Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: YES P1788/M0029.02:Level3-InterfaceOnly, *BUT*



Baker wrote:
>  On 12/29/2011 05:00 PM, Michel Hack wrote:
>  > Lee Winter wrote:
>  >> I believe the exclusion of overflows of the same sign is unnecessary
>  >> for the interchange format.
>  >
>  > I suppose you mean two infinities of the same sign.  Infinities may
>  > indeed be the result of containing overflow, but they can also simply
>  > mean "unbounded".  Decorations should tell these cases apart.  (In 754
>  > it is the overflow flag that does the job -- unfortunately globally.)
>
>  Although we have been discussing this for a while, the distinction
>  is still not clear to me, at least on a philosophical level.

Remember my claim that there are two philosophically different interpretations
of intervals: (a) imprecise single values, and (b) ranges of multiple values.
In case (a) the concept of unbounded only arises when dividing by an imprecise
value which could be zero -- and perhaps an exception should be noted.  In
case (b) a range may be naturally unbounded on one or both sides.

> Can't an overflow be viewed simply as a "round to unbounded?"

Yes, but only because we have no other way to preserve containment when
using traditional floating-point representations.  There are in fact
representations that avoid overflow, such as level-index representations.

Michel.
---Sent: 2011-12-31 00:12:42 UTC