/[pcre]/code/trunk/doc/Tech.Notes
ViewVC logotype

Contents of /code/trunk/doc/Tech.Notes

Parent Directory Parent Directory | Revision Log Revision Log


Revision 75 - (hide annotations) (download)
Sat Feb 24 21:40:37 2007 UTC (7 years, 1 month ago) by nigel
File size: 12771 byte(s)
Load pcre-5.0 into code/trunk.

1 nigel 41 Technical Notes about PCRE
2     --------------------------
3    
4 nigel 75 Historical note 1
5     -----------------
6    
7 nigel 41 Many years ago I implemented some regular expression functions to an algorithm
8     suggested by Martin Richards. These were not Unix-like in form, and were quite
9     restricted in what they could do by comparison with Perl. The interesting part
10     about the algorithm was that the amount of space required to hold the compiled
11     form of an expression was known in advance. The code to apply an expression did
12 nigel 63 not operate by backtracking, as the original Henry Spencer code and current
13     Perl code does, but instead checked all possibilities simultaneously by keeping
14     a list of current states and checking all of them as it advanced through the
15 nigel 75 subject string. In the terminology of Jeffrey Friedl's book, it was a "DFA
16     algorithm". When the pattern was all used up, all remaining states were
17 nigel 63 possible matches, and the one matching the longest subset of the subject string
18     was chosen. This did not necessarily maximize the individual wild portions of
19     the pattern, as is expected in Unix and Perl-style regular expressions.
20 nigel 41
21 nigel 75 Historical note 2
22     -----------------
23    
24 nigel 41 By contrast, the code originally written by Henry Spencer and subsequently
25     heavily modified for Perl actually compiles the expression twice: once in a
26     dummy mode in order to find out how much store will be needed, and then for
27     real. The execution function operates by backtracking and maximizing (or,
28     optionally, minimizing in Perl) the amount of the subject that matches
29     individual wild portions of the pattern. This is an "NFA algorithm" in Friedl's
30     terminology.
31    
32 nigel 75 OK, here's the real stuff
33     -------------------------
34    
35 nigel 43 For the set of functions that forms PCRE (which are unrelated to those
36     mentioned above), I tried at first to invent an algorithm that used an amount
37     of store bounded by a multiple of the number of characters in the pattern, to
38     save on compiling time. However, because of the greater complexity in Perl
39     regular expressions, I couldn't do this. In any case, a first pass through the
40 nigel 63 pattern is needed, for a number of reasons. PCRE works by running a very
41     degenerate first pass to calculate a maximum store size, and then a second pass
42     to do the real compile - which may use a bit less than the predicted amount of
43     store. The idea is that this is going to turn out faster because the first pass
44     is degenerate and the second pass can just store stuff straight into the
45     vector. It does make the compiling functions bigger, of course, but they have
46     got quite big anyway to handle all the Perl stuff.
47 nigel 41
48     The compiled form of a pattern is a vector of bytes, containing items of
49     variable length. The first byte in an item is an opcode, and the length of the
50 nigel 75 item is either implicit in the opcode or contained in the data bytes that
51     follow it.
52 nigel 41
53 nigel 75 In many cases below "two-byte" data values are specified. This is in fact just
54     a default. PCRE can be compiled to use 3-byte or 4-byte values (impairing the
55     performance). This is necessary only when patterns whose compiled length is
56     greater than 64K are going to be processed. In this description, we assume the
57     "normal" compilation options.
58    
59     A list of all the opcodes follows:
60    
61 nigel 41 Opcodes with no following data
62     ------------------------------
63    
64     These items are all just one byte long
65    
66     OP_END end of pattern
67     OP_ANY match any character
68 nigel 75 OP_ANYBYTE match any single byte, even in UTF-8 mode
69 nigel 41 OP_SOD match start of data: \A
70 nigel 71 OP_SOM, start of match (subject + offset): \G
71 nigel 41 OP_CIRC ^ (start of data, or after \n in multiline)
72     OP_NOT_WORD_BOUNDARY \W
73     OP_WORD_BOUNDARY \w
74     OP_NOT_DIGIT \D
75     OP_DIGIT \d
76     OP_NOT_WHITESPACE \S
77     OP_WHITESPACE \s
78     OP_NOT_WORDCHAR \W
79     OP_WORDCHAR \w
80     OP_EODN match end of data or \n at end: \Z
81     OP_EOD match end of data: \z
82     OP_DOLL $ (end of data, or before \n in multiline)
83 nigel 75 OP_EXTUNI match an extended Unicode character
84    
85 nigel 41
86     Repeating single characters
87     ---------------------------
88    
89 nigel 75 The common repeats (*, +, ?) when applied to a single character use the
90     following opcodes:
91 nigel 41
92     OP_STAR
93     OP_MINSTAR
94     OP_PLUS
95     OP_MINPLUS
96     OP_QUERY
97     OP_MINQUERY
98    
99 nigel 75 In ASCII mode, these are two-byte items; in UTF-8 mode, the length is variable.
