Re: Motion P1788/M0006.04_Level_2_Multi-format NO
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Dear John,
John Pryce wrote:
> Dear Frédéric & P1788
>
> On 8 Sep 2009, at 06:36, Frédéric Goualard wrote:
>> I am still confused here. I understand your point regarding the various
>> levels (also raised by Prof. Neumaier in one of his recent mails).
>> However, if, say, [-oo, -oo] is considered an invalid *notation* that
>> shall not be identified with the empty set, what are our options to
>> having it handled when time comes? I believe it can either be identified
>> with some "exceptional" interval (viz. NaI), or the program has to
>> "break" (flag raising, exception raising, or whatever is available in
>> the programming language considered). Both ways do not make me happy.
>> Hence my NO vote on this motion; I cannot foresee a satisfying way of
>> defining the levels that are not the subject of this motion given what
>> would be defined in the levels it considers. I would be more than happy
>> to be proved wrong, in which case I would gladly reverse my vote.
>
> Ah yes. I share your concerns. But I don't see that voting no on motion
> 6 avoids the difficulties.
>
> - There are bound to be interval constructors, for instance in a C++
> "double" version, we surely can't avoid one that looks like
> interval(double x, double y)
>
> - Then you can't avoid the fact that possible calls are
> interval(Inf, Inf)
> interval(NaN,3)
> interval(4,3)
>
> The system has to respond to these invalid calls in SOME way. Frédéric:
> what would YOU have it do?
State that any invalid form is mapped to the empty set, which Motion 6
seems to forbid (at least for interval(inf, inf)). No exceptional
behavior, no NaI.
>
> I think George Corliss has made some some crucial points in his email
> "Tagged intervals (Was Branch & bound for not everywhere defined
> constraints)" of 5 September. These are relevant to this discussion and
> have made me modify my views on returning NaI, raising flags, etc. I
> will respond to that in more detail as soon as I can.
I completely agree with Dr. Corliss conclusion in that bell and
whistles, however useful they might be, should be left out of the
present standard. In an Einstein-esque way, P1788 should stay as simple
as possible (but not simpler) and serve as a foundation for more evolved
interval types (e.g., with decorations). The real difficulty is then to
identify what shall be included to avoid closing doors on possible
extensions out of the standard.
F.
- --
Frédéric Goualard LINA - UMR CNRS 6241
Tel.: +33 2 51 12 58 38 Univ. of Nantes - Ecole des Mines de Nantes
Fax.: +33 2 51 12 58 12 2, rue de la Houssinière - BP 92208
http://goualard.frederic.free.fr/ F-44322 NANTES CEDEX 3
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iD8DBQFKpiAhEJvxJgN8HkgRAkEHAJ413YWem0IWZByFoRq6YKGTFnwd2QCgyMFc
fvsj1GboalRi+gQRuzjMtwc=
=C5Z5
-----END PGP SIGNATURE-----