You are right, grep will not do captures in such a situation. I might not have been clear, but I only mentionned is as an additional useful tool in similar synctactic context, for filtering purpose.

Having said that you can do something akin to captures in a grep, although it is not very clean. Consider this:

  DB<1> @a = qw (foobar barfoo foobaz, bar, foobor);

  DB<2> @c = grep {s/fo(ob.).+/$1/} @a;

  DB<3> x @c
0  'oba'
1  'oba'
2  'obo'
[download]

Not clean, I would not really recommend it, but not impossible.

Not clean ...

... in part, perhaps, because the  @a input array is affected (via aliasing) and winds up 'oba-oba-obo-ed':

>perl -wMstrict -le
"my @a = qw(foobar barfoo foobaz bar foobor);
 my @c = grep {s/fo(ob.).+/$1/} @a;
 ;;
 printf qq{'$_'  } for @c;
 print qq{\n};
 printf qq{'$_'  } for @a;
"
'oba'  'oba'  'obo'

'oba'  'barfoo'  'oba'  'bar'  'obo'
[download]

Yes, you are absolutely right, this is definitely one of the main reasons why it is not clean in my mind (and I upvoted your comment). But you can also make a temporary copy if you want to keep the original array intact, no big deal if the data is nor too large. But again, I am not suggesting this as a good solution. It might just be a possible useful alternative in some edge cases or very specific context, although I fail to see any.