3 # pragma for controlling the regexp engine
8 our @ISA = qw(Exporter);
9 our @EXPORT_OK = ('regmust',
10 qw(is_regexp regexp_pattern
11 regname regnames regnames_count));
12 our %EXPORT_OK = map { $_ => 1 } @EXPORT_OK;
15 taint => 0x00100000, # HINT_RE_TAINT
16 eval => 0x00200000, # HINT_RE_EVAL
19 my $flags_hint = 0x02000000; # HINT_RE_FLAGS
22 m => 1 << ($PMMOD_SHIFT + 0),
23 s => 1 << ($PMMOD_SHIFT + 1),
24 i => 1 << ($PMMOD_SHIFT + 2),
25 x => 1 << ($PMMOD_SHIFT + 3),
26 n => 1 << ($PMMOD_SHIFT + 5),
27 p => 1 << ($PMMOD_SHIFT + 6),
28 strict => 1 << ($PMMOD_SHIFT + 10),
38 eval { # Ignore errors
41 my $terminal = Tgetent Term::Cap ({OSPEED => 9600}); # Avoid warning.
42 my $props = $ENV{PERL_RE_TC} || 'md,me,so,se,us,ue';
43 my @props = split /,/, $props;
44 my $colors = join "\t", map {$terminal->Tputs($_,1)} @props;
47 $ENV{PERL_RE_COLORS} = $colors;
50 $ENV{PERL_RE_COLORS} ||= qq'\t\t> <\t> <\t\t';
72 OFFSETSDBG => 0x040000,
74 OPTIMISEM => 0x100000,
79 $flags{ALL} = -1 & ~($flags{OFFSETS}|$flags{OFFSETSDBG}|$flags{BUFFERS});
80 $flags{All} = $flags{all} = $flags{DUMP} | $flags{EXECUTE};
81 $flags{Extra} = $flags{EXECUTE} | $flags{COMPILE} | $flags{GPOS};
82 $flags{More} = $flags{MORE} = $flags{All} | $flags{TRIEC} | $flags{TRIEM} | $flags{STATE};
83 $flags{State} = $flags{DUMP} | $flags{EXECUTE} | $flags{STATE};
84 $flags{TRIE} = $flags{DUMP} | $flags{EXECUTE} | $flags{TRIEC};
86 if (defined &DynaLoader::boot_DynaLoader) {
91 # We need to work for miniperl, because the XS toolchain uses Text::Wrap, which
97 # We call install() every time, as if we didn't, we wouldn't
98 # "see" any changes to the color environment var since
99 # the last time it was called.
101 # install() returns an integer, which if casted properly
102 # in C resolves to a structure containing the regexp
103 # hooks. Setting it to a random integer will guarantee
105 $^H{regcomp} = install();
114 my $turning_all_off = ! @_ && ! $on;
115 my %seen; # Has flag already been seen?
116 if ($turning_all_off) {
118 # Pretend were called with certain parameters, which are best dealt
120 push @_, keys %bitmask; # taint and eval
124 # Process each subpragma parameter
126 foreach my $idx (0..$#_){
128 if ($s eq 'Debug' or $s eq 'Debugcolor') {
129 setcolor() if $s =~/color/i;
130 ${^RE_DEBUG_FLAGS} = 0 unless defined ${^RE_DEBUG_FLAGS};
131 for my $idx ($idx+1..$#_) {
132 if ($flags{$_[$idx]}) {
134 ${^RE_DEBUG_FLAGS} |= $flags{$_[$idx]};
136 ${^RE_DEBUG_FLAGS} &= ~ $flags{$_[$idx]};
140 Carp::carp("Unknown \"re\" Debug flag '$_[$idx]', possible flags: ",
141 join(", ",sort keys %flags ) );
144 _load_unload($on ? 1 : ${^RE_DEBUG_FLAGS});
146 } elsif ($s eq 'debug' or $s eq 'debugcolor') {
147 setcolor() if $s =~/color/i;
150 } elsif (exists $bitmask{$s}) {
151 $bits |= $bitmask{$s};
152 } elsif ($EXPORT_OK{$s}) {
154 re->export_to_level(2, 're', $s);
155 } elsif ($s eq 'strict') {
157 $^H{reflags} |= $reflags{$s};
158 warnings::warnif('experimental::re_strict',
159 "\"use re 'strict'\" is experimental");
161 # Turn on warnings if not already done.
162 if (! warnings::enabled('regexp')) {
164 warnings->import('regexp');
169 $^H{reflags} &= ~$reflags{$s} if $^H{reflags};
171 # Turn off warnings if we turned them on.
