Re: Motion P1788/M0006.04_Level_2_Multi-format NO

[Arnold Neumaier <Arnold.Neumaier@xxxxxxxxxxxx>]

Frédéric Goualard schrieb:
> On 9/7/09 8:46 AM, "Vincent Lefevre" <vincent@xxxxxxxxxx> wrote:
> > On 2009-09-07 09:46:46 +0200, Frédéric Goualard wrote:
> > > My vote is NO because the associated position paper definitely
> > > assumes---even if peripherally---the implicit (e.g., [-oo, -oo] and
> > > [+oo, +oo] have no meaning, which should make them NaIs, whatever
> > > that object is) or explicit existence of NaIs, a concept I am not in
> > > favour of.
> > This is not what the paper says. "No meaning" currently implies
> > undefined behavior, not the existence of some form of NaI.
> > Having to deal with such meaningless representations could
> > imply performance penalties.
>
> Indeed. Anyway, I wonder how such an undefined behavior would be handled
> practically by programming languages that lack exception facilities.
> External flags (such as the infamous C errno) do not seem that good an
> option. Should we really put this burden on implementors (the Standard
> being language agnostic) while identifying [-oo,-oo] and [+oo, +oo] to
> the empty set appears to have so few drawbacks (correct me on that one:
> I have seen many discussions in the Motion 7 thread on that topic and
> may have overlooked some definitive argument against it)?

I don't understand the problem discussed.

Since there has not been any motion on conversion of [x,y] to intervals,
and intervals are agreed by Motion 3 to be closed. connected sets of 
reals in their usual notation, [-oo,-oo] is meaningless on the 
mathematical level. And only that was asserted by this motion.

Nothing at all has been said on its meaning as a pair of IEEE datums.
This would have to be discussed together with the meaning and support
of nonstandard intervals and/or of conversion/constructors, but not
as part of this motion.


Arnold Neumaier