1 <?php rcs_id('$Id: TextSearchQuery.php,v 1.22 2005-11-14 22:30:17 rurban Exp $');
3 * A text search query, converting queries to PCRE and SQL matchers.
5 * This represents an enhanced "Google-like" text search query:
7 * <dt> default: case-insensitive glob-style search with special operators OR AND NOT -
9 * <dd> Match strings containing the substring 'wiki', and NOT containing the
11 * <dt> wiki word or page
12 * <dd> Match strings containing the substring 'wiki' AND either the substring
13 * 'word' OR the substring 'page'.
14 * <dt> auto-detect regex hints, glob-style or regex-style, and converts them
15 * to PCRE and SQL matchers:
16 * <dd> "^word$" => EXACT(word)
17 * <dd> "^word" => STARTS_WITH(word)
18 * <dd> "word*" => STARTS_WITH(word)
19 * <dd> "*word" => ENDS_WITH(word)
20 * <dd> "/^word.* /" => REGEX(^word.*)
21 * <dd> "word*word" => REGEX(word.*word)
24 * The full query syntax, in order of precedence, is roughly:
26 * The unary 'NOT' or '-' operator (they are equivalent) negates the
27 * following search clause.
29 * Search clauses may be joined with the (left-associative) binary operators
30 * 'AND' and 'OR'. (case-insensitive)
32 * Two adjoining search clauses are joined with an implicit 'AND'. This has
33 * lower precedence than either an explicit 'AND' or 'OR', so "a b OR c"
34 * parses as "a AND ( b OR c )", while "a AND b OR c" parses as
35 * "( a AND b ) OR c" (due to the left-associativity of 'AND' and 'OR'.)
37 * Search clauses can be grouped with parentheses.
39 * Phrases (or other things which don't look like words) can be forced to
40 * be interpreted as words by quoting them, either with single (') or double (")
41 * quotes. If you wan't to include the quote character within a quoted string,
42 * double-up on the quote character: 'I''m hungry' is equivalent to
45 * Force regex on "re:word" => posix-style, "/word/" => pcre-style
46 * or use regex='glob' to use file wildcard-like matching. (not yet)
48 * The parsed tree is then converted to the needed PCRE (highlight,
49 * simple backends) or SQL functions.
51 * @author: Jeff Dairiki
52 * @author: Reini Urban (case and regex detection, enhanced sql callbacks)
55 // regex-style: 'auto', 'none', 'glob', 'posix', 'pcre', 'sql'
56 define ('TSQ_REGEX_NONE', 0);
57 define ('TSQ_REGEX_AUTO', 1);
58 define ('TSQ_REGEX_POSIX', 2);
59 define ('TSQ_REGEX_GLOB', 4);
60 define ('TSQ_REGEX_PCRE', 8);
61 define ('TSQ_REGEX_SQL', 16);
63 class TextSearchQuery {
67 * @param $search_query string The query. Syntax is as described above.
68 * Note that an empty $search_query will match anything.
69 * @param $case_exact boolean
70 * @param $regex string one of 'auto', 'none', 'glob', 'posix', 'pcre', 'sql'
71 * @see TextSearchQuery
73 function TextSearchQuery($search_query, $case_exact=false, $regex='auto') {
74 if ($regex == 'none' or !$regex)
76 elseif (defined("TSQ_REGEX_".strtoupper($regex)))
77 $this->_regex = constant("TSQ_REGEX_".strtoupper($regex));
79 trigger_error(fmt("Unsupported argument: %s=%s", 'regex', $regex));
82 $this->_case_exact = $case_exact;
83 $parser = new TextSearchQuery_Parser;
84 $this->_tree = $parser->parse($search_query, $case_exact, $this->_regex);
85 $this->_optimize(); // broken under certain circumstances: "word -word -word"
86 $this->_stoplist = '(A|An|And|But|By|For|From|In|Is|It|Of|On|Or|The|To|With)';
89 function _optimize() {
90 $this->_tree = $this->_tree->optimize();
94 * Get a PCRE regexp which matches the query.
