Scala 3.3.x Lazy Vals regression with AnyVal value ( scala.MatchError: scala.runtime.LazyVals$Evaluating )

[soronpo]

I decided to report this before I find the time to minimize, because this is a severe issue for anyone using AnyVal values.
They should avoid upgrading to 3.3.x until this is found and resolved.

Compiler version

v3.3.0-RC2

Minimized code

I still don't have a minimized error and the errors are occurring randomly, so it makes it very problematic to minimize.
In general where it fails is that it tries to match against the underlying value of the AnyVal, and it gets a MatchError.

TBD

Output

 scala.MatchError: scala.MatchError: scala.runtime.LazyVals$Evaluating$@6ce08dec (of class scala.runtime.LazyVals$Evaluating$) (of class scala.MatchError)

Expectation

No error.

[szymon-rd]

I would be really grateful if you could provide some minimization, even if it fails only a percentage of time.

[soronpo]

I have it failing quite consistently but in different stack traces (the final part is the same).
I guess that it might also be because of a concurrency issue, and only when I run multiple tests then it fails.
I'll try to force tests to run sequentially and see if indeed that's the case.

[szymon-rd]

This snippet works ok, for an example:

class LazyHolder {
    lazy val xInstant: AnyVal = {
        21
    }
}

val ec = Executors.newFixedThreadPool(Runtime.getRuntime().availableProcessors() * 4)

@main
def main() = 
    val holders = List.fill(1_000_000)(new LazyHolder())
    val counter = new AtomicInteger()
    for (
        holder <- holders
    ) {
        ec.execute(
            () => holder.xInstant match {
                case 21 => counter.incrementAndGet()
            }
        )
    }

    new Thread() {
        override def run(): Unit = {
            while(true) {
                println(counter.get())
                Thread.sleep(1000)
            }
        }
    }.start()

[sjrd]

lazy val xInstant: AnyVal

That erases to Object like AnyRef. Try an actual value class, one that extends AnyVal. Not AnyVal itself.

[szymon-rd]

Fair point. But I tried that, and it passed as well.

[sjrd]

Can you show the generated code? Perhaps we can already spot it there.

[soronpo]

I pushed my (not-yet-minimized) codebase into https://github.com/DFiantHDL/DFiant under the branch lazyexception
I use AdoptOpenJDK Java 16.

Just clone, checkout lazyexception and run sbt run.
Let me know if this helps and you replicate the issue on your computer.
Note:

• This codebase uses compiler plugins
• This codebase was not meant to be used by "normal" people in its current state.

[szymon-rd]

I pulled this repo, and it fails on this branch for me too. But are you sure that it's not the custom compiler plugin that makes it fail? Maybe you are relying on the lazy val holder (foo$lzy<n>) in it, instead of the accessor? If that were true, then this error could occur.

[soronpo]

I pulled this repo, and it fails on this branch for me too. But are you sure that it's not the custom compiler plugin that makes it fail? Maybe you are relying on the lazy val holder (foo$lzy<n>) in it, instead of the accessor? If that were true, then this error could occur.

That possibility was i my mind and that is why I want find away I can minimize without the plugin or rule it out. However, if I remove the AnyVal dependency everything works correctly. Additionally, the compiler does not distinguish different runtimes, so if the fault was in the plugin, it should have tripped in a single test as well.