Path: utzoo!mnetor!uunet!husc6!mailrus!ames!amdahl!pyramid!decwrl!16bits.dec.com!kruger
From: kruger@16bits.dec.com (Bear with me)
Newsgroups: comp.arch
Subject: Dusty decks
Message-ID: <8803282017.AA04296@decwrl.dec.com>
Date: 28 Mar 88 22:32:00 GMT
Organization: Digital Equipment Corporation
Lines: 26

1. A new FORTRAN compiler on a new machine is, if anything, buggier and harder
to maintain than a new C compiler. The syntax is ugly, and that affects more
than the theoretical sensibilities of computer science.

2. If you write in a language that allows the specification of data attributes
in such a way that the compiler has more information and can do more
parallelizing, that is syntactically clean, and designed for clarity, you will
in the long run save a lot of scientist-time.

3. The Tangible benefits are in faster porting to new machines, lower product
cost, wider availability AND increased reliability after an initial
"settle down." A source to source translator could at least make programs run,
although it obviously could not extract more information than was present in
the original program.

The problem with this idea is that the initial cost is high. But then, that is
true with most technology, and this is not deterring younger scientists from
building new tools in other languages. I personally feel this is a generation
problem, much as cards continue to be used (albeit thankfully rare now).

I am certainly not proposing a modified C, or even C++ as the replacement of 
choice. Though these languages have tremendous advantages for science in their
compactness, they are somewhat ambiguous, and too oriented for less efficient,
dynamic control.

dov