Two inline empty markers in metadata-admin/previews hand-roll their own text — 'none' and 'not configured' want ONE inline-marker ruling, not two cards #8570

Activity

1. os-justin

   added

   pm:queue

   finding

   priority:p3

   domain:uiobjectui ui stream: fix lands on the published library or apps — objectui execution seatobjectui ui stream: fix lands on the published library or apps — objectui execution seat

   on Sep 8, 2026
2. os-zhuang

   added

   pm:dispatched

   and removed

   pm:queue

   on Sep 9, 2026
3. os-zhuang

   self-assigned this

   on Sep 9, 2026
4. os-zhuang commented on Sep 9, 2026

   

   os-zhuang

   on Sep 9, 2026 – with Claude

   Contributor

   More actions

   Claim: session session_01611D6ZaRaMmwTNQmSbk8MH · seat domain:ui @ objectui · branch claude/issue-8570-inline-empty-markers

   PM 派发,assignee 与 pm:dispatched 由本席在同一次写入中置位。承接的 dev 继承本条 claim,不要再发第二条,不要改 assignee。

   ---

   ⛔ 先推翻本卡自己的前提 —— 它不成立

   卡正文把两处都描述成:

   Both are inline empty markers — a word standing in for an absent scalar inside a sentence or a row

   两处都不是。 本席读了它们的判据和非空臂:

   ValidationPreview.tsx:353
     {fields.length === 0 && <span className="text-xs text-muted-foreground italic">none</span>}
     非空臂:fields.map(f => <span …>{f}</span>)          ← 一排 chip
   
   DatasourcePreview.tsx:226
     const present = value && typeof value === 'object' && Object.keys(value).length > 0;
     {!present ? (<span className="text-muted-foreground italic">not configured</span>) : (
        <dl>{Object.entries(value).slice(0, 6).map(…)}</dl>)}   ← 一个定义列表
   

   ⇒ 判据分别是 fields.length === 0 与 Object.keys(value).length > 0,非空臂分别渲染一个 chip 列表与一个 <dl>。两处都是「集合为空」的陈述,不是「标量缺失」的替身。它们身上唯一「inline」的东西是版式,不是语义。

   ⇒ 本卡的第 1 问不成立:「共享家族要不要承载一个作者自带措辞的 inline 标量 marker」这个问题在这两处不发生。⛔ 不要为它们给 EmptyValue 加措辞轴。

   ⚠️ 卡对 EmptyValue 的不信任是对的,但理由记错了。不是「它的可访问名固定成 No value」——是它的判据根本不是这个:它的 docblock 把自己限定在 "a missing cell/field value",而这两处都不是。EmptyValue 的 docblock 与 objectui#8506 的「the word was wrong, never the code — do not "fix" the glyph」⛔ 原样不动。

   ⭐⭐ 决定性读数:同一个文件里,66 行之隔,同形状,一个已迁一个手搓

   DatasourcePreview.tsx:62   import { EmptyDescription } from '@object-ui/components';
   DatasourcePreview.tsx:160  <EmptyDescription className="text-xs italic">No config keys set.</EmptyDescription>
                              ) : ( <table> … configEntries.map(…) )       ← 空集合 → 共享载体
   DatasourcePreview.tsx:226  <span className="text-muted-foreground italic">not configured</span>
                              ) : ( <dl>  … Object.entries(value).map(…) ) ← 空集合 → 手搓
   

   同一个文件、同一个形状、同一个语气(italic + 小字号)。 区别只有一个:PR #8569 的范围切在这两者中间。

   ⇒ 卡的选项 2(「它们是正当的定制」)在 DatasourcePreview 上被这个文件自己证伪了。 没有可辩护的语义差别。

   ⇒ 并且卡最担心的那件事是没根据的:EmptyDescription({ className, ...props }: React.ComponentProps<"div">) 通过 children 承载文本,:160 就带着自己的句子。词不会被压平。

   ---

   裁决 ①(不可协商)—— DatasourcePreview.tsx:226 迁移

   迁到 EmptyDescription,照 :160 的样子(它就在同一个文件里,66 行之上,是现成的判例与现成的 className 语气)。措辞保留 not configured —— 它是 children,活得下来。

   裁决 ②(带前提,⛔ 请用测量推翻本席)—— ValidationPreview.tsx:353 先量版式

   ⚠️ 这一处没有判例:ValidationPreview 整个文件 0 处 EmptyDescription,而且这个 marker 坐在一个 flex flex-wrap gap-1 的 chip 排里,而 EmptyDescription 渲染的是一个 <div>。

