What, exactly, would be warned against? Addition of something that is too small to cause a change in the result would be easy to check for, it could be made part of the addition routine; if warnings are on, issue a warning when $d != 0 and ($c + $d) == $c this could be checked before the operation, by comparing the exponent of $d against (the exponent of $c minus the bits of accuracy in out representations) In the e example, the result would get pretty screwy before the warning kicks in; maybe a threshold could be added into the condition that causes the warning. "John Nicol (WEBTV)" wrote: > > Sounds like a useful warning, all right. > > -----Original Message----- > From: David L. Nicol [mailto:[EMAIL PROTECTED]] > Sent: Friday, March 16, 2001 2:56 PM > To: John Nicol (WEBTV) > Subject: A funny thing about e > > I wrote a geometricly progressing e generator > > perl -le '$n=1; print "$n \t",((1 + (1/$n))** $n) while $n*=1.001' > > and initially, it converges on e. Then it runs out of accuracy and > re converges on 1. There might be something interesting there. I don't > know if the behavior of overloaded double precision floating point is > interesting or not: it is certainly a popular form of numeric failure, > as people forget about its limits of usefulness. When to throw away > a result as meaningless is certainly an important piece of wisdom, > I do not know any programming languages that do it for you -- issue a > warning when you've overloaded your accuracy instead of merrily > returning > you your noise -- what do you think? > > -- > David Nicol 816.235.1187 [EMAIL PROTECTED] > If God had meant us to compute securely, He'd have given > us more prime numbers! -- Casey Schaufler -- David Nicol 816.235.1187 [EMAIL PROTECTED] If God had meant us to compute securely, He'd have given us more prime numbers! -- Casey Schaufler
