{
    "mode": "man",
    "parameter": "perlreguts",
    "section": "1",
    "url": "https://www.chedong.com/phpMan.php/man/perlreguts/1/json",
    "generated": "2026-06-15T16:42:32Z",
    "sections": {
        "NAME": {
            "content": "perlreguts - Description of the Perl regular expression engine.\n",
            "subsections": []
        },
        "DESCRIPTION": {
            "content": "This document is an attempt to shine some light on the guts of the regex engine and how it\nworks. The regex engine represents a significant chunk of the perl codebase, but is\nrelatively poorly understood. This document is a meagre attempt at addressing this situation.\nIt is derived from the author's experience, comments in the source code, other papers on the\nregex engine, feedback on the perl5-porters mail list, and no doubt other places as well.\n\nNOTICE! It should be clearly understood that the behavior and structures discussed in this\nrepresents the state of the engine as the author understood it at the time of writing. It is\nNOT an API definition, it is purely an internals guide for those who want to hack the regex\nengine, or understand how the regex engine works. Readers of this document are expected to\nunderstand perl's regex syntax and its usage in detail. If you want to learn about the basics\nof Perl's regular expressions, see perlre. And if you want to replace the regex engine with\nyour own, see perlreapi.\n",
            "subsections": []
        },
        "OVERVIEW": {
            "content": "",
            "subsections": [
                {
                    "name": "A quick note on terms",
                    "content": "There is some debate as to whether to say \"regexp\" or \"regex\". In this document we will use\nthe term \"regex\" unless there is a special reason not to, in which case we will explain why.\n\nWhen speaking about regexes we need to distinguish between their source code form and their\ninternal form. In this document we will use the term \"pattern\" when we speak of their\ntextual, source code form, and the term \"program\" when we speak of their internal\nrepresentation. These correspond to the terms S-regex and B-regex that Mark Jason Dominus\nemploys in his paper on \"Rx\" ([1] in \"REFERENCES\").\n"
                },
                {
                    "name": "What is a regular expression engine?",
                    "content": "A regular expression engine is a program that takes a set of constraints specified in a mini-\nlanguage, and then applies those constraints to a target string, and determines whether or\nnot the string satisfies the constraints. See perlre for a full definition of the language.\n\nIn less grandiose terms, the first part of the job is to turn a pattern into something the\ncomputer can efficiently use to find the matching point in the string, and the second part is\nperforming the search itself.\n\nTo do this we need to produce a program by parsing the text. We then need to execute the\nprogram to find the point in the string that matches. And we need to do the whole thing\nefficiently.\n"
                },
                {
                    "name": "Structure of a Regexp Program",
                    "content": "High Level\n\nAlthough it is a bit confusing and some people object to the terminology, it is worth taking\na look at a comment that has been in regexp.h for years:\n\nThis is essentially a linear encoding of a nondeterministic finite-state machine (aka syntax\ncharts or \"railroad normal form\" in parsing technology).\n\nThe term \"railroad normal form\" is a bit esoteric, with \"syntax diagram/charts\", or \"railroad\ndiagram/charts\" being more common terms.  Nevertheless it provides a useful mental image of a\nregex program: each node can be thought of as a unit of track, with a single entry and in\nmost cases a single exit point (there are pieces of track that fork, but statistically not\nmany), and the whole forms a layout with a single entry and single exit point. The matching\nprocess can be thought of as a car that moves along the track, with the particular route\nthrough the system being determined by the character read at each possible connector point. A\ncar can fall off the track at any point but it may only proceed as long as it matches the\ntrack.\n\nThus the pattern \"/foo(?:\\w+|\\d+|\\s+)bar/\" can be thought of as the following chart:\n\n[start]\n|\n<foo>\n|\n+-----+-----+\n|     |     |\n<\\w+> <\\d+> <\\s+>\n|     |     |\n+-----+-----+\n|\n<bar>\n|\n[end]\n\nThe truth of the matter is that perl's regular expressions these days are much more complex\nthan this kind of structure, but visualising it this way can help when trying to get your\nbearings, and it matches the current implementation pretty closely.