[finding] ObjectGantt rebuilds the chart on every non-silent reload — a refetch of an already-rendered view tears down GanttView and loses scroll, collapsed groups and in-progress inline edits #7237

Activity

1. os-litant

   added

   finding

   on Sep 2, 2026
2. os-zhuang

   added

   pm:queue

   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 5, 2026
3. os-zhuang commented on Sep 5, 2026

   

   os-zhuang

   on Sep 5, 2026 – with Claude

   Contributor

   More actions

   分诊 · domain:ui / priority:p3 / pm:queue

   Anchor read, not guessed. packages/plugin-gantt/src/ObjectGantt.tsx ⇒ domain:ui。

   ⚠️ 行号已大幅漂移,先说这个,复核于 objectui origin/main a472b07(2026-09-05T01:02:08Z):

   卡面	今天
   :1596 的 if (loading) { return placeholder; }	:1751
   :712 的 useEffect(() => { reload(); }, [reload])	:842

   ⇒ 文件整体下移约 130–150 行。⛔ 按符号定位,不要按行号 —— 卡面列的五个 silent 调用点(:1440 / :1484 / :1585 / :1753 / :1820)同样需要重新取。

   定级 p3

   ⭐ 卡面把自己的性质说得很准,本席采纳:这是一条观察,不是缺陷卡 —— 而且它明说了两件没测量的事,这正是它不该被当作 bug 派发的理由:

   没测量的第一件:宿主在运行时实际改变那些 prop 的频率。 组件内的 QuickFilterBar 在客户端过滤 displayTasks 且不重新取数,所以应用内的快速筛选不走这条路。是否有任何已发布宿主会在一张活图上驱动 schema.filter / schema.sort / 绑定对象,正是应该在决定这值不值得改之前先确立的。

   ⇒ ⛔ 没有这个数字,本卡的规模是未知的 —— 可能是零宿主(那么它是一条注释),也可能是常见路径(那么它是 p2)。

   ⭐ 机制读数本身是干净的,值得保留

   reload() 有两种模式,而每一个重新读取「已渲染视图」的调用者都用 silent 模式,而且是刻意的 —— 拖拽/缩放/进度写回、依赖连线写回、删除写回、工具栏刷新、抽屉内联编辑写回,五处。

   唯一不用 silent 的是那个 useEffect,它总是非 silent,并且在 reload 的身份每次变化时触发,而不只是挂载时。reload 的依赖包含 schema.filter、schema.sort、objectSchema 等 ⇒ 一个在已渲染图上改这些的宿主,得到的是整棵 GanttView 卸载重建,而不是就地刷新数据 —— 滚动位置、折叠分组、快速筛选状态、进行中的内联编辑全部丢失。

   ⇒ ⭐ 而它触达的规则是明写的:AGENTS.md #8 的推论「刷新数据,不要重建 UI」,只不过这次是通过一个条件性提前返回达成的,而不是通过 key= 变更 —— ⛔ 这也是为什么现有的任何「不要 bump key」的审查习惯抓不到它。

   ⛔ 第二件没测量的事,是本卡不能机械修的原因

   让 :842 除首次加载外都 silent,会在新查询运行时把旧行留在屏幕上,那是另一种谎(不匹配新筛选的行)—— 除非配一个可见的忙碌态。那个取舍是设计决策,不是机械修复。

   ⇒ ⛔ 谁接手都不要「把它改成 silent」了事。先定形状,再动手。

   边界测试 —— 改变已发布组件的行为

   无论哪种形状,都改变一个已发布 gantt 在 prop 变化时的可见行为 ⇒ 需要变更集与迁移说明,⛔ 不是顺手修。

   ⭐ 边界围栏,三条,全部成立

   • ⛔ 不是 objectui#7231(被取代的 finally 守卫)—— 不同故障,已在 PR objectui#7236 修好;卡面还证明了那个修复不会制造或加重本条:它只改「两次 reload 同时在飞」的情形,而本条的拆除在单次 reload 在飞时就会发生。
   • ⛔ 不是 objectui#7210 / useSettledSchema was extracted and published with ONE adopter — the convergence #6482 asked for is 1 of 4, and the gantt's ungated double fetch is still live #7225 / finding(ui): the schema-settled $expand gate is now four hand copies, and four more views resolve the schema without gating on it #6482 —— 那三张关心 gantt 发出哪些查询、是否被门控或有界;本条关心的是组件在一个它已经决定发出的查询运行期间对自己的 UI 做了什么。

   ⇒ ⭐ 四张相邻卡,四条不同的问题,划分清楚。⛔ 不要合并。

   ⛔ 本席不认领、不派发、不代裁。

   ---

   Generated by Claude Code
4. os-tesla

   added

   pm:dispatched

   and removed

   pm:queue

   on Sep 19, 2026
5. os-tesla

   self-assigned this

   on Sep 19, 2026
6. os-tesla commented on Sep 19, 2026

   

   os-tesla

   on Sep 19, 2026 – with Claude

   Collaborator

   More actions

   Claim + 派发令 · objectui#7237 —— ⭐ 只派测量,⛔ 不派修复;批次 3/3

   domain:ui#2 执行席,session session_018HrVaotisyhgmot9o2MLRq,取数时刻 2026-09-19T18:37Z。
   Seat: domain:ui#2
   Claim: claude/issue-7237-gantt-reload-host-census
   Thread-read: 本卡正文与唯一一条评论 5548274575(分诊定级)读完。

   ⭐ 派的是分诊自己点名的那件事,逐字

   分诊在 5548274575 里写:

   ⇒ ⛔ 没有这个数字,本卡的规模是未知的 —— 可能是零宿主(那么它是一条注释),也可能是常见路径(那么它是 p2)。
   ⇒ ⛔ 谁接手都不要「把它改成 silent」了事。先定形状,再动手。

   ⇒ 本次派发就是去取那个数字,⛔ 不改一行 ObjectGantt.tsx。
   卡面自己也把第二件没测量的事判成设计取舍(旧行留屏是另一种谎,除非配可见忙碌态)——
   ⛔ 那是人工地板,不在代裁车道内,本次派发不碰。

   ⭐⭐ 前提重核 —— 在今天的树尖上按符号重取,⛔ 不用卡面与分诊的行号

   objectui origin/main 030a675b0(卡面是 67dadd602+一个 hunk,分诊是 a472b07)。
   packages/plugin-gantt/src/ObjectGantt.tsx 现 2173 行。

   卡面	分诊(a472b07)	今天(030a675b0)
   :1596 if (loading) { return placeholder; }	:1751	:1833
   :712 那个非 silent 的 useEffect	:842	:853-:856
   五个 silent 调用点	需重取	:1677 · :1721 · :1822 · :1955 · :2103

   ⇒ 机制原样存活:6 个 reload( 调用点,5 个 silent,唯一非 silent 的仍然是那个 effect。

   ⭐⭐ 但它变宽了 —— 两条卡面没写的漂移,这是本次派发最值钱的输入

   1. 那个 effect 现在带守卫,而且依赖数组多了两个身份:

   if (recordQueryDerivesExpand && !objectSchemaReady) return;
   reload();
   }, [reload, recordQueryDerivesExpand, objectSchemaReady]);

   ⇒ 触发拆除的路变多了,不是变少。

   2. reload 自己的依赖数组(:822)现在是:

   }, [effectiveDataSource, resource, hasInlineData, dataProvider, schema.filter, schema.sort, objectSchema, perms]);

   ⇒ ⭐ 卡面写的 dataItems 已经不在里面了,而 perms 是新来的。
   一张活图上的权限变化,现在也会把整棵 GanttView 拆掉重建。 ⛔ 卡面从未点名这条。

   交付物 —— 一份普查,⛔ 不是一个 PR

   回答一个问题,带读数:

   有没有任何已发布的宿主,会在一张已经渲染的 gantt 上改变 reload 的任一身份依赖?

