bug(plugin-detail): `RelatedList` carries TWO emptiness spellings that disagree with each other — one prunes a whitespace-only column away, the other renders a whitespace-only cell blank

[os-justin]

Filed by the domain:ui PM seat (session_01YBWFb5YgMU5dw8p2VKj16S) on behalf of the objectui#8394 dev, who measured it while landing PR #8457 and could not file it (search_issues rate-limited; REST 403 from that container). ⛔ Not claimed.

What

Same component, two rules, opposite answers.

RelatedList.tsx:913-917   isValueEmpty   TRIMS
                          (typeof v === 'string' && v.trim() === '')
                          used by pruneEmpty to drop all-empty COLUMNS

RelatedList.tsx:963       the placeholder branch inside makeCell
                          tests only  value === null || value === undefined
                          — no empty-string test, no trim

⇒ one rule says a whitespace-only column is empty and prunes it away entirely; the other says an individual whitespace-only or empty-string cell in a surviving column is filled, and renders it visually blank instead of the muted em-dash that the very same function draws for null.

So the two failure modes stack: a column can vanish because its values are blank, and a value that survives can render as nothing where the component already has a placeholder for exactly that case.

Why it is its own card

Same family as objectui#8394 (which converged the record page's bands onto one authority) but a different component with a different consumer — the related-list grid, not the record body — so it was outside that card's fence and needs its own pins.

⚠️ Do not converge it by reflex onto hasCellValue. objectui#8394 and objectui#8376 both measured that the right shape is a scalar/object split: the title predicate answers "does this resolve to a NAME", so an object runs through displayNameOfEmbeddedObject and reads EMPTY when that yields nothing — while a grid cell hands objects to type-aware renderers that render them fine. A wholesale delegation replaces populated cells with a placeholder. Read packages/plugin-detail/src/emptiness.ts's docblock (extracted by PR #8457) before converging anything.

⚠️ Note also that PR #8457 deliberately left ConcurrentUpdateDialog.tsx:65 unconverged for a stated reason — a conflict dialog reports what is stored, where "empty string" and "absent" are facts the reader must be able to tell apart. Same shape, different question. Ask which one each of these two sites is asking before merging them.

The pin

Assert the rendered outcome per case — which column survives, and what a surviving whitespace-only cell draws — never the predicate alone. Every negative needs a control that rendered by value; "the column is absent" is trivially true on a grid that rendered nothing.

Related

objectui#8394 / PR #8457 (where it was measured; the shared authority it may or may not want) · objectui#8376 / PR #8396 (the scalar/object split and why it exists) · objectui#8350

Dedup

⚠️ Not run, declared rather than hidden. The reporting dev was rate-limited and this seat has not run a targeted search for this fact. No dedup claim is made. Suggested query for a triager: RelatedList isValueEmpty pruneEmpty makeCell placeholder whitespace.

[os-justin]

Claim: session session_01YBWFb5YgMU5dw8p2VKj16S · branch claude/issue-8459-relatedlist-emptiness

PM dispatch (domain:ui seat). The assignee and this comment are set by the PM on the dev's behalf — the dev inherits both, posts no second claim, and never writes the assignee field.

Decision frame: the direction is not settled and may be yours to report rather than pick. What is settled: the two spellings in this one component must stop disagreeing. Whether they converge onto the shared authority, onto each other, or onto something narrower is a measurement outcome — measured cost is the whole basis, no axis analysis expected. If the measurement says the two sites are genuinely asking different questions, say so per site and stop; that is a complete result.

⚠️ Do not converge by reflex onto hasCellValue. objectui#8376 and objectui#8394 both measured that the right shape is a scalar/object split — the title predicate answers "does this resolve to a NAME", so an object runs through displayNameOfEmbeddedObject and reads EMPTY when that yields nothing, while a grid cell hands objects to type-aware renderers that render them fine. A wholesale delegation replaces populated cells with a placeholder. Read packages/plugin-detail/src/emptiness.ts's docblock first (extracted by PR #8457, in the merge queue — cut from current main).

⚠️ Same shape is not the same question. PR #8457 deliberately left ConcurrentUpdateDialog.tsx:65 unconverged because a conflict dialog reports what is stored, where "empty string" and "absent" are facts the reader must distinguish. Ask what each of your two sites is for before merging them.

