2009/2/5 Brian Barker <[email protected]> > At 18:38 05/02/2009 -0600, Ely Schoenfeld wrote: > >> The precision error didn't happen when I added 2.34 plus -2.73, it only >> happened when adding 18282.34 plus -18282.73 so it must have to be related >> with the "size" of the operated numbers. >> > > The numbers will be stored as "floating-point" quantities, so they will > have a fixed number of significant figures (*not* fractional places). There > appear to be about fifteen significant decimal digits, which corresponds to > about fifty significant binary digits, so I'm guessing that the so-called > mantissa is stored in six bytes. And so yes: the larger the number, the > sooner along the fractional part any rounding error will appear. In fact, > you have to go up only to 4.34 and -4.73 to see the effect appear. > > Brian Barker > > You are right.
Thank you for your explanation. Ely. > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > >
