Trouble with double

Question

Hi all,
I tried to compile and run this code :
public class TestDouble {
     public static void main(String ar[]) {
          //The total balance
          double balance =19620.9d;
        balance -= 3542.9d;
        //Expecting the balance to be 16078
          System.out.println("Remaining balance : "+balance);
          balance -= 6144.0d;
          //Expecting the balance to be 9934
          System.out.println("Remaining balance : "+balance);
          balance -= 1200.0d;
          //Expecting the balance to be 8734
          System.out.println("Remaining balance : "+balance);
          balance -= 1200.0d;
          //Expecting the balance to be 7534
          System.out.println("Remaining balance : "+balance);
          balance -= 5134.0d;
          //Expecting the balance to be 2400
          System.out.println("Remaining balance : "+balance);
          balance -= 1200.0d;
          //Expecting the balance to be 1200
          System.out.println("Remaining balance : "+balance);
          balance -= 1200.0d;
          //Expecting the balance to be 0
          System.out.println("Remaining balance : "+balance);
          System.out.println("Remaining balance > 0 : "+(balance>0));
}I expect the final balance to be printed as zero.
However this is the out put that I am getting
G:\Ws\Test>javac -version
javac 1.6.0_14
G:\Ws\Test>javac G:\Ws\Test\TestDouble.java
G:\Ws\Test>java -version
java version "1.6.0_15"
Java(TM) SE Runtime Environment (build 1.6.0_15-b03)
Java HotSpot(TM) Client VM (build 14.1-b02, mixed mode, sharing)
G:\Ws\Test>java TestDouble
Remaining balance : 16078.000000000002
Remaining balance : 9934.000000000002
Remaining balance : 8734.000000000002
Remaining balance : 7534.000000000002
Remaining balance : 2400.000000000002
Remaining balance : 1200.0000000000018
Remaining balance : 1.8189894035458565E-12
Remaining balance > 0 : trueCan someone help me in understanding what's wrong.
Thanks,
Anil

Answer 1

For accounting, the thought of using rounding is not acceptable. Accounting typical only requires add, substract and multiplied by a quantity. An alternative to BigDecimal is to use long values for the smallest unit. e.g. use cents instead of dollars. The risk of error is important and BigDecimal is likely to be the best choice for this reason.
However in finance or banking, rounding is better understood as it is used, often unavoidable. You will find double is typically used for simplicity (often speed is also given as justification) While double is dramatically faster than BigDecimal, this speed improvement is usually not required IMHO.
Only large amounts of money are likely to have a significant rounding error, however large amounts of money are typically estimates. The problem with double is its limited precision, however money is a limit resource and very rare do you need more than 12 digits of accuracy. (double has 15-16 digits of accuracy) Even the US national debt can be represented to the cent using double, yet this is just an estimate. If you attempted to add up all the amounts involved in the US national debt, it would change while you are doing it.
I worked at a bank where in one dept the unexplained PnL was measuring in "millions of pounds per month" (PnL being Profit and Loss) I can assure you if you has mentioned an error 1.8e-12 dollars it would be given a proportionate amount of consideration, if not amusement.
In summary: There is no realistic, hard amount of money that would get an error using double which could not be handled by appropriate rounding. Any estimate of money is not accurate enough to require more precision than double.
Edited by: Peter__Lawrey on 13-Dec-2009 14:38