website_jukni/dokuwiki/inc/fulltext.php
2017-12-29 15:51:59 +01:00

816 lines
27 KiB
PHP

<?php
/**
* DokuWiki fulltextsearch functions using the index
*
* @license GPL 2 (http://www.gnu.org/licenses/gpl.html)
* @author Andreas Gohr <andi@splitbrain.org>
*/
if(!defined('DOKU_INC')) die('meh.');
/**
* create snippets for the first few results only
*/
if(!defined('FT_SNIPPET_NUMBER')) define('FT_SNIPPET_NUMBER',15);
/**
* The fulltext search
*
* Returns a list of matching documents for the given query
*
* refactored into ft_pageSearch(), _ft_pageSearch() and trigger_event()
*
* @param string $query
* @param array $highlight
* @return array
*/
function ft_pageSearch($query,&$highlight){
$data = array();
$data['query'] = $query;
$data['highlight'] =& $highlight;
return trigger_event('SEARCH_QUERY_FULLPAGE', $data, '_ft_pageSearch');
}
/**
* Returns a list of matching documents for the given query
*
* @author Andreas Gohr <andi@splitbrain.org>
* @author Kazutaka Miyasaka <kazmiya@gmail.com>
*
* @param array $data event data
* @return array matching documents
*/
function _ft_pageSearch(&$data) {
$Indexer = idx_get_indexer();
// parse the given query
$q = ft_queryParser($Indexer, $data['query']);
$data['highlight'] = $q['highlight'];
if (empty($q['parsed_ary'])) return array();
// lookup all words found in the query
$lookup = $Indexer->lookup($q['words']);
// get all pages in this dokuwiki site (!: includes nonexistent pages)
$pages_all = array();
foreach ($Indexer->getPages() as $id) {
$pages_all[$id] = 0; // base: 0 hit
}
// process the query
$stack = array();
foreach ($q['parsed_ary'] as $token) {
switch (substr($token, 0, 3)) {
case 'W+:':
case 'W-:':
case 'W_:': // word
$word = substr($token, 3);
$stack[] = (array) $lookup[$word];
break;
case 'P+:':
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){
$evdata = array(
'id' => $id,
'phrase' => $phrase,
'text' => rawWiki($id)
);
$evt = new Doku_Event('FULLTEXT_PHRASE_MATCH',$evdata);
if ($evt->advise_before() && $evt->result !== true) {
$text = utf8_strtolower($evdata['text']);
if (strpos($text, $phrase) !== false) {
$evt->result = true;
}
}
$evt->advise_after();
if ($evt->result === true) {
$pages_matched[$id] = 0; // phrase: always 0 hit
}
}
$stack[] = $pages_matched;
break;
case 'N+:':
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 (empty($docs)) return array();
// check: settings, acls, existence
foreach (array_keys($docs) as $id) {
if (isHiddenPage($id) || auth_quickaclcheck($id) < AUTH_READ || !page_exists($id, '', false)) {
unset($docs[$id]);
}
}
// sort docs by count
arsort($docs);
return $docs;
}
/**
* Returns the backlinks for a given page
*
* Uses the metadata index.
*
* @param string $id The id for which links shall be returned
* @param bool $ignore_perms Ignore the fact that pages are hidden or read-protected
* @return array The pages that contain links to the given page
*/
function ft_backlinks($id, $ignore_perms = false){
$result = idx_get_indexer()->lookupKey('relation_references', $id);
if(!count($result)) return $result;
// check ACL permissions
foreach(array_keys($result) as $idx){
if(($ignore_perms !== true && (
isHiddenPage($result[$idx]) || auth_quickaclcheck($result[$idx]) < AUTH_READ
)) || !page_exists($result[$idx], '', false)){
unset($result[$idx]);
}
}
sort($result);
return $result;
}
/**
* Returns the pages that use a given media file
*
* Uses the relation media metadata property and the metadata index.
*
* Note that before 2013-07-31 the second parameter was the maximum number of results and
* permissions were ignored. That's why the parameter is now checked to be explicitely set
* to true (with type bool) in order to be compatible with older uses of the function.
