From: Paul Brannan <pbrannan@...>
Date: 2002-09-12T00:01:33+09:00
Subject: Re: how to make proxy objects

On Wed, Sep 11, 2002 at 07:27:16PM +0900, Tom Sawyer wrote:
> class Module
>   public :undef_method
> end
> 
> class Proxy_Object
>     def initialize(peer, oid)
>       methods.each { |mid|
>         if not ['inspect',
>                 'instance_eval',
>                 'instance_variables',
>                 '__id__',
>                 'id',
>                 'new',
>                 'methods',
>                 'method_missing',
>                 'undef_method',
>                 'class',
>                 '__send__',
>                 'send'].include?(mid)

I think this belongs in the body of the class definition, and not
in the constructor; otherwise, you have to iterate over all the methods
in Proxy_Object every time you create a new Proxy_Object; this will slow
down construction (the same problem that plagues SimpleDelegator).

Also, new() and undef_method() are class methods and not instance
methods, so I don't think they belong in this list.

If you don't override inspect(), it might be nice to still be able to
inspect the remote object.

Note that Ruby's Delegate takes a slightly different approach; instead
of keeping a static list of preserved methods, it assumes that all
methods in Kernel will be preserved except 6 specific methods.

You might also want to look at drb and see how its proxy objects are
implemented.

>           self.class.undef_method(mid.intern)

Instead of making undef_method public, you could use:
            self.class.instance_eval { undef_method(mid.intern) }

If you move the entire loop outside initialize(), then you can call it
directly:
            undef_method(mid.intern)

>         end
>       }
>     end
> 
>     def method_missing(mid, *args)
>       ...act as proxy...
>     end
> end