Confusing behavior of Monoid[Xor[A, B]]#combine

[notxcain]

Don’t you feel that it’s strange that Monoid[Xor[A, B]] requires A: Semigroup?
I expected combine to be implemented like this

def combine(x: A Xor B, y: A Xor B): A Xor B =
  x.flatMap { xb =>
    y.map { yb => Monoid[B].combine(xb, yb) }
  }

In current implementation it also tries to combine Lefts

[benhutchison]

It seems that two lefts combine, and two right combine, but a left and a right combined is a left. I think that is consistent with Scalaz, eg [https://gist.github.com/travisbrown/3732844]. It's an interpretation where a failure of a part fails the combined group.

But I can definitely imagine cases where you'd want to opposite, to discard failures. Like a lot of types, there are several monoids for Xor and one gets blessed as the default. But IMO its a mistake to think of it as the monoid for Xor.

Welcome to the Ammonite Repl 0.5.2
(Scala 2.11.7 Java 1.8.0_51)
@ load.ivy("org.typelevel" %% "cats" % "0.4.1")
@ import cats._
import cats._
@ import cats.data._
import cats.data._
@ import cats.syntax.all._
import cats.syntax.all._
@ val left1 = Xor.left[String, Int]("fail1")
left1: Xor[String, Int] = Left(fail1)
@ val right = Xor.right[String, Int](7)
right: Xor[String, Int] = Right(7)
@ import cats.std.all._
import cats.std.all._
@ left1 |+| left1
res7: Xor[String, Int] = Left(fail1fail1)
@ left1 |+| right
res8: Xor[String, Int] = Left(fail1)
@ right |+| right
res9: Xor[String, Int] = Right(14)

[ceedubs]

@notxcain sorry I missed this issue when you first opened it, but I was thinking the same thing the other day. I would expect the errors to combine for Validated but not for Xor.

There was a conversation about this on Gitter yesterday.