[cfp-interest 3597] Re: Annex F Changes Discussion Paper

[Damian McGuckin]

On Mon, 15 Sep 2025, Jim Thomas wrote:

> 1.
>> nextafter(x, y) raises the "underflow" floating?point exception when x ??y
>> and the result is zero or subnormal
>
>
> Disambiguation of whether ?and the result is zero or subnormal? is a 
> qualification or a consequence depends on knowing the form.
>
> Did you consider the form
>
> 	qualification: ?

I do not quite understand. Let's discuss on Wednesday.
>
> 2.
>> Notes that some of these cases refer to odd and not odd, integer n, 
>> while others refer to odd and even, where the latter implies that 0 is 
>> a positive even number. Except for that implication, it is tempting to 
>> leave them alone. But essentially, this consistency is unresolved. 
>> Suggestions?
>
>
> Indicating specific cases would be helpful for understanding the issue.

Will do.

> 3.
>> Here Annex F currently relies (without explanation) on the implication 
>> of an inclusive upper bound of positive infinity. To avoid this lack of 
>> clarity,
>
> What?s not clear? Is it not clear that positive infinity is greater than 0?

I originally thought that

 	x > 1

implied 1 < x <= Inf(inity). But based on feedback from a few weeks ago, 
it was obvious that that the inclusive infinity limit was not clear.

Also, can I ask how would you read?

 	|x| >= 0

> 4.
>> Its negation, i.e., is not a NaN, also appears.
>
>
> Yes, this is a problem, given that NaN generally means quiet NaN.

> 5.
>> Consistency will be enforced by ensuring that the use of NaN in such cases always appear with both the ?is?
>> and the indefinite article.

> ?x is a NaN? reads better after ?if? or ?when?. ?x a NaN? reads better 
> after ?for?. Why not incorporate that be the rule?

I agree but I wanted a single way of expressing it.

> 6.
>> The mathematical domain which covers all of those eight cases is the combination of (2) and (4):
>>
>> 	(8)	0 ? |x| ? ? or x is a NaN
>
>
> If we change F.2.2 #3:
>
> ?? This annex does not require the full support for signaling NaNs 
> specified in ISO/IEC 60559. This annex uses the term NaN, unless 
> explicitly qualified, to denote quiet NaNs. ??
>
> to something like:
>
> ----- This annex does not require the full support for signaling NaNs 
> specified in ISO/IEC 60559. Signaling NaNs are generally excluded from 
> the specification. Unless explicitly stated otherwise, phrases such as 
> ?for all x? do not include signaling NaNs, and the term ?NaN" denotes 
> only quiet NaNs. ??
>
> then (8) could be replaced by ?all x?.

> The change would also give a fuller characterization of the status of 
> signaling NaNs in the annex.

I have deliberately not provided the changes yet.

And there is a cross reference that I need to find in Section 5.

Thanks - Damian