Skip to end of metadata
Go to start of metadata

The values of boxed primitives cannot be directly compared using the == and != operators because these operators compare object references rather than object values. Programmers can find this behavior surprising because autoboxing memoizes, or caches, the values of some primitive variables. Consequently, reference comparisons and value comparisons produce identical results for the subset of values that are memoized.

Autoboxing automatically wraps a value of a primitive type with the corresponding wrapper object. The Java Language Specification (JLS), §5.1.7, "Boxing Conversion" [JLS 2015], explains which primitive values are memoized during autoboxing:

If the value p being boxed is true, false, a byte, a char in the range \u0000 to \u007f, or an int or short number between -128 and 127, then let r1 and r2 be the results of any two boxing conversions of p. It is always the case that r1 == r2.

Primitive Type

Boxed Type

Fully Memoized

boolean, byte

Boolean, Byte

Yes

char, short, int

Char, Short, Int

No

Use of the == and != operators for comparing the values of fully memoized boxed primitive types is permitted.

Use of the == and != operators for comparing the values of boxed primitive types that are not fully memoized is permitted only when the range of values represented is guaranteed to be within the ranges specified by the JLS to be fully memoized.

Use of the == and != operators for comparing the values of boxed primitive types is not allowed in all other cases.

Note that Java Virtual Machine (JVM) implementations are allowed, but not required, to memoize additional values [JLS 2015]:

Less memory-limited implementations could, for example, cache all characters and shorts, as well as integers and longs in the range of −32K to +32K. (§5.1.7)

Code that depends on implementation-defined behavior is nonportable. It is permissible to depend on implementation-specific ranges of memoized values provided that all targeted implementations support these greater ranges.

Noncompliant Code Example

This noncompliant code example defines a Comparator with a compare() method [Bloch 2009]. The compare() method accepts two boxed primitives as arguments. The == operator is used to compare the two boxed primitives. In this context, however, it compares the references to the wrapper objects rather than comparing the values held in those objects.

import java.util.Comparator;
 
static Comparator<Integer> cmp = new Comparator<Integer>() {
  public int compare(Integer i, Integer j) {
    return i < j ? -1 : (i == j ? 0 : 1);
  } 
};

Note that primitive integers are also accepted by this declaration because they are autoboxed at the call site.

Compliant Solution

This compliant solution uses the comparison operators, <, >, <=, or >=, because these cause automatic unboxing of the primitive values. The == and != operators should not be used to compare boxed primitives.

import java.util.Comparator;
 
static Comparator<Integer> cmp = new Comparator<Integer>() { 
  public int compare(Integer i, Integer j) {
    return i < j ? -1 : (i > j ? 1 : 0) ;
  }
};

Noncompliant Code Example

This noncompliant code example uses the == operator in an attempt to compare the values of pairs of Integer objects. However, the == operator compares object references rather than object values.

public class Wrapper {
  public static void main(String[] args) {
    Integer i1 = 100;
    Integer i2 = 100;
    Integer i3 = 1000;
    Integer i4 = 1000;
    System.out.println(i1 == i2);
    System.out.println(i1 != i2);
    System.out.println(i3 == i4);
    System.out.println(i3 != i4);
  }
}

The Integer class is guaranteed to cache only integer values from -128 to 127, which can result in equivalent values outside this range comparing as unequal when tested using the equality operators. For example, a JVM that did not cache any other values when running this program would output

true
false
false
true

Compliant Solution

This compliant solution uses the equals() method instead of the == operator to compare the values of the objects. The program now prints true, false, true, false on all platforms, as expected.

public class Wrapper {
  public static void main(String[] args) {
    Integer i1 = 100;
    Integer i2 = 100;
    Integer i3 = 1000;
    Integer i4 = 1000;
    System.out.println(i1.equals(i2));
    System.out.println(!i1.equals(i2));
    System.out.println(i3.equals(i4));
    System.out.println(!i3.equals(i4));
  }
}

Noncompliant Code Example

Java Collections contain only objects; they cannot contain primitive types. Further, the type parameters of all Java generics must be object types rather than primitive types. That is, attempting to declare an ArrayList<int> (which, presumably, would contain values of type int) fails at compile time because type int is not an object type. The appropriate declaration would be ArrayList<Integer>, which makes use of the wrapper classes and autoboxing.

This noncompliant code example attempts to count the number of indices in arrays list1 and list2 that have equivalent values. Recall that class Integer is required to memoize only those integer values in the range −128 to 127; it might return a nonunique object for any value outside that range. Consequently, when comparing autoboxed integer values outside that range, the == operator might return false and the example could deceptively output 0.

public class Wrapper {
  public static void main(String[] args) {
    // Create an array list of integers, where each element 
    // is greater than 127
    ArrayList<Integer> list1 = new ArrayList<Integer>();
    for (int i = 0; i < 10; i++) {
      list1.add(i + 1000);
    }

    // Create another array list of integers, where each element
    // has the same value as the first list
    ArrayList<Integer> list2 = new ArrayList<Integer>();
    for (int i = 0; i < 10; i++) {
      list2.add(i + 1000);
    }

    // Count matching values
    int counter = 0;
    for (int i = 0; i < 10; i++) {
      if (list1.get(i) == list2.get(i)) {  // Uses '=='
        counter++;
      }
    }

    // Print the counter: 0 in this example
    System.out.println(counter);
  }

}

However, if the particular JVM running this code memoized integer values from −32,768 to 32,767, all of the int values in the example would have been autoboxed to the corresponding Integer objects, and the example code would have operated as expected. Using reference equality instead of object equality requires that all values encountered fall within the interval of values memoized by the JVM. The JLS lacks a specification of this interval; rather, it specifies a minimum range that must be memoized. Consequently, successful prediction of this program's behavior would require implementation-specific details of the JVM.

