The conditional operator ?: uses the boolean value of its first operand to decide which of the other two expressions will be evaluated. (See §15.25, "Conditional Operator ? :" of the Java Language Specification (JLS) [JLS 2013].)

The general form of a Java conditional expression is operand1 ? operand2 : operand3.

  • If the value of the first operand (operand1) is true, then the second operand expression (operand2) is chosen.
  • If the value of the first operand is false, then the third operand expression (operand3) is chosen.

The conditional operator is syntactically right-associative. For example, a?b:c?d:e?f:g is equivalent to a?b:(c?d:(e?f:g)).

The JLS rules for determining the result type of a conditional expression (see following table) are complicated; programmers could be surprised by the type conversions required for expressions they have written.

Result type determination begins from the top of the table; the compiler applies the first matching rule. The Operand 2 and Operand 3 columns refer to operand2 and operand3 (from the previous definition) respectively. In the table, constant int refers to constant expressions of type int (such as '0' or variables declared final).

For the final table row, S1 and S2 are the types of the second and third operands respectively. T1 is the type that results from applying boxing conversion to S1, and T2 is the type that results from applying boxing conversion to S2. The type of the conditional expression is the result of applying capture conversion to S2. The type of the conditional expression is the result of applying capture conversion to the least upper bound of T1 and T2. See §5.1.7, "Boxing Conversion," §5.1.10, "Capture Conversion," and §15.12.2.7, "Inferring Type Arguments Based on Actual Arguments," of the JLS for additional information [JLS 2013].

Determining the Result Type of a Conditional Expression

Rule

Operand 2

Operand 3

Resultant Type

1

Type T

Type T

Type T

2

boolean

Boolean

boolean

3

Boolean

boolean

boolean

4

null

reference

reference

5

reference

null

reference

6

byte or Byte

short or Short

short

7

short or Short

byte or Byte

short

8

byte, short, char, Byte, Short, Character

constant int

byte, short, char if value of int is representable

9

constant int

byte, short, char, Byte, Short, Character

byte, short, char if value of int is representable

10

Other numeric

Other numeric

Promoted type of the second and third operands

11

T1 = boxing conversion(S1)

T2 = boxing conversion(S2)

Apply capture conversion to lub(T1,T2)

The complexity of the rules that determine the result type of a conditional expression can lead to unintended type conversions. Consequently, the second and third operands of each conditional expression should have identical types. This recommendation also applies to boxed primitives.

Noncompliant Code Example

In this noncompliant code example, the programmer expects that both print statements will print the value of alpha as a char:

public class Expr {
  public static void main(String[] args) {
    char alpha = 'A';
    int i = 0;
    // Other code. Value of i may change 
    boolean trueExp = true; // Some expression that evaluates to true
    System.out.print(trueExp ? alpha : 0); // prints A
    System.out.print(trueExp ? alpha : i); // prints 65
  }
}

The first print statement prints A because the compiler applies rule 8 from the result type determination table to determine that the second and third operands of the conditional expression are, or are converted to, type char. However, the second print statement prints 65—the value of alpha as an int. The first matching rule from the table is rule 10. Consequently, the compiler promotes the value of alpha to type int.

Compliant Solution

This compliant solution uses identical types for the second and third operands of each conditional expression; the explicit casts specify the type expected by the programmer:

public class Expr {
  public static void main(String[] args) {
    char alpha = 'A';
    int i = 0;
    boolean trueExp = true; // Expression that evaluates to true
    System.out.print(trueExp ? alpha : ((char) 0)); // Prints A
    // Deliberate narrowing cast of i; possible truncation OK
    System.out.print(trueExp ? alpha : ((char) i)); // Prints A
  }
}

When the value of i in the second conditional expression falls outside the range that can be represented as a char, the explicit cast will truncate its value. This usage complies with exception NUM12-J-EX0 of NUM12-J. Ensure conversions of numeric types to narrower types do not result in lost or misinterpreted data.

