I think I’m encountering a bug in `Resource.parZip...

# arrow

[phldavies]

I think I’m encountering a bug in Resource.parZip when `parZip`ing a `parZip`’d resource 🙂

suspend fun main() = int(1)
    .parZip(int(2), Int::plus)
    .parZip(int(3), Int::times)
    .use { println("using $it") }

private fun int(n: Int) = resource {
    n.also { println("acquiring $it") }
} release {
    println("releasing $it")
}

outputs

acquiring 2
acquiring 1
acquiring 3
using 9
releasing 3

and never releases 1 and 2

[simon.vergauwen]

Ouch, that is strange..
Resource#parZip was rewritten to be build on top of the resource { } DSL in 1.1.0.

Can you give that a shot there?

[phldavies]

same code (including deprecation) on 1.1.2 results in this:

acquiring 2
acquiring 1
acquiring 3
using 9
releasing 1
releasing 3
releasing 2

[simon.vergauwen]

That looks better 😄

[phldavies]

indeed - I was waiting for the release notes / change-log for 1.1.x before making the switch. I came across this when looking to write a Iterable<A>.parTraverseResource method but can get away with traverseResource for now to unblock my task and look into an upgrade to 1.1.x next 😉

[simon.vergauwen]

You can safely do this;

[simon.vergauwen]

resource {
  listOf(..).parTraverse {
    resource(it).bind()
  }
}

For doing parTraverseResource

[phldavies]

With the new resource { } DSL, sure - that’ll be my go-to replacement once I’ve made the jump. 🙏

[simon.vergauwen]

If you look at the new implementation of Resource#parZip it's also simply resource DSL + the regular parZip operator.

The goal of Arrow is to make all its functional DSLs and constructs interchangeable.

[phldavies]

I can definitely see the move to that with the effect { } block 🙂 - I assume most/all usage of the old computation blocks will still work in 11.x but I’d rather switch them all to effect as part of upgrading

[simon.vergauwen]

1.1.x has nothing weird or special that is required for the transition.

TL;DR it adds resource DSL and a better abstraction for either {}. Which you can migrate by finding and replacing arrow.core.computation to arrow.core.continuation and the API usage will not break anywere.

[simon.vergauwen]

Yes, there is no breakage anywhere in the API and you can thus safely just find and replace the package.

[simon.vergauwen]

We ran into some weird issues with Github Actions that we need to figure out, thee release post will happen asap.

[phldavies]

I did notice that Resource.tap is broken (removed the unnecessary <B> type argument, which was seemingly always required to invoke it) I had a few .tap<Nothing> { … } that would need updating.

[simon.vergauwen]

Ah yes, that indeed needs updating

[simon.vergauwen]

There was no way for us to fix that in a non-breaking way 😞

At least the binary is not breaking, but the Kotlin API is.

It requires you to indeed manually change from .tap<Nothing> { to .tap {

[simon.vergauwen]

So at least it's still backwards compatible

[phldavies]

I wonder if something like

@Deprecated(message = "Use `tap(f)`", replaceWith = ReplaceWith(expression = "tap(f)"))
  public inline fun <B> tap(marker: Nothing? = null, f: suspend (A) -> Unit): Resource<A> = tap(f)

would work for maintaining source compatibility without impacting binary compatibility 🤔

[phldavies]

wouldn’t work for .tap<Nothing>({ … }) I suppose.