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);
99 class TextSearchQuery {
101 * Create a new query.
103 * @param $search_query string The query. Syntax is as described above.
104 * Note that an empty $search_query will match anything.
105 * @param $case_exact boolean
106 * @param $regex string one of 'auto', 'none', 'glob', 'posix', 'pcre', 'sql'
107 * @see TextSearchQuery
109 function TextSearchQuery($search_query, $case_exact=false, $regex='auto') {
110 if ($regex == 'none' or !$regex)
112 elseif (defined("TSQ_REGEX_".strtoupper($regex)))
113 $this->_regex = constant("TSQ_REGEX_".strtoupper($regex));
115 trigger_error(fmt("Unsupported argument: %s=%s", 'regex', $regex));
118 $this->_regex_modifier = ($case_exact?'':'i').'sS';
119 $this->_case_exact = $case_exact;
120 if ($regex != 'pcre') {
121 $parser = new TextSearchQuery_Parser;
122 $this->_tree = $parser->parse($search_query, $case_exact, $this->_regex);
123 $this->_optimize(); // broken under certain circumstances: "word -word -word"
124 if (defined("FULLTEXTSEARCH_STOPLIST"))
125 $this->_stoplist = FULLTEXTSEARCH_STOPLIST;
126 else // default stoplist, localizable.
127 $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
138 function getType() { return 'text'; }
140 function _optimize() {
141 $this->_tree = $this->_tree->optimize();
145 * Get a PCRE regexp which matches the query.
147 function asRegexp() {
148 if (!isset($this->_regexp)) {
149 if (!isset($this->_regex_modifier))
150 $this->_regex_modifier = ($this->_case_exact?'':'i').'sS';
152 $this->_regexp = '/' . $this->_tree->regexp() . '/'.$this->_regex_modifier;
154 $this->_regexp = '/^' . $this->_tree->regexp() . '/'.$this->_regex_modifier;
156 return $this->_regexp;
160 * Match query against string.
161 * EXACT ("Term") ignores the case_exact setting.
163 * @param $string string The string to match.
164 * @return boolean True if the string matches the query.
166 function match($string) {
167 if ($this->_tree->_op == TSQ_TOK_ALL) return true;
168 if ($this->_tree->_op == TSQ_TOK_EXACT) return $this->_tree->word == $string;
169 return preg_match($this->asRegexp(), $string);
172 /* How good does it match? Returns a number */
173 function score($string) {
176 foreach (array_unique($this->_tree->highlight_words()) as $word) {
177 if ($nummatch = preg_match_all("/".preg_quote($word, '/')."/".
178 $this->_regex_modifier,
180 $score += ($i-- * $nummatch);
182 return min(1.0, $score / 10.0);
187 * Get a regular expression suitable for highlighting matched words.
189 * This returns a PCRE regular expression which matches any non-negated
192 * @return string The PCRE regexp.
194 function getHighlightRegexp() {
195 if (!isset($this->_hilight_regexp)) {
196 $words = array_unique($this->_tree->highlight_words());
198 $this->_hilight_regexp = false;
200 foreach ($words as $key => $word)
201 $words[$key] = preg_quote($word, '/');
202 $this->_hilight_regexp = '(?'.($this->_case_exact?'':'i').':'
203 . join('|', $words) . ')';
206 return $this->_hilight_regexp;
210 * Make an SQL clause which matches the query.
211 * Deprecated, use makeSqlClauseObj instead.
213 * @param $make_sql_clause_cb WikiCallback
214 * A callback which takes a single word as an argument and
215 * returns an SQL clause which will match exactly those records
216 * containing the word. The word passed to the callback will always
217 * be in all lower case.
219 * Support db-specific extensions, like MATCH AGAINST or REGEX
220 * mysql => 4.0.1 can also do Google: MATCH AGAINST IN BOOLEAN MODE
221 * by using makeSqlClauseObj
225 * function sql_title_match($word) {
226 * return sprintf("LOWER(title) like '%s'",
227 * addslashes($word));
232 * $query = new TextSearchQuery("wiki -page");
233 * $cb = new WikiFunctionCb('sql_title_match');
234 * $sql_clause = $query->makeSqlClause($cb);
236 * This will result in $sql_clause containing something like
237 * "(LOWER(title) like 'wiki') AND NOT (LOWER(title) like 'page')".
