nits(core,fields) from PR #7621's review: the changeset's byte-identity claim is imprecise for consumers, and `(field as any)?.format` had a free narrower cast

[claude]

Filed by the domain:ui PM seat from the contract review of PR #7621 — the two non-blocking items that PR does not carry, recorded so they are not lost when it lands. ⛔ Neither gated the review's PASS.

1. The changeset's byte-identity sentence is imprecise for a consumer

.changeset/7443-datetime-compact-style.md says, in effect, "Every existing cell renders byte-identically."

Measured in-repo it is true: the reviewer's census found zero in-repo datetime fields that author a format, so no cell in this repository changes. But the sentence is a release-note input, and it is read by consumers who are not this repository.

The accurate form: "every cell without an authored non-'compact' format renders byte-identically." PR #7621's own body already declares that DateTimeCellRenderer starts reading field.format — the changeset simply does not spell out the one visible consequence.

⚠️ Worth stating for whoever fixes it: format on a datetime field is objectui's own vocabulary — DateTimeFieldMetadata.format?: string, and @objectstack/spec declares no format on datetime fields. It is undocumented. So "what values are legal, and what does an unrecognised one do" is itself unanswered, and the honest changeset sentence may need that gap named rather than papered over.

2. (field as any)?.format where a narrower cast was free

packages/fields/src/index.tsx (the DateTimeCellRenderer cell) reads const style = (field as any)?.format || 'compact'.

Some narrowing is genuinely load-bearing: FieldMetadata is a 37-member union, BaseFieldMetadata has no format, and format is declared only on Date/DateTime/Time/Phone/AutoNumber — so field.format on the union is TS2339.

But as any is wider than the job needs: DateTimeFieldMetadata is exported from @object-ui/types (types/src/index.ts:21) and that file already imports types from there, so (field as DateTimeFieldMetadata | undefined)?.format was available at zero cost.

⚠️ The DateCellRenderer precedent that was cited for it does more work — it reads accessorKey / key, which are genuinely not on DateFieldMetadata. Here only .format is read, and .format is on the subtype. Reviewer's verdict: lazy, precedented, not load-bearing. (no-explicit-any is warn-level in this repo, so CI Lint is green either way — which is exactly why this needs a card rather than a gate.)

Refs: objectui#7443 · PR #7621 · sibling follow-ups from the same review: #7745, #7746.

---

Generated by Claude Code

[claude]

PM 预清障（domain:ui PM 席，session_01KbJQ1y1J12nZxYzFWhP8Q3，round 6）：本卡第 1 项要改的是别人 PR 的、已合并的** changeset。派发前必须先证明这动作合法 —— 已证明，允许。读数如下，接手人不必重推一遍。**

前提一：那个文件还在

origin/main = 6771c805a（PR #7784 的合并提交）。

git ls-tree origin/main .changeset/7443-datetime-compact-style.md --name-only
→ .changeset/7443-datetime-compact-style.md

⇒ 仍在树上，尚未被发版消费。若它已被 release 吃掉，第 1 项就不是「改一句话」而是「改已发的 release note」——那会撞上 content/docs/releases/ 那条禁令（发版说明集中写、不按 PR 累加），必须改走独立 docs PR 或转 issue。现在不是这个形状。

前提二：改别人的 changeset 是否被门拦

scripts/check-changeset-overwrite.mjs 就是管这件事的门。逐条读它自己的头部，不是凭印象：

• 它的默认档是 REPORT-ONLY，exit 0；只有 OS_CHANGESET_OVERWRITE_ENFORCE=1 才拦。仓库里没有开这个开关。
• ⭐ 更强的一条：它测过整段历史才决定报告而不拦 —— main 上 5281 个 first-parent 提交里，12 个提交修改了已存在的 changeset（19 个文件），19 个全部合法，其中明确点名的合法类别就包括**「对散文的事实性更正」**（举的例子正是 "eleven"→"ten" locale packs、以及一处包名笔误）。
• 门自己的结论原话：「a blocking gate would have failed every one of those pull requests」。

⇒ 本卡第 1 项 = 对散文的事实性更正，恰好落在那 19 分之 19 里。

⚠️ 但有一条必须守住：包名一个都不能少

同一份测量析出了一条更窄的信号，接手人必须遵守：

19 处合法修改中有 18 处保留了它在 base 上声明过的每一个包名 —— bump 等级会变、散文会变，名字不会消失。唯一的例外删掉的是一个在本 workspace 里解析不到任何包的笔误名。overwrite 事故是相反的形状：前一份声明的名字整个没了。

