0*oo in recent discussion of M7
The discussion involved mainly Vincent LefÛvre, John Pryce, Arnold
Neumaier and myself.
I had indeed misunderstood the context of Vincent's original remark
in reply to John Pryce (Date: Mon, 24 Aug 2009 17:28:40 +0200):
it was 754 and not IA. Of course, as a recently revived thread in
stds-754 mentions, there are different concepts of zero in 754 too.
In IA the issue of 0*oo only arises in computations with endpoints,
or rather, with bounds; again I was sloppy in my terminology. The
bounds may be oo (with appropriate sign only), and if so they are
not endpoints, because of Motion 3 (sets of real numbers). Those
computations are implementation artifacts and not themselves IA
computations.
Finally, the Vienna-preferred rule 0*oo = 0 need not disturb 754 too
much, as it only applies to two directed rounding modes, or possibly
all three, but not to the default rounding mode. There would still
have to be an IA-mode flag somewhere of course... ideally, part of
the arithmetic opcode together with the rounding direction.
(Another way to look at it is two new rounding modes, IA+ and IA-.)
Michel.
---Sent: 2009-08-25 12:03:35 UTC