3 perlreguts - Description of the Perl regular expression engine.
7 This document is an attempt to shine some light on the guts of the regex
8 engine and how it works. The regex engine represents a significant chunk
9 of the perl codebase, but is relatively poorly understood. This document
10 is a meagre attempt at addressing this situation. It is derived from the
11 author's experience, comments in the source code, other papers on the
12 regex engine, feedback on the perl5-porters mail list, and no doubt other
15 B<WARNING!> It should be clearly understood that this document represents
16 the state of the regex engine as the author understands it at the time of
17 writing. Unless stated otherwise it is B<NOT> an API definition; it is
18 purely an internals guide for those who want to hack the regex engine, or
19 understand how the regex engine works. Readers of this document are
20 expected to understand perl's regex syntax and its usage in detail. If you
21 want to learn about the basics of Perl's regular expressions, see
26 =head2 A quick note on terms
28 There is some debate as to whether to say "regexp" or "regex". In this
29 document we will use the term "regex" unless there is a special reason
30 not to, in which case we will explain why.
32 When speaking about regexes we need to distinguish between their source
33 code form and their internal form. In this document we will use the term
34 "pattern" when we speak of their textual, source code form, the term
35 "program" when we speak of their internal representation. These
36 correspond to the terms I<S-regex> and I<B-regex> that Mark Jason
37 Dominus employs in his paper on "Rx" ([1] in L</REFERENCES>).
39 =head2 What is a regular expression engine?
41 A regular expression engine is a program that takes a set of constraints
42 specified in a mini-language, and then applies those constraints to a
43 target string, and determines whether or not the string satisfies the
44 constraints. See L<perlre> for a full definition of the language.
46 So in less grandiose terms the first part of the job is to turn a pattern into
47 something the computer can efficiently use to find the matching point in
48 the string, and the second part is performing the search itself.
50 To do this we need to produce a program by parsing the text. We then
51 need to execute the program to find the point in the string that
52 matches. And we need to do the whole thing efficiently.
54 =head2 Structure of a Regexp Program
58 Although it is a bit confusing and some people object to the terminology, it
59 is worth taking a look at a comment that has
60 been in F<regexp.h> for years:
62 I<This is essentially a linear encoding of a nondeterministic
63 finite-state machine (aka syntax charts or "railroad normal form" in
66 The term "railroad normal form" is a bit esoteric, with "syntax
67 diagram/charts", or "railroad diagram/charts" being more common terms.
68 Nevertheless it provides a useful mental image of a regex program: each
69 node can be thought of as a unit of track, with a single entry and in
70 most cases a single exit point (there are pieces of track that fork, but
71 statistically not many), and the whole forms a layout with a
72 single entry and single exit point. The matching process can be thought
73 of as a car that moves along the track, with the particular route through
74 the system being determined by the character read at each possible
75 connector point. A car can fall off the track at any point but it may
76 only proceed as long as it matches the track.
78 Thus the pattern C</foo(?:\w+|\d+|\s+)bar/> can be thought of as the
95 The truth of the matter is that perl's regular expressions these days are
96 much more complex than this kind of structure, but visualising it this way
97 can help when trying to get your bearings, and it matches the
98 current implementation pretty closely.
100 To be more precise, we will say that a regex program is an encoding
101 of a graph. Each node in the graph corresponds to part of
102 the original regex pattern, such as a literal string or a branch,
103 and has a pointer to the nodes representing the next component
104 to be matched. Since "node" and "opcode" already have other meanings in the
105 perl source, we will call the nodes in a regex program "regops".
107 The program is represented by an array of C<regnode> structures, one or
108 more of which represent a single regop of the program. Struct
109 C<regnode> is the smallest struct needed, and has a field structure which is
110 shared with all the other larger structures.
112 The "next" pointers of all regops except C<BRANCH> implement concatenation;
113 a "next" pointer with a C<BRANCH> on both ends of it is connecting two
114 alternatives. [Here we have one of the subtle syntax dependencies: an
115 individual C<BRANCH> (as opposed to a collection of them) is never
116 concatenated with anything because of operator precedence.]
118 The operand of some types of regop is a literal string; for others,
119 it is a regop leading into a sub-program. In particular, the operand
120 of a C<BRANCH> node is the first regop of the branch.