97 if (!isset($this->_regexp)) {
99 $this->_regexp = '/' . $this->_tree->regexp() . '/'.($this->_case_exact?'':'i').'sS';
101 $this->_regexp = '/^' . $this->_tree->regexp() . '/'.($this->_case_exact?'':'i').'sS';
103 return $this->_regexp;
107 * Match query against string.
109 * @param $string string The string to match.
110 * @return boolean True if the string matches the query.
112 function match($string) {
113 return preg_match($this->asRegexp(), $string);
118 * Get a regular expression suitable for highlighting matched words.
120 * This returns a PCRE regular expression which matches any non-negated
123 * @return string The PCRE regexp.
125 function getHighlightRegexp() {
126 if (!isset($this->_hilight_regexp)) {
127 $words = array_unique($this->_tree->highlight_words());
129 $this->_hilight_regexp = false;
131 foreach ($words as $key => $word)
132 $words[$key] = preg_quote($word, '/');
133 $this->_hilight_regexp = '(?:' . join('|', $words) . ')';
136 return $this->_hilight_regexp;
140 * Make an SQL clause which matches the query. (deprecated, use makeSqlClause instead)
142 * @param $make_sql_clause_cb WikiCallback
143 * A callback which takes a single word as an argument and
144 * returns an SQL clause which will match exactly those records
145 * containing the word. The word passed to the callback will always
146 * be in all lower case.
148 * TODO: support db-specific extensions, like MATCH AGAINST or REGEX
149 * mysql => 4.0.1 can also do Google: MATCH AGAINST IN BOOLEAN MODE
150 * How? WikiDB backend method?
154 * function sql_title_match($word) {
155 * return sprintf("LOWER(title) like '%s'",
156 * addslashes($word));
161 * $query = new TextSearchQuery("wiki -page");
162 * $cb = new WikiFunctionCb('sql_title_match');
163 * $sql_clause = $query->makeSqlClause($cb);
165 * This will result in $sql_clause containing something like
166 * "(LOWER(title) like 'wiki') AND NOT (LOWER(title) like 'page')".
168 * @return string The SQL clause.
170 function makeSqlClause($sql_clause_cb) {
171 $this->_sql_clause_cb = $sql_clause_cb;
172 return $this->_sql_clause($this->_tree);
174 // deprecated: use _sql_clause_obj now.
175 function _sql_clause($node) {
177 case 'WORD': // word => %word%
178 return $this->_sql_clause_cb->call($node->word);
180 return "NOT (" . $this->_sql_clause($node->leaves[0]) . ")";
183 $subclauses = array();
184 foreach ($node->leaves as $leaf)
185 $subclauses[] = "(" . $this->_sql_clause($leaf) . ")";
186 return join(" $node->op ", $subclauses);
188 assert($node->op == 'VOID');
193 /** Get away with the callback and use a db-specific search class instead.
194 * @see WikiDB_backend_PearDB_search
196 function makeSqlClauseObj(&$sql_search_cb) {
197 $this->_sql_clause_cb = $sql_search_cb;
198 return $this->_sql_clause_obj($this->_tree);
201 function _sql_clause_obj($node) {
204 return "NOT (" . $this->_sql_clause_cb->call($node->leaves[0]) . ")";
207 $subclauses = array();
208 foreach ($node->leaves as $leaf)
209 $subclauses[] = "(" . $this->_sql_clause_obj($leaf) . ")";
210 return join(" $node->op ", $subclauses);
216 return $this->_sql_clause_cb->call($node);
221 postgresql tsearch2 uses no WHERE operators, just & | and ! in the searchstring
223 function makeTsearch2SqlClauseObj(&$sql_search_cb) {
224 $this->_sql_clause_cb = $sql_search_cb;
225 return $this->_Tsearch2Sql_clause_obj($this->_tree);
228 function _Tsearch2Sql_clause_obj($node) {
229 // TODO: "such a phrase"
232 return "!" . $node->leaves[0];
234 $subclauses = array();
235 foreach ($node->leaves as $leaf)
236 $subclauses[] = $this->_Tsearch2Sql_clause_obj($leaf);
237 return join("&", $subclauses);
239 $subclauses = array();
240 foreach ($node->leaves as $leaf)
241 $subclauses[] = $this->_Tsearch2Sql_clause_obj($leaf);
242 return join("|", $subclauses);
248 return $this->_sql_clause_cb->call($node);
252 function sql() { return '%'.$this->_sql_quote($this->word).'%'; }
255 * Get printable representation of the parse tree.
