Re: Edge case conversions, exceptions to IEEE FPA
Michel Hack schrieb:
On (Mon, 17 Nov 2008 13:31:27 +0100) Sylvain Pion wrote:
Further thinking about this, I am now of the opinion that
intval(1e309) = Empty is not safe enough.
Since intval(1e-1) returns an interval not containing the result,
without any warning, there cannot be any complete safety in the
use of intval.
Users of IA must learn to avoid these construct anyway;
there is no need to give the unlikely use of intval(1e309)
special attention in the standard when the likely use of
intval(1e-1) escapes for free.
Indeed, Empty is NOT propagated by convexHull, and probably not by max
and min either (as these are supposedly defined in terms of IEEE 754-2008
maxNum and minNum, which ignore NaNs), so that max(xx, EMpty) would be xx.
In my proposal, max and min are treated as arithmetic operations,
hence return Empty when an argument is Empty.
Hull is different since it is not an arithmetic operation.
Arnold Neumaier