细谈Slick（6）－ Projection：ProvenShape，强类型的Query结果类型

在Slick官方文档中描述：连接后台数据库后，需要通过定义Projection，即def * 来进行具体库表列column的选择和排序。通过Projection我们可以选择库表中部分列、也可以增加一些自定义列computed column。具体来说Projection提供了数据库表列与Scala值的对应。例如def * = (column1,column2)把库表的column1和column2与(Int,String)对应，column1[Int]，column2[String]。也可以说是与定义column的类参数进行对应。从Slick源代码中我们可以找到Projection定义：

abstract class AbstractTable[T](val tableTag: Tag, val schemaName: Option[String], val tableName: String) extends Rep[T] {
  /** The client-side type of the table as defined by its * projection */
  type TableElementType
...
  /** The * projection of the table used as default for queries and inserts.
    * Should include all columns as a tuple, HList or custom shape and optionally
    * map them to a custom entity type using the <> operator.
    * The `ProvenShape` return type ensures that
    * there is a `Shape` available for translating between the `Column`-based
    * type in * and the client-side type without `Column` in the table's type
    * parameter. */
  def * : ProvenShape[T]
...
}

我们看到Projection是个ProvenShape[T]类。再看看ProvenShape是怎么定义的：

/** A limited version of ShapedValue which can be constructed for every type
  * that has a valid shape. We use it to enforce that a table's * projection
  * has a valid shape. A ProvenShape has itself a Shape so it can be used in
  * place of the value that it wraps for purposes of packing and unpacking. */
trait ProvenShape[U] {
  def value: Any
  val shape: Shape[_ <: FlatShapeLevel, _, U, _]
  def packedValue[R](implicit ev: Shape[_ <: FlatShapeLevel, _, U, R]): ShapedValue[R, U]
  def toNode = packedValue(shape).toNode
}

object ProvenShape {
  /** Convert an appropriately shaped value to a ProvenShape */
  implicit def proveShapeOf[T, U](v: T)(implicit sh: Shape[_ <: FlatShapeLevel, T, U, _]): ProvenShape[U] =
    new ProvenShape[U] {
      def value = v
      val shape: Shape[_ <: FlatShapeLevel, _, U, _] = sh.asInstanceOf[Shape[FlatShapeLevel, _, U, _]]
      def packedValue[R](implicit ev: Shape[_ <: FlatShapeLevel, _, U, R]): ShapedValue[R, U] = ShapedValue(sh.pack(value).asInstanceOf[R], sh.packedShape.asInstanceOf[Shape[FlatShapeLevel, R, U, _]])
    }

  /** The Shape for a ProvenShape */
  implicit def provenShapeShape[T, P](implicit shape: Shape[_ <: FlatShapeLevel, T, T, P]): Shape[FlatShapeLevel, ProvenShape[T], T, P] = new Shape[FlatShapeLevel, ProvenShape[T], T, P] {
    def pack(value: Mixed): Packed =
      value.shape.pack(value.value.asInstanceOf[value.shape.Mixed]).asInstanceOf[Packed]
    def packedShape: Shape[FlatShapeLevel, Packed, Unpacked, Packed] =
      shape.packedShape.asInstanceOf[Shape[FlatShapeLevel, Packed, Unpacked, Packed]]
    def buildParams(extract: Any => Unpacked): Packed =
      shape.buildParams(extract.asInstanceOf[Any => shape.Unpacked])
    def encodeRef(value: Mixed, path: Node) =
      value.shape.encodeRef(value.value.asInstanceOf[value.shape.Mixed], path)
    def toNode(value: Mixed): Node =
      value.shape.toNode(value.value.asInstanceOf[value.shape.Mixed])
  }
}

从implicit def proveShapeOf[T,U](v：T)：ProvenShape[U]可以得出对于任何T，如果能提供Shape[_,_,T,U,_]的隐式实例implicit instance的话就能构建出ProvenShape[U]。我们再看看什么是Shape：

/** A type class that encodes the unpacking `Mixed => Unpacked` of a
 * `Query[Mixed]` to its result element type `Unpacked` and the packing to a
 * fully packed type `Packed`, i.e. a type where everything which is not a
 * transparent container is wrapped in a `Column[_]`.
 *
 * =Example:=
 * - Mixed: (Column[Int], Column[(Int, String)], (Int, Option[Double]))
 * - Unpacked: (Int, (Int, String), (Int, Option[Double]))
 * - Packed: (Column[Int], Column[(Int, String)], (Column[Int], Column[Option[Double]]))
 * - Linearized: (Int, Int, String, Int, Option[Double])
 */
abstract class Shape[Level <: ShapeLevel, -Mixed, Unpacked_, Packed_] {...}

