1 <?php rcs_id('$Id: TextSearchQuery.php,v 1.30 2007-07-14 12:31:00 rurban Exp $');
3 * A text search query, converting queries to PCRE and SQL matchers.
5 * This represents an enhanced "Google-like" text search query:
7 * <dt> default: case-insensitive glob-style search with special operators OR AND NOT -
9 * <dd> Match strings containing the substring 'wiki', and NOT containing the
11 * <dt> wiki word or page
12 * <dd> Match strings containing the substring 'wiki' AND either the substring
13 * 'word' OR the substring 'page'.
14 * <dt> auto-detect regex hints, glob-style or regex-style, and converts them
15 * to PCRE and SQL matchers:
16 * <dd> "^word$" => EXACT(word)
17 * <dd> "^word" => STARTS_WITH(word)
18 * <dd> "word*" => STARTS_WITH(word)
19 * <dd> "*word" => ENDS_WITH(word)
20 * <dd> "/^word.* /" => REGEX(^word.*)
21 * <dd> "word*word" => REGEX(word.*word)
24 * The full query syntax, in order of precedence, is roughly:
26 * The unary 'NOT' or '-' operator (they are equivalent) negates the
27 * following search clause.
29 * Search clauses may be joined with the (left-associative) binary operators
30 * 'AND' and 'OR'. (case-insensitive)
32 * Two adjoining search clauses are joined with an implicit 'AND'. This has
33 * lower precedence than either an explicit 'AND' or 'OR', so "a b OR c"
34 * parses as "a AND ( b OR c )", while "a AND b OR c" parses as
35 * "( a AND b ) OR c" (due to the left-associativity of 'AND' and 'OR'.)
37 * Search clauses can be grouped with parentheses.
39 * Phrases (or other things which don't look like words) can be forced to
40 * be interpreted as words by quoting them, either with single (') or double (")
41 * quotes. If you wan't to include the quote character within a quoted string,
42 * double-up on the quote character: 'I''m hungry' is equivalent to
45 * Force regex on "re:word" => posix-style, "/word/" => pcre-style
46 * or use regex='glob' to use file wildcard-like matching. (not yet)
48 * The parsed tree is then converted to the needed PCRE (highlight,
49 * simple backends) or SQL functions.
51 * @author: Jeff Dairiki
52 * @author: Reini Urban (case and regex detection, enhanced sql callbacks)
55 // regex-style: 'auto', 'none', 'glob', 'posix', 'pcre', 'sql'
56 define ('TSQ_REGEX_NONE', 0);
57 define ('TSQ_REGEX_AUTO', 1);
58 define ('TSQ_REGEX_POSIX',2);
59 define ('TSQ_REGEX_GLOB', 4);
60 define ('TSQ_REGEX_PCRE', 8);
61 define ('TSQ_REGEX_SQL', 16);
63 class TextSearchQuery {
67 * @param $search_query string The query. Syntax is as described above.
68 * Note that an empty $search_query will match anything.
69 * @param $case_exact boolean
70 * @param $regex string one of 'auto', 'none', 'glob', 'posix', 'pcre', 'sql'
71 * @see TextSearchQuery
73 function TextSearchQuery($search_query, $case_exact=false, $regex='auto') {
74 if ($regex == 'none' or !$regex)
76 elseif (defined("TSQ_REGEX_".strtoupper($regex)))
77 $this->_regex = constant("TSQ_REGEX_".strtoupper($regex));
79 trigger_error(fmt("Unsupported argument: %s=%s", 'regex', $regex));
82 $this->_case_exact = $case_exact;
83 if ($regex != 'pcre') {
84 $parser = new TextSearchQuery_Parser;
85 $this->_tree = $parser->parse($search_query, $case_exact, $this->_regex);
86 $this->_optimize(); // broken under certain circumstances: "word -word -word"
87 if (defined("FULLTEXTSEARCH_STOPLIST"))
88 $this->_stoplist = FULLTEXTSEARCH_STOPLIST;
89 else // default stoplist, localizable.
90 $this->_stoplist = _("(A|An|And|But|By|For|From|In|Is|It|Of|On|Or|The|To|With)");
93 $this->_tree = new TextSearchQuery_node_regex_pcre($search_query);
94 if (preg_match("/^\/(.*)\/(\w*)$/", $search_query, $m)) {
95 $this->_tree->word = $m[1];
96 $this->_regex_modifier = $m[2]; // overrides case_exact
101 function getType() { return 'text'; }
103 function _optimize() {
104 $this->_tree = $this->_tree->optimize();
108 * Get a PCRE regexp which matches the query.