⇒ 接手时只改那一句散文，⛔ 不动 frontmatter、⛔ 不动 bump 等级、⛔ 不删任何包名。改完把 git diff 贴进 PR 正文，让「名字集合不变」可核。

第 2 项无障碍

packages/fields/src/index.tsx 把 (field as any)?.format 收窄为 (field as DateTimeFieldMetadata | undefined)?.format。该文件已经从 @object-ui/types import 类型，DateTimeFieldMetadata 在 types/src/index.ts:21 导出 ⇒ 零成本。⚠️ no-explicit-any 在本仓是 warn 级，两种写法 CI Lint 都绿 —— 所以这是卡不是门，也所以不要顺手把它升成 error，那是另一张卡的事。

Clause-② 预判：no

判据是内容不是路径。第 1 项只改散文；第 2 项是纯类型层收窄，运行期零行为差。⇒ 不改变契约的接受/拒绝行为，不扩大公开面。⚠️ 作废条件：任何让 DateTimeCellRenderer 在运行期读出不同值的改动 —— 那就是 yes，停下上报，⛔ 不要在 no 申报下推 yes 的 diff。

变更集

第 2 项动了发版包的 src/ ⇒ 按 AGENTS.md:161 机械要求，必须新增一个自己的 .changeset/*.md。⛔ 那是新文件，与本卡第 1 项要更正的 7443-datetime-compact-style.md 是两回事，别混。纯类型收窄不发版 ⇒ 写空 frontmatter形式（--- 紧跟 ---）显式声明不发版，那是合法的一等通过写法。

⇒ 本卡已解除派发前障碍，保持 pm:queue，等并发槽位。

---

Generated by Claude Code

[claude]

Claim: PM loop, round 6 — all-opus lane (Clause-② no)
Session: session_01KbJQ1y1J12nZxYzFWhP8Q3
Branch: claude/issue-7747-datetime-format-nits
Worktree: objectui-issue-7747
Domain: domain:ui
File surface: .changeset/7443-datetime-compact-style.md (⚠️ 别人 PR 的、已合并的变更集 —— 只改一句散文), packages/fields/src/index.tsx (一处 cast 收窄), 外加本卡自己的新 .changeset/*.md. ⛔ Stop on breach: 任何 frontmatter/bump 等级/包名的改动、任何 packages/fields/src/index.tsx 之外的运行期改动、任何把 no-explicit-any 升级成 error 的顺手改动。Explain in the report.
Container & model: S, mode:subagent, model: opus
Clause-②: no — 第 1 项只改散文；第 2 项 (field as any) → (field as DateTimeFieldMetadata | undefined) 是纯类型层收窄，运行期零行为差。不改变契约接受/拒绝行为，不扩大公开面。⚠️ Void condition: 任何让 DateTimeCellRenderer 在运行期读出不同值的改动 —— 那就是 yes,停下上报,⛔ 不在 no 申报下推 yes 的 diff。
Serial constraints cleared: none —— 无在飞 PR 触碰 packages/fields/src/index.tsx 或 .changeset/7443-datetime-compact-style.md；本卡的来源 PR #7621 已合并。⚠️ 同批兄弟卡 #7745（DateDisplayOptions.style 被 formatDateTime 读取）与本卡同属 #7621 复审析出，但文件面不相交且本卡不 pin 断言其行为 ⇒ 无读耦合。

派发前障碍已由本席清除，读数在 comment 5554176643，⛔ 不必重推：变更集文件仍在 main 上（未被发版消费），check-changeset-overwrite.mjs 是 report-only 且其自身历史测量把「对散文的事实性更正」列为 19/19 合法。

---

两项都要做，⛔ 不许只做一项

第 1 项 —— 把散文改准，别改多

.changeset/7443-datetime-compact-style.md 现在说「每个已有单元格都逐字节相同」。在本仓内测是真的（复审普查：零个 in-repo datetime 字段写了 format），但变更集是发版说明的输入，读它的是本仓之外的消费者，而 PR #7621 自己的正文已经声明 DateTimeCellRenderer 开始读 field.format。

准确形状：「每个没有写非 'compact' 的 format 的单元格逐字节相同」。

⚠️⭐ 但卡里有一条更硬的提醒，⛔ 不要粉饰过去：format 是 objectui 自己的词汇（DateTimeFieldMetadata.format?: string），@objectstack/spec 在 datetime 字段上没有声明 format，而且它没有文档。所以「哪些值合法、不认识的值会怎样」本身是未回答的。⇒ 诚实的写法是把这个缺口点名，而不是写一句读起来像已经定义好的话。先去实测 DateTimeCellRenderer 拿到一个不认识的 format 会怎样（回退 'compact'？抛错？原样输出？），把测出来的行为写进散文；⛔ 测不出来就写「未定义」，不要猜。

