1 Technical Notes about PCRE
2 --------------------------
4 These are very rough technical notes that record potentially useful information
5 about PCRE internals. For information about testing PCRE, see the pcretest
6 documentation and the comment at the head of the RunTest file.
12 Many years ago I implemented some regular expression functions to an algorithm
13 suggested by Martin Richards. These were not Unix-like in form, and were quite
14 restricted in what they could do by comparison with Perl. The interesting part
15 about the algorithm was that the amount of space required to hold the compiled
16 form of an expression was known in advance. The code to apply an expression did
17 not operate by backtracking, as the original Henry Spencer code and current
18 Perl code does, but instead checked all possibilities simultaneously by keeping
19 a list of current states and checking all of them as it advanced through the
20 subject string. In the terminology of Jeffrey Friedl's book, it was a "DFA
21 algorithm", though it was not a traditional Finite State Machine (FSM). When
22 the pattern was all used up, all remaining states were possible matches, and
23 the one matching the longest subset of the subject string was chosen. This did
24 not necessarily maximize the individual wild portions of the pattern, as is
25 expected in Unix and Perl-style regular expressions.
31 By contrast, the code originally written by Henry Spencer (which was
32 subsequently heavily modified for Perl) compiles the expression twice: once in
33 a dummy mode in order to find out how much store will be needed, and then for
34 real. (The Perl version probably doesn't do this any more; I'm talking about
35 the original library.) The execution function operates by backtracking and
36 maximizing (or, optionally, minimizing in Perl) the amount of the subject that
37 matches individual wild portions of the pattern. This is an "NFA algorithm" in
41 OK, here's the real stuff
42 -------------------------
44 For the set of functions that form the "basic" PCRE library (which are
45 unrelated to those mentioned above), I tried at first to invent an algorithm
46 that used an amount of store bounded by a multiple of the number of characters
47 in the pattern, to save on compiling time. However, because of the greater
48 complexity in Perl regular expressions, I couldn't do this. In any case, a
49 first pass through the pattern is helpful for other reasons.
52 Support for 16-bit data strings
53 -------------------------------
55 From release 8.30, PCRE supports 16-bit as well as 8-bit data strings, by being
56 compilable in either 8-bit or 16-bit modes, or both. Thus, two different
57 libraries can be created. In the description that follows, the word "short" is
58 used for a 16-bit data quantity, and the word "unit" is used for a quantity
59 that is a byte in 8-bit mode and a short in 16-bit mode. However, so as not to
60 over-complicate the text, the names of PCRE functions are given in 8-bit form
64 Computing the memory requirement: how it was
65 --------------------------------------------
67 Up to and including release 6.7, PCRE worked by running a very degenerate first
68 pass to calculate a maximum store size, and then a second pass to do the real
69 compile - which might use a bit less than the predicted amount of memory. The
70 idea was that this would turn out faster than the Henry Spencer code because
71 the first pass is degenerate and the second pass can just store stuff straight
72 into the vector, which it knows is big enough.
75 Computing the memory requirement: how it is
76 -------------------------------------------
78 By the time I was working on a potential 6.8 release, the degenerate first pass
79 had become very complicated and hard to maintain. Indeed one of the early
80 things I did for 6.8 was to fix Yet Another Bug in the memory computation. Then
81 I had a flash of inspiration as to how I could run the real compile function in
82 a "fake" mode that enables it to compute how much memory it would need, while
83 actually only ever using a few hundred bytes of working memory, and without too
84 many tests of the mode that might slow it down. So I refactored the compiling
85 functions to work this way. This got rid of about 600 lines of source. It
86 should make future maintenance and development easier. As this was such a major
87 change, I never released 6.8, instead upping the number to 7.0 (other quite
88 major changes were also present in the 7.0 release).
90 A side effect of this work was that the previous limit of 200 on the nesting
91 depth of parentheses was removed. However, there is a downside: pcre_compile()
92 runs more slowly than before (30% or more, depending on the pattern) because it
93 is doing a full analysis of the pattern. My hope was that this would not be a
94 big issue, and in the event, nobody has commented on it.
97 Traditional matching function
98 -----------------------------
100 The "traditional", and original, matching function is called pcre_exec(), and
101 it implements an NFA algorithm, similar to the original Henry Spencer algorithm
102 and the way that Perl works. This is not surprising, since it is intended to be
103 as compatible with Perl as possible. This is the function most users of PCRE
104 will use most of the time. From release 8.20, if PCRE is compiled with
105 just-in-time (JIT) support, and studying a compiled pattern with JIT is
106 successful, the JIT code is run instead of the normal pcre_exec() code, but the
110 Supplementary matching function
111 -------------------------------
113 From PCRE 6.0, there is also a supplementary matching function called
114 pcre_dfa_exec(). This implements a DFA matching algorithm that searches
115 simultaneously for all possible matches that start at one point in the subject
116 string. (Going back to my roots: see Historical Note 1 above.) This function
117 intreprets the same compiled pattern data as pcre_exec(); however, not all the
118 facilities are available, and those that are do not always work in quite the
119 same way. See the user documentation for details.
