Repository navigation
Studio: the publish sheet and load errors speak in plain words, and a load error carries the server's real reason #29704
Description
epic:#28130#1 · hotlong · session_040dfd80c4c9411b895c698e785a6848
Filing gate: ② filed on the maintainer's word in direct chat with the epic PM seat, 2026-11-03: 「第二批也立卡」. Priority on 「相关卡 p1, 让车道处理」. ⛔ Not a claim.
Reader: domain:ui (ui/packages/app-shell/src/preview/DraftChangesPanel.tsx; the list load error at console/AppContent.tsx:2112).
Dedupe: publish sheet, loadErrorMessage. #29673 is the case that exposed the second part, and its fix does not cover it. New.
type: enhancement
Business reason. The publish sheet is where a person decides whether to make a change live. It should say what will change in their terms. When a page cannot load, it should say why.
What happens
Measured on main at 3482f3d8d2, zh-CN.
- The sheet's diff for a customized dashboard reads
+ overridesand~ Label. A view reads a raw key list, and the extension's type group readsobject_extension. An independent review also saw+ isDefault,+ viewKindand 平台生成 (6) on a new item. - An object that failed to register (Studio in zh-CN: a new object publishes "successfully" and never reaches the data plane, because its echoed queue column carries a translated label #29673) has its records page ask the person to check their network (请检查网络连接后重试), though the server answered 404
OBJECT_NOT_FOUNDwith a clear sentence.AppContent.tsx:2112uses the generic sentence.
Direction
- 新增: no key list. 更新: top-level keys through the inspector's existing label catalog, and declaration or platform-generated keys are not listed as the person's changes. Type groups use the type's translated name.
- The load-error state shows the server's code and sentence when one came back. The network sentence is only for a request that got no answer. One rule, on the list and record pages alike.
- No new diff capability.
Activity
objectstack-fleet commentedon Nov 3, 2026
domain:ui#5 · zhuangjianguo · session_01QsR2jQjvWkZ7is6GXbC6DB
Claim: PM loop round 7 · 2026-11-03T02:38Z
Session:
session_01QsR2jVQc3uTU5ZhjP31Sh8(card thread, os-dev); dispatched by seat sessionsession_01QsR2jQjvWkZ7is6GXbC6DBAccount:
zhuangjianguoBranch:
claude/project-thread-jehz9h(platform-assigned to the card thread; it can push only this branch)Worktree:
objectstack-issue-29704Domain:
domain:uiSeat:
domain:ui#5Scope: the publish sheet names a person's changes in plain words (no key list on 新增, labelled top-level keys on 更新, no declaration or platform-generated keys, translated type-group names), and a list or record load error shows the server's code and sentence when one came back, keeping the network sentence for a request that got no answer.
Fixes #29704.File surface:
ui/packages/app-shell/src/preview/DraftChangesPanel.tsxand its helpers, the load-error lines ofui/packages/app-shell/src/console/AppContent.tsxand of the record page, one shared error-reading helper, the i18n packs, their tests, and a changeset. The dev names the exact files after measuring the premise, and the seat amends this line before any edit outside it (stop on breach).Container & model: M,
mode:thread,model: default tier (no Fable per the maintainer's standing rule),effort: highClause-②: yes (widening) — provisional: new i18n keys are expected. The dev runs on the diff, and the seat amends this line if its output differs (for example, a retired key).
Metadata-protocol: none
Responsibility:
a builder publishing a change, or a member opening a list or record that failed to load | the publish sheet shows raw keys and engine names, and a 404 with a clear server sentence reads as a network fault | builders deciding what goes live, and support chasing a network fault that is not thereThread-read: none
Premise: to prove first on current main, per the brief (/mnt/project-files/issue-29704/brief.md). The dev reproduces each symptom before editing and drops or reports any that does not reproduce.
Serial constraints cleared: seat 4's #29693 and #29692 (StudioItemHeader row) and seat 1's #29691 (saved-to sentence on the pending-changes sheet) are named in the brief. Seat 5's #29704, #29705 and #29706 share the studio i18n packs on different keys. The dev merges main right before pr_create.
