NetNews Usenet Archive 1993 #3/NN_1993_3.iso/spool/comp/object/4986

Text File | 1993-01-21 | 10.8 KB | 208 lines

Newsgroups: comp.object
Path: sparky!uunet!paladin.american.edu!howland.reston.ans.net!sol.ctr.columbia.edu!The-Star.honeywell.com!umn.edu!csus.edu!netcom.com!objsys
From: Bob Hathaway <[e-mail]>
Subject: Re: FAQ  Part 1 (of 2) [ polymorphism or latent typing? ]
Message-ID: <[e-mail]>
Sender: [e-mail] (Object Systems)
Organization: Object Systems
References: <[e-mail]> <[e-mail]>
Date: Fri, 22 Jan 1993 00:44:20 GMT
Lines: 196

In article <[e-mail]>, [e-mail] (Piercarlo Grandi) writes:
>The discussion on polymorphism is in another article, but let me
>register here my deep dissent over the idea that variables (or even
>worse, objects, except wheere they can change class over time) can be
>polymorphic; procedure implementations only can be polymoprphic
>(procedure interfaces too if one accepts the misleading but popular term
>"ad hoc polymorphism").

This is only with your procedural view of object-orientation.  Polymorphic
objects are the rule in an object-oriented [read "strong equivalence"] view,
which I believe is preferred as pointed out in my previous posting.  Also,
[I have done work in this area], the cases of objects becoming and assuming
(through something like pointer semantics) other objects can be unified, as
I have done in my definition of polymorphism.  In use there is a difference
(assigning or becoming), but not in definition.  Also, I show below that the
canonical use of the term "polymorphism" refers to objects that can assume
(be assigned to) many objects and does *not* usually refer to objects that
can become (or change themselves into) other objects.  This is more of
an interesting special case, usually referred to as metamorphosis and not
polymorphism, as I recall.

>I think that by polymorphic variables you mean variables that are not
>restricted at compile time to denote values of one class only, as in
>the various cases (bounded, unbounded) of latent "typing".

But this is polymorphism in programming, since it is objects assuming or
becoming many different objects, or the essence of polymorphism.  If achieved
thru latent typing, fine.  Since Strachey defined parametrically polymorphic
functions (functions parameterized by objects of different types), the
formal parameters assume many forms and can be referred to as polymorphic.
Strachey did define polymorphism in passing, he didn't give elaborate
technical definitions that I'm aware of.  Its a shame he's not around to
discuss his intentions further.

>objsys> I disagree.  Polymorphic really refers to objects that can
>objsys> assume many forms; this was standard usage (and the "morphemic"
>objsys> definition) long before Strachey.
>
>You are right in this, but Strachey redefined 'polymorphic' in the
>context of technical discussions on typing, so by now 'polymorphic'
>usually is meant to extend only to proper and ad-hoc polymorphism for
>procedures (except in special cases like value polymorphism, in the
>sense of values that can change shape).

Not true.  Most static OO languages equate polymorphism with dynamic
binding.  For instance in the C++ ARM (p209): "... the ability to call
a variety of functions using exactly the same interface - as is provided
by virtual functions - is sometimes called polymorphism."

Or even better, try Eiffel [Meyer 88:p224] ""Polymorphism" means the ability
to take several forms.  In object-oriented programming, this refers to the
ability of an entity to refer at run-time to instances of various classes.
In a typed environment such as Eiffel, this is constrained by inheritance:..."

This is almost identical to my proposed definition [by coincidence, BTW]
and is also almost identical to the definition quoted from Booch, below.

>                                             On the other hand I think
>that 'polymorphism' is not used everywhere it should; I would call
>'compiletime polymorphism' what others call 'templates'.

Cardelli and Wegner (85) even equate/rename parametric polymorphism as
generics/templates, so you're not the first.  Generics clearly provide a
static parametric case of polymorphism.

>objsys> In a nutshell, polymorphism is the ability of an object to be a
>objsys> placeholder (or formal object) for many different objects, and
>objsys> for that formal object to behave differently based on the actual
>objsys> object present.
>
>What you actually wrote is actually more like the ability of an *object*
>to change class (as in Self), which is quite a different thing from the
>ability of a variable to denote objects of different classes.

That's not what I wrote; I wrote and meant the case of assignment.  I've
unified the two cases now, although I think "becomes" is really a special
case of polymorphism.  Most modern languages (I already gave references)
use polymorphism to refer to the ability of an object to assume (or be
assigned to) many different forms of objects (Meyer is quite clear on this),
and the ability for an object to change itself, as opposed to an explicit
external assignment to a reference to an object, makes things much simpler
but can be viewed as a variation on the theme.

