typelevel/cats-effect

Most discussion is on Typelevel Discord: https://discord.gg/bSQBZA3Ced

Adam Rosien

@arosien

they are awesome!

Luis Rodero-Merino

@lrodero

Basically what I'm saying is it's hard to write tutorials :-)

Yes, absolutely :) . So, ok, agreed. @marcraminv and I will try to work on it. Thanks for the feedback!

rnd4222

@rnd4222_gitlab

Do you guys use explicit Kleisli in a "tagless-final" code, or hide it behind another parametric type, having (say) F and G for a types with and without context?

I'm aware of cats-mtl, but what if you need to eventually provide a context and convert reader-like G to a non-reader F?

Adam Rosien

@arosien

@rnd4222_gitlab like

trait Train[F[_]] {
  def toot(): F[Unit]
}

object Train {
  def apply: Train[Kleisli[IO, Whistle]] = ???
}

vs.

trait Train[F[_]] {
  def toot(): Kleisli[F, Whistle, Unit]
}

i usually use the former, and usually add

trait Train[F[_]] {
  def toot(): F[Unit]
  def mapK[G[_]](f: F ~> G): Train[G]
}

(cats-tagless can auto-generate these also)

rnd4222

@rnd4222_gitlab

@arosien more like

trait Train[F[_]]
trait Sound[F[_]]
class Whistle[F[_] : ApplicativeAsk[*[_], Volume], G[_]]

Adam Rosien

@arosien

does Whistle[F, G] extends Sound[F]?

sorry, i must be missing something

there's using Kleisli vs. ApplicativeAsk, that's one choice

rnd4222

@rnd4222_gitlab

I'm now working in a project where "tagless-final-to-the-max" approach was taken.
There is always different types for Fs with different contexts.
Lots of custom Lift/Unlift classes for conversion back and forth.
Oh, and it uses doobie, so add another two types - one is instantiated to ConnectionIO and second to Kleisli[ConnectionIO, ...], both with its own lifts-unlifts

So I wanted to know if its usual approach or not

Adam Rosien

@arosien

hmm. so for the impls/interpreters of those algebras there isn't 1 type like IO that's used for F, there are multiple, so you end up translating a lot?

is there some intermediate type like class BizLogic[F[_]](alg1: Alg1[F], alg2[Kleisli[F, Dependency]])? because something like that would be difficult.

or at least more complicated than it probably should be. but i'm just guessing because i'm not sure of the structure you are describing.

rnd4222

@rnd4222_gitlab

It's usually in a places where some context is added/eliminated, for example in HttpRoutes or SomethingRepository

typical signature is

def withRequestId[F[_], G[_] : WithRun[*[_], F, RequestId[F]](routes: HttpRoutes[F]): HttpRoutes[G]

Adam Rosien

@arosien

i definitely know that there are good reasons to need to lift/unlift between types, especially with Kleisli. but where that happens is the question.

rnd4222

@rnd4222_gitlab

I'm thinking that using Kleisli in this example would be much better:

def withRequestId[F[_], G[_] : WithRun[*[_], F, RequestId[F]](routes: HttpRoutes[F]): HttpRoutes[G]

// vs

def withRequestId[F[_]](routes: HttpRoutes[F]): HttpRoutes[Kleisli[F, RequestId[F], *]

Adam Rosien

@arosien

yeah that signature makes sense, sort of.

i see now what you're saying about monad transformers vs. typeclass constraints.

the former is more flexible, although it is likely that only Kleisli has a WithRun instance i'm guessing.

it's a familiarity issue, like you said. i think more folks would understand the Kleisli version since it is more "concrete"

Adam Rosien

@arosien

but i can see how some business logic is going to depend on [F[_]: WithRun[...]](routes: HttpRoutes[F]) which makes the context bound version more congruent

so... i feel ya.

Georgi Krastev

@joroKr21

It depends. With ApplicativeAsk it was easy to migrate from Kleisli to Monix TaskLocal. But it's better to have higher level interfaces like Logger. When converting to routes it sounds like a place to use concrete types though.

Adam Rosien

@arosien

there's the evalMap(f: A => F[B]): Resource[F, A] => Resource[F, B] = _.flatMap(a => Resource.liftF(f(a))) combinator that lifts the output of f into an "inner" Resource.

what about a different combinator that uses the first Resource to create the next, where the first can be released: evalUse(f: A => F[B]): Resource[F, A] => Resource[F, B] = rfa => Resource.liftF(rfa.use(f))?

or is there a better way?

Fabio Labella

@SystemFw

that's it

Keir Lawson

@keirlawson

Is there a nice pre-existing fake impl of Clock for use with tests? Obviously I can make my own but figured there might be something already?

Fabio Labella

@SystemFw

TestContext

Gavin Bisesi

@Daenyth

@keirlawson example integration with scalatest here https://gist.github.com/Daenyth/67575575b5c1acc1d6ea100aae05b3a9

Adam Rosien

@arosien

very quiet today in the gitterverse.

Michael Pilquist

@mpilquist

@arosien_twitter you’ve gotta hang out in cats-effect-dev channel

Adam Rosien

@arosien

Fabio Labella

@SystemFw

yeah I think this channel is going to be for user questions (which I think is a good thing)

Aleksander Sumowski

@aleksandersumowski

hi all,
what would be a good way to extend IO.race to more then 2 tasks?

Fabio Labella

@SystemFw

@aleksandersumowski do they all return the same thing?

things of the same type I mean

Aleksander Sumowski

@aleksandersumowski

yes

This message was deleted

Fabio Labella

@SystemFw

@aleksandersumowski this should work

def multiRace[F[_]: Concurrent, A](fas: List[F[A]]): F[A] = {
  def spawn[B](fa: F[B]): Resource[F, Unit] =
    Resource.make(fa.start)(_.cancel).void

  def finish(fa: F[A], d: Deferred[F, Either[Throwable, A]]): F[Unit] =
    fa.attempt.flatMap(d.complete)

  Deferred[F, Either[Throwable, A]]
    .flatMap { result =>
      fas
        .traverse(fa => spawn(finish(fa, result)))
        .use(_ => result.get.rethrow)
    }
}

def program(i: Int) = {
  for {
    wait <- IO(scala.util.Random.nextInt(1000))
    _ <- IO(println(s"program $i waiting for $wait millis"))
    _ <- IO.sleep(wait.millis)
    _ <- IO(println(s"program $i finished"))
  } yield i
}.guarantee(IO(println(s"program $i finalised")))

def test = multiRace(List.range(0, 5).map(program)).unsafeRunSync

Aleksander Sumowski

@aleksandersumowski

thanks a lot!

Daniel Ochoa Rodríguez

@tzimisce012

About this PR: typelevel/cats-effect#365 Does it miss ConcurrentEffectfor IorT or am I missing something?

Vasil Vasilev

@vasilmkd

surprise, it was traverse

Daniel Ochoa Rodríguez

@tzimisce012

Anton Sviridov

@keynmol

just a running gag about .traverse being the answer to pretty much any cats-effect question online :)

Paul Snively

@paul-snively

It's really interesting how often questions reduce to "How do I turn a Traverse of A into a B[Traverse] given an A => F[B]? And enough Fs form Applicatives that the answer really is traverse.

Anton Sviridov

@keynmol

I actually wonder if Future.sequence pops up with same frequency. May be it's because it's part of a smaller public Future interface that it's discovered more easily

Daniel Ochoa Rodríguez

@tzimisce012

oh, it was related with the previous question, I was out of context

Vasil Vasilev

@vasilmkd

ah, sorry for the confusion :)

Where communities thrive

People

Repo info

Activity