The interfaces of the Java Collections Framework [JCF 2014] use generically typed, parameterized methods, such as add(E e)
and put(K key, V value)
, to insert objects into the collection or map, but they have other methods, such as contains()
, remove()
, and get()
, that accept an argument of type Object
rather than a parameterized type. Consequently, these methods accept an object of any type. The collections framework interfaces were designed in this manner to maximize backwards compatibility, but this design can also lead to coding errors. Programmers must ensure that arguments passed to methods such as Map<K,V>
get()
, Collection<E>
contains()
, and remove()
have the same type as the parameterized type of the corresponding class instance.
Noncompliant Code Example
After adding and removing 10 elements, the HashSet
in this noncompliant code example still contains 10
elements and not the expected 0. Java's type checking requires that only values of type Short
can be inserted into s. Consequently, the programmer has added a cast to short
so that the code will compile. However, the Collections<E>.remove()
method accepts an argument of type Object
rather than of type E
, allowing a programmer to attempt to remove an object of any type. In this noncompliant code example, the programmer has neglected to also cast the variable i
before passing it to the remove()
method, which is autoboxed into an object of type Integer
rather than one of type Short
. The HashSet
contains only values of type Short
; the code attempts to remove objects of type Integer
. Consequently, the remove()
method has no effect.
Code Block | ||
---|---|---|
| ||
import java.util.HashSet;
public class ShortSet {
public static void main(String[] args) {
HashSet<Short> s = new HashSet<Short>();
for |
Autoboxing can automatically wrap the primitive type to the corresponding wrapper object, which can be convenient in many cases and avoid clutters in your own code. But you should always be careful about this process, especially when comparision. Considerthe follwing code:
Code Example
Code Block |
---|
public class TestWrapper2 {
 public static void main(String[] args) {
 Â
  Integer i1 = 100;
    Integer i2 = 100;
    Integer i3 = 1000;
    Integer i4 = 1000;
    System.out.println(i1==i2);
    System.out.println(i3==i4);
   Â
 }
}
|
Output of this code
Code Block |
---|
true
false
|
It is because that in JDK 5.0, if the value p being boxed is true, false, a byte, an ASCII character, or an integer 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. And the reason for this rule explained in criterion for autoboxing:
"Ideally, boxing a given primitive value p, would always yield an identical reference. In practice, this may not be feasible using existing implementation techniques. The rules above are a pragmatic compromise. The final clause above requires that certain common values always be boxed into indistinguishable objects. The implementation may cache these, lazily or eagerly."
To convince our idea, we can take an insight of the source code of Integer of JDK 1.6.0_10:
Code Block |
---|
 private static class IntegerCache {  private IntegerCache(){}  static final Integer cache[] = new Integer[-(-128) + 127 + 1];  static {     for(int i = 0; i < cache.length10; i++)   cache[i] = new Integer(i - 128);  }    } |
Here there exists a cache in the Integer, which clearly explains the result of above code. It also means that if we have enough memory, we could caches all the integer value(-32K-32K), which means that all the int value could be autoboxing to the same Integer object. However, actually it is impractical, so we should be careful about using the following code.
Code Example
...
{
s.add((short)i); // Cast required so that the code compiles
s.remove(i); // Tries to remove an Integer
}
System.out.println(s.size());
}
}
|
This noncompliant code example also violates EXP00-J. Do not ignore values returned by methods because the remove()
method returns a Boolean value indicating its success.
Compliant Solution
Objects removed from a collection must share the type of the collection elements. Numeric promotion and autoboxing can produce unexpected object types. This compliant solution uses an explicit cast to short
that matches the intended boxed type.
Code Block | ||
---|---|---|
| ||
import java.util.HashSet;
public class ShortSet {
public | ||
Code Block | ||
import java.util.ArrayList; public class TestWrapper1 {  public static void main(String[] args) {   //create an array list ofHashSet<Short> integers, which each element   //is more than 127     ArrayList<Integer> list1 s = new ArrayList<Integer>HashSet<Short>();     for for(int i = 0;i<10; i < 10; i++) {      list1s.add((short)i);   //create another array list of integers, which each element   //is the same with the first one     ArrayList<Integer> list2 = new ArrayList<Integer>();     for(int i=0;i<10;i++)      list2.add(i);                int counter = 0;     for(int i=0;i<10;i++)      if(list1.get(i) == list2.get(i)) counter++;     //output the total equal number     System.out.println(counter);  } } |
JDK 5.0, the output of this code is 0. But if we set more caches inside Integer, then the result may be different!
Risk Assessment
The result is an undefined behavior, so it will exert a potential security risk.
// Remove a Short
if (s.remove((short)i) == false) {
System.err.println("Error removing " + i);
}
}
System.out.println(s.size());
}
}
|
Exceptions
EXP04-J-EX1: The collections framework equals()
method also takes an argument of type Object
, but it is acceptable to pass an object of a different type from that of the underlying collection/map to the equals()
method. Doing so cannot cause any confusion because the contract of the equals()
method stipulates that objects of different classes will never be equivalent (see MET08-J. Preserve the equality contract when overriding the equals() method for more information).
EXP04-J-EX2: Some Java programs, particularly legacy programs, may iterate through a collection of variously typed objects with the expectation that only those objects with the same type as the collection parameter will be operated on. An exception is allowed when there is no expectation that the operation is not a no-op.
Risk Assessment
Passing arguments to certain Java Collection Framework methods that are of a different type from that of the class instance can cause silent failures, resulting in unintended object retention, memory leaks, or incorrect program operation [Techtalk 2007].
Rule |
---|
Severity | Likelihood | Remediation Cost | Priority | Level |
---|
EXP04-J |
Low |
Probable |
Low |
P6 |
L2 |
Automated Detection
TODO
Related Vulnerabilities
Search for vulnerabilities resulting from the violation of this rule on the CERT website.
References
Detection of invocations of Collection.remove()
whose operand fails to match the type of the elements of the underlying collection is straightforward. It is possible, although unlikely, that some of these invocations could be intended. The remainder are heuristically likely to be in error. Automated detection for other APIs could be possible.
Tool | Version | Checker | Description | ||||||
---|---|---|---|---|---|---|---|---|---|
PVS-Studio |
| V6066 | |||||||
SonarQube |
| S2175 | Inappropriate "Collection" calls should not be made |
Bibliography
Chapter 5, "Inheritance" | |
[JCF 2014] | The Java Collections Framework |
[JLS 2015] | |
[Seacord 2015] | |
"The Joy of Sets" |
...
Chapter 5, Core Java⢠2 Volume I - Fundamentals, Seventh Edition