1. Общ преглед
В този задълбочен урок ще преминем през практическото използване на Java 8 Streams от създаването до паралелното изпълнение.
За да разберат този материал, читателите трябва да имат основни познания за Java 8 (ламбда изрази, незадължителни, препратки към методите) и за API на Stream. Ако не сте запознати с тези теми, моля, разгледайте предишните ни статии - Нови функции в Java 8 и Въведение в Java 8 Streams.
2. Създаване на поток
Има много начини за създаване на екземпляр на поток от различни източници. Веднъж създаден, екземплярът няма да модифицира своя източник, поради което позволява създаването на множество екземпляри от един източник.
2.1. Празен поток
Методът празен () трябва да се използва в случай на създаване на празен поток:
Stream streamEmpty = Stream.empty();Често се случва методът empty () да се използва при създаването, за да се избегне връщането на null за потоци без елемент:
public Stream streamOf(List list) return list == null 2.2. Поток на колекция
Потокът може да бъде създаден и от всякакъв тип колекция ( Collection, List, Set ):
Collection collection = Arrays.asList("a", "b", "c"); Stream streamOfCollection = collection.stream();2.3. Поток от масив
Масивът също може да бъде източник на поток:
Stream streamOfArray = Stream.of("a", "b", "c");Те също могат да бъдат създадени от съществуващ масив или от част от масив:
String[] arr = new String[]{"a", "b", "c"}; Stream streamOfArrayFull = Arrays.stream(arr); Stream streamOfArrayPart = Arrays.stream(arr, 1, 3);2.4. Stream.builder ()
Когато се използва конструктор, желаният тип трябва да бъде допълнително посочен в дясната част на израза, в противен случай методът build () ще създаде екземпляр на потока:
Stream streamBuilder = Stream.builder().add("a").add("b").add("c").build();2.5. Stream.generate ()
Методът create () приема доставчик за генериране на елементи. Тъй като полученият поток е безкраен, разработчикът трябва да посочи желания размер или методът create () ще работи, докато достигне лимита на паметта:
Stream streamGenerated = Stream.generate(() -> "element").limit(10);Кодът по-горе създава последователност от десет низа със стойността - „елемент“ .
2.6. Stream.iterate ()
Друг начин за създаване на безкраен поток е чрез използване на метода iterate () :
Stream streamIterated = Stream.iterate(40, n -> n + 2).limit(20);Първият елемент от получения поток е първи параметър на метода iterate () . За създаване на всеки следващ елемент посочената функция се прилага към предишния елемент. В примера по-горе вторият елемент ще бъде 42.
2.7. Поток от примитиви
Java 8 предлага възможност за създаване на потоци от три примитивни типа: int, long и double. Тъй като Stream е общ интерфейс и няма начин да се използват примитиви като параметър на типа с генерици, бяха създадени три нови специални интерфейса: IntStream, LongStream, DoubleStream.
Използването на новите интерфейси облекчава ненужното автоматично боксиране позволява повишена производителност:
IntStream intStream = IntStream.range(1, 3); LongStream longStream = LongStream.rangeClosed(1, 3);Методът range (int startInclusive, int endExclusive) създава подреден поток от първия параметър до втория параметър. Той увеличава стойността на следващите елементи със стъпка, равна на 1. Резултатът не включва последния параметър, той е само горна граница на последователността.
Методът rangeClosed (int startInclusive, int endInclusive) прави същото само с една разлика - включен е вторият елемент. Тези два метода могат да се използват за генериране на който и да е от трите вида потоци от примитиви.
От Java 8 класът Random предоставя широк спектър от методи за генериране на потоци от примитиви. Например следният код създава DoubleStream, който има три елемента:
Random random = new Random(); DoubleStream doubleStream = random.doubles(3);2.8. Поток от струни
String може да се използва и като източник за създаване на поток.
С помощта на метода chars () от класа String . Тъй като в JDK няма интерфейс CharStream , вместо това се използва IntStream, за да представлява поток от символи.
