Remove redundant Monad constraints from Parallel syntax

[joroKr21]

Binary compatible version of #2180
Fixes #2254

[joroKr21]

Not sure if this should be 2.5.1 or 2.6.0

[joroKr21]

Since this is not working, the other option is to create new AnyVals - should I go for it?

[tnielens]

Would something along this line be binary compatible?

final class ParallelTraversableOps[T[_], A](private val ta: T[A]) extends AnyVal {
  def parTraverse[M[_]: Monad, B](f: A => M[B])(implicit T: Traverse[T], P: Parallel[M]): M[T[B]] =
    Parallel.parTraverse(ta)(f)
}

final class ParallelTraversableOps2[T[_], A](private val ta: T[A]) extends AnyVal {
  def parTraverse[M[_], B](f: A => M[B])(implicit T: Traverse[T], P: Parallel[M]): M[T[B]] =
    Parallel.parTraverse(ta)(f)
}

[joroKr21]

Would something along this line be binary compatible?

Yes, that's what I meant. Unfortunately adding an overload in Scala 2.12 changes the name mangling of extension methods.
There is also an opportunity to consolidate extension methods (if we drop the constraints from the implicit conversions).

[tnielens]

shouldn't this have a second third parameter list instead?

[joroKr21]

WDYM? 🤔

[tnielens]

You moved the monad constraint in the first position of the implicit parameter list. My understanding of context bounds is that a new parameter list is appended to the method definition in this way:

   def parTraverse[M[_]: Monad, B](f: A => M[B])(implicit T: Traverse[T], P: Parallel[M]): M[T[B]] = 
   // is equivalent to
    def parTraverse[M[_], B](f: A => M[B])(implicit T: Traverse[T], P: Parallel[M])(implicit M: Monad[M]): M[T[B]] = 

[joroKr21]

Nope, in Scala 2 you can have only one implicit list and the context bounds come first.

[tnielens]

Didn't know that, thanks for the explanation.

[joroKr21]

But in the bytecode there is only one parameter list anyway (and no implicits) which is why both are binary compatible:

def parTraverse[M[_]: Monad, B](f: A => M[B])(implicit T: Traverse[T], P: Parallel[M]): M[T[B]]
def parTraverse[M[_], B](f: A => M[B], M: Monad[M], T: Traverse[T], P: Parallel[M]): M[T[B]]

That's a neat trick for library authors 😄

[larsrh]

LGTM, but I think this warrants a bump to 2.6.0