Sprint 208: Asynchronous Asset Streaming & Background WGPU Queue Worker 🚀
Welcome to Sprint 208! We have reached a vital milestone for real-time 3D execution: eliminating frame stuttering when loading massive assets. By introducing asynchronous asset streaming, the WGPU main thread remains entirely fluent, while heavy file reads and image decoding are offloaded to background worker threads.
⚡ Non-Blocking Texture Loading
Previously, calling registry_load_texture(path) would block the main thread while opening and decoding image data (using the image crate). In a real-time 3D environment or game, this blocked frame rendering, causing noticeable spikes.
In Sprint 208, registry_load_texture has been made completely non-blocking:
- The FFI call increments an atomic counter (`TEXTURE_ID_COUNTER`) and returns the texture ID immediately to the script, without waiting.
- The heavy work — reading the image file and decoding it to a raw RGBA byte buffer — is spawned in a background thread via
std::thread::spawn.
⚙️ Deferred GPU Injection
Once the background thread finishes decoding, it sends the raw pixel buffer back to the main render thread via a render queue channel command: RenderCommand::LoadTexture { id, width, height, rgba }.
The main thread picks up the command asynchronously on the next frame and updates the GPU VRAM texture via WGPU queues. This means image decoding runs in parallel on separate CPU cores, resulting in zero frame hitches!
🧪 Verification & Testing
To guarantee zero regressions, we added the integration test test_asset_streaming_non_blocking in `tests/sandbox_tests.rs`. The test simulates five concurrent texture load requests, verifying that all IDs are assigned immediately and the entire operation stays under 200ms, proving the non-blocking execution model.