IntStream streamOfChars = "abc".chars();Следващият пример разбива String на поднизове съгласно указания RegEx :
Stream streamOfString = Pattern.compile(", ").splitAsStream("a, b, c");2.9. Поток от файл
Java NIO клас Files позволява да се генерира поток от текстов файл чрез метода lines () . Всеки ред от текста се превръща в елемент от потока:
Path path = Paths.get("C:\\file.txt"); Stream streamOfStrings = Files.lines(path); Stream streamWithCharset = Files.lines(path, Charset.forName("UTF-8"));The Charset can be specified as an argument of the lines() method.
3. Referencing a Stream
It is possible to instantiate a stream and to have an accessible reference to it as long as only intermediate operations were called. Executing a terminal operation makes a stream inaccessible.
To demonstrate this we will forget for a while that the best practice is to chain sequence of operation. Besides its unnecessary verbosity, technically the following code is valid:
Stream stream = Stream.of("a", "b", "c").filter(element -> element.contains("b")); Optional anyElement = stream.findAny();But an attempt to reuse the same reference after calling the terminal operation will trigger the IllegalStateException:
Optional firstElement = stream.findFirst();As the IllegalStateException is a RuntimeException, a compiler will not signalize about a problem. So, it is very important to remember that Java 8 streams can't be reused.
This kind of behavior is logical because streams were designed to provide an ability to apply a finite sequence of operations to the source of elements in a functional style, but not to store elements.
So, to make previous code work properly some changes should be done:
List elements = Stream.of("a", "b", "c").filter(element -> element.contains("b")) .collect(Collectors.toList()); Optional anyElement = elements.stream().findAny(); Optional firstElement = elements.stream().findFirst();4. Stream Pipeline
To perform a sequence of operations over the elements of the data source and aggregate their results, three parts are needed – the source, intermediate operation(s) and a terminal operation.
Intermediate operations return a new modified stream. For example, to create a new stream of the existing one without few elements the skip() method should be used:
Stream onceModifiedStream = Stream.of("abcd", "bbcd", "cbcd").skip(1);If more than one modification is needed, intermediate operations can be chained. Assume that we also need to substitute every element of current Stream with a sub-string of first few chars. This will be done by chaining the skip() and the map() methods:
Stream twiceModifiedStream = stream.skip(1).map(element -> element.substring(0, 3));As you can see, the map() method takes a lambda expression as a parameter. If you want to learn more about lambdas take a look at our tutorial Lambda Expressions and Functional Interfaces: Tips and Best Practices.
A stream by itself is worthless, the real thing a user is interested in is a result of the terminal operation, which can be a value of some type or an action applied to every element of the stream. Only one terminal operation can be used per stream.
The right and most convenient way to use streams are by a stream pipeline, which is a chain of stream source, intermediate operations, and a terminal operation. For example:
List list = Arrays.asList("abc1", "abc2", "abc3"); long size = list.stream().skip(1) .map(element -> element.substring(0, 3)).sorted().count();5. Lazy Invocation
Intermediate operations are lazy. This means that they will be invoked only if it is necessary for the terminal operation execution.
To demonstrate this, imagine that we have method wasCalled(), which increments an inner counter every time it was called:
private long counter; private void wasCalled() { counter++; }Let's call method wasCalled() from operation filter():
List list = Arrays.asList(“abc1”, “abc2”, “abc3”); counter = 0; Stream stream = list.stream().filter(element -> { wasCalled(); return element.contains("2"); });As we have a source of three elements we can assume that method filter() will be called three times and the value of the counter variable will be 3. But running this code doesn't change counter at all, it is still zero, so, the filter() method wasn't called even once. The reason why – is missing of the terminal operation.
Let's rewrite this code a little bit by adding a map() operation and a terminal operation – findFirst(). We will also add an ability to track an order of method calls with a help of logging:
Optional stream = list.stream().filter(element -> { log.info("filter() was called"); return element.contains("2"); }).map(element -> { log.info("map() was called"); return element.toUpperCase(); }).findFirst();Resulting log shows that the filter() method was called twice and the map() method just once. It is so because the pipeline executes vertically. In our example the first element of the stream didn't satisfy filter's predicate, then the filter() method was invoked for the second element, which passed the filter. Without calling the filter() for third element we went down through pipeline to the map() method.
