Repository navigation
fix(console): the docs portal's book sidebar keeps a doc placed by its own group key from outside the book's package (objectui#11245) - #11274
Conversation
…key from outside the book's package (objectui#11245) `scopeDocsToBook` dropped every doc outside the book's packages before `resolveBookTree` ran, so a doc whose own `group` names one of the book's group keys was missing from the sidebar while `GET /meta/book/:name/tree` (the spec resolver: explicit `group` carries no package scope) listed it. It now keeps such a doc whatever its package. The package filter still narrows what `include` and the synthetic Uncategorized group can collect. Pins: the helper answer against the spec resolver's for the same docs, the count, a group-level `package` override, the Uncategorized control, and the rendered sidebar / book landing through the real route table. Claude-Session: https://claude.ai/code/session_0122Knsowci76D2rBWReCzzZ Co-authored-by: Claude <noreply@anthropic.com>
✅ Console Performance Budget
The eager closure is every chunk the entry reaches through static imports — what the browser fetches and parses before the app renders. The entry chunk on its own is a small fraction of it. 📦 Bundle Size Report
Size Limits
|
Contract reviewServed-tier: Inputs read: card objectui#11245 (body; comments ① Derived judgmentsThe diff is one hunk in
② Semver level
③ Boundary flags
Implemented-by: VERDICT: PASS |
…, mirror two chatbot keys, file eighteen runtime slots by name (objectui#6152, round 4) (objectstack-ai#11285) Refs objectstack-ai#6152 Clause-②: yes (narrowing). The two mirrors widen `@object-ui/types/zod` and the strict authoring face to keys the TypeScript face already invited and the renderers read (`requestBody`, `floatingConfig`). The four retirements (`label.content`, `report.chartConfig`, `report.reportType`, `detail-view.autoDiscoverRelated`) narrow both faces at once, under the startup no-gradualism rule. It is priced in `.changeset/6152-unmirrored-round4.md` as `minor`, with the break stated. Round 4 of objectui#6152, dispatched under claim `5919877490` (the `domain:devx` seat 2, session `session_01TdiauJaVCHuj45EzZGUxHh`, on the maintainer's direct order). The rulings are the seat review `5914490915`: Q1 B, Q2 A, Q3 B. ⛔ This is a draft from a dispatched dev: do not mark it ready and do not enable auto-merge. The card stays open, because `ObjectGridSchema`, the spec-derived pairs and three `DataTableSchema` keys (below) are not this round's. ## What changed 1. **Q1 B: `LabelSchema.content` is retired.** It is `?: never` on `form.ts` and a `retirementTombstone` on `form.zod.ts` that names the kept spellings `text` and `label`. The `label` renderer drops its third read limb (`schema.text || schema.label`). Re-measured: 0 documents author it (JSON, Markdown JSON fences and typed literals); its only author was a test in `form-renderers.test.tsx`, which now pins the refusal. Which of `text` / `label` is canonical is not decided here. 2. **Q2 A: `DetailViewSchema` `activities`, `comments`, `history` and `recordNavigation` go to `RuntimeOnlyDeclared` by name.** Each has a row with its reason on `RuntimeOnlyNamedAllowList`, in round 3's shape. They carry host-fetched runtime data, and `recordNavigation.onNavigate` is a required function. 3. **Q3 B, under objectui#7654 ruling B: `displayMode` keeps its TypeScript-only refusal.** It gets no zod arm, because stored designer documents carry `displayMode: 'floating'` (pinned by `chatbot-display-mode-retired.test.ts`). The ledger entries no longer call the mirror half "owed"; the key stays listed by decision. Separately, after a reader census for each: - `ChatbotSchema.requestBody` is mirrored through the shared pick. All three chatbot registrations read it (`body: schema.requestBody`). - `ChatbotFloatingSchema.floatingConfig` is mirrored through a module-private restatement of `FloatingChatbotConfig`. That restatement judges the six members `FloatingChatbot` destructures, and carries the `triggerIcon` tombstone's runtime half that the objectui#7654 tripwire asked for. 4. **The three zero-read candidates are retired on both faces.** They are `ReportComponentSchema.chartConfig`, `ReportComponentSchema.reportType` and `DetailViewSchema.autoDiscoverRelated`, each `?: never` plus a `retirementTombstone`. `reportType`'s one in-code producer, `specReportToPresentation` in `spec-report.ts`, no longer writes it, and `LegacyReportPresentationLike` no longer declares it. 5. **`DataTableSchema`: 14 of 17 keys go to `RuntimeOnlyDeclared` by name.