fix(nitrogen): match Kotlin's is prefix rule for JVM accessors

[giaBaoJS]

What breaks

packages/nitrogen/src/syntax/Property.ts:103 (getSetterName) and :74 (getGetterName) modelled Kotlin's is prefix rule as "the property is a boolean and name.startsWith('is')".

Kotlin's actual rule (JvmAbi.startsWithIsPrefix) is different in two ways:

1. The character after is must not be a lowercase letter. isBlue is shortened, isolatedBoolean is not.
2. It applies to every property type, not just Boolean. var isTextValue: String is shortened too.

Because the JNI bridge looks methods up by name at runtime, a mismatch is not caught by the Android build. It throws on first access:

Kotlin property	kotlinc/javap emits	nitrogen generated (before)
var isolatedBoolean: Boolean	getIsolatedBoolean() / setIsolatedBoolean(..)	isolatedBoolean() / setOlatedBoolean(..)
var isTextValue: String	isTextValue() / setTextValue(..)	getIsTextValue() / setIsTextValue(..)

Note setOlatedBoolean: the old code did name.replace('is', ''), so isolatedBoolean became OlatedBoolean.

iOS is unaffected. Swift/C++ interop really does use a plain is/has prefix check and really is boolean only, so that branch is kept as it was, just moved out of the shared boolean gate.

Fix

Property.ts now has a single hasJvmIsPrefix() helper that mirrors JvmAbi.startsWithIsPrefix, used by both the JVM getter and setter name. The Swift branch keeps its own (boolean, is/has) rule.

Test

Two properties on SharedTestObjectProps (next to the existing has*/is* block added in #1223), one per shape, plus a set + get assertion for each in apps/example/src/getTests.ts, so the Harness workflows cover them on both platforms:

isolatedBoolean: boolean
isTextValue: string

isTextValue rather than isStringValue, because isStringValue collides with the existing stringValue property on the same class (both would be setStringValue(String), which kotlinc rejects with a platform declaration clash).

How I proved it

Ground truth from the real Kotlin compiler, on the property declarations nitrogen generates in HybridTestObjectSwiftKotlinSpec.kt:

$ javap -cp outspec Spec
public abstract class Spec {
  public abstract java.lang.String getStringValue();
  public abstract void setStringValue(java.lang.String);
  public abstract boolean getHasBoolean();
  public abstract boolean isBoolean();
  public abstract boolean getHasBooleanWritable();
  public abstract void setHasBooleanWritable(boolean);
  public abstract boolean isBooleanWritable();
  public abstract void setBooleanWritable(boolean);
  public abstract boolean getIsolatedBoolean();
  public abstract void setIsolatedBoolean(boolean);
  public abstract java.lang.String isTextValue();
  public abstract void setTextValue(java.lang.String);
}

With the fix, bun specs generates exactly those names in JHybridTestObjectSwiftKotlinSpec.cpp:

getMethod<jboolean()>("getIsolatedBoolean");
getMethod<void(jboolean /* isolatedBoolean */)>("setIsolatedBoolean");
getMethod<jni::local_ref<jni::JString>()>("isTextValue");
getMethod<void(jni::alias_ref<jni::JString> /* isTextValue */)>("setTextValue");

Reverting only Property.ts and re-running bun specs gives four lookups that do not exist on the Kotlin class:

getMethod<jboolean()>("isolatedBoolean");
getMethod<void(jboolean /* isolatedBoolean */)>("setOlatedBoolean");
getMethod<jni::local_ref<jni::JString>()>("getIsTextValue");
getMethod<void(jni::alias_ref<jni::JString> /* isTextValue */)>("setIsTextValue");

I also checked the iOS side against swiftc -cxx-interoperability-mode=default -emit-clang-header-path, which emits isolatedBoolean() / setIsolatedBoolean(..) and getIsTextValue() / setIsTextValue(..), matching the generated HybridTestObjectSwiftKotlinSpecSwift.hpp.

Checks

• bun specs run, generated files committed. Apart from the two new properties, no generated output changed.
• bun run build, bun typecheck, bun lint, bun lint-ci in every package, bun lint-cpp, bun lint-swift, bun lint-kotlin: all clean, no files changed by the formatters.
• I could not run the Harness suite (no device available here), so the two new runtime assertions are verified by CI, not locally.

Breaking changes

A Nitro module with a Kotlin is* property in one of these two shapes changes its generated JNI lookup names. That code could not have worked before, so nothing that currently runs is affected.