*
* @param string $id The media id to look for
* @param bool $ignore_perms Ignore hidden pages and acls (optional, default: false)
* @return array A list of pages that use the given media file
*/
function ft_mediause($id, $ignore_perms = false){
$result = idx_get_indexer()->lookupKey('relation_media', $id);
if(!count($result)) return $result;
// check ACL permissions
foreach(array_keys($result) as $idx){
if(($ignore_perms !== true && (
isHiddenPage($result[$idx]) || auth_quickaclcheck($result[$idx]) < AUTH_READ
)) || !page_exists($result[$idx], '', false)){
unset($result[$idx]);
}
}
sort($result);
return $result;
}
/**
* Quicksearch for pagenames
*
* By default it only matches the pagename and ignores the
* namespace. This can be changed with the second parameter.
* The third parameter allows to search in titles as well.
*
* The function always returns titles as well
*
* @triggers SEARCH_QUERY_PAGELOOKUP
* @author Andreas Gohr <andi@splitbrain.org>
* @author Adrian Lang <lang@cosmocode.de>
*
* @param string $id page id
* @param bool $in_ns match against namespace as well?
* @param bool $in_title search in title?
* @return string[]
*/
function ft_pageLookup($id, $in_ns=false, $in_title=false){
$data = compact('id', 'in_ns', 'in_title');
$data['has_titles'] = true; // for plugin backward compatibility check
return trigger_event('SEARCH_QUERY_PAGELOOKUP', $data, '_ft_pageLookup');
}
/**
* Returns list of pages as array(pageid => First Heading)
*
* @param array &$data event data
* @return string[]
*/
function _ft_pageLookup(&$data){
// split out original parameters
$id = $data['id'];
if (preg_match('/(?:^| )(?:@|ns:)([\w:]+)/', $id, $matches)) {
$ns = cleanID($matches[1]) . ':';
$id = str_replace($matches[0], '', $id);
}
$in_ns = $data['in_ns'];
$in_title = $data['in_title'];
$cleaned = cleanID($id);
$Indexer = idx_get_indexer();
$page_idx = $Indexer->getPages();
$pages = array();
if ($id !== '' && $cleaned !== '') {
foreach ($page_idx as $p_id) {
if ((strpos($in_ns ? $p_id : noNSorNS($p_id), $cleaned) !== false)) {
if (!isset($pages[$p_id]))
$pages[$p_id] = p_get_first_heading($p_id, METADATA_DONT_RENDER);
}
}
if ($in_title) {
foreach ($Indexer->lookupKey('title', $id, '_ft_pageLookupTitleCompare') as $p_id) {
if (!isset($pages[$p_id]))
$pages[$p_id] = p_get_first_heading($p_id, METADATA_DONT_RENDER);
}
}
}
if (isset($ns)) {
foreach (array_keys($pages) as $p_id) {
if (strpos($p_id, $ns) !== 0) {
unset($pages[$p_id]);
}
}
}
// discard hidden pages
// discard nonexistent pages
// check ACL permissions
foreach(array_keys($pages) as $idx){
if(!isVisiblePage($idx) || !page_exists($idx) ||
auth_quickaclcheck($idx) < AUTH_READ) {
unset($pages[$idx]);
}
}
uksort($pages,'ft_pagesorter');
return $pages;
}
/**
* Tiny helper function for comparing the searched title with the title
* from the search index. This function is a wrapper around stripos with
* adapted argument order and return value.
*
* @param string $search searched title
* @param string $title title from index
* @return bool
*/
function _ft_pageLookupTitleCompare($search, $title) {
return stripos($title, $search) !== false;
}
/**
* Sort pages based on their namespace level first, then on their string
* values. This makes higher hierarchy pages rank higher than lower hierarchy
* pages.
*
* @param string $a
* @param string $b
* @return int Returns < 0 if $a is less than $b; > 0 if $a is greater than $b, and 0 if they are equal.
