Feature overview: shapeless 2.0.0

// size is a function from Ints or Strings or pairs to a 'size' defined
// by type specific cases

object size extends Poly1 {
  implicit def caseInt = at[Int](x => 1)
  implicit def caseString = at[String](_.length)
  implicit def caseTuple[T, U]
    (implicit st : Case.Aux[T, Int], su : Case.Aux[U, Int]) =
      at[(T, U)](t => size(t._1)+size(t._2))
}

scala> size(23)
res4: Int = 1

scala> size("foo")
res5: Int = 3

scala> size((23, "foo"))
res6: Int = 4

scala> size(((23, "foo"), 13))
res7: Int = 5

Heterogenous lists

shapeless provides a comprehensive Scala HList which has many features not shared by other HList implementations.

It has a map operation, applying a polymorphic function value across its elements. This means that it subsumes both typical HList's and also KList's (HList's whose elements share a common outer type constructor).

import poly._

// The same definition of choose as above
object choose extends (Set ~> Option) {
  def apply[T](s : Set[T]) = s.headOption
}

scala> val sets = Set(1) :: Set("foo") :: HNil
sets: Set[Int] :: Set[String] :: HNil = Set(1) :: Set(foo) :: HNil

scala> val opts = sets map choose   // map selects cases of choose for each HList element
opts: Option[Int] :: Option[String] :: HNil = Some(1) :: Some(foo) :: HNil

It also has a flatMap operation,

import poly.identity

scala> val l = (23 :: "foo" :: HNil) :: HNil :: (true :: HNil) :: HNil
l: ((Int :: String :: HNil) :: HNil :: (Boolean :: HNil) :: HNil
   = (23 :: foo :: HNil) :: HNil :: (true :: HNil) :: HNil

scala> l flatMap identity
res0: Int :: String :: Boolean :: HNil = 23 :: foo :: true :: HNil

It has a set of fully polymorphic fold operations which take a polymorphic binary function value. The fold is sensitive to the static types of all of the elements of the HList. Given the earlier definition of size,

object addSize extends Poly2 {
  implicit  def default[T](implicit st: size.Case.Aux[T, Int]) =
    at[Int, T]{ (acc, t) => acc+size(t) }
}

scala> val l = 23 :: "foo" :: (13, "wibble") :: HNil
l: Int :: String :: (Int, String) :: HNil = 23 :: foo :: (13,wibble) :: HNil

scala> l.foldLeft(0)(addSize)
res1: Int = 11

It also has a zipper for traversal and persistent update,

import syntax.zipper._

scala> val l = 1 :: "foo" :: 3.0 :: HNil
l: Int :: String :: Double :: HNil = 1 :: foo :: 3.0 :: HNil

scala> l.toZipper.right.put(("wibble", 45)).reify
res0: Int :: (String, Int) :: Double :: HNil = 1 :: (wibble,45) :: 3.0 :: HNil

scala> l.toZipper.right.delete.reify
res1: Int :: Double :: HNil = 1 :: 3.0 :: HNil

scala> l.toZipper.last.left.insert("bar").reify
res2: Int :: String :: String :: Double :: HNil = 1 :: foo :: bar :: 3.0 :: HNil

It is covariant,

trait Fruit
case class Apple() extends Fruit
case class Pear() extends Fruit

type FFFF = Fruit :: Fruit :: Fruit :: Fruit :: HNil
type APAP = Apple :: Pear :: Apple :: Pear :: HNil

val a : Apple = Apple()
val p : Pear = Pear()

val apap : APAP = a :: p :: a :: p :: HNil
val ffff : FFFF = apap  // APAP <: FFFF

And it has a unify operation which converts it to an HList of elements of the least upper bound of the original types,

scala> apap.unify
res0: Fruit :: Fruit :: Fruit :: Fruit :: HNil = Apple() :: Pear() :: Apple() :: Pear() :: HNil

It supports conversion to an ordinary Scala List of elements of the least upper bound of the original types,

scala> apap.toList
res0: List[Fruit] = List(Apple(), Pear(), Apple(), Pear())

