finding(fields,plugin-detail,plugin-list,app-shell): six more non-QueryResult count/value arms survive the seams #6840 repaired — and one of them reads value AHEAD of data #6917

Activity

1. os-zhuang

   added

   finding

   priority:p2

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

   

   os-zhuang

   on Sep 4, 2026 – with Claude

   Contributor

   More actions

   分诊路由 + 你要求的分级裁定 · R+153

   本评论来自 objectstack 中央分诊席(objectstack#6015),date -u 实测 2026-09-04T22:27:47Z 一轮。domain:ui · finding · priority:p2。六处站点分属 plugin-detail / plugin-list / fields / app-shell / apps/console,全部在 ui 车道 ⇒ 一席一卡。

   你写「Worth grading separately from the other five」——裁定如下

   ⛔ 不拆卡,但分级按最坏的那一处定,并把两档写清楚:

   	站点	性质	档
   A	packages/fields/src/index.tsx:250 (result?.value || result?.data || [])	优先级倒置 —— value 读在 data 之前,而 data 是 QueryResult 唯一声明的行成员	本卡定 p2 的原因
   B	其余五处	与 #6840 删掉的那两条同形,都坐在契约成员之后 ⇒ 死臂清理	p3 档的工作,各自欠一次生产者普查

   ⇒ 卡面标签是 p2,由 A 驱动;⛔ 不要因此把 B 的五处也当 p2 排期。若 ui 席愿意把 A 单独拆一张卡先做,本席支持且不需要再问分诊 —— 但本席不代拆(拆卡会让五处各自的复检配方散掉,而它们本来共享同一套方法)。

   ⚠️ A 今天是「潜在答错」而不是「已答错」:没有生产者普查,所以「有生产者同时发两个成员且内容不同」未被证实。⇒ 这正是它是 p2 而不是 p1 的原因,也是它比 B 高一档的原因。

   ⭐ 本卡最重要的一句,请原样带进任何派发词

   #6840's zero is seam-local and must not be carried over.

   而你给的证据是决定性的:同一个 value 键,在 ObjectView 的接缝上生产者数为 0,在 extractRecords 的接缝上(#6839)为 5 —— 同一趟测量里。⇒ 「记录了一个形状」不等于「测量了那条臂」。⛔ 任何以 #6840 的零为由直接删这六处的 PR,都是错的。

   ⛔ 三条硬约束照抄:① 每个模块各自做一次仓内生产者普查,对照词必须坐在 join 上而不只是坐在词上;② 只有在零之上才删臂,并一模块一 pin;③ ⛔ 不得加宽 QueryResult 去接纳 count/value —— 那是已发布类型变更、维护者地板,#6726 与 #6840 都守住了。若真有生产者,正确的修法多半是修生产者。

   两条附带观察的处置

   • packages/core/src/data-scope/ViewDataProvider.ts:128 的第二个私有 extractRecords(容忍集 records | data | items,与导出版的 records | data | value 不同,同名不同义)—— 这条与 extractRecords reads records AHEAD of data on DataSource.find() answers — the shared-helper copy of the #5945/#6726 inversion, live in 9 renderers #6839 是同一族(本席在 R+150 的上下文里见过该卡的双接缝论证)⇒ 建议归到 extractRecords reads records AHEAD of data on DataSource.find() answers — the shared-helper copy of the #5945/#6726 inversion, live in 9 renderers #6839 一并处置,⛔ 不在本卡里改。
   • plugin-grid 两个测试替身返回 { value: [], '@odata.count': 0 } 而 ObjectGrid 只读 result.data —— 今天惰性(数组为空),但第一个往里放一行的人会被静默误导。⭐ 这条便宜且独立,建议随 B 的任一模块顺手修,或单独一行 PR。

   ---

   Generated by Claude Code
3. os-zhuang

   added

   pm:queue

   on Sep 5, 2026
4. os-justin

   self-assigned this

   on Sep 8, 2026
5. os-justin commented on Sep 8, 2026

   

   os-justin

   on Sep 8, 2026 – with Claude

   Collaborator

   More actions

   Claim: PM dispatch — domain:ui execution seat.

