Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!samsung!usc!zaphod.mps.ohio-state.edu!usenet
From: vidynath@function.mps.ohio-state.edu (Vidhyarath K Rao)
Newsgroups: comp.theory
Subject: Procedures with multiple out parameters: why not?
Message-ID: <1990Aug15.231838.5664@zaphod.mps.ohio-state.edu>
Date: 15 Aug 90 23:18:38 GMT
Sender: usenet@zaphod.mps.ohio-state.edu
Organization: Me? Organized?
Lines: 55


[Line eaters are extinct, aren't they?]

I hope that this is the right newsgroup for this article.

I was reading E. Dijkstra, "The humble programmer", Comm ACM 15(1972), 
pp. 859--866, as reprinted in "Classics in Software Engineering" edited by
E. N. Yourdon. I came across the following [para 2 on p. 121 in the reprint]:
	A number of rules have been discovered, voilation of which will
	either seriously impair or totally destory the intellectual
	managebility of the program. These are of two kinds. Those of the first
	kind ... Examples are the exclusion of the goto-statements and of
	procedures with more than one output parameter.
	^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

While there is a lot of well-known [i.e. I have heard of them :-^] articles
on the goto statement, I have never seen heard of the theoreticians' case
against having more than one out parameter. None of the reasons I could
think of relates to theory. Can somebody out there give some pointers?

Let me list the reasons I came up with for not having it in an existing
compiler:
	1. It is hard to implement: I do not doubt this: I have heard of
functions returning records/structures being implemented in an non-reentrant
way. But that does not explain why it shold not be used in psuedo-code or in
an ideal language. [In fact R. E. Tarjan uses them in Data Structures and
Network Algorithms.]
	2. It does not increase efficiency over var parameters in accumulator
based machines or in machines with few general purpose registers: This is a
practical issue. In any case [using Tarjan's terminology] parameters being
modified shold have been initialized, where as parameters being returned
need not be. This is useful in lint and its brethern.
	3. It does not fit in ``typed Omega calculus'' [I don't know if
I can say Lamda calculus here. As far as I know, there is no such thing as
Omega calculus.]: Why not get rid of Omega calculus? Is any stronger theory
inconsistent?

  Two further remarks: Firstly, I do not question the fact that having a
single return parameter of pre-specified type allows the use of functions
in expressions. But why prohibit more than one at all times?
  Second point, in regard to 3 above: In some otherwise forgettable article
in an old issue of BYTE, I saw something to this effect: "Functions can
return only a simple value because functions in mathemtatics are required
to have a unique result". I immediately performed a gedanken experiment:
Go around telling mathemtaicians that they cannot say [in TeX mode]
	You cannot say things like "Now we will construct a function $f$
	from $X$ to $Y \times Z$."
At that time I simple chalked it upto the writer's ignorance. But reading
the cited article of Dijkstra brought the question again.

Eagerly awaiting enlightenment
--
Vidhyanth Rao			It is the man, not the method, that solves
function.mps.ohio-state.edu	the problem. - Henri Poincare
    (614)-366-9341		[as paraphrased by E. T. Bell]