Path: news.cs.au.dk!olm
From: olm@cs.au.dk (Ole Lehrmann Madsen)
Newsgroups: comp.lang.beta
Subject: Re: Is Beta a statically typed language?
Date: 10 Jan 1999 21:52:44 GMT
Organization: University of Aarhus, Department of Computer Science (DAIMI)
Lines: 101
Message-ID: <77b7fc$l2o$1@xinwen.cs.au.dk>
References: <VA.000001c4.01626e6a@enterprise>
NNTP-Posting-Host: sunrise.cs.au.dk
X-Trace: xinwen.cs.au.dk 916005164 21592 255.255.255.255 (10 Jan 1999 21:52:44 GMT)
X-Complaints-To: news@cs.au.dk
NNTP-Posting-Date: 10 Jan 1999 21:52:44 GMT
X-Newsreader: NN version 6.5.3 (NOV)
Xref: news.cs.au.dk comp.lang.beta:11786

In <VA.000001c4.01626e6a@enterprise> Jean-Louis Leroy <jll@skynet.be> writes:

>In my explorations I came up with the following oddity:

>(#
>   Fruit: (#  #);
>   Apple: Fruit (#  #);
>   Banana: (#  #);
>   Animal: (# Food:< Object; f: ^Food enter f[] do INNER #);
>   Monkey: Animal (# Food::< Banana #);
>   cheetah: ^Animal
>do
>   &Monkey[]->cheetah[];
>   &Banana[]->cheetah;
>   &Apple[]->cheetah
>#)  

>This program compiles but fails at runtime because of a 'qualification 
>error in reference assignment'. In fact my program contains an implicit 
>downcast which is quite well hidden.

The compiler gives you a warning about this unless you have used
the compiler siwtch '-q' which eliminates warnings.

>This experiment was inspired by the C++ FAQ that goes 'why doesn't 
>Monkey::eat(Banana&) override virtual Animal::eat(Fruit&)?'. It is also 
>a Frequently Proposed Extension, sometimes under the moniker 'covariant 
>method signatures'.

>The canonical reply/rebuttal is 'because it would open a hole in the 
>type system'.

>This posits Beta somewhere between typeless and statically typed 
>languages, doesn't it?

You are right. BETA is not completely statically typed.
The issues goes all the way back to Simula where so-called
reverse assignments were allowed. Consider declarations

   F: ^Fruit; A: ^Apple
   
The assignment 

   A[] -> F[]
   
is of course legal, and this can be detected at compile-time.
The assignment

   F[] -> A[]
   
is legal if F refers to an Apple or a subpattern of Apple.
This can in general only be checked at run-time. Simula and
BETA allows such assignments and inserts a run-time check.

You (covariant) example above is another example of reverse assignment
where a run-time check is needed.

We have discussed these issues in several papers
(the references below are as precise as I remember them)

	Madsen, Moller-Pedersen,Magnusson:
	Strong typing of object-oriented languages revisited
	OOPSLA'90
	
	Madsen: 
	Open issues in object-oriented programming
	Software Practice & Experience, vol. 25(S4), December 95

	Thorup
	Genericity in Java with Virtual types
	ECOOP'97
	
	Torgersen:
	Virtual types are statically safe
	FOOL'5
	
	Madsen: 
	BETA, a block-structured, concurrent, object-oriented language
	In Handbook of Object Technology (S. Zamir ed.)
	CRC Press, 1999
	
	and I think it is discussed in the BETA book also:
	Madsen, Moller-Pedersen, Nygaard:
	Object-oriented programming in the BETA Programming language
	ACM Press/Addison Wesley, 1993
	
Finally I am giving an invited a talk on these issues at

	The Sixth International Workshop on
        Foundations of Object-Oriented Languages
	FOOL 6, January 23, 1999
	San Antonio, Texas, USA
	http://www.cs.williams.edu/~kim/FOOL/FOOL6.html
	
Thanks for this opportunity to provide some commercials:-)

---olm

>Jean-Louis Leroy
>http://ourworld.compuserve.com/homepages/jl_leroy/