Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!mailrus!accuvax.nwu.edu!krulwich
From: krulwich@ils.nwu.edu (Bruce Krulwich)
Newsgroups: comp.lang.scheme
Subject: RETURN-ARITY of a function (was Re: Request for Comments: ...)
Message-ID: <1671@accuvax.nwu.edu>
Date: 28 Nov 89 18:09:30 GMT
References: <891127-144857-1155@Xerox> <8911272329.AA07264@garlic.Stanford.EDU>
Sender: news@accuvax.nwu.edu
Reply-To: krulwich@ils.nwu.edu (Bruce Krulwich)
Organization: Institute for the Learning Sciences, Northwestern University, Evanston, IL 60201
Lines: 32
In-reply-to: mkatz@GARLIC.STANFORD.EDU (Morris Katz)

In article <8911272329.AA07264@garlic.Stanford.EDU>, mkatz@GARLIC (Morris
Katz) writes:
>For symetry reasons, a language which supports multiple return values would
>require two types of arity functions: call-arity and return-arity.  (Please
>don't flame about the names.  I just hate using foo and bar as names when I
>can think of something semantically more meaningful.)  Call-arity is the
>procedure which is more commonly called arity and returns information about
>the number of parameters which a function expect.  Return-arity is the
>procedure which returns information about the number of return values to be
>returned by a function.

I think that RETURN-ARITY is not computable in enough cases to be worth
having.  First of all, anytime the returned value(s) come from a call to
another function, RETURN-ARITY would have to descend through the call tree.
Secondly, there is no assurance that every branch of a conditional will return
the same number of arguments, so there will probably be a combinatorial blowup
in descending this tree.  Thirdly, and perhaps most obviously, the number of
values that a function returns may only be specified at run-time (such as when
they come from calling a passed in procedure, or when they vary according to
the number or contents of passed-in arguments), in which case it is impossible
to determine RETURN-ARITY altogether.

Of course, the RETURN-ARITY could be given by a programmer as a declaration
of some sort, and intelligent early-binding compilers could probably use this
information to make better code, but this is a seperate issue and certainly
can't be assumed to exist for all functions.


Bruce Krulwich
Institute for the Learning Sciences