Noncompliant Code Example

This noncompliant code example prints 100 as the size of the HashSet rather than the expected result (some value between 0 and 50):

public class ShortSet {
  public static void main(String[] args) {
    HashSet<Short> s = new HashSet<Short>();
    for (short i = 0; i < 100; i++) {
      s.add(i);
      // Cast of i-1 is safe because value is always representable
      Short workingVal = (short) (i-1);
      // ... Other code may update workingVal

      s.remove(((i % 2) == 1) ? i-1 : workingVal);
    }
    System.out.println(s.size());
  }
}

The combination of values of types short and int in the second argument of the conditional expression (the operation i-1) causes the result to be an int, as specified by the integer promotion rules. Consequently, the Short object in the third argument is unboxed into a short, which is then promoted into an int. The result of the conditional expression is then autoboxed into an object of type Integer. Because the HashSet contains only values of type Short, the call to HashSet.remove() has no effect.

Compliant Solution

This compliant solution casts the second operand to type short, then explicitly invokes the Short.valueOf() method to create a Short instance whose value is i-1:

public class ShortSet {
  public static void main(String[] args) {
    HashSet<Short> s = new HashSet<Short>();
    for (short i = 0; i < 100; i++) {
      s.add(i);
      // Cast of i-1 is safe because the resulting value is always representable
      Short workingVal = (short) (i-1);
      // ... Other code may update workingVal

      // Cast of i-1 is safe because the resulting value is always representable
      s.remove(((i % 2) == 1) ? Short.valueOf((short) (i-1)) : workingVal);
    }
    System.out.println(s.size());
  }
}

As a result of the cast, the second and third operands of the conditional expression both have type Short, and the remove() call has the expected result.

Writing the conditional expression as ((i % 2) == 1) ? (short) (i-1)) : workingVal also complies with this guideline because both the second and third operands in this form have type short. However, this alternative is less efficient because it forces unboxing of workingVal on each even iteration of the loop and autoboxing of the result of the conditional expression (from short to Short) on every iteration of the loop.

Applicability

When the second and third operands of a conditional expression have different types, they can be subject to unexpected type conversions.

Automated detection of condition expressions whose second and third operands are of different types is straightforward.

Automated Detection

ToolVersionCheckerDescription
Parasoft Jtest
2023.1
CERT.EXP55.COMTAvoid using the conditional operator with mismatched numeric types

Bibliography

[Bloch 2005]

Puzzle 8, "Dos Equis"

[Findbugs 2008]

"Bx: Primitive Value Is Unboxed and Coerced for Ternary Operator"

[JLS 2013]

§15.25, "Conditional Operator ? :"



6 Comments

  1. I move that this be promoted to a rule.

    1. What are the security implications of violating this guideline? To promote this to a rule, we need to demonstrate that violating this guideline can have detrimental effects on security. That is, you can get a vulnerability, rather than just a bug.

      1. Ah! I didn't understand that all "Rules" are security-related. This is quite interesting. 

        Back on the subject of EXP55-J., paranoiacs would say that any time you get unexpected results it's a vulnerability. But my personal standards are a little more stringent. (smile)

        So I withdraw the motion.

        1. The Rules versus Recommendations (Java) page should help explain the difference.

          There are many who say every bug is a potential vulnerability. So we do assume that there are bugs that are not vuls. So to be more precise, you would have to show that violating this guideline is likely to result in a vulnerability, whereas violating other guidelines can still result in non-vulnerability bugs.

          1. Well, now I'm questioning this again. The Rules versus Recommendations page says (emphasis added)

            "likely to result in a defect that may adversely affect the safety, reliability, or security of a system, for example, by introducing a security flaw that may result in an exploitable vulnerability."

            ...?

            1. Well, the point is that if you violate this rule, you will usually not affect the safety, reliability, or security. There are cases where you would (eg if you made this mistake while writing cryptography), but in the common case this has no effect on safety, reliability or security. 

              This is prob a case where no one has built a proof-of-concept exploitable code based on this bug, rather than that someone has proved it is not possible. Also no cases where someone pwned a program b/c of this bug have come to light.