This code does what I want but I don't why the numbers are so weird. Is there anyway to count:

-0
-0.01
-0.02
-0.03
-0.04
-0.05
-0.06
-0.07

instead of

-0
-0.01
-0.02
-0.03
-0.04
-0.04999997
-0.05999995
-0.06999999
-0.07999999


Quote:

...which is not to say you're out of luck, though.  Depending on your needs, you could either use integers (just interpret them as being shifted by two decimal places - this requires care when doing multiplications or divisions) or use Java's BigDecimal class (see http://java.sun.com/j2se/1.4.2/docs/api/java/math/BigDecimal.html for the details on that).

Neither method is ideal, as you can no longer just write out your math the natural way, but this is an inherent limitation of Java, caused by the fact that Sun thinks Java programmers en masse are too stupid and foolish to handle such a subtle language feature as operator overloading.  Then again, they might be right (see the many abuses of operator overloading in C++ if you doubt this), so I shouldn't complain too much...

Here's a simple example, to count down by .1:

Code:

import java.math.BigDecimal;

BigDecimal bd = new BigDecimal(0.0);
// To get .1 exactly, pass it as a string to the BigDecimal constructor
BigDecimal oneTenth = new BigDecimal(".1");

for (int i=0; i<100; ++i) {
  bd = bd.subtract(oneTenth);
  println(bd);
}

// Divide result by 40 - we have to set the number of decimal places to 2 to
// allow for exact computation, which is the second argument to divide()
println(bd.divide(new BigDecimal("40"),2,BigDecimal.ROUND_UNNECESSARY));
 

To do more, I'd suggest you read the BigDecimal docs - there are a lot of subtleties when it comes to setting scale properly under division/multiplication, and handling rounding.  But this class is very useful, especially if you ever do any sort of financial programming, where floating point error means actual missing money.  The following snippet will indicate that even though floating point calculations are useful, for real world amounts of money (to a company or rich person, at least), they are totally insufficient:

Code:

float number = 0.0;

while( number != (++number) ){
//not as infinite a loop as you might think...
}

println("Floating point precision is less than 1 at "+number);
 

triscuit wrote on May 4^th, 2008, 7:45am:


No, you can't use == even if they're both BigDecimals, in Java "==" on objects just compares to see if they're the same absolute object, not that the objects have the same value.

You can get a float value out with myBigDecimal.floatValue();