Re: bug in functions 'roundup' and 'ceil'

[Kjetil Torgrim Homme <kjetilho ifi uio no>]

[Morten Welinder]

>   I have attempted to "fix" this in current cvs sources.  The real
>   problem is lack of understanding of binary math, but I guess we
>   cannot expect to pass that message to end users.
>   
>   Therefore the "fix" is to be wrong on different numbers that most
>   users are not likely to stumble upon.

Isn't the right fix to run ceil(x) and ceil(x-epsilon), where epsilon
is the smallest possible difference to x (x/2**52 for doubles), and
return the value closest to x?


Kjetil T.