Effective Java Generics

This chapter is from the book

Effective Java, 3rd Edition

Learn More Buy

This chapter is from the book

This chapter is from the book 

Effective Java, 3rd Edition

Learn More Buy

Item 32: Combine generics and varargs judiciously

Varargs methods (Item 53) and generics were both added to the platform in Java 5, so you might expect them to interact gracefully; sadly, they do not. The purpose of varargs is to allow clients to pass a variable number of arguments to a method, but it is a leaky abstraction: when you invoke a varargs method, an array is created to hold the varargs parameters; that array, which should be an implementation detail, is visible. As a consequence, you get confusing compiler warnings when varargs parameters have generic or parameterized types.

Recall from Item 28 that a non-reifiable type is one whose runtime representation has less information than its compile-time representation, and that nearly all generic and parameterized types are non-reifiable. If a method declares its varargs parameter to be of a non-reifiable type, the compiler generates a warning on the declaration. If the method is invoked on varargs parameters whose inferred type is non-reifiable, the compiler generates a warning on the invocation too. The warnings look something like this:

warning: [unchecked] Possible heap pollution from
    parameterized vararg type List<String>

Heap pollution occurs when a variable of a parameterized type refers to an object that is not of that type [JLS, 4.12.2]. It can cause the compiler’s automatically generated casts to fail, violating the fundamental guarantee of the generic type system.

For example, consider this method, which is a thinly disguised variant of the code fragment on page 127:

// Mixing generics and varargs can violate type safety!
static void dangerous(List<String>... stringLists) {
    List<Integer> intList = List.of(42);
    Object[] objects = stringLists;
    objects[0] = intList;             // Heap pollution
    String s = stringLists[0].get(0); // ClassCastException
}

This method has no visible casts yet throws a ClassCastException when invoked with one or more arguments. Its last line has an invisible cast that is generated by the compiler. This cast fails, demonstrating that type safety has been compromised, and it is unsafe to store a value in a generic varargs array parameter.

This example raises an interesting question: Why is it even legal to declare a method with a generic varargs parameter, when it is illegal to create a generic array explicitly? In other words, why does the method shown previously generate only a warning, while the code fragment on page 127 generates an error? The answer is that methods with varargs parameters of generic or parameterized types can be very useful in practice, so the language designers opted to live with this inconsistency. In fact, the Java libraries export several such methods, including Arrays.asList(T... a), Collections.addAll(Collection<? super T> c, T... elements), and EnumSet.of(E first, E... rest). Unlike the dangerous method shown earlier, these library methods are typesafe.

Prior to Java 7, there was nothing the author of a method with a generic varargs parameter could do about the warnings at the call sites. This made these APIs unpleasant to use. Users had to put up with the warnings or, preferably, to eliminate them with @SuppressWarnings("unchecked") annotations at every call site (Item 27). This was tedious, harmed readability, and hid warnings that flagged real issues.

In Java 7, the SafeVarargs annotation was added to the platform, to allow the author of a method with a generic varargs parameter to suppress client warnings automatically. In essence, the SafeVarargs annotation constitutes a promise by the author of a method that it is typesafe. In exchange for this promise, the compiler agrees not to warn the users of the method that calls may be unsafe.

It is critical that you do not annotate a method with @SafeVarargs unless it actually is safe. So what does it take to ensure this? Recall that a generic array is created when the method is invoked, to hold the varargs parameters. If the method doesn’t store anything into the array (which would overwrite the parameters) and doesn’t allow a reference to the array to escape (which would enable untrusted code to access the array), then it’s safe. In other words, if the varargs parameter array is used only to transmit a variable number of arguments from the caller to the method—which is, after all, the purpose of varargs—then the method is safe.

It is worth noting that you can violate type safety without ever storing anything in the varargs parameter array. Consider the following generic varargs method, which returns an array containing its parameters. At first glance, it may look like a handy little utility:

// UNSAFE - Exposes a reference to its generic parameter array!
static <T> T[] toArray(T... args) {
    return args;
}

This method simply returns its varargs parameter array. The method may not look dangerous, but it is! The type of this array is determined by the compile-time types of the arguments passed in to the method, and the compiler may not have enough information to make an accurate determination. Because this method returns its varargs parameter array, it can propagate heap pollution up the call stack.

To make this concrete, consider the following generic method, which takes three arguments of type T and returns an array containing two of the arguments, chosen at random:

static <T> T[] pickTwo(T a, T b, T c) {
    switch(ThreadLocalRandom.current().nextInt(3)) {
      case 0: return toArray(a, b);
      case 1: return toArray(a, c);
      case 2: return toArray(b, c);
    }
    throw new AssertionError(); // Can't get here
}

This method is not, in and of itself, dangerous and would not generate a warning except that it invokes the toArray method, which has a generic varargs parameter.

