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$" => EXACT(phrase)
17 * <dd> "^word" => STARTS_WITH(phrase)
18 * <dd> "word$" => ENDS_WITH(phrase)
19 * <dd> "^word" ... => STARTS_WITH(word)
20 * <dd> "word$" ... => ENDS_WITH(word)
21 * <dd> "word*" => STARTS_WITH(word)
22 * <dd> "*word" => ENDS_WITH(word)
23 * <dd> "/^word.* /" => REGEX(^word.*)
24 * <dd> "word*word" => REGEX(word.*word)
27 * The full query syntax, in order of precedence, is roughly:
29 * The unary 'NOT' or '-' operator (they are equivalent) negates the
30 * following search clause.
32 * Search clauses may be joined with the (left-associative) binary operators
33 * 'AND' and 'OR'. (case-insensitive)
35 * Two adjoining search clauses are joined with an implicit 'AND'. This has
36 * lower precedence than either an explicit 'AND' or 'OR', so "a b OR c"
37 * parses as "a AND ( b OR c )", while "a AND b OR c" parses as
38 * "( a AND b ) OR c" (due to the left-associativity of 'AND' and 'OR'.)
40 * Search clauses can be grouped with parentheses.
42 * Phrases (or other things which don't look like words) can be forced to
43 * be interpreted as words by quoting them, either with single (') or double (")
44 * quotes. If you wan't to include the quote character within a quoted string,
45 * double-up on the quote character: 'I''m hungry' is equivalent to
48 * Force regex on "re:word" => posix-style, "/word/" => pcre-style
49 * or use regex='glob' to use file wildcard-like matching. (not yet)
51 * The parsed tree is then converted to the needed PCRE (highlight,
52 * simple backends) or SQL functions.
54 * @author: Jeff Dairiki
55 * @author: Reini Urban (case and regex detection, enhanced sql callbacks)
58 // regex-style: 'auto', 'none', 'glob', 'posix', 'pcre', 'sql'
59 define ('TSQ_REGEX_NONE', 0);
60 define ('TSQ_REGEX_AUTO', 1);
61 define ('TSQ_REGEX_POSIX', 2);
62 define ('TSQ_REGEX_GLOB', 4);
63 define ('TSQ_REGEX_PCRE', 8);
64 define ('TSQ_REGEX_SQL', 16);
66 define ('TSQ_TOK_VOID', 0);
67 define ('TSQ_TOK_BINOP', 1);
68 define ('TSQ_TOK_NOT', 2);
69 define ('TSQ_TOK_LPAREN', 4);
70 define ('TSQ_TOK_RPAREN', 8);
71 define ('TSQ_TOK_WORD', 16);
72 define ('TSQ_TOK_STARTS_WITH', 32);
73 define ('TSQ_TOK_ENDS_WITH', 64);
74 define ('TSQ_TOK_EXACT', 128);
75 define ('TSQ_TOK_REGEX', 256);
76 define ('TSQ_TOK_REGEX_GLOB', 512);
77 define ('TSQ_TOK_REGEX_PCRE', 1024);
78 define ('TSQ_TOK_REGEX_SQL', 2048);
79 define ('TSQ_TOK_ALL', 4096);
80 // all bits from word to the last.
81 define ('TSQ_ALLWORDS', (4096 * 2) - 1 - (16 - 1));
85 private $sql_clause_cb;
91 * @param $search_query string The query. Syntax is as described above.
92 * Note that an empty $search_query will match anything.
93 * @param $case_exact boolean
94 * @param $regex string one of 'auto', 'none', 'glob', 'posix', 'pcre', 'sql'
95 * @see TextSearchQuery
97 function TextSearchQuery($search_query, $case_exact = false, $regex = 'auto')
99 if ($regex == 'none' or !$regex)
101 elseif (defined("TSQ_REGEX_" . strtoupper($regex)))
102 $this->_regex = constant("TSQ_REGEX_" . strtoupper($regex)); else {
103 trigger_error(fmt("Unsupported argument: %s=%s", 'regex', $regex));
106 $this->_regex_modifier = ($case_exact ? '' : 'i') . 'sS';
107 $this->_case_exact = $case_exact;
108 if ($regex != 'pcre') {
109 $parser = new TextSearchQuery_Parser;
110 $this->_tree = $parser->parse($search_query, $case_exact, $this->_regex);
111 $this->optimize(); // broken under certain circumstances: "word -word -word"
112 if (defined("FULLTEXTSEARCH_STOPLIST"))
113 $this->_stoplist = FULLTEXTSEARCH_STOPLIST;
114 else // default stoplist, localizable.
