From: Christoph <swap@...> (news_chr)
Date: 2003-10-03T12:29:08+09:00
Subject: Re: Instance- and class-variables (was Re: mixing in class methods)

<dblack@superlink.net> wrote:
....
> Also, every object should have the right to maintain state in instance
> variables that are not (easily) visible to other objects.  Class
> objects should not be penalized in this regard (i.e., by having their
> instance variables conflated with something else) just because they
> are Class objects.
>
> Another way to look at it is that Matz clearly wants these two very
> different things to exist, so if @@var were an alias for class <<
> self; @var; end, there would presumably be something else ($$var or
> ^^var or whatever) to represent class variables.  So then the question
> would arise again: should ^^var be equivalent to class << self; @var;
> end, and so on infinitely.  I guess the @-similarity leads more quickly
> to this kind of question than might be the case otherwise, but it's
> really just a cosmetic detail.


The behavior (scoping rules) of class variables changed a lot (for the
better) between 1.6 and 1.8 - still given murky differences like

---
class A
    @@rules = false
end

class B < A
    @@rules ||= false
    @@ruler ||= false # doesn't throw an exception but
   begin
      @@ruled &= true  # does
    rescue => mes
        p  mes
    end
end

class A
    @@rules = true
    @@ruler = true
end

class B
    p @@rules # true
   p  @@ruler # false
end
---


I can't see why ``class_attribute_accessor like'' methods in
Module with a behavior similar to

---
class Module
    def class_attr_accessor(sym)
        sym = sym.to_s
        mod = self
        module_eval <<-Body
            protected
            def #{sym}
                #{mod}.instance_eval { @#{sym} }
            end
            def #{sym}=(rhs)
                #{mod}.instance_eval { @#{sym} = rhs }
            end
        Body
    end
end

class B
  class_attr_accessor :sym
  def set=(rhs)
    self.sym= rhs
  end
end

class C < B
  def get
    self.sym
  end
end

B.new.set = "okay"
p C.new.get # okay
---

wouldn't have been enough. Considering that accommodating
class variables complicates Ruby's scoping  rules considerably,
I feel that class variables didn't buy enough to justify their
existence.



/Christoph