Class::Multimethods - Support multimethods and function overloading in Perl

This document describes version 1.70 of Class::Multimethods, released April 9, 2000.

 # IMPORT THE multimethod DECLARATION SUB...

    use Class::Multimethods;

 # DECLARE VARIOUS MULTIMETHODS CALLED find...

 # 1. DO THIS IF find IS CALLED WITH A Container REF AND A Query REF...

    multimethod find => (Container, Query) 
                     => sub { $_[0]->findquery($_[1]) };

 # 2. DO THIS IF find IS CALLED WITH A Container REF AND A Sample REF...

    multimethod find => (Container, Sample)
                     => sub { $_[0]->findlike($_[1]) };

 # 3. DO THIS IF find IS CALLED WITH AN Index REF AND A Word REF...

    multimethod find => (Index, Word)      
                     => sub { $_[0]->lookup_word($_[1]) };

 # 4. DO THIS IF find IS CALLED WITH AN Index REF AND A qr// PATTERN

    multimethod find => (Index, Regexp)    
                     => sub { $_[0]->lookup_rx($_[1]) };

 # 5. DO THIS IF find IS CALLED WITH AN Index REF AND A NUMERIC SCALAR

    multimethod find => (Index, '#')       
                     => sub { $_[0]->lookup_elem($_[1]) };

 # 6. DO THIS IF find IS CALLED WITH AN Index REF AND A NON-NUMERIC SCALAR

    multimethod find => (Index, '$')       
                     => sub { $_[0]->lookup_str($_[1]) };

 # 7. DO THIS IF find IS CALLED WITH AN Index REF AND AN UNBLESSED ARRAY REF
 #    (NOTE THE RECURSIVE CALL TO THE find MULTIMETHOD)

    multimethod find => (Index, ARRAY)     
                     => sub { map { find($_[0],$_) } @{$_[1]} };

 # SET UP SOME OBJECTS...

        my $cntr = new Container ('./datafile');
        my $indx = $cntr->get_index();

 # ...AND SOME INHERITANCE...

        @BadWord::ISA = qw( Word );
        my $badword = new BadWord("fubar");

 # ...AND EXERCISE THEM...

        print find($cntr, new Query('cpan OR Perl'));           # CALLS 1.
        print find($cntr, new Example('by a committee'));       # CALLS 2.

        print find($indx, new Word('sugar'));                   # CALLS 3.
        print find($indx, $badword);                            # CALLS 3.
        print find($indx, qr/another brick in the Wall/);       # CALLS 4.
        print find($indx, 7);                                   # CALLS 5.
        print find($indx, 'But don't do that.');                # CALLS 6.
        print find($indx, [1,"one"]);                           # CALLS 7,
                                                                # THEN 5 & 6.

The Class:Multimethod module exports a subroutine (&multimethod) that can be used to declare other subroutines that are dispatched using a algorithm different from the normal Perl subroutine or method dispatch mechanism.

Normal Perl subroutines are dispatched by finding the appropriately-named subroutine in the current (or specified) package and calling that. Normal Perl methods are dispatched by attempting to find the appropriately-named subroutine in the package into which the invoking object is blessed or, failing that, recursively searching for it in the packages listed in the appropriate CW@ISA arrays.

Class::Multimethods multimethods are dispatched quite differently. The dispatch mechanism looks at the classes or types of each argument to the multimethod (by calling CWref on each) and determines the closest matching variant of the multimethod, according to the argument types specified in the variants' definitions (see Finding the nearest multimethod for a definition of closest).

The result is something akin to 's function overloading, but more intelligent, since multimethods take the inheritance relationships of each argument into account. Another way of thinking of the mechanism is that it performs polymorphic dispatch on every argument of a method, not just the first.

The Class::Multimethods module exports a subroutine called CWmultimethod, which can be used to specify multimethod variants with the dispatch behaviour described above. The CWmultimethod subroutine takes the name of the desired multimethod, a list of class names, and a subroutine reference, and generates a corresponding multimethod variant within the current package.

