Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!bellcore!decvax!decwrl!glacier!kestrel!ladkin
From: ladkin@kestrel.ARPA (Peter Ladkin)
Newsgroups: net.cse
Subject: Re: Value of Computer Science degree
Message-ID: <5284@kestrel.ARPA>
Date: Fri, 28-Feb-86 19:08:47 EST
Article-I.D.: kestrel.5284
Posted: Fri Feb 28 19:08:47 1986
Date-Received: Sat, 1-Mar-86 23:34:25 EST
References: <1404@ames.UUCP> <3367@umcp-cs.UUCP>
Organization: Kestrel Institute, Palo Alto, CA
Lines: 61

In article <3367@umcp-cs.UUCP>, mangoe@umcp-cs.UUCP (Charley Wingate) writes:
> [...] it seems clear to me that
> courses in software engineering ought to be part of CS curricula, and they
> should include group projects.

Agreed, and they are all too rarely. Group programming is possible
without a special course. Students in my compiler and OS courses
are encouraged to work in groups. The projects are stiff enough that
they have to.
I have long been puzzled as to what should be in a software
engineering course. It shouldn't just be Programming Methods N
(for large N) else it will bore teacher and student alike.
My favorite current candidate is Specification. There is 
much more to this than I had realised before I worked with
(and designed) specification languages. 
Many of the problems are general enough to attract anyone's
attention.

A sample:
Is first order logic an adequate specification language?
(Traditional verification vs Gordon's HOL hardware
verification examples).
Are algebraic specifications adequate for Abstract Data Types?
Does one need some set theory?
Pick a problem - what is the most abstract specification of
this problem? Is there one? Does it generalise?
Pick an area (e.g. time modelling). How does one model
temporal information in a given setting?
(Temporal Logic, Interval Calculi, First order axiomatisations).

These issues are not at all theoretical. I am currently
working in an environment which allows me to define
functions assertionally (with a fragment of first order
logic and set theory) or procedurally. I have designed
a time calculus, and can implement it using transformations
from intervals to sets, defined assertionally. I would 
also claim (with theorems to back it up) that in a precise
sense the time calculus is most general for countable
time modelling (and who cares about higher infinities
for applications).
I am an experienced logician, therefore I have been used
to specifying in first order logic. I nevertheless found
that finding good specifications of problems in computer
science was a subtle art that could be as difficult as
you could wish for.
I also know that most of my students, even the good ones,
lack the ability and training in specification that I wished
they had. I used to think a training in logic was adequate,
but now I believe the problem is more general.

I would appreciate suggestions for extending my sample
list above. I would prefer a curriculum oriented to
mental problem-solving, since I believe that in ten years
our software engineering environments will be much more
sophisticated, and somewhat similar to the one in which
I currently work. In such an environment, training in
*abstract* specification is likely to be much more
important than having programmed a large editor in Pascal
10 years before.

Peter Ladkin