re::engine::PCRE2 - PCRE2 regular expression engine with jit


    use re::engine::PCRE2;

    if ("Hello, world" =~ /(?<=Hello|Hi), (world)/) {
        print "Greetings, $1!";


Replaces perl's regex engine in a given lexical scope with PCRE2 regular expressions provided by libpcre2-8.

This provides jit support and faster matching, but may fail in corner cases. See pcre2compat. It is typically 50% faster than the core regex engine. See "BENCHMARKS".

The goal is to pass the full core re testsuite, identify all problematic patterns and fall-back to the core re engine. From the 1330 core tests, 46 currently fail. 90% of the most popular cpan modules do work fine already. Note that older perl version do fail more regression tests. See "FAILING TESTS".

Note that some packaged libpcre2-8 libraries do not enable the jit compiler. CFLAGS=-fPIC cmake -DPCRE2_SUPPORT_JIT=ON; make PCRE2 then silently falls back to the normal PCRE2 compiler and matcher.

Check with:

  perl -Mre::engine::PCRE2 -e'print re::engine::PCRE2::JIT'


Since re::engine::PCRE2 derives from the Regexp package, you can call compiled qr// objects with these methods. See PCRE2 NATIVE API MATCH CONTEXT FUNCTIONS and INFORMATION ABOUT A COMPILED PATTERN.

With older library versions which do not support a particular info method, undef is returned. E.g. hasbackslashc and framesize.

_alloptions (RX)

The result of pcre2_pattern_info(PCRE2_INFO_ALLOPTIONS) as unsigned integer.

   my $q=qr/(a)/; print $q->_alloptions
   => 64

64 stands for PCRE2_DUPNAMES which is always set. See pcre2.h

_argoptions (RX)

The result of pcre2_pattern_info(PCRE2_INFO_ARGOPTIONS) as unsigned integer.

   my $q=qr/(a)/i; print $q->_argoptions
   => 72

72 = 64+8 64 stands for PCRE2_DUPNAMES which is always set. 8 for PCRE2_CASELESS. See pcre2.h

backrefmax (RX)

Return the number of the highest back reference in the pattern.

  my $q=qr/(a)\1/; print $q->backrefmax
  => 1
  my $q=qr/(a)(?(1)a|b)/; print $q->backrefmax
  => 1
bsr (RX)

What character sequences the \R escape sequence matches. 1 means that \R matches any Unicode line ending sequence; 2 means that \R matches only CR, LF, or CRLF.

capturecount (RX)

Return the highest capturing subpattern number in the pattern. In patterns where (?| is not used, this is also the total number of capturing subpatterns.

  my $q=qr/(a(b))/; print $q->capturecount
  => 2
firstbitmap (RX)

In the absence of a single first code unit for a non-anchored pattern, pcre2_compile() may construct a 256-bit table that defines a fixed set of values for the first code unit in any match. For example, a pattern that starts with [abc] results in a table with three bits set. When code unit values greater than 255 are supported, the flag bit for 255 means "any code unit of value 255 or above". If such a table was constructed, it is returned as string.

firstcodetype (RX)

Return information about the first code unit of any matched string, for a non-anchored pattern. If there is a fixed first value, for example, the letter "c" from a pattern such as (cat|cow|coyote), 1 is returned, and the character value can be retrieved using "firstcodeunit". If there is no fixed first value, but it is known that a match can occur only at the start of the subject or following a newline in the subject, 2 is returned. Otherwise, and for anchored patterns, 0 is returned.

firstcodeunit (RX)

Return the value of the first code unit of any matched string in the situation where "firstcodetype (RX)" returns 1; otherwise return 0. The value is always less than 256.

  my $q=qr/(cat|cow|coyote)/; print $q->firstcodetype, $q->firstcodeunit
  => 1 99
framesize (RX)

Undocumented. Only available since pcre-10.24. Returns undef on older versions. The pcre2_match() frame size.

hasbackslashc (RX)

