1<html> 2<head> 3<title>pcreperform specification</title> 4</head> 5<body bgcolor="#FFFFFF" text="#00005A" link="#0066FF" alink="#3399FF" vlink="#2222BB"> 6<h1>pcreperform man page</h1> 7<p> 8Return to the <a href="index.html">PCRE index page</a>. 9</p> 10<p> 11This page is part of the PCRE HTML documentation. It was generated automatically 12from the original man page. If there is any nonsense in it, please consult the 13man page, in case the conversion went wrong. 14<br> 15<br><b> 16PCRE PERFORMANCE 17</b><br> 18<P> 19Two aspects of performance are discussed below: memory usage and processing 20time. The way you express your pattern as a regular expression can affect both 21of them. 22</P> 23<br><b> 24COMPILED PATTERN MEMORY USAGE 25</b><br> 26<P> 27Patterns are compiled by PCRE into a reasonably efficient byte code, so that 28most simple patterns do not use much memory. However, there is one case where 29the memory usage of a compiled pattern can be unexpectedly large. If a 30parenthesized subpattern has a quantifier with a minimum greater than 1 and/or 31a limited maximum, the whole subpattern is repeated in the compiled code. For 32example, the pattern 33<pre> 34 (abc|def){2,4} 35</pre> 36is compiled as if it were 37<pre> 38 (abc|def)(abc|def)((abc|def)(abc|def)?)? 39</pre> 40(Technical aside: It is done this way so that backtrack points within each of 41the repetitions can be independently maintained.) 42</P> 43<P> 44For regular expressions whose quantifiers use only small numbers, this is not 45usually a problem. However, if the numbers are large, and particularly if such 46repetitions are nested, the memory usage can become an embarrassment. For 47example, the very simple pattern 48<pre> 49 ((ab){1,1000}c){1,3} 50</pre> 51uses 51K bytes when compiled. When PCRE is compiled with its default internal 52pointer size of two bytes, the size limit on a compiled pattern is 64K, and 53this is reached with the above pattern if the outer repetition is increased 54from 3 to 4. PCRE can be compiled to use larger internal pointers and thus 55handle larger compiled patterns, but it is better to try to rewrite your 56pattern to use less memory if you can. 57</P> 58<P> 59One way of reducing the memory usage for such patterns is to make use of PCRE's 60<a href="pcrepattern.html#subpatternsassubroutines">"subroutine"</a> 61facility. Re-writing the above pattern as 62<pre> 63 ((ab)(?2){0,999}c)(?1){0,2} 64</pre> 65reduces the memory requirements to 18K, and indeed it remains under 20K even 66with the outer repetition increased to 100. However, this pattern is not 67exactly equivalent, because the "subroutine" calls are treated as 68<a href="pcrepattern.html#atomicgroup">atomic groups</a> 69into which there can be no backtracking if there is a subsequent matching 70failure. Therefore, PCRE cannot do this kind of rewriting automatically. 71Furthermore, there is a noticeable loss of speed when executing the modified 72pattern. Nevertheless, if the atomic grouping is not a problem and the loss of 73speed is acceptable, this kind of rewriting will allow you to process patterns 74that PCRE cannot otherwise handle. 75</P> 76<br><b> 77STACK USAGE AT RUN TIME 78</b><br> 79<P> 80When <b>pcre_exec()</b> is used for matching, certain kinds of pattern can cause 81it to use large amounts of the process stack. In some environments the default 82process stack is quite small, and if it runs out the result is often SIGSEGV. 83This issue is probably the most frequently raised problem with PCRE. Rewriting 84your pattern can often help. The 85<a href="pcrestack.html"><b>pcrestack</b></a> 86documentation discusses this issue in detail. 87</P> 88<br><b> 89PROCESSING TIME 90</b><br> 91<P> 92Certain items in regular expression patterns are processed more efficiently 93than others. It is more efficient to use a character class like [aeiou] than a 94set of single-character alternatives such as (a|e|i|o|u). In general, the 95simplest construction that provides the required behaviour is usually the most 96efficient. Jeffrey Friedl's book contains a lot of useful general discussion 97about optimizing regular expressions for efficient performance. This document 98contains a few observations about PCRE. 99</P> 100<P> 101Using Unicode character properties (the \p, \P, and \X escapes) is slow, 102because PCRE has to scan a structure that contains data for over fifteen 103thousand characters whenever it needs a character's property. If you can find 104an alternative pattern that does not use character properties, it will probably 105be faster. 106</P> 107<P> 108When a pattern begins with .* not in parentheses, or in parentheses that are 109not the subject of a backreference, and the PCRE_DOTALL option is set, the 110pattern is implicitly anchored by PCRE, since it can match only at the start of 111a subject string. However, if PCRE_DOTALL is not set, PCRE cannot make this 112optimization, because the . metacharacter does not then match a newline, and if 113the subject string contains newlines, the pattern may match from the character 114immediately following one of them instead of from the very start. For example, 115the pattern 116<pre> 117 .*second 118</pre> 119matches the subject "first\nand second" (where \n stands for a newline 120character), with the match starting at the seventh character. In order to do 121this, PCRE has to retry the match starting after every newline in the subject. 122</P> 123<P> 124If you are using such a pattern with subject strings that do not contain 125newlines, the best performance is obtained by setting PCRE_DOTALL, or starting 126the pattern with ^.* or ^.*? to indicate explicit anchoring. That saves PCRE 127from having to scan along the subject looking for a newline to restart at. 128</P> 129<P> 130Beware of patterns that contain nested indefinite repeats. These can take a 131long time to run when applied to a string that does not match. Consider the 132pattern fragment 133<pre> 134 ^(a+)* 135</pre> 136This can match "aaaa" in 16 different ways, and this number increases very 137rapidly as the string gets longer. (The * repeat can match 0, 1, 2, 3, or 4 138times, and for each of those cases other than 0 or 4, the + repeats can match 139different numbers of times.) When the remainder of the pattern is such that the 140entire match is going to fail, PCRE has in principle to try every possible 141variation, and this can take an extremely long time, even for relatively short 142strings. 143</P> 144<P> 145An optimization catches some of the more simple cases such as 146<pre> 147 (a+)*b 148</pre> 149where a literal character follows. Before embarking on the standard matching 150procedure, PCRE checks that there is a "b" later in the subject string, and if 151there is not, it fails the match immediately. However, when there is no 152following literal this optimization cannot be used. You can see the difference 153by comparing the behaviour of 154<pre> 155 (a+)*\d 156</pre> 157with the pattern above. The former gives a failure almost instantly when 158applied to a whole line of "a" characters, whereas the latter takes an 159appreciable time with strings longer than about 20 characters. 160</P> 161<P> 162In many cases, the solution to this kind of performance issue is to use an 163atomic group or a possessive quantifier. 164</P> 165<br><b> 166AUTHOR 167</b><br> 168<P> 169Philip Hazel 170<br> 171University Computing Service 172<br> 173Cambridge CB2 3QH, England. 174<br> 175</P> 176<br><b> 177REVISION 178</b><br> 179<P> 180Last updated: 07 March 2010 181<br> 182Copyright © 1997-2010 University of Cambridge. 183<br> 184<p> 185Return to the <a href="index.html">PCRE index page</a>. 186</p> 187