121 The algorithm that is used for pcre_dfa_exec() is not a traditional FSM,
122 because it may have a number of states active at one time. More work would be
123 needed at compile time to produce a traditional FSM where only one state is
124 ever active at once. I believe some other regex matchers work this way.
130 The /i, /m, or /s options (PCRE_CASELESS, PCRE_MULTILINE, PCRE_DOTALL) may
131 change in the middle of patterns. From PCRE 8.13, their processing is handled
132 entirely at compile time by generating different opcodes for the different
133 settings. The runtime functions do not need to keep track of an options state
137 Format of compiled patterns
138 ---------------------------
140 The compiled form of a pattern is a vector of units (bytes in 8-bit mode, or
141 shorts in 16-bit mode), containing items of variable length. The first unit in
142 an item contains an opcode, and the length of the item is either implicit in
143 the opcode or contained in the data that follows it.
145 In many cases listed below, LINK_SIZE data values are specified for offsets
146 within the compiled pattern. LINK_SIZE always specifies a number of bytes. The
147 default value for LINK_SIZE is 2, but PCRE can be compiled to use 3-byte or
148 4-byte values for these offsets, although this impairs the performance. (3-byte
149 LINK_SIZE values are available only in 8-bit mode.) Specifing a LINK_SIZE
150 larger than 2 is necessary only when patterns whose compiled length is greater
151 than 64K are going to be processed. In this description, we assume the "normal"
152 compilation options. Data values that are counts (e.g. for quantifiers) are
153 always just two bytes long (one short in 16-bit mode).
155 Opcodes with no following data
156 ------------------------------
158 These items are all just one unit long
160 OP_END end of pattern
161 OP_ANY match any one character other than newline
162 OP_ALLANY match any one character, including newline
163 OP_ANYBYTE match any single byte, even in UTF-8 mode
164 OP_SOD match start of data: \A
165 OP_SOM, start of match (subject + offset): \G
166 OP_SET_SOM, set start of match (\K)
167 OP_CIRC ^ (start of data)
168 OP_CIRCM ^ multiline mode (start of data or after newline)
169 OP_NOT_WORD_BOUNDARY \W
181 OP_EODN match end of data or \n at end: \Z
182 OP_EOD match end of data: \z
183 OP_DOLL $ (end of data, or before final newline)
184 OP_DOLLM $ multiline mode (end of data or before newline)
185 OP_EXTUNI match an extended Unicode character
186 OP_ANYNL match any Unicode newline sequence
188 OP_ACCEPT ) These are Perl 5.10's "backtracking control
189 OP_COMMIT ) verbs". If OP_ACCEPT is inside capturing
190 OP_FAIL ) parentheses, it may be preceded by one or more
191 OP_PRUNE ) OP_CLOSE, followed by a 2-byte number,
192 OP_SKIP ) indicating which parentheses must be closed.
195 Backtracking control verbs with (optional) data
196 -----------------------------------------------
198 (*THEN) without an argument generates the opcode OP_THEN and no following data.
199 OP_MARK is followed by the mark name, preceded by a one-unit length, and
200 followed by a binary zero. For (*PRUNE), (*SKIP), and (*THEN) with arguments,
201 the opcodes OP_PRUNE_ARG, OP_SKIP_ARG, and OP_THEN_ARG are used, with the name
202 following in the same format.
205 Matching literal characters
206 ---------------------------
208 The OP_CHAR opcode is followed by a single character that is to be matched
209 casefully. For caseless matching, OP_CHARI is used. In UTF-8 or UTF-16 modes,
210 the character may be more than one unit long.
213 Repeating single characters
214 ---------------------------
216 The common repeats (*, +, ?), when applied to a single character, use the
217 following opcodes, which come in caseful and caseless versions:
221 OP_MINSTAR OP_MINSTARI
222 OP_POSSTAR OP_POSSTARI
224 OP_MINPLUS OP_MINPLUSI
225 OP_POSPLUS OP_POSPLUSI
227 OP_MINQUERY OP_MINQUERYI
228 OP_POSQUERY OP_POSQUERYI
230 Each opcode is followed by the character that is to be repeated. In ASCII mode,
231 these are two-unit items; in UTF-8 or UTF-16 modes, the length is variable.