\n\nTo be more precise, we will say that a regex program is an encoding of a graph. Each node in\nthe graph corresponds to part of the original regex pattern, such as a literal string or a\nbranch, and has a pointer to the nodes representing the next component to be matched. Since\n\"node\" and \"opcode\" already have other meanings in the perl source, we will call the nodes in\na regex program \"regops\".\n\nThe program is represented by an array of \"regnode\" structures, one or more of which\nrepresent a single regop of the program. Struct \"regnode\" is the smallest struct needed, and\nhas a field structure which is shared with all the other larger structures.  (Outside this\ndocument, the term \"regnode\" is sometimes used to mean \"regop\", which could be confusing.)\n\nThe \"next\" pointers of all regops except \"BRANCH\" implement concatenation; a \"next\" pointer\nwith a \"BRANCH\" on both ends of it is connecting two alternatives.  [Here we have one of the\nsubtle syntax dependencies: an individual \"BRANCH\" (as opposed to a collection of them) is\nnever concatenated with anything because of operator precedence.]\n\nThe operand of some types of regop is a literal string; for others, it is a regop leading\ninto a sub-program.  In particular, the operand of a \"BRANCH\" node is the first regop of the\nbranch.\n\nNOTE: As the railroad metaphor suggests, this is not a tree structure:  the tail of the\nbranch connects to the thing following the set of \"BRANCH\"es.  It is a like a single line of\nrailway track that splits as it goes into a station or railway yard and rejoins as it comes\nout the other side.\n\nRegops\n\nThe base structure of a regop is defined in regexp.h as follows:\n\nstruct regnode {\nU8  flags;    /* Various purposes, sometimes overridden */\nU8  type;     /* Opcode value as specified by regnodes.h */\nU16 nextoff; /* Offset in size regnode */\n};\n\nOther larger \"regnode\"-like structures are defined in regcomp.h. They are almost like\nsubclasses in that they have the same fields as \"regnode\", with possibly additional fields\nfollowing in the structure, and in some cases the specific meaning (and name) of some of base\nfields are overridden. The following is a more complete description.\n\n\"regnode1\"\n\"regnode2\"\n\"regnode1\" structures have the same header, followed by a single four-byte argument;\n\"regnode2\" structures contain two two-byte arguments instead:\n\nregnode1                U32 arg1;\nregnode2                U16 arg1;  U16 arg2;\n\n\"regnodestring\"\n\"regnodestring\" structures, used for literal strings, follow the header with a one-byte\nlength and then the string data. Strings are padded on the tail end with zero bytes so\nthat the total length of the node is a multiple of four bytes:\n\nregnodestring           char string[1];\nU8 strlen; /* overrides flags */\n\n\"regnodecharclass\"\nBracketed character classes are represented by \"regnodecharclass\" structures, which have\na four-byte argument and then a 32-byte (256-bit) bitmap indicating which characters in\nthe Latin1 range are included in the class.\n\nregnodecharclass        U32 arg1;\nchar bitmap[ANYOFBITMAPSIZE];\n\nVarious flags whose names begin with \"ANYOF\" are used for special situations.  Above\nLatin1 matches and things not known until run-time are stored in \"Perl's pprivate\nstructure\".\n\n\"regnodecharclassposixl\"\nThere is also a larger form of a char class structure used to represent POSIX char\nclasses under \"/l\" matching, called \"regnodecharclassposixl\" which has an additional\n32-bit bitmap indicating which POSIX char classes have been included.\n\nregnodecharclassposixl U32 arg1;\nchar bitmap[ANYOFBITMAPSIZE];\nU32 classflags;\n\nregnodes.h defines an array called \"regarglen[]\" which gives the size of each opcode in units\nof \"size regnode\" (4-byte). A macro is used to calculate the size of an \"EXACT\" node based on\nits \"strlen\" field.\n\nThe regops are defined in regnodes.h which is generated from regcomp.sym by regcomp.pl.\nCurrently the maximum possible number of distinct regops is restricted to 256, with about a\nquarter already used.\n\nA set of macros makes accessing the fields easier and more consistent. These include \"OP()\",\nwhich is used to determine the type of a \"regnode\"-like structure; \"NEXTOFF()\", which is the\noffset to the next node (more on this later); \"ARG()\", \"ARG1()\", \"ARG2()\", \"ARGSET()\", and\nequivalents for reading and setting the arguments; and \"STRLEN()\", \"STRING()\" and\n\"OPERAND()\" for manipulating strings and regop bearing types.\n\nWhat regop is next?\n\nThere are three distinct concepts of \"next\" in the regex engine, and it is important to keep\nthem clear.