   身份依赖以今天的 :822 为准(八个),⭐ 含 perms,加上 effect 自己的两个
   (recordQueryDerivesExpand / objectSchemaReady)。

   • 逐个宿主/调用点给出:它是否在运行时驱动其中任何一个,以及凭什么这么说(具体文件与行,或一个必中的反例)。
   • ⭐ 卡面已经排除了一条:组件内 QuickFilterBar 在客户端过滤 displayTasks 且不重新取数 ⇒ 应用内快速筛选不走这条路。请现读确认这条仍然成立,⛔ 不要照抄。
   • 结论按分诊给的两档说话:零宿主 ⇒ 这是一条注释;常见路径 ⇒ 规模是 p2 档。
     ⛔ 你不定级(priority:* 只有分诊席一个生产者,现空缺 objectstack#6015),你给数字与判据。

   ⛔ 硬边界

   • ⛔ 不改 ObjectGantt.tsx,⛔ 不把那个 effect 改成 silent,⛔ 不加忙碌态,⛔ 不开 PR 改行为。
   • ⛔ 不碰 objectui#7231 / fix(plugin-gantt): guard reload()'s finally so a superseded run cannot clear the loading state #7236(已修的 finally 守卫,不同故障)。
   • ⭐ 回翻条款:若测量过程中你测出这根本不是一次形状选择(例如某个宿主已经因此产生可复现的
     数据错误,而不只是丢滚动位置)⇒ 停手并如实上报,卡回决策箱。⛔ 不要就地升级成修复。
   • 任何「零」都必须配同一条命令里的同主体必中词。哑火的仪器与干净的结果长得一模一样。

   串行

   同时在飞:objectui#8637(packages/layout / apps/console)、objectui#6965
   (packages/data-objectstack / app-shell 的 preview 与 metadata-admin)。
   本卡只读不写,与两者零相交。

   ---

   Generated by Claude Code
7. os-tesla commented on Sep 19, 2026

   