239 * @return string The SQL clause.
241 function makeSqlClause($sql_clause_cb) {
242 $this->_sql_clause_cb = $sql_clause_cb;
243 return $this->_sql_clause($this->_tree);
245 // deprecated: use _sql_clause_obj now.
246 function _sql_clause($node) {
247 switch ($node->_op) {
248 case TSQ_TOK_WORD: // word => %word%
249 return $this->_sql_clause_cb->call($node->word);
251 return "NOT (" . $this->_sql_clause($node->leaves[0]) . ")";
253 $subclauses = array();
254 foreach ($node->leaves as $leaf)
255 $subclauses[] = "(" . $this->_sql_clause($leaf) . ")";
256 return join(" $node->op ", $subclauses);
258 assert($node->_op == TSQ_TOK_VOID);
263 /** Get away with the callback and use a db-specific search class instead.
264 * @see WikiDB_backend_PearDB_search
266 function makeSqlClauseObj(&$sql_search_cb) {
267 $this->_sql_clause_cb = $sql_search_cb;
268 return $this->_sql_clause_obj($this->_tree);
271 function _sql_clause_obj($node) {
272 switch ($node->_op) {
274 return "NOT (" . $this->_sql_clause_cb->call($node->leaves[0]) . ")";
276 $subclauses = array();
277 foreach ($node->leaves as $leaf)
278 $subclauses[] = "(" . $this->_sql_clause_obj($leaf) . ")";
279 return join(" $node->op ", $subclauses);
285 return $this->_sql_clause_cb->call($node);
290 postgresql tsearch2 uses no WHERE operators, just & | and ! in the searchstring
292 function makeTsearch2SqlClauseObj(&$sql_search_cb) {
293 $this->_sql_clause_cb = $sql_search_cb;
294 return $this->_Tsearch2Sql_clause_obj($this->_tree);
297 function _Tsearch2Sql_clause_obj($node) {
298 // TODO: "such a phrase"
299 switch ($node->_op) {
301 return "!" . $node->leaves[0];
303 $subclauses = array();
304 foreach ($node->leaves as $leaf)
305 $subclauses[] = $this->_Tsearch2Sql_clause_obj($leaf);
306 return join($node->_op == 'AND' ? "&" : "|", $subclauses);
312 return $this->_sql_clause_cb->call($node);
316 function sql() { return '%'.$this->_sql_quote($this->word).'%'; }
319 * Get printable representation of the parse tree.
321 * This is for debugging only.
322 * @return string Printable parse tree.
324 function asString() {
325 return $this->_as_string($this->_tree);
328 function _as_string($node, $indent = '') {
329 switch ($node->_op) {
331 return $indent . "WORD: $node->word";
333 return $indent . "VOID";
335 return $indent . "ALL";
337 $lines = array($indent . $node->op . ":");
339 foreach ($node->leaves as $leaf)
340 $lines[] = $this->_as_string($leaf, $indent);
341 return join("\n", $lines);
347 * This is a TextSearchQuery which matches nothing.
349 class NullTextSearchQuery extends TextSearchQuery {
351 * Create a new query.
353 * @see TextSearchQuery
355 function NullTextSearchQuery() {}
356 function asRegexp() { return '/^(?!a)a/x'; }
357 function match($string) { return false; }
358 function getHighlightRegexp() { return ""; }
359 function makeSqlClause($make_sql_clause_cb) { return "(1 = 0)"; }
360 function asString() { return "NullTextSearchQuery"; }
364 * A simple algebraic matcher for numeric attributes.
365 * NumericSearchQuery can do ("population < 20000 and area > 1000000", array("population", "area"))
366 * ->match(array('population' => 100000, 'area' => 10000000))
368 * Supports all mathematical PHP comparison operators, plus ':=' for equality.
369 * "(x < 2000000 and x >= 10000) or (x >= 100 and x < 2000)"
370 * "x := 100000" is the same as "x == 100000"
372 * Since this is basic numerics only, we simply try to get away with
373 * replacing the variable values at the right positions and do an eval then.