115 $this->_stoplist = _("(A|An|And|But|By|For|From|In|Is|It|Of|On|Or|The|To|With)");
117 $this->_tree = new TextSearchQuery_node_regex_pcre($search_query);
118 if (preg_match("/^\/(.*)\/(\w*)$/", $search_query, $m)) {
119 $this->_tree->word = $m[1];
120 $this->_regex_modifier = $m[2]; // overrides case_exact
130 private function optimize()
132 $this->_tree = $this->_tree->optimize();
136 * Get a PCRE regexp which matches the query.
140 if (!isset($this->_regexp)) {
141 if (!isset($this->_regex_modifier))
142 $this->_regex_modifier = ($this->_case_exact ? '' : 'i') . 'sS';
144 $this->_regexp = '/' . $this->_tree->regexp() . '/' . $this->_regex_modifier;
146 $this->_regexp = '/^' . $this->_tree->regexp() . '/' . $this->_regex_modifier;
148 return $this->_regexp;
152 * Match query against string.
153 * EXACT ("Term") ignores the case_exact setting.
155 * @param $string string The string to match.
156 * @return boolean True if the string matches the query.
158 function match($string)
160 if ($this->_tree->_op == TSQ_TOK_ALL) return true;
161 if ($this->_tree->_op == TSQ_TOK_EXACT) return $this->_tree->word == $string;
162 return preg_match($this->asRegexp(), $string);
165 /* How good does it match? Returns a number */
166 function score($string)
170 foreach (array_unique($this->_tree->highlight_words()) as $word) {
171 if ($nummatch = preg_match_all("/" . preg_quote($word, '/') . "/" .
172 $this->_regex_modifier,
175 $score += ($i-- * $nummatch);
177 return min(1.0, $score / 10.0);
181 * Get a regular expression suitable for highlighting matched words.
183 * This returns a PCRE regular expression which matches any non-negated
186 * @return string The PCRE regexp.
188 function getHighlightRegexp()
190 if (!isset($this->_hilight_regexp)) {
191 $words = array_unique($this->_tree->highlight_words());
193 $this->_hilight_regexp = false;
195 foreach ($words as $key => $word)
196 $words[$key] = preg_quote($word, '/');
197 $this->_hilight_regexp = '(?' . ($this->_case_exact ? '' : 'i') . ':'
198 . join('|', $words) . ')';
201 return $this->_hilight_regexp;
204 /** Get away with the callback and use a db-specific search class instead.
205 * @see WikiDB_backend_PearDB_search
207 function makeSqlClauseObj(&$sql_search_cb)
209 $this->sql_clause_cb = $sql_search_cb;
210 return $this->sql_clause_obj($this->_tree);
213 private function sql_clause_obj($node)
215 switch ($node->_op) {
217 return "NOT (" . $this->sql_clause_cb->call($node->leaves[0]) . ")";
219 $subclauses = array();
220 foreach ($node->leaves as $leaf)
221 $subclauses[] = "(" . $this->sql_clause_obj($leaf) . ")";
222 return join(" $node->op ", $subclauses);
228 return $this->sql_clause_cb->call($node);
233 postgresql tsearch2 uses no WHERE operators, just & | and ! in the searchstring
235 function makeTsearch2SqlClauseObj(&$sql_search_cb)
237 $this->sql_clause_cb = $sql_search_cb;
238 return $this->Tsearch2Sql_clause_obj($this->_tree);
241 private function Tsearch2Sql_clause_obj($node)
243 // TODO: "such a phrase"
244 switch ($node->_op) {
246 return "!" . $node->leaves[0];
248 $subclauses = array();
249 foreach ($node->leaves as $leaf)
250 $subclauses[] = $this->Tsearch2Sql_clause_obj($leaf);
251 return join($node->_op == 'OR' ? "|" : "&", $subclauses);
257 return $this->sql_clause_cb->call($node);
263 return '%' . $this->_sql_quote($this->word) . '%';
267 * Get printable representation of the parse tree.
269 * This is for debugging only.
270 * @return string Printable parse tree.
