1Technical Notes about PCRE
2--------------------------
3
4These are very rough technical notes that record potentially useful information
5about PCRE internals. For information about testing PCRE, see the pcretest
6documentation and the comment at the head of the RunTest file.
7
8
9Historical note 1
10-----------------
11
12Many years ago I implemented some regular expression functions to an algorithm
13suggested by Martin Richards. These were not Unix-like in form, and were quite
14restricted in what they could do by comparison with Perl. The interesting part
15about the algorithm was that the amount of space required to hold the compiled
16form of an expression was known in advance. The code to apply an expression did
17not operate by backtracking, as the original Henry Spencer code and current
18Perl code does, but instead checked all possibilities simultaneously by keeping
19a list of current states and checking all of them as it advanced through the
20subject string. In the terminology of Jeffrey Friedl's book, it was a "DFA
21algorithm", though it was not a traditional Finite State Machine (FSM). When
22the pattern was all used up, all remaining states were possible matches, and
23the one matching the longest subset of the subject string was chosen. This did
24not necessarily maximize the individual wild portions of the pattern, as is
25expected in Unix and Perl-style regular expressions.
26
27
28Historical note 2
29-----------------
30
31By contrast, the code originally written by Henry Spencer (which was
32subsequently heavily modified for Perl) compiles the expression twice: once in
33a dummy mode in order to find out how much store will be needed, and then for
34real. (The Perl version probably doesn't do this any more; I'm talking about
35the original library.) The execution function operates by backtracking and
36maximizing (or, optionally, minimizing in Perl) the amount of the subject that
37matches individual wild portions of the pattern. This is an "NFA algorithm" in
38Friedl's terminology.
39
40
41OK, here's the real stuff
42-------------------------
43
44For the set of functions that form the "basic" PCRE library (which are
45unrelated to those mentioned above), I tried at first to invent an algorithm
46that used an amount of store bounded by a multiple of the number of characters
47in the pattern, to save on compiling time. However, because of the greater
48complexity in Perl regular expressions, I couldn't do this. In any case, a
49first pass through the pattern is helpful for other reasons.
50
51
52Support for 16-bit and 32-bit data strings
53-------------------------------------------
54
55From release 8.30, PCRE supports 16-bit as well as 8-bit data strings; and from
56release 8.32, PCRE supports 32-bit data strings. The library can be compiled
57in any combination of 8-bit, 16-bit or 32-bit modes, creating up to three
58different libraries. In the description that follows, the word "short" is used
59for a 16-bit data quantity, and the word "unit" is used for a quantity that is
60a byte in 8-bit mode, a short in 16-bit mode and a 32-bit word in 32-bit mode.
61However, so as not to over-complicate the text, the names of PCRE functions are
62given in 8-bit form only.
63
64
65Computing the memory requirement: how it was
66--------------------------------------------
67
68Up to and including release 6.7, PCRE worked by running a very degenerate first
69pass to calculate a maximum store size, and then a second pass to do the real
70compile - which might use a bit less than the predicted amount of memory. The
71idea was that this would turn out faster than the Henry Spencer code because
72the first pass is degenerate and the second pass can just store stuff straight
73into the vector, which it knows is big enough.
74
75
76Computing the memory requirement: how it is
77-------------------------------------------
78
79By the time I was working on a potential 6.8 release, the degenerate first pass
80had become very complicated and hard to maintain. Indeed one of the early
81things I did for 6.8 was to fix Yet Another Bug in the memory computation. Then
82I had a flash of inspiration as to how I could run the real compile function in
83a "fake" mode that enables it to compute how much memory it would need, while
84actually only ever using a few hundred bytes of working memory, and without too
85many tests of the mode that might slow it down. So I refactored the compiling
86functions to work this way. This got rid of about 600 lines of source. It
87should make future maintenance and development easier. As this was such a major
88change, I never released 6.8, instead upping the number to 7.0 (other quite
89major changes were also present in the 7.0 release).
90
91A side effect of this work was that the previous limit of 200 on the nesting
92depth of parentheses was removed. However, there is a downside: pcre_compile()
93runs more slowly than before (30% or more, depending on the pattern) because it
94is doing a full analysis of the pattern. My hope was that this would not be a
95big issue, and in the event, nobody has commented on it.
