From: James Y K. <fo...@fu...> - 2005-11-14 18:41:01
|
On Nov 14, 2005, at 1:09 PM, Raymond Toy wrote: > Seems to me that if you're playing with float infinities, you'd have > overflow disabled, and, most-likely, divide-by-zero too. Not necessarily -- it's perfectly natural to have traps enabled, yet still want to introduce an explicit infinity into your data. The trapping rules allow this to occur sensibly -- most manipulation of the introduced infinity doesn't cause additional traps. A representation for infinity that requires fp traps to be off is not workable. (I don't think anyone has proposed one of those yet). James |