Stream是Java 8 提供的高效操作集合类(Collection)数据的API。
1. 从Iterator到Stream
有一个字符串的list,要统计其中长度大于7的字符串的数量,用迭代来实现:
List<String> wordList = Arrays.asList("regular", "expression", "specified", "as", "a", "string", "must");
int countByIterator = 0;
for (String word: wordList) {
if (word.length() > 7) {
countByIterator++;
}
}
用Stream实现:
long countByStream= wordList.stream().filter(w -> w.length() > 7).count();
显然,用stream实现更简洁,不仅如此,stream很容易实现并发操作,比如:
long countByParallelStream = wordList.parallelStream().filter(w -> w.length() > 7).count();
stream遵循的原则是:告诉我做什么,不用管我怎么做。比如上例:告诉stream通过多线程统计字符串长度,至于以什么顺序、在哪个线程中执行,由stream来负责;而在迭代实现中,由于计算的方式已确定,很难优化了。
Stream和Collection的区别主要有:
- stream本身并不存储数据,数据是存储在对应的collection里,或者在需要的时候才生成的;
- stream不会修改数据源,总是返回新的stream;
- stream的操作是懒执行(lazy)的:仅当最终的结果需要的时候才会执行,比如上面的例子中,结果仅需要前3个长度大于7的字符串,那么在找到前3个长度符合要求的字符串后,
filter()
将停止执行;
使用stream的步骤如下:
- 创建stream;
- 通过一个或多个中间操作(intermediate operations)将初始stream转换为另一个stream;
- 通过中止操作(terminal operation)获取结果;该操作触发之前的懒操作的执行,中止操作后,该stream关闭,不能再使用了;
在上面的例子中, wordList.stream()
和 wordList.parallelStream()
是创建stream, filter()
是中间操作,过滤后生成一个新的stream, count()
是中止操作,获取结果。
2. 创建Stream的方式
- 从array或list创建stream:
Stream<Integer> integerStream = Stream.of(10, 20, 30, 40);
String[] cityArr = {"Beijing", "Shanghai", "Chengdu"};
Stream<String> cityStream = Stream.of(cityArr);
Stream<String> nameStream = Arrays.asList("Daniel", "Peter", "Kevin").stream();
Stream<String> cityStream2 = Arrays.stream(cityArr, 0, 1);
Stream<String> emptyStream = Stream.empty();
- 通过
generate
和iterate
创建无穷stream:
Stream<String> echos = Stream.generate(() -> "echo");
Stream<Integer> integers = Stream.iterate(0, num -> num + 1);
- 通过其它API创建stream:
Stream<String> lines = Files.lines(Paths.get("test.txt"))
String content = "AXDBDGXC";
Stream<String> contentStream = Pattern.compile("[ABC]{1,3}").splitAsStream(content);
3. Stream转换
filter()
用于过滤,即使原stream中满足条件的元素构成新的stream:
List<String> langList = Arrays.asList("Java", "Python", "Swift", "HTML");
Stream<String> filterStream = langList.stream().filter(lang -> lang.equalsIgnoreCase("java"));
map()
用于映射,遍历原stream中的元素,转换后构成新的stream:
List<String> langList = Arrays.asList("Java", "Python", "Swift", "HTML");
Stream<String> mapStream = langList.stream().map(String::toUpperCase);
flatMap()
用于将[["ABC", "DEF"], ["FGH", "IJK"]]
的形式转换为["ABC", "DEF", "FGH", "IJK"]
:
Stream<String> cityStream = Stream.of("Beijing", "Shanghai", "Shenzhen");
// [['B', 'e', 'i', 'j', 'i', 'n', 'g'], ['S', 'h', 'a', 'n', 'g', 'h', 'a', 'i'], ...]
Stream<Stream<Character>> characterStream1 = cityStream.map(city -> characterStream(city));
Stream<String> cityStreamCopy = Stream.of("Beijing", "Shanghai", "Shenzhen");
// ['B', 'e', 'i', 'j', 'i', 'n', 'g', 'S', 'h', 'a', 'n', 'g', 'h', 'a', 'i', ...]
