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));
84 define ('TSQ_NODE_ALL', 0);
85 define ('TSQ_NODE_EXACT', 1);
86 define ('TSQ_NODE_NOT', 2);
87 define ('TSQ_NODE_WORD', 3);
88 define ('TSQ_NODE_STARTS_WITH', 4);
89 define ('TSQ_NODE_ENDS_WITH', 5);
90 define ('TSQ_NODE_REGEX', 6);
91 define ('TSQ_NODE_REGEX_GLOB', 7);
92 define ('TSQ_NODE_REGEX_PCRE', 8);
93 define ('TSQ_NODE_REGEX_SQL', 9);
94 define ('TSQ_NODE_AND', 10);
95 define ('TSQ_NODE_OR', 11);
102 * Create a new query.
104 * @param $search_query string The query. Syntax is as described above.
105 * Note that an empty $search_query will match anything.
106 * @param $case_exact boolean
107 * @param $regex string one of 'auto', 'none', 'glob', 'posix', 'pcre', 'sql'
108 * @see TextSearchQuery
110 function TextSearchQuery($search_query, $case_exact = false, $regex = 'auto')
112 if ($regex == 'none' or !$regex)
114 elseif (defined("TSQ_REGEX_" . strtoupper($regex)))
115 $this->_regex = constant("TSQ_REGEX_" . strtoupper($regex)); else {
116 trigger_error(fmt("Unsupported argument: %s=%s", 'regex', $regex));
119 $this->_regex_modifier = ($case_exact ? '' : 'i') . 'sS';
120 $this->_case_exact = $case_exact;
121 if ($regex != 'pcre') {
122 $parser = new TextSearchQuery_Parser;
123 $this->_tree = $parser->parse($search_query, $case_exact, $this->_regex);
124 $this->_optimize(); // broken under certain circumstances: "word -word -word"
125 if (defined("FULLTEXTSEARCH_STOPLIST"))
126 $this->_stoplist = FULLTEXTSEARCH_STOPLIST;
127 else // default stoplist, localizable.
128 $this->_stoplist = _("(A|An|And|But|By|For|From|In|Is|It|Of|On|Or|The|To|With)");
130 $this->_tree = new TextSearchQuery_node_regex_pcre($search_query);
131 if (preg_match("/^\/(.*)\/(\w*)$/", $search_query, $m)) {
132 $this->_tree->word = $m[1];
133 $this->_regex_modifier = $m[2]; // overrides case_exact
145 $this->_tree = $this->_tree->optimize();
149 * Get a PCRE regexp which matches the query.
153 if (!isset($this->_regexp)) {
154 if (!isset($this->_regex_modifier))
155 $this->_regex_modifier = ($this->_case_exact ? '' : 'i') . 'sS';
157 $this->_regexp = '/' . $this->_tree->regexp() . '/' . $this->_regex_modifier;
159 $this->_regexp = '/^' . $this->_tree->regexp() . '/' . $this->_regex_modifier;
161 return $this->_regexp;
165 * Match query against string.
166 * EXACT ("Term") ignores the case_exact setting.
168 * @param $string string The string to match.
169 * @return boolean True if the string matches the query.
171 function match($string)
173 if ($this->_tree->_op == TSQ_TOK_ALL) return true;
174 if ($this->_tree->_op == TSQ_TOK_EXACT) return $this->_tree->word == $string;
175 return preg_match($this->asRegexp(), $string);
178 /* How good does it match? Returns a number */
179 function score($string)
183 foreach (array_unique($this->_tree->highlight_words()) as $word) {
184 if ($nummatch = preg_match_all("/" . preg_quote($word, '/') . "/" .
185 $this->_regex_modifier,
188 $score += ($i-- * $nummatch);
190 return min(1.0, $score / 10.0);
195 * Get a regular expression suitable for highlighting matched words.
197 * This returns a PCRE regular expression which matches any non-negated
200 * @return string The PCRE regexp.
202 function getHighlightRegexp()
204 if (!isset($this->_hilight_regexp)) {
205 $words = array_unique($this->_tree->highlight_words());
207 $this->_hilight_regexp = false;
209 foreach ($words as $key => $word)
210 $words[$key] = preg_quote($word, '/');
211 $this->_hilight_regexp = '(?' . ($this->_case_exact ? '' : 'i') . ':'
212 . join('|', $words) . ')';
215 return $this->_hilight_regexp;
219 * Make an SQL clause which matches the query.