100 nigel 41 Those with "MIN" in their name are the minimizing versions. Each is followed by
101     the character that is to be repeated. Other repeats make use of
102    
103     OP_UPTO
104     OP_MINUPTO
105     OP_EXACT
106    
107     which are followed by a two-byte count (most significant first) and the
108     repeated character. OP_UPTO matches from 0 to the given number. A repeat with a
109     non-zero minimum and a fixed maximum is coded as an OP_EXACT followed by an
110     OP_UPTO (or OP_MINUPTO).
111    
112    
113     Repeating character types
114     -------------------------
115    
116     Repeats of things like \d are done exactly as for single characters, except
117     that instead of a character, the opcode for the type is stored in the data
118     byte. The opcodes are:
119    
120     OP_TYPESTAR
121     OP_TYPEMINSTAR
122     OP_TYPEPLUS
123     OP_TYPEMINPLUS
124     OP_TYPEQUERY
125     OP_TYPEMINQUERY
126     OP_TYPEUPTO
127     OP_TYPEMINUPTO
128     OP_TYPEEXACT
129    
130    
131 nigel 75 Match by Unicode property
132     -------------------------
133    
134     OP_PROP and OP_NOTPROP are used for positive and negative matches of a
135     character by testing its Unicode property (the \p and \P escape sequences).
136     Each is followed by a single byte that encodes the desired property value.
137    
138     Repeats of these items use the OP_TYPESTAR etc. set of opcodes, followed by two
139     bytes: OP_PROP or OP_NOTPROP and then the desired property value.
140    
141    
142     Matching literal characters
143 nigel 41 ---------------------------
144    
145 nigel 75 The OP_CHAR opcode is followed by a single character that is to be matched
146     casefully. For caseless matching, OP_CHARNC is used. In UTF-8 mode, the
147     character may be more than one byte long. (Earlier versions of PCRE used
148     multi-character strings, but this was changed to allow some new features to be
149     added.)
150 nigel 41
151    
152     Character classes
153     -----------------
154    
155 nigel 75 If there is only one character, OP_CHAR or OP_CHARNC is used for a positive
156     class, and OP_NOT for a negative one (that is, for something like [^a]).
157     However, in UTF-8 mode, the use of OP_NOT applies only to characters with
158     values < 128, because OP_NOT is confined to single bytes.
159 nigel 41
160 nigel 63 Another set of repeating opcodes (OP_NOTSTAR etc.) are used for a repeated,
161     negated, single-character class. The normal ones (OP_STAR etc.) are used for a
162     repeated positive single-character class.
163    
164 nigel 71 When there's more than one character in a class and all the characters are less
165 nigel 75 than 256, OP_CLASS is used for a positive class, and OP_NCLASS for a negative
166 nigel 71 one. In either case, the opcode is followed by a 32-byte bit map containing a 1
167     bit for every character that is acceptable. The bits are counted from the least
168     significant end of each byte.
169 nigel 41
170 nigel 75 The reason for having both OP_CLASS and OP_NCLASS is so that, in UTF-8 mode,
171     subject characters with values greater than 256 can be handled correctly. For
172 nigel 71 OP_CLASS they don't match, whereas for OP_NCLASS they do.
173    
174 nigel 63 For classes containing characters with values > 255, OP_XCLASS is used. It
175     optionally uses a bit map (if any characters lie within it), followed by a list
176 nigel 75 of pairs and single characters. There is a flag character than indicates
177 nigel 63 whether it's a positive or a negative class.
178 nigel 41
179 nigel 63
180 nigel 41 Back references
181     ---------------
182    
183 nigel 53 OP_REF is followed by two bytes containing the reference number.
184 nigel 41
185    
186     Repeating character classes and back references
187     -----------------------------------------------
188    
189     Single-character classes are handled specially (see above). This applies to
190     OP_CLASS and OP_REF. In both cases, the repeat information follows the base
191     item. The matching code looks at the following opcode to see if it is one of
192    
193     OP_CRSTAR
194     OP_CRMINSTAR
195     OP_CRPLUS
196     OP_CRMINPLUS
197     OP_CRQUERY
198     OP_CRMINQUERY
199     OP_CRRANGE
200     OP_CRMINRANGE
201    
202     All but the last two are just single-byte items. The others are followed by
203     four bytes of data, comprising the minimum and maximum repeat counts.
204    
205    
206     Brackets and alternation
207     ------------------------
208    
209 nigel 43 A pair of non-capturing (round) brackets is wrapped round each expression at
210 nigel 41 compile time, so alternation always happens in the context of brackets.
211 nigel 53
212 nigel 43 Non-capturing brackets use the opcode OP_BRA, while capturing brackets use
213 nigel 41 OP_BRA+1, OP_BRA+2, etc. [Note for North Americans: "bracket" to some English
214 nigel 43 speakers, including myself, can be round, square, curly, or pointy. Hence this
215     usage.]
216 nigel 41
217 nigel 53 Originally PCRE was limited to 99 capturing brackets (so as not to use up all