375 * @package NumericSearchQuery
376 * @author Reini Urban
377 * @see SemanticAttributeSearchQuery
379 class NumericSearchQuery
382 * Create a new query.
383 * NumericSearchQuery("population > 20000 or population < 200", "population")
384 * NumericSearchQuery("population < 20000 and area > 1000000", array("population", "area"))
386 * With a single variable it is easy: The valid name must be matched elsewhere, just
387 * replace the given number in match in the query.
390 * With matching a struct we need strict names, no * as name is allowed.
391 * So always when the placeholder is an array, the names of the target struct must match
392 * and all vars be defined. Use the method can_match($struct) therefore.
395 * @param $search_query string A numerical query with placeholders as variable.
396 * @param $placeholders array or string All placeholders in the query must be defined
397 * here, and will be replaced by the matcher.
399 function NumericSearchQuery($search_query, $placeholders) {
400 // added some basic security checks against user input
401 $this->_query = $search_query;
402 $this->_placeholders = $placeholders;
404 // we should also allow the M_ constants
405 $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');
406 $this->_allowed_operators = explode(',', '-,<,<=,>,>=,==,!=,*,+,/,(,),%,and,or,xor,<<,>>,===,!==,&,^,|,&&,||');
407 $this->_parser_check = array();
408 // check should be fast, so make a hash
409 foreach ($this->_allowed_functions as $f)
410 $this->_parser_check[$f] = 1;
411 foreach ($this->_allowed_operators as $f)
412 $this->_parser_check[$f] = 1;
413 if (is_array($placeholders))
414 foreach ($placeholders as $f)
415 $this->_parser_check[$f] = 1;
416 else $this->_parser_check[$placeholders] = 1;
418 // This is a speciality: := looks like the attribute definition and is
419 // therefore a dummy check for this definition.
420 // php-4.2.2 has a problem with /\b:=\b/ matching "population := 1223400"
421 $this->_query = preg_replace("/:=/", "==", $this->_query);
422 $this->_query = $this->check_query($this->_query);
425 function getType() { return 'numeric'; }
428 * Check the symbolic definition query against unwanted functions and characters.
429 * "population < 20000 and area > 1000000" vs
430 * "area > 1000000 and mail($me,file("/etc/passwd"),...)"
431 * http://localhost/wikicvs/SemanticSearch?attribute=*&attr_op=<0 and find(1)>&s=-0.01&start_debug=1
433 function check_query ($query) {
434 $tmp = $query; // check for all function calls, in case the tokenizer is not available.
435 while (preg_match("/([a-z][a-z0-9]+)\s*\((.*)$/i", $tmp, $m)) {
436 if (!in_array($m[1], $this->_allowed_functions)
437 and !in_array($m[1], $this->_allowed_operators))
439 trigger_error("Illegal function in query: ".$m[1], E_USER_WARNING);
445 // Strictly check for illegal functions and operators, which are no placeholders.
446 if (function_exists('token_get_all')) {
447 $parsed = token_get_all("<?$query?>");
448 foreach ($parsed as $x) { // flat, non-recursive array
449 if (is_string($x) and !isset($this->_parser_check[$x])) {
450 // single char op or name
451 trigger_error("Illegal string or operator in query: \"$x\"", E_USER_WARNING);
454 elseif (is_array($x)) {
455 $n = token_name($x[0]);
456 if ($n == 'T_OPEN_TAG' or $n == 'T_WHITESPACE'
457 or $n == 'T_CLOSE_TAG' or $n == 'T_LNUMBER'
458 or $n == 'T_CONST' or $n == 'T_DNUMBER' ) continue;
459 if ($n == 'T_VARIABLE') { // but we do allow consts
460 trigger_error("Illegal variable in query: \"$x[1]\"", E_USER_WARNING);
463 if (is_string($x[1]) and !isset($this->_parser_check[$x[1]])) {
464 // multi-char char op or name
465 trigger_error("Illegal $n in query: \"$x[1]\"", E_USER_WARNING);
470 //echo "$query <br>";
471 //$this->_parse_token($parsed);
476 { T_OPEN_TAG "<?"} { T_STRING "_x"} { T_WHITESPACE " "} ">" { T_WHITESPACE " "} { T_LNUMBER "0"} { T_CLOSE_TAG "?>"}
477 Interesting: on-char ops, as ">" are not tokenized.
479 { T_OPEN_TAG "< ?" } { T_STRING "_x" } { T_WHITESPACE " " } { T_IS_SMALLER_OR_EQUAL "<=" } { T_WHITESPACE " " } { T_LNUMBER "0" } { T_CLOSE_TAG "?>" }
482 // Detect illegal characters besides nums, words and ops.
