Re: [platform-testers] Release candidate mpc-1.3.0rc2

[Paul Zimmermann]

       Dear Rob,

mpc_abs(NaN+I*Inf) = Inf is consistent with hypot(qNaN,Inf) = Inf in IEEE 754,
with the meaning that mpc_abs(x+I*Inf) = Inf for any input x (either normal
or +Inf or -Inf).

Best regards,
Paul

> From: sisyphus <sisyphus359@gmail.com>
> Date: Tue, 6 Dec 2022 22:09:57 +1100
> 
> 
> [1:text/plain Show]
> 
> 
> [2:text/html Hide Save:noname (4kB)]
> 
> In fiddling about with mpc_eta_fund() I noticed that, for an mpc_t that has
> real part NaN and imaginary part Inf, mpc_abs() sets the absolute value to
> Inf.
> I would have expected the absolute value to instead be Nan - because sqrt
> ((NaN^2) + (Inf^2)) is NaN.
> 
> As it stands, with that return value of Inf, and the current documentation
> of mpc_eta_fund(), that mpc_t lies "in the fundamental domain".
> 
> Similarly, if the real and imaginary parts of this mpc_t are interchanged,
> then mpc_abs() still sets an absolute value of Inf.
> 
> I'm feeling a bit lazy and at this stage I've only tested this using the
> perl interface ... I don't think that's where the problem lies ... apologies
> if it turns out I'm wrong about that.
> 
> >perl -MMath::MPC=":mpc" -le "print MPC_VERSION_STRING; $rop =
> Math::MPFR->new(); $op = Math::MPC->new('NaN' + 0, 'Inf' + 0); print $op;
> Rmpc_abs($rop, $op, 0); print $rop;"
> 1.3.0rc2
> (@NaN@ @Inf@)
> Inf
> 
> On looking a bit further, I see mpc_abs() just calls on mpfr_hypot().
> Again, using the perli interface only, it seems that mpfr_hypot() is the
> culprit:
> 
> perl -MMath::MPFR=":mpfr" -le "$rop = Math::MPFR->new(); $op1 =
> Math::MPFR->new(); $op2 = Math::MPFR->new('Inf' + 0); print $op1; print
> $op2; Rmpfr_hypot($rop, $op1, $op2, 0); print $rop;"
> NaN
> Inf
> Inf
> 
> Cheers,
> Rob