110 function asRegexp() {
111 if (!isset($this->_regexp)) {
112 if (!isset($this->_regex_modifier))
113 $this->_regex_modifier = ($this->_case_exact?'':'i').'sS';
115 $this->_regexp = '/' . $this->_tree->regexp() . '/'.$this->_regex_modifier;
117 $this->_regexp = '/^' . $this->_tree->regexp() . '/'.$this->_regex_modifier;
119 return $this->_regexp;
123 * Match query against string.
125 * @param $string string The string to match.
126 * @return boolean True if the string matches the query.
128 function match($string) {
129 return preg_match($this->asRegexp(), $string);
132 /* How good does it match? Returns a number */
133 function score($string) {
136 foreach (array_unique($this->_tree->highlight_words()) as $word) {
137 if ($nummatch = preg_match_all("/".preg_quote($word, '/')."/".$this->_regex_modifier, $string, $out))
138 $score += ($i-- * $nummatch);
140 return min(1.0, $score / 10.0);
145 * Get a regular expression suitable for highlighting matched words.
147 * This returns a PCRE regular expression which matches any non-negated
150 * @return string The PCRE regexp.
152 function getHighlightRegexp() {
153 if (!isset($this->_hilight_regexp)) {
154 $words = array_unique($this->_tree->highlight_words());
156 $this->_hilight_regexp = false;
158 foreach ($words as $key => $word)
159 $words[$key] = preg_quote($word, '/');
160 $this->_hilight_regexp = '(?'.($this->_case_exact?'':'i').':' . join('|', $words) . ')';
163 return $this->_hilight_regexp;
167 * Make an SQL clause which matches the query.
168 * Deprecated, use makeSqlClauseObj instead.
170 * @param $make_sql_clause_cb WikiCallback
171 * A callback which takes a single word as an argument and
172 * returns an SQL clause which will match exactly those records
173 * containing the word. The word passed to the callback will always
174 * be in all lower case.
176 * Support db-specific extensions, like MATCH AGAINST or REGEX
177 * mysql => 4.0.1 can also do Google: MATCH AGAINST IN BOOLEAN MODE
178 * by using makeSqlClauseObj
182 * function sql_title_match($word) {
183 * return sprintf("LOWER(title) like '%s'",
184 * addslashes($word));
189 * $query = new TextSearchQuery("wiki -page");
190 * $cb = new WikiFunctionCb('sql_title_match');
191 * $sql_clause = $query->makeSqlClause($cb);
193 * This will result in $sql_clause containing something like
194 * "(LOWER(title) like 'wiki') AND NOT (LOWER(title) like 'page')".
196 * @return string The SQL clause.
198 function makeSqlClause($sql_clause_cb) {
199 $this->_sql_clause_cb = $sql_clause_cb;
200 return $this->_sql_clause($this->_tree);
202 // deprecated: use _sql_clause_obj now.
203 function _sql_clause($node) {
205 case 'WORD': // word => %word%
206 return $this->_sql_clause_cb->call($node->word);
208 return "NOT (" . $this->_sql_clause($node->leaves[0]) . ")";
211 $subclauses = array();
212 foreach ($node->leaves as $leaf)
213 $subclauses[] = "(" . $this->_sql_clause($leaf) . ")";
214 return join(" $node->op ", $subclauses);
216 assert($node->op == 'VOID');
221 /** Get away with the callback and use a db-specific search class instead.
222 * @see WikiDB_backend_PearDB_search
224 function makeSqlClauseObj(&$sql_search_cb) {
225 $this->_sql_clause_cb = $sql_search_cb;
226 return $this->_sql_clause_obj($this->_tree);
229 function _sql_clause_obj($node) {
232 return "NOT (" . $this->_sql_clause_cb->call($node->leaves[0]) . ")";
235 $subclauses = array();
236 foreach ($node->leaves as $leaf)
237 $subclauses[] = "(" . $this->_sql_clause_obj($leaf) . ")";
238 return join(" $node->op ", $subclauses);
244 return $this->_sql_clause_cb->call($node);
249 postgresql tsearch2 uses no WHERE operators, just & | and ! in the searchstring
251 function makeTsearch2SqlClauseObj(&$sql_search_cb) {
252 $this->_sql_clause_cb = $sql_search_cb;
253 return $this->_Tsearch2Sql_clause_obj($this->_tree);
256 function _Tsearch2Sql_clause_obj($node) {
257 // TODO: "such a phrase"
260 return "!" . $node->leaves[0];
262 $subclauses = array();
263 foreach ($node->leaves as $leaf)
264 $subclauses[] = $this->_Tsearch2Sql_clause_obj($leaf);
265 return join("&", $subclauses);
267 $subclauses = array();
268 foreach ($node->leaves as $leaf)
269 $subclauses[] = $this->_Tsearch2Sql_clause_obj($leaf);
270 return join("|", $subclauses);
276 return $this->_sql_clause_cb->call($node);
280 function sql() { return '%'.$this->_sql_quote($this->word).'%'; }
283 * Get printable representation of the parse tree.
