我们在上一节讨论了scalaz Future，我们说它是一个不完善的类型，最起码没有完整的异常处理机制，只能用在构建类库之类的内部环境。如果scalaz在Future类定义中增加异常处理工具的话，用户就会经常遇到Future[Throwable\/A]这样的类型，那么在进行Monadic编程时就必须使用Monad Transformer来匹配类型，程序就会变得不必要的复杂。scalaz的解决方案就是把Future[Throwable\/A]包嵌在Task类里，然后把所有Future都统一升格成Task。Task是个Monad, 这样，我们就可以统一方便地用Task来进行多线程函数式编程了。我们先看看Task的定义：scalaz.concurrent/Task.scala
class Task[+A](val get: Future[Throwable \/ A]) { def flatMap[B](f: A => Task[B]): Task[B] = new Task(get flatMap { case -\/(e) => Future.now(-\/(e)) case \/-(a) => Task.Try(f(a)) match { case e @ -\/(_) => Future.now(e) case \/-(task) => task.get } }) def map[B](f: A => B): Task[B] = new Task(get map { _ flatMap {a => Task.Try(f(a))} })... Task实现了flatMap，所以是个Monad，我们可以在for-comprehension中使用Task。
Task的构建方式与Future一样：
1 val tnow = Task.now { println("run now ..."); 3+4 }2 //> run now ...3 //| tnow : scalaz.concurrent.Task[Int] = scalaz.concurrent.Task@1a968a594 val tdelay = Task.delay { println("run delay ..."); 3+4 }5 //> tdelay : scalaz.concurrent.Task[Int] = scalaz.concurrent.Task@13deb50e6 val tapply = Task { println("run apply ..."); 3+4 }7 //> tapply : scalaz.concurrent.Task[Int] = scalaz.concurrent.Task@59494225 同样，now函数是即时运算的。它就是一个lifter，能把一个普通运算直接升格为Task。
针对Task有几种运算方法：
1 tnow.unsafePerformSync //> res0: Int = 7 2 tdelay.unsafePerformSync //> run delay ... 3 //| res1: Int = 7 4 tnow.unsafePerformAsync { 5 case \/-(a) => println(s"the result is: $a") 6 case -\/(e) => println(e.getMessage) 7 } //> the result is: 7 8 tdelay.unsafePerformAsync { 9 case \/-(a) => println(s"the result is: $a")10 case -\/(e) => println(e.getMessage)11 } //> run delay ...12 //| the result is: 713 tapply.unsafePerformAsync {14 case \/-(a) => println(s"the result is: $a")15 case -\/(e) => println(e.getMessage)16 }17 Thread.sleep(1000) //> run apply ...18 //| the result is: 7 从上面的例子我们可以得出：tnow已经完成了运算，因为运算结果没有"run now ..."提示了。tdelay和tapply都是存在trampoline结构里的。但tapply存在更深一层的结构里，所以我们必须拖时间来等待tapply的运算结果。tdelay存放在Future.Suspend结构里，而tapply是存放在Future.Async结构里的，所以tdelay是一种延迟运算，而tapply就是异步运算了：
1 def delay[A](a: => A): Task[A] = suspend(now(a)) 2 def suspend[A](a: => Task[A]): Task[A] = new Task(Future.suspend( 3 Try(a.get) match { 4 case -\/(e) => Future.now(-\/(e)) 5 case \/-(f) => f 6 })) 7 //Future.suspend: 8 def suspend[A](f: => Future[A]): Future[A] = Suspend(() => f) 9 10 def apply[A](a: => A)(implicit pool: ExecutorService = Strategy.DefaultExecutorService): Task[A] =11 new Task(Future(Try(a))(pool))12 //Future.apply13 def apply[A](a: => A)(implicit pool: ExecutorService = Strategy.DefaultExecutorService): Future[A] = Async { cb =>14 pool.submit { new Callable[Unit] { def call = cb(a).run }}15 } 好了，我们再看看Task是怎样处理异常情况的：
