Java SE 8's New Streams API
Java SE 8 strengthens the Java language by introducing lambda expressions and additional language features. Although lambdas help you write functional code that's more compact than the anonymous class equivalent, they more importantly facilitate parallel processing in the context of the Java SE 8 Streams API.
To learn about lambdas, check out my article Java SE 8's New Language Features, Part 1: Interface Default/Static Methods and Lambda Expressions.
In this article, I first introduce you to the Streams API, where you learn about API fundamentals, receive an overview of key types, and discover the need for this API. Then we explore various API operations that you can perform on stream sources; for example, performing actions on all stream elements, filtering, mapping, and reduction.
I developed this article's applications with the 64-bit version of JDK 8 build 132 on a Windows 7 platform. You can download the code from this article here.
Introducing the Streams API
The Streams API facilitates the processing of data items sequentially or in parallel. This section first explores several fundamentals on which this API is based; for example, what is a stream, and what is a source? Next, I present an overview of the Streams API in terms of package and key types. Finally, we examine the need for streams.
Streams API Fundamentals
The Streams API is based on the concept of a stream, which is a sequence of elements originating from a source and supporting sequential and parallel aggregate operations. A source is an entity that stores elements (for example, a collection) or generates elements (such as a random number generator). An aggregate is a single result calculated from multiple input values.
Streams support various operations that are either intermediate or terminal. An intermediate operation returns a new stream, whereas a terminal operation consumes the stream. Operations are connected into a pipeline, which starts with a source followed by zero or more intermediate operations and ending with a terminal operation. Figure 1 illustrates this arrangement.
Figure 1 A pipeline of operations in which an intermediate operation's output becomes the input to the next operation
Intermediate operations are always lazy in that they don't do anything until a terminal operation is executed. Instead, an intermediate operation returns a new stream that is traversed after the pipeline's terminal operation is executed. Note that an intermediate operation doesn't have to finish before its output becomes available to the next operation in the pipeline.
Intermediate operations are classified as stateless and stateful. Stateless operations retain no state from a previously seen element when processing a new element. Each element can be processed independently of operations on other elements. Stateful operations may incorporate state from previously seen elements when processing new elements.
Terminal operations may traverse the stream to produce a side-effect or result. A side-effect is the execution of a code block. Although side-effects are generally discouraged because they can often lead to thread-safety hazards, some side-effects (such as printing elements for debugging purposes) are typically harmless.
After a terminal operation completes, the pipeline is considered to be consumed and no longer can be used. To traverse the same source again, you must obtain a new stream from the source. In most cases, terminal operations are eager in that they complete their source traversal and pipeline processing before returning.
Some operations are classified as short-circuiting. An intermediate operation is short-circuiting if it produces a finite stream when presented with infinite input. A terminal operation is short-circuiting if it terminates in finite time when presented with infinite input. Having a short-circuiting operation in the pipeline is a necessary (but not sufficient) condition for processing an infinite stream to terminate normally in finite time.
Some streams have a defined encounter order that specifies the order in which elements are returned. Whether a stream has an encounter order depends on its source and intermediate operations. Stream sources such as java.util.List and arrays are intrinsically ordered, whereas sources such as java.util.HashSet are not.
Some intermediate operations may impose an encounter order on an otherwise unordered stream, whereas other intermediate operations may return an unordered stream for an ordered stream. Also, terminal operations might ignore an encounter order. For an ordered stream, most operations are constrained to operate on the elements in their encounter order.
Streams API Overview
The Streams API is associated with the java.util.stream package. This package consists of two classes, one enum, and several interfaces, with BaseStream<T,S extends BaseStream<T,S>> being the fundamental stream interface that's extended by the following interfaces:
- Stream<T>: A sequence of object elements of type T supporting sequential and parallel aggregate operations.
- DoubleStream: A sequence of primitive double-valued elements supporting sequential and parallel aggregate operations. This is the double primitive specialization of Stream.
- IntStream: A sequence of primitive int-valued elements supporting sequential and parallel aggregate operations. This is the int primitive specialization of Stream.
- LongStream: A sequence of primitive long-valued elements supporting sequential and parallel aggregate operations. This is the long primitive specialization of Stream.
Stream<T> describes streams of objects (of type T). Although you can use this interface to describe streams of ints, longs, and doubles, doing so isn't efficient because of the need for autoboxing and unboxing (and java.lang.Double, java.lang.Integer, and java.lang.Long wrapper objects created behind the scenes). This is the reason for DoubleStream, IntStream, and LongStream.