And it has a Typeable type class instance (see below), allowing, eg. vanilla List[Any]'s or HList's with elements of type Any to be safely cast to precisely typed HList's.

import syntax.typeable._

scala> val ffff : FFFF = apap.unify
ffff: FFFF = Apple() :: Pear() :: Apple() :: Pear() :: HNil

scala> val precise: Option[APAP] = ffff.cast[APAP]
precise: Option[APAP] = Some(Apple() :: Pear() :: Apple() :: Pear() :: HNil)

These last three features make this HList dramatically more practically useful than HList's are typically thought to be: normally the full type information required to work with them is too fragile to cross subtyping or I/O boundaries. This implementation supports the discarding of precise information where necessary (eg. to serialize a precisely typed record after construction), and its later reconstruction (eg. a weakly typed deserialized record with a known schema can have it's precise typing reestabilished).

HList-style operations on standard Scala tuples

shapeless allows standard Scala tuples to be manipulated in exactly the same ways as HLists,

import syntax.std.tuple._

// head, tail, take, drop, split
scala> (23, "foo", true).head
res0: Int = 23

scala> (23, "foo", true).tail
res1: (String, Boolean) = (foo,true)

scala> (23, "foo", true).drop(2)
res2: (Boolean,) = (true,)

scala> (23, "foo", true).take(2)
res3: (Int, String) = (23,foo)

scala> (23, "foo", true).split(1)
res4: ((Int,), (String, Boolean)) = ((23,),(foo,true))

// prepend, append, concatenate
scala> 23 +: ("foo", true)
res5: (Int, String, Boolean) = (23,foo,true)

scala> (23, "foo") :+ true
res6: (Int, String, Boolean) = (23,foo,true)

scala> (23, "foo") ++ (true, 2.0)
res7: (Int, String, Boolean, Double) = (23,foo,true,2.0)

// map, flatMap
import poly._

object option extends (Id ~> Option) {
  def apply[T](t: T) = Option(t)
}

scala> (23, "foo", true) map option
res8: (Option[Int], Option[String], Option[Boolean]) = (Some(23),Some(foo),Some(true))

scala> ((23, "foo"), (), (true, 2.0)) flatMap identity
res9: (Int, String, Boolean, Double) = (23,foo,true,2.0)

// fold (using previous definition of addSize)
scala> (23, "foo", (13, "wibble")).foldLeft(0)(addSize)
res10: Int = 11

// conversion to `HList`s and ordinary Scala `List`s
scala> (23, "foo", true).productElements
res11: Int :: String :: Boolean :: HNil = 23 :: foo :: true :: HNil

scala> (23, "foo", true).toList
res12: List[Any] = List(23, foo, true)

// zipper
import syntax.zipper._

scala> (23, ("foo", true), 2.0).toZipper.right.down.put("bar").root.reify
res13: (Int, (String, Boolean), Double) = (23,(bar,true),2.0)

Facilities for abstracting over arity

Conversions between tuples and HList's, and between ordinary Scala functions of arbitrary arity and functions which take a single corresponding HList argument allow higher order functions to abstract over the arity of the functions and values they are passed,

import syntax.std.function._
import ops.function._

def applyProduct[P <: Product, F, L <: HList, R](p: P)(f: F)
  (implicit gen: Generic.Aux[P, L], fp: FnToProduct.Aux[F, L => R]) =
    f.toProduct(gen.to(p))

scala> applyProduct(1, 2)((_: Int)+(_: Int))
res0: Int = 3

scala> applyProduct(1, 2, 3)((_: Int)*(_: Int)*(_: Int))
res1: Int = 6

Heterogenous maps

Shapeless provides a heterogenous map which supports an arbitrary relation between the key type and the corresponding value type,

// Key/value relation to be enforced: Strings map to Ints and vice versa
class BiMapIS[K, V]
implicit val intToString = new BiMapIS[Int, String]
implicit val stringToInt = new BiMapIS[String, Int]

val hm = HMap[BiMapIS](23 -> "foo", "bar" -> 13)
//val hm2 = HMap[BiMapIS](23 -> "foo", 23 -> 13)   // Does not compile

scala> hm.get(23)
res0: Option[String] = Some(foo)

scala> hm.get("bar")
res1: Option[Int] = Some(13)

