Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!apple!snorkelwacker!think!yale!cmcl2!lanl!lambda!jlg
From: jlg@lambda.UUCP (Jim Giles)
Newsgroups: comp.lang.misc
Subject: Re: Pointers as 3-tuples
Message-ID: <14324@lambda.UUCP>
Date: 10 Apr 90 20:38:08 GMT
References: <6531@brazos.Rice.edu>
Lines: 20

From article <6531@brazos.Rice.edu>, by preston@titan.rice.edu (Preston Briggs):
> In article <Ekg#.d5@cs.psu.edu> flee@shire.cs.psu.edu (Felix Lee) writes:
>>If pointers were implemented as 3-tuples,
>>then the compiler could generate run-time checks for aliasing, and
>>then switch to either a fast- or slow-coded loop.  
> 
> Jim Giles has suggested this in terms of arrays and fortran
> (presumably extended to allow aliasing).

Actually, I recommended it as something _NOT_ to do.  The only way
to make it genuinely useful is to have special versions of the code
code each combination of aliased/non-aliased arguments.  Since this
increases combinatorially (by the definition of combinatorial :-)
with the number of _prtentially_ aliased items, this leads to an
unacceptably large number of alternatives.  You can't efficiently
implement this as a run-time test.  Hence (as I've said before),
the only solutions are to outlaw aliasing or to insist on sophisticated
implementations which do interprocedural dataflow analysis.

J. Giles