Java provides various ways to obtain a BaseStream from collections and other sources:
- From a collection via java.util.Collection's Stream<E> stream() (return a sequential stream) and Stream<E> parallelStream() (return a parallel stream—a sequential stream might be returned) default methods.
- From an array via the various stream() methods of the java.util.Arrays class.
- From static factory methods on the stream classes, such as <T> Stream<T> of(T t), IntStream range(int startInclusive, int endExclusive), and <T> Stream<T> iterate(T seed, UnaryOperator<T> f).
- From java.io.BufferedReader's Stream<String> lines() method.
- From java.nio.file.Files methods such as Stream<String> lines(Path path).
- From a stream of random numbers generated by java.util.Random methods such as IntStream ints().
- From additional JDK stream-oriented methods such as java.util.BitSet's IntStream stream() and java.util.regex.Pattern's Stream<String> splitAsStream(CharSequence input).
Additionally, stream sources can be provided by third-party libraries.
The Need for Streams
Multicore processors have raised awareness of parallelism, which is the ability to execute multiple tasks at the same time, greatly improving performance. The introduction of the Fork/Join Framework in Java SE 7 gave Java developers the ability to start leveraging parallelism in their applications. But this was only the beginning.
The Java Collections Framework was introduced (as part of JDK 1.2) before today's emphasis on parallelism. Although this framework could benefit from leveraging multicore processors, it isn't amenable to parallelism. A significant impediment is the need to use external iteration to retrieve objects from a collection. The following example demonstrates external iteration:
List<String> birds = Arrays.asList("Robin", "Bluejay", "Penguin", "Ostrich", "Canary"); // Employ external iteration. for (String bird: birds) System.out.println(bird);
This example creates a list of strings, using the enhanced for loop statement to iterate over this list and subsequently output each object's string representation on its own line. Although external iteration is easy to specify, it's problematic for the following reasons:
- It's inherently sequential. External iteration is sequential and must process the elements in the order that the collection specifies. Significant refactoring would be required to leverage parallelism and improve performance, especially for huge collections.
- It's client-centric. External iteration is part of client code and not part of a library. If the code was generalized and stored in a library, you might have opportunities to exploit parallelism and other features that improve performance. As it stands, you would have to refactor the code and develop a library that hides the iteration and provides you with these performance-improvement opportunities. This isn't an easy task if you're unfamiliar with Fork/Join, which you would probably use to exploit parallelism. Also, a redeployment of the code might prove costly or otherwise problematic.
Ideally, Collections Framework types should support internal iteration that can exploit parallelism and other performance-enhancing features. Because introducing this capability would break legacy code, Java SE 8 introduced the Streams API (and lambdas) and enhanced the Collection interface with stream() and parallelStream() default methods to make the Collections Framework parallel-friendly.
The following example demonstrates internal iteration in sequential and parallel contexts, contrasting the previous example's verbosity with Streams API/lambda elegance:
// Employ internal iteration. birds.stream().forEach(System.out::println); birds.parallelStream().forEach(System.out::println);
I first invoke stream() or parallelStream() on birds to obtain a stream for the list. I then invoke forEach() on the resulting stream to iterate over the list objects. I pass a System.out::println method reference (a compact representation of a lambda) argument to forEach(), which it executes to output the object. (The output order of a parallel stream will probably differ from that of a sequential stream.)
You might be tempted to view a stream as a collection. However, streams differ from collections in the following (and other) ways:
- Lack of storage. A stream isn't a data structure that stores or generates elements. Instead, a stream transports elements from a collection or other source that stores or generates elements.
- Tendency to be lazy. Many stream operations (such as filtering) are or can be implemented lazily, which exposes optimization opportunities. For example, a "Find the first String starting with the letter J" stream operation doesn't have to examine all input strings. In contrast, collection iteration is always eager, returning every element.
- Potentially unbounded. Collections have a finite size, but streams don't need to be finite and are often referred to as infinite streams. Various operations can allow computations on infinite streams to complete in finite time.
Finally, although the Streams API is largely present to make the Collections Framework parallel-friendly, this isn't its only reason for existence. As you've previously learned, other entities (such as arrays) can serve as stream sources. There is definitely a need for the Streams API.