1PCREPERFORM(3) Library Functions Manual PCREPERFORM(3)
2
6 PCRE - Perl-compatible regular expressions
7
9 Two aspects of performance are discussed below: memory usage and pro‐
11 cessing time. The way you express your pattern as a regular expression
12 can affect both of them.
13
15 Patterns are compiled by PCRE into a reasonably efficient interpretive
17 code, so that most simple patterns do not use much memory. However,
18 there is one case where the memory usage of a compiled pattern can be
19 unexpectedly large. If a parenthesized subpattern has a quantifier with
20 a minimum greater than 1 and/or a limited maximum, the whole subpattern
21 is repeated in the compiled code. For example, the pattern
22 (abc|def){2,4}
24 is compiled as if it were
26 (abc|def)(abc|def)((abc|def)(abc|def)?)?
28 (Technical aside: It is done this way so that backtrack points within
30 each of the repetitions can be independently maintained.)
31 For regular expressions whose quantifiers use only small numbers, this
33 is not usually a problem. However, if the numbers are large, and par‐
34 ticularly if such repetitions are nested, the memory usage can become
35 an embarrassment. For example, the very simple pattern
36 ((ab){1,1000}c){1,3}
38 uses 51K bytes when compiled using the 8-bit library. When PCRE is com‐
40 piled with its default internal pointer size of two bytes, the size
41 limit on a compiled pattern is 64K data units, and this is reached with
42 the above pattern if the outer repetition is increased from 3 to 4.
43 PCRE can be compiled to use larger internal pointers and thus handle
44 larger compiled patterns, but it is better to try to rewrite your pat‐
45 tern to use less memory if you can.
46 One way of reducing the memory usage for such patterns is to make use
48 of PCRE's "subroutine" facility. Re-writing the above pattern as
49 ((ab)(?2){0,999}c)(?1){0,2}
51 reduces the memory requirements to 18K, and indeed it remains under 20K
53 even with the outer repetition increased to 100. However, this pattern
54 is not exactly equivalent, because the "subroutine" calls are treated
55 as atomic groups into which there can be no backtracking if there is a
56 subsequent matching failure. Therefore, PCRE cannot do this kind of
57 rewriting automatically. Furthermore, there is a noticeable loss of
58 speed when executing the modified pattern. Nevertheless, if the atomic
59 grouping is not a problem and the loss of speed is acceptable, this
60 kind of rewriting will allow you to process patterns that PCRE cannot
61 otherwise handle.
62
64 When pcre_exec() or pcre[16|32]_exec() is used for matching, certain
66 kinds of pattern can cause it to use large amounts of the process
67 stack. In some environments the default process stack is quite small,
68 and if it runs out the result is often SIGSEGV. This issue is probably
69 the most frequently raised problem with PCRE. Rewriting your pattern
70 can often help. The pcrestack documentation discusses this issue in
71 detail.
72
74 Certain items in regular expression patterns are processed more effi‐
76 ciently than others. It is more efficient to use a character class like
77 [aeiou] than a set of single-character alternatives such as
78 (a|e|i|o|u). In general, the simplest construction that provides the
79 required behaviour is usually the most efficient. Jeffrey Friedl's book
80 contains a lot of useful general discussion about optimizing regular
81 expressions for efficient performance. This document contains a few
82 observations about PCRE.
83 Using Unicode character properties (the \p, \P, and \X escapes) is
85 slow, because PCRE has to use a multi-stage table lookup whenever it
86 needs a character's property. If you can find an alternative pattern
87 that does not use character properties, it will probably be faster.
88 By default, the escape sequences \b, \d, \s, and \w, and the POSIX
90 character classes such as [:alpha:] do not use Unicode properties,
91 partly for backwards compatibility, and partly for performance reasons.
92 However, you can set PCRE_UCP if you want Unicode character properties
93 to be used. This can double the matching time for items such as \d,
94 when matched with a traditional matching function; the performance loss
95 is less with a DFA matching function, and in both cases there is not
96 much difference for \b.
97 When a pattern begins with .* not in parentheses, or in parentheses
99 that are not the subject of a backreference, and the PCRE_DOTALL option
100 is set, the pattern is implicitly anchored by PCRE, since it can match
101 only at the start of a subject string. However, if PCRE_DOTALL is not
102 set, PCRE cannot make this optimization, because the . metacharacter
103 does not then match a newline, and if the subject string contains new‐
104 lines, the pattern may match from the character immediately following
105 one of them instead of from the very start. For example, the pattern
106 .*second
108 matches the subject "first\nand second" (where \n stands for a newline
110 character), with the match starting at the seventh character. In order
111 to do this, PCRE has to retry the match starting after every newline in
112 the subject.
113 If you are using such a pattern with subject strings that do not con‐
115 tain newlines, the best performance is obtained by setting PCRE_DOTALL,
116 or starting the pattern with ^.* or ^.*? to indicate explicit anchor‐
117 ing. That saves PCRE from having to scan along the subject looking for
118 a newline to restart at.
119 Beware of patterns that contain nested indefinite repeats. These can
121 take a long time to run when applied to a string that does not match.
122 Consider the pattern fragment
123 ^(a+)*
125 This can match "aaaa" in 16 different ways, and this number increases
127 very rapidly as the string gets longer. (The * repeat can match 0, 1,
128 2, 3, or 4 times, and for each of those cases other than 0 or 4, the +
129 repeats can match different numbers of times.) When the remainder of
130 the pattern is such that the entire match is going to fail, PCRE has in
131 principle to try every possible variation, and this can take an
132 extremely long time, even for relatively short strings.
133 An optimization catches some of the more simple cases such as
135 (a+)*b
137 where a literal character follows. Before embarking on the standard
139 matching procedure, PCRE checks that there is a "b" later in the sub‐
140 ject string, and if there is not, it fails the match immediately. How‐
141 ever, when there is no following literal this optimization cannot be
142 used. You can see the difference by comparing the behaviour of
143 (a+)*\d
145 with the pattern above. The former gives a failure almost instantly
147 when applied to a whole line of "a" characters, whereas the latter
148 takes an appreciable time with strings longer than about 20 characters.
149 In many cases, the solution to this kind of performance issue is to use
151 an atomic group or a possessive quantifier.
152
154 Philip Hazel
156 University Computing Service
157 Cambridge CB2 3QH, England.
158
160 Last updated: 25 August 2012
162 Copyright (c) 1997-2012 University of Cambridge.
163PCRE 8.30 09 January 2012 PCREPERFORM(3)