483 // So attribute names can not be utf-8
484 $c = "/([^\d\w.,\s".preg_quote(join("",$this->_allowed_operators),"/")."])/";
485 if (preg_match($c, $query, $m)) {
486 trigger_error("Illegal character in query: ".$m[1], E_USER_WARNING);
494 * Check the bound, numeric-only query against unwanted functions and sideeffects.
495 * "4560000 < 20000 and 1456022 > 1000000"
497 function _live_check () {
498 // TODO: check $this->_workquery again?
499 return !empty($this->_workquery);
503 * A numeric query can only operate with predefined variables. "x < 0 and y < 1"
505 * @return array The names as array of strings. => ('x', 'y') the placeholders.
508 if(is_array($this->_placeholders)) return $this->_placeholders;
509 else return array($this->_placeholders);
513 * Strip non-numeric chars from the variable (as the groupseperator) and replace
514 * it in the symbolic query for evaluation.
517 * @param $value number A numerical value: integer, float or string.
518 * @param $x string The variable name to be replaced in the query.
521 function _bind($value, $x) {
522 // TODO: check is_number, is_float, is_integer and do casting
523 $this->_bound[] = array('linkname' => $x,
524 'linkvalue' => $value);
525 $value = preg_replace("/[^-+0123456789.,]/", "", $value);
526 //$c = "/\b".preg_quote($x,"/")."\b/";
527 $this->_workquery = preg_replace("/\b".preg_quote($x,"/")."\b/", $value, $this->_workquery);
528 // FIXME: do again a final check. now only numbers and some operators are allowed.
529 return $this->_workquery;
532 /* array of successfully bound vars, and in case of success, the resulting vars
535 return $this->_bound;
539 * With an array of placeholders we need a hash to check against, if all required names are given.
540 * Purpose: Be silent about missing vars, just return false.
543 * @param $variable string or hash of name => value The keys must satisfy all placeholders in the definition.
544 * We want the full hash and not just the keys because a hash check is faster than the array of keys check.
547 function can_match(&$variables) {
548 if (empty($this->_query))
550 $p =& $this->_placeholders;
551 if (!is_array($variables) and !is_array($p))
552 return $variables == $p; // This was easy.
553 // Check if all placeholders have definitions. can be overdefined but not underdefined.
555 if (!isset($variables[$p])) return false;
558 if (!isset($variables[$x])) return false;
565 * We can match against a single variable or against a hash of variables.
566 * With one placeholder we need just a number.
567 * With an array of placeholders we need a hash.
570 * @param $variable number or array of name => value The keys must satisfy all placeholders in the definition.
573 function match(&$variable) {
574 $p =& $this->_placeholders;
575 $this->_workquery = $this->_query;
577 if (is_array($variable)) { // which var to match? we cannot decide this here
578 if (!isset($variable[$p]))
579 trigger_error("Required NumericSearchQuery->match variable $x not defined.", E_USER_ERROR);
580 $this->_bind($variable[$p], $p);
582 $this->_bind($variable, $p);
586 if (!isset($variable[$x]))
587 trigger_error("Required NumericSearchQuery->match variable $x not defined.", E_USER_ERROR);
588 $this->_bind($variable[$x], $x);
591 if (!$this->_live_check()) // check returned an error
593 $search = $this->_workquery;
595 //if (DEBUG & _DEBUG_VERBOSE)
596 // trigger_error("\$result = (boolean)($search);", E_USER_NOTICE);
597 // We might have a numerical problem:
598 // php-4.2.2 eval'ed as module: "9.636e+08 > 1000" false;
599 // php-5.1.2 cgi true, 4.2.2 cgi true
600 eval("\$result = (boolean)($search);");
601 if ($result and is_array($p)) {
602 return $this->_bound();
608 ////////////////////////////////////////////////////////////////
610 // Remaining classes are private.