274 return $this->as_string($this->_tree);
277 private function as_string($node, $indent = '')
279 switch ($node->_op) {
281 return $indent . "WORD: $node->word";
283 return $indent . "VOID";
285 return $indent . "ALL";
287 $lines = array($indent . $node->op . ":");
289 foreach ($node->leaves as $leaf)
290 $lines[] = $this->as_string($leaf, $indent);
291 return join("\n", $lines);
297 * This is a TextSearchQuery which matches nothing.
299 class NullTextSearchQuery extends TextSearchQuery
302 * Create a new query.
304 * @see TextSearchQuery
306 function NullTextSearchQuery()
315 function match($string)
320 function getHighlightRegexp()
327 return "NullTextSearchQuery";
332 * A simple algebraic matcher for numeric attributes.
333 * NumericSearchQuery can do ("population < 20000 and area > 1000000", array("population", "area"))
334 * ->match(array('population' => 100000, 'area' => 10000000))
336 * Supports all mathematical PHP comparison operators, plus ':=' for equality.
337 * "(x < 2000000 and x >= 10000) or (x >= 100 and x < 2000)"
338 * "x := 100000" is the same as "x == 100000"
340 * Since this is basic numerics only, we simply try to get away with
341 * replacing the variable values at the right positions and do an eval then.
343 * @package NumericSearchQuery
344 * @author Reini Urban
345 * @see SemanticAttributeSearchQuery
347 class NumericSearchQuery
350 protected $workquery;
353 * Create a new query.
354 * NumericSearchQuery("population > 20000 or population < 200", "population")
355 * NumericSearchQuery("population < 20000 and area > 1000000", array("population", "area"))
357 * With a single variable it is easy: The valid name must be matched elsewhere, just
358 * replace the given number in match in the query.
361 * With matching a struct we need strict names, no * as name is allowed.
362 * So always when the placeholder is an array, the names of the target struct must match
363 * and all vars be defined. Use the method can_match($struct) therefore.
366 * @param $search_query string A numerical query with placeholders as variable.
367 * @param $placeholders array or string All placeholders in the query must be defined
368 * here, and will be replaced by the matcher.
370 function NumericSearchQuery($search_query, $placeholders)
372 // added some basic security checks against user input
373 $this->_query = $search_query;
374 $this->_placeholders = $placeholders;
376 // we should also allow the M_ constants
377 $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');
378 $this->_allowed_operators = explode(',', '-,<,<=,>,>=,==,!=,*,+,/,(,),%,and,or,xor,<<,>>,===,!==,&,^,|,&&,||');
379 $this->_parser_check = array();
380 // check should be fast, so make a hash
381 foreach ($this->_allowed_functions as $f)
382 $this->_parser_check[$f] = 1;
383 foreach ($this->_allowed_operators as $f)
384 $this->_parser_check[$f] = 1;
385 if (is_array($placeholders))
386 foreach ($placeholders as $f)
387 $this->_parser_check[$f] = 1;
388 else $this->_parser_check[$placeholders] = 1;
390 // This is a speciality: := looks like the attribute definition and is
391 // therefore a dummy check for this definition.
392 // php-4.2.2 has a problem with /\b:=\b/ matching "population := 1223400"
393 $this->_query = preg_replace("/:=/", "==", $this->_query);
394 $this->_query = $this->check_query($this->_query);
403 * Check the symbolic definition query against unwanted functions and characters.
404 * "population < 20000 and area > 1000000" vs
405 * "area > 1000000 and mail($me,file("/etc/passwd"),...)"
406 * http://localhost/wikicvs/SemanticSearch?attribute=*&attr_op=<0 and find(1)>&s=-0.01&start_debug=1
408 function check_query($query)
410 $tmp = $query; // check for all function calls, in case the tokenizer is not available.
411 while (preg_match("/([a-z][a-z0-9]+)\s*\((.*)$/i", $tmp, $m)) {
412 if (!in_array($m[1], $this->_allowed_functions)
413 and !in_array($m[1], $this->_allowed_operators)
415 trigger_error("Illegal function in query: " . $m[1], E_USER_WARNING);
421 // Strictly check for illegal functions and operators, which are no placeholders.
