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 and 32-bit data strings
53 -------------------------------------------
55 From release 8.30, PCRE supports 16-bit as well as 8-bit data strings; and from
56 release 8.32, PCRE supports 32-bit data strings. The library can be compiled
57 in any combination of 8-bit, 16-bit or 32-bit modes, creating up to three
58 different libraries. In the description that follows, the word "short" is used
59 for a 16-bit data quantity, and the word "unit" is used for a quantity that is
60 a byte in 8-bit mode, a short in 16-bit mode and a 32-bit word in 32-bit mode.
61 However, so as not to over-complicate the text, the names of PCRE functions are
62 given in 8-bit form only.
65 Computing the memory requirement: how it was
66 --------------------------------------------
68 Up to and including release 6.7, PCRE worked by running a very degenerate first
69 pass to calculate a maximum store size, and then a second pass to do the real
70 compile - which might use a bit less than the predicted amount of memory. The
71 idea was that this would turn out faster than the Henry Spencer code because
72 the first pass is degenerate and the second pass can just store stuff straight
73 into the vector, which it knows is big enough.
76 Computing the memory requirement: how it is
77 -------------------------------------------
79 By the time I was working on a potential 6.8 release, the degenerate first pass
80 had become very complicated and hard to maintain. Indeed one of the early
81 things I did for 6.8 was to fix Yet Another Bug in the memory computation. Then
82 I had a flash of inspiration as to how I could run the real compile function in
83 a "fake" mode that enables it to compute how much memory it would need, while
84 actually only ever using a few hundred bytes of working memory, and without too
85 many tests of the mode that might slow it down. So I refactored the compiling
86 functions to work this way. This got rid of about 600 lines of source. It
87 should make future maintenance and development easier. As this was such a major
88 change, I never released 6.8, instead upping the number to 7.0 (other quite
89 major changes were also present in the 7.0 release).
91 A side effect of this work was that the previous limit of 200 on the nesting
92 depth of parentheses was removed. However, there is a downside: pcre_compile()
93 runs more slowly than before (30% or more, depending on the pattern) because it
94 is doing a full analysis of the pattern. My hope was that this would not be a
95 big issue, and in the event, nobody has commented on it.
97 At release 8.34, a limit on the nesting depth of parentheses was re-introduced
98 (default 250, settable at build time) so as to put a limit on the amount of
99 system stack used by pcre_compile(). This is a safety feature for environments
100 with small stacks where the patterns are provided by users.
103 Traditional matching function
104 -----------------------------
106 The "traditional", and original, matching function is called pcre_exec(), and
107 it implements an NFA algorithm, similar to the original Henry Spencer algorithm
108 and the way that Perl works. This is not surprising, since it is intended to be
109 as compatible with Perl as possible. This is the function most users of PCRE
110 will use most of the time. From release 8.20, if PCRE is compiled with
111 just-in-time (JIT) support, and studying a compiled pattern with JIT is
112 successful, the JIT code is run instead of the normal pcre_exec() code, but the
116 Supplementary matching function
117 -------------------------------
119 From PCRE 6.0, there is also a supplementary matching function called
120 pcre_dfa_exec(). This implements a DFA matching algorithm that searches
121 simultaneously for all possible matches that start at one point in the subject
122 string. (Going back to my roots: see Historical Note 1 above.) This function
123 intreprets the same compiled pattern data as pcre_exec(); however, not all the
124 facilities are available, and those that are do not always work in quite the
125 same way. See the user documentation for details.
127 The algorithm that is used for pcre_dfa_exec() is not a traditional FSM,
128 because it may have a number of states active at one time. More work would be
129 needed at compile time to produce a traditional FSM where only one state is
130 ever active at once. I believe some other regex matchers work this way. JIT
131 support is not available for this kind of matching.
137 The /i, /m, or /s options (PCRE_CASELESS, PCRE_MULTILINE, PCRE_DOTALL, and some
138 others) may change in the middle of patterns. From PCRE 8.13, their processing
139 is handled entirely at compile time by generating different opcodes for the
140 different settings. The runtime functions do not need to keep track of an
141 options state any more.
