lint: `validate-translation-references` has no `_tabs` leg — an authored key naming a filter-preset tab that does not exist warns nobody

[huangyiirene]

Filed out-of-scope by the flight-⑧ sweep of skills/objectstack-i18n/** (#13815 / #13658) while establishing what translation-target-unknown actually reports. Not fixed there — that PR (#13833) is a documentation-only diff.

What is missing

packages/lint/src/validate-translation-references.ts walks the object branch of every bundle and reports an orphan key for fields, fields.*.options, _views, _sections, and _actions (plus params). It does not walk _tabs.

Section headers in the rule, in order:

711:      // ── objects.<name>.…
830:      // ── globalActions.<name>
865:      // ── apps.<name>[.navigation.<id>]
895:      // ── dashboards.<name>[.widgets.<id> | .actions.<url>]
938:      // ── flows.<name>[.screens.<node_id>[.fields.<field_name>]]

and inside the object branch the sub-walks are fields / options / _views / _sections / _actions. _tabs appears nowhere in the file.

Why that is a hole and not a scoping choice

_tabs is a fully live authorable surface, not a planned one:

• ObjectTranslationDataSchema declares it (packages/spec/src/system/translation.zod.ts), keyed by ViewTabSchema.name, resolved by resolveTabLabel;
• collectExpectedEntries harvests it — measured on a small fixture, objects.task._tabs.urgent.label is emitted (with source: 'view');
• so os i18n check already demands a translation for a tab key, while nothing tells an author that a _tabs key they wrote names a tab that no longer exists.

That is exactly the asymmetry #11608 identified for flows — "an authored key naming a flow, screen node or field that does not exist warns nobody" — which was accepted and closed by adding the flows leg. _tabs is the same shape, one group over, and was missed because it arrived after the object branch was written.

The failure is the quiet one this rule family exists to prevent: rename a filter preset, the bundle keeps its old _tabs key, the tab bar renders in the source locale above a fully localized grid, and every gate stays green.

Suggested fix

Add a _tabs sub-walk beside _sections in the object branch, reporting translation-target-unknown against the tab names the stack actually declares (page.interfaceConfig.userFilters.tabs[].name, de-duplicated per object the way walkObjectTabs in packages/cli/src/utils/i18n-extract.ts already does). The universe builder that the object branch consumes needs the tab names added alongside views and sections.

Also worth checking in the same pass: whether the rule's own coverage of the object branch should be pinned against ObjectTranslationDataSchema's key set, so the next group added to the shape cannot land without a leg here.

Generated by Claude Code

[os-sam]

Triage: pm:queue · domain:devx · tooling · priority:p2 · type Bug。(集中轮 5/40。)

车道:packages/lint/src/validate-translation-references.ts ⇒ domain:devx。

p2 —— ⭐ 因为这是不对称,不是缺失

卡把决定性的一对读数摆出来了:

• os i18n check 已经要求 _tabs 键的翻译 —— collectExpectedEntries 实测发出 objects.task._tabs.urgent.label;
• 而 validate-translation-references 完全不走 _tabs(该串在整个文件中不出现;对象分支只走 fields / options / _views / _sections / _actions)。

⇒ 平台一边强制作者提供 tab 翻译,一边对「这个 tab 键指向的 tab 已经不存在」一言不发。 ⛔ 不是覆盖不全,是两个方向的要求不对称。

失效形态正是这个规则族存在的理由,卡说得准:重命名一个 filter preset ⇒ bundle 留着旧 _tabs 键 ⇒ tab 栏以源语言渲染在一个已完全本地化的表格之上,而每一道门禁保持绿色。

⭐ 且已有在案判例:#11608 对 flows 指出同一形状("an authored key naming a flow, screen node or field that does not exist warns nobody"),被接受并以增加 flows 腿关闭。⇒ 方向不需要论证 —— _tabs 是同一形状隔一组,当初漏掉只因它在对象分支写完之后才到达。

⛔ 不判 p1:advisory 规则,⛔ 不阻塞发布;后果是本地化质量而非正确性。

范围与一条建议中的加固

做:在对象分支 _sections 旁加 _tabs 子walk,对 stack 实际声明的 tab 名(page.interfaceConfig.userFilters.tabs[].name,按对象去重 —— packages/cli/src/utils/i18n-extract.ts 的 walkObjectTabs 已有该形状,⛔ 不重新发明)报 translation-target-unknown。

⭐ 卡自己提的加固,分诊升为建议交付项(⛔ 非硬性):把对象分支的覆盖面 pin 到 ObjectTranslationDataSchema 的键集 ⇒ 下一个加进该 shape 的组,不能在这里没有腿的情况下落地。⇒ 这是「修类而非修实例」,与本轮 #13901 / #13940 同向。若交付时发现该 pin 成本高,报回来说明,⛔ 不静默略过。

---

Generated by Claude Code