Scala State monad - 结合不同的状态类型

Question

我正在围绕着 State monad。琐碎的例子很容易理解。我现在正在转向域对象是复合的现实世界案例。例如，使用以下域对象（它们没有多大意义，只是纯粹的例子）：

case class Master(workers: Map[String, Worker])
case class Worker(elapsed: Long, result: Vector[String])
case class Message(workerId: String, work: String, elapsed: Long)

将Worker 视为S 类型在State[S, +A] monad 中，编写一些像这样的组合子很容易：

type WorkerState[+A] = State[Worker, A]
def update(message: Message): WorkerState[Unit] = State.modify { w =>
    w.copy(elapsed = w.elapsed + message.elapsed,
           result = w.result :+ message.work)
}
def getWork: WorkerState[Vector[String]] = State { w => (w.result, w) }
def getElapsed: WorkerState[Long] = State { w => (w.elapsed, w) }
def updateAndGetElapsed(message: Message): WorkerState[Long] = for {
    _ <- update(message)
    elapsed <- getElapsed
} yield elapsed
// etc.

将这些与Master 状态组合器结合起来的惯用方式是什么？例如

type MasterState[+A] = State[Master, A]
def updateAndGetElapsedTime(message: Message): MasterState[Option[Long]]

我可以这样实现：

def updateAndGetElapsedTime(message: Message): MasterState[Option[Long]] =   
    State { m =>
        m.workers.get(message.workerId) match {
            case None => (None, m)
            case Some(w) =>
                val (t, newW) = updateAndGetElapsed(message).run(w)
                (Some(t), m.copy(m.workers.updated(message.workerId, newW))
        }
    }

我不喜欢的是我必须在最后一个转换器中手动运行 State monad。我的现实世界的例子涉及更多。使用这种方法，它很快就会变得混乱。

有没有更惯用的方式来运行这种增量更新？

Answer 1

通过结合镜头和状态单子可以很好地做到这一点。首先是设置（我已经稍微编辑了你的设置以使其与 Scalaz 7.1 一起编译）：

case class Master(workers: Map[String, Worker])
case class Worker(elapsed: Long, result: Vector[String])
case class Message(workerId: String, work: String, elapsed: Long)

import scalaz._, Scalaz._

type WorkerState[A] = State[Worker, A]

def update(message: Message): WorkerState[Unit] = State.modify { w =>
  w.copy(
    elapsed = w.elapsed + message.elapsed,
    result = w.result :+ message.work
  )
}

def getWork: WorkerState[Vector[String]] = State.gets(_.result)
def getElapsed: WorkerState[Long] = State.gets(_.elapsed)
def updateAndGetElapsed(message: Message): WorkerState[Long] = for {
  _ <- update(message)
  elapsed <- getElapsed
} yield elapsed

现在有几个通用镜头可以让我们看到Master的内部：

val workersLens: Lens[Master, Map[String, Worker]] = Lens.lensu(
  (m, ws) => m.copy(workers = ws),
  _.workers
)

def workerLens(workerId: String): PLens[Master, Worker] =
  workersLens.partial andThen PLens.mapVPLens(workerId)

然后我们基本上就完成了：

def updateAndGetElapsedTime(message: Message): State[Master, Option[Long]] =
  workerLens(message.workerId) %%= updateAndGetElapsed(message)

这里的%%= 只是告诉我们一旦我们通过镜头放大到适当的工作人员后要执行什么状态操作。