ViewVC logotype

Contents of /code/trunk/doc/pcreperform.3

Parent Directory Parent Directory | Revision Log Revision Log

Revision 99 - (hide annotations) (download)
Tue Mar 6 12:27:42 2007 UTC (8 years, 1 month ago) by ph10
File size: 6187 byte(s)
1. Move the comment about version numbers from pcre.h.in to configure.ac 
because that's where they are now set.
2. Update all the man pages to remove the use of .br and .in because this
causes trouble for some HTML converters. Also standardised the final sections 
giving author information and revision date.
3. Update the maintain/132html man page converter to handle .nf/.fi and to barf 
at .br/.in.

1 nigel 79 .TH PCREPERFORM 3
2 nigel 63 .SH NAME
3     PCRE - Perl-compatible regular expressions
5 nigel 63 .rs
6     .sp
7 nigel 93 Two aspects of performance are discussed below: memory usage and processing
8     time. The way you express your pattern as a regular expression can affect both
9     of them.
10     .
12     .rs
13     .sp
14     Patterns are compiled by PCRE into a reasonably efficient byte code, so that
15     most simple patterns do not use much memory. However, there is one case where
16     memory usage can be unexpectedly large. When a parenthesized subpattern has a
17     quantifier with a minimum greater than 1 and/or a limited maximum, the whole
18     subpattern is repeated in the compiled code. For example, the pattern
19     .sp
20     (abc|def){2,4}
21     .sp
22     is compiled as if it were
23     .sp
24     (abc|def)(abc|def)((abc|def)(abc|def)?)?
25     .sp
26     (Technical aside: It is done this way so that backtrack points within each of
27     the repetitions can be independently maintained.)
28     .P
29     For regular expressions whose quantifiers use only small numbers, this is not
30     usually a problem. However, if the numbers are large, and particularly if such
31     repetitions are nested, the memory usage can become an embarrassment. For
32     example, the very simple pattern
33     .sp
34     ((ab){1,1000}c){1,3}
35     .sp
36     uses 51K bytes when compiled. When PCRE is compiled with its default internal
37     pointer size of two bytes, the size limit on a compiled pattern is 64K, and
38     this is reached with the above pattern if the outer repetition is increased
39     from 3 to 4. PCRE can be compiled to use larger internal pointers and thus
40     handle larger compiled patterns, but it is better to try to rewrite your
41     pattern to use less memory if you can.
42     .P
43     One way of reducing the memory usage for such patterns is to make use of PCRE's
44     .\" HTML <a href="pcrepattern.html#subpatternsassubroutines">
45     .\" </a>
46     "subroutine"
47     .\"
48     facility. Re-writing the above pattern as
49     .sp
50     ((ab)(?2){0,999}c)(?1){0,2}
51     .sp
52     reduces the memory requirements to 18K, and indeed it remains under 20K even
53     with the outer repetition increased to 100. However, this pattern is not
54     exactly equivalent, because the "subroutine" calls are treated as
55     .\" HTML <a href="pcrepattern.html#atomicgroup">
56     .\" </a>
57     atomic groups
58     .\"
59     into which there can be no backtracking if there is a subsequent matching
60     failure. Therefore, PCRE cannot do this kind of rewriting automatically.
61     Furthermore, there is a noticeable loss of speed when executing the modified
62     pattern. Nevertheless, if the atomic grouping is not a problem and the loss of
63     speed is acceptable, this kind of rewriting will allow you to process patterns
64     that PCRE cannot otherwise handle.
65     .
67     .rs
68     .sp
69     Certain items in regular expression patterns are processed more efficiently
70 nigel 63 than others. It is more efficient to use a character class like [aeiou] than a
71 nigel 93 set of single-character alternatives such as (a|e|i|o|u). In general, the
72     simplest construction that provides the required behaviour is usually the most
73     efficient. Jeffrey Friedl's book contains a lot of useful general discussion
74     about optimizing regular expressions for efficient performance. This document
75     contains a few observations about PCRE.
76 nigel 75 .P
77     Using Unicode character properties (the \ep, \eP, and \eX escapes) is slow,
78     because PCRE has to scan a structure that contains data for over fifteen
79     thousand characters whenever it needs a character's property. If you can find
80     an alternative pattern that does not use character properties, it will probably
81     be faster.
82     .P
83 nigel 63 When a pattern begins with .* not in parentheses, or in parentheses that are
84     not the subject of a backreference, and the PCRE_DOTALL option is set, the
85     pattern is implicitly anchored by PCRE, since it can match only at the start of
86     a subject string. However, if PCRE_DOTALL is not set, PCRE cannot make this
87     optimization, because the . metacharacter does not then match a newline, and if
88     the subject string contains newlines, the pattern may match from the character
89     immediately following one of them instead of from the very start. For example,
90     the pattern
91 nigel 75 .sp
92 nigel 63 .*second
93 nigel 75 .sp
94     matches the subject "first\enand second" (where \en stands for a newline
95 nigel 63 character), with the match starting at the seventh character. In order to do
96     this, PCRE has to retry the match starting after every newline in the subject.
97 nigel 75 .P
98 nigel 63 If you are using such a pattern with subject strings that do not contain
99     newlines, the best performance is obtained by setting PCRE_DOTALL, or starting
100 nigel 77 the pattern with ^.* or ^.*? to indicate explicit anchoring. That saves PCRE
101     from having to scan along the subject looking for a newline to restart at.
102 nigel 75 .P
103 nigel 63 Beware of patterns that contain nested indefinite repeats. These can take a
104     long time to run when applied to a string that does not match. Consider the
105     pattern fragment
106 nigel 75 .sp
107 nigel 93 ^(a+)*
108 nigel 75 .sp
109 nigel 93 This can match "aaaa" in 16 different ways, and this number increases very
110 nigel 63 rapidly as the string gets longer. (The * repeat can match 0, 1, 2, 3, or 4
111 nigel 93 times, and for each of those cases other than 0 or 4, the + repeats can match
112 nigel 63 different numbers of times.) When the remainder of the pattern is such that the
113     entire match is going to fail, PCRE has in principle to try every possible
114 nigel 93 variation, and this can take an extremely long time, even for relatively short
115     strings.
116 nigel 75 .P
117 nigel 63 An optimization catches some of the more simple cases such as
118 nigel 75 .sp
119 nigel 63 (a+)*b
120 nigel 75 .sp
121 nigel 63 where a literal character follows. Before embarking on the standard matching
122     procedure, PCRE checks that there is a "b" later in the subject string, and if
123     there is not, it fails the match immediately. However, when there is no
124     following literal this optimization cannot be used. You can see the difference
125     by comparing the behaviour of
126 nigel 75 .sp
127     (a+)*\ed
128     .sp
129 nigel 63 with the pattern above. The former gives a failure almost instantly when
130     applied to a whole line of "a" characters, whereas the latter takes an
131     appreciable time with strings longer than about 20 characters.
132 nigel 75 .P
133     In many cases, the solution to this kind of performance issue is to use an
134     atomic group or a possessive quantifier.
135 ph10 99 .
136     .
137     .SH AUTHOR
138     .rs
139     .sp
140     .nf
141     Philip Hazel
142     University Computing Service
143     Cambridge CB2 3QH, England.
144     .fi
145     .
146     .
147     .SH REVISION
148     .rs
149     .sp
150     .nf
151     Last updated: 06 March 2007
152     Copyright (c) 1997-2007 University of Cambridge.
153     .fi


Name Value
svn:eol-style native
svn:keywords "Author Date Id Revision Url"

ViewVC Help
Powered by ViewVC 1.1.12