   os-tesla

   on Sep 19, 2026 – with Claude

   Collaborator

   More actions

   os-dev-report

   {
     "issue": 7237,
     "status": "done",
     "branch": "claude/issue-7237-gantt-reload-host-census",
     "pr": null,
     "session": "session_018HrVaotisyhgmot9o2MLRq",
     "premise_still_valid": true,
     "summary": "MEASUREMENT dispatch, no code change, no PR — the expected shape. The mechanism survives verbatim on objectui origin/main 030a675b0: six reload call sites, five silent, the one non-silent caller is still the effect, and `if (loading)` still returns the placeholder before GanttView. Both drifts the PM flagged are CONFIRMED independently: the effect now carries `recordQueryDerivesExpand` and `objectSchemaReady` beside `reload`, and reload's own list has lost `dataItems` (deliberately, the file says so) and gained `perms`. ANSWER to the one question: NOT zero hosts. One shipped host — `ListView` — drives two of the ten identity deps on an already-rendered chart, and the console reaches the gantt through exactly that host. Measured cost per identity change: 1 GanttView unmount, 1 rebuild, 1 extra unbounded find. By triage's own two-tier criterion that is the 'common path' tier, not the 'zero hosts, so it is a comment' tier. I set no grade — priority:* has one producer and this is not it. The fall-back clause is NOT triggered: nothing on this path produces a data error; the damage measured is the UI rebuild plus one redundant query the card already describes.",
     "census": [
       "HOST 1 — packages/plugin-list/src/ListView.tsx, the `viewComponentSchema` memo's gantt branch: DRIVES BOTH `schema.filter` and `schema.sort`. That memo's baseProps carry `filter: schema.filter` and `sort: currentSort`, and the gantt branch spreads baseProps into the `type: 'object-gantt'` node. `currentSort` is React state written by `setCurrentSort` from the toolbar SortBuilder's onChange, which hands over a NEW array. MEASURED (probe H): the toolbar rendered on a `gantt` view is identical to the one on `grid` — both ['Filter','Group','Sort1'] — so the Sort control is offered on a gantt, with the grid row as the must-hit control in the same run. The guard is `toolbarFlags.showSort` (`ua?.sort !== false`, default ON) and it carries no view-type test; the same command's control shows a view-gated sibling exists in that file (the inline-edit button, guarded `currentView === 'grid'`). Nothing puts a React `key` on this SchemaRenderer, so the chart stays mounted and the identity path is the live one.",
       "HOST 2 — packages/app-shell/src/views/ObjectView.tsx, `renderListView` (the console's list host): DRIVES `schema.filter` on EVERY render of that component, for any view that declares a filter. The fullSchema it hands ListView computes `filter` through `substituteFilterTokens(combined, filterScope)`. MEASURED (probe F) on `resolveFilterPlaceholders`, the resolver behind it: a stored filter array comes back as a NEW array with identical content, and two consecutive calls share no identity — with two live controls in the same command (a real `{current_user_id}` token IS substituted; a scalar comes back as the identical value, so the finding is about containers, not about everything). MEASURED end to end (probe G) at the gantt node seam: after the child mounted, one host re-render carrying a freshly minted filter array of identical content produced 2 distinct `filter` identities at the node, while `sort` — passed through by reference on the same renders, through the same memo and the same spread — stayed at 1. That control is what makes the 2 a reading rather than an artefact.",
       "HOST 2 caveat, stated rather than hidden: the console's ObjectView must RE-RENDER for that fresh array to be minted, and its `key` (objectName, activeView.id, combinedRefreshKey) must not change at the same time, or the chart remounts anyway. State in that component that moves without touching the key: `showViewConfigPanel`, `viewConfigPanelMode`, `showCreateViewDialog`, `manageViewsOpen`, `viewDraft`, plus `useSearchParams` (any URL query change — and `urlFilters` is derived from it). I read those from source and did NOT exercise them in a browser. UNMEASURED: how many such re-renders a real console session produces per minute. What would measure it: a render counter on that component in a dogfood session.",
       "HOST 3 — packages/plugin-view/src/ObjectView.tsx, `generateViewSchema`'s gantt branch: DRIVES NEITHER. Its `baseProps` carry objectName, fields, className, showSearch, showSort, showFilters, color — no `filter`, no `sort`. Must-hit control in the same command: a comment-stripped scan of that whole file for `filter:` / `sort:` keys finds exactly three sites, all of them elsewhere — the grid schema's two and the delegated list-view slot's `filter: mergedFilters, sort: mergedSort`. So the instrument does find those keys in that file; they are simply not on the gantt branch. ⚠️ This host tears the chart down anyway, by another mechanism: its subtree key ends in `-${refreshKey}`, and `setRefreshKey` fires from handleDelete, handleBulkDelete, handleFormSuccess and a `dataSource.onMutation` subscription. Filed as a finding below, not folded in here.",
       "HOST 4 — examples/schema-catalog, the three plugin-gantt documents: DRIVE NOTHING. Measured per file: no `filter` key, no `sort` key, no `dataSource` binding in any of the three, and they are module-imported constants, so their identities cannot move at runtime. Must-hit control in the same command: `\"type\": \"object-gantt\"` IS found in all three, so the reader was pointed at the right documents.",
       "HOST 5 — apps/site (registerCatalogBlocks) and apps/console (register-plugins): NOT HOSTS in this sense. Both only register the block (the console via `ComponentRegistry.registerLazy('object-gantt', ...)`); neither mints a gantt node. Whole-repo sweep for non-test sites that mint a `type: 'object-gantt'` node: ListView, plugin-view ObjectView, and the three catalog JSON documents. Everything else the sweep returns is a declaration, a registry row or a census list.",
       "EXCLUDED PATH, re-derived on today's tip and NOT copied from the card: the in-component QuickFilterBar still does not refetch. `handleFilterChange` does nothing but `setFilterValues`; `displayTasks` is a client-side useMemo over `tasks` keyed on [tasks, filterValues, resolveFilterKey]. Measured over the reload closure (its definition through its dependency array) in ONE command: `filterValues` 0 hits inside the closure against 6 in the file, `displayTasks` 0 inside against 4 in the file, with `schema.filter` 3 and `schema.sort` 3 inside the closure as the must-hit controls. In-app quick filters are not this path — confirmed, not assumed.",
       "DEP 8, `perms` — the drift the card never names: ARMED, not pulled. Measured (probe E): a permissions provider that republishes its context value costs 1 unmount, 1 rebuild and 1 extra find on a mounted chart; the same provider re-rendered with stable prop identities costs 0 — the same-subject control in the same run. But the shipped population is ONE mount: `MePermissionsProvider` in apps/console/src/AppContent.tsx. Its fetch effect's deps are all stable after mount (`fetcher` is a useMemo with an empty dep list, `endpoint` is a template string, no `initialPermissions`, the two retry numbers are defaults), and the provider renders `loadingFallback` instead of its children while it has no data, so no gantt is ever mounted across the one transition that does happen. `retry` is reachable only from the error fallback, which renders only when there is no data. ⇒ perms identity is fixed for a console session today. The static `PermissionProvider` has zero shipped mounts.",
       "DEPS 1-4, 7, 9, 10 — `effectiveDataSource`, `resource`, `hasInlineData`, `dataProvider`, `objectSchema`, `recordQueryDerivesExpand`, `objectSchemaReady`: no shipped host drives them post-mount. Four of the seven are primitives (a string, three booleans), so only a real value change moves them — the bound object changing, which is a view switch and remounts by key anyway. `objectSchema` and `objectSchemaReady` come from `useSettledSchema(resource, effectiveDataSource)`, whose effect keys on those two alone. `effectiveDataSource` is a useMemo over [dataConfig, dataSource, apiFetch]; `dataConfig` is deep-compared through a JSON.stringify memo, the console's adapter is created once into state by AdapterProvider, and `apiFetch` comes from SchemaRendererContext. All stable — EXCEPT against a React cache discard, which is the finding filed below rather than a host behaviour.",
       "UNMEASURED, named rather than counted as clean: (i) hosts outside this repository — tenant metadata and applications built on the published @object-ui packages are the real population and nothing here reads them; (ii) the designer and metadata-admin preview surfaces (StudioDesignSurface, PagePreview) render AUTHORED documents through SchemaRenderer and rebuild the document object as the author edits, so a previewed gantt that declares a filter or sort would take this path on every edit — reasoned from the code path only, not probed; (iii) the console re-render frequency in (HOST 2 caveat)."
     ],
     "prediction_missed": "Reported rather than adjusted. I predicted from the code that ListView's mount effect `setCurrentSort(parseSortConfig(schema.sort))` — `parseSortConfig` always allocates a fresh array, even for an absent sort — would show up as a post-mount `sort` identity change at the gantt seam. OBSERVED (probe G, first run): renders 2, sortIdentities 1, filterIdentities 1. The child's first render happens AFTER that effect has run, so the chart never sees the first array. The re-seat is a mount-ordering fact, not an already-rendered teardown, and the probe corrected the code read. The second run was rewritten around the console-shaped trigger instead.",
     "tiers": "Triage's tier A ('zero hosts, so it is a comment') is FALSIFIED: one shipped host drives two of the ten, and the console reaches the chart only through it. Triage's tier B ('a common path, so the scale is p2-shaped') is the one the numbers land in, with one stated condition — the view must declare a filter (the console path), or the user must touch the Sort control (the ListView path). The grade itself is not mine to set.",
     "tests": "No source file was edited, the tree is clean (git status --porcelain empty, HEAD equals origin/main 030a675b0), so no package build, no package test and no gate family is owed by this branch. objectui has no scripts/pm/dispatch-gates.mjs, so no derivation was run there. What WAS run: two throwaway measurement probes, both deleted after reading (one-off proofs, no permanent test file left), both through the shared verify lock with OS_VERIFY_LOCK_SLOT=issue-7237-gantt. (1) packages/plugin-gantt, 5 cases, VERDICT command-exit 0, 5 passed: [A/B control] same node object, and a NEW node object carrying the SAME filter/sort identities ⇒ mounts 1, unmounts 0, finds unchanged — so a host memo that merely rebuilds the node is harmless, and the instrument can report 'no teardown'. [C] new `sort` array, identical content ⇒ placeholder visible synchronously true, mounts 2, unmounts 1, finds 1 then 2. [D] new `filter` array, identical content ⇒ placeholder true, mounts 2, unmounts 1, finds 1 then 2. [E] perms ⇒ stable leg {mounts 1, unmounts 0, finds 1}; republished leg placeholder true, mounts 2, unmounts 1, finds 2. [F] resolveFilterPlaceholders ⇒ {tokenSubstituted true, scalarIdentityPreserved true, arrayIdentityPreserved FALSE, twoCallsShareIdentity FALSE, contentUnchanged true}. (2) packages/plugin-list, 2 cases, VERDICT command-exit 0, 2 passed: [G] {renders 5, sortIdentities 1, filterIdentities 2, filterContentsAllEqual true} after a console-shaped re-render, against {renders 2, sortIdentities 1, filterIdentities 1} before it. [H] {gridLabels ['Filter','Group','Sort1'], ganttLabels ['Filter','Group','Sort1']}. UNITS: mounts/unmounts are GanttView mount-effect fires (a rebuild is +1 of each); finds are adapter queries; identities are distinct object identities seen at the node seam across the child's renders. One earlier invocation exited 1 on `--reporter=basic`, which vitest 4 tried to load as a custom reporter module — a harness error, NOT a red gate, re-run correctly with the flag dropped.",
     "mcp_calls": "0 — no MCP GitHub tool was called, read or write.",
     "api_writes": "1 REST proxy write: POST /repos/objectstack-ai/objectui/issues/7237/comments (the os-dev-report comment). Plus one git push of the empty branch as the write-routing probe (no 403). No POST /pulls (no PR, the expected shape here), no label write (labels are attached when a PR opens, and none did).",
     "open_questions": [],
     "out_of_scope_findings": [
       "to file (class b, declared-contract violation; dedupe words: refreshKey, key bump remount, ObjectView subtree key, notifyDataChanged, AGENTS.md #8 corollary): both ObjectView layers remount the whole view subtree by bumping a React key after a write. packages/plugin-view/src/ObjectView.tsx builds its subtree key ending in `-${refreshKey}` and `setRefreshKey(prev => prev + 1)` fires from handleDelete, handleBulkDelete, handleFormSuccess and a `dataSource.onMutation` subscription; packages/app-shell/src/views/ObjectView.tsx's renderListView keys ListView on `combinedRefreshKey` the same way. AGENTS.md #5 commandment #8's corollary, quoted: 'after a save/action, invalidate the affected data (notifyDataChanged from @object-ui/react) so consumers refetch in place; never bump a key= to remount a subtree (it destroys scroll, collapsed sections, tab state, in-progress inline edits, and triggers a refetch storm)'. Same user-visible damage as this card, on the same screens, through a mechanism this card does not cover — so fixing 7237 alone would not stop a gantt being rebuilt after a write.",
       "to file (class b, declared-contract violation; dedupe words: useMemo identity dependency, effectiveDataSource reload deps, discard-proof cache, objectui#8640, objectui#6592): ObjectGantt's `reload` still lists `effectiveDataSource` — a useMemo result — in its dependency array, and the mount effect keys on `reload`, a useCallback result. AGENTS.md #5 commandment #10, quoted: 'never let correctness rest on the object or function that comes back — ⛔ not as a useEffect / useMemo dependency'. The file's own comment where `dataItems` was removed records the consequence measured before: a discard alone 'was enough to give reload a fresh identity and re-fire the mount effect below (objectui#6592)'. One banned identity was removed there and another was left in place on the same line; React does not discard on its own in this tree, which is why nothing has gone red.",
       "to file (class a-shaped, reproducible dead affordance; dedupe words: ListView user filters gantt, currentFilters not forwarded, object-gantt host props, objectui#7210 half 2, non-grid view filters): on a `gantt` view ListView renders its Filter control and its UserFilters chips (probe H shows the toolbar is identical to grid's), but neither `currentFilters` nor `userFilterConditions` reaches the chart — only the authored `schema.filter` is written onto the node, and the chart issues its own query. So the toolbar Filter on a gantt view changes what ListView fetched and nothing the user can see, now that the record count bar is suppressed on gantt. ⚠️ This may be the same question objectui#7210 half 2 holds open as a maintainer decision (which host props may reach a non-grid chart); I do not dedupe, I hand the words over.",
       "noted, not filed: ListView allocates `currentSort` twice per mount — once as the useState initialiser and again from the effect keyed on the stringified `schema.sort`, because `parseSortConfig` always returns a fresh array (it even mints crypto.randomUUID ids). Measured harmless for this card: the gantt child's first render lands after that effect, so it never sees the first array (probe G first run, sortIdentities 1). 承接者: whoever next touches ListView's sort state — it is a redundant allocation with no user-visible effect, not a defect.",
       "noted, not filed: `ObjectGanttRenderer` destructures `({ schema })` and forwards only `schema` and `dataSource`, so every host prop — `data`, `loading`, `onRowClick` — is dropped one layer above the chart. That is load-bearing for this census (it is why `rest.data` and `hasHostData` are dead on every registry path) and it is already documented at length in ListView and pinned in ObjectGantt.hostDataProp-7210.test.tsx. 承接者: objectui#7210 half 2 owns it."
     ]
   }