Stream<Character> characterStreamCopy = cityStreamCopy.flatMap(city -> characterStream(city));
其中, characterStream()
返回有参数字符串的字符构成的Stream ;
limit()
表示限制stream中元素的数量,skip()
表示跳过stream中前几个元素,concat
表示将多个stream连接起来,peek()
主要用于debug时查看stream中元素的值:
Stream<Integer> limitStream = Stream.of(18, 20, 12, 35, 89).sorted().limit(3);
Stream<Integer> skipStream = Stream.of(18, 20, 12, 35, 89).sorted(Comparator.reverseOrder()).skip(1);
Stream<Integer> concatStream = Stream.concat(Stream.of(1, 2, 3), Stream.of(4, 5, 6));
concatStream.peek(i -> System.out.println(i)).count();
peek()
是 intermediate operation ,所以后面需要一个 terminal operation ,如 count()
才能在输出中看到结果;
- 有状态的(stateful)转换,即元素之间有依赖关系,如
distinct()
返回由唯一元素构成的stream,sorted()
返回排序后的stream:
Stream<String> distinctStream = Stream.of("Beijing", "Tianjin", "Beijing").distinct();
Stream<String> sortedStream = Stream.of("Beijing", "Shanghai", "Chengdu").sorted(Comparator.comparing
(String::length).reversed());
4. Stream reduction
reduction
就是从stream中取出结果,是 terminal operation
,因此经过 reduction
后的stream不能再使用了。
4.1 Optional
Optional 表示或者有一个T类型的对象,或者没有值;
- 创建Optional对象
直接通过Optional的类方法: of()
/ empty()
/ ofNullable()
:
Optional<Integer> intOpt = Optional.of(10);
Optional<String> emptyOpt = Optional.empty();
Optional<Double> doubleOpt = Optional.ofNullable(5.5);
- 使用Optional对象
你当然可以这么使用:
if (intOpt.isPresent()) {
intOpt.get();
}
但是,最好这么使用:
doubleOpt.orElse(0.0);
doubleOpt.orElseGet(() -> 1.0);
doubleOpt.orElseThrow(RuntimeException::new);
List<Double> doubleList = new ArrayList<>();
doubleOpt.ifPresent(doubleList::add);
map()
方法与 ifPresent()
用法相同,就是多个返回值, flatMap()
用于Optional的链式表达:
Optional<Boolean> addOk = doubleOpt.map(doubleList::add);
Optional.of(4.0).flatMap(num -> Optional.ofNullable(num * 100)).flatMap(num -> Optional.ofNullable(Math.sqrt
(num)));
4.2 简单的reduction
主要包含以下操作: findFirst()
/ findAny()
/ allMatch
/ anyMatch()
/ noneMatch
,比如:
Optional<String> firstWord = wordStream.filter(s -> s.startsWith("Y")).findFirst();
Optional<String> anyWord = wordStream.filter(s -> s.length() > 3).findAny();
wordStream.allMatch(s -> s.length() > 3);
wordStream.anyMatch(s -> s.length() > 3);
wordStream.noneMatch(s -> s.length() > 3);
4.3 reduce方法
reduce(accumulator)
:参数是一个执行双目运算的Functional Interface
,假如这个参数表示的操作为op,stream中的元素为x, y, z, …,则reduce()
执行的就是x op y op z ...
,所以要求op这个操作具有结合性(associative),即满足:(x op y) op z = x op (y op z)
,满足这个要求的操作主要有:求和、求积、求最大值、求最小值、字符串连接、集合并集和交集等。另外,该函数的返回值是Optional的:
Optional<Integer> sum1 = numStream.reduce((x, y) -> x + y);
reduce(identity, accumulator)
:可以认为第一个参数为默认值,但需要满足identity op x = x
,所以对于求和操作,identity
的值为0,对于求积操作,identity
的值为1。返回值类型是stream元素的类型:
Integer sum2 = numStream.reduce(0, Integer::sum);
5. collect结果
collect()
方法
reduce()
和 collect()
的区别是:
reduce()
的结果是一个值;collect()
可以对stream中的元素进行各种处理后,得到stream中元素的值;
Collectors
接口提供了很方便的创建 Collector
对象的工厂方法:
// collect to Collection
Stream.of("You", "may", "assume").collect(Collectors.toList());
Stream.of("You", "may", "assume").collect(Collectors.toSet());
Stream.of("You", "may", "assume").collect(Collectors.toCollection(TreeSet::new));
// join element
Stream.of("You", "may", "assume").collect(Collectors.joining());
Stream.of("You", "may", "assume").collect(Collectors.joining(", "));
// summarize element
IntSummaryStatistics summary = Stream.of("You", "may", "assume").collect(Collectors.summarizingInt(String::length));
summary.getMax();
foreach()
方法
foreach()
用于遍历stream中的元素,属于terminal operation
;-
forEachOrdered()
是按照stream中元素的顺序遍历,也就无法利用并发的优势;Stream.of(“You”, “may”, “assume”, “you”, “can”, “fly”).parallel().forEach(w -> System.out.println(w));Stream.of(“You”, “may”, “assume”, “you”, “can”, “fly”).forEachOrdered(w -> System.out.println(w));
toArray()