122 B<NOTE>: As the railroad metaphor suggests, this is B<not> a tree
123 structure: the tail of the branch connects to the thing following the
124 set of C<BRANCH>es. It is a like a single line of railway track that
125 splits as it goes into a station or railway yard and rejoins as it comes
130 The base structure of a regop is defined in F<regexp.h> as follows:
133 U8 flags; /* Various purposes, sometimes overridden */
134 U8 type; /* Opcode value as specified by regnodes.h */
135 U16 next_off; /* Offset in size regnode */
138 Other larger C<regnode>-like structures are defined in F<regcomp.h>. They
139 are almost like subclasses in that they have the same fields as
140 C<regnode>, with possibly additional fields following in
141 the structure, and in some cases the specific meaning (and name)
142 of some of base fields are overridden. The following is a more
143 complete description.
151 C<regnode_1> structures have the same header, followed by a single
152 four-byte argument; C<regnode_2> structures contain two two-byte
156 regnode_2 U16 arg1; U16 arg2;
158 =item C<regnode_string>
160 C<regnode_string> structures, used for literal strings, follow the header
161 with a one-byte length and then the string data. Strings are padded on
162 the end with zero bytes so that the total length of the node is a
163 multiple of four bytes:
165 regnode_string char string[1];
166 U8 str_len; /* overrides flags */
168 =item C<regnode_charclass>
170 Character classes are represented by C<regnode_charclass> structures,
171 which have a four-byte argument and then a 32-byte (256-bit) bitmap
172 indicating which characters are included in the class.
174 regnode_charclass U32 arg1;
175 char bitmap[ANYOF_BITMAP_SIZE];
177 =item C<regnode_charclass_class>
179 There is also a larger form of a char class structure used to represent
180 POSIX char classes called C<regnode_charclass_class> which has an
181 additional 4-byte (32-bit) bitmap indicating which POSIX char class
184 regnode_charclass_class U32 arg1;
185 char bitmap[ANYOF_BITMAP_SIZE];
186 char classflags[ANYOF_CLASSBITMAP_SIZE];
190 F<regnodes.h> defines an array called C<regarglen[]> which gives the size
191 of each opcode in units of C<size regnode> (4-byte). A macro is used
192 to calculate the size of an C<EXACT> node based on its C<str_len> field.
194 The regops are defined in F<regnodes.h> which is generated from
195 F<regcomp.sym> by F<regcomp.pl>. Currently the maximum possible number
196 of distinct regops is restricted to 256, with about a quarter already
199 A set of macros makes accessing the fields
200 easier and more consistent. These include C<OP()>, which is used to determine
201 the type of a C<regnode>-like structure; C<NEXT_OFF()>, which is the offset to
202 the next node (more on this later); C<ARG()>, C<ARG1()>, C<ARG2()>, C<ARG_SET()>,
203 and equivalents for reading and setting the arguments; and C<STR_LEN()>,
204 C<STRING()> and C<OPERAND()> for manipulating strings and regop bearing
207 =head3 What regop is next?
209 There are three distinct concepts of "next" in the regex engine, and
210 it is important to keep them clear.
216 There is the "next regnode" from a given regnode, a value which is
217 rarely useful except that sometimes it matches up in terms of value
218 with one of the others, and that sometimes the code assumes this to
223 There is the "next regop" from a given regop/regnode. This is the
224 regop physically located after the the current one, as determined by
225 the size of the current regop. This is often useful, such as when
226 dumping the structure we use this order to traverse. Sometimes the code
227 assumes that the "next regnode" is the same as the "next regop", or in
228 other words assumes that the sizeof a given regop type is always going
229 to be one regnode large.
233 There is the "regnext" from a given regop. This is the regop which
234 is reached by jumping forward by the value of C<NEXT_OFF()>,
235 or in a few cases for longer jumps by the C<arg1> field of the C<regnode_1>
236 structure. The subroutine C<regnext()> handles this transparently.
237 This is the logical successor of the node, which in some cases, like
238 that of the C<BRANCH> regop, has special meaning.
242 =head1 Process Overview
244 Broadly speaking, performing a match of a string against a pattern
245 involves the following steps:
253 =item 1. Parsing for size
255 =item 2. Parsing for construction
257 =item 3. Peep-hole optimisation and analysis
265 =item 4. Start position and no-match optimisations
267 =item 5. Program execution
274 Where these steps occur in the actual execution of a perl program is
275 determined by whether the pattern involves interpolating any string
276 variables. If interpolation occurs, then compilation happens at run time. If it
277 does not, then compilation is performed at compile time. (The C</o> modifier changes this,
278 as does C<qr//> to a certain extent.) The engine doesn't really care that
283 This code resides primarily in F<regcomp.c>, along with the header files
284 F<regcomp.h>, F<regexp.h> and F<regnodes.h>.
286 Compilation starts with C<pregcomp()>, which is mostly an initialisation
287 wrapper which farms work out to two other routines for the heavy lifting: the
288 first is C<reg()>, which is the start point for parsing; the second,
289 C<study_chunk()>, is responsible for optimisation.
291 Initialisation in C<pregcomp()> mostly involves the creation and data-filling
292 of a special structure, C<RExC_state_t> (defined in F<regcomp.c>).
