enhanced full-text search function

Ignore-this: cb05f50ca4de12e1cdf3a6cfb0e1b8bc

  - better search experience in Asian language
  - sophisticated search query syntax (OR, grouping, etc.)

darcs-hash:20090920121116-9b77a-2718be7a043374669037b10d94101fc70efb95e3.gz

1 files changed, 339 insertions, 126 deletions

diff --git a/inc/fulltext.php b/inc/fulltext.php
index a1b8a1541..636eb3f76 100644
--- a/inc/fulltext.php
+++ b/inc/fulltext.php
@@ -25,105 +25,89 @@ function ft_pageSearch($query,&$highlight){
 
   return trigger_event('SEARCH_QUERY_FULLPAGE', $data, '_ft_pageSearch');
 }
-function _ft_pageSearch(&$data){
-    // split out original parameters
-    $query = $data['query'];
-    $highlight =& $data['highlight'];
 
-    $q = ft_queryParser($query);
-
-    $highlight = array();
+/**
+ * Returns a list of matching documents for the given query
+ *
+ * @author Andreas Gohr <andi@splitbrain.org>
+ * @author Kazutaka Miyasaka <kazmiya@gmail.com>
+ */
+function _ft_pageSearch(&$data) {
+    // parse the given query
+    $q = ft_queryParser($data['query']);
+    $data['highlight'] = $q['highlight'];
 
-    // remember for hilighting later
-    foreach($q['words'] as $wrd){
-        $highlight[] =  str_replace('*','',$wrd);
-    }
+    if (empty($q['parsed_ary'])) return array();
 
     // lookup all words found in the query
-    $words  = array_merge($q['and'],$q['not']);
-    if(!count($words)) return array();
-    $result = idx_lookup($words);
-    if(!count($result)) return array();
-
-    // merge search results with query
-    foreach($q['and'] as $pos => $w){
-        $q['and'][$pos] = $result[$w];
-    }
-    // create a list of unwanted docs
-    $not = array();
-    foreach($q['not'] as $pos => $w){
-        $not = array_merge($not,array_keys($result[$w]));
-    }
+    $lookup = idx_lookup($q['words']);
 
-    // combine and-words
-    if(count($q['and']) > 1){
-        $docs = ft_resultCombine($q['and']);
-    }else{
-        $docs = $q['and'][0];
-    }
-    if(!count($docs)) return array();
-
-    // create a list of hidden pages in the result
-    $hidden = array();
-    $hidden = array_filter(array_keys($docs),'isHiddenPage');
-    $not = array_merge($not,$hidden);
-
-    // filter unmatched namespaces
-    if(!empty($q['ns'])) {
-        $pattern = implode('|^',$q['ns']);
-        foreach($docs as $key => $val) {
-            if(!preg_match('/^'.$pattern.'/',$key)) {
-                unset($docs[$key]);
-            }
-        }
+    // get all pages in this dokuwiki site (!: includes nonexistent pages)
+    $pages_all = array();
+    foreach (idx_getIndex('page', '') as $id) {
+        $pages_all[trim($id)] = 0; // base: 0 hit
     }
 
