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));
104 trigger_error(fmt("Unsupported argument: %s=%s", 'regex', $regex));
107 $this->_regex_modifier = ($case_exact ? '' : 'i') . 'sS';
108 $this->_case_exact = $case_exact;
109 if ($regex != 'pcre') {
110 $parser = new TextSearchQuery_Parser;
111 $this->_tree = $parser->parse($search_query, $case_exact, $this->_regex);
112 $this->optimize(); // broken under certain circumstances: "word -word -word"
113 if (defined("FULLTEXTSEARCH_STOPLIST"))
114 $this->_stoplist = FULLTEXTSEARCH_STOPLIST;
115 else // default stoplist, localizable.
116 $this->_stoplist = _("(A|An|And|But|By|For|From|In|Is|It|Of|On|Or|The|To|With)");
118 $this->_tree = new TextSearchQuery_node_regex_pcre($search_query);
119 if (preg_match("/^\/(.*)\/(\w*)$/", $search_query, $m)) {
120 $this->_tree->word = $m[1];
121 $this->_regex_modifier = $m[2]; // overrides case_exact
131 private function optimize()
133 $this->_tree = $this->_tree->optimize();
137 * Get a PCRE regexp which matches the query.
141 if (!isset($this->_regexp)) {
142 if (!isset($this->_regex_modifier))
143 $this->_regex_modifier = ($this->_case_exact ? '' : 'i') . 'sS';
145 $this->_regexp = '/' . $this->_tree->regexp() . '/' . $this->_regex_modifier;
147 $this->_regexp = '/^' . $this->_tree->regexp() . '/' . $this->_regex_modifier;
149 return $this->_regexp;
153 * Match query against string.
154 * EXACT ("Term") ignores the case_exact setting.
156 * @param $string string The string to match.
157 * @return boolean True if the string matches the query.
159 function match($string)
161 if ($this->_tree->_op == TSQ_TOK_ALL) return true;
162 if ($this->_tree->_op == TSQ_TOK_EXACT) return $this->_tree->word == $string;
163 return preg_match($this->asRegexp(), $string);
166 /* How good does it match? Returns a number */
167 function score($string)
171 foreach (array_unique($this->_tree->highlight_words()) as $word) {
172 if ($nummatch = preg_match_all("/" . preg_quote($word, '/') . "/" .
173 $this->_regex_modifier,
176 $score += ($i-- * $nummatch);
178 return min(1.0, $score / 10.0);
182 * Get a regular expression suitable for highlighting matched words.
184 * This returns a PCRE regular expression which matches any non-negated
187 * @return string The PCRE regexp.
189 function getHighlightRegexp()
191 if (!isset($this->_hilight_regexp)) {
192 $words = array_unique($this->_tree->highlight_words());
194 $this->_hilight_regexp = false;
196 foreach ($words as $key => $word)
197 $words[$key] = preg_quote($word, '/');
198 $this->_hilight_regexp = '(?' . ($this->_case_exact ? '' : 'i') . ':'
199 . join('|', $words) . ')';
202 return $this->_hilight_regexp;
205 /** Get away with the callback and use a db-specific search class instead.
206 * @see WikiDB_backend_PearDB_search
208 function makeSqlClauseObj(&$sql_search_cb)
210 $this->sql_clause_cb = $sql_search_cb;
211 return $this->sql_clause_obj($this->_tree);
214 private function sql_clause_obj($node)
216 switch ($node->_op) {
218 return "NOT (" . $this->sql_clause_cb->call($node->leaves[0]) . ")";
220 $subclauses = array();
221 foreach ($node->leaves as $leaf)
222 $subclauses[] = "(" . $this->sql_clause_obj($leaf) . ")";
223 return join(" $node->op ", $subclauses);
229 return $this->sql_clause_cb->call($node);
234 postgresql tsearch2 uses no WHERE operators, just & | and ! in the searchstring
236 function makeTsearch2SqlClauseObj(&$sql_search_cb)
238 $this->sql_clause_cb = $sql_search_cb;
239 return $this->Tsearch2Sql_clause_obj($this->_tree);
242 private function Tsearch2Sql_clause_obj($node)
244 // TODO: "such a phrase"
245 switch ($node->_op) {
247 return "!" . $node->leaves[0];
249 $subclauses = array();
250 foreach ($node->leaves as $leaf)
251 $subclauses[] = $this->Tsearch2Sql_clause_obj($leaf);
252 return join($node->_op == 'OR' ? "|" : "&", $subclauses);
258 return $this->sql_clause_cb->call($node);
264 return '%' . $this->sql_quote($this->word) . '%';
268 * Get printable representation of the parse tree.
