xref: /openbsd-src/gnu/usr.bin/perl/pod/perlreguts.pod (revision 91f110e064cd7c194e59e019b83bb7496c1c84d4)
1=head1 NAME
2
3perlreguts - Description of the Perl regular expression engine.
4
5=head1 DESCRIPTION
6
7This document is an attempt to shine some light on the guts of the regex
8engine and how it works. The regex engine represents a significant chunk
9of the perl codebase, but is relatively poorly understood. This document
10is a meagre attempt at addressing this situation. It is derived from the
11author's experience, comments in the source code, other papers on the
12regex engine, feedback on the perl5-porters mail list, and no doubt other
13places as well.
14
15B<NOTICE!> It should be clearly understood that the behavior and
16structures discussed in this represents the state of the engine as the
17author understood it at the time of writing. It is B<NOT> an API
18definition, it is purely an internals guide for those who want to hack
19the regex engine, or understand how the regex engine works. Readers of
20this document are expected to understand perl's regex syntax and its
21usage in detail. If you want to learn about the basics of Perl's
22regular expressions, see L<perlre>. And if you want to replace the
23regex engine with your own, see L<perlreapi>.
24
25=head1 OVERVIEW
26
27=head2 A quick note on terms
28
29There is some debate as to whether to say "regexp" or "regex". In this
30document we will use the term "regex" unless there is a special reason
31not to, in which case we will explain why.
32
33When speaking about regexes we need to distinguish between their source
34code form and their internal form. In this document we will use the term
35"pattern" when we speak of their textual, source code form, and the term
36"program" when we speak of their internal representation. These
37correspond to the terms I<S-regex> and I<B-regex> that Mark Jason
38Dominus employs in his paper on "Rx" ([1] in L</REFERENCES>).
39
40=head2 What is a regular expression engine?
41
42A regular expression engine is a program that takes a set of constraints
43specified in a mini-language, and then applies those constraints to a
44target string, and determines whether or not the string satisfies the
45constraints. See L<perlre> for a full definition of the language.
46
47In less grandiose terms, the first part of the job is to turn a pattern into
48something the computer can efficiently use to find the matching point in
49the string, and the second part is performing the search itself.
50
51To do this we need to produce a program by parsing the text. We then
52need to execute the program to find the point in the string that
53matches. And we need to do the whole thing efficiently.
54
55=head2 Structure of a Regexp Program
56
57=head3 High Level
58
59Although it is a bit confusing and some people object to the terminology, it
60is worth taking a look at a comment that has
61been in F<regexp.h> for years:
62
63I<This is essentially a linear encoding of a nondeterministic
64finite-state machine (aka syntax charts or "railroad normal form" in
65parsing technology).>
66
67The term "railroad normal form" is a bit esoteric, with "syntax
68diagram/charts", or "railroad diagram/charts" being more common terms.
69Nevertheless it provides a useful mental image of a regex program: each
70node can be thought of as a unit of track, with a single entry and in
71most cases a single exit point (there are pieces of track that fork, but
72statistically not many), and the whole forms a layout with a
73single entry and single exit point. The matching process can be thought
74of as a car that moves along the track, with the particular route through
75the system being determined by the character read at each possible
76connector point. A car can fall off the track at any point but it may
77only proceed as long as it matches the track.
78
79Thus the pattern C</foo(?:\w+|\d+|\s+)bar/> can be thought of as the
80following chart:
81
82                      [start]
83                         |
84                       <foo>
85                         |
86                   +-----+-----+
87                   |     |     |
88                 <\w+> <\d+> <\s+>
89                   |     |     |
90                   +-----+-----+
91                         |
92                       <bar>
93                         |
94                       [end]
95
96The truth of the matter is that perl's regular expressions these days are
97much more complex than this kind of structure, but visualising it this way
98can help when trying to get your bearings, and it matches the
99current implementation pretty closely.
100
101To be more precise, we will say that a regex program is an encoding
102of a graph. Each node in the graph corresponds to part of
103the original regex pattern, such as a literal string or a branch,
104and has a pointer to the nodes representing the next component
105to be matched. Since "node" and "opcode" already have other meanings in the
106perl source, we will call the nodes in a regex program "regops".
107
108The program is represented by an array of C<regnode> structures, one or
109more of which represent a single regop of the program. Struct
110C<regnode> is the smallest struct needed, and has a field structure which is
111shared with all the other larger structures.
112
113The "next" pointers of all regops except C<BRANCH> implement concatenation;
114a "next" pointer with a C<BRANCH> on both ends of it is connecting two
115alternatives.  [Here we have one of the subtle syntax dependencies: an
116individual C<BRANCH> (as opposed to a collection of them) is never
117concatenated with anything because of operator precedence.]