** They are produced only in code, by `ObjectGrid`'s in-code `data-table` node or `RelatedList`'s, and authored by no document. Three keys fit none of the classes, so the round stopped at them (see "Stopped" below). 6. **The pagination corpus fix, carried from round 3.** `components-basic-pagination/with-item-count.json` no longer authors `pageSize` / `totalItems`, which neither face declares and nothing reads. The plaintext interface block on `content/docs/components/basic/pagination.mdx` stops listing them, and its section is retitled from "With Page Size" to "On a Later Page". 7. **Changesets.** One new: `.changeset/6152-unmirrored-round4.md` (`@object-ui/types` minor, `@object-ui/components` minor, with the break stated). Four dated, append-only corrections were added, each 2 lines with frontmatter unchanged: `6639-objectgrid-title-mirrored.md` (the stale "97 + 1 + 23" sentence), `7654-floating-chatbot-trigger-icon-tombstone.md`, `7655-chatbot-registration-authoring-faces.md` and `9447-detail-view-handler-key-refusal.md`. Every other pending changeset was re-read after the merges, including the 8 that landed since the base, and none is made false by this round. ## Ledger totals, before and after These are read by the file's own pins (`objectui#7279` derives every figure) and re-counted with the TypeScript parser on the final head. | ledger | before (`origin/main` `e420df3`) | after (`2cc4ab9`) | | --- | --- | --- | | `UnmirroredDeclared` | 9 entries / 44 keys: 3 / 14 spec-derived, 6 / 30 local | **6 / 20**: 3 / 14 spec-derived, 3 / 6 local | | `RuntimeOnlyDeclared` | 4 / 9 | **4 / 27** | | `RuntimeOnlyNamedAllowList` | 1 entry / 2 rows | 3 / 20 | | cross-ledger | 2 of the 4 are in `UnmirroredDeclared`; union 11 = 9 + 2 | **1 of the 4**; union **9** = 6 + 3 | | `KnownDrift` | 49 / 86 | 49 / 86, unchanged. This round does not edit its total line; PR objectstack-ai#11269's 87 to 86 edit is kept as merged. | | `WiderThanDeclared`, `MirroredUndeclared` | 3 / 3, 2 / 2 | unchanged | The 24 keys that left `UnmirroredDeclared` moved by three routes, which must not be added up as one repair: 2 were mirrored, 4 were retired, and 18 were reclassified by name. The reclassification is one move seen from both ledgers: it is not new debt, and it mirrors nothing. ## Per-key table | pair · key | measured (type-checker read census over 5548 sources; authored census over tracked JSON, Markdown JSON fences and `type` literals) | route | | --- | --- | --- | | `label` · `content` | 1 read (the third limb in `label.tsx`), 0 documents (1 test) | retired on both faces; the read is dropped | | `report` · `chartConfig` | 0 typed reads. The untyped `widget?.chartConfig` hits are plugin-dashboard's `DatasetWidget` (`widget: any`, a dashboard widget), which plugin-report never imports. 0 documents | retired on both faces | | `report` · `reportType` | 0 reads; one in-code producer (`specReportToPresentation`); 0 documents | retired on both faces; the producer write is removed | | `detail-view` · `autoDiscoverRelated` | 0 reads (typed or untyped), 0 documents | retired on both faces | | `chatbot` · `requestBody` | 3 reads (all three chatbot registrations) | mirrored (shared pick) | | `chatbot-floating` · `floatingConfig` | 1 read (`floatingConfig={schema.floatingConfig}`); 6 of 7 `FloatingChatbotConfig` members read, and `triggerIcon` 0 | mirrored, with a `FloatingChatbotConfig` restatement | | `chatbot`, `chatbot-floating` · `displayMode` | 0 reads; stored documents carry it | unchanged by ruling (TypeScript-only refusal) | | `detail-view` · `activities` (7 reads), `comments` (4), `history` (7) | host-fetched rows; 0 documents | `RuntimeOnlyDeclared` by name | | `detail-view` · `recordNavigation` | 11 reads; its value carries a required function | `RuntimeOnlyDeclared` by name | | `data-table` · `disableInnerScroll`, `manualPagination`, `manualSearch`, `manualSorting`, `page`, `rowActionDefs`, `rowClassName`, `rowCount`, `rowStyle`, `search`, `selectionResetKey`, `showAddRow`, `showSelectionCount`, `sort` | read by `data-table`; produced only in code (`ObjectGrid`, `RelatedList`); 0 docs and 0 catalog documents | `RuntimeOnlyDeclared` by name; the reasons are on the allow-list | | `data-table` · `editable`, `singleClickEdit` | read; produced in