Have a need to create method overloading, so I put together an
implementation. Curious what other think of it; how to improve it; or
do it a better way.
Thanks,
T.
# ---- overload.rb
class Module
def method_overloads @method_overloads ||= {}
end
def overload( name, *signiture, &block )
name = name.to_sym
if method_overloads.key?( name )
method_overloads[name][signiture] = block
if method_defined?( name ) #method_overloads[name][nil] = instance_method( name ) #true
alias_method( "#{name}Generic", name )
has_generic = true
else
has_generic = false
end
define_method( name ) do |*args|
ovr = self.class.method_overloads[:"#{name}"]
sig = args.collect{ |a| a.class }
hit = nil
faces = ovr.keys.sort { |a,b| b.size <=> a.size }
faces.each do |cmp|
next unless cmp.size == sig.size
cmp.size.times { |i|
next unless cmp[i] < sig[i]
}
hit = cmp
end
if hit
ovr[hit].call(*args)
else
if has_generic #ovr[nil]
send( "#{name}Generic", *args ) #ovr[nil].bind(self).call(*args)
else
raise NoMethodError
end
end
I'm afraid I don't care for how the strongtyping lib handles
overloading with the #overload call internal to the method. It
undermines dynamic capabilities. Eg.
I'm afraid I don't care for how the strongtyping lib handles
overloading with the #overload call internal to the method. It
undermines dynamic capabilities. Eg.
It would be tricky to add another overload post defnition of the
original.
I would agree. You loose the most powerful part - to define methods after the definition. One good example for multimethods are in event callbacks. You can capture the event you want without a bunch of switch code:
overload :mouse_down, LeftButton, ButtonDown do |button, state, x, y|
puts "left mouse button down"
end
Unfortunately it looks a bit odd. If I was using it in a library internally it might be cool. But exposing library users to it would be something else. I wonder if it could be done like this:
class Test
generic :foo
def foo.Integer_String( x, y )
return x * y
end
end
t = Test.new
t.foo( 10, 20 )
Here's a prototype, but doesn't quite work:
def generic( symbol )
class_eval do #instance_variable_set( "@@#{symbol.to_s}", Object.new )
@@foo = Object.new
define_method( symbol ) do | *args |
# Find applicable method and call
@@foo.Integer_String( *args )
end
end
end
class Test #generic :foo (would generate the code below)
@@foo = Object.new
define_method :foo do |*args|
method = args.map { |a| a.class }.join( "_" )
@@foo.send( method, *args )
end
end
class Test
def @@foo.Fixnum_Fixnum( x, y )
return x * y
end
def @@foo.Fixnum_Float( x, y )
return x / y
end
end
t = Test.new
t.foo( 10, 20 )
t.foo( 10, 2.0 )
It's inefficient and all that, but I think it looks pretty nice
Mike
Trans wrote:
···
Thanks Andrew.
I'm afraid I don't care for how the strongtyping lib handles
overloading with the #overload call internal to the method. It
undermines dynamic capabilities. Eg.
The "methods" are executed in the context of an Object object,
instead of in the context of a Test object... And it behaves
like a singleton... And it doesn't handle subclasses of a
Class....
No, I don't think it's a usable solution...
The code below would probably work better. It uses the
monitor-function "def_overload" which uses Module#wrap_method
[1] to add functionality to an already existing method. If the
arguments of a call matches the fingerprint, the given block
will be executed, otherwise the previous definition of the same
method is checked/called.
class Module
def def_overload(method, *klasses, &block)
wrap_method(method) do |org_method, args|
nok =
args.zip(klasses).find do |a, k|
if k.kind_of?(Class)
not a.kind_of?(k)
else
not a.respond_to?(k)
end
end
if nok
if org_method
org_method.call(*args)
else
raise NotImplementedError, "Method #{self}##{method} not
implemented for arguments #{args.inspect}."
end
else
block.call(*args)
end
end
end
end
class Test
def_overload(:foo, Fixnum, :to_f) do |x, y|
x + y.to_f
end
def_overload(:foo, Fixnum, Fixnum) do |x, y|
x * y
end
def_overload(:foo, Fixnum, Float) do |x, y|
x / y
end
end
t = Test.new
p t.foo(10, 20)
p t.foo(10, 2.0)
p t.foo(10, "2")
p t.foo(10, :two) # Should fail...
Points taken for the problems of the implementation, and the metameta solution is nicely done. However, I would like to see if it's possible to use the 'def' syntax for defining multimethods. Below is a version I have been toying with, keep in mind the 'self' problem still exists and it is very inefficient. Is it possible to execute a method in a specific context, or is there a way to make the following work correctly?
class Generic
def find_method( args )
methods(false).sort.reverse.each do |method_name|
signature = method_name.split( "_" ).map do |i| eval i end
result = args.zip( signature ).map do |m|
m[1] == nil ? true : m[0].kind_of?( m[1] )
end
return method_name if result.inject( true ) { |a, i| a and i }
end
raise NotImplementedError
end
end
class MultimethodTest
@@foo = Generic.new
def @@foo.Fixnum( number )
puts 'foo( Fixnum )'
end
def @@foo.String_Numeric( string, number )
puts 'foo( String, Number )'
end
def foo( *args )
@@foo.send( @@foo.find_method( args ), *args )
end
end
t = MultimethodTest.new
t.foo( 10 )
t.foo( "Hello", 10 )
Erik Veenstra wrote:
···
The "methods" are executed in the context of an Object object,
instead of in the context of a Test object... And it behaves
like a singleton... And it doesn't handle subclasses of a
Class....
No, I don't think it's a usable solution...
The code below would probably work better. It uses the
monitor-function "def_overload" which uses Module#wrap_method
[1] to add functionality to an already existing method. If the
arguments of a call matches the fingerprint, the given block
will be executed, otherwise the previous definition of the same
method is checked/called.
class Module
def def_overload(method, *klasses, &block)
wrap_method(method) do |org_method, args|
nok =
args.zip(klasses).find do |a, k|
if k.kind_of?(Class)
not a.kind_of?(k)
else
not a.respond_to?(k)
end
end
if nok
if org_method
org_method.call(*args)
else
raise NotImplementedError, "Method #{self}##{method} not
implemented for arguments #{args.inspect}."
end
else
block.call(*args)
end
end
end
end
class Test
def_overload(:foo, Fixnum, :to_f) do |x, y|
x + y.to_f
end
def_overload(:foo, Fixnum, Fixnum) do |x, y|
x * y
end
def_overload(:foo, Fixnum, Float) do |x, y|
x / y
end
end
t = Test.new
p t.foo(10, 20)
p t.foo(10, 2.0)
p t.foo(10, "2")
p t.foo(10, :two) # Should fail...
Is it possible to execute a method in a specific context,
As long as the objects are of the same class, you could use
Method#unbind and UnboundMethod#bind. If they aren't, you could
experiment with Method#to_proc.
Is it possible to execute a method in a specific context,
As long as the objects are of the same class, you could use
Method#unbind and UnboundMethod#bind. If they aren't, you could
experiment with Method#to_proc.
Thanks, I'll give it a try.
@@foo = Generic.new
Why?...
I was trying to keep the class namespace clean and to consolidate the methods into their own space. The unbind/bind technique is interesting - now I tried making the generic object the same class, and doing this:
But alas, I get:
generic.rb:33:in `bind': singleton method called for a different object (TypeError)
I guess I don't fully understand how singleton methods differ from normal methods. Thanks for all your time, this is just a curiosity not for a project per se.