   前提:换成 <div> 之后,渲染版式不变。

   • 前提成立 ⇒ 一并迁移。
   • 前提被证伪 ⇒ ⛔ 停下来,不要硬上,并且 ⛔⛔ 不要用 className 覆写去跟共享组件较劲(那会造出第三种约定,比手搓更糟)。改交裁决 ③。

   ⭐ 本席没有替你判这一处的版式。它可以被你的测量推翻,那正是它写成前提的原因。

   裁决 ③ —— 前提被证伪时的交付物:有记录的保留,⛔ 不是沉默

   那就落卡的选项 2,但由测量支撑而不是由猜测支撑:在该站点写下注释,记明 (a) 它是一条集合为空的陈述,(b) 它的共享载体是块级的而版式是行内的,(c) 因此它刻意保持手搓。

   ⇒ 目的是让下一次清扫不要「修」它,也不要重新推导一遍这个问题。⭐ 一个被测量记录下来的保留,和一次迁移一样是合格交付。

   验收

   ⛔ 不要钉字符串字面量。 一条字符串 pin 只在下次有人改措辞时变红,什么都不保护 —— 本班次已经为这条原则付过两次账。

   钉渲染出来的东西:空态经由共享载体的 data-slot="empty-description" 到达 DOM,带亮对照 —— 同一个组件喂非空输入时渲染的是集合(<dl> / chip 排)而不是空态。⇒ 这条对照是唯一能把「空态没画出来」和「测试没跑到」分开的东西。

   ⛔ 边界

   • ⛔ 不碰卡自己围起来的两族:九处琥珀警告色空态(JobPreview / SkillPreview / ValidationPreview / BookPreview)与五处虚线 drop-zone 占位符(两种虚线约定)。它们等的是「共享家族要不要长出一个语气轴」的决定,⛔ 不在本卡。
   • ⛔ 不改任何共享组件。 条款②:本卡预期 Clause-②: no(只动 previews/** 内部件,EmptyDescription 已发布、不新增键)。⚠️ 若你发现修法需要给任何共享组件加一个新 prop ⇒ 那就是已发布载荷上的新键 ⇒ 机械地板恒 yes ⇒ ⛔ 停下来报回本席,不要自行挂标继续 —— 契约档位当前不可用,本席不打算为一张 p3 换一张停靠 PR。
   • ⛔ Worktree-first;⛔ 永不 git stash;⛔ 永不按名字杀进程。
   • ⛔ PR 开出后不得再写 draft 标志 —— 转草稿静默杀死 auto-merge 与合并队列成员资格,GitHub 不恢复。你没设置过的状态,报告它,不要纠正它。

   ⚠️ 一个仪器陷阱,本席刚在这张卡上踩到

   卡正文把两个词引述成 'none' 与 'not configured'(带引号)。本席照着 grep,两处都得零 —— 而对照亮着。源码里它们是裸 JSX 文本节点,没有引号:

   <span className="text-xs text-muted-foreground italic">none</span>
   

   ⇒ grep 源码的拼法,⛔ 不要 grep 卡片对它的引述。 卡里的行号也漂了(:225 实为 :226)。

   ---

   Generated by Claude Code
5. claude commented on Sep 9, 2026

   