For example, the declaration:

        package LargeInt;   @ISA = (LargeNumeric);
        package LargeFloat; @ISA = (LargeNumeric);

        package LargeNumeric;
        use Class::Multimethods;

        multimethod divide => (LargeInt, LargeInt) => sub
        {
                LargeInt::divide($_[0],$_[1]);
        };

        multimethod divide => (LargeInt, LargeFloat) => sub
        {
                LargeFloat::divide($_[0]->AsLargeFloat(),$_[1]));
        };

creates a (single!) multimethod CW&LargeNumeric::divide with two variants. If the multimethod is called with two references to CWLargeInt objects as arguments, the first variant (i.e. anonymous subroutine) is invoked. If the multimethod is called with a CWLargeInt reference and a CWLargeFloat reference, the second variant is called.

Note that if you're running under CWuse strict, the list of bareword class names in each variant definition will cause problems. In that case you'll need to say:

        multimethod divide => ('LargeInt', 'LargeInt') => sub
        {
                LargeInt::divide($_[0],$_[1]);
        };

        multimethod divide => ('LargeInt', 'LargeFloat') => sub
        {
                LargeFloat::divide($_[0]->AsLargeFloat(),$_[1]));
        };

or better still:

        multimethod divide => qw( LargeInt LargeInt ) => sub
        {
                LargeInt::divide($_[0],$_[1]);
        };

        multimethod divide => qw( LargeInt LargeFloat ) => sub
        {
                LargeFloat::divide($_[0]->AsLargeFloat(),$_[1]));
        };

or best of all (;-):

        {
            no strict;

            multimethod divide => (LargeInt, LargeInt) => sub
            {
                LargeInt::divide($_[0],$_[1]);
            };

            multimethod divide => (LargeInt, LargeFloat) => sub
            {
                LargeFloat::divide($_[0]->AsLargeFloat(),$_[1]));
            };
        }

Calling the multimethod with any other combination of CWLargeNumeric reference arguments (e.g. a reference to a CWLargeFloat and a reference to a CWLargeInt, or two CWLargeFloat referencess) results in an exception being thrown, with the message:

        No viable candidate for call to
        multimethod LargeNumeric::divide at ...

To avoid this, we could provide a catch-all variant:

        multimethod divide => (LargeNumeric, LargeNumeric) => sub
        {
                LargeFloat::divide($_[0]->AsLargeFloat(),$_[1]->AsLargeFloat));
        }

Now, calling CW&LargeNumeric::divide with either a CWLargeFloat reference and a CWLargeInt reference or two CWLargeFloat references results in this third variant being invoked. Note that, adding this third alternative doesn't affect calls to the other two, since Class::Multimethods always selects the nearest match (see Finding the nearest multimethod below for details of what nearest means).

This best fit behaviour is extremely useful, because it means you can code the specific cases you want to handle, and the one or more catch-all cases to deal with any other combination of arguments. Of course, the usefulness of the entire system depends on how intelligently Class::Multimethods decides which version of a multimethod is nearest to the set of arguments you provided. This decision process is called dispatch resolution, and Class::Multimethods does it like this:

1.

If the types of the arguments given (as determined by CWref) exactly match the types specified in any variant of the multimethod, that variant is the one called.

2.

Otherwise, Class::Multimethods compiles a list of viable targets. A viable target is a variant of the multimethod with the correct number of parameters, such that for each parameter the specified parameter type is a base class of the actual type of the corresponding argument in the actual call.

3.

If there is only one viable target, it is immediately called. if there are no viable targets, an exception is thrown indicating the fact.

4.

Otherwise, Class::Multimethod examines each viable target and computes its distance to the actual set of arguments. The distance of a target is the sum of the distances of each of its parameters. The distance of an individual parameter is the number of inheritance steps between its class and the actual class of the corresponding argument. Hence, if a specific argument is of the same class as the corresponding parameter type, the distance to that parameter is zero. If the argument is of a class that is an immediate child of the parameter type, the distance is 1. If the argument is of a class which is a grandchild of the parameter type, the distance is 2. Et cetera.

5.

Class::Multimethod then chooses the viable target with the smallest distance as the final target. If there is more than one viable target with an equally smallest distance, an exception is thrown indicating that the call is ambiguous. If there is only a single final target Class::Multimethod records its identity (so the distance computations don't have to be repeated next time the same set of argument types is used), and then calls that final target.

Class::Multimethods doesn't care which packages the individual variants of a multimethod are defined in. Every variant of a multimethod is visible to the underlying multimethod dispatcher, no matter where it was defined.

For example, the three variants for the CWdivide multimethod shown above could all be defined in the LargeNumeric package, or the LargeFloat package or the LargeInt package, or in CWmain, or in a separate package of their own.

