How to implement the equals/hashCode methods for classes that contain double fields

Question

I am overriding equals and hashCode in a class that includes double fields. My first approach was to use the epsilon test in the equals method, and Double.hashCode( double ) in hashCode, but that can result in equal objects having different hash codes; here is a simplified example:

public class DoubleHashTest2
{
    public static void main(String[] args)
    {
        double  base1   = .9;
        double  base2   = .7;
        Test    test1   = new Test( base1 - .1 );
        Test    test2   = new Test( base2 + .1 );

        System.out.println( test1.equals( test2 ) );
        System.out.println( test1.hashCode() );
        System.out.println( test2.hashCode() );
    }

    private static class Test
    {
        private double  dnum1;

        public Test( double dnum1 )
        {
            this.dnum1 = dnum1;
        }

        public boolean equals( Test other )
        {
            final double    epsilon = .0001;
            boolean         result  = false;
            if ( this == other )
                result = true;
            else if ( other == null )
                result = false;
            else
                result  = Math.abs( this.dnum1 - other.dnum1 ) < epsilon;
            return result;
        }

        public int hashCode()
        {
            int hash    = Double.hashCode( dnum1 );
            return hash;
        }
    }
}

I've thought of several solutions, including converting to BigDecimal, but I'm not really happy with any of them. I finally settled on rounding:

public boolean equals( Test other )
{
    boolean         result  = false;
    if ( this == other )
        result = true;
    else if ( other == null )
        result = false;
    else
    {
        double  test1   = round( dnum1 );
        double  test2   = round( other.dnum1 );
        result  = test1 == test2;
    }
    return result;
}

public int hashCode()
{
    double  temp    = round( dnum1 );
    int hash    = Double.hashCode( temp );
    return hash;
}

private double round( double dnum )
{
    // tests for NaN and +/-infinity omitted for brevity
    final int       places      = 4;
    final double    round_const = Math.pow( 10, places );
    double result   = ((int)(dnum * round_const + .5)) / round_const;
    return result;
}

But choosing a good rounding algorithm is difficult, and this seems kind of expensive. I looked at similar classes, such as Point2D.Double, but equals in this class fails, for example, when comparing .8 and 0.7999999999999999.

Is there a recommended way for dealing with this issue?

Answer 1

Answering the main question

You don't need any custom rounding, as Double class has doubleToLongBits() method, which simply converts double to long (both of them are 64-bit values).

Also, for your equals() method, you can compare two double values with Double#compare().

Possible equals() and hashCode() for your example:

public boolean equals(Object other) {
    if (this == other) {
        return true;
    }
    if (null == other
            || this.getClass() != other.getClass()) {
        return false;
    }

    return Double.compare(this.dnum1, ((Test) other).dnum1) == 0;
}

public int hashCode() {
    long bits = Double.doubleToLongBits(this.dnum1);
    return (int) (bits ^ (bits >>> 32));
}

About floating point arithmetic

Your example shows the disadvantage of double usage for floating point calculations - even values with the same magnitude can give close, but different results. Maybe you should use BigDecimal?

Also, see this question answers.