232 Those with "MIN" in their names are the minimizing versions. Those with "POS"
233 in their names are possessive versions. Other repeats make use of these
238 OP_MINUPTO OP_MINUPTOI
239 OP_POSUPTO OP_POSUPTOI
242 Each of these is followed by a two-byte (one short) count (most significant
243 byte first in 8-bit mode) and then the repeated character. OP_UPTO matches from
244 0 to the given number. A repeat with a non-zero minimum and a fixed maximum is
245 coded as an OP_EXACT followed by an OP_UPTO (or OP_MINUPTO or OPT_POSUPTO).
248 Repeating character types
249 -------------------------
251 Repeats of things like \d are done exactly as for single characters, except
252 that instead of a character, the opcode for the type is stored in the data
253 unit. The opcodes are:
270 Match by Unicode property
271 -------------------------
273 OP_PROP and OP_NOTPROP are used for positive and negative matches of a
274 character by testing its Unicode property (the \p and \P escape sequences).
275 Each is followed by two units that encode the desired property as a type and a
278 Repeats of these items use the OP_TYPESTAR etc. set of opcodes, followed by
279 three units: OP_PROP or OP_NOTPROP, and then the desired property type and
286 If there is only one character in the class, OP_CHAR or OP_CHARI is used for a
287 positive class, and OP_NOT or OP_NOTI for a negative one (that is, for
288 something like [^a]). However, OP_NOT[I] can be used only with single-unit
289 characters, so in UTF-8 (UTF-16) mode, the use of OP_NOT[I] applies only to
290 characters whose code points are no greater than 127 (0xffff).
292 Another set of 13 repeating opcodes (called OP_NOTSTAR etc.) are used for
293 repeated, negated, single-character classes. The normal single-character
294 opcodes (OP_STAR, etc.) are used for repeated positive single-character
297 When there is more than one character in a class and all the characters are
298 less than 256, OP_CLASS is used for a positive class, and OP_NCLASS for a
299 negative one. In either case, the opcode is followed by a 32-byte (16-short)
300 bit map containing a 1 bit for every character that is acceptable. The bits are
301 counted from the least significant end of each unit. In caseless mode, bits for
304 The reason for having both OP_CLASS and OP_NCLASS is so that, in UTF-8/16 mode,
305 subject characters with values greater than 255 can be handled correctly. For
306 OP_CLASS they do not match, whereas for OP_NCLASS they do.
308 For classes containing characters with values greater than 255, OP_XCLASS is
309 used. It optionally uses a bit map (if any characters lie within it), followed
310 by a list of pairs (for a range) and single characters. In caseless mode, both
311 cases are explicitly listed. There is a flag character than indicates whether
312 it is a positive or a negative class.
318 OP_REF (caseful) or OP_REFI (caseless) is followed by two bytes (one short)
319 containing the reference number.
322 Repeating character classes and back references
323 -----------------------------------------------
325 Single-character classes are handled specially (see above). This section
326 applies to OP_CLASS and OP_REF[I]. In both cases, the repeat information
327 follows the base item. The matching code looks at the following opcode to see
339 All but the last two are just single-unit items. The others are followed by
340 four bytes (two shorts) of data, comprising the minimum and maximum repeat
341 counts. There are no special possessive opcodes for these repeats; a possessive
342 repeat is compiled into an atomic group.
345 Brackets and alternation
346 ------------------------
348 A pair of non-capturing (round) brackets is wrapped round each expression at
349 compile time, so alternation always happens in the context of brackets.
351 [Note for North Americans: "bracket" to some English speakers, including
352 myself, can be round, square, curly, or pointy. Hence this usage rather than
355 Non-capturing brackets use the opcode OP_BRA. Originally PCRE was limited to 99
356 capturing brackets and it used a different opcode for each one. From release
357 3.5, the limit was removed by putting the bracket number into the data for
358 higher-numbered brackets. From release 7.0 all capturing brackets are handled
359 this way, using the single opcode OP_CBRA.
361 A bracket opcode is followed by LINK_SIZE bytes which give the offset to the
362 next alternative OP_ALT or, if there aren't any branches, to the matching
363 OP_KET opcode. Each OP_ALT is followed by LINK_SIZE bytes giving the offset to
364 the next one, or to the OP_KET opcode. For capturing brackets, the bracket
365 number immediately follows the offset, always as a 2-byte (one short) item.
