Re: Motion P1788/M007.01_NaI: NO, several NaIs please
I also vote NO (I currently don't see the need for NaI's).
On 2009-09-01 05:19:06 -0700, Dan Zuras Intervals wrote:
> John is quite correct & it is this last that is my meaning.
>
> NaNs in 754 have been 'historically useless' (IHMO) precisely
> BECAUSE we could not settle on a common behavior across
> platforms. No common behavior. No common code written.
> No utility for NaNs. (Well, beyond their function as a
> placeholder for undefined results.)
I don't think this is the reason. You would also need language support
and bindings. Without language support, such different NaNs would be
useless. Even the two kinds of NaN's (qNaN and sNaN) are hardly
supported by C. I don't think this is much better in most other
languages. Also, IEEE 754-1985 would have had to cover other functions
than the arithmetic operations and the square root (even if correct
rounding were not provided), to define their behavior on special
values (there's a similar problem for pow(0,0)).
You also need examples, implementations and so on, in order to show
that what is standardized works / is useful in practice.
> Still, it is early in the game for 1788. We should have that
> discussion now. If we can agree then I would vote in favor
> of that agreement.
IMHO the question whether a NaI is needed or not is too early. NaI
currently appears only in the interval constructor and doesn't seem
to really be useful here (what except detecting bugs in programs?).
I'd rather see more interesting examples based on what NaI's could
be.
> Vincent & I were trying to do just that privately & even the
> two of us could not agree. Collectively we have much more
> variance in opinion.
However some majority could more easily obtained.
Now I think we should currently focus on general cases rather than on
error-like cases.
> So, I will also cast a NO vote for this (George please note).
> For much the same reason as John but with a little bit of
> doubt thrown in there concerning our ability to come to an
> agreement that 754 could not. :-)
--
Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arénaire project (LIP, ENS-Lyon)