\n\n•   There is the \"next regnode\" from a given regnode, a value which is rarely useful except\nthat sometimes it matches up in terms of value with one of the others, and that sometimes\nthe code assumes this to always be so.\n\n•   There is the \"next regop\" from a given regop/regnode. This is the regop physically\nlocated after the current one, as determined by the size of the current regop. This is\noften useful, such as when dumping the structure we use this order to traverse. Sometimes\nthe code assumes that the \"next regnode\" is the same as the \"next regop\", or in other\nwords assumes that the sizeof a given regop type is always going to be one regnode large.\n\n•   There is the \"regnext\" from a given regop. This is the regop which is reached by jumping\nforward by the value of \"NEXTOFF()\", or in a few cases for longer jumps by the \"arg1\"\nfield of the \"regnode1\" structure. The subroutine \"regnext()\" handles this\ntransparently.  This is the logical successor of the node, which in some cases, like that\nof the \"BRANCH\" regop, has special meaning.\n"
                },
                {
                    "name": "Process Overview",
                    "content": "Broadly speaking, performing a match of a string against a pattern involves the following\nsteps:\n\nA. Compilation\n1. Parsing\n2. Peep-hole optimisation and analysis\nB. Execution\n3. Start position and no-match optimisations\n4. Program execution\n\nWhere these steps occur in the actual execution of a perl program is determined by whether\nthe pattern involves interpolating any string variables. If interpolation occurs, then\ncompilation happens at run time. If it does not, then compilation is performed at compile\ntime. (The \"/o\" modifier changes this, as does \"qr//\" to a certain extent.) The engine\ndoesn't really care that much.\n"
                },
                {
                    "name": "Compilation",
                    "content": "This code resides primarily in regcomp.c, along with the header files regcomp.h, regexp.h and\nregnodes.h.\n\nCompilation starts with \"pregcomp()\", which is mostly an initialisation wrapper which farms\nwork out to two other routines for the heavy lifting: the first is \"reg()\", which is the\nstart point for parsing; the second, \"studychunk()\", is responsible for optimisation.\n\nInitialisation in \"pregcomp()\" mostly involves the creation and data-filling of a special\nstructure, \"RExCstatet\" (defined in regcomp.c).  Almost all internally-used routines in\nregcomp.h take a pointer to one of these structures as their first argument, with the name\n\"pRExCstate\".  This structure is used to store the compilation state and contains many\nfields. Likewise there are many macros which operate on this variable: anything that looks\nlike \"RExCxxxx\" is a macro that operates on this pointer/structure.\n\n\"reg()\" is the start of the parse process. It is responsible for parsing an arbitrary chunk\nof pattern up to either the end of the string, or the first closing parenthesis it encounters\nin the pattern.  This means it can be used to parse the top-level regex, or any section\ninside of a grouping parenthesis. It also handles the \"special parens\" that perl's regexes\nhave. For instance when parsing \"/x(?:foo)y/\", \"reg()\" will at one point be called to parse\nfrom the \"?\" symbol up to and including the \")\".\n\nAdditionally, \"reg()\" is responsible for parsing the one or more branches from the pattern,\nand for \"finishing them off\" by correctly setting their next pointers. In order to do the\nparsing, it repeatedly calls out to \"regbranch()\", which is responsible for handling up to\nthe first \"|\" symbol it sees.\n\n\"regbranch()\" in turn calls \"regpiece()\" which handles \"things\" followed by a quantifier. In\norder to parse the \"things\", \"regatom()\" is called. This is the lowest level routine, which\nparses out constant strings, character classes, and the various special symbols like \"$\". If\n\"regatom()\" encounters a \"(\" character it in turn calls \"reg()\".\n\nThere used to be two main passes involved in parsing, the first to calculate the size of the\ncompiled program, and the second to actually compile it.  But now there is only one main\npass, with an initial crude guess based on the length of the input pattern, which is\nincreased if necessary as parsing proceeds, and afterwards, trimmed to the actual amount\nused.\n\nHowever, it may happen that parsing must be restarted at the beginning when various\ncircumstances occur along the way.  An example is if the program turns out to be so large\nthat there are jumps in it that won't fit in the normal 16 bits available.  There are two\nspecial regops that can hold bigger jump destinations, BRANCHJ and LONGBRANCH.  