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 << ($PMMOD_SHIFT + 3),
26 xx => 1 << ($PMMOD_SHIFT + 4),
27 n => 1 << ($PMMOD_SHIFT + 5),
28 p => 1 << ($PMMOD_SHIFT + 6),
29 strict => 1 << ($PMMOD_SHIFT + 10),
39 eval { # Ignore errors
42 my $terminal = Tgetent Term::Cap ({OSPEED => 9600}); # Avoid warning.
43 my $props = $ENV{PERL_RE_TC} || 'md,me,so,se,us,ue';
44 my @props = split /,/, $props;
45 my $colors = join "\t", map {$terminal->Tputs($_,1)} @props;
48 $ENV{PERL_RE_COLORS} = $colors;
51 $ENV{PERL_RE_COLORS} ||= qq'\t\t> <\t> <\t\t';
73 OFFSETSDBG => 0x040000,
75 OPTIMISEM => 0x100000,
80 $flags{ALL} = -1 & ~($flags{OFFSETS}|$flags{OFFSETSDBG}|$flags{BUFFERS});
81 $flags{All} = $flags{all} = $flags{DUMP} | $flags{EXECUTE};
82 $flags{Extra} = $flags{EXECUTE} | $flags{COMPILE} | $flags{GPOS};
83 $flags{More} = $flags{MORE} = $flags{All} | $flags{TRIEC} | $flags{TRIEM} | $flags{STATE};
84 $flags{State} = $flags{DUMP} | $flags{EXECUTE} | $flags{STATE};
85 $flags{TRIE} = $flags{DUMP} | $flags{EXECUTE} | $flags{TRIEC};
87 if (defined &DynaLoader::boot_DynaLoader) {
92 # We need to work for miniperl, because the XS toolchain uses Text::Wrap, which
98 # We call install() every time, as if we didn't, we wouldn't
99 # "see" any changes to the color environment var since
100 # the last time it was called.
102 # install() returns an integer, which if casted properly
103 # in C resolves to a structure containing the regexp
104 # hooks. Setting it to a random integer will guarantee
106 $^H{regcomp} = install();
115 my $turning_all_off = ! @_ && ! $on;
116 if ($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 = $^H{reflags} || 0;
184 while ($s =~ m/( . )/gx) {
187 # The 'a' may be repeated; hide this from the rest of the
188 # code by counting and getting rid of all of them, then
189 # changing to 'aa' if there is a repeat.
191 my $sav_pos = pos $s;
192 my $a_count = $s =~ s/a//g;
193 pos $s = $sav_pos - 1; # -1 because got rid of the 'a'
197 qq 'The "a" flag may only appear a maximum of twice'
200 elsif ($a_count == 2) {
207 if ($seen_charset ne $_) {
209 qq 'The "$seen_charset" and "$_" flags '
215 qq 'The "$seen_charset" flag may not appear '
220 $^H{reflags_charset} = $reflags{$_};
224 delete $^H{reflags_charset}
225 if defined $^H{reflags_charset}
226 && $^H{reflags_charset} == $reflags{$_};
228 } elsif (exists $reflags{$_}) {
234 qq 'The "x" flag may only appear a maximum of twice'
237 elsif ($x_count == 2) {
238 $_ = 'xx'; # First time through got the /x
243 ? $reflags |= $reflags{$_}
244 : ($reflags &= ~$reflags{$_});
248 qq'Unknown regular expression flag "$_"'
253 ($^H{reflags} = $reflags or defined $^H{reflags_charset})
255 : ($^H &= ~$flags_hint);
258 Carp::carp("Unknown \"re\" subpragma '$s' (known ones are: ",
259 join(', ', map {qq('$_')} 'debug', 'debugcolor', sort keys %bitmask),
264 if ($turning_all_off) {
267 $^H{reflags_charset} = 0;
281 $^H &= ~ bits(0, @_);
290 re - Perl pragma to alter regular expression behaviour
295 ($x) = ($^X =~ /^(.*)$/s); # $x is tainted here
297 $pat = '(?{ $foo = 1 })';
299 /foo${pat}bar/; # won't fail (when not under -T
303 no re 'taint'; # the default
304 ($x) = ($^X =~ /^(.*)$/s); # $x is not tainted here
306 no re 'eval'; # the default
307 /foo${pat}bar/; # disallowed (with or without -T
311 use re 'strict'; # Raise warnings for more conditions
314 "FOO" =~ / foo /; # /ix implied
316 "FOO" =~ /foo/; # just /i implied
318 use re 'debug'; # output debugging info during
319 /^(.*)$/s; # compile and run time
322 use re 'debugcolor'; # same as 'debug', but with colored
