@Janus :

isn't that PROCINFO prec-max thing for MPFR-only max prec ?

gawk-with-GMP's max exponent is approximately 2^36/37 ?

used to be +/- 2^30 i guess they upped it

On 29.03.2022 03:20, Kpop 2GM wrote:
> @Janus :
>
> isn't that PROCINFO prec-max thing for MPFR-only max prec ?
>
> gawk-with-GMP's max exponent is approximately 2^36/37 ?
>
> used to be +/- 2^30 i guess they upped it
>
>

You're replying to an old Usenet post and without context I cannot
see to what you were actually referring to and what concrete issue
your reply is supposed to address.

Janis

On Friday, April 1, 2022 at 4:33:13 AM UTC-4, Janis Papanagnou wrote:
> On 29.03.2022 03:20, Kpop 2GM wrote:
> > @Janus :
> >
> > isn't that PROCINFO prec-max thing for MPFR-only max prec ?
> >
> > gawk-with-GMP's max exponent is approximately 2^36/37 ?
> >
> > used to be +/- 2^30 i guess they upped it
> >
> >
> You're replying to an old Usenet post and without context I cannot
> see to what you were actually referring to and what concrete issue
> your reply is supposed to address.
>
> Janis

No I actually read the thread first.

you wanna operate on them ? sure :

gawk -nMbe '{ print PROCINFO["prec_max"]-1 }'
9223372036854775807

As for shifting, there's no inconsistency - you're still only getting double-mantissa worth of precision.

% gawk -Mnbe '{ print PROCINFO["prec_max"]-1; print sprintf("%.16X",or(7223372068549377581,1999)) }' <<<'';
% gawk -be '{ print sprintf("%.16X",or(7223372068549377581,0)) }' <<<''
% gawk -be '{ print PROCINFO["prec_max"]-1; print sprintf("%.16X",or(7223372068549377581,1999)) }' <<<''

9223372036854775807
643E92A0125D 3FEF
643E92A0125D 4000
001E92A0125D 47CF
9223372036854775808

1999 contributes the 7CF. bit-wise OR function took the best 52-bits of each input, OR them in a 64-bit fashion, then give you the lower 1+52 in the end.

print sprintf("%.16X",lshift(7223372068549377581,0)) }' <<<''
643E92A0125D4000

Same with shifting - the precision is already lost even upon input, before any shifting occurs. You're just shifting those same 52-bits.

prec_min of 2 ?

if it's less than 2, then what exactly do you want to do with a 1-bit wide exponent, especially when considering it still needs to reserve space for NaN and InF.