Recipe 13.14 Overloading Operators

13.14.1 Problem

You want to use familiar operators like = = or + on objects from a class you've written, or you want to define the print interpolation value for objects.

13.14.2 Solution

Use the use overload pragma. Here are two of the more commonly overloaded operators:

use overload    '<=>' => \&threeway_compare; sub threeway_compare {     my ($s1, $s2) = @_;     return uc($s1->{NAME}) cmp uc($s2->{NAME}); } use overload      '""'  => \&stringify; sub stringify {     my $self = shift;     return sprintf "%s (%05d)",             ucfirst(lc($self->{NAME})),             $self->{IDNUM}; }

13.14.3 Discussion

When you use built-in types, certain operators apply, like + for addition or . for string concatenation. With the use overload pragma, you can customize these operators so they do something special on your own objects.

This pragma takes a list of operator/function call pairs, such as:

package TimeNumber; use overload         '+' => \&my_plus,                      '-' => \&my_minus,                      '*' => \&my_star,                      '/' => \&my_slash;

Those four operators can now be used with objects of class TimeNumber, and the listed functions will be called as method invocations. These functions can do anything you'd like.

Here's a simple example of an overload of + for use with an object that holds hours, minutes, and seconds. It assumes that both operands are of a class that has a new method that can be invoked as an object method, and that the structure names are as shown:

sub my_plus {     my($left, $right) = @_;     my $answer = $left->new( );     $answer->{SECONDS} = $left->{SECONDS} + $right->{SECONDS};     $answer->{MINUTES} = $left->{MINUTES} + $right->{MINUTES};     $answer->{HOURS}   = $left->{HOURS}   + $right->{HOURS};     if ($answer->{SECONDS} >= 60) {         $answer->{SECONDS} %= 60;         $answer->{MINUTES} ++;     }     if ($answer->{MINUTES} >= 60) {         $answer->{MINUTES} %= 60;         $answer->{HOURS}   ++;     }     return $answer; }

It's probably best to overload numeric operators only when the objects themselves are mirroring some sort of inherently numeric construct, such as complex or infinite precision numbers, vectors, or matrices. Otherwise, the code becomes hard to understand and might lead users to invalid assumptions. Imagine a class that modeled a country. If you can add one country to another, couldn't you subtract one country from another? Applying overloaded mathematical operators for non-mathematical objects rapidly becomes ridiculous.

You may compare objects (and, in fact, any reference) using either = = or eq, but this only tells you whether the addresses are the same. (Using = = is about 10 times faster than eq though.) Because an object is a higher-level notion than a raw machine address, you often want to define your own notion of what it takes for two of them to be considered equal.

Two operators frequently overloaded even for a non-numeric class are the comparison and string interpolation operators. Both the <=> and the cmp operators can be overloaded, although the former is more prevalent. Once the spaceship operator <=> is defined for an object, you can use = =, !=, <, <=, >, and >= as well. This lets objects be collated. If ordering is not desired, overload only = =. Similarly, an overloaded cmp is used for lt, gt, and other string comparisons if they aren't explicitly overloaded.

The string interpolation operator goes by the unlikely name of "", that is, two double quotes. This operator is triggered whenever a conversion to a string is called for, such as within double or back quotes or when passed to the print function.

Read the documentation on the overload pragma that comes with Perl or Chapter 13 of Programming Perl. Perl's operator overloading has some elaborate features, such as string, numeric, and Boolean conversion methods, autogeneration of missing methods, and reversing operands if needed, as in 5 + $a where $a is an object.

13.14.4 Example: Overloaded StrNum Class

Here's a StrNum class that lets you use strings with numeric operators. Yes, we're about to do something we advised against that is, use numeric operators on non-numeric entities but programmers from other backgrounds are always expecting + and = = to work on strings. This is a simple way to demonstrate operator overloading. We almost certainly wouldn't use this in a time-critical production program due to performance concerns. It's also an interesting illustration of using a constructor of the same name as the class, something that C++ and Python programmers may take some small comfort in.

#!/usr/bin/perl # show_strnum - demo operator overloading use StrNum; $x = StrNum("Red"); $y = StrNum("Black"); $z = $x + $y; $r = $z * 3; print "values are $x, $y, $z, and $r\n"; print "$x is ", $x < $y ? "LT" : "GE", " $y\n"; values are Red, Black, RedBlack, and RedBlackRedBlackRedBlack Red is GE Black

The class is shown in Example 13-1.