And in much the same way that an ordinary monomorphic Scala map can be viewed as a monomorphic function value, so too can a heterogenous shapeless map be viewed as a polymorphic function value,

scala> import hm._
import hm._

scala> val l = 23 :: "bar" :: HNil
l: Int :: String :: HNil = 23 :: bar :: HNil

scala> l map hm
res2: String :: Int :: HNil = foo :: 13 :: HNil

Singleton-typed literals

Although Scala's typechecker has always represented singleton types for literal values internally, there has not previously been syntax available to express them, other than by modifying the compiler. shapeless adds support for singleton-typed literals via implicit macros.

Singleton types bridge the gap between the value level and the type level and hence allow the exploration in Scala of techniques which would typically only be available in languages with support for full-spectrum dependent types. The latest iteration of shapeless records (see next bullet) makes a start on that. Another simpler application is the use of Int literals to index into HLists and tuples,

import syntax.std.tuple._

scala> val l = 23 :: "foo" :: true :: HNil
l: Int :: String :: Boolean :: HNil = 23 :: foo :: true :: HNil

scala> l(1)
res0: String = foo

scala> val t = (23, "foo", true)
t: (Int, String, Boolean) = (23,foo,true)

scala> t(1)
res1: String = foo

The examples in the tests and the following illustrate other possibilities,

scala> import shapeless._, syntax.singleton._
import shapeless._
import syntax.singleton._

scala> 23.narrow
res0: Int(23) = 23

scala> "foo".narrow
res1: String("foo") = foo

scala> val (wTrue, wFalse) = (Witness(true), Witness(false))
wTrue: shapeless.Witness{type T = Boolean(true)} = $1$$1@212b9eca
wFalse: shapeless.Witness{type T = Boolean(false)} = $2$$1@36c5f0c9

scala> type True = wTrue.T
defined type alias True

scala> type False = wFalse.T
defined type alias False

scala> trait Select[B] { type Out }
defined trait Select

scala> implicit val selInt = new Select[True] { type Out = Int }
selInt: Select[True]{type Out = Int} = $anon$1@2c7b5e2a

scala> implicit val selString = new Select[False] { type Out = String }
selString: Select[False]{type Out = String} = $anon$2@57632e36

scala> def select(b: WitnessWith[Select])(t: b.instance.Out) = t
select: (b: shapeless.WitnessWith[Select])(t: b.instance.Out)b.instance.Out

scala> select(true)(23)
res2: Int = 23

scala> select(true)("foo")
<console>:18: error: type mismatch;
 found   : String("foo")
 required: Int
              select(true)("foo")
                           ^

scala> select(false)(23)
<console>:18: error: type mismatch;
 found   : Int(23)
 required: String
              select(false)(23)
                            ^

scala> select(false)("foo")
res3: String = foo

Singleton-typed Symbols

Scala's Symbol type, despite having its own syntax and being isomorphic to the String type, isn't equipped with useful singleton-typed literals. An encoding of singleton types for Symbol literals has proven to valuable (see below), and is represented by tagging the non-singleton type with the singleton type of the corresponding String literal,

scala> import syntax.singleton._
import syntax.singleton._

scala> 'foo          // non-singleton type
res0: Symbol = 'foo

scala> 'foo.narrow   // singleton type
res1: Symbol with shapeless.tag.Tagged[String("foo")] = 'foo

Extensible records

shapeless provides an implementation of extensible records modelled as HLists of values tagged with the singleton types of their keys. This means that there is no concrete representation needed at all for the keys. Amongst other things this will allow subsequent work on Generic to map case classes directly to records with their member names encoded in their element types.

