Comments on Motion 36 (Interval flavors)
Note: I am quite late, and these are my current comments on Motion 36
(Interval flavors) without taking into account the other comments yet.
The standard document must at least specify set-based intervals (as
done until now). I am OK with the support and specification of other
flavors, but they shouldn't delay the approval of the standard.
In the rationale:
A typo: change
m(X op Y) = mx op my
to
m(X op Y) = mX op mY
About the domain, I would change
for all intervals X,Y in C.
to
for all intervals X,Y in C such that sX op sY is in sC and/or
mX op mY is in mC.
Basically, this means that if an operation is internal in one flavor,
then it must be internal and give a compatible result in all the other
flavors. Otherwise there should be two separate operations.
> Queries: Should there be an "agreed" minimum subset of TP that each
> implementation shall support (say, at least the binary64 inf-sup type)?
I don't think so, except for 754-conforming implementations.
> The result I'm aiming for is that
>
> (*) A Level 2 program that only ever uses common intervals
> runs identically whether it is linked with the "modal"
> or the "set-based" library.
I agree, but assuming that whenever an output is an interval,
it is a (well-defined) common interval.
--
Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <http://www.vinc17.net/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.net/blog/>
Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)