118
119The operand of some types of regop is a literal string; for others,
120it is a regop leading into a sub-program.  In particular, the operand
121of a C<BRANCH> node is the first regop of the branch.
122
123B<NOTE>: As the railroad metaphor suggests, this is B<not> a tree
124structure:  the tail of the branch connects to the thing following the
125set of C<BRANCH>es.  It is a like a single line of railway track that
126splits as it goes into a station or railway yard and rejoins as it comes
127out the other side.
128
129=head3 Regops
130
131The base structure of a regop is defined in F<regexp.h> as follows:
132
133    struct regnode {
134        U8  flags;    /* Various purposes, sometimes overridden */
135        U8  type;     /* Opcode value as specified by regnodes.h */
136        U16 next_off; /* Offset in size regnode */
137    };
138
139Other larger C<regnode>-like structures are defined in F<regcomp.h>. They
140are almost like subclasses in that they have the same fields as
141C<regnode>, with possibly additional fields following in
142the structure, and in some cases the specific meaning (and name)
143of some of base fields are overridden. The following is a more
144complete description.
145
146=over 4
147
148=item C<regnode_1>
149
150=item C<regnode_2>
151
152C<regnode_1> structures have the same header, followed by a single
153four-byte argument; C<regnode_2> structures contain two two-byte
154arguments instead:
155
156    regnode_1                U32 arg1;
157    regnode_2                U16 arg1;  U16 arg2;
158
159=item C<regnode_string>
160
161C<regnode_string> structures, used for literal strings, follow the header
162with a one-byte length and then the string data. Strings are padded on
163the end with zero bytes so that the total length of the node is a
164multiple of four bytes:
165
166    regnode_string           char string[1];
167                             U8 str_len; /* overrides flags */
168
169=item C<regnode_charclass>
170
171Character classes are represented by C<regnode_charclass> structures,
172which have a four-byte argument and then a 32-byte (256-bit) bitmap
173indicating which characters are included in the class.
174
175    regnode_charclass        U32 arg1;
176                             char bitmap[ANYOF_BITMAP_SIZE];
177
178=item C<regnode_charclass_class>
179
180There is also a larger form of a char class structure used to represent
181POSIX char classes called C<regnode_charclass_class> which has an
182additional 4-byte (32-bit) bitmap indicating which POSIX char classes
183have been included.
184
185   regnode_charclass_class  U32 arg1;
186                            char bitmap[ANYOF_BITMAP_SIZE];
187                            char classflags[ANYOF_CLASSBITMAP_SIZE];
188
189=back
190
191F<regnodes.h> defines an array called C<regarglen[]> which gives the size
192of each opcode in units of C<size regnode> (4-byte). A macro is used
193to calculate the size of an C<EXACT> node based on its C<str_len> field.
194
195The regops are defined in F<regnodes.h> which is generated from
196F<regcomp.sym> by F<regcomp.pl>. Currently the maximum possible number
197of distinct regops is restricted to 256, with about a quarter already
198used.
199
200A set of macros makes accessing the fields
201easier and more consistent. These include C<OP()>, which is used to determine
202the type of a C<regnode>-like structure; C<NEXT_OFF()>, which is the offset to
203the next node (more on this later); C<ARG()>, C<ARG1()>, C<ARG2()>, C<ARG_SET()>,
204and equivalents for reading and setting the arguments; and C<STR_LEN()>,
205C<STRING()> and C<OPERAND()> for manipulating strings and regop bearing
206types.
207
208=head3 What regop is next?
209
210There are three distinct concepts of "next" in the regex engine, and
211it is important to keep them clear.
212
213=over 4
214
215=item *
216
217There is the "next regnode" from a given regnode, a value which is
218rarely useful except that sometimes it matches up in terms of value
219with one of the others, and that sometimes the code assumes this to
220always be so.
221
222=item *
223
224There is the "next regop" from a given regop/regnode. This is the
225regop physically located after the current one, as determined by
226the size of the current regop. This is often useful, such as when
227dumping the structure we use this order to traverse. Sometimes the code
228assumes that the "next regnode" is the same as the "next regop", or in
229other words assumes that the sizeof a given regop type is always going
230to be one regnode large.
231
232=item *
233
234There is the "regnext" from a given regop. This is the regop which
235is reached by jumping forward by the value of C<NEXT_OFF()>,
236or in a few cases for longer jumps by the C<arg1> field of the C<regnode_1>
237structure. The subroutine C<regnext()> handles this transparently.