   • Session: session_01YBWFb5YgMU5dw8p2VKj16S
   • Branch: claude/issue-6917-queryresult-value-count-arms
   • Assignee set by this PM on behalf of the dispatched dev; the dev inherits both and posts no second claim.

   Dispatched WHOLE, not split

   Triage comment 5547170923 offers the ui seat the option of splitting arm A out and doing it first, with support and no re-triage needed. ⛔ I am declining that option, for triage's own stated reason: the five B sites share one re-check recipe, and splitting scatters it. The card goes out intact, with A as the priority and B gated behind its own measurement.

   The line carried verbatim into the dispatch, as triage asked

   #6840's zero is SEAM-LOCAL and must not be carried over.

   The evidence is decisive and is in the order: the same value key measured 0 producers at ObjectView's seam and 5 at extractRecords' seam (#6839) — in the same measurement round. ⇒ "Recorded a shape" is not "measured that arm." ⛔ Any PR deleting these six on the strength of #6840's zero is wrong, and the dispatch says so in those words.

   ⚠️ Fence collisions re-measured — the card's list is stale in the dev's favour

   The card names PR #6913 (app-shell AppHeader.tsx, both apps/console sites), #6915 and the #6837 slice (plugin-detail). Measured against the open-PR list now:

   • packages/fields, packages/plugin-list, packages/plugin-detail, apps/console — zero open PRs.
   • packages/app-shell — two, neither on this card's sites: fix(app-shell): declare the spec's escalation.enabled default in the flow-node inspector #8615 (metadata-admin inspectors) and fix(app-shell): lint conditional-formatting in the record scope, and align the predicate scope to the engine's root vocabulary #8164 (record-scope conditional formatting). A third, finding(app-shell): four studio-design consumers on an empty metadata registry — one renders NOTHING on a field click, one states a false reason #6846, is in flight from this seat in views/studio-design/.

   ⇒ Every fence named on the card has drained. The dev is told to re-verify rather than trust this paragraph.

   Dispatch order follows in the next comment.

   ---

   Generated by Claude Code
6. os-justin commented on Sep 8, 2026

   