257 * This is for debugging only.
258 * @return string Printable parse tree.
260 function asString() {
261 return $this->_as_string($this->_tree);
264 function _as_string($node, $indent = '') {
267 return $indent . "WORD: $node->word";
269 return $indent . "VOID";
271 return $indent . "ALL";
273 $lines = array($indent . $node->op . ":");
275 foreach ($node->leaves as $leaf)
276 $lines[] = $this->_as_string($leaf, $indent);
277 return join("\n", $lines);
283 * This is a TextSearchQuery which matches nothing.
285 class NullTextSearchQuery extends TextSearchQuery {
287 * Create a new query.
289 * @see TextSearchQuery
291 function NullTextSearchQuery() {}
292 function asRegexp() { return '/^(?!a)a/x'; }
293 function match($string) { return false; }
294 function getHighlightRegexp() { return ""; }
295 function makeSqlClause($make_sql_clause_cb) { return "(1 = 0)"; }
296 function asString() { return "NullTextSearchQuery"; }
300 ////////////////////////////////////////////////////////////////
302 // Remaining classes are private.
304 ////////////////////////////////////////////////////////////////
306 * Virtual base class for nodes in a TextSearchQuery parse tree.
308 * Also serves as a 'VOID' (contentless) node.
310 class TextSearchQuery_node
315 * Optimize this node.
316 * @return object Optimized node.
318 function optimize() {
323 * @return regexp matching this node.
330 * @param bool True if this node has been negated (higher in the parse tree.)
331 * @return array A list of all non-negated words contained by this node.
333 function highlight_words($negated = false) {
337 function sql() { return $this->word; }
343 class TextSearchQuery_node_word
344 extends TextSearchQuery_node
348 function TextSearchQuery_node_word($word) {
352 return '(?=.*' . preg_quote($this->word, '/') . ')';
354 function highlight_words ($negated = false) {
355 return $negated ? array() : array($this->word);
357 function _sql_quote() {
358 $word = preg_replace('/(?=[%_\\\\])/', "\\", $this->word);
359 return $GLOBALS['request']->_dbi->qstr($word);
361 function sql() { return '%'.$this->_sql_quote($this->word).'%'; }
364 class TextSearchQuery_node_all
365 extends TextSearchQuery_node {
367 function regexp() { return '(?=.*)'; }
368 function sql() { return '%'; }
370 class TextSearchQuery_node_starts_with
371 extends TextSearchQuery_node_word {
372 var $op = "STARTS_WITH";
373 function regexp() { return '(?=.*\b' . preg_quote($this->word, '/') . ')'; }
374 function sql () { return $this->_sql_quote($this->word).'%'; }
377 class TextSearchQuery_node_ends_with
378 extends TextSearchQuery_node_word {
379 var $op = "ENDS_WITH";
380 function regexp() { return '(?=.*' . preg_quote($this->word, '/') . '\b)'; }
381 function sql () { return '%'.$this->_sql_quote($this->word); }
384 class TextSearchQuery_node_exact
385 extends TextSearchQuery_node_word {
387 function regexp() { return '(?=\b' . preg_quote($this->word, '/') . '\b)'; }
388 function sql () { return $this->_sql_squote($this->word); }
391 class TextSearchQuery_node_regex // posix regex. FIXME!
