console-starter 的 vite 别名表漏了 5 个源码实际 import 的工作区包,未先构建就 pnpm dev 得到白屏

越界发现,记录于 #3520(给 examples/console-starter 写 README)期间的实测启动。未在该 PR 中顺手修改;PR #3524 的 README 已把"必须先 pnpm -w build"如实写进"How to run",所以这是打磨类发现,不是今天会挡住按文档操作的读者的缺陷 —— 标 finding,不排队。

现象

examples/console-starter/vite.config.ts 把 24 个 @object-ui/* specifier 别名到 packages/*/src,注释写明用意是让插件的 side-effect 注册命中同一个 ComponentRegistry 单例。但这些 src 文件自己又 import 了 5 个既不在别名表、也不在 examples/console-starter/package.json 里的工作区包:

未覆盖的包	import 它的文件
`@object-ui/mobile`	`packages/plugin-list/src/ListView.tsx`
`@object-ui/providers`	`packages/app-shell/src/views/RecordAttachmentsPanel.tsx`
`@object-ui/sdui-parser`	`packages/components/src/renderers/layout/page.tsx`
`@object-ui/plugin-editor`	`packages/app-shell/src/views/metadata-admin/widgets.tsx`
`@object-ui/react-runtime`	`packages/components/src/renderers/layout/react-page.tsx`

这 5 个包只能退回 node 解析,落到 packages/*/dist —— 而 dist 只有构建之后才存在。

实测

在 origin/main 的干净 worktree 里,只 pnpm install(不 pnpm -w build)后 pnpm dev(自选空闲端口 5197):

VITE v8.2.0  ready in 389 ms
(!) Failed to run dependency scan. Skipping dependency pre-bundling. Error: The following dependencies are imported but could not be resolved:
  @object-ui/mobile (imported by .../packages/plugin-list/src/ListView.tsx)
  @object-ui/providers (imported by .../packages/app-shell/src/views/RecordAttachmentsPanel.tsx)
  @object-ui/sdui-parser (imported by .../packages/components/src/renderers/layout/page.tsx)
  @object-ui/plugin-editor (imported by .../packages/app-shell/src/views/metadata-admin/widgets.tsx)
  @object-ui/react-runtime (imported by .../packages/components/src/renderers/layout/react-page.tsx)

src/main.tsx 里 import '@object-ui/plugin-list' 是急切的,而 plugin-list/src/index.tsx 又急切 import { ListView } from './ListView',所以这条链一定会被浏览器走到:

$ curl -s -o /dev/null -w "%{http_code}" http://localhost:5197/@fs/.../packages/plugin-list/src/ListView.tsx
500   → Failed to resolve import "@object-ui/mobile" from "../../packages/plugin-list/src/ListView.tsx". Does the file exist?

无头 Chromium 打开 http://localhost:5197/:HTTP 200,但 #root 的 innerHTML 长度为 0,document.body.innerText 为空字符串,控制台是 6 条 500 —— 纯白屏,页面上没有任何提示。

补上这 5 个包(及其依赖)的构建后重启,同一页面正常渲染出品牌加载屏(ObjectOS / Initializing application... / Connecting to data source)。所以根因就是 dist 缺失,不是别的。

为什么标 `finding` 而不是缺陷

examples/README.md 与根 README.md 都已写明启动顺序是 pnpm install → pnpm -w build → cd examples/... && pnpm dev,按文档走不会踩到。踩到的是跳过构建的人 —— 而失败现场(白屏 + 无文案)完全不指认根因,这是它值得记一笔的地方。

可选的处理方向(不预设结论)

把这 5 个包补进 vite.config.ts 的别名表 —— 与现有 24 条同构,pnpm dev 从此不依赖 dist,对 example 的"开箱即跑"最直接。
把它们补进 examples/console-starter/package.json 的 dependencies —— 与 vite.config.ts 注释里那句"Customers shipping a real app can drop these aliases once the @object-ui/* packages are published to a registry"更一致:真去掉别名时,少了这 5 个声明的 fork 是缺依赖的(pnpm/npm 的 auto-install-peers 可能兜住 @object-ui/mobile 这类 peer,但 @object-ui/sdui-parser / @object-ui/react-runtime 是普通依赖链,值得单独核实)。
什么都不改,只靠文档 —— PR docs(examples): 给 console-starter 补 README #3524 的 README 已经是这一档。

方向 1 与 2 不互斥;哪个是长期正确取决于这个 example 到底承诺"在 monorepo 里能跑"还是"fork 出去能跑",这是维护者的判断,故此单不预设。

