plugin-editor `test` task red on main: vite.config.ts alias table missing @object-ui/providers

[yinlianghui]

Finding

@object-ui/plugin-editor's test task is already red on main, independent of any in-flight change. Found by the #4178 dev while measuring turbo test inputs (flagged in PR #4188's body for triage; this is that card).

Mechanism: packages/plugin-editor/vite.config.ts aliases @object-ui/fields to source but does not alias @object-ui/providers. The package's heavy setup graph (root vitest.setup.tsx via setupFiles: ['../../vitest.setup.tsx']) reaches @object-ui/providers, so resolution fails:

Failed to resolve import "@object-ui/providers" from ...

This is masked in day-to-day CI because ci.yml runs the canonical root Vitest (sharded), not per-package turbo run test — the red only bites local turbo runs and any future CI use of the cached task.

Direction

This is the same defect family as #3240 (console-starter alias table missing 5 packages — @object-ui/providers among them), and #3240's ruling applies by analogy: complete the alias table to match what the code actually reaches, same shape as the existing entries, rather than trimming the reachable graph. Follow the ruled direction recorded on #3240; consider batching with #3240's quiet-window execution since both are alias-table completions.

Acceptance

• Premise check first: reproduce the red on current main (pnpm turbo run test --filter=@object-ui/plugin-editor --force) before touching anything — if it has been fixed in the interim, close with the evidence.
• Fix: add the missing alias entr(ies) so the setup graph resolves; check whether plugin-markdown (the other package reaching out only via setupFiles, per PR ci(turbo): derive the test inputs guard from each package's Vitest config program (#4178) #4188's derivation) has the same gap while in there.
• Evidence: red-before / green-after of the same command, plus the root-Vitest invocation staying green (no regression in the canonical path).

Refs: #3240 (ruled direction), #4178 / PR #4188 (discovery context).

[yinlianghui]

CLAIM (objectui whole-repo seat PM, session session_017Qqyix2QcnpUC9XeYVDzx3) — dispatching a dev agent now.

• Branch claude/issue-4194-plugin-editor-alias, worktree objectui-issue-4194 (S container)
• Scope per the card: premise-check the red on current main first (turbo run test --filter=@object-ui/plugin-editor --force), complete the alias table per 17 个包各自带 vitest.config.ts 与根 vitest.config.mts 行为分叉:先回答「哪些包真的需要本地配置」 #3240's ruled direction, check plugin-markdown for the same gap while there, red-before/green-after evidence + the canonical root-Vitest path staying green.

---

Generated by Claude Code

[yinlianghui]

ACCEPT (objectui seat PM, session session_017Qqyix2QcnpUC9XeYVDzx3, review of record) — PR #4218.

The card asked for one alias; the measurement returned six per config, correctly split by kind: three hard reds (providers, mobile, permissions — dist never exists in the ^build closure) and three SILENT DIVERGENCES (i18n, sdui-parser, data-objectstack resolved to dist where CI tests src — the per-package path was quietly testing different code, now aligned). The deliberate non-add of @object-ui/types/zod carries its reasoning (Vite prefix-order hazard, nothing in the reach set imports it). plugin-markdown swept as the card asked — identical gap, same PR, own before/after, and both are honestly characterized as no-load→loading (setup 0ms → real setup), not merely red→green. The reverse verification recorded its own precondition first (providers/dist present: no), closing the false-green door before walking through it. Changeset verdict arbitrated by the gate and confirmed by CI's own job: none owed. The #3240 non-collision note (neither package is among the 17 per-package vitest configs ruled for the quiet-window batch) is exactly the scope hygiene the packing note wanted. CI converged, zero failures. Flipping ready + auto-merge.

---

Generated by Claude Code