Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site ut-sally.UUCP
Path: utzoo!linus!gatech!ut-sally!crandell
From: crandell@ut-sally.UUCP (Jim Crandell)
Newsgroups: net.bizarre
Subject: Re: Bizarre Code
Message-ID: <2710@ut-sally.UUCP>
Date: Mon, 19-Aug-85 20:06:16 EDT
Article-I.D.: ut-sally.2710
Posted: Mon Aug 19 20:06:16 1985
Date-Received: Wed, 21-Aug-85 06:41:34 EDT
References: <462@moncol.UUCP> <29712@lanl.ARPA>
Organization: U. Texas CS Dept., Austin, Texas
Lines: 33

>   The most bizarre code I ever had to program follows (in FORTRAN)
> 
>   100 IF (IREQ.EQ.0) GO TO 100

A former colleague of mine once found it necessary to write a program
with just such a line in it for a now ancient Datacraft (now Harris)
machine.  To my taste, however, the line was not nearly so bizarre as
the object code to which the Datacrock FORTRAN compiler converted it.
This compiler was exceptionally stupid, even by the standards of its
time, to such an extent that it didn't recognize that a statement of
the form IF (<logical expression>) GO TO <label> would require only
a conditional branch to <label> following the code to evaluate
<logical expression> to a result in the condition register.  Instead,
it simple-mindedly generated a conditional branch to a forward, followed
by an unconditional branch to <label> and finally the definition of the
forward.  Now it happens that branches were among the machine's fastest
instructions, so the performance impact of this asininity wasn't really
severe in most instances, but in the interesting case where the <label>
following GO TO was identical with the one in columns 1-5, the compiler
became so confused that it forgot to generate the unconditional branch,
thus reducing the statement to a conditional no-op.  Now this compiler
was endowed with such a magnificent array of charming little quirks and
truly endearing features -- I deeply regret that time doesn't permit me
to enumerate even a fraction of them -- that I came to regard the idea of
modifying it as the moral equivalent of blasphemy.  I therefore proposed
the creation of an object-code pessimizer, whose primary function was
examining FORTRAN compiler output for sequences of this type and inserting
the necessary but omitted instruction.  Unfortunately, I never had time
to reduce this revolutionary idea to practice.
-- 

    Jim Crandell, C. S. Dept., The University of Texas at Austin
               {ihnp4,seismo,ctvax}!ut-sally!crandell