*/
function ft_pagesorter($a, $b){
$ac = count(explode(':',$a));
$bc = count(explode(':',$b));
if($ac < $bc){
return -1;
}elseif($ac > $bc){
return 1;
}
return strcmp ($a,$b);
}
/**
* Creates a snippet extract
*
* @author Andreas Gohr <andi@splitbrain.org>
* @triggers FULLTEXT_SNIPPET_CREATE
*
* @param string $id page id
* @param array $highlight
* @return mixed
*/
function ft_snippet($id,$highlight){
$text = rawWiki($id);
$text = str_replace("\xC2\xAD",'',$text); // remove soft-hyphens
$evdata = array(
'id' => $id,
'text' => &$text,
'highlight' => &$highlight,
'snippet' => '',
);
$evt = new Doku_Event('FULLTEXT_SNIPPET_CREATE',$evdata);
if ($evt->advise_before()) {
$match = array();
$snippets = array();
$utf8_offset = $offset = $end = 0;
$len = utf8_strlen($text);
// build a regexp from the phrases to highlight
$re1 = '('.join('|',array_map('ft_snippet_re_preprocess', array_map('preg_quote_cb',array_filter((array) $highlight)))).')';
$re2 = "$re1.{0,75}(?!\\1)$re1";
$re3 = "$re1.{0,45}(?!\\1)$re1.{0,45}(?!\\1)(?!\\2)$re1";
for ($cnt=4; $cnt--;) {
if (0) {
} else if (preg_match('/'.$re3.'/iu',$text,$match,PREG_OFFSET_CAPTURE,$offset)) {
} else if (preg_match('/'.$re2.'/iu',$text,$match,PREG_OFFSET_CAPTURE,$offset)) {
} else if (preg_match('/'.$re1.'/iu',$text,$match,PREG_OFFSET_CAPTURE,$offset)) {
} else {
break;
}
list($str,$idx) = $match[0];
// convert $idx (a byte offset) into a utf8 character offset
$utf8_idx = utf8_strlen(substr($text,0,$idx));
$utf8_len = utf8_strlen($str);
// establish context, 100 bytes surrounding the match string
// first look to see if we can go 100 either side,
// then drop to 50 adding any excess if the other side can't go to 50,
$pre = min($utf8_idx-$utf8_offset,100);
$post = min($len-$utf8_idx-$utf8_len,100);
if ($pre>50 && $post>50) {
$pre = $post = 50;
} else if ($pre>50) {
$pre = min($pre,100-$post);
} else if ($post>50) {
$post = min($post, 100-$pre);
} else if ($offset == 0) {
// both are less than 50, means the context is the whole string
// make it so and break out of this loop - there is no need for the
// complex snippet calculations
$snippets = array($text);
break;
}
// establish context start and end points, try to append to previous
// context if possible
$start = $utf8_idx - $pre;
$append = ($start < $end) ? $end : false; // still the end of the previous context snippet
$end = $utf8_idx + $utf8_len + $post; // now set it to the end of this context
if ($append) {
$snippets[count($snippets)-1] .= utf8_substr($text,$append,$end-$append);
} else {
$snippets[] = utf8_substr($text,$start,$end-$start);
}
// set $offset for next match attempt
// continue matching after the current match
// if the current match is not the longest possible match starting at the current offset
// this prevents further matching of this snippet but for possible matches of length
// smaller than match length + context (at least 50 characters) this match is part of the context
$utf8_offset = $utf8_idx + $utf8_len;
$offset = $idx + strlen(utf8_substr($text,$utf8_idx,$utf8_len));
$offset = utf8_correctIdx($text,$offset);
}
$m = "\1";
$snippets = preg_replace('/'.$re1.'/iu',$m.'$1'.$m,$snippets);
$snippet = preg_replace('/'.$m.'([^'.$m.']*?)'.$m.'/iu','<strong class="search_hit">$1</strong>',hsc(join('... ',$snippets)));
$evdata['snippet'] = $snippet;
}
$evt->advise_after();
unset($evt);
return $evdata['snippet'];
}
/**
* Wraps a search term in regex boundary checks.
*
* @param string $term
* @return string
*/
function ft_snippet_re_preprocess($term) {
// do not process asian terms where word boundaries are not explicit
if(preg_match('/'.IDX_ASIAN.'/u',$term)){
return $term;
}
if (UTF8_PROPERTYSUPPORT) {
// unicode word boundaries
// see http://stackoverflow.com/a/2449017/172068
$BL = '(?<!\pL)';
$BR = '(?!\pL)';
} else {
// not as correct as above, but at least won't break
$BL = '\b';
$BR = '\b';
}
if(substr($term,0,2) == '\\*'){
$term = substr($term,2);
}else{
$term = $BL.$term;
}
if(substr($term,-2,2) == '\\*'){
$term = substr($term,0,-2);
}else{
$term = $term.$BR;
}
if($term == $BL || $term == $BR || $term == $BL.$BR) $term = '';
return $term;
}
/**
* Combine found documents and sum up their scores
*
* This function is used to combine searched words with a logical
* AND. Only documents available in all arrays are returned.
