typelevel/cats-effect

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

Fabio Labella

@SystemFw

e.g. imagine wanting to return different http responses on success or failure of an operation

or perhaps wanting to retry something when there's an error

like, I don't see why this model is confusing

you have

IO.unit.handleError(...) >> boom, which doesn't handle errors in boom

just like

try {
  thing
} catch {
   ...
}

boom

doesn't handle errors in boom

Arnau Abella

@monadplus

It's not confusing now at all

Fabio Labella

@SystemFw

ah, nice ;)

Arnau Abella

@monadplus

but I couldn't understand it the first time :S

so thanks

Fabio Labella

@SystemFw

heh, that's ok

Arnau Abella

@monadplus

How do you handle situation where you fork then cancelled (e.g. the resource is not available) but there is a join afterwards ?

Fabio Labella

@SystemFw

you are going to have to be a bit more specific

that use of the word "cancelled" doesn't necessarily map to what I would think in the context of a fork

i.e. it's unclear if you are talking about some logical notion of "I don't want to do this anymore", or actual fiber interruption

Arnau Abella

@monadplus

Fiber interruption (e.g. Fiber.cancel)

I mean I dont want to do that anymore as .cancel

Fabio Labella

@SystemFw

who cancels? the thing that would do the join otherwise or something else?

Arnau Abella

@monadplus

The thread you forked from with .start

Fabio Labella

@SystemFw

and who joins?

I think it would be better if you could sketch an actual example. This can be tricky so it's easier that way

Arnau Abella

@monadplus

  (for {
    fiber <- readFile(new File("data/lore.txt")).start
    _     <- IO.sleep(100.millis) *> fiber.cancel
    lore  <- fiber.join // blocks
  } yield lore).map { println(_) }.unsafeRunSync()

@SystemFw

Fabio Labella

@SystemFw

well, that's kinda contrived given that this specific control flow doesn't really make a lot of sense, you alwasy join after cancel has happened

there are three things I'd say

if you just want to put a timeout on something

you should just readFile.timeout(5.seconds) (assuming that your readfile implementation is cancelable)

number 2 would be to only call .join if you haven't called cancel. In the simplest case this is an if, in more complicated cases you might need some coordination between the interrupter and the joiner

number 3 would be to put the result in a Ref + Deferred thing (in this case a simple Deferred), which has some form of default, this simplest case of which would be Deferred[Option[Thing]]

then you need to bracketCase or guaranteeCase the started operation so that it puts None on that Deferred when it gets cancelled

thoughts?

Arnau Abella

@monadplus

All three points are clear

nothing to add

John A. De Goes

@jdegoes

@monadplus Instead of fiber.join, call fiber.await. Join says, "the outcome of the fiber will be my outcome", await says, "just let me know what happened".

Arnau Abella

@monadplus

1.5.0 cats.effect.fiber does only expose cancel and join

Fabio Labella

@SystemFw

a quickly sketched prototype might look like this (requires testing ofc)

def await[F[_]: Concurrent, A](fa: F[A]): F[(F[Option[Either[Throwable, A]]], F[Unit])] =
  Deferred[F, Option[Either[Throwable, A]]].flatMap { result =>
       val action = {
          fa.attempt.flatMap(r => result.complete(r.some).uncancelable)
       }.guaranteeCase {
          case ExitCase.Canceled => result.complete(None)
          case _ => ().pure[F]
      }

      action.start.bracketCase(fiber => result.get.pure[F] -> fiber.cancel){
         case (fiber, ExitCase.Canceled) => fiber.cancel
         case (_, _) => ().pure[F]
      }
    }

obviously with a proper datatype instead of that monstrosity, and after testing (this isn't really the sort of code you want to sketch on gitter in 5 mins)

John A. De Goes

@jdegoes

@monadplus I sketched it in the design for Cats Effect 2.0. Until then you can get it in ZIO or with a patch like the above ^^^

Jakub Kozłowski

@kubukoz

I'm trying to implement Async for slick's DBIO, not passing one law now (bracket release is called on Completed or Error)... can someone roughly explain what the implementation needs to care about in the context of that law?

I have a rough idea that it means release should not be possible to cancel, but I'm not sure how to ensure that

Gavin Bisesi

@Daenyth

@kubukoz how did you encode Sync?

Jakub Kozłowski

@kubukoz

duude

you're going to love this

Gavin Bisesi

@Daenyth

the from(future) is strict in its argument

so I couldn't figure how to do Sync

Jakub Kozłowski

@kubukoz

behold delay(thunk): DBIO.successful(()).flatMap(_ => thunk)

Gavin Bisesi

@Daenyth

:ok_hand:

Jakub Kozłowski

@kubukoz

and it's even stack safe since 3.2.2

I'll try to get Async soon, and in any case submit something to slickcats

Where communities thrive

People

Repo info

Activity