ViewVC logotype

Contents of /code/trunk/HACKING

Parent Directory Parent Directory | Revision Log Revision Log

Revision 460 - (hide annotations) (download)
Sun Oct 4 09:27:20 2009 UTC (5 years, 6 months ago) by ph10
File size: 18258 byte(s)
Documentation update.

1 nigel 41 Technical Notes about PCRE
2     --------------------------
4 nigel 91 These are very rough technical notes that record potentially useful information
5     about PCRE internals.
7 nigel 75 Historical note 1
8     -----------------
10 nigel 41 Many years ago I implemented some regular expression functions to an algorithm
11     suggested by Martin Richards. These were not Unix-like in form, and were quite
12     restricted in what they could do by comparison with Perl. The interesting part
13     about the algorithm was that the amount of space required to hold the compiled
14     form of an expression was known in advance. The code to apply an expression did
15 nigel 63 not operate by backtracking, as the original Henry Spencer code and current
16     Perl code does, but instead checked all possibilities simultaneously by keeping
17     a list of current states and checking all of them as it advanced through the
18 nigel 75 subject string. In the terminology of Jeffrey Friedl's book, it was a "DFA
19 nigel 93 algorithm", though it was not a traditional Finite State Machine (FSM). When
20     the pattern was all used up, all remaining states were possible matches, and
21     the one matching the longest subset of the subject string was chosen. This did
22     not necessarily maximize the individual wild portions of the pattern, as is
23     expected in Unix and Perl-style regular expressions.
24 nigel 41
25 nigel 75 Historical note 2
26     -----------------
28 nigel 91 By contrast, the code originally written by Henry Spencer (which was
29     subsequently heavily modified for Perl) compiles the expression twice: once in
30     a dummy mode in order to find out how much store will be needed, and then for
31     real. (The Perl version probably doesn't do this any more; I'm talking about
32     the original library.) The execution function operates by backtracking and
33     maximizing (or, optionally, minimizing in Perl) the amount of the subject that
34     matches individual wild portions of the pattern. This is an "NFA algorithm" in
35     Friedl's terminology.
36 nigel 41
37 nigel 75 OK, here's the real stuff
38     -------------------------
40 nigel 77 For the set of functions that form the "basic" PCRE library (which are
41     unrelated to those mentioned above), I tried at first to invent an algorithm
42     that used an amount of store bounded by a multiple of the number of characters
43     in the pattern, to save on compiling time. However, because of the greater
44     complexity in Perl regular expressions, I couldn't do this. In any case, a
45 nigel 93 first pass through the pattern is helpful for other reasons.
46 nigel 41
47 nigel 93 Computing the memory requirement: how it was
48     --------------------------------------------
50     Up to and including release 6.7, PCRE worked by running a very degenerate first
51     pass to calculate a maximum store size, and then a second pass to do the real
52     compile - which might use a bit less than the predicted amount of memory. The
53     idea was that this would turn out faster than the Henry Spencer code because
54     the first pass is degenerate and the second pass can just store stuff straight
55     into the vector, which it knows is big enough.
57     Computing the memory requirement: how it is
58     -------------------------------------------
60     By the time I was working on a potential 6.8 release, the degenerate first pass
61     had become very complicated and hard to maintain. Indeed one of the early
62     things I did for 6.8 was to fix Yet Another Bug in the memory computation. Then
63     I had a flash of inspiration as to how I could run the real compile function in
64     a "fake" mode that enables it to compute how much memory it would need, while
65     actually only ever using a few hundred bytes of working memory, and without too
66     many tests of the mode that might slow it down. So I re-factored the compiling
67     functions to work this way. This got rid of about 600 lines of source. It
68     should make future maintenance and development easier. As this was such a major
69     change, I never released 6.8, instead upping the number to 7.0 (other quite
70 ph10 456 major changes were also present in the 7.0 release).
71 nigel 93
72 ph10 456 A side effect of this work was that the previous limit of 200 on the nesting
73 nigel 93 depth of parentheses was removed. However, there is a downside: pcre_compile()
74     runs more slowly than before (30% or more, depending on the pattern) because it
75 ph10 456 is doing a full analysis of the pattern. My hope was that this would not be a
76     big issue, and in the event, nobody has commented on it.