293 Almost all internally-used routines in F<regcomp.h> take a pointer to one
294 of these structures as their first argument, with the name C<pRExC_state>.
295 This structure is used to store the compilation state and contains many
296 fields. Likewise there are many macros which operate on this
297 variable: anything that looks like C<RExC_xxxx> is a macro that operates on
298 this pointer/structure.
300 =head3 Parsing for size
302 In this pass the input pattern is parsed in order to calculate how much
303 space is needed for each regop we would need to emit. The size is also
304 used to determine whether long jumps will be required in the program.
306 This stage is controlled by the macro C<SIZE_ONLY> being set.
308 The parse proceeds pretty much exactly as it does during the
309 construction phase, except that most routines are short-circuited to
310 change the size field C<RExC_size> and not do anything else.
312 =head3 Parsing for construction
314 Once the size of the program has been determined, the pattern is parsed
315 again, but this time for real. Now C<SIZE_ONLY> will be false, and the
316 actual construction can occur.
318 C<reg()> is the start of the parse process. It is responsible for
319 parsing an arbitrary chunk of pattern up to either the end of the
320 string, or the first closing parenthesis it encounters in the pattern.
321 This means it can be used to parse the top-level regex, or any section
322 inside of a grouping parenthesis. It also handles the "special parens"
323 that perl's regexes have. For instance when parsing C</x(?:foo)y/> C<reg()>
324 will at one point be called to parse from the "?" symbol up to and
327 Additionally, C<reg()> is responsible for parsing the one or more
328 branches from the pattern, and for "finishing them off" by correctly
329 setting their next pointers. In order to do the parsing, it repeatedly
330 calls out to C<regbranch()>, which is responsible for handling up to the
331 first C<|> symbol it sees.
333 C<regbranch()> in turn calls C<regpiece()> which
334 handles "things" followed by a quantifier. In order to parse the
335 "things", C<regatom()> is called. This is the lowest level routine which
336 parses out constant strings, character classes, and the
337 various special symbols like C<$>. If C<regatom()> encounters a "("
338 character it in turn calls C<reg()>.
340 The routine C<regtail()> is called by both C<reg()>, C<regbranch()>
341 in order to "set the tail pointer" correctly. When executing and
342 we get to the end of a branch, we need to go to the node following the
343 grouping parens. When parsing, however, we don't know where the end will
344 be until we get there, so when we do we must go back and update the
345 offsets as appropriate. C<regtail> is used to make this easier.
347 A subtlety of the parsing process means that a regex like C</foo/> is
348 originally parsed into an alternation with a single branch. It is only
349 afterwards that the optimiser converts single branch alternations into the
352 =head3 Parse Call Graph and a Grammar
354 The call graph looks like this:
356 reg() # parse a top level regex, or inside of parens
357 regbranch() # parse a single branch of an alternation
358 regpiece() # parse a pattern followed by a quantifier
359 regatom() # parse a simple pattern
360 regclass() # used to handle a class
361 reg() # used to handle a parenthesised subpattern
364 regtail() # finish off the branch
366 regtail() # finish off the branch sequence. Tie each
367 # branch's tail to the tail of the sequence
368 # (NEW) In Debug mode this is
371 A grammar form might be something like this:
373 atom : constant | class
374 quant : '*' | '+' | '?' | '{min,max}'
380 group : '(' branch ')'
387 In the 5.9.x development version of perl you can C<< use re Debug => 'PARSE'; >> to see some trace
388 information about the parse process. We will start with some simple
389 patterns and build up to more complex patterns.
391 So when we parse C</foo/> we see something like the following table. The
392 left shows what is being parsed, and the number indicates where the next regop
393 would go. The stuff on the right is the trace output of the graph. The
394 names are chosen to be short to make it less dense on the screen. 'tsdy'
395 is a special form of C<regtail()> which does some extra analysis.
401 >< 4 tsdy~ EXACT <foo> (EXACT) (1)
402 ~ attach to END (3) offset to 2
404 The resulting program then looks like:
409 As you can see, even though we parsed out a branch and a piece, it was ultimately
410 only an atom. The final program shows us how things work. We have an C<EXACT> regop,
411 followed by an C<END> regop. The number in parens indicates where the C<regnext> of
412 the node goes. The C<regnext> of an C<END> regop is unused, as C<END> regops mean
413 we have successfully matched. The number on the left indicates the position of
414 the regop in the regnode array.
