2015-04-30

Introducing Emblem - TypeKeys

This is the second post in the Introducing Emblem series, where I discuss a reflection-based utility library I have developed called emblem. The first post presented a high-level overview of the library. In this post, we look at the core building block of the emblem library: the TypeKey.

Because Scala generics are built on top of Java generics in the JVM, they suffer the same problems with type-erasure. We effectively lose track of the actual values of the type arguments for the generic type. Thankfully, Scala provides TypeTags to get around this problem. Type tags are explained in detail in other places, such as on the Scala website, and in this Stack Overflow answer, so I won't go into the details here.

Type tags are great, and I use them all the time. Maybe I use them a little too much, because sometimes I go beyond their intended usage, and run into problems. This time around, I tried to use them as keys in a map. As the Stack Overflow answer above points out, type tags are not necessarily equal, as in == equal, even though the two type tags represent an equivalent type.

To see this in action, let's boot up the REPL (Scala's interactive shell). I'm going to start it up inside SBT in the emblem project, so later on I can import emblem stuff as well. This procedure should work for any project that declares a dependency on emblem. Prompts are in black, input in blue, non-prompt output in purple, and superfluous output is omitted:

bash% git clone https://github.com/sullivan-/longevity.git
bash% cd longevity
bash% sbt
> project emblem
> console
Welcome to Scala version 2.11.5 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0_25).

scala> 

Okay, now we're in the REPL. Let's import all the type-taggy stuff from the Scala reflection library:

scala> import scala.reflect.runtime.universe._
import scala.reflect.runtime.universe._

Let's start with a simple type alias as an example. The type tags are not equal:

scala> trait A
defined trait A

scala> type B = A
defined type alias B

scala> typeTag[A] == typeTag[B]
res0: Boolean = false

The proper way to determine if these are equivalent types is with the =:= operator on scala.reflect.api.Types.Type, which can be accessed with method TypeTag.tpe:

scala> typeTag[A].tpe =:= typeTag[B].tpe
res1: Boolean = true

Or, equivalently:

scala> typeOf[A] =:= typeOf[B]
res2: Boolean = true

Note that TypeTag.== will fail us not only when the types are equivalent as above, but in some circumstances when they are exactly the same type:

scala> object x {
     |   trait C
     |   val ctag = typeTag[C]
     | }
defined object x

scala> x.ctag == typeTag[x.C]
res3: Boolean = false

scala> x.ctag.tpe =:= typeTag[x.C].tpe
res4: Boolean = true

Thankfully, we don't get a new, non-equal type tag every time the typeTag method is called:

scala> typeTag[A] == typeTag[A]
res5: Boolean = true

Clearly, type tags are not going to work very well as keys in a map. But emblem's TypeKeys are. Let's give it a try:

scala> import emblem.imports._
import emblem.imports._

scala> typeKey[A] == typeKey[B]
res6: Boolean = true

scala> object x {
     |   trait C
     |   val ckey = typeKey[C]
     | }
defined object x

scala> x.ckey == typeKey[x.C]
res7: Boolean = true

Because there is an implicit conversion from TypeTag to TypeKey, we can basically use a key anywhere we can use a tag. Common usage of type tags is like so:

scala> def grokTag[A : TypeTag] = println(typeTag[A].tpe)
grokTag: [A](implicit evidence$1: reflect.runtime.universe.TypeTag[A])Unit

scala> grokTag[List[_]]
scala.List[_]

scala> grokTag[List[Int]]
scala.List[Int]

We can do the exact same thing with type keys:

scala> def grokKey[A : TypeKey] = println(typeKey[A].tpe)
grokKey: [A](implicit evidence$1: emblem.imports.TypeKey[A])Unit

scala> grokKey[List[_]]
scala.List[_]

scala> grokKey[List[Int]]

scala.List[Int]

Either of the above grokking methods could have alternatively made use of Scala's implicitly:

scala> def grokTag[A : TypeTag] =
     |   println(implicitly[TypeTag[A]].tpe)
grokTag: [A](implicit evidence$1: reflect.runtime.universe.TypeTag[A])Unit

scala> def grokKey[A : TypeKey] = 
     |   println(implicitly[TypeKey[A]].tpe)
grokKey: [A](implicit evidence$1: emblem.imports.TypeKey[A])Unit