392 extends TextSearchQuery_node_word {
393 var $op = "REGEX"; // using REGEXP or ~ extension
394 function regexp() { return '(?=.*\b' . $this->word . '\b)'; }
395 function sql () { return $this->_sql_quote($this->word); }
398 class TextSearchQuery_node_regex_glob
399 extends TextSearchQuery_node_regex {
400 var $op = "REGEX_GLOB";
401 function regexp() { return '(?=.*\b' . glob_to_pcre($this->word) . '\b)'; }
404 class TextSearchQuery_node_regex_pcre // how to handle pcre modifiers? /i
405 extends TextSearchQuery_node_regex {
406 var $op = "REGEX_PCRE";
407 function regexp() { return $this->word; }
410 class TextSearchQuery_node_regex_sql
411 extends TextSearchQuery_node_regex {
412 var $op = "REGEX_SQL"; // using LIKE
413 function regexp() { return str_replace(array("/%/","/_/"), array(".*","."), $this->word); }
414 function sql() { return $this->word; }
420 class TextSearchQuery_node_not
421 extends TextSearchQuery_node
425 function TextSearchQuery_node_not($leaf) {
426 $this->leaves = array($leaf);
429 function optimize() {
430 $leaf = &$this->leaves[0];
431 $leaf = $leaf->optimize();
432 if ($leaf->op == 'NOT')
433 return $leaf->leaves[0]; // ( NOT ( NOT x ) ) -> x
438 $leaf = &$this->leaves[0];
439 return '(?!' . $leaf->regexp() . ')';
442 function highlight_words ($negated = false) {
443 return $this->leaves[0]->highlight_words(!$negated);
448 * Virtual base class for 'AND' and 'OR conjoins.
450 class TextSearchQuery_node_binop
451 extends TextSearchQuery_node
453 function TextSearchQuery_node_binop($leaves) {
454 $this->leaves = $leaves;
457 function _flatten() {
458 // This flattens e.g. (AND (AND a b) (OR c d) e)
459 // to (AND a b e (OR c d))
461 foreach ($this->leaves as $leaf) {
462 $leaf = $leaf->optimize();
463 if ($this->op == $leaf->op)
464 $flat = array_merge($flat, $leaf->leaves);
468 $this->leaves = $flat;
471 function optimize() {
473 assert(!empty($this->leaves));
474 if (count($this->leaves) == 1)
475 return $this->leaves[0]; // (AND x) -> x
479 function highlight_words($negated = false) {
481 foreach ($this->leaves as $leaf)
482 array_splice($words,0,0,
483 $leaf->highlight_words($negated));
489 * A (possibly multi-argument) 'AND' conjoin.
491 class TextSearchQuery_node_and
492 extends TextSearchQuery_node_binop
496 function optimize() {
499 // Convert (AND (NOT a) (NOT b) c d) into (AND (NOT (OR a b)) c d).
500 // Since OR's are more efficient for regexp matching:
501 // (?!.*a)(?!.*b) vs (?!.*(?:a|b))
503 // Suck out the negated leaves.
505 foreach ($this->leaves as $key => $leaf) {
506 if ($leaf->op == 'NOT') {
507 $nots[] = $leaf->leaves[0];
508 unset($this->leaves[$key]);
512 // Combine the negated leaves into a single negated or.
514 $node = ( new TextSearchQuery_node_not
515 (new TextSearchQuery_node_or($nots)) );
516 array_unshift($this->leaves, $node->optimize());
519 assert(!empty($this->leaves));
520 if (count($this->leaves) == 1)
521 return $this->leaves[0]; // (AND x) -> x
526 * Either we need all combinations of all words to be position independent,
527 * or we have to use multiple match calls for each AND
528 * (AND x y) => /(?(:x)(:y))|(?(:y)(:x))/
532 foreach ($this->leaves as $leaf)
533 $regexp .= $leaf->regexp();
539 * A (possibly multi-argument) 'OR' conjoin.
541 class TextSearchQuery_node_or
542 extends TextSearchQuery_node_binop
547 // We will combine any of our direct descendents which are WORDs
548 // into a single (?=.*(?:word1|word2|...)) regexp.
