I’m playing around with kotlin context receivers a...

# arrow

[Erik Dreyer]

I’m playing around with kotlin context receivers and Arrow’s EffectScope by migrating some existing code. I’m having a bit of trouble getting something to work. See the screenshot below.

I think I have an instance of Result<E>, and the compiler agrees, but at runtime, I’m seeing some odd and conflicting information. It seems in reality, I have a Result<Result<E>>

I’ve tried to figure out how we ended up with nested Result, but haven’t been able to see where this is happening.

[Wilerson Oliveira]

val bar = it.invoke(command) as E

I would guess this casting is the culprit. Wouldn’t val bar = it.invoke(command).bind() work?

[simon.vergauwen]

I was going to say the same, there is a good chance that as E passes through the Success(E) because the Success case is completely inlined.

Can you debug what it.invoke(command) returns?

[Erik Dreyer]

Ahh, I’ll try. I get a compile error without the casting, but not sure why.

[Erik Dreyer]

The signature of that invoke method:

context(ResultEffectScope)
  suspend fun invoke(request: R): E

[simon.vergauwen]

Maybe a rewrite involving less inline code.

result {
  val foo = ensureNotNull(commandHandlers[command::class]) { IllegalStateException("...") }
  foo.invoke(command)
}

[simon.vergauwen]

That's also a much shorter, an clearer encoding IMO.

[Erik Dreyer]

oh, nice. Thx!

[Erik Dreyer]

@simon.vergauwen, btw, just found your FP course for Kotlin. I’m looking forward to taking that

[simon.vergauwen]

This one? https://academy.47deg.com/courses/functional-programming-in-kotlin-and-arrow

I'm not the trainer in the videos, but collaborated on the content. We'll also be working on new courses to add over time. If you have any questions about the curriculum I'd be happy to clarify ☺️

[Erik Dreyer]

OK, I finally got back to this, and have been playing with the debugger. I’m pretty good with debugging, but I can’t make sense of this.

In the screenshot below I have a variable foo of type Result<UserRegisteredEvent>

The debugger shows me the variable is this same type.
But foo.toString() shows: Result<Result<UserRegisteredEvent>>

ok, so what does foo.getOrNull() give me?
java.lang.ClassCastException : kotlin.Result cannot be cast to io.liquidsoftware.base.user.application.port.in.UserRegisteredEvent

Which would indicate it is indeed a double-nested Result

BUT, and here’s the kicker.

If I force cast it to the type I think it should be, it works

(foo as Result<UserRegisteredEvent>).getOrNull() works like I expect it to?

So my question is, WTF is going on here?

Please check out this screenshot

[Erik Dreyer]

I removed the context receivers for a bit to make this more explicit for now

[Erik Dreyer]

@simon.vergauwen Any chance you could spare a minute to look at this?

[Erik Dreyer]

I feel like I’m taking crazy pills 🙂

[simon.vergauwen]

Wow, so toString and the debugger (toString) are showing two completely different things 🤯

I am making the assumption here that IDEA also uses toString in the debugger,

[simon.vergauwen]

What version of Kotlin were you using? Potentially conflicting Kotlin version on the classpath through Spring?

I've had some odd things when working with @JvmInline and value class in the past year, and in some production projects we switched to using data class rather than value class due to this reason. To avoid the value class you could perhaps try using Either.catch instead of runCatching and see if Either doesn't have this issue.

It's hard to say much beyond that without actually being able to look around in all the code to see what's going on. I've also had a bunch of issues with inline fun and value class, so you might also want to get rid of that in some places surrounding this code.

[Erik Dreyer]

yeah, right? Mind blown.

latest and greatest IntelliJ, with Kotlin 1.7.20

I’m not using @JvmInline or value classes. I may switch to Either just to see if I get different behavior.

I’m close to stumped. I’ll keep poking at it to see if I can figure out where the root cause is

[simon.vergauwen]

Result is value + @JvmInline and I've seen this fail as well in combination with suspend and inline fun.
There are a couple edge-cases, and not sure if all of them are fixed yet.

[simon.vergauwen]

Also interface + suspend + inline fun + Result / value + @JvmInline can cause some issues through inlining and dynamic dispatching etc, but I don't know all the compiler/low-level details. I've only got some experience debugging, and spotting problematic code 😅

[Erik Dreyer]

oh, interesting. OK, I’ll definitely try Either

[Erik Dreyer]

Thanks for the insights. I’ll post back here what happens

[simon.vergauwen]

My pleasure ☺️

[Erik Dreyer]

To isolate this change, and reduce the amount of updates required to test either, I did the following.

1. replaced a result { … } block with either { … }
2. added an Either.toResult() extension method, so I didn’t have to change the return type of the method. (does one already exist in the library? I didn’t see it)
This fixed the issue I was seeing, which probably verifies @simon.vergauwen’s suspicions above

[simon.vergauwen]

Thanks for the update! There is no Either<Throwable, A> -> Result<A> conversion method, nor the other way around 🤔 That would be a welcoming addition.