326 use re qw(Debug All); # Same as "use re 'debug'", but you
327 # can use "Debug" with things other
329 use re qw(Debug More); # 'All' plus output more details
330 no re qw(Debug ALL); # Turn on (almost) all re debugging
333 use re qw(is_regexp regexp_pattern); # import utility functions
334 my ($pat,$mods)=regexp_pattern(qr/foo/i);
335 if (is_regexp($obj)) {
336 print "Got regexp: ",
337 scalar regexp_pattern($obj); # just as perl would stringify
338 } # it but no hassle with blessed
341 (We use $^X in these examples because it's tainted by default.)
347 When C<use re 'taint'> is in effect, and a tainted string is the target
348 of a regexp, the regexp memories (or values returned by the m// operator
349 in list context) are tainted. This feature is useful when regexp operations
350 on tainted data aren't meant to extract safe substrings, but to perform
351 other transformations.
355 When C<use re 'eval'> is in effect, a regexp is allowed to contain
356 C<(?{ ... })> zero-width assertions and C<(??{ ... })> postponed
357 subexpressions that are derived from variable interpolation, rather than
358 appearing literally within the regexp. That is normally disallowed, since
360 potential security risk. Note that this pragma is ignored when the regular
361 expression is obtained from tainted data, i.e. evaluation is always
362 disallowed with tainted regular expressions. See L<perlre/(?{ code })>
363 and L<perlre/(??{ code })>.
365 For the purpose of this pragma, interpolation of precompiled regular
366 expressions (i.e., the result of C<qr//>) is I<not> considered variable
371 I<is> allowed if $pat is a precompiled regular expression, even
372 if $pat contains C<(?{ ... })> assertions or C<(??{ ... })> subexpressions.
376 Note that this is an experimental feature which may be changed or removed in a
379 When C<use re 'strict'> is in effect, stricter checks are applied than
380 otherwise when compiling regular expressions patterns. These may cause more
381 warnings to be raised than otherwise, and more things to be fatal instead of
382 just warnings. The purpose of this is to find and report at compile time some
383 things, which may be legal, but have a reasonable possibility of not being the
384 programmer's actual intent. This automatically turns on the C<"regexp">
385 warnings category (if not already on) within its scope.
387 As an example of something that is caught under C<"strict'>, but not
388 otherwise, is the pattern
392 The C<"\x"> construct without curly braces should be followed by exactly two
393 hex digits; this one is followed by three. This currently evaluates as
398 that is, the character whose code point value is C<0xAB>, followed by the
399 letter C<C>. But since C<C> is a a hex digit, there is a reasonable chance
404 that is the single character at C<0xABC>. Under C<'strict'> it is an error to
405 not follow C<\x> with exactly two hex digits. When not under C<'strict'> a
406 warning is generated if there is only one hex digit, and no warning is raised
407 if there are more than two.
409 It is expected that what exactly C<'strict'> does will evolve over time as we
410 gain experience with it. This means that programs that compile under it in
411 today's Perl may not compile, or may have more or fewer warnings, in future
412 Perls. There is no backwards compatibility promises with regards to it. Also
413 there are already proposals for an alternate syntax for enabling it. For
414 these reasons, using it will raise a C<experimental::re_strict> class warning,
415 unless that category is turned off.