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated
nitro-docs	Skipped		Sep 7, 2026 11:11am UTC

[nitro-modules-bot]

Performance Report

⚠️ Advisory: Results do not fail this PR while the baseline is being calibrated.

iOS

Benchmark	Before	After	Difference
C++ copy(1 MiB)	128.42 µs	96.93 µs	🟢 -24.52% faster
C++ synchronousCallback()	404.6 ns	325.6 ns	🟢 -19.52% faster
C++ number | string variant	246.0 ns	205.9 ns	🟢 -16.31% faster
Swift copy(1 MiB)	112.92 µs	93.36 µs	🟢 -17.32% faster
Swift copy(4 KiB)	2.07 µs	1.54 µs	🟢 -25.59% faster
Swift large array (1,024)	84.99 µs	67.79 µs	🟢 -20.24% faster
Swift synchronousCallback()	462.3 ns	419.9 ns	🟢 -9.16% faster
Swift create()	44.43 µs	38.21 µs	🟢 -14% faster
Swift simpleFunc()	106.8 ns	122.1 ns	🔴 +14.25% slower
Swift number | string variant	409.3 ns	312.9 ns	🟢 -23.54% faster

All Benchmarks

Benchmark	Before	After	Difference
JavaScript addNumbers()	38.1 ns	39.6 ns	🟡 +3.98% slower (noisy)
C++ bounce(1 MiB)	389.9 ns	386.4 ns	🟡 -0.88% faster (noisy)
C++ bounce(4 KiB)	465.7 ns	421.8 ns	🟡 -9.43% faster (noisy)
C++ copy(4 KiB)	1.71 µs	1.47 µs	🟡 -13.8% faster (noisy)
C++ large array (1,024)	66.99 µs	62.77 µs	🟡 -6.3% faster (noisy)
C++ small array (16)	1.27 µs	1.25 µs	🟡 -1.27% faster (noisy)
C++ create()	676.9 ns	687.9 ns	🟡 +1.62% slower (noisy)
C++ returnExisting()	321.9 ns	340.7 ns	🟡 +5.84% slower (noisy)
C++ typed map (8 entries)	2.84 µs	2.93 µs	🟡 +3.15% slower (noisy)
C++ optional trailing string	268.3 ns	278.1 ns	🟡 +3.65% slower (noisy)
C++ addNumbers()	146.3 ns	144.1 ns	🟡 -1.48% faster (noisy)
C++ simpleFunc()	120.3 ns	133.7 ns	🟡 +11.09% slower (noisy)
C++ immediatePromise()	1.36 µs	1.42 µs	ℹ️ +4.38% slower (advisory)
C++ number property get/set	234.9 ns	215.9 ns	🟡 -8.12% faster (noisy)
C++ short ASCII string	251.9 ns	204.4 ns	🟡 -18.83% faster (noisy)
C++ Unicode string	462.9 ns	480.0 ns	🟡 +3.7% slower (noisy)
C++ nested Car struct	3.58 µs	3.33 µs	🟡 -6.97% faster (noisy)
Swift bounce(1 MiB)	508.3 ns	487.5 ns	🟡 -4.1% faster (noisy)
Swift bounce(4 KiB)	529.3 ns	465.6 ns	🟡 -12.04% faster (noisy)
Swift small array (16)	1.52 µs	1.48 µs	🟡 -3.08% faster (noisy)
Swift returnExisting()	497.8 ns	419.6 ns	🟡 -15.72% faster (noisy)
Swift typed map (8 entries)	3.45 µs	3.37 µs	🟡 -2.31% faster (noisy)
Swift optional trailing string	291.1 ns	260.4 ns	🟡 -10.55% faster (noisy)
Swift addNumbers()	133.3 ns	166.0 ns	🟡 +24.54% slower (noisy)
Swift immediatePromise()	2.36 µs	2.05 µs	ℹ️ -13.24% faster (advisory)
Swift number property get/set	264.1 ns	332.7 ns	🟡 +26% slower (noisy)
Swift short ASCII string	293.5 ns	311.9 ns	🟡 +6.26% slower (noisy)
Swift Unicode string	657.4 ns	615.4 ns	🟡 -6.39% faster (noisy)
Swift nested Car struct	3.71 µs	3.24 µs	🟡 -12.62% faster (noisy)
TurboModule addNumbers()	1.56 µs	1.63 µs	🟡 +4.38% slower (noisy)