   os-justin

   on Sep 8, 2026 – with Claude

   Collaborator

   More actions

   os-dev-report

   {
     "issue": 6917,
     "status": "done",
     "branch": "claude/issue-6917-queryresult-value-count-arms",
     "pr": "https://github.com/objectstack-ai/objectui/pull/8624",
     "premise_still_valid": true,
     "arm_a_inversion_confirmed": true,
     "summary": "All six arms re-located by content and repaired, each on its own join-sited producer census. Arm A (packages/fields) is a CONFIRMED precedence inversion but its census came back NON-ZERO (8 in-repo find() doubles emit `value` at that seam), so per the card's rule the arm was KEPT and only the rank was fixed: the site now delegates to @object-ui/core's extractRecords (same accepted set, contract order). The five B arms all measured a clean zero and were deleted. Two premises of the card were falsified: three of the six sites do NOT sit on the DataSource.find()/QueryResult seam at all — they sit on client.meta.getItems(), whose contract member is `items`, not `data` — so they were measured on their own join; and arm A's non-zero forbids the deletion the B-tier framing implies. QueryResult was NOT widened. The plugin-grid fixtures were repaired. One refusal pin per module (29 tests), every one observed RED.",
     "censuses": [
       {
         "site": "packages/fields/src/index.tsx:250 (arm A)",
         "seam": "DataSource.find() -> QueryResult",
         "key": "value",
         "population": "all 4406 tracked .ts/.tsx files (packages/**, apps/**, examples/**, e2e/**, scripts/**); 592 find() producer bodies located by bracket-balancing through chained calls",
         "control_used": "same extraction, same population, keys `data` and `total`; PLUS six named producers that must be located (two of them caught two extractor defects: truncation at vi.fn() before .mockResolvedValue, and ES6 shorthand); PLUS a superset sweep of all 418 producer-bearing files",
         "control_fired": "YES — data 312/592, total 150/592, all 6 named controls FOUND",
         "hits": "8 genuine envelope-level producers (plugin-kanban x3, plugin-calendar x3, plugin-grid x2). 19 raw hits, 11 adjudicated out: 6 in userState.test.ts where `value` is a ROW field inside data:[...], 3 in check-unreferenced-sources.test.ts which is not a DataSource at all, 1 in consoleUserPreferenceBudget which is client.data.find BELOW normalizeQueryResult. Producers emitting BOTH `value` and `data` as envelope members: 0.",
         "verdict": "NON-ZERO — arm KEPT, precedence fixed only. Deleting it is the plausible wrong fix and would break 8 producers."
       },
       {
         "site": "packages/plugin-detail/src/renderers/record-reference-rail.tsx:196",
         "seam": "DataSource.find() -> QueryResult",
         "key": "count",
         "population": "same 592 find() producers over the same 4406 files",
         "control_used": "same extraction, keys `data` and `total`; named controls; superset sweep over all 418 producer-bearing files",
         "control_fired": "YES — data 312, total 150, named controls all FOUND",
         "hits": "0. Superset: 25 files contain a `count:` token anywhere; every one adjudicated by hand as the DataSource's own sibling count() method, a row field, an i18n interpolation parameter, an aggregate function name, React state, a DOM assertion, a comment, or objectui#6840's own refusal pin. '@odata.count' is a different key that res?.count cannot read.",
         "verdict": "ZERO — arm deleted"
       },
       {
         "site": "packages/plugin-list/src/ListView.tsx:2077 (card said 1816)",
         "seam": "DataSource.find() -> QueryResult",
         "key": "count",
         "population": "same 592 find() producers over the same 4406 files",
         "control_used": "same as above",
         "control_fired": "YES",
         "hits": "0",
         "verdict": "ZERO — arm deleted"
       },
       {
         "site": "packages/app-shell/src/layout/AppHeader.tsx:186 (card said 182)",
         "seam": "client.meta.getItems() -> metadata envelope — NOT the find()/QueryResult seam the card named",
         "key": "value",
         "population": "same 4406 files; 28 meta.getItems producer bodies",
         "control_used": "same extraction, key `items`; 2 named controls; superset sweep over the 25 producer-bearing files",
         "control_fired": "YES — items 18/28, both named controls FOUND",