422 if (function_exists('token_get_all')) {
423 $parsed = token_get_all("<?$query?>");
424 foreach ($parsed as $x) { // flat, non-recursive array
425 if (is_string($x) and !isset($this->_parser_check[$x])) {
426 // single char op or name
427 trigger_error("Illegal string or operator in query: \"$x\"", E_USER_WARNING);
429 } elseif (is_array($x)) {
430 $n = token_name($x[0]);
431 if ($n == 'T_OPEN_TAG' or $n == 'T_WHITESPACE'
432 or $n == 'T_CLOSE_TAG' or $n == 'T_LNUMBER'
433 or $n == 'T_CONST' or $n == 'T_DNUMBER'
435 if ($n == 'T_VARIABLE') { // but we do allow consts
436 trigger_error("Illegal variable in query: \"$x[1]\"", E_USER_WARNING);
439 if (is_string($x[1]) and !isset($this->_parser_check[$x[1]])) {
440 // multi-char char op or name
441 trigger_error("Illegal $n in query: \"$x[1]\"", E_USER_WARNING);
446 //echo "$query <br>";
447 //$this->_parse_token($parsed);
452 { T_OPEN_TAG "<?"} { T_STRING "_x"} { T_WHITESPACE " "} ">" { T_WHITESPACE " "} { T_LNUMBER "0"} { T_CLOSE_TAG "?>"}
453 Interesting: on-char ops, as ">" are not tokenized.
455 { T_OPEN_TAG "< ?" } { T_STRING "_x" } { T_WHITESPACE " " } { T_IS_SMALLER_OR_EQUAL "<=" } { T_WHITESPACE " " } { T_LNUMBER "0" } { T_CLOSE_TAG "?>" }
458 // Detect illegal characters besides nums, words and ops.
459 // So attribute names can not be utf-8
460 $c = "/([^\d\w.,\s" . preg_quote(join("", $this->_allowed_operators), "/") . "])/";
461 if (preg_match($c, $query, $m)) {
462 trigger_error("Illegal character in query: " . $m[1], E_USER_WARNING);
470 * Check the bound, numeric-only query against unwanted functions and sideeffects.
471 * "4560000 < 20000 and 1456022 > 1000000"
473 private function live_check()
475 // TODO: check $this->_workquery again?
476 return !empty($this->workquery);
480 * A numeric query can only operate with predefined variables. "x < 0 and y < 1"
482 * @return array The names as array of strings. => ('x', 'y') the placeholders.
486 if (is_array($this->_placeholders)) return $this->_placeholders;
487 else return array($this->_placeholders);
491 * Strip non-numeric chars from the variable (as the groupseperator) and replace
492 * it in the symbolic query for evaluation.
495 * @param $value number A numerical value: integer, float or string.
496 * @param $x string The variable name to be replaced in the query.
499 private function bind($value, $x)
501 // TODO: check is_number, is_float, is_integer and do casting
502 $this->bound[] = array('linkname' => $x,
503 'linkvalue' => $value);
504 $value = preg_replace("/[^-+0123456789.,]/", "", $value);
505 //$c = "/\b".preg_quote($x,"/")."\b/";
506 $this->workquery = preg_replace("/\b" . preg_quote($x, "/") . "\b/", $value, $this->workquery);
507 // FIXME: do again a final check. now only numbers and some operators are allowed.
508 return $this->workquery;
511 /* array of successfully bound vars, and in case of success, the resulting vars
513 private function bound()
519 * With an array of placeholders we need a hash to check against, if all required names are given.
520 * Purpose: Be silent about missing vars, just return false.
523 * @param string $variables string or hash of name => value The keys must satisfy all placeholders in the definition.
524 * We want the full hash and not just the keys because a hash check is faster than the array of keys check.
527 function can_match(&$variables)
529 if (empty($this->_query))
531 $p =& $this->_placeholders;
532 if (!is_array($variables) and !is_array($p))
533 return $variables == $p; // This was easy.
534 // Check if all placeholders have definitions. can be overdefined but not underdefined.
536 if (!isset($variables[$p])) return false;
539 if (!isset($variables[$x])) return false;
546 * We can match against a single variable or against a hash of variables.
547 * With one placeholder we need just a number.
548 * With an array of placeholders we need a hash.
551 * @param $variable number or array of name => value The keys must satisfy all placeholders in the definition.
