Re: Real OO

[jhd@herold.franken.de (Joachim Durchholz)]

bobduff@world.std.com wrote 08.04.96 on Re: Real OO:

> In Eiffel, there's something called "system validity check" or "system
> goodness check" or some such thing (I can't remember what it's called).

It is "system validity". It is a check wether a call to a routine  
with a redefined parameter can result in an object being mistaken  
for a descendant one. Such systems (programs) are not system- 
valid.

The notion "system validity" is problematic in itself, because it  
can make sets of modules invalid just by combining them. Bertrand  
Meyer has proposed a different rule that is a bit more  
restrictive than the current definition of system validity, but  
which can be checked on a by-class level.

Details are on http://www.eiffel.com, in a paper called "Beware  
of polymorphic catcalls".


Anyway, I'm not so sure something like congruency can be  
naturally done in Eiffel. Eiffel deliberately lacks mechanisms  
for polymorphism in routine parameters (apart from the usual  
polymorphism in the implicit Current or this or self or however- 
your-language-names-it parameter). My OOA/OOD "bible" says this  
can always be achieved by something like
  muli-polymorphic(parm)
  begin
    parm.function()
  end
but I don't think this applies to congruency or other symmetric  
polymorphic functions. I can't comment on this without some  
serious thinking though (and I've been trying to avoid that for  
the past holidays <g>).


-Joachim

--
Im speaking for myself here.