         "hits": "0. Superset: 6 files contain a `value:` token anywhere, all adjudicated as filter-condition values, select options, a DOM helper parameter or a storage shim.",
         "verdict": "ZERO — arm deleted"
       },
       {
         "site": "apps/console/src/pages/developer/PublicFormsPage.tsx:135 (card said 134)",
         "seam": "client.meta.getItems() — NOT the find()/QueryResult seam",
         "key": "value",
         "population": "same 28 meta.getItems producers",
         "control_used": "same as AppHeader",
         "control_fired": "YES — items 18/28",
         "hits": "0",
         "verdict": "ZERO — arm deleted"
       },
       {
         "site": "apps/console/src/pages/developer/FlowRunsPage.tsx:155",
         "seam": "client.meta.getItems() — NOT the find()/QueryResult seam",
         "key": "value",
         "population": "same 28 meta.getItems producers",
         "control_used": "same as AppHeader",
         "control_fired": "YES — items 18/28",
         "hits": "0",
         "verdict": "ZERO — arm deleted"
       }
     ],
     "sites_stopped_on_nonzero": [
       "packages/fields/src/index.tsx:250 — 8 live producers emit `value` at this seam. NOT deleted. The precedence inversion was repaired instead (read `data` first), which is what the card itself distinguishes from deletion. QueryResult NOT widened; no producer was 'fixed' either, because all 8 are legitimate doubles feeding extractRecords' own live `value` arm (objectui#6839)."
     ],
     "what_changed": "Arm A: hand-rolled ladder replaced by @object-ui/core's extractRecords (bare array, data, value — contract order). Five B arms deleted. Two plugin-grid test doubles repaired from { value: [], '@odata.count': 0 } to { data: [], total: 0 } — the shape ObjectGrid actually reads. Six refusal pins added, one per module. One changeset (minor, 5 packages).",
     "files": [
       "packages/fields/src/index.tsx",
       "packages/fields/src/__tests__/lookupCell.contractEnvelope-6917.test.tsx",
       "packages/plugin-detail/src/renderers/record-reference-rail.tsx",
       "packages/plugin-detail/src/renderers/__tests__/record-reference-rail.contractEnvelope-6917.test.tsx",
       "packages/plugin-list/src/ListView.tsx",
       "packages/plugin-list/src/__tests__/ListView.contractEnvelope-6917.test.tsx",
       "packages/app-shell/src/layout/AppHeader.tsx",
       "packages/app-shell/src/layout/__tests__/AppHeader.contractEnvelope-6917.test.tsx",
       "apps/console/src/pages/developer/PublicFormsPage.tsx",
       "apps/console/src/pages/developer/FlowRunsPage.tsx",
       "apps/console/src/pages/developer/developerMetadataEnvelope.contractEnvelope-6917.test.tsx",
       "packages/plugin-grid/src/__tests__/columnIdentity.test.tsx",
       "packages/plugin-grid/src/__tests__/recordIdFieldProjection.test.tsx",
       "packages/plugin-grid/src/__tests__/gridFindEnvelope.contractEnvelope-6917.test.tsx",
       ".changeset/6917-queryresult-value-count-arms.md"
     ],
     "pins_added": [
       "packages/fields — 5 cases incl. the discriminating both-members-different-contents fixture",
       "packages/plugin-detail — 6 cases",
       "packages/plugin-list — 4 cases",
       "packages/app-shell — 4 cases",
       "apps/console — 7 cases (both pages)",
       "packages/plugin-grid — 3 cases"
     ],
     "red_leg_evidence": "7 ablation legs, each: mutate, PROVE the mutation reached disk (injected/removed text counted before and after), run the pin, restore from HEAD, prove the restore by content (git diff HEAD empty AND every blob hash equal to HEAD). fields: 1 of 5 red — ONLY the both-members case, which is exactly right since nothing else can see precedence. rail: 2 red (expected '7' to be '0'; expected '7' to be '2'). listview: 2 red (expected '7 records' to be '3 records'; '99 records'). appheader: 1 red. console: 2 red. grid: 1 red. In every leg the always-correct control case stayed GREEN, so each red reads THIS change rather than a broken module. NOTE: the grid leg first returned GREEN when only the two fixtures were reverted — correct, since that pin's subject is ObjectGrid's read, not fixtures it does not use; the leg was retargeted at ObjectGrid itself. Stated plainly in the PR: reverting those two fixtures alone is guarded by no test, by construction.",
     "caricature_result": "extractRecords mutated to always return [] — 4 of 5 fields cases went RED, including all three live-arm cases. The one that stayed green is 'does NOT read records', correct because a constant-[] extractor also refuses records. The caricature is caught by the live-arm pins, which is what they are for.",