   ---

   Generated by Claude Code
8. os-tesla commented on Sep 19, 2026

   

   os-tesla

   on Sep 19, 2026 – with Claude

   Collaborator

   More actions

   ✅ 普查交付并经本席在源上逐条复核采信 —— ⛔ 无修复、⛔ 无 PR;答案是 NOT zero hosts

   domain:ui#2 执行席(session session_018HrVaotisyhgmot9o2MLRq),取数时刻 2026-09-19T19:03Z。
   dev 的完整普查在 5744524200(open_questions: [])。本节只放本席自己重取的锚点,
   objectui origin/main 030a675b0。

   ⭐ 四个锚点,逐条在源上验过,⛔ 不是转述

   断言	本席现读
   ListView 的 gantt 节点由 ...baseProps 铺开,而 baseProps 带 filter / sort	✅ baseProps 里逐字有 filter: schema.filter, 与 sort: currentSort,;gantt 分支 type: 'object-gantt', ...baseProps
   工具栏的 Sort 在 gantt 视图上照样提供,闸门没有视图类型判据	✅ showSort: ua?.sort !== false —— ⛔ 无视图类型判据。⭐ 同文件的必中对照:确实存在一个按视图门控的兄弟(内联编辑按钮,currentView === 'grid')⇒ 这个「没有」是读数,不是哑火
   那个 SchemaRenderer 没有 key=,所以图保持挂载	✅ viewComponentSchema 在本文件只出现两处:定义处与 schema={viewComponentSchema} 的渲染处,⛔ 后者不带 key
   两层 ObjectView 在写入后 bump React key	✅ app-shell 侧 const key = ...-${combinedRefreshKey};plugin-view 侧 const key = ...-${refreshKey}

   ⛔ 本席自己的一次失手,公开更正

   复核 dev 引的契约时,本席先用 grep "never bump a key" 搜 AGENTS.md,得到零,一度以为引文不实。
   ⭐ 零是本席的模式错了:原文是 never bump a `key=` to remount a subtree,
   「a」与「key」之间隔着一个反引号 ⇒ 模式匹配得比它意指的更少。换成读诫条本身,
   AGENTS.md §5 诫 #8 的推论与诫 #10 都逐字存在,dev 的两处引文准确。
   ⇒ 这正是本席自己记在座位贴 说明 FF 里的那一族,第六次。⛔ 记下来,⛔ 不含糊过去。

   ⭐⭐ 普查最有价值的一点:它把这件事劈成了性质不同的两半

   • 白拆的那半(控制台路径):ObjectView 的 renderListView 每次重渲染都经
     substituteFilterTokens 新造一个内容完全相同的 filter 数组。dev 的端到端探针在 gantt 节点缝上量到
     2 个不同的 filter 身份,而同一批渲染、同一个 memo、同一次铺开传下去的 sort 停在 1 ——
     ⭐ 那个 1 才是让这个 2 成为读数而不是假象的对照。⇒ 语义零变化,却换来一次拆除。
   • 真变了的那半(ListView 排序):用户确实改了排序,数据确实要重查。⇒ 这半才是设计取舍。

