Re: Sparse + complex arithmetic inconsistency

[Michael Goffioul]

On 3/30/07, Rafael Laboissiere <address@hidden> wrote:
> Does the gcc/g++ bug have something to do with the fact that the two
> following lines produce different results in Octave (but shouldn't):
>
>    octave2.9:1> ([1+i 1+i] ./ [0 i]) (1)
>    ans = NaN - NaNi
>    octave2.9:2> (1+i) / 0
>    warning: division by zero
>    ans = Inf + Infi
>
> It looks like a problem in Octave.

The first one will call

std::complex<double>::operator / (std::complex<double>, double)

the second will call

std::complex<double>::operator / (std::complex<double>, std::complex<double>)

The fact is that in current C++ implementations, these 2 methods
return different results when the second argument is 0 (0+0i in the
second case). IMO, it's more a implementation problem in standard
C++ libraries, than a problem in Octave.

Michael.