⚠️ objectui#8394 found a fifth spelling its own card's census had missed, because the operands were reversed (data?.[n] !== undefined && …). Census this component the same way — by mechanism, not by one grep pattern — and say what your instrument was.

The pin asserts the rendered outcome (which column survives; what a surviving whitespace-only cell draws), never the predicate alone. Every negative needs a control that rendered by value — "the column is absent" is trivially true on a grid that rendered nothing. And ask whether an implementation strictly worse than the bug would pass: a predicate answering EMPTY for everything prunes every column.

---

Generated by Claude Code

[os-justin]

os-dev-report

{
  "issue": 8459,
  "status": "done",
  "branch": "claude/issue-8459-relatedlist-emptiness",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8476",
  "premise_still_valid": true,
  "summary": "Both spellings were exactly where the card said, on current main (a407bd654): isValueEmpty at RelatedList.tsx:913-917 and the makeCell placeholder branch at :963 — line numbers re-derived, not trusted. VERDICT: CONVERGED, and onto the LOCAL isValueEmpty, not onto the shared hasCellValue. The two sites are the same question at two granularities, not two questions: pruneEmpty keeps a column when .some() cell is not empty, so it is defined in terms of the cell test, and its promise to the reader is only true while both agree. The old cell test was a strict SUBSET of isValueEmpty, so the fix is a pure widening — no cell that drew the em-dash before stops drawing it. Measured in real DOM, the set difference maps one-to-one onto the observed blank cells: '   ' rendered as three literal spaces and [] rendered as nothing, in columns that survived because another row had a value. TWO THINGS CONTRADICT THE BRIEF. (1) packages/plugin-detail/src/emptiness.ts DOES NOT EXIST on main — PR #8457 has not landed (no commit mentions it), so the brief's and the card's instruction to read its docblock first could not be followed as written; I read the equivalent hasCellValue docblock in DetailSection.tsx on main AND the extracted file from refs/pull/8457/head, and the predicate is byte-identical, so nothing turns on which lands. (2) The brief's core warning was to avoid a reflex convergence onto hasCellValue because of the scalar/object split. That warning is right but its stated reason is incomplete, and measurement shows converging would have REGRESSED for a reason the split does not cover: hasCellValue opens 'if (value !== null && typeof value === object) return true' and typeof [] is object, so it calls an EMPTY ARRAY a VALUE, while isValueEmpty calls it empty. An all-[] column is correctly pruned today; delegating would have made it survive and render blank cells — introducing the very defect. The #8457 docblock reasons only about POPULATED arrays rendering as badges and never considers []. Ablation C makes this a measurement, not an argument.",
  "census_instrument": "Seven mechanism sweeps over the whole file, not one grep for the known pattern: (1) null/undefined comparison in BOTH operand orders; (2) empty-string comparison in both orders; (3) .trim(); (4) .length === 0 and Object.keys().length; (5) Array.isArray guards; (6) String() coercion used as a presence guard; (7) Boolean() and filter(Boolean). Sweep 1 in the reversed/present direction is what caught the two extra spellings a single isValueEmpty grep would miss — the same shape objectui#8394's own census missed.",
  "third_spelling_exists": true,
  "per_site_verdict": [
    {"site": "RelatedList.tsx:913-917 isValueEmpty, feeding pruneEmpty", "asks": "is every cell in this column empty", "verdict": "CONVERGED — this is now the one authority for both rendering decisions"},
    {"site": "RelatedList.tsx:963 makeCell placeholder branch", "asks": "does this cell draw the placeholder", "verdict": "CONVERGED onto isValueEmpty; old test was a strict subset, so pure widening"},
    {"site": "RelatedList.tsx:685-687 lookup id harvesting (v == null, then typeof v === string && v)", "asks": "is this a fetchable foreign key", "verdict": "GENUINELY DIFFERENT — its answer never reaches the reader as a rendered value. A whitespace id yields a doomed filter, not a blank cell. Loose == null plus a truthiness test that rejects empty string but accepts whitespace. NOT converged."},
