feat(engine,nodes): add cross-node control messaging via param_bridge

[staging-devin-ai-integration]

Summary

Adds a new core::param_bridge terminal node that bridges data-plane packets to control-plane UpdateParams messages, enabling cross-node parameter control without custom code. This powers use cases like live subtitle overlays where transcription text drives a Slint UI property.

Key changes

New node: core::param_bridge (crates/nodes/src/core/param_bridge.rs)

• Three operating modes: template (JSON template with {{text}} placeholders), raw (forward full packet payload), auto_map (wrap in { "properties": {...} } for Slint compatibility)
• Configurable debounce (debounce_ms) using tokio::select! with a pinned sleep timer to coalesce rapid updates
• Whitespace-insensitive placeholder normalization ({{ text }} and {{text}} both work; prevents double-replacement when substituted text contains literal {{text}})
• Schema endpoint via register_static_with_description (avoids FFI instantiation)
• Emits stt.result telemetry events with text_preview when forwarding text, so transcribed text appears in the stream view's telemetry timeline
• Monitors control_rx for Shutdown to support graceful termination (avoids 5s timeout on RemoveNode)

Engine support (crates/engine/)

• New BestEffort connection mode — connects when both endpoints exist, silently skips otherwise
• connection_mode: best_effort in pipeline YAML

tune_sibling() error handling (crates/core/src/node.rs)

• Returns Result<(), StreamKitError> instead of String for consistency with the rest of the codebase

Sample pipeline (samples/pipelines/dynamic/video_moq_webcam_subtitles.yml)

• Full webcam → MoQ → whisper → param_bridge → Slint subtitle → compositor pipeline
• Silero VAD filtering on Whisper (vad_threshold: 0.4, min_silence_duration_ms: 600) for faster transcription
• core::telemetry_out as best_effort side branch surfaces STT results in stream view
• Documents required plugin builds and model dependencies

Slint subtitle component (samples/slint/system/subtitle.slint)

• Semi-transparent panel with word-wrapped text and fade-in animation
• Uses show property (not visible, which is a reserved Slint built-in — original visible caused a silent compilation failure at the FFI boundary)

Commits

1. feat(engine,nodes): core param_bridge implementation + BestEffort connections
2. fix(nodes): skip unsupported packets in template mode, add unit tests
3. style(nodes): fix clippy lints
4. fix(nodes): address Devin Review findings (debounce, error types, placeholders, auto_map docs, schema endpoint, sample pipeline fixes)
5. fix(nodes): normalize template placeholders before substitution (prevents double-replacement bug)
6. fix(slint): rename reserved visible property in subtitle.slint
7. feat(nodes): emit telemetry from param_bridge for stream view visibility
8. feat(pipeline): add VAD filtering and telemetry_out to subtitle pipeline
9. fix(nodes): handle control_rx Shutdown in param_bridge select loop

Review & Testing Checklist for Human

• Verify subtitle.slint loads — the visible → show rename fixes a Slint compilation error that was silently swallowed at the FFI boundary. Run the subtitle pipeline and confirm the Slint node initializes
• Test param_bridge template mode — send transcription packets through the pipeline and verify the Slint subtitle text updates via UpdateParams
• Verify telemetry in stream view — stt_telemetry (telemetry_out) and param_bridge telemetry emission should surface transcribed text in the stream view's telemetry timeline
• Test VAD filtering — with vad_threshold: 0.4 and min_silence_duration_ms: 600, silence should be filtered before Whisper inference, improving responsiveness
• Test graceful shutdown — removing the param_bridge node should complete within the 5s timeout (not force-abort)

Notes

• Slow transcription on CPU is expected when Whisper runs without GPU (whisper_backend_init_gpu: no GPU found). The VAD filter reduces wasted inference on silence. min_silence_duration_ms and max_segment_duration_secs can be further tuned.
• The just lint failure in plugins/native/aac-encoder/src/lib.rs is a pre-existing formatting issue unrelated to this PR.
• 27 unit tests cover all param_bridge modes, template replacement, debounce config, and the placeholder normalization regression.
• The FFI boundary in native_source_plugin_entry! macro silently discards plugin new() errors — the visible property bug was invisible because of this. A follow-up to surface these errors would improve debuggability.

Link to Devin session: https://staging.itsdev.in/sessions/a750af18ee254481a97c4ac581ba129e
Requested by: @streamer45

---

[staging-devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring

[staging-devin-ai-integration]

Devin Review found 1 new potential issue.

View 5 additional findings in Devin Review.

Debug

Playground

[staging-devin-ai-integration]

🚩 Blocking send in tune_sibling may delay shutdown by up to 2 seconds

The tune_sibling method at crates/core/src/node.rs:381-386 uses tx.send().await on the engine's main control channel (engine_control_tx). During engine shutdown (crates/engine/src/dynamic_actor.rs:1550-1639), the engine stops reading from control_rx while waiting for nodes to exit. If a param_bridge node is mid-send and the channel buffer (128 slots) happens to be full, the node blocks and cannot exit its loop until the engine's 2-second timeout fires and the task is aborted. This produces a misleading warning: "Node did not shut down within 2s, this indicates a bug".

In practice this is unlikely (the 128-capacity channel is rarely full), and the timeout+abort mechanism handles it correctly, but using try_send instead of send().await would make the param_bridge more shutdown-friendly and avoid the false warning. The sample YAML already marks the connection as best_effort, so dropping a single UpdateParams during shutdown is acceptable.

---

Was this helpful? React with 👍 or 👎 to provide feedback.

Debug

Playground

[staging-devin-ai-integration]

Valid observation. In practice the 128-slot buffer makes this unlikely, and the timeout+abort handles the edge case correctly. Using try_send is a reasonable improvement but would change the semantics slightly (silently dropping messages when the buffer is full during normal operation, not just shutdown). I'd defer this to human review — if the maintainer prefers try_send semantics here I'm happy to switch it.

[staging-devin-ai-integration]

Slint subtitle fix (1560532)

Root cause identified: visible is a built-in property on all Slint elements (including Window). Declaring in property <bool> visible: true; in subtitle.slint caused the Slint compiler to reject the file with Cannot override property 'visible'.

This error was invisible because the native_source_plugin_entry! macro's __plugin_create_instance trampoline silently discards Err(_) from new() (returns null without logging the error message). The host side then sees only the generic "Plugin failed to create instance".

Fix: Renamed visible → show (consistent with lower_third.slint's show property). Updated both the .slint file and the pipeline YAML template.

Verified with a standalone Slint compiler test — subtitle.slint now compiles cleanly and exposes show (Bool) and text (String) properties.