Repository navigation
test(types): pin NamedListView's per-member liveness census (objectui#7924) - #8933
Merged
Merged
Conversation
…#7924)
Re-derive both halves of the `listViews` measurement on today's head and pin
them per member, replacing the regex instruments with the TypeScript parser.
- declared set: walked with `ts.createSourceFile` over the interface's own
`PropertySignature` members, not a regex and not a brace-depth count. The two
regexes are kept and asserted against the parser, so the instrument swap is
itself a reading: strict regex 47 = parser 47, loose regex 59.
- read set: derived through every route from `schema.listViews` to a named
view, with every occurrence of that record classified. A new route fails as
UNCLASSIFIED instead of silently shrinking the reading. This found a
named-view read the old `currentNamedViewConfig?.KEY` regex cannot see —
`{view.label || key}` on the tab strip.
- each of the 47 members is pinned by name on its side of the partition
(6 read, 41 unread), so a member moving between the sets fails in either
direction; `data` stays read-but-undeclared, reached via an `as any` cast.
A member reached through `activeView?.KEY` — the host's `views` prop — is not
read off the named view; `rowHeight: activeView?.rowHeight` is the pinned
counter-control.
Measurement only: no declaration face moved and no renderer read set moved.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Jmxdo7bmeqCQHLSfmLVX9w
Contributor
✅ 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
|
This was referenced Sep 10, 2026
os-zhuang
marked this pull request as ready for review
September 10, 2026 04:20
Collaborator
Author
Green on
|
os-zhuang
deleted the
claude/issue-7924-namedlistview-liveness-census
branch
September 10, 2026 04:37
akarma-synetal
pushed a commit
to akarma-synetal/objectui
that referenced
this pull request
Sep 28, 2026
…t the protocol (objectui#7924) (objectstack-ai#9720) Part of objectstack-ai#7924 — the card is NOT settled by this PR; see "What this does not do". Clause-②: no — no declaration face moves. `packages/types/src/objectql.ts`, every `*.zod.ts` and every renderer read set are untouched. ## Why The liveness census (PR objectstack-ai#8933) answers one question per member: *is it declared, and is it read off a named view?* The disposition this card is waiting on turns on a **second** question the census never pinned: **does `@objectstack/spec` declare the member?** That is what separates "the renderer sources it from the wrong place" from "objectui invented a spelling", and the 2026-09-16 director-seat ruling is written entirely on that axis — as **prose, with no assertion that can fail**. That is the exact failure mode this card has already produced twice: `about 52 / seven / ~45` → `47 / six / 41` → **stale again** at `64 / 21 / 43` since objectui#8980. The protocol-side partition is today in the same state the member count was in then: measured once, on a **branch** revision (`d4733f27e4`) in a shallow checkout with a stale `origin/main`, and carried forward in prose. The sweep said *"Re-sample before dispatching bucket ②"*; the ruling repeated it. This PR re-samples it **and makes it an assertion**. ## Figures, each with its instrument — all re-derived on `78a9c6744`, none inherited | figure | value | instrument | | --- | --- | --- | | `NamedListView` declared members | **64** | TypeScript parser (`ts.createSourceFile`, the interface's own `PropertySignature` members). No heritage clause, no index signature, no computed member — so the property signatures ARE the population. One required member: `label`. | | read off a named view | **21** | the pin's AST derivation, every route from `schema.listViews` to a named view classified | | declared and unread | **43** | set difference of the two above | | protocol keys for a named list view | **50** | `ObjectListViewSchema.shape` off `@objectstack/spec@17.4.0` as installed — read off the schema object, not a hand list | | protocol retirement tombstones | **5** | derived by the slot description's own `[REMOVED]` marker, then each refused BY NAME through a real `safeParse` | | live protocol keys | **45** | 50 minus the 5 tombstones | | live protocol keys objectui does NOT declare | **0** | set difference over those two corpora | | objectui-only members | **19** | the same set difference, other direction — the control that the 0 above is a reading | | unread partition | **23 + 8 + 10 + 2 = 43** | derived per member from the protocol key set and the runtime fold | ### Disagreement with the figures the card body carries The card's corrected block says **47 declared / six read / 41 unread**. On today's head those read **64 / 21 / 43**. This is not a new correction by me: objectui#8980 (PR objectstack-ai#9534) declared the seventeen protocol members objectui was missing, and the census pin, `objectql.zod.ts`, `zod-mirror-parity.test.ts` and `.changeset/object-view-unmirrored-keys-7779.md` were all updated with it. **The card body and its title were not.** Not repaired here — a card body is the PM/filing seat's to edit. ## What lands A new `describe` block in the existing census pin (extended, not duplicated — same corpus, same helpers), with every figure **derived at test time** and the literals as the pinned reading: - **the protocol declares 50 keys, 5 of them tombstones** — each refused by name with the protocol's own `[REMOVED]` prescription, with the accept leg (`{ label, columns }` parses) as the control. So objectui not declaring those five is **agreement, not narrowness**; a raw key-count diff reads it the other way and is wrong to. - **objectui declares all 45 live protocol keys** — the narrower-than-protocol direction is **empty today**. It was **seventeen** when objectui#8979 was filed. The control for that zero is the same set difference over the same two key sets in the other direction: **19** objectui-only members. - **exactly one objectui-only member is read** (`options`) — the whole cost of a by-reference mirror in one name, with the strict protocol value's refusal of it measured. - **`.omit().extend().superRefine()` still refuses unknown keys under zod 4** — executed with `safeParse` plus an accept control. This was carried as NOT MEASURED by the 2026-09-10 sweep and by the triage that answered it; if it ever flips, bucket ③'s "the protocol refuses these" weakens to "does not declare these". - **the 43 unread members partition four ways, disjoint and exhaustive**, each member pinned by name: ① 23 protocol-declared, ② 8 legacy `show*` spellings, ③ 10 objectui-only, plus the 2 that objectui#8980 declared inert by ruling. A member changing bucket now fails **by name** instead of staling a prose list. - **the fold is read off the fold**: `SHOW_FLAG_TO_USER_ACTION` is parsed out of `packages/core/src/utils/normalize-list-view.ts` and each of the seven is asserted against its canonical protocol twin inside `userActions`; the eighth (`showDescription` to `appearance.showDescription`) by its own branch. Control: a flag the fold does not know is absent from the map on the same query. ## Ablation — both legs, mutation proven on disk BEFORE any result was read The subject reaches the assertions by `readFileSync` of the source (plus the installed schema object), so there is no `dist` hop to prove; the proof is the on-disk text count plus the blob hash. **Leg A — the fold derivation** (`packages/core/src/utils/normalize-list-view.ts`, HEAD blob `59dc77b6`): removed-text count 1 to 0, injected 0 to 1, blob `59dc77b6` to `1567eae0`. Result: `Tests 1 failed | 213 passed (214)`, and the **only** red row is *"the runtime fold maps seven of the eight legacy spellings…"*. Restored with `git checkout HEAD -- ABSOLUTE_PATH` from a `trap … EXIT INT TERM`; restoration proven by blob equality back to `59dc77b6` **and** an empty `git diff HEAD` for the path. **Leg B — the declaration and the protocol difference** (`packages/types/src/objectql.ts`, HEAD blob `033c24f9`): one bucket ① member renamed (`navigation` to `navigationZZZ`), counts 1 to 0 and 0 to 1, blob `033c24f9` to `147d6bbd`. Result: `Tests 5 failed | 209 passed (214)`, red rows exactly: the declared-set pin, the 21+43=64 partition, `UNREAD — navigation`, **`objectui declares every LIVE protocol key … is EMPTY today`**, and **`BUCKET ① — navigation`**. So the zero in this PR's headline row is a live query, not a dead one. Restored and proven the same way; blob back to `033c24f9`, `git diff HEAD` empty. Neither ablation file is edited by this PR. ## Gates — exit code captured to a file