96
97At 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
99system stack used by pcre_compile(). This is a safety feature for environments
100with small stacks where the patterns are provided by users.
101
102
103Traditional matching function
104-----------------------------
105
106The "traditional", and original, matching function is called pcre_exec(), and
107it implements an NFA algorithm, similar to the original Henry Spencer algorithm
108and the way that Perl works. This is not surprising, since it is intended to be
109as compatible with Perl as possible. This is the function most users of PCRE
110will use most of the time. From release 8.20, if PCRE is compiled with
111just-in-time (JIT) support, and studying a compiled pattern with JIT is
112successful, the JIT code is run instead of the normal pcre_exec() code, but the
113result is the same.
114
115
116Supplementary matching function
117-------------------------------
118
119From PCRE 6.0, there is also a supplementary matching function called
120pcre_dfa_exec(). This implements a DFA matching algorithm that searches
121simultaneously for all possible matches that start at one point in the subject
122string. (Going back to my roots: see Historical Note 1 above.) This function
123intreprets the same compiled pattern data as pcre_exec(); however, not all the
124facilities are available, and those that are do not always work in quite the
125same way. See the user documentation for details.
126
127The algorithm that is used for pcre_dfa_exec() is not a traditional FSM,
128because it may have a number of states active at one time. More work would be
129needed at compile time to produce a traditional FSM where only one state is
130ever active at once. I believe some other regex matchers work this way. JIT
131support is not available for this kind of matching.
132
133
134Changeable options
135------------------
136
137The /i, /m, or /s options (PCRE_CASELESS, PCRE_MULTILINE, PCRE_DOTALL, and some
138others) may change in the middle of patterns. From PCRE 8.13, their processing
139is handled entirely at compile time by generating different opcodes for the
140different settings. The runtime functions do not need to keep track of an
141options state any more.
142
143
144Format of compiled patterns
145---------------------------
146
147The compiled form of a pattern is a vector of unsigned units (bytes in 8-bit
148mode, shorts in 16-bit mode, 32-bit words in 32-bit mode), containing items of
149variable length. The first unit in an item contains an opcode, and the length
150of the item is either implicit in the opcode or contained in the data that
151follows it.
152
153In many cases listed below, LINK_SIZE data values are specified for offsets
154within the compiled pattern. LINK_SIZE always specifies a number of bytes. The
155default value for LINK_SIZE is 2, but PCRE can be compiled to use 3-byte or
1564-byte values for these offsets, although this impairs the performance. (3-byte
157LINK_SIZE values are available only in 8-bit mode.) Specifing a LINK_SIZE
158larger than 2 is necessary only when patterns whose compiled length is greater
159than 64K are going to be processed. In this description, we assume the "normal"
160compilation options. Data values that are counts (e.g. quantifiers) are two
161bytes long in 8-bit mode (most significant byte first), or one unit in 16-bit
162and 32-bit modes.
163
164
165Opcodes with no following data
166------------------------------
167
168These items are all just one unit long
169
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
180  OP_WORD_BOUNDARY       \w
181  OP_NOT_DIGIT           \D
182  OP_DIGIT               \d
183  OP_NOT_HSPACE          \H
184  OP_HSPACE              \h
185  OP_NOT_WHITESPACE      \S
186  OP_WHITESPACE          \s
187  OP_NOT_VSPACE          \V
188  OP_VSPACE              \v
189  OP_NOT_WORDCHAR        \W
190  OP_WORDCHAR            \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
197
198  OP_ASSERT_ACCEPT       )
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.
204  OP_THEN                )
205
206OP_ASSERT_ACCEPT is used when (*ACCEPT) is encountered within an assertion.
207This ends the assertion, not the entire pattern match.
208
209
210Backtracking control verbs with optional data
211---------------------------------------------
212
213(*THEN) without an argument generates the opcode OP_THEN and no following data.
214OP_MARK is followed by the mark name, preceded by a one-unit length, and
215followed by a binary zero. For (*PRUNE), (*SKIP), and (*THEN) with arguments,
216the opcodes OP_PRUNE_ARG, OP_SKIP_ARG, and OP_THEN_ARG are used, with the name
217following in the same format as OP_MARK.
218
219
220Matching literal characters
221---------------------------
222
223The OP_CHAR opcode is followed by a single character that is to be matched
224casefully. For caseless matching, OP_CHARI is used. In UTF-8 or UTF-16 modes,
225the character may be more than one unit long. In UTF-32 mode, characters
226are always exactly one unit long.
