...is a variant on the Type Class Pattern where Scala copies a Haskell idiom. A good article is here. The idea is that there is just one method that takes a variety of objects with the same super-trait depending on what parameters were used in calling that method. This replaces having many overloaded methods.
[Just to refresh one's memory: the Type Class pattern is a method of ad hoc polymorphism but it differs from the Pimp My Library pattern. A good comparison of the two can be found here. Essentially, the Pimp My Library pattern appears to add methods to classes that were not there when they were first written. The key thing about the Type Class pattern is that it asks for a witness to evidence that something is possible for a certain type. Think about Scala's TraversableOnce.sum method that asks for evidence that the type it contains has an associated Numeric. In Pimp My Library, you're creating the functionality per type. In the Type Class pattern your providing the functionality for a more general function to do its work.]
Implicits
What is apparent immediately is that you really need to know about implicit rules. Here are some notes but a really good reference lies here.
Let's take a fairly pointless class:
class AContainer
class PimpMe {
def pimpedMethodOnClass: String = "pimpedMethodOnClass"
}
implicit def pimpedWithADefCreatingClass(aContainter: AContainer): PimpMe = new PimpMe
Now, we can call this method as if it were on the class itself:
println(aContainer.pimpedMethodOnClass) // "pimpedWithADef"
More efficiently (see below) we could write:
implicit class PimpMeEfficiently(val aContainer: AContainer) extends AnyVal {
def pimpedMethodOnEfficientClass: String = "pimpedMethodOnEfficientClass"
}
.
.
println(aContainer.pimpedMethodOnEfficientClass) // "pimpedMethodOnEfficientClass"
The reason for AnyVal is "properly-defined user value classes provide a way to improve performance on user-defined types by avoiding object allocation at runtime, and by replacing virtual method invocations with static method invocations" (from the ScalaDocs).
The Magnet Pattern Simplified
A simplified magnet pattern implementation is a single method. Let's use a slightly more interesting magnet class whose toString method yields a different value. In real life, each subclass would have a more interesting implementation:
class Magnet(string: String) {
override def toString() = string
}
And let's have a very trivial implementation of the method taking the magnet:
def magnetMethod(magnet: Magnet): Unit = println(magnet)
Now, we can call this method with what look like different arguments, for example:
implicit def fromStringAndInt(tuple: (String, Int)) = new Magnet(s"Implementation X: (String, Int) = (${tuple._1}, ${tuple._2})")
.
.
magnetMethod("aString", 1) // "Implementation X: (String, Int) = (aString, 1)" (thanks to fromStringAndInt)
magnetMethod(("aString", 1)) // ditto
Note calling the magnet method with a tuple has the same effect as calling it with individual arguments.
Now, we could have a different implementation but let's be simple and just pretend there is different functionality. But we really do have a different call site:
implicit def fromDouble(double: Double) = new Magnet(s"Implementation Y: double = $double")
.
.
magnetMethod(1.1d) // "Implementation Y: double = 1.1" (thanks to fromDouble)
We're calling the same method with different arguments (cardinality as well as types)!
No comments:
Post a Comment