The parse is\nrestarted, and these are used instead of the normal shorter ones.  Whenever restarting the\nparse is required, the function returns failure and sets a flag as to what needs to be done.\nThis is passed up to the top level routine which takes the appropriate action and restarts\nfrom scratch.  In the case of needing longer jumps, the \"RExCuseBRANCHJ\" flag is set in the\n\"RExCstatet\" structure, which the functions know to inspect before deciding how to do\nbranches.\n\nIn most instances, the function that discovers the issue sets the causal flag and returns\nfailure immediately.  \"Parsing complications\" contains an explicit example of how this works.\nIn other cases, such as a forward reference to a numbered parenthetical grouping, we need to\nfinish the parse to know if that numbered grouping actually appears in the pattern.  In those\ncases, the parse is just redone at the end, with the knowledge of how many groupings occur in\nit.\n\nThe routine \"regtail()\" is called by both \"reg()\" and \"regbranch()\" in order to \"set the tail\npointer\" correctly. When executing and we get to the end of a branch, we need to go to the\nnode following the grouping parens. When parsing, however, we don't know where the end will\nbe until we get there, so when we do we must go back and update the offsets as appropriate.\n\"regtail\" is used to make this easier.\n\nA subtlety of the parsing process means that a regex like \"/foo/\" is originally parsed into\nan alternation with a single branch. It is only afterwards that the optimiser converts single\nbranch alternations into the simpler form.\n\nParse Call Graph and a Grammar\n\nThe call graph looks like this:\n\nreg()                        # parse a top level regex, or inside of\n# parens\nregbranch()              # parse a single branch of an alternation\nregpiece()           # parse a pattern followed by a quantifier\nregatom()        # parse a simple pattern\nregclass()   #   used to handle a class\nreg()        #   used to handle a parenthesised\n#   subpattern\n....\n...\nregtail()            # finish off the branch\n...\nregtail()                # finish off the branch sequence. Tie each\n# branch's tail to the tail of the\n# sequence\n# (NEW) In Debug mode this is\n# regtailstudy().\n\nA grammar form might be something like this:\n\natom  : constant | class\nquant : '*' | '+' | '?' | '{min,max}'\nbranch: piece\n| piece branch\n| nothing\nbranch: branch\n| branch '|' branch\ngroup : '(' branch ')'\npiece: atom | group\npiece : piece\n| piece quant\n\nParsing complications\n\nThe implication of the above description is that a pattern containing nested parentheses will\nresult in a call graph which cycles through \"reg()\", \"regbranch()\", \"regpiece()\",\n\"regatom()\", \"reg()\", \"regbranch()\" etc multiple times, until the deepest level of nesting is\nreached. All the above routines return a pointer to a \"regnode\", which is usually the last\nregnode added to the program. However, one complication is that reg() returns NULL for\nparsing \"(?:)\" syntax for embedded modifiers, setting the flag \"TRYAGAIN\". The \"TRYAGAIN\"\npropagates upwards until it is captured, in some cases by \"regatom()\", but otherwise\nunconditionally by \"regbranch()\". Hence it will never be returned by \"regbranch()\" to\n\"reg()\". This flag permits patterns such as \"(?i)+\" to be detected as errors (Quantifier\nfollows nothing in regex; marked by <-- HERE in m/(?i)+ <-- HERE /).\n\nAnother complication is that the representation used for the program differs if it needs to\nstore Unicode, but it's not always possible to know for sure whether it does until midway\nthrough parsing. The Unicode representation for the program is larger, and cannot be matched\nas efficiently. (See \"Unicode and Localisation Support\" below for more details as to why.)\nIf the pattern contains literal Unicode, it's obvious that the program needs to store\nUnicode. Otherwise, the parser optimistically assumes that the more efficient representation\ncan be used, and starts sizing on this basis.  However, if it then encounters something in\nthe pattern which must be stored as Unicode, such as an \"\\x{...}\" escape sequence\nrepresenting a character literal, then this means that all previously calculated sizes need\nto be redone, using values appropriate for the Unicode representation.  This is another\ninstance where the parsing needs to be restarted, and it can and is done immediately.  The\nfunction returns failure, and sets the flag \"RESTARTUTF8\" (encapsulated by using the macro\n\"REQUIREUTF8\").  