pcrestack.3 8.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215
  1. .TH PCRESTACK 3 "24 June 2012" "PCRE 8.30"
  2. .SH NAME
  3. PCRE - Perl-compatible regular expressions
  4. .SH "PCRE DISCUSSION OF STACK USAGE"
  5. .rs
  6. .sp
  7. When you call \fBpcre[16|32]_exec()\fP, it makes use of an internal function
  8. called \fBmatch()\fP. This calls itself recursively at branch points in the
  9. pattern, in order to remember the state of the match so that it can back up and
  10. try a different alternative if the first one fails. As matching proceeds deeper
  11. and deeper into the tree of possibilities, the recursion depth increases. The
  12. \fBmatch()\fP function is also called in other circumstances, for example,
  13. whenever a parenthesized sub-pattern is entered, and in certain cases of
  14. repetition.
  15. .P
  16. Not all calls of \fBmatch()\fP increase the recursion depth; for an item such
  17. as a* it may be called several times at the same level, after matching
  18. different numbers of a's. Furthermore, in a number of cases where the result of
  19. the recursive call would immediately be passed back as the result of the
  20. current call (a "tail recursion"), the function is just restarted instead.
  21. .P
  22. The above comments apply when \fBpcre[16|32]_exec()\fP is run in its normal
  23. interpretive manner. If the pattern was studied with the
  24. PCRE_STUDY_JIT_COMPILE option, and just-in-time compiling was successful, and
  25. the options passed to \fBpcre[16|32]_exec()\fP were not incompatible, the matching
  26. process uses the JIT-compiled code instead of the \fBmatch()\fP function. In
  27. this case, the memory requirements are handled entirely differently. See the
  28. .\" HREF
  29. \fBpcrejit\fP
  30. .\"
  31. documentation for details.
  32. .P
  33. The \fBpcre[16|32]_dfa_exec()\fP function operates in an entirely different way,
  34. and uses recursion only when there is a regular expression recursion or
  35. subroutine call in the pattern. This includes the processing of assertion and
  36. "once-only" subpatterns, which are handled like subroutine calls. Normally,
  37. these are never very deep, and the limit on the complexity of
  38. \fBpcre[16|32]_dfa_exec()\fP is controlled by the amount of workspace it is given.
  39. However, it is possible to write patterns with runaway infinite recursions;
  40. such patterns will cause \fBpcre[16|32]_dfa_exec()\fP to run out of stack. At
  41. present, there is no protection against this.
  42. .P
  43. The comments that follow do NOT apply to \fBpcre[16|32]_dfa_exec()\fP; they are
  44. relevant only for \fBpcre[16|32]_exec()\fP without the JIT optimization.
  45. .
  46. .
  47. .SS "Reducing \fBpcre[16|32]_exec()\fP's stack usage"
  48. .rs
  49. .sp
  50. Each time that \fBmatch()\fP is actually called recursively, it uses memory
  51. from the process stack. For certain kinds of pattern and data, very large
  52. amounts of stack may be needed, despite the recognition of "tail recursion".
  53. You can often reduce the amount of recursion, and therefore the amount of stack
  54. used, by modifying the pattern that is being matched. Consider, for example,
  55. this pattern:
  56. .sp
  57. ([^<]|<(?!inet))+
  58. .sp
  59. It matches from wherever it starts until it encounters "<inet" or the end of
  60. the data, and is the kind of pattern that might be used when processing an XML
  61. file. Each iteration of the outer parentheses matches either one character that
  62. is not "<" or a "<" that is not followed by "inet". However, each time a
  63. parenthesis is processed, a recursion occurs, so this formulation uses a stack
  64. frame for each matched character. For a long string, a lot of stack is
  65. required. Consider now this rewritten pattern, which matches exactly the same
  66. strings:
  67. .sp
  68. ([^<]++|<(?!inet))+
  69. .sp
  70. This uses very much less stack, because runs of characters that do not contain
  71. "<" are "swallowed" in one item inside the parentheses. Recursion happens only
  72. when a "<" character that is not followed by "inet" is encountered (and we
  73. assume this is relatively rare). A possessive quantifier is used to stop any
  74. backtracking into the runs of non-"<" characters, but that is not related to
  75. stack usage.
  76. .P
  77. This example shows that one way of avoiding stack problems when matching long
  78. subject strings is to write repeated parenthesized subpatterns to match more
  79. than one character whenever possible.
  80. .
  81. .
  82. .SS "Compiling PCRE to use heap instead of stack for \fBpcre[16|32]_exec()\fP"
  83. .rs
  84. .sp
  85. In environments where stack memory is constrained, you might want to compile
  86. PCRE to use heap memory instead of stack for remembering back-up points when
  87. \fBpcre[16|32]_exec()\fP is running. This makes it run a lot more slowly, however.
  88. Details of how to do this are given in the
  89. .\" HREF
  90. \fBpcrebuild\fP
  91. .\"
  92. documentation. When built in this way, instead of using the stack, PCRE obtains
  93. and frees memory by calling the functions that are pointed to by the
  94. \fBpcre[16|32]_stack_malloc\fP and \fBpcre[16|32]_stack_free\fP variables. By
  95. default, these point to \fBmalloc()\fP and \fBfree()\fP, but you can replace
  96. the pointers to cause PCRE to use your own functions. Since the block sizes are
  97. always the same, and are always freed in reverse order, it may be possible to
  98. implement customized memory handlers that are more efficient than the standard
  99. functions.
