[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
Re: Zeroes and infinities
David Hough 754R work <754r@xxxxxxxxxxx> wrote:
I think the discussions of zeros and infinities and pow have all served
to underline the observation that no matter what default is prescribed
for the exceptional situations in which they arise, somebody will take
exception to it (because it's wrong for his application). That's one
reason they are called exceptions.
Indeed. One can even prove, mathematically and rigorously, that no
single default will be best for all mathematical criteria :-)
So correct exception handling necessarily encompasses
1) default handling that best serves most applications most of the time
2) portable facilities for alternate exception handling for all the other
applications and times
Grrk. I would say (and I know that most of the IEEE 754 originators
disagree) that there is a more critical point:
0) There is an abstract model, consistent set of conventions, call it
what you will, so that all failures are detectable without more than
rare false positives. For some consistent meaning of "failures".
I'm not surprised that 1) continues to generate a lot of discussion
since there are elements of judgment involved in some of the decisions.
Yes, but a lot of THOSE arise because of the lack of (0).
754 punted on 2) by specifying optional traps that are useless in portable
programs. I'm surprised at the resistance expressed since 2001 to attempts
to begin to address that in 754R. I'd understand the resistance better if
it were expressed in terms of needs of providers and users of mathematical
software.
Part of the reason is that a large number (actually, I believe, the
vast majority) of the clued-up subset of those people have disagreements
of fundamental principle with IEEE 754. Any attempt to meet their needs
would imply changing some of the basics of IEEE 754. I can see why
that didn't happen, which is not to say that I agree with 754R.
The one exception, which can only be a political objection, is to not
include the LIA-1 mandatory mode "Terminate with message". It is, after
all, the oldest and most widespread exception handling mode of all ....
Regards,
Nick Maclaren,
University of Cambridge Computing Service,
New Museums Site, Pembroke Street, Cambridge CB2 3QH, England.
Email: nmm1@xxxxxxxxx
Tel.: +44 1223 334761 Fax: +44 1223 334679