1 <?php rcs_id('$Id: TextSearchQuery.php,v 1.31 2007-09-12 19:36:47 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, '/')."/".
138 $this->_regex_modifier,
140 $score += ($i-- * $nummatch);
142 return min(1.0, $score / 10.0);
147 * Get a regular expression suitable for highlighting matched words.
149 * This returns a PCRE regular expression which matches any non-negated
152 * @return string The PCRE regexp.
154 function getHighlightRegexp() {
155 if (!isset($this->_hilight_regexp)) {
156 $words = array_unique($this->_tree->highlight_words());
158 $this->_hilight_regexp = false;
160 foreach ($words as $key => $word)
161 $words[$key] = preg_quote($word, '/');
162 $this->_hilight_regexp = '(?'.($this->_case_exact?'':'i').':'
163 . join('|', $words) . ')';
166 return $this->_hilight_regexp;
170 * Make an SQL clause which matches the query.
171 * Deprecated, use makeSqlClauseObj instead.
173 * @param $make_sql_clause_cb WikiCallback
174 * A callback which takes a single word as an argument and
175 * returns an SQL clause which will match exactly those records
176 * containing the word. The word passed to the callback will always
177 * be in all lower case.
179 * Support db-specific extensions, like MATCH AGAINST or REGEX
180 * mysql => 4.0.1 can also do Google: MATCH AGAINST IN BOOLEAN MODE
181 * by using makeSqlClauseObj
185 * function sql_title_match($word) {
186 * return sprintf("LOWER(title) like '%s'",
187 * addslashes($word));
192 * $query = new TextSearchQuery("wiki -page");
193 * $cb = new WikiFunctionCb('sql_title_match');
194 * $sql_clause = $query->makeSqlClause($cb);
196 * This will result in $sql_clause containing something like
197 * "(LOWER(title) like 'wiki') AND NOT (LOWER(title) like 'page')".
199 * @return string The SQL clause.
201 function makeSqlClause($sql_clause_cb) {
202 $this->_sql_clause_cb = $sql_clause_cb;
203 return $this->_sql_clause($this->_tree);
205 // deprecated: use _sql_clause_obj now.
206 function _sql_clause($node) {
208 case 'WORD': // word => %word%
209 return $this->_sql_clause_cb->call($node->word);
211 return "NOT (" . $this->_sql_clause($node->leaves[0]) . ")";
214 $subclauses = array();
215 foreach ($node->leaves as $leaf)
216 $subclauses[] = "(" . $this->_sql_clause($leaf) . ")";
217 return join(" $node->op ", $subclauses);
219 assert($node->op == 'VOID');
224 /** Get away with the callback and use a db-specific search class instead.
225 * @see WikiDB_backend_PearDB_search
227 function makeSqlClauseObj(&$sql_search_cb) {
228 $this->_sql_clause_cb = $sql_search_cb;
229 return $this->_sql_clause_obj($this->_tree);
232 function _sql_clause_obj($node) {
235 return "NOT (" . $this->_sql_clause_cb->call($node->leaves[0]) . ")";
238 $subclauses = array();
239 foreach ($node->leaves as $leaf)
240 $subclauses[] = "(" . $this->_sql_clause_obj($leaf) . ")";
241 return join(" $node->op ", $subclauses);
247 return $this->_sql_clause_cb->call($node);
252 postgresql tsearch2 uses no WHERE operators, just & | and ! in the searchstring
254 function makeTsearch2SqlClauseObj(&$sql_search_cb) {
255 $this->_sql_clause_cb = $sql_search_cb;
256 return $this->_Tsearch2Sql_clause_obj($this->_tree);
259 function _Tsearch2Sql_clause_obj($node) {
260 // TODO: "such a phrase"
263 return "!" . $node->leaves[0];
265 $subclauses = array();
266 foreach ($node->leaves as $leaf)
267 $subclauses[] = $this->_Tsearch2Sql_clause_obj($leaf);
268 return join("&", $subclauses);
270 $subclauses = array();
271 foreach ($node->leaves as $leaf)
272 $subclauses[] = $this->_Tsearch2Sql_clause_obj($leaf);
273 return join("|", $subclauses);
279 return $this->_sql_clause_cb->call($node);
283 function sql() { return '%'.$this->_sql_quote($this->word).'%'; }
286 * Get printable representation of the parse tree.