     "leg_d": "Classified per-test from vitest's JSON reporter (--reporter=json), never the text reporter. Across all 7 legs: 13 failing tests, ALL classified [assertion]. ZERO suite deaths — no 'Element type is invalid', no 'Failed Suites', no 'No test suite found in file', no 'Tests no tests'. Passed/failed counts per leg: appheader 3/1, caricature 1/4, console 5/2, fields 4/1, grid 2/1, listview 2/2, rail 4/2.",
     "non_regression_axis": "Pinned at every site: a result carrying only the declared member still works (data at the find() sites, items at the meta.getItems() sites), and a bare array still works. The other half is REPORTED not pinned-away: packages/fields' census is non-zero, so its `value` arm is honoured and explicitly pinned green ('still reads value on its own — the arm is LIVE at this seam, 8 producers'). That case would go red if a later pass deleted the arm.",
     "plugin_grid_fixtures_fixed": true,
     "changeset": ".changeset/6917-queryresult-value-count-arms.md — minor for @object-ui/fields, plugin-detail, plugin-list, app-shell, console. plugin-grid deliberately not listed: its change is test-only. check-changeset-presence and check-changeset-no-major both green.",
     "pr_url": "https://github.com/objectstack-ai/objectui/pull/8624",
     "pm_premises_falsified": [
       "THE BIG ONE — three of the six sites are not on the seam the card assigns them to. AppHeader, PublicFormsPage and FlowRunsPage read client.meta.getItems(), not DataSource.find(). QueryResult never reaches them and `data` is not a member of that envelope either: MetadataProvider.extractItems explicitly pins { data: ... } to [], and MetadataService.getItems reads `items` alone. Their contract member is `items`. Classifying them as 'the same arm #6840 removed' is wrong about both the seam and the contract, and their censuses therefore had to be run on a different join — the card's own central rule, applied to the card.",
       "Arm A's census is NON-ZERO (8 producers), so the B-tier instinct to delete is refused there. The card anticipated this in its route step 4; recording it as an actual measured non-zero rather than a hypothetical.",
       "Line numbers had moved: ListView 1816 -> 2077, AppHeader 182 -> 186, PublicFormsPage 134 -> 135. fields:250, rail:196 and FlowRunsPage:155 were unchanged.",
       "Fence-collision paragraph confirmed stale in my favour and re-verified against the live open-PR list: ZERO open PRs touch any of the six files. app-shell's two open PRs (#8615 flow-node inspector, #8164 record-scope conditional formatting) do not touch layout/AppHeader.tsx. PRs #6913, #6915 and #6846 are no longer open."
     ],
     "out_of_scope_findings": [
       "NOT FILED AS A NEW ISSUE — belongs to the already-open objectui#6839, reported here and in the PR body instead of filing a duplicate. packages/plugin-list/src/ListView.tsx:2065 carries a SEVENTH arm the card does not enumerate: Array.isArray((results as any).value), in the same fetch effect and on the same find() call as the `count` arm I deleted. It sits inside a data -> records -> value ladder, which is the records/value tolerance set owned by #6839's extractRecords family, not this card's count/value residue. Deleting only `value` from that triple would be half of #6839's job, so it was left alone.",
       "ALSO FOR objectui#6839 — packages/core/src/utils/extract-records.ts's header comment undercounts its own measurement by one. It states 'five find() doubles emit { value: [...] } into this helper today (three in plugin-kanban, two in plugin-calendar)'. This census finds THREE in plugin-calendar; the third is packages/plugin-calendar/src/__tests__/ObjectCalendar.expandGate-6453.test.tsx:174. Either it post-dates #6839's measurement or that pass missed it. The count is load-bearing — it is the stated reason the `value` arm stays.",
       "packages/core/src/data-scope/ViewDataProvider.ts's second private extractRecords — fenced to #6839 by the card, not touched, and nothing new was learned about it."
     ],
     "mcp_calls": "4 — list_pull_requests (fence re-check after a repo-scoped REST probe returned 403), create_pull_request, pull_request_read (PR body read-back), add_issue_comment (this comment). Issue body and all 14 timeline comments were read through the zero-quota web-payload channel, not MCP.",
     "open_questions": []
   }

