From: Pit Capitain <pit@...>
Date: 2001-08-16T22:17:11+09:00
Subject: [ruby-talk:19893] Re: Why not?: Assigning to self

On 16 Aug 2001, at 20:57, hipster wrote:

> On Thu, 16 Aug 2001  17:40:48 +0900, Pit Capitain wrote:
> [snip]
> > Take the famous Shape class hierarchy:
> > 
> > class Shape; ...; end
> > class Rectangle < Shape; ...; end
> > class Square < Rectangle; ...; end
> > 
> > Suppose you have a Rectangle instance and use one of its resizing 
> > methods so that the new height equals the new width. After that it 
> > would be nice if the Object had become a Square instance.
> > 
> 
> FYI:
> 
> The C++ FAQ discusses an identical inheritance problem: circle
> vs. ellipse -- and provides some valuable insights along the way.
> 
> http://www.parashift.com/c++-faq-lite/proper-inheritance.html#[21.6]
> 
>         Michel

Thanks for the link. It clearly shows the conflict between the 
intuition (a circle is a kind of an ellipse) on one side and models 
implemented in languages without #become on the other side.

To make it clear what I was thinking of (substituting rectangle for 
ellipse and square for circle):

class Rectangle
  def set_size( width, height )
    if width == height
      become Square.new( width )
    else
      @width, @height = width, height
    end
  end
end

class Square < Rectangle
  def set_size( width, height )
    if width == height
      super
    else
      become Rectangle.new( width, height )
    end
  end
end

This implementation meets the (at least my) intuition.

Now this code should be improved so that Rectangle doesn't need 
to know Square, and better yet so that the "predicate subclass" 
(as it seems to be called in cecil), that states that a Square is a 
Rectangle with @width == @height, is coded only once. But I think 
I don't know how to do that, and since we don't have #become 
(yet?), all this is only theory anyway.

Regards, Pit