1PCREMATCHING(3) Library Functions Manual PCREMATCHING(3)
2
6 PCRE - Perl-compatible regular expressions
7
9 This document describes the two different algorithms that are available
11 in PCRE for matching a compiled regular expression against a given sub‐
12 ject string. The "standard" algorithm is the one provided by the
13 pcre_exec() function. This works in the same was as Perl's matching
14 function, and provides a Perl-compatible matching operation.
15 An alternative algorithm is provided by the pcre_dfa_exec() function;
17 this operates in a different way, and is not Perl-compatible. It has
18 advantages and disadvantages compared with the standard algorithm, and
19 these are described below.
20 When there is only one possible way in which a given subject string can
22 match a pattern, the two algorithms give the same answer. A difference
23 arises, however, when there are multiple possibilities. For example, if
24 the pattern
25 ^<.*>
27 is matched against the string
29 <something> <something else> <something further>
31 there are three possible answers. The standard algorithm finds only one
33 of them, whereas the alternative algorithm finds all three.
34
36 The set of strings that are matched by a regular expression can be rep‐
38 resented as a tree structure. An unlimited repetition in the pattern
39 makes the tree of infinite size, but it is still a tree. Matching the
40 pattern to a given subject string (from a given starting point) can be
41 thought of as a search of the tree. There are two ways to search a
42 tree: depth-first and breadth-first, and these correspond to the two
43 matching algorithms provided by PCRE.
44
46 In the terminology of Jeffrey Friedl's book "Mastering Regular Expres‐
48 sions", the standard algorithm is an "NFA algorithm". It conducts a
49 depth-first search of the pattern tree. That is, it proceeds along a
50 single path through the tree, checking that the subject matches what is
51 required. When there is a mismatch, the algorithm tries any alterna‐
52 tives at the current point, and if they all fail, it backs up to the
53 previous branch point in the tree, and tries the next alternative
54 branch at that level. This often involves backing up (moving to the
55 left) in the subject string as well. The order in which repetition
56 branches are tried is controlled by the greedy or ungreedy nature of
57 the quantifier.
58 If a leaf node is reached, a matching string has been found, and at
60 that point the algorithm stops. Thus, if there is more than one possi‐
61 ble match, this algorithm returns the first one that it finds. Whether
62 this is the shortest, the longest, or some intermediate length depends
63 on the way the greedy and ungreedy repetition quantifiers are specified
64 in the pattern.
65 Because it ends up with a single path through the tree, it is rela‐
67 tively straightforward for this algorithm to keep track of the sub‐
68 strings that are matched by portions of the pattern in parentheses.
69 This provides support for capturing parentheses and back references.
70
72 This algorithm conducts a breadth-first search of the tree. Starting
74 from the first matching point in the subject, it scans the subject
75 string from left to right, once, character by character, and as it does
76 this, it remembers all the paths through the tree that represent valid
77 matches. In Friedl's terminology, this is a kind of "DFA algorithm",
78 though it is not implemented as a traditional finite state machine (it
79 keeps multiple states active simultaneously).
80 Although the general principle of this matching algorithm is that it
82 scans the subject string only once, without backtracking, there is one
83 exception: when a lookaround assertion is encountered, the characters
84 following or preceding the current point have to be independently
85 inspected.
86 The scan continues until either the end of the subject is reached, or
88 there are no more unterminated paths. At this point, terminated paths
89 represent the different matching possibilities (if there are none, the
90 match has failed). Thus, if there is more than one possible match,
91 this algorithm finds all of them, and in particular, it finds the long‐
92 est. There is an option to stop the algorithm after the first match
93 (which is necessarily the shortest) is found.
94 Note that all the matches that are found start at the same point in the
96 subject. If the pattern
97 cat(er(pillar)?)
99 is matched against the string "the caterpillar catchment", the result
101 will be the three strings "cat", "cater", and "caterpillar" that start
102 at the fourth character of the subject. The algorithm does not automat‐
103 ically move on to find matches that start at later positions.
104 There are a number of features of PCRE regular expressions that are not
106 supported by the alternative matching algorithm. They are as follows:
107 1. Because the algorithm finds all possible matches, the greedy or
109 ungreedy nature of repetition quantifiers is not relevant. Greedy and
110 ungreedy quantifiers are treated in exactly the same way. However, pos‐
111 sessive quantifiers can make a difference when what follows could also
112 match what is quantified, for example in a pattern like this:
113 ^a++\w!
115 This pattern matches "aaab!" but not "aaa!", which would be matched by
117 a non-possessive quantifier. Similarly, if an atomic group is present,
118 it is matched as if it were a standalone pattern at the current point,
119 and the longest match is then "locked in" for the rest of the overall
120 pattern.
121 2. When dealing with multiple paths through the tree simultaneously, it
123 is not straightforward to keep track of captured substrings for the
124 different matching possibilities, and PCRE's implementation of this
125 algorithm does not attempt to do this. This means that no captured sub‐
126 strings are available.
127 3. Because no substrings are captured, back references within the pat‐
129 tern are not supported, and cause errors if encountered.
130 4. For the same reason, conditional expressions that use a backrefer‐
132 ence as the condition or test for a specific group recursion are not
133 supported.
134 5. Because many paths through the tree may be active, the \K escape
136 sequence, which resets the start of the match when encountered (but may
137 be on some paths and not on others), is not supported. It causes an
138 error if encountered.
139 6. Callouts are supported, but the value of the capture_top field is
141 always 1, and the value of the capture_last field is always -1.
142 7. The \C escape sequence, which (in the standard algorithm) matches a
144 single byte, even in UTF-8 mode, is not supported because the alterna‐
145 tive algorithm moves through the subject string one character at a
146 time, for all active paths through the tree.
147 8. Except for (*FAIL), the backtracking control verbs such as (*PRUNE)
149 are not supported. (*FAIL) is supported, and behaves like a failing
150 negative assertion.
151
153 Using the alternative matching algorithm provides the following advan‐
155 tages:
156 1. All possible matches (at a single point in the subject) are automat‐
158 ically found, and in particular, the longest match is found. To find
159 more than one match using the standard algorithm, you have to do kludgy
160 things with callouts.
161 2. Because the alternative algorithm scans the subject string just
163 once, and never needs to backtrack, it is possible to pass very long
164 subject strings to the matching function in several pieces, checking
165 for partial matching each time. The pcrepartial documentation gives
166 details of partial matching.
167
169 The alternative algorithm suffers from a number of disadvantages:
171 1. It is substantially slower than the standard algorithm. This is
173 partly because it has to search for all possible matches, but is also
174 because it is less susceptible to optimization.
175 2. Capturing parentheses and back references are not supported.
177 3. Although atomic groups are supported, their use does not provide the
179 performance advantage that it does for the standard algorithm.
180
182 Philip Hazel
184 University Computing Service
185 Cambridge CB2 3QH, England.
186
188 Last updated: 29 September 2009
190 Copyright (c) 1997-2009 University of Cambridge.
191 PCREMATCHING(3)