288 * This is for debugging only.
289 * @return string Printable parse tree.
291 function asString() {
292 return $this->_as_string($this->_tree);
295 function _as_string($node, $indent = '') {
298 return $indent . "WORD: $node->word";
300 return $indent . "VOID";
302 return $indent . "ALL";
304 $lines = array($indent . $node->op . ":");
306 foreach ($node->leaves as $leaf)
307 $lines[] = $this->_as_string($leaf, $indent);
308 return join("\n", $lines);
314 * This is a TextSearchQuery which matches nothing.
316 class NullTextSearchQuery extends TextSearchQuery {
318 * Create a new query.
320 * @see TextSearchQuery
322 function NullTextSearchQuery() {}
323 function asRegexp() { return '/^(?!a)a/x'; }
324 function match($string) { return false; }
325 function getHighlightRegexp() { return ""; }
326 function makeSqlClause($make_sql_clause_cb) { return "(1 = 0)"; }
327 function asString() { return "NullTextSearchQuery"; }
331 * A simple algebraic matcher for numeric attributes.
332 * NumericSearchQuery can do ("population < 20000 and area > 1000000", array("population", "area"))
333 * ->match(array('population' => 100000, 'area' => 10000000))
335 * Supports all mathematical PHP comparison operators, plus ':=' for equality.
336 * "(x < 2000000 and x >= 10000) or (x >= 100 and x < 2000)"
337 * "x := 100000" is the same as "x == 100000"
339 * Since this is basic numerics only, we simply try to get away with
340 * replacing the variable values at the right positions and do an eval then.
342 * @package NumericSearchQuery
343 * @author Reini Urban
344 * @see SemanticAttributeSearchQuery
346 class NumericSearchQuery
349 * Create a new query.
350 * NumericSearchQuery("population > 20000 or population < 200", "population")
351 * NumericSearchQuery("population < 20000 and area > 1000000", array("population", "area"))
353 * With a single variable it is easy: The valid name must be matched elsewhere, just
354 * replace the given number in match in the query.
357 * With matching a struct we need strict names, no * as name is allowed.
358 * So always when the placeholder is an array, the names of the target struct must match
359 * and all vars be defined. Use the method can_match($struct) therefore.
362 * @param $search_query string A numerical query with placeholders as variable.
363 * @param $placeholders array or string All placeholders in the query must be defined
364 * here, and will be replaced by the matcher.
366 function NumericSearchQuery($search_query, $placeholders) {
367 // added some basic security checks against user input
368 $this->_query = $search_query;
369 $this->_placeholders = $placeholders;
371 // we should also allow the M_ constants
372 $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');
373 $this->_allowed_operators = explode(',', '-,<,<=,>,>=,==,!=,*,+,/,(,),%,and,or,xor,<<,>>,===,!==,&,^,|,&&,||');
374 $this->_parser_check = array();
375 // check should be fast, so make a hash
376 foreach ($this->_allowed_functions as $f)
377 $this->_parser_check[$f] = 1;
378 foreach ($this->_allowed_operators as $f)
379 $this->_parser_check[$f] = 1;
380 if (is_array($placeholders))
381 foreach ($placeholders as $f)
382 $this->_parser_check[$f] = 1;
383 else $this->_parser_check[$placeholders] = 1;
385 // This is a speciality: := looks like the attribute definition and is
386 // therefore a dummy check for this definition.