code by `ObjectGrid`; 0 documents, but `data-table.mdx` lists both and teaches `editable: true` | **stopped**, ruling needed | | `data-table` · `selectionStyle` | read; 0 producers and 0 documents (0 hits in objectstack too) | **stopped**, ruling needed | ## Runtime probe for the three zero-read retirements AGENTS.md requires more than a source census, because leftover keys reach a renderer as spread props. So the probe used the real `SchemaRenderer` and the real registry, and varied only the key. It was a one-off: the files were deleted afterwards and `git diff HEAD` is empty. - `report`: the DOM is identical with and without `chartConfig` plus `reportType` (797 = 797 characters, byte-equal). Lit control: changing `title` changes the DOM (805). - `detail-view`: the DOM is identical with and without `autoDiscoverRelated: true`, after normalising React's `useId` tokens. The only raw difference was `radix-_r_5_` against `radix-_r_d_`. The data-source and `fetch` call log is identical: `getObjectSchema(account)` and two `security/explain` fetches. Lit control: changing `title` changes the DOM (5943 against 5951). ## Ablations The predictions were written at `4c272dc` before any run, and every leg was restored with the blob equal to `HEAD` and `git diff HEAD` empty. - **A1:** plant a `LabelSchema` literal carrying `content` into `label.tsx`. Components `tsc --noEmit` exits 2 with exactly one TS2322 there (`'string' is not assignable to type 'undefined'`), as predicted. This red direction also shows the consumer read the rebuilt `.d.ts`. - **A2:** restore the four `zod/*.zod.ts` files to base `f61dab1`. `unmirrored-declared-round4-6152.test.ts` reads `27 failed | 6 passed (33)`, exactly as predicted, and `tsc -p tsconfig.test.json` exits 2 with its only error in the ledger reconciliation of `zod-mirror-parity.test.ts`. - **A3:** rename the allow-list row `rowCount` to `rowCountRenamed`. `tsc -p tsconfig.test.json` exits 2 with exactly two TS2344, at the two allow-list pins, as predicted. ## Verification on the final head `2cc4ab9` This round resumed after two container restarts, so every run below was redone on `2cc4ab9`. That head carries `origin/main` `e420df3`, merged twice (never rebased): once through PR objectstack-ai#11275, and before that through PRs objectstack-ai#11267, objectstack-ai#11268, objectstack-ai#11269, objectstack-ai#11271, objectstack-ai#11273 and objectstack-ai#11274. Each run went through the shared verify lock with at most 2 workers. - `@object-ui/types`: `type-check` (the `--noEmit`, `tsconfig.examples.json` and `tsconfig.test.json` legs) exits 0; `vitest run packages/types/` gives 299 files and 7498 tests passed; `build` exits 0 (138 emitted files verified). - Consumers, from the type-checker census: `type-check` exits 0 with 0 TS errors for `components`, `plugin-report`, `plugin-detail` and `plugin-chatbot`. `vitest run packages/plugin-report/ packages/plugin-chatbot/` plus `form-renderers.test.tsx` gives 78 files, 827 passed and 7 skipped. All 228 `plugin-detail` test files, run in two halves, give 227 files passed plus 1 skipped, and 2264 tests passed plus 8 skipped. - Zod-face consumers: `vitest run examples/schema-catalog/ packages/cli/` plus the 17 app-shell files that import `@object-ui/types/zod` or sit in `metadata-admin/previews/__tests__` give 77 files and 2741 tests passed. - Gates, all exit 0: `check:changeset-claims`, `check:pending-changeset-literals`, `check-changeset-presence` (16 source files of 2 released packages, 1 changeset), `check-changeset-no-major`, `check-changeset-fixed`, `check-changeset-overwrite` (report-only; its 4 findings are the 4 dated appends above), `check:control-bytes` (9717 files), `check:doc-fences`, `check:doc-snippets` (697 of 697 blocks judged, 0 failed), `check:doc-examples`, `check:doc-types`, `check:doc-example-ids`, `check:doc-example-readers`, `check:skill-examples`, `check:handler-key-reads`, `check:component-surface-parity` (report-only), `check:designer-field-key-parity`, `check:element-data-source-declaration`, `check:new-line-citations` (0 new), `check:prompt-keys`, `check:readme-exports`, `check:spec-symbols`, `check:test-path-roots`, `check:unreferenced-sources` and `check-type-check-coverage`. `doc-snippets` and `doc-examples` ran after the scoped build their own `--build-filter` names (35 turbo tasks, all successful). - Lint, narrowed: `eslint --no-inline-config --format json` over