import shapeless._ ; import syntax.singleton._ ; import record._

val book =
  ("author" ->> "Benjamin Pierce") ::
  ("title"  ->> "Types and Programming Languages") ::
  ("id"     ->>  262162091) ::
  ("price"  ->>  44.11) ::
  HNil

scala> book("author")  // Note result type ...
res0: String = Benjamin Pierce

scala> book("title")   // Note result type ...
res1: String = Types and Programming Languages

scala> book("id")      // Note result type ...
res2: Int = 262162091

scala> book("price")   // Note result type ...
res3: Double = 44.11

scala> book.keys       // Keys are materialized from singleton types encoded in value type
res4: String("author") :: String("title") :: String("id") :: String("price") :: HNil =
  author :: title :: id :: price :: HNil

scala> book.values
res5: String :: String :: Int :: Double :: HNil =
  Benjamin Pierce :: Types and Programming Languages :: 262162091 :: 44.11 :: HNil

scala> val newPrice = book("price")+2.0
newPrice: Double = 46.11

scala> val updated = book +("price" ->> newPrice)  // Update an existing field
updated: ... complex type elided ... =
  Benjamin Pierce :: Types and Programming Languages :: 262162091 :: 46.11 :: HNil

scala> updated("price")
res6: Double = 46.11

scala> val extended = updated + ("inPrint" ->> true)  // Add a new field
extended: ... complex type elided ... =
  Benjamin Pierce :: Types and Programming Languages :: 262162091 :: 46.11 :: true :: HNil

scala> val noId = extended - "id"  // Removed a field
noId: ... complex type elided ... =
  Benjamin Pierce :: Types and Programming Languages :: 46.11 :: true :: HNil

scala> noId("id")  // Attempting to access a missing field is a compile time error
<console>:25: error: could not find implicit value for parameter selector ...
              noId("id")
                  ^

Joni Freeman's (@jonifreeman) sqltyped library makes extensive use of shapeless records.

Coproducts and discriminated unions

shapeless has a Coproduct type, a generalization of Scala's Either to an arbitrary number of choices. Currently it exists primarily to support Generic (see the next section), but will be expanded analogously to HList in later releases. Currently Coproduct supports mapping, selection and unification,

scala> type ISB = Int :+: String :+: Boolean :+: CNil
defined type alias ISB

scala> val isb = Coproduct[ISB]("foo")
isb: ISB = foo

scala> isb.select[Int]
res0: Option[Int] = None

scala> isb.select[String]
res1: Option[String] = Some(foo)

object size extends Poly1 {
  implicit def caseInt = at[Int](i => (i, i))
  implicit def caseString = at[String](s => (s, s.length))
  implicit def caseBoolean = at[Boolean](b => (b, 1))
}

scala> isb map size
res2: (Int, Int) :+: (String, Int) :+: (Boolean, Int) :+: CNil = (foo,3)

scala> res2.select[(String, Int)]
res3: Option[(String, Int)] = Some((foo,3))

In the same way that adding labels to the elements of an HList gives us a record, adding labels to the elements of a Coproduct gives us a discriminated union,

scala> import record.RecordType, syntax.singleton._, union._
import record.RecordType
import syntax.singleton._
import union._

scala> val uSchema = RecordType.like('i ->> 23 :: 's ->> "foo" :: 'b ->> true :: HNil)
scala> type U = uSchema.Union
defined type alias U

scala> val u = Coproduct[U]('s ->> "foo")  // Inject a String into the union at label 's
u: U = foo

scala> u.get('i)   // Nothing at 'i
res0: Option[Int] = None

scala> u.get('s)   // Something at 's
res1: Option[String] = Some(foo)

scala> u.get('b)   // Nothing at 'b
res2: Option[Boolean] = None

Currently these exist primarily to support LabelledGeneric but, like Coproducts and records, will be further developed in future releases.

Generic representation of (sealed families of) case classes

The Isos of earlier shapeless releases have been completely reworked as the new Generic type, which closely resembles the generic programming capabilities introduced to GHC 7.2.