238This is the logical successor of the node, which in some cases, like
239that of the C<BRANCH> regop, has special meaning.
240
241=back
242
243=head1 Process Overview
244
245Broadly speaking, performing a match of a string against a pattern
246involves the following steps:
247
248=over 5
249
250=item A. Compilation
251
252=over 5
253
254=item 1. Parsing for size
255
256=item 2. Parsing for construction
257
258=item 3. Peep-hole optimisation and analysis
259
260=back
261
262=item B. Execution
263
264=over 5
265
266=item 4. Start position and no-match optimisations
267
268=item 5. Program execution
269
270=back
271
272=back
273
274
275Where these steps occur in the actual execution of a perl program is
276determined by whether the pattern involves interpolating any string
277variables. If interpolation occurs, then compilation happens at run time. If it
278does not, then compilation is performed at compile time. (The C</o> modifier changes this,
279as does C<qr//> to a certain extent.) The engine doesn't really care that
280much.
281
282=head2 Compilation
283
284This code resides primarily in F<regcomp.c>, along with the header files
285F<regcomp.h>, F<regexp.h> and F<regnodes.h>.
286
287Compilation starts with C<pregcomp()>, which is mostly an initialisation
288wrapper which farms work out to two other routines for the heavy lifting: the
289first is C<reg()>, which is the start point for parsing; the second,
290C<study_chunk()>, is responsible for optimisation.
291
292Initialisation in C<pregcomp()> mostly involves the creation and data-filling
293of a special structure, C<RExC_state_t> (defined in F<regcomp.c>).
294Almost all internally-used routines in F<regcomp.h> take a pointer to one
295of these structures as their first argument, with the name C<pRExC_state>.
296This structure is used to store the compilation state and contains many
297fields. Likewise there are many macros which operate on this
298variable: anything that looks like C<RExC_xxxx> is a macro that operates on
299this pointer/structure.
300
301=head3 Parsing for size
302
303In this pass the input pattern is parsed in order to calculate how much
304space is needed for each regop we would need to emit. The size is also
305used to determine whether long jumps will be required in the program.
306
307This stage is controlled by the macro C<SIZE_ONLY> being set.
308
309The parse proceeds pretty much exactly as it does during the
310construction phase, except that most routines are short-circuited to
311change the size field C<RExC_size> and not do anything else.
312
313=head3 Parsing for construction
314
315Once the size of the program has been determined, the pattern is parsed
316again, but this time for real. Now C<SIZE_ONLY> will be false, and the
317actual construction can occur.
318
319C<reg()> is the start of the parse process. It is responsible for
320parsing an arbitrary chunk of pattern up to either the end of the
321string, or the first closing parenthesis it encounters in the pattern.
322This means it can be used to parse the top-level regex, or any section
323inside of a grouping parenthesis. It also handles the "special parens"
324that perl's regexes have. For instance when parsing C</x(?:foo)y/> C<reg()>
325will at one point be called to parse from the "?" symbol up to and
326including the ")".
327
328Additionally, C<reg()> is responsible for parsing the one or more
329branches from the pattern, and for "finishing them off" by correctly
330setting their next pointers. In order to do the parsing, it repeatedly
331calls out to C<regbranch()>, which is responsible for handling up to the
332first C<|> symbol it sees.
333
334C<regbranch()> in turn calls C<regpiece()> which
335handles "things" followed by a quantifier. In order to parse the
336"things", C<regatom()> is called. This is the lowest level routine, which
337parses out constant strings, character classes, and the
338various special symbols like C<$>. If C<regatom()> encounters a "("
339character it in turn calls C<reg()>.
340
341The routine C<regtail()> is called by both C<reg()> and C<regbranch()>
342in order to "set the tail pointer" correctly. When executing and
343we get to the end of a branch, we need to go to the node following the
344grouping parens. When parsing, however, we don't know where the end will
345be until we get there, so when we do we must go back and update the
346offsets as appropriate. C<regtail> is used to make this easier.
347
348A subtlety of the parsing process means that a regex like C</foo/> is
349originally parsed into an alternation with a single branch. It is only
350afterwards that the optimiser converts single branch alternations into the
351simpler form.
352
353=head3 Parse Call Graph and a Grammar
354
355The call graph looks like this:
356
357 reg()                        # parse a top level regex, or inside of
358                              # parens
359     regbranch()              # parse a single branch of an alternation
360         regpiece()           # parse a pattern followed by a quantifier
361             regatom()        # parse a simple pattern
362                 regclass()   #   used to handle a class
363                 reg()        #   used to handle a parenthesised
364                              #   subpattern
365                 ....