This restart request is propagated up the call chain in a similar fashion,\nuntil it is \"caught\" in \"Perlreopcompile()\", which marks the pattern as containing\nUnicode, and restarts the sizing pass. It is also possible for constructions within run-time\ncode blocks to turn out to need Unicode representation., which is signalled by\n\"Scompileruntimecode()\" returning false to \"Perlreopcompile()\".\n\nThe restart was previously implemented using a \"longjmp\" in \"regatom()\" back to a \"setjmp\" in\n\"Perlreopcompile()\", but this proved to be problematic as the latter is a large function\ncontaining many automatic variables, which interact badly with the emergent control flow of\n\"setjmp\".\n\nDebug Output\n\nStarting in the 5.9.x development version of perl you can \"use re Debug => 'PARSE'\" to see\nsome trace information about the parse process. We will start with some simple patterns and\nbuild up to more complex patterns.\n\nSo when we parse \"/foo/\" we see something like the following table. The left shows what is\nbeing parsed, and the number indicates where the next regop would go. The stuff on the right\nis the trace output of the graph. The names are chosen to be short to make it less dense on\nthe screen. 'tsdy' is a special form of \"regtail()\" which does some extra analysis.\n\n>foo<             1    reg\nbrnc\npiec\natom\n><                4      tsdy~ EXACT <foo> (EXACT) (1)\n~ attach to END (3) offset to 2\n\nThe resulting program then looks like:\n\n1: EXACT <foo>(3)\n3: END(0)\n\nAs you can see, even though we parsed out a branch and a piece, it was ultimately only an\natom. The final program shows us how things work. We have an \"EXACT\" regop, followed by an\n\"END\" regop. The number in parens indicates where the \"regnext\" of the node goes. The\n\"regnext\" of an \"END\" regop is unused, as \"END\" regops mean we have successfully matched. The\nnumber on the left indicates the position of the regop in the regnode array.\n\nNow let's try a harder pattern. We will add a quantifier, so now we have the pattern\n\"/foo+/\". We will see that \"regbranch()\" calls \"regpiece()\" twice.\n\n>foo+<            1    reg\nbrnc\npiec\natom\n>o+<              3        piec\natom\n><                6        tail~ EXACT <fo> (1)\n7      tsdy~ EXACT <fo> (EXACT) (1)\n~ PLUS (END) (3)\n~ attach to END (6) offset to 3\n\nAnd we end up with the program:\n\n1: EXACT <fo>(3)\n3: PLUS(6)\n4:   EXACT <o>(0)\n6: END(0)\n\nNow we have a special case. The \"EXACT\" regop has a \"regnext\" of 0. This is because if it\nmatches it should try to match itself again. The \"PLUS\" regop handles the actual failure of\nthe \"EXACT\" regop and acts appropriately (going to regnode 6 if the \"EXACT\" matched at least\nonce, or failing if it didn't).\n\nNow for something much more complex: \"/x(?:foo*|b[a][rR])(foo|bar)$/\"\n\n>x(?:foo*|b...    1    reg\nbrnc\npiec\natom\n>(?:foo*|b[...    3        piec\natom\n>?:foo*|b[a...                 reg\n>foo*|b[a][...                   brnc\npiec\natom\n>o*|b[a][rR...    5                piec\natom\n>|b[a][rR])...    8                tail~ EXACT <fo> (3)\n>b[a][rR])(...    9              brnc\n10                piec\natom\n>[a][rR])(f...   12                piec\natom\n>a][rR])(fo...                         clas\n>[rR])(foo|...   14                tail~ EXACT <b> (10)\npiec\natom\n>rR])(foo|b...                         clas\n>)(foo|bar)...   25                tail~ EXACT <a> (12)\ntail~ BRANCH (3)\n26              tsdy~ BRANCH (END) (9)\n~ attach to TAIL (25) offset to 16\ntsdy~ EXACT <fo> (EXACT) (4)\n~ STAR (END) (6)\n~ attach to TAIL (25) offset to 19\ntsdy~ EXACT <b> (EXACT) (10)\n~ EXACT <a> (EXACT) (12)\n~ ANYOF[Rr] (END) (14)\n~ attach to TAIL (25) offset to 11\n>(foo|bar)$<               tail~ EXACT <x> (1)\npiec\natom\n>foo|bar)$<                    reg\n28              brnc\npiec\natom\n>|bar)$<         31              tail~ OPEN1 (26)\n>bar)$<                          brnc\n32                piec\natom\n>)$<             34              tail~ BRANCH (28)\n36              tsdy~ BRANCH (END) (31)\n~ attach to CLOSE1 (34) offset to 3\ntsdy~ EXACT <foo> (EXACT) (29)\n~ attach to CLOSE1 (34) offset to 5\ntsdy~ EXACT <bar> (EXACT) (32)\n~ attach to CLOSE1 (34) offset to 2\n>$<                        tail~ BRANCH (3)\n~ BRANCH (9)\n~ TAIL (25)\npiec\natom\n><               37        tail~ OPEN1 (26)\n~ BRANCH (28)\n~ BRANCH (31)\n~ CLOSE1 (34)\n38      tsdy~ EXACT <x> (EXACT) (1)\n~ BRANCH (END) (3)\n~ BRANCH (END) (9)\n~ TAIL (END) (25)\n~ OPEN1 (END) (26)\n~ BRANCH (END) (28)\n~ BRANCH (END) (31)\n~ CLOSE1 (END) (34)\n~ EOL (END) (36)\n~ attach to END (37) offset to 1\n\nResulting in the program\n\n1: EXACT <x>(3)\n3: BRANCH(9)\n4:   EXACT <fo>(6)\n6:   STAR(26)\n7:     EXACT <o>(0)\n9: BRANCH(25)\n10:   EXACT <ba>(14)\n12:   OPTIMIZED (2 nodes)\n14:   ANYOF[Rr](26)\n25: TAIL(26)\n26: OPEN1(28)\n28:   TRIE-EXACT(34)\n[StS:1 Wds:2 Cs:6 Uq:5 #Sts:7 Mn:3 Mx:3 Stcls:bf]\n<foo>\n<bar>\n30:   OPTIMIZED (4 nodes)\n34: CLOSE1(36)\n36: EOL(37)\n37: END(0)\n\nHere we can see a much more complex program, with various optimisations in play. At regnode\n10 we see an example where a character class with only one character in it was turned into an\n\"EXACT\" node. We can also see where an entire alternation was turned into a \"TRIE-EXACT\"\nnode. As a consequence, some of the regnodes have been marked as optimised away. We can see\nthat the \"$\" symbol has been converted into an \"EOL\" regop, a special piece of code that\nlooks for \"\\n\" or the end of the string.