Return 1 if the pattern contains any instances of \C, otherwise 0. Note that \C is forbidden since perl 5.26 (?). With an older pcre2 library undef will be returned.

hascrorlf (RX)

Return 1 if the pattern contains any explicit matches for CR or LF characters, otherwise 0. An explicit match is either a literal CR or LF character, or \r or \n.

heaplimit (RX, [INT])

Get or set the backtracking heap limit in a match context. If the option is not set, build-time 'HEAPLIMIT' option is in effect, which is 20000000. See "config (OPTION)". Added only since 10.30, with earlier versions it will return undef. The setter method is not yet implemented.

jchanged (RX)

Return 1 if the (?J) or (?-J) option setting is used in the pattern, otherwise 0. (?J) and (?-J) set and unset the local PCRE2_DUPNAMES option, respectively.

jitsize (RX)

If the compiled pattern was successfully processed by pcre2_jit_compile(), return the size of the JIT compiled code, otherwise return zero.

lastcodetype (RX)

Returns 1 if there is a rightmost literal code unit that must exist in any matched string, other than at its start. If there is no such value, 0 is returned. When 1 is returned, the code unit value itself can be retrieved using "lastcodeunit (RX)". For anchored patterns, a last literal value is recorded only if it follows something of variable length. For example, for the pattern /^a\d+z\d+/ the returned value is 1 (with "z" returned from lastcodeunit), but for /^a\dz\d/ the returned value is 0.

lastcodeunit (RX)

Return the value of the rightmost literal data unit that must exist in any matched string, other than at its start, if such a value has been recorded. If there is no such value, 0 is returned.

matchempty (RX)

Return 1 if the pattern might match an empty string, otherwise 0. When a pattern contains recursive subroutine calls it is not always possible to determine whether or not it can match an empty string. PCRE2 takes a cautious approach and returns 1 in such cases.

matchlimit (RX, [INT])

Get or set the match_limit match context. Corresponds to the pcre-specific (*LIMIT_MATCH=nnnn) option. If the option is not set, build-time 'MATCHLIMIT' option is in effect, which is 10000000. See "config (OPTION)".

maxlookbehind (RX)

Return the number of characters (not code units) in the longest lookbehind assertion in the pattern. This information is useful when doing multi-segment matching using the partial matching facilities. Note that the simple assertions \b and \B require a one-character lookbehind. \A also registers a one-character lookbehind, though it does not actually inspect the previous character. This is to ensure that at least one character from the old segment is retained when a new segment is processed. Otherwise, if there are no lookbehinds in the pattern, \A might match incorrectly at the start of a new segment.

minlength (RX)

If a minimum length for matching subject strings was computed, its value is returned. Otherwise the returned value is 0. The value is a number of characters, which in UTF mode may be different from the number of code units. The value is a lower bound to the length of any matching string. There may not be any strings of that length that do actually match, but every string that does match is at least that long.

namecount (RX)
nameentrysize (RX)

PCRE2 supports the use of named as well as numbered capturing parentheses. The names are just an additional way of identifying the parentheses, which still acquire numbers. Several convenience functions such as pcre2_substring_get_byname() are provided for extracting captured substrings by name. It is also possible to extract the data directly, by first converting the name to a number in order to access the correct pointers in the output vector. To do the conversion, you need to use the name-to-number map, which is described by these three values.

The map consists of a number of fixed-size entries. namecount gives the number of entries, and nameentrysize gives the size of each entry in code units; The entry size depends on the length of the longest name.