270 * This is for debugging only.
271 * @return string Printable parse tree.
275 return $this->as_string($this->_tree);
278 private function as_string($node, $indent = '')
280 switch ($node->_op) {
282 return $indent . "WORD: $node->word";
284 return $indent . "VOID";
286 return $indent . "ALL";
288 $lines = array($indent . $node->op . ":");
290 foreach ($node->leaves as $leaf)
291 $lines[] = $this->as_string($leaf, $indent);
292 return join("\n", $lines);
298 * This is a TextSearchQuery which matches nothing.
300 class NullTextSearchQuery extends TextSearchQuery
303 * Create a new query.
305 * @see TextSearchQuery
307 function __construct()
316 function match($string)
321 function getHighlightRegexp()
328 return "NullTextSearchQuery";
333 * A simple algebraic matcher for numeric attributes.
334 * NumericSearchQuery can do ("population < 20000 and area > 1000000", array("population", "area"))
335 * ->match(array('population' => 100000, 'area' => 10000000))
337 * Supports all mathematical PHP comparison operators, plus ':=' for equality.
338 * "(x < 2000000 and x >= 10000) or (x >= 100 and x < 2000)"
339 * "x := 100000" is the same as "x == 100000"
341 * Since this is basic numerics only, we simply try to get away with
342 * replacing the variable values at the right positions and do an eval then.
344 * @package NumericSearchQuery
345 * @author Reini Urban
346 * @see SemanticAttributeSearchQuery
348 class NumericSearchQuery
351 protected $workquery;
354 * Create a new query.
355 * NumericSearchQuery("population > 20000 or population < 200", "population")
356 * NumericSearchQuery("population < 20000 and area > 1000000", array("population", "area"))
358 * With a single variable it is easy: The valid name must be matched elsewhere, just
359 * replace the given number in match in the query.
362 * With matching a struct we need strict names, no * as name is allowed.
363 * So always when the placeholder is an array, the names of the target struct must match
364 * and all vars be defined. Use the method can_match($struct) therefore.
366 * @param $search_query string A numerical query with placeholders as variable.
367 * @param $placeholders array or string All placeholders in the query must be defined
368 * here, and will be replaced by the matcher.
370 public function NumericSearchQuery($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.
494 * @param $value number A numerical value: integer, float or string.
495 * @param $x string The variable name to be replaced in the query.
498 private function bind($value, $x)
500 // TODO: check is_number, is_float, is_integer and do casting
501 $this->bound[] = array('linkname' => $x,
502 'linkvalue' => $value);
503 $value = preg_replace("/[^-+0123456789.,]/", "", $value);
504 //$c = "/\b".preg_quote($x,"/")."\b/";
505 $this->workquery = preg_replace("/\b" . preg_quote($x, "/") . "\b/", $value, $this->workquery);
506 // FIXME: do again a final check. now only numbers and some operators are allowed.
507 return $this->workquery;
510 /* array of successfully bound vars, and in case of success, the resulting vars
512 private function bound()
518 * With an array of placeholders we need a hash to check against, if all required names are given.
519 * Purpose: Be silent about missing vars, just return false.
521 * @param string $variables string or hash of name => value The keys must satisfy all placeholders in the definition.
522 * We want the full hash and not just the keys because a hash check is faster than the array of keys check.
525 public function can_match(&$variables)
527 if (empty($this->_query))
529 $p =& $this->_placeholders;
530 if (!is_array($variables) and !is_array($p))
531 return $variables == $p; // This was easy.
532 // Check if all placeholders have definitions. can be overdefined but not underdefined.
534 if (!isset($variables[$p])) return false;
537 if (!isset($variables[$x])) return false;
544 * We can match against a single variable or against a hash of variables.
545 * With one placeholder we need just a number.
546 * With an array of placeholders we need a hash.
548 * @param $variable number or array of name => value The keys must satisfy all placeholders in the definition.