1 def eval(value: => Int) = Task { Thread.sleep(1000); value } 2 //> eval: (value: => Int)scalaz.concurrent.Task[Int] 3 eval( 3 * 7 ).onFinish { 4 case None => Task { println("finished calculation successfully.") } 5 case Some(e) => Task { println(s"caught error [${e.getMessage}]") } 6 }.unsafePerformSyncAttempt match { 7 case -\/(e) => println(s"calculation error [${e.getMessage}]") 8 case \/-(a) => println(s"the result is: $a") 9 } //> finished calculation successfully.10 //| the result is: 2111 // 异常处理12 eval( 3 * 7 / 0 ).onFinish {13 case None => Task { println("finished calculation successfully.") }14 case Some(e) => Task { println(s"caught error [${e.getMessage}]") }15 }.unsafePerformAsync {16 case -\/(e) => println(s"calculation error [${e.getMessage}]")17 case \/-(a) => println(s"the result is: $a")18 }19 Thread.sleep(2000) //> caught error [/ by zero]20 //| calculation error [/ by zero] 精准异常处理例子：
1 import java.util.concurrent._2 val timedTask = Task {Thread.sleep(2000); 3+4} 3 //> timedTask : scalaz.concurrent.Task[Int] = scalaz.concurrent.Task@3d921e204 timedTask.timed(1 second).handleWith {5 case e: TimeoutException => Task { println(s"calculation exceeding time limit: ${e.getMessage}") }6 }.unsafePerformSync //> calculation exceeding time limit: Timed out after 1000 milliseconds7 //| res2: AnyVal{def getClass(): Class[_ >: Int with Unit <: AnyVal]} = () 再看一些多线程编程例子：
1 val tasks = (1 |-> 5).map(n => Task{ Thread.sleep(100); n }) 2 //> tasks : List[scalaz.concurrent.Task[Int]] = List(scalaz.concurrent.Task@61 3 //| 8b19ad, scalaz.concurrent.Task@2d3379b4, scalaz.concurrent.Task@30c15d8b, s 4 //| calaz.concurrent.Task@5e0e82ae, scalaz.concurrent.Task@6771beb3) 5 //并行运算list of tasks 6 Task.gatherUnordered(tasks).unsafePerformSync //> res3: List[Int] = List(1, 2, 3, 4, 5) 7 val sb = new StringBuffer //> sb : StringBuffer = 8 val t1 = Task.fork { Thread.sleep(100); sb.append("a"); Task.now("a")} 9 //> t1 : scalaz.concurrent.Task[String] = scalaz.concurrent.Task@6200f9cb10 val t2 = Task.fork { Thread.sleep(800); sb.append("b"); Task.now("b")}11 //> t2 : scalaz.concurrent.Task[String] = scalaz.concurrent.Task@2002fc1d12 val t3 = Task.fork { Thread.sleep(200); sb.append("c"); Task.now("c")}13 //> t3 : scalaz.concurrent.Task[String] = scalaz.concurrent.Task@69453e3714 val t4 = Task.fork { Thread.sleep(100); sb.append("d"); Task.now("d")}15 //> t4 : scalaz.concurrent.Task[String] = scalaz.concurrent.Task@6f4a47c716 val t5 = Task.fork { Thread.sleep(400); sb.append("e"); Task.now("e")}17 //> t5 : scalaz.concurrent.Task[String] = scalaz.concurrent.Task@ae1354418 val t6 = Task.fork { Thread.sleep(100); sb.append("f"); Task.now("f")}19 //> t6 : scalaz.concurrent.Task[String] = scalaz.concurrent.Task@3d34d21120 val r = Nondeterminism[Task].nmap6(t1,t2,t3,t4,t5,t6)(List(_,_,_,_,_,_))21 //> r : scalaz.concurrent.Task[List[String]] = scalaz.concurrent.Task@394df057 22 r.unsafePerformSync //> res4: List[String] = List(a, b, c, d, e, f) 看个耗时算法的并行运算吧：