Example 13-1. StrNum

  package StrNum;   use Exporter ( );   @ISA = "Exporter";   @EXPORT = qw(StrNum);  # unusual   use overload        (           '<=>'   => \&spaceship,           "cmp"   => \&spaceship,           '""'    => \&stringify,           "bool"  => \&boolify,           '0+'    => \&nummify,           '+'     => \&concat,           '*'     => \&repeat,   );   # constructor   sub StrNum {       my ($value) = @_;       return bless \$value;   }   sub stringify { ${ $_[0] } }   sub nummify    { ${ $_[0] } }   sub boolify   { ${ $_[0] } }   # providing <=> gives us <, =  =, etc. for free.   sub spaceship {       my ($s1, $s2, $inverted) = @_;       return $inverted ? $$s2 cmp $$s1 : $$s1 cmp $$s2;   }   # this uses stringify   sub concat {       my ($s1, $s2, $inverted) = @_;       return StrNum($inverted ? ($s2 . $s1) : ($s1 . $s2));   }   # this uses stringify   sub repeat {       my ($s1, $s2, $inverted) = @_;       return StrNum($inverted ? ($s2 x $s1) : ($s1 x $s2));   }   1;

13.14.5 Example: Overloaded FixNum Class

This class uses operator overloading to control the number of decimal places in output. It still uses full precision for its operations. A places method can be used on the class or a particular object to set the number of places of output to the right of the decimal point.

#!/usr/bin/perl # demo_fixnum - show operator overloading use FixNum; FixNum->places(5); $x = FixNum->new(40); $y = FixNum->new(12); print "sum of $x and $y is ", $x + $y, "\n"; print "product of $x and $y is ", $x * $y, "\n"; $z = $x / $y; printf "$z has %d places\n", $z->places; $z->places(2) unless $z->places; print "div of $x by $y is $z\n"; print "square of that is ", $z * $z, "\n"; sum of STRFixNum: 40 and STRFixNum: 12 is STRFixNum: 52 product of STRFixNum: 40 and STRFixNum: 12 is STRFixNum: 480 STRFixNum: 3 has 0 places div of STRFixNum: 40 by STRFixNum: 12 is STRFixNum: 3.33 square of that is STRFixNum: 11.11

The class itself is shown in Example 13-2. It overloads only the addition, multiplication, and division operations for math operators. Other overloaded operators are the spaceship operator (which handles all comparisons), the string-interpolation operator, and the numeric conversion operator. The string interpolation operator is given a distinctive look for debugging purposes.

Example 13-2. FixNum

  package FixNum;   use strict;   my $PLACES = 0;   sub new {       my $proto   = shift;       my $class   = ref($proto) || $proto;       my $parent  = ref($proto) && $proto;       my $v = shift;       my $self = {           VALUE  => $v,           PLACES => undef,       };       if ($parent && defined $parent->{PLACES}) {           $self->{PLACES} = $parent->{PLACES};       } elsif ($v =~ /(\.\d*)/) {           $self->{PLACES} = length($1) - 1;       }  else {           $self->{PLACES} = 0;       }       return bless $self, $class;   }   sub places {       my $proto = shift;       my $self  = ref($proto) && $proto;       my $type  = ref($proto) || $proto;       if (@_) {           my $places = shift;           ($self ? $self->{PLACES} : $PLACES) = $places;       }       return $self ? $self->{PLACES} : $PLACES;   }   sub _max { $_[0] > $_[1] ? $_[0] : $_[1] }   use overload '+'    => \&add,                '*'    => \&multiply,                '/'    => \&divide,                '<=>'  => \&spaceship,                '""'   => \&as_string,                '0+'   => \&as_number;   sub add {       my ($this, $that, $flipped) = @_;       my $result = $this->new( $this->{VALUE} + $that->{VALUE} );       $result->places( _max($this->{PLACES}, $that->{PLACES} ));       return $result;   }   sub multiply {       my ($this, $that, $flipped) = @_;       my $result = $this->new( $this->{VALUE} * $that->{VALUE} );       $result->places( _max($this->{PLACES}, $that->{PLACES} ));       return $result;   }   sub divide {       my ($this, $that, $flipped) = @_;       my $result = $this->new( $this->{VALUE} / $that->{VALUE} );       $result->places( _max($this->{PLACES}, $that->{PLACES} ));       return $result;   }   sub as_string {       my $self = shift;       return sprintf("STR%s: %.*f", ref($self),           defined($self->{PLACES}) ? $self->{PLACES} : $PLACES,   }   sub as_number {       my $self = shift;       return $self->{VALUE};   }   sub spaceship {       my ($this, $that, $flipped) = @_;       $this->{VALUE} <=> $that->{VALUE};   }   1;

13.14.6 See Also

The documentation for the standard overload, bigint, and bigrat pragmata and the standard Math::BigInt, Math::BigFloat, and Math::Complex modules; also Chapters 13, 31, and 32 of Programming Perl