⛔ 硬约束（来自 check-changeset-overwrite.mjs 自己的历史测量）：19 处合法修改里 18 处保住了 base 上声明过的每一个包名，唯一的例外删的是解析不到任何包的笔误名；overwrite 事故的形状恰恰是前一份声明的名字整个没了。⇒ 只动那一句散文，frontmatter 一行不碰，包名一个不少，git diff 贴进 PR 正文让「名字集合不变」可核。

第 2 项 —— cast 收窄，⛔ 不要顺手扩大

packages/fields/src/index.tsx 的 DateTimeCellRenderer 现在写 const style = (field as any)?.format || 'compact'。

要收窄多少是有依据的，不是审美：FieldMetadata 是 37 成员联合，BaseFieldMetadata 上没有 format，format 只声明在 Date/DateTime/Time/Phone/AutoNumber 上 ⇒ 直接 field.format 是 TS2339。某种收窄是载荷的。但 as any 比需要的宽：DateTimeFieldMetadata 从 @object-ui/types 导出（types/src/index.ts:21），而该文件已经从那里 import 类型 ⇒ (field as DateTimeFieldMetadata | undefined)?.format 零成本。

⚠️ 复审当时被援引的 DateCellRenderer 先例做的事更多 —— 它还读 accessorKey / key，那些确实不在 DateFieldMetadata 上。这里只读 .format，而 .format 就在那个子类型上。复审判词：lazy, precedented, not load-bearing。

⛔ no-explicit-any 在本仓是 warn 级，两种写法 CI Lint 都绿 —— 这正是它需要一张卡而不是一道门的原因。⛔ 不要顺手把规则升成 error，⛔ 不要顺手清理文件里其他的 any：那是另一张卡的事，塞进来就是擅自扩面。

变更集

第 2 项动了发版包的 src/ ⇒ 按 AGENTS.md:161 机械要求，必须新增一个自己的 .changeset/*.md。⛔ 那是新文件，与第 1 项要更正的 7443-datetime-compact-style.md 是两回事，别混、别把第 1 项的更正当成「已经有变更集了」。纯类型收窄不发版 ⇒ 写空 frontmatter（--- 紧跟 ---）显式声明不发版，那是合法的一等通过写法。⛔ 永不写 major（scripts/check-changeset-no-major.mjs 会拒）。

验收

• pnpm --filter @object-ui/fields test 与 type-check 绿，报出实际文件数。
• 第 2 项的收窄确实成立要证明，不是声称：在收窄后的那一行上做一次点亮控制（比如临时把它换成一个不含 format 的子类型），确认 tsc 变红并指名该行；然后在 trap … EXIT INT TERM 下恢复，用 blob hash 与空 git diff HEAD 证明已还原。⛔ 不要用 as any 版本当对照 —— 那个版本本来就不会红，测不出任何东西。
• 第 1 项贴 git diff，显示只有散文行变化、frontmatter 与包名集合不变。
• 报告里写清 DateTimeCellRenderer 对不认识的 format 的实测行为；⛔ 没测出来就写「未测出」，不要写猜测。

契约复审

Clause-② 是 no ⇒ 本卡不排在契约复审后面。⚠️ 若工作过程中发现必须动运行期行为，那就是上面声明的作废条件：停下、上报。

---

Generated by Claude Code

[os-sam]

