1. Introduction
This tutorial is a guide to different functional interfaces present in Java 8, as well as their general use cases, and usage in the standard JDK library.
2. Lambdas in Java 8
Java 8 brought a powerful new syntactic improvement in the form of lambda expressions. A lambda is an anonymous function that we can handle as a first-class language citizen. For instance, we can pass it to or return it from a method.
Before Java 8, we would usually create a class for every case where we needed to encapsulate a single piece of functionality. This implied a lot of unnecessary boilerplate code to define something that served as a primitive function representation.
The article “Lambda Expressions and Functional Interfaces: Tips and Best Practices” describes in more detail the functional interfaces and best practices of working with lambdas. This guide focuses on some particular functional interfaces that are present in the java.util.function package.
3. Functional Interfaces
It’s recommended that all functional interfaces have an informative @FunctionalInterface annotation. This clearly communicates the purpose of the interface, and also allows a compiler to generate an error if the annotated interface does not satisfy the conditions.
Any interface with a SAM(Single Abstract Method) is a functional interface, and its implementation may be treated as lambda expressions.
Note that Java 8’s default methods are not abstract and do not count; a functional interface may still have multiple default methods. We can observe this by looking at the Function’s documentation.
4. Functions
The most simple and general case of a lambda is a functional interface with a method that receives one value and returns another. This function of a single argument is represented by the Function interface, which is parameterized by the types of its argument and a return value:
public interface Function<T, R> { … }
One of the usages of the Function type in the standard library is the Map.computeIfAbsent method. This method returns a value from a map by key, but calculates a value if a key is not already present in a map. To calculate a value, it uses the passed Function implementation:
Map<String, Integer> nameMap = new HashMap<>();
Integer value = nameMap.computeIfAbsent("John", s -> s.length());
In this case, we will calculate a value by applying a function to a key, put inside a map, and also returned from a method call. W****e may replace the lambda with a method reference that matches passed and returned value types.
Remember that an object we invoke the method on is, in fact, the implicit first argument of a method. This allows us to cast an instance method length reference to a Function interface:
Integer value = nameMap.computeIfAbsent("John", String::length);
The Function interface also has a default compose method that allows us to combine several functions into one and execute them sequentially:
Function<Integer, String> intToString = Object::toString;
Function<String, String> quote = s -> "'" + s + "'";
Function<Integer, String> quoteIntToString = quote.compose(intToString);
assertEquals("'5'", quoteIntToString.apply(5));
The quoteIntToString function is a combination of the quote function applied to a result of the intToString function.
5. Primitive Function Specializations
Since a primitive type can’t be a generic type argument, there are versions of the Function interface for the most used primitive types double, int, long, and their combinations in argument and return types:
- IntFunction, LongFunction, DoubleFunction: arguments are of specified type, return type is parameterized
- ToIntFunction, ToLongFunction, ToDoubleFunction: return type is of specified type, arguments are parameterized
- DoubleToIntFunction, DoubleToLongFunction, IntToDoubleFunction, IntToLongFunction, LongToIntFunction, LongToDoubleFunction: having both argument and return type defined as primitive types, as specified by their names
As an example, there is no out-of-the-box functional interface for a function that takes a short and returns a byte, but nothing stops us from writing our own:
@FunctionalInterface
public interface ShortToByteFunction {
byte applyAsByte(short s);
}
Now we can write a method that transforms an array of short to an array of byte using a rule defined by a ShortToByteFunction:
public byte[] transformArray(short[] array, ShortToByteFunction function) {
byte[] transformedArray = new byte[array.length];
for (int i = 0; i < array.length; i++) {
transformedArray[i] = function.applyAsByte(array[i]);
}
return transformedArray;
}
Here’s how we could use it to transform an array of shorts to an array of bytes multiplied by 2:
short[] array = {(short) 1, (short) 2, (short) 3};
byte[] transformedArray = transformArray(array, s -> (byte) (s * 2));
byte[] expectedArray = {(byte) 2, (byte) 4, (byte) 6};
assertArrayEquals(expectedArray, transformedArray);
6. Two-Arity Function Specializations
To define lambdas with two arguments, we have to use additional interfaces that contain “Bi” keyword in their names: BiFunction, ToDoubleBiFunction, ToIntBiFunction, and ToLongBiFunction.
BiFunction has both arguments and a return type generified, while ToDoubleBiFunction and others allow us to return a primitive value.
One of the typical examples of using this interface in the standard API is in the Map.replaceAll method, which allows replacing all values in a map with some computed value.
Let’s use a BiFunction implementation that receives a key and an old value to calculate a new value for the salary and return it.