220 * Deprecated, use makeSqlClauseObj instead.
222 * @param $make_sql_clause_cb WikiCallback
223 * A callback which takes a single word as an argument and
224 * returns an SQL clause which will match exactly those records
225 * containing the word. The word passed to the callback will always
226 * be in all lower case.
228 * Support db-specific extensions, like MATCH AGAINST or REGEX
229 * mysql => 4.0.1 can also do Google: MATCH AGAINST IN BOOLEAN MODE
230 * by using makeSqlClauseObj
234 * function sql_title_match($word) {
235 * return sprintf("LOWER(title) like '%s'",
236 * addslashes($word));
241 * $query = new TextSearchQuery("wiki -page");
242 * $cb = new WikiFunctionCb('sql_title_match');
243 * $sql_clause = $query->makeSqlClause($cb);
245 * This will result in $sql_clause containing something like
246 * "(LOWER(title) like 'wiki') AND NOT (LOWER(title) like 'page')".
248 * @return string The SQL clause.
250 function makeSqlClause($sql_clause_cb)
252 $this->_sql_clause_cb = $sql_clause_cb;
253 return $this->_sql_clause($this->_tree);
256 // deprecated: use _sql_clause_obj now.
257 function _sql_clause($node)
259 switch ($node->_op) {
260 case TSQ_TOK_WORD: // word => %word%
261 return $this->_sql_clause_cb->call($node->word);
263 return "NOT (" . $this->_sql_clause($node->leaves[0]) . ")";
265 $subclauses = array();
266 foreach ($node->leaves as $leaf)
267 $subclauses[] = "(" . $this->_sql_clause($leaf) . ")";
268 return join(" $node->op ", $subclauses);
270 assert($node->_op == TSQ_TOK_VOID);
275 /** Get away with the callback and use a db-specific search class instead.
276 * @see WikiDB_backend_PearDB_search
278 function makeSqlClauseObj(&$sql_search_cb)
280 $this->_sql_clause_cb = $sql_search_cb;
281 return $this->_sql_clause_obj($this->_tree);
284 function _sql_clause_obj($node)
286 switch ($node->_op) {
288 return "NOT (" . $this->_sql_clause_cb->call($node->leaves[0]) . ")";
290 $subclauses = array();
291 foreach ($node->leaves as $leaf)
292 $subclauses[] = "(" . $this->_sql_clause_obj($leaf) . ")";
293 return join(" $node->op ", $subclauses);
299 return $this->_sql_clause_cb->call($node);
304 postgresql tsearch2 uses no WHERE operators, just & | and ! in the searchstring
306 function makeTsearch2SqlClauseObj(&$sql_search_cb)
308 $this->_sql_clause_cb = $sql_search_cb;
309 return $this->_Tsearch2Sql_clause_obj($this->_tree);
312 function _Tsearch2Sql_clause_obj($node)
314 // TODO: "such a phrase"
315 switch ($node->_op) {
317 return "!" . $node->leaves[0];
319 $subclauses = array();
320 foreach ($node->leaves as $leaf)
321 $subclauses[] = $this->_Tsearch2Sql_clause_obj($leaf);
322 return join($node->_op == 'AND' ? "&" : "|", $subclauses);
328 return $this->_sql_clause_cb->call($node);
334 return '%' . $this->_sql_quote($this->word) . '%';
338 * Get printable representation of the parse tree.
340 * This is for debugging only.
341 * @return string Printable parse tree.
345 return $this->_as_string($this->_tree);
348 function _as_string($node, $indent = '')
350 switch ($node->_op) {
352 return $indent . "WORD: $node->word";
354 return $indent . "VOID";
356 return $indent . "ALL";
358 $lines = array($indent . $node->op . ":");
360 foreach ($node->leaves as $leaf)
361 $lines[] = $this->_as_string($leaf, $indent);
362 return join("\n", $lines);
368 * This is a TextSearchQuery which matches nothing.
370 class NullTextSearchQuery extends TextSearchQuery
373 * Create a new query.
