Java theory and practice: Hashing it out --hashCode

Summary:   Every Java object has a hashCode() and an equals() method. Many classes override the default implementations of these methods to provide a higher degree of semantic comparability between object instances. In this installment of Java theory and practice , Java developer Brian Goetz shows you the rules and guidelines you should follow when creating Java classes in order to define hashCode() and equals() effectively and appropriately.

 

While the Java language does not provide direct support for associative arrays -- arrays that can take any object as an index -- the presence of the hashCode() method in the root Object class clearly anticipates the ubiquitous use of HashMap (and its predecessor, Hashtable ). Under ideal conditions, hash-based containers offer both efficient insertion and efficient retrieval; supporting hashing directly in the object model facilitates the development and use of hash-based containers.

 

Defining equality

 

The Object class has two methods for making inferences about an object's identity: equals() and hashCode() . In general, if you override one of these methods, you must override both , as there are important relationships between them that must be maintained. In particular, if two objects are equal according to the equals() method, they must have the same hashCode() value (although the reverse is not generally true).

The semantics of equals() for a given class are left to the implementer; defining what equals() means for a given class is part of the design work for that class. The default implementation, provided by Object , is simply reference equality:

  public boolean equals(Object obj) { 
    return (this == obj); 
  }

 

Under this default implementation, two references are equal only if they refer to the exact same object . Similarly, the default implementation of hashCode() provided by Object is derived by mapping the memory address of the object to an integer value . Because on some architectures the address space is larger than the range of values for int , it is possible that two distinct objects could have the same hashCode() . If you override hashCode() , you can still use the System.identityHashCode() method to access this default value.

 

Overriding equals() -- a simple example

 

An identity-based implementation for equals() and hashCode() is a sensible default, but for some classes, it is desirable to relax the definition of equality somewhat. For example, the Integer class defines equals() similarly to this:

  public boolean equals(Object obj) {
    return (obj instanceof Integer 
            && intValue() == ((Integer) obj).intValue());
  }

 

Under this definition, two Integer objects are equal only if they contain the same integer value. This, along with Integer being immutable, makes it practical to use an Integer as a key in a HashMap . This value-based approach to equality is used by all the primitive wrapper classes in the Java class library, such as Integer , Float , Character , and Boolean , as well as String (two String objects are equal if they contain the same sequence of characters). Because these classes are immutable and implement hashCode() and equals() sensibly, they all make good hash keys.

 

Why override equals() and hashCode()?

 

What would happen if Integer did not override equals() and hashCode() ? Nothing, if we never used an Integer as a key in a HashMap or other hash-based collection . However, if we were to use such an Integer object for a key in a HashMap , we would not be able to reliably retrieve the associated value, unless we used the exact same Integer instance in the get() call as we did in the put() call. This would require ensuring that we only use a single instance of the Integer object corresponding to a particular integer value throughout our program. Needless to say, this approach would be inconvenient and error prone.

The interface contract for Object requires that if two objects are equal according to equals() , then they must have the same hashCode() value . Why does our root object class need hashCode() , when its discriminating ability is entirely subsumed by that of equals() ? The hashCode() method exists purely for efficiency . The Java platform architects anticipated the importance of hash-based collection classes -- such as Hashtable , HashMap , and HashSet -- in typical Java applications, and comparing against many objects with equals() can be computationally expensive. Having every Java object support hashCode() allows for efficient storage and retrieval using hash-based collections.

---

Back to top

 

Requirements for implementing equals() and hashCode()

 

There are some restrictions placed on the behavior of equals() and hashCode() , which are enumerated in the documentation for Object . In particular, the equals() method must exhibit the following properties:

• Symmetry[对称] : For two references, a and b , a.equals(b) if and only if b.equals(a)
• Reflexivity[自反性] : For all non-null references, a.equals(a)
• Transitivity[传递性] : If a.equals(b) and b.equals(c) , then a.equals(c)
• Consistency with hashCode() : Two equal objects must have the same hashCode() value

The specification for Object offers a vague[含糊的] guideline that equals() and hashCode() be consistent -- that their results will be the same for subsequent invocations, provided that "no information used in equals comparison on the object is modified." This sounds sort of like "the result of the calculation shouldn't change, unless it does." This vague statement is generally interpreted to mean that equality and hash value calculations should be a deterministic function of an object's state and nothing else.

---

Back to top

 

What should equality mean?

 

The requirements for equals() and hashCode() imposed by the Object class specification are fairly simple to follow. Deciding whether, and how, to override equals() requires a little more judgment. In the case of simple immutable value classes, such as Integer (and in fact for nearly all immutable classes), the choice is fairly obvious -- equality should be based on the equality of the underlying object state. In the case of Integer , the object's only state is the underlying integer value.

For mutable objects, the answer is not always so clear. Should equals() and hashCode() be based on the object's identity (like the default implementation) or the object's state (like Integer and String)? There's no easy answer -- it depends on the intended use of the class. For containers like List and Map , one could have made a reasonable argument either way. Most classes in the Java class library, including container classes, err on the side of providing an equals() and hashCode() implementation based on the object state.

If an object's hashCode() value can change based on its state, then we must be careful when using such objects as keys in hash-based collections to ensure that we don't allow their state to change when they are being used as hash keys . All hash-based collections assume that an object's hash value does not change while it is in use as a key in the collection. If a key's hash code were to change while it was in a collection, some unpredictable and confusing consequences could follow. This is usually not a problem in practice -- it is not common practice to use a mutable object like a List as a key in a HashMap .

