Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!spool.mu.edu!uunet!timbuk!raphael!wws
From: wws@raphael.cray.com (Walter Spector)
Newsgroups: comp.lang.fortran
Subject: Re: I/O
Message-ID: <134822.25178@timbuk.cray.com>
Date: 30 Jan 91 20:14:35 GMT
References: <27a6aacc@ThreeL.co.uk>
Reply-To: wws@raphael.cray.com (Walter Spector)
Organization: Cray Research, Inc.
Lines: 60

In article <27a6aacc@ThreeL.co.uk>, jf@threel.co.uk (John Fisher) writes:
> mkh6317@rigel.tamu.edu (HOWARD, MATTHEW KENDALL) asks:
> 
> > Ok, ok. Uncle.  Since I don't have a copy of the "standard" I
> > will have to take your word that "equivalencing a character
> > variable to a non-character variable" is a violation of the
> > standard.    Now tell me why.
> 
> Why the sneer quotes round standard?
> 
> Reason: because it's very unportable.  Why?
> 
> 1) Endian problems
> 2) Word-length problems
> 3) Character-code problems

All three very good reasons from a portablility point of view.
Much of the use of this 'feature' is to avoid rewriting old
Fortran-66 '8HHOLLERIT,1HH' based programs when adding new code.
The new code uses character data type (for obvious maintainability
reasons), but can still get at the old existing data structures.
Porting large (.gt. 10**5 lines) Hollerith-based codes can be
a nightmare.

A fourth reason is word *alignment* problems.  Not all the world
is a VAX. (Actually, *none* of it was when the '77 standard was
created.)  Consider something approximating the following:

	character*100 string
	integer i
	equivalence (string(2:2),i)

This would, if legal, force the integer to be equivalenced to the
second character of the character variable.  Since the integer
would no longer be word-aligned, performance penalties would
result on word-addressable machines.  (I understand this is even
true for most byte-addressable machines, since the hardware still
doubles your memory references and then shifts to align - even if
invisible to the programmer.)

The same situation exists when mixing character and non-character
data in common blocks.  The '77 standard does not allow mixing
in common either.

Many compilers do support mixing of character/non-character types
in equivalence/common.  Some implementations only allow it when
alignment rules are followed to avoid the performance problems.
Consider:

	character*100 string
	integer i
	equivalence (string(1:1),i)

The integer is word-aligned, and equivalencing is possible with
no performance problems.

Walt Spector
(wws@renaissance.cray.com)
Sunnyvale, California
_._ _._ _.... _. ._.