553 foreach ($this->leaves as $leaf) {
554 if ($leaf->op == 'WORD')
555 $words[] = preg_quote($leaf->word, '/');
557 $regexps[] = $leaf->regexp();
561 array_unshift($regexps,
562 '(?=.*' . $this->_join($words) . ')');
564 return $this->_join($regexps);
567 function _join($regexps) {
568 assert(count($regexps) > 0);
570 if (count($regexps) > 1)
571 return '(?:' . join('|', $regexps) . ')';
578 ////////////////////////////////////////////////////////////////
581 // op's (and, or, not) are forced to lowercase in the tokenizer.
583 ////////////////////////////////////////////////////////////////
584 define ('TSQ_TOK_BINOP', 1);
585 define ('TSQ_TOK_NOT', 2);
586 define ('TSQ_TOK_LPAREN', 4);
587 define ('TSQ_TOK_RPAREN', 8);
588 define ('TSQ_TOK_WORD', 16);
589 define ('TSQ_TOK_STARTS_WITH', 32);
590 define ('TSQ_TOK_ENDS_WITH', 64);
591 define ('TSQ_TOK_EXACT', 128);
592 define ('TSQ_TOK_REGEX', 256);
593 define ('TSQ_TOK_REGEX_GLOB', 512);
594 define ('TSQ_TOK_REGEX_PCRE', 1024);
595 define ('TSQ_TOK_REGEX_SQL', 2048);
596 define ('TSQ_TOK_ALL', 4096);
597 // all bits from word to the last.
598 define ('TSQ_ALLWORDS', (4096*2)-1 - (16-1));
600 class TextSearchQuery_Parser
603 * This is a simple recursive descent parser, based on the following grammar:
618 * atom : '(' list ')'
623 * The terminal tokens are:
630 * /[^-()\s][^()\s]* WORD
641 function parse ($search_expr, $case_exact=false, $regex=TSQ_REGEX_AUTO) {
642 $this->lexer = new TextSearchQuery_Lexer($search_expr, $case_exact, $regex);
643 $this->_regex = $regex;
644 $tree = $this->get_list('toplevel');
645 assert($this->lexer->eof());
650 function get_list ($is_toplevel = false) {
653 // token types we'll accept as words (and thus expr's) for the
654 // purpose of error recovery:
655 $accept_as_words = TSQ_TOK_NOT | TSQ_TOK_BINOP;
657 $accept_as_words |= TSQ_TOK_LPAREN | TSQ_TOK_RPAREN;
659 while ( ($expr = $this->get_expr())
660 || ($expr = $this->get_word($accept_as_words)) ) {
666 return new TextSearchQuery_node;
670 return new TextSearchQuery_node_and($list);
673 function get_expr () {
674 if ( !($expr = $this->get_atom()) )
677 $savedpos = $this->lexer->tell();
678 while ( ($op = $this->lexer->get(TSQ_TOK_BINOP)) ) {
679 if ( ! ($right = $this->get_atom()) ) {
684 $expr = new TextSearchQuery_node_and(array($expr, $right));
687 $expr = new TextSearchQuery_node_or(array($expr, $right));
690 $savedpos = $this->lexer->tell();
692 $this->lexer->seek($savedpos);
698 function get_atom() {
699 if ($word = $this->get_word(TSQ_ALLWORDS))
702 $savedpos = $this->lexer->tell();
703 if ( $this->lexer->get(TSQ_TOK_LPAREN) ) {
704 if ( ($list = $this->get_list()) && $this->lexer->get(TSQ_TOK_RPAREN) )
707 elseif ( $this->lexer->get(TSQ_TOK_NOT) ) {
708 if ( ($atom = $this->get_atom()) )
709 return new TextSearchQuery_node_not($atom);
711 $this->lexer->seek($savedpos);
715 function get_word($accept = TSQ_ALLWORDS) {
716 foreach (array("WORD","STARTS_WITH","ENDS_WITH","EXACT",
717 "REGEX","REGEX_GLOB","REGEX_PCRE","ALL") as $tok) {