*
* based upon PEAR's PHP_Compat function for array_intersect_key()
*
* @param array $args An array of page arrays
* @return array
*/
function ft_resultCombine($args){
$array_count = count($args);
if($array_count == 1){
return $args[0];
}
$result = array();
if ($array_count > 1) {
foreach ($args[0] as $key => $value) {
$result[$key] = $value;
for ($i = 1; $i !== $array_count; $i++) {
if (!isset($args[$i][$key])) {
unset($result[$key]);
break;
}
$result[$key] += $args[$i][$key];
}
}
}
return $result;
}
/**
* Unites found documents and sum up their scores
*
* based upon ft_resultCombine() function
*
* @param array $args An array of page arrays
* @return array
*
* @author Kazutaka Miyasaka <kazmiya@gmail.com>
*/
function ft_resultUnite($args) {
$array_count = count($args);
if ($array_count === 1) {
return $args[0];
}
$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 $args An array of page arrays
* @return array
*
* @author Kazutaka Miyasaka <kazmiya@gmail.com>
*/
function ft_resultComplement($args) {
$array_count = count($args);
if ($array_count === 1) {
return $args[0];
}
$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>
*
* @param Doku_Indexer $Indexer
* @param string $query search query
* @return array of search formulas
*/
function ft_queryParser($Indexer, $query){
/**
* 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 (or pipe symbol '|', 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+:, W-:, W_: - word (underscore: no need to highlight)
* P+:, P-: - phrase (minus sign: logically in NOT group)
* N+:, 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 = $not.ft_termParser($Indexer, $matches[2], 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 pipe symbols as 'OR' operators
$term = str_replace('|', ' or ', $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
$parsed .= 'NOT('.ft_termParser($Indexer, $matches[1]).')';
} else {
// word-include
$parsed .= ft_termParser($Indexer, $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);
// adjustment: make highlightings right
$parens_level = 0;
$notgrp_levels = array();
$parsed_query_new = '';
$tokens = preg_split('/(NOT\(|[()])/u', $parsed_query, -1, PREG_SPLIT_DELIM_CAPTURE | PREG_SPLIT_NO_EMPTY);
foreach ($tokens as $token) {
if ($token === 'NOT(') {
$notgrp_levels[] = ++$parens_level;
} elseif ($token === '(') {
++$parens_level;
} elseif ($token === ')') {
if ($parens_level-- === end($notgrp_levels)) array_pop($notgrp_levels);
} elseif (count($notgrp_levels) % 2 === 1) {
// turn highlight-flag off if terms are logically in "NOT" group
$token = preg_replace('/([WPN])\+\:/u', '$1-:', $token);
}
$parsed_query_new .= $token;
}
$parsed_query = $parsed_query_new;
/**
* 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 ($last_ope !== false && $ope_precedence[$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 'N-:':
$q['notns'][] = $body; // for backward compatibility
break;
case 'W_:':
$q['words'][] = $body;
break;
case 'W-:':
$q['words'][] = $body;
$q['not'][] = $body; // for backward compatibility
break;
case 'W+:':
$q['words'][] = $body;
$q['highlight'][] = $body;
$q['and'][] = $body; // for backward compatibility
break;
case 'P-:':
$q['phrases'][] = $body;
break;
case 'P+:':
$q['phrases'][] = $body;
$q['highlight'][] = $body;
break;
}
}
foreach (array('words', 'phrases', 'highlight', 'ns', 'notns', 'and', 'not') as $key) {
$q[$key] = empty($q[$key]) ? array() : array_values(array_unique($q[$key]));
}
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>
*
* @param Doku_Indexer $Indexer
* @param string $term
* @param bool $consider_asian
* @param bool $phrase_mode
* @return string
*/
function ft_termParser($Indexer, $term, $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) {
$phrase_mode = $phrase_mode ? true : preg_match('/'.IDX_ASIAN.'/u', $word);
$parsed .= ft_termParser($Indexer, $word, false, $phrase_mode);
}
} else {
$term_noparen = str_replace(array('(', ')'), ' ', $term);
$words = $Indexer->tokenizer($term_noparen, true);
// 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 :