227
228If there is only one character in a character class, OP_CHAR or OP_CHARI is
229used for a positive class, and OP_NOT or OP_NOTI for a negative one (that is,
230for something like [^a]).
231
232
233Repeating single characters
234---------------------------
235
236The common repeats (*, +, ?), when applied to a single character, use the
237following opcodes, which come in caseful and caseless versions:
238
239  Caseful         Caseless
240  OP_STAR         OP_STARI
241  OP_MINSTAR      OP_MINSTARI
242  OP_POSSTAR      OP_POSSTARI
243  OP_PLUS         OP_PLUSI
244  OP_MINPLUS      OP_MINPLUSI
245  OP_POSPLUS      OP_POSPLUSI
246  OP_QUERY        OP_QUERYI
247  OP_MINQUERY     OP_MINQUERYI
248  OP_POSQUERY     OP_POSQUERYI
249
250Each opcode is followed by the character that is to be repeated. In ASCII mode,
251these are two-unit items; in UTF-8 or UTF-16 modes, the length is variable; in
252UTF-32 mode these are one-unit items. Those with "MIN" in their names are the
253minimizing versions. Those with "POS" in their names are possessive versions.
254Other repeats make use of these opcodes:
255
256  Caseful         Caseless
257  OP_UPTO         OP_UPTOI
258  OP_MINUPTO      OP_MINUPTOI
259  OP_POSUPTO      OP_POSUPTOI
260  OP_EXACT        OP_EXACTI
261
262Each of these is followed by a count and then the repeated character. OP_UPTO
263matches from 0 to the given number. A repeat with a non-zero minimum and a
264fixed maximum is coded as an OP_EXACT followed by an OP_UPTO (or OP_MINUPTO or
265OPT_POSUPTO).
266
267Another set of matching repeating opcodes (called OP_NOTSTAR, OP_NOTSTARI,
268etc.) are used for repeated, negated, single-character classes such as [^a]*.
269The normal single-character opcodes (OP_STAR, etc.) are used for repeated
270positive single-character classes.
271
272
273Repeating character types
274-------------------------
275
276Repeats of things like \d are done exactly as for single characters, except
277that instead of a character, the opcode for the type is stored in the data
278unit. The opcodes are:
279
280  OP_TYPESTAR
281  OP_TYPEMINSTAR
282  OP_TYPEPOSSTAR
283  OP_TYPEPLUS
284  OP_TYPEMINPLUS
285  OP_TYPEPOSPLUS
286  OP_TYPEQUERY
287  OP_TYPEMINQUERY
288  OP_TYPEPOSQUERY
289  OP_TYPEUPTO
290  OP_TYPEMINUPTO
291  OP_TYPEPOSUPTO
292  OP_TYPEEXACT
293
294
295Match by Unicode property
296-------------------------
297
298OP_PROP and OP_NOTPROP are used for positive and negative matches of a
299character by testing its Unicode property (the \p and \P escape sequences).
300Each is followed by two units that encode the desired property as a type and a
301value. The types are a set of #defines of the form PT_xxx, and the values are
302enumerations of the form ucp_xx, defined in the ucp.h source file. The value is
303relevant only for PT_GC (General Category), PT_PC (Particular Category), and
304PT_SC (Script).
305
306Repeats of these items use the OP_TYPESTAR etc. set of opcodes, followed by
307three units: OP_PROP or OP_NOTPROP, and then the desired property type and
308value.
309
310
311Character classes
312-----------------
313
314If there is only one character in a class, OP_CHAR or OP_CHARI is used for a
315positive class, and OP_NOT or OP_NOTI for a negative one (that is, for
316something like [^a]).
317
318A set of repeating opcodes (called OP_NOTSTAR etc.) are used for repeated,
319negated, single-character classes. The normal single-character opcodes
320(OP_STAR, etc.) are used for repeated positive single-character classes.
321
322When there is more than one character in a class, and all the code points are
323less than 256, OP_CLASS is used for a positive class, and OP_NCLASS for a
324negative one. In either case, the opcode is followed by a 32-byte (16-short,
3258-word) bit map containing a 1 bit for every character that is acceptable. The
326bits are counted from the least significant end of each unit. In caseless mode,
327bits for both cases are set.
