You can't use + on parens like that to "remember" any number of matches - that's one of the reasons to have split. As for why it only gives the last part, is that matches are greedy (unless you tell them not to be) and will get the largest match they can - here the end of your string.

Although I suspect you know, for anybody else reading this a slight clarification:

is that matches are greedy (unless you tell them not to be) and will get the largest match they can

This is not strictly true with perls NFA based regex engine. They will match the leftmost longest match that they can. This doesnt mean the longest possible match as a DFA based regex engine (egrep) would provide. Thus

"AAABBBBBBBBAAAAAAAAAAAA"=~m/A+|A+B+A+/;
[download]

Will match "AAA" and not the entire string. But a DFA based regex engine would match the entire string.

OTOH reversing the option

"AAABBBBBBBBAAAAAAAAAAAA"=~m/A+B+A+|A+/;
[download]

Would match the entire string using either engine.

Yves / DeMerphq
---
Writing a good benchmark isn't as easy it might look.

what I believe you want to do is match all wordparts followed by an ' or the end of line like so:

@words = $word =~ /(\w+)['\z]?/g;
[download]

   larryk                                          
perl -le "s,,reverse killer,e,y,rifle,lycra,,print"

Doesn't really explain why I get only the last two parts as my regex started whith '^' and therefore can't just start matching in the middle of the string...

But the correct answer has already been given (thanks, erikharrison). I just wasn't aware that ()+ gives you just one item, no matter how often it matched.

pike

The point is not whether it can start matching in the middle of the string but whether it can start _capturing_ in the middle of the string. By adding a + to the first part you allow the RE engine to effectively ignore the ^ anchor while still forcing it to make a match at the end of the string.

   larryk                                          
perl -le "s,,reverse killer,e,y,rifle,lycra,,print"

Actually, it matches all of them, but since it is repeating, it overwrites the first match:

$ perl -lwe"q<d'aujourd'hui> =~ /^((\w+')(?{print$+}))+([\w-]+)$/"
d'
aujourd'
$
[download]

    p

I suppose that if you really wanted to use a non-split solution, you could try something like this:

#in scalar context:
  my @words =();
  push @words, $1 while $word =~ /(\w+'?)/g;

#in list context:
  my @words = ( $word =~ /(\w+'?)/g );
[download]

Check it again: your first idea doesnt work either for d'aujourd'hui (but does for aujourd'hui, unless I am myself doing some typos ;-)

The problem there is that your regexp is not.. what should I say.. "recursive": /^(\w+')([\w-]+)$) can only return two matches (i.e.: $1 and $2 and not $1, $2, $3 to $n, etc)