Difference between "parallelStream()" and "asSequence().asStream().parallel()"-CodePudding

Using kotlin 1.6.10 it seems that parallelStream() and asSequence().asStream().parallel() are not behaving the same. The latter doesn't seem to parallelize. Here is an example code snippet, demonstrating the issue:

import java.util.stream.Collectors.toList
import kotlin.streams.asStream
import java.util.Calendar

fun main() {
    println("Available processors: ${Runtime.getRuntime().availableProcessors()}")

    val numbers = (10 downTo 0).toList()
    
    println("Using parallelStream:")
    val start1=System.currentTimeMillis()
    numbers.parallelStream().map { println("${it} - ${Thread.currentThread().getId()}"); Thread.sleep(100); it }.collect(toList<Int>())
    println("Execution time: ${System.currentTimeMillis()-start1}")
    
    println("Using asSequence().asStream().parallel():")
    val start2=System.currentTimeMillis()
    numbers.asSequence().asStream().parallel().map { println("${it} - ${Thread.currentThread().getId()}"); Thread.sleep(100); it }.collect(toList<Int>())
    println("Execution time: ${System.currentTimeMillis()-start2}")
}

You can execute it in the kotlin playground: https://pl.kotl.in/z3sxKHKvL

The output is

Available processors: 2
Using parallelStream:
4 - 1
8 - 10
3 - 1
7 - 10
5 - 1
6 - 10
1 - 1
10 - 10
0 - 1
9 - 10
2 - 1
Execution time: 617
Using asSequence().asStream().parallel():
10 - 1
9 - 1
8 - 1
7 - 1
6 - 1
5 - 1
4 - 1
3 - 1
2 - 1
1 - 1
0 - 1
Execution time: 1105

I don't understand the reason for this difference in behavior.

CodePudding user response：

Calling stream() or parallelStream() directly on your List means under the hood it is able to call spliterator() on the underlying List directly and get a functional Spliterator that can break up the work in parallel.

If you convert the List to a Sequence, you no longer have direct access to the List when calling asStream() on it. Calling asStream() on the Sequence creates a Stream with an ad hoc Spliterator that doesn't have access to the size of the collection (because Sequences do not have a size property), and so it cannot break up the work for parallel execution.

Basically, asSequence() strips away size information and the built-in Spliterator implementation of List.