328
329The reason for having both OP_CLASS and OP_NCLASS is so that, in UTF-8/16/32
330mode, subject characters with values greater than 255 can be handled correctly.
331For OP_CLASS they do not match, whereas for OP_NCLASS they do.
332
333For 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
335are less than 256, followed by a list of pairs (for a range) and single
336characters. In caseless mode, both cases are explicitly listed.
337
338OP_XCLASS is followed by a unit containing flag bits: XCL_NOT indicates that
339this is a negative class, and XCL_MAP indicates that a bit map is present.
340There follows the bit map, if XCL_MAP is set, and then a sequence of items
341coded as follows:
342
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
348
349If a range starts with a code point less than 256 and ends with one greater
350than 256, an XCL_RANGE item is used, without setting any bits in the bit map.
351This means that if no other items in the class set bits in the map, a map is
352not needed.
353
354
355Back references
356---------------
357
358OP_REF (caseful) or OP_REFI (caseless) is followed by a count containing the
359reference number if the reference is to a unique capturing group (either by
360number or by name). When named groups are used, there may be more than one
361group with the same name. In this case, a reference by name generates OP_DNREF
362or OP_DNREFI. These are followed by two counts: the index (not the byte offset)
363in the group name table of the first entry for the requred name, followed by
364the number of groups with the same name.
365
366
367Repeating character classes and back references
368-----------------------------------------------
369
370Single-character classes are handled specially (see above). This section
371applies to other classes and also to back references. In both cases, the repeat
372information follows the base item. The matching code looks at the following
373opcode to see if it is one of
374
375  OP_CRSTAR
376  OP_CRMINSTAR
377  OP_CRPOSSTAR
378  OP_CRPLUS
379  OP_CRMINPLUS
380  OP_CRPOSPLUS
381  OP_CRQUERY
382  OP_CRMINQUERY
383  OP_CRPOSQUERY
384  OP_CRRANGE
385  OP_CRMINRANGE
386  OP_CRPOSRANGE
387
388All but the last three are single-unit items, with no data. The others are
389followed by the minimum and maximum repeat counts.
390
391
392Brackets and alternation
393------------------------
394
395A pair of non-capturing round brackets is wrapped round each expression at
396compile time, so alternation always happens in the context of brackets.
397
398[Note for North Americans: "bracket" to some English speakers, including
399myself, can be round, square, curly, or pointy. Hence this usage rather than
400"parentheses".]
401
402Non-capturing brackets use the opcode OP_BRA. Originally PCRE was limited to 99
403capturing brackets and it used a different opcode for each one. From release
4043.5, the limit was removed by putting the bracket number into the data for
405higher-numbered brackets. From release 7.0 all capturing brackets are handled
406this way, using the single opcode OP_CBRA.
407
408A bracket opcode is followed by LINK_SIZE bytes which give the offset to the
409next alternative OP_ALT or, if there aren't any branches, to the matching
410OP_KET opcode. Each OP_ALT is followed by LINK_SIZE bytes giving the offset to
411the next one, or to the OP_KET opcode. For capturing brackets, the bracket
412number is a count that immediately follows the offset.
413
414OP_KET is used for subpatterns that do not repeat indefinitely, and OP_KETRMIN
415and OP_KETRMAX are used for indefinite repetitions, minimally or maximally
416respectively (see below for possessive repetitions). All three are followed by
417LINK_SIZE bytes giving (as a positive number) the offset back to the matching
418bracket opcode.
419
420If a subpattern is quantified such that it is permitted to match zero times, it
421is preceded by one of OP_BRAZERO, OP_BRAMINZERO, or OP_SKIPZERO. These are
422single-unit opcodes that tell the matcher that skipping the following
423subpattern entirely is a valid branch. In the case of the first two, not
424skipping the pattern is also valid (greedy and non-greedy). The third is used
425when a pattern has the quantifier {0,0}. It cannot be entirely discarded,
426because it may be called as a subroutine from elsewhere in the regex.
427
428A subpattern with an indefinite maximum repetition is replicated in the
429compiled data its minimum number of times (or once with OP_BRAZERO if the
430minimum is zero), with the final copy terminating with OP_KETRMIN or OP_KETRMAX
431as appropriate.