387 // php-4.2.2 has a problem with /\b:=\b/ matching "population := 1223400"
388 $this->_query = preg_replace("/:=/", "==", $this->_query);
389 $this->_query = $this->check_query($this->_query);
392 function getType() { return 'numeric'; }
395 * Check the symbolic definition query against unwanted functions and characters.
396 * "population < 20000 and area > 1000000" vs
397 * "area > 1000000 and mail($me,file("/etc/passwd"),...)"
398 * http://localhost/wikicvs/SemanticSearch?attribute=*&attr_op=<0 and find(1)>&s=-0.01&start_debug=1
400 function check_query ($query) {
401 $tmp = $query; // check for all function calls, in case the tokenizer is not available.
402 while (preg_match("/([a-z][a-z0-9]+)\s*\((.*)$/i", $tmp, $m)) {
403 if (!in_array($m[1], $this->_allowed_functions)
404 and !in_array($m[1], $this->_allowed_operators))
406 trigger_error("Illegal function in query: ".$m[1], E_USER_WARNING);
412 // Strictly check for illegal functions and operators, which are no placeholders.
413 if (function_exists('token_get_all')) {
414 $parsed = token_get_all("<?$query?>");
415 foreach ($parsed as $x) { // flat, non-recursive array
416 if (is_string($x) and !isset($this->_parser_check[$x])) {
417 // single char op or name
418 trigger_error("Illegal string or operator in query: \"$x\"", E_USER_WARNING);
421 elseif (is_array($x)) {
422 $n = token_name($x[0]);
423 if ($n == 'T_OPEN_TAG' or $n == 'T_WHITESPACE'
424 or $n == 'T_CLOSE_TAG' or $n == 'T_LNUMBER'
425 or $n == 'T_CONST' or $n == 'T_DNUMBER' ) continue;
426 if ($n == 'T_VARIABLE') { // but we do allow consts
427 trigger_error("Illegal variable in query: \"$x[1]\"", E_USER_WARNING);
430 if (is_string($x[1]) and !isset($this->_parser_check[$x[1]])) {
431 // multi-char char op or name
432 trigger_error("Illegal $n in query: \"$x[1]\"", E_USER_WARNING);
437 //echo "$query <br>";
438 //$this->_parse_token($parsed);
443 { T_OPEN_TAG "<?"} { T_STRING "_x"} { T_WHITESPACE " "} ">" { T_WHITESPACE " "} { T_LNUMBER "0"} { T_CLOSE_TAG "?>"}
444 Interesting: on-char ops, as ">" are not tokenized.
446 { T_OPEN_TAG "< ?" } { T_STRING "_x" } { T_WHITESPACE " " } { T_IS_SMALLER_OR_EQUAL "<=" } { T_WHITESPACE " " } { T_LNUMBER "0" } { T_CLOSE_TAG "?>" }
449 // Detect illegal characters besides nums, words and ops.
450 // So attribute names can not be utf-8
451 $c = "/([^\d\w.,\s".preg_quote(join("",$this->_allowed_operators),"/")."])/";
452 if (preg_match($c, $query, $m)) {
453 trigger_error("Illegal character in query: ".$m[1], E_USER_WARNING);
461 * Check the bound, numeric-only query against unwanted functions and sideeffects.
462 * "4560000 < 20000 and 1456022 > 1000000"
464 function _live_check () {
465 // TODO: check $this->_workquery again?
466 return !empty($this->_workquery);
470 * A numeric query can only operate with predefined variables. "x < 0 and y < 1"
472 * @return array The names as array of strings. => ('x', 'y') the placeholders.
475 if(is_array($this->_placeholders)) return $this->_placeholders;
476 else return array($this->_placeholders);
480 * Strip non-numeric chars from the variable (as the groupseperator) and replace
481 * it in the symbolic query for evaluation.
484 * @param $value number A numerical value: integer, float or string.
485 * @param $x string The variable name to be replaced in the query.
