From:  Samuel Mize <[log in to unmask]>

>
> I personally wouldn't limit the term "language construct" to
> syntax-level entities, so I can agree with both of you.

I think Robert and I agree with each other, so...

> - You don't see a syntax-level "thing" that encapsulates a class.
>
> - Robert sees a semantic "thing" that provides the functionality
>   of a class (abstraction, inheritance, polymorphism, etc.)

Mm-hmm.  Me too.

> For this reason, Ada has a PR problem among people who look at the
> very shallowest surface of the language.  Some people in the Ada 9X
> process argued strongly that Ada 95 should have a keyword or
> syntax construct called "class," just to make it easier to sell
> the language's O-O features -- for absolutely no technical reason.
>
> I feel the Ada construct is the right technical choice.  The
> fact is that Ada's system is MORE powerful.
>
> - To fully and formally encapsulate classes, you use Ada's
>   encapsulation mechanism -- the package.  Put each class in
>   its own package.
>
> - To tightly couple some classes, you can put them into the
>   same package, so each can see the other's implementation.
>   (This is often a bad idea, but it is occasionally quite
>   useful -- rather like keeping GOTO in the language.)
>
> - I have also used tagged types to simplify coding in a purely
>   procedural program.  From what I've read about typical C++
>   usage, this is not uncommon -- but in C++ it's more confusing,
>   because you have to define bogus "classes" for non-objects.

Yes to all that.  Well put.

There's also the fact that in C++, dispatching is coupled to whether the
call is by reference or by value, which seems arbitrary and encumbering.
In Ada, control over dispatching is orthogonal to reference/value
semantics.

> Note that I'm not a O-O totalitarian.  Some people thing
> EVERYTHING should be object oriented, NOTHING should be
> procedural.  They will knock a language that lets you organize
> your design in any other way.
>
> I personally think that's hogwash, and I very much like the
> integrated procedural and object-oriented facilities of Ada.

Right.  I think pure OO was an interesting way to explore the concepts
of encapsulation, inheritance, and polymorphism.  Those concepts aren't
ends in themselves but techniques for making programs better.  I would
say that good factoring, good cohesion and appropriate coupling are ends
in themselves.  I could be dogmatic about those...

Alexander Stepanov (of C++ STL fame) notes that saying
"everything is an object", like other philosophies that claim
"everything is an X", isn't that interesting and is of little use.
It's inherently question-begging.  When someone says "everything is an
X", they haven't said all that much, 'cause we already know what
"everything" means... :-);

-- Mark


    Q. What did the Zen monk say to the hot-dog man?
    A. "Make me one with everything!"

    badump-bump!