Path: news.daimi.aau.dk!olm
From: olm@daimi.aau.dk (Ole Lehrmann Madsen)
Newsgroups: comp.lang.beta
Subject: Re: Mantatory argument
Date: 14 Jun 1997 16:54:01 GMT
Organization: DAIMI, Computer Science Dept. at Aarhus University
Lines: 80
Distribution: y
Message-ID: <5nuib9$lrb$1@gjallar.daimi.aau.dk>
References: <wogg1up64ix.fsf@byleist.ifi.uio.no>
NNTP-Posting-Host: ariel.daimi.aau.dk
X-Newsreader: NN version 6.5.1 (NOV)
Xref: news.daimi.aau.dk comp.lang.beta:11146

In <wogg1up64ix.fsf@byleist.ifi.uio.no> Alf-Ivar Holm <alfh@ifi.uio.no> writes:

You are right, that this kind of error may be a problem, and
it would be nice to catch it at compile-time. There is actually
a suggestion for a syntax for this:

Optional enter/exit parts:

	foo: (# ...
	     enter?(x,y)
	     do ...
	     exit? z
	     #)

Mandatory enter/exit parts:

	bar: (#
	     enter(x,y)
	     do ...
	     exit z
	     #)


The enter/exit lists of foo are defined as optional, in the
sense that input/output arguments may be specified.

The enter/exit list  of bar are mandatory meaning that enter/exit 
arguments must be supplied.

The current enter/exit parts of BETA thus correspond to the
optional enter/exit parts, even if the current syntax is the same
as the one for mandatory.

Of course an optional enter-part can be combined with a mandatory
exit part and vice versa.

This proposal was, however, never implemented. BETA has always
beeen guided by a minimalistic approach, i.e. as few constructs 
as possible. For mandatory enter/exit parts we decided to
find out how much of a problem it would be. So far few people
have complained. I have been beaten by it myself and maybe others
have the same experience?

---olm

>I have a library with some functions.  These functions does only
>provide something useful if given an argument.  If I, by mistake,
>don't assign any values to these functions, the program continues, but
>the value(s) produces by the functions may break something else later
>on.

>Two problems: 

>	1) The error is not catched where it occurred, and
>	2) The error is not found before run-time.

>Problem 1) may by fixed by adding something extra to each function
>that give a warning or error at runtime if the functions are not
>called with an argument.  Something like:

>	myFunc:
>	(# in, out: @integer;
>   	   entered: @boolean;
>   	enter (# enter in do true->entered #)
>   	do (if not(entered) then notification(# ... #) else ... if);
>	exit out
>	#);

>This is still no remedy for problem 2).  I have to wait for the
>erroneous statement to occur run-time.  What I would like were some
>kind of construct that would let me say that some patterns must be
>assigned values.  It may be a bit non-Beta-ish to prevent patterns from
>being used as any evaluation type.  OTOH, Beta does catch a lot of
>errors compile-time, and this seems to be another case where it would
>not be too hard for the compiler to find the error, given some
>indication that the pattern must evaluate it's enter part.

>Is it only me who would like this?  Have I missed something?

>	Affi