488 function _bind($value, $x) {
489 // TODO: check is_number, is_float, is_integer and do casting
490 $this->_bound[] = array('linkname' => $x,
491 'linkvalue' => $value);
492 $value = preg_replace("/[^-+0123456789.,]/", "", $value);
493 //$c = "/\b".preg_quote($x,"/")."\b/";
494 $this->_workquery = preg_replace("/\b".preg_quote($x,"/")."\b/", $value, $this->_workquery);
495 // FIXME: do again a final check. now only numbers and some operators are allowed.
496 return $this->_workquery;
499 /* array of successfully bound vars, and in case of success, the resulting vars
502 return $this->_bound;
506 * With an array of placeholders we need a hash to check against, if all required names are given.
507 * Purpose: Be silent about missing vars, just return false.
510 * @param $variable string or hash of name => value The keys must satisfy all placeholders in the definition.
511 * We want the full hash and not just the keys because a hash check is faster than the array of keys check.
514 function can_match(&$variables) {
515 if (empty($this->_query))
517 $p =& $this->_placeholders;
518 if (!is_array($variables) and !is_array($p))
519 return $variables == $p; // This was easy.
520 // Check if all placeholders have definitions. can be overdefined but not underdefined.
522 if (!isset($variables[$p])) return false;
525 if (!isset($variables[$x])) return false;
532 * We can match against a single variable or against a hash of variables.
533 * With one placeholder we need just a number.
534 * With an array of placeholders we need a hash.
537 * @param $variable number or array of name => value The keys must satisfy all placeholders in the definition.
540 function match(&$variable) {
541 $p =& $this->_placeholders;
542 $this->_workquery = $this->_query;
544 if (is_array($variable)) { // which var to match? we cannot decide this here
545 if (!isset($variable[$p]))
546 trigger_error("Required NumericSearchQuery->match variable $x not defined.", E_USER_ERROR);
547 $this->_bind($variable[$p], $p);
549 $this->_bind($variable, $p);
553 if (!isset($variable[$x]))
554 trigger_error("Required NumericSearchQuery->match variable $x not defined.", E_USER_ERROR);
555 $this->_bind($variable[$x], $x);
558 if (!$this->_live_check()) // check returned an error
560 $search = $this->_workquery;
562 //if (DEBUG & _DEBUG_VERBOSE)
563 // trigger_error("\$result = (boolean)($search);", E_USER_NOTICE);
564 // We might have a numerical problem:
565 // php-4.2.2 eval'ed as module: "9.636e+08 > 1000" false;
566 // php-5.1.2 cgi true, 4.2.2 cgi true
567 eval("\$result = (boolean)($search);");
568 if ($result and is_array($p)) {
569 return $this->_bound();
576 ////////////////////////////////////////////////////////////////
578 // Remaining classes are private.
580 ////////////////////////////////////////////////////////////////
582 * Virtual base class for nodes in a TextSearchQuery parse tree.
584 * Also serves as a 'VOID' (contentless) node.
586 class TextSearchQuery_node
591 * Optimize this node.
592 * @return object Optimized node.
594 function optimize() {
599 * @return regexp matching this node.
606 * @param bool True if this node has been negated (higher in the parse tree.)
607 * @return array A list of all non-negated words contained by this node.