The nametable itself is not yet returned.

newline (RX, [INT]))

Get or set the newline regime. The default is the build-time 'NEWLINE' option, i.e. 2 on non-windows systems. See "config (OPTION)". The setter method is not yet implemented.

recursionlimit (RX, [INT])

Get or set a recursion limit, i.e. the pcre specific (*LIMIT_RECURSION=nnnn) option. The default is the build-time 'RECURSIONLIMIT' option. See "config (OPTION)". The setter method is not yet implemented.

size (RX)

Return the size of the compiled pattern in bytes. This value includes the size of the general data block that precedes the code units of the compiled pattern itself. The value that is used when pcre2_compile() is getting memory in which to place the compiled pattern may be slightly larger than the value returned by this option, because there are cases where the code that calculates the size has to over-estimate. Processing a pattern with the JIT compiler does not alter the value returned by this option.



import lexically sets the PCRE2 engine to be active.

import will later accept compile context options. See PCRE2 NATIVE API COMPILE CONTEXT FUNCTIONS.

  bsr            => INT (default: 1)
  max_pattern_length => INT
  newline        => INT (default: 2)
  parenslimit    => INT (default: 250)
  matchlimit     => INT (default: 10000000)
  offsetlimit    => INT (default: ?)
  recursionlimit => INT (default: 10000000) i.e. the depthlimit
  heaplimit      => INT (default: 20000000) ony since 10.30

unimport sets the regex engine to the previous one. If PCRE2 with the previous context options.

offsetlimit ([INT])

Get or set the offset_limit in the match context. The method is not yet implemented.

parenslimit ([INT])

Get or set the parens_nest_limit in the match context. The default is the build-time 'PARENSLIMIT' option, 250. See "config (OPTION)". The method is not yet implemented.


Returns a pointer to the internal PCRE2 engine, suitable for the XS API (regexp*)re->engine field.


Returns 1 or 0, if the JIT engine is available or not.

config (OPTION)

Returns build-time information about libpcre2. Note that some of these options may later be set'able at run-time.

OPTIONS can be one of the following strings:


    HEAPLIMIT       (Only since 10.30)
    DEPTHLIMIT      (Not always defined)
    RECURSIONLIMIT  (Obsolete synonym for DEPTHLIMIT)
    STACKRECURSE    (Obsolete. Always 0 in newer libs)

The first three options return a string, the rest an integer. In case of internal errors, e.g. the new option is not yet supported by libpcre, undef is returned. See

NEWLINE returns an integer, representing:

   PCRE2_NEWLINE_CR          1
   PCRE2_NEWLINE_LF          2
   PCRE2_NEWLINE_CRLF        3
   PCRE2_NEWLINE_ANY         4  Any Unicode line ending
   PCRE2_NEWLINE_ANYCRLF     5  Any of CR, LF, or CRLF

The default is OS specific.

BSR returns an integer, representing:

   PCRE2_BSR_UNICODE         1
   PCRE2_BSR_ANYCRLF         2

A value of PCRE2_BSR_UNICODE means that \R matches any Unicode line ending sequence; a value of PCRE2_BSR_ANYCRLF means that \R matches only CR, LF, or CRLF.

The default is 1 for UNICODE, as all libpcre2 libraries are now compiled with unicode support builtin. (--enable-unicode).


    time perl5.24.1 -Mblib t/perl/regexp.t 10000 >/dev/null

Without PCRE2:


With PCRE2:

    17.922s - 50% faster


About 90% of all core tests and cpan modules do work with re::engine::PCRE2 already, but there are still some unresolved problems. Esp. when the pattern is not detectable or marked as UTF-8 but the subject is, the match will be performed without UTF-8.

Try the new faster matcher with export PERL5OPT=-Mre::engine::PCRE2.

Known problematic popular modules are: Test-Harness-3.38, Params-Util-1.07 t/12_main.t 552-553, 567-568, HTML-Parser-3.72 (unicode), DBI-1.636 (EUMM problem), DBD-SQLite-1.54 (xsubpp), Sub-Name-0.21 t/exotic_names.t:105, XML-LibXML-2.0129 (local charset), Module-Install-1.18 unrecognized character after (? or (?-, Text-CSV_XS-1.28 (unicode), YAML-Syck-1.29, MD5-2.03, XML-Parser-2.44, Module-Build-0.4222, libwww-perl-6.25.