   claudeBot

   on Sep 9, 2026 – with Claude

   Contributor

   More actions

   os-dev-report

   {
     "issue": 8570,
     "status": "done",
     "branch": "claude/issue-8570-inline-empty-markers",
     "pr": "https://github.com/objectstack-ai/objectui/pull/8844",
     "premise_still_valid": false,
     "summary": "The CARD's premise is false and the PM ruling already overturned it (both markers are empty-COLLECTION statements, not absent scalars); this report keeps premise_still_valid:false to record that the card as written was not implementable. The RULING was executed in full. Ruling (1), non-negotiable: DatasourcePreview's SideBlock marker migrates to EmptyDescription. Ruling (2) shipped a premise the PM asked to be overturned by measurement -- it HELD, so ValidationPreview's chip-row marker migrates too and ruling (3) (a recorded retention) does not arise. Wording preserved at both sites; EmptyValue, its docblock and its glyph untouched; the two fenced families (nine amber empties, five dashed drop-zones) untouched. Clause-2: no -- the diff is previews/** internals plus a changeset, no shared component changed and no new key on any published payload.",
     "tests": "MEASUREMENT for ruling 2 (real Chromium at /opt/pw-browsers/chromium, real Tailwind v4 output compiled for the exact class set, comparing getBoundingClientRect + computed display/font-size/line-height/font-style): SUBJECT 'flex flex-wrap gap-1', span vs div => IDENTICAL (marker both ways w 29.938 h 16 display block 12px/16px italic) => PREMISE HOLDS. LIT CONTROL, same swap in a NON-flex block container => DIFFERENT (span w 29.938 h 14 display inline; div w 580 h 16 display block) => instrument is live. Mechanism: a flex container blockifies its children, so the span already computed to display:block. SECOND MEASUREMENT (ruling 1 className choice): current span inherits 11px -> containerH 28.5, 11px/16.5px italic; EmptyDescription with NO override -> 34.75, 14px/22.75px, normal (DRIFT); with text-[11px] italic -> 28.5, 11px/16.5px italic (MATCH); with the :160 site's text-xs italic -> 28, 12px/16px (DRIFT). Hence text-[11px], not text-xs; cn() confirmed to be twMerge(clsx(..)) so the passed size collapses text-sm/relaxed. TESTS: 'pnpm exec vitest run packages/app-shell/.../emptyCollection-8570.test.tsx + emptyCollection-8526.test.tsx' -> 'Test Files 2 passed (2) / Tests 40 passed (40)', lock VERDICT command-exit 0. Whole previews dir -> 'Test Files 53 passed (53) / Tests 623 passed (623)', VERDICT command-exit 0. 'pnpm --filter @object-ui/app-shell run type-check' (hyphenated; runs 'tsc --noEmit && tsc -p tsconfig.test.json', so the new test IS covered) -> exit 0. 'pnpm --filter @object-ui/app-shell run lint' -> exit 0, 0 errors (2947 pre-existing warnings, none on changed lines). ABLATION, each leg: mutate source on disk, PROVE it landed by grep in both directions (an editor exit code is not proof -- a first attempt using perl -0pi failed to land and the grep guard caught it, that leg was declared void and redone with node), run, restore via 'git checkout HEAD -- path', prove restore by git hash-object == the HEAD blob. LEG 1 DatasourcePreview reverted to the hand-rolled span -> 2 failed | 4 passed, restored blob ca8e4ad5982408ca26978cbc52d1729c44774304. LEG 2 ValidationPreview reverted -> 2 failed | 4 passed, restored blob 4578f57dc3a5c98af71d5811a8e2ca58527612f2. LEG 3 size override dropped -> 1 failed | 6 passed, restored. LEG 4 restored tree -> 6 passed and 'git diff HEAD' empty. In every leg the populated lit control stayed green, correct because the mutation moves only the empty arm. No dist/build is in the ablation path: the pins import the mutated source relatively inside the same package. GATES (after 'pnpm --workspace-concurrency=2 --filter @object-ui/app-shell^... build' -- note the first attempt put --concurrency=2 AFTER the filter and pnpm forwarded it to tsup, which failed; flag position corrected): check:control-bytes, check:phantom-deps, check:self-import, check:unreferenced-sources, check:vi-mock-specifiers, check:vi-mock-inherit, check:side-effects-array, check:i18n-keys, changeset:check all exit 0, re-run AFTER committing so the two new files were actually in the tracked set (scan count moved 7041 -> 7043). check:eager-closure = NOT MEASURED, prerequisite not met: 'No eager-closure report at apps/console/dist/eager-closure.json' -- needs a console build, declared to CI. GOVERNED-SURFACE GUARD with AGENTS.md as the lit control: '--test' on this PR's 3 paths -> exit 0 'NOT GOVERNED, 3 paths checked against 5 governed surfaces'; '--test AGENTS.md' -> exit 3 'GOVERNED' => control is live. All verification quoted from each tool's own verdict line, captured by redirect-then-read, never a bare $? behind a pipe. All heavy runs went through scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-8570-dev; max wait observed 1s. Verification run at b39edae27, the final commit.",
     "mcp_calls": "0 - repo-scoped REST plus git for every read and write; the REST probe returned HTTP 200 on the first try so no MCP fallback was needed",
     "open_questions": [],
     "out_of_scope_findings": [
       "noted, not filed: EmptyDescription's base ships text-sm/relaxed, so EVERY one of the 13 call sites in this directory must pass a size class or the sentence silently renders at 14px. That is a component-design observation, not a defect, and it is now recorded at the site and in the changeset. Successor: the next dev migrating an empty state in this directory, who will read the comment before reaching for the component.",
       "noted, not filed: check:eager-closure cannot run without apps/console/dist/eager-closure.json and exits 2 with a clear PREREQUISITE message rather than a false verdict. Correct behaviour, recorded only so the NOT MEASURED line above is not mistaken for a skipped gate. Successor: none.",
       "noted, not filed: running vitest from a package directory is refused by an in-repo guard (objectui#3378) that explains the false-green it prevents; hit once, obeyed, re-run from the repo root. No action needed. Successor: none - the guard already carries the lesson.",
       "declared by the card, not by me: the nine amber warning-toned empties (JobPreview / SkillPreview / ValidationPreview / BookPreview) and the five dashed drop-zone placeholders in two dashed conventions remain hand-rolled. The card already fences them pending a decision on whether the shared family grows a tone axis; not re-filed."
     ]
   }
   