the 16 lintable changed files gives 0 errors. All 16 were judged and 0 were ignored, a population read from eslint's own output. The 39 warnings are pre-existing `no-explicit-any`, and the diff adds no explicit `any`. `eslint.config.js` declares no `parserOptions.project` or `projectService`, and no custom rule in `eslint-rules/` reads the disk, so this diff cannot move any untouched file's verdict. The repo-wide `pnpm lint` is CI's. - Governed surface: `check-governed-queue-guard.mjs --test` over the 23 paths reports NOT GOVERNED. - NOT MEASURED: - `check:sdui-registration-pins`: it needs a console build, and this diff changes no registration. - The full `pnpm test` farm and repo-wide lint: CI's. - `plugin-grid`: its code and tests are untouched. The 14 reclassified keys moved between two test ledgers and no declaration changed. ## Stopped (ruling needed; reported in the card report with options) - `DataTableSchema.editable` and `singleClickEdit`. `data-table` reads both and `ObjectGrid` sets both in code, while no document authors them. But `content/docs/components/complex/data-table.mdx` lists both in its interface block and teaches `editable: true`. Filing them runtime-only would contradict that page. Mirroring them would rest on no authored document, and, from the source, an authored `editable: true` stages cell edits locally and persists nothing without the `onRowSave` / `onCellChange` function slots. - `DataTableSchema.selectionStyle`. `data-table` reads it (`'always'` or `'hover'`), and nothing authors or produces it in either repository. - Noted, not narrowed: `ChatbotSchema.floatingConfig` has 0 reads on the `chatbot` node (only `chatbot-floating` reads it), which makes it an ADR-0049 narrowing candidate. The `chatbot` half of `floating-chatbot-trigger-icon-retired.test.ts` pins its absence as a tripwire. ## Parallel discipline PR objectstack-ai#11269 and PR objectstack-ai#11273 edited `zod-mirror-parity.test.ts` in disjoint hunks: the `KnownDrift` total line, the Kanban entry and one `EXCLUSIONS` row. Both merges were clean. This round's ledger edits stay in the `UnmirroredDeclared` / `RuntimeOnlyDeclared` / `RuntimeOnlyNamedAllowList` blocks, their prose totals and its own pairs' entries. Every prose total was reconciled to the measured count above. The diff does not touch `zod/objectql.zod.ts`, `zod/index.zod.ts`, `zod/nested-component-walk.ts`, `strict-authoring-face.ts`, `zod/README.md`, `field-types.ts`, the `DetailViewField` member, or `data-table.tsx` (read only). ## Acceptance notes - 23 files, under the ~25-file size guard, so no pair was cut. - Every pending changeset this round made false carries a dated append. The ones it leaves true were re-read: `7654-chatbot-display-mode-tombstone.md`, `6150-…`, `7804-…`, `6175-…`, `7722-…`, `7947-…` and `8072-…`. Session: `https://claude.ai/code/session_01TdiauJaVCHuj45EzZGUxHh` --- _Generated by [Claude Code](https://claude.ai/code/session_01TdiauJaVCHuj45EzZGUxHh)_ --------- Co-authored-by: Claude <noreply@anthropic.com>
2 participants
Fixes #11245
Clause-②: no
What
scopeDocsToBookinapps/console/src/pages/book-nav.tsnow keeps a doc whose owngroupnames one of the book's group keys, whatever package the doc lives in. Everything else about the scoping step is unchanged. When the book declares a package, docs from other packages are still dropped before resolving, so they never reach the synthetic Uncategorized group or anincluderule.includeis still scoped by the localresolveBookTreeport itself, togroup.packageor else the book's package.resolveBookTree,homeBookand the direct-route refusal are not touched. There is no new helper.DocPage.tsx,BookPage.tsxandcountBookDocspick up the fix through the helper they already call.Why
The spec's
resolveBookTree(packages/spec/src/system/book.zod.tsin objectstack, byte-identical at the@objectstack/spec@17.5.0tag this repo installs and on objectstackmain) places a doc in a group byd.group === group.keywith no package scope.GET /meta/book/:name/treeanswers that resolver over every doc (createMetaBookTreeAnswer→readableTree, withbookPackage = book._packageId). The portal dropped the doc before resolving, so a doc an admin placed from the doc editor was listed by the endpoint and missing from the sidebar.Measured (the order's Zone 2)
These readings come from a scratch suite that ran the port and the spec resolver on the same fixtures. The suite was not committed.