As of 0.05 the following core regression tests still fail:

    perl -C -Mblib t/perl/regexp.t | grep -a TODO

    # new patterns and pcre2 fails: need to fallback
    143..146, # \B{gcb} \B{lb} \B{sb} \B{wb}
    352,      # '^'i:ABC:y:$&:
    402,      # '(a+|b){0,1}?'i
    409,      # 'a*'i $&
    578,      # '(b.)c(?!\N)'s:a
    654,655,664, # unicode
    667,      # '[[:^cntrl:]]+'u:a\x80:y:$&:a

    # Pathological patterns that run into run-time PCRE_ERROR_MATCHLIMIT,
    # even with huge set_match_limit 512mill
    880 .. 897, # .X(.+)+[X][X]:bbbbXXXaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa

    # aba =~ ^(a(b)?)+$ and aabbaa =~ ^(aa(bb)?)+$
    941, # ^(a(b)?)+$:aba:y:-$1-$2-:-a-- => `-a-b-', match=1
    942, # ^(aa(bb)?)+$:aabbaa:y:-$1-$2-:-aa-- => `-aa-bb-', match=1
    947, # ^(a\1?){4}$:aaaaaa:y:$1:aa => `', match=

    # empty codeblock
    1005, #TODO (??{}):x:y:-:- => error `Eval-group not allowed at runtime, use re 'eval' in regex m/(??{})/ at (eval 5663) line 1.'

    # XXX: <<<>>> pattern
    1096, # ^(<(?:[^<>]+|(?3)|(?1))*>)()(!>!>!>)$:<<!>!>!>><>>!>!>!>:y:$1:<<!>!>!>><>> => `', match=
    1126, # /^(?'main'<(?:[^<>]+|(?&crap)|(?&main))*>)(?'empty')(?'crap'!>!>!>)$/:<<!>!>!>><>>!>!>!>:yM:$+{main}:<<!>!>!>><>> => `', match=

    # XXX: \R doesn't match an utf8::upgraded \x{85}, we need to
    # always convert the subject and pattern to utf-8 for these cases
    # to work
    1378, # (utf8::upgrade($subject)) foo(\R+)bar:foo\r
    1380, # (utf8::upgrade($subject)) (\R+)(\V):foo\r
    1381, # (utf8::upgrade($subject)) foo(\R)bar:foo\x{85}bar:y:$1:\x{85} => `', match=
    1382, # (utf8::upgrade($subject)) (\V)(\R):foo\x{85}bar:y:$1-$2:o-\x{85} => `�-�', match=1
    1394, # (utf8::upgrade($subject)) foo(\v+)bar:foo\r
    1396..1398, # (utf8::upgrade($subject)) (\v+)(\V):foo\r
    1405,1407..1409, # (utf8::upgrade($subject)) foo(\h+)bar:foo\t\x{A0}bar:y:$1:\t\x{A0} => `', match=