285 * This is for debugging only.
286 * @return string Printable parse tree.
288 function asString() {
289 return $this->_as_string($this->_tree);
292 function _as_string($node, $indent = '') {
295 return $indent . "WORD: $node->word";
297 return $indent . "VOID";
299 return $indent . "ALL";
301 $lines = array($indent . $node->op . ":");
303 foreach ($node->leaves as $leaf)
304 $lines[] = $this->_as_string($leaf, $indent);
305 return join("\n", $lines);
311 * This is a TextSearchQuery which matches nothing.
313 class NullTextSearchQuery extends TextSearchQuery {
315 * Create a new query.
317 * @see TextSearchQuery
319 function NullTextSearchQuery() {}
320 function asRegexp() { return '/^(?!a)a/x'; }
321 function match($string) { return false; }
322 function getHighlightRegexp() { return ""; }
323 function makeSqlClause($make_sql_clause_cb) { return "(1 = 0)"; }
324 function asString() { return "NullTextSearchQuery"; }
328 * A simple algebraic matcher for numeric attributes.
329 * NumericSearchQuery can do ("population < 20000 and area > 1000000", array("population", "area"))
330 * ->match(array('population' => 100000, 'area' => 10000000))
332 * Supports all mathematical PHP comparison operators, plus ':=' for equality.
333 * "(x < 2000000 and x >= 10000) or (x >= 100 and x < 2000)"
334 * "x := 100000" is the same as "x == 100000"
336 * Since this is basic numerics only, we simply try to get away with
337 * replacing the variable values at the right positions and do an eval then.
339 * @package NumericSearchQuery
340 * @author Reini Urban
341 * @see SemanticAttributeSearchQuery
343 class NumericSearchQuery
346 * Create a new query.
347 * NumericSearchQuery("population > 20000 or population < 200", "population")
348 * NumericSearchQuery("population < 20000 and area > 1000000", array("population", "area"))
350 * With a single variable it is easy: The valid name must be matched elsewhere, just
351 * replace the given number in match in the query.
354 * With matching a struct we need strict names, no * as name is allowed.
355 * So always when the placeholder is an array, the names of the target struct must match
356 * and all vars be defined. Use the method can_match($struct) therefore.
359 * @param $search_query string A numerical query with placeholders as variable.
360 * @param $placeholders array or string All placeholders in the query must be defined
361 * here, and will be replaced by the matcher.
363 function NumericSearchQuery($search_query, $placeholders) {
364 // added some basic security checks against user input
365 $this->_query = $search_query;
366 $this->_placeholders = $placeholders;
368 // we should also allow the M_ constants
369 $this->_allowed_functions = explode(':','abs:acos:acosh:asin:asinh:atan2:atan:atanh:base_convert:bindec:ceil:cos:cosh:decbin:dechex:decoct:deg2rad:exp:expm1:floor:fmod:getrandmax:hexdec:hypot:is_finite:is_infinite:is_nan:lcg_value:log10:log1p:log:max:min:mt_getrandmax:mt_rand:mt_srand:octdec:pi:pow:rad2deg:rand:round:sin:sinh:sqrt:srand:tan:tanh');
370 $this->_allowed_operators = explode(',', '-,<,<=,>,>=,==,!=,*,+,/,(,),%,and,or,xor,<<,>>,===,!==,&,^,|,&&,||');
371 $this->_parser_check = array();
372 // check should be fast, so make a hash
373 foreach ($this->_allowed_functions as $f)
374 $this->_parser_check[$f] = 1;
375 foreach ($this->_allowed_operators as $f)
376 $this->_parser_check[$f] = 1;
377 if (is_array($placeholders))
378 foreach ($placeholders as $f)
379 $this->_parser_check[$f] = 1;
380 else $this->_parser_check[$placeholders] = 1;
382 // This is a speciality: := looks like the attribute definition and is
383 // therefore a dummy check for this definition.