609 function highlight_words($negated = false) {
613 function sql() { return $this->word; }
615 function _sql_quote() {
617 $word = preg_replace('/(?=[%_\\\\])/', "\\", $this->word);
618 return $request->_dbi->_backend->qstr($word);
625 class TextSearchQuery_node_word
626 extends TextSearchQuery_node
630 function TextSearchQuery_node_word($word) {
634 return '(?=.*' . preg_quote($this->word, '/') . ')';
636 function highlight_words ($negated = false) {
637 return $negated ? array() : array($this->word);
639 function sql() { return '%'.$this->_sql_quote($this->word).'%'; }
642 class TextSearchQuery_node_all
643 extends TextSearchQuery_node {
645 function regexp() { return '(?=.*)'; }
646 function sql() { return '%'; }
648 class TextSearchQuery_node_starts_with
649 extends TextSearchQuery_node_word {
650 var $op = "STARTS_WITH";
651 function regexp() { return '(?=.*\b' . preg_quote($this->word, '/') . ')'; }
652 function sql () { return $this->_sql_quote($this->word).'%'; }
655 class TextSearchQuery_node_ends_with
656 extends TextSearchQuery_node_word {
657 var $op = "ENDS_WITH";
658 function regexp() { return '(?=.*' . preg_quote($this->word, '/') . '\b)'; }
659 function sql () { return '%'.$this->_sql_quote($this->word); }
662 class TextSearchQuery_node_exact
663 extends TextSearchQuery_node_word {
665 function regexp() { return '(?=\b' . preg_quote($this->word, '/') . '\b)'; }
666 function sql () { return $this->_sql_squote($this->word); }
669 class TextSearchQuery_node_regex // posix regex. FIXME!
670 extends TextSearchQuery_node_word {
671 var $op = "REGEX"; // using REGEXP or ~ extension
672 function regexp() { return '(?=.*\b' . $this->word . '\b)'; }
673 function sql () { return $this->_sql_quote($this->word); }
676 class TextSearchQuery_node_regex_glob
677 extends TextSearchQuery_node_regex {
678 var $op = "REGEX_GLOB";
679 function regexp() { return '(?=.*\b' . glob_to_pcre($this->word) . '\b)'; }
682 class TextSearchQuery_node_regex_pcre // how to handle pcre modifiers? /i
683 extends TextSearchQuery_node_regex {
684 var $op = "REGEX_PCRE";
685 function regexp() { return $this->word; }
688 class TextSearchQuery_node_regex_sql
689 extends TextSearchQuery_node_regex {
690 var $op = "REGEX_SQL"; // using LIKE
691 function regexp() { return str_replace(array("/%/","/_/"), array(".*","."), $this->word); }
692 function sql() { return $this->word; }
698 class TextSearchQuery_node_not
699 extends TextSearchQuery_node
703 function TextSearchQuery_node_not($leaf) {
704 $this->leaves = array($leaf);
707 function optimize() {
708 $leaf = &$this->leaves[0];
709 $leaf = $leaf->optimize();
710 if ($leaf->op == 'NOT')
711 return $leaf->leaves[0]; // ( NOT ( NOT x ) ) -> x
716 $leaf = &$this->leaves[0];
717 return '(?!' . $leaf->regexp() . ')';
720 function highlight_words ($negated = false) {
721 return $this->leaves[0]->highlight_words(!$negated);
726 * Virtual base class for 'AND' and 'OR conjoins.
728 class TextSearchQuery_node_binop
729 extends TextSearchQuery_node
731 function TextSearchQuery_node_binop($leaves) {
732 $this->leaves = $leaves;
735 function _flatten() {
736 // This flattens e.g. (AND (AND a b) (OR c d) e)
737 // to (AND a b e (OR c d))
739 foreach ($this->leaves as $leaf) {
740 $leaf = $leaf->optimize();
741 if ($this->op == $leaf->op)
742 $flat = array_merge($flat, $leaf->leaves);
746 $this->leaves = $flat;
749 function optimize() {
751 assert(!empty($this->leaves));
752 if (count($this->leaves) == 1)
753 return $this->leaves[0]; // (AND x) -> x
757 function highlight_words($negated = false) {
759 foreach ($this->leaves as $leaf)
760 array_splice($words,0,0,
761 $leaf->highlight_words($negated));
767 * A (possibly multi-argument) 'AND' conjoin.
769 class TextSearchQuery_node_and
770 extends TextSearchQuery_node_binop
774 function optimize() {
777 // Convert (AND (NOT a) (NOT b) c d) into (AND (NOT (OR a b)) c d).