    # regressions in 5.8.x (only) introduced by change 30638
    1433, # /^\s*i.*?o\s*$/s:io
    1446, #/\N{}\xe4/i:\xc4:y:$&:\xc4 => error `Unknown charname '' is deprecated. Its use will be fatal in Perl 5.28 at (eval 7892) line 2.'
    1484, # /abc\N {U+41}/x:-:c:-:Missing braces => `-', match=
    1485, # /abc\N {SPACE}/x:-:c:-:Missing braces => `-', match=
    1490, # /\N{U+BEEF.BEAD}/:-:c:-: => `-', match=
    1495, # \c`:-:ac:-:\"\\c`\" is more clearly written simply as \"\\ \" => `-', match=
    1496, # \c1:-:ac:-:\"\\c1\" is more clearly written simply as \"q\" => `-', match=
    1514, # \c?:\x9F:ey:$&:\x9F => `\', match=
    1575, # [\8\9]:\000:Sn:-:- => `-', match=
    1576, # [\8\9]:-:sc:$&:Unrecognized escape \\8 in character class => `[', match=
    1582, # [\0]:-:sc:-:Need exactly 3 octal digits => `-', match=
    1584, # [\07]:-:sc:-:Need exactly 3 octal digits => `-', match=
    1585, # [\07]:7\000:Sn:-:- => `-', match=
    1586, # [\07]:-:sc:-:Need exactly 3 octal digits => `-', match=
    1599, # /\xe0\pL/i:\xc0a:y:$&:\xc0a => `/', match=
    1618, # ^_?[^\W_0-9]\w\z:\xAA\x{100}:y:$&:\xAA\x{100} => `^', match=
    1621, # /s/ai:\x{17F}:y:$&:\x{17F} => `/', match=
    1630, # /[^\x{1E9E}]/i:\x{DF}:Sn:-:- => `-', match=
    1639, # /^\p{L}/:\x{3400}:y:$&:\x{3400} => `�', match=1
    1642, # /[s\xDF]a/ui:ssa:Sy:$&:ssa => `sa', match=1
    1648, # /ff/i:\x{FB00}\x{FB01}:y:$&:\x{FB00} => `/', match=
    1649, # /ff/i:\x{FB01}\x{FB00}:y:$&:\x{FB00} => `/', match=
    1650, # /fi/i:\x{FB01}\x{FB00}:y:$&:\x{FB01} => `/', match=
    1651, # /fi/i:\x{FB00}\x{FB01}:y:$&:\x{FB01} => `/', match=

    # These test that doesn't cut-off matching too soon in the string for
    # multi-char folds
    1669, # /ffiffl/i:abcdef\x{FB03}\x{FB04}:y:$&:\x{FB03}\x{FB04} => `/', match=
    1670, # /\xdf\xdf/ui:abcdefssss:y:$&:ssss => `/', match=
    1672, # /st/i:\x{DF}\x{FB05}:y:$&:\x{FB05} => `/', match=
    1673, # /ssst/i:\x{DF}\x{FB05}:y:$&:\x{DF}\x{FB05} => `/', match=
    # [perl #101970]
    1678, # /[[:lower:]]/i:\x{100}:y:$&:\x{100} => `/', match=
    1679, # /[[:upper:]]/i:\x{101}:y:$&:\x{101} => `/', match=
    # Was matching 'ss' only and failing the entire match, not seeing the
    # alternative that would succeed
    1683, # /s\xDF/ui:\xDFs:y:$&:\xDFs => `/', match=
    1684, # /sst/ui:s\N{LATIN SMALL LIGATURE ST}:y:$&:s\N{LATIN SMALL LIGATURE ST} => `/', match=
    1685, # /sst/ui:s\N{LATIN SMALL LIGATURE LONG S T}:y:$&:s\N{LATIN SMALL LIGATURE LONG S T} => `/', match=
    # [perl #111400].  Tests the first Y/N boundary above 255 for each of these.
    1699, # /[[:alnum:]]/:\x{2c1}:y:-:- => `-', match=
    1701, # /[[:alpha:]]/:\x{2c1}:y:-:- => `-', match=
    1703, # /[[:graph:]]/:\x{377}:y:-:- => `-', match=
    1706, # /[[:lower:]]/:\x{101}:y:-:- => `-', match=
    1708, # /[[:print:]]/:\x{377}:y:-:- => `-', match=
    1711, # /[[:punct:]]/:\x{37E}:y:-:- => `-', match=
    1713, # /[[:upper:]]/:\x{100}:y:-:- => `-', match=
    1715, # /[[:word:]]/:\x{2c1}:y:-:- => `-', match=

    # $^N, $+ on backtrackracking
    # BRANCH
    1739, # ^(.)(?:(..)|B)[CX]:ABCDE:y:$^N-$+:A-A => `-', match=1
    # TRIE
    1741, # ^(.)(?:BC(.)|B)[CX]:ABCDE:y:$^N-$+:A-A => `-', match=1
    # CURLYX
    1743, # ^(.)(?:(.)+)*[BX]:ABCDE:y:$^N-$+:A-A => `-', match=1
    # CURLYM
    1746, # ^(.)(BC)*[BX]:ABCDE:y:$^N-$+:A-A => `-', match=1
    # CURLYN
    1749, # ^(.)(B)*.[CX]:ABCDE:y:$^N-$+:A-A => `-', match=1