384 // php-4.2.2 has a problem with /\b:=\b/ matching "population := 1223400"
385 $this->_query = preg_replace("/:=/", "==", $this->_query);
386 $this->_query = $this->check_query($this->_query);
389 function getType() { return 'numeric'; }
392 * Check the symbolic definition query against unwanted functions and characters.
393 * "population < 20000 and area > 1000000" vs
394 * "area > 1000000 and mail($me,file("/etc/passwd"),...)"
395 * http://localhost/wikicvs/SemanticSearch?attribute=*&attr_op=<0 and find(1)>&s=-0.01&start_debug=1
397 function check_query ($query) {
398 $tmp = $query; // check for all function calls, in case the tokenizer is not available.
399 while (preg_match("/([a-z][a-z0-9]+)\s*\((.*)$/i", $tmp, $m)) {
400 if (!in_array($m[1], $this->_allowed_functions)
401 and !in_array($m[1], $this->_allowed_operators))
403 trigger_error("Illegal function in query: ".$m[1], E_USER_WARNING);
409 // Strictly check for illegal functions and operators, which are no placeholders.
410 if (function_exists('token_get_all')) {
411 $parsed = token_get_all("<?$query?>");
412 foreach ($parsed as $x) { // flat, non-recursive array
413 if (is_string($x) and !isset($this->_parser_check[$x])) {
414 // single char op or name
415 trigger_error("Illegal string or operator in query: \"$x\"", E_USER_WARNING);
418 elseif (is_array($x)) {
419 $n = token_name($x[0]);
420 if ($n == 'T_OPEN_TAG' or $n == 'T_WHITESPACE'
421 or $n == 'T_CLOSE_TAG' or $n == 'T_LNUMBER'
422 or $n == 'T_CONST' or $n == 'T_DNUMBER' ) continue;
423 if ($n == 'T_VARIABLE') { // but we do allow consts
424 trigger_error("Illegal variable in query: \"$x[1]\"", E_USER_WARNING);
427 if (is_string($x[1]) and !isset($this->_parser_check[$x[1]])) {
428 // multi-char char op or name
429 trigger_error("Illegal $n in query: \"$x[1]\"", E_USER_WARNING);
434 //echo "$query <br>";
435 //$this->_parse_token($parsed);
440 { T_OPEN_TAG "<?"} { T_STRING "_x"} { T_WHITESPACE " "} ">" { T_WHITESPACE " "} { T_LNUMBER "0"} { T_CLOSE_TAG "?>"}
441 Interesting: on-char ops, as ">" are not tokenized.
443 { T_OPEN_TAG "< ?" } { T_STRING "_x" } { T_WHITESPACE " " } { T_IS_SMALLER_OR_EQUAL "<=" } { T_WHITESPACE " " } { T_LNUMBER "0" } { T_CLOSE_TAG "?>" }
446 // Detect illegal characters besides nums, words and ops.
447 // So attribute names can not be utf-8
448 $c = "/([^\d\w.,\s".preg_quote(join("",$this->_allowed_operators),"/")."])/";
449 if (preg_match($c, $query, $m)) {
450 trigger_error("Illegal character in query: ".$m[1], E_USER_WARNING);
458 * Check the bound, numeric-only query against unwanted functions and sideeffects.
459 * "4560000 < 20000 and 1456022 > 1000000"
461 function _live_check () {
462 // TODO: check $this->_workquery again?
463 return !empty($this->_workquery);
467 * A numeric query can only operate with predefined variables. "x < 0 and y < 1"
469 * @return array The names as array of strings. => ('x', 'y') the placeholders.
472 if(is_array($this->_placeholders)) return $this->_placeholders;
473 else return array($this->_placeholders);
477 * Strip non-numeric chars from the variable (as the groupseperator) and replace
478 * it in the symbolic query for evaluation.
481 * @param $value number A numerical value: integer, float or string.
482 * @param $x string The variable name to be replaced in the query.
485 function _bind($value, $x) {
486 // TODO: check is_number, is_float, is_integer and do casting
487 $this->_bound[] = array('linkname' => $x,
488 'linkvalue' => $value);
489 $value = preg_replace("/[^-+0123456789.,]/", "", $value);
490 //$c = "/\b".preg_quote($x,"/")."\b/";
491 $this->_workquery = preg_replace("/\b".preg_quote($x,"/")."\b/", $value, $this->_workquery);
492 // FIXME: do again a final check. now only numbers and some operators are allowed.