554 function match(&$variable)
556 $p =& $this->_placeholders;
557 $this->workquery = $this->_query;
559 if (is_array($variable)) { // which var to match? we cannot decide this here
560 if (!isset($variable[$p]))
561 trigger_error("Required NumericSearchQuery->match variable $p not defined.", E_USER_ERROR);
562 $this->bind($variable[$p], $p);
564 $this->bind($variable, $p);
568 if (!isset($variable[$x]))
569 trigger_error("Required NumericSearchQuery->match variable $x not defined.", E_USER_ERROR);
570 $this->bind($variable[$x], $x);
573 if (!$this->live_check()) // check returned an error
575 $search = $this->workquery;
577 //if (DEBUG & _DEBUG_VERBOSE)
578 // trigger_error("\$result = (boolean)($search);", E_USER_NOTICE);
579 // We might have a numerical problem:
580 // php-4.2.2 eval'ed as module: "9.636e+08 > 1000" false;
581 // php-5.1.2 cgi true, 4.2.2 cgi true
582 eval("\$result = (boolean)($search);");
583 if ($result and is_array($p)) {
584 return $this->bound();
590 ////////////////////////////////////////////////////////////////
592 // Remaining classes are private.
594 ////////////////////////////////////////////////////////////////
596 * Virtual base class for nodes in a TextSearchQuery parse tree.
598 * Also serves as a 'VOID' (contentless) node.
600 class TextSearchQuery_node
607 * Optimize this node.
608 * @return object Optimized node.
616 * @return string regexp matching this node.
624 * @param bool $negated True if this node has been negated (higher in the parse tree.)
625 * @return array A list of all non-negated words contained by this node.
627 function highlight_words($negated = false)
637 protected function sql_quote()
640 $word = preg_replace('/(?=[%_\\\\])/', "\\", $this->word);
641 return $request->_dbi->_backend->qstr($word);
646 * A word. Exact or substring?
648 class TextSearchQuery_node_word
649 extends TextSearchQuery_node
652 public $_op = TSQ_TOK_WORD;
654 function TextSearchQuery_node_word($word)
661 return '(?=.*\b' . preg_quote($this->word, '/') . '\b)';
664 function highlight_words($negated = false)
666 return $negated ? array() : array($this->word);
671 return '%' . $this->sql_quote($this->word) . '%';
675 class TextSearchQuery_node_all
676 extends TextSearchQuery_node
679 public $_op = TSQ_TOK_ALL;
692 class TextSearchQuery_node_starts_with
693 extends TextSearchQuery_node_word
695 public $op = "STARTS_WITH";
696 public $_op = TSQ_TOK_STARTS_WITH;
700 return '(?=.*\b' . preg_quote($this->word, '/') . ')';
705 return $this->sql_quote($this->word) . '%';
709 // ^word: full phrase starts with
710 class TextSearchQuery_phrase_starts_with
711 extends TextSearchQuery_node_starts_with
715 return '(?=^' . preg_quote($this->word, '/') . ')';
719 class TextSearchQuery_node_ends_with
720 extends TextSearchQuery_node_word
722 public $op = "ENDS_WITH";
723 public $_op = TSQ_TOK_ENDS_WITH;
727 return '(?=.*' . preg_quote($this->word, '/') . '\b)';
732 return '%' . $this->sql_quote($this->word);
736 // word$: full phrase ends with
737 class TextSearchQuery_phrase_ends_with
738 extends TextSearchQuery_node_ends_with
742 return '(?=' . preg_quote($this->word, '/') . '$)';
746 class TextSearchQuery_node_exact
747 extends TextSearchQuery_node_word
749 public $op = "EXACT";
750 public $_op = TSQ_TOK_EXACT;
754 return '(?=\b' . preg_quote($this->word, '/') . '\b)';
759 return $this->_sql_squote($this->word);
763 class TextSearchQuery_node_regex // posix regex. FIXME!