An example of a simple mutable class that defines equals() and hashCode() based on its state is Point . Two Point objects are equal if they refer to the same (x, y) coordinates, and the hash value of a Point is derived from the IEEE 754-bit representation of the x and y coordinate values.

For more complex classes, the behavior of equals() and hashCode() may even be imposed by the specification of a superclass or interface. For example, the List interface requires that a List object is equal to another object if and only if the other object is also a List and they contain the same elements (defined by Object.equals() on the elements) in the same order. The requirements for hashCode() are defined with even more specificity -- the hashCode() value of a list must conform to the following calculation:

  hashCode = 1;
  Iterator i = list.iterator();
  while (i.hasNext()) {
      Object obj = i.next();
      hashCode = 31*hashCode + (obj==null ? 0 : obj.hashCode());
  }

 

Not only is the hash value dependent on the contents of the list, but the specific algorithm for combining the hash values of the individual elements is specified as well. (The String class specifies a similar algorithm to be used for computing the hash value of a String .)

---

Back to top

 

Writing your own equals() and hashCode() methods

 

Overriding the default equals() method is fairly easy, but overriding an already overridden equals() method can be extremely tricky to do without violating either the symmetry or transitivity requirement. When overriding equals() , you should always include some Javadoc comments on equals() to help those who might want to extend your class do so correctly.

As a simple example, consider the following class:

  class A {
    final B someNonNullField;
    C someOtherField;
    int someNonStateField;
  }

 

How would we write the equals() method for this class? This way is suitable for many situations:

  public boolean equals(Object other) {
    // Not strictly necessary, but often a good optimization
    if (this == other)
      return true;
    if (!(other instanceof A))
      return false;
    A otherA = (A) other;
    return 
      (someNonNullField.equals(otherA.someNonNullField))
        && ((someOtherField == null) 
            ? otherA.someOtherField == null 
            : someOtherField.equals(otherA.someOtherField)));
  }

 

Now that we've defined equals() , we have to define hashCode() in a compatible manner. One compatible, but not all that useful, way to define hashCode() is like this:

  public int hashCode() { return 0; }

 

This approach will yield horrible performance for HashMap s with a large number of entries, but it does conform to the specification. A more sensible implementation of hashCode() for A would be like this:

  public int hashCode() { 
    int hash = 1;
    hash = hash * 31 + someNonNullField.hashCode();
    hash = hash * 31 
                + (someOtherField == null ? 0 : someOtherField.hashCode());
    return hash;
  }

 

Note that both of these implementations delegate a portion of the computation to the equals() or hashCode() method of the state fields of the class. Depending on your class, you may also want to delegate part of the computation to the equals() or hashCode() function of the superclass. For primitive fields, there are helper functions in the associated wrapper classes that can help in creating hash values, such as Float.floatToIntBits .

Writing an equals() method is not without pitfalls. In general, it is impractical[不切实际的] to cleanly override equals() when extending an instantiable class that itself overrides equals() , and writing an equals() method that is intended to be overridden (such as in an abstract class) is done differently than writing an equals() method for a concrete class. See Effective Java Programming Language Guide , Item 7 (in Resources ) for some examples and more details about why this is so.

---

Back to top

 

Room for improvement?

 

Building hashing into the root object class of the Java class library was a very sensible design compromise -- it makes using hash-based containers so much easier and more efficient. However, several criticisms have been made of the approach to and implementation of hashing and equality in the Java class library. The hash-based containers in java.util are very convenient and easy to use, but may not be suitable for applications that require very high performance. While most of these will never be changed, it is worthwhile to keep in mind when you're designing applications that rely heavily on the efficiency of hash-based containers. These criticisms include:

• Too small a hash range. Using int , instead of long , for the return type of hashCode() increases the possibility of hash collisions.
  
  
• Bad distribution of hash values. The hash values for short strings and small integers are themselves small integers, and are close to the hash values of other "nearby" objects. A more well-behaved hash function would distribute the hash values more evenly across the hash range.
  
  
• No defined hashing operations. While some classes, such as String and List , define a hash algorithm to be used in combining the hash values of its constituent elements into a single hash value, the language specification does not define any approved means of combining the hash values of multiple objects into a new hash value. The trick used by List , String , or the example class A discussed earlier in Writing your own equals() and hashCode() methods are simple, but far from mathematically ideal. Nor does the class library offer convenience implementations of any hashing algorithm that would simplify the creation of more sophisticated hashCode() implementations.
  
  
• Difficulty writing equals() when extending an instantiable class that already overrides equals() . The "obvious" ways to define equals() when extending an instantiable class that already overrides equals() all fail to meet the symmetry or transitivity requirements of the equals() method. This means that you must understand the structure and implementation details of classes you are extending when overriding equals() , and may even need to expose private fields in the base class as protected to do so, which violates principles of good object-oriented design.

---

Back to top

 

Summary

 

By defining equals() and hashCode() consistently, you can improve the usability of your classes as keys in hash-based collections. There are two approaches to defining equality and hash value: identity-based, which is the default provided by Object , and state-based, which requires overriding both equals() and hashCode() . If an object's hash value can change when its state changes, be sure you don't allow its state to change while it is being used as a hash key.