    # [perl #114220]
    1793, # (utf8::upgrade($subject)) /[\H]/:\x{BF}:y:$&:\xBF => `�', match=1
    1794, # (utf8::upgrade($subject)) /[\H]/:\x{A0}:n:-:- => false positive
    1795, # (utf8::upgrade($subject)) /[\H]/:\x{A1}:y:$&:\xA1 => `�', match=1

    # \W in pattern -> !UTF8: add UTF if subject is UTF8 [#15]
    1804..1807, # \w:\x{200C}:y:$&:\x{200C} => `\', match=
    #1805, # \W:\x{200C}:n:-:- => false positive
    #1806, # \w:\x{200D}:y:$&:\x{200D} => `\', match=
    #1807, # \W:\x{200D}:n:-:- => false positive
    # again missing UTF [#15]
    1818..1820, # /^\D{11}/a:\x{10FFFF}\x{10FFFF}\x{10FFFF}\x{10FFFF}\x{10FFFF}\x{10FFFF}\x{10FFFF}\x{10FFFF}\x{10FFFF}\x{10FFFF}:n:-:- => false positive
    1823, # (utf8::upgrade($subject)) \Vn:\xFFn/:y:$&:\xFFn => `�n', match=1
    1830, # a?\X:a\x{100}:y:$&:a\x{100} => `a�', match=1
    1892, # /^\S+=/d:\x{3a3}=\x{3a0}:y:$&:\x{3a3}= => `Σ=', match=1
    1893, # /^\S+=/u:\x{3a3}=\x{3a0}:y:$&:\x{3a3}= => `Σ=', match=1
    1936, # /[a-z]/i:\N{KELVIN SIGN}:y:$&:\N{KELVIN SIGN} => `/', match=
    1937, # /[A-Z]/ia:\N{KELVIN SIGN}:y:$&:\N{KELVIN SIGN} => `/', match=
    1939, # /[A-Z]/i:\N{LATIN SMALL LETTER LONG S}:y:$&:\N{LATIN SMALL LETTER LONG S} => `/', match=
    1964, # \N(?#comment){SPACE}:A:c:-:Missing braces on \\N{} => `-', match=
    1983, # /(?xx:[a b])/x:\N{SPACE}:n:-:- => false positive
    1985, # /(?xx)[a b]/x:\N{SPACE}:n:-:- => false positive

    # [perl #125825]
    1945, # /(a+){1}+a/:aaa:n:-:- => false positive
    # [perl 128420] recursive matches
    1976, # aa$|a(?R)a|a:aaa:y:$&:aaa => `a', match=1

Note that core tests suite also reveals that about a similar number of fails occur with older perls, without PCRE2. Many of them pass with PCRE2.

Failures in older perls:

    -5.12:  629, 1367 (fatal)
    -5.10:  40..51, 90..91, 93..94, 96..97, 105, 356, 539,
            541, 543, 577, 1360, 1416, 1418, 1456..1457,
    -5.12:  1448, 1521, 1524, 1577..1578, 1594..1596,
            1598, 1674..1675
    -5.14:  1633..1634
    -5.16:  871, 1745, 1789, 1816
    -5.18:  1674..1675, 1856..1857, 1885..1886, 1889
    -5.20:  138..142
    -5.22:  139, 1958, 1965
    -5.24:  1977

Invalid tests for older perls (fatal):

    -5.14: 1684..1996
    -5.20: 1896..1996
    -5.26: 1981..1996


Reini Urban <>


Copyright 2007 Ævar Arnfjörð Bjarmason. Copyright 2017 Reini Urban.

The original version was copyright 2006 Audrey Tang <> and Yves Orton.

This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.