Of course, to make a specific multimethod visible within a given package you still need to tell that package about it. That can be done by specifying the name of the multimethod only (i.e. no argument list or variant code):

        package Some::Other::Package::That::Wants::To::Use::divide;

        use Class::Multimethods;
        multimethod "divide";

For convenience, the declaration itself can be abbreviated to:

        package Some::Other::Package::That::Wants::To::Use::divide;

        use Class::Multimethods "divide";

Similarly, Class::Multimethod doesn't actually care whether multimethods are called as methods or as regular subroutines. This is quite different from the behaviour of normal Perl methods and subroutines, where how you call them, determines how they are dispatched.

With multimethods, since all arguments participate in the polymorphic resolution of a call (instead of just the first), it make no difference whether a multimethod is called as a subroutine:

        numref3 = divide($numref1, $numref2);

or a method:

        numref3 = $numref1->divide($numref2);

(so long as the multimethod has been declared in the appropriate place: the current package for subroutine-like calls, or the invoking object's package for method-like calls).

In other words, Class::Multimethods also provides general subroutine overloading. For example:

        package main;
        use IO;
        use Class::Multimethods;

        multimethod debug => (IO::File) => sub
        {
                print $_[0] "This should go in a file\n";
        }

        multimethod debug => (IO::Pipe) => sub
        {
                print $_[0] "This should go down a pipe\n";
        }

        multimethod debug => (IO::Socket) => sub
        {
                print $_[0] "This should go out a socket\n";
        }

        # and later

        debug($some_io_handle);

Yet another thing Class::Multimethods doesn't care about is whether the parameter types for each multimethod variant are the names of real classes or just the identifiers returned when raw Perl data types are passed to the built-in CWref function. That means you could also define multimethod variants like this:

        multimethod stringify => (ARRAY) => sub
        {
                my @arg = @{$_[0]};
                return "[" .  join(", ",@arg) . "]";
        }

        multimethod stringify => (HASH) => sub
        {
                my %arg = %{$_[0]};
                return "{" . join(", ", map("$_=>$arg{$_}",keys %arg)) . "}";
        }

        multimethod stringify => (CODE) => sub
        {
                return "sub {???}";
        }

        # and later

        print stringify( [1,2,3] ), "\n";
        print stringify( {a=>1,b=>2,c=>3} ), "\n"; 
        print stringify( $array_or_hash_ref ), "\n";

Provided you remember that the parameter types ARRAY, HASH, and CODE really mean reference to array, reference to hash, and reference to subroutine, the names of built-in types (i.e. those returned by CWref) are perfectly acceptable as multimethod parameters.

That's a nice bonus, but there's a problem. Because CWref returns an empty string when given any literal string or numeric value, the following code:

        print stringify( 2001 ), "\n";
        print stringify( "a multiple dispatch oddity" ), "\n";

will produce a nasty surprise:

        No viable candidate for call to multimethod stringify() at line 1

That's because the dispatch resolution process first calls CWref(2) to get the class name for the first argument, and therefore thinks it's of class CW"". Since there's no CWstringify variant with an empty string as its parameter type, there are no viable targets for the multimethod call. Hence the exception.

To overcome this limitation, Class::Multimethods allows three special pseudo-type names within the parameter lists of multimethod variants. The first pseudo-type - CW"$" - is the class that Class::Multimethods pretends that any scalar value (except a reference) belongs to. Hence, you can make the two recalcitrant stringifications of scalars work by defining:

        multimethod stringify => ("$")
                => sub { return qq{"$_[0]"} }

With that definition in place, the two calls:

        print stringify( 2001 ), "\n";
        print stringify( "a multiple dispatch oddity" ), "\n";

would produce:

        "2001"
        "a multiple dispatch oddity"

That solves the problem, but not as elegantly as it might. It would be better if numeric values were left unquoted. To this end, Class::Multimethods offers a second pseudo-type - CW"#" - which is the class it pretends numeric scalar values belong to (where a scalar value is numeric if it's truly a numerical value (without implicit coercions):

        $var = 0        # numeric --> '$'
        $var = 0.0      # numeric --> '$'
        $var = "0";     # string  --> '#'

Hence you could now also define:

        multimethod stringify => ("#")
                => sub { return "+$_[0]" }

the two calls to CW&stringify now produce:

        +2001
        "a multiple dispatch oddity"

