SearchSqlite.php

Go to the documentation of this file.

<?php
use MediaWiki\MediaWikiServices;
use Wikimedia\AtEase\AtEase;
use Wikimedia\Rdbms\IDatabase;
 
class SearchSqlite extends SearchDatabase {
    private function fulltextSearchSupported() {
        $dbr = $this->dbProvider->getReplicaDatabase();
        $sql = (string)$dbr->newSelectQueryBuilder()
            ->select( 'sql' )
            ->from( $dbr->addIdentifierQuotes( 'sqlite_master' ) )
            ->where( [ 'tbl_name' => $dbr->tableName( 'searchindex', 'raw' ) ] )
            ->caller( __METHOD__ )->fetchField();
 
        return ( stristr( $sql, 'fts' ) !== false );
    }
 
    private function parseQuery( $filteredText, $fulltext ) {
        $lc = $this->legalSearchChars( self::CHARS_NO_SYNTAX ); // Minus syntax chars (" and *)
        $searchon = '';
        $this->searchTerms = [];
 
        $m = [];
        if ( preg_match_all( '/([-+<>~]?)(([' . $lc . ']+)(\*?)|"[^"]*")/',
                $filteredText, $m, PREG_SET_ORDER ) ) {
            foreach ( $m as $bits ) {
                AtEase::suppressWarnings();
                [ /* all */, $modifier, $term, $nonQuoted, $wildcard ] = $bits;
                AtEase::restoreWarnings();
 
                if ( $nonQuoted != '' ) {
                    $term = $nonQuoted;
                    $quote = '';
                } else {
                    $term = str_replace( '"', '', $term );
                    $quote = '"';
                }
 
                if ( $searchon !== '' ) {
                    $searchon .= ' ';
                }
 
                // 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.
 
                $converter = MediaWikiServices::getInstance()->getLanguageConverterFactory()
                    ->getLanguageConverter();
                $convertedVariants = $converter->autoConvertToAllVariants( $term );
                if ( is_array( $convertedVariants ) ) {
                    $variants = array_unique( array_values( $convertedVariants ) );
                } else {
                    $variants = [ $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(
                    [ MediaWikiServices::getInstance()->getContentLanguage(),
                        '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 .= '(';
                }
                $count = 0;
                foreach ( $strippedVariants as $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 ) . '"';
                    }
                    if ( $count > 0 ) {
                        $searchon .= " OR ";
                    }
                    $searchon .= "$quote$stripped$quote$wildcard ";
                    ++$count;
                }
                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;
            }
 
        } else {
            wfDebug( __METHOD__ . ": Can't understand search query '{$filteredText}'" );
        }
 
        $dbr = $this->dbProvider->getReplicaDatabase();
        $searchon = $dbr->addQuotes( $searchon );
        $field = $this->getIndexField( $fulltext );
 
        return " $field MATCH $searchon ";
    }
 
    private function regexTerm( $string, $wildcard ) {
        $regex = preg_quote( $string, '/' );
        if ( MediaWikiServices::getInstance()->getContentLanguage()->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 function legalSearchChars( $type = self::CHARS_ALL ) {
        $searchChars = parent::legalSearchChars( $type );
        if ( $type === self::CHARS_ALL ) {
            // " for phrase, * for wildcard
            $searchChars = "\"*" . $searchChars;
        }
        return $searchChars;
    }
 
    protected function doSearchTextInDB( $term ) {
        return $this->searchInternal( $term, true );
    }
 
    protected function doSearchTitleInDB( $term ) {
        return $this->searchInternal( $term, false );
    }
 
    protected function searchInternal( $term, $fulltext ) {
        if ( !$this->fulltextSearchSupported() ) {
            return null;
        }
 