416 Now let's try a harder pattern. We will add a quantifier, so now we have the pattern
417 C</foo+/>. We will see that C<regbranch()> calls C<regpiece()> twice.
425 >< 6 tail~ EXACT <fo> (1)
426 7 tsdy~ EXACT <fo> (EXACT) (1)
428 ~ attach to END (6) offset to 3
430 And we end up with the program:
437 Now we have a special case. The C<EXACT> regop has a C<regnext> of 0. This is
438 because if it matches it should try to match itself again. The C<PLUS> regop
439 handles the actual failure of the C<EXACT> regop and acts appropriately (going
440 to regnode 6 if the C<EXACT> matched at least once, or failing if it didn't).
442 Now for something much more complex: C</x(?:foo*|b[a][rR])(foo|bar)$/>
448 >(?:foo*|b[... 3 piec
454 >o*|b[a][rR... 5 piec
456 >|b[a][rR])... 8 tail~ EXACT <fo> (3)
457 >b[a][rR])(... 9 brnc
460 >[a][rR])(f... 12 piec
463 >[rR])(foo|... 14 tail~ EXACT <b> (10)
467 >)(foo|bar)... 25 tail~ EXACT <a> (12)
469 26 tsdy~ BRANCH (END) (9)
470 ~ attach to TAIL (25) offset to 16
471 tsdy~ EXACT <fo> (EXACT) (4)
473 ~ attach to TAIL (25) offset to 19
474 tsdy~ EXACT <b> (EXACT) (10)
475 ~ EXACT <a> (EXACT) (12)
476 ~ ANYOF[Rr] (END) (14)
477 ~ attach to TAIL (25) offset to 11
478 >(foo|bar)$< tail~ EXACT <x> (1)
485 >|bar)$< 31 tail~ OPEN1 (26)
489 >)$< 34 tail~ BRANCH (28)
490 36 tsdy~ BRANCH (END) (31)
491 ~ attach to CLOSE1 (34) offset to 3
492 tsdy~ EXACT <foo> (EXACT) (29)
493 ~ attach to CLOSE1 (34) offset to 5
494 tsdy~ EXACT <bar> (EXACT) (32)
495 ~ attach to CLOSE1 (34) offset to 2
501 >< 37 tail~ OPEN1 (26)
505 38 tsdy~ EXACT <x> (EXACT) (1)
514 ~ attach to END (37) offset to 1
516 Resulting in the program
525 12: OPTIMIZED (2 nodes)
530 [StS:1 Wds:2 Cs:6 Uq:5 #Sts:7 Mn:3 Mx:3 Stcls:bf]
533 30: OPTIMIZED (4 nodes)
538 Here we can see a much more complex program, with various optimisations in
539 play. At regnode 10 we see an example where a character class with only
540 one character in it was turned into an C<EXACT> node. We can also see where
541 an entire alternation was turned into a C<TRIE-EXACT> node. As a consequence,
542 some of the regnodes have been marked as optimised away. We can see that
543 the C<$> symbol has been converted into an C<EOL> regop, a special piece of
544 code that looks for C<\n> or the end of the string.
546 The next pointer for C<BRANCH>es is interesting in that it points at where
547 execution should go if the branch fails. When executing if the engine
548 tries to traverse from a branch to a C<regnext> that isn't a branch then
549 the engine will know that the entire set of branches have failed.
551 =head3 Peep-hole Optimisation and Analysis
553 The regular expression engine can be a weighty tool to wield. On long
554 strings and complex patterns it can end up having to do a lot of work
555 to find a match, and even more to decide that no match is possible.
556 Consider a situation like the following pattern.
558 'ababababababababababab' =~ /(a|b)*z/
560 The C<(a|b)*> part can match at every char in the string, and then fail
561 every time because there is no C<z> in the string. So obviously we can
562 avoid using the regex engine unless there is a C<z> in the string.
563 Likewise in a pattern like:
567 In this case we know that the string must contain a C<foo> which must be
568 followed by C<bar>. We can use Fast Boyer-Moore matching as implemented
569 in C<fbm_instr()> to find the location of these strings. If they don't exist
570 then we don't need to resort to the much more expensive regex engine.
571 Even better, if they do exist then we can use their positions to
572 reduce the search space that the regex engine needs to cover to determine
573 if the entire pattern matches.
