[fields] the readonly `datetime` WIDGET faces are the same omission one type over — a bare `toLocaleDateString` + `toLocaleTimeString` pair, not `formatDateTime`

[os-justin]

Found while implementing objectui#8194 (converging the four readonly date widget faces onto formatDate). Render-measured on origin/main 158d75bb3. Filing unassigned — recording, not claiming.

⛔ Deliberately NOT folded into #8194: that card's landing point is formatDate's year-dropping decision, and the maintainer ruling it follows (objectui#7620, option A) is about formatDate and nothing else. These two sites are datetime faces whose one home is formatDateTime, and picking one of ITS named faces is a second display-convention decision.

The claim

#8194's population is defined by OMISSION — a date rendered with no options bag — so it was re-enumerated by mechanism rather than by spelling. Across the 78 non-test source files of @object-ui/fields the complete set of "a Date becomes user-visible text" mechanisms is toLocaleDateString (7 occurrences), toLocaleTimeString (2), and the shared formatDate / formatDateTime family. There is no Intl.DateTimeFormat, no toDateString / toUTCString, and no date-fns / dayjs / luxon / moment anywhere in the package. (toLocaleString, 4 occurrences, is number formatting.)

That enumeration returns six bare no-bag sites, not the four #8194 names. The two extra ones are these:

site	spelling
packages/fields/src/widgets/DateTimeField.tsx (readonly branch)	date.toLocaleDateString(locale) and date.toLocaleTimeString(locale), joined by a space
packages/fields/src/widgets/GridField.tsx (temporalText, the datetime branch)	the same pair, same join

Neither ever calls formatDateTime, which since objectui#7443 is the declared ONE home for the datetime display convention.

Measured, five locales, origin/main, TZ=UTC

Value 2026-07-04T07:00:00.000Z, rendered in the readonly DateTimeField and in the sub-grid datetime cell (both sites produce the identical string):

locale	these two sites render	formatDateTime 'compact' (what the datetime CELL renders)
en	7/4/2026 7:00:00 AM	7/4/2026 7:00 am
de	4.7.2026 07:00:00	4.7.2026 7:00 am
zh	2026/7/4 07:00:00	2026/7/4 上午7:00
ja	2026/7/4 7:00:00	2026/7/4 午前7:00
ar	4/7/2026 7:00:00 ص	4/7/2026 7:00 ص

(The ar forms carry U+200F marks around the date separators; they are omitted here so the table stays readable.)

The date half agrees with the compact face by coincidence — formatDateTimeCompactParts asks for { month: 'numeric', day: 'numeric', year: 'numeric' }, which is what Intl's no-options default happens to produce for these locales. The time half does not: these sites carry seconds and, where the locale writes one, a differently-cased meridiem.

⭐ One thing a triager should notice in that right-hand column rather than take from me: the compact face declares hour12: true, so the German session renders 7:00 am. date-display.ts records that as design ("the 12-hour face is its design, not a locale artefact"), which is exactly why "obviously converge on compact" is not a free move for a readonly FORM field.

Why it is worth a card

Same mechanism as objectui#7443, objectui#7620, objectui#8194 and objectui#4576 — more than one option bag for one concept (here: one bag and one absence of a bag), kept in step by nothing. #7443 gave datetime one home and converged the CELL onto it under a constraint that no pixel move elsewhere; these two widget faces are what that constraint left behind, and nothing says out loud that they opted out.

What the answer is NOT, yet

Converging them requires choosing WHICH formatDateTime face, and that is the decision:

• 'compact' — what every datetime CELL renders. Matches the cell beside it; drops the seconds these sites currently show, and imports the forced 12-hour face above.
• the verbose default (Jul 4, 2026, 07:00 AM in en) — what a non-cell caller (dataset measure, gantt tooltip, data-table) gets. Arguably the right register for a readonly FORM field, and it also carries the year-dropping question date has the same two-convention split datetime just had: data-table's date-only cell keeps a year that formatDate drops #7620 answered for date but nobody has asked for datetime.

Either way pixels move on the readonly form / detail surfaces and in the sub-grid, so it wants the treatment #7620 and #8194 got — a ruling, then a pin against a verbatim copy of the former face — not a drive-by edit. ⛔ In particular the answer is NOT "obviously compact, to match the cell": DateTimeCellRenderer reads field.format and defaults to 'compact', while these widgets read no style at all, so "match the cell" is itself a proposal to give the widgets a style vocabulary.

Grading

Observation-class. Every face is locale-correct, no data is lost or misread. objectui#8194's PR leaves both sites untouched on purpose and pins that boundary in packages/fields/src/__tests__/fields-date-widget-convention-8194.test.tsx (SCOPE FENCE block), so this card has a measured starting point rather than a prose one.

Related: objectui#8194, objectui#7443, objectui#7620, objectui#4576.

[os-justin]

转入决策箱 —— 卡片自己写了「答案还不是什么」,余下的是一次显示约定裁决

domain:ui 执行席 session_017bzjQQsBrHdwBy17iHVVe9,2026-09-16T10:0xZ。本卡带 pm:queue 坐在可派发队列里,但它的 ## What the answer is NOT, yet 一节明写:收敛这两个面需要先选定 formatDateTime 的哪一张脸,而那是一次显示约定决策 ⇒ 状态修为 needs-user-decision。⛔ 无代码改动,⛔ 无重新测量——卡片的五语种渲染表是 render-measured 的,且 objectui#8194 的 PR 已经用 SCOPE FENCE 把这两个站点的当前面逐字钉住,所以本卡有一个测量过的起点。

Governing text: 四棱决策框架第三轴与第四轴(创业阶段聚焦,维护者 2026-08-04)· 本项目基本裁决原则(维护者 2026-09-10:spec 声明 > 实现 > 文档面)· objectui#7443 已裁「datetime 只有一个家 = formatDateTime」。

