fix(views): fill Props::rawProps on Android when overriding cloneProps

[ronickg]

Fixes #1656.

The problem

ViewComponentDescriptor::cloneProps overrides React Native's implementation and
does not call initializeDynamicProps. Up to RN 0.85 that was harmless, because
Props::Props called it for you via Props::initialize. RN 0.86 removed that, so
nothing fills Props::rawProps for a Hybrid View any more — and on Android that
map is exactly what gets serialized to Java and applied by
ViewManager.updateProperties.

The effect: a Hybrid View silently receives no backgroundColor, border*,
transform, opacity, testID or accessibility props on Android. iOS is
unaffected, because RCTViewComponentView reads the typed C++ fields directly.

Nothing warns, because React Native never sees a prop to report as unsupported.

Left is a plain RN View, right is TestView, same style object, on RN 0.87.1
with no feature flags:

Props::rawProps has exactly one writer per RN version:

	writers of Props::rawProps
RN 0.85.3	Props::initialize (from the constructor) + ConcreteComponentDescriptor::cloneProps
RN 0.87.1	ConcreteComponentDescriptor::cloneProps only

(ShadowNode::propsForClonedShadowNode also touches it, but it is gated on
!rawProps.empty() — it merges, it cannot create.)

A corroborating detail: nitrogen generates a per-view filterObjectKeys and threads
it through ViewProps → BaseViewProps → YogaStylableProps → Props, where React
Native's only consumer of it is rawProps.toDynamic(filterObjectKeys) inside
initializeDynamicProps. Without that call it is dead code.

The fix

auto newProps = TShadowNode::Props(context, /* & */ rawProps, props);
#ifdef RN_SERIALIZABLE_STATE
  TShadowNode::initializeDynamicProps(newProps, rawProps, props);
#endif
return newProps;

This does add a rawProps.toDynamic() per props clone, which may be part of what
the override was avoiding. If you'd prefer something narrower — serializing only
the base ViewProps keys — I'm happy to rework it that way.

Why the harness didn't catch it

It couldn't. On the pinned RN 0.85.3 the constructor still fills the map, so the
bug has no observable effect. Turning on enableExclusivePropsUpdateAndroid makes
0.85 take the same path 0.86+ takes unconditionally, and then it does — and not
subtly. queryByTestId resolves against the native view tree, and testID is
itself a base view prop, so most of the file can't even find the view it rendered:

full nitro.views.harness, Pixel 9 API 36 emulator, flag on

  without this fix ....  8 failed,  2 passed
  with it .............  10 passed

Six of those eight are existing tests, not ones added here:

● TestView › renders with native props, layout, pixels, methods, and callbacks
● TestView › updates every prop, changes pixels, and resizes the same native view
● TestView › unmounts and creates a fresh native view when remounted
● multiple RecyclableTestViews › keeps instances isolated while one is updated and recycled
● RecyclableTestView › renders, changes pixels, and resizes the same native view
● RecyclableTestView › reuses and resets the native view across recycling cycles

What's in this PR

file	why
cpp/views/ViewComponentDescriptor.hpp	the fix
apps/example/android/.../MainApplication.kt	turns on enableExclusivePropsUpdateAndroid so the harness exercises the 0.86+ path. Droppable once the floor is >= 0.86, where it's unconditional.
apps/example/__tests__/nitro.views.harness.tsx	a dedicated pair of tests naming the cause (applies opacity to the native view, plus a control)

The opacity test asserts the blue surface washes out at 10% alpha rather than
asserting on PNG alpha — a screen capture is always alpha 255, so that would
have passed either way.

Verification

• bun specs — no codegen drift
• bun typecheck — clean
• bun example lint-ci — clean
• clang-format -style=file:./config/.clang-format on the changed header — clean
• bun lint-kotlin — ktlint not installed locally; its scope is packages/*/android, and the Kotlin change here is under apps/example
• nitro.harness (the other file) shows 11 pre-existing ArrayBuffer/HardwareBuffer
  failures on this emulator — identical with and without the flag, verified with a
  control build

A clean, fix-free reproduction on RN 0.87.1 is on
bump/rn-0.87: build it and
open the example's View tab — that's the screenshot above.

Tested on a Pixel 9 API 36 emulator (Android 16, arm64). I have not run this on
physical hardware or on iOS; the iOS claim above is from reading
RCTViewComponentView, not from a measurement.

🤖 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		Sep 21, 2026 8:53pm UTC