575 There are various aspects of the pattern that can be used to facilitate
576 optimisations along these lines:
580 =item * anchored fixed strings
582 =item * floating fixed strings
584 =item * minimum and maximum length requirements
588 =item * Beginning/End of line positions
592 Another form of optimisation that can occur is post-parse "peep-hole"
593 optimisations, where inefficient constructs are replaced by
594 more efficient constructs. An example of this are C<TAIL> regops which are used
595 during parsing to mark the end of branches and the end of groups. These
596 regops are used as place-holders during construction and "always match"
597 so they can be "optimised away" by making the things that point to the
598 C<TAIL> point to thing that the C<TAIL> points to, thus "skipping" the node.
600 Another optimisation that can occur is that of "C<EXACT> merging" which is
601 where two consecutive C<EXACT> nodes are merged into a single
602 regop. An even more aggressive form of this is that a branch
603 sequence of the form C<EXACT BRANCH ... EXACT> can be converted into a
606 All of this occurs in the routine C<study_chunk()> which uses a special
607 structure C<scan_data_t> to store the analysis that it has performed, and
608 does the "peep-hole" optimisations as it goes.
610 The code involved in C<study_chunk()> is extremely cryptic. Be careful. :-)
614 Execution of a regex generally involves two phases, the first being
615 finding the start point in the string where we should match from,
616 and the second being running the regop interpreter.
618 If we can tell that there is no valid start point then we don't bother running
619 interpreter at all. Likewise, if we know from the analysis phase that we
620 cannot detect a short-cut to the start position, we go straight to the
623 The two entry points are C<re_intuit_start()> and C<pregexec()>. These routines
624 have a somewhat incestuous relationship with overlap between their functions,
625 and C<pregexec()> may even call C<re_intuit_start()> on its own. Nevertheless
626 other parts of the the perl source code may call into either, or both.
628 Execution of the interpreter itself used to be recursive. Due to the
629 efforts of Dave Mitchell in the 5.9.x development track, it is now iterative. Now an
630 internal stack is maintained on the heap and the routine is fully
631 iterative. This can make it tricky as the code is quite conservative
632 about what state it stores, with the result that that two consecutive lines in the
633 code can actually be running in totally different contexts due to the
636 =head3 Start position and no-match optimisations
638 C<re_intuit_start()> is responsible for handling start points and no-match
639 optimisations as determined by the results of the analysis done by
640 C<study_chunk()> (and described in L<Peep-hole Optimisation and Analysis>).
642 The basic structure of this routine is to try to find the start- and/or
643 end-points of where the pattern could match, and to ensure that the string
644 is long enough to match the pattern. It tries to use more efficient
645 methods over less efficient methods and may involve considerable
646 cross-checking of constraints to find the place in the string that matches.
647 For instance it may try to determine that a given fixed string must be
648 not only present but a certain number of chars before the end of the
651 It calls several other routines, such as C<fbm_instr()> which does
652 Fast Boyer Moore matching and C<find_byclass()> which is responsible for
653 finding the start using the first mandatory regop in the program.
655 When the optimisation criteria have been satisfied, C<reg_try()> is called
656 to perform the match.
658 =head3 Program execution
660 C<pregexec()> is the main entry point for running a regex. It contains
661 support for initialising the regex interpreter's state, running
662 C<re_intuit_start()> if needed, and running the interpreter on the string
663 from various start positions as needed. When it is necessary to use
664 the regex interpreter C<pregexec()> calls C<regtry()>.
666 C<regtry()> is the entry point into the regex interpreter. It expects
667 as arguments a pointer to a C<regmatch_info> structure and a pointer to
668 a string. It returns an integer 1 for success and a 0 for failure.
669 It is basically a set-up wrapper around C<regmatch()>.
671 C<regmatch> is the main "recursive loop" of the interpreter. It is
672 basically a giant switch statement that implements a state machine, where
673 the possible states are the regops themselves, plus a number of additional
674 intermediate and failure states. A few of the states are implemented as
675 subroutines but the bulk are inline code.
679 =head2 Unicode and Localisation Support
681 When dealing with strings containing characters that cannot be represented
682 using an eight-bit character set, perl uses an internal representation
683 that is a permissive version of Unicode's UTF-8 encoding[2]. This uses single
684 bytes to represent characters from the ASCII character set, and sequences
685 of two or more bytes for all other characters. (See L<perlunitut>
686 for more information about the relationship between UTF-8 and perl's
687 encoding, utf8 -- the difference isn't important for this discussion.)