375 * @see TextSearchQuery
377 function NullTextSearchQuery()
386 function match($string)
391 function getHighlightRegexp()
396 function makeSqlClause($make_sql_clause_cb)
403 return "NullTextSearchQuery";
410 * A simple algebraic matcher for numeric attributes.
411 * NumericSearchQuery can do ("population < 20000 and area > 1000000", array("population", "area"))
412 * ->match(array('population' => 100000, 'area' => 10000000))
414 * Supports all mathematical PHP comparison operators, plus ':=' for equality.
415 * "(x < 2000000 and x >= 10000) or (x >= 100 and x < 2000)"
416 * "x := 100000" is the same as "x == 100000"
418 * Since this is basic numerics only, we simply try to get away with
419 * replacing the variable values at the right positions and do an eval then.
421 * @package NumericSearchQuery
422 * @author Reini Urban
423 * @see SemanticAttributeSearchQuery
425 class NumericSearchQuery
428 * Create a new query.
429 * NumericSearchQuery("population > 20000 or population < 200", "population")
430 * NumericSearchQuery("population < 20000 and area > 1000000", array("population", "area"))
432 * With a single variable it is easy: The valid name must be matched elsewhere, just
433 * replace the given number in match in the query.
436 * With matching a struct we need strict names, no * as name is allowed.
437 * So always when the placeholder is an array, the names of the target struct must match
438 * and all vars be defined. Use the method can_match($struct) therefore.
441 * @param $search_query string A numerical query with placeholders as variable.
442 * @param $placeholders array or string All placeholders in the query must be defined
443 * here, and will be replaced by the matcher.
445 function NumericSearchQuery($search_query, $placeholders)
447 // added some basic security checks against user input
448 $this->_query = $search_query;
449 $this->_placeholders = $placeholders;
451 // we should also allow the M_ constants
452 $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');
453 $this->_allowed_operators = explode(',', '-,<,<=,>,>=,==,!=,*,+,/,(,),%,and,or,xor,<<,>>,===,!==,&,^,|,&&,||');
454 $this->_parser_check = array();
455 // check should be fast, so make a hash
456 foreach ($this->_allowed_functions as $f)
457 $this->_parser_check[$f] = 1;
458 foreach ($this->_allowed_operators as $f)
459 $this->_parser_check[$f] = 1;
460 if (is_array($placeholders))
461 foreach ($placeholders as $f)
462 $this->_parser_check[$f] = 1;
463 else $this->_parser_check[$placeholders] = 1;
465 // This is a speciality: := looks like the attribute definition and is
466 // therefore a dummy check for this definition.
467 // php-4.2.2 has a problem with /\b:=\b/ matching "population := 1223400"
468 $this->_query = preg_replace("/:=/", "==", $this->_query);
469 $this->_query = $this->check_query($this->_query);
478 * Check the symbolic definition query against unwanted functions and characters.
479 * "population < 20000 and area > 1000000" vs
480 * "area > 1000000 and mail($me,file("/etc/passwd"),...)"
481 * http://localhost/wikicvs/SemanticSearch?attribute=*&attr_op=<0 and find(1)>&s=-0.01&start_debug=1
483 function check_query($query)
485 $tmp = $query; // check for all function calls, in case the tokenizer is not available.
486 while (preg_match("/([a-z][a-z0-9]+)\s*\((.*)$/i", $tmp, $m)) {
487 if (!in_array($m[1], $this->_allowed_functions)
488 and !in_array($m[1], $this->_allowed_operators)
490 trigger_error("Illegal function in query: " . $m[1], E_USER_WARNING);
496 // Strictly check for illegal functions and operators, which are no placeholders.