    {"site": "RelatedList.tsx:746 client text filter (val !== null && val !== undefined && String(val)...)", "asks": "does this value contribute searchable text", "verdict": "GENUINELY DIFFERENT — reversed/present operand direction, and a different DOMAIN: it reads Object.values(row), i.e. ALL fields including ones no column shows, and decides which ROWS appear. Its guard exists so String(null) does not become the searchable text 'null'. NOT converged."}
  ],
  "pin": "packages/plugin-detail/src/__tests__/RelatedList.emptinessAgreement-8459.test.tsx — 6 cases, every one asserting RENDERED OUTCOME (real SchemaRenderer, real data-table, real th/td text), never the predicate. window.innerWidth pinned to 1280 per objectui#8399: RelatedList swaps the whole table for an object-gallery on mobile, which draws none of this. Every negative carries a control that rendered BY VALUE — the pruning case also asserts Product is present AND Widget reached a cell.",
  "empty_for_everything_would_pass": false,
  "ablations": [
    {"name": "A — revert the fix (isValueEmpty(value) back to null/undefined)", "head_blob": "6352a946114de31950574883e0e2240f075316a4", "mutated_blob": "81ff1624fb6a01dc1d2618a2c0d00d0c819f16a1", "result": "2 red BY NAME: 'A SURVIVING WHITESPACE-ONLY CELL DRAWS THE EM-DASH', 'A SURVIVING EMPTY-ARRAY CELL DRAWS THE EM-DASH'"},
    {"name": "B — isValueEmpty = () => true (EMPTY for everything, strictly worse than the bug)", "head_blob": "6352a946114de31950574883e0e2240f075316a4", "mutated_blob": "3f699b311d106ee52448cad712286580f107c9cc", "result": "6 of 6 RED. It prunes every column; both non-regression axes plus every by-value control catch it. This is the direct answer to the brief's question."},
    {"name": "C — the hasCellValue shape (drop the array clause, so [] is a VALUE)", "head_blob": "6352a946114de31950574883e0e2240f075316a4", "mutated_blob": "546226ed2b3b69b4597c9f360c60b2ec0b79d7b7", "result": "1 red BY NAME: 'A SURVIVING EMPTY-ARRAY CELL DRAWS THE EM-DASH'. This is what makes 'do not delegate to hasCellValue' a measurement rather than a claim."}
  ],
  "ablation_method": "Each mutation proved it reached disk by hash (git hash-object differs from git rev-parse HEAD:PATH) AND by removed/injected grep counts in BOTH directions, with the mutated line printed from disk. Each restore verified BY STATE (hash equal to the HEAD blob AND git diff HEAD empty), never by an exit code. Run from a COMMITTED implementation, under trap on EXIT INT TERM with absolute paths. No rebuild step was needed and none is claimed: the pin imports ../RelatedList relatively from source, so no dist artifact is on the resolution path — stating that rather than reporting a dist preflight that did not apply.",
  "non_regression_axes": "0 IS A VALUE (an all-zero column survives AND renders 0) and AN OBJECT IS A VALUE (a location column renders '51.5000, -0.1200'). The second is also the objectui#8376 / objectui#8394 axis — red for a wholesale delegation to the display-name authority. Ablation B shows these two plus the by-value controls are what make an EMPTY-for-everything predicate fail all 6.",
  "tests": "Run from the repo root, paths relative to the root, nothing after a bare double-dash. (1) pnpm exec vitest run packages/plugin-detail/ -> exit 0, 'Test Files 138 passed (138) / Tests 1239 passed (1239)'. (2) the new pin alone -> exit 0, 6/6. (3) pnpm --filter @object-ui/plugin-detail run type-check -> exit 0 (note the hyphenated script name; it runs tsc --noEmit && tsc -p tsconfig.test.json). (4) pnpm --filter @object-ui/plugin-detail run lint -> exit 0, verdict line 'x 904 problems (0 errors, 904 warnings)'; the 3 warnings in the new test file are no-explicit-any, matching every sibling test in the package. (5) closure build FIRST: pnpm --workspace-concurrency=2 --filter '@object-ui/plugin-detail^...' build -> exit 0. Ablations A/B/C as above. DECLARED NARROWING: app-shell is a genuinely affected package (RecordDetailView and AssignedUsersSection consume RelatedList), but the full packages/app-shell/ suite exceeded the 10-minute foreground cap twice producing no output, so I ran the tests that actually mount a RelatedList surface — relatedListFilter-4664, relatedListInheritedSort-5795, richtextSurfaceParity, AssignedUsersSection, both RelatedRecordActionsBridge tests, deriveRelatedLists -> exit 0, 'Test Files 7 passed (7) / Tests 64 passed (64)'. The rest of app-shell is left to CI; declared, not silently skipped.",