612 ////////////////////////////////////////////////////////////////
614 * Virtual base class for nodes in a TextSearchQuery parse tree.
616 * Also serves as a 'VOID' (contentless) node.
618 class TextSearchQuery_node
624 * Optimize this node.
625 * @return object Optimized node.
627 function optimize() {
632 * @return regexp matching this node.
639 * @param bool True if this node has been negated (higher in the parse tree.)
640 * @return array A list of all non-negated words contained by this node.
642 function highlight_words($negated = false) {
646 function sql() { return $this->word; }
648 function _sql_quote() {
650 $word = preg_replace('/(?=[%_\\\\])/', "\\", $this->word);
651 return $request->_dbi->_backend->qstr($word);
656 * A word. Exact or substring?
658 class TextSearchQuery_node_word
659 extends TextSearchQuery_node
662 var $_op = TSQ_TOK_WORD;
664 function TextSearchQuery_node_word($word) {
668 return '(?=.*\b' . preg_quote($this->word, '/') . '\b)';
670 function highlight_words ($negated = false) {
671 return $negated ? array() : array($this->word);
673 function sql() { return '%'.$this->_sql_quote($this->word).'%'; }
676 class TextSearchQuery_node_all
677 extends TextSearchQuery_node {
679 var $_op = TSQ_TOK_ALL;
680 function regexp() { return '(?=.*)'; }
681 function sql() { return '%'; }
684 class TextSearchQuery_node_starts_with
685 extends TextSearchQuery_node_word {
686 var $op = "STARTS_WITH";
687 var $_op = TSQ_TOK_STARTS_WITH;
688 function regexp() { return '(?=.*\b' . preg_quote($this->word, '/') . ')'; }
689 function sql () { return $this->_sql_quote($this->word).'%'; }
692 // ^word: full phrase starts with
693 class TextSearchQuery_phrase_starts_with
694 extends TextSearchQuery_node_starts_with {
695 function regexp() { return '(?=^' . preg_quote($this->word, '/') . ')'; }
698 class TextSearchQuery_node_ends_with
699 extends TextSearchQuery_node_word {
700 var $op = "ENDS_WITH";
701 var $_op = TSQ_TOK_ENDS_WITH;
702 function regexp() { return '(?=.*' . preg_quote($this->word, '/') . '\b)'; }
703 function sql () { return '%'.$this->_sql_quote($this->word); }
706 // word$: full phrase ends with
707 class TextSearchQuery_phrase_ends_with
708 extends TextSearchQuery_node_ends_with {
709 function regexp() { return '(?=' . preg_quote($this->word, '/') . '$)'; }
712 class TextSearchQuery_node_exact
713 extends TextSearchQuery_node_word {
715 var $_op = TSQ_TOK_EXACT;
716 function regexp() { return '(?=\b' . preg_quote($this->word, '/') . '\b)'; }
717 function sql () { return $this->_sql_squote($this->word); }
720 class TextSearchQuery_node_regex // posix regex. FIXME!
721 extends TextSearchQuery_node_word {
722 var $op = "REGEX"; // using REGEXP or ~ extension
723 var $_op = TSQ_TOK_REGEX;
724 function regexp() { return '(?=.*\b' . $this->word . '\b)'; }
725 function sql () { return $this->_sql_quote($this->word); }
728 class TextSearchQuery_node_regex_glob
729 extends TextSearchQuery_node_regex {
730 var $op = "REGEX_GLOB";
731 var $_op = TSQ_TOK_REGEX_GLOB;
732 function regexp() { return '(?=.*\b' . glob_to_pcre($this->word) . '\b)'; }
735 class TextSearchQuery_node_regex_pcre // how to handle pcre modifiers? /i
736 extends TextSearchQuery_node_regex {
737 var $op = "REGEX_PCRE";
738 var $_op = TSQ_TOK_REGEX_PCRE;
739 function regexp() { return $this->word; }
742 class TextSearchQuery_node_regex_sql
743 extends TextSearchQuery_node_regex {
744 var $op = "REGEX_SQL"; // using LIKE
745 var $_op = TSQ_TOK_REGEX_SQL;
746 function regexp() { return str_replace(array("/%/","/_/"), array(".*","."), $this->word); }
747 function sql() { return $this->word; }
753 class TextSearchQuery_node_not
754 extends TextSearchQuery_node
757 var $_op = TSQ_TOK_NOT;
759 function TextSearchQuery_node_not($leaf) {
760 $this->leaves = array($leaf);
763 function optimize() {
764 $leaf = &$this->leaves[0];
765 $leaf = $leaf->optimize();
766 if ($leaf->_op == TSQ_TOK_NOT)
767 return $leaf->leaves[0]; // ( NOT ( NOT x ) ) -> x
772 $leaf = &$this->leaves[0];
773 return '(?!' . $leaf->regexp() . ')';
776 function highlight_words ($negated = false) {
777 return $this->leaves[0]->highlight_words(!$negated);
782 * Virtual base class for 'AND' and 'OR conjoins.