366         ...
367         regtail()            # finish off the branch
368     ...
369     regtail()                # finish off the branch sequence. Tie each
370                              # branch's tail to the tail of the
371                              # sequence
372                              # (NEW) In Debug mode this is
373                              # regtail_study().
374
375A grammar form might be something like this:
376
377    atom  : constant | class
378    quant : '*' | '+' | '?' | '{min,max}'
379    _branch: piece
380           | piece _branch
381           | nothing
382    branch: _branch
383          | _branch '|' branch
384    group : '(' branch ')'
385    _piece: atom | group
386    piece : _piece
387          | _piece quant
388
389=head3 Parsing complications
390
391The implication of the above description is that a pattern containing nested
392parentheses will result in a call graph which cycles through C<reg()>,
393C<regbranch()>, C<regpiece()>, C<regatom()>, C<reg()>, C<regbranch()> I<etc>
394multiple times, until the deepest level of nesting is reached. All the above
395routines return a pointer to a C<regnode>, which is usually the last regnode
396added to the program. However, one complication is that reg() returns NULL
397for parsing C<(?:)> syntax for embedded modifiers, setting the flag
398C<TRYAGAIN>. The C<TRYAGAIN> propagates upwards until it is captured, in
399some cases by by C<regatom()>, but otherwise unconditionally by
400C<regbranch()>. Hence it will never be returned by C<regbranch()> to
401C<reg()>. This flag permits patterns such as C<(?i)+> to be detected as
402errors (I<Quantifier follows nothing in regex; marked by <-- HERE in m/(?i)+
403<-- HERE />).
404
405Another complication is that the representation used for the program differs
406if it needs to store Unicode, but it's not always possible to know for sure
407whether it does until midway through parsing. The Unicode representation for
408the program is larger, and cannot be matched as efficiently. (See L</Unicode
409and Localisation Support> below for more details as to why.)  If the pattern
410contains literal Unicode, it's obvious that the program needs to store
411Unicode. Otherwise, the parser optimistically assumes that the more
412efficient representation can be used, and starts sizing on this basis.
413However, if it then encounters something in the pattern which must be stored
414as Unicode, such as an C<\x{...}> escape sequence representing a character
415literal, then this means that all previously calculated sizes need to be
416redone, using values appropriate for the Unicode representation. Currently,
417all regular expression constructions which can trigger this are parsed by code
418in C<regatom()>.
419
420To avoid wasted work when a restart is needed, the sizing pass is abandoned
421- C<regatom()> immediately returns NULL, setting the flag C<RESTART_UTF8>.
422(This action is encapsulated using the macro C<REQUIRE_UTF8>.) This restart
423request is propagated up the call chain in a similar fashion, until it is
424"caught" in C<Perl_re_op_compile()>, which marks the pattern as containing
425Unicode, and restarts the sizing pass. It is also possible for constructions
426within run-time code blocks to turn out to need Unicode representation.,
427which is signalled by C<S_compile_runtime_code()> returning false to
428C<Perl_re_op_compile()>.
429
430The restart was previously implemented using a C<longjmp> in C<regatom()>
431back to a C<setjmp> in C<Perl_re_op_compile()>, but this proved to be
432problematic as the latter is a large function containing many automatic
433variables, which interact badly with the emergent control flow of C<setjmp>.
434
435=head3 Debug Output
436
437In the 5.9.x development version of perl you can C<< use re Debug => 'PARSE' >>
438to see some trace information about the parse process. We will start with some
439simple patterns and build up to more complex patterns.
440
441So when we parse C</foo/> we see something like the following table. The
442left shows what is being parsed, and the number indicates where the next regop
443would go. The stuff on the right is the trace output of the graph. The
444names are chosen to be short to make it less dense on the screen. 'tsdy'
445is a special form of C<regtail()> which does some extra analysis.
446
447 >foo<             1    reg
448                          brnc
449                            piec
450                              atom
451 ><                4      tsdy~ EXACT <foo> (EXACT) (1)
452                              ~ attach to END (3) offset to 2
453
454The resulting program then looks like:
455
456   1: EXACT <foo>(3)
457   3: END(0)
458
459As you can see, even though we parsed out a branch and a piece, it was ultimately
460only an atom. The final program shows us how things work. We have an C<EXACT> regop,
461followed by an C<END> regop. The number in parens indicates where the C<regnext> of
462the node goes. The C<regnext> of an C<END> regop is unused, as C<END> regops mean
463we have successfully matched. The number on the left indicates the position of
464the regop in the regnode array.