  "changeset": ".changeset/8459-relatedlist-emptiness-agreement.md at '@object-ui/plugin-detail: minor', matching the landed precedent PR #8457 for a plugin-detail render change. Gate verdict line quoted verbatim: 'OK  1 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s): .changeset/8459-relatedlist-emptiness-agreement.md.' skip-changeset NOT applied (phantom label in this repo, and the change is user-visible anyway); major is forbidden here. Also ran node scripts/check-governed-queue-guard.mjs --test with the three changed paths passed EXPLICITLY (never bare, which is a silent false green): 'NOT GOVERNED — 3 path(s) checked against 5 governed surface(s); none matched.'",
  "mcp_calls": "9 — issue_read get, issue_read get_comments, 2x search_issues (dedup, both returned #8459 as a control hit so the readings are real), 3x issue_write create (the three findings), create_pull_request, add_issue_comment. REST was 403 from this container (probed before choosing), so the dedup search ran over MCP as a declared channel switch; PR body read-back used the zero-quota public web page instead of a further MCP call.",
  "open_questions": [],
  "out_of_scope_findings": [
    "filed as #8474: the SHARED authority hasCellValue calls an EMPTY ARRAY a value, so the RECORD PAGE paints a blank row for []. Measured on main: a DetailSection with tags:[] renders 'SNameWidgetTags' with ZERO em-dashes, plus the full objectui#8376 triple (escapes emptyCount, canCopy offers to copy it, and it can arm auto-hide alone). Survives PR #8457 verbatim — the extracted predicate is identical.",
    "filed as #8475: RelatedList draws a hand-rolled em-dash span (class 'text-muted-foreground/50 text-xs italic', captured from real DOM) instead of the shared EmptyValue from @object-ui/components, so its empty cells carry NO aria-label while cells that fall through to a renderer's own empty branch do. DetailSection's hand-rolled one at least has aria-label and title; RelatedList's is the only one of the three with no accessible name.",
    "filed as #8477: a RelatedList column with no resolvable field type gets no cell renderer at all (makeCell returns undefined), so pruning and the placeholder STILL disagree on that path even after this fix. Measured on the fix branch: an undeclared 'memo' column survives and renders three literal spaces. Left unfixed deliberately — attaching a cell where there deliberately is none is a different mechanism with a new verification surface, outside this card's fence."
  ],
  "contradicts_brief": [
    "packages/plugin-detail/src/emptiness.ts does not exist on main — PR #8457 has NOT landed (verified: no commit in origin/main mentions 8457). Both the card and the brief instructed reading its docblock first. Read the equivalent on main (DetailSection.tsx hasCellValue) and the extracted file from refs/pull/8457/head; predicates identical, so the analysis is unaffected.",
    "The brief said the settled right shape is the scalar/object split and warned against hasCellValue on those grounds. The split is right, but its stated reason ('objects go to type-aware renderers that draw them fine') does not cover the EMPTY array, which has nothing to draw. hasCellValue's blanket 'every non-null object is a value' is therefore too coarse by exactly one case, and RelatedList's local isValueEmpty is strictly better-shaped for this surface than the shared authority. Ablation C measures it; #8474 files the consequence on the record page.",
    "PR #8457 does not touch RelatedList.tsx (verified against refs/pull/8457/head: it touches DetailSection, DetailView, HeaderHighlight, HistoryTimeline, RecordMetaFooter and adds emptiness.ts), so despite the merge-queue warning there is no file collision with this branch.",
    "Empty string '' was already rendering the em-dash BEFORE this fix, so the card's framing is one case narrower than stated for that value — but not by this component's doing: TextCellRenderer independently tests coerceToSafeValue(value) === '' and returns EmptyValue. That agreement was a downstream coincidence, not a decision, and it stayed green under ablation A. After this fix the '' answer is decided here rather than depending on each renderer re-implementing emptiness.",
    "Process note, not a code finding: the backgrounded app-shell suite was started TWICE by the harness (two live vitest trees, PIDs 23289 and 24189, same command, ~4 min apart, both writing to one truncated log). Verified both were mine and no other agent's vitest was running before terminating them by PID, never by process name."
  ]
}

---

Generated by Claude Code