172 warnings->unimport('regexp') if $^H{re_strict};
180 } elsif ($s =~ s/^\///) {
181 my $reflags = $^H{reflags} || 0;
183 while ($s =~ m/( . )/gx) {
186 # The 'a' may be repeated; hide this from the rest of the
187 # code by counting and getting rid of all of them, then
188 # changing to 'aa' if there is a repeat.
190 my $sav_pos = pos $s;
191 my $a_count = $s =~ s/a//g;
192 pos $s = $sav_pos - 1; # -1 because got rid of the 'a'
196 qq 'The "a" flag may only appear a maximum of twice'
199 elsif ($a_count == 2) {
206 if ($seen_charset ne $_) {
208 qq 'The "$seen_charset" and "$_" flags '
214 qq 'The "$seen_charset" flag may not appear '
219 $^H{reflags_charset} = $reflags{$_};
223 delete $^H{reflags_charset}
224 if defined $^H{reflags_charset}
225 && $^H{reflags_charset} == $reflags{$_};
227 } elsif (exists $reflags{$_}) {
230 ? $reflags |= $reflags{$_}
231 : ($reflags &= ~$reflags{$_});
235 qq'Unknown regular expression flag "$_"'
240 ($^H{reflags} = $reflags or defined $^H{reflags_charset})
242 : ($^H &= ~$flags_hint);
245 Carp::carp("Unknown \"re\" subpragma '$s' (known ones are: ",
246 join(', ', map {qq('$_')} 'debug', 'debugcolor', sort keys %bitmask),
250 if (exists $seen{'x'} && $seen{'x'} > 1
251 && (warnings::enabled("deprecated")
252 || warnings::enabled("regexp")))
254 my $message = "Having more than one /x regexp modifier is deprecated";
255 if (warnings::enabled("deprecated")) {
256 warnings::warn("deprecated", $message);
259 warnings::warn("regexp", $message);
263 if ($turning_all_off) {
266 $^H{reflags_charset} = 0;
280 $^H &= ~ bits(0, @_);
289 re - Perl pragma to alter regular expression behaviour
294 ($x) = ($^X =~ /^(.*)$/s); # $x is tainted here
296 $pat = '(?{ $foo = 1 })';
298 /foo${pat}bar/; # won't fail (when not under -T
302 no re 'taint'; # the default
303 ($x) = ($^X =~ /^(.*)$/s); # $x is not tainted here
305 no re 'eval'; # the default
306 /foo${pat}bar/; # disallowed (with or without -T
310 use re 'strict'; # Raise warnings for more conditions
313 "FOO" =~ / foo /; # /ix implied
315 "FOO" =~ /foo/; # just /i implied
317 use re 'debug'; # output debugging info during
318 /^(.*)$/s; # compile and run time
321 use re 'debugcolor'; # same as 'debug', but with colored
325 use re qw(Debug All); # Same as "use re 'debug'", but you
326 # can use "Debug" with things other
328 use re qw(Debug More); # 'All' plus output more details
329 no re qw(Debug ALL); # Turn on (almost) all re debugging
332 use re qw(is_regexp regexp_pattern); # import utility functions
333 my ($pat,$mods)=regexp_pattern(qr/foo/i);
334 if (is_regexp($obj)) {
335 print "Got regexp: ",
336 scalar regexp_pattern($obj); # just as perl would stringify
337 } # it but no hassle with blessed
340 (We use $^X in these examples because it's tainted by default.)
346 When C<use re 'taint'> is in effect, and a tainted string is the target
347 of a regexp, the regexp memories (or values returned by the m// operator
348 in list context) are tainted. This feature is useful when regexp operations
349 on tainted data aren't meant to extract safe substrings, but to perform
350 other transformations.
354 When C<use re 'eval'> is in effect, a regexp is allowed to contain
355 C<(?{ ... })> zero-width assertions and C<(??{ ... })> postponed
356 subexpressions that are derived from variable interpolation, rather than
357 appearing literally within the regexp. That is normally disallowed, since
359 potential security risk. Note that this pragma is ignored when the regular
360 expression is obtained from tainted data, i.e. evaluation is always
361 disallowed with tainted regular expressions. See L<perlre/(?{ code })>
362 and L<perlre/(??{ code })>.
364 For the purpose of this pragma, interpolation of precompiled regular
365 expressions (i.e., the result of C<qr//>) is I<not> considered variable
370 I<is> allowed if $pat is a precompiled regular expression, even
371 if $pat contains C<(?{ ... })> assertions or C<(??{ ... })> subexpressions.