The findFirst() operation satisfies by just one element. So, in this particular example the lazy invocation allowed to avoid two method calls – one for the filter() and one for the map().
6. Order of Execution
From the performance point of view, the right order is one of the most important aspects of chaining operations in the stream pipeline:
long size = list.stream().map(element -> { wasCalled(); return element.substring(0, 3); }).skip(2).count();Execution of this code will increase the value of the counter by three. This means that the map() method of the stream was called three times. But the value of the size is one. So, resulting stream has just one element and we executed the expensive map() operations for no reason twice out of three times.
If we change the order of the skip() and the map() methods, the counter will increase only by one. So, the method map() will be called just once:
long size = list.stream().skip(2).map(element -> { wasCalled(); return element.substring(0, 3); }).count();This brings us up to the rule: intermediate operations which reduce the size of the stream should be placed before operations which are applying to each element. So, keep such methods as skip(), filter(), distinct() at the top of your stream pipeline.
7. Stream Reduction
The API has many terminal operations which aggregate a stream to a type or to a primitive, for example, count(), max(), min(), sum(), but these operations work according to the predefined implementation. And what if a developer needs to customize a Stream's reduction mechanism? There are two methods which allow to do this – the reduce()and the collect() methods.
7.1. The reduce() Method
There are three variations of this method, which differ by their signatures and returning types. They can have the following parameters:
identity – the initial value for an accumulator or a default value if a stream is empty and there is nothing to accumulate;
accumulator – a function which specifies a logic of aggregation of elements. As accumulator creates a new value for every step of reducing, the quantity of new values equals to the stream's size and only the last value is useful. This is not very good for the performance.
combiner – a function which aggregates results of the accumulator. Combiner is called only in a parallel mode to reduce results of accumulators from different threads.
So, let's look at these three methods in action:
OptionalInt reduced = IntStream.range(1, 4).reduce((a, b) -> a + b);reduced = 6 (1 + 2 + 3)
int reducedTwoParams = IntStream.range(1, 4).reduce(10, (a, b) -> a + b);reducedTwoParams = 16 (10 + 1 + 2 + 3)
int reducedParams = Stream.of(1, 2, 3) .reduce(10, (a, b) -> a + b, (a, b) -> { log.info("combiner was called"); return a + b; });The result will be the same as in the previous example (16) and there will be no login which means, that combiner wasn't called. To make a combiner work, a stream should be parallel:
int reducedParallel = Arrays.asList(1, 2, 3).parallelStream() .reduce(10, (a, b) -> a + b, (a, b) -> { log.info("combiner was called"); return a + b; });The result here is different (36) and the combiner was called twice. Here the reduction works by the following algorithm: accumulator ran three times by adding every element of the stream to identity to every element of the stream. These actions are being done in parallel. As a result, they have (10 + 1 = 11; 10 + 2 = 12; 10 + 3 = 13;). Now combiner can merge these three results. It needs two iterations for that (12 + 13 = 25; 25 + 11 = 36).
7.2. The collect() Method
Reduction of a stream can also be executed by another terminal operation – the collect() method. It accepts an argument of the type Collector, which specifies the mechanism of reduction. There are already created predefined collectors for most common operations. They can be accessed with the help of the Collectors type.
In this section we will use the following List as a source for all streams:
List productList = Arrays.asList(new Product(23, "potatoes"), new Product(14, "orange"), new Product(13, "lemon"), new Product(23, "bread"), new Product(13, "sugar"));Converting a stream to the Collection (Collection, List or Set):
List collectorCollection = productList.stream().map(Product::getName).collect(Collectors.toList());Reducing to String:
String listToString = productList.stream().map(Product::getName) .collect(Collectors.joining(", ", "[", "]"));The joiner() method can have from one to three parameters (delimiter, prefix, suffix). The handiest thing about using joiner() – developer doesn't need to check if the stream reaches its end to apply the suffix and not to apply a delimiter. Collector will take care of that.