784 class TextSearchQuery_node_binop
785 extends TextSearchQuery_node
787 var $_op = TSQ_TOK_BINOP;
788 function TextSearchQuery_node_binop($leaves) {
789 $this->leaves = $leaves;
792 function _flatten() {
793 // This flattens e.g. (AND (AND a b) (OR c d) e)
794 // to (AND a b e (OR c d))
796 foreach ($this->leaves as $leaf) {
797 $leaf = $leaf->optimize();
798 if ($this->op == $leaf->op)
799 $flat = array_merge($flat, $leaf->leaves);
803 $this->leaves = $flat;
806 function optimize() {
808 assert(!empty($this->leaves));
809 if (count($this->leaves) == 1)
810 return $this->leaves[0]; // (AND x) -> x
814 function highlight_words($negated = false) {
816 foreach ($this->leaves as $leaf)
817 array_splice($words,0,0,
818 $leaf->highlight_words($negated));
824 * A (possibly multi-argument) 'AND' conjoin.
826 class TextSearchQuery_node_and
827 extends TextSearchQuery_node_binop
831 function optimize() {
834 // Convert (AND (NOT a) (NOT b) c d) into (AND (NOT (OR a b)) c d).
835 // Since OR's are more efficient for regexp matching:
836 // (?!.*a)(?!.*b) vs (?!.*(?:a|b))
838 // Suck out the negated leaves.
840 foreach ($this->leaves as $key => $leaf) {
841 if ($leaf->_op == TSQ_TOK_NOT) {
842 $nots[] = $leaf->leaves[0];
843 unset($this->leaves[$key]);
847 // Combine the negated leaves into a single negated or.
849 $node = ( new TextSearchQuery_node_not
850 (new TextSearchQuery_node_or($nots)) );
851 array_unshift($this->leaves, $node->optimize());
854 assert(!empty($this->leaves));
855 if (count($this->leaves) == 1)
856 return $this->leaves[0]; // (AND x) -> x
861 * Either we need all combinations of all words to be position independent,
862 * or we have to use multiple match calls for each AND
863 * (AND x y) => /(?(:x)(:y))|(?(:y)(:x))/
867 foreach ($this->leaves as $leaf)
868 $regexp .= $leaf->regexp();
874 * A (possibly multi-argument) 'OR' conjoin.
876 class TextSearchQuery_node_or
877 extends TextSearchQuery_node_binop
882 // We will combine any of our direct descendents which are WORDs
883 // into a single (?=.*(?:word1|word2|...)) regexp.
888 foreach ($this->leaves as $leaf) {
889 if ($leaf->op == TSQ_TOK_WORD)
890 $words[] = preg_quote($leaf->word, '/');
892 $regexps[] = $leaf->regexp();
896 array_unshift($regexps,
897 '(?=.*' . $this->_join($words) . ')');
899 return $this->_join($regexps);
902 function _join($regexps) {
903 assert(count($regexps) > 0);
905 if (count($regexps) > 1)
906 return '(?:' . join('|', $regexps) . ')';
912 ////////////////////////////////////////////////////////////////
915 // op's (and, or, not) are forced to lowercase in the tokenizer.