497 if (function_exists('token_get_all')) {
498 $parsed = token_get_all("<?$query?>");
499 foreach ($parsed as $x) { // flat, non-recursive array
500 if (is_string($x) and !isset($this->_parser_check[$x])) {
501 // single char op or name
502 trigger_error("Illegal string or operator in query: \"$x\"", E_USER_WARNING);
504 } elseif (is_array($x)) {
505 $n = token_name($x[0]);
506 if ($n == 'T_OPEN_TAG' or $n == 'T_WHITESPACE'
507 or $n == 'T_CLOSE_TAG' or $n == 'T_LNUMBER'
508 or $n == 'T_CONST' or $n == 'T_DNUMBER'
510 if ($n == 'T_VARIABLE') { // but we do allow consts
511 trigger_error("Illegal variable in query: \"$x[1]\"", E_USER_WARNING);
514 if (is_string($x[1]) and !isset($this->_parser_check[$x[1]])) {
515 // multi-char char op or name
516 trigger_error("Illegal $n in query: \"$x[1]\"", E_USER_WARNING);
521 //echo "$query <br>";
522 //$this->_parse_token($parsed);
527 { T_OPEN_TAG "<?"} { T_STRING "_x"} { T_WHITESPACE " "} ">" { T_WHITESPACE " "} { T_LNUMBER "0"} { T_CLOSE_TAG "?>"}
528 Interesting: on-char ops, as ">" are not tokenized.
530 { T_OPEN_TAG "< ?" } { T_STRING "_x" } { T_WHITESPACE " " } { T_IS_SMALLER_OR_EQUAL "<=" } { T_WHITESPACE " " } { T_LNUMBER "0" } { T_CLOSE_TAG "?>" }
533 // Detect illegal characters besides nums, words and ops.
534 // So attribute names can not be utf-8
535 $c = "/([^\d\w.,\s" . preg_quote(join("", $this->_allowed_operators), "/") . "])/";
536 if (preg_match($c, $query, $m)) {
537 trigger_error("Illegal character in query: " . $m[1], E_USER_WARNING);
545 * Check the bound, numeric-only query against unwanted functions and sideeffects.
546 * "4560000 < 20000 and 1456022 > 1000000"
548 function _live_check()
550 // TODO: check $this->_workquery again?
551 return !empty($this->_workquery);
555 * A numeric query can only operate with predefined variables. "x < 0 and y < 1"
557 * @return array The names as array of strings. => ('x', 'y') the placeholders.
561 if (is_array($this->_placeholders)) return $this->_placeholders;
562 else return array($this->_placeholders);
566 * Strip non-numeric chars from the variable (as the groupseperator) and replace
567 * it in the symbolic query for evaluation.
570 * @param $value number A numerical value: integer, float or string.
571 * @param $x string The variable name to be replaced in the query.
574 function _bind($value, $x)
576 // TODO: check is_number, is_float, is_integer and do casting
577 $this->_bound[] = array('linkname' => $x,
578 'linkvalue' => $value);
579 $value = preg_replace("/[^-+0123456789.,]/", "", $value);
580 //$c = "/\b".preg_quote($x,"/")."\b/";
581 $this->_workquery = preg_replace("/\b" . preg_quote($x, "/") . "\b/", $value, $this->_workquery);
582 // FIXME: do again a final check. now only numbers and some operators are allowed.
583 return $this->_workquery;
586 /* array of successfully bound vars, and in case of success, the resulting vars
590 return $this->_bound;
594 * With an array of placeholders we need a hash to check against, if all required names are given.
595 * Purpose: Be silent about missing vars, just return false.
598 * @param $variable string or hash of name => value The keys must satisfy all placeholders in the definition.
599 * We want the full hash and not just the keys because a hash check is faster than the array of keys check.
602 function can_match(&$variables)
604 if (empty($this->_query))
606 $p =& $this->_placeholders;
607 if (!is_array($variables) and !is_array($p))
608 return $variables == $p; // This was easy.
609 // Check if all placeholders have definitions. can be overdefined but not underdefined.
611 if (!isset($variables[$p])) return false;
614 if (!isset($variables[$x])) return false;
621 * We can match against a single variable or against a hash of variables.
622 * With one placeholder we need just a number.
623 * With an array of placeholders we need a hash.
626 * @param $variable number or array of name => value The keys must satisfy all placeholders in the definition.