77 nigel 93
78 nigel 77 Traditional matching function
79     -----------------------------
81     The "traditional", and original, matching function is called pcre_exec(), and
82     it implements an NFA algorithm, similar to the original Henry Spencer algorithm
83 ph10 456 and the way that Perl works. This is not surprising, since it is intended to be
84     as compatible with Perl as possible. This is the function most users of PCRE
85     will use most of the time.
86 nigel 77
87     Supplementary matching function
88     -------------------------------
90     From PCRE 6.0, there is also a supplementary matching function called
91     pcre_dfa_exec(). This implements a DFA matching algorithm that searches
92     simultaneously for all possible matches that start at one point in the subject
93     string. (Going back to my roots: see Historical Note 1 above.) This function
94     intreprets the same compiled pattern data as pcre_exec(); however, not all the
95 nigel 91 facilities are available, and those that are do not always work in quite the
96 nigel 77 same way. See the user documentation for details.
98 nigel 93 The algorithm that is used for pcre_dfa_exec() is not a traditional FSM,
99     because it may have a number of states active at one time. More work would be
100     needed at compile time to produce a traditional FSM where only one state is
101     ever active at once. I believe some other regex matchers work this way.
104 nigel 77 Format of compiled patterns
105     ---------------------------
107 nigel 41 The compiled form of a pattern is a vector of bytes, containing items of
108     variable length. The first byte in an item is an opcode, and the length of the
109 nigel 75 item is either implicit in the opcode or contained in the data bytes that
110     follow it.
111 nigel 41
112 ph10 212 In many cases below LINK_SIZE data values are specified for offsets within the
113     compiled pattern. The default value for LINK_SIZE is 2, but PCRE can be
114 nigel 93 compiled to use 3-byte or 4-byte values for these offsets (impairing the
115 nigel 75 performance). This is necessary only when patterns whose compiled length is
116 nigel 93 greater than 64K are going to be processed. In this description, we assume the
117 ph10 212 "normal" compilation options. Data values that are counts (e.g. for
118     quantifiers) are always just two bytes long.
119 nigel 75
120 ph10 212 A list of the opcodes follows:
121 nigel 75
122 ph10 456
123 nigel 41 Opcodes with no following data
124     ------------------------------
126     These items are all just one byte long
128     OP_END end of pattern
129 ph10 342 OP_ANY match any one character other than newline
130     OP_ALLANY match any one character, including newline
131 nigel 75 OP_ANYBYTE match any single byte, even in UTF-8 mode
132 nigel 41 OP_SOD match start of data: \A
133 nigel 71 OP_SOM, start of match (subject + offset): \G
134 ph10 181 OP_SET_SOM, set start of match (\K)
135 nigel 41 OP_CIRC ^ (start of data, or after \n in multiline)
