Allow non-constant arbitrary functions for (Co)Kleisli, State, Func

[ceedubs]

Resolves #1605.

This is a replacement for #1606.

Now that Arbitrary instances for functions have been improved with the introduction of Cogen in scalacheck, we can pick up better arbitrary functions than the constant ones we were using.

[ceedubs]

Requesting a review from @non. As you are the one who introduced Cogen into Scalacheck, I'm interested in your input on the decision that was made in #1606 (review) to require implicit arbitrary instances for functions as opposed to requiring implicit Gen and Cogen instances and deriving the arbitrary functions ourselves.

[edmundnoble]

LGTM. Curious about @non's input.

[ceedubs]

Oops there's a compile error in the free module. I'll fix it.

[ceedubs]

@edmundnoble hmm actually I guess this compile error is one potential reason to favor the direct Cogen/Arbitrary implicit parameters as opposed to the function Arbitrary instance. In some cases the implicit search for the function Arbitrary instance fails while the Arbitrary[A] and Cogen[A] implicit search succeeds. I also kind of like the idea that conceptually the Cogen and Arbitrary instances are the fundamental things that we need to provide our instance and the fact that we get there through scalacheck's Function1 instance is an implementation detail, but I suppose one could argue this in the opposite direction.

Now I'm second-guessing this approach :\

[edmundnoble]

Huh. I can't see why that'd be the case, the implicit error that is. TBH I am more inclined toward the argument in the other direction; users can provide their own Arbitrary function instances that way, and the Function1 instance is what I'd see people choosing to use or not use. Having different Cogen and output Arbitrary instances in scope will already feed into the Function1 instance generator. Also... is there not an error in the Kleisli monoid in the tests? https://travis-ci.org/typelevel/cats/jobs/224346639#L1658

[ceedubs]

@edmundnoble hmm I guess that the implicit divergence only is happening in the scala.js build:

[info] Compiling 7 Scala sources to /home/travis/build/typelevel/cats/free/.js/target/scala-2.10/test-classes...
[error] /home/travis/build/typelevel/cats/free/src/test/scala/cats/free/FreeTTests.scala:211: diverging implicit expansion for type org.scalacheck.Arbitrary[Int => cats.Eval[(Int, A)]]
[error] starting with method arbContainer2 in trait ArbitraryLowPriority
[error]   implicit def intStateArb[A: Arbitrary]: Arbitrary[IntState[A]] = catsLawArbitraryForState[Int, A]
[error]                                                                                            ^
[error] one error found
[error] (freeJS/test:compileIncremental) Compilation failed

And yeah you are right, it is reporting a problem with the Kleisli Monoid. That's concerning. I'll try to take a look at this this evening (Eastern US time).

[ceedubs]

Closing in favor of #1666. @edmundnoble sorry I'm departing from your preferred approach, but I seem to need to in order for this to compile in the scala.js build 😬