1 def seqFib(n: Int): Task[Int] = n match { 2 case 0 | 1 => Task now 1 3 case n => { 4 for { 5 x <- seqFib(n-1) 6 y <- seqFib(n-2) 7 } yield x + y 8 } 9 } //> seqFib: (n: Int)scalaz.concurrent.Task[Int]10 //并行算法11 def parFib(n: Int): Task[Int] = n match {12 case 0 | 1 => Task now 113 case n => {14 val ND = Nondeterminism[Task]15 for {16 pair <- ND.both(parFib(n-1), parFib(n-2))17 (x,y) = pair18 } yield x + y19 }20 } //> parFib: (n: Int)scalaz.concurrent.Task[Int]21 def msFib(n: Int, fib: Int => Task[Int]) = for {22 b <- Task now { System.currentTimeMillis() }23 a <- fib(n)24 e <- Task now { System.currentTimeMillis() }25 } yield (a, (e-b)) //> msFib: (n: Int, fib: Int => scalaz.concurrent.Task[Int])scalaz.concurrent.T26 //| ask[(Int, Long)]27 28 msFib(20, parFib).unsafePerformSync //> res3: (Int, Long) = (10946,373)29 msFib(20, seqFib).unsafePerformSync //> res4: (Int, Long) = (10946,17) 哎呀！奇怪了，为什么并行算法要慢很多呢？这个问题暂且放一放，以后再研究。当然，如果有读者能给出个解释就太感激了。
Task的线程池是如何分配的呢？看看Task.apply和Task.fork：/** Create a `Task` that will evaluate `a` using the given `ExecutorService`. */ def apply[A](a: => A)(implicit pool: ExecutorService = Strategy.DefaultExecutorService): Task[A] = new Task(Future(Try(a))(pool))def fork[A](a: => Task[A])(implicit pool: ExecutorService = Strategy.DefaultExecutorService): Task[A] = apply(a).join//Future.apply/** Create a `Future` that will evaluate `a` using the given `ExecutorService`. */ def apply[A](a: => A)(implicit pool: ExecutorService = Strategy.DefaultExecutorService): Future[A] = Async { cb => pool.submit { new Callable[Unit] { def call = cb(a).run }} 这两个函数都包括了一个隐式参数implicit pool: ExecutorService。默认值是Strategy.DefultExecutorService。我们可以这样指定线程池：
1 Task {longProcess}(myExecutorService)2 Task.fork { Task {longProcess} }(myExecutorService) 下面是一个动态指定线程池的例子：
1 import java.util.concurrent.{ExecutorService,Executors} 2 type Delegated[A] = Kleisli[Task,ExecutorService,A] 3 def delegate: Delegated[ExecutorService] = Kleisli(e => Task.now(e)) 4 //> delegate: => demo.ws.task.Delegated[java.util.concurrent.ExecutorService] 5 implicit def delegateTaskToPool[A](ta: Task[A]): Delegated[A] = Kleisli(x => ta) 6 //> delegateTaskToPool: [A](ta: scalaz.concurrent.Task[A])demo.ws.task.Delegated[A] 7 val tPrg = for { 8 p <- delegate 9 b <- Task("x")(p)10 c <- Task("y")(p)11 } yield c //> tPrg : scalaz.Kleisli[scalaz.concurrent.Task,java.util.concurrent.Executor12 //| Service,String] = Kleisli( )13 tPrg.run(Executors.newFixedThreadPool(3)).unsafePerformSync14 //> res3: String = y 当然，Task和scala Future之间是可以相互转换的：
1 import scala.concurrent.{Future => sFuture} 2 import scala.util.{Success,Failure} 3 import scala.concurrent.ExecutionContext 4 def futureToTask[A](fut: sFuture[A])(implicit ec: ExecutionContext): Task[A] = 5 Task.async { 6 cb => 7 fut.onComplete { 8 case Success(a) => cb(a.right) 9 case Failure(e) => cb(e.left)10 }11 }12 def taskToFuture[A](ta: Task[A]): sFuture[A] = {13 val prom = scala.concurrent.Promise[A]14 ta.unsafePerformAsync {15 case -\/(e) => prom.failure(e)16 case \/-(a) => prom.success(a)17 }18 prom.future19 } 与Future不同的是：Task增加了异常处理机制。