138     OP_NOT_DIGIT \D
139     OP_DIGIT \d
140 ph10 181 OP_NOT_HSPACE \H
141     OP_HSPACE \h
142 nigel 41 OP_NOT_WHITESPACE \S
143     OP_WHITESPACE \s
144 ph10 181 OP_NOT_VSPACE \V
145     OP_VSPACE \v
146 nigel 41 OP_NOT_WORDCHAR \W
147     OP_WORDCHAR \w
148     OP_EODN match end of data or \n at end: \Z
149     OP_EOD match end of data: \z
150     OP_DOLL $ (end of data, or before \n in multiline)
151 nigel 75 OP_EXTUNI match an extended Unicode character
152 nigel 93 OP_ANYNL match any Unicode newline sequence
153 nigel 75
154 ph10 456 OP_ACCEPT ) These are Perl 5.10's "backtracking
155     OP_COMMIT ) control verbs". If OP_ACCEPT is inside
156     OP_FAIL ) capturing parentheses, it may be preceded
157     OP_PRUNE ) by one or more OP_CLOSE, followed by a 2-byte
158     OP_SKIP ) number, indicating which parentheses must be
159     OP_THEN ) closed.
160 ph10 212
161 nigel 41
162     Repeating single characters
163     ---------------------------
165 nigel 75 The common repeats (*, +, ?) when applied to a single character use the
166     following opcodes:
167 nigel 41
168     OP_STAR
169     OP_MINSTAR
170 nigel 93 OP_POSSTAR
171 nigel 41 OP_PLUS
172     OP_MINPLUS
173 nigel 93 OP_POSPLUS
174 nigel 41 OP_QUERY
176 nigel 93 OP_POSQUERY
177 nigel 41
178 nigel 75 In ASCII mode, these are two-byte items; in UTF-8 mode, the length is variable.
179 nigel 93 Those with "MIN" in their name are the minimizing versions. Those with "POS" in
180     their names are possessive versions. Each is followed by the character that is
181     to be repeated. Other repeats make use of
182 nigel 41
183     OP_UPTO
184     OP_MINUPTO
185 nigel 93 OP_POSUPTO
186 nigel 41 OP_EXACT
188     which are followed by a two-byte count (most significant first) and the
189     repeated character. OP_UPTO matches from 0 to the given number. A repeat with a
190     non-zero minimum and a fixed maximum is coded as an OP_EXACT followed by an
191 nigel 93 OP_UPTO (or OP_MINUPTO or OPT_POSUPTO).
192 nigel 41
194     Repeating character types
195     -------------------------
197     Repeats of things like \d are done exactly as for single characters, except
198     that instead of a character, the opcode for the type is stored in the data
199     byte. The opcodes are:
203 nigel 93 OP_TYPEPOSSTAR
204 nigel 41 OP_TYPEPLUS
206 nigel 93 OP_TYPEPOSPLUS
207 nigel 41 OP_TYPEQUERY
209 nigel 93 OP_TYPEPOSQUERY
210 nigel 41 OP_TYPEUPTO
212 nigel 93 OP_TYPEPOSUPTO
213 nigel 41 OP_TYPEEXACT
216 nigel 75 Match by Unicode property
217     -------------------------
219     OP_PROP and OP_NOTPROP are used for positive and negative matches of a
220     character by testing its Unicode property (the \p and \P escape sequences).
221 nigel 91 Each is followed by two bytes that encode the desired property as a type and a
222     value.
223 nigel 75
224 nigel 91 Repeats of these items use the OP_TYPESTAR etc. set of opcodes, followed by
225     three bytes: OP_PROP or OP_NOTPROP and then the desired property type and
226     value.
227 nigel 75
229     Matching literal characters
230 nigel 41 ---------------------------
232 nigel 75 The OP_CHAR opcode is followed by a single character that is to be matched
233     casefully. For caseless matching, OP_CHARNC is used. In UTF-8 mode, the
234     character may be more than one byte long. (Earlier versions of PCRE used
235     multi-character strings, but this was changed to allow some new features to be
236     added.)
237 nigel 41
239     Character classes
240     -----------------
242 nigel 75 If there is only one character, OP_CHAR or OP_CHARNC is used for a positive
243     class, and OP_NOT for a negative one (that is, for something like [^a]).
244     However, in UTF-8 mode, the use of OP_NOT applies only to characters with
245     values < 128, because OP_NOT is confined to single bytes.
246 nigel 41
247 nigel 63 Another set of repeating opcodes (OP_NOTSTAR etc.) are used for a repeated,
248     negated, single-character class. The normal ones (OP_STAR etc.) are used for a
249     repeated positive single-character class.
251 nigel 71 When there's more than one character in a class and all the characters are less
252 nigel 75 than 256, OP_CLASS is used for a positive class, and OP_NCLASS for a negative
253 nigel 71 one. In either case, the opcode is followed by a 32-byte bit map containing a 1
254     bit for every character that is acceptable. The bits are counted from the least
255     significant end of each byte.