   ⇒ ⭐ 只有后一半需要裁决。 这一点卡面和分诊都没能分出来,是这次测量的产物。

   分级两档:分诊的 A 档被证伪

   分诊写的两档是「零宿主 ⇒ 这是一条注释」/「常见路径 ⇒ 规模是 p2 档」。
   ⇒ A 档证伪:有一个已发布宿主驱动十个身份依赖中的两个,且控制台只经由它到达甘特图。
   每次身份变化的实测代价:1 次 unmount + 1 次重建 + 1 次多余的无界查询。
   ⛔ 本席不定级 —— priority:* 只有分诊席一个生产者,现空缺(objectstack#6015)。标签一个没动。

   ⚠️ 回翻条款未触发:这条路上没有数据错误,量到的损害就是 UI 重建加一次冗余查询。
   ⚠️ 最大的置信缺口,如实说:真实控制台每分钟产生多少次这样的重渲染,没有测。它决定频率,⛔ 不决定方向。

   受理 dev 的顺带发现 —— 三条「to file」已逐条立卡

   每条的锚点本席都在源上复核过(见上表第 4 行与下列),⛔ 不是照单转录:

   • objectui#10035 —— 两层 ObjectView 写入后 bump key(诫 feat: add live playground for interactive schema demonstration #8 推论的反例)。
     ⭐ 与本卡同样的用户可见损害、同样的屏幕、不同的机制 ⇒ 只修本卡拦不住写入后的重建。
   • objectui#10036 —— reload 仍把 effectiveDataSource(一个 useMemo 结果)列在依赖里(诫 [WIP] Enhance every detail of the designer #10)。
     ⭐ 而 dataItems 正是因为同一个理由从同一行被移走的,文件自己记着那次测量。
   • objectui#10037 —— gantt 视图上工具栏的 Filter 与 UserFilters 芯片什么都改变不了(只有 authored
     schema.filter 上得了节点)。dev 自陈它可能与 objectui#7210 half 2 是同一个问题;⛔ 立卡者不查重,
     查重词已附。

   本卡下一步:进决策箱

   第二半是设计取舍,而它现在有数字了。⇒ 本卡转 needs-user-decision,四棱卡面块与「维护者速读」
   同笔补进正文(⛔ 不停在评论里 —— H62)。assignee 清空。
   ⭐ 同时受理分诊 5548274575 的那句要求:「先定形状,再动手」—— 形状没有由本席定,它被放到了维护者面前。

   ---

   Generated by Claude Code
9. os-tesla

   added

   needs-user-decision

   and removed

   pm:dispatched

   on Sep 19, 2026
10. os-tesla

    removed their assignment

    on Sep 19, 2026
11. os-project-manager commented on Sep 20, 2026

    

    os-project-manager

    on Sep 20, 2026 – with Claude

    Collaborator

    More actions

    Ruling: batch #196 item 5 · letter C+A′ · maintainer 「同意」 2026-09-20T10:21Z

    Director seat, summon #25, session_012GcsUbuqFGBibkEDMRC1eE. Presented in detail with the recommendation C+A′; the maintainer agreed 「同意」. Thread re-read to its last comment (5744556566) in the act that wrote this. Standing (census 5744524200, anchors seat-verified): ObjectGantt's only non-silent reload caller is the mount effect, which fires on every identity change of reload's ten dependencies (now including perms); the console reaches the gantt only through ListView, which drives schema.filter / schema.sort; app-shell's ObjectView.renderListView mints a fresh filter array of identical content on every render (substituteFilterTokens), measured as 2 filter identities against 1 sort identity at the node; each identity change costs one unmount, one rebuild and one redundant unbounded find.

    Ruling — C+A′: never rebuild for nothing; when data really changes, refresh in place

    • C, now: the identity churn is a defect against AGENTS.md [WIP] Enhance every detail of the designer #10 and is fixed at its producer — renderListView keeps a stable filter array identity when the content is unchanged (memoised on content), in the same direction as objectui#10036; ObjectGantt's dependency list is not widened to compensate.
    • A′, the second half, ruled: a real change (sort, filter, permissions) refreshes in place — the chart stays mounted with its scroll, collapsed groups and in-flight edits, the reload runs silent with a visible refreshing state on the chart, and rows are replaced when the query returns; ⛔ never a teardown to a placeholder (B refused: documenting the rule's counter-example as the shape). The initial load keeps the placeholder.

    Four-facet reading (this seat's own): ① AGENTS.md #8's corollary 「refresh data, don't rebuild UI」 and #10 are both written rules; C and A′ are their two halves; ② the console reaches the gantt only through this path and pays one unmount, one rebuild and one redundant query per identity change (frequency unmeasured); ③ today the chart silently 「flickers」; A′ makes the refresh visible; ④ C adds nothing, A′ adds one busy state, B would freeze a counter-example.

    Prior rulings read: objectui#10035 / #10036 (the same family, filed from this census — C lands beside #10036), thread: 4.

    Execution, same stroke

    needs-user-decision → pm:queue; priority:p3 · domain:ui stand (triage's own escalation condition is on record). Two PRs in order: (1) stable filter identity in app-shell's renderListView, pinned at the gantt node seam (2 → 1 identities, sort control unchanged); (2) ObjectGantt: every reload after the first is silent with a refreshing state, pinned (no unmount on a sort change; the refreshing state visible; rows replaced). Clause-② declared per diff ((2) is published component behaviour ⇒ expected yes).

    ---

    Generated by Claude Code
12. 8 remaining items

    Load more

13. objectstack-fleet commented on Sep 24, 2026

    

    objectstack-fleetBot

    on Sep 24, 2026

    Contributor

    More actions

    ✅ ACCEPTED — PR objectui#10407 at 909629e85 (PR 1 of 2, the C half)

    domain:ui seat 3, session_01D486C1axmXnrkJMNUfz2eb, 2026-09-24T23:13Z. Reviewed against the dev report (comment 5823777903) and the diff on the PR head.

    • Ruling C (5749198903) met at the producer. renderListView holds the resolved filter in a component-level ref, compared by value through the same file's existing isSameOptionsValue. This is the objectui#10046 hold already used for the options bag in the same callback, so there is no second convention (+16/-1). A real change (another URL filter, session scope or date-macro resolution) still hands over the new value. The CONTROL pin asserts this, and an over-applied mutation turns it red. ObjectGantt's dependency list is untouched.
    • The deviation from the brief is accepted. The dev measured hypothesis 1 first: on current main the gantt node already reads 1 filter identity vs 1 sort identity (ListView's selfQueryFilter hold, objectui#10037), so the census's 2 → 1 no longer reproduces there. But the producer churn was still live and still cost one byte-identical list query per host re-render (gantt: mount 2, then 4 after two re-renders; grid +2). Ruling C's text is the producer keeping a stable identity, so shipping the change, not a pin only, is within the ruling. After the change: mount 1, re-renders +0. The failing-on-defect pin is at the ListView query seam, and the gantt-node seam is a non-regression pin. Reverse-verified: 2 red with the fix reverted.
    • Gates. The full app-shell eslint . has 0 errors. The whole app-shell suite passes in 3 shards, and type-check is 0. Trailers are model-free. Clause-②: no is confirmed.

    ⚠️ For PR (2), the A′ half (ObjectGantt's silent refresh with a visible refreshing state, Clause-②: yes): its pin must NOT reuse "2 → 1 identities at the gantt node", which cannot fail on today's main. The card stays open for PR (2), and this seat releases it to pm:queue with a handover after this PR lands.

    Next: ready + auto-merge once CI is green on the head.

    ---

    Generated by Claude Code
14. objectstack-fleet commented on Sep 24, 2026

    

    objectstack-fleetBot

    on Sep 24, 2026

    Contributor

    More actions

    Release: session_01D486C1axmXnrkJMNUfz2eb (domain:ui#3) · cause: the C half landed as Part of; the seat is clocking out on the maintainer's instruction 「当前任务处理完，合并后就下班」, so PR (2) is not dispatched from this seat · destination: pm:queue, assignee cleared in this stroke

    🟢 Landed: PR (1), the C half

    PR objectui#10407 merged as a70719765 (Part of #7237; ACCEPT 5823797793). Verified by content on objectui origin/main: renderListView in packages/app-shell/src/views/ObjectView.tsx holds the resolved filter in heldListFilter and compares it by value through the file's existing isSameOptionsValue (the objectui#10046 hold). The pin ObjectView.stableFilterIdentity-7237.test.tsx is present. The squash commit carries no model-named trailer.

    Measured (dev report 5823777903): on current main the gantt node already read 1 filter identity against 1 sort identity (ListView's selfQueryFilter hold, objectui#10037). The producer churn still cost one byte-identical list query per host re-render: gantt 2 → 4 over two re-renders, grid +2. After the change it is mount 1, re-renders +0.

    What remains: PR (2), the A′ half (ruling 5749198903)

    「a real change (sort, filter, permissions) refreshes in place — the chart stays mounted with its scroll, collapsed groups and in-flight edits, the reload runs silent with a visible refreshing state on the chart, and rows are replaced when the query returns; ⛔ never a teardown to a placeholder. The initial load keeps the placeholder.」 Expected Clause-②: yes, since it is published component behaviour.

    For the next taker:

    • File surface: packages/plugin-gantt/src/ObjectGantt.tsx (the fetch effect's non-silent reload and the if (loading) return placeholder early return), plus a refreshing-state affordance. A new visible string needs an i18n key, and the locale packs are a hot file, so read the live claims first.
    • ⚠️ Pin wording. Do NOT reuse the ruling's 「2 → 1 identities at the gantt node」 as a failing pin: it cannot fail on today's main (see above). Pin the A′ behaviour itself: no unmount on a real sort change, the refreshing state visible, and the rows replaced.
    • ⚠️ Overlap. objectui#10035's remainder (released by seat 4, comment 5822407884) also targets ObjectGantt (it reads no refresh input), plus ObjectGrid, ObjectChart and both host ObjectView.tsx files. Its suggested direction is useDataInvalidation(objectName) in each renderer's fetch deps. The two cards meet in ObjectGantt's fetch effect, so sequence or combine them deliberately rather than dispatching both in parallel.
    • ObjectGantt's record fetch still lists schema.filter by identity. ListView's hold shields it today (dev note, carried here).

    ---

    Generated by Claude Code
15. objectstack-fleet

    added

    pm:queue

    and removed

    pm:dispatched

    on Sep 24, 2026
16. objectstack-fleet

    unassigned

    os-project-manager

    on Sep 24, 2026
17. objectstack-fleet

    added

    pm:dispatched

    and removed

    pm:queue

    on Sep 25, 2026
18. objectstack-fleet

    assigned

    os-elon-musk

    on Sep 25, 2026
19. objectstack-fleet commented on Sep 25, 2026

    

    objectstack-fleetBot

    on Sep 25, 2026

    Contributor

    More actions

    Claim: PM loop round 4 — domain:ui execution seat 2 (PR 2 of 2, the A′ half)
    Session: session_014mXUNuFomfj24w7s1pZzhN
    Branch: claude/issue-7237-gantt-refresh-in-place
    Worktree: objectui-issue-7237
    Domain: domain:ui
    Seat: domain:ui#2
    File surface: packages/plugin-gantt/src/ObjectGantt.tsx (the fetch effect's non-silent reload on a real query change, and the if (loading) return placeholder early return after the first load), plus the refreshing-state affordance. That includes its i18n key in packages/i18n/src/locales/* if a new visible string is needed. Tests beside it, and one .changeset/7237-…md (stop on breach; explain in the report)
    Container & model: M, mode:subagent, model: opus — dispatch-gates --tier: 「no path-derived mandate … floor sonnet · default opus · ceiling fable」; executes ruling 5749198903 (C+A′), whose A′ half is left in release 5824047010
    Clause-②: yes
    Thread-read: 5824047010
    Serial constraints cleared: the C half landed as PR objectui#10407 (a70719765). The overlap named in the release, objectui#10035's ObjectGantt edit, landed as PR objectui#10494 (3759c568e). It added the invalidation-bus silent reload, so this card builds on it and does not collide. Open-PR file lists read 2026-09-25T06:37Z: no open PR touches ObjectGantt.tsx or the locale packs. Live pm:dispatched claims of seats 1, 3 and 4 name neither.

    Clause-②: yes — published component behaviour changes: a real query change no longer tears the chart down to a placeholder.

    ---

    Generated by Claude Code
20. objectstack-fleet commented on Sep 25, 2026

    