689 No matter how you look at it, Unicode support is going to be a pain in a
690 regex engine. Tricks that might be fine when you have 256 possible
691 characters often won't scale to handle the size of the UTF-8 character
692 set. Things you can take for granted with ASCII may not be true with
693 Unicode. For instance, in ASCII, it is safe to assume that
694 C<sizeof(char1) == sizeof(char2)>, but in UTF-8 it isn't. Unicode case folding is
695 vastly more complex than the simple rules of ASCII, and even when not
696 using Unicode but only localised single byte encodings, things can get
697 tricky (for example, GERMAN-SHARP-ESS should match 'SS' in localised
698 case-insensitive matching).
700 Making things worse is that UTF-8 support was a later addition to the
701 regex engine (as it was to perl) and this necessarily made things a lot
702 more complicated. Obviously it is easier to design a regex engine with
703 Unicode support in mind from the beginning than it is to retrofit it to
706 Nearly all regops that involve looking at the input string have
707 two cases, one for UTF-8, and one not. In fact, it's often more complex
708 than that, as the pattern may be UTF-8 as well.
710 Care must be taken when making changes to make sure that you handle
711 UTF-8 properly, both at compile time and at execution time, including
712 when the string and pattern are mismatched.
714 The following comment in F<regcomp.h> gives an example of exactly how
717 Two problematic code points in Unicode casefolding of EXACT nodes:
719 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
720 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
726 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
727 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
729 This means that in case-insensitive matching (or "loose matching",
730 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded
731 byte length of the above casefolded versions) can match a target
732 string of length two (the byte length of UTF-8 encoded U+0390 or
733 U+03B0). This would rather mess up the minimum length computation.
735 What we'll do is to look for the tail four bytes, and then peek
736 at the preceding two bytes to see whether we need to decrease
737 the minimum length by four (six minus two).
739 Thanks to the design of UTF-8, there cannot be false matches:
740 A sequence of valid UTF-8 bytes cannot be a subsequence of
741 another valid sequence of UTF-8 bytes.
744 =head2 Base Structures
746 There are two structures used to store a compiled regular expression.
747 One, the regexp structure is considered to be perl's property, and the
748 other is considered to be the property of the regex engine which
749 compiled the regular expression; in the case of the stock engine this
750 structure is called regexp_internal.
752 The regexp structure contains all the data that perl needs to be aware of
753 to properly work with the regular expression. It includes data about
754 optimisations that perl can use to determine if the regex engine should
755 really be used, and various other control info that is needed to properly
756 execute patterns in various contexts such as is the pattern anchored in
757 some way, or what flags were used during the compile, or whether the
758 program contains special constructs that perl needs to be aware of.
760 In addition it contains two fields that are intended for the private use
761 of the regex engine that compiled the pattern. These are the C<intflags>
762 and pprivate members. The C<pprivate> is a void pointer to an arbitrary
763 structure whose use and management is the responsibility of the compiling
764 engine. perl will never modify either of these values.
766 As mentioned earlier, in the case of the default engines, the C<pprivate>
767 will be a pointer to a regexp_internal structure which holds the compiled
768 program and any additional data that is private to the regex engine
771 =head3 Perl Inspectable Data About Pattern
773 F<regexp.h> contains the "public" structure definition. All regex engines
774 must be able to correctly build a regexp structure.
776 typedef struct regexp {
777 /* what engine created this regexp? */
778 const struct regexp_engine* engine;
780 /* Information about the match that the perl core uses to manage things */
781 U32 extflags; /* Flags used both externally and internally */
782 I32 minlen; /* mininum possible length of string to match */
783 I32 minlenret; /* mininum possible length of $& */
784 U32 gofs; /* chars left of pos that we search from */
785 struct reg_substr_data *substrs; /* substring data about strings that must appear
786 in the final match, used for optimisations */
787 U32 nparens; /* number of capture buffers */
789 /* private engine specific data */
790 U32 intflags; /* Engine Specific Internal flags */
791 void *pprivate; /* Data private to the regex engine which
792 created this object. */
794 /* Data about the last/current match. These are modified during matching*/
795 U32 lastparen; /* last open paren matched */
796 U32 lastcloseparen; /* last close paren matched */
797 I32 *startp; /* Array of offsets from start of string (@-) */
798 I32 *endp; /* Array of offsets from start of string (@+) */
799 char *subbeg; /* saved or original string
800 so \digit works forever. */
801 I32 sublen; /* Length of string pointed by subbeg */
802 SV_SAVED_COPY /* If non-NULL, SV which is COW from original */
805 /* Information about the match that isn't often used */
806 char *precomp; /* pre-compilation regular expression */
807 I32 prelen; /* length of precomp */
808 I32 seen_evals; /* number of eval groups in the pattern - for security checks */
809 HV *paren_names; /* Optional hash of paren names */
811 /* Refcount of this regexp */
812 I32 refcnt; /* Refcount of this regexp */
815 The fields are discussed in more detail below:
822 The number of times the structure is referenced. When this falls to 0
823 the regexp is automatically freed by a call to pregfree.