before any pipe | gate | exit | printed verdict | | --- | --- | --- | | `pnpm exec vitest run packages/types/src/__tests__/object-view-unmirrored-keys-7779.test.ts` | 0 | `Test Files 1 passed (1)` · `Tests 214 passed (214)` (164 before) | | `pnpm --filter @object-ui/types test` | 0 | `Test Files 202 passed (202)` · `Tests 4748 passed (4748)` | | `pnpm --filter @object-ui/types run type-check` | 0 | all three `tsc` invocations echoed, `tsconfig.test.json` among them, so the test file IS typechecked | | `eslint --no-inline-config` on the changed test file | 0 | `--format json`: 1 file linted, 0 errors, 0 warnings | | `check:control-bytes` | 0 | | | `check-changeset-presence` | 0 | *"Every one of them has an EMPTY frontmatter — declared as releasing nothing, which is the explicit exemption"* | | `check-changeset-no-major` | 0 | `No changeset declares a major bump.` | | `check-changeset-fixed` | 0 | `All workspace packages are in the changeset fixed group.` | | `check-changeset-overwrite` | 0 | **report-only finding, named on purpose — see below** | | `check:changeset-claims` · `check:pending-changeset-literals` | 0 | | | `check:new-line-citations` | 0 | this PR adds no cross-file line address | | `check:test-path-roots` · `check:installed-pin-claims` | 0 | | | `check-governed-queue-guard --test` on the changed test path | 0 | `NOT GOVERNED — 1 path(s) checked against 5 governed surface(s); none matched.` | Every heavy run went through the shared verify lock (`OS_VERIFY_LOCK_SLOT=dev-7924`), each printing its own `VERDICT command-exit` line. **Declared narrowing of the lint run.** Repo-wide `pnpm lint` is `turbo run lint` and belongs to CI; locally only the changed file was linted. The narrowing is measurable: the file count is read off eslint's own `--format json` output (1), and `eslint.config.js` configures **no** type-aware linting (no `project` / `projectService`), so this diff cannot move the verdict for any untouched file. Figures taken at `4794f4786`, the final commit. ## The in-place repair, named with its evidence `.changeset/7924-named-list-view-liveness-census.md` — the census's own, still unreleased — carried the retired `47 declared / 6 read / 41 unread`. The other three sites that carry the pair were re-taken at objectui#8980; this was the fourth and last, and it would have shipped those figures into the release notes. Repaired in the same defect class as this card. **Frontmatter untouched** (empty before, empty after), so no package's release changes; `check-changeset-overwrite` reports the edit (`declared at base: nothing` / `declares now: nothing`) and it is that gate's own documented case 2, *"CORRECTING a declaration on purpose — prose that no longer matches the change"*. ## What this does not do - **No member is retired and no member is declared.** Buckets ② and ③ of the 2026-09-16 ruling retire 18 members from a published TS face — that narrows a published accept set (`Clause-②: yes`) and is outside this dispatch's remit. This PR only makes the bucket each of those 18 sits in a **derived, failing** assertion, which is the input that PR needs and the thing that was still prose. - **Bucket ① is not implemented.** Making the delegation read the 23 protocol-declared members off the named view is capability growth, on `packages/plugin-view/src/ObjectView.tsx`. - **The card is not closed** and its body's figures are not edited. ## 维护者速读(草稿) - **改了什么**:只加测试。把「这 43 个成员分别属于哪一类处置」从评论里的文字,变成每次跑测试都会重新推导的断言 —— 推导来源是安装着的 `@objectstack/spec`、声明文件自己的 AST、以及运行时的折叠表。 - **为什么改**:这张卡的数字已经错过两轮(52 → 47 → 今天的 64)。裁决所依据的「协议有没有声明这个键」这一面,至今只在一条分支上量过一次,从没有任何断言守着。现在有了:某个成员换了桶,测试按名字红。 - **风险与代价(含回滚)**:不动任何已发布面,风险仅限于测试变严;回滚就是 revert 这个 PR。 - **席位意见**:(留空) - **你要做的**:不需要做什么。若要推进退休(八个 `show*` 拼法 + 十个本地键),那是另一张 `Clause-②: yes` 的 PR。 --- _Generated by [Claude Code](https://claude.ai/code/session_01UanLVj6xvbS6puBCewLr8L)_ --------- Co-authored-by: Claude <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Refs #7924 — this PR lands the census only and decides nothing. The two dispositions the card names for the unread members (
?: nevertombstones per the objectui#7129 route, or making the delegation read them from the named view) are maintainer rulings and stay with objectui#7928. ⛔ No declaration face moved:packages/types/src/objectql.ts, every*.zod.tsand every renderer read set are byte-identical tomainin this diff, which touches one test file and one changeset.Why none of the card's figures are quoted