417 Note that if a pattern compiled within C<'strict'> is recompiled, say by
418 interpolating into another pattern, outside of C<'strict'>, it is not checked
419 again for strictness. This is because if it works under strict it must work
424 When C<use re '/I<flags>'> is specified, the given I<flags> are automatically
425 added to every regular expression till the end of the lexical scope.
426 I<flags> can be any combination of
441 C<no re '/I<flags>'> will turn off the effect of C<use re '/I<flags>'> for the
444 For example, if you want all your regular expressions to have /msxx on by
449 at the top of your code.
451 The character set C</adul> flags cancel each other out. So, in this example,
458 the second C<use re> does an implicit C<no re '/u'>.
462 use re "/xx"; # Doubled-x
464 use re "/x"; # Single x from here on
467 Turning on one of the character set flags with C<use re> takes precedence over the
468 C<locale> pragma and the 'unicode_strings' C<feature>, for regular
469 expressions. Turning off one of these flags when it is active reverts to
470 the behaviour specified by whatever other pragmata are in scope. For
473 use feature "unicode_strings";
474 no re "/u"; # does nothing
476 no re "/l"; # reverts to unicode_strings behaviour
480 When C<use re 'debug'> is in effect, perl emits debugging messages when
481 compiling and using regular expressions. The output is the same as that
482 obtained by running a C<-DDEBUGGING>-enabled perl interpreter with the
483 B<-Dr> switch. It may be quite voluminous depending on the complexity
484 of the match. Using C<debugcolor> instead of C<debug> enables a
485 form of output that can be used to get a colorful display on terminals
486 that understand termcap color sequences. Set C<$ENV{PERL_RE_TC}> to a
487 comma-separated list of C<termcap> properties to use for highlighting
488 strings on/off, pre-point part on/off.
489 See L<perldebug/"Debugging Regular Expressions"> for additional info.
491 As of 5.9.5 the directive C<use re 'debug'> and its equivalents are
492 lexically scoped, as the other directives are. However they have both
493 compile-time and run-time effects.
495 See L<perlmodlib/Pragmatic Modules>.
499 Similarly C<use re 'Debug'> produces debugging output, the difference
500 being that it allows the fine tuning of what debugging output will be
501 emitted. Options are divided into three groups, those related to
502 compilation, those related to execution and those related to special
503 purposes. The options are as follows:
507 =item Compile related options
513 Turns on all compile related debug options.
517 Turns on debug output related to the process of parsing the pattern.
521 Enables output related to the optimisation phase of compilation.
525 Detailed info about trie compilation.
529 Dump the final program out after it is compiled and optimised.
533 Dump the flags associated with the program
537 Print output intended for testing the internals of the compile process
541 =item Execute related options
547 Turns on all execute related debug options.
551 Turns on debugging of the main matching loop.
555 Extra debugging of how tries execute.
559 Enable debugging of start-point optimisations.
563 =item Extra debugging options
569 Turns on all "extra" debugging options.
573 Enable debugging the capture group storage during match. Warning,
574 this can potentially produce extremely large output.
578 Enable enhanced TRIE debugging. Enhances both TRIEE
583 Enable debugging of states in the engine.
587 Enable debugging of the recursion stack in the engine. Enabling
588 or disabling this option automatically does the same for debugging
589 states as well. This output from this can be quite large.
593 Enable debugging of the \G modifier.
597 Enable enhanced optimisation debugging and start-point optimisations.
598 Probably not useful except when debugging the regexp engine itself.
602 Dump offset information. This can be used to see how regops correlate
603 to the pattern. Output format is
605 NODENUM:POSITION[LENGTH]
607 Where 1 is the position of the first char in the string. Note that position
608 can be 0, or larger than the actual length of the pattern, likewise length
613 Enable debugging of offsets information. This emits copious
614 amounts of trace information and doesn't mesh well with other
617 Almost definitely only useful to people hacking
618 on the offsets part of the debug engine.