\n\nThe next pointer for \"BRANCH\"es is interesting in that it points at where execution should go\nif the branch fails. When executing, if the engine tries to traverse from a branch to a\n\"regnext\" that isn't a branch then the engine will know that the entire set of branches has\nfailed.\n\nPeep-hole Optimisation and Analysis\n\nThe regular expression engine can be a weighty tool to wield. On long strings and complex\npatterns it can end up having to do a lot of work to find a match, and even more to decide\nthat no match is possible.  Consider a situation like the following pattern.\n\n'ababababababababababab' =~ /(a|b)*z/\n\nThe \"(a|b)*\" part can match at every char in the string, and then fail every time because\nthere is no \"z\" in the string. So obviously we can avoid using the regex engine unless there\nis a \"z\" in the string.  Likewise in a pattern like:\n\n/foo(\\w+)bar/\n\nIn this case we know that the string must contain a \"foo\" which must be followed by \"bar\". We\ncan use Fast Boyer-Moore matching as implemented in \"fbminstr()\" to find the location of\nthese strings. If they don't exist then we don't need to resort to the much more expensive\nregex engine.  Even better, if they do exist then we can use their positions to reduce the\nsearch space that the regex engine needs to cover to determine if the entire pattern matches.\n\nThere are various aspects of the pattern that can be used to facilitate optimisations along\nthese lines:\n\n•    anchored fixed strings\n\n•    floating fixed strings\n\n•    minimum and maximum length requirements\n\n•    start class\n\n•    Beginning/End of line positions\n\nAnother form of optimisation that can occur is the post-parse \"peep-hole\" optimisation, where\ninefficient constructs are replaced by more efficient constructs. The \"TAIL\" regops which are\nused during parsing to mark the end of branches and the end of groups are examples of this.\nThese regops are used as place-holders during construction and \"always match\" so they can be\n\"optimised away\" by making the things that point to the \"TAIL\" point to the thing that \"TAIL\"\npoints to, thus \"skipping\" the node.\n\nAnother optimisation that can occur is that of \"\"EXACT\" merging\" which is where two\nconsecutive \"EXACT\" nodes are merged into a single regop. An even more aggressive form of\nthis is that a branch sequence of the form \"EXACT BRANCH ... EXACT\" can be converted into a\n\"TRIE-EXACT\" regop.\n\nAll of this occurs in the routine \"studychunk()\" which uses a special structure\n\"scandatat\" to store the analysis that it has performed, and does the \"peep-hole\"\noptimisations as it goes.\n\nThe code involved in \"studychunk()\" is extremely cryptic. Be careful. :-)\n"
                },
                {
                    "name": "Execution",
                    "content": "Execution of a regex generally involves two phases, the first being finding the start point\nin the string where we should match from, and the second being running the regop interpreter.\n\nIf we can tell that there is no valid start point then we don't bother running the\ninterpreter at all. Likewise, if we know from the analysis phase that we cannot detect a\nshort-cut to the start position, we go straight to the interpreter.\n\nThe two entry points are \"reintuitstart()\" and \"pregexec()\". These routines have a somewhat\nincestuous relationship with overlap between their functions, and \"pregexec()\" may even call\n\"reintuitstart()\" on its own. Nevertheless other parts of the perl source code may call\ninto either, or both.\n\nExecution of the interpreter itself used to be recursive, but thanks to the efforts of Dave\nMitchell in the 5.9.x development track, that has changed: now an internal stack is\nmaintained on the heap and the routine is fully iterative. This can make it tricky as the\ncode is quite conservative about what state it stores, with the result that two consecutive\nlines in the code can actually be running in totally different contexts due to the simulated\nrecursion.