493 return $this->_workquery;
496 /* array of successfully bound vars, and in case of success, the resulting vars
499 return $this->_bound;
503 * With an array of placeholders we need a hash to check against, if all required names are given.
504 * Purpose: Be silent about missing vars, just return false.
507 * @param $variable string or hash of name => value The keys must satisfy all placeholders in the definition.
508 * We want the full hash and not just the keys because a hash check is faster than the array of keys check.
511 function can_match(&$variables) {
512 if (empty($this->_query))
514 $p =& $this->_placeholders;
515 if (!is_array($variables) and !is_array($p))
516 return $variables == $p; // This was easy.
517 // Check if all placeholders have definitions. can be overdefined but not underdefined.
519 if (!isset($variables[$p])) return false;
522 if (!isset($variables[$x])) return false;
529 * We can match against a single variable or against a hash of variables.
530 * With one placeholder we need just a number.
531 * With an array of placeholders we need a hash.
534 * @param $variable number or array of name => value The keys must satisfy all placeholders in the definition.
537 function match(&$variable) {
538 $p =& $this->_placeholders;
539 $this->_workquery = $this->_query;
541 if (is_array($variable)) { // which var to match? we cannot decide this here
542 if (!isset($variable[$p]))
543 trigger_error("Required NumericSearchQuery->match variable $x not defined.", E_USER_ERROR);
544 $this->_bind($variable[$p], $p);
546 $this->_bind($variable, $p);
550 if (!isset($variable[$x]))
551 trigger_error("Required NumericSearchQuery->match variable $x not defined.", E_USER_ERROR);
552 $this->_bind($variable[$x], $x);
555 if (!$this->_live_check()) // check returned an error
557 $search = $this->_workquery;
559 //if (DEBUG & _DEBUG_VERBOSE)
560 // trigger_error("\$result = (boolean)($search);", E_USER_NOTICE);
561 // We might have a numerical problem:
562 // php-4.2.2 eval'ed as module: "9.636e+08 > 1000" false;
563 // php-5.1.2 cgi true, 4.2.2 cgi true
564 eval("\$result = (boolean)($search);");
565 if ($result and is_array($p)) {
566 return $this->_bound();
573 ////////////////////////////////////////////////////////////////
575 // Remaining classes are private.
577 ////////////////////////////////////////////////////////////////
579 * Virtual base class for nodes in a TextSearchQuery parse tree.
581 * Also serves as a 'VOID' (contentless) node.
583 class TextSearchQuery_node
588 * Optimize this node.
589 * @return object Optimized node.
591 function optimize() {
596 * @return regexp matching this node.
603 * @param bool True if this node has been negated (higher in the parse tree.)
604 * @return array A list of all non-negated words contained by this node.
606 function highlight_words($negated = false) {
610 function sql() { return $this->word; }
612 function _sql_quote() {
614 $word = preg_replace('/(?=[%_\\\\])/', "\\", $this->word);
615 return $request->_dbi->_backend->qstr($word);
622 class TextSearchQuery_node_word
623 extends TextSearchQuery_node
627 function TextSearchQuery_node_word($word) {
631 return '(?=.*' . preg_quote($this->word, '/') . ')';
633 function highlight_words ($negated = false) {
634 return $negated ? array() : array($this->word);
636 function sql() { return '%'.$this->_sql_quote($this->word).'%'; }
639 class TextSearchQuery_node_all
640 extends TextSearchQuery_node {
642 function regexp() { return '(?=.*)'; }
643 function sql() { return '%'; }
645 class TextSearchQuery_node_starts_with
646 extends TextSearchQuery_node_word {
647 var $op = "STARTS_WITH";
648 function regexp() { return '(?=.*\b' . preg_quote($this->word, '/') . ')'; }
649 function sql () { return $this->_sql_quote($this->word).'%'; }
652 class TextSearchQuery_node_ends_with
653 extends TextSearchQuery_node_word {
654 var $op = "ENDS_WITH";
655 function regexp() { return '(?=.*' . preg_quote($this->word, '/') . '\b)'; }
656 function sql () { return '%'.$this->_sql_quote($this->word); }
659 class TextSearchQuery_node_exact
660 extends TextSearchQuery_node_word {
662 function regexp() { return '(?=\b' . preg_quote($this->word, '/') . '\b)'; }
663 function sql () { return $this->_sql_squote($this->word); }
666 class TextSearchQuery_node_regex // posix regex. FIXME!