629 function match(&$variable)
631 $p =& $this->_placeholders;
632 $this->_workquery = $this->_query;
634 if (is_array($variable)) { // which var to match? we cannot decide this here
635 if (!isset($variable[$p]))
636 trigger_error("Required NumericSearchQuery->match variable $x not defined.", E_USER_ERROR);
637 $this->_bind($variable[$p], $p);
639 $this->_bind($variable, $p);
643 if (!isset($variable[$x]))
644 trigger_error("Required NumericSearchQuery->match variable $x not defined.", E_USER_ERROR);
645 $this->_bind($variable[$x], $x);
648 if (!$this->_live_check()) // check returned an error
650 $search = $this->_workquery;
652 //if (DEBUG & _DEBUG_VERBOSE)
653 // trigger_error("\$result = (boolean)($search);", E_USER_NOTICE);
654 // We might have a numerical problem:
655 // php-4.2.2 eval'ed as module: "9.636e+08 > 1000" false;
656 // php-5.1.2 cgi true, 4.2.2 cgi true
657 eval("\$result = (boolean)($search);");
658 if ($result and is_array($p)) {
659 return $this->_bound();
665 ////////////////////////////////////////////////////////////////
667 // Remaining classes are private.
669 ////////////////////////////////////////////////////////////////
671 * Virtual base class for nodes in a TextSearchQuery parse tree.
673 * Also serves as a 'VOID' (contentless) node.
675 class TextSearchQuery_node
681 * Optimize this node.
682 * @return object Optimized node.
690 * @return regexp matching this node.
698 * @param bool True if this node has been negated (higher in the parse tree.)
699 * @return array A list of all non-negated words contained by this node.
701 function highlight_words($negated = false)
711 function _sql_quote()
714 $word = preg_replace('/(?=[%_\\\\])/', "\\", $this->word);
715 return $request->_dbi->_backend->qstr($word);
720 * A word. Exact or substring?
722 class TextSearchQuery_node_word
723 extends TextSearchQuery_node
726 var $_op = TSQ_TOK_WORD;
728 function TextSearchQuery_node_word($word)
735 return '(?=.*\b' . preg_quote($this->word, '/') . '\b)';
738 function highlight_words($negated = false)
740 return $negated ? array() : array($this->word);
745 return '%' . $this->_sql_quote($this->word) . '%';
749 class TextSearchQuery_node_all
750 extends TextSearchQuery_node
753 var $_op = TSQ_TOK_ALL;
766 class TextSearchQuery_node_starts_with
767 extends TextSearchQuery_node_word
769 var $op = "STARTS_WITH";
770 var $_op = TSQ_TOK_STARTS_WITH;
774 return '(?=.*\b' . preg_quote($this->word, '/') . ')';
779 return $this->_sql_quote($this->word) . '%';
783 // ^word: full phrase starts with
784 class TextSearchQuery_phrase_starts_with
785 extends TextSearchQuery_node_starts_with
789 return '(?=^' . preg_quote($this->word, '/') . ')';
793 class TextSearchQuery_node_ends_with
794 extends TextSearchQuery_node_word
796 var $op = "ENDS_WITH";
797 var $_op = TSQ_TOK_ENDS_WITH;
801 return '(?=.*' . preg_quote($this->word, '/') . '\b)';
806 return '%' . $this->_sql_quote($this->word);
810 // word$: full phrase ends with
811 class TextSearchQuery_phrase_ends_with
812 extends TextSearchQuery_node_ends_with
816 return '(?=' . preg_quote($this->word, '/') . '$)';
820 class TextSearchQuery_node_exact
821 extends TextSearchQuery_node_word
824 var $_op = TSQ_TOK_EXACT;
828 return '(?=\b' . preg_quote($this->word, '/') . '\b)';
833 return $this->_sql_squote($this->word);
837 class TextSearchQuery_node_regex // posix regex. FIXME!
838 extends TextSearchQuery_node_word
840 var $op = "REGEX"; // using REGEXP or ~ extension
841 var $_op = TSQ_TOK_REGEX;
845 return '(?=.*\b' . $this->word . '\b)';
850 return $this->_sql_quote($this->word);
854 class TextSearchQuery_node_regex_glob
855 extends TextSearchQuery_node_regex
857 var $op = "REGEX_GLOB";
858 var $_op = TSQ_TOK_REGEX_GLOB;
862 return '(?=.*\b' . glob_to_pcre($this->word) . '\b)';
866 class TextSearchQuery_node_regex_pcre // how to handle pcre modifiers? /i
867 extends TextSearchQuery_node_regex
869 var $op = "REGEX_PCRE";
870 var $_op = TSQ_TOK_REGEX_PCRE;
878 class TextSearchQuery_node_regex_sql
879 extends TextSearchQuery_node_regex
881 var $op = "REGEX_SQL"; // using LIKE
882 var $_op = TSQ_TOK_REGEX_SQL;
886 return str_replace(array("/%/", "/_/"), array(".*", "."), $this->word);
898 class TextSearchQuery_node_not
899 extends TextSearchQuery_node
902 var $_op = TSQ_TOK_NOT;
904 function TextSearchQuery_node_not($leaf)
906 $this->leaves = array($leaf);
911 $leaf = &$this->leaves[0];
912 $leaf = $leaf->optimize();
913 if ($leaf->_op == TSQ_TOK_NOT)
914 return $leaf->leaves[0]; // ( NOT ( NOT x ) ) -> x
920 $leaf = &$this->leaves[0];
921 return '(?!' . $leaf->regexp() . ')';
924 function highlight_words($negated = false)
926 return $this->leaves[0]->highlight_words(!$negated);
931 * Virtual base class for 'AND' and 'OR conjoins.
