Re: Difficulties in implementation of textToInterval(s)
Michel,
> That's an issue for full 1788 (where we have rational literals). It does
> not arise in 1788.1 where all implementations use (or appear to use) b64
> endpoints.
I should be more clear. I consider interchanging between 1788 and 1788.1 .
A program in 1788 computes and outputs result for further processing by
other programs without knowledge which interval datatypes are native for other program.
The first program doesn't want to loose precision in conversions.
So it outputs interval literals by s=intervalToExact(x).
If datatype of the first program contains intervals with rational ends,
it will outputs interval literals with rational number literals.
Some of the second programs are in 1788.1 .
They should be able to read the output of the first program containing rational number literals.
-Dima
----- Original Message -----
From: mhack@xxxxxxx
To: stds-1788@xxxxxxxxxxxxxxxxx
Sent: Tuesday, May 24, 2016 7:37:08 AM GMT +03:00 Iraq
Subject: Re: Difficulties in implementation of textToInterval(s)
On Mon, 23 May 2016 21:11:53 -0700 (PDT), Dmitry Nadezhin wrote:
> I remind why I insist on rational number literals. I want interchange
> between different interval types [in] different implementations.
That's an issue for full 1788 (where we have rational literals). It does
not arise in 1788.1 where all implementations use (or appear to use) b64
endpoints.
The trouble with rationals is that there is no natural limit to p and q,
and they could be a Continued Fraction approximation of a b64 rounding
threshold, making it difficult to return the correct (tightest) hull.
If (as Dima asks later) the result of conversion were relaxed to just
"accurate" this would not be an issue.
Michel.
---Sent: 2016-05-24 04:35:58 UTC