778 // Since OR's are more efficient for regexp matching:
779 // (?!.*a)(?!.*b) vs (?!.*(?:a|b))
781 // Suck out the negated leaves.
783 foreach ($this->leaves as $key => $leaf) {
784 if ($leaf->op == 'NOT') {
785 $nots[] = $leaf->leaves[0];
786 unset($this->leaves[$key]);
790 // Combine the negated leaves into a single negated or.
792 $node = ( new TextSearchQuery_node_not
793 (new TextSearchQuery_node_or($nots)) );
794 array_unshift($this->leaves, $node->optimize());
797 assert(!empty($this->leaves));
798 if (count($this->leaves) == 1)
799 return $this->leaves[0]; // (AND x) -> x
804 * Either we need all combinations of all words to be position independent,
805 * or we have to use multiple match calls for each AND
806 * (AND x y) => /(?(:x)(:y))|(?(:y)(:x))/
810 foreach ($this->leaves as $leaf)
811 $regexp .= $leaf->regexp();
817 * A (possibly multi-argument) 'OR' conjoin.
819 class TextSearchQuery_node_or
820 extends TextSearchQuery_node_binop
825 // We will combine any of our direct descendents which are WORDs
826 // into a single (?=.*(?:word1|word2|...)) regexp.
831 foreach ($this->leaves as $leaf) {
832 if ($leaf->op == 'WORD')
833 $words[] = preg_quote($leaf->word, '/');
835 $regexps[] = $leaf->regexp();
839 array_unshift($regexps,
840 '(?=.*' . $this->_join($words) . ')');
842 return $this->_join($regexps);
845 function _join($regexps) {
846 assert(count($regexps) > 0);
848 if (count($regexps) > 1)
849 return '(?:' . join('|', $regexps) . ')';
856 ////////////////////////////////////////////////////////////////
859 // op's (and, or, not) are forced to lowercase in the tokenizer.
861 ////////////////////////////////////////////////////////////////
862 define ('TSQ_TOK_BINOP', 1);
863 define ('TSQ_TOK_NOT', 2);
864 define ('TSQ_TOK_LPAREN', 4);
865 define ('TSQ_TOK_RPAREN', 8);
866 define ('TSQ_TOK_WORD', 16);
867 define ('TSQ_TOK_STARTS_WITH', 32);
868 define ('TSQ_TOK_ENDS_WITH', 64);
869 define ('TSQ_TOK_EXACT', 128);
870 define ('TSQ_TOK_REGEX', 256);
871 define ('TSQ_TOK_REGEX_GLOB', 512);
872 define ('TSQ_TOK_REGEX_PCRE', 1024);
873 define ('TSQ_TOK_REGEX_SQL', 2048);
874 define ('TSQ_TOK_ALL', 4096);
875 // all bits from word to the last.