764 extends TextSearchQuery_node_word
766 public $op = "REGEX"; // using REGEXP or ~ extension
767 public $_op = TSQ_TOK_REGEX;
771 return '(?=.*\b' . $this->word . '\b)';
776 return $this->sql_quote($this->word);
780 class TextSearchQuery_node_regex_glob
781 extends TextSearchQuery_node_regex
783 public $op = "REGEX_GLOB";
784 public $_op = TSQ_TOK_REGEX_GLOB;
788 return '(?=.*\b' . glob_to_pcre($this->word) . '\b)';
792 class TextSearchQuery_node_regex_pcre // how to handle pcre modifiers? /i
793 extends TextSearchQuery_node_regex
795 public $op = "REGEX_PCRE";
796 public $_op = TSQ_TOK_REGEX_PCRE;
804 class TextSearchQuery_node_regex_sql
805 extends TextSearchQuery_node_regex
807 public $op = "REGEX_SQL"; // using LIKE
808 public $_op = TSQ_TOK_REGEX_SQL;
812 return str_replace(array("/%/", "/_/"), array(".*", "."), $this->word);
824 class TextSearchQuery_node_not
825 extends TextSearchQuery_node
828 public $_op = TSQ_TOK_NOT;
830 function TextSearchQuery_node_not($leaf)
832 $this->leaves = array($leaf);
837 $leaf = &$this->leaves[0];
838 $leaf = $leaf->optimize();
839 if ($leaf->_op == TSQ_TOK_NOT)
840 return $leaf->leaves[0]; // ( NOT ( NOT x ) ) -> x
846 $leaf = &$this->leaves[0];
847 return '(?!' . $leaf->regexp() . ')';
850 function highlight_words($negated = false)
852 return $this->leaves[0]->highlight_words(!$negated);
857 * Virtual base class for 'AND' and 'OR conjoins.
859 class TextSearchQuery_node_binop
860 extends TextSearchQuery_node
862 public $_op = TSQ_TOK_BINOP;
864 function TextSearchQuery_node_binop($leaves)
866 $this->leaves = $leaves;
869 protected function flatten()
871 // This flattens e.g. (AND (AND a b) (OR c d) e)
872 // to (AND a b e (OR c d))
874 foreach ($this->leaves as $leaf) {
875 $leaf = $leaf->optimize();
876 if ($this->op == $leaf->op)
877 $flat = array_merge($flat, $leaf->leaves);
881 $this->leaves = $flat;
887 assert(!empty($this->leaves));
888 if (count($this->leaves) == 1)
889 return $this->leaves[0]; // (AND x) -> x
893 function highlight_words($negated = false)
896 foreach ($this->leaves as $leaf)
897 array_splice($words, 0, 0,
898 $leaf->highlight_words($negated));
904 * A (possibly multi-argument) 'AND' conjoin.
906 class TextSearchQuery_node_and
907 extends TextSearchQuery_node_binop
915 // Convert (AND (NOT a) (NOT b) c d) into (AND (NOT (OR a b)) c d).
916 // Since OR's are more efficient for regexp matching:
917 // (?!.*a)(?!.*b) vs (?!.*(?:a|b))
919 // Suck out the negated leaves.
921 foreach ($this->leaves as $key => $leaf) {
922 if ($leaf->_op == TSQ_TOK_NOT) {
923 $nots[] = $leaf->leaves[0];
924 unset($this->leaves[$key]);
928 // Combine the negated leaves into a single negated or.
930 $node = (new TextSearchQuery_node_not
931 (new TextSearchQuery_node_or($nots)));
932 array_unshift($this->leaves, $node->optimize());
935 assert(!empty($this->leaves));
936 if (count($this->leaves) == 1)
937 return $this->leaves[0]; // (AND x) -> x
942 * Either we need all combinations of all words to be position independent,
943 * or we have to use multiple match calls for each AND
944 * (AND x y) => /(?(:x)(:y))|(?(:y)(:x))/
949 foreach ($this->leaves as $leaf)
950 $regexp .= $leaf->regexp();
956 * A (possibly multi-argument) 'OR' conjoin.
958 class TextSearchQuery_node_or
959 extends TextSearchQuery_node_binop
965 // We will combine any of our direct descendents which are WORDs
966 // into a single (?=.*(?:word1|word2|...)) regexp.
971 foreach ($this->leaves as $leaf) {
972 if ($leaf->op == TSQ_TOK_WORD)
973 $words[] = preg_quote($leaf->word, '/');
975 $regexps[] = $leaf->regexp();
979 array_unshift($regexps,
980 '(?=.*' . $this->join($words) . ')');
982 return $this->join($regexps);
985 private function join($regexps)
987 assert(count($regexps) > 0);
989 if (count($regexps) > 1)
990 return '(?:' . join('|', $regexps) . ')';
996 ////////////////////////////////////////////////////////////////
999 // op's (and, or, not) are forced to lowercase in the tokenizer.