Processing the average value of all numeric elements of the stream:
double averagePrice = productList.stream() .collect(Collectors.averagingInt(Product::getPrice));Processing the sum of all numeric elements of the stream:
int summingPrice = productList.stream() .collect(Collectors.summingInt(Product::getPrice));Methods averagingXX(), summingXX() and summarizingXX() can work as with primitives (int, long, double) as with their wrapper classes (Integer, Long, Double). One more powerful feature of these methods is providing the mapping. So, developer doesn't need to use an additional map() operation before the collect() method.
Collecting statistical information about stream’s elements:
IntSummaryStatistics statistics = productList.stream() .collect(Collectors.summarizingInt(Product::getPrice));By using the resulting instance of type IntSummaryStatistics developer can create a statistical report by applying toString() method. The result will be a String common to this one “IntSummaryStatistics{count=5, sum=86, min=13, average=17,200000, max=23}”.
It is also easy to extract from this object separate values for count, sum, min, average by applying methods getCount(), getSum(), getMin(), getAverage(), getMax(). All these values can be extracted from a single pipeline.
Grouping of stream’s elements according to the specified function:
Map
collectorMapOfLists = productList.stream() .collect(Collectors.groupingBy(Product::getPrice));
In the example above the stream was reduced to the Map which groups all products by their price.
Dividing stream’s elements into groups according to some predicate:
Map
mapPartioned = productList.stream() .collect(Collectors.partitioningBy(element -> element.getPrice() > 15));
Pushing the collector to perform additional transformation:
Set unmodifiableSet = productList.stream() .collect(Collectors.collectingAndThen(Collectors.toSet(), Collections::unmodifiableSet));In this particular case, the collector has converted a stream to a Set and then created the unmodifiable Set out of it.
Custom collector:
If for some reason, a custom collector should be created, the most easier and the less verbose way of doing so – is to use the method of() of the type Collector.
Collector
toLinkedList = Collector.of(LinkedList::new, LinkedList::add, (first, second) -> { first.addAll(second); return first; }); LinkedList linkedListOfPersons = productList.stream().collect(toLinkedList);
In this example, an instance of the Collector got reduced to the LinkedList.
Parallel Streams
Before Java 8, parallelization was complex. Emerging of the ExecutorService and the ForkJoin simplified developer’s life a little bit, but they still should keep in mind how to create a specific executor, how to run it and so on. Java 8 introduced a way of accomplishing parallelism in a functional style.
The API allows creating parallel streams, which perform operations in a parallel mode. When the source of a stream is a Collection or an array it can be achieved with the help of the parallelStream() method:
Stream streamOfCollection = productList.parallelStream(); boolean isParallel = streamOfCollection.isParallel(); boolean bigPrice = streamOfCollection .map(product -> product.getPrice() * 12) .anyMatch(price -> price > 200);If the source of stream is something different than a Collection or an array, the parallel() method should be used:
IntStream intStreamParallel = IntStream.range(1, 150).parallel(); boolean isParallel = intStreamParallel.isParallel();Under the hood, Stream API automatically uses the ForkJoin framework to execute operations in parallel. By default, the common thread pool will be used and there is no way (at least for now) to assign some custom thread pool to it. This can be overcome by using a custom set of parallel collectors.
When using streams in parallel mode, avoid blocking operations and use parallel mode when tasks need the similar amount of time to execute (if one task lasts much longer than the other, it can slow down the complete app’s workflow).
The stream in parallel mode can be converted back to the sequential mode by using the sequential() method:
IntStream intStreamSequential = intStreamParallel.sequential(); boolean isParallel = intStreamSequential.isParallel();Conclusions
API на Stream е мощен, но лесен за разбиране набор от инструменти за обработка на последователност от елементи. Позволява ни да намалим огромно количество шаблонни кодове, да създадем по-четливи програми и да подобрим производителността на приложението, когато се използва правилно.
В по-голямата част от примерните кодове, показани в тази статия, потоците не са използвани (не сме приложили метода close () или терминална операция). В истинско приложение не оставяйте инстанцирани потоци неконсумирани, тъй като това ще доведе до изтичане на памет.
Пълните примерни кодове, които придружават статията, са достъпни в GitHub.