375 Note that this is an experimental feature which may be changed or removed in a
378 When C<use re 'strict'> is in effect, stricter checks are applied than
379 otherwise when compiling regular expressions patterns. These may cause more
380 warnings to be raised than otherwise, and more things to be fatal instead of
381 just warnings. The purpose of this is to find and report at compile time some
382 things, which may be legal, but have a reasonable possibility of not being the
383 programmer's actual intent. This automatically turns on the C<"regexp">
384 warnings category (if not already on) within its scope.
386 As an example of something that is caught under C<"strict'>, but not
387 otherwise, is the pattern
391 The C<"\x"> construct without curly braces should be followed by exactly two
392 hex digits; this one is followed by three. This currently evaluates as
397 that is, the character whose code point value is C<0xAB>, followed by the
398 letter C<C>. But since C<C> is a a hex digit, there is a reasonable chance
403 that is the single character at C<0xABC>. Under C<'strict'> it is an error to
404 not follow C<\x> with exactly two hex digits. When not under C<'strict'> a
405 warning is generated if there is only one hex digit, and no warning is raised
406 if there are more than two.
408 It is expected that what exactly C<'strict'> does will evolve over time as we
409 gain experience with it. This means that programs that compile under it in
410 today's Perl may not compile, or may have more or fewer warnings, in future
411 Perls. There is no backwards compatibility promises with regards to it. Also
412 there are already proposals for an alternate syntax for enabling it. For
413 these reasons, using it will raise a C<experimental::re_strict> class warning,
414 unless that category is turned off.
416 Note that if a pattern compiled within C<'strict'> is recompiled, say by
417 interpolating into another pattern, outside of C<'strict'>, it is not checked
418 again for strictness. This is because if it works under strict it must work
423 When C<use re '/flags'> is specified, the given flags are automatically
424 added to every regular expression till the end of the lexical scope.
426 C<no re '/flags'> will turn off the effect of C<use re '/flags'> for the
429 For example, if you want all your regular expressions to have /msx on by
434 at the top of your code.
436 The character set /adul flags cancel each other out. So, in this example,
443 the second C<use re> does an implicit C<no re '/u'>.
445 Turning on one of the character set flags with C<use re> takes precedence over the
446 C<locale> pragma and the 'unicode_strings' C<feature>, for regular
447 expressions. Turning off one of these flags when it is active reverts to
448 the behaviour specified by whatever other pragmata are in scope. For
451 use feature "unicode_strings";
452 no re "/u"; # does nothing
454 no re "/l"; # reverts to unicode_strings behaviour
458 When C<use re 'debug'> is in effect, perl emits debugging messages when
459 compiling and using regular expressions. The output is the same as that
460 obtained by running a C<-DDEBUGGING>-enabled perl interpreter with the
461 B<-Dr> switch. It may be quite voluminous depending on the complexity
462 of the match. Using C<debugcolor> instead of C<debug> enables a
463 form of output that can be used to get a colorful display on terminals
464 that understand termcap color sequences. Set C<$ENV{PERL_RE_TC}> to a
465 comma-separated list of C<termcap> properties to use for highlighting
466 strings on/off, pre-point part on/off.
467 See L<perldebug/"Debugging Regular Expressions"> for additional info.
469 As of 5.9.5 the directive C<use re 'debug'> and its equivalents are
470 lexically scoped, as the other directives are. However they have both
471 compile-time and run-time effects.
473 See L<perlmodlib/Pragmatic Modules>.
477 Similarly C<use re 'Debug'> produces debugging output, the difference
478 being that it allows the fine tuning of what debugging output will be
479 emitted. Options are divided into three groups, those related to
480 compilation, those related to execution and those related to special
481 purposes. The options are as follows:
485 =item Compile related options
491 Turns on all compile related debug options.
495 Turns on debug output related to the process of parsing the pattern.
499 Enables output related to the optimisation phase of compilation.
503 Detailed info about trie compilation.
507 Dump the final program out after it is compiled and optimised.
511 Dump the flags associated with the program
515 Print output intended for testing the internals of the compile process
519 =item Execute related options
525 Turns on all execute related debug options.
529 Turns on debugging of the main matching loop.
533 Extra debugging of how tries execute.
537 Enable debugging of start-point optimisations.
541 =item Extra debugging options
547 Turns on all "extra" debugging options.
551 Enable debugging the capture group storage during match. Warning,
552 this can potentially produce extremely large output.