551 public function match(&$variable)
553 $p =& $this->_placeholders;
554 $this->workquery = $this->_query;
556 if (is_array($variable)) { // which var to match? we cannot decide this here
557 if (!isset($variable[$p]))
558 trigger_error("Required NumericSearchQuery->match variable $p not defined.", E_USER_ERROR);
559 $this->bind($variable[$p], $p);
561 $this->bind($variable, $p);
565 if (!isset($variable[$x]))
566 trigger_error("Required NumericSearchQuery->match variable $x not defined.", E_USER_ERROR);
567 $this->bind($variable[$x], $x);
570 if (!$this->live_check()) // check returned an error
572 $search = $this->workquery;
574 //if (DEBUG & _DEBUG_VERBOSE)
575 // trigger_error("\$result = (boolean)($search);", E_USER_NOTICE);
576 // We might have a numerical problem:
577 // php-4.2.2 eval'ed as module: "9.636e+08 > 1000" false;
578 // php-5.1.2 cgi true, 4.2.2 cgi true
579 eval("\$result = (boolean)($search);");
580 if ($result and is_array($p)) {
581 return $this->bound();
587 ////////////////////////////////////////////////////////////////
589 // Remaining classes are private.
591 ////////////////////////////////////////////////////////////////
593 * Virtual base class for nodes in a TextSearchQuery parse tree.
595 * Also serves as a 'VOID' (contentless) node.
597 class TextSearchQuery_node
604 * Optimize this node.
605 * @return object Optimized node.
613 * @return string regexp matching this node.
621 * @param bool $negated True if this node has been negated (higher in the parse tree.)
622 * @return array A list of all non-negated words contained by this node.
624 function highlight_words($negated = false)
637 $word = preg_replace('/(?=[%_\\\\])/', "\\", $this->word);
638 return $request->_dbi->_backend->qstr($word);
643 * A word. Exact or substring?
645 class TextSearchQuery_node_word
646 extends TextSearchQuery_node
649 public $_op = TSQ_TOK_WORD;
651 function __construct($word)
658 return '(?=.*\b' . preg_quote($this->word, '/') . '\b)';
661 function highlight_words($negated = false)
663 return $negated ? array() : array($this->word);
668 return '%' . $this->sql_quote($this->word) . '%';
672 class TextSearchQuery_node_all
673 extends TextSearchQuery_node
676 public $_op = TSQ_TOK_ALL;
689 class TextSearchQuery_node_starts_with
690 extends TextSearchQuery_node_word
692 public $op = "STARTS_WITH";
693 public $_op = TSQ_TOK_STARTS_WITH;
697 return '(?=.*\b' . preg_quote($this->word, '/') . ')';
702 return $this->sql_quote($this->word) . '%';
706 // ^word: full phrase starts with
707 class TextSearchQuery_phrase_starts_with
708 extends TextSearchQuery_node_starts_with
712 return '(?=^' . preg_quote($this->word, '/') . ')';
716 class TextSearchQuery_node_ends_with
717 extends TextSearchQuery_node_word
719 public $op = "ENDS_WITH";
720 public $_op = TSQ_TOK_ENDS_WITH;
724 return '(?=.*' . preg_quote($this->word, '/') . '\b)';
729 return '%' . $this->sql_quote($this->word);
733 // word$: full phrase ends with
734 class TextSearchQuery_phrase_ends_with
735 extends TextSearchQuery_node_ends_with
739 return '(?=' . preg_quote($this->word, '/') . '$)';
743 class TextSearchQuery_node_exact
744 extends TextSearchQuery_node_word
746 public $op = "EXACT";
747 public $_op = TSQ_TOK_EXACT;
751 return '(?=\b' . preg_quote($this->word, '/') . '\b)';
756 return $this->_sql_squote($this->word);
760 class TextSearchQuery_node_regex // posix regex. FIXME!