Generic[T], where T is a case class or an abstract type at the root of a case class hierarchy, maps between values of T and a generic sum of products representation (HLists and Coproducts),

scala> case class Foo(i: Int, s: String, b: Boolean)
defined class Foo

scala> val fooGen = Generic[Foo]
fooGen: shapeless.Generic[Foo]{ type Repr = Int :: String :: Boolean :: HNil } = $1$$1@724d2dfe

scala> val foo = Foo(23, "foo", true)
foo: Foo = Foo(23,foo,true)

scala> fooGen.to(foo)
res0: fooGen.Repr = 23 :: foo :: true :: HNil

scala> 13 :: res0.tail
res1: Int :: String :: Boolean :: HNil = 13 :: foo :: true :: HNil

scala> fooGen.from(res1)
res2: Foo = Foo(13,foo,true)

Typically values of Generic for a given case class are materialized using an implicit macro, allowing a wide variety of structural programming problems to be solved with no or minimal boilerplate. In particular the existing lens, Scrap Your Boilerplate and generic zipper implementations are now available for any case class family (recursive families included, as illustrated below) without any additional boilerplate being required,

// Simple recursive case class family
sealed trait Tree[T]
case class Leaf[T](t: T) extends Tree[T]
case class Node[T](left: Tree[T], right: Tree[T]) extends Tree[T]

// Polymorphic function which adds 1 to any Int and is the identity
// on all other values
object inc extends ->((i: Int) => i+1)

val tree: Tree[Int] =
  Node(
    Node(
      Node(
        Leaf(1),
        Node(
          Leaf(2),
          Leaf(3)
        )
      ),
      Leaf(4)
    ),
    Node(
      Leaf(5),
      Leaf(6)
    )
  )

// Transform tree by applying inc everywhere
everywhere(inc)(tree)

// result:
//   Node(
//     Node(
//       Node(
//         Leaf(2),
//         Node(
//           Leaf(3),
//           Leaf(4)
//         )
//       ),
//       Leaf(5)
//     ),
//     Node(
//       Leaf(6),
//       Leaf(7)
//     )
//   )

A natural extension of Generic's mapping of the content of data types onto a sum of products representation is to a mapping of the data type including its constructor and field names onto a labelled sum of products representation, ie. a representation in terms of the discriminated unions and records that we saw above. This is provided by LabelledGeneric. Currently it provides the underpinnings for the use of shapeless lenses with symbolic path selectors (see next section) and it is expected that it will support many scenarios which would otherwise require the support of hard to maintain special case macros.

scala> import record._, syntax.singleton._
import record._
import syntax.singleton._

scala> case class Book(author: String, title: String, id: Int, price: Double)
defined class Book

scala> val bookGen = LabelledGeneric[Book]

scala> val tapl = Book("Benjamin Pierce", "Types and Programming Languages", 262162091, 44.11)
tapl: Book = Book(Benjamin Pierce,Types and Programming Languages,262162091,44.11)

scala> val rec = bookGen.to(tapl) // Convert case class value to generic representation
rec: bookGen.Repr = Benjamin Pierce :: Types and Programming Languages :: 262162091 :: 44.11 :: HNil

scala> rec('price) // Access the price field symbolically, maintaining type information
res0: Double = 44.11

scala> bookGen.from(rec.updateWith('price)(_+2.0)) // type safe operations on fields
res1: Book = Book(Benjamin Pierce,Types and Programming Languages,262162091,46.11)

scala> case class ExtendedBook(author: String, title: String, id: Int, price: Double, inPrint: Boolean)
defined class ExtendedBook

scala> val bookExtGen = LabelledGeneric[ExtendedBook]

scala> bookExtGen.from(rec + ('inPrint ->> true))  // map values between case classes via generic representation
res2: ExtendedBook = ExtendedBook(Benjamin Pierce,Types and Programming Languages,262162091,44.11,true)

Boilerplate-free lenses for arbitrary case classes

A combination of LabelledGeneric and singleton-typed Symbol literals supports boilerplate-free lens creation for arbitrary case classes,

import shapeless._

// A pair of ordinary case classes ...
case class Address(street : String, city : String, postcode : String)
case class Person(name : String, age : Int, address : Address)