-    // filter unwanted namespaces
-    if(!empty($q['notns'])) {
-        $pattern = implode('|^',$q['notns']);
-        foreach($docs as $key => $val) {
-            if(preg_match('/^'.$pattern.'/',$key)) {
-                unset($docs[$key]);
-            }
-        }
-    }
-
-    // remove negative matches
-    foreach($not as $n){
-        unset($docs[$n]);
-    }
-
-    if(!count($docs)) return array();
-    // handle phrases
-    if(count($q['phrases'])){
-        $q['phrases'] = array_map('utf8_strtolower',$q['phrases']);
-        // use this for higlighting later:
-        $highlight = array_merge($highlight,$q['phrases']);
-        $q['phrases'] = array_map('preg_quote_cb',$q['phrases']);
-        // check the source of all documents for the exact phrases
-        foreach(array_keys($docs) as $id){
-            $text  = utf8_strtolower(rawWiki($id));
-            foreach($q['phrases'] as $phrase){
-                if(!preg_match('/'.$phrase.'/usi',$text)){
-                    unset($docs[$id]); // no hit - remove
-                    break;
+    // process the query
+    $stack = array();
+    foreach ($q['parsed_ary'] as $token) {
+        switch (substr($token, 0, 3)) {
+            case 'W+:':
+            case 'W-:': // word
+                $word    = substr($token, 3);
+                $stack[] = (array) $lookup[$word];
+                break;
+            case 'P_:': // phrase
+                $phrase = substr($token, 3);
+                // since phrases are always parsed as ((W1)(W2)...(P)),
+                // the end($stack) always points the pages that contain
+                // all words in this phrase
+                $pages  = end($stack);
+                $pages_matched = array();
+                foreach(array_keys($pages) as $id){
+                    $text = utf8_strtolower(rawWiki($id));
+                    if (strpos($text, $phrase) !== false) {
+                        $pages_matched[$id] = 0; // phrase: always 0 hit
+                    }
                 }
-            }
+                $stack[] = $pages_matched;
+                break;
+            case 'N_:': // namespace
+                $ns = substr($token, 3);
+                $pages_matched = array();
+                foreach (array_keys($pages_all) as $id) {
+                    if (strpos($id, $ns) === 0) {
+                        $pages_matched[$id] = 0; // namespace: always 0 hit
+                    }
+                }
+                $stack[] = $pages_matched;
+                break;
+            case 'AND': // and operation
+                list($pages1, $pages2) = array_splice($stack, -2);
+                $stack[] = ft_resultCombine(array($pages1, $pages2));
+                break;
+            case 'OR':  // or operation
+                list($pages1, $pages2) = array_splice($stack, -2);
+                $stack[] = ft_resultUnite(array($pages1, $pages2));
+                break;
+            case 'NOT': // not operation (unary)
+                $pages   = array_pop($stack);
+                $stack[] = ft_resultComplement(array($pages_all, $pages));
+                break;
         }
     }
+    $docs = array_pop($stack);
 
-    if(!count($docs)) return array();
+    if (empty($docs)) return array();
 
-    // check ACL permissions
-    foreach(array_keys($docs) as $doc){
-        if(auth_quickaclcheck($doc) < AUTH_READ){
-            unset($docs[$doc]);
+    // check: settings, acls, existence
+    foreach (array_keys($docs) as $id) {
+        if (isHiddenPage($id) || auth_quickaclcheck($id) < AUTH_READ || !page_exists($id, '', false)) {
+            unset($docs[$id]);
         }
     }
 
-    if(!count($docs)) return array();
-
-    // if there are any hits left, sort them by count
+    // sort docs by count
     arsort($docs);
 
     return $docs;
@@ -404,61 +388,290 @@ function ft_resultCombine($args){
 }
 
 /**
- * Builds an array of search words from a query
+ * Unites found documents and sum up their scores
+ *
+ * based upon ft_resultCombine() function
  *
- * @todo support OR and parenthesises?
+ * @param array $args An array of page arrays
+ * @author Kazutaka Miyasaka <kazmiya@gmail.com>
  */
-function ft_queryParser($query){
-    global $conf;
-    $swfile   = DOKU_INC.'inc/lang/'.$conf['lang'].'/stopwords.txt';
-    if(@file_exists($swfile)){
-        $stopwords = file($swfile);
-    }else{
-        $stopwords = array();
+function ft_resultUnite($args) {
+    $array_count = count($args);
+    if ($array_count === 1) {
+        return $args[0];
     }
 
-    $q = array();
-    $q['query']   = $query;
-    $q['ns']      = array();
-    $q['notns']   = array();
-    $q['phrases'] = array();
-    $q['words']   = array();
-    $q['and']     = array();
-    $q['not']     = array();
-
-    // handle phrase searches
-    while(preg_match('/"(.*?)"/',$query,$match)){
-        $q['phrases'][] = $match[1];
-        $q['and'] = array_merge($q['and'], idx_tokenizer($match[0],$stopwords));
-        $query = preg_replace('/"(.*?)"/','',$query,1);
+    $result = $args[0];
+    for ($i = 1; $i !== $array_count; $i++) {
+        foreach (array_keys($args[$i]) as $id) {
+            $result[$id] += $args[$i][$id];
+        }
+    }
+    return $result;
+}
+
+/**
+ * Computes the difference of documents using page id for comparison
+ *
+ * nearly identical to PHP5's array_diff_key()
+ *
+ * @param array $args An array of page arrays
+ * @author Kazutaka Miyasaka <kazmiya@gmail.com>
+ */
+function ft_resultComplement($args) {
+    $array_count = count($args);
+    if ($array_count === 1) {
+        return $args[0];
     }
 