备注:查重未能完成

立单时 list_issues 持续返回 API rate limit already exceeded(共享 GitHub 身份,多 agent 并行),约 20 分钟冷却后仍未恢复,因此未能对 open issues 做关键字/文件路径查重。若已有同源单,请按惯例 race-close 本单。

Generated by Claude Code

发现分诊轮(持有):文档化的启动顺序(pnpm -w build 先行,PR #3524 的 README 已写明白屏症状与五个包名)绕开了该坑;三个修向的取舍取决于"console-starter 是否要支持 fork 出仓独立跑"——维护者裁量,不预设。README 已落地为第一道防线,伤害面收窄,继续持有。(objectui 车道 PM,会话 session_01GTRjn8xBqp75dk7kFupVRt)

Generated by Claude Code

Triage (maintainer-authorized one-off pass, 2026-08-09, registered on objectstack#6015): promoted finding → pm:queue. A silent white screen with zero on-page diagnosis is the failure shape, and the fix is measured and mechanical: add the 5 missing workspace packages to the alias table (direction 1, symmetric with the existing 24) and to the example's package.json dependencies (direction 2 — they are real imports of the fork story the config's own comment promises). The README workaround stays as documentation, not as the fix.

Generated by Claude Code

Claim: PM loop round 5 (objectui whole-repo seat, non-v17 queue wave)
Session: session_017Qqyix2QcnpUC9XeYVDzx3
Branch: claude/issue-3528-console-starter-aliases
Worktree: objectui-issue-3528
File surface: examples/console-starter only (vite.config.ts, package.json, and its README line if the run story changes). Serial constraints: disjoint from the other four dispatches of this batch.
Container & model: S, mode:subagent, model: opus

Delegated ruling on the card's open fork (authorization: maintainer's 2026-08-10 delegation to this seat; veto window open until merge): do both direction 1 and direction 2 — the card itself notes they are not mutually exclusive, and they answer different promises: the aliases keep pnpm dev working in-monorepo without a prior build (the example's demo purpose), the dependency declarations keep the fork-out story coherent with the config's own "drop these aliases once published" comment. Doing both removes the white-screen trap without betting on which promise the example is for. Direction 3 (docs-only) stays as the floor that PR #3524 already laid.

Generated by Claude Code

ACCEPT (objectui seat PM, session session_017Qqyix2QcnpUC9XeYVDzx3, review of record) — PR #4103, both directions of the delegated ruling implemented.

The acceptance-bar evidence is exactly the card reversed: fresh worktree, no build, zero dist dirs → dev server up, the five formerly-500 modules all 200, #root renders 2506 bytes where the card measured 0, zero 4xx/5xx. The closure test is the real prize — it re-derives the alias set from the actual import graph (with the fixed-point argument for why 29 is stable), so this class of drift now fails a test instead of a reader. Both CI laps were real defects correctly absorbed: the new test had to enter a tsc program (the type-check-coverage gate exists precisely to catch that), and the shard-4 pin's own comment asked for the member-list update it got — with the "must arrive wired up" clause now asserted per member.

Correction to my ruling's rationale, on the record: direction 2's justification does not survive the dev's verification — every consuming manifest already declares the five packages itself, so a fork dropping the aliases resolves them today; only @object-ui/mobile (peerDep of plugin-list) is genuinely load-bearing, the other four are self-documentation. The ruling STANDS as implemented — workspace:* declarations in a private example cost nothing and make the example's real surface explicit — but the record should show the dev's facts beat my rationale, and any future revisit starts from their table, not my claim. Rider findings #4101 (two more hand-maintained lists vs the real graph) and #4102 (README's pre-#4042 root-route claim — correctly NOT folded in, different cause) are well-filed.

Flipping ready + auto-merge.

Generated by Claude Code

fix(console-starter): close the vite alias table over its real import…

Navigation Menu

Repository navigation

console-starter 的 vite 别名表漏了 5 个源码实际 import 的工作区包,未先构建就 pnpm dev 得到白屏 #3528

现象

实测

为什么标 `finding` 而不是缺陷

可选的处理方向(不预设结论)

备注:查重未能完成

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

console-starter 的 vite 别名表漏了 5 个源码实际 import 的工作区包,未先构建就 pnpm dev 得到白屏 #3528

Description

现象

实测

为什么标 finding 而不是缺陷

可选的处理方向(不预设结论)

备注:查重未能完成

Activity

yinlianghui commented on Aug 7, 2026

os-project-manager commented on Aug 9, 2026

yinlianghui commented on Aug 10, 2026

yinlianghui commented on Aug 10, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

为什么标 `finding` 而不是缺陷