144 Format of compiled patterns
145 ---------------------------
147 The compiled form of a pattern is a vector of unsigned units (bytes in 8-bit
148 mode, shorts in 16-bit mode, 32-bit words in 32-bit mode), containing items of
149 variable length. The first unit in an item contains an opcode, and the length
150 of the item is either implicit in the opcode or contained in the data that
153 In many cases listed below, LINK_SIZE data values are specified for offsets
154 within the compiled pattern. LINK_SIZE always specifies a number of bytes. The
155 default value for LINK_SIZE is 2, but PCRE can be compiled to use 3-byte or
156 4-byte values for these offsets, although this impairs the performance. (3-byte
157 LINK_SIZE values are available only in 8-bit mode.) Specifing a LINK_SIZE
158 larger than 2 is necessary only when patterns whose compiled length is greater
159 than 64K are going to be processed. In this description, we assume the "normal"
160 compilation options. Data values that are counts (e.g. quantifiers) are two
161 bytes long in 8-bit mode (most significant byte first), or one unit in 16-bit
165 Opcodes with no following data
166 ------------------------------
168 These items are all just one unit long
170 OP_END end of pattern
171 OP_ANY match any one character other than newline
172 OP_ALLANY match any one character, including newline
173 OP_ANYBYTE match any single unit, even in UTF-8/16 mode
174 OP_SOD match start of data: \A
175 OP_SOM, start of match (subject + offset): \G
176 OP_SET_SOM, set start of match (\K)
177 OP_CIRC ^ (start of data)
178 OP_CIRCM ^ multiline mode (start of data or after newline)
179 OP_NOT_WORD_BOUNDARY \W
191 OP_EODN match end of data or newline at end: \Z
192 OP_EOD match end of data: \z
193 OP_DOLL $ (end of data, or before final newline)
194 OP_DOLLM $ multiline mode (end of data or before newline)
195 OP_EXTUNI match an extended Unicode grapheme cluster
196 OP_ANYNL match any Unicode newline sequence
199 OP_ACCEPT ) These are Perl 5.10's "backtracking control
200 OP_COMMIT ) verbs". If OP_ACCEPT is inside capturing
201 OP_FAIL ) parentheses, it may be preceded by one or more
202 OP_PRUNE ) OP_CLOSE, each followed by a count that
203 OP_SKIP ) indicates which parentheses must be closed.
206 OP_ASSERT_ACCEPT is used when (*ACCEPT) is encountered within an assertion.
207 This ends the assertion, not the entire pattern match.
210 Backtracking control verbs with optional data
211 ---------------------------------------------
213 (*THEN) without an argument generates the opcode OP_THEN and no following data.
214 OP_MARK is followed by the mark name, preceded by a one-unit length, and
215 followed by a binary zero. For (*PRUNE), (*SKIP), and (*THEN) with arguments,
216 the opcodes OP_PRUNE_ARG, OP_SKIP_ARG, and OP_THEN_ARG are used, with the name
217 following in the same format as OP_MARK.
220 Matching literal characters
221 ---------------------------
223 The OP_CHAR opcode is followed by a single character that is to be matched
224 casefully. For caseless matching, OP_CHARI is used. In UTF-8 or UTF-16 modes,
225 the character may be more than one unit long. In UTF-32 mode, characters
226 are always exactly one unit long.
228 If there is only one character in a character class, OP_CHAR or OP_CHARI is
229 used for a positive class, and OP_NOT or OP_NOTI for a negative one (that is,
230 for something like [^a]).
233 Repeating single characters
234 ---------------------------
236 The common repeats (*, +, ?), when applied to a single character, use the
237 following opcodes, which come in caseful and caseless versions:
241 OP_MINSTAR OP_MINSTARI
242 OP_POSSTAR OP_POSSTARI
244 OP_MINPLUS OP_MINPLUSI
245 OP_POSPLUS OP_POSPLUSI
247 OP_MINQUERY OP_MINQUERYI
248 OP_POSQUERY OP_POSQUERYI
250 Each opcode is followed by the character that is to be repeated. In ASCII mode,
251 these are two-unit items; in UTF-8 or UTF-16 modes, the length is variable; in
252 UTF-32 mode these are one-unit items. Those with "MIN" in their names are the
253 minimizing versions. Those with "POS" in their names are possessive versions.
254 Other repeats make use of these opcodes:
258 OP_MINUPTO OP_MINUPTOI
259 OP_POSUPTO OP_POSUPTOI
262 Each of these is followed by a count and then the repeated character. OP_UPTO
263 matches from 0 to the given number. A repeat with a non-zero minimum and a
264 fixed maximum is coded as an OP_EXACT followed by an OP_UPTO (or OP_MINUPTO or
267 Another set of matching repeating opcodes (called OP_NOTSTAR, OP_NOTSTARI,
268 etc.) are used for repeated, negated, single-character classes such as [^a]*.
269 The normal single-character opcodes (OP_STAR, etc.) are used for repeated
270 positive single-character classes.
273 Repeating character types
274 -------------------------
276 Repeats of things like \d are done exactly as for single characters, except
277 that instead of a character, the opcode for the type is stored in the data
278 unit. The opcodes are:
295 Match by Unicode property
296 -------------------------
298 OP_PROP and OP_NOTPROP are used for positive and negative matches of a
299 character by testing its Unicode property (the \p and \P escape sequences).
300 Each is followed by two units that encode the desired property as a type and a
301 value. The types are a set of #defines of the form PT_xxx, and the values are
302 enumerations of the form ucp_xx, defined in the ucp.h source file. The value is
303 relevant only for PT_GC (General Category), PT_PC (Particular Category), and
306 Repeats of these items use the OP_TYPESTAR etc. set of opcodes, followed by
307 three units: OP_PROP or OP_NOTPROP, and then the desired property type and
314 If there is only one character in a class, OP_CHAR or OP_CHARI is used for a
315 positive class, and OP_NOT or OP_NOTI for a negative one (that is, for
316 something like [^a]).
318 A set of repeating opcodes (called OP_NOTSTAR etc.) are used for repeated,
319 negated, single-character classes. The normal single-character opcodes
320 (OP_STAR, etc.) are used for repeated positive single-character classes.
322 When there is more than one character in a class, and all the code points are
323 less than 256, OP_CLASS is used for a positive class, and OP_NCLASS for a
324 negative one. In either case, the opcode is followed by a 32-byte (16-short,
325 8-word) bit map containing a 1 bit for every character that is acceptable. The
326 bits are counted from the least significant end of each unit. In caseless mode,
327 bits for both cases are set.
329 The reason for having both OP_CLASS and OP_NCLASS is so that, in UTF-8/16/32
330 mode, subject characters with values greater than 255 can be handled correctly.
331 For OP_CLASS they do not match, whereas for OP_NCLASS they do.
333 For classes containing characters with values greater than 255 or that contain
334 \p or \P, OP_XCLASS is used. It optionally uses a bit map if any code points
335 are less than 256, followed by a list of pairs (for a range) and single
336 characters. In caseless mode, both cases are explicitly listed.
338 OP_XCLASS is followed by a unit containing flag bits: XCL_NOT indicates that
339 this is a negative class, and XCL_MAP indicates that a bit map is present.
340 There follows the bit map, if XCL_MAP is set, and then a sequence of items
343 XCL_END marks the end of the list
344 XCL_SINGLE one character follows
345 XCL_RANGE two characters follow
346 XCL_PROP a Unicode property (type, value) follows
347 XCL_NOTPROP a Unicode property (type, value) follows
349 If a range starts with a code point less than 256 and ends with one greater
350 than 256, an XCL_RANGE item is used, without setting any bits in the bit map.
351 This means that if no other items in the class set bits in the map, a map is
358 OP_REF (caseful) or OP_REFI (caseless) is followed by a count containing the
359 reference number if the reference is to a unique capturing group (either by
360 number or by name). When named groups are used, there may be more than one
361 group with the same name. In this case, a reference by name generates OP_DNREF
362 or OP_DNREFI. These are followed by two counts: the index (not the byte offset)
363 in the group name table of the first entry for the requred name, followed by
364 the number of groups with the same name.
367 Repeating character classes and back references
368 -----------------------------------------------
370 Single-character classes are handled specially (see above). This section
371 applies to other classes and also to back references. In both cases, the repeat
372 information follows the base item. The matching code looks at the following
373 opcode to see if it is one of
388 All but the last three are single-unit items, with no data. The others are
389 followed by the minimum and maximum repeat counts.
392 Brackets and alternation
393 ------------------------
395 A pair of non-capturing round brackets is wrapped round each expression at
396 compile time, so alternation always happens in the context of brackets.
398 [Note for North Americans: "bracket" to some English speakers, including
399 myself, can be round, square, curly, or pointy. Hence this usage rather than
402 Non-capturing brackets use the opcode OP_BRA. Originally PCRE was limited to 99
403 capturing brackets and it used a different opcode for each one. From release
404 3.5, the limit was removed by putting the bracket number into the data for
405 higher-numbered brackets. From release 7.0 all capturing brackets are handled
406 this way, using the single opcode OP_CBRA.
408 A bracket opcode is followed by LINK_SIZE bytes which give the offset to the
409 next alternative OP_ALT or, if there aren't any branches, to the matching
410 OP_KET opcode. Each OP_ALT is followed by LINK_SIZE bytes giving the offset to
411 the next one, or to the OP_KET opcode. For capturing brackets, the bracket
412 number is a count that immediately follows the offset.
414 OP_KET is used for subpatterns that do not repeat indefinitely, and OP_KETRMIN
415 and OP_KETRMAX are used for indefinite repetitions, minimally or maximally
416 respectively (see below for possessive repetitions). All three are followed by
417 LINK_SIZE bytes giving (as a positive number) the offset back to the matching
420 If a subpattern is quantified such that it is permitted to match zero times, it
421 is preceded by one of OP_BRAZERO, OP_BRAMINZERO, or OP_SKIPZERO. These are
422 single-unit opcodes that tell the matcher that skipping the following
423 subpattern entirely is a valid branch. In the case of the first two, not
424 skipping the pattern is also valid (greedy and non-greedy). The third is used
425 when a pattern has the quantifier {0,0}. It cannot be entirely discarded,
426 because it may be called as a subroutine from elsewhere in the regex.
