Implement lens generation macros

[liufengyun]

The following macros are used to cut down boilerplate when generating lens.

• GenLens
• GenPrism
• GenIso
  • 0 field
  • 1 field
  • n fields (blocked by missing support for whitebox macros)

Links

• http://julien-truffaut.github.io/Monocle/optics/prism.html#prism-generation
• http://julien-truffaut.github.io/Monocle/optics/lens.html#lens-generation
• https://github.com/julien-truffaut/Monocle/tree/master/macro/shared/src/main/scala/monocle/macros

Thanks to @milessabin for refering the macros to us.

[julien-truffaut]

Thanks for looking into this! Another macro I wanted to implement but never got the time is GenLens equivalent for common fields of Coproduct e.g.

sealed trait Foo
case class Foo1(name: String, i: Int) extends Foo
case class Foo2(name: String, b: Boolean) extends Foo

val name = GenLens[Foo](_.name)

but it would fail if they were a 3 implementation of Foo like:

case class Foo3(x: Int) extends Foo

[milessabin]

I'm pretty sure the answer is "yes", but can you confirm for me that this generalizes to nested selectors, ie. _.address.street?

[julien-truffaut]

good question, it would be awesome if it was.

[milessabin]

It's pattern matching on the Tasty trees, so I think it could be generalized and Just Work, but it'd be nice to be sure.

[liufengyun]

@julien-truffaut There is some difficulty to implement the Coproduct lens in Dotty macros:

Dotty macros should type check without expansion.

Even for the non-Coproduct GenLens macro, programmers have to write GenLens[Foo, String](_.name) in Dotty instead of GenLens[Foo](_.name) in Scala2.

Nested selectors _.address.street should be able to be supported.

[milessabin]

Could you get it to typecheck by having the target type inferred as Nothing in the unexpanded context?

[liufengyun]

@milessabin The problem is that given the following macro signature:

object GenLens {
    inline def apply[S, T](get: S => T): Lens[S, T] = ~impl('(get))
}

Dotty does not support only supplying S:

5 |    val len = GenLens[Address](_.streetNumber)
  |              ^^^^^^^^^^^^^^^^^
  |              Not enough type arguments for GenLens.apply[S, T]
  |              expected: [S, T]
  |              actual:   [Address]

[liufengyun]

@nicolasstucki Any thoughts about this?

[milessabin]

In vanilla Scala the usual trick is something like,

object GenLens {
  def apply[S] = new MkGenLens[S]
  class MkGenLens[S] {
    def apply[T](get: S => T): Lens[S, T] = ...
  }
}

Then S can be explicit and T inferred. Could you do something like that here?

[liufengyun]

@milessabin Thanks a lot for sharing the trick, it works like a charm 👍 73524a7

[nafg]

I thought Dotty was supposed to solve those issues

…

On Mon, Feb 18, 2019, 4:15 PM Fengyun Liu ***@***.***> wrote: @milessabin <https://github.com/milessabin> Thanks a lot for sharing the trick, it works like a charm 👍 73524a7 <73524a7> — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#5941 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAGAUJm3K3LFbyhjTf6kcNGbY5pwlETaks5vOxftgaJpZM4bAlx5> .

[smarter]

Dotty does not support only supplying S:

You can write:

val len = GenLens[S=Address](_.streetNumber)

cf http://dotty.epfl.ch/docs/reference/other-new-features/named-typeargs.html

[milessabin]

You can write

Yeah, but that's really quite awkward in this sort of context (which is really quite common). It's a shame ... we ought to be able to do better.

This works with poly functions,

trait Lens[S, T]

def GenLens[S] = [T] -> (get: S => T) => new Lens[S, T] {}

case class Foo(i: Int)

object Test {
  GenLens[Foo](_.i)
}

But I think it would be nicer if we could support multiple type parameter lists, ie.,

def GenLens[S][T](get: S => T) = new Lens[S, T] {}

[smarter]

Make a SIP :).

[milessabin]

Make a SIP :)

With a PR attached ;-)

[liufengyun]

The GenPrism and GenLens macros are implemented in #5944.

Unfortunately, we are unable to implement GenIso.fields due to missing support for whitebox macros.

Code examples:

    val len2 = GenLens[Employee](_.addr.streetNumber)
    val employee = Employee("Bob", Address(10, "High Street"))
    assert(len2.get(employee) == 10)
    val employee2 = len2.set(5, employee)
    assert(employee2.name == "Bob")
    assert(len2.get(employee2) == 5)

    val jNum: Prism[Json, Double] = GenPrism[Json, JNum] composeIso GenIso[JNum, Double]
    assert(jNum(3.5) == JNum(3.5))
    assert(jNum.getOption(JNum(3.5)) == Some(3.5))
    assert(jNum.getOption(JNull) == None)

The implementation of GenIso[JNum, Double] is supposed to be more robust than the Scala2 implementation, as it handles the case where the case classes are inner classes.

[milessabin]

@liufengyun the work I'm doing on generics will cover GenIso.

[liufengyun]

It would be nice to implement GenIso.fields based on @milessabin 's work. Could you please let us know how to do that once your work is merged @milessabin ?

[julien-truffaut]

I started the implementation of Monocle for Dotty in the following repo: https://github.com/julien-truffaut/Monocly

It would be amazing if any of you would like to participate or point me to some documentation. I tried to adapt the macro in https://github.com/lampepfl/dotty/blob/master/tests/run-macros/i5941/macro_1.scala but I am running into some issues. It is probably because I am using polymorphic optics (4 or 5 type parameters) and a curried set method.

[liufengyun]

@julien-truffaut The following is the latest prototype implementation in Dotty https://github.com/lampepfl/dotty/blob/master/tests/run-macros/i5941/macro_1.scala . The documentation and the paper may also be useful.

In your prototype, I think you should remove the toolbox line, as they are not intended for macros.

  def impl[A: Type, B: Type](getter: Expr[A => B])(given qctx: QuoteContext): Expr[Lens[A, B]] = {
    implicit val toolbox: scala.quoted.staging.Toolbox = scala.quoted.staging.Toolbox.make(this.getClass.getClassLoader)
    import qctx.tasty.{_, given}