/[pcre]/code/trunk/doc/html/pcrestack.html
ViewVC logotype

Contents of /code/trunk/doc/html/pcrestack.html

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1194 - (hide annotations) (download) (as text)
Wed Oct 31 17:42:29 2012 UTC (2 years, 1 month ago) by ph10
File MIME type: text/html
File size: 9594 byte(s)
More documentation updates

1 nigel 91 <html>
2     <head>
3     <title>pcrestack specification</title>
4     </head>
5     <body bgcolor="#FFFFFF" text="#00005A" link="#0066FF" alink="#3399FF" vlink="#2222BB">
6     <h1>pcrestack man page</h1>
7     <p>
8     Return to the <a href="index.html">PCRE index page</a>.
9     </p>
10 ph10 111 <p>
11 nigel 91 This page is part of the PCRE HTML documentation. It was generated automatically
12     from the original man page. If there is any nonsense in it, please consult the
13     man page, in case the conversion went wrong.
14 ph10 111 <br>
15 nigel 91 <br><b>
16     PCRE DISCUSSION OF STACK USAGE
17     </b><br>
18     <P>
19 ph10 1194 When you call <b>pcre[16|32]_exec()</b>, it makes use of an internal function
20 ph10 869 called <b>match()</b>. This calls itself recursively at branch points in the
21     pattern, in order to remember the state of the match so that it can back up and
22     try a different alternative if the first one fails. As matching proceeds deeper
23     and deeper into the tree of possibilities, the recursion depth increases. The
24 ph10 654 <b>match()</b> function is also called in other circumstances, for example,
25     whenever a parenthesized sub-pattern is entered, and in certain cases of
26     repetition.
27 nigel 91 </P>
28     <P>
29     Not all calls of <b>match()</b> increase the recursion depth; for an item such
30     as a* it may be called several times at the same level, after matching
31     different numbers of a's. Furthermore, in a number of cases where the result of
32     the recursive call would immediately be passed back as the result of the
33     current call (a "tail recursion"), the function is just restarted instead.
34     </P>
35     <P>
36 ph10 1194 The above comments apply when <b>pcre[16|32]_exec()</b> is run in its normal
37 ph10 691 interpretive manner. If the pattern was studied with the
38     PCRE_STUDY_JIT_COMPILE option, and just-in-time compiling was successful, and
39 ph10 1194 the options passed to <b>pcre[16|32]_exec()</b> were not incompatible, the matching
40 ph10 691 process uses the JIT-compiled code instead of the <b>match()</b> function. In
41     this case, the memory requirements are handled entirely differently. See the
42     <a href="pcrejit.html"><b>pcrejit</b></a>
43     documentation for details.
44     </P>
45     <P>
46 ph10 1194 The <b>pcre[16|32]_dfa_exec()</b> function operates in an entirely different way,
47 ph10 869 and uses recursion only when there is a regular expression recursion or
48     subroutine call in the pattern. This includes the processing of assertion and
49     "once-only" subpatterns, which are handled like subroutine calls. Normally,
50     these are never very deep, and the limit on the complexity of
51 ph10 1194 <b>pcre[16|32]_dfa_exec()</b> is controlled by the amount of workspace it is given.
52 ph10 869 However, it is possible to write patterns with runaway infinite recursions;
53 ph10 1194 such patterns will cause <b>pcre[16|32]_dfa_exec()</b> to run out of stack. At
54 ph10 869 present, there is no protection against this.
55 nigel 91 </P>
56     <P>
57 ph10 1194 The comments that follow do NOT apply to <b>pcre[16|32]_dfa_exec()</b>; they are
58     relevant only for <b>pcre[16|32]_exec()</b> without the JIT optimization.
59 nigel 91 </P>
60 ph10 487 <br><b>
61 ph10 1194 Reducing <b>pcre[16|32]_exec()</b>'s stack usage
62 ph10 487 </b><br>
63 nigel 91 <P>
64     Each time that <b>match()</b> is actually called recursively, it uses memory
65     from the process stack. For certain kinds of pattern and data, very large
66     amounts of stack may be needed, despite the recognition of "tail recursion".