objectstack-fleet commentedon Nov 3, 2026
epic:#28130#1 · hotlong · session_040dfd80c4c9411b895c698e785a6848
2026-11-03T02:52Z
Re-measured live by the epic PM seat on
mainatc9636f1e50, on the maintainer's word in direct chat, 2026-11-03: 「需要复测的内容，你直接处理」. This updates the card body where it differs. ⛔ Not a claim.Confirmed, zh-CN. The sheet for a new object lists
+ Label,+ Name,+ Sharing Modeland 平台生成（6）, and a new flow lists+ Edges,+ Label,+ Name,+ Status,+ Type, each under 包含：. The field rows above them are readable (+ 名称 name). TheisDefaultandviewKindlines belong to a new view, which was not created here; expect the same list.Records page after a failed registration:
GET /data/OBJanswers 404OBJECT_NOT_FOUNDwith a clear sentence. #29673 is still open and still reproduces on this build, so the case stays available to test against.objectstack-fleet commentedon Nov 3, 2026
domain:ui#5 · zhuangjianguo · session_01QsR2jQjvWkZ7is6GXbC6DB
Claim amended · 2026-11-03T02:52Z · #29704 file surface named after the premise held
The dev measured the premise at
7b55ea95(file-surface.md in issue-29704).OBJECT_NOT_FOUNDrendersdata-error-kind=networkand "Check your connection". So the list site is plugin-list'sListView, not AppContent ~L2112: in Studio in zh-CN: a new object publishes "successfully" and never reaches the data plane, because its echoed queue column carries a translated label #29673 the meta read answered 200 and the data read 404.RecordDetailView, andPageView's record scope), an answered 5xx reads as a network fault. A 404 there already reads "Record not found".overridesand English key labels raw in zh-CN, and groupsobject_extensionwithout a translated name.File surface (replaces the claim's provisional line; stop on breach):
ui/packages/app-shell/src/preview/DraftChangesPanel.tsx,preview/changeKeyLabels.ts, newviews/metadata-admin/schema-field-label.ts(a leaf, with members moved verbatim),views/metadata-admin/i18n.ts(re-exports), andviews/metadata-admin/metadata-form-i18n.ts. Tests:preview/__tests__/DraftChangesPanel.namedChanges-27835,.generatedGroup-25890,.protectionEnvelope-12594and.groupHeading-13221.ui/packages/react/src/utils/error-message.ts(one shared answered-vs-unanswered helper, lifted from RelatedList'sserverSentencerule),react/src/index.ts(its export) andutils/error-message.test.ts. Alsoui/packages/plugin-list/src/ListView.tsxwith new testsrc/__tests__/ListView.loadErrorKind.test.tsx,ui/packages/plugin-detail/src/RelatedList.tsx(the local rule replaced by the shared helper),app-shell/src/views/RecordDetailView.tsxwithRecordDetailView.recordReadFailure-11902.test.tsx, andapp-shell/src/views/PageView.tsxwith__tests__/PageView.appPageRecord-12931.test.tsx.ui/packages/i18n/src/locales/*.tspacks: thepreview.changes.typeCount.object_extensioncount family andpreview.changes.detailNewItem. No key is removed..changeset/29704-plain-publish-sheet-and-load-errors.md, with patch bumps for app-shell, react, plugin-list, plugin-detail and i18n.⛔ Out: the console route's
RouteObjectReadFailed(AppContent.tsx,MetadataProvider.tsx,appObjectScope.ts). Its defect is by reading, not measured. The fix would carry the error through MetadataProvider's by-name read state, andappObjectScope.tsis the landing site of seat 2's #29681 and #29552. The seat files it as a separate card.Clause-②: yes (widening) — new i18n keys and one new
@object-ui/reactexport. The dev runsscripts/pm/clause2-line.mjson the diff, and the seat amends again if its output differs.Metadata-protocol: none
Serial:
DraftChangesPanel.tsxis the sheet that seat 1's #29691 edits, on branchclaude/project-thread-pkn482. Seat 1 edits the props tail, a new export, the destructure and the header. This card editsTYPE_COUNT,EntryDetail,KeyRowand the imports. In the packs, seat 1 adds atchanges.titleand this card atdetailNewKeys/typeCount. The dev merges main right before pr_create.objectstack-fleet commentedon Nov 3, 2026
domain:ui#5 · zhuangjianguo · session_01QsR2jVQc3uTU5ZhjP31Sh8
os-dev-report
dispatch-gates --ranreconciles to 0 NOT-MEASURED, 0 UNRUN.node scripts/check-sdui-manifest.mjsexit 0. The body gatescheck-changeset-no-major,check-adr-0087-registration,check-closing-keyword-parityandcheck-partof-closing-keywordeach exit 0.git diff HEADempty afterwards. The log is in issue-29704/measure/ablations.log.Clause-②: yes (widening)andMetadata-protocol: none, both read as declared by clause2-linereadClause2Line/readMetadataProtocolLine. The changeset is a patch for app-shell, react, plugin-list, plugin-detail and i18n.top=0) as an empty collection, so the route answered 404 only on the list's expand read (populate=), the raw path Studio in zh-CN: a new object publishes "successfully" and never reaches the data plane, because its echoed queue column carries a translated label #29673 took. On that plain path, an answered 404 still reads as 「这里还没有数据」 and never reaches the error panel. This is adapter behaviour outside this card.objectstack-fleet commentedon Nov 3, 2026
domain:ui#5 · zhuangjianguo · session_01QsR2jQjvWkZ7is6GXbC6DB
Landed · 2026-11-03T05:09Z
69dbb3cc. The seat ACCEPT is 6595631277, and the director's same-head PASS is 6595471560 oneae9d624.Fixes #29704. No other card was closed by the body.pm:dispatchedis removed and the card stays closed. Out of scope: the route-levelRouteObjectReadFailedis console: a route whose object read was answered (5xx, or a 404 with an unknown code) says "check your connection" instead of the server's reason #29728, and the adapter memoising a list's own 404 as empty is data-objectstack: a plain list read answered 404 OBJECT_NOT_FOUND reads as an empty list ("no data yet"), so the list never shows the server's reason #29782.