256 nigel 41
257 nigel 75 The reason for having both OP_CLASS and OP_NCLASS is so that, in UTF-8 mode,
258     subject characters with values greater than 256 can be handled correctly. For
259 nigel 71 OP_CLASS they don't match, whereas for OP_NCLASS they do.
261 nigel 63 For classes containing characters with values > 255, OP_XCLASS is used. It
262     optionally uses a bit map (if any characters lie within it), followed by a list
263 nigel 75 of pairs and single characters. There is a flag character than indicates
264 nigel 63 whether it's a positive or a negative class.
265 nigel 41
266 nigel 63
267 nigel 41 Back references
268     ---------------
270 nigel 53 OP_REF is followed by two bytes containing the reference number.
271 nigel 41
273     Repeating character classes and back references
274     -----------------------------------------------
276 nigel 93 Single-character classes are handled specially (see above). This section
277     applies to OP_CLASS and OP_REF. In both cases, the repeat information follows
278     the base item. The matching code looks at the following opcode to see if it is
279     one of
280 nigel 41
281     OP_CRSTAR
283     OP_CRPLUS
285     OP_CRQUERY
287     OP_CRRANGE
290     All but the last two are just single-byte items. The others are followed by
291 nigel 93 four bytes of data, comprising the minimum and maximum repeat counts. There are
292     no special possessive opcodes for these repeats; a possessive repeat is
293     compiled into an atomic group.
294 nigel 41
296     Brackets and alternation
297     ------------------------
299 nigel 43 A pair of non-capturing (round) brackets is wrapped round each expression at
300 nigel 41 compile time, so alternation always happens in the context of brackets.
301 nigel 53
302 nigel 93 [Note for North Americans: "bracket" to some English speakers, including
303     myself, can be round, square, curly, or pointy. Hence this usage.]
304 nigel 41
305 nigel 93 Non-capturing brackets use the opcode OP_BRA. Originally PCRE was limited to 99
306     capturing brackets and it used a different opcode for each one. From release
307     3.5, the limit was removed by putting the bracket number into the data for
308     higher-numbered brackets. From release 7.0 all capturing brackets are handled
309     this way, using the single opcode OP_CBRA.
310 nigel 53
311 nigel 77 A bracket opcode is followed by LINK_SIZE bytes which give the offset to the
312     next alternative OP_ALT or, if there aren't any branches, to the matching
313     OP_KET opcode. Each OP_ALT is followed by LINK_SIZE bytes giving the offset to
314 nigel 93 the next one, or to the OP_KET opcode. For capturing brackets, the bracket
315     number immediately follows the offset, always as a 2-byte item.
316 nigel 41
317     OP_KET is used for subpatterns that do not repeat indefinitely, while
318     OP_KETRMIN and OP_KETRMAX are used for indefinite repetitions, minimally or
319 nigel 77 maximally respectively. All three are followed by LINK_SIZE bytes giving (as a
320 nigel 93 positive number) the offset back to the matching bracket opcode.
321 nigel 41
322     If a subpattern is quantified such that it is permitted to match zero times, it
323 ph10 335 is preceded by one of OP_BRAZERO, OP_BRAMINZERO, or OP_SKIPZERO. These are
324     single-byte opcodes that tell the matcher that skipping the following
325     subpattern entirely is a valid branch. In the case of the first two, not
326     skipping the pattern is also valid (greedy and non-greedy). The third is used
327     when a pattern has the quantifier {0,0}. It cannot be entirely discarded,
328     because it may be called as a subroutine from elsewhere in the regex.
329 nigel 41
330     A subpattern with an indefinite maximum repetition is replicated in the
331 nigel 75 compiled data its minimum number of times (or once with OP_BRAZERO if the
332     minimum is zero), with the final copy terminating with OP_KETRMIN or OP_KETRMAX
333     as appropriate.
334 nigel 41
335     A subpattern with a bounded maximum repetition is replicated in a nested
336 nigel 75 fashion up to the maximum number of times, with OP_BRAZERO or OP_BRAMINZERO
337     before each replication after the minimum, so that, for example, (abc){2,5} is
338 nigel 93 compiled as (abc)(abc)((abc)((abc)(abc)?)?)?, except that each bracketed group
339     has the same number.