667 extends TextSearchQuery_node_word {
668 var $op = "REGEX"; // using REGEXP or ~ extension
669 function regexp() { return '(?=.*\b' . $this->word . '\b)'; }
670 function sql () { return $this->_sql_quote($this->word); }
673 class TextSearchQuery_node_regex_glob
674 extends TextSearchQuery_node_regex {
675 var $op = "REGEX_GLOB";
676 function regexp() { return '(?=.*\b' . glob_to_pcre($this->word) . '\b)'; }
679 class TextSearchQuery_node_regex_pcre // how to handle pcre modifiers? /i
680 extends TextSearchQuery_node_regex {
681 var $op = "REGEX_PCRE";
682 function regexp() { return $this->word; }
685 class TextSearchQuery_node_regex_sql
686 extends TextSearchQuery_node_regex {
687 var $op = "REGEX_SQL"; // using LIKE
688 function regexp() { return str_replace(array("/%/","/_/"), array(".*","."), $this->word); }
689 function sql() { return $this->word; }
695 class TextSearchQuery_node_not
696 extends TextSearchQuery_node
700 function TextSearchQuery_node_not($leaf) {
701 $this->leaves = array($leaf);
704 function optimize() {
705 $leaf = &$this->leaves[0];
706 $leaf = $leaf->optimize();
707 if ($leaf->op == 'NOT')
708 return $leaf->leaves[0]; // ( NOT ( NOT x ) ) -> x
713 $leaf = &$this->leaves[0];
714 return '(?!' . $leaf->regexp() . ')';
717 function highlight_words ($negated = false) {
718 return $this->leaves[0]->highlight_words(!$negated);
723 * Virtual base class for 'AND' and 'OR conjoins.
725 class TextSearchQuery_node_binop
726 extends TextSearchQuery_node
728 function TextSearchQuery_node_binop($leaves) {
729 $this->leaves = $leaves;
732 function _flatten() {
733 // This flattens e.g. (AND (AND a b) (OR c d) e)
734 // to (AND a b e (OR c d))
736 foreach ($this->leaves as $leaf) {
737 $leaf = $leaf->optimize();
738 if ($this->op == $leaf->op)
739 $flat = array_merge($flat, $leaf->leaves);
743 $this->leaves = $flat;
746 function optimize() {
748 assert(!empty($this->leaves));
749 if (count($this->leaves) == 1)
750 return $this->leaves[0]; // (AND x) -> x
754 function highlight_words($negated = false) {
756 foreach ($this->leaves as $leaf)
757 array_splice($words,0,0,
758 $leaf->highlight_words($negated));
764 * A (possibly multi-argument) 'AND' conjoin.
766 class TextSearchQuery_node_and
767 extends TextSearchQuery_node_binop
771 function optimize() {
774 // Convert (AND (NOT a) (NOT b) c d) into (AND (NOT (OR a b)) c d).
775 // Since OR's are more efficient for regexp matching:
776 // (?!.*a)(?!.*b) vs (?!.*(?:a|b))
778 // Suck out the negated leaves.
780 foreach ($this->leaves as $key => $leaf) {
781 if ($leaf->op == 'NOT') {
782 $nots[] = $leaf->leaves[0];
783 unset($this->leaves[$key]);
787 // Combine the negated leaves into a single negated or.
789 $node = ( new TextSearchQuery_node_not
790 (new TextSearchQuery_node_or($nots)) );
791 array_unshift($this->leaves, $node->optimize());
794 assert(!empty($this->leaves));
795 if (count($this->leaves) == 1)
796 return $this->leaves[0]; // (AND x) -> x
801 * Either we need all combinations of all words to be position independent,
802 * or we have to use multiple match calls for each AND
803 * (AND x y) => /(?(:x)(:y))|(?(:y)(:x))/
807 foreach ($this->leaves as $leaf)
808 $regexp .= $leaf->regexp();
814 * A (possibly multi-argument) 'OR' conjoin.
816 class TextSearchQuery_node_or
817 extends TextSearchQuery_node_binop
822 // We will combine any of our direct descendents which are WORDs
823 // into a single (?=.*(?:word1|word2|...)) regexp.
828 foreach ($this->leaves as $leaf) {
829 if ($leaf->op == 'WORD')
830 $words[] = preg_quote($leaf->word, '/');
832 $regexps[] = $leaf->regexp();
836 array_unshift($regexps,
837 '(?=.*' . $this->_join($words) . ')');
839 return $this->_join($regexps);
842 function _join($regexps) {
843 assert(count($regexps) > 0);
845 if (count($regexps) > 1)
846 return '(?:' . join('|', $regexps) . ')';
853 ////////////////////////////////////////////////////////////////
856 // op's (and, or, not) are forced to lowercase in the tokenizer.