>obj>   class D: public B{};
>obj>
>obj>   D d;
>obj>
>obj>   b = d;    // b is an object, not a formal object, and d is sliced
>obj>   rb = &d;  // rb is a formal object (and hence, polymorphic) ref to d.
>
>This example really confirms me in my opinion that your notion of
>'formal object' is really that of 'variable that can denote objects of
>different classes', as in 'variables with latent/dynamic typing'.

[It also confirms presentation is not programming, since rb was a reference
and in C++ must be initialized with its antecedent!]

Absolutely, as do most languages designers, although I am quite familiar with
the "becomes" facility.  Even Smalltalk defines it.

>It is true that latent typing, whether bounded/inclusive or not, is the
>norm in many OO languages, but also in many non OO ones. It is also true

True, but the classical OO definition of polymorphism, or what Booch
simply defines as polymorphism [Booch 91, p517] is:  "A concept in type
theory, according to which a name (such as a variable declaration) may
denote objects of many different classes that are related by some common
superclass; thus, any object denoted by this name is able to respond to
some common set of operations in different ways."

Again, the fact that OO has polymorphism does not infer that polymorphism
is unique to OO, although the common structure in OO is usually defined in
terms of class membership.  And again, Booch and Meyer's definitions seem
to be awfully close to mine.

>that latent typing results in both overloading and polymorphism, but
>this is more a side effect than an essential characteristic. I have the

I show below why polymorphism is not a side-effect of latent typing.

>feeling that most languages with latent typing of variables are such
>because this avoided the trouble of defining a proper static checker,
>and declaring each variable, when inferential type checkers a la
>ML/Russell did not yet exist.

This may have been the reason why Smalltalk is statically typeless, but
dynamic typing is required in many circumstances.  Programmers must implement
such facilities themselves if a language does not provide support for such
cases, which undesirably adds to program complexity and size.

>I would find it rather hard to agree with those who say that latent
>typing of variables is an essential aspect of an OO language, as IMNHO
>one can still apply the OO paradigm without it.

I think that depends on what you mean by the OO paradigm.  Many consider
dynamic binding (and hence polymorphism) to be a requirement of object-
orientation, possibly deriving in origin from something like C+W's (85) 3rd
criterion for an OO programming language:

  - Types [classes] may inherit attributes from supertypes [superclasses].

where overriding can occur with specialized attributes in almost all OO
languages.

>objsys> But this occurs for polymorphic objects, not just formal
>objsys> parameters.  See the paper on Emerald (sorry, don't have the ref
>objsys> with me) for their approach to polymorphic objects; its one
>objsys> among many.
>
>Indeed these guys do not describe object polymorphism; the describe the
>ability of a variable to denote objects of different classes (and if
>they call it 'object polymorphism' they are as guilty as you are of
>inflicting confusing terminology on the public).

I think you're the one inflicting confusing terminology.  Everyone in recent
history seems to define polymorphism as a property of objects (Stroustrup,
Meyer, Booch, ...) and I've never seen anyone equate polymorphism with
structural and behavioral replacement alone - noone, never, anywhere - except
you.  That doesn't mean that everyone else is right and that you're wrong,
but this requires some thought and requires you to really prove your case.

>This has nothing to do with OO programming, and with objects. Let me
>insist (I hope that repetition works) that it usually called
>'latent/dynamic typing', as in Lisp for example, not 'object
>polymorphism'; I have seen the latter term denoting the ability of an
>object to change its class and therefore its shape after it has been
>created, as is possible in Self.

Can you provide references?  I have never seen such terminology, although
I have provided many, many cases where object polymorphism is used as I have
defined it.  I've seen "behavioral replacement" used to name a "becomes"
facility [Kim 89], but never polymorphism.

Lets look at the real world.  There is some class of objects, B, that can
have many forms, lets assume as given the case that they don't change once
a particular form is taken.  Now lets define some kinds of forms that a B
can take, and call these classes D1, D2, ...  Now take the case of where any
of the D's can occur.  In the sciences, B is refered to as "polymorphic" or
as a "polymorphic object".  I can declare a B object that can stand for any
of the D forms.  Since B can take on many differing forms, it is defined as
"polymorphic".  The property that B can assume different forms (and not the
ability to become other forms, as in metamorphosis) is the canonical use of
the term polymorphism.  So sorry, it doesn't look as though structural and
behavioral replacement (or a becomes facility) is the standard usage of the
term polymorphism.

This is also how most OO literature defines polymorphism.  You call this
latent/dynamic typing, although I clearly disagree with the dynamic typing
part, because dynamic binding can occur with static typing.  Since I believe
OO languages should support structural and behavioral replacement, I should
be an easy one to convince, yet convinced I am not, since based on the
canonical use of term polymorphism in both the natural and computer science
literatures, I believe you are wrong.

bob
[e-mail]