933 class TextSearchQuery_node_binop
934 extends TextSearchQuery_node
936 var $_op = TSQ_TOK_BINOP;
938 function TextSearchQuery_node_binop($leaves)
940 $this->leaves = $leaves;
945 // This flattens e.g. (AND (AND a b) (OR c d) e)
946 // to (AND a b e (OR c d))
948 foreach ($this->leaves as $leaf) {
949 $leaf = $leaf->optimize();
950 if ($this->op == $leaf->op)
951 $flat = array_merge($flat, $leaf->leaves);
955 $this->leaves = $flat;
961 assert(!empty($this->leaves));
962 if (count($this->leaves) == 1)
963 return $this->leaves[0]; // (AND x) -> x
967 function highlight_words($negated = false)
970 foreach ($this->leaves as $leaf)
971 array_splice($words, 0, 0,
972 $leaf->highlight_words($negated));
978 * A (possibly multi-argument) 'AND' conjoin.
980 class TextSearchQuery_node_and
981 extends TextSearchQuery_node_binop
989 // Convert (AND (NOT a) (NOT b) c d) into (AND (NOT (OR a b)) c d).
990 // Since OR's are more efficient for regexp matching:
991 // (?!.*a)(?!.*b) vs (?!.*(?:a|b))
993 // Suck out the negated leaves.
995 foreach ($this->leaves as $key => $leaf) {
996 if ($leaf->_op == TSQ_TOK_NOT) {
997 $nots[] = $leaf->leaves[0];
998 unset($this->leaves[$key]);
1002 // Combine the negated leaves into a single negated or.
1004 $node = (new TextSearchQuery_node_not
1005 (new TextSearchQuery_node_or($nots)));
1006 array_unshift($this->leaves, $node->optimize());
1009 assert(!empty($this->leaves));
1010 if (count($this->leaves) == 1)
1011 return $this->leaves[0]; // (AND x) -> x
1016 * Either we need all combinations of all words to be position independent,
1017 * or we have to use multiple match calls for each AND
1018 * (AND x y) => /(?(:x)(:y))|(?(:y)(:x))/
1023 foreach ($this->leaves as $leaf)
1024 $regexp .= $leaf->regexp();
1030 * A (possibly multi-argument) 'OR' conjoin.
1032 class TextSearchQuery_node_or
1033 extends TextSearchQuery_node_binop
1039 // We will combine any of our direct descendents which are WORDs
1040 // into a single (?=.*(?:word1|word2|...)) regexp.
1045 foreach ($this->leaves as $leaf) {
1046 if ($leaf->op == TSQ_TOK_WORD)
1047 $words[] = preg_quote($leaf->word, '/');
1049 $regexps[] = $leaf->regexp();
1053 array_unshift($regexps,
1054 '(?=.*' . $this->_join($words) . ')');
1056 return $this->_join($regexps);
1059 function _join($regexps)
1061 assert(count($regexps) > 0);
1063 if (count($regexps) > 1)
1064 return '(?:' . join('|', $regexps) . ')';
1070 ////////////////////////////////////////////////////////////////
1073 // op's (and, or, not) are forced to lowercase in the tokenizer.