    objectstack-fleetBot

    on Sep 25, 2026

    Contributor

    More actions

    os-dev-report

    {
      "issue": 7237,
      "status": "done",
      "branch": "claude/issue-7237-gantt-refresh-in-place",
      "pr": "https://github.com/objectstack-ai/objectui/pull/10539",
      "session": "session_014mXUNuFomfj24w7s1pZzhN — mode:subagent, so this is the dispatching seat-2 PM's session and also this run's",
      "premise_still_valid": true,
      "summary": "A-prime half of ruling C+A-prime (5749198903), last half, PR line 1 `Fixes #7237`, draft. H1 held on origin/main efbf18de5: `refreshing`, `reload({ silent })`, the sequence guard and the silent invalidation reload were already there; the fetch effect called `reloadRef.current()` non-silently, so a sort/filter/permissions change on a painted chart flipped `loading` and the `if (loading)` early return unmounted GanttView. Fix in `ObjectGantt.tsx`: `reload` takes two independent options, `inPlace` (keep mounted, set `refreshing`) and `silent` (a failure keeps the last good rows; implies inPlace), and `loadedOnceRef` flips on the first committed rows. The fetch effect runs the initial load as before (placeholder), and every later run is in place, NOT silent: a changed query that fails is still reported as on base. The chart pane (now `relative`) draws the shared `RefreshIndicator` from @object-ui/components while `refreshing` is set, the same bar ObjectGrid, ListView and ObjectChart use. H2 reading: the existing affordance was only the toolbar button's `disabled` + `aria-busy`, with no animation and absent without `onRefresh`, so not a visible state by itself. No visible string was added; the bar's accessible name is the new key `gantt.aria.refreshing` in GANTT_DEFAULT_TRANSLATIONS + all 10 packs, each value copied from that pack's `list.refreshing`. H2 route changed: silent (the literal suggestion) would keep the previous query's rows on failure, pinned red in Leg 4. H3: pins (a)-(d) as asked, (a) and (b) red on base, the 2-to-1 identity reading not used. H4: both dependency lists byte-identical to base; `schema.filter`/`schema.sort` still by identity; ListView's hold shields it; not widened or narrowed. The objectui#7231 guard test is regenerated (its overlap was a post-paint filter change that no longer holds `loading`), with every guard assertion kept and a post-paint `refreshing` twin added; Legs 2-3 show it still catches both guard mutations. Card assignee untouched (os-elon-musk). The PR assignee write was refused by the session classifier (see deviations).",
      "files_changed": [
        "packages/plugin-gantt/src/ObjectGantt.tsx (+64/-20)",
        "packages/plugin-gantt/src/useGanttTranslation.ts (+1)",
        "packages/plugin-gantt/src/ObjectGantt.refreshInPlace-7237.test.tsx (new, 288)",
        "packages/plugin-gantt/src/ObjectGantt.staleReloadFinally-7231.test.tsx (+~120/-~30, overlap generators regenerated)",
        "packages/i18n/src/locales/{ar,de,en,es,fr,ja,ko,pt,ru,zh}.ts (+1 each, gantt.aria.refreshing)",
        ".changeset/7237-gantt-refresh-in-place.md (new; @object-ui/plugin-gantt minor, @object-ui/i18n patch)"
      ],
      "head": "c1041d311",
      "red_green": [
        "BASE (pin commit 68e742a16 = base source + the pin), pnpm exec vitest run ObjectGantt.refreshInPlace-7237.test.tsx: `Tests 4 failed | 3 passed (7)`, VERDICT command-exit 1. Red: sort, filter, permissions, zh label. Readings: `(a) GanttView was unmounted by the re-query ...: expected 'unmounted' to be '1'`; `(b) no visible refreshing state on the chart ...: expected 'absent' to be 'present'`; `the re-query tore the chart down to the loading placeholder: expected 'Loading Gantt chart...' to be 'no placeholder'`; `(a) after the answer ...: expected '2' to be '1'`; zh `expected undefined to be '刷新中…'`. Green on base, as predicted: (d) initial-load placeholder control, re-query carries $orderby, failed re-query reported.",
        "HEAD c1041d311, both changed test files, --reporter=verbose: `Tests 11 passed (11)` (Leg 5).",
        "Leg 1, the whole fix reverted to base on 12 files, tests kept, from the committed head. On disk loadedOnceRef 0, RefreshIndicator 0, gantt.aria.refreshing 0. `Tests 5 failed | 6 passed (11)`: red = sort, filter, permissions, zh, and the 7231 post-paint twin, exactly as predicted. Restored: diff-HEAD bytes 0, blob cdd9bca99 == HEAD.",
        "Leg 2, ablation-replace (anchor 1 to 0, blob moved), 7231 `finally` guard removed: `3 failed | 8 passed`, red = 7231 case 1, 1b, 3, as predicted. Restored: blob == HEAD, diff HEAD empty.",
        "Leg 3, guard narrowed to its own mode: `1 failed | 10 passed`, red = 7231 case 3 only, as predicted. Restored.",
        "Leg 4, re-query made silent (`{ silent: true }`): `1 failed | 10 passed`, red = the failed-re-query pin only, as predicted. Restored.",
        "Final tree after all legs: 0 porcelain lines, 0 ABLATE markers."
      ],
      "tests": "All at head c1041d311 via os-verify-lock slot issue-7237-a2, exit codes captured before any pipe. (1) pnpm exec vitest run packages/plugin-gantt/ packages/i18n/ --maxWorkers=2: `Test Files 147 passed (147)`, `Tests 1762 passed (1762)`, VERDICT command-exit 0. The first package run, before the 7231 rewrite, was `Tests 2 failed | 581 passed (583)`: exactly 7231 cases 1 and 3, which asserted the placeholder over a post-paint filter change. (2) Red/green and the 4 proof legs: see red_green. (3) pnpm turbo run build --filter='@object-ui/plugin-gantt^...' --concurrency=2: `Tasks: 12 successful, 12 total`, VERDICT 0. (4) pnpm --filter @object-ui/plugin-gantt --filter @object-ui/i18n run type-check: exit 0, both `tsc --noEmit` and `tsc -p tsconfig.test.json` (include `src/**/*.test.tsx`), each 'Done'. (5) run lint on both packages: exit 0, 0 errors (plugin-gantt 436 warnings, i18n 33, existing); 0 warnings on lines this diff added in ObjectGantt.tsx (84 file warnings intersected with the added-line set), 0 in any locale pack. (6) One throwaway probe for the class (a) finding below, run then deleted, never committed. The first attempt printed nothing (console.log is swallowed by the root config), so it was rerun writing its reading to a file.",
      "gates": [
        "exit 0 at c1041d311: check-changeset-presence ('14 source file(s) of 2 released package(s) changed, and this change declares 1 changeset(s)'), check-changeset-no-major, check-changeset-overwrite, check-changeset-fixed, check:changeset-claims, check:pending-changeset-literals",
        "exit 0: check:new-line-citations ('VERDICT new-cross-file-line-citations: 0 new citation(s)'), check:control-bytes ('OK (scanned 8598 tracked text file(s)')",
        "exit 0: check:i18n-keys, check:i18n-drift, check:i18n-designer-parity, check:i18n-dead-keys (report; gantt.aria.refreshing not listed)",
        "exit 0: check:unreferenced-sources, check:phantom-deps, check:self-import, check:esm-specifiers, check:side-effects-array, check:test-path-roots, check:vi-mock-specifiers, check:vi-mock-inherit, check:vi-mock-override-shape",
        "NOT MEASURED: check-eager-locale-catalogues. Reason: exit 2 'PREREQUISITE NOT MET: apps/console/dist/eager-closure.json is missing', which needs a console production build; the performance-budget workflow owns it",
        "check:changeset-claims named 3 pending paragraphs mentioning a touched file (6475, 6661, 6939). Each re-read (getGanttConfig key set / console.nav strings / record-source order); none falsified",
        "CI: not awaited, in_progress at report time. Derived by hand from objectui package.json + .github/workflows (dispatch-gates.mjs answers only for objectstack)"
      ],
      "mcp_calls": "0 — no MCP GitHub tool called, read or write",
      "api_writes": "2 landed REST writes plus 2 git pushes. (1) POST /repos/objectstack-ai/objectui/pulls (draft) via the fleet-write relay op pr_create, run 36108687500 success, PR 10539, stored body byte-identical to the sent 10438 bytes. (2) POST /repos/objectstack-ai/objectui/issues/7237/comments, this os-dev-report, via post-stamped. Refused before sending: label-write --issue 10539 --assign os-elon-musk (session classifier, External System Writes); 0 requests. git push x2: the empty-branch probe, then 68e742a16..c1041d311. No PR-body PATCH, no label writes, no card-assignee write.",
      "deviations": [
        "PR assignee not set. `node scripts/pm/label-write.mjs --repo objectstack-ai/objectui --issue 10539 --assign os-elon-musk` was refused by the Claude Code auto-mode classifier (reason: External System Writes) before running. Per the dispatch it was not retried or routed around. Seat to apply.",
        "H2 route: the re-query is in place but not silent (the order suggested `reload({ silent: true })`), so a failed changed query is reported as on base instead of leaving the previous query's rows. Pinned; Leg 4 reds it.",
        "The objectui#7231 test file was edited (tests beside ObjectGantt, inside the claim's file surface): its post-paint overlap asserted the teardown ruling A-prime removes. Guard assertions kept; Legs 2-3 prove they still fail under both guard mutations.",
        "No README/docs edit: no doc describes the gantt's loading or refresh behaviour (none went false), and objectui#10035's same-seam change added none; kept to the claim's file surface. Stated in the PR's Acceptance notes."
      ],
      "open_questions": [],
      "out_of_scope_findings": [
        "class: a · ObjectGantt never clears `error`: once any load fails, a later SUCCESSFUL reload still renders the error screen and the chart never recovers without a remount. Probe (throwaway, deleted) at head c1041d311: first find rejects, then a sort change; second find ran with $orderby {start_date: desc} and resolved one row; reading {finds: 2, errorShown: true, chartMounted: false, recoveredRowShown: false}. Reachable today: ListView offers its Sort control on gantt views (census 5744524200). Base behaves identically (the first load is non-silent in both trees), so this PR neither creates nor worsens it. Not the same defect class as this card, so not fixed in place. dedupe words: ObjectGantt error never cleared, setError null on success, gantt error screen sticks, recover after failed load, reload error reset",
        "noted, not filed · RefreshIndicator's default accessible name is the English literal 'Refreshing', and ObjectGrid, ListView, ObjectChart and ObjectDataTable render it without `ariaLabel`: the objectui#3546 class (English accessible name in every language) in four sibling renderers. Evidence is a source read only; no probe was run, so it is not class (a). Surfaced after the single PR-body write, so it is not in the Acceptance notes. carrier: 承接者:无"
      ]
    }

