A third way is to call pipe before fork to set up a reader and writer handle. You'll want to set the writer to autoflush, and to close the unwanted end of the pipe in each process.

That is low level, so is very flexible and powerful. Writing C in Perl is not always a bad thing.

Here's how to set up the child as a coprocess, taking STDIN from the parent, and writing back STDOUT to the parent. That needs two pipes, one each direction.

my ($reader,$writer,$cpid);
{
    pipe $reader, local *STDOUT or die $!;
    pipe local(*STDIN), $writer or die $!;

    $cpid = fork;
    defined $cpid or die $!;
    last if $cpid;

    # child
    close $reader or die $!;
    close $writer or die $!;
    $| = 1;
    exec '/path/to/foo', @foo_args;
    die 'Failed to launch foo: ', $!;
}

# parent
select((select($writer), $| = 1)[0]); # see 'perldoc -f select'

# builtin 4-arg select or IO::Select may be handy here
# . . .
waitpid $cpid;
[download]

Use wait (if it returns -1, then you have no childs left).

If you wish to gather its output and wait for the child there are two simple options. The first runs the child to completion then gets its output

perl -le '$result=`echo "hello world"`;print $result; print "OK"'
[download]

perl -le 'open CHLD, "ls -l |";while(<CHLD>){chomp; print "got: $_"} p
+rint "OK"'
[download]

Cheers,
R.