-    $words = explode(' ',$query);
-    foreach($words as $w){
-        if($w{0} == '-'){
-            $token = idx_tokenizer($w,$stopwords,true);
-            if(count($token)) $q['not'] = array_merge($q['not'],$token);
-        } else if ($w{0} == '@') { // Namespace to search?
-            $w = substr($w,1);
-            $q['ns'] = array_merge($q['ns'],(array)$w);
-        } else if ($w{0} == '^') { // Namespace not to search?
-            $w = substr($w,1);
-            $q['notns'] = array_merge($q['notns'],(array)$w);
-        }else{
-            // asian "words" need to be searched as phrases
-            if(@preg_match_all('/(('.IDX_ASIAN.')+)/u',$w,$matches)){
-                $q['phrases'] = array_merge($q['phrases'],$matches[1]);
+    $result = $args[0];
+    foreach (array_keys($result) as $id) {
+        for ($i = 1; $i !== $array_count; $i++) {
+            if (isset($args[$i][$id])) unset($result[$id]);
+        }
+    }
+    return $result;
+}
 
+/**
+ * Parses a search query and builds an array of search formulas
+ *
+ * @author Andreas Gohr <andi@splitbrain.org>
+ * @author Kazutaka Miyasaka <kazmiya@gmail.com>
+ */
+function ft_queryParser($query){
+    global $conf;
+    $swfile    = DOKU_INC.'inc/lang/'.$conf['lang'].'/stopwords.txt';
+    $stopwords = @file_exists($swfile) ? file($swfile) : array();
+
+    /**
+     * parse a search query and transform it into intermediate representation
+     *
+     * in a search query, you can use the following expressions:
+     *
+     *   words:
+     *     include
+     *     -exclude
+     *   phrases:
+     *     "phrase to be included"
+     *     -"phrase you want to exclude"
+     *   namespaces:
+     *     @include:namespace (or ns:include:namespace)
+     *     ^exclude:namespace (or -ns:exclude:namespace)
+     *   groups:
+     *     ()
+     *     -()
+     *   operators:
+     *     and ('and' is the default operator: you can always omit this)
+     *     or  (lower precedence than 'and')
+     *
+     * e.g. a query [ aa "bb cc" @dd:ee ] means "search pages which contain
+     *      a word 'aa', a phrase 'bb cc' and are within a namespace 'dd:ee'".
+     *      this query is equivalent to [ -(-aa or -"bb cc" or -ns:dd:ee) ]
+     *      as long as you don't mind hit counts.
+     *
+     * intermediate representation consists of the following parts:
+     *
+     *   ( ) - group
+     *   AND - logical and
+     *   OR  - logical or
+     *   NOT - logical not
+     *   W+: - word (needs to be highlighted)
+     *   W-: - word (no need to highlight)
+     *   P_: - phrase
+     *   N_: - namespace
+     */
+    $parsed_query = '';
+    $parens_level = 0;
+    $terms = preg_split('/(-?".*?")/u', utf8_strtolower($query), -1, PREG_SPLIT_DELIM_CAPTURE | PREG_SPLIT_NO_EMPTY);
+
+    foreach ($terms as $term) {
+        $parsed = '';
+        if (preg_match('/^(-?)"(.+)"$/u', $term, $matches)) {
+            // phrase-include and phrase-exclude
+            $not = $matches[1] ? 'NOT' : '';
+            $parsed = $not.ft_termParser($matches[2], $stopwords, false, true);
+        } else {
+            // fix incomplete phrase
+            $term = str_replace('"', ' ', $term);
+
+            // fix parentheses
+            $term = str_replace(')'  , ' ) ', $term);
+            $term = str_replace('('  , ' ( ', $term);
+            $term = str_replace('- (', ' -(', $term);
+
+            // treat ideographic spaces (U+3000) as search term separators
+            // FIXME: some more separators?
+            $term = preg_replace('/[ \x{3000}]+/u', ' ',  $term);
+            $term = trim($term);
+            if ($term === '') continue;
+
+            $tokens = explode(' ', $term);
+            foreach ($tokens as $token) {
+                if ($token === '(') {
+                    // parenthesis-include-open
+                    $parsed .= '(';
+                    ++$parens_level;
+                } elseif ($token === '-(') {
+                    // parenthesis-exclude-open
+                    $parsed .= 'NOT(';
+                    ++$parens_level;
+                } elseif ($token === ')') {
+                    // parenthesis-any-close
+                    if ($parens_level === 0) continue;
+                    $parsed .= ')';
+                    $parens_level--;
+                } elseif ($token === 'and') {
+                    // logical-and (do nothing)
+                } elseif ($token === 'or') {
+                    // logical-or
+                    $parsed .= 'OR';
+                } elseif (preg_match('/^(?:\^|-ns:)(.+)$/u', $token, $matches)) {
+                    // namespace-exclude
