为什么Scala的Future.onComplete需要一个ExecutionContext

Question

能否仅运行Future.onComplete的回调函数，而不通过ExecutionContext并将Future.onComplete的回调函数计划安排在稍后，可能在另一个线程中运行？

- Yuri Geinish

Answer 1

您可以提供自己的 ExecutionContext 给 onComplete，这将在同一线程上运行代码：

val immediateContext: ExecutionContext = new ExecutionContext {
  def execute(runnable: Runnable) {
    runnable.run()
  }
  def reportFailure(cause: Throwable) {}
}

你甚至可以将其做成隐式，并且对于想要在另一个线程中执行的情况，你可以提供scala.concurrent.ExecutionContext.global或其他上下文。

以下是一个测试，它的工作原理：

val immediateContext: ExecutionContext = new ExecutionContext {
  override def reportFailure(cause: Throwable): Unit = {}
  override def execute(runnable: Runnable): Unit = {
    println("Executing")
    runnable.run()
    println("Executed")
  }
}

def testContext(): Unit = {
  println("Scheduling on an uncompleted future")
  val p = Promise[Int]()
  println("Scheduling")
  p.future.onComplete { _ => println("Completed") }(immediateContext)
  println("Scheduled")
  p.complete(Success(5))

  println()

  println("Scheduling on an already completed future")
  val p2 = Promise[Int]().complete(Success(5))
  println("Scheduling")
  p2.future.map { n =>
    println("Mapping")
    n * 2
  }(immediateContext).onComplete{
    case Success(n) => println(s"Completed: $n") 
    case _ =>
  }(immediateContext)
  println("Scheduled")

  println()

  println("Using scala.concurrent.ExecutionContext.global for comparison")
  val p3 = Promise[Int]().complete(Success(5))
  println("Scheduling")
  p3.future.onComplete {
    _ => println("Completed")
  }(concurrent.ExecutionContext.global)
  println("Scheduled")
}

运行testContext()将会输出：

Scheduling on an uncompleted future
Scheduling
Scheduled
Executing
Completed
Executed

Scheduling on an already completed future
Scheduling
Executing
Mapping
Executed
Executing
Completed: 10
Executed
Scheduled

Using scala.concurrent.ExecutionContext.global for comparison
Scheduling
Scheduled
Completed

- Kolmar

Answer 2

这是实现Scala Futures的设计决策，所有操作（map、flatMap等）都需要隐式ExecutionContext。如果您希望具有更好的线程可重用性和更少的线程上下文切换，我建议您查看Scalaz Task，这是一种略有不同的延迟计算抽象：http://timperrett.com/2014/07/20/scalaz-task-the-missing-documentation/。

- Eugene Zhulenev