827 This field points at a regexp_engine structure which contains pointers
828 to the subroutine that are to be used for performing a match. It
829 is the compiling routines responsibility to populate this field before
830 returning the regexp object.
832 =item C<precomp> C<prelen>
834 Used for debugging purposes. C<precomp> holds a copy of the pattern
839 This is used to store various flags about the pattern, such as whether it
840 contains a \G or a ^ or $ symbol.
842 =item C<minlen> C<minlenret>
844 C<minlen> is the minimum string length required for the pattern to match.
845 This is used to prune the search space by not bothering to match any
846 closer to the end of a string than would allow a match. For instance
847 there is no point in even starting the regex engine if the minlen is
848 10 but the string is only 5 characters long. There is no way that the
851 C<minlenret> is the minimum length of the string that would be found
854 The difference between C<minlen> and C<minlenret> can be seen in the
859 where the C<minlen> would be 3 but the minlen ret would only be 2 as
860 the \d is required to match but is not actually included in the matched
861 content. This distinction is particularly important as the substitution
862 logic uses the C<minlenret> to tell whether it can do in-place substition
863 which can result in considerable speedup.
867 Left offset from pos() to start match at.
869 =item C<nparens>, C<lasparen>, and C<lastcloseparen>
871 These fields are used to keep track of how many paren groups could be matched
872 in the pattern, which was the last open paren to be entered, and which was
873 the last close paren to be entered.
877 This is a hash used internally to track named capture buffers and their
878 offsets. The keys are the names of the buffers the values are dualvars,
879 with the IV slot holding the number of buffers with the given name and the
880 pv being an embedded array of I32. The values may also be contained
881 independently in the data array in cases where named backreferences are
884 =item C<reg_substr_data>
886 Holds information on the longest string that must occur at a fixed
887 offset from the start of the pattern, and the longest string that must
888 occur at a floating offset from the start of the pattern. Used to do
889 Fast-Boyer-Moore searches on the string to find out if its worth using
890 the regex engine at all, and if so where in the string to search.
892 =item C<startp>, C<endp>,
894 These fields store arrays that are used to hold the offsets of the begining
895 and end of each capture group that has matched. -1 is used to indicate no match.
897 These are the source for @- and @+.
899 =item C<subbeg> C<sublen> C<saved_copy>
901 These are used during execution phase for managing search and replace
906 This stores the number of eval groups in the pattern. This is used
907 for security purposes when embedding compiled regexes into larger
912 =head3 Engine Private Data About Pattern
914 Additionally regexp.h contains the following "private" definition which is perl
915 specific and is only of curiosity value to other engine implementations.
917 typedef struct regexp_internal {
918 regexp_paren_ofs *swap; /* Swap copy of *startp / *endp */
919 U32 *offsets; /* offset annotations 20001228 MJD
920 data about mapping the program to the
922 regnode *regstclass; /* Optional startclass as identified or constructed
924 struct reg_data *data; /* Additional miscellaneous data used by the program.
925 Used to make it easier to clone and free arbitrary
926 data that the regops need. Often the ARG field of
927 a regop is an index into this structure */
928 regnode program[1]; /* Unwarranted chumminess with compiler. */
935 C<swap> is an extra set of startp/endp stored in a C<regexp_paren_ofs>
936 struct. This is used when the last successful match was from same pattern
937 as the current pattern, so that a partial match doesn't overwrite the
938 previous match's results. When this field is data filled the matching
939 engine will swap buffers before every match attempt. If the match fails,
940 then it swaps them back. If it's successful it leaves them. This field
941 is populated on demand and is by default null.
945 Offsets holds a mapping of offset in the C<program>
946 to offset in the C<precomp> string. This is only used by ActiveStates
947 visual regex debugger.
951 Special regop that is used by C<re_intuit_start()> to check if a pattern
952 can match at a certain position. For instance if the regex engine knows
953 that the pattern must start with a 'Z' then it can scan the string until
954 it finds one and then launch the regex engine from there. The routine
955 that handles this is called C<find_by_class()>. Sometimes this field
956 points at a regop embedded in the program, and sometimes it points at
957 an independent synthetic regop that has been constructed by the optimiser.
