问题
I have implementations of type-level arithmetics capable of doing some compile time arithmetic validation, namely <,>,=
in two ways:
- simple implementation
- rigorous implementation
With these, I can have a getFoo
function that I can call like this:
getFoo[_2,_3]
With _2
and _3
being the type-level equivalents of integer values 2 and 3. Now Ideally I would like my getFoo
function to take integer values as arguments and attempt to infer _2
from the value 2
.
My plan was to add the following associatedInt information to the Nat
base class:
trait Nat {
val associatedInt: Int
type AssociatedInt = associatedInt.type
}
So that the subsequent types would be defined as:
type _1 = Succ[_0] {
override val associatedInt: Int = 1
}
And then change getFoo's signature so that it takes an integer:
def getFoo(i:Int)(implicit ...)
Based on which, we would do our type level arithmetic assertions with types associated to the AssociatedInt
type. Ie, something like:
def getFoo(i: Integer)(implicit ev: Nat{type I = i.type } =:= ExpectedType)
Which is not working. Ie:
trait Nat {
val i: Integer
type I = i.type
}
type ExpectedType = _1
trait _1 extends Nat {
override val i: Integer = 1
}
def getFoo(i: Integer)
(implicit ev: Nat{type I = i.type } =:= ExpectedType)= ???
getFoo(1) //this fails to prove the =:= implicit.
On reflection, I shouldn't have expected it to. Since if we have:
val x : Integer = 1
val y : Integer = 1
type X = x.type
type Y = y.type
def foo(implicit ev: X =:= Y) = 123
foo //would fail to compile.
I.e. the singleton types of the different "objects" with the same values is different. (I guess the reason is that currently in Scala the singleton types are for objects and are distinct from literal type)
So with this background information, I would like to know if there is any way to achieve what I am trying to do, namely inferring the a type from an associated value, through other methods.
回答1:
The answer seems to be just "No". The values exist at runtime. The type checking happens at compile time. These two time intervals do not intersect, the runtime always comes strictly after the compile time, so there is no way to propagate the information about a value back in time to obtain some additional information about a type.
Note that if an ordering is all you want (you don't want to add or subtract version numbers), then you can simply reuse the subtyping relation as follows:
sealed trait Num
class _9 extends Num
class _8 extends _9
class _7 extends _8
class _6 extends _7
class _5 extends _6
class _4 extends _5
class _3 extends _4
class _2 extends _3
class _1 extends _2
class _0 extends _1
trait Version[+Major <: Num, +Minor <: Num]
println(implicitly[Version[_2, _4] =:= Version[_2, _4]])
println(implicitly[Version[_2, _3] <:< Version[_2, _4]])
回答2:
The trick is to realise that values can have type fields and that type information is available at compile time. With this in mind, we can define:
sealed trait NatEnum{
type Nat_ <:Nat
}
and define enum "values" for these types like:
object __0 extends NatEnum{ override type Nat_ = _0 }
object __1 extends NatEnum{ override type Nat_ = _1 }
object __2 extends NatEnum{ override type Nat_ = _2 }
object __3 extends NatEnum{ override type Nat_ = _3 }
and refactor getFoo
as below:
def getFoo(maj: NatEnum, min: NatEnum)(implicit
maj_check: FooClient.Major =:= maj.Nat_,
min_check: FooClient.Minor <:< min.Nat_
) = FooClient.foo
which we can test with:
getFoo(__2,__2) //compiles
getFoo(__1,__0)// doesn't compile
here is the updated version of the gists: simple and rigorous
来源:https://stackoverflow.com/questions/50762710/making-connections-between-types-and-values