465
466Now let's try a harder pattern. We will add a quantifier, so now we have the pattern
467C</foo+/>. We will see that C<regbranch()> calls C<regpiece()> twice.
468
469 >foo+<            1    reg
470                          brnc
471                            piec
472                              atom
473 >o+<              3        piec
474                              atom
475 ><                6        tail~ EXACT <fo> (1)
476                   7      tsdy~ EXACT <fo> (EXACT) (1)
477                              ~ PLUS (END) (3)
478                              ~ attach to END (6) offset to 3
479
480And we end up with the program:
481
482   1: EXACT <fo>(3)
483   3: PLUS(6)
484   4:   EXACT <o>(0)
485   6: END(0)
486
487Now we have a special case. The C<EXACT> regop has a C<regnext> of 0. This is
488because if it matches it should try to match itself again. The C<PLUS> regop
489handles the actual failure of the C<EXACT> regop and acts appropriately (going
490to regnode 6 if the C<EXACT> matched at least once, or failing if it didn't).
491
492Now for something much more complex: C</x(?:foo*|b[a][rR])(foo|bar)$/>
493
494 >x(?:foo*|b...    1    reg
495                          brnc
496                            piec
497                              atom
498 >(?:foo*|b[...    3        piec
499                              atom
500 >?:foo*|b[a...                 reg
501 >foo*|b[a][...                   brnc
502                                    piec
503                                      atom
504 >o*|b[a][rR...    5                piec
505                                      atom
506 >|b[a][rR])...    8                tail~ EXACT <fo> (3)
507 >b[a][rR])(...    9              brnc
508                  10                piec
509                                      atom
510 >[a][rR])(f...   12                piec
511                                      atom
512 >a][rR])(fo...                         clas
513 >[rR])(foo|...   14                tail~ EXACT <b> (10)
514                                    piec
515                                      atom
516 >rR])(foo|b...                         clas
517 >)(foo|bar)...   25                tail~ EXACT <a> (12)
518                                  tail~ BRANCH (3)
519                  26              tsdy~ BRANCH (END) (9)
520                                      ~ attach to TAIL (25) offset to 16
521                                  tsdy~ EXACT <fo> (EXACT) (4)
522                                      ~ STAR (END) (6)
523                                      ~ attach to TAIL (25) offset to 19
524                                  tsdy~ EXACT <b> (EXACT) (10)
525                                      ~ EXACT <a> (EXACT) (12)
526                                      ~ ANYOF[Rr] (END) (14)
527                                      ~ attach to TAIL (25) offset to 11
528 >(foo|bar)$<               tail~ EXACT <x> (1)
529                            piec
530                              atom
531 >foo|bar)$<                    reg
532                  28              brnc
533                                    piec
534                                      atom
535 >|bar)$<         31              tail~ OPEN1 (26)
536 >bar)$<                          brnc
537                  32                piec
538                                      atom
539 >)$<             34              tail~ BRANCH (28)
540                  36              tsdy~ BRANCH (END) (31)
541                                     ~ attach to CLOSE1 (34) offset to 3
542                                  tsdy~ EXACT <foo> (EXACT) (29)
543                                     ~ attach to CLOSE1 (34) offset to 5
544                                  tsdy~ EXACT <bar> (EXACT) (32)
545                                     ~ attach to CLOSE1 (34) offset to 2
546 >$<                        tail~ BRANCH (3)
547                                ~ BRANCH (9)
548                                ~ TAIL (25)
549                            piec
550                              atom
551 ><               37        tail~ OPEN1 (26)
552                                ~ BRANCH (28)
553                                ~ BRANCH (31)
554                                ~ CLOSE1 (34)
555                  38      tsdy~ EXACT <x> (EXACT) (1)
556                              ~ BRANCH (END) (3)
557                              ~ BRANCH (END) (9)
558                              ~ TAIL (END) (25)
559                              ~ OPEN1 (END) (26)
560                              ~ BRANCH (END) (28)
561                              ~ BRANCH (END) (31)
562                              ~ CLOSE1 (END) (34)
563                              ~ EOL (END) (36)
564                              ~ attach to END (37) offset to 1
565
566Resulting in the program
567
568   1: EXACT <x>(3)
569   3: BRANCH(9)
570   4:   EXACT <fo>(6)
571   6:   STAR(26)
572   7:     EXACT <o>(0)
573   9: BRANCH(25)
574  10:   EXACT <ba>(14)
575  12:   OPTIMIZED (2 nodes)
576  14:   ANYOF[Rr](26)
577  25: TAIL(26)
578  26: OPEN1(28)
579  28:   TRIE-EXACT(34)
580        [StS:1 Wds:2 Cs:6 Uq:5 #Sts:7 Mn:3 Mx:3 Stcls:bf]
581          <foo>
582          <bar>
583  30:   OPTIMIZED (4 nodes)
584  34: CLOSE1(36)
585  36: EOL(37)
586  37: END(0)
587
588Here we can see a much more complex program, with various optimisations in
589play. At regnode 10 we see an example where a character class with only
590one character in it was turned into an C<EXACT> node. We can also see where
591an entire alternation was turned into a C<TRIE-EXACT> node. As a consequence,
592some of the regnodes have been marked as optimised away. We can see that
593the C<$> symbol has been converted into an C<EOL> regop, a special piece of
594code that looks for C<\n> or the end of the string.