The card body says "about 52 members" / "exactly seven read" / "~45", measured on
6a9ee323. Every figure below is re-derived on this branch's head (3b5053d45). The reading is 47 declared, 6 declared-and-read, 41 unread, plus one read-but-undeclared key. "About 52" reproduces only as a hand figure sitting between two instruments and is neither of them — the instrument table below pins that gap as an assertion rather than a sentence.The census
NamedListView(packages/types/src/objectql.ts:2025–:2202today — re-located by AST, not by the card's stale line numbers) declares 47 top-level members. Theobject-viewnode renderer reads 7 names off a named view; 6 of them are declared members.activeView?.KEY— the host'sviewsprop — is not read off the named view. It is included because it is what the ruling needs: it separates "unread here" from "unread anywhere".activeViewlabelactiveView?.labeltypeactiveView?.typecolumnsactiveView?.columnsfilteractiveView?.filtersortactiveView?.sortoptionsshowSearchactiveView?.showSearchshowSortactiveView?.showSortshowFiltersactiveView?.showFiltersshowHideFieldsshowGroupshowColorshowDensitycompactToolbaractiveView?.compactToolbarallowExportactiveView?.allowExportcoloractiveView?.colorinlineEditactiveView?.inlineEditwrapHeadersactiveView?.wrapHeadersclickIntoRecordDetailsactiveView?.clickIntoRecordDetailsaddRecordViaFormactiveView?.addRecordViaFormaddDeleteRecordsInlineactiveView?.addDeleteRecordsInlinecollapseAllByDefaultactiveView?.collapseAllByDefaultfieldTextColoractiveView?.fieldTextColorprefixFieldactiveView?.prefixFielddescriptionshowDescriptionactiveView?.showDescriptionnavigationactiveView?.navigationselectionactiveView?.selectionpaginationactiveView?.paginationsearchableFieldsactiveView?.searchableFieldsfilterableFieldsactiveView?.filterableFieldsresizableactiveView?.resizabledensityModeactiveView?.densityModerowHeightactiveView?.rowHeighthiddenFieldsactiveView?.hiddenFieldsexportOptionsrowActionsactiveView?.rowActionsbulkActionsactiveView?.bulkActionsbulkActionDefssharingactiveView?.sharingaddRecordactiveView?.addRecordconditionalFormattingactiveView?.conditionalFormattinguserFiltersactiveView?.userFiltersshowRecordCountactiveView?.showRecordCountallowPrintingactiveView?.allowPrintingemptyStateactiveView?.emptyStateariaactiveView?.ariadataas anycastactiveView?.dataTotals: 47 declared = 6 read + 41 unread. Read set off the named view = 7 names (the 6 above plus the undeclared
data)."fold only" means the member has no⚠️ Worth stating because a named-property probe alone reports those four as having no reader at all, which is false — and that mistake would have gone straight into the ruling.
activeView?.KEYaccess, but IS consumed when it appears onactiveView, throughnormalizeListViewSchema(activeView)(packages/core/src/utils/normalize-list-view.ts, itsSHOW_FLAG_TO_USER_ACTIONmap).Three readings the ruling request should carry
datais still read-but-undeclared. It reaches the renderer asdata: (currentNamedViewConfig as any)?.data ?? …. objectui#7928 requires it to be declared or its cast removed and names this card as its home; it is still neither. It is why the arithmetic subtracts six, not seven, from 47.activeView; 4 more are folded fromactiveView; 3 have no reader on any path —description,exportOptions,bulkActionDefs. Of those,exportOptionsandbulkActionDefsdo not appear in the renderer at all (pinned), anddescriptionappears only offschema/schema.table/schema.form, never off a view.labelis the only required member, and it is read — so a tombstone route would leave the sole required member intact.Instruments, and the control that makes each zero a reading
namedListViewMembers()(new)ts.createSourceFile, the interface's ownPropertySignaturemembersnamedListViewMemberCount()(kept)namedListViewLooseMemberCount()(new)⛔ Not a brace-depth count: a brace parser on this repo has already failed to terminate at the end of an object literal (objectui#8071). The three are now pinned against each other, so the instrument swap is itself a measurement and "52 is between two instruments and is neither" stops being prose.