917 ////////////////////////////////////////////////////////////////
918 class TextSearchQuery_Parser
921 * This is a simple recursive descent parser, based on the following grammar:
936 * atom : '(' list ')'
941 * The terminal tokens are:
948 * /[^-()\s][^()\s]* WORD
952 * ^WORD TextSearchQuery_phrase_starts_with
959 function parse ($search_expr, $case_exact=false, $regex=TSQ_REGEX_AUTO) {
960 $this->lexer = new TextSearchQuery_Lexer($search_expr, $case_exact, $regex);
961 $this->_regex = $regex;
962 $tree = $this->get_list('toplevel');
963 // Assert failure when using the following URL in debug mode.
964 // /TitleSearch?action=FullTextSearch&s=WFXSSProbe'")/>&case_exact=1®ex=sql
965 // assert($this->lexer->eof());
970 function get_list ($is_toplevel = false) {
973 // token types we'll accept as words (and thus expr's) for the
974 // purpose of error recovery:
975 $accept_as_words = TSQ_TOK_NOT | TSQ_TOK_BINOP;
977 $accept_as_words |= TSQ_TOK_LPAREN | TSQ_TOK_RPAREN;
979 while ( ($expr = $this->get_expr())
980 || ($expr = $this->get_word($accept_as_words)) )
987 return new TextSearchQuery_node;
991 if ($is_toplevel and count($list) == 1) {
992 if ($this->lexer->query_str[0] == '^')
993 return new TextSearchQuery_phrase_starts_with($list[0]->word);
997 return new TextSearchQuery_node_and($list);
1000 function get_expr () {
1001 if ( ($expr = $this->get_atom()) === false ) // protect against '0'
1004 $savedpos = $this->lexer->tell();
1005 // Bug#1791564: allow string '0'
1006 while ( ($op = $this->lexer->get(TSQ_TOK_BINOP)) !== false) {
1007 if ( ! ($right = $this->get_atom()) ) {
1012 $expr = new TextSearchQuery_node_and(array($expr, $right));
1014 assert($op == 'or');
1015 $expr = new TextSearchQuery_node_or(array($expr, $right));
1018 $savedpos = $this->lexer->tell();
1020 $this->lexer->seek($savedpos);
1026 function get_atom() {
1027 if ($atom = $this->get_word(TSQ_ALLWORDS)) // Bug#1791564 not involved: '*'
1030 $savedpos = $this->lexer->tell();
1031 if ( $this->lexer->get(TSQ_TOK_LPAREN) ) {
1032 if ( ($list = $this->get_list()) && $this->lexer->get(TSQ_TOK_RPAREN) ) {
1036 // Handle " ( " or "(test" without closing ")" as plain word
1037 $this->lexer->seek($savedpos);
1038 return new TextSearchQuery_node_word($this->lexer->get(-1));
1041 elseif ( $this->lexer->get(TSQ_TOK_NOT) ) {
1042 if ( ($atom = $this->get_atom()) )
1043 return new TextSearchQuery_node_not($atom);
1045 $this->lexer->seek($savedpos);
1049 function get_word($accept = TSQ_ALLWORDS) {
1050 // Performance shortcut for ( and ). This is always false
1051 if (!empty($this->lexer->tokens[$this->lexer->pos])) {
1052 list ($type, $val) = $this->lexer->tokens[$this->lexer->pos];
1053 if ($type == TSQ_TOK_LPAREN or $type == TSQ_TOK_RPAREN)
1056 foreach (array("WORD","STARTS_WITH","ENDS_WITH","EXACT",
1057 "REGEX","REGEX_GLOB","REGEX_PCRE","ALL") as $tok) {
1058 $const = constant("TSQ_TOK_".$tok);
1059 // Bug#1791564: allow word '0'
1060 if ( $accept & $const and
1061 (($word = $this->lexer->get($const)) !== false))
1063 // phrase or word level?