// Some lenses over Person/Address ...
val nameLens     = lens[Person] >> 'name
val ageLens      = lens[Person] >> 'age
val addressLens  = lens[Person] >> 'address
val streetLens   = lens[Person] >> 'address >> 'street
val cityLens     = lens[Person] >> 'address >> 'city
val postcodeLens = lens[Person] >> 'address >> 'postcode

scala> val person = Person("Joe Grey", 37, Address("Southover Street", "Brighton", "BN2 9UA"))
person: Person = Person(Joe Grey,37,Address(Southover Street,Brighton,BN2 9UA))

scala> val age1 = ageLens.get(person)               // Read field, note inferred type
age1: Int = 37

scala> val person2 = ageLens.set(person)(38)        // Update field
person2: Person = Person(Joe Grey,38,Address(Southover Street,Brighton,BN2 9UA))

scala> val person3 = ageLens.modify(person2)(_ + 1) // Transform field
person3: Person = Person(Joe Grey,39,Address(Southover Street,Brighton,BN2 9UA))

scala> val street = streetLens.get(person3)         // Read nested field
street: String = Southover Street

scala> val person4 = streetLens.set(person3)("Montpelier Road")  // Update nested field
person4: Person = Person(Joe Grey,39,Address(Montpelier Road,Brighton,BN2 9UA))

Automatic type class instance derivation

Based on and extending Generic and LabelledGeneric, Lars Hupel (@larsr_h) has contributed the TypeClass family of type classes, which provide automatic type class derivation facilities roughly equivalent to those available with GHC as described in "A Generic Deriving Mechanism for Haskell". There is a description of an earlier iteration of the Scala mechanism here, and examples of its use deriving Show and Monoid instances here and here for labelled coproducts and unlabelled products respectively.

For example, in the Monoid case, once the general deriving infrastructure for monoids is in place, instances are automatically available for arbitrary case classes without any additional boilerplate,

import MonoidSyntax._
import Monoid.auto._

// A pair of arbitrary case classes
case class Foo(i : Int, s : String)
case class Bar(b : Boolean, s : String, d : Double)

scala> Foo(13, "foo") |+| Foo(23, "bar")
res0: Foo = Foo(36,foobar)

scala> Bar(true, "foo", 1.0) |+| Bar(false, "bar", 3.0)
res1: Bar = Bar(true,foobar,4.0)

The shapeless-contrib project also contains automatically derived type class instances for Scalaz, Spire and Scalacheck.

First class lazy values tie implicit recursive knots

Traversals and transformations of recursive types (eg. cons lists or trees) must themselves be recursive. Consequently type class instances which perform such operations must be recursive values in turn. This is problematic in Scala at both the value and the type levels: at the value level the issue is that recursive type class instances would have to be constructed lazily, whilst Scala doesn't natively support lazy implicit arguments; at the type level the issue is that during the type checking of expressions constructing recursive implicit values the implicit resolution mechanism would revisit types in a way that would trip the divergence checker.

The Lazy[T] type constructor and associated macro in shapeless addresses both of these problems in many cases. It is similar to Scalaz's Need[T] and adds lazy implicit construction and suppression of divergence checking. This supports constructions such as,

// Simple cons list
sealed trait List[+T]
case class Cons[T](hd: T, tl: List[T]) extends List[T]
sealed trait Nil extends List[Nothing]
case object Nil extends Nil

trait Show[T] {
  def apply(t: T): String
}

object Show {
  // Base case for Int
  implicit def showInt: Show[Int] = new Show[Int] {
    def apply(t: Int) = t.toString
  }

  // Base case for Nil
  implicit def showNil: Show[Nil] = new Show[Nil] {
    def apply(t: Nil) = "Nil"
  }

  // Case for Cons[T]: note (mutually) recursive implicit argument referencing Show[List[T]]
  implicit def showCons[T](implicit st: Lazy[Show[T]], sl: Lazy[Show[List[T]]]): Show[Cons[T]] = new Show[Cons[T]] {
    def apply(t: Cons[T]) = s"Cons(${show(t.hd)(st.value)}, ${show(t.tl)(sl.value)})"
  }

  // Case for List[T]: note (mutually) recursive implicit argument referencing Show[Cons[T]]
  implicit def showList[T](implicit sc: Lazy[Show[Cons[T]]]): Show[List[T]] = new Show[List[T]] {
    def apply(t: List[T]) = t match {
      case n: Nil => show(n)
      case c: Cons[T] => show(c)(sc.value)
    }
  }
}

def show[T](t: T)(implicit s: Show[T]) = s(t)

val l: List[Int] = Cons(1, Cons(2, Cons(3, Nil)))

// Without the Lazy wrappers above the following would diverge ...

show(l)  // "Cons(1, Cons(2, Cons(3, Nil)))"

which would otherwise be impossible in Scala.

