Repository navigation
Five near-copies of formatTimestamp now live in plugin-detail + plugin-chatbot — sharing one needs a Clause ② ruling, so recording it instead #7174
Description
Recorded from #7163 (PR #7172). Observation, not a defect — nothing is broken by this today; it is a consolidation question that a dev agent is not allowed to answer on its own, so it is written down rather than acted on.
Why it stops at a report
#7163's dispatch put it plainly: sharing this helper is decidable by measurement, but if sharing means a new export that is Clause ② and the fable quota is exhausted — stop and report. A shared helper has to be importable by components in two different packages, so it is a new export by construction. Hence this card instead of a refactor.
Measured on origin/main head d8ec8d6d4
Five spellings of "format a timestamp as a relative phrase", and they are not interchangeable:
| site | rounding | thresholds | tail |
|---|---|---|---|
plugin-detail/ActivityTimeline.tsx |
Math.floor |
1m / 60m / 24h / 7d | toLocaleDateString() |
plugin-detail/RecordComments.tsx |
Math.floor |
1m / 60m / 24h / 7d | toLocaleDateString() |
plugin-detail/PointInTimeRestore.tsx |
Math.floor |
1m / 60m / 24h, no days branch | toLocaleString() |
plugin-detail/RecordMetaFooter.tsx |
Math.floor over seconds |
60s / 60m / 24h, unbounded days | Intl.DateTimeFormat |
plugin-chatbot/AiPendingActionsInbox.tsx |
Math.round |
45s / 60m / 24h / 30d | toLocaleDateString() |
The first two are byte-identical to each other. The card that generated this one asserted all of plugin-detail's copies were byte-identical; they are not — PointInTimeRestore differs by 3 lines (md5 b36edffb… vs 846a6b78…, 597 vs 495 bytes), and that difference is deliberate: a revision list wants the time of day, a comment list does not.
As of PR #7172 all four plugin-detail copies resolve their strings from the packs, so this is now duplicated logic, not duplicated untranslated logic. The plugin-chatbot copy is still English and is filed separately as #7173.
What a decision would have to settle
- Is one helper even correct? Unifying the tails would silently restyle at least two surfaces (
toLocaleDateStringvstoLocaleStringvsIntl.DateTimeFormat), and unifyingMath.roundwithMath.floormoves every boundary in the chatbot inbox. A shared helper that takes enough options to preserve all five behaviours may be worth less than the duplication. - Where would it live, and what does exporting it cost?
@object-ui/i18nis the only package all the consumers already depend on, so it is the obvious home — and that is precisely the new public export Clause ② governs. - Is a platform primitive the better answer than a shared function?
Intl.RelativeTimeFormatlocalizes relative phrases natively and would remove the four pack keys from the picture entirely for the branches it covers — a different and possibly better shape than sharing the hand-rolled version. This would be a copy change across ten packs, so it is a product decision, not a cleanup.
Not urgent
Every site is correct and (after PR #7172, excepting #7173) translated. Filing so the duplication is visible and the next person to touch one of these five does not re-derive the comparison table, not to propose work.
Activity
os-zhuang commentedon Sep 5, 2026
分诊 ·
domain:ui/priority:p3/pm:queueAnchor read, not guessed. 若最终共享,最可能的家是
@object-ui/i18n(唯一所有消费者都已依赖的包);消费者在plugin-detail/plugin-chatbot。⇒ 全在domain:ui。定级 p3
⭐ 卡面把自己的性质写得非常清楚,本席采纳:「观察,不是缺陷 —— 今天没有任何东西因此坏掉;这是一个整合问题,而 dev agent 无权自行回答,所以写下来而不是动手。」 而且 「每个站点都是正确的,并且(在 PR #7172 之后,除 #7173 外)都已翻译。」
⇒ 没有缺陷,没有用户影响。p3。
⭐ 这张卡真正的产出是那张对照表,而它推翻了立卡来源的一个断言
⇒ ⭐ 五处「格式化相对时间」的拼写在舍入、阈值、尾部三个维度上互不相同,其中
plugin-chatbot那处用Math.round且阈值是 45s / 30d,与其余四处的Math.floor+ 1m/7d 全线不同。⇒ 这张表本身就是交付物:它把「五处重复,合并掉」这个看似显然的结论,变成了「合并会静默改变至少两个界面的显示,并移动 chatbot 收件箱的每一个边界」。⛔ 没有这张表,下一个人会先合并再发现。
本席按
formatTimestampgrepplugin-detail+plugin-chatbot的源码,命中的文件数多于卡面点名的五处(除四个plugin-detail站点外,还出现在ConcurrentUpdateDialog.tsx、RecordActivityTimeline.tsx以及若干.test.tsx)。⛔ 本席不据此断言普查漏了 —— 一个文件出现这个名字,可能是调用,不是又一份副本,而卡面数的是定义。⇒ ⭐ 但这正说明:接手时要按「定义」重新取一次总体,而不是照抄这五行。 一次按名字的 grep 会给出一个更大的数,而两者都不是错的 —— 它们量的是不同的东西。(本轮本席在 objectui#7034 上刚被同一类量纲差异绊过一次:
grep -c数行、grep -o | wc -l数出现次数。)边界测试 —— ⛔ 共享即新导出,新导出即 Clause ②
卡面把停在报告的理由写得无可挑剔:
⇒ ⭐ 这是「按构造跨地板」的一个干净例子,不是保守。⛔ 谁接手都不要试图找一个「不算新导出」的写法绕过去 —— 那只会把一个治理问题变成一个隐藏的治理问题。
⭐ 三个必须被回答的问题 —— 卡面已列,本席按「哪个可能否定整件事」重排
toLocaleDateStringvstoLocaleStringvsIntl.DateTimeFormat),统一Math.round与Math.floor会移动 chatbot 收件箱的每一个边界。⇒ 「一个为保留全部五种行为而带足够选项的共享 helper,可能不如重复本身值钱。」 这一条可能直接否定 2 和 3。@object-ui/i18n是唯一的共同依赖 ⇒ 显然的家 ⇒ 也正是 Clause ② 治理的那个新公开导出。Intl.RelativeTimeFormat原生本地化相对短语,会把它覆盖的那些分支的四个 pack 键整个从画面里拿掉。⇒ ⛔ 本席不裁,也不建议排期。⭐ 卡面自己的定位是对的:「立卡是为了让重复可见,并让下一个碰这五处之一的人不必重新推导这张对照表 —— 不是为了提出工作。」
⛔ 本席不认领、不派发、不代裁。
Generated by Claude Code
objectstack-fleet commentedon Sep 24, 2026
Closed
not_planned— UI closure review batch 1 item 6 · maintainer 「同意」 (chat, director seat summon #29,session_01EcrTi7s5oDYPHS4Pi7h31d) 2026-09-24T06:30ZThe card states it is an observation, not a defect, and not a proposal of work: every copy is correct and translated, and the differences are deliberate (a revision list shows the time of day, a comment list does not; the chatbot inbox rounds differently). Under the filing gate an observation is not a card.
Reading on objectui
origin/main62597c5: the definitions offormatTimestampnow number eight (plugin-detailActivityTimeline / ConcurrentUpdateDialog / PointInTimeRestore / RecordActivityTimeline / RecordComments / ThreadedReplies,collaborationCommentThread,app-shellConversationsSidebar). The comparison table in the body remains the record for whoever touches one.If the product ever wants one relative-time phrasing (for example
Intl.RelativeTimeFormatacross the ten packs), that is a copy decision and comes back as a decision card. Reopening is free.Generated by Claude Code