scopeDocsToBook→ port, the book gavestart=[docprobe_gs_welcome] reference=[]. The spec resolver on the endpoint's inputs gavereference=[live_guide,probe_curl]. After the fix the two agree.groupleg. But neither the port nor the spec scopes the orphan catch-all. Both put another package's ungrouped doc into Uncategorized:uncategorized=[other_note]from each. So "stop pre-filtering and pass the scope" would break the ruled control. The fix therefore keeps the pre-filter and lets explicit members through it, as the ruling words it: "scopeDocsToBookkeeps any doc whosegroupnames one of the book's group keys". The portal's scoped Uncategorized is a deliberate, ruled narrowing of the endpoint's. The pins compare the authored groups with the spec's answer.packageoverride. In both resolvers the override replaces the book's scope for that group'sinclude; it does not add to it. Fixture:owngets only package a,extgets package b throughincludeplus c_1 through its owngroup. The portal's answer equals the spec's, and one pin holds it.homeBook/findBookContainingDoc. Nothing changes for a book that carries a stampedpackageId.homeBookkeeps only books of the doc's own package, and that package is already in scope for them. One edge case moves. The book must carry no stampedpackageId, its name prefix must equal the doc's package, and its only package must be a grouppackageoverride. If such a doc names one of that book's groups, it is now claimed by that book.homeBookthen answers that book instead of the package's first book (the measured fixture moved fromq_othertoq_manual). That followshomeBook's documented rule (the home is the book that claims the doc), andhomeBookwas not changed.Divergences beyond the explicit-
groupleg, reported and not widened:include. The spec admits a doc with nopackageIdinto any scopedinclude. The port falls back to the name prefix (pkgOf) and refuses it. Measured: portall=[x_a], specall=[x_a,y_unstamped].pagespin naming a doc from another package. The pre-filter drops that doc's header, so the entry keeps its place but loses its label and description. Measured: port{"doc":"c_9"}, spec{"doc":"c_9","label":"C9 Label"}.Tests
book-nav.explicitGroup-11245.test.ts: the explicit member is listed in the group it names; the authored groups equal the spec resolver's answer (@objectstack/spec/system) for the same docs;countBookDocscounts it; the group-override case; control: another package's ungrouped doc, and one naming a key the book lacks, stay out, with no Uncategorized.docs-portal.explicitGroup-11245.test.tsxrenders through the real route table. The Reference group of the rendered sidebar links the placed doc, and marks it current when reading it. A book whose only member is placed that way opens to it instead of "This book has no documents yet". Control: another package's ungrouped doc is absent from the sidebar.ablation-replace.mjswith a disk-verified mutation and restore:docs-portal.test.tsxcases (that is the existing pin the triage names).git diff HEAD.Gate verdicts and the head they ran on are in the
os-dev-reportcomment on the card.Acceptance notes
homeBookedge case (reading 4 above): the canonical URL moves for a doc placed into an own-package book that has no stampedpackageIdand a group-levelpackageoverride.pkgOffalls back to the name prefix on an unstamped doc, and apagespin from another package loses its label. Carrier: none.referenceplaces a doc in every book that has a group with that key. The spec resolver and the tree endpoint do the same; the pins do not cover it.Generated by Claude Code