        $filteredTerm =
            $this->filter( MediaWikiServices::getInstance()->getContentLanguage()->lc( $term ) );
        $dbr = $this->dbProvider->getReplicaDatabase();
        // The real type is still IDatabase, but IReplicaDatabase is used for safety.
        '@phan-var IDatabase $dbr';
        // phpcs:ignore MediaWiki.Usage.DbrQueryUsage.DbrQueryFound
        $resultSet = $dbr->query( $this->getQuery( $filteredTerm, $fulltext ), __METHOD__ );
 
        $total = null;
        // phpcs:ignore MediaWiki.Usage.DbrQueryUsage.DbrQueryFound
        $totalResult = $dbr->query( $this->getCountQuery( $filteredTerm, $fulltext ), __METHOD__ );
        $row = $totalResult->fetchObject();
        if ( $row ) {
            $total = intval( $row->c );
        }
        $totalResult->free();
 
        return new SqlSearchResultSet( $resultSet, $this->searchTerms, $total );
    }
 
    private function queryNamespaces() {
        if ( $this->namespaces === null ) {
            return '';  # search all
        }
        if ( $this->namespaces === [] ) {
            $namespaces = NS_MAIN;
        } else {
            $dbr = $this->dbProvider->getReplicaDatabase();
            $namespaces = $dbr->makeList( $this->namespaces );
        }
        return 'AND page_namespace IN (' . $namespaces . ')';
    }
 
    private function limitResult( $sql ) {
        return $this->dbProvider->getReplicaDatabase()->limitResult( $sql, $this->limit, $this->offset );
    }
 
    private function getQuery( $filteredTerm, $fulltext ) {
        return $this->limitResult(
            $this->queryMain( $filteredTerm, $fulltext ) . ' ' .
            $this->queryNamespaces()
        );
    }
 
    private function getIndexField( $fulltext ) {
        return $fulltext ? 'si_text' : 'si_title';
    }
 
    private function queryMain( $filteredTerm, $fulltext ) {
        $match = $this->parseQuery( $filteredTerm, $fulltext );
        $dbr = $this->dbProvider->getReplicaDatabase();
        $page = $dbr->tableName( 'page' );
        $searchindex = $dbr->tableName( 'searchindex' );
        return "SELECT $searchindex.rowid, page_namespace, page_title " .
            "FROM $searchindex CROSS JOIN $page " .
            "WHERE page_id=$searchindex.rowid AND $match";
    }
 
    private function getCountQuery( $filteredTerm, $fulltext ) {
        $match = $this->parseQuery( $filteredTerm, $fulltext );
        $dbr = $this->dbProvider->getReplicaDatabase();
        $page = $dbr->tableName( 'page' );
        $searchindex = $dbr->tableName( 'searchindex' );
        return "SELECT COUNT(*) AS c " .
            "FROM $searchindex CROSS JOIN $page " .
            "WHERE page_id=$searchindex.rowid AND $match " .
            $this->queryNamespaces();
    }
 
    public function update( $id, $title, $text ) {
        if ( !$this->fulltextSearchSupported() ) {
            return;
        }
        // @todo find a method to do it in a single request,
        // couldn't do it so far due to typelessness of FTS3 tables.
        $dbw = $this->dbProvider->getPrimaryDatabase();
        $dbw->newDeleteQueryBuilder()
            ->deleteFrom( 'searchindex' )
            ->where( [ 'rowid' => $id ] )
            ->caller( __METHOD__ )->execute();
        $dbw->newInsertQueryBuilder()
            ->insertInto( 'searchindex' )
            ->row( [ 'rowid' => $id, 'si_title' => $title, 'si_text' => $text ] )
            ->caller( __METHOD__ )->execute();
    }
 
    public function updateTitle( $id, $title ) {
        if ( !$this->fulltextSearchSupported() ) {
            return;
        }
 
        $dbw = $this->dbProvider->getPrimaryDatabase();
        $dbw->newUpdateQueryBuilder()
            ->update( 'searchindex' )
            ->set( [ 'si_title' => $title ] )
            ->where( [ 'rowid' => $id ] )
            ->caller( __METHOD__ )->execute();
    }
}