fix(FN-6035): align workflow dispatch and broad suite

[gsxdsm]

Summary

• Fix scheduler admission so default in-progress dispatch respects the global semaphore limit as well as project WIP capacity.
• Update broad-suite test harnesses for Vitest 4 constructable mocks and file URL imports.
• Refresh workflow cutover test expectations for the workflow-graph review handoff path.

Validation

• pnpm test
• pnpm --filter @fusion/engine test
• pnpm --filter @fusion/engine test:core
• pnpm typecheck
• pnpm lint
• pnpm build

Notes

The concurrency fix covers the case where rows are already in in-progress but executors have not acquired semaphore slots yet; scheduler now treats those rows as consuming global dispatch capacity before admitting more work.

Summary by CodeRabbit

• Bug Fixes

  • Fixed task scheduler concurrency calculation to properly enforce global semaphore limits when dispatching concurrent tasks, preventing exceeding available capacity.

• Tests

  • Enhanced test reliability with improved mock cleanup and reset between test cases to prevent test interference.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: b9010df1-5b0b-4b44-8057-cba9560f035d

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• ✅ Review completed - (🔄 Check again to review again)

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feature/workflow-mapping

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[greptile-apps]

Greptile Summary

This PR addresses a race condition in the scheduler's global semaphore accounting and modernizes the broad test suite for Vitest 4 compatibility. The concurrency fix ensures tasks already in in-progress but whose executors have not yet acquired a semaphore slot are counted against global dispatch capacity before new todo tasks are admitted.

• Scheduler fix (scheduler.ts): semaphoreAvailable is now Math.min(availableCount, limit - agentSlots), capping available slots by both the semaphore's own acquired-count view and the number of currently in-progress rows — preventing over-admission when executors lag behind in slot acquisition.
• Vitest 4 constructable mock migration: All vi.fn().mockImplementation(() => ({...})) arrow-function factories replaced with function() { return {...}; } across 28 test files, required for proper new-call semantics in Vitest 4.
• Test expectation updates: interpreter-merge-seam corrected to assert onMerge receives a second {} options argument (matching production code); executor-task-done-invariant updated to workflow-graph-review reason for the graph review handoff path.

Confidence Score: 4/5

The scheduler change is a targeted, well-reasoned fix with a direct regression test; the remaining 29 files are test-only Vitest 4 compatibility updates.

The concurrency fix in scheduler.ts is logically sound: taking the minimum of the semaphore's own available-slot view and the remaining capacity computed from in-progress row count correctly prevents over-admission. The new test covers the precise race scenario described in the PR. The broad test migration is mechanical and low-risk. One assertion in heartbeat-monitor-per-agent-config.test.ts was weakened without explanation, which slightly reduces test fidelity on the warning-emit path.

packages/engine/src/tests/heartbeat-monitor-per-agent-config.test.ts (weakened warn-call assertion) and packages/engine/src/tests/executor-task-done-invariant.test.ts (reason string change to workflow-graph-review — the corresponding production-side change is not visible in this diff).

Important Files Changed

Filename	Overview
packages/engine/src/scheduler.ts	Core concurrency fix: semaphoreAvailable now takes the minimum of the semaphore's own available count and (limit - agentSlots), correctly accounting for in-progress tasks that haven't acquired semaphore slots yet.
packages/engine/src/tests/scheduler.test.ts	New test validates the semaphore cap: 3 in-progress tasks with a limit-3 semaphore and no acquired slots correctly blocks further dispatch from todo.
packages/engine/src/tests/interpreter-merge-seam.test.ts	Test corrected to assert onMerge is called with the second options argument ({}) matching the production implementation.
packages/engine/src/tests/executor-task-done-invariant.test.ts	Handoff audit reason updated from fn_task_done to workflow-graph-review to match the new workflow-graph review path; no production source change visible in this diff.
packages/engine/src/tests/heartbeat-monitor-per-agent-config.test.ts	Assertion relaxed from toHaveBeenCalledTimes(1) to toHaveBeenCalled(), which weakens the warning-count guard.
packages/engine/src/tests/project-engine.test.ts	All constructor mocks converted to function() { return {...}; } form for Vitest 4 constructable mock compatibility.
packages/engine/src/tests/reliability-interactions/node-settings-sync-auth.test.ts	Cross-package imports switched to file URL form (new URL(..., import.meta.url).href) for reliable ESM resolution in Vitest 4.
packages/engine/src/tests/reliability-interactions/branch-recovery-live-zero-commits.test.ts	Added afterEach(() => vi.restoreAllMocks()) for proper spy teardown between tests.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["schedule() called"] --> B["agentSlots = inProgress.length"]
    B --> C{"Semaphore provided?"}
    C -- No --> D["semaphoreAvailable = Infinity"]
    C -- Yes --> E["availableCount = limit minus active"]
    E --> F["capacityByRows = limit minus agentSlots"]
    F --> G["semaphoreAvailable = min of both"]
    G --> H["available = min of maxConcurrent gap, worktree gap, semaphoreAvailable"]
    D --> H
    H --> I{"available <= 0?"}
    I -- Yes --> J["Return early - no dispatch"]
    I -- No --> K["Admit tasks from todo to in-progress"]

Loading

_{Reviews (1): Last reviewed commit: "fix(FN-6035): cap scheduler dispatch by ..." | Re-trigger Greptile}

[greptile-apps]

The assertion was relaxed from toHaveBeenCalledTimes(1) to toHaveBeenCalled(), which no longer guards against the warning being emitted multiple times. If a refactor introduces duplicate warn calls on the fallback path this test would still pass silently. The exact-count assertion was the tighter, more useful signal here unless there's a known reason it fires more than once in this scenario.

Suggested change

	expect(warnSpy).toHaveBeenCalled();
	expect(warnSpy).toHaveBeenCalledTimes(1);

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!