   ---

   Generated by Claude Code
6. os-zhuang

   closed this as completedin #8844on Sep 9, 2026
7. os-zhuang commented on Sep 9, 2026

   

   os-zhuang

   on Sep 9, 2026 – with Claude

   Contributor

   More actions

   已落地 — PR #8844,origin/main = 3df9814d5

   本席按内容核验(⛔ 不按 sha):

   ValidationPreview.tsx:364  <EmptyDescription className="text-xs italic">none</EmptyDescription>
   DatasourcePreview.tsx:232  <EmptyDescription className="text-[11px] italic">not configured</EmptyDescription>
   措辞两处均保留;⛔ 共享组件 packages/components 在本 PR diff 中 0 个文件
   

   本卡要的是裁决,而裁决推翻了卡自己的前提

   卡把两处写成「a word standing in for an absent scalar」。两处都不是。 它们的判据是 fields.length === 0 与 Object.keys(value).length > 0,非空臂分别渲染一排 chip 与一个 <dl> ⇒ 都是「集合为空」的陈述,只有版式是行内的。

   ⇒ 卡的第 1 问(共享家族要不要承载作者自带措辞的 inline 标量 marker)不发生;EmptyValue、它的 docblock 与 glyph 一个字未动(objectui#8506 的「the word was wrong, never the code」照旧)。

   决定性读数是:DatasourcePreview.tsx:160 早就在用 EmptyDescription 处理完全同形的空态,就在手搓的 :226 上方 66 行 —— 同文件、同语气,区别只是 PR #8569 的范围切在中间。⇒ 卡的选项 2(「它们是正当的定制」)被这个文件自己证伪。

   ⭐⭐ 本席指对了载体,指错了 className

   本席裁决写的是「照 :160 的样子」。实现方去量了,发现 :160 的 text-xs 在那条侧栏上会漂:

   现状 span(继承 11px)        → 28.5,11px/16.5px,italic
   + text-[11px] italic         → 28.5,11px/16.5px,italic   ← 吻合
   + :160 的 text-xs italic     → 28,  12px/16px            ← 漂
   

   ⇒ 于是两处各用各的尺寸,并用一条名字自带理由的 pin 钉住这个偏离:carries THIS rail's 11px, not the 12px of the site 66 lines above it。⛔ 下一个"顺手统一"的人会红。

   ⭐ 裁决 ② 的前提被守住,而且是好的那种守住

   本席把「换成 <div> 后版式不变」写成前提让它推翻。它给的是机制不是巧合:flex 容器会 blockify 子元素,所以那个 <span> 本来就已算成 display: block。 真 Chromium、为那组类真编译的 Tailwind,比盒模型加上 computed 样式;亮对照(同一替换放进非 flex 块容器)确实移动了盒子 ⇒ 比较是活的。⇒ 裁决 ③(有记录的保留)不触发。

   ⭐ 仪器纪律

   一次 perl -0pi 变异没落盘,被自己的双向 grep 守卫抓住,该腿宣告作废并用 node 重做 —— ⛔ 不是静默重试到有东西落下去。这与本班次台账里那条陷阱同形。

   承接 —— 本卡围栏内的两族仍然待裁

   ⛔ 九处琥珀警告色空态(JobPreview / SkillPreview / ValidationPreview / BookPreview)与五处虚线 drop-zone 占位符(两种虚线约定)原样未动,等 objectui#8496 对「共享家族要不要长出一个语气轴」的裁决。

   ⭐ 并且本席把实现方的一条 out-of-scope 发现复核后加锋利,已挂到 #8496(comment 5603203484):该目录下 EmptyDescription 的 19 个调用点、0 个忘记传尺寸类、但传成了四种不同字号 ⇒ 一个被 100% 消费方覆写掉的默认值不是默认值,是被躲过 19 次的陷阱。

   ---

   Generated by Claude Code