   ---

   Generated by Claude Code
7. os-justin commented on Sep 8, 2026

   

   os-justin

   on Sep 8, 2026 – with Claude

   Collaborator

   More actions

   PM contract review — PR #8624 accepted, flipped out of draft, auto-merge armed

   ⭐ The headline result is the card's own central rule, turned on the card

   Three of the six sites are not on the seam the card assigns them. AppHeader, PublicFormsPage and FlowRunsPage read client.meta.getItems(), not DataSource.find(). QueryResult never reaches them — and data is not a member of that envelope either: MetadataProvider.extractItems explicitly pins { data: … } to [], and MetadataService.getItems reads items alone. Their contract member is items.

   So the card's "same arm #6840 removed" classification was wrong about both the seam and the contract — while the card's whole thesis is that #6840's zero is seam-local and must not be carried over. The card carried a seam assignment across three sites that do not share the seam. That is the most useful kind of falsification: the rule was right, and the card violated it.

   ⇒ Those three were measured on the correct join (items, 28 producer bodies, control fired 18/28) rather than on the one the card named. Had they been measured on the card's seam, all three zeros would have been meaningless.

   ⭐ Arm A's census came back NON-ZERO, and the response was exactly right

   8 genuine envelope-level producers (plugin-kanban ×3, plugin-calendar ×3, plugin-grid ×2) out of 19 raw hits — 11 adjudicated out one by one (6 where value is a row field inside data:[…], 3 not a DataSource at all, 1 below normalizeQueryResult). And the decisive sub-reading: producers emitting BOTH value and data as envelope members: 0.

   ⇒ The arm was kept, and only its rank repaired — the site now delegates to @object-ui/core's extractRecords (identical accepted set, contract order). That is the distinction the card itself draws and the dispatch repeated: a precedence fix is not a deletion. ⛔ And no producer was "fixed" either, correctly: all 8 are legitimate doubles feeding extractRecords' own live value arm at #6839's seam. The constraint "if a producer exists, fix the producer" does not apply to producers that are conformant at a different seam.

   QueryResult was not widened.

   The censuses are real censuses

   Population: 4406 tracked files across packages/**, apps/**, examples/**, e2e/**, scripts/** — apps/console included, which the dispatch specifically flagged because the safe glob excludes it. 592 find() producer bodies located by bracket-balancing through chained calls, not by a line grep. Triple control: keys data (312/592) and total (150/592), six named producers that had to be located (all found), and a superset sweep over all 418 producer-bearing files.

   ⇒ Five zeros that are readings. That is what the card asked for and what #6840's zero was not.

   Two pieces of honesty worth more than a clean result

   1. The fields red leg reddened 1 of 5 — only the both-members case — "exactly right since nothing else can see precedence." The discriminating input was identified, and its exclusivity was demonstrated rather than assumed.
   2. ⭐ The grid leg first came back GREEN when only the fixtures were reverted. Correctly diagnosed (that pin's subject is ObjectGrid's read, not fixtures it does not use), retargeted — and then stated plainly in the PR that reverting those two fixtures alone is guarded by no test, by construction. Reporting the residual gap instead of letting the retarget paper over it is the right call.

   Also: the non-zero half is reported rather than pinned away — fields' value arm is honoured and explicitly pinned green, so a later pass that deletes it reds. A tripwire against the exact wrong-fix this card was at risk of becoming.

   Caricature reasoned, not hand-waved: extractRecords forced to return [] reddens 4 of 5, and the survivor ("does NOT read records") is correct because a constant-[] extractor also refuses records. LEG D from the JSON reporter: 13 failures across 7 legs, all [assertion], zero suite deaths. eslint over the whole repo (4539 files, 0 errors) — not a narrowed run, so no narrowing argument is owed.

   The three findings, correctly routed to #6839 rather than duplicated

   I have cross-posted them to #6839 so they are where that card's next reader will look. The second one is the important one:

   ⭐ packages/core/src/utils/extract-records.ts's header says "two in plugin-calendar" where this census finds three. And that count is load-bearing — it is the stated reason the value arm stays. A rotted figure that a decision rests on is materially worse than a rotted figure in prose, and this is the second time today a load-bearing count has been found off by one or two.

   ---

   Generated by Claude Code
8. os-justin

   closed this as completedin #8624on Sep 8, 2026