[Aymericr]

What does this PR do?

Fixes the quality job on main, which has been red since #903. It also fixes 22 order-dependent nodes tests.

Root cause on CI (core, 5 verify_scene tests). Bun runs all of a package's test files in one process, in filesystem enumeration order. That order differs between machines and changes when a file is added. verify_scene looks up a kind's schema in the node registry before falling back to core's own schemas. services/drag-session.test.ts and services/alignment-anchors.test.ts register a stub wall kind with schema z.object({ type: z.literal('wall') }) and never restore the registry. So when those files run before agent-operations/operations.test.ts, malformed walls pass validation: start: null, or a pool retyped as wall.

On the last green run, hosting.test.ts happened to run in between, and its beforeEach wiped the registry by accident. #903 changed no logic involved here. It only added a test file, which reordered the runner's file list so nothing cleaned up before operations.test.ts. Running core in CI's exact 336-file order reproduces exactly those 5 failures locally.

Changes

• core: add a test preload (packages/core/test-setup.ts) that snapshots nodeRegistry before each test and restores it after. Preload hooks wrap each file's own hooks, so registrations made inside a test or its beforeEach (including after a _reset()) end with that test. Stub registrations made at file scope (top level or beforeAll), which a per-test hook can't reach, now restore in afterAll: validate-dwv, system-graph, port-connectivity, stair-surface-snap and procedural-items/mutations.
• nodes, the 22 failures: wall-batch-system.test.ts set useViewer's selection.selectedIds and previewSelectedIds to Sets. The store holds arrays. It never restored them, so later stair, column and floor-plan tests threw on .includes. The test now uses arrays and restores the viewer state it started with.
• nodes, leaks found while verifying:
  • The stair paint test now loads the built-in plugin itself instead of relying on an earlier test in its file.
  • cabinet/hosting and shelf-hosting start a mounted move session that writes the placement preview and alignment guides, and never commit or cancel it. They now clear both in afterEach. Leftover guides made a later stair tool mount render DOM labels.
  • Six hosting tests restored document by assignment, which left an own document: undefined property instead of removing it. They now restore by property descriptor, the way they already restore window.
• nodes guardrail: a test preload (packages/nodes/test-setup.ts) snapshots and restores, around every test, the node registry and the useScene, useLiveNodeOverrides, useLiveTransforms, useViewer, useEditor, usePlacementPreview and useAlignmentGuides stores (the built @pascal-app/* instances).

No timeouts were raised and no tests were skipped.

Not done: a random-order CI step. bun test --randomize still finds a few latent dependencies unrelated to these fixes, so as a required gate it would flake today:

• a test-to-test dependency inside core/src/lib/structure-kernel.test.ts (seed 12);
• the core spatialGridManager singleton in nodes frame-pipeline.test.tsx (seed 6).

The per-test reset preloads are the guardrail in this PR.

How to test

1. bunx turbo run build --filter=@pascal-app/nodes..., so core's and nodes' self-imports resolve to dist.
2. In packages/core, run bun test ./src/services/drag-session.test.ts ./src/agent-operations/operations.test.ts. On main this reports the same 5 failures as CI; on this branch it reports 0 failures.
3. In packages/core, run bun test src. Expected: 5414 pass, 0 fail. Under --randomize, seeds 1, 2, 3, 11, 13 and 14 also pass.
4. In packages/nodes, run bun test. On main: 22 failures. On this branch: 5021 pass, 0 fail. Under --randomize, seeds 1, 2, 3, 4, 5, 7 and 8 also pass.

Screenshots / screen recording

N/A — non-visual change (tests only).

Checklist

• I've tested this locally with bun dev
• My code follows the existing code style (run bun check to verify)
• I've updated relevant documentation (if applicable)
• This PR targets the main branch

  

---

Note

Low Risk
Test-only harness and teardown changes; no production runtime behavior is modified.

^{Reviewed by Cursor Bugbot for commit 7314993. Bugbot is set up for automated code reviews on this repo. Configure here.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-21T22:28:53.678037Z	7314993	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.