+                    $parsed .= 'NOT(N_:'.$matches[1].')';
+                } elseif (preg_match('/^(?:@|ns:)(.+)$/u', $token, $matches)) {
+                    // namespace-include
+                    $parsed .= '(N_:'.$matches[1].')';
+                } elseif (preg_match('/^-(.+)$/', $token, $matches)) {
+                    // word-exclude
+                    $parsed .= 'NOT('.ft_termParser($matches[1], $stopwords).')';
+                } else {
+                    // word-include
+                    $parsed .= ft_termParser($token, $stopwords);
+                }
             }
-            $token = idx_tokenizer($w,$stopwords,true);
-            if(count($token)){
-                $q['and']   = array_merge($q['and'],$token);
-                $q['words'] = array_merge($q['words'],$token);
+        }
+        $parsed_query .= $parsed;
+    }
+
+    // cleanup (very sensitive)
+    $parsed_query .= str_repeat(')', $parens_level);
+    do {
+        $parsed_query_old = $parsed_query;
+        $parsed_query = preg_replace('/(NOT)?\(\)/u', '', $parsed_query);
+    } while ($parsed_query !== $parsed_query_old);
+    $parsed_query = preg_replace('/(NOT|OR)+\)/u', ')'      , $parsed_query);
+    $parsed_query = preg_replace('/(OR)+/u'      , 'OR'     , $parsed_query);
+    $parsed_query = preg_replace('/\(OR/u'       , '('      , $parsed_query);
+    $parsed_query = preg_replace('/^OR|OR$/u'    , ''       , $parsed_query);
+    $parsed_query = preg_replace('/\)(NOT)?\(/u' , ')AND$1(', $parsed_query);
+
+    /**
+     * convert infix notation string into postfix (Reverse Polish notation) array
+     * by Shunting-yard algorithm
+     *
+     * see: http://en.wikipedia.org/wiki/Reverse_Polish_notation
+     * see: http://en.wikipedia.org/wiki/Shunting-yard_algorithm
+     */
+    $parsed_ary     = array();
+    $ope_stack      = array();
+    $ope_precedence = array(')' => 1, 'OR' => 2, 'AND' => 3, 'NOT' => 4, '(' => 5);
+    $ope_regex      = '/([()]|OR|AND|NOT)/u';
+
+    $tokens = preg_split($ope_regex, $parsed_query, -1, PREG_SPLIT_DELIM_CAPTURE | PREG_SPLIT_NO_EMPTY);
+    foreach ($tokens as $token) {
+        if (preg_match($ope_regex, $token)) {
+            // operator
+            $last_ope = end($ope_stack);
+            while ($ope_precedence[$token] <= $ope_precedence[$last_ope] && $last_ope != '(') {
+                $parsed_ary[] = array_pop($ope_stack);
+                $last_ope = end($ope_stack);
+            }
+            if ($token == ')') {
+                array_pop($ope_stack); // this array_pop always deletes '('
+            } else {
+                $ope_stack[] = $token;
             }
+        } else {
+            // operand
+            $token_decoded = str_replace(array('OP', 'CP'), array('(', ')'), $token);
+            $parsed_ary[] = $token_decoded;
         }
     }
+    $parsed_ary = array_values(array_merge($parsed_ary, array_reverse($ope_stack)));
+
+    // cleanup: each double "NOT" in RPN array actually does nothing
+    $parsed_ary_count = count($parsed_ary);
+    for ($i = 1; $i < $parsed_ary_count; ++$i) {
+        if ($parsed_ary[$i] === 'NOT' && $parsed_ary[$i - 1] === 'NOT') {
+            unset($parsed_ary[$i], $parsed_ary[$i - 1]);
+        }
+    }
+    $parsed_ary = array_values($parsed_ary);
+
+    // build return value
+    $q = array();
+    $q['query']      = $query;
+    $q['parsed_str'] = $parsed_query;
+    $q['parsed_ary'] = $parsed_ary;
+
+    foreach ($q['parsed_ary'] as $token) {
+        if ($token[2] !== ':') continue;
+        $body = substr($token, 3);
+
+        switch (substr($token, 0, 3)) {
+            case 'N_:':
+                $q['ns'][]        = $body; // for backward compatibility
+                break;
+            case 'W-:':
+                $q['words'][]     = $body;
+                break;
+            case 'W+:':
+                $q['words'][]     = $body;
+                $q['highlight'][] = str_replace('*', '', $body);
+                break;
+            case 'P_:':
+                $q['phrases'][]   = $body;
+                $q['highlight'][] = str_replace('*', '', $body);
+                break;
+        }
+    }
+    foreach (array('words', 'phrases', 'highlight', 'ns') as $key) {
+        $q[$key] = empty($q[$key]) ? array() : array_values(array_unique($q[$key]));
+    }
+
+    // keep backward compatibility (to some extent)
+    // this part can be deleted if no plugins use ft_queryParser() directly
+    $q['and']   = $q['words'];
+    $q['not']   = array(); // difficult to set: imagine [ aaa -(bbb -ccc) ]
+    $q['notns'] = array(); // same as above
 