340 nigel 41
341 nigel 93 When a repeated subpattern has an unbounded upper limit, it is checked to see
342     whether it could match an empty string. If this is the case, the opcode in the
343     final replication is changed to OP_SBRA or OP_SCBRA. This tells the matcher
344     that it needs to check for matching an empty string when it hits OP_KETRMIN or
345     OP_KETRMAX, and if so, to break the loop.
346 nigel 41
347 nigel 93
348 nigel 41 Assertions
349     ----------
351     Forward assertions are just like other subpatterns, but starting with one of
352     the opcodes OP_ASSERT or OP_ASSERT_NOT. Backward assertions use the opcodes
353     OP_ASSERTBACK and OP_ASSERTBACK_NOT, and the first opcode inside the assertion
354     is OP_REVERSE, followed by a two byte count of the number of characters to move
355 nigel 49 back the pointer in the subject string. When operating in UTF-8 mode, the count
356     is a character count rather than a byte count. A separate count is present in
357     each alternative of a lookbehind assertion, allowing them to have different
358     fixed lengths.
359 nigel 41
361 nigel 93 Once-only (atomic) subpatterns
362     ------------------------------
363 nigel 41
364     These are also just like other subpatterns, but they start with the opcode
365 nigel 93 OP_ONCE. The check for matching an empty string in an unbounded repeat is
366     handled entirely at runtime, so there is just this one opcode.
367 nigel 41
369     Conditional subpatterns
370     -----------------------
372 nigel 93 These are like other subpatterns, but they start with the opcode OP_COND, or
373     OP_SCOND for one that might match an empty string in an unbounded repeat. If
374 nigel 41 the condition is a back reference, this is stored at the start of the
375 nigel 53 subpattern using the opcode OP_CREF followed by two bytes containing the
376 ph10 460 reference number. OP_NCREF is used instead if the reference was generated by
377     name (so that the runtime code knows to check for duplicate names).
378 nigel 41
379 ph10 460 If the condition is "in recursion" (coded as "(?(R)"), or "in recursion of
380     group x" (coded as "(?(Rx)"), the group number is stored at the start of the
381     subpattern using the opcode OP_RREF or OP_NRREF (cf OP_NCREF), and a value of
382     zero for "the whole pattern". For a DEFINE condition, just the single byte
383     OP_DEF is used (it has no associated data). Otherwise, a conditional subpattern
384     always starts with one of the assertions.
385 nigel 41
386 ph10 460
387 nigel 71 Recursion
388     ---------
390     Recursion either matches the current regex, or some subexpression. The opcode
391     OP_RECURSE is followed by an value which is the offset to the starting bracket
392 nigel 87 from the start of the whole pattern. From release 6.5, OP_RECURSE is
393     automatically wrapped inside OP_ONCE brackets (because otherwise some patterns
394     broke it). OP_RECURSE is also used for "subroutine" calls, even though they
395     are not strictly a recursion.
396 nigel 71
398     Callout
399     -------
401 nigel 75 OP_CALLOUT is followed by one byte of data that holds a callout number in the
402     range 0 to 254 for manual callouts, or 255 for an automatic callout. In both
403     cases there follows a two-byte value giving the offset in the pattern to the
404     start of the following item, and another two-byte item giving the length of the
405     next item.
406 nigel 71
408 nigel 41 Changing options
409     ----------------
411 nigel 63 If any of the /i, /m, or /s options are changed within a pattern, an OP_OPT
412     opcode is compiled, followed by one byte containing the new settings of these
413     flags. If there are several alternatives, there is an occurrence of OP_OPT at
414     the start of all those following the first options change, to set appropriate
415     options for the start of the alternative. Immediately after the end of the
416     group there is another such item to reset the flags to their previous values. A
417     change of flag right at the very start of the pattern can be handled entirely
418     at compile time, and so does not cause anything to be put into the compiled
419     data.
420 nigel 41
421     Philip Hazel
422 ph10 456 October 2009

ViewVC Help
Powered by ViewVC 1.1.12