367 OP_KET is used for subpatterns that do not repeat indefinitely, and
368 OP_KETRMIN and OP_KETRMAX are used for indefinite repetitions, minimally or
369 maximally respectively (see below for possessive repetitions). All three are
370 followed by LINK_SIZE bytes giving (as a positive number) the offset back to
371 the matching bracket opcode.
373 If a subpattern is quantified such that it is permitted to match zero times, it
374 is preceded by one of OP_BRAZERO, OP_BRAMINZERO, or OP_SKIPZERO. These are
375 single-unit opcodes that tell the matcher that skipping the following
376 subpattern entirely is a valid branch. In the case of the first two, not
377 skipping the pattern is also valid (greedy and non-greedy). The third is used
378 when a pattern has the quantifier {0,0}. It cannot be entirely discarded,
379 because it may be called as a subroutine from elsewhere in the regex.
381 A subpattern with an indefinite maximum repetition is replicated in the
382 compiled data its minimum number of times (or once with OP_BRAZERO if the
383 minimum is zero), with the final copy terminating with OP_KETRMIN or OP_KETRMAX
386 A subpattern with a bounded maximum repetition is replicated in a nested
387 fashion up to the maximum number of times, with OP_BRAZERO or OP_BRAMINZERO
388 before each replication after the minimum, so that, for example, (abc){2,5} is
389 compiled as (abc)(abc)((abc)((abc)(abc)?)?)?, except that each bracketed group
392 When a repeated subpattern has an unbounded upper limit, it is checked to see
393 whether it could match an empty string. If this is the case, the opcode in the
394 final replication is changed to OP_SBRA or OP_SCBRA. This tells the matcher
395 that it needs to check for matching an empty string when it hits OP_KETRMIN or
396 OP_KETRMAX, and if so, to break the loop.
401 When a repeated group (capturing or non-capturing) is marked as possessive by
402 the "+" notation, e.g. (abc)++, different opcodes are used. Their names all
403 have POS on the end, e.g. OP_BRAPOS instead of OP_BRA and OP_SCPBRPOS instead
404 of OP_SCBRA. The end of such a group is marked by OP_KETRPOS. If the minimum
405 repetition is zero, the group is preceded by OP_BRAPOSZERO.
411 Forward assertions are just like other subpatterns, but starting with one of
412 the opcodes OP_ASSERT or OP_ASSERT_NOT. Backward assertions use the opcodes
413 OP_ASSERTBACK and OP_ASSERTBACK_NOT, and the first opcode inside the assertion
414 is OP_REVERSE, followed by a two byte (one short) count of the number of
415 characters to move back the pointer in the subject string. In ASCII mode, the
416 count is a number of units, but in UTF-8/16 mode each character may occupy more
417 than one unit. A separate count is present in each alternative of a lookbehind
418 assertion, allowing them to have different fixed lengths.
421 Once-only (atomic) subpatterns
422 ------------------------------
424 These are also just like other subpatterns, but they start with the opcode
425 OP_ONCE. The check for matching an empty string in an unbounded repeat is
426 handled entirely at runtime, so there is just this one opcode.
429 Conditional subpatterns
430 -----------------------
432 These are like other subpatterns, but they start with the opcode OP_COND, or
433 OP_SCOND for one that might match an empty string in an unbounded repeat. If
434 the condition is a back reference, this is stored at the start of the
435 subpattern using the opcode OP_CREF followed by two bytes (one short)
436 containing the reference number. OP_NCREF is used instead if the reference was
437 generated by name (so that the runtime code knows to check for duplicate
440 If the condition is "in recursion" (coded as "(?(R)"), or "in recursion of
441 group x" (coded as "(?(Rx)"), the group number is stored at the start of the
442 subpattern using the opcode OP_RREF or OP_NRREF (cf OP_NCREF), and a value of
443 zero for "the whole pattern". For a DEFINE condition, just the single unit
444 OP_DEF is used (it has no associated data). Otherwise, a conditional subpattern
445 always starts with one of the assertions.
451 Recursion either matches the current regex, or some subexpression. The opcode
452 OP_RECURSE is followed by an value which is the offset to the starting bracket
453 from the start of the whole pattern. From release 6.5, OP_RECURSE is
454 automatically wrapped inside OP_ONCE brackets (because otherwise some patterns
455 broke it). OP_RECURSE is also used for "subroutine" calls, even though they
456 are not strictly a recursion.
462 OP_CALLOUT is followed by one unit of data that holds a callout number in the
463 range 0 to 254 for manual callouts, or 255 for an automatic callout. In both
464 cases there follows a two-byte (one short) value giving the offset in the
465 pattern to the start of the following item, and another two-byte (one short)
466 item giving the length of the next item.
~matthewv / pcre3.git / blob