上面的Mixed就是ProvenShape的T，Unpacked就是U。如此看来T代表Query[T]的T，而U就是返回结果类型了。如果我们能提供T的Shape隐式实例就能把U升格成ProvenShape[U]。我们来看看Slick官方文件上的例子：

  import scala.reflect.ClassTag
  // A custom record class
  case class Pair[A, B](a: A, b: B)

  // A Shape implementation for Pair
  final class PairShape[Level <: ShapeLevel, M <: Pair[_,_], U <: Pair[_,_] : ClassTag, P <: Pair[_,_]](
                                                                     val shapes: Seq[Shape[_, _, _, _]])
    extends MappedScalaProductShape[Level, Pair[_,_], M, U, P] {
    def buildValue(elems: IndexedSeq[Any]) = Pair(elems(), elems())
    def copy(shapes: Seq[Shape[_ <: ShapeLevel, _, _, _]]) = new PairShape(shapes)
  }

  implicit def pairShape[Level <: ShapeLevel, M1, M2, U1, U2, P1, P2](
                implicit s1: Shape[_ <: Level, M1, U1, P1], s2: Shape[_ <: Level, M2, U2, P2]
               ) = new PairShape[Level, Pair[M1, M2], Pair[U1, U2], Pair[P1, P2]](Seq(s1, s2))

  // Use it in a table definition
  class A(tag: Tag) extends Table[Pair[Int, String]](tag, "shape_a") {
    def id = column[Int]("id", O.PrimaryKey)
    def s = column[String]("s")
    def * = Pair(id, s)
  }
  val as = TableQuery[A]

现在Projection可以写成Pair(id,s)。也就是说因为有了implicit def pairShape[...](...)：PairShape所以Pair(id,s)被升格成ProvenShape[Pair]。这样Query的返回类型就是Seq[Pair]了。实际上Slick本身提供了Tuple、Case Class、HList等类型的默认Shape隐式实例，所以我们可以把Projection直接写成 def * = (...) 或 Person(...) 或 Int::String::HNil。下面是Tuple的默认Shape：

trait TupleShapeImplicits {
  @inline
  implicit final def tuple1Shape[Level <: ShapeLevel, M1, U1, P1](implicit u1: Shape[_ <: Level, M1, U1, P1]): Shape[Level, Tuple1[M1], Tuple1[U1], Tuple1[P1]] =
    new TupleShape[Level, Tuple1[M1], Tuple1[U1], Tuple1[P1]](u1)
  @inline
  implicit final def tuple2Shape[Level <: ShapeLevel, M1,M2, U1,U2, P1,P2](implicit u1: Shape[_ <: Level, M1, U1, P1], u2: Shape[_ <: Level, M2, U2, P2]): Shape[Level, (M1,M2), (U1,U2), (P1,P2)] =
    new TupleShape[Level, (M1,M2), (U1,U2), (P1,P2)](u1,u2)
...

回到主题，下面是一个典型的Slick数据库表读取例子：

   class TupleTypedPerson(tag: Tag) extends Table[(
      Option[Int],String,Int,Option[String])](tag,"PERSON") {
     def id = column[Int]("id",O.PrimaryKey,O.AutoInc)
     def name = column[String]("name")
     def age = column[Int]("age")
     def alias = column[Option[String]]("alias")
     def * = (id.?,name,age,alias)
   }
   val tupleTypedPerson = TableQuery[TupleTypedPerson]

   val db = Database.forURL("jdbc:h2:mem:test1;DB_CLOSE_DELAY=-1", driver = "org.h2.Driver")
   val createSchemaAction = tupleTypedPerson.schema.create
   Await.ready(db.run(createSchemaAction),Duration.Inf)
   val initDataAction = DBIO.seq {
     tupleTypedPerson ++= Seq(
       (Some(),"Tiger Chan", , Some("Tiger_XC")),
       (Some(),"Johnny Cox", , None),
       (Some(),"Cathy Williams", , Some("Catty")),
       (Some(),"David Wong", , None)
     )
   }
   Await.ready(db.run(initDataAction),Duration.Inf)
   val queryAction = tupleTypedPerson.result

   Await.result(db.run(queryAction),Duration.Inf).foreach {row =>
     println(s"${row._1.get} ${row._2} ${row._4.getOrElse("")}, ${row._3}")
   }

