Evidence for absence of implicit

alexandru · August 13, 2019, 7:00am

Given a type, I want an implicit that proves there is no implicit value in scope for it, e.g…

def notANumber[A](a: A)(implicit ev: Not[Numeric[A]]) = 
  println(s"Definitely not a number: $a")

Towards this purpose I used:

gist.github.com

https://gist.github.com/alexandru/11375046

evidence-for-absence-of-implicit.scala

// CONTEXT:  https://groups.google.com/forum/#!topic/scala-user/8YpX1VkIkDs

import scala.annotation.implicitNotFound
import scala.language.experimental.macros
import scala.reflect.macros.whitebox

/**
  * Macro-driven type for proving that an implicit parameter does not exist
  * in scope.
  */

This file has been truncated. show original

Question: can we do this without a macro?

I seem to remember that there exists a solution based just on implicits hackery, which works by creating a divergence in case the implicit value exists. And I think it’s included in Dotty’s standard lib. But my google-fu is failing me now. Can you help me find it?

Thanks,

Jasper-M · August 13, 2019, 7:25am

I think this is the pattern you’re looking for: https://github.com/Jasper-M/implicitlogic/blob/master/src/main/scala/implicitlogic/Not.scala

Here’s the dotty version, but I think it is actually implemented in the compiler, because something changed in dotty’s implicit search that makes this pattern no longer work.

andreak · August 13, 2019, 7:39am

Maybe this wil help you: https://milessabin.com/blog/2011/06/09/scala-union-types-curry-howard/

alexandru · August 13, 2019, 2:27pm

Thanks, that’s the stuff

nob · August 20, 2019, 7:35am

In one project I had the same problem to generate a type class when some Type T is not a tuple. In the end the solution was like here (making the implicit resolution ambigiuous), but in order to make IntelliJ happy, it was necessary to not rely on Macros or typeclasses relying on Macros.