[mfazekas]

On the new Android backend, every ViewModelImageProperty.set() decoded a new ImageAsset on the command server and never closed it, even when setting the same image again. Swapping images grew GPU memory from 304 MB to 4.8 GB over 211 swaps (2048×2048 PNGs, emulator). The handle is now closed right after setImage (the property keeps its own ref), and memory stays flat at ~311 MB.

set(undefined) now clears the image instead of being ignored. Sets run on a single-threaded scope so they apply in call order. Before, a clear could be overtaken by an earlier set that was still decoding.

Adds a harness test for set/clear. Memory isn't asserted automatically; use the AndroidMemoryLeak reproducer from #426 (image swap scenarios) to check.

[Copilot]

🟡 Changes recommended

An Android coroutine race can restore an image after clearing, and the regression test neither waits for mutation completion nor verifies the cleared state.

2 open issues

• Serialize image decoding and clearing as whole operations · New
• Wait for mutation completion and assert the image was cleared · New

What changed in this PR

This PR addresses decoded-image handle leaks in the new Android backend, complementing the legacy-backend fixes in #426.

Changes:

• Closes temporary decoded image handles after assignment.
• Adds image clearing via set(undefined) and attempts to preserve setter call order.
• Adds a set/clear harness test, skipping clears on experimental iOS.

File	Description
example/​__tests__/​image-property-clear.harness.ts	Adds repeated image set/clear calls and resource cleanup.
android/​src/​new/​java/​com/​margelo/​nitro/​rive/​HybridViewModelImageProperty.kt	Releases decoded handles, supports clearing, and changes coroutine dispatching.

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

The Android setter still allows a suspended image decode to overwrite a later clear, while the test neither waits for native completion nor verifies the final cleared state.

1 open issue

• Serialize image decoding and clearing as whole operations

1 resolved since last review

• Wait for mutation completion and assert the image was cleared

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

Independent sequencing state across image-property lookups allows an earlier decode to overwrite a later clear, while the harness can pass without verifying asynchronous set/clear results.

No open issues 🎉

1 previously missed issue

In code that hasn't changed since last review

Reuse image property state across repeated lookups

android/​src/​new/​java/​com/​margelo/​nitro/​rive/​HybridViewModelImageProperty.kt:30

HybridViewModelInstance.imageProperty() creates a fresh wrapper on every lookup (HybridViewModelInstance.kt:79–81), so repeated lookups use independent generation counters and dispatchers. Calling instance.imageProperty(path).set(image) followed by instance.imageProperty(path).set(undefined) can therefore let the earlier decode finish after the clear and restore the image. Share the sequencing and generation state per underlying instance/path, or reuse the property wrapper across lookups.

1 resolved since last review

• Serialize image decoding and clearing as whole operations

_{Review effort: Balanced}

[mfazekas]

Re the previously missed issue in Copilot's last review (independent sequencing per imageProperty() lookup): fixed in 9a27f55. HybridViewModelInstance now keeps one sequencer per path (a serial scope plus a generation counter), and every wrapper it returns for that path shares it.

[Copilot]

🔵 Needs a closer look

The Android image set/clear harness can pass despite failed native writes because it neither waits for asynchronous operations nor verifies their results before disposing resources.

No open issues 🎉

_{Review effort: Balanced}

[HayesGordon]

LGTM