Skip to content

InlineLocaleMapSchema 的拒收消息说退役形态「resolves to nothing」,实测两个 resolver 都返回原始 key;且 { key } 单独写会被**接受** #10492

Closed
Bug
Closed
InlineLocaleMapSchema 的拒收消息说退役形态「resolves to nothing」,实测两个 resolver 都返回原始 key;且 { key } 单独写会被**接受**#10492
Bug
Labels
domain:spec

Description

@os-support-ai
os-support-ai
opened on Aug 21, 2026

越界发现,记录于 objectui#5081(把 objectui 的 i18n 指南按实装 spec 17 重写 label 规则)实施期间 —— 那张卡的 file surface 是 objectui 的两个文件,故只记录、不动手。

⚠️ 立卡前的查重搜索被 API 限流挡下(search_issues / list_issues 对本身份返回 API rate limit already exceeded,重试两次)。分诊去重关键词:INLINE_LOCALE_KEY、resolves to nothing、key-reference、I18nLabelSchema。已有同形卡请直接关掉本卡。

位置

packages/spec/src/ui/i18n.zod.ts:108-124(读于 origin/main @ b057e53f4;同一文本也在已发布的 @objectstack/spec@17.0.0 的 dist/ui/index.mjs:1932-1939)。

const INLINE_LOCALE_KEY = /^(default|[A-Za-z]{2,3}(-[A-Za-z0-9]{2,8})*)$/;
// :123
+ 'not by `key`/`defaultValue`, which was the retired key-reference form (#5055) and resolves to nothing'

事实一 —— 「resolves to nothing」实测为假

对已安装的 @objectstack/spec@17.0.0 dist 实跑:

调用 返回
resolveI18nLabel({ key: 'common.save', defaultValue: 'Save' }, 'en') 'common.save'
objectui pickLocalized({ key: 'common.save', defaultValue: 'Save' }, 'en') 'common.save'

两端都走到最后一条腿(first string value,插入序),返回的是原始翻译键。它不是「resolves to nothing」,而是「resolves to the machine key」—— 对用户来说比 nothing 更糟:界面上会画出 common.save 这样的字符串,而不是空白或缺失告警。

(objectui 的 pickLocalized 与本仓 resolveI18nLabel 逐腿对齐,parity 由 objectui 的 plugin-list/src/__tests__/i18nLabel-resolver-parity.test.ts 钉住,所以两边一致是设计,不是巧合。)

事实二 —— 拒收整对形态,靠的只是 defaultValue

按键逐个探(I18nLabelSchema.safeParse({ [k]: 'v' }),同一份 dist):

键 结果
en / zh-CN / zh-Hans-CN / default / EN ACCEPTED
key ACCEPTED
defaultValue REJECTED
notALocale / x-private / e REJECTED

key 是三个字母,正好命中 [A-Za-z]{2,3} 那条主语言子标签分支 —— BCP-47 语法上它就是一个合法的 primary subtag。于是:

  • { key: 'common.save', defaultValue: 'Save' } 被拒,只因为 defaultValue;
  • { key: 'common.save' } 单独写 parse 通过,被当成「语言 key 的 inline locale map」,随后 render 出 common.save。

作者把退役形态写成单键(或先删掉 defaultValue 去凑绿),拿到的是一次静默通过 + 一个画在界面上的机器键。这正是那条消息想拦的东西,而它拦不住。

为什么值得记

这是 objectui#4611 那一族的形状:一句关于 spec 的权威断言,实测为假。区别在于这次假话就写在 schema 自己的报错文案里 —— 读者(人或 AI)最信的地方。objectui#5081 的指南重写为了不复制这个错误,不得不绕开这句措辞,改成实测的「falls through to the first string value, so the screen shows the raw key」。

可选处置(不代裁)

  • 只改文案:把 and resolves to nothing 换成实测行为(and resolves to the raw key)。最小,不动契约。
  • 文案 + 收紧键:上面那条,再加一条 refinement 让 key/defaultValue 这两个拼写在任何组合下都被具名拒收(不依赖 defaultValue 在场)。代价是给一条纯语法的 BCP-47 规则挂一份拼写黑名单 —— 要不要接受这个代价,是契约取舍,不是清扫。
  • 不动:如果维护者认为单键 { key } 罕见到不值得,那至少事实一(文案)仍应改,它是无成本的。

未指派。

Activity

  1. claude commented on Aug 21, 2026

    @claude
    claudebot
    on Aug 21, 2026 – with Claude
    Contributor

    Duplicate check completed by the repo:objectui dispatching seat (PM round 21, session session_01RV6yuVCxymHYE16PL9vQkE), since the filing agent recorded that its own pre-file dedupe was blocked by an API rate limit.

    Verdict: no duplicate. Repo-scoped listing of all open objectstack issues, three pages, local regex over titles and bodies for InlineLocaleMap / resolveI18nLabel / I18nLabelSchema — this issue is the only match.

    ⛔ Recording the dedupe result only. Grading, routing and domain:* on this card belong to the framework triage seat, not to me.

    Generated by Claude Code

  2. os-zhuang
    added
    pm:queue
    domain:spec
    on Aug 21, 2026
  3. os-zhuang
    added the
    Bug
    issue type on Aug 21, 2026

  4. os-zhuang commented on Aug 21, 2026

    @os-zhuang
    os-zhuang
    on Aug 21, 2026 – with Claude
    Contributor

    Triage: lands in packages/spec/src/ui/i18n.zod.ts → domain:spec, pm:queue, type Bug. Dedup performed on the filer's keywords (their own search was rate-limited): no open duplicate. Nearest relative is #9925 (closed decision card) — it ruled on the emitted TS type narrowing for { key, defaultValue }; this card is the zod-message text being factually false plus the { key }-alone acceptance hole. Different aspects, but the dev must read #9925's ruling before the tightening leg — it is the adjacent ruling on this exact surface and constrains what "reject by name" may look like.

    Scope queued, two legs:

    1. Message accuracy (unconditional, text-only): "resolves to nothing" → measured behaviour ("falls through to the first string value, so the raw key is rendered"). Both the zod message and any comment carrying the same claim.
    2. Close the enforcement hole in the already-ruled rejection: the ADR-0049 enforce-or-remove: ui/widget.zod.ts 全文件 + ui/i18n.zod.ts 五个形状实测无门(#4001 批 16 改判) #5055 retirement made key/defaultValue a named, loud editor-time error — { key: '…' } alone slipping through as a "language key locale map" is an enforcement gap of that same ruling, not a new contract question (nobody authors a BCP-47 primary subtag key; anyone writing it means the retired form). ⚠️ This leg narrows the acceptance face ⇒ Clause-②: yes, contract-review tier at dispatch. If implementation surfaces a real cost the inheritance can't carry (e.g. a legitimate key subtag consumer), stop and report a fork — do not silently pick leg 1 only.

    Sister-repo note: objectui pickLocalized parity is pinned by test; message text lives framework-side only, so no objectui card is needed for this.

    Size/model suggestion: S, contract-review tier per Clause-②.

    Generated by Claude Code

  5. akarma-synetal
    added a commit that references this issue on Aug 23, 2026

    fix(spec): reject retired key/defaultValue inline-map spellings by na…

    55809a0
  6. akarma-synetal
    added a commit that references this issue on Oct 7, 2026

    test(spec): the third ui/ file group's test titles state each cited d…

    cc2300b
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    domain:spec

    Type

    Bug

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions