I think Abigail-II has your answer, but please allow me to make a suggestion about RecDescent grammar style. You could probably improve the legibility of your grammar by slightly altering the your grammar notation. The idea is to divide the expressions into groups which will represent the individual items in the returned array.

my $prd_grammar = q(

  startrule   : PlaceHolder Comment | PlaceHolder

  PlaceHolder : Xopen | Sprintf

  Comment     : DD_Style

  DD_Style    : '--' /.*/
                { print "Comment  : \n\t" . $item[2] . "\n\n" }

  Sprintf     : '=' /.*/ '%s' /.*/
                { print "After  : \n\t%%s" . $item[4] . "\n\n" }

  Xopen       : /.*/ '?' /.*/
                { print "After  : \n\t%s", $item[3], "\n\n" }
);

new Parse::RecDescent($prd_grammar);
[download]

Often just putting everything into one regex is fine, but your problem with whitespace made me think of this. This notational style allows you (mostly) to avoid explicitly accounting for spaces in your grammar.

I don't think I like the '.*' notation here (maybe /\w+/?), but the notational style is the idea I'm trying to get across, so I'll leave your regexen up to you (mostly). :) There's always one more way to do it.