761 extends TextSearchQuery_node_word
763 public $op = "REGEX"; // using REGEXP or ~ extension
764 public $_op = TSQ_TOK_REGEX;
768 return '(?=.*\b' . $this->word . '\b)';
773 return $this->sql_quote($this->word);
777 class TextSearchQuery_node_regex_glob
778 extends TextSearchQuery_node_regex
780 public $op = "REGEX_GLOB";
781 public $_op = TSQ_TOK_REGEX_GLOB;
785 return '(?=.*\b' . glob_to_pcre($this->word) . '\b)';
789 class TextSearchQuery_node_regex_pcre // how to handle pcre modifiers? /i
790 extends TextSearchQuery_node_regex
792 public $op = "REGEX_PCRE";
793 public $_op = TSQ_TOK_REGEX_PCRE;
801 class TextSearchQuery_node_regex_sql
802 extends TextSearchQuery_node_regex
804 public $op = "REGEX_SQL"; // using LIKE
805 public $_op = TSQ_TOK_REGEX_SQL;
809 return str_replace(array("/%/", "/_/"), array(".*", "."), $this->word);
821 class TextSearchQuery_node_not
822 extends TextSearchQuery_node
825 public $_op = TSQ_TOK_NOT;
827 function TextSearchQuery_node_not($leaf)
829 $this->leaves = array($leaf);
834 $leaf = &$this->leaves[0];
835 $leaf = $leaf->optimize();
836 if ($leaf->_op == TSQ_TOK_NOT)
837 return $leaf->leaves[0]; // ( NOT ( NOT x ) ) -> x
843 $leaf = &$this->leaves[0];
844 return '(?!' . $leaf->regexp() . ')';
847 function highlight_words($negated = false)
849 return $this->leaves[0]->highlight_words(!$negated);
854 * Virtual base class for 'AND' and 'OR conjoins.
856 class TextSearchQuery_node_binop
857 extends TextSearchQuery_node
859 public $_op = TSQ_TOK_BINOP;
861 function __construct($leaves)
863 $this->leaves = $leaves;
866 protected function flatten()
868 // This flattens e.g. (AND (AND a b) (OR c d) e)
869 // to (AND a b e (OR c d))
871 foreach ($this->leaves as $leaf) {
872 $leaf = $leaf->optimize();
873 if ($this->op == $leaf->op)
874 $flat = array_merge($flat, $leaf->leaves);
878 $this->leaves = $flat;
884 assert(!empty($this->leaves));
885 if (count($this->leaves) == 1)
886 return $this->leaves[0]; // (AND x) -> x
890 function highlight_words($negated = false)
893 foreach ($this->leaves as $leaf)
894 array_splice($words, 0, 0,
895 $leaf->highlight_words($negated));
901 * A (possibly multi-argument) 'AND' conjoin.
903 class TextSearchQuery_node_and
904 extends TextSearchQuery_node_binop
912 // Convert (AND (NOT a) (NOT b) c d) into (AND (NOT (OR a b)) c d).
913 // Since OR's are more efficient for regexp matching:
914 // (?!.*a)(?!.*b) vs (?!.*(?:a|b))
916 // Suck out the negated leaves.
918 foreach ($this->leaves as $key => $leaf) {
919 if ($leaf->_op == TSQ_TOK_NOT) {
920 $nots[] = $leaf->leaves[0];
921 unset($this->leaves[$key]);
925 // Combine the negated leaves into a single negated or.
927 $node = (new TextSearchQuery_node_not
928 (new TextSearchQuery_node_or($nots)));
929 array_unshift($this->leaves, $node->optimize());
932 assert(!empty($this->leaves));
933 if (count($this->leaves) == 1)
934 return $this->leaves[0]; // (AND x) -> x
939 * Either we need all combinations of all words to be position independent,
940 * or we have to use multiple match calls for each AND
941 * (AND x y) => /(?(:x)(:y))|(?(:y)(:x))/
946 foreach ($this->leaves as $leaf)
947 $regexp .= $leaf->regexp();
953 * A (possibly multi-argument) 'OR' conjoin.
955 class TextSearchQuery_node_or
956 extends TextSearchQuery_node_binop
962 // We will combine any of our direct descendents which are WORDs
963 // into a single (?=.*(?:word1|word2|...)) regexp.
968 foreach ($this->leaves as $leaf) {
969 if ($leaf->op == TSQ_TOK_WORD)
970 $words[] = preg_quote($leaf->word, '/');
972 $regexps[] = $leaf->regexp();
976 array_unshift($regexps,
977 '(?=.*' . $this->join($words) . ')');
979 return $this->join($regexps);
982 private function join($regexps)
984 assert(count($regexps) > 0);
986 if (count($regexps) > 1)
987 return '(?:' . join('|', $regexps) . ')';
993 ////////////////////////////////////////////////////////////////
996 // op's (and, or, not) are forced to lowercase in the tokenizer.