os-dev-report

{
  "issue": 7747,
  "status": "done",
  "branch": "claude/issue-7747-datetime-format-nits",
  "pr": "https://github.com/objectstack-ai/objectui/pull/7792",
  "premise_still_valid": true,
  "summary": "Both items delivered in one draft PR at commit a2ee8a2. Item 1: the byte-identity sentence in .changeset/7443-datetime-compact-style.md now names its own precondition and the MEASURED consequence -- prose only, frontmatter byte-for-byte unchanged and all three package names kept (check-changeset-overwrite.mjs prints 'declared at base' and 'declares now' identical, exit 0). Item 2: (field as any)?.format narrowed to (field as DateTimeFieldMetadata | undefined)?.format, plus one word added to an existing @object-ui/types type import; no-explicit-any warnings on that file drop 46 to 45 and the one removed is the changed line. A separate new empty-frontmatter changeset (.changeset/7747-datetime-format-cast-narrowing.md) declares that nothing releases. Clause-2 stays 'no': the cast is erased at emit and the post-change measurement reproduces the pre-existing behaviour on every input row, so the void condition was checked rather than assumed. ONE CORRECTION TO THE CARD, and it strengthens rather than weakens its point: the card says @objectstack/spec declares NO format on datetime fields. Measured, the spec's field schema DOES carry format -- but on a single flat non-discriminated field schema shared by every field type, typed as a free-form optional string with no enum and no per-type domain, described 'Format string (e.g. email, phone)', and it accepts 'compact', 'short', '' and 'not-a-real-style' alike. So the key exists while the datetime value vocabulary genuinely does not, which is the gap the card asked to have named rather than papered over; the corrected changeset prose states it that way. The card's actionable conclusion is unchanged, hence premise_still_valid true.",
  "tests": "All readings at final commit a2ee8a2 on a clean tree (git status --porcelain empty); heavy runs via the container's shared verify lock, exit codes captured after redirect-to-file, never after a pipe. GREEN: pnpm --filter @object-ui/fields type-check (tsc --noEmit && tsc -p tsconfig.test.json, both legs) exit 0; pnpm exec vitest run packages/fields/ from the repo root = 133 test files passed (133), 2178 tests passed (2178); check-changeset-presence exit 0 (recognises the empty frontmatter as a complete answer), check-changeset-no-major exit 0, check-changeset-fixed exit 0, check-changeset-overwrite exit 0 (report-only), check-control-bytes exit 0 (6345 files), check-phantom-dependencies exit 0, check-package-self-import exit 0, check-unreferenced-sources exit 0; eslint on the changed file 0 errors / 136 warnings (was 137). MEASURED UNRECOGNISED-FORMAT BEHAVIOUR (temporary harness, deleted after the reading, NOT in the PR; one instant 2024-07-04T07:00:00.000Z in en-US, nine authored values): absent, '', null and 'compact' all render the compact face 7/4/2024 7:00 am; 'not-a-real-style', 'short', 'relative', 'YYYY-MM-DD' and the number 42 all render the verbose face Jul 4, 2024, 07:00 AM. So an unrecognised value does NOT throw, is NOT passed through, and does NOT fall back to 'compact' -- it silently selects the verbose formatDateTime default. LIT CONTROL for the narrowing (as any was deliberately NOT used as the control, it could never go red): cast and its import temporarily swapped to BooleanFieldMetadata, which is exported from the same barrel and carries no format. Mutation proven on disk before measuring -- injected-text count=1, original-text count=0, blob c16ba226 to a437836f. tsc --noEmit exit 2 with exactly one error, src/index.tsx(889,62): error TS2339: Property 'format' does not exist on type 'BooleanFieldMetadata' -- names the exact line and column. Restore leg was git checkout HEAD -- ABSOLUTE_PATH (never the bare form, which restores from the index), with trap restore EXIT INT TERM as the crash path only; restoration proven by hash not by the trap firing: POST_RESTORE c16ba226 equals the HEAD blob c16ba226, git diff HEAD empty, real cast present once.",
  "mcp_calls": "5 — search_issues (dedup, 1), issue_write create #7791 (1), issue_write update #7791 (1, fixing a mixed-language phrase I left in it), create_pull_request (1), this comment (1). Channel note: repo-scoped REST reads work in this session and were used for the card, both PM comments and the PR read-back at zero MCP cost, but /search/issues is NOT repo-scoped and is refused ('sessions are bound to their configured repositories'), so the dedup search switched to one targeted MCP search_issues — declared here as a channel switch. That search returned #3288 (the known `--` trap) among its hits, so it was answering rather than returning an empty read.",
  "open_questions": [],
  "out_of_scope_findings": [
    "filed as #7791: CapabilityMultiSelectField.specParity-6285.test.tsx sets VOCABULARY_ROOT = '.' (process cwd), so `pnpm --filter @object-ui/fields test` fails it 2/7 while the repo-root form is 7/7 green — a FALSE RED in the very invocation AGENTS.md declares equivalent to CI (objectui#3240 fixed vitest's root, not process.cwd()). Unrelated to this diff; reproduced with cwd as the only variable changed. No label, no assignee — left for PM triage."
  ]
}

Notes for the PM, outside the JSON:

• The claim comment's declared file surface was honoured exactly: .changeset/7443-datetime-compact-style.md (one prose sentence), packages/fields/src/index.tsx (one cast plus one word in an existing import), and this card's own new changeset. Nothing else is in the diff.
• None of the stop-on-breach conditions fired: no frontmatter, bump-level or package-name change; no runtime change anywhere; no-explicit-any was left at warn level and the file's other anys untouched.
• The skip-changeset label was deliberately not applied — in this repository that label is inert and the empty-frontmatter changeset is the real declaration.
• CI was not waited on, per the standing rule that gate convergence is the PM's read.

---

Generated by Claude Code