feat(inference): inference-llm PR-3b — InferenceLlmModule auto-publishes via bus hook (pure Rust)

[joelteply]

Summary

PR-3b of inference-llm. Wires the bus helpers from PR-3a (#1392) INTO InferenceLlmModule::handle_command so every successful inference response auto-publishes InferenceComplete + FirstTokenEmitted to the trace bus.

Closes the inference-llm bus loop end-to-end, fully in Rust:

producer (command) → engine (stub for now) → response (CommandResult)
                                            ↓
                              bus dispatch (complete + first_token)
                                            ↓
                              subscribers (sentinel / VDD / audit-recorder)

Zero TS. Zero shim. Zero fallback.

What lands

• BusHook private struct holding Arc<MessageBus> + Arc<ModuleRegistry>. Same shape as my genome BusHook (feat(genome): working-set-manager PR-5 — LocalWorkingSetManager auto-publishes via bus hook (+MessageBus borrow fix) #1362).
• InferenceLlmModule.bus_hook: Option<BusHook> — None = bus-less PR-2 behavior; Some = auto-publish.
• with_bus(bus, registry) constructor — wires both Arcs at module construction (prevents the "bus added mid-service" race).
• handle_request body: on success, spawns publish_inference_complete + publish_first_token_emitted into the current tokio runtime via Handle::try_current. Spawn pattern (not await) avoids the DashMap borrow-across-await lifetime issue inside Send-bounded async_trait — same workaround as my genome LocalWorkingSetManager (feat(genome): working-set-manager PR-5 — LocalWorkingSetManager auto-publishes via bus hook (+MessageBus borrow fix) #1362).
• spawn_publish_inference_complete + spawn_publish_first_token_emitted module-private helpers.

Design choices

• Publishing is best-effort observability. Authoritative response goes back through CommandResult regardless of publish success.
• Error paths do NOT publish. Unknown command + invalid payload silent on the bus — typed Err in Result is the authoritative signal. Tests pin this.
• Two separate spawns (one per event) rather than one bundled publish. Lets subscribers see first_token even if complete hasn't dispatched (race-tolerant TTFT observability).

Test plan

• cargo test --lib --features metal,accelerate inference::llm_module_service — 12/12 pass:
  • 4 new PR-3b bus tests:
    • handle_command_with_bus_auto_publishes_complete_and_first_token (end-to-end)
    • handle_command_without_bus_does_not_publish (backwards-compat with PR-2)
    • handle_command_unknown_with_bus_does_not_publish (error path silent)
    • handle_command_invalid_payload_with_bus_does_not_publish (error path silent)
  • 8 PR-2 tests still pass unchanged
• No regressions across other 2957 lib tests

Stack

• feat(inference): inference-llm PR-1 — typed event surface (MODULE-CATALOG §II) #1387 — inference-llm PR-1: typed event surface
• feat(inference): inference-llm PR-2 — InferenceLlmModule ServiceModule impl (stub-backed) #1391 — inference-llm PR-2: ServiceModule impl (stub-backed)
• feat(inference): inference-llm PR-3a — canonical ArtifactKeys + publishing helpers #1392 — inference-llm PR-3a: bus keys + publishing helpers
• This PR — inference-llm PR-3b: auto-publish wiring
• NEXT — PR-4: real LlamaCppAdapter invoke (the stub stays in place; PR-4 swaps under the same external contract)

🤖 Generated with Claude Code