[mrousavy]

Second of five stacked PRs: #1997 → #1998 (this one) → #1999 → #2008 → #2009.

Kotlin variant case classes stored the type JNI uses instead of the public Kotlin type. For UInt64 | string, when (v) { is X.First -> v.value } gave a signed Long, so values above 2^63 came out negative, and X.First(5uL) didn't compile. A UInt64[] case held a LongArray, and a callback case held the generated Callback_* class, so constructing that case meant typing a Nitrogen specialization name.

Cases now hold their public type, e.g. ULong, ULongArray or Callback<...>, like Swift. fromCpp_<n> converts from the JNI value. JNI reads the value field directly when it already has the JNI representation, which covers ULong and ULongArray. Otherwise it calls a private getValue_cxx() accessor, like structs do. For callback cases that accessor uses Callback_*.fromKotlin(), which returns a callback that came from C++ unchanged, so create() no longer wraps the callback up front.

JNI calls these accessors, and the struct get<Field>_cxx() getters, with getNonvirtualMethod. Both are private and final, like the hybrid object _cxx bridges that already use it, so the virtual lookup was wasted.

New TestObject methods bounce UInt64 | string and UInt64[] | string variants. The Kotlin fixture rebuilds the ULong case through its constructor.

Validation: Kotlin compiles without warnings, javap shows the long and long[] fields and the createFirst(long) JVM name JNI expects, and the variant JNI sources compile with the NDK. I did not run the runtime tests locally.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
nitro-docs	Skipped		Oct 23, 2026 5:51pm UTC