountQuery( $filteredTerm, $fulltext ) );
+                       $row = $totalResult->fetchObject();
+                       if( $row ) {
+                               $total = intval( $row->c );
+                       }
+                       $totalResult->free();
+               }
+               
+               return new MySQLSearchResultSet( $resultSet, $this->searchTerms, $total );
+       }
+
+
+       /**
+        * Return a partial WHERE clause to exclude redirects, if so set
+        * @return String
+        */
+       function queryRedirect() {
+               if( $this->showRedirects ) {
+                       return '';
+               } else {
+                       return 'AND page_is_redirect=0';
+               }
+       }
+
+       /**
+        * Return a partial WHERE clause to limit the search to the given namespaces
+        * @return String
+        */
+       function queryNamespaces() {
+               if( is_null($this->namespaces) )
+                       return '';  # search all
+               if ( !count( $this->namespaces ) ) {
+                       $namespaces = '0';
+               } else {
+                       $namespaces = $this->db->makeList( $this->namespaces );
+               }
+               return 'AND page_namespace IN (' . $namespaces . ')';
+       }
+
+       /**
+        * Return a LIMIT clause to limit results on the query.
+        * @return String
+        */
+       function queryLimit() {
+               return $this->db->limitResult( '', $this->limit, $this->offset );
+       }
+
+       /**
+        * Does not do anything for generic search engine
+        * subclasses may define this though
+        * @return String
+        */
+       function queryRanking( $filteredTerm, $fulltext ) {
+               return '';
+       }
+
+       /**
+        * Construct the full SQL query to do the search.
+        * The guts shoulds be constructed in queryMain()
+        * @param $filteredTerm String
+        * @param $fulltext Boolean
+        */
+       function getQuery( $filteredTerm, $fulltext ) {
+               return $this->queryMain( $filteredTerm, $fulltext ) . ' ' .
+                       $this->queryRedirect() . ' ' .
+                       $this->queryNamespaces() . ' ' .
+                       $this->queryRanking( $filteredTerm, $fulltext ) . ' ' .
+                       $this->queryLimit();
+       }
+       
+       /**
+        * Picks which field to index on, depending on what type of query.
+        * @param $fulltext Boolean
+        * @return String
+        */
+       function getIndexField( $fulltext ) {
+               return $fulltext ? 'si_text' : 'si_title';
+       }
+
+       /**
+        * Get the base part of the search query.
+        * The actual match syntax will depend on the server
+        * version; MySQL 3 and MySQL 4 have different capabilities
+        * in their fulltext search indexes.
+        *
+        * @param $filteredTerm String
+        * @param $fulltext Boolean
+        * @return String
+        */
+       function queryMain( $filteredTerm, $fulltext ) {
+               $match = $this->parseQuery( $filteredTerm, $fulltext );
+               $page        = $this->db->tableName( 'page' );
+               $searchindex = $this->db->tableName( 'searchindex' );
+               return 'SELECT page_id, page_namespace, page_title ' .
+                       "FROM $page,$searchindex " .
+                       'WHERE page_id=si_page AND ' . $match;
+       }
+
+       function getCountQuery( $filteredTerm, $fulltext ) {
+               $match = $this->parseQuery( $filteredTerm, $fulltext );
+               $page        = $this->db->tableName( 'page' );
+               $searchindex = $this->db->tableName( 'searchindex' );
+               return "SELECT COUNT(*) AS c " .
+                       "FROM $page,$searchindex " .
+                       'WHERE page_id=si_page AND ' . $match .
+                       $this->queryRedirect() . ' ' .
+                       $this->queryNamespaces();
+       }
+
+       /**
+        * Create or update the search index record for the given page.
+        * Title and text should be pre-processed.
