That assigns "merlyn" and "kudra" to @who, but it's not really useful. What if I want the query to be something like "who owns something valuable?" I can do this:

use Data::Dumper;
use constant SUCCEED => qr{(?=)};
use constant FAIL    => qr{(?!)};                                     
+                                                                     
+                                                  

my $owns = <<END_FACTS;
owns merlyn gold
owns Ovid books
owns kudra gold
END_FACTS
                                                                      
+                                                                     
+                 
my $valuable = <<END_VALUABLE;
valuable Perl
valuable gold
valuable Corinna
END_VALUABLE                                                          
+                                                                     
+                             

my $is_valuable = qr[(\S+)(??{match($valuable, "valuable $2")})];
my $owns_query = qr/owns (\S+) $is_valuable/;

# here's the magic!
my @who = $owns =~ /$owns_query/mg;

print Dumper \@who;                                                   
+                                                                     
+                                    

sub match {
    my ($facts, $query) = @_;
    my $sleep_count = 0;
    my $pid;
    do {
        $pid = open KID_TO_READ, "-|";
        unless (defined $pid) {
            die "Couldn't fork" if $sleep_count++ > 6;
            sleep 1;
        }
    } until defined $pid;                                             
+                                                                     
+                                          

    unless ($pid) { # child
        if ($facts =~ $query) {
            print SUCCEED;
        }
        else {
            print FAIL;
        }
    }
    else { # parent
        my $result = <KID_TO_READ>;
        close KID_TO_READ;
        return $result;
    }
}
[download]

The prints out:

$VAR1 = [
          'merlyn',
          'gold',
          'kudra',
          'gold'
        ];
[download]

That's great. Once I can autogenerate those regexes and clean up the output I can wrap this up in a pretty framework and do all of the predicate logic programming I want.

Uh, no I can't. Regular expressions are not reentrant, thus forcing me to fork off a new process for every regex "nested" inside of the main query (such as the "is_valuable" condition). This winds up being so ridiculously slow that it's probably not worth much more than a toy. I also suspect that a complicated query will suck memory like a Hoover. Any suggestions to get around this, or am I stuck?