Apache Commons HashCodeBuilder: null vs zero for Numeric types

Question

Recently I have encountered the following hashcode "equality" scenario in a Java codebase using Apache Commons Lang 3, and was surprised that I could not find much information on how to handle what seems like it could be a common problem:

MyObject one = new MyObject();
one.setFoo("foo");
one.setBar(null);

MyObject two = new MyObject();
two.setFoo("foo");
two.setBar((short) 0);

int oneHash = HashCodeBuilder.reflectionHashCode(one);
int twoHash = HashCodeBuilder.reflectionHashCode(two);

System.out.println("oneHash: " + oneHash);
System.out.println("twoHash: " + twoHash);
System.out.println("Bar equality: " + Objects.equals(one.getBar(), two.getBar()));

The preceding code produces the following output, which shows that both objects have the same hashcode, despite being un-equal:

oneHash: 3781511
twoHash: 3781511
Bar equality: false

MyObject definition:

public class MyObject {
    private String foo;
    private Short bar;

    public String getFoo() {
        return foo;
    }

    public void setFoo(String foo) {
        this.foo = foo;
    }

    public Short getBar() {
        return bar;
    }

    public void setBar(Short bar) {
        this.bar = bar;
    }
}

While I could maybe understand a null Numeric and 0 Numeric having the same hash in a purely mathematical sense, in any practical setting this causes non-equal objects to have the same hashcode, which can lead to fairly major collision problems.

Clarification/Complication: While I would love to be able to just call equals() or hashcode() on the object, the codebase I am working with is unfortunately comparing two Objects, which means I have no insight to whether equals() or hashcode() is actually defined for any given input, and I am not able to edit the class definitions to add these methods in cases where it is missing. This is likely why the original author of this code chose to use reflectionHashCode(). With this in mind, is there a programmatic / code-based solution or workaround to this problem, such as an alternative library that wouldn't require equals() or hashcode() to be defined on the objects being compared?

So you are working with code that uses a function from an open source library that does not fit your needs (HashCodeBuilder.reflectionHashCode) and you cannot change the code to make it use another function. Is that correct description of your situation? — Joni, Apr 18 '20 at 00:13
The opposite: I can change the code doing the hashing and comparisons, but I cannot change the underlying objects to add `equals()` and `hashcode()`. — Matsu Q., Apr 18 '20 at 21:12
Does that mean you can make the code that computes the hash to treat `null` as a number other than 0? — Joni, Apr 19 '20 at 15:46
Yes, I could, in theory, reimplement `reflectionHashCode()` entirely on my own, but that doesn't seem like a practical solution. — Matsu Q., Apr 20 '20 at 15:46

score 1 · Answer 1 · answered Apr 17 '20 at 22:14

There is always a possibility of hash codes collisions even for objects with totally different values. After all you are mapping an infinite number of all possible object values to a 32 bit integer. This still works because data structures that take advantage of has codes like sets and maps additionally use .equals to check for object equality.

Joni · Answer 2 · 2020-04-18T21:37:27.253

0

The code-based solution is to implement a hash function in a way that distinguishes between null and 0. There are lots of ways to do it, here is one:

// this could be called hashCode, but you don't want to override hashCode
public int yourCustomHashFunction() {
    if (bar == null) {
        return Objects.hashCode(foo, 1234567);
    } else {
        return Objects.hashCode(foo, bar);
    }
}

Since bar is a Short, a value outside the valid range for short like 1234567 is unlikely cause collisions with valid short values.

edited Apr 18 '20 at 21:37

answered Apr 17 '20 at 22:15

Joni

108,737
14
143
193

The first thing you need to do is stop using `HashCodeBuilder.reflectionHashCode()` since that method does not give you sufficient control. You can use the other methods in HashCodeBuilder if for some reason you don't want to use the standard library Objects class – Joni Apr 17 '20 at 22:30

Apache Commons HashCodeBuilder: null vs zero for Numeric types

2 Answers2