718 $const = constant("TSQ_TOK_".$tok);
719 if ( $accept & $const and ($word = $this->lexer->get($const)) ) {
720 $classname = "TextSearchQuery_node_".strtolower($tok);
721 return new $classname($word);
728 class TextSearchQuery_Lexer {
729 function TextSearchQuery_Lexer ($query_str, $case_exact=false, $regex=TSQ_REGEX_AUTO) {
730 $this->tokens = $this->tokenize($query_str, $case_exact, $regex);
738 function seek($pos) {
743 return $this->pos == count($this->tokens);
747 * TODO: support more regex styles, esp. prefer the forced ones over auto
750 function tokenize($string, $case_exact=false, $regex=TSQ_REGEX_AUTO) {
752 $buf = $case_exact ? ltrim($string) : strtolower(ltrim($string));
753 while (!empty($buf)) {
754 if (preg_match('/^(and|or)\b\s*/i', $buf, $m)) {
755 $val = strtolower($m[1]);
756 $type = TSQ_TOK_BINOP;
758 elseif (preg_match('/^(-|not\b)\s*/i', $buf, $m)) {
759 $val = strtolower($m[1]);
762 elseif (preg_match('/^([()])\s*/', $buf, $m)) {
764 $type = $m[1] == '(' ? TSQ_TOK_LPAREN : TSQ_TOK_RPAREN;
768 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
769 and preg_match('/^\*\s*/', $buf, $m)) {
774 elseif ($regex & (TSQ_REGEX_PCRE)
775 and preg_match('/^\.\*\s*/', $buf, $m)) {
780 elseif ($regex & (TSQ_REGEX_SQL)
781 and preg_match('/^%\s*/', $buf, $m)) {
787 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
788 and preg_match('/^\^([^-()][^()\s]*)\s*/', $buf, $m)) {
790 $type = TSQ_TOK_STARTS_WITH;
793 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
794 and preg_match('/^([^-()][^()\s]*)\*\s*/', $buf, $m)) {
796 $type = TSQ_TOK_STARTS_WITH;
799 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_GLOB)
800 and preg_match('/^\*([^-()][^()\s]*)\s*/', $buf, $m)) {
802 $type = TSQ_TOK_ENDS_WITH;
805 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
806 and preg_match('/^([^-()][^()\s]*)\$\s*/', $buf, $m)) {
808 $type = TSQ_TOK_ENDS_WITH;
811 elseif ($regex & (TSQ_REGEX_AUTO|TSQ_REGEX_POSIX|TSQ_REGEX_PCRE)
812 and preg_match('/^\^([^-()][^()\s]*)\$\s*/', $buf, $m)) {
814 $type = TSQ_TOK_EXACT;
818 elseif (preg_match('/^ " ( (?: [^"]+ | "" )* ) " \s*/x', $buf, $m)) {
819 $val = str_replace('""', '"', $m[1]);
820 $type = TSQ_TOK_WORD;
823 elseif (preg_match("/^ ' ( (?:[^']+|'')* ) ' \s*/x", $buf, $m)) {
824 $val = str_replace("''", "'", $m[1]);
825 $type = TSQ_TOK_WORD;
828 elseif (preg_match('/^([^-()][^()\s]*)\s*/', $buf, $m)) {
830 $type = TSQ_TOK_WORD;
836 $buf = substr($buf, strlen($m[0]));
838 /* refine the simple parsing from above: bla*bla, bla?bla, ...
839 if ($regex and $type == TSQ_TOK_WORD) {
840 if (substr($val,0,1) == "^")
841 $type = TSQ_TOK_STARTS_WITH;
842 elseif (substr($val,0,1) == "*")
843 $type = TSQ_TOK_ENDS_WITH;
844 elseif (substr($val,-1,1) == "*")
845 $type = TSQ_TOK_STARTS_WITH;
848 $tokens[] = array($type, $val);
853 function get($accept) {
854 if ($this->pos >= count($this->tokens))
857 list ($type, $val) = $this->tokens[$this->pos];
858 if (($type & $accept) == 0)
870 // c-hanging-comment-ender-p: nil
871 // indent-tabs-mode: nil