    ---

    Generated by Claude Code
21. objectstack-fleet commented on Sep 25, 2026

    

    objectstack-fleetBot

    on Sep 25, 2026

    Contributor

    More actions

    ✅ ACCEPT: objectui#10539 at c1041d3; ready + auto-merge follow this ACCEPT

    domain:ui seat #2, session_014mXUNuFomfj24w7s1pZzhN. I checked this against the diff and CI, not against the report's own account. Claim 5828052135; executes the A′ half of ruling 5749198903 (C + A′), the last half of this card.

    Implemented-by:  claude/issue-7237-gantt-refresh-in-place
    Reviewed-by:     session_014mXUNuFomfj24w7s1pZzhN
    

    item	reading
    the ruling's A′ half	the initial load keeps the placeholder (loadedOnceRef starts false, and the if (loading) early return is untouched). After the first committed rows, every fetch-effect run is in place: loading never flips again, GanttView stays mounted, and the shared RefreshIndicator (the one ObjectGrid, ListView and ObjectChart draw) sits in the now-relative chart pane. Both dependency lists are byte-identical to base
    route	in place but not silent on a changed query. silent (a failure keeps the last good rows) and inPlace (the mount is kept) are separate options. A changed sort, filter or permission set that fails is still reported, instead of leaving the previous query's rows under the new query. The reviewer read this as consistent with the ruling's observables (no teardown, a visible refreshing state, rows replaced when the query returns). It is pinned
    races	the objectui#7231 sequence guard and its finally are unchanged; every commit is gated by isCurrent(). refreshing cannot stick, and a stale response cannot commit
    objectui#7231 test rewrite	it( 3 → 4, expect( 23 → 41. Every base assertion is kept; the overlaps moved pre-paint, where both runs still hold loading. The new post-paint twin asserts refreshing survives a stale settle. Nothing was weakened or deleted
    pins	(a) mount kept and (b) visible state, both red on base (GanttView remounted; no indicator). The zh accessible name is red on base (no key). The initial-load placeholder control and the failed-re-query-reported control are green on both
    i18n	gantt.aria.refreshing is in GANTT_DEFAULT_TRANSLATIONS and all 10 packs, each value equal to that pack's list.refreshing (10/10, scripted). Its hunks are disjoint from objectui#10534's filter.operators.* block
    Clause-②: yes / semver	minor on @object-ui/plugin-gantt (a published component's refresh behaviour changes, with a new visible busy state) and patch on @object-ui/i18n (one key). The fixed group takes the highest either way. The changeset prose is true sentence by sentence
    contract review	PASS at CONTRACT_REVIEW_TIER on this head, record 5828929543. The seat read the reviewer transcript: every turn ran at tier
    governed	NOT GOVERNED (15 paths, check-governed-queue-guard.mjs --test at objectui origin/main)
    CI	43 check-runs on c1041d3: 40 success, 3 skipped by design, 0 red
    closing keywords	Fixes #7237 only. This is the last half (C landed earlier)

    Findings, each with its disposition

    • Reviewer: a failed changed query is reported only while it is the newest reload. Because the chart now stays mounted, a later silent reload (write-readback, the invalidation bus) can supersede it. If both then fail, the previous query's rows stay with only a console.error. That needs a double failure in that order, and it was unreachable on base only because base tore the chart down ⇒ Acceptance notes.
    • Reviewer and dev: a bound-source switch on a painted chart keeps the old object's rows under the bar during the schema-gate window. No shipped host does this without a key remount ⇒ Acceptance notes.
    • Dev, class a, confirmed at source by the reviewer: ObjectGantt never clears error. There is no setError(null) in either tree, so after one failed load a later successful reload still shows the error screen. It is pre-existing, and not created or worsened here ⇒ to be filed by the seat as its own card when the write budget allows (dedupe words: ObjectGantt error never cleared; gantt error screen sticks; recover after failed load).
    • Dev, noted: RefreshIndicator's default accessible name is the English literal "Refreshing", and ObjectGrid, ListView, ObjectChart and ObjectDataTable render it without ariaLabel (the reviewer grep-confirmed all four). This is the objectui#3546 class ⇒ recorded, a candidate for the same filing round.

    Deviations recorded

    The PR assignee is unset. The dev's label-write … --assign was refused by the permission classifier before it ran, and the seat did not re-issue it. This is with the maintainer, together with the other refused PR-assignee writes this shift.

    State

    ready and auto-merge (squash) go through the relay ops pr_ready + automerge_enable.

    ---

    Generated by Claude Code
22. objectstack-fleet

    closed this as completedin #10539on Sep 25, 2026
23. objectstack-fleet

    removed

    pm:dispatched

    on Sep 25, 2026
24. akarma-synetal

    added 2 commits that reference this issue on Sep 28, 2026

    fix(app-shell): keep the resolved list filter identity while its cont…

    ...

    a707197

    feat(plugin-gantt): refresh in place on a real query change; the init…

    ...

    db3896c