很多函数式编程爱好者都把FP称为Monadic Programming,意思是用Monad进行编程。我想FP作为一种比较成熟的编程模式,应该有一套比较规范的操作模式吧。因为Free能把任何F[A]升格成Monad,所以Free的算式(AST)、算法(Interpreter)关注分离(separation of concern)模式应该可以成为一种规范的FP编程模式。我们在前面的几篇讨论中都涉及了一些AST的设计和运算,但都是一些功能单一,离散的例子。如果希望通过Free获取一个完整可用的程序,就必须想办法把离散的Free
AST组合成一体运算。我们先从单一的Free AST例子开始:
import scalaz._ import Scalaz._ import scala.language.higherKinds import scala.language.implicitConversions object FreeModules { object FreeInteract { trait Interact[+A] type FreeInteract[A] = Free.FreeC[Interact,A] object Interact { case class Ask(prompt: String) extends Interact[String] case class Tell(msg: String) extends Interact[Unit] implicit def interactToFreeC[A](ia: Interact[A]) = Free.liftFC(ia) object InteractConsole extends (Interact ~> Id) { def apply[A](ia: Interact[A]): Id[A] = ia match { case Ask(p) => println(p); readLine case Tell(m) => println(m) } } } import Interact._ val interactScript = for { first <- Ask("What's your first name?") last <- Ask("What's your last name?") _ <- Tell(s"Hello ${first} ${last}, nice to meet you!") } yield () } }
这是一个我们在前面讨论中重复描述几次的简单交互例子,包括了ADT、AST和Interpreter。我们可以直接运行这个程序:
object freePrgDemo extends App { import FreeModules._ import FreeInteract._ import Interact._ Free.runFC(interactScript)(InteractConsole) }
运算结果如下:
What's your first name? Tiger What's your last name? Chan Hello Tiger Chan, nice to meet you!
就是简单的两句界面提示和键盘输入,然后提示输入结果,没什么意义。作为测试,我们也可以模拟Console交互:用Map[String,String]来模拟Map[提问,回答],然后把这个Map提供给Interpreter,返回结果(List[String],A),其中List[String]是运行跟踪记录,A是模拟的键盘输入:
type InteractMapTester[A] = Map[String,String] => (List[String], A) implicit val mapTesterMonad = new Monad[InteractMapTester] { def point[A](a: => A) = _ => (List(), a) def bind[A,B](ia: InteractMapTester[A])(f: A => InteractMapTester[B]): InteractMapTester[B] = m => { val (o1,a1) = ia(m) val (o2,a2) = f(a1)(m) (o1 ++ o2, a2) } } object InteractTesterMap extends (Interact ~> InteractMapTester) { def apply[A](ia: Interact[A]): InteractMapTester[A] = ia match { case Ask(p) => { m => (List(), m(p)) } //m(p)返回提问对应的答案作为键盘输入 case Tell(s) => { m => (List(s), ()) } //List(s)在bind函数中的o1++o2形成跟踪记录 //在运算AST时就会调用InteractMapTester的bind函数 } }
使用模拟Console的Interpreter来运行:
object freePrgDemo extends App { import FreeModules._ import FreeInteract._ import Interact._ //Free.runFC(interactScript)(InteractConsole) val result = Free.runFC(interactScript)(InteractTesterMap).apply( Map( "What's your first name?" -> "tiger", "What's your last name?" -> "chan" )) println(result) } //产生以下输出结果 (List(Hello tiger chan, nice to meet you!),())
从mapTesterMonad定义中的bind看到了这句:o1++o2,是Logger的典型特征。那么用Writer能不能实现同等效果呢?我们先看看WriterT:
final case class WriterT[F[_], W, A](run: F[(W, A)]) { self => ...