When compiling this method, the compiler generates code to create a varargs parameter array in which to pass two T instances to toArray. This code allocates an array of type Object[], which is the most specific type that is guaranteed to hold these instances, no matter what types of objects are passed to pickTwo at the call site. The toArray method simply returns this array to pickTwo, which in turn returns it to its caller, so pickTwo will always return an array of type Object[].

Now consider this main method, which exercises pickTwo:

public static void main(String[] args) {
    String[] attributes = pickTwo("Good", "Fast", "Cheap");
}

There is nothing at all wrong with this method, so it compiles without generating any warnings. But when you run it, it throws a ClassCastException, though it contains no visible casts. What you don’t see is that the compiler has generated a hidden cast to String[] on the value returned by pickTwo so that it can be stored in attributes. The cast fails, because Object[] is not a subtype of String[]. This failure is quite disconcerting because it is two levels removed from the method that actually causes the heap pollution (toArray), and the varargs parameter array is not modified after the actual parameters are stored in it.

This example is meant to drive home the point that it is unsafe to give another method access to a generic varargs parameter array, with two exceptions: it is safe to pass the array to another varargs method that is correctly annotated with @SafeVarargs, and it is safe to pass the array to a non-varargs method that merely computes some function of the contents of the array.

Here is a typical example of a safe use of a generic varargs parameter. This method takes an arbitrary number of lists as arguments and returns a single list containing the elements of all of the input lists in sequence. Because the method is annotated with @SafeVarargs, it doesn’t generate any warnings, on the declaration or at its call sites:

// Safe method with a generic varargs parameter
@SafeVarargs
static <T> List<T> flatten(List<? extends T>... lists) {
    List<T> result = new ArrayList<>();
    for (List<? extends T> list : lists)
        result.addAll(list);
    return result;
}

The rule for deciding when to use the SafeVarargs annotation is simple: Use @SafeVarargs on every method with a varargs parameter of a generic or parameterized type, so its users won’t be burdened by needless and confusing compiler warnings. This implies that you should never write unsafe varargs methods like dangerous or toArray. Every time the compiler warns you of possible heap pollution from a generic varargs parameter in a method you control, check that the method is safe. As a reminder, a generic varargs methods is safe if:

1. it doesn’t store anything in the varargs parameter array, and

2. it doesn’t make the array (or a clone) visible to untrusted code.

If either of these prohibitions is violated, fix it.

Note that the SafeVarargs annotation is legal only on methods that can’t be overridden, because it is impossible to guarantee that every possible overriding method will be safe. In Java 8, the annotation was legal only on static methods and final instance methods; in Java 9, it became legal on private instance methods as well.

An alternative to using the SafeVarargs annotation is to take the advice of Item 28 and replace the varargs parameter (which is an array in disguise) with a List parameter. Here’s how this approach looks when applied to our flatten method. Note that only the parameter declaration has changed:

// List as a typesafe alternative to a generic varargs parameter
static <T> List<T> flatten(List<List<? extends T>> lists) {
    List<T> result = new ArrayList<>();
    for (List<? extends T> list : lists)
        result.addAll(list);
    return result;
}

This method can then be used in conjunction with the static factory method List.of to allow for a variable number of arguments. Note that this approach relies on the fact that the List.of declaration is annotated with @SafeVarargs:

audience = flatten(List.of(friends, romans, countrymen));

The advantage of this approach is that the compiler can prove that the method is typesafe. You don’t have to vouch for its safety with a SafeVarargs annotation, and you don’t have worry that you might have erred in determining that it was safe. The main disadvantage is that the client code is a bit more verbose and may be a bit slower.

This trick can also be used in situations where it is impossible to write a safe varargs method, as is the case with the toArray method on page 147. Its List analogue is the List.of method, so we don’t even have to write it; the Java libraries authors have done the work for us. The pickTwo method then becomes this:

static <T> List<T> pickTwo(T a, T b, T c) {
    switch(rnd.nextInt(3)) {
      case 0: return List.of(a, b);
      case 1: return List.of(a, c);
      case 2: return List.of(b, c);
    }
    throw new AssertionError();
}

and the main method becomes this:

public static void main(String[] args) {
    List<String> attributes = pickTwo("Good", "Fast", "Cheap");
}

The resulting code is typesafe because it uses only generics, and not arrays.

In summary, varargs and generics do not interact well because the varargs facility is a leaky abstraction built atop arrays, and arrays have different type rules from generics. Though generic varargs parameters are not typesafe, they are legal. If you choose to write a method with a generic (or parameterized) varargs parameter, first ensure that the method is typesafe, and then annotate it with @SafeVarargs so it is not unpleasant to use.