876 define ('TSQ_ALLWORDS', (4096*2)-1 - (16-1));
878 class TextSearchQuery_Parser
881 * This is a simple recursive descent parser, based on the following grammar:
896 * atom : '(' list ')'
901 * The terminal tokens are:
908 * /[^-()\s][^()\s]* WORD
919 function parse ($search_expr, $case_exact=false, $regex=TSQ_REGEX_AUTO) {
920 $this->lexer = new TextSearchQuery_Lexer($search_expr, $case_exact, $regex);
921 $this->_regex = $regex;
922 $tree = $this->get_list('toplevel');
923 assert($this->lexer->eof());
928 function get_list ($is_toplevel = false) {
931 // token types we'll accept as words (and thus expr's) for the
932 // purpose of error recovery:
933 $accept_as_words = TSQ_TOK_NOT | TSQ_TOK_BINOP;
935 $accept_as_words |= TSQ_TOK_LPAREN | TSQ_TOK_RPAREN;
937 while ( ($expr = $this->get_expr())
938 || ($expr = $this->get_word($accept_as_words)) )
945 return new TextSearchQuery_node;
949 return new TextSearchQuery_node_and($list);
952 function get_expr () {
953 if ( ($expr = $this->get_atom()) === false ) // protect against '0'
956 $savedpos = $this->lexer->tell();
957 // Bug#1791564: allow string '0'
958 while ( ($op = $this->lexer->get(TSQ_TOK_BINOP)) !== false) {
959 if ( ! ($right = $this->get_atom()) ) {
964 $expr = new TextSearchQuery_node_and(array($expr, $right));
967 $expr = new TextSearchQuery_node_or(array($expr, $right));
970 $savedpos = $this->lexer->tell();
972 $this->lexer->seek($savedpos);
978 function get_atom() {
979 if ($atom = $this->get_word(TSQ_ALLWORDS)) // Bug#1791564 not involved: '*'
982 $savedpos = $this->lexer->tell();
983 if ( $this->lexer->get(TSQ_TOK_LPAREN) ) {
984 if ( ($list = $this->get_list()) && $this->lexer->get(TSQ_TOK_RPAREN) ) {
988 // Handle " ( " or "(test" without closing ")" as plain word
989 $this->lexer->seek($savedpos);
990 return new TextSearchQuery_node_word($this->lexer->get(-1));
993 elseif ( $this->lexer->get(TSQ_TOK_NOT) ) {
994 if ( ($atom = $this->get_atom()) )
995 return new TextSearchQuery_node_not($atom);
997 $this->lexer->seek($savedpos);
1001 function get_word($accept = TSQ_ALLWORDS) {
1002 // Performance shortcut for ( and ). This is always false
1003 if (!empty($this->lexer->tokens[$this->lexer->pos])) {
1004 list ($type, $val) = $this->lexer->tokens[$this->lexer->pos];
1005 if ($type == TSQ_TOK_LPAREN or $type == TSQ_TOK_RPAREN)
1008 foreach (array("WORD","STARTS_WITH","ENDS_WITH","EXACT",
1009 "REGEX","REGEX_GLOB","REGEX_PCRE","ALL") as $tok) {
1010 $const = constant("TSQ_TOK_".$tok);
1011 // Bug#1791564: allow word '0'
1012 if ( $accept & $const and
1013 (($word = $this->lexer->get($const)) !== false))
1015 $classname = "TextSearchQuery_node_".strtolower($tok);
1016 return new $classname($word);
1023 class TextSearchQuery_Lexer {
1024 function TextSearchQuery_Lexer ($query_str, $case_exact=false,
1025 $regex=TSQ_REGEX_AUTO)
1027 $this->tokens = $this->tokenize($query_str, $case_exact, $regex);
1035 function seek($pos) {
1040 return $this->pos == count($this->tokens);
1044 * TODO: support more regex styles, esp. prefer the forced ones over auto
1047 function tokenize($string, $case_exact=false, $regex=TSQ_REGEX_AUTO) {
1049 $buf = $case_exact ? ltrim($string) : strtolower(ltrim($string));
1050 while (!empty($buf)) {
1051 if (preg_match('/^(and|or)\b\s*/i', $buf, $m)) {
1052 $val = strtolower($m[1]);
1053 $type = TSQ_TOK_BINOP;
1055 elseif (preg_match('/^(-|not\b)\s*/i', $buf, $m)) {
