Private vs. protected, naming conventions etc.

Question

I've been grappling with a couple of frustrations with Forte, and I'm
interested in feedback from others on this list regarding approaches
they may have adopted to address these.
One is that in the Forte workshops there is no way to view only the
public methods and attributes of a class (we're still using V2 here; I'm
assuming that V3 has not changed this). While referring to appropriate
technical documentation for a class is obviously important, I still find
myself opening up classes in the workshops to inspect the methods and
attributes available. (What I really want to see is an integrated class
browser. I sure hope Forte is working on something like this, or will
open up their development environment to support third-party extensions.
But that's an aside.)
A convention I just recently adopted in my work is to name private
methods and attributes with a beginning underscore ("_"). That way the
private elements are sorted to the top of the list and can be easily
differentiated from public elements. I'm curious, though, whether others
have adopted similar or different approaches.
I've also felt a bit frustrated over the lack of support for protected
attributes/methods for TOOL classes. This strikes me as a rather
bothersome shortcoming. The only approach I can think of is to make such
elements public, but adopt the same or similar naming conventions as a
strong hint to developers to avoid using these in clients of these
classes. Again, I'd be very interested in hearing how others have dealt
with this issue.
Thanks.
Michael Brennan
Programmer/Analyst
Amgen Inc.

Answer 1

I sent this once before, but the list seemed to be having trouble late last
week. If you get two copies of it... my apologies.
OK, I couldn't resist joining the fray...
At 10:56 AM 11/6/97 -0800, Michael Brennan wrote:
>
A convention I just recently adopted in my work is to name private
methods and attributes with a beginning underscore ("_"). That way the
private elements are sorted to the top of the list and can be easily
differentiated from public elements. I'm curious, though, whether others
have adopted similar or different approaches.You might even designate a single character before the underscore to denote
that, just in case some environment (CORBA) doesn't like the "_". You could
make it something like "Q" or "Z" or something that wouldn't normally be
used alone at the start of a name.
>
I've also felt a bit frustrated over the lack of support for protected
attributes/methods for TOOL classes. This strikes me as a rather
bothersome shortcoming. The only approach I can think of is to make such
elements public, but adopt the same or similar naming conventions as a
strong hint to developers to avoid using these in clients of these
classes. I share your desire for protected methods, but I have to disagree about
protected attributes. Philosophically speaking, protected and public
attributes are EVIL!! (I say "philosophically speaking" because, in the
Forte environment, there are some valid reasons for using them based upon
the visibility constraints of the language. In other languages, C++ and
Java, for example, it's not even philosophically speaking - they're just
evil!!)
One of the principal reasons for adopting the object paradigm is to
tightly control the impact of change - to provide good boundaries of
encapsulation that change does not ripple beyond. If you think about it,
one of the measures of the success of a superclass is the number of
subclasses that it has (especially for a good dabstract interface). This
says you have very nicely captured the semantics of the application domain
in the interface of the superclass. So, let's imagine a superclass with
protected attributes that are used by each of its 100 subclasses (probably
more than you would have, but I'm illustrating my point - incidentally, I'm
not talking about a hierarchy 100 deep; I'm talking about 100 subclasses
that are all direct decendants of the superclass). Now you go and change
one of the attributes. You must go look at all 100 subclasses to determine
the impact of change. This is exactly the kind of thing the object paradigm
was designed to eliminate.
Protected methods, on the other hand, would be nice.
And At 12:06 PM 11/6/97 -0800, Mark S. Potts wrote:
>
Forte inherits in a strange way when attributes are private. A
superclass attribute that is made private is not accessible from any of
its subclasses - this means that many of what you would consider private
attributes in fact have to be public. Well, the definition of private means "not visible outside of the class
where it is defined". I find it useful to think of the level of visibility
the same as secrets. There are things that are not really secrets at all -
it's ok if anyone knows them ("My name is Stephen"). These are public.
Then, there are things that it's ok if my family knows, but I don't want
the world to know - familial secrets, if you will ("I belch at the dinner
table when I'm at home"). These are visible to descendant classes and we
call them protected. Finally, there are things we don't want anyone else
to know, no matter who they are ("I poisoned my mother-in-law"). These are
private. We don't want anyone outside of ourselves to know these things.
These are the classic definitions of public, protected and private (perhaps
classic only because C++ defined them that way and everyone else just
copied what it meant).
Private attributes are not meant to be inherited by their subclasses.
That's why they're private. And, yes, I would argue that that is completely
correct. What you want, if you want them to be visible to subclasses, is
"protected". Now, Forte doesn't support protected, but that's a different
arguement - perhaps even an enhancement request.
We also should not confuse what we need to do in a language/environment
with what good OO principles are. For example, good OO design principles
state that you do not have public or protected attributes. Period! You
access them via accessors and mutators defined on the appropriate class.
Now, in some environments, this will not give you the performance you need,
so you open things up a bit. But, you shouldn't convince yourself that
doing this is the ideal design, just that it was necessary for performance.
The real problem here is that the performance of accessor and mutator
functions is not fast enough. That's why we open it up. Not because it is
good design. The proper way to fix the problem is to make accessors and
mutators fast enough so that they can be used (C++, for example, does this
with "inline" - not that C++ is my favorite language, it's not. But they
have fixed this one area nicely.)
Some would argue that this is correct and that inheritance does break thepure rules
of encapsulation I don't think inheritance, properly handled (and Forte does properly
handle it) breaks any rules of encapsulation. I would argure that the way
they treat private attributes is quite correct.
- but these people dont build applications!Hmmm... let's see... started building OO applications in 1985 (and building
them ever since) in complex application domains like CAD/CAM/CAE, Air
Traffic Control, Graphics/Imaging, Telecommunications, e-commerce,
entertainment,... ...wrote (and teach) the Forte OO Analysis and Design
course.
I guess you're right. I don't build applications. I build robust,
maintainable, extendable applications. ( ;-) ...nudge, nudge!)
Stephen