218     the opcodes). From release 3.5, there is no limit. What happens is that the
219     first ones, up to EXTRACT_BASIC_MAX are handled with separate opcodes, as
220     above. If there are more, the opcode is set to EXTRACT_BASIC_MAX+1, and the
221     first operation in the bracket is OP_BRANUMBER, followed by a 2-byte bracket
222     number. This opcode is ignored while matching, but is fished out when handling
223     the bracket itself. (They could have all been done like this, but I was making
224     minimal changes.)
225    
226 nigel 41 A bracket opcode is followed by two bytes which give the offset to the next
227 nigel 75 alternative OP_ALT or, if there aren't any branches, to the matching OP_KET
228 nigel 41 opcode. Each OP_ALT is followed by two bytes giving the offset to the next one,
229 nigel 75 or to the OP_KET opcode.
230 nigel 41
231     OP_KET is used for subpatterns that do not repeat indefinitely, while
232     OP_KETRMIN and OP_KETRMAX are used for indefinite repetitions, minimally or
233     maximally respectively. All three are followed by two bytes giving (as a
234 nigel 75 positive number) the offset back to the matching OP_BRA opcode.
235 nigel 41
236     If a subpattern is quantified such that it is permitted to match zero times, it
237     is preceded by one of OP_BRAZERO or OP_BRAMINZERO. These are single-byte
238     opcodes which tell the matcher that skipping this subpattern entirely is a
239     valid branch.
240    
241     A subpattern with an indefinite maximum repetition is replicated in the
242 nigel 75 compiled data its minimum number of times (or once with OP_BRAZERO if the
243     minimum is zero), with the final copy terminating with OP_KETRMIN or OP_KETRMAX
244     as appropriate.
245 nigel 41
246     A subpattern with a bounded maximum repetition is replicated in a nested
247 nigel 75 fashion up to the maximum number of times, with OP_BRAZERO or OP_BRAMINZERO
248     before each replication after the minimum, so that, for example, (abc){2,5} is
249     compiled as (abc)(abc)((abc)((abc)(abc)?)?)?.
250 nigel 41
251    
252     Assertions
253     ----------
254    
255     Forward assertions are just like other subpatterns, but starting with one of
256     the opcodes OP_ASSERT or OP_ASSERT_NOT. Backward assertions use the opcodes
257     OP_ASSERTBACK and OP_ASSERTBACK_NOT, and the first opcode inside the assertion
258     is OP_REVERSE, followed by a two byte count of the number of characters to move
259 nigel 49 back the pointer in the subject string. When operating in UTF-8 mode, the count
260     is a character count rather than a byte count. A separate count is present in
261     each alternative of a lookbehind assertion, allowing them to have different
262     fixed lengths.
263 nigel 41
264    
265     Once-only subpatterns
266     ---------------------
267    
268     These are also just like other subpatterns, but they start with the opcode
269     OP_ONCE.
270    
271    
272     Conditional subpatterns
273     -----------------------
274    
275     These are like other subpatterns, but they start with the opcode OP_COND. If
276     the condition is a back reference, this is stored at the start of the
277 nigel 53 subpattern using the opcode OP_CREF followed by two bytes containing the
278 nigel 63 reference number. If the condition is "in recursion" (coded as "(?(R)"), the
279     same scheme is used, with a "reference number" of 0xffff. Otherwise, a
280     conditional subpattern always starts with one of the assertions.
281 nigel 41
282    
283 nigel 71 Recursion
284     ---------
285    
286     Recursion either matches the current regex, or some subexpression. The opcode
287     OP_RECURSE is followed by an value which is the offset to the starting bracket
288     from the start of the whole pattern.
289    
290    
291     Callout
292     -------
293    
294 nigel 75 OP_CALLOUT is followed by one byte of data that holds a callout number in the
295     range 0 to 254 for manual callouts, or 255 for an automatic callout. In both
296     cases there follows a two-byte value giving the offset in the pattern to the
297     start of the following item, and another two-byte item giving the length of the
298     next item.
299 nigel 71
300    
301 nigel 41 Changing options
302     ----------------
303    
304 nigel 63 If any of the /i, /m, or /s options are changed within a pattern, an OP_OPT
305     opcode is compiled, followed by one byte containing the new settings of these
306     flags. If there are several alternatives, there is an occurrence of OP_OPT at
307     the start of all those following the first options change, to set appropriate
308     options for the start of the alternative. Immediately after the end of the
309     group there is another such item to reset the flags to their previous values. A
310     change of flag right at the very start of the pattern can be handled entirely
311     at compile time, and so does not cause anything to be put into the compiled
312     data.
313 nigel 41
314     Philip Hazel
315 nigel 75 September 2004

webmaster@exim.org
ViewVC Help
Powered by ViewVC 1.1.12