(?!) explained:

From perlre: "A zero-width negative look-ahead assertion. For example /foo(?!bar)/ matches any occurrence of "foo" that isn't followed by "bar"." So in other words, (?!) is a negative lookahead assertion that "must not match nothing" (an impossibility).

YAPE::Regex::Explain describes it simply like this:

perl -MYAPE::Regex::Explain -e "printYAPE::Regex::Explain->new( qr/(?!
+)/ )->explain();"

The regular expression:

(?-imsx:(?!))

matches as follows:

NODE                     EXPLANATION
----------------------------------------------------------------------
(?-imsx:                 group, but do not capture (case-sensitive)
                         (with ^ and $ matching normally) (with . not
                         matching \n) (matching whitespace and #
                         normally):
----------------------------------------------------------------------
  (?!)                     fail
----------------------------------------------------------------------
)                        end of grouping
----------------------------------------------------------------------
[download]

"must not match nothing"

I understand that at some level, but my brain is yelling "double negative!!". It might be better to say:

(?!) is the negative look-ahead assertion for the null string. Since the null string matches anything and everything (including nothing), (?!) always fails.

[I was tempted to find a symbolic representation for the null string, but I thougt that would only confuse the issue.]

-QM
--
Quantum Mechanics: The dreams stuff is made of

Regarding the length $^N > 1 test, I added that because while refining this RE I discovered that it was capable of considering a single-character string to be palindromic (new word?), since "a" in reverse is still "a".

Update: Now that I think about it, the length $^N > 1 test is better written as ..+ earlier in the RE, like this:

qr/(..+)
   (??{
       ( lc $^N eq reverse lc $^N )
       ? '' : '(?!)'
   })
/ix
[download]

...thus forcing at least two characters to be tested in the first place, rather than hand-coding a test for two or more characters via 'length'. Duh!

It must be the late hour, because I missed that (.+) would give trivial palindromes of length 1.

I prefer (.{2,}) over (..+), as those dots tend to fade into the background.

-QM
--
Quantum Mechanics: The dreams stuff is made of

As you point out, the regexp engine is not re-entrant. But tr/// isn't a regular expression, and that can be used to strip away whitespace.

use strict;
use warnings;
use vars qw/ $N /;

my $string = "God dog";

my $re = qr/^
            (..+)
            (??{
              local $N = lc $^N;
              $N =~ tr!\t \f!!d;
              (
                $N eq reverse $N
              ) ? ''
              :   '(?!)'
            })
          $/ix;

print "$string is",
      ( $string =~ $re )
      ? ""
      : "n't"
      , " a palindrome\n";
[download]

Does this find all non-trivial palindrome phrases on each input line?

Does it fail on some interesting input?

#!/your/perl/here

# find all palindrome phrases on each line
#     only phrases of two or more alpha
#     a phrase starts and ends on a word boundary
#
# nested palindrome phrases will also be found
# (including single words)

use strict;
use warnings;
our $N;

my $re = 
  qr/(                         # start $1
      \b                       # left word edge
      ([a-z].*[a-z])           # at least 2 alpha
      \b                       # right word edge 
      (??{                     # start code
          local $N = lc $^N;   # save capture
          $N =~ tr!a-zA-Z!!dc; # remove non-alpha
                               # fail if not pal
          '(?!)' if (lc($N) ne reverse lc($N));
         })                    # end code
     )                         # end $1
    /ix;

while (<DATA>)
{
  my $found;
  while ( /$re/g )
  {
    print "line $.:\n" unless $found;
    pos = pos() - length($1);  # find nested pals
    print "(",pos,") \"$1\"\n";
    $found = 1;                # \n between groups
    pos = pos() + 1;           # advance one
  }
  print "\n" if $found;
}

exit;

__DATA__
god dog
alpha beta gamma stop pots wonka wonka wonka
bookkeeper raisinhead
what is a bookkeeper pop repeekkoob tomorrow?
boob tube A man, a plan, a canal, Panama! kook peep aha
aba abba abbba aabbbaa abbba abba aba
[download]

produces

__OUTPUT__
line 1:
(0) "god dog"

line 2:
(17) "stop pots"

line 4:
(10) "bookkeeper pop repeekkoob"
(21) "pop"

line 5:
(0) "boob"
(10) "A man, a plan, a canal, Panama"
(42) "kook"
(47) "peep"
(52) "aha"

line 6:
(0) "aba abba   abbba   aabbbaa abbba abba aba"
(4) "abba       abbba   aabbbaa abbba abba"
(10) "abbba     aabbbaa abbba"
(17) "aabbbaa"
(25) "abbba"
(31) "abba"
(36) "aba"
[download]

BTW, if '(?!)' is changed to (?!) (no single quotes), I get a "panic: top_env" from the compiler. Is this a known bug?

-QM
--
Quantum Mechanics: The dreams stuff is made of

From perldiag:
Panic: top_env
The compiler attempted to do a goto, or something wierd like that.

"Panic" messages fall into that category of errors that "you should never see¹"

I've seen that one before, and it seems to me I was mucking around with (??{code}) constructs at the time. They don't call them "highly experimental" for nothing. ;)




Dave

This breaks on overlapping palindromes:

adding these two lines to DATA:
nested testest detsen nested
i prefer pi ip referp

produces this output:
line 7:
(0) "nested testest detsen nested"
(7) "testest detsen nested"
(15) "detsen nested"
(22) "nested"

line 8:
(0) "i prefer pi ip referp"
(2) "prefer pi ip referp"
(9) "pi ip referp"
(12) "ip referp"
(15) "referp"

Be Appropriate && Follow Your Curiosity