方法
得到由stream中的元素得到的数组,默认是Object[],可以通过参数设置需要结果的类型:
Object[] words1 = Stream.of("You", "may", "assume").toArray();
String[] words2 = Stream.of("You", "may", "assume").toArray(String[]::new);
toMap()
方法
toMap
: 将stream中的元素映射为 的形式,两个参数分别用于生成对应的key和value的值。比如有一个字符串stream,将首字母作为key,字符串值作为value,得到一个map:
Stream<String> introStream = Stream.of("Get started with UICollectionView and the photo library".split(" "));
Map<String, String> introMap = introStream.collect(Collectors.toMap(s -> s.substring(0, 1), s -> s));
如果一个key对应多个value,则会抛出异常,需要使用第三个参数设置如何处理冲突,比如仅使用原来的value、使用新的value,或者合并:
Stream<String> introStream = Stream.of("Get started with UICollectionView and the photo library".split(" "));
Map<Integer, String> introMap2 = introStream.collect(Collectors.toMap(s -> s.length(), s -> s, (existingValue, newValue) -> existingValue));
如果value是一个集合,即将key对应的所有value放到一个集合中,则需要使用第三个参数,将多个value合并:
Stream<String> introStream3 = Stream.of("Get started with UICollectionView and the photo library".split(" "));
Map<Integer, Set<String>> introMap3 = introStream3.collect(Collectors.toMap(s -> s.length(),
s -> Collections.singleton(s), (existingValue, newValue) -> {
HashSet<String> set = new HashSet<>(existingValue);
set.addAll(newValue);
return set;
}));
introMap3.forEach((k, v) -> System.out.println(k + ": " + v));
如果value是对象自身,则使用 Function.identity()
,如:
Map<Integer, Person> idToPerson = people.collect(Collectors.toMap(Person::getId, Function.identity()));
toMap()
默认返回的是HashMap,如果需要其它类型的map,比如TreeMap,则可以在第四个参数指定构造方法:
Map<Integer, String> introMap2 = introStream.collect(Collectors.toMap(s -> s.length(), s -> s, (existingValue, newValue) -> existingValue, TreeMap::new));
6. Grouping和Partitioning
groupingBy()
表示根据某一个字段或条件进行分组,返回一个Map,其中key为分组的字段或条件,value默认为list,groupingByConcurrent()
是其并发版本:
Map<String, List<Locale>> countryToLocaleList = Stream.of(Locale.getAvailableLocales())
.collect(Collectors.groupingBy(l -> l.getDisplayCountry()));
- 如果
groupingBy()
分组的依据是一个bool条件,则key的值为true/false,此时与partitioningBy()
等价,且partitioningBy()
的效率更高:
// predicate
Map<Boolean, List<Locale>> englishAndOtherLocales = Stream.of(Locale.getAvailableLocales())
.collect(Collectors.groupingBy(l -> l.getDisplayLanguage().equalsIgnoreCase("English")));
// partitioningBy
Map<Boolean, List<Locale>> englishAndOtherLocales2 = Stream.of(Locale.getAvailableLocales())
.collect(Collectors.partitioningBy(l -> l.getDisplayLanguage().equalsIgnoreCase("English")));
groupingBy()
提供第二个参数,表示downstream
,即对分组后的value作进一步的处理
- 返回set,而不是list:
Map<String, Set<Locale>> countryToLocaleSet = Stream.of(Locale.getAvailableLocales()).collect(Collectors.groupingBy(l -> l.getDisplayCountry(), Collectors.toSet()));
- 返回value集合中元素的数量:
Map<String, Long> countryToLocaleCounts = Stream.of(Locale.getAvailableLocales())
.collect(Collectors.groupingBy(l -> l.getDisplayCountry(), Collectors.counting()));
- 对value集合中的元素求和:
Map<String, Integer> cityToPopulationSum = Stream.of(cities)
.collect(Collectors.groupingBy(City::getName, Collectors.summingInt(City::getPopulation)));
- 对value的某一个字段求最大值,注意value是Optional的:
Map<String, Optional<City>> cityToPopulationMax = Stream.of(cities)
.collect(Collectors.groupingBy(City::getName, Collectors.maxBy(Comparator.comparing(City::getPopulation))));
- 使用mapping对value的字段进行map处理:
Map<String, Optional<String>> stateToNameMax = Stream.of(cities)
.collect(Collectors.groupingBy(City::getState, Collectors.mapping(City::getName, Collectors.maxBy
(Comparator.comparing(String::length)))));
Map<String, Set<String>> stateToNameSet = Stream.of(cities)
.collect(Collectors.groupingBy(City::getState, Collectors.mapping(City::getName, Collectors.toSet())));
- 通过
summarizingXXX
获取统计结果:
Map<String, IntSummaryStatistics> stateToPopulationSummary = Stream.of(cities)
.collect(Collectors.groupingBy(City::getState, Collectors.summarizingInt(City::getPopulation)));
reducing()
可以对结果作更复杂的处理,但是reducing()
却并不常用:
Map<String, String> stateToNameJoining = Stream.of(cities)
.collect(Collectors.groupingBy(City::getState, Collectors.reducing("", City::getName,
(s, t) -> s.length() == 0 ? t : s + ", " + t)));
比如上例可以通过mapping达到同样的效果:
Map<String, String> stateToNameJoining2 = Stream.of(cities)
.collect(Collectors.groupingBy(City::getState, Collectors.mapping(City::getName, Collectors.joining(", ")
)));
7. Primitive Stream
Stream<Integer>
对应的Primitive Stream就是 IntStream
,类似的还有 DoubleStream
和 LongStream
。
- Primitive Stream的构造:
of()
,range()
,rangeClosed()
,Arrays.stream()
:
IntStream intStream = IntStream.of(10, 20, 30);
IntStream zeroToNintyNine = IntStream.range(0, 100);
IntStream zeroToHundred = IntStream.rangeClosed(0, 100);
double[] nums = {10.0, 20.0, 30.0};
DoubleStream doubleStream = Arrays.stream(nums, 0, 3);
- Object Stream与Primitive Stream之间的相互转换,通过
mapToXXX()
和boxed()
:
// map to
Stream<String> cityStream = Stream.of("Beijing", "Tianjin", "Chengdu");
IntStream lengthStream = cityStream.mapToInt(String::length);
// box
Stream<Integer> oneToNine = IntStream.range(0, 10).boxed();
- 与Object Stream相比,Primitive Stream的特点:
toArray()
方法返回的是对应的Primitive类型:
int[] intArr = intStream.toArray();
- 自带统计类型的方法,如:
max()
,average()
,summaryStatistics()
:
OptionalInt maxNum = intStream.max();
IntSummaryStatistics intSummary = intStream.summaryStatistics();
8. Parallel Stream
- Stream支持并发操作,但需要满足以下几点:
- 构造一个paralle stream,默认构造的stream是顺序执行的,调用
paralle()
构造并行的stream:
IntStream scoreStream = IntStream.rangeClosed(10, 30).parallel();
- 要执行的操作必须是可并行执行的,即并行执行的结果和顺序执行的结果是一致的,而且必须保证stream中执行的操作是线程安全的:
int[] wordLength = new int[12];
Stream.of("It", "is", "your", "responsibility").parallel().forEach(s -> {
if (s.length() < 12) wordLength[s.length()]++;
});
这段程序的问题在于,多线程访问共享数组 wordLength
,是非线程安全的。解决的思路有:1)构造AtomicInteger数组;2)使用 groupingBy()
根据length统计;
- 可以通过并行提高效率的常见场景
- 使stream无序:对于
distinct()
和limit()
等方法,如果不关心顺序,则可以使用并行:
LongStream.rangeClosed(5, 10).unordered().parallel().limit(3);
IntStream.of(14, 15, 15, 14, 12, 81).unordered().parallel().distinct();
- 在
groupingBy()
的操作中,map的合并操作是比较重的,可以通过groupingByConcurrent()
来并行处理,不过前提是parallel stream:
Stream.of(cities).parallel().collect(Collectors.groupingByConcurrent(City::getState));
- 在执行stream操作时不能修改stream对应的collection
stream本身是不存储数据的,数据保存在对应的collection中,所以在执行stream操作的同时修改对应的collection,结果是未定义的:
// ok
Stream<String> wordStream = wordList.stream();
wordList.add("number");
wordStream.distinct().count();
// ConcurrentModificationException
Stream<String> wordStream = wordList.stream();
wordStream.forEach(s -> { if (s.length() >= 6) wordList.remove(s);});
9. Functional Interface
仅包含一个抽象方法的interface被成为 Functional Interface
,比如: Predicate
, Function
, Consumer
等。此时我们一般传入一个lambda表达式或 Method Reference
。
常见的 Functional Interface
有:
Functional Interface Parameter Return Type Description Types
Supplier<T> None T Supplies a value of type T
Consumer<T> T void Consumes a value of type T
BiConsumer<T, U> T,U void Consumes values of types T and U
Predicate<T> T boolean A Boolean-valued function
ToIntFunction<T> T int An int-, long-, or double-valued function
ToLongFunction<T> T long
ToDoubleFunction<T> T double
IntFunction<R> int R A function with argument of type int, long, or double
LongFunction<R> long
DoubleFunction<R> double
Function<T, R> T R A function with argument of type T
BiFunction<T, U, R> T,U R A function with arguments of types T and U
UnaryOperator<T> T T A unary operator on the type T
BinaryOperator<T> T,T T A binary operator on the type T