The problem with your makeLens
is that we want e.g. makeLens[Content]('foo)
to fail at compile time, and that's not possible with an ordinary Symbol
argument. You need some extra implicit arguments to track the singleton type for the given name and to provide evidence that it's the name of a member of the case class:
import shapeless._, ops.record.{ Selector, Updater }, record.FieldType
class MakeLens[T <: Product] {
def apply[K, V, R <: HList](s: Witness.Aux[K])(implicit
gen: LabelledGeneric.Aux[T, R],
sel: Selector.Aux[R, K, V],
upd: Updater.Aux[R, FieldType[K, V], R]
): Lens[T, V] = lens[T] >> s
}
def makeLens[T <: Product] = new MakeLens[T]
And then:
scala> case class Content(field: Int)
defined class Content
scala> makeLens[Content]('field)
res0: shapeless.Lens[Content,Int] = shapeless.Lens$$anon$6@7d7ec2b0
But makeLens[Content]('foo)
won't compile (which is what we want).
You need the same kind of tracking for your nestedMapLens
:
import scalaz._, Scalaz._
import shapeless.contrib.scalaz._
case class LensesFor[T <: Product]() {
def nestedMapLens[K, V, R <: HList](
outerKey: String,
innerKey: Int,
s: Witness.Aux[K]
)(implicit
gen: LabelledGeneric.Aux[T, R],
sel: Selector.Aux[R, K, V],
upd: Updater.Aux[R, FieldType[K, V], R]
): PLens[Map[String, Map[Int, T]], V] =
(lens[T] >> s).asScalaz.partial.compose(
PLens.mapVPLens(innerKey)
).compose(
PLens.mapVPLens(outerKey)
)
}
Note that I'm assuming a build.sbt
like this:
scalaVersion := "2.11.2"
libraryDependencies ++= Seq(
"com.chuusai" %% "shapeless" % "2.0.0",
"org.typelevel" %% "shapeless-scalaz" % "0.3"
)
Now let's define an example map and some lenses:
val myMap = Map("foo" -> Map(1 -> Content(13)))
val myFoo1Lens = LensesFor[Content].nestedMapLens("foo", 1, 'field)
val myBar2Lens = LensesFor[Content].nestedMapLens("bar", 2, 'field)
And then:
scala> myFoo1Lens.get(myMap)
res4: Option[Int] = Some(13)
scala> myBar2Lens.get(myMap)
res5: Option[Int] = None
This is about as "boilerplate-free" as you're going to get. The messy implicit argument lists are intimidating at first, but you get used to them pretty quickly, and their role in pulling together different bits of evidence about the types you're working with becomes fairly intuitive after a little practice.