The final pseudo-type - CW"*" - is a wild-card or don't care type specifier, which matches any argument type exactly. For example, we could provide a catch-all CWstringify variant (to handle GLOB or IO references, for example):

        multimethod stringify => ("*")
                => sub { croak "can't stringify a " . ref($_[0]) }

The CW"*" pseudo-type can also be used in multiple-argument multimethods. For example:

        # General case...

            multimethod handle => (Window, Event, Mode)
                => sub { ... }

        # Special cases...

            multimethod handle => (MovableWindow, MoveEvent, NormalMode)
                => sub { ... }

            multimethod handle => (ScalableWindow, ResizeEvent, NormalMode)
                => sub { ... }

        # Very special case
        # (ignore any event in any window in PanicMode)

            multimethod handle => ("*", "*", PanicMode)
                => sub { ... }

It's relatively easy to set up a multimethod such that particular combinations of argument types cannot be correctly dispatched. For example, consider the following variants of a multimethod called CWput_peg:

        multimethod put_peg => (RoundPeg,Hole) => sub
        {
                print "a round peg in any old hole\n";
        };

        multimethod put_peg => (Peg,SquareHole) => sub
        {
                print "any old peg in a square hole\n";
        };

        multimethod put_peg => (Peg,Hole) => sub
        {
                print "any old peg in any old hole\n";
        };

If CWput_peg is called like so:

        put_peg( RoundPeg->new(), SquareHole->new() );

then Class::Multimethods can't dispatch the call, because it cannot decide between the CW(RoundPeg,Hole) and CW(Peg,SquareHole) variants, each of which is the same distance (i.e. 1 derivation) from the actual arguments.

The default behaviour is to throw an exception (i.e. die) like this:

        Cannot resolve call to multimethod put_peg(RoundPeg,SquareHole).
        The multimethods:
                put_peg(RoundPeg,Hole)
                put_peg(Peg,SquareHole)
        are equally viable at ...

Sometimes, however, the more specialized variants are only optimizations, and a more general case (e.g. the CW(Peg,Hole) variant) would suffice as a default where such an ambiguity exists. If that is the case, it's possible to tell Class::Multimethods to resolve the ambiguity by calling that variant, using the CWresolve_ambiguous subroutine. CWresolve_ambiguous is automatically exported by Class::Multimethods and is used like this:

        resolve_ambiguous put_peg => (Peg,Hole);

That is, you specify the name of the multimethod being disambiguated, and the signature of the variant to be used in ambiguous cases. Of course, the specified variant must actually exist at the time of the call. If it doesn't, Class::Multimethod ignores it and throws the usual exception.

Alternatively, if no variant is suitable as a default, you can register a reference to a subroutine that is to be called instead:

        resolve_ambiguous put_peg => \&disambiguator;

Now, whenever CWput_peg can't dispatch a call because it's ambiguous, CWdisambiguator will be called instead, with the same argument list as CWput_peg was given.

Of course, CWresolve_ambiguous doesn't care what subroutine it's given a reference to, so you can also use an anonymous subroutine:

        resolve_ambiguous put_peg
                => sub
                   {
                        print "can't put a ", ref($_[0]),
                              " into a ", ref($_[1]), "\n";
                   };

Dispatch can also fail if there are no suitable variants available to handle a particular call. For example:

        put_peg( JPEG->new(), Loophole->new() );

which would normally produce the exception:

        No viable candidate for call to
        multimethod put_peg(JPeg,Loophole) at ...

since classes JPEG and Loophole are't in the Peg and Hole hierarchies, so there's no inheritance path back to a more general variant.

To handle cases like this, you can use the <resolve_no_match> subroutine, which is also exported from Class::Multimethods. CWresolve_no_match registers a multimethod variant, or a reference to some other subroutine, that is then used whenever the dispatch mechanism can't find a suitable variant for a given multimethod call.

For example:

        resolve_no_match put_peg
                => sub
                   {
                        put_jpeg(@_)
                                if ref($_[0]) eq 'JPEG';
                        shift()->hang(@_)
                                if ref($_[0]) eq 'ClothesPeg';
                        hammer(@_)
                                if ref($_[0]) eq 'TentPeg';
                        # etc.

                   };

As with CWresolve_ambiguous the registered variant or subroutine is called with the same set of arguments that were passed to the original multimethod call.