\n\nStart position and no-match optimisations\n\n\"reintuitstart()\" is responsible for handling start points and no-match optimisations as\ndetermined by the results of the analysis done by \"studychunk()\" (and described in \"Peep-\nhole Optimisation and Analysis\").\n\nThe basic structure of this routine is to try to find the start- and/or end-points of where\nthe pattern could match, and to ensure that the string is long enough to match the pattern.\nIt tries to use more efficient methods over less efficient methods and may involve\nconsiderable cross-checking of constraints to find the place in the string that matches.  For\ninstance it may try to determine that a given fixed string must be not only present but a\ncertain number of chars before the end of the string, or whatever.\n\nIt calls several other routines, such as \"fbminstr()\" which does Fast Boyer Moore matching\nand \"findbyclass()\" which is responsible for finding the start using the first mandatory\nregop in the program.\n\nWhen the optimisation criteria have been satisfied, \"regtry()\" is called to perform the\nmatch.\n\nProgram execution\n\n\"pregexec()\" is the main entry point for running a regex. It contains support for\ninitialising the regex interpreter's state, running \"reintuitstart()\" if needed, and\nrunning the interpreter on the string from various start positions as needed. When it is\nnecessary to use the regex interpreter \"pregexec()\" calls \"regtry()\".\n\n\"regtry()\" is the entry point into the regex interpreter. It expects as arguments a pointer\nto a \"regmatchinfo\" structure and a pointer to a string.  It returns an integer 1 for\nsuccess and a 0 for failure.  It is basically a set-up wrapper around \"regmatch()\".\n\n\"regmatch\" is the main \"recursive loop\" of the interpreter. It is basically a giant switch\nstatement that implements a state machine, where the possible states are the regops\nthemselves, plus a number of additional intermediate and failure states. A few of the states\nare implemented as subroutines but the bulk are inline code.\n"
                }
            ]
        },
        "MISCELLANEOUS": {
            "content": "",
            "subsections": [
                {
                    "name": "Unicode and Localisation Support",
                    "content": "When dealing with strings containing characters that cannot be represented using an eight-bit\ncharacter set, perl uses an internal representation that is a permissive version of Unicode's\nUTF-8 encoding[2]. This uses single bytes to represent characters from the ASCII character\nset, and sequences of two or more bytes for all other characters. (See perlunitut for more\ninformation about the relationship between UTF-8 and perl's encoding, utf8. The difference\nisn't important for this discussion.)\n\nNo matter how you look at it, Unicode support is going to be a pain in a regex engine. Tricks\nthat might be fine when you have 256 possible characters often won't scale to handle the size\nof the UTF-8 character set.  Things you can take for granted with ASCII may not be true with\nUnicode. For instance, in ASCII, it is safe to assume that \"sizeof(char1) == sizeof(char2)\",\nbut in UTF-8 it isn't. Unicode case folding is vastly more complex than the simple rules of\nASCII, and even when not using Unicode but only localised single byte encodings, things can\nget tricky (for example, LATIN SMALL LETTER SHARP S (U+00DF, ß) should match 'SS' in\nlocalised case-insensitive matching).\n\nMaking things worse is that UTF-8 support was a later addition to the regex engine (as it was\nto perl) and this necessarily  made things a lot more complicated. Obviously it is easier to\ndesign a regex engine with Unicode support in mind from the beginning than it is to retrofit\nit to one that wasn't.\n\nNearly all regops that involve looking at the input string have two cases, one for UTF-8, and\none not. In fact, it's often more complex than that, as the pattern may be UTF-8 as well.\n\nCare must be taken when making changes to make sure that you handle UTF-8 properly, both at\ncompile time and at execution time, including when the string and pattern are mismatched.\n"
                },
                {
                    "name": "Base Structures",