The read set is derived by walking every route from
schema.listViewsto a named view — a binding annotatedNamedListView; the value binding ofObject.entries(record).map(...)(through an intermediateconst entries = …too); and element access straight off the record — stripping( ),!andas anyat each access, which is the only reasondatais visible at all.⭐ That walk found a named-view read the old
currentNamedViewConfig?.KEYregex cannot see:{view.label || key}on the named-view tab strip.labelis read at two sites, and the count is asserted.Completeness, not just coverage: every occurrence of the record binding is classified into a role (
DECLARATION,ELEMENT_ACCESS,OBJECT_KEYS,OBJECT_ENTRIES,COMPARISON,CONDITION,HOOK_DEPENDENCY). A new syntactic route lands asUNCLASSIFIEDand fails before the census can silently under-count — the test says in its own message not to add the role to make it green.Firing controls (a silent probe is a parser that found nothing, not a reading of zero):
columns— asserted present in both the declared set and the read set;rowHeightis asserted declared and not read off the named view, withrowHeight: activeView?.rowHeight,pinned as the text that makes it so;stickyHeader, in neither set, is non-vacuity for both probes at once;grepthat finds nothing for them is asserted to findrowHeight.Ablation — the pin is non-vacuous in both directions
Each leg: mutation proven on disk before any result was read (anchor counts on both the removed and the injected text, plus a
git hash-objectdiffering from the HEAD blob), restore from atrap … EXIT INT TERMusinggit checkout HEAD -- ABSOLUTE_PATH, restoration proven by blob hash, never by an exit code. HEAD blob ofObjectView.tsx=c1a807af683a0147796780c801ac34ffa97e8c70.Leg A — unread becomes read.
navigation: activeView?.navigation ?? …rewritten to read the named view first. Removed-text count 1 to 0, injected 0 to 1, blobc1a807aftofdfe41ac. Result: 4 failed / 143 passed, and the only per-member row that moved is the expected one:UNREAD — navigation is declared and NOT read off a named view⇒ red6 read + 41 unread = 47partition ⇒ redLeg B — read becomes unread. The
currentNamedViewConfig?.typeread deleted. Blobc1a807aftoc9e79701. Result: 4 failed / 143 passed:READ — type is declared AND read off a named view⇒ redif (activeView?.type) return activeView.type;) already existed on the next line, so the on-disk proof read injected-count 2, expected 1, and the script aborted withexit 94before running anything and restored the file. The reading was void, not green. Re-anchored on a unique marker and re-run; both blob hashes above are from the run that actually happened.Restoration verified after every leg:
git hash-objectback toc1a807af683a0147796780c801ac34ffa97e8c70andgit diff HEADempty for that path.Readings — commands and exit codes
Exit codes captured into a file before any pipe (
cmd > out 2>&1; RC=$?), never off the end of a pipeline.pnpm exec vitest run packages/types/src/__tests__/object-view-unmirrored-keys-7779.test.tsTest Files 1 passed (1)·Tests 147 passed (147)pnpm --filter @object-ui/types type-checktsc --noEmit,tsconfig.examples.json,tsconfig.test.jsonechoed and green — the test file is typecheckedpnpm exec eslint packages/types/src/__tests__/object-view-unmirrored-keys-7779.test.tsnode scripts/check-changeset-presence.mjsnode scripts/check-changeset-no-major.mjsmajorpnpm check:control-bytespnpm check:phantom-depstypescriptis already a declared devDependency of@object-ui/typesnode scripts/check-governed-queue-guard.mjs --test(both changed paths)NOT GOVERNED— 2 paths, no governed surface matchedAll test runs went through the container's shared heavy-verify lock; each printed
VERDICT command-exit 0.Declared narrowing. Repo-wide
pnpm lintandpnpm testare CI's runs, not this seat's. The lint reading above is narrowed to the one changed file;eslint.config.jsconfigures no type-aware linting (noproject/projectService), so this diff cannot move the verdict on any file it does not touch. Test scope is the single pin file because the diff changes exactly that file plus a changeset, and no published byte moves.Scope
⛔ The three live branches' files were not touched:
packages/types/src/complex.ts/plugin-kanban/plugin-gantt(PR #8865),packages/types/src/layout.ts/page-breadcrumbs-refusal-8871.test.ts(PR #8930),registry-inputs-spec-parity.test.ts/plugin-calendar(PR #8872).The census extends the existing pin rather than adding a file. That file already owned both halves of this measurement (the 47 count and the seven-name read set) and is where the regex instrument being improved on lives; a second file would create a second instrument for the same number, which is the staleness shape objectui#7947 was filed about. The header's prose figures are now backed by the assertions underneath them.
⛔ This PR stays draft — the PM lands it. The card is not closed by it: the disposition is still unruled.
🤖 Generated with Claude Code
https://claude.ai/code/session_01Jmxdo7bmeqCQHLSfmLVX9w
Generated by Claude Code
Generated by Claude Code