858 ////////////////////////////////////////////////////////////////
859 define ('TSQ_TOK_BINOP', 1);
860 define ('TSQ_TOK_NOT', 2);
861 define ('TSQ_TOK_LPAREN', 4);
862 define ('TSQ_TOK_RPAREN', 8);
863 define ('TSQ_TOK_WORD', 16);
864 define ('TSQ_TOK_STARTS_WITH', 32);
865 define ('TSQ_TOK_ENDS_WITH', 64);
866 define ('TSQ_TOK_EXACT', 128);
867 define ('TSQ_TOK_REGEX', 256);
868 define ('TSQ_TOK_REGEX_GLOB', 512);
869 define ('TSQ_TOK_REGEX_PCRE', 1024);
870 define ('TSQ_TOK_REGEX_SQL', 2048);
871 define ('TSQ_TOK_ALL', 4096);
872 // all bits from word to the last.
873 define ('TSQ_ALLWORDS', (4096*2)-1 - (16-1));
875 class TextSearchQuery_Parser
878 * This is a simple recursive descent parser, based on the following grammar:
893 * atom : '(' list ')'
898 * The terminal tokens are:
905 * /[^-()\s][^()\s]* WORD
916 function parse ($search_expr, $case_exact=false, $regex=TSQ_REGEX_AUTO) {
917 $this->lexer = new TextSearchQuery_Lexer($search_expr, $case_exact, $regex);
918 $this->_regex = $regex;
919 $tree = $this->get_list('toplevel');
920 assert($this->lexer->eof());
925 function get_list ($is_toplevel = false) {
928 // token types we'll accept as words (and thus expr's) for the
929 // purpose of error recovery:
930 $accept_as_words = TSQ_TOK_NOT | TSQ_TOK_BINOP;
932 $accept_as_words |= TSQ_TOK_LPAREN | TSQ_TOK_RPAREN;
934 while ( ($expr = $this->get_expr())
935 || ($expr = $this->get_word($accept_as_words)) ) {
941 return new TextSearchQuery_node;
945 return new TextSearchQuery_node_and($list);
948 function get_expr () {
949 if ( !($expr = $this->get_atom()) )
952 $savedpos = $this->lexer->tell();
953 while ( ($op = $this->lexer->get(TSQ_TOK_BINOP)) ) {
954 if ( ! ($right = $this->get_atom()) ) {
959 $expr = new TextSearchQuery_node_and(array($expr, $right));
962 $expr = new TextSearchQuery_node_or(array($expr, $right));
965 $savedpos = $this->lexer->tell();
967 $this->lexer->seek($savedpos);
973 function get_atom() {
974 if ($word = $this->get_word(TSQ_ALLWORDS))
977 $savedpos = $this->lexer->tell();
978 if ( $this->lexer->get(TSQ_TOK_LPAREN) ) {
979 if ( ($list = $this->get_list()) && $this->lexer->get(TSQ_TOK_RPAREN) )
982 elseif ( $this->lexer->get(TSQ_TOK_NOT) ) {
983 if ( ($atom = $this->get_atom()) )
984 return new TextSearchQuery_node_not($atom);
986 $this->lexer->seek($savedpos);
990 function get_word($accept = TSQ_ALLWORDS) {
991 foreach (array("WORD","STARTS_WITH","ENDS_WITH","EXACT",
992 "REGEX","REGEX_GLOB","REGEX_PCRE","ALL") as $tok) {
993 $const = constant("TSQ_TOK_".$tok);
994 if ( $accept & $const and ($word = $this->lexer->get($const)) ) {
995 $classname = "TextSearchQuery_node_".strtolower($tok);
996 return new $classname($word);
1003 class TextSearchQuery_Lexer {
1004 function TextSearchQuery_Lexer ($query_str, $case_exact=false, $regex=TSQ_REGEX_AUTO) {
1005 $this->tokens = $this->tokenize($query_str, $case_exact, $regex);
1013 function seek($pos) {
1018 return $this->pos == count($this->tokens);
1022 * TODO: support more regex styles, esp. prefer the forced ones over auto
1025 function tokenize($string, $case_exact=false, $regex=TSQ_REGEX_AUTO) {
1027 $buf = $case_exact ? ltrim($string) : strtolower(ltrim($string));
1028 while (!empty($buf)) {
1029 if (preg_match('/^(and|or)\b\s*/i', $buf, $m)) {
1030 $val = strtolower($m[1]);