961 This field points at a reg_data structure, which is defined as follows
969 This structure is used for handling data structures that the regex engine
970 needs to handle specially during a clone or free operation on the compiled
971 product. Each element in the data array has a corresponding element in the
972 what array. During compilation regops that need special structures stored
973 will add an element to each array using the add_data() routine and then store
974 the index in the regop.
978 Compiled program. Inlined into the structure so the entire struct can be
979 treated as a single blob.
983 =head2 Pluggable Interface
985 As of Perl 5.9.5 there is a new interface for using other regexp engines
986 than the default one. Each engine is supposed to provide access to
987 a constant structure of the following format:
989 typedef struct regexp_engine {
990 regexp* (*comp) (pTHX_ char* exp, char* xend, PMOP* pm);
991 I32 (*exec) (pTHX_ regexp* prog, char* stringarg, char* strend,
992 char* strbeg, I32 minend, SV* screamer,
993 void* data, U32 flags);
994 char* (*intuit) (pTHX_ regexp *prog, SV *sv, char *strpos,
995 char *strend, U32 flags,
996 struct re_scream_pos_data_s *data);
997 SV* (*checkstr) (pTHX_ regexp *prog);
998 void (*free) (pTHX_ struct regexp* r);
1000 void* (*dupe) (pTHX_ const regexp *r, CLONE_PARAMS *param);
1004 When a regexp is compiled its C<engine> field is then set to point at
1005 the appropriate structure so that when it needs to be used Perl can find
1006 the right routines to do so.
1008 In order to install a new regexp handler, C<$^H{regcomp}> is set
1009 to an integer which (when casted appropriately) resolves to one of these
1010 structures. When compiling the C<comp> method is executed, and the
1011 resulting regexp structures engine field is expected to point back at
1014 The pTHX_ symbol in the definition is a macro used by perl under threading
1015 to provide an extra argument to the routine holding a pointer back to
1016 the interpreter that is executing the regexp. So under threading all
1017 routines get an extra argument.
1019 The routines are as follows:
1025 regexp* comp(char *exp, char *xend, PMOP pm);
1027 Compile the pattern between exp and xend using the flags contained in
1028 pm and return a pointer to a prepared regexp structure that can perform
1033 I32 exec(regexp* prog,
1034 char *stringarg, char* strend, char* strbeg,
1035 I32 minend, SV* screamer,
1036 void* data, U32 flags);
1042 char* intuit( regexp *prog,
1043 SV *sv, char *strpos, char *strend,
1044 U32 flags, struct re_scream_pos_data_s *data);
1046 Find the start position where a regex match should be attempted,
1047 or possibly whether the regex engine should not be run because the
1048 pattern can't match. This is called as appropriate by the core
1049 depending on the values of the extflags member of the regexp
1054 SV* checkstr(regexp *prog);
1056 Return a SV containing a string that must appear in the pattern. Used
1057 for optimising matches.
1061 void free(regexp *prog);
1063 Called by perl when it is freeing a regexp pattern so that the engine
1064 can release any resources pointed to by the C<pprivate> member of the
1065 regexp structure. This is only responsible for freeing private data,
1066 perl will handle releasing anything else contained in the regexp structure.
1070 void* dupe(const regexp *r, CLONE_PARAMS *param);
1072 On threaded builds a regexp may need to be duplicated so that the pattern
1073 can be used by mutiple threads. This routine is expected to handle the
1074 duplication of any private data pointed to by the C<pprivate> member of
1075 the regexp structure. It will be called with the preconstructed new
1076 regexp structure as an argument, the C<pprivate> member will point at
1077 the B<old> private structue, and it is this routines responsibility to
1078 construct a copy and return a pointer to it (which perl will then use to
1079 overwrite the field as passed to this routine.)
1081 This allows the engine to dupe its private data but also if necessary
1082 modify the final structure if it really must.
1084 On unthreaded builds this field doesn't exist.
1089 =head2 De-allocation and Cloning
1091 Any patch that adds data items to the regexp will need to include
1092 changes to F<sv.c> (C<Perl_re_dup()>) and F<regcomp.c> (C<pregfree()>). This
1093 involves freeing or cloning items in the regexes data array based
1094 on the data item's type.
1104 by Yves Orton, 2006.
1106 With excerpts from Perl, and contributions and suggestions from
1107 Ronald J. Kimball, Dave Mitchell, Dominic Dunlop, Mark Jason Dominus,
1108 Stephen McCamant, and David Landgren.
1116 [1] L<http://perl.plover.com/Rx/paper/>
1118 [2] L<http://www.unicode.org>