595
596The next pointer for C<BRANCH>es is interesting in that it points at where
597execution should go if the branch fails. When executing, if the engine
598tries to traverse from a branch to a C<regnext> that isn't a branch then
599the engine will know that the entire set of branches has failed.
600
601=head3 Peep-hole Optimisation and Analysis
602
603The regular expression engine can be a weighty tool to wield. On long
604strings and complex patterns it can end up having to do a lot of work
605to find a match, and even more to decide that no match is possible.
606Consider a situation like the following pattern.
607
608   'ababababababababababab' =~ /(a|b)*z/
609
610The C<(a|b)*> part can match at every char in the string, and then fail
611every time because there is no C<z> in the string. So obviously we can
612avoid using the regex engine unless there is a C<z> in the string.
613Likewise in a pattern like:
614
615   /foo(\w+)bar/
616
617In this case we know that the string must contain a C<foo> which must be
618followed by C<bar>. We can use Fast Boyer-Moore matching as implemented
619in C<fbm_instr()> to find the location of these strings. If they don't exist
620then we don't need to resort to the much more expensive regex engine.
621Even better, if they do exist then we can use their positions to
622reduce the search space that the regex engine needs to cover to determine
623if the entire pattern matches.
624
625There are various aspects of the pattern that can be used to facilitate
626optimisations along these lines:
627
628=over 5
629
630=item * anchored fixed strings
631
632=item * floating fixed strings
633
634=item * minimum and maximum length requirements
635
636=item * start class
637
638=item * Beginning/End of line positions
639
640=back
641
642Another form of optimisation that can occur is the post-parse "peep-hole"
643optimisation, where inefficient constructs are replaced by more efficient
644constructs. The C<TAIL> regops which are used during parsing to mark the end
645of branches and the end of groups are examples of this. These regops are used
646as place-holders during construction and "always match" so they can be
647"optimised away" by making the things that point to the C<TAIL> point to the
648thing that C<TAIL> points to, thus "skipping" the node.
649
650Another optimisation that can occur is that of "C<EXACT> merging" which is
651where two consecutive C<EXACT> nodes are merged into a single
652regop. An even more aggressive form of this is that a branch
653sequence of the form C<EXACT BRANCH ... EXACT> can be converted into a
654C<TRIE-EXACT> regop.
655
656All of this occurs in the routine C<study_chunk()> which uses a special
657structure C<scan_data_t> to store the analysis that it has performed, and
658does the "peep-hole" optimisations as it goes.
659
660The code involved in C<study_chunk()> is extremely cryptic. Be careful. :-)
661
662=head2 Execution
663
664Execution of a regex generally involves two phases, the first being
665finding the start point in the string where we should match from,
666and the second being running the regop interpreter.
667
668If we can tell that there is no valid start point then we don't bother running
669interpreter at all. Likewise, if we know from the analysis phase that we
670cannot detect a short-cut to the start position, we go straight to the
671interpreter.
672
673The two entry points are C<re_intuit_start()> and C<pregexec()>. These routines
674have a somewhat incestuous relationship with overlap between their functions,
675and C<pregexec()> may even call C<re_intuit_start()> on its own. Nevertheless
676other parts of the perl source code may call into either, or both.
677
678Execution of the interpreter itself used to be recursive, but thanks to the
679efforts of Dave Mitchell in the 5.9.x development track, that has changed: now an
680internal stack is maintained on the heap and the routine is fully
681iterative. This can make it tricky as the code is quite conservative
682about what state it stores, with the result that two consecutive lines in the
683code can actually be running in totally different contexts due to the
684simulated recursion.
685
686=head3 Start position and no-match optimisations
687
688C<re_intuit_start()> is responsible for handling start points and no-match
689optimisations as determined by the results of the analysis done by
690C<study_chunk()> (and described in L<Peep-hole Optimisation and Analysis>).