432
433A subpattern with a bounded maximum repetition is replicated in a nested
434fashion up to the maximum number of times, with OP_BRAZERO or OP_BRAMINZERO
435before each replication after the minimum, so that, for example, (abc){2,5} is
436compiled as (abc)(abc)((abc)((abc)(abc)?)?)?, except that each bracketed group
437has the same number.
438
439When a repeated subpattern has an unbounded upper limit, it is checked to see
440whether it could match an empty string. If this is the case, the opcode in the
441final replication is changed to OP_SBRA or OP_SCBRA. This tells the matcher
442that it needs to check for matching an empty string when it hits OP_KETRMIN or
443OP_KETRMAX, and if so, to break the loop.
444
445
446Possessive brackets
447-------------------
448
449When a repeated group (capturing or non-capturing) is marked as possessive by
450the "+" notation, e.g. (abc)++, different opcodes are used. Their names all
451have POS on the end, e.g. OP_BRAPOS instead of OP_BRA and OP_SCPBRPOS instead
452of OP_SCBRA. The end of such a group is marked by OP_KETRPOS. If the minimum
453repetition is zero, the group is preceded by OP_BRAPOSZERO.
454
455
456Once-only (atomic) groups
457-------------------------
458
459These are just like other subpatterns, but they start with the opcode
460OP_ONCE or OP_ONCE_NC. The former is used when there are no capturing brackets
461within the atomic group; the latter when there are. The distinction is needed
462for when there is a backtrack to before the group - any captures within the
463group must be reset, so it is necessary to retain backtracking points inside
464the group even after it is complete in order to do this. When there are no
465captures in an atomic group, all the backtracking can be discarded when it is
466complete. This is more efficient, and also uses less stack.
467
468The check for matching an empty string in an unbounded repeat is handled
469entirely at runtime, so there are just these two opcodes for atomic groups.
470
471
472Assertions
473----------
474
475Forward assertions are also just like other subpatterns, but starting with one
476of the opcodes OP_ASSERT or OP_ASSERT_NOT. Backward assertions use the opcodes
477OP_ASSERTBACK and OP_ASSERTBACK_NOT, and the first opcode inside the assertion
478is OP_REVERSE, followed by a count of the number of characters to move back the
479pointer in the subject string. In ASCII mode, the count is a number of units,
480but in UTF-8/16 mode each character may occupy more than one unit; in UTF-32
481mode each character occupies exactly one unit. A separate count is present in
482each alternative of a lookbehind assertion, allowing them to have different
483fixed lengths.
484
485
486Conditional subpatterns
487-----------------------
488
489These are like other subpatterns, but they start with the opcode OP_COND, or
490OP_SCOND for one that might match an empty string in an unbounded repeat. If
491the condition is a back reference, this is stored at the start of the
492subpattern using the opcode OP_CREF followed by a count containing the
493reference number, provided that the reference is to a unique capturing group.
494If the reference was by name and there is more than one group with that name,
495OP_DNCREF is used instead. It is followed by two counts: the index in the group
496names table, and the number of groups with the same name.
497
498If the condition is "in recursion" (coded as "(?(R)"), or "in recursion of
499group x" (coded as "(?(Rx)"), the group number is stored at the start of the
500subpattern using the opcode OP_RREF (with a value of zero for "the whole
501pattern") or OP_DNRREF (with data as for OP_DNCREF). For a DEFINE condition,
502just the single unit OP_DEF is used (it has no associated data). Otherwise, a
503conditional subpattern always starts with one of the assertions.
504
505
506Recursion
507---------
508
509Recursion either matches the current regex, or some subexpression. The opcode
510OP_RECURSE is followed by aLINK_SIZE value that is the offset to the starting
511bracket from the start of the whole pattern. From release 6.5, OP_RECURSE is
512automatically wrapped inside OP_ONCE brackets, because otherwise some patterns
513broke it. OP_RECURSE is also used for "subroutine" calls, even though they are
514not strictly a recursion.
515
516
517Callout
518-------
519
520OP_CALLOUT is followed by one unit of data that holds a callout number in the
521range 0 to 254 for manual callouts, or 255 for an automatic callout. In both
522cases there follows a count giving the offset in the pattern string to the
523start of the following item, and another count giving the length of this item.
524These values make is possible for pcretest to output useful tracing information
525using automatic callouts.
526
527Philip Hazel
528November 2013
529