1001 ////////////////////////////////////////////////////////////////
1002 class TextSearchQuery_Parser
1005 * This is a simple recursive descent parser, based on the following grammar:
1020 * atom : '(' list ')'
1025 * The terminal tokens are:
1032 * /[^-()\s][^()\s]* WORD
1036 * ^WORD TextSearchQuery_phrase_starts_with
1046 function parse($search_expr, $case_exact = false, $regex = TSQ_REGEX_AUTO)
1048 $this->lexer = new TextSearchQuery_Lexer($search_expr, $case_exact, $regex);
1049 $this->regex = $regex;
1050 $tree = $this->get_list('toplevel');
1051 // Assert failure when using the following URL in debug mode.
1052 // /TitleSearch?action=FullTextSearch&s=WFXSSProbe'")/>&case_exact=1®ex=sql
1053 // assert($this->lexer->eof());
1054 unset($this->lexer);
1058 function get_list($is_toplevel = false)
1062 // token types we'll accept as words (and thus expr's) for the
1063 // purpose of error recovery:
1064 $accept_as_words = TSQ_TOK_NOT | TSQ_TOK_BINOP;
1066 $accept_as_words |= TSQ_TOK_LPAREN | TSQ_TOK_RPAREN;
1068 while (($expr = $this->get_expr())
1069 || ($expr = $this->get_word($accept_as_words))) {
1075 return new TextSearchQuery_node;
1079 if ($is_toplevel and count($list) == 1) {
1080 if ($this->lexer->query_str[0] == '^')
1081 return new TextSearchQuery_phrase_starts_with($list[0]->word);
1085 return new TextSearchQuery_node_and($list);
1090 if (($expr = $this->get_atom()) === false) // protect against '0'
1093 $savedpos = $this->lexer->tell();
1094 // Bug#1791564: allow string '0'
1095 while (($op = $this->lexer->get(TSQ_TOK_BINOP)) !== false) {
1096 if (!($right = $this->get_atom())) {
1101 $expr = new TextSearchQuery_node_and(array($expr, $right));
1103 assert($op == 'or');
1104 $expr = new TextSearchQuery_node_or(array($expr, $right));
1107 $savedpos = $this->lexer->tell();
1109 $this->lexer->seek($savedpos);
1116 if ($atom = $this->get_word(TSQ_ALLWORDS)) // Bug#1791564 not involved: '*'
1119 $savedpos = $this->lexer->tell();
1120 if ($this->lexer->get(TSQ_TOK_LPAREN)) {
1121 if (($list = $this->get_list()) && $this->lexer->get(TSQ_TOK_RPAREN)) {
1125 // Handle " ( " or "(test" without closing ")" as plain word
1126 $this->lexer->seek($savedpos);
1127 return new TextSearchQuery_node_word($this->lexer->get(-1));
1129 } elseif ($this->lexer->get(TSQ_TOK_NOT)) {
1130 if (($atom = $this->get_atom()))
1131 return new TextSearchQuery_node_not($atom);
1133 $this->lexer->seek($savedpos);
1137 function get_word($accept = TSQ_ALLWORDS)
1139 // Performance shortcut for ( and ). This is always false
1140 if (!empty($this->lexer->tokens[$this->lexer->pos])) {
1141 list ($type, $val) = $this->lexer->tokens[$this->lexer->pos];
1142 if ($type == TSQ_TOK_LPAREN or $type == TSQ_TOK_RPAREN)
1145 foreach (array("WORD", "STARTS_WITH", "ENDS_WITH", "EXACT",
1146 "REGEX", "REGEX_GLOB", "REGEX_PCRE", "ALL") as $tok) {
1147 $const = constant("TSQ_TOK_" . $tok);
1148 // Bug#1791564: allow word '0'
1149 if ($accept & $const and
1150 (($word = $this->lexer->get($const)) !== false)
1152 // phrase or word level?
