[lemcoder]

Problem

The JNI generator passed every pointer to a struct as a bare Long. Any Long, even a channel number, compiled where a tsf* was expected. A failed call's NULL came back as 0L, ready to be dereferenced by the next call, which crashes the process.

Change

Each named struct or union reached by a pointer gets a typed handle:

@JvmInline internal value class TsfPointer(val address: Long)

internal fun tsf_load_filename(filename: String?): TsfPointer?            // NULL → null
internal fun tsf_note_on(f: TsfPointer, preset_index: Int, key: Int, vel: Float): Int

• Naming: by the record's tag, else (anonymous record) its typedef. node_t* and struct node* are one NodePointer; const is ignored.
• Nullability: parameters are non-null, results are nullable.
• JNI: the external keeps the Long behind the <name>$kni wrapper that strings already use. Kotlin mangles the JVM name of functions with value-class parameters, and JNI couldn't resolve it.
• Unchanged: typedef'd void*, function pointers and pointers to pointers (LongArray?).
• Escape hatch: address gives the raw pointer, and TsfPointer(0L) passes NULL deliberately.

Breaking for consumers: code passing these pointers as Long must switch to the handle type.

Ownership (freeing, use after free) is out of scope here. It needs a DSL to declare the release function.

Tests

• GenerateFromNativeIndexTest asserts the value class, the wrapper/external pair, tag naming through a typedef, and a nullable result (sample_create added to sample.h). Full suite: 66 tests, 0 failures.
• Checked end to end in MikroSoundFont (Android-KMP, AGP 9.3.1) via includeBuild. Delegates moved to TsfPointer compile, and an instrumented test on a physical arm64 device (Android 17) passed 2/2: load, render, tsf_close, and garbage input rejected.

🤖 Generated with Claude Code