typelevel/cats-effect

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

Jakub Kozłowski

@kubukoz

you can make sure it starts (and cancels immediately afterwards) with a promise

Deferred[IO, Unit].flatMap { deff =>
  mod(deff.complete(()) *> io).start.flatMap(deff.get *> _.cancel)
}

The default implementation of uncancelable will have cancel wait for the effect

IO's impl overrides that to make cancel a no-op

I was experimenting just now, see https://scastie.scala-lang.org/cTQWIR8mRPOT0m4VPwRuRA

sadly the scastie isn't consistent with master since your change ;)

local scores

0ms: good: Sleep completed
7ms: good: Cancel completed

0ms: good alt: Sleep completed
1ms: good alt: Cancel completed

0ms: bad: Cancel completed
5003ms: bad: Sleep completed

(low diff means cancel kept waiting)

Jakub Kozłowski

@kubukoz

here's a polymorphic version in case you want to give other effects a try https://scastie.scala-lang.org/gL9rkl9rQp2v2fgYOV88Ew

Piotr Gawryś

@Avasil

I thought you meant something like this

  val io = for {
    pa <- Deferred[IO, Unit]
    fiber <- (pa.complete(()) >> IO(Thread.sleep(5000))).start
    _ <- pa.get
    _ <- fiber.cancel
  } yield ()

(no bracket)

and nevermind, it seems to be waiting for zio

Jakub Kozłowski

@kubukoz

well, the bracket in the examples was just to show the difference between the standard and IO's impls

the easiest thing to do to fix IO's uncancelable would be to... stop overriding the default :D

it'll reduce the amount of code, too

the whole IOCancel seems to be redundant then

I'll submit a PR with some tests and we'll see where it goes

Jakub Kozłowski

@kubukoz

hmm are law tests broken?

  def uncancelableIsDerivedFromBracket[A](fa: F[A]) =
    F.uncancelable(fa) <-> fa

this passes for IO...

Piotr Gawryś

@Avasil

Can uncancelable be written in terms of bracket if Bracket's implementation is using it?

Jakub Kozłowski

@kubukoz

if the implementation is indeed using it, then it's a problem...

and I can see it does

I'm out of my depth here...

and someone just messed in that code! ;)

Piotr Gawryś

@Avasil

It's a bit tricky :P I want to trim my backlog first and then I could try to attempt this but I feel like it will require major changes in cancelation implementation

Jakub Kozłowski

@kubukoz

yeah... and honestly the cancelation code feels quite complex

Jules Ivanic

@guizmaii

Hi everyone :)
Happy New Year :tada:

Daniel Spiewak

@djspiewak

🎉

Jules Ivanic

@guizmaii

I have a question.about the diff between Sync[F].pure and Sync[F].delay.

Fabio Labella

@SystemFw

shoot

Jules Ivanic

@guizmaii

In my head, the first one should only be used with values

Fabio Labella

@SystemFw

yep

Jules Ivanic

@guizmaii

and the second one should be used to purify a function call

Fabio Labella

@SystemFw

that's not the main difference (talking about the function call)

well, it depends on your definition of "value"

do you mean val, or something else?

Jules Ivanic

@guizmaii

so something like that is a bug: Sync[F].pure(logger.info(“…"))

Am I right?

Fabio Labella

@SystemFw

that's right

Jules Ivanic

@guizmaii

Hum. Maybe the best definition of a value I can give you is that it’s something already computed

Fabio Labella

@SystemFw

I would define a value as a referentially transparent expression. Applying pure to a value yields another value (another referentially transparent expression). Applying pure to a non-value yields a non-value. delay is magic because you can apply it to a non-value and get a value out

does that make any sense?

Jules Ivanic

@guizmaii

Yes, it does. Totally

if something like that Sync[F].pure(logger.info(“…")) is used, the logger will only be called once, right?

Fabio Labella

@SystemFw

informally yes. Slightly more formally, the semantics of execution are now linked to the semantics of evaluation

i.e. now performing effects depend on whether you have a def or val, normal parameters vs by-name, etc

Jules Ivanic

@guizmaii

now performing effects depend on whether you have a def or val

Not sure to understand that part

🤔

Adam Rosien

@arosien_twitter

a val is a value, but a def is something that produces a value (when it is evaluated)

Jules Ivanic

@guizmaii

Can’t we consider def a = 1 as a value?

Adam Rosien

@arosien_twitter

in what way?

like, is a a value? or the literal expression you wrote?

Gavin Bisesi

@Daenyth

@guizmaii It's a method that returns a value. Coincidentally, it returns the same value every time you call it

Consider the difference between def x = Random.nextInt vs val b = Random.nextInt when you see x + x vs b + b

Where communities thrive

People

Repo info

Activity