1031 $type = TSQ_TOK_BINOP;
1033 elseif (preg_match('/^(-|not\b)\s*/i', $buf, $m)) {
1034 $val = strtolower($m[1]);
1035 $type = TSQ_TOK_NOT;
1037 elseif (preg_match('/^([()])\s*/', $buf, $m)) {
1039 $type = $m[1] == '(' ? TSQ_TOK_LPAREN : TSQ_TOK_RPAREN;
1043 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
1044 and preg_match('/^\*\s*/', $buf, $m)) {
1046 $type = TSQ_TOK_ALL;
1049 elseif ($regex & (TSQ_REGEX_PCRE)
1050 and preg_match('/^\.\*\s*/', $buf, $m)) {
1052 $type = TSQ_TOK_ALL;
1055 elseif ($regex & (TSQ_REGEX_SQL)
1056 and preg_match('/^%\s*/', $buf, $m)) {
1058 $type = TSQ_TOK_ALL;
1062 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
1063 and preg_match('/^\^([^-()][^()\s]*)\s*/', $buf, $m)) {
1065 $type = TSQ_TOK_STARTS_WITH;
1068 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
1069 and preg_match('/^([^-()][^()\s]*)\*\s*/', $buf, $m)) {
1071 $type = TSQ_TOK_STARTS_WITH;
1074 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
1075 and preg_match('/^\*([^-()][^()\s]*)\s*/', $buf, $m)) {
1077 $type = TSQ_TOK_ENDS_WITH;
1080 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
1081 and preg_match('/^([^-()][^()\s]*)\$\s*/', $buf, $m)) {
1083 $type = TSQ_TOK_ENDS_WITH;
1086 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
1087 and preg_match('/^\^([^-()][^()\s]*)\$\s*/', $buf, $m)) {
1089 $type = TSQ_TOK_EXACT;
1093 elseif (preg_match('/^ " ( (?: [^"]+ | "" )* ) " \s*/x', $buf, $m)) {
1094 $val = str_replace('""', '"', $m[1]);
1095 $type = TSQ_TOK_WORD;
1098 elseif (preg_match("/^ ' ( (?:[^']+|'')* ) ' \s*/x", $buf, $m)) {
1099 $val = str_replace("''", "'", $m[1]);
1100 $type = TSQ_TOK_WORD;
1103 elseif (preg_match('/^([^-()][^()\s]*)\s*/', $buf, $m)) {
1105 $type = TSQ_TOK_WORD;
1108 assert(empty($buf));
1111 $buf = substr($buf, strlen($m[0]));
1113 /* refine the simple parsing from above: bla*bla, bla?bla, ...
1114 if ($regex and $type == TSQ_TOK_WORD) {
1115 if (substr($val,0,1) == "^")
1116 $type = TSQ_TOK_STARTS_WITH;
1117 elseif (substr($val,0,1) == "*")
1118 $type = TSQ_TOK_ENDS_WITH;
1119 elseif (substr($val,-1,1) == "*")
1120 $type = TSQ_TOK_STARTS_WITH;
1123 $tokens[] = array($type, $val);
1128 function get($accept) {
1129 if ($this->pos >= count($this->tokens))
1132 list ($type, $val) = $this->tokens[$this->pos];
1133 if (($type & $accept) == 0)
1141 // $Log: not supported by cvs2svn $
1142 // Revision 1.29 2007/07/14 12:03:38 rurban
1143 // support ranked search: simple score() function
1145 // Revision 1.28 2007/03/18 17:35:26 rurban
1148 // Revision 1.27 2007/01/21 23:27:32 rurban
1149 // Fix ->_backend->qstr()
1151 // Revision 1.26 2007/01/04 16:41:52 rurban
1152 // Improve error description. Fix the function parser for illegal functions, when the tokenizer cannot be used.
1154 // Revision 1.25 2007/01/03 21:22:34 rurban
1155 // add getType(). NumericSearchQuery::check Improve hacker detection using token_get_all(). Better support for multiple attributes. Add getVars().
1157 // Revision 1.24 2007/01/02 13:19:05 rurban
1158 // add NumericSearchQuery. change on pcre: no parsing done, detect modifiers
1160 // Revision 1.23 2006/04/13 19:30:44 rurban
1161 // make TextSearchQuery->_stoplist localizable and overridable within config.ini
1167 // c-basic-offset: 4
1168 // c-hanging-comment-ender-p: nil
1169 // indent-tabs-mode: nil