Re: Exception Handling in Draft Standard Text, V02.1
Marco, Jürgen
On 16 Mar 2010, at 16:21, Prof.Dr. Jürgen Wolff von Gudenberg wrote:
> 2.
> the examples on page 16 are wrong or at least misleading
> they set valid=0
The text was:
[Example. For the bare interval x = [0,+oo], valid(), defined(), continuous() and bounded() return 0, 0, 0 and − respectively. For x = Empty they return 0, +, + and +.]
>
> 3.
> in our opinion the bounded attribute has to be set to 0 in case of overflow.
> that is floats2interval(1,maxreal*2) =[1,oo] bounded = +
> but in usual language environments that will call
> floats2interval(1,oo) =[1,oo] bounded = -
> hence we suggest that is floats2interval(1,maxreal*2) =[1,oo] bounded = 0
> or what else ???
Thanks for this rapid evidence of careful analysis.
On (2.), this comes from the Neumaier-Hayes "least informative" rule. I hadn't noticed this produces valid=0, which elsewhere is excluded. If an interval is not known to be either valid or invalid, should it be set invalid? Or should we "do what it says on the tin" and set "possibly valid", i.e.valid=0? I tend to the latter.
I shall have to think about (3.), but at first reading it seems to be bringing a non-mathematical reason into the concept of "mathematically bounded". I guess "bounded" is really a level 1 concept that we are trying to model at level 2. That's not necessarily foolish, but the more feedback I get about "bounded", the more I feel this property is making a rod for our own back.
John