Collections with statically known sizes

shapeless provides collection types with statically known sizes. These can prevent runtime errors such as those that would result from attempting to take the head of an empty list, and can also verify more complex relationships.

In the example below we define a method csv whose signature guarantees at compile time that there are exactly as many column headers provided as colums,

def row(cols : Seq[String]) =
  cols.mkString("\\"", "\\", \\"", "\\"")

def csv[N <: Nat]
  (hdrs : Sized[Seq[String], N],
   rows : List[Sized[Seq[String], N]]) = row(hdrs) :: rows.map(row(_))

val hdrs = Sized("Title", "Author")

val rows = List(
  Sized("Types and Programming Languages", "Benjamin Pierce"),
  Sized("The Implementation of Functional Programming Languages", "Simon Peyton-Jones")
)

// hdrs and rows statically known to have the same number of columns
val formatted = csv(hdrs, rows)                        // Compiles

// extendedHdrs has the wrong number of columns for rows
val extendedHdrs = Sized("Title", "Author", "ISBN")
val badFormatted = csv(extendedHdrs, rows)             // Does not compile

Type safe cast

shapeless provides a Typeable type class which provides a type safe cast operation. cast returns an Option of the target type rather than throwing an exception if the value is of the incorrect type, as can happen with separate isInstanceOf and asInstanceOf operations. Typeable handles primitive values correctly and will recover erased types in many circumstances,

import syntax.typeable._

val l: Any = List(Vector("foo", "bar", "baz"), Vector("wibble"))
l: Any = List(Vector(foo, bar, baz), Vector(wibble))

scala> l.cast[List[Vector[String]]]
res0: Option[List[Vector[String]]] = Some(List(Vector(foo, bar, baz), Vector(wibble)))

scala> l.cast[List[Vector[Int]]]
res1: Option[List[Vector[Int]]] = None

scala> l.cast[List[List[String]]]
res2: Option[List[List[String]]] = None

An extractor based on Typeable is also available, allowing more precision in pattern matches,

scala> val `List[String]` = TypeCase[List[String]]
List[String]: shapeless.TypeCase[List[String]] = shapeless.TypeCase$$anon$16@14a9d20a

scala> val `List[Int]` = TypeCase[List[Int]]
List[Int]: shapeless.TypeCase[List[Int]] = shapeless.TypeCase$$anon$16@5810c269

scala> val l = List(1, 2, 3)
l: List[Int] = List(1, 2, 3)

scala> (l: Any) match {
     |   case `List[String]`(List(s, _*)) => s.length
     |   case `List[Int]`(List(i, _*))    => i+1
     | }
res0: Int = 2

The equivalent pattern match without Typeable/TypeCase would result in a compile-time warning about the erasure of the list's type parameter, then at runtime spuriously match the List[String] case and fail with a ClassCastException while attempting to evaluate its right hand side.

Be aware that the increased precision and safety provided by Typeable/TypeCase don't alter the fact that type caseing should be avoided in general other than at boundaries with external components which are intrinsically untyped (eg. serialization points) or which otherwise have poor type discipline.

Testing for non-compilation

Libraries like shapeless which make extensive use of type-level computation and implicit resolution often need to provide guarantees that certain expressions don't typecheck. Testing these guarantees is supported in shapeless via the illTyped macro,

import shapeless.test.illTyped

scala> illTyped { """1+1 : Boolean""" }

scala> illTyped { """1+1 : Int""" }
<console>:19: error: Type-checking succeeded unexpectedly.
Expected some error.
       illTyped { """1+1 : Int""" }
                ^

Provide feedback

Saved searches

Use saved searches to filter your results more quickly