实际上这个W就可以满足Logger的功能,因为在WriterT的flatMap中实现了W|+|W:
def flatMap[B](f: A => WriterT[F, W, B])(implicit F: Bind[F], s: Semigroup[W]): WriterT[F, W, B] = flatMapF(f.andThen(_.run)) def flatMapF[B](f: A => F[(W, B)])(implicit F: Bind[F], s: Semigroup[W]): WriterT[F, W, B] = writerT(F.bind(run){wa => val z = f(wa._2) F.map(z)(wb => (s.append(wa._1, wb._1), wb._2)) })
那么如何把Map[提问,回答]传人呢?我们可以通过WriterT[F[_],W,A]的F[]来实现这一目的:
type WriterTF[A] = Map[String,String] => A type InteractWriterTester[A] = WriterT[WriterTF,List[String],A]
然后我们可以用WriterT的参数run来传人Map[String,String]:run:WriterTF[(W,A)] == Map[String,String]=>(W,A)。
以下是用WriterT实现的Interpreter版本:
type WriterTF[A] = Map[String,String] => A type InteractWriterTester[A] = WriterT[WriterTF,List[String],A] def testerToWriter[A](f: Map[String,String] => (List[String], A)) = WriterT[WriterTF,List[String],A](f) implicit val writerTesterMonad = WriterT.writerTMonad[WriterTF, List[String]] object InteractTesterWriter extends (Interact ~> InteractWriterTester) { def apply[A](ia: Interact[A]): InteractWriterTester[A] = ia match { case Ask(p) => testerToWriter { m => (List(), m(p)) } case Tell(s) => testerToWriter { m => (List(s), ())} } }
我们可以这样运行:
object freePrgDemo extends App { import FreeModules._ import FreeInteract._ import Interact._ //Free.runFC(interactScript)(InteractConsole) //val result = Free.runFC(interactScript)(InteractTesterMap).apply( val result = Free.runFC(interactScript)(InteractTesterWriter).run( Map( "What's your first name?" -> "tiger", "What's your last name?" -> "chan" )) println(result) }
我们再设计另一个用户登录Login的例子:
object FreeUserLogin { import Dependencies._ trait UserLogin[+A] type FreeUserLogin[A] = Free.FreeC[UserLogin,A] object UserLogin { case class Login(user: String, pswd: String) extends UserLogin[Boolean] implicit def loginToFree[A](ul: UserLogin[A]) = Free.liftFC(ul) type LoginService[A] = Reader[PasswordControl,A] object LoginInterpreter extends (UserLogin ~> LoginService) { def apply[A](ul: UserLogin[A]): LoginService[A] = ul match { case Login(u,p) => Reader( cr => cr.matchPassword(u, p)) } } } import UserLogin._ val loginScript = for { b <- Login("Tiger","1234") } yield b }
这个例子里只有Login一个ADT,它的功能是把输入的User和Password与一个用户登录管理系统内的用户身份信息进行验证。由于如何进行用户密码验证不是这个ADT的功能,它可能涉及另一特殊功能系统的调用,刚好用来做个Reader依赖注入示范。以下是这项依赖定义:
object Dependencies { trait PasswordControl { type User = String type Password = String val pswdMap: Map[User, Password] def matchPassword(u: User, p: Password): Boolean } }
对loginScript进行测试运算时必须先获取PasswordControl实例,然后注入运算:
import Dependencies._ import FreeUserLogin._ import UserLogin._ object Passwords extends PasswordControl { //依赖实例 val pswdMap = Map ( "Tiger" -> "1234", "John" -> "0332" ) def matchPassword(u: User, p: Password) = pswdMap.getOrElse(u, p+"!") === p } val result = Free.runFC(loginScript)(LoginInterpreter).run(Passwords) //注入依赖 println(result)
不过即使能够运行,loginScsript的功能明显不完整,还需要像Interact那样的互动部分来获取用户输入信息。那么我们是不是考虑在ADT层次上把Interact和UserLogin合并起来,像这样:
case class Ask(prompt: String) extends Interact[String] case class Tell(msg: String) extends Interact[Unit] case class Login(user: String, pswd: String) extends Interact[Boolean]
明显这是可行的。但是,Interact和Login被紧紧捆绑在了一起形成了一个新的ADT。如果我们设计另一个同样需要互动的ADT,我们就需要重复同样的Interact功能设计,显然这样做违背了FP的原则:从功能单一的基本计算开始,按需要对基本函数进行组合实现更复杂的功能。Interact和UserLogin都是基础ADT,从编程语言角度描述Interact和UserLogin属于两种类型的编程语句。我们最终需要的AST是这样的:
val interLogin: Free[???, A] = for { user <- Ask("Enter User ID:") //Free[Interact,A] pswd <- Ask("Enter Password:") //Free[Interact,A] ok <- Login(user,pswd) //Free[UserLogin,A] } yield ok
不过明显类型对不上,因为Interact和UserLogin是两种语句。scalaz的Coproduct类型可以帮助我们实现两种Monadic语句的语义(sematics)合并。Coproduct是这样定义的:scalaz/Coproduct.scala
/** `F` on the left, and `G` on the right, of [[scalaz.\/]]. * * @param run The underlying [[scalaz.\/]]. */ final case class Coproduct[F[_], G[_], A](run: F[A] \/ G[A]) { import Coproduct._ def map[B](f: A => B)(implicit F: Functor[F], G: Functor[G]): Coproduct[F, G, B] = Coproduct(run.bimap(F.map(_)(f), G.map(_)(f))) ...
从run:F[A]\/G[A]可以理解Coproduct是两种语句F,G的联合(union)。在我们上面的例子里我们可以用下面的表达方式代表Interact和UserLogin两种语句的联合(union):
type InteractLogin[A] = Coproduct[Interact,UserLogin,A]
这是一个语义更广泛的类型:包含了Interact和UserLogin语义。我们可以用Inject类型来把Interact和UserLogin语句集“注入”到一个更大的句集。Inject是这样定义的:scalaz/Inject.scala
/** * Inject type class as described in "Data types a la carte" (Swierstra 2008). * * @see [[http://www.staff.science.uu.nl/~swier004/Publications/DataTypesALaCarte.pdf]] */ sealed abstract class Inject[F[_], G[_]] { def inj[A](fa: F[A]): G[A] def prj[A](ga: G[A]): Option[F[A]] } sealed abstract class InjectInstances { implicit def reflexiveInjectInstance[F[_]] = new Inject[F, F] { def inj[A](fa: F[A]) = fa def prj[A](ga: F[A]) = some(ga) } implicit def leftInjectInstance[F[_], G[_]] = new Inject[F, ({type λ[α] = Coproduct[F, G, α]})#λ] { def inj[A](fa: F[A]) = Coproduct.leftc(fa) def prj[A](ga: Coproduct[F, G, A]) = ga.run.fold(some(_), _ => none) } implicit def rightInjectInstance[F[_], G[_], H[_]](implicit I: Inject[F, G]) = new Inject[F, ({type λ[α] = Coproduct[H, G, α]})#λ] { def inj[A](fa: F[A]) = Coproduct.rightc(I.inj(fa)) def prj[A](ga: Coproduct[H, G, A]) = ga.run.fold(_ => none, I.prj(_)) } } ...