在这个例子的表结构定义里默认的Projection是个Tuple。造成的后果是返回的结果行不含字段名，只有字段位置。使用这样的行数据很容易错误对应，或者重复确认正确的列值会影响工作效率。如果返回的结果类型是Seq[Person]这样的话：Person是个带属性的对象如case class，那么我们就可以通过IDE提示的字段名称来选择字段了。上面提过返回结果类型可以通过ProvenShape来确定，如果能实现ProvenShape[A] => ProvenShape[B]这样的转换处理，那么我们就可以把返回结果行类型从Tuple变成有字段名的类型了：

   class Person(val id: Option[Int],
                val name: String, val age: Int, val alias: Option[String])
   def toPerson(t: (Option[Int],String,Int,Option[String])) = new Person (
     t._1,t._2,t._3,t._4
   )
   def fromPerson(p: Person) = Some((p.id,p.name,p.age,p.alias))
   class TupleMappedPerson(tag: Tag) extends Table[
     Person](tag,"PERSON") {
     def id = column[Int]("id",O.PrimaryKey,O.AutoInc)
     def name = column[String]("name")
     def age = column[Int]("age")
     def alias = column[Option[String]]("alias")
     def * = (id.?,name,age,alias) <> (toPerson,fromPerson)
   }
   val tupleMappedPerson = TableQuery[TupleMappedPerson]

   Await.result(db.run(tupleMappedPerson.result),Duration.Inf).foreach {row =>
     println(s"${row.id.get} ${row.name} ${row.alias.getOrElse("")}, ${row.age}")
   }

我们用<>函数进行了Tuple=>Person转换。注意toPerson和fromPerson这两个相互转换函数。如果Person是个case class，那么Person.tupled和Person.unapply就是它自备的转换函数，我们可以用case class来构建MappedProjection：

   case class Person(id: Option[Int]=None, name: String, age: Int, alias: Option[String])

   class MappedTypePerson(tag: Tag) extends Table[Person](tag,"PERSON") {
     def id = column[Int]("id",O.PrimaryKey,O.AutoInc)
     def name = column[String]("name")
     def age = column[Int]("age")
     def alias = column[Option[String]]("alias")
     def * = (id.?,name,age,alias) <> (Person.tupled,Person.unapply)
   }
   val mappedPeople = TableQuery[MappedTypePerson]

从上面两个例子里我们似乎可以得出ProvenShape[T]的T类型就是Table[T]的T，也就是返回结果行的类型了。我们可以用同样方式来进行HList与Person转换：

   def hlistToPerson(hl: Option[Int]::String::Int::(Option[String])::HNil) =
     new Person(hl(),hl(),hl(),hl())
   def personToHList(p: Person) = Some(p.id::p.name::p.age::p.alias::HNil)
   class HListPerson(tag: Tag) extends Table[Person](tag,"PERSON") {
     def id = column[Int]("id",O.PrimaryKey,O.AutoInc)
     def name = column[String]("name")
     def age = column[Int]("age")
     def alias = column[Option[String]]("alias")
     def * = (id.?)::name::age::alias::HNil <> (hlistToPerson,personToHList)
   }
   val hlistPerson = TableQuery[HListPerson]
   Await.result(db.run(hlistPerson.result),Duration.Inf).foreach {row =>
     println(s"${row.id.get} ${row.name} ${row.alias.getOrElse("")}, ${row.age}")
   }

同样，必须首先实现hlistToPerson和personToHList转换函数。现在Table的类型参数必须是Person。上面的Projection都是对Table默认Projection的示范。实际上我们可以针对每个Query来自定义Projection，如下：

  case class YR(name: String, yr: Int)

   val qYear = for {
     p <- hlistPerson
   } yield ((p.name, p.age) <> (YR.tupled,YR.unapply))

   Await.result(db.run(qYear.result),Duration.Inf).foreach {row =>
     println(s"${row.name} ${row.yr}")
   }

上面这个例子里我们构建了基于case class YR的projection。在join table query情况下只能通过这种方式来构建Projection，看看下面这个例子：

   case class Title(id: Int, title: String)
   class PersonTitle(tag: Tag) extends Table[Title](tag,"TITLE") {
     def id = column[Int]("id")
     def title = column[String]("title")
     def * = (id,title) <> (Title.tupled,Title.unapply)
   }
   val personTitle = TableQuery[PersonTitle]
   val createTitleAction = personTitle.schema.create
    Await.ready(db.run(createTitleAction),Duration.Inf)
    val initTitleData = DBIO.seq {
      personTitle ++= Seq(
        Title(,"Manager"),
        Title(,"Programmer"),
        Title(,"Clerk")
      )
    }
    Await.ready(db.run(initTitleData),Duration.Inf)

   case class Titles(id: Int, name: String, title: String)
   val qPersonWithTitle = for {
     p <- hlistPerson
     t <- personTitle if p.id === t.id
   } yield ((p.id,p.name,t.title) <> (Titles.tupled,Titles.unapply))
   Await.result(db.run(qPersonWithTitle.result),Duration.Inf).foreach {row =>
     println(s"${row.id} ${row.name}, ${row.title}")
   }

