11.4. Taking References to Functions

Problem

You need to manipulate a subroutine by reference. This might happen if you need to create a signal handler, a Tk callback, or a hash of function pointers.

Solution

To get a code reference:

$cref = \&func;
$cref = sub { ... };

To call a code reference:

@returned = $cref->(@arguments);
@returned = &$cref(@arguments);

Discussion

If the name of a function is func , you can produce a reference to this code by preceding that name with \& . You can also create anonymous functions using the sub {} notation. These code references can be stored just like any other reference.

Perl 5.004 introduced the postfix arrow notation for dereferencing a code reference. Prior to that, to call a subroutine by reference, you had to say &$funcname(@ARGS) , where $funcname contained the name of a function. Although it is still possible to store the name of a function in a variable, such as:

$funcname = "thefunc";
&$funcname();

that's not a very good solution for several reasons. First, it uses symbolic references, not real (hard) references, so it is forbidden under the use strict 'refs' pragma. Symbolic references to variables are usually a bad idea, since they can't access lexical variables, only globals, and aren't reference counted.

Second, it doesn't include package information, so if executed in a different package, it would try to call the wrong function. Finally, in the odd case that the function were redefined at some point, the symbolic reference would get whatever the current definition for the function was, whereas the hard reference would retain its old definition.

Instead of placing the name of the function in the variable, use the backslash operator to create a reference to the function. This is the normal way to store a function in a variable or pass it to another function. You can mix and match references to named functions with references to unnamed ones:

my %commands = (
    "happy" => \&joy,
    "sad"   => \&sullen,
    "done"  => sub { die "See ya!" },
    "mad"   => \&angry,
);

print "How are you? ";
chomp($string = <STDIN>);
if ($commands{$string}) {
    $commands{$string}->();
} else {
    print "No such command: $string\n";
} 

If you create an anonymous function that refers to a lexical ( my ) variable from an enclosing scope, Perl's reference counting ensures that the lexical variable is never deallocated so long as that function reference exists:

sub counter_maker {
    my $start = 0;
    return sub {                      # this is a closure
        return $start++;              # lexical from enclosing scope
    };
}

$counter = counter_maker();

for ($i = 0; $i < 5; $i ++) {
    print &$counter, "\n";
}

Even though counter_maker has ended and $start has gone out of scope, Perl doesn't free it because the anonymous subroutine referenced by $counter still has a reference to $start . If we call counter_maker again, it'll return another anonymous subroutine reference that uses a different $start :

$counter1 = counter_maker();
$counter2 = counter_maker();

for ($i = 0; $i < 5; $i ++) {
    print &$counter1, "\n";
}

print &$counter1, " ", &$counter2, "\n";




0








1








2








3








4








5 0



Closures are often used in callback routines. In graphical and other event-based programming, you associate code with a keypress, mouse click, window expose event, etc. The code will be called much later, probably from an entirely different scope. Variables mentioned in the closure must be available when it's finally called. To work properly, such variables must be lexicals, not globals.

Another use for closures is function generators, that is, functions that create and return brand new functions. counter_maker is a function generator. Here's another simple one:

sub timestamp {
    my $start_time = time(); 
    return sub { return time() - $start_time };
} 
$early = timestamp(); 
sleep 20; 
$later = timestamp(); 
sleep 10;
printf "It's been %d seconds since early.\n", $early->();
printf "It's been %d seconds since later.\n", $later->();




It's been 30 seconds since early.








It's been 10 seconds since later.



Each call to timestamp generates and returns a brand new function. The timestamp function creates a lexical called $start_time that contains the current clock time (in epoch seconds). Every time that closure is called, it returns how many seconds have passed since it was created by subtracting its starting time from the current time.

See Also

The section on "Closures" in Chapter 4 of Programming Perl and the discussion on closures in perlref (1); Recipe 10.11 ; Recipe 11.4