how does java.math.RoundingMode work?

Question

I'm having trouble with rounding. Specifically, after reading all the javadoc, I was expecting the following code:

int n = (integer between 0 and 9, included)
new BigDecimal(n + 0.555d).setScale(2, RoundingMode.HALF_UP).doubleValue()

to return n + 0.56. Instead, these are the return values for n from 0 to 4:

 new BigDecimal(0.555d).setScale(2, RoundingMode.HALF_UP).doubleValue()
 0.56
 new BigDecimal(1.555d).setScale(2, RoundingMode.HALF_UP).doubleValue()
 1.55
 new BigDecimal(2.555d).setScale(2, RoundingMode.HALF_UP).doubleValue()
 2.56
 new BigDecimal(3.555d).setScale(2, RoundingMode.HALF_UP).doubleValue()
 3.56
 new BigDecimal(4.555d).setScale(2, RoundingMode.HALF_UP).doubleValue()
 4.55

I have also tried to change the rounding mode:

int n = (integer between 0 and 9, included)
new BigDecimal(n + 0.555d).setScale(2, RoundingMode.HALF_DOWN).doubleValue()

expecting n + 0.55 as a result for each and every n. Instead, the return values are exactly the same as the previous example:

 new BigDecimal(0.555d).setScale(2, RoundingMode.HALF_DOWN).doubleValue()
 0.56
 new BigDecimal(1.555d).setScale(2, RoundingMode.HALF_DOWN).doubleValue()
 1.55
 new BigDecimal(2.555d).setScale(2, RoundingMode.HALF_DOWN).doubleValue()
 2.56
 new BigDecimal(3.555d).setScale(2, RoundingMode.HALF_DOWN).doubleValue()
 3.56
 new BigDecimal(4.555d).setScale(2, RoundingMode.HALF_DOWN).doubleValue()
 4.55

Am I missing something?

Answer 1

@Peter Lawrey I looked at your examples and wrote a quick program comparing the simple approach that you posted along with all the methods for RoundingMode. The code is here for anyone interested, this will clearly show the differences:

[RoundingMode.java] https://gitlab.com/bobby.estey/java/-/blob/master/maven/jdk14/src/main/java/mathematics/RoundingModeExamples.java

Results:
lowDouble:                             1.55553
simple - lowDouble:                    1.5555
RoundingMode.UP - lowDouble:           1.5556
RoundingMode.DOWN - lowDouble:         1.5555
RoundingMode.CEILING - lowDouble:      1.5556
RoundingMode.FLOOR - lowDouble:        1.5555
RoundingMode.HALF_UP - lowDouble:      1.5555
RoundingMode.HALF_DOWN - lowDouble:    1.5555
RoundingMode.HALF_EVEN - lowDouble:    1.5555
highDouble:                            1.55555
simple - highDouble:                   1.5556
RoundingMode.UP - highDouble:          1.5556
RoundingMode.DOWN - highDouble:        1.5555
RoundingMode.CEILING - highDouble:     1.5556
RoundingMode.FLOOR - highDouble:       1.5555
RoundingMode.HALF_UP - highDouble:     1.5555
RoundingMode.HALF_DOWN - highDouble:   1.5555
RoundingMode.HALF_EVEN - highDouble:   1.5555

Answer 2

The problem you have is that double is not a precise representation and you are round based on this imprecise number.

BigDecimal bd = new BigDecimal(1.555d);
System.out.println("bd=" + bd);
bd = bd.setScale(2, RoundingMode.HALF_UP);
System.out.println("after rounding bd=" + bd);
double d = bd.doubleValue();
System.out.println("after rounding d=" + d);

prints

bd=1.5549999999999999378275106209912337362766265869140625
after rounding bd=1.55
after rounding d=1.55

however

BigDecimal bd = BigDecimal.valueOf(1.555d);
System.out.println("bd=" + bd);
bd = bd.setScale(2, RoundingMode.HALF_UP);
System.out.println("after rounding bd=" + bd);
double d = bd.doubleValue();
System.out.println("after rounding d=" + d);

prints

bd=1.555
after rounding bd=1.56
after rounding d=1.56

This works because BigDecimal.valueOf does some extra rounding based on how double would appear if you printed it.

---

However I wouldn't use BigDecimal unless performance/simplicity is not an issue.

double d = 1.555d;
System.out.println("d=" + d);
d = roundToTwoPlaces(d);
System.out.println("after rounding d=" + d);

public static double roundToTwoPlaces(double d) {
    return ((long) (d < 0 ? d * 100 - 0.5 : d * 100 + 0.5)) / 100.0;
}

prints

d=1.555
after rounding d=1.56

For more details Double your money again compares the performance of different ways of rounding.