998 ////////////////////////////////////////////////////////////////
999 class TextSearchQuery_Parser
1002 * This is a simple recursive descent parser, based on the following grammar:
1017 * atom : '(' list ')'
1022 * The terminal tokens are:
1029 * /[^-()\s][^()\s]* WORD
1033 * ^WORD TextSearchQuery_phrase_starts_with
1043 function parse($search_expr, $case_exact = false, $regex = TSQ_REGEX_AUTO)
1045 $this->lexer = new TextSearchQuery_Lexer($search_expr, $case_exact, $regex);
1046 $this->regex = $regex;
1047 $tree = $this->get_list('toplevel');
1048 // Assert failure when using the following URL in debug mode.
1049 // /TitleSearch?action=FullTextSearch&s=WFXSSProbe'")/>&case_exact=1®ex=sql
1050 // assert($this->lexer->eof());
1051 unset($this->lexer);
1055 function get_list($is_toplevel = false)
1059 // token types we'll accept as words (and thus expr's) for the
1060 // purpose of error recovery:
1061 $accept_as_words = TSQ_TOK_NOT | TSQ_TOK_BINOP;
1063 $accept_as_words |= TSQ_TOK_LPAREN | TSQ_TOK_RPAREN;
1065 while (($expr = $this->get_expr())
1066 || ($expr = $this->get_word($accept_as_words))) {
1072 return new TextSearchQuery_node;
1076 if ($is_toplevel and count($list) == 1) {
1077 if ($this->lexer->query_str[0] == '^')
1078 return new TextSearchQuery_phrase_starts_with($list[0]->word);
1082 return new TextSearchQuery_node_and($list);
1087 if (($expr = $this->get_atom()) === false) // protect against '0'
1090 $savedpos = $this->lexer->tell();
1091 // Bug#1791564: allow string '0'
1092 while (($op = $this->lexer->get(TSQ_TOK_BINOP)) !== false) {
1093 if (!($right = $this->get_atom())) {
1098 $expr = new TextSearchQuery_node_and(array($expr, $right));
1100 assert($op == 'or');
1101 $expr = new TextSearchQuery_node_or(array($expr, $right));
1104 $savedpos = $this->lexer->tell();
1106 $this->lexer->seek($savedpos);
1113 if ($atom = $this->get_word(TSQ_ALLWORDS)) // Bug#1791564 not involved: '*'
1116 $savedpos = $this->lexer->tell();
1117 if ($this->lexer->get(TSQ_TOK_LPAREN)) {
1118 if (($list = $this->get_list()) && $this->lexer->get(TSQ_TOK_RPAREN)) {
1122 // Handle " ( " or "(test" without closing ")" as plain word
1123 $this->lexer->seek($savedpos);
1124 return new TextSearchQuery_node_word($this->lexer->get(-1));
1126 } elseif ($this->lexer->get(TSQ_TOK_NOT)) {
1127 if (($atom = $this->get_atom()))
1128 return new TextSearchQuery_node_not($atom);
1130 $this->lexer->seek($savedpos);
1134 function get_word($accept = TSQ_ALLWORDS)
1136 // Performance shortcut for ( and ). This is always false
1137 if (!empty($this->lexer->tokens[$this->lexer->pos])) {
1138 list ($type, $val) = $this->lexer->tokens[$this->lexer->pos];
1139 if ($type == TSQ_TOK_LPAREN or $type == TSQ_TOK_RPAREN)
1142 foreach (array("WORD", "STARTS_WITH", "ENDS_WITH", "EXACT",
1143 "REGEX", "REGEX_GLOB", "REGEX_PCRE", "ALL") as $tok) {
1144 $const = constant("TSQ_TOK_" . $tok);
1145 // Bug#1791564: allow word '0'
1146 if ($accept & $const and
1147 (($word = $this->lexer->get($const)) !== false)
1149 // phrase or word level?