Compliant Solution

This compliant solution uses the equals() method to perform value comparisons of wrapped objects. It produces the correct output, 10.

public class Wrapper {
  public static void main(String[] args) {
    // Create an array list of integers
    ArrayList<Integer> list1 = new ArrayList<Integer>();

    for (int i = 0; i < 10; i++) {
      list1.add(i + 1000);
    }

    // Create another array list of integers, where each element
    // has the same value as the first one
    ArrayList<Integer> list2 = new ArrayList<Integer>();
    for (int i = 0; i < 10; i++) {
      list2.add(i + 1000);
    }
 
    // Count matching values
    int counter = 0;
    for (int i = 0; i < 10; i++) {
      if (list1.get(i).equals(list2.get(i))) {  // Uses 'equals()'
        counter++;
      }
    }
 
    // Print the counter: 10 in this example
    System.out.println(counter);
  }
}

Noncompliant Code Example (Boolean)

In this noncompliant code example, constructors for class Boolean return distinct newly instantiated objects. Using the reference equality operators in place of value comparisons will yield unexpected results.

public void exampleEqualOperator(){
  Boolean b1 = new Boolean("true");
  Boolean b2 = new Boolean("true");

  if (b1 == b2) {    // Never equal
    System.out.println("Never printed");
  }
}

Compliant Solution (Boolean)

Boolean.TRUEBoolean.FALSE, or the values of autoboxed true and false literals, may be compared using the reference equality operators because the Java language guarantees that the Boolean type is fully memoized. Consequently, these objects are guaranteed to be singletons.

public void exampleEqualOperator(){
  Boolean b1 = true;
  Boolean b2 = true; 
	
  if (b1 == b2) {   // Always equal
    System.out.println("Always printed");
  }
 
  b1 = Boolean.TRUE;
  if (b1 == b2) {   // Always equal
    System.out.println("Always printed");
  }
}

Risk Assessment

Using the equivalence operators to compare values of boxed primitives can lead to erroneous comparisons.

Rule

Severity

Likelihood

Remediation Cost

Priority

Level

EXP03-J

Low

Likely

Medium

P6

L2

Automated Detection

Detection of all uses of the reference equality operators on boxed primitive objects is straightforward. Determining the correctness of such uses is infeasible in the general case.

Tool
Version
Checker
Description
Coverity7.5

BAD_EQ
FB.EQ_ABSTRACT_SELF
FB.EQ_ALWAYS_FALSE
FB.EQ_ALWAYS_TRUE
FB.EQ_CHECK_FOR_OPERAND_NOT_ COMPATIBLE_WITH_THIS
FB.EQ_COMPARETO_USE_OBJECT_ EQUALS
FB.EQ_COMPARING_CLASS_NAMES
FB.EQ_DOESNT_OVERRIDE_EQUALS
FB.EQ_DONT_DEFINE_EQUALS_ FOR_ENUM
FB.EQ_GETCLASS_AND_CLASS_ CONSTANT
FB.EQ_OTHER_NO_OBJECT
FB.EQ_OTHER_USE_OBJECT
FB.EQ_OVERRIDING_EQUALS_ NOT_SYMMETRIC
FB.EQ_SELF_NO_OBJECT
FB.EQ_SELF_USE_OBJECT
FB.EQ_UNUSUAL
FB.ES_COMPARING_PARAMETER_ STRING_WITH_EQ
FB.ES_COMPARING_STRINGS_ WITH_EQ
FB.ES_COMPARING_PARAMETER_ STRING_WITH_EQ

Implemented
Parasoft Jtest
10.3
PB.CUB.UEIC
SonarQube
6.7
S1698"==" and "!=" should not be used when "equals" is overridden

Related Guidelines

MITRE CWE

CWE-595, Comparison of Object References Instead of Object Contents
CWE-597, Use of Wrong Operator in String Comparison

Bibliography

[Bloch 2009]

Puzzle 4, "Searching for the One"

[JLS 2015]

§5.1.7, "Boxing Conversion"

[Pugh 2009]

Using == to Compare Objects Rather than .equals

[Seacord 2015]



4 Comments

  1. However, if the particular JVM running this code memoized integer values from -32,768 to 32,767, all of the int values in the example would have been autoboxed to singleton Integer objects,

    What do you mean by "singleton"?
    I think it's better to rephrase it to some like:

    However, if the particular JVM running this code memoized integer values from -32,768 to 32,767, all of the int values in the example would have been autoboxed to the corresponding Integer objects,

    1. Agreed, I made this change.

    2. I think it is something like the multiton pattern.

      http://en.wikipedia.org/wiki/Multiton_pattern

      While in the wiki article it is realized in the Java language, here it may be realized in a different way, because its integral part of the JRE, but I think it is satisfying the same purpose.

      1. Yes, I guess you could say Java uses memoization using a 'multiton paradign'. But you are replying to a comment which is about text that was removed from this rule several months ago. One of the points of this rule is that the memoization is not part of Java's public API, so you shouldn't rely too heavily on it...its details can vary between JVMs, or even different versions of the same JVM.