     return $q;
 }
 
+/**
+ * Transforms given search term into intermediate representation
+ *
+ * This function is used in ft_queryParser() and not for general purpose use.
+ *
+ * @author Kazutaka Miyasaka <kazmiya@gmail.com>
+ */
+function ft_termParser($term, &$stopwords, $consider_asian = true, $phrase_mode = false) {
+    $parsed = '';
+    if ($consider_asian) {
+        // successive asian characters need to be searched as a phrase
+        $words = preg_split('/('.IDX_ASIAN.'+)/u', $term, -1, PREG_SPLIT_DELIM_CAPTURE | PREG_SPLIT_NO_EMPTY);
+        foreach ($words as $word) {
+            if (preg_match('/'.IDX_ASIAN.'/u', $word)) $phrase_mode = true;
+            $parsed .= ft_termParser($word, $stopwords, false, $phrase_mode);
+        }
+    } else {
+        $term_noparen = str_replace(array('(', ')'), ' ', $term);
+        $words = idx_tokenizer($term_noparen, $stopwords, true);
+
+        // W+: needs to be highlighted, W-: no need to highlight
+        if (empty($words)) {
+            $parsed = '()'; // important: do not remove
+        } elseif ($words[0] === $term) {
+            $parsed = '(W+:'.$words[0].')';
+        } elseif ($phrase_mode) {
+            $term_encoded = str_replace(array('(', ')'), array('OP', 'CP'), $term);
+            $parsed = '((W-:'.implode(')(W-:', $words).')(P_:'.$term_encoded.'))';
+        } else {
+            $parsed = '((W+:'.implode(')(W+:', $words).'))';
+        }
+    }
+    return $parsed;
+}
+
 //Setup VIM: ex: et ts=4 enc=utf-8 :