Thread Links | Date Links | ||||
---|---|---|---|---|---|
Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
On 6/14/2013 7:58 AM, Ned Nedialkov
wrote:
Actually, I do not find this convincing. If you have a program VNODE-LP, that is supposed to work on various computers and with various underlying pieces of software e.g. FILIB++ and PROFIL, and VNODE-LP produces different answers (or the same answers but in different formats), to the extent that it matters to you, then VNODE-LP was not properly "ported" to these different environments. In fact, I would expect that VNODE-LP could extract the inf and sup values from intervals computed by FILIB++ and PROFIL, and present them in a uniform manner, if the issue were raised to the appropriate level of concern to the program author. One could similarly object to things not being exactly the same if one computer were IEEE754 double compliant and the other used IEEE754-double-extended, and gave more digits. I think it would be unreasonable to insist that the interval library be changed so the results were printed the same, although I suppose one could argue, just as one could argue (as essentially has been done) that the interval package should decimal arithmetic and hence believes that 0.1 = 1/10, even on a binary-float machine. RJF . |