691
692The basic structure of this routine is to try to find the start- and/or
693end-points of where the pattern could match, and to ensure that the string
694is long enough to match the pattern. It tries to use more efficient
695methods over less efficient methods and may involve considerable
696cross-checking of constraints to find the place in the string that matches.
697For instance it may try to determine that a given fixed string must be
698not only present but a certain number of chars before the end of the
699string, or whatever.
700
701It calls several other routines, such as C<fbm_instr()> which does
702Fast Boyer Moore matching and C<find_byclass()> which is responsible for
703finding the start using the first mandatory regop in the program.
704
705When the optimisation criteria have been satisfied, C<reg_try()> is called
706to perform the match.
707
708=head3 Program execution
709
710C<pregexec()> is the main entry point for running a regex. It contains
711support for initialising the regex interpreter's state, running
712C<re_intuit_start()> if needed, and running the interpreter on the string
713from various start positions as needed. When it is necessary to use
714the regex interpreter C<pregexec()> calls C<regtry()>.
715
716C<regtry()> is the entry point into the regex interpreter. It expects
717as arguments a pointer to a C<regmatch_info> structure and a pointer to
718a string.  It returns an integer 1 for success and a 0 for failure.
719It is basically a set-up wrapper around C<regmatch()>.
720
721C<regmatch> is the main "recursive loop" of the interpreter. It is
722basically a giant switch statement that implements a state machine, where
723the possible states are the regops themselves, plus a number of additional
724intermediate and failure states. A few of the states are implemented as
725subroutines but the bulk are inline code.
726
727=head1 MISCELLANEOUS
728
729=head2 Unicode and Localisation Support
730
731When dealing with strings containing characters that cannot be represented
732using an eight-bit character set, perl uses an internal representation
733that is a permissive version of Unicode's UTF-8 encoding[2]. This uses single
734bytes to represent characters from the ASCII character set, and sequences
735of two or more bytes for all other characters. (See L<perlunitut>
736for more information about the relationship between UTF-8 and perl's
737encoding, utf8. The difference isn't important for this discussion.)
738
739No matter how you look at it, Unicode support is going to be a pain in a
740regex engine. Tricks that might be fine when you have 256 possible
741characters often won't scale to handle the size of the UTF-8 character
742set.  Things you can take for granted with ASCII may not be true with
743Unicode. For instance, in ASCII, it is safe to assume that
744C<sizeof(char1) == sizeof(char2)>, but in UTF-8 it isn't. Unicode case folding is
745vastly more complex than the simple rules of ASCII, and even when not
746using Unicode but only localised single byte encodings, things can get
747tricky (for example, B<LATIN SMALL LETTER SHARP S> (U+00DF, E<szlig>)
748should match 'SS' in localised case-insensitive matching).
749
750Making things worse is that UTF-8 support was a later addition to the
751regex engine (as it was to perl) and this necessarily  made things a lot
752more complicated. Obviously it is easier to design a regex engine with
753Unicode support in mind from the beginning than it is to retrofit it to
754one that wasn't.
755
756Nearly all regops that involve looking at the input string have
757two cases, one for UTF-8, and one not. In fact, it's often more complex
758than that, as the pattern may be UTF-8 as well.
759
760Care must be taken when making changes to make sure that you handle
761UTF-8 properly, both at compile time and at execution time, including
762when the string and pattern are mismatched.
763
764The following comment in F<regcomp.h> gives an example of exactly how
765tricky this can be:
766
767    Two problematic code points in Unicode casefolding of EXACT nodes:
768
769    U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
770    U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
771
772    which casefold to
773
774    Unicode                      UTF-8
775
776    U+03B9 U+0308 U+0301         0xCE 0xB9 0xCC 0x88 0xCC 0x81
777    U+03C5 U+0308 U+0301         0xCF 0x85 0xCC 0x88 0xCC 0x81
778
779    This means that in case-insensitive matching (or "loose matching",
780    as Unicode calls it), an EXACTF of length six (the UTF-8 encoded
781    byte length of the above casefolded versions) can match a target
782    string of length two (the byte length of UTF-8 encoded U+0390 or
783    U+03B0). This would rather mess up the minimum length computation.
784
785    What we'll do is to look for the tail four bytes, and then peek
786    at the preceding two bytes to see whether we need to decrease
787    the minimum length by four (six minus two).
788
789    Thanks to the design of UTF-8, there cannot be false matches:
790    A sequence of valid UTF-8 bytes cannot be a subsequence of
791    another valid sequence of UTF-8 bytes.