67     You can often reduce the amount of recursion, and therefore the amount of stack
68     used, by modifying the pattern that is being matched. Consider, for example,
69     this pattern:
70     <pre>
71     ([^&#60;]|&#60;(?!inet))+
72     </pre>
73     It matches from wherever it starts until it encounters "&#60;inet" or the end of
74     the data, and is the kind of pattern that might be used when processing an XML
75     file. Each iteration of the outer parentheses matches either one character that
76     is not "&#60;" or a "&#60;" that is not followed by "inet". However, each time a
77     parenthesis is processed, a recursion occurs, so this formulation uses a stack
78     frame for each matched character. For a long string, a lot of stack is
79     required. Consider now this rewritten pattern, which matches exactly the same
80     strings:
81     <pre>
82 ph10 123 ([^&#60;]++|&#60;(?!inet))+
83 nigel 91 </pre>
84     This uses very much less stack, because runs of characters that do not contain
85     "&#60;" are "swallowed" in one item inside the parentheses. Recursion happens only
86     when a "&#60;" character that is not followed by "inet" is encountered (and we
87     assume this is relatively rare). A possessive quantifier is used to stop any
88     backtracking into the runs of non-"&#60;" characters, but that is not related to
89     stack usage.
90     </P>
91     <P>
92 nigel 93 This example shows that one way of avoiding stack problems when matching long
93     subject strings is to write repeated parenthesized subpatterns to match more
94     than one character whenever possible.
95     </P>
96 ph10 358 <br><b>
97 ph10 1194 Compiling PCRE to use heap instead of stack for <b>pcre[16|32]_exec()</b>
98 ph10 358 </b><br>
99 nigel 93 <P>
100 nigel 91 In environments where stack memory is constrained, you might want to compile
101 ph10 487 PCRE to use heap memory instead of stack for remembering back-up points when
102 ph10 1194 <b>pcre[16|32]_exec()</b> is running. This makes it run a lot more slowly, however.
103 ph10 487 Details of how to do this are given in the
104 nigel 91 <a href="pcrebuild.html"><b>pcrebuild</b></a>
105 ph10 182 documentation. When built in this way, instead of using the stack, PCRE obtains
106     and frees memory by calling the functions that are pointed to by the
107 ph10 1194 <b>pcre[16|32]_stack_malloc</b> and <b>pcre[16|32]_stack_free</b> variables. By
108 ph10 869 default, these point to <b>malloc()</b> and <b>free()</b>, but you can replace
109     the pointers to cause PCRE to use your own functions. Since the block sizes are
110     always the same, and are always freed in reverse order, it may be possible to
111     implement customized memory handlers that are more efficient than the standard
112     functions.
113 nigel 91 </P>
114 ph10 358 <br><b>
115 ph10 1194 Limiting <b>pcre[16|32]_exec()</b>'s stack usage
116 ph10 358 </b><br>
117 nigel 91 <P>
118 ph10 487 You can set limits on the number of times that <b>match()</b> is called, both in
119 ph10 1194 total and recursively. If a limit is exceeded, <b>pcre[16|32]_exec()</b> returns an
120 ph10 487 error code. Setting suitable limits should prevent it from running out of
121     stack. The default values of the limits are very large, and unlikely ever to
122     operate. They can be changed when PCRE is built, and they can also be set when
123 ph10 1194 <b>pcre[16|32]_exec()</b> is called. For details of these interfaces, see the
124 ph10 358 <a href="pcrebuild.html"><b>pcrebuild</b></a>
125 ph10 487 documentation and the
126 ph10 1194 <a href="pcreapi.html#extradata">section on extra data for <b>pcre[16|32]_exec()</b></a>
127 ph10 487 in the
128 ph10 358 <a href="pcreapi.html"><b>pcreapi</b></a>
129     documentation.
130     </P>
131     <P>
132     As a very rough rule of thumb, you should reckon on about 500 bytes per
133 ph10 903 recursion. Thus, if you want to limit your stack usage to 8Mb, you should set
134     the limit at 16000 recursions. A 64Mb stack, on the other hand, can support
