V125KJG8%UBVMS.BITNET@wiscvm.wisc.EDU (04/17/87)
>Think twice before advocating fuzzy equality -- it may solve some >problems but it causes others. For instance, given > > a == b > >and > > b == c > >shouldn't one be able to assume > > a == c > >? I use a microcomputer C and its floating point accumulators are kept in memory and are always 64 bits long for comparisons. As an individual at watmath.uucp pointed out to me, it's possible on a processor, say, the 68020/68881 combination, where the floating-point processor internally computes to 80 bits and it's moved back out to a 64-bit area. We now compare a 64-bit float to an 80-bit float, and if, for example, the IEEE INX (Inexact Result) flag is set, we ARE NOT going to be equal here! As long as we are doing 64-bit to 64-bit comparisions, and a companion INX is cleared, that is the only time we are guaranteed of equality. Once values go beyond the range of the mantissa, the INX flag will be set, and floating-point equality is lost. Just my few cents' worth before Easter... --Curtis R. Anderson State University of New York at Buffalo (V125KJG8@UBVMS.BITNET) (C125KJG8@UBVM.BITNET)