[Oberon] [EXT] Re: Wrong output of Texts.WriteReal and Texts.WriteRealFix in PO
hellwig.geisse at mni.thm.de
Sun Oct 4 08:15:56 CEST 2020
On Sa, 2020-10-03 at 23:30 +0000, Skulski, Wojciech wrote:
> I do not think it is right. Numerical results must be either correct or none. Returning a NaN or a
> trap is tolerable, though troublesome. Returning a wrong result cannot be accepted as valid.
I fully second this statement. And the definitive standard
how to do that correctly is IEEE 754. If you don't believe
me, read the failure stories in, e.g., Hennessy/Patterson's
"Computer Organization & Design".
> There are enough experts in the community to fix this problem. Perhaps Robert's Component Pascal
> numerical libraries can help?
I suspect the floating-point hardware itself isn't fit for
IEEE 754, although I didn't investigate that. If RISC5
is meant to be a serious tool for floating-point tasks,
this area must be cleared up, IMHO.
More information about the Oberon