135     around 128000 recursions.
136 ph10 358 </P>
137 ph10 487 <P>
138     In Unix-like environments, the <b>pcretest</b> test program has a command line
139     option (<b>-S</b>) that can be used to increase the size of its stack. As long
140     as the stack is large enough, another option (<b>-M</b>) can be used to find the
141     smallest limits that allow a particular pattern to match a given subject
142 ph10 1194 string. This is done by calling <b>pcre[16|32]_exec()</b> repeatedly with different
143 ph10 487 limits.
144     </P>
145 ph10 358 <br><b>
146 ph10 903 Obtaining an estimate of stack usage
147     </b><br>
148     <P>
149     The actual amount of stack used per recursion can vary quite a lot, depending
150     on the compiler that was used to build PCRE and the optimization or debugging
151     options that were set for it. The rule of thumb value of 500 bytes mentioned
152     above may be larger or smaller than what is actually needed. A better
153     approximation can be obtained by running this command:
154     <pre>
155     pcretest -m -C
156     </pre>
157     The <b>-C</b> option causes <b>pcretest</b> to output information about the
158     options with which PCRE was compiled. When <b>-m</b> is also given (before
159     <b>-C</b>), information about stack use is given in a line like this:
160     <pre>
161     Match recursion uses stack: approximate frame size = 640 bytes
162     </pre>
163     The value is approximate because some recursions need a bit more (up to perhaps
164     16 more bytes).
165     </P>
166     <P>
167     If the above command is given when PCRE is compiled to use the heap instead of
168     the stack for recursion, the value that is output is the size of each block
169     that is obtained from the heap.
170     </P>
171     <br><b>
172 ph10 358 Changing stack size in Unix-like systems
173     </b><br>
174     <P>
175 nigel 93 In Unix-like environments, there is not often a problem with the stack unless
176     very long strings are involved, though the default limit on stack size varies
177     from system to system. Values from 8Mb to 64Mb are common. You can find your
178     default limit by running the command:
179 nigel 91 <pre>
180     ulimit -s
181     </pre>
182 nigel 93 Unfortunately, the effect of running out of stack is often SIGSEGV, though
183     sometimes a more explicit error message is given. You can normally increase the
184     limit on stack size by code such as this:
185 nigel 91 <pre>
186     struct rlimit rlim;
187     getrlimit(RLIMIT_STACK, &rlim);
188     rlim.rlim_cur = 100*1024*1024;
189     setrlimit(RLIMIT_STACK, &rlim);
190     </pre>
191     This reads the current limits (soft and hard) using <b>getrlimit()</b>, then
192     attempts to increase the soft limit to 100Mb using <b>setrlimit()</b>. You must
193 ph10 1194 do this before calling <b>pcre[16|32]_exec()</b>.
194 nigel 91 </P>
195 ph10 358 <br><b>
196     Changing stack size in Mac OS X
197     </b><br>
198 nigel 91 <P>
199 ph10 358 Using <b>setrlimit()</b>, as described above, should also work on Mac OS X. It
200     is also possible to set a stack size when linking a program. There is a
201     discussion about stack sizes in Mac OS X at this web site:
202     <a href="http://developer.apple.com/qa/qa2005/qa1419.html">http://developer.apple.com/qa/qa2005/qa1419.html.</a>
203 nigel 91 </P>
204 ph10 99 <br><b>
205     AUTHOR
206     </b><br>
207 nigel 91 <P>
208 ph10 99 Philip Hazel
209 nigel 91 <br>
210 ph10 99 University Computing Service
211     <br>
212     Cambridge CB2 3QH, England.
213     <br>
214     </P>
215     <br><b>
216     REVISION
217     </b><br>
218     <P>
219 ph10 1194 Last updated: 24 June 2012
220 ph10 99 <br>
221 ph10 869 Copyright &copy; 1997-2012 University of Cambridge.
222 ph10 99 <br>
223 nigel 91 <p>
224     Return to the <a href="index.html">PCRE index page</a>.
225     </p>

Properties

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

webmaster@exim.org
ViewVC Help
Powered by ViewVC 1.1.12