Rounded Numbers Dont Add Up

Question

I have a user who is complaining that the numbers on her report don't add up to the total that's printed at the bottom. I haven't seen a sample of this yet but I wanted to run this by you guys and see if this code makes sense.
  function() {
    NumberFormat nf = new DecimalFormat("#,##0.00");
    double total = 0D;
    for (...) {
      double d = Math.random();
      total += round(d);
      print(nf.format(d));
    print(nf.format(total));
  double round(double d) {
    return (double)(Math.round(d*100)/100D);
  }Is my round() sound?

Answer 1

I have a user who is complaining that the numbers on
her report don't add up to the total that's printed at
the bottom. This could be a problem with the user in that they don't understand numerics.
Or it could b a problem with you in that you do not understand the problem domain.
Floating point numbers are always inaccurate. It doesn't matter whether humans or computers do this. That is because floating point numbers (when used correctly) are an approximate measure of something which can not be measured exactly.
For example a pound of hamburger might be marked as weighing 1.01 pounds. The inaccuracy is in that it might actually weigh 1.01015678 pounds (but is unlikely to weigh even that.) There is no way to get the exact weight.
Now if that is your situation then it is problem with the user - the user doesn't understand numerics and the inherit inaccuracies.
And it would probably be a bad idea for you to "fix" it for one user, when other users are expecting correct approximations.
(One way to fix this is to 'fiddle' with the numbers that make up your sum. Adjust one so everything 'looks' right.)
On the other hand maybe it is your problem. For example money. Money is not a floating point number, because money is always exact. Even when caculating interest it is always exact. You will never end up with a partial penny in your savings account, no matter how the interest rate is computed.
And since money is accurate a floating point number is not necessarily the best way to represent it. Care must be used in doing so. As a previous poster suggested the BigDecimal class can be used.
Or if your numbers are small then simply convert the money to integers by multiplying by 100 and rounding and then compute everything using that. And simply reformat the integer values so they look like floating point numbers rather than integers when you print them out.