Sometimes a polymorphic method in a derived class is used to add functionality to an inherited method. For example, a derived class's CWprint_me method might call it's base class's CWprint_me, making use of Perl's special CW$obj-SUPER::method()> construct:

        class Base;

        sub print_me
        {       
                my ($self) = @_;
                print "Base stuff\n";
        }

        class Derived; @ISA = qw( Base );

        sub print_me
        {
                my ($self) = @_;
                $self->SUPER::print_me();       # START LOOKING IN ANCESTORS
                print "Derived stuff\n";
        }

If the CWprint_me methods are implemented as multimethods, it's still possible to reinvoke an ancestral method, using the automatically exported CWClass::Multimethods::superclass subroutine:

        use Class::Multimethods;

        multimethod print_me => (Base) => sub
        {       
                my ($self) = @_;
                print "Base stuff\n";
        }

        multimethod print_me => (Derived) => sub
        {
                my ($self) = @_;
                print_me( superclass($self) );  # START LOOKING IN ANCESTORS
                print "Derived stuff\n";
        }
        }

Applying CWsuperclass to the multimethod argument tells Class::Multimethod to start looking for parameter types amongst the ancestors of Derived.

It's also possible in regular Perl to explcitly tell the polymorphic dispacther where to start looking, by explicitly qualifying the method name:

        sub Derived::print_me
        {
                my ($self) = @_;
                $self->Base::print_me();        # START LOOKING IN Base CLASS
                print "Derived stuff\n";
        }

The same is possible with multimethods. CWsuperclass takes an optional second argument that tells Class::Multimethods exactly where to start looking:

        multimethod print_me => (Derived) => sub
        {
                my ($self) = @_;
                print_me( superclass($self => Base) );  # START LOOKING IN Base
                print "Derived stuff\n";
        }

Note that, unlike regular method calls, with multimethods you can apply the CWsuperclass subroutine to any or all of a multimethod's arguments. For example:

        multimethod handle => (MovableWindow, MoveEvent, NormalMode) => sub
        {
                my ($w, $e, $m) = @_;

                # Do any special stuff,
                # then redispatch to more general handler...

                handle(superclass($w), $e, superclass($m => Mode) );
        }

In this case the redispatch would start looking for variants which matched CW(CIany of MovableWindow's ancestorsCW, MoveEvent, Mode).

It's also important to remember that, as with regular methods, the class of the actual arguments doesn't change just because we subverted the dispatch sequence. That means if the above redispatch called the handle variant that takes arguments (Window, MoveEvent, Mode), the actual arguments would still be of types (MovableWindow, MoveEvent, NormalMode).

If you call CWmultimethod and forget to provide a code reference as the last argument, it CWdies with the message:

        "multimethod: last arg must be a code reference at %s"

If the dispatch mechanism cannot find any multimethod with a signature matching the actual arguments, it CWdies with the message:

        "No viable candidate for call to multimethod %s at %s"

If the dispatch mechanism finds two or more multimethods with signatures equally close to the actual arguments (see The dispatch resolution process), it CWdies with the message:

        "Cannot resolve call to multimethod %s. The multimethods:
                %s
         are equally viable at %s"

If you specify two variants with the same parameter lists, Class::Multimethods warns:

        "Multimethod %s redefined at %s"

but only if $^W is true (i.e. under the CW-w flag).

Damian Conway (damian@conway.org)

There are undoubtedly serious bugs lurking somewhere in code this complex :-) Bug reports and other feedback are most welcome.

Ongoing annoyances include:

•

The module uses qr// constructs to improve performance. Hence it won't run under Perls earlier than 5.005.

•

Multimethod dispatch is much slower than regular dispatch when the resolution has to resort to the more generic cases (though it's actually as very nearly as fast as doing the equivalent type resolution by hand, and certainly more reliable and maintainable)

•

The cache management is far too dumb. Adding any new multimethod clobbers the entire cache, when it should only expunge those entries upstream from the the new multimethod's actual parameter types. It's unclear, however, under what circumstances the expense of a more careful cache correction algorithm would ever be recouped by the savings in dispatch (well, obviously, when the installion of multimethods is a rare event and multimethod dispatching is frequent, but where is the breakeven point?)

        Copyright (c) 1998-2000, Damian Conway. All Rights Reserved.
      This module is free software. It may be used, redistributed
      and/or modified under the terms of the Perl Artistic License
           (see http://www.perl.com/perl/misc/Artistic.html)