SemigroupK[IO] about cats-effect HOT 11 CLOSED

:) I feel like the expectation mismatch leads to https://github.com/quchen/articles/blob/master/monad_fail.md.

Anyway, I'm 👍 on a SemigroupK[IO] as proposed.

from cats-effect.

  implicit val IOPlus: SemigroupK[IO] = new SemigroupK[IO] {
    def combineK[A](x: IO[A], y: IO[A]): IO[A] = x.handleErrorWith(_ => y)
  }

from cats-effect.

There was a horrid suggestion on the gitter channel to provide an Alternative by returning a failed IO as empty. I guess that works.

from cats-effect.

For Strings the concatenation operation makes the canonical Semigroup / Monoid, same for Lists.

Not having a problem with adding a SemigroupK[IO], however given that we try to be coherent in our type-class definitions, I'm not sure what makes x.handleErrorWith(_ => y) canonical. It's not very obvious to me, especially the IO[A] values aren't both evaluated.

A more obvious definition for me would be something that evaluates both:

x.flatMap(_ => y)

x.flatMap(x => y.map(_ => x))

But maybe this isn't obvious enough either. More obvious is Semigroup[IO[A]] for any A : Semigroup[A], which is already defined.

So I'm not sure how to help. Somebody more knowledgeable with this stuff needs to comment.

from cats-effect.

Also, I'm not sure if my expectation for evaluating both helps, since you want "OR logic", with short-circuiting, so I'm assuming that evaluating both is precisely what you don't want?

I guess that what I'm asking is why would anybody expect x.handleErrorWith(_ => y) to be the implementation of a SemigroupK[IO]?

Is there some theory or precedent in other libraries?

from cats-effect.

Plus (or as cats have renamed it, SemigroupK) has "pick the first winner, ignore errors" semantics. Although it is not expressed in its laws.

from cats-effect.

Semigroup comes from category theory, I'm only familiar with the laws I learned in my 12th grade — kids in Romania learn about semigroups and monoids in high-school, guess we got lucky 😀

Those semantics are interesting. If that's the case, guess it makes sense, but are there examples of such instances in other libraries? Cats, Scalaz, Haskell, PureScript? If this can be expected given prior work, then we could have a SemigroupK[IO] instance.

I know that issues aren't for tutoring, but I saw this issue lingering since a week ago, so I'm asking in absence of somebody else more knowledgeable confirming this.

from cats-effect.

Plus (or as cats have renamed it, SemigroupK) has "pick the first winner, ignore errors" semantics. Although it is not expressed in its laws.

This is a property of certain lawful instances but not of the type class itself. The type class only defines a universally quantified associative combine operation. For example, consider MonoidK[List].

from cats-effect.

Here's another example which is more fundamental:

implicit def semigroupKConst[A](implicit A: Semigroup[A]): SemigroupK[Const[A,?]] = new SemigroupK[Const[A,?]] {
  def combineK[B](x: Const[A,B], y: Const[A,B]): Const[A,B] = Const(A.combine(x.getConst, y.getConst))
}

from cats-effect.

I thought the "pick the first winner, ignore errors" semantics might be expected whenever there is an ApplicativeError to define a "winner". Not always:

scala> def law[F[_]](implicit AE: ApplicativeError[F, String], SK: SemigroupK[F], A: Arbitrary[F[Int]]) = forAll { (e1: F[Int], e2: F[Int]) => e1.handleErrorWith(_ => e2) == e1 <+> e2 }.check
law: [F[_]](implicit AE: cats.ApplicativeError[F,String], implicit SK: cats.SemigroupK[F], implicit A: org.scalacheck.Arbitrary[F[Int]])Unit

scala> law[Either[String, ?]]
+ OK, passed 100 tests.

scala> law[EitherT[List, String, ?]]
+ OK, passed 100 tests.

scala> law[OptionT[Either[String, ?], ?]]
! Falsified after 0 passed tests.
> ARG_0: OptionT(Right(None))
> ARG_1: OptionT(Left())

from cats-effect.

well, yes, I guess it's more like "pick the first winner, ignoring failures, unless there are no winners. then we all lose"

from cats-effect.