556 Enable enhanced TRIE debugging. Enhances both TRIEE
561 Enable debugging of states in the engine.
565 Enable debugging of the recursion stack in the engine. Enabling
566 or disabling this option automatically does the same for debugging
567 states as well. This output from this can be quite large.
571 Enable debugging of the \G modifier.
575 Enable enhanced optimisation debugging and start-point optimisations.
576 Probably not useful except when debugging the regexp engine itself.
580 Dump offset information. This can be used to see how regops correlate
581 to the pattern. Output format is
583 NODENUM:POSITION[LENGTH]
585 Where 1 is the position of the first char in the string. Note that position
586 can be 0, or larger than the actual length of the pattern, likewise length
591 Enable debugging of offsets information. This emits copious
592 amounts of trace information and doesn't mesh well with other
595 Almost definitely only useful to people hacking
596 on the offsets part of the debug engine.
601 =item Other useful flags
603 These are useful shortcuts to save on the typing.
609 Enable all options at once except OFFSETS, OFFSETSDBG and BUFFERS.
610 (To get every single option without exception, use both ALL and EXTRA.)
614 Enable DUMP and all execute options. Equivalent to:
622 Enable the options enabled by "All", plus STATE, TRIEC, and TRIEM.
628 As of 5.9.5 the directive C<use re 'debug'> and its equivalents are
629 lexically scoped, as are the other directives. However they have both
630 compile-time and run-time effects.
632 =head2 Exportable Functions
634 As of perl 5.9.5 're' debug contains a number of utility functions that
635 may be optionally exported into the caller's namespace. They are listed
640 =item is_regexp($ref)
642 Returns true if the argument is a compiled regular expression as returned
643 by C<qr//>, false if it is not.
645 This function will not be confused by overloading or blessing. In
646 internals terms, this extracts the regexp pointer out of the
647 PERL_MAGIC_qr structure so it cannot be fooled.
649 =item regexp_pattern($ref)
651 If the argument is a compiled regular expression as returned by C<qr//>,
652 then this function returns the pattern.
654 In list context it returns a two element list, the first element
655 containing the pattern and the second containing the modifiers used when
656 the pattern was compiled.
658 my ($pat, $mods) = regexp_pattern($ref);
660 In scalar context it returns the same as perl would when stringifying a raw
661 C<qr//> with the same pattern inside. If the argument is not a compiled
662 reference then this routine returns false but defined in scalar context,
663 and the empty list in list context. Thus the following
665 if (regexp_pattern($ref) eq '(?^i:foo)')
667 will be warning free regardless of what $ref actually is.
669 Like C<is_regexp> this function will not be confused by overloading
670 or blessing of the object.
674 If the argument is a compiled regular expression as returned by C<qr//>,
675 then this function returns what the optimiser considers to be the longest
676 anchored fixed string and longest floating fixed string in the pattern.
678 A I<fixed string> is defined as being a substring that must appear for the
679 pattern to match. An I<anchored fixed string> is a fixed string that must
680 appear at a particular offset from the beginning of the match. A I<floating
681 fixed string> is defined as a fixed string that can appear at any point in
682 a range of positions relative to the start of the match. For example,
684 my $qr = qr/here .* there/x;
685 my ($anchored, $floating) = regmust($qr);
686 print "anchored:'$anchored'\nfloating:'$floating'\n";
693 Because the C<here> is before the C<.*> in the pattern, its position
694 can be determined exactly. That's not true, however, for the C<there>;
695 it could appear at any point after where the anchored string appeared.
696 Perl uses both for its optimisations, preferring the longer, or, if they are
699 B<NOTE:> This may not necessarily be the definitive longest anchored and
700 floating string. This will be what the optimiser of the Perl that you
701 are using thinks is the longest. If you believe that the result is wrong
702 please report it via the L<perlbug> utility.
704 =item regname($name,$all)
706 Returns the contents of a named buffer of the last successful match. If
707 $all is true, then returns an array ref containing one entry per buffer,
708 otherwise returns the first defined buffer.
712 Returns a list of all of the named buffers defined in the last successful
713 match. If $all is true, then it returns all names defined, if not it returns
714 only names which were involved in the match.
716 =item regnames_count()
718 Returns the number of distinct names defined in the pattern used
719 for the last successful match.
721 B<Note:> this result is always the actual number of distinct
722 named buffers defined, it may not actually match that which is
723 returned by C<regnames()> and related routines when those routines
724 have not been called with the $all parameter set.
730 L<perlmodlib/Pragmatic Modules>.