Android

Performance is unchanged! 😎

All Benchmarks

Benchmark	Before	After	Difference
JavaScript addNumbers()	43.7 ns	43.6 ns	⚪ ~0.18% unchanged
C++ bounce(1 MiB)	619.7 ns	626.5 ns	🟡 +1.1% slower (noisy)
C++ bounce(4 KiB)	621.6 ns	602.7 ns	🟡 -3.05% faster (noisy)
C++ copy(1 MiB)	33.32 µs	33.27 µs	⚪ ~0.15% unchanged
C++ copy(4 KiB)	1.47 µs	1.56 µs	🟡 +6.7% slower (noisy)
C++ large array (1,024)	83.22 µs	83.18 µs	⚪ ~0.05% unchanged
C++ small array (16)	1.69 µs	1.70 µs	⚪ ~0.46% unchanged
C++ synchronousCallback()	504.6 ns	499.5 ns	🟡 -1% faster (noisy)
C++ create()	991.3 ns	1.00 µs	🟡 +1.09% slower (noisy)
C++ returnExisting()	372.8 ns	372.0 ns	⚪ ~0.21% unchanged
C++ typed map (8 entries)	3.39 µs	3.31 µs	⚪ ~2.2% unchanged
C++ optional trailing string	274.5 ns	276.9 ns	⚪ ~0.87% unchanged
C++ addNumbers()	166.6 ns	169.3 ns	⚪ ~1.61% unchanged
C++ simpleFunc()	132.3 ns	133.8 ns	⚪ ~1.16% unchanged
C++ immediatePromise()	1.71 µs	1.72 µs	ℹ️ +0.81% slower (advisory)
C++ number property get/set	271.1 ns	279.8 ns	⚪ ~3.2% unchanged
C++ short ASCII string	242.0 ns	244.0 ns	⚪ ~0.84% unchanged
C++ Unicode string	544.8 ns	540.2 ns	⚪ ~0.85% unchanged
C++ nested Car struct	4.40 µs	4.40 µs	⚪ ~0.01% unchanged
C++ number | string variant	261.1 ns	255.9 ns	⚪ ~2.01% unchanged
Kotlin bounce(1 MiB)	1.84 µs	1.81 µs	🟡 -1.73% faster (noisy)
Kotlin bounce(4 KiB)	1.87 µs	1.79 µs	🟡 -4.06% faster (noisy)
Kotlin copy(1 MiB)	460.73 µs	462.70 µs	⚪ ~0.43% unchanged
Kotlin copy(4 KiB)	8.60 µs	8.66 µs	⚪ ~0.77% unchanged
Kotlin large array (1,024)	91.56 µs	91.45 µs	⚪ ~0.12% unchanged
Kotlin small array (16)	2.27 µs	2.25 µs	⚪ ~0.59% unchanged
Kotlin synchronousCallback()	1.72 µs	1.73 µs	⚪ ~0.43% unchanged
Kotlin create()	3.12 µs	3.00 µs	🟡 -3.96% faster (noisy)
Kotlin returnExisting()	1.23 µs	1.23 µs	⚪ ~0.22% unchanged
Kotlin typed map (8 entries)	13.00 µs	13.40 µs	⚪ ~3.08% unchanged
Kotlin optional trailing string	692.5 ns	708.1 ns	⚪ ~2.26% unchanged
Kotlin addNumbers()	269.8 ns	266.5 ns	⚪ ~1.21% unchanged
Kotlin simpleFunc()	207.4 ns	217.3 ns	⚪ ~4.78% unchanged
Kotlin immediatePromise()	4.65 µs	4.63 µs	ℹ️ -0.4% faster (advisory)
Kotlin number property get/set	536.5 ns	551.2 ns	⚪ ~2.74% unchanged
Kotlin short ASCII string	920.3 ns	919.9 ns	⚪ ~0.04% unchanged
Kotlin Unicode string	1.48 µs	1.48 µs	⚪ ~0.59% unchanged
Kotlin nested Car struct	9.62 µs	9.60 µs	🟡 -0.17% faster (noisy)
Kotlin number | string variant	868.5 ns	851.7 ns	⚪ ~1.94% unchanged
TurboModule addNumbers()	528.7 ns	533.6 ns	⚪ ~0.93% unchanged

Benchmarking Code Diff fe50d8fc...4fcbe425 (view raw output)

[mrousavy]

yea okay that performance bot is not really reliable yet lol