1150 if ($tok == 'STARTS_WITH' and $this->lexer->query_str[0] == '^')
1151 $classname = "TextSearchQuery_phrase_" . strtolower($tok);
1152 elseif ($tok == 'ENDS_WITH' and
1153 string_ends_with($this->lexer->query_str, '$')
1155 $classname = "TextSearchQuery_phrase_" . strtolower($tok); else
1156 $classname = "TextSearchQuery_node_" . strtolower($tok);
1157 return new $classname($word);
1164 class TextSearchQuery_Lexer
1166 function TextSearchQuery_Lexer($query_str, $case_exact = false,
1167 $regex = TSQ_REGEX_AUTO)
1169 $this->tokens = $this->tokenize($query_str, $case_exact, $regex);
1170 $this->query_str = $query_str;
1186 return $this->pos == count($this->tokens);
1190 * TODO: support more regex styles, esp. prefer the forced ones over auto
1193 function tokenize($string, $case_exact = false, $regex = TSQ_REGEX_AUTO)
1196 $buf = $case_exact ? ltrim($string) : strtolower(ltrim($string));
1197 while (!empty($buf)) {
1198 if (preg_match('/^([()])\s*/', $buf, $m)) {
1200 $type = $m[1] == '(' ? TSQ_TOK_LPAREN : TSQ_TOK_RPAREN;
1202 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_GLOB)
1203 and preg_match('/^\*\s*/', $buf, $m)
1206 $type = TSQ_TOK_ALL;
1208 elseif ($regex & (TSQ_REGEX_PCRE)
1209 and preg_match('/^\.\*\s*/', $buf, $m)
1212 $type = TSQ_TOK_ALL;
1214 elseif ($regex & (TSQ_REGEX_SQL)
1215 and preg_match('/^%\s*/', $buf, $m)
1218 $type = TSQ_TOK_ALL;
1220 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_PCRE)
1221 and preg_match('/^\^([^-()][^()\s]*)\s*/', $buf, $m)
1224 $type = TSQ_TOK_STARTS_WITH;
1226 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_GLOB)
1227 and preg_match('/^([^-()][^()\s]*)\*\s*/', $buf, $m)
1230 $type = TSQ_TOK_STARTS_WITH;
1232 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_GLOB)
1233 and preg_match('/^\*([^-()][^()\s]*)\s*/', $buf, $m)
1236 $type = TSQ_TOK_ENDS_WITH;
1238 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_PCRE)
1239 and preg_match('/^([^-()][^()\s]*)\$\s*/', $buf, $m)
1242 $type = TSQ_TOK_ENDS_WITH;
1244 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_PCRE)
1245 and preg_match('/^\^([^-()][^()\s]*)\$\s*/', $buf, $m)
1248 $type = TSQ_TOK_EXACT;
1249 } elseif (preg_match('/^(and|or)\b\s*/i', $buf, $m)) {
1250 $val = strtolower($m[1]);
1251 $type = TSQ_TOK_BINOP;
1252 } elseif (preg_match('/^(-|not\b)\s*/i', $buf, $m)) {
1253 $val = strtolower($m[1]);
1254 $type = TSQ_TOK_NOT;
1256 elseif (preg_match('/^ " ( (?: [^"]+ | "" )* ) " \s*/x', $buf, $m)) {
1257 $val = str_replace('""', '"', $m[1]);
1258 $type = TSQ_TOK_WORD;
1260 elseif (preg_match("/^ ' ( (?:[^']+|'')* ) ' \s*/x", $buf, $m)) {
1261 $val = str_replace("''", "'", $m[1]);
1262 $type = TSQ_TOK_WORD;
1264 elseif (preg_match('/^([^-()][^()\s]*)\s*/', $buf, $m)) {
1266 $type = TSQ_TOK_WORD;
1268 assert(empty($buf));
1271 $buf = substr($buf, strlen($m[0]));
1273 /* refine the simple parsing from above: bla*bla, bla?bla, ...
1274 if ($regex and $type == TSQ_TOK_WORD) {
1275 if (substr($val,0,1) == "^")
1276 $type = TSQ_TOK_STARTS_WITH;
1277 elseif (substr($val,0,1) == "*")
1278 $type = TSQ_TOK_ENDS_WITH;
1279 elseif (substr($val,-1,1) == "*")
1280 $type = TSQ_TOK_STARTS_WITH;
1283 $tokens[] = array($type, $val);
1288 function get($accept)
1290 if ($this->pos >= count($this->tokens))
1293 list ($type, $val) = $this->tokens[$this->pos];
1294 if (($type & $accept) == 0)
1305 // c-basic-offset: 4
1306 // c-hanging-comment-ender-p: nil
1307 // indent-tabs-mode: nil