Sometimes double negation is difficult to understand ...

More than just sometimes, IMHO, but it's tolerable if taken in moderation. E.g., if you need a "digit boundary" assertion analogous to  \b in that it also matches at the start/end of a string, then  (?<! \d) and  (?! \d) are very attractive. Then  (?<! \d) \d{4} (?! \d) matches  '1234' 'x1234' '1234x' 'x1234x' but none of  '12345' 'x12345x' etc. Extend this to  (?<! \D) and  (?! \D) and you have a sometimes-useful double-negation asserting "non-digit boundary".

Wow!
But it seems that (?<! \D) and (?! \D) is not "equivalent" to \B, because \B don't match beginning or ending of string:

# I've changed all x-es to spaces (for comparison).
for my $line ('1234', ' 1234', '1234 ', ' 1234 ', 
            '12345', ' 12345 ',, ' 123456 '){
    print map { sprintf "%10s: $_\n", "'$line'" } join ' ', 
        map { $line =~ qr/$_/x ? 'OK' : 'NO' } 
            '(?<! \d) \d{4} (?! \d)', 
            '(?<! \D) \d{4} (?! \D)', 
            '\b       \d{4} \b     ', 
            '\B       \d{4} \B     ', 
    }
__END__
    '1234': OK OK OK NO
   ' 1234': OK NO OK NO
   '1234 ': OK NO OK NO
  ' 1234 ': OK NO OK NO
   '12345': NO OK NO NO
 ' 12345 ': NO NO NO NO
' 123456 ': NO OK NO OK
[download]

... (?<! \D) and (?! \D) is not "equivalent" to \B ...

That's why they're only sometimes useful :)

---

Give a man a fish:  <%-{-{-{-<