  100. .
  101. .
  102. .SS "Limiting \fBpcre[16|32]_exec()\fP's stack usage"
  103. .rs
  104. .sp
  105. You can set limits on the number of times that \fBmatch()\fP is called, both in
  106. total and recursively. If a limit is exceeded, \fBpcre[16|32]_exec()\fP returns an
  107. error code. Setting suitable limits should prevent it from running out of
  108. stack. The default values of the limits are very large, and unlikely ever to
  109. operate. They can be changed when PCRE is built, and they can also be set when
  110. \fBpcre[16|32]_exec()\fP is called. For details of these interfaces, see the
  111. .\" HREF
  112. \fBpcrebuild\fP
  113. .\"
  114. documentation and the
  115. .\" HTML <a href="pcreapi.html#extradata">
  116. .\" </a>
  117. section on extra data for \fBpcre[16|32]_exec()\fP
  118. .\"
  119. in the
  120. .\" HREF
  121. \fBpcreapi\fP
  122. .\"
  123. documentation.
  124. .P
  125. As a very rough rule of thumb, you should reckon on about 500 bytes per
  126. recursion. Thus, if you want to limit your stack usage to 8Mb, you should set
  127. the limit at 16000 recursions. A 64Mb stack, on the other hand, can support
  128. around 128000 recursions.
  129. .P
  130. In Unix-like environments, the \fBpcretest\fP test program has a command line
  131. option (\fB-S\fP) that can be used to increase the size of its stack. As long
  132. as the stack is large enough, another option (\fB-M\fP) can be used to find the
  133. smallest limits that allow a particular pattern to match a given subject
  134. string. This is done by calling \fBpcre[16|32]_exec()\fP repeatedly with different
  135. limits.
  136. .
  137. .
  138. .SS "Obtaining an estimate of stack usage"
  139. .rs
  140. .sp
  141. The actual amount of stack used per recursion can vary quite a lot, depending
  142. on the compiler that was used to build PCRE and the optimization or debugging
  143. options that were set for it. The rule of thumb value of 500 bytes mentioned
  144. above may be larger or smaller than what is actually needed. A better
  145. approximation can be obtained by running this command:
  146. .sp
  147. pcretest -m -C
  148. .sp
  149. The \fB-C\fP option causes \fBpcretest\fP to output information about the
  150. options with which PCRE was compiled. When \fB-m\fP is also given (before
  151. \fB-C\fP), information about stack use is given in a line like this:
  152. .sp
  153. Match recursion uses stack: approximate frame size = 640 bytes
  154. .sp
  155. The value is approximate because some recursions need a bit more (up to perhaps
  156. 16 more bytes).
  157. .P
  158. If the above command is given when PCRE is compiled to use the heap instead of
  159. the stack for recursion, the value that is output is the size of each block
  160. that is obtained from the heap.
  161. .
  162. .
  163. .SS "Changing stack size in Unix-like systems"
  164. .rs
  165. .sp
  166. In Unix-like environments, there is not often a problem with the stack unless
  167. very long strings are involved, though the default limit on stack size varies
  168. from system to system. Values from 8Mb to 64Mb are common. You can find your
  169. default limit by running the command:
  170. .sp
  171. ulimit -s
  172. .sp
  173. Unfortunately, the effect of running out of stack is often SIGSEGV, though
  174. sometimes a more explicit error message is given. You can normally increase the
  175. limit on stack size by code such as this:
  176. .sp
  177. struct rlimit rlim;
  178. getrlimit(RLIMIT_STACK, &rlim);
  179. rlim.rlim_cur = 100*1024*1024;
  180. setrlimit(RLIMIT_STACK, &rlim);
  181. .sp
  182. This reads the current limits (soft and hard) using \fBgetrlimit()\fP, then
  183. attempts to increase the soft limit to 100Mb using \fBsetrlimit()\fP. You must
  184. do this before calling \fBpcre[16|32]_exec()\fP.
  185. .
  186. .
  187. .SS "Changing stack size in Mac OS X"
  188. .rs
  189. .sp
  190. Using \fBsetrlimit()\fP, as described above, should also work on Mac OS X. It
  191. is also possible to set a stack size when linking a program. There is a
  192. discussion about stack sizes in Mac OS X at this web site:
  193. .\" HTML <a href="http://developer.apple.com/qa/qa2005/qa1419.html">
  194. .\" </a>
  195. http://developer.apple.com/qa/qa2005/qa1419.html.
  196. .\"
  197. .
  198. .
  199. .SH AUTHOR
  200. .rs
  201. .sp
  202. .nf
  203. Philip Hazel
  204. University Computing Service
  205. Cambridge CB2 3QH, England.
  206. .fi
  207. .
  208. .
  209. .SH REVISION
  210. .rs
  211. .sp
  212. .nf
  213. Last updated: 24 June 2012
  214. Copyright (c) 1997-2012 University of Cambridge.
  215. .fi