Skip to end of metadata
Go to start of metadata

Method and constructor overloading allows declaration of methods or constructors with the same name but with different parameter lists. The compiler inspects each call to an overloaded method or constructor and uses the declared types of the method parameters to decide which method to invoke. In some cases, however, confusion may arise because of the presence of relatively new language features such as autoboxing and generics.

Furthermore, methods or constructors with the same parameter types that differ only in their declaration order are typically not flagged by Java compilers. Errors can result when a developer fails to consult the documentation at each use of a method or constructor. A related pitfall is to associate different semantics with each of the overloaded methods or constructors. Defining different semantics sometimes necessitates different orderings of the same method parameters, creating a vicious circle. Consider, for example, an overloaded getDistance() method in which one overloaded method returns the distance traveled from the source while another (with reordered parameters) returns the remaining distance to the destination. Implementers may fail to realize the difference unless they consult the documentation at each use.

Noncompliant Code Example (Constructor)

Constructors cannot be overridden and can only be overloaded. This noncompliant code example shows the class Con with three overloaded constructors:

class Con {
  public Con(int i, String s) { 
    // Initialization Sequence #1
  }
  public Con(String s, int i) {
    // Initialization Sequence #2 
  } 
  public Con(Integer i, String s) {
    // Initialization Sequence #3 
  } 
}

Failure to exercise caution while passing arguments to these constructors can create confusion because calls to these constructors contain the same number of similarly typed actual parameters. Overloading must also be avoided when the overloaded constructors or methods provide distinct semantics for formal parameters of the same types, differing solely in their declaration order.

Compliant Solution (Constructor)

This compliant solution avoids overloading by declaring public static factory methods having distinct names in place of the public class constructors:

public static Con createCon1(int i, String s) { 
  /* Initialization Sequence #1 */
}
public static Con createCon2(String s, int i) { 
  /* Initialization Sequence #2 */
}
public static Con createCon3(Integer i, String s) {
  /* Initialization Sequence #3 */
}

Noncompliant Code Example (Method)

In this noncompliant code example, the OverLoader class holds a HashMap instance and has overloaded getData() methods. One getData() method chooses the record to return on the basis of its key value in the map; the other chooses on the basis of the actual mapped value. 

class OverLoader extends HashMap<Integer,Integer> {
  HashMap<Integer,Integer> hm;
  public OverLoader() {
    hm = new HashMap<Integer, Integer>();
    // SSN records
    hm.put(1, 111990000);
    hm.put(2, 222990000);
    hm.put(3, 333990000);	  
  }
  public String getData(Integer i) { // Overloading sequence #1
    String s = get(i).toString(); // Get a particular record
    return (s.substring(0, 3) + "-" + s.substring(3, 5) + "-" + 
            s.substring(5, 9));
	  
  }
  public Integer getData(int i) { // Overloading sequence #2
    return hm.get(i); // Get record at position 'i'
  }
  // Checks whether the ssn exists
  @Override public Integer get(Object data) {
    // SecurityManagerCheck()

    for (Map.Entry<Integer, Integer> entry : hm.entrySet()) {
      if(entry.getValue().equals(data)) {
        return entry.getValue();  // Exists 
      }
    }
    return null;
  }
  public static void main(String[] args) {
    OverLoader bo = new OverLoader();
    // Get record at index '3'
    System.out.println(bo.getData(3));
    //Get record containing data '111990000'
    System.out.println(bo.getData((Integer)111990000));
  }
}

For purposes of overload resolution, the signatures of the getData() methods differ only in the static type of their formal parameters. The OverLoader class inherits from java.util.HashMap and overrides its get() method to provide the checking functionality. This implementation can be extremely confusing to the client who expects both getData() methods to behave in a similar fashion and not depend on whether an index of the record or the value to be retrieved is specified.

Although the client programmer might eventually deduce such behavior, other cases, such as with the List interface, may go unnoticed, as Joshua Bloch [Bloch 2008] describes:

The List<E> interface has two overloadings of the remove method: remove(E) and remove(int). Prior to release 1.5 when it was "generified," the List interface had a remove(Object) method in place of remove(E), and the corresponding parameter types, Object and int, were radically different. But in the presence of generics and autoboxing, the two parameter types are no longer radically different.

Consequently, a client programmer may fail to realize that the wrong element has been removed from the list.

A further problem is that in the presence of autoboxing, adding a new overloaded method definition can break previously working client code. This can happen when a new overloaded method with a more specific type is added to an API whose methods used less specific types in earlier versions. For example, if an earlier version of the OverLoader class provided only the getData(Integer) method, the client could correctly invoke this method by passing a parameter of type int; the result would be selected on the basis of its value because the int parameter would be autoboxed to Integer. Subsequently, when the getData(int) method is added, the compiler resolves all calls whose parameter is of type int to invoke the new getData(int) method, thereby changing their semantics and potentially breaking previously correct code. The compiler is entirely correct in such cases; the actual problem is an incompatible change to the API.

Compliant Solution (Method)

Naming the two related methods differently eliminates both the overloading and the confusion.

public Integer getDataByIndex(int i) { 
  // No longer overloaded
}

public String getDataByValue(Integer i) {
  // No longer overloaded 
}

Applicability

Ambiguous or confusing uses of overloading can lead to unexpected results.

Bibliography

[API 2013]Interface Collection<E>
[Bloch 2008]Item 41, "Use Overloading Judiciously"

 


6 Comments

  1. I tried to make this rule somewhat more normative, but I'm not entirely convinced this should be a requirement for conformance because a violation of this guidelines doesn't necessarily indicate the presence of an error.

    1. In the examples, because of autoboxing, the incorrect constructor/method may be called. Isn't that an acceptable violation? The NCEs can be made more comple but were kept simple to demonstrate the point clearly. The List interface quote describes the problem that happened because of violating this guideline.

      1. For starters, that would be a violation of EXP10-J. Ensure that autoboxed values have the intended type

        Even that rule is mostly non-normative because it includes the word "intended".

        Currently, the following code cannot be present in a conforming system:

        class Con {
          public Con(int i, String s) { /* Initialization Sequence #1 */ }
          public Con(String s, int i) { /* Initialization Sequence #2 */ } 
        }
        

        Although clearly this code could be used without error.

  2. Why does the 2nd NCCE both inherit from HashMap and contain its own HashMap?

  3. Class name "Con" cannot be used on Windows platform because it's a reserved device name.

    Writing to Con.class means writing to console device. (I suppose there's a similar problem for "PRN", "AUX", ...)

    It may confuse some readers using Windows.

    how about changing to "cCon" or such ?

     

    1. I changed the class name to Ctor (a common abbrev for constructor in C++).