Make stableNull public in the stdlib

[noti0na1]

The PR removes private[scala] from stableNull definition.

It was first introduced by #23528 for 3.8 stdlib migration. We want to make it public as part of explicit nulls.

How much have you relied on LLM-based tools in this contribution?

Minimally for document polishing

How was the solution tested?

It was already tested by #23528

Note

Need a minor release

[olhotak]

LGTM

[SolalPirelli]

If we're making stuff public for explicit nulls, should nullForGC also be public? Not strictly needed but convenient (Seb's PR removing unsafe nulls in the compiler could use it in a few places, for instance) https://github.com/scala/scala3/blob/main/library/src/scala/runtime/ScalaRunTime.scala#L324

[sjrd]

I wouldn't make nullForGC public, no. That makes it less safe, contrary to @stableNull, which is safer than the alternative.

[bishabosha]

so this is i guess just as safe as compiletime.uninitialized, i.e. not at all, but can there be hope for actual static analysis for checking "set-once"

[sjrd]

It's not "not at all". Both require you to provide the reasoning/guarantee for one property. Then they do the rest of the job.

• uninitialized: you need to guarantee that you will not read it before setting it the first time.
• @stableNull: you need to guarantee that only one thread accesses it.

If you provide those guarantees, then they guarantee the rest of the safety.

nullForGC however gives you nothing besides local documentation.

[olhotak]

For @stableNull, the required guarantee is stronger than just only one thread accessing it. You also need to guarantee that the mutable var will not be mutated in other methods or through aliases, even in the same thread.

[noti0na1]

but can there be hope for actual static analysis for checking "set-once"

1. A special "nullable" type, and Null is not its subtype;
2. A runtime check it disallowing assigning null after initiallization.

Then, we can guarantee: if a field/variable has been initialized, it will never become null again.

[noti0na1]

For question about Kotlin's behaviour: no, they don't track mutable fields for smart cast.

class A(var s: String?)

fun main() {
    val a = A("a")
    if (a.s != null) {
        println(a.s.length) // rejected by smart cast: 's' is a mutable property that could be mutated concurrently
    }
}

cc @odersky

[SethTisue]

Core agreed today that this can become public, modulo a possible name change.