iterator - Java 8 迭代器流到迭代器会导致对 hasNext() 的冗余调用

Question

在以下场景中，我注意到了一些奇怪的行为:

迭代器 -> 流 -> map() -> iterator() -> 迭代

原始迭代器的 hasNext() 在已经返回 false 后被额外调用一次。

这是正常的吗？

package com.test.iterators;

import java.util.Iterator;
import java.util.Spliterators;
import java.util.stream.Stream;
import java.util.stream.StreamSupport;

public class TestIterator {

    private static int counter = 2;

    public static void main(String[] args) {

        class AdapterIterator implements Iterator<Integer> {
            boolean active = true;

            @Override
            public boolean hasNext() {
                System.out.println("hasNext() called");

                if (!active) {
                    System.out.println("Ignoring duplicate call to hasNext!!!!");
                    return false;
                }

                boolean hasNext = counter >= 0;
                System.out.println("actually has next:" + active);

                if (!hasNext) {
                    active = false;
                }

                return hasNext;
            }

            @Override
            public Integer next() {
                System.out.println("next() called");
                return counter--;
            }
        }

        Stream<Integer> stream = StreamSupport.stream(Spliterators.spliteratorUnknownSize(new AdapterIterator(), 0), false);
        stream.map(num -> num + 1).iterator().forEachRemaining(num -> {
            System.out.println(num);
        });
    }
}

如果我删除 map() 或用类似 count() 或 collect() 之类的东西替换最终的 itearator() ，它就可以在没有冗余调用的情况下工作。

输出

hasNext() called
actually has next:true
next() called
3
hasNext() called
actually has next:true
next() called
2
hasNext() called
actually has next:true
next() called
1
hasNext() called
actually has next:true
hasNext() called
Ignoring duplicate call to hasNext!!!!

Answer 1

是的，这是正常的。冗余调用发生在 StreamSpliterators.AbstractWrappingSpliterator.fillBuffer() ，从 hasNext() 调用stream.map(num -> num + 1).iterator() 返回的迭代器的方法.来自 JDK 8 源代码:

/**
 * If the buffer is empty, push elements into the sink chain until
 * the source is empty or cancellation is requested.
 * @return whether there are elements to consume from the buffer
 */
private boolean fillBuffer() {
    while (buffer.count() == 0) {
        if (bufferSink.cancellationRequested() || !pusher.getAsBoolean()) {
            if (finished)
                return false;
            else {
                bufferSink.end(); // might trigger more elements
                finished = true;
            }
        }
    }
    return true;
}

调用 pusher.getAsBoolean()电话 hasNext()上原 AdapterIterator实例。如果为真，它将下一个元素添加到 bufferSink并返回真，否则返回假。当原始迭代器用完项目并返回 false 时，此方法调用 bufferSink.end()并重试填充缓冲区，这导致了多余的 hasNext()称呼。

在这种情况下， bufferSink.end()没有效果，第二次尝试填充缓冲区是不必要的，但正如源评论所解释的那样，它在另一种情况下“可能会触发更多元素”。这只是一个深埋在 Java 8 流复杂内部工作中的实现细节。