623 =item Other useful flags
625 These are useful shortcuts to save on the typing.
631 Enable all options at once except OFFSETS, OFFSETSDBG and BUFFERS.
632 (To get every single option without exception, use both ALL and EXTRA.)
636 Enable DUMP and all execute options. Equivalent to:
644 Enable the options enabled by "All", plus STATE, TRIEC, and TRIEM.
650 As of 5.9.5 the directive C<use re 'debug'> and its equivalents are
651 lexically scoped, as are the other directives. However they have both
652 compile-time and run-time effects.
654 =head2 Exportable Functions
656 As of perl 5.9.5 're' debug contains a number of utility functions that
657 may be optionally exported into the caller's namespace. They are listed
662 =item is_regexp($ref)
664 Returns true if the argument is a compiled regular expression as returned
665 by C<qr//>, false if it is not.
667 This function will not be confused by overloading or blessing. In
668 internals terms, this extracts the regexp pointer out of the
669 PERL_MAGIC_qr structure so it cannot be fooled.
671 =item regexp_pattern($ref)
673 If the argument is a compiled regular expression as returned by C<qr//>,
674 then this function returns the pattern.
676 In list context it returns a two element list, the first element
677 containing the pattern and the second containing the modifiers used when
678 the pattern was compiled.
680 my ($pat, $mods) = regexp_pattern($ref);
682 In scalar context it returns the same as perl would when stringifying a raw
683 C<qr//> with the same pattern inside. If the argument is not a compiled
684 reference then this routine returns false but defined in scalar context,
685 and the empty list in list context. Thus the following
687 if (regexp_pattern($ref) eq '(?^i:foo)')
689 will be warning free regardless of what $ref actually is.
691 Like C<is_regexp> this function will not be confused by overloading
692 or blessing of the object.
696 If the argument is a compiled regular expression as returned by C<qr//>,
697 then this function returns what the optimiser considers to be the longest
698 anchored fixed string and longest floating fixed string in the pattern.
700 A I<fixed string> is defined as being a substring that must appear for the
701 pattern to match. An I<anchored fixed string> is a fixed string that must
702 appear at a particular offset from the beginning of the match. A I<floating
703 fixed string> is defined as a fixed string that can appear at any point in
704 a range of positions relative to the start of the match. For example,
706 my $qr = qr/here .* there/x;
707 my ($anchored, $floating) = regmust($qr);
708 print "anchored:'$anchored'\nfloating:'$floating'\n";
715 Because the C<here> is before the C<.*> in the pattern, its position
716 can be determined exactly. That's not true, however, for the C<there>;
717 it could appear at any point after where the anchored string appeared.
718 Perl uses both for its optimisations, preferring the longer, or, if they are
721 B<NOTE:> This may not necessarily be the definitive longest anchored and
722 floating string. This will be what the optimiser of the Perl that you
723 are using thinks is the longest. If you believe that the result is wrong
724 please report it via the L<perlbug> utility.
726 =item regname($name,$all)
728 Returns the contents of a named buffer of the last successful match. If
729 $all is true, then returns an array ref containing one entry per buffer,
730 otherwise returns the first defined buffer.
734 Returns a list of all of the named buffers defined in the last successful
735 match. If $all is true, then it returns all names defined, if not it returns
736 only names which were involved in the match.
738 =item regnames_count()
740 Returns the number of distinct names defined in the pattern used
741 for the last successful match.
743 B<Note:> this result is always the actual number of distinct
744 named buffers defined, it may not actually match that which is
745 returned by C<regnames()> and related routines when those routines
746 have not been called with the $all parameter set.
752 L<perlmodlib/Pragmatic Modules>.