1064 if ($tok == 'STARTS_WITH' and $this->lexer->query_str[0] == '^')
1065 $classname = "TextSearchQuery_phrase_".strtolower($tok);
1066 elseif ($tok == 'ENDS_WITH' and
1067 string_ends_with($this->lexer->query_str,'$'))
1068 $classname = "TextSearchQuery_phrase_".strtolower($tok);
1070 $classname = "TextSearchQuery_node_".strtolower($tok);
1071 return new $classname($word);
1078 class TextSearchQuery_Lexer {
1079 function TextSearchQuery_Lexer ($query_str, $case_exact=false,
1080 $regex=TSQ_REGEX_AUTO)
1082 $this->tokens = $this->tokenize($query_str, $case_exact, $regex);
1083 $this->query_str = $query_str;
1091 function seek($pos) {
1096 return $this->pos == count($this->tokens);
1100 * TODO: support more regex styles, esp. prefer the forced ones over auto
1103 function tokenize($string, $case_exact=false, $regex=TSQ_REGEX_AUTO) {
1105 $buf = $case_exact ? ltrim($string) : strtolower(ltrim($string));
1106 while (!empty($buf)) {
1107 if (preg_match('/^([()])\s*/', $buf, $m)) {
1109 $type = $m[1] == '(' ? TSQ_TOK_LPAREN : TSQ_TOK_RPAREN;
1112 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
1113 and preg_match('/^\*\s*/', $buf, $m)) {
1115 $type = TSQ_TOK_ALL;
1118 elseif ($regex & (TSQ_REGEX_PCRE)
1119 and preg_match('/^\.\*\s*/', $buf, $m)) {
1121 $type = TSQ_TOK_ALL;
1124 elseif ($regex & (TSQ_REGEX_SQL)
1125 and preg_match('/^%\s*/', $buf, $m)) {
1127 $type = TSQ_TOK_ALL;
1131 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
1132 and preg_match('/^\^([^-()][^()\s]*)\s*/', $buf, $m)) {
1134 $type = TSQ_TOK_STARTS_WITH;
1137 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
1138 and preg_match('/^([^-()][^()\s]*)\*\s*/', $buf, $m)) {
1140 $type = TSQ_TOK_STARTS_WITH;
1143 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
1144 and preg_match('/^\*([^-()][^()\s]*)\s*/', $buf, $m)) {
1146 $type = TSQ_TOK_ENDS_WITH;
1149 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
1150 and preg_match('/^([^-()][^()\s]*)\$\s*/', $buf, $m)) {
1152 $type = TSQ_TOK_ENDS_WITH;
1155 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
1156 and preg_match('/^\^([^-()][^()\s]*)\$\s*/', $buf, $m)) {
1158 $type = TSQ_TOK_EXACT;
1160 elseif (preg_match('/^(and|or)\b\s*/i', $buf, $m)) {
1161 $val = strtolower($m[1]);
1162 $type = TSQ_TOK_BINOP;
1164 elseif (preg_match('/^(-|not\b)\s*/i', $buf, $m)) {
1165 $val = strtolower($m[1]);
1166 $type = TSQ_TOK_NOT;
1170 elseif (preg_match('/^ " ( (?: [^"]+ | "" )* ) " \s*/x', $buf, $m)) {
1171 $val = str_replace('""', '"', $m[1]);
1172 $type = TSQ_TOK_WORD;
1175 elseif (preg_match("/^ ' ( (?:[^']+|'')* ) ' \s*/x", $buf, $m)) {
1176 $val = str_replace("''", "'", $m[1]);
1177 $type = TSQ_TOK_WORD;
1180 elseif (preg_match('/^([^-()][^()\s]*)\s*/', $buf, $m)) {
1182 $type = TSQ_TOK_WORD;
1185 assert(empty($buf));
1188 $buf = substr($buf, strlen($m[0]));
1190 /* refine the simple parsing from above: bla*bla, bla?bla, ...
1191 if ($regex and $type == TSQ_TOK_WORD) {
1192 if (substr($val,0,1) == "^")
1193 $type = TSQ_TOK_STARTS_WITH;
1194 elseif (substr($val,0,1) == "*")
1195 $type = TSQ_TOK_ENDS_WITH;
1196 elseif (substr($val,-1,1) == "*")
1197 $type = TSQ_TOK_STARTS_WITH;
1200 $tokens[] = array($type, $val);
1205 function get($accept) {
1206 if ($this->pos >= count($this->tokens))
1209 list ($type, $val) = $this->tokens[$this->pos];
1210 if (($type & $accept) == 0)
1221 // c-basic-offset: 4
1222 // c-hanging-comment-ender-p: nil
1223 // indent-tabs-mode: nil