Trouble with doubles (adding, setting precision)

Question

Firstly,
I have a loop that adds 0.3 three times. For some reason it comes out as .8999999999999!
Why is this?
Second, does anyone know where I can download a class or view source code that allows you to set decimal precision of a double? I started writing one myself but it's more complex than I thought and I'd rather just use someone elses if I could...

Answer 1

The precision of calculations and how numerical
floating point numbers are handled in computers are is
very relevant for business situations. Actually the
problems with decimal numbers themselves often need to
be understood by those creating the business rules (ie
like the sales people, audit departements, CEOs, etc.)
The precision of the java double itself is unlikely to
ever have an impact on that though.
For example, if a business person asks you to
calculate a mortgage payement then they must
understand exactly the impact that the imprecision of
such decimal calculations will have on the business.
And they must understand it enough so that they can
decide the correct way for handling it. The rules
for handling this case, at least in the
mortgage/banking industry, is going to be far less
the possible precision that a java double can have.
(The reason of course being that how this is handled
d impacts the credits of the lender and the debits of
the lendee.)
Given the above it matters little what data types are
used to handle the calculation itself. But rather
that the calculation is done in such a way that the
business people understand it and that they understand
the limitations of it as well.Okay, I don't disagree with any of that. Still not sure about your overall point though.
* P-L said "don't add doubles in a loop because it compounds rounding errors."
* I said "but if the number of addtions multiplied by the maximum error is still within your error tolerance, it's okay." In other words, Java's double's precision is not necessarily going to be a problem.
* You said some things that seem to agree with that final point, but your "that is fine if your job consists of nothing but evaluating computers, but for the rest of us the errors in the data that we are using..." comment makes it sound like you're disagreeing with my point: "The errors in Java's double may not cause you problems. Know it's limitations, and how they relate to your requirements."
Are you looking for an argument, now that "one exit point" has fizzled? :-) Or am I just being particularly dense today?
¶