现在对任何形式的Query结果我们都能使用强类型（strong typed）的字段名称来进行操作了。

下面是本次示范的源代码：

 import slick.collection.heterogeneous.{ HList, HCons, HNil }
 import slick.collection.heterogeneous.syntax._
 import slick.driver.H2Driver.api._

 import scala.concurrent.ExecutionContext.Implicits.global
 import scala.concurrent.duration._
 import scala.concurrent.{Await, Future}

 object chkProjection {

   class TupleTypedPerson(tag: Tag) extends Table[(
      Option[Int],String,Int,Option[String])](tag,"PERSON") {
     def id = column[Int]("id",O.PrimaryKey,O.AutoInc)
     def name = column[String]("name")
     def age = column[Int]("age")
     def alias = column[Option[String]]("alias")
     def * = (id.?,name,age,alias)
   }
   val tupleTypedPerson = TableQuery[TupleTypedPerson]

   val db = Database.forURL("jdbc:h2:mem:test1;DB_CLOSE_DELAY=-1", driver = "org.h2.Driver")
   val createSchemaAction = tupleTypedPerson.schema.create
   Await.ready(db.run(createSchemaAction),Duration.Inf)
   val initDataAction = DBIO.seq {
     tupleTypedPerson ++= Seq(
       (Some(),"Tiger Chan", , Some("Tiger_XC")),
       (Some(),"Johnny Cox", , None),
       (Some(),"Cathy Williams", , Some("Catty")),
       (Some(),"David Wong", , None)
     )
   }
   Await.ready(db.run(initDataAction),Duration.Inf)

   val queryAction = tupleTypedPerson.result

   Await.result(db.run(queryAction),Duration.Inf).foreach {row =>
     println(s"${row._1.get} ${row._2} ${row._4.getOrElse("")}, ${row._3}")
   }

   class Person(val id: Option[Int],
                val name: String, val age: Int, val alias: Option[String])
   def toPerson(t: (Option[Int],String,Int,Option[String])) = new Person (
     t._1,t._2,t._3,t._4
   )
   def fromPerson(p: Person) = Some((p.id,p.name,p.age,p.alias))
   class TupleMappedPerson(tag: Tag) extends Table[
     Person](tag,"PERSON") {
     def id = column[Int]("id",O.PrimaryKey,O.AutoInc)
     def name = column[String]("name")
     def age = column[Int]("age")
     def alias = column[Option[String]]("alias")
     def * = (id.?,name,age,alias) <> (toPerson,fromPerson)
   }
   val tupleMappedPerson = TableQuery[TupleMappedPerson]

   Await.result(db.run(tupleMappedPerson.result),Duration.Inf).foreach {row =>
     println(s"${row.id.get} ${row.name} ${row.alias.getOrElse("")}, ${row.age}")
   }

   def hlistToPerson(hl: Option[Int]::String::Int::(Option[String])::HNil) =
     new Person(hl(),hl(),hl(),hl())
   def personToHList(p: Person) = Some(p.id::p.name::p.age::p.alias::HNil)
   class HListPerson(tag: Tag) extends Table[Person](tag,"PERSON") {
     def id = column[Int]("id",O.PrimaryKey,O.AutoInc)
     def name = column[String]("name")
     def age = column[Int]("age")
     def alias = column[Option[String]]("alias")
     def * = (id.?)::name::age::alias::HNil <> (hlistToPerson,personToHList)
   }
   val hlistPerson = TableQuery[HListPerson]
   Await.result(db.run(hlistPerson.result),Duration.Inf).foreach {row =>
     println(s"${row.id.get} ${row.name} ${row.alias.getOrElse("")}, ${row.age}")
   }

   case class YR(name: String, yr: Int)

   val qYear = for {
     p <- hlistPerson
   } yield ((p.name, p.age) <> (YR.tupled,YR.unapply))

   Await.result(db.run(qYear.result),Duration.Inf).foreach {row =>
     println(s"${row.name} ${row.yr}")
   }

   case class Title(id: Int, title: String)
   class PersonTitle(tag: Tag) extends Table[Title](tag,"TITLE") {
     def id = column[Int]("id")
     def title = column[String]("title")
     def * = (id,title) <> (Title.tupled,Title.unapply)
   }
   val personTitle = TableQuery[PersonTitle]
   val createTitleAction = personTitle.schema.create
    Await.ready(db.run(createTitleAction),Duration.Inf)
    val initTitleData = DBIO.seq {
      personTitle ++= Seq(
        Title(,"Manager"),
        Title(,"Programmer"),
        Title(,"Clerk")
      )
    }
    Await.ready(db.run(initTitleData),Duration.Inf)

   case class Titles(id: Int, name: String, title: String)
   val qPersonWithTitle = for {
     p <- hlistPerson
     t <- personTitle if p.id === t.id
   } yield ((p.id,p.name,t.title) <> (Titles.tupled,Titles.unapply))
   Await.result(db.run(qPersonWithTitle.result),Duration.Inf).foreach {row =>
     println(s"${row.id} ${row.name}, ${row.title}")
   }

 }