You can also manually convert between tags and keys yourself:

scala> val tag = typeTag[A]
tag: reflect.runtime.universe.TypeTag[A] = TypeTag[A]

scala> TypeKey(tag)
res8: emblem.TypeKey[A] = TypeKey[A]

scala> val key = typeKey[A]
key: emblem.imports.TypeKey[A] = TypeKey[A]

scala> key.tag
res9: reflect.runtime.universe.TypeTag[A] = TypeTag[A]

Of course, the TypeKey.hashCode method is consistent with equals, so we can use them as keys in sets and maps. To see this, let's set up some types to test with:

scala> trait A1
scala> type A2 = A1
scala> trait B1
scala> type B2 = B1
scala> trait C1
scala> type C2 = C1

And now, some sets:

scala> val tagset1 = Set(
     |   typeTag[A1], typeTag[A2],
     |   typeTag[B1], typeTag[B2],
     |   typeTag[C1], typeTag[C2])
scala> val tagset2 = Set(typeTag[A1], typeTag[B1], typeTag[C1])
scala> val keyset1 = Set(
     |   typeKey[A1], typeKey[A2],
     |   typeKey[B1], typeKey[B2],
     |   typeKey[C1], typeKey[C2])
scala> val keyset2 = Set(typeKey[A1], typeKey[B1], typeKey[C1])

The size of these sets are 6, 3, 3 and 3, respectively. This is how they equal:


scala> tagset1 == tagset2
res10: Boolean = false

scala> keyset1 == keyset2
res11: Boolean = true

scala> tagset1.map(TypeKey(_)) == keyset1
res12: Boolean = true

Clearly, a set of type tags does not have any conceptual correlation to a set of types. But the set of type keys does.

Above, when we called methods grokTag and grokKey, we specified the type argument explicitly, with calls like grokTag[A] and grokKey[A]. We can also lock down the type argument another way: by explicitly specifying the implicit arguments for TypeTag and TypeKey:

scala> grokTag(typeTag[List[Int]])
scala.List[Int]

scala> grokKey(typeKey[List[Int]])
scala.List[Int]

We can do the same thing with the tags and keys in our sets:

scala> tagset1.foreach(grokTag(_))
C2
A1
B2
A2
B1
C1

scala> tagset2.foreach(grokTag(_))
A1
B1
C1

scala> keyset1.foreach(grokKey(_))
A1
B1
C1

scala> keyset2.foreach(grokKey(_))
A1
B1
C1

We can also use type keys as keys in maps. For instance:

scala> var counter = Map[TypeKey[_], Int]()
counter: scala.collection.immutable.Map[emblem.TypeKey[_],Int] = Map()

Here's a function to increment the counter for a given key:

scala> def countKey[A : TypeKey]: Unit =
     |   counter += typeKey[A] ->
     |     (counter.getOrElse(typeKey[A], 0) + 1)
countKey: [A](implicit evidence$1: emblem.imports.TypeKey[A])Unit

Let's see how it works:

scala> counter(typeKey[A1])
java.util.NoSuchElementException: key not found: TypeKey[A1]
  at scala.collection.MapLike$class.default(MapLike.scala:228)
  at scala.collection.AbstractMap.default(Map.scala:59)
  at scala.collection.MapLike$class.apply(MapLike.scala:141)
  at scala.collection.AbstractMap.apply(Map.scala:59)
  ... 43 elided

scala> counter(typeKey[A2])
java.util.NoSuchElementException: key not found: TypeKey[A2]

scala> countKey[A1]

scala> counter(typeKey[A1])
res13: Int = 1

scala> counter(typeKey[A2])
res14: Int = 1

scala> countKey[A1]

scala> countKey[A2]

scala> countKey[A1]

scala> countKey[A2]

scala> counter(typeKey[A1])
res15: Int = 5

scala> counter(typeKey[A2])
res16: Int = 5

By this point, we've exercised all the basic functionality of a TypeKey.  Do you have any ideas of real-life scenarios where these might come in handy? If you do, please comment, I would love to hear about it. We'll start looking at a more realistic use-case in the next post, when we begin to investigate TypeKeyMaps.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.