From: "Chr. Rippel" <chr_news@...>
Date: 2002-02-08T00:03:47+09:00
Subject: Class variable bug

It seems that the following class variable bug feel through the
reporting cracks -  the problem is that the two different class method
definition styles potentially effect the visibility of class variables
in conjunction with mixins in subclasses.

class A 
    @@var = nil
   
    class <<  self
        def first_form; @@var end
    end

   def self.second_form; @@var end

end


The following longish example illustrates this bug (1.6.5  & cvs). It
is  worthwhile to point out that once this situation is resolved then 
including modules becomes a valuable(don't know?)  idiom for  
``invalidating class variables'' in the subclass hierarchy.
 
----
class Gods
  @@rule = "Uranus"
  def ruler
    puts @@rule
  end

  def self.ruler                 # <= per method definition style
        puts @@rule
  end
end

module Olympians
 @@rule ="Zeus"
 def ruler
    puts @@rule
  end
end

class Titans < Gods
  @@rule = "Cronus"
  include Olympians
end

Atlas = Titans.new

Gods.new.ruler    # => Cronus
Gods.ruler           # => Cronus
Atlas.ruler           # => Zeus
Titans.ruler         # => Zeus      # <= I like this better

-----
versus
-----

class Gods
  @@rule = "Uranus"
  def ruler
    puts @@rule
  end

  class << self              # <= multiple method definition style    
      def ruler
            puts @@rule
      end
   end
end

module Olympians
 @@rule ="Zeus"
 def ruler
    puts @@rule
  end
end

class Titans < Gods
  @@rule = "Cronus"
  include Olympians
end

Atlas = Titans.new

Gods.new.ruler  # => Cronus
Gods.ruler         # => Cronus
Atlas.ruler         # => Zeus
Titans.ruler        # => Cronus <=  here is the difference


# extended example

def Atlas.ruler
  puts @@rule
end

Atlas.ruler          # => Zeus

class Gods 
  def Atlas.ruler
    puts @@rule
  end
end

Atlas.ruler          # => Zeus
----


/Christoph