Re: [bug-gawk] MPFR rounding issue

[arnold]

I agree with Andy that the code is using MPFR routines as
requested, and I don't think there's a bug in gawk.

I agree that the doc may be at fault here. That chapter was written
by the person who did the MPFR integration, but said person is no
longer involved with the project and hasn't been for close to 5 years.
I have edited that chapter to improve the prose and content flow, but
the technical statements therein I took at face value.

Andy - sounds like we need to revise the doc to say "rounds away
from zero" in all appropriate places.

Arnold

"Andrew J. Schorr" <address@hidden> wrote:

> On Mon, Nov 20, 2017 at 10:52:25AM -0700, david kerns wrote:
> > I'm sorry I can't assist with actual test data, but compare the results of
> > 2.21 vs (3.21 or 2.31)
>
> bash-4.2$ printf "2.21\n3.21\n2.31\n" | gawk -M -v ROUNDMODE=A '{printf 
> "%.1f\n", $1}'
> 2.3
> 3.3
> 2.4
>
> It rounds up in all cases. Isn't that the intended behavior of
> the MPFR_RNDA? The mpfr docs define it as:
>       * `MPFR_RNDA': round away from zero.
>
> > reference https://en.wikipedia.org/wiki/Rounding?#Round_half_to_even ...
>
> But this is not "half-to-even" rounding mode. It is "round away from zero".
>
> > BTW I might be completely off kilter on this :P
> > but as Andy notes, it's just using the value the MPFR library returned
>
> I'm not an expert in this stuff, but the behavior seems consistent with
> the MPFR docs.
>
> Maybe the problem is in the gawk docs. In the "Setting the Rounding Mode"
> section of the manual, it says:
>
> https://www.gnu.org/software/gawk/manual/html_node/Setting-the-rounding-mode.html
>
> Rounding mode IEEE name       ROUNDMODE
> Round to nearest, ties to even        roundTiesToEven "N" or "n"
> Round toward positive infinity        roundTowardPositive     "U" or "u"
> Round toward negative infinity        roundTowardNegative     "D" or "d"
> Round toward zero     roundTowardZero "Z" or "z"
> Round to nearest, ties away from zero roundTiesToAway "A" or "a"
>
> So maybe that's the confusion. The gawk documentation of "A" says
> "Round to nearest, ties away from zero", but the actual implementation uses
> MPFR_RNDA, which rounds away from zero.
>
> Perhaps this is simply a bug in the gawk documentation?
>
> Regards,
> Andy