实现函数inj(fa:F[A]):G[A]代表把F[A]并入G[A]。这里还提供了三个类型的实例:
1、reflexiceInjectInstance[F[_]]:自我注入
2、leftInjectInstance[F[_],G[_]]:把F[A]注入Coproduct[F,G,A]的left(-\/)
3、rightInjectInstance[F[_],G[_],H[_]]:把F[A]注入Coproduct的right(\/-)。需要先把F注入G(inj(F[A]):G[A])
我们可以用implicitly来证明Interact和UserLogin的Inject实例存在:
val selfInj = implicitly[Inject[Interact,Interact]] type LeftInterLogin[A] = Coproduct[Interact,UserLogin,A] val leftInj = implicitly[Inject[Interact,LeftInterLogin]] type RightInterLogin[A] = Coproduct[UserLogin,LeftInterLogin,A] val rightInj = implicitly[Inject[Interact,RightInterLogin]]
现在我们需要把Coproduct[F,G,A]的F与G合并然后把F[A]升格成Free[G,A]:
object coproduct { def lift[F[_],G[_],A](fa: F[A])(implicit I: Inject[F,G]): Free.FreeC[G,A] = Free.liftFC(I.inj(fa)) }
我们可以用这个lift把Interact和UserLogin的ADT统一升格成Free[G,A]:
object coproduct { import FreeInteract._ import Interact._ import FreeUserLogin._ import UserLogin._ def lift[F[_],G[_],A](fa: F[A])(implicit I: Inject[F,G]): Free.FreeC[G,A] = Free.liftFC(I.inj(fa)) class Interacts[G[_]](implicit I: Inject[Interact,G]) { def ask(prompt: String): Free.FreeC[G,String] = lift(Ask(prompt)) def tell(msg: String): Free.FreeC[G,Unit] = lift(Tell(msg)) } class Logins[G[_]](implicit I: Inject[UserLogin,G]) { def login(u: String, p: String): Free.FreeC[G,Boolean] = lift(Login(u,p)) } }
我们用lift把基础Interact和UserLogin的语句注入了联合的语句集G[A],然后升格成FreeC[G,A]。现在我们可以把Interact,UserLogin这两种语句用在同一个for-comprehension里了:
def loginScript[G[_]](implicit I: Interacts[G], L: Logins[G]) ={ import I._ import L._ for { uid <- ask("ya id?") pwd <- ask("password?") login <- login(uid,pwd) _ <- if (login) tell("ya lucky bastard!") else tell("geda fk outa here!") } yield() }
有了Inject和Lift,现在已经成功的用两种ADT集成了一个AST。不过我们还必须提供Interacts[G]和Logins[G]实例:
object CoproductModules { object CoproductFunctions { import FreeInteract._ import Interact._ import FreeUserLogin._ import UserLogin._ def lift[F[_],G[_],A](fa: F[A])(implicit I: Inject[F,G]): Free.FreeC[G,A] = Free.liftFC(I.inj(fa)) class Interacts[G[_]](implicit I: Inject[Interact,G]) { def ask(prompt: String): Free.FreeC[G,String] = lift(Ask(prompt)) def tell(msg: String): Free.FreeC[G,Unit] = lift(Tell(msg)) } object Interacts { implicit def instance[G[_]](implicit I: Inject[Interact,G]) = new Interacts[G] } class Logins[G[_]](implicit I: Inject[UserLogin,G]) { def login(u: String, p: String): Free.FreeC[G,Boolean] = lift(Login(u,p)) } object Logins { implicit def instance[G[_]](implicit I: Inject[UserLogin,G]) = new Logins[G] } }
现在我们的语句集(AST)是一个联合的语句集(Coproduct)。那么,我们应该怎么去运算它呢?我们应该如何实现它的Interpreter?现在我们面对的Monadic程序类型是个Coproduct:
type InteractLogin[A] = Coproduct[Interact,UserLogin,A] val loginPrg = loginScript[InteractLogin]
现在语句集Interact和UserLogin是分别放在Coproduce的左右两边。那么我们可以历遍这个Coproduct来分别运算Interact和UserLogin语句:
def or[F[_],G[_],H[_]](fg: F ~> G, hg: H ~> G): ({type l[x] = Coproduct[F,H,x]})#l ~> G = new (({type l[x] = Coproduct[F,H,x]})#l ~> G) { def apply[A](ca: Coproduct[F,H,A]): G[A] = ca.run match { case -\/(fa) => fg(fa) case \/-(ha) => hg(ha) } }
值得注意的是如果or函数用在Interact和UserLogin上时它们自然转换(NaturalTransformation)的目标类型必须一致,应该是一个更大的类型,而且必须是Monad,这是NaturalTransformation的要求。所以我们可以把InteractInterpreter的转换目标类型由Id变成Reader,也就是LoginInterpreter的转换目标类型:
object InteractReader extends (Interact ~> LoginService) { def apply[A](ia: Interact[A]): LoginService[A] = ia match { case Ask(p) => println(p); Reader(cr => readLine) case Tell(m) => println(m); Reader(cr => ()) } }
好了,现在我们可以这样来测试运算:
object freePrgDemo extends App { import FreeModules._ import FreeInteract._ import Interact._ //Free.runFC(interactScript)(InteractConsole) //val result = Free.runFC(interactScript)(InteractTesterMap).apply( /* val result = Free.runFC(interactScript)(InteractTesterWriter).run( Map( "What's your first name?" -> "tiger", "What's your last name?" -> "chan" )) println(result) */ import Dependencies._ import FreeUserLogin._ import UserLogin._ object Passwords extends PasswordControl { val pswdMap = Map ( "Tiger" -> "1234", "John" -> "0332" ) def matchPassword(u: User, p: Password) = pswdMap.getOrElse(u, p+"!") === p } /* val result = Free.runFC(loginScript)(LoginInterpreter).run(Passwords) println(result) */ import CoproductDemo._ Free.runFC(loginPrg)(or(InteractReader,LoginInterpreter)).run(Passwords) }
我们把密码管理依赖也注入进去了。看看结果:
ya id? Tiger password? 2012 geda fk outa here! ya id? Tiger password? 1234 ya lucky bastard! ya id? John password? 0332 ya lucky bastard!
OK, 把这节示范源代码提供在下面:
package demos import scalaz._ import Scalaz._ import scala.language.higherKinds import scala.language.implicitConversions object FreeModules { object FreeInteract { trait Interact[+A] type FreeInteract[A] = Free.FreeC[Interact,A] object Interact { case class Ask(prompt: String) extends Interact[String] case class Tell(msg: String) extends Interact[Unit] implicit def interactToFreeC[A](ia: Interact[A]) = Free.liftFC(ia) object InteractConsole extends (Interact ~> Id) { def apply[A](ia: Interact[A]): Id[A] = ia match { case Ask(p) => println(p); readLine case Tell(m) => println(m) } } type InteractMapTester[A] = Map[String,String] => (List[String], A) implicit val mapTesterMonad = new Monad[InteractMapTester] { def point[A](a: => A) = _ => (List(), a) def bind[A,B](ia: InteractMapTester[A])(f: A => InteractMapTester[B]): InteractMapTester[B] = m => { val (o1,a1) = ia(m) val (o2,a2) = f(a1)(m) (o1 ++ o2, a2) } } object InteractTesterMap extends (Interact ~> InteractMapTester) { def apply[A](ia: Interact[A]): InteractMapTester[A] = ia match { case Ask(p) => { m => (List(), m(p)) } //m(p)返回提问对应的答案作为键盘输入 case Tell(s) => { m => (List(s), ()) } //List(s)在bind函数中的o1++o2形成跟踪记录 //在运算AST时会用到InteractMapTester的bind } } type WriterTF[A] = Map[String,String] => A type InteractWriterTester[A] = WriterT[WriterTF,List[String],A] def testerToWriter[A](f: Map[String,String] => (List[String], A)) = WriterT[WriterTF,List[String],A](f) implicit val writerTesterMonad = WriterT.writerTMonad[WriterTF, List[String]] object InteractTesterWriter extends (Interact ~> InteractWriterTester) { def apply[A](ia: Interact[A]): InteractWriterTester[A] = ia match { case Ask(p) => testerToWriter { m => (List(), m(p)) } case Tell(s) => testerToWriter { m => (List(s), ())} } } } import Interact._ val interactScript = for { first <- Ask("What's your first name?") last <- Ask("What's your