+        *
+        * @param $id Integer
+        * @param $title String
+        * @param $text String
+        */
+       function update( $id, $title, $text ) {
+               $dbw = wfGetDB( DB_MASTER );
+               $dbw->replace( 'searchindex',
+                       array( 'si_page' ),
+                       array(
+                               'si_page' => $id,
+                               'si_title' => $this->normalizeText( $title ),
+                               'si_text' => $this->normalizeText( $text )
+                       ), __METHOD__ );
+       }
+
+       /**
+        * Update a search index record's title only.
+        * Title should be pre-processed.
+        *
+        * @param $id Integer
+        * @param $title String
+        */
+    function updateTitle( $id, $title ) {
+               $dbw = wfGetDB( DB_MASTER );
+
+               $dbw->update( 'searchindex',
+                       array( 'si_title' => $this->normalizeText( $title ) ),
+                       array( 'si_page'  => $id ),
+                       __METHOD__,
+                       array( $dbw->lowPriorityOption() ) );
+       }
+
+       /**
+        * Converts some characters for MySQL's indexing to grok it correctly,
+        * and pads short words to overcome limitations.
+        */
+       function normalizeText( $string ) {
+               global $wgContLang;
+
+               wfProfileIn( __METHOD__ );
+               
+               // Some languages such as Chinese require word segmentation
+               $out = $wgContLang->wordSegmentation( $string );
+
+               // MySQL fulltext index doesn't grok utf-8, so we
+               // need to fold cases and convert to hex
+               $out = preg_replace_callback(
+                       "/([\\xc0-\\xff][\\x80-\\xbf]*)/",
+                       array( $this, 'stripForSearchCallback' ),
+                       $wgContLang->lc( $out ) );
+
+               // And to add insult to injury, the default indexing
+               // ignores short words... Pad them so we can pass them
+               // through without reconfiguring the server...
+               $minLength = $this->minSearchLength();
+               if( $minLength > 1 ) {
+                       $n = $minLength - 1;
+                       $out = preg_replace(
+                               "/\b(\w{1,$n})\b/",
+                               "$1u800",
+                               $out );
+               }
+
+               // Periods within things like hostnames and IP addresses
+               // are also important -- we want a search for "example.com"
+               // or "192.168.1.1" to work sanely.
+               //
+               // MySQL's search seems to ignore them, so you'd match on
+               // "example.wikipedia.com" and "192.168.83.1" as well.
+               $out = preg_replace(
+                       "/(\w)\.(\w|\*)/u",
+                       "$1u82e$2",
+                       $out );
+
+               wfProfileOut( __METHOD__ );
+               
+               return $out;
+       }
+
+       /**
+        * Armor a case-folded UTF-8 string to get through MySQL's
+        * fulltext search without being mucked up by funny charset
+        * settings or anything else of the sort.
+        */
+       protected function stripForSearchCallback( $matches ) {
+               return 'u8' . bin2hex( $matches[1] );
+       }
+
+       /**
+        * Check MySQL server's ft_min_word_len setting so we know
+        * if we need to pad short words...
+        * 
+        * @return int
+        */
+       protected function minSearchLength() {
+               if( is_null( self::$mMinSearchLength ) ) {
+                       $sql = "SHOW GLOBAL VARIABLES LIKE 'ft\\_min\\_word\\_len'";
+
+                       $dbr = wfGetDB( DB_SLAVE );
+                       $result = $dbr->query( $sql );
+                       $row = $result->fetchObject();
+                       $result->free();
+
+                       if( $row && $row->Variable_name == 'ft_min_word_len' ) {
+                               self::$mMinSearchLength = intval( $row->Value );
+                       } else {
+                               self::$mMinSearchLength = 0;
+                       }
+               }
+               return self::$mMinSearchLength;
+       }
+}
+
+/**
+ * @ingroup Search
+ */
+class MySQLSearchResultSet extends SqlSearchResultSet {
+       function MySQLSearchResultSet( $resultSet, $terms, $totalHits=null ) {
+               parent::__construct( $resultSet, $terms );
+               $this->mTotalHits = $totalHits;
+       }
+
+       function getTotalHits() {
+               return $this->mTotalHits;
+       }
+}
\ No newline at end of file