Floating point errors?
John W. Baxter
jwbaxter at olympus.net
Fri May 11 11:33:48 EDT 2001
In article <mailman.989545217.22380.python-list at python.org>, Tim Peters
<tim.one at home.com> wrote:
> Harrumph. Of course they're exact! Selling them as some sort of fuzzy
> mystery is what keeps people confused about them 50 years after they should
> have learned better.
Yes, but they only exactly represent a few of the many possible values.
And our friend 3.0/10.0 isn't one of them, when the floating point is
in binary. 25.0 / 100.0 "happens" to be one which is exactly
represented. Of course it doesn't "happen"...one quarter is a nice
fraction in binary.
Murphy's law suggests that the particular value which needs to be
exactly represented to keep the boiler from exploding can't be. ;-)
--John
More information about the Python-list
mailing list