From: kaufmann@CLI.COM (Matt Kaufmann)
Date: Mon, 22 May 95 15:43:05 CDT
Message-Id: <9505222043.AA04830@thunder.cli.com>
To: bshults@fireant.ma.utexas.edu
Cc: John.Harrison@cl.cam.ac.uk, qed@mcs.anl.gov
In-Reply-To: <199505222032.PAA14060@pythagoras.ma.utexas.edu> (message from Benjamin Price Shults on Mon, 22 May 1995 15:32:07 -0500)
Subject: Re: Undefined terms
Sender: owner-qed@mcs.anl.gov
Precedence: bulk

Actually, I think there may still be a difference between your [5] and
Harrison's [2].  I interpreted [2] to allow different "default" values for
different function applications, but I interpreted your [5] as a proposal to
use one error value for all occasions, kind of a "bottom" (in the Scott sense).
In that sense, your [5] is a sort of strengthening of [1], in that you are
fixing one value for all "erroneous" applications.  I suppose that one could
also view your [5] as a sort of [4], i.e., "bottom" is a way to make sense out
partiality.  Now to confuse the matter, I can't help but mention that this ties
into [3], in the sense that one could conceive of a "bottom" of each type.

Stop me before I write again!

-- Matt