Re: Please vote against Motion P1788/0019.01: Explicit/Implicit idatatypes
Svetoslav Markov wrote:
The only new argument seems to be the contradiction
to motion 16, which is procedural question.
Indeed, the old arguments are already compelling enough
to reject the motion.
Where is an efficiency analysis for an outward rounding
exponential or square root, say?
Where is a public pilot implementation of all elementary functions
that shows which requirements on a midrad implementation beyond
simple valid enclosure can be imposed meaningfully?
Two years of discussion of the standard didn't even produce a
position paper on the matter that addresses this sort of questions.
50 years of interval computing that have not produced a single
outward rounding midrad implementation including some of the
standard functions would be reason enough not to create a standard
that allows it while not providing what already proved useful and
tested.
It would be a severe mistake to let this poor state of affairs
lead to an impotent standard which imposes essentially no restriction
upon what conforms to the standard.
To have a label that allows everything instead of only the best
is worse than having no standard at all.
If the current standard prohibits mid-rad-only then better this is
metioned explicitely in the name of the standard:
"IEEE Standard for Inf-sup Interval Arithmetic".
Otherwise it will be not good for the reputation of the standard.
Only in the eye of some midrad afficionados.
For the majority of interval users (and probably for _all_ people
outside the interval community), a strong = simple and restrictive
standard will make for a much better reputation.
Arnold Neumaier