Map<String, Integer> salaries = new HashMap<>();
salaries.put("John", 40000);
salaries.put("Freddy", 30000);
salaries.put("Samuel", 50000);
salaries.replaceAll((name, oldValue) ->
name.equals("Freddy") ? oldValue : oldValue + 10000);
7. Suppliers
The Supplier functional interface is yet another Function specialization that does not take any arguments. We typically use it for lazy generation of values. For instance, let’s define a function that squares a double value. It will not receive a value itself, but a Supplier of this value:
public double squareLazy(Supplier<Double> lazyValue) {
return Math.pow(lazyValue.get(), 2);
}
This allows us to lazily generate the argument for invocation of this function using a Supplier implementation. This can be useful if the generation of the argument takes a considerable amount of time. We’ll simulate that using Guava’s sleepUninterruptibly method:
Supplier<Double> lazyValue = () -> {
Uninterruptibles.sleepUninterruptibly(1000, TimeUnit.MILLISECONDS);
return 9d;
};
Double valueSquared = squareLazy(lazyValue);
Another use case for the Supplier is defining logic for sequence generation. To demonstrate it, let’s use a static Stream.generate method to create a Stream of Fibonacci numbers:
int[] fibs = {0, 1};
Stream<Integer> fibonacci = Stream.generate(() -> {
int result = fibs[1];
int fib3 = fibs[0] + fibs[1];
fibs[0] = fibs[1];
fibs[1] = fib3;
return result;
});
The function that we pass to the Stream.generate method implements the Supplier functional interface. Notice that to be useful as a generator, the Supplier usually needs some sort of external state. In this case, its state comprises the last two Fibonacci sequence numbers.
To implement this state, we use an array instead of a couple of variables because all external variables used inside the lambda have to be effectively final.
Other specializations of the Supplier functional interface include BooleanSupplier, DoubleSupplier, LongSupplier and IntSupplier, whose return types are corresponding primitives.
8. Consumers
As opposed to the Supplier, the Consumer accepts a generified argument and returns nothing. It is a function that is representing side effects.
For instance, let’s greet everybody in a list of names by printing the greeting in the console. The lambda passed to the List.forEach method implements the Consumer functional interface:
List<String> names = Arrays.asList("John", "Freddy", "Samuel");
names.forEach(name -> System.out.println("Hello, " + name));
There are also specialized versions of the Consumer — DoubleConsumer, IntConsumer and LongConsumer — that receive primitive values as arguments. More interesting is the BiConsumer interface. One of its use cases is iterating through the entries of a map:
Map<String, Integer> ages = new HashMap<>();
ages.put("John", 25);
ages.put("Freddy", 24);
ages.put("Samuel", 30);
ages.forEach((name, age) -> System.out.println(name + " is " + age + " years old"));
Another set of specialized BiConsumer versions is comprised of ObjDoubleConsumer, ObjIntConsumer, and ObjLongConsumer, which receive two arguments; one of the arguments is generified, and the other is a primitive type.
9. Predicates
In mathematical logic, a predicate is a function that receives a value and returns a boolean value.
The Predicate functional interface is a specialization of a Function that receives a generified value and returns a boolean. A typical use case of the Predicate lambda is to filter a collection of values:
List<String> names = Arrays.asList("Angela", "Aaron", "Bob", "Claire", "David");
List<String> namesWithA = names.stream()
.filter(name -> name.startsWith("A"))
.collect(Collectors.toList());
In the code above, we filter a list using the Stream API and keep only the names that start with the letter “A”. The Predicate implementation encapsulates the filtering logic.
As in all of the previous examples, there are IntPredicate, DoublePredicate and LongPredicate versions of this function that receive primitive values.
10. Operators
Operator interfaces are special cases of a function that receive and return the same value type. The UnaryOperator interface receives a single argument. One of its use cases in the Collections API is to replace all values in a list with some computed values of the same type:
List<String> names = Arrays.asList("bob", "josh", "megan");
names.replaceAll(name -> name.toUpperCase());
The List.replaceAll function returns void as it replaces the values in place. To fit the purpose, the lambda used to transform the values of a list has to return the same result type as it receives. This is why the UnaryOperator is useful here.
Of course, instead of name -> name.toUpperCase(), we can simply use a method reference:
names.replaceAll(String::toUpperCase);
One of the most interesting use cases of a BinaryOperator is a reduction operation. Suppose we want to aggregate a collection of integers in a sum of all values. With Stream API, we could do this using a collector*,* but a more generic way to do it would be to use the reduce method:
List<Integer> values = Arrays.asList(3, 5, 8, 9, 12);
int sum = values.stream()
.reduce(0, (i1, i2) -> i1 + i2);
The reduce method receives an initial accumulator value and a BinaryOperator function. The arguments of this function are a pair of values of the same type; the function itself also contains a logic for joining them in a single value of the same type. The passed function must be associative, which means that the order of value aggregation does not matter, i.e. the following condition should hold:
op.apply(a, op.apply(b, c)) == op.apply(op.apply(a, b), c)
The associative property of a BinaryOperator operator function allows us to easily parallelize the reduction process.
Of course, there are also specializations of UnaryOperator and BinaryOperator that can be used with primitive values, namely DoubleUnaryOperator, IntUnaryOperator, LongUnaryOperator, DoubleBinaryOperator, IntBinaryOperator, and LongBinaryOperator.
11. Legacy Functional Interfaces
Not all functional interfaces appeared in Java 8. Many interfaces from previous versions of Java conform to the constraints of a FunctionalInterface, and we can use them as lambdas. Prominent examples include the Runnable and Callable interfaces that are used in concurrency APIs. In Java 8, these interfaces are also marked with a @FunctionalInterface annotation. This allows us to greatly simplify concurrency code:
Thread thread = new Thread(() -> System.out.println("Hello From Another Thread"));
thread.start();
12. Conclusion
In this article, we examined different functional interfaces present in the Java 8 API that we can use as lambda expressions. The source code for the article is available over on GitHub.