1153 if ($tok == 'STARTS_WITH' and $this->lexer->query_str[0] == '^')
1154 $classname = "TextSearchQuery_phrase_" . strtolower($tok);
1155 elseif ($tok == 'ENDS_WITH' and
1156 string_ends_with($this->lexer->query_str, '$')
1158 $classname = "TextSearchQuery_phrase_" . strtolower($tok); else
1159 $classname = "TextSearchQuery_node_" . strtolower($tok);
1160 return new $classname($word);
1167 class TextSearchQuery_Lexer
1169 function TextSearchQuery_Lexer($query_str, $case_exact = false,
1170 $regex = TSQ_REGEX_AUTO)
1172 $this->tokens = $this->tokenize($query_str, $case_exact, $regex);
1173 $this->query_str = $query_str;
1189 return $this->pos == count($this->tokens);
1193 * TODO: support more regex styles, esp. prefer the forced ones over auto
1196 function tokenize($string, $case_exact = false, $regex = TSQ_REGEX_AUTO)
1199 $buf = $case_exact ? ltrim($string) : strtolower(ltrim($string));
1200 while (!empty($buf)) {
1201 if (preg_match('/^([()])\s*/', $buf, $m)) {
1203 $type = $m[1] == '(' ? TSQ_TOK_LPAREN : TSQ_TOK_RPAREN;
1205 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_GLOB)
1206 and preg_match('/^\*\s*/', $buf, $m)
1209 $type = TSQ_TOK_ALL;
1211 elseif ($regex & (TSQ_REGEX_PCRE)
1212 and preg_match('/^\.\*\s*/', $buf, $m)
1215 $type = TSQ_TOK_ALL;
1217 elseif ($regex & (TSQ_REGEX_SQL)
1218 and preg_match('/^%\s*/', $buf, $m)
1221 $type = TSQ_TOK_ALL;
1223 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_PCRE)
1224 and preg_match('/^\^([^-()][^()\s]*)\s*/', $buf, $m)
1227 $type = TSQ_TOK_STARTS_WITH;
1229 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_GLOB)
1230 and preg_match('/^([^-()][^()\s]*)\*\s*/', $buf, $m)
1233 $type = TSQ_TOK_STARTS_WITH;
1235 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_GLOB)
1236 and preg_match('/^\*([^-()][^()\s]*)\s*/', $buf, $m)
1239 $type = TSQ_TOK_ENDS_WITH;
1241 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_PCRE)
1242 and preg_match('/^([^-()][^()\s]*)\$\s*/', $buf, $m)
1245 $type = TSQ_TOK_ENDS_WITH;
1247 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_PCRE)
1248 and preg_match('/^\^([^-()][^()\s]*)\$\s*/', $buf, $m)
1251 $type = TSQ_TOK_EXACT;
1252 } elseif (preg_match('/^(and|or)\b\s*/i', $buf, $m)) {
1253 $val = strtolower($m[1]);
1254 $type = TSQ_TOK_BINOP;
1255 } elseif (preg_match('/^(-|not\b)\s*/i', $buf, $m)) {
1256 $val = strtolower($m[1]);
1257 $type = TSQ_TOK_NOT;
1259 elseif (preg_match('/^ " ( (?: [^"]+ | "" )* ) " \s*/x', $buf, $m)) {
1260 $val = str_replace('""', '"', $m[1]);
1261 $type = TSQ_TOK_WORD;
1263 elseif (preg_match("/^ ' ( (?:[^']+|'')* ) ' \s*/x", $buf, $m)) {
1264 $val = str_replace("''", "'", $m[1]);
1265 $type = TSQ_TOK_WORD;
1267 elseif (preg_match('/^([^-()][^()\s]*)\s*/', $buf, $m)) {
1269 $type = TSQ_TOK_WORD;
1271 assert(empty($buf));
1274 $buf = substr($buf, strlen($m[0]));
1276 /* refine the simple parsing from above: bla*bla, bla?bla, ...
1277 if ($regex and $type == TSQ_TOK_WORD) {
1278 if (substr($val,0,1) == "^")
1279 $type = TSQ_TOK_STARTS_WITH;
1280 elseif (substr($val,0,1) == "*")
1281 $type = TSQ_TOK_ENDS_WITH;
1282 elseif (substr($val,-1,1) == "*")
1283 $type = TSQ_TOK_STARTS_WITH;
1286 $tokens[] = array($type, $val);
1291 function get($accept)
1293 if ($this->pos >= count($this->tokens))
1296 list ($type, $val) = $this->tokens[$this->pos];
1297 if (($type & $accept) == 0)
1308 // c-basic-offset: 4
1309 // c-hanging-comment-ender-p: nil
1310 // indent-tabs-mode: nil