428 A subpattern with an indefinite maximum repetition is replicated in the
429 compiled data its minimum number of times (or once with OP_BRAZERO if the
430 minimum is zero), with the final copy terminating with OP_KETRMIN or OP_KETRMAX
433 A subpattern with a bounded maximum repetition is replicated in a nested
434 fashion up to the maximum number of times, with OP_BRAZERO or OP_BRAMINZERO
435 before each replication after the minimum, so that, for example, (abc){2,5} is
436 compiled as (abc)(abc)((abc)((abc)(abc)?)?)?, except that each bracketed group
439 When a repeated subpattern has an unbounded upper limit, it is checked to see
440 whether it could match an empty string. If this is the case, the opcode in the
441 final replication is changed to OP_SBRA or OP_SCBRA. This tells the matcher
442 that it needs to check for matching an empty string when it hits OP_KETRMIN or
443 OP_KETRMAX, and if so, to break the loop.
449 When a repeated group (capturing or non-capturing) is marked as possessive by
450 the "+" notation, e.g. (abc)++, different opcodes are used. Their names all
451 have POS on the end, e.g. OP_BRAPOS instead of OP_BRA and OP_SCPBRPOS instead
452 of OP_SCBRA. The end of such a group is marked by OP_KETRPOS. If the minimum
453 repetition is zero, the group is preceded by OP_BRAPOSZERO.
456 Once-only (atomic) groups
457 -------------------------
459 These are just like other subpatterns, but they start with the opcode
460 OP_ONCE or OP_ONCE_NC. The former is used when there are no capturing brackets
461 within the atomic group; the latter when there are. The distinction is needed
462 for when there is a backtrack to before the group - any captures within the
463 group must be reset, so it is necessary to retain backtracking points inside
464 the group even after it is complete in order to do this. When there are no
465 captures in an atomic group, all the backtracking can be discarded when it is
466 complete. This is more efficient, and also uses less stack.
468 The check for matching an empty string in an unbounded repeat is handled
469 entirely at runtime, so there are just these two opcodes for atomic groups.
475 Forward assertions are also just like other subpatterns, but starting with one
476 of the opcodes OP_ASSERT or OP_ASSERT_NOT. Backward assertions use the opcodes
477 OP_ASSERTBACK and OP_ASSERTBACK_NOT, and the first opcode inside the assertion
478 is OP_REVERSE, followed by a count of the number of characters to move back the
479 pointer in the subject string. In ASCII mode, the count is a number of units,
480 but in UTF-8/16 mode each character may occupy more than one unit; in UTF-32
481 mode each character occupies exactly one unit. A separate count is present in
482 each alternative of a lookbehind assertion, allowing them to have different
486 Conditional subpatterns
487 -----------------------
489 These are like other subpatterns, but they start with the opcode OP_COND, or
490 OP_SCOND for one that might match an empty string in an unbounded repeat. If
491 the condition is a back reference, this is stored at the start of the
492 subpattern using the opcode OP_CREF followed by a count containing the
493 reference number, provided that the reference is to a unique capturing group.
494 If the reference was by name and there is more than one group with that name,
495 OP_DNCREF is used instead. It is followed by two counts: the index in the group
496 names table, and the number of groups with the same name.
498 If the condition is "in recursion" (coded as "(?(R)"), or "in recursion of
499 group x" (coded as "(?(Rx)"), the group number is stored at the start of the
500 subpattern using the opcode OP_RREF (with a value of zero for "the whole
501 pattern") or OP_DNRREF (with data as for OP_DNCREF). For a DEFINE condition,
502 just the single unit OP_DEF is used (it has no associated data). Otherwise, a
503 conditional subpattern always starts with one of the assertions.
509 Recursion either matches the current regex, or some subexpression. The opcode
510 OP_RECURSE is followed by aLINK_SIZE value that is the offset to the starting
511 bracket from the start of the whole pattern. From release 6.5, OP_RECURSE is
512 automatically wrapped inside OP_ONCE brackets, because otherwise some patterns
513 broke it. OP_RECURSE is also used for "subroutine" calls, even though they are
514 not strictly a recursion.
520 OP_CALLOUT is followed by one unit of data that holds a callout number in the
521 range 0 to 254 for manual callouts, or 255 for an automatic callout. In both
522 cases there follows a count giving the offset in the pattern string to the
523 start of the following item, and another count giving the length of this item.
524 These values make is possible for pcretest to output useful tracing information
525 using automatic callouts.
~ian / pcre3.git / blob