1056 $val = strtolower($m[1]);
1057 $type = TSQ_TOK_NOT;
1059 elseif (preg_match('/^([()])\s*/', $buf, $m)) {
1061 $type = $m[1] == '(' ? TSQ_TOK_LPAREN : TSQ_TOK_RPAREN;
1065 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
1066 and preg_match('/^\*\s*/', $buf, $m)) {
1068 $type = TSQ_TOK_ALL;
1071 elseif ($regex & (TSQ_REGEX_PCRE)
1072 and preg_match('/^\.\*\s*/', $buf, $m)) {
1074 $type = TSQ_TOK_ALL;
1077 elseif ($regex & (TSQ_REGEX_SQL)
1078 and preg_match('/^%\s*/', $buf, $m)) {
1080 $type = TSQ_TOK_ALL;
1084 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
1085 and preg_match('/^\^([^-()][^()\s]*)\s*/', $buf, $m)) {
1087 $type = TSQ_TOK_STARTS_WITH;
1090 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
1091 and preg_match('/^([^-()][^()\s]*)\*\s*/', $buf, $m)) {
1093 $type = TSQ_TOK_STARTS_WITH;
1096 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
1097 and preg_match('/^\*([^-()][^()\s]*)\s*/', $buf, $m)) {
1099 $type = TSQ_TOK_ENDS_WITH;
1102 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
1103 and preg_match('/^([^-()][^()\s]*)\$\s*/', $buf, $m)) {
1105 $type = TSQ_TOK_ENDS_WITH;
1108 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
1109 and preg_match('/^\^([^-()][^()\s]*)\$\s*/', $buf, $m)) {
1111 $type = TSQ_TOK_EXACT;
1115 elseif (preg_match('/^ " ( (?: [^"]+ | "" )* ) " \s*/x', $buf, $m)) {
1116 $val = str_replace('""', '"', $m[1]);
1117 $type = TSQ_TOK_WORD;
1120 elseif (preg_match("/^ ' ( (?:[^']+|'')* ) ' \s*/x", $buf, $m)) {
1121 $val = str_replace("''", "'", $m[1]);
1122 $type = TSQ_TOK_WORD;
1125 elseif (preg_match('/^([^-()][^()\s]*)\s*/', $buf, $m)) {
1127 $type = TSQ_TOK_WORD;
1130 assert(empty($buf));
1133 $buf = substr($buf, strlen($m[0]));
1135 /* refine the simple parsing from above: bla*bla, bla?bla, ...
1136 if ($regex and $type == TSQ_TOK_WORD) {
1137 if (substr($val,0,1) == "^")
1138 $type = TSQ_TOK_STARTS_WITH;
1139 elseif (substr($val,0,1) == "*")
1140 $type = TSQ_TOK_ENDS_WITH;
1141 elseif (substr($val,-1,1) == "*")
1142 $type = TSQ_TOK_STARTS_WITH;
1145 $tokens[] = array($type, $val);
1150 function get($accept) {
1151 if ($this->pos >= count($this->tokens))
1154 list ($type, $val) = $this->tokens[$this->pos];
1155 if (($type & $accept) == 0)
1163 // $Log: not supported by cvs2svn $
1164 // Revision 1.30 2007/07/14 12:31:00 rurban
1165 // fix bug#1752172 undefined method TextSearchQuery_node_or::_sql_quote()
1167 // Revision 1.29 2007/07/14 12:03:38 rurban
1168 // support ranked search: simple score() function
1170 // Revision 1.28 2007/03/18 17:35:26 rurban
1173 // Revision 1.27 2007/01/21 23:27:32 rurban
1174 // Fix ->_backend->qstr()
1176 // Revision 1.26 2007/01/04 16:41:52 rurban
1177 // Improve error description. Fix the function parser for illegal functions, when the tokenizer cannot be used.
1179 // Revision 1.25 2007/01/03 21:22:34 rurban
1180 // add getType(). NumericSearchQuery::check Improve hacker detection using token_get_all(). Better support for multiple attributes. Add getVars().
1182 // Revision 1.24 2007/01/02 13:19:05 rurban
1183 // add NumericSearchQuery. change on pcre: no parsing done, detect modifiers
1185 // Revision 1.23 2006/04/13 19:30:44 rurban
1186 // make TextSearchQuery->_stoplist localizable and overridable within config.ini
1192 // c-basic-offset: 4
1193 // c-hanging-comment-ender-p: nil
1194 // indent-tabs-mode: nil