792
793
794=head2 Base Structures
795
796The C<regexp> structure described in L<perlreapi> is common to all
797regex engines. Two of its fields that are intended for the private use
798of the regex engine that compiled the pattern. These are the
799C<intflags> and pprivate members. The C<pprivate> is a void pointer to
800an arbitrary structure whose use and management is the responsibility
801of the compiling engine. perl will never modify either of these
802values. In the case of the stock engine the structure pointed to by
803C<pprivate> is called C<regexp_internal>.
804
805Its C<pprivate> and C<intflags> fields contain data
806specific to each engine.
807
808There are two structures used to store a compiled regular expression.
809One, the C<regexp> structure described in L<perlreapi> is populated by
810the engine currently being. used and some of its fields read by perl to
811implement things such as the stringification of C<qr//>.
812
813
814The other structure is pointed to be the C<regexp> struct's
815C<pprivate> and is in addition to C<intflags> in the same struct
816considered to be the property of the regex engine which compiled the
817regular expression;
818
819The regexp structure contains all the data that perl needs to be aware of
820to properly work with the regular expression. It includes data about
821optimisations that perl can use to determine if the regex engine should
822really be used, and various other control info that is needed to properly
823execute patterns in various contexts such as is the pattern anchored in
824some way, or what flags were used during the compile, or whether the
825program contains special constructs that perl needs to be aware of.
826
827In addition it contains two fields that are intended for the private use
828of the regex engine that compiled the pattern. These are the C<intflags>
829and pprivate members. The C<pprivate> is a void pointer to an arbitrary
830structure whose use and management is the responsibility of the compiling
831engine. perl will never modify either of these values.
832
833As mentioned earlier, in the case of the default engines, the C<pprivate>
834will be a pointer to a regexp_internal structure which holds the compiled
835program and any additional data that is private to the regex engine
836implementation.
837
838=head3 Perl's C<pprivate> structure
839
840The following structure is used as the C<pprivate> struct by perl's
841regex engine. Since it is specific to perl it is only of curiosity
842value to other engine implementations.
843
844 typedef struct regexp_internal {
845         U32 *offsets;           /* offset annotations 20001228 MJD
846                                  * data about mapping the program to
847                                  * the string*/
848         regnode *regstclass;    /* Optional startclass as identified or
849                                  * constructed by the optimiser */
850         struct reg_data *data;  /* Additional miscellaneous data used
851                                  * by the program.  Used to make it
852                                  * easier to clone and free arbitrary
853                                  * data that the regops need. Often the
854                                  * ARG field of a regop is an index
855                                  * into this structure */
856         regnode program[1];     /* Unwarranted chumminess with
857                                  * compiler. */
858 } regexp_internal;
859
860=over 5
861
862=item C<offsets>
863
864Offsets holds a mapping of offset in the C<program>
865to offset in the C<precomp> string. This is only used by ActiveState's
866visual regex debugger.
867
868=item C<regstclass>
869
870Special regop that is used by C<re_intuit_start()> to check if a pattern
871can match at a certain position. For instance if the regex engine knows
872that the pattern must start with a 'Z' then it can scan the string until
873it finds one and then launch the regex engine from there. The routine
874that handles this is called C<find_by_class()>. Sometimes this field
875points at a regop embedded in the program, and sometimes it points at
876an independent synthetic regop that has been constructed by the optimiser.
877
878=item C<data>
879
880This field points at a reg_data structure, which is defined as follows
881
882    struct reg_data {
883        U32 count;
884        U8 *what;
885        void* data[1];
886    };
887
888This structure is used for handling data structures that the regex engine
889needs to handle specially during a clone or free operation on the compiled
890product. Each element in the data array has a corresponding element in the
891what array. During compilation regops that need special structures stored
892will add an element to each array using the add_data() routine and then store
893the index in the regop.
894
895=item C<program>
896
897Compiled program. Inlined into the structure so the entire struct can be
898treated as a single blob.
899
900=back
901
902=head1 SEE ALSO
903
904L<perlreapi>
905
906L<perlre>
907
908L<perlunitut>
909
910=head1 AUTHOR
911
912by Yves Orton, 2006.
913
914With excerpts from Perl, and contributions and suggestions from
915Ronald J. Kimball, Dave Mitchell, Dominic Dunlop, Mark Jason Dominus,
916Stephen McCamant, and David Landgren.
917
918=head1 LICENCE
919
920Same terms as Perl.
921
922=head1 REFERENCES
923
924[1] L<http://perl.plover.com/Rx/paper/>
925
926[2] L<http://www.unicode.org>
927
928=cut
929