Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!cs.utexas.edu!sm.unisys.com!ism780c!news
From: news@ism780c.isc.com (News system)
Newsgroups: comp.arch
Subject: Re: Semantics (was Software Distribution)
Message-ID: <17934@ism780c.isc.com>
Date: 13 Oct 88 20:49:26 GMT
References: <13914@mimsy.UUCP> <1000@esunix.UUCP>
Reply-To: marv@ism780.UUCP (Marvin Rubinstein)
Organization: Interactive Systems Corp., Santa Monica CA
Lines: 23

In article <1000@esunix.UUCP> bpendlet@esunix.UUCP (Bob Pendleton) writes:
>Defined as undefined is a perfectly valid way to define the actions of
>an operand on particular subranges, like * on NULL and square root on
>negative numbers.  But, from the point of view of trying to make
>programs machine independent it is a useless definition. This kind of
>definition forces me to test a program on every existing machine
>architecture, using every existing compiler for each architecture, and
>even on every new release of each compiler, before I can claim that
>the program is machine independent.

Defining it won't make it so.  Even if I wrote programs in a language where
the semantics for every legal syntactical construct was *EXACTLY* specified,
I would still test my program on every combination of machine and compiler
for which I would like to guarantee that my program runs correctly.  For
although I never make mistakes :-), I cannot say the same for the people who
write operating systems and compilers.  (I have even encountered hardware
that does not work as advertised.)

Would anyone guarantee a program distributed in a form that required that the
customer compile it and link it before using it?


   Marv Rubinstein