1075 ////////////////////////////////////////////////////////////////
1076 class TextSearchQuery_Parser
1079 * This is a simple recursive descent parser, based on the following grammar:
1094 * atom : '(' list ')'
1099 * The terminal tokens are:
1106 * /[^-()\s][^()\s]* WORD
1110 * ^WORD TextSearchQuery_phrase_starts_with
1117 function parse($search_expr, $case_exact = false, $regex = TSQ_REGEX_AUTO)
1119 $this->lexer = new TextSearchQuery_Lexer($search_expr, $case_exact, $regex);
1120 $this->_regex = $regex;
1121 $tree = $this->get_list('toplevel');
1122 // Assert failure when using the following URL in debug mode.
1123 // /TitleSearch?action=FullTextSearch&s=WFXSSProbe'")/>&case_exact=1®ex=sql
1124 // assert($this->lexer->eof());
1125 unset($this->lexer);
1129 function get_list($is_toplevel = false)
1133 // token types we'll accept as words (and thus expr's) for the
1134 // purpose of error recovery:
1135 $accept_as_words = TSQ_TOK_NOT | TSQ_TOK_BINOP;
1137 $accept_as_words |= TSQ_TOK_LPAREN | TSQ_TOK_RPAREN;
1139 while (($expr = $this->get_expr())
1140 || ($expr = $this->get_word($accept_as_words))) {
1146 return new TextSearchQuery_node;
1150 if ($is_toplevel and count($list) == 1) {
1151 if ($this->lexer->query_str[0] == '^')
1152 return new TextSearchQuery_phrase_starts_with($list[0]->word);
1156 return new TextSearchQuery_node_and($list);
1161 if (($expr = $this->get_atom()) === false) // protect against '0'
1164 $savedpos = $this->lexer->tell();
1165 // Bug#1791564: allow string '0'
1166 while (($op = $this->lexer->get(TSQ_TOK_BINOP)) !== false) {
1167 if (!($right = $this->get_atom())) {
1172 $expr = new TextSearchQuery_node_and(array($expr, $right));
1174 assert($op == 'or');
1175 $expr = new TextSearchQuery_node_or(array($expr, $right));
1178 $savedpos = $this->lexer->tell();
1180 $this->lexer->seek($savedpos);
1188 if ($atom = $this->get_word(TSQ_ALLWORDS)) // Bug#1791564 not involved: '*'
1191 $savedpos = $this->lexer->tell();
1192 if ($this->lexer->get(TSQ_TOK_LPAREN)) {
1193 if (($list = $this->get_list()) && $this->lexer->get(TSQ_TOK_RPAREN)) {
1197 // Handle " ( " or "(test" without closing ")" as plain word
1198 $this->lexer->seek($savedpos);
1199 return new TextSearchQuery_node_word($this->lexer->get(-1));
1201 } elseif ($this->lexer->get(TSQ_TOK_NOT)) {
1202 if (($atom = $this->get_atom()))
1203 return new TextSearchQuery_node_not($atom);
1205 $this->lexer->seek($savedpos);
1209 function get_word($accept = TSQ_ALLWORDS)
1211 // Performance shortcut for ( and ). This is always false
1212 if (!empty($this->lexer->tokens[$this->lexer->pos])) {
1213 list ($type, $val) = $this->lexer->tokens[$this->lexer->pos];
1214 if ($type == TSQ_TOK_LPAREN or $type == TSQ_TOK_RPAREN)
1217 foreach (array("WORD", "STARTS_WITH", "ENDS_WITH", "EXACT",
1218 "REGEX", "REGEX_GLOB", "REGEX_PCRE", "ALL") as $tok) {
1219 $const = constant("TSQ_TOK_" . $tok);
1220 // Bug#1791564: allow word '0'
1221 if ($accept & $const and
1222 (($word = $this->lexer->get($const)) !== false)
1224 // phrase or word level?