                    "content": "The \"regexp\" structure described in perlreapi is common to all regex engines. Two of its\nfields are intended for the private use of the regex engine that compiled the pattern. These\nare the \"intflags\" and pprivate members. The \"pprivate\" is a void pointer to an arbitrary\nstructure whose use and management is the responsibility of the compiling engine. perl will\nnever modify either of these values. In the case of the stock engine the structure pointed to\nby \"pprivate\" is called \"regexpinternal\".\n\nIts \"pprivate\" and \"intflags\" fields contain data specific to each engine.\n\nThere are two structures used to store a compiled regular expression.  One, the \"regexp\"\nstructure described in perlreapi is populated by the engine currently being. used and some of\nits fields read by perl to implement things such as the stringification of \"qr//\".\n\nThe other structure is pointed to by the \"regexp\" struct's \"pprivate\" and is in addition to\n\"intflags\" in the same struct considered to be the property of the regex engine which\ncompiled the regular expression;\n\nThe regexp structure contains all the data that perl needs to be aware of to properly work\nwith the regular expression. It includes data about optimisations that perl can use to\ndetermine if the regex engine should really be used, and various other control info that is\nneeded to properly execute patterns in various contexts such as is the pattern anchored in\nsome way, or what flags were used during the compile, or whether the program contains special\nconstructs that perl needs to be aware of.\n\nIn addition it contains two fields that are intended for the private use of the regex engine\nthat compiled the pattern. These are the \"intflags\" and pprivate members. The \"pprivate\" is a\nvoid pointer to an arbitrary structure whose use and management is the responsibility of the\ncompiling engine. perl will never modify either of these values.\n\nAs mentioned earlier, in the case of the default engines, the \"pprivate\" will be a pointer to\na regexpinternal structure which holds the compiled program and any additional data that is\nprivate to the regex engine implementation.\n\nPerl's \"pprivate\" structure\n\nThe following structure is used as the \"pprivate\" struct by perl's regex engine. Since it is\nspecific to perl it is only of curiosity value to other engine implementations.\n\ntypedef struct regexpinternal {\nU32 *offsets;           /* offset annotations 20001228 MJD\n* data about mapping the program to\n* the string*/\nregnode *regstclass;    /* Optional startclass as identified or\n* constructed by the optimiser */\nstruct regdata *data;  /* Additional miscellaneous data used\n* by the program.  Used to make it\n* easier to clone and free arbitrary\n* data that the regops need. Often the\n* ARG field of a regop is an index\n* into this structure */\nregnode program[1];     /* Unwarranted chumminess with\n* compiler. */\n} regexpinternal;\n\n\"offsets\"\nOffsets holds a mapping of offset in the \"program\" to offset in the \"precomp\" string.\nThis is only used by ActiveState's visual regex debugger.\n\n\"regstclass\"\nSpecial regop that is used by \"reintuitstart()\" to check if a pattern can match at a\ncertain position. For instance if the regex engine knows that the pattern must start\nwith a 'Z' then it can scan the string until it finds one and then launch the regex\nengine from there. The routine that handles this is called \"findbyclass()\". Sometimes\nthis field points at a regop embedded in the program, and sometimes it points at an\nindependent synthetic regop that has been constructed by the optimiser.\n\n\"data\"\nThis field points at a \"regdata\" structure, which is defined as follows\n\nstruct regdata {\nU32 count;\nU8 *what;\nvoid* data[1];\n};\n\nThis structure is used for handling data structures that the regex engine needs to\nhandle specially during a clone or free operation on the compiled product. Each element\nin the data array has a corresponding element in the what array. During compilation\nregops that need special structures stored will add an element to each array using the\nadddata() routine and then store the index in the regop.\n\n\"program\"\nCompiled program. Inlined into the structure so the entire struct can be treated as a\nsingle blob.\n"
                }
            ]
        },
        "SEE ALSO": {
            "content": "perlreapi\n\nperlre\n\nperlunitut\n",
            "subsections": []
        },
        "AUTHOR": {
            "content": "by Yves Orton, 2006.\n\nWith excerpts from Perl, and contributions and suggestions from Ronald J. Kimball, Dave\nMitchell, Dominic Dunlop, Mark Jason Dominus, Stephen McCamant, and David Landgren.\n\nNow maintained by Perl 5 Porters.\n",
            "subsections": []
        },
        "LICENCE": {
            "content": "Same terms as Perl.\n",
            "subsections": []
        },
        "REFERENCES": {
            "content": "[1] <https://perl.plover.com/Rx/paper/>\n\n[2] <https://www.unicode.org/>\n\n\n\nperl v5.34.0                                 2025-07-25                                PERLREGUTS(1)",
            "subsections": []
        }
    },
    "summary": "perlreguts - Description of the Perl regular expression engine.",
    "flags": [],
    "examples": [],
    "see_also": []
}