https://kotlinlang.org logo
Channels
Powered by
#arrow
  • d

    Derek Berner

    06/04/2019, 3:05 PM
    which is an effect
  • d

    Derek Berner

    06/04/2019, 3:06 PM
    Bugger. So, the JVM is the problem!
  • b

    Bob Glamm

    06/04/2019, 3:06 PM
    Even Haskell has 
    _|_
  • b

    Bob Glamm

    06/04/2019, 3:07 PM
    This conversation is wandering into Turing machine territory
  • b

    Bob Glamm

    06/04/2019, 3:08 PM
    and decidability
  • b

    Bob Glamm

    06/04/2019, 3:09 PM
    maybe think of it as an engineering compromise for computing specific, well-bounded problems deterministically, most of the time
  • d

    Derek Berner

    06/04/2019, 3:09 PM
    Is this semantics at this point? All heap allocation by definition requires a system call to reserve a location in RAM and return the pointer
  • r

    raulraja

    06/04/2019, 3:09 PM
    effects is what you want to track and exceptions is one Arrow cares about tracking
  • r

    raulraja

    06/04/2019, 3:10 PM
    we don't care for example about allocations which is also an effect
  • s

    simon.vergauwen

    06/04/2019, 3:10 PM
    We only care about allocation when it's mutable state like 
    Ref
    👍 1
  • d

    Derek Berner

    06/04/2019, 3:12 PM
    So that means, everything should eventually terminate in IO, even if it's not effectful, so that we can guarantee runtime error safety?
  • s

    simon.vergauwen

    06/04/2019, 3:16 PM
    I'm not sure what you mean. Kotlins mpp main is the same haskells.
  • s

    simon.vergauwen

    06/04/2019, 3:16 PM
    suspend fun main(): Unit
    ~> 
    fun main(): IO<Unit>
  • d

    Derek Berner

    06/04/2019, 3:24 PM
    Well specifically, there are certain runtime errors that can theoretically be thrown by any code. So to reason about those errors, you can do that at the edge of your code.
  • s

    simon.vergauwen

    06/04/2019, 3:26 PM
    & 
    IO<E, A>
    will give you much better fine grained control over business errors and runtime stuff.
  • d

    Derek Berner

    06/05/2019, 2:27 PM
    Maybe I'm approaching this wrong, but my understanding of the use case for 
    ReaderT
    is an action that requires some external dependency. And then 
    local
    is used to narrow down a larger context of dependencies into a specific one.
  • d

    Derek Berner

    06/05/2019, 2:27 PM
    So if one effect needs a Redis client and another effect needs a JSON converter, I can use 
    local
    to compose those into an effect that requires both
  • d

    Derek Berner

    06/05/2019, 2:28 PM
    But - if that's right, I must be missing something because it's causing my type annotations to explode, type inference to fail utterly, and is requiring lots of messy transformative boilerplate.
  • d

    Derek Berner

    06/05/2019, 2:28 PM
    I have a meeting now but I'll try to post code later. The code in its current state is a hot mess.
  • b

    Bob Glamm

    06/05/2019, 2:30 PM
    If that's the case I really misunderstood 
    local
    . I thought 
    local
    was a way to modify the external dependency before working with it
  • d

    Derek Berner

    06/05/2019, 2:30 PM
    That's an orthogonal but not wrong way of looking at 
    local
    , but yeah, it's essentially a contravariant functor
  • d

    Derek Berner

    06/05/2019, 4:25 PM
    Do I need to use a separate repository for the 0.9.1 snapshot dependency?
  • p

    pakoito

    06/05/2019, 4:26 PM
    you shouldn't, although there are so many changes it'll be a pain to go back and forth
  • b

    Bob Glamm

    06/05/2019, 4:43 PM
    So here we do a lot of Spring web applications - REST controllers + service layer + repositories. If I wanted to add Arrow incrementally, I think I would start adding ops/syntax based on 
    MonadDefer<F>
    for the transactional pieces and instantiate to 
    IO
    using 
    unsafeRunSync
    in the controllers. Is that reasonable or is there a better pattern to use?
  • b

    Bob Glamm

    06/05/2019, 4:44 PM
    (I mean, ideally there is Spring library support for controllers in 
    IO
    but that isn't there yet as far as I'm aware)
  • p

    pakoito

    06/05/2019, 6:49 PM
    if you have other asynchronous calls you'll need 
    Async<F>
    , and if you want parallelism you need 
    Concurrent<F>
  • p

    pakoito

    06/05/2019, 6:49 PM
    otherwise it seems reasonable, yes
  • s

    simon.vergauwen

    06/05/2019, 6:58 PM
    In the next release there'll be an api to run an 
    IO
    as a suspend function so you can also use the Spring coroutine support.
  • r

    raulraja

    06/05/2019, 6:58 PM
    @Bob Glamm you can also just use IO since in 0.10 it will be bifunctor and it will have a means to inject dependencies
  • s

    simon.vergauwen

    06/05/2019, 6:59 PM
    Or any 
    F
    if you prefer polymorphic code :)
1...313314315...416Latest