Kover coverage

[tonytonycoder11]

Scoped in M15 and not shipped. Kover 0.9.1 is not yet compatible with the Kotlin 2.4 Gradle plugin.

[tonytonycoder11]

Retested today against the current toolchain rather than taking the 1.0.0 note on trust. Still broken, and here is the exact failure so nobody has to repeat the experiment:

Kover 0.9.1 (still the latest published version) applied to kmemo-core on Kotlin 2.4.10:

Could not determine the dependencies of task ':kmemo-core:koverGenerateArtifactJvm'.
> Could not get unknown property 'compileKotlinTask' for compilation 'main' (target  (jvm))
  of type org.jetbrains.kotlin.gradle.plugin.mpp.KotlinWithJavaCompilation_Decorated.

Kover reaches for an internal property the Kotlin 2.4 Gradle plugin no longer exposes. Nothing on our side fixes that, so this stays blocked on an upstream release.

Worth a decision, though: the issue is titled after the tool, but what it is actually for is coverage. Kover is a wrapper over JaCoCo and the IntelliJ agent, and JaCoCo works with Kotlin bytecode directly. Adopting JaCoCo would satisfy the intent now, at the cost of a less Kotlin-aware report (inline functions and coroutine state machines read worse) and of moving off the tool the rest of the JetBrains toolchain assumes.

Three ways forward, none of them urgent:

1. Wait for a Kover release that supports Kotlin 2.4. Costs nothing, unbounded wait.
2. Adopt JaCoCo now and revisit if Kover catches up. Coverage exists; the report is less idiomatic.
3. Close as Declined and lean on what already carries the quality bar: the corpus regression gate with floors on all three splits, property-based tests, ktlint and detekt. That is the honest option if nobody is actually going to read a coverage number.

Leaving it open and blocked is only right if the answer is (1).

[tonytonycoder11]

Declined.

Retested today: Kover 0.9.1 is still the latest release and still breaks against the Kotlin 2.4 Gradle plugin, reaching for a compileKotlinTask property it no longer exposes. Nothing on our side fixes that.

Declined rather than left open, because what carries the quality bar here is not a coverage percentage: the corpus regression gate with floors on all three blind splits, property-based tests on the vector maths and the tokenizer, ktlint and detekt. A number nobody reads would not have changed a single decision on this repository.

Reopen if a Kover release lands with Kotlin 2.4 support and there is an actual use for the figure, such as a CI threshold.