last name?") _ <- Tell(s"Hello ${first} ${last}, nice to meet you!") } yield () } object FreeUserLogin { import Dependencies._ trait UserLogin[+A] type FreeUserLogin[A] = Free.FreeC[UserLogin,A] object UserLogin { case class Login(user: String, pswd: String) extends UserLogin[Boolean] implicit def loginToFree[A](ul: UserLogin[A]) = Free.liftFC(ul) type LoginService[A] = Reader[PasswordControl,A] object LoginInterpreter extends (UserLogin ~> LoginService) { def apply[A](ul: UserLogin[A]): LoginService[A] = ul match { case Login(u,p) => Reader( cr => cr.matchPassword(u, p)) } } } import UserLogin._ val loginScript = for { b <- Login("Tiger","1234") } yield b } } object Dependencies { trait PasswordControl { type User = String type Password = String val pswdMap: Map[User, Password] def matchPassword(u: User, p: Password): Boolean } } object CoproductDemo { import FreeModules._ import FreeUserLogin._ import UserLogin._ import FreeInteract._ import Interact._ import Dependencies._ def lift[F[_],G[_],A](fa: F[A])(implicit I: Inject[F,G]): Free.FreeC[G,A] = Free.liftFC(I.inj(fa)) class Interacts[G[_]](implicit I: Inject[Interact,G]) { def ask(prompt: String) = lift(Ask(prompt)) def tell(msg: String) = lift(Tell(msg)) } object Interacts { implicit def instance[F[_]](implicit I: Inject[Interact,F]) = new Interacts[F] } class Logins[G[_]](implicit I: Inject[UserLogin,G]) { def login(user: String, pswd: String) = lift(Login(user,pswd)) } object Logins { implicit def instance[F[_]](implicit I: Inject[UserLogin,F]) = new Logins[F] } def loginScript[G[_]](implicit I: Interacts[G], L: Logins[G]) ={ import I._ import L._ for { uid <- ask("ya id?") pwd <- ask("password?") login <- login(uid,pwd) _ <- if (login) tell("ya lucky bastard!") else tell("geda fk outa here!") } yield() } def or[F[_],G[_],H[_]](fg: F ~> G, hg: H ~> G): ({type l[x] = Coproduct[F,H,x]})#l ~> G = new (({type l[x] = Coproduct[F,H,x]})#l ~> G) { def apply[A](ca: Coproduct[F,H,A]): G[A] = ca.run match { case -\/(fa) => fg(fa) case \/-(ha) => hg(ha) } } type InteractLogin[A] = Coproduct[Interact,UserLogin,A] val loginPrg = loginScript[InteractLogin] object InteractReader extends (Interact ~> LoginService) { def apply[A](ia: Interact[A]): LoginService[A] = ia match { case Ask(p) => println(p); Reader(cr => readLine) case Tell(m) => println(m); Reader(cr => ()) } } } object freePrgDemo extends App { import FreeModules._ import FreeInteract._ import Interact._ //Free.runFC(interactScript)(InteractConsole) //val result = Free.runFC(interactScript)(InteractTesterMap).apply( /* val result = Free.runFC(interactScript)(InteractTesterWriter).run( Map( "What's your first name?" -> "tiger", "What's your last name?" -> "chan" )) println(result) */ import Dependencies._ import FreeUserLogin._ import UserLogin._ object Passwords extends PasswordControl { val pswdMap = Map ( "Tiger" -> "1234", "John" -> "0332" ) def matchPassword(u: User, p: Password) = pswdMap.getOrElse(u, p+"!") === p } /* val result = Free.runFC(loginScript)(LoginInterpreter).run(Passwords) println(result) */ import CoproductDemo._ Free.runFC(loginPrg)(or(InteractReader,LoginInterpreter)).run(Passwords) }