---

维护者速读

事情: 只读的日期时间字段,在表单/详情页上和在表格单元格里,显示得不一样 —— 而它们本该是同一个约定。

两个站点(只读的 DateTimeField、子表格的 datetime 格子)自己拼 toLocaleDateString + toLocaleTimeString,从不走项目声明的唯一入口 formatDateTime。后果是它们带秒,而单元格不带;并且在写「上午/下午」的语种里大小写也不一致。

语种	这两个站点	单元格
en	7/4/2026 7:00:00 AM	7/4/2026 7:00 am
de	4.7.2026 07:00:00	4.7.2026 7:00 am
zh	2026/7/4 07:00:00	2026/7/4 上午7:00

⚠️ 没有数据错误,没有语种错误 —— 每一种写法本身都是对的。问题是一个概念两套写法,靠什么都不维系。

两个选项 —— 收敛到 formatDateTime 的哪张脸:

• A —— 'compact'(单元格用的那张)。 和旁边的格子长得一样。代价:丢掉现在显示的秒,并且把强制 12 小时制带进来 —— 德语会话会渲染成 7:00 am。卡片查到 date-display.ts 把这条记为有意设计,不是语种瑕疵。
• B —— 冗长默认面(席位推荐),en 下是 Jul 4, 2026, 07:00 AM。这是所有非单元格调用方(数据集度量、甘特悬浮、数据表)今天已经拿到的那张脸。

⚠️ 答案不是「显然用 compact,好跟单元格一致」 —— 单元格是读 field.format 之后默认到 compact 的,而这两个 widget 根本不读任何 style。所以「跟单元格一致」本身就是一个提案:给 widget 新增一套 style 词汇。

四棱分析:

轴	读数
实际业务需求	⚠️ 未实测哪一张脸有用户拉动。但只读表单字段与密集表格是不同的显示register,这是产品判断不是技术判断 —— 这一轴是你的,不是席位的。
项目长远合理性	objectui#7443 已裁「一个概念一个家」。两条路都收敛到那个家 ⇒ 这一轴不区分 A 和 B,只否决「维持现状」。
防 AI 写错	⭐ 指向 B。A 需要先给 widget 一套 style 词汇才能自圆其说,那是新增声明面;B 用的是已经存在、已经是默认的那张脸,零新增。
创业阶段不扩散	⭐ 同样指向 B。A 的隐含成本是一套新的 widget style 词汇 + 把一个「有意的」12 小时制约束扩散到只读表单面。零拉动 ⇒ 默认从紧。

⇒ 席位推荐 B(冗长默认面),三、四两轴同向,第二轴中立。⚠️ 两条路都会动像素(只读表单/详情面与子表格),所以无论选哪条,都应照 objectui#7620 / objectui#8194 的待遇:先裁决,再对旧面留一份逐字拷贝的 pin。⚠️ 另有一个 B 会顺带撞上的问题:date 的省略年份规则(objectui#7620 已答)没有人替 datetime 问过。选 B 请顺便说一句 datetime 要不要省略年份。

你要做的: 回 A(compact,跟单元格一致)或 B(冗长默认面,推荐);选 B 请附一句 datetime 是否省略年份。

_{domain:ui execution seat · session_017bzjQQsBrHdwBy17iHVVe9 · 2026-09-16T10:0xZ}

---

Generated by Claude Code

[hotlong]

Ruling: batch #142 item 2 · letter = per register (readonly form → B default face; sub-grid cell → A compact) · maintainer 「同意」 2026-09-16T11:41Z

Director seat, summon #24, session_01Wj1HUjzyeiBQ8atRf1ZhaL. Presented as batch #142 item 2 with the recommendation each site takes the face of its register, both through the one home formatDateTime; the maintainer's reply, verbatim: 「同意」.

Ruling

site	register	face	call shape
packages/fields/src/widgets/DateTimeField.tsx (readonly branch)	form / detail	default (verbose) — Jul 4, 2026, 07:00 AM in en, what every non-cell caller already gets	formatDateTime(value, { locale })
packages/fields/src/widgets/GridField.tsx (temporalText, datetime branch)	grid cell	'compact' — the face the sibling datetime cell renders (DateTimeCellRenderer reads field.format, default 'compact')	formatDateTime(value, { style: field.format ?? 'compact', locale })

• Both sites stop composing toLocaleDateString + toLocaleTimeString bags; objectui#7443's 「datetime has one home」 is now true for all six bare sites the card enumerated.
• ⛔ No new widget style vocabulary: the sub-grid cell reads field.format the way DateTimeCellRenderer does — reuse, not a new declaration.
• datetime does not drop the current year. The year-drop is formatDate's (date-only) cell rule per objectui#7620; the verbose default face carries the year, and nothing here extends the drop to datetime.
• Precedent basis: objectui#7443 ruling B (signature unchanged, 'compact' inside options), objectui#7620 ruling A (one face per register within the grid family), date-display.ts:264 (hour12 is the compact cell's design).

Execution (domain:ui seat)

• Card: needs-user-decision → pm:queue.
• Pixels move on the readonly form/detail surface and in the sub-grid ⇒ objectui#7620 / [fields] the readonly date WIDGET faces are a third convention — a bare toLocaleDateString(locale), not formatDate #8194 treatment: first a verbatim pin of the former face (objectui#8194's SCOPE FENCE block already pins today's strings — convert it to a flip pin), then the change, then the five-locale render table re-taken.
• Fallback recorded for the record: both sites on the default face, if the per-register split proves unimplementable at the sub-grid (it has the field, so this is not expected).

Four-facet reading: as presented in batch #142.

---

Generated by Claude Code