1225 if ($tok == 'STARTS_WITH' and $this->lexer->query_str[0] == '^')
1226 $classname = "TextSearchQuery_phrase_" . strtolower($tok);
1227 elseif ($tok == 'ENDS_WITH' and
1228 string_ends_with($this->lexer->query_str, '$')
1230 $classname = "TextSearchQuery_phrase_" . strtolower($tok); else
1231 $classname = "TextSearchQuery_node_" . strtolower($tok);
1232 return new $classname($word);
1239 class TextSearchQuery_Lexer
1241 function TextSearchQuery_Lexer($query_str, $case_exact = false,
1242 $regex = TSQ_REGEX_AUTO)
1244 $this->tokens = $this->tokenize($query_str, $case_exact, $regex);
1245 $this->query_str = $query_str;
1261 return $this->pos == count($this->tokens);
1265 * TODO: support more regex styles, esp. prefer the forced ones over auto
1268 function tokenize($string, $case_exact = false, $regex = TSQ_REGEX_AUTO)
1271 $buf = $case_exact ? ltrim($string) : strtolower(ltrim($string));
1272 while (!empty($buf)) {
1273 if (preg_match('/^([()])\s*/', $buf, $m)) {
1275 $type = $m[1] == '(' ? TSQ_TOK_LPAREN : TSQ_TOK_RPAREN;
1277 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_GLOB)
1278 and preg_match('/^\*\s*/', $buf, $m)
1281 $type = TSQ_TOK_ALL;
1283 elseif ($regex & (TSQ_REGEX_PCRE)
1284 and preg_match('/^\.\*\s*/', $buf, $m)
1287 $type = TSQ_TOK_ALL;
1289 elseif ($regex & (TSQ_REGEX_SQL)
1290 and preg_match('/^%\s*/', $buf, $m)
1293 $type = TSQ_TOK_ALL;
1295 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_PCRE)
1296 and preg_match('/^\^([^-()][^()\s]*)\s*/', $buf, $m)
1299 $type = TSQ_TOK_STARTS_WITH;
1301 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_GLOB)
1302 and preg_match('/^([^-()][^()\s]*)\*\s*/', $buf, $m)
1305 $type = TSQ_TOK_STARTS_WITH;
1307 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_GLOB)
1308 and preg_match('/^\*([^-()][^()\s]*)\s*/', $buf, $m)
1311 $type = TSQ_TOK_ENDS_WITH;
1313 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_PCRE)
1314 and preg_match('/^([^-()][^()\s]*)\$\s*/', $buf, $m)
1317 $type = TSQ_TOK_ENDS_WITH;
1319 elseif ($regex & (TSQ_REGEX_AUTO | TSQ_REGEX_POSIX | TSQ_REGEX_PCRE)
1320 and preg_match('/^\^([^-()][^()\s]*)\$\s*/', $buf, $m)
1323 $type = TSQ_TOK_EXACT;
1324 } elseif (preg_match('/^(and|or)\b\s*/i', $buf, $m)) {
1325 $val = strtolower($m[1]);
1326 $type = TSQ_TOK_BINOP;
1327 } elseif (preg_match('/^(-|not\b)\s*/i', $buf, $m)) {
1328 $val = strtolower($m[1]);
1329 $type = TSQ_TOK_NOT;
1331 elseif (preg_match('/^ " ( (?: [^"]+ | "" )* ) " \s*/x', $buf, $m)) {
1332 $val = str_replace('""', '"', $m[1]);
1333 $type = TSQ_TOK_WORD;
1335 elseif (preg_match("/^ ' ( (?:[^']+|'')* ) ' \s*/x", $buf, $m)) {
1336 $val = str_replace("''", "'", $m[1]);
1337 $type = TSQ_TOK_WORD;
1339 elseif (preg_match('/^([^-()][^()\s]*)\s*/', $buf, $m)) {
1341 $type = TSQ_TOK_WORD;
1343 assert(empty($buf));
1346 $buf = substr($buf, strlen($m[0]));
1348 /* refine the simple parsing from above: bla*bla, bla?bla, ...
1349 if ($regex and $type == TSQ_TOK_WORD) {
1350 if (substr($val,0,1) == "^")
1351 $type = TSQ_TOK_STARTS_WITH;
1352 elseif (substr($val,0,1) == "*")
1353 $type = TSQ_TOK_ENDS_WITH;
1354 elseif (substr($val,-1,1) == "*")
1355 $type = TSQ_TOK_STARTS_WITH;
1358 $tokens[] = array($type, $val);
1363 function get($accept)
1365 if ($this->pos >= count($this->tokens))
1368 list ($type, $val) = $this->tokens[$this->pos];
1369 if (($type & $accept) == 0)
1380 // c-basic-offset: 4
1381 // c-hanging-comment-ender-p: nil
1382 // indent-tabs-mode: nil