From: Dave Thomas <Dave@...>
Date: 2002-02-18T16:17:35+09:00
Subject: Re: alias danger

Joel VanderWerf <vjoel@PATH.Berkeley.EDU> writes:

> But now I remember why that makes me itch. It's tempting to think of
> this alias-and-redefine as an operation that is insulated from the
> rest of the world, but the binding effect of alias is, of course,
> not local.  Unless you choose a very obscure name for you alias, you
> run the risk of a stack overflow:

That's always been an issue which has worried me.

An alternative suggestion: have two new keywords, 'redef' or 'prior'

  module ExtraLong
    redef length
      prior * 2
    end
  end

  class String
    include ExtraLong
  end

  "cat".length  #=>  6


redef is a method definition, like def, but it keeps the prior version
lying around (in the same way alias does). The 'prior' keyword can
then be used to invoke that previous definition. Like super, prior
with no parameters would pass on its caller's parameters, and prior
with parameters would just pass them.

Is the problem really worth adding two new keywords to the language?
Personally, I think it is. The current alias/def scheme is a bit of a
mess. As you point out, it isn't safe. It also isn't great at imparting
intent -- is the alias there to give the method two names (alias size
length) or as preparation for redefining the method?

Would the feature be used? Look through the libraries and you'll find
method redefinitions galore, so I think the answer is yes.

Matz? Would this stand a chance as an RCR?


Dave