finding(react): five spellings of one "is this a real config bag?" predicate in packages/react

[os-sales]

Observation-class, measured while implementing objectui#6752. No behaviour is wrong; this is one question with several spellings.

The fact, measured on b76ca6764

"Is this value a real config bag (an object, not an array, not null)?" is asked in packages/react in five places, in four different spellings:

1. SchemaRenderer.tsx — the properties evaluation guard;
2. SchemaRenderer.tsx — propsWithoutCanonicalKeys, inline;
3. SchemaRenderer.tsx — the winningVisibilityKey caller block, inline;
4. utils/propsBagDiagnostic.ts — a private isConfigBag helper (objectui#6708);
5. utils/unevaluatedExpression.ts — the early return in scanBag (objectui#4795).

objectui#6752 unified 1 and 2 behind a single isConfigBag in SchemaRenderer.tsx, because it had to touch both. Sites 3, 4 and 5 were left alone deliberately: they live in other cards' modules, and merging them is a refactor across cards rather than that one's to make. The new helper's docblock says so rather than leaving it implied.

Why it is worth a line

This repo has repeatedly paid for one question having several spellings — hasDeclaredPredicate exists because "is a gate DECLARED?" had three (objectui#3842 / objectui#3849 / objectui#3850), and the SchemaRenderer comments name a "fourth dialect of one question" as the thing to avoid. The five sites agree today. The cost is that they can drift apart silently: each is a boolean expression whose disagreement produces no error, only a different answer on one channel.

Arms, if anyone takes it

• export one predicate (a small shared util in packages/react/src) and have all five read it, with a pin that a new spelling fails;
• or decide the duplication is acceptable at this size and record why, so the next reader does not re-open it.

No behaviour change either way. Not urgent.

Related: objectui#6752 (where sites 1 and 2 were merged), objectui#6708, objectui#4795.

[os-sales]

Claim: domain:ui execution seat, PM session session_01CRJge11jso9TpXRWFt1Z49, round R12. Branch: claude/issue-6761-config-bag-predicate.

Scope

Five spellings of one predicate — "is this value a real config bag (object, not array, not null)?" — across packages/react. #6752 (PR #6763) already merged sites 1 and 2 behind a single isConfigBag in SchemaRenderer.tsx, and deliberately left 3, 4 and 5 alone because merging across modules was a refactor beyond that card. This card is that refactor.

Take either arm and say which you took:

• Converge: export one predicate from a small shared util in packages/react/src, have all five read it, and add a pin that fails when a sixth spelling appears. The pin is the part that matters — without it the card fixes today and not tomorrow.
• Record and close: if measurement shows the duplication is acceptable at this size, write down why, in the place the next reader will look, so this does not get re-opened a fourth time.

⚠️ Re-derive the five sites on your own base; #6763 has landed since the card was written and moved SchemaRenderer.tsx.

Fences

• ⛔ No behaviour change. The five sites agree today. The point is that they can drift apart silently — each is a boolean expression whose disagreement produces no error, only a different answer on one channel. If your convergence changes any answer, you have found a real bug: stop, report it, and do not fold the fix into this card.
• Your file surface is packages/react/src/. Sites 4 and 5 are utils/propsBagDiagnostic.ts (finding(react): a key authored under the props envelope never reaches a renderer that reads schema — measured as a silent empty data-table next to a working properties twin #6708) and utils/unevaluatedExpression.ts (除 content 外没有任何文本键既被求值又被读回 —— statistic.value / card.title / button.label 无法绑定表达式 #4795) — both in your surface, both fair game.
• Four sibling devs hold packages/plugin-timeline/, packages/components/src/renderers/, apps/console/ and packages/fields/src/widgets/ this round. Stay out.

One reading from #6752 worth carrying

That card's dev ran an isolated ablation and overturned the properties guard's own stated reason — ablating it left the hoist's indexed keys identical, so the guard's reason turned out to be channel-independent rather than hoist-specific, and the false comment was corrected. Reading a diff shows structure; it does not show which half is load-bearing. If you converge five sites, ablate each one rather than trusting that they look alike.

---

Generated by Claude Code

[os-sales]

os-dev-report

{
  "issue": 6761,
  "status": "done",
  "branch": "claude/issue-6761-config-bag-predicate",
  "pr": "https://github.com/objectstack-ai/objectui/pull/6769",
  "premise_still_valid": true,
  "summary": "Arm taken: CONVERGE. Re-derived on my own base (b98352a15): the count is SIX occurrences in four spellings, not five — utils/unevaluatedExpression.ts carries TWO (scanBag's negated early return, and the `hoisted` dedupe read 40 lines below it), and it was six on the card's base b76ca6764 as well; the file was counted once. All six now read one exported predicate, packages/react/src/utils/configBag.ts (9 call sites), and packages/react/src/utils/configBag.pin.test.ts fails when a seventh appears — it scans every production source under packages/react/src for the typeof-plus-Array.isArray conjunction on the same operand, either order, either polarity, and requires the set to equal a declared allowlist; a second leg pins that the three converged modules still import the predicate. The two sites asking the same SHAPE question about a data ROW (SchemaRenderer's boundRecord, usePredicateRecordContext) are deliberately NOT merged and are named in the allowlist with that reason: their 'no row means bind NOTHING' rule is a separate ruling and must not move if 'config bag' ever narrows. No behaviour change: every spelling computed the same answer (the one textual outlier, `!= null` vs truthiness, cannot disagree because every falsy value except document.all fails the next conjunct). #6752's pin SchemaRenderer.degeneratePropsBag.test.tsx is UNTOUCHED by this PR (not in git diff --name-only) and green — and it is demonstrably the oracle, going red in three of the nine ablations. Changeset has EMPTY frontmatter (declared as releasing nothing). Two readings that are recorded in the PR body rather than filed as issues, because both are about lines inside this card's own surface: winningVisibilityKey's guard is NOT load-bearing (doubly dominated), and collectDroppedPropsKeys' outgoing guard is unreachable from the renderer though load-bearing for a direct caller.",
  "tests": "All heavy runs serialized through the container's shared verify lock; verdicts quoted from the gate's own line, never a bare $?. At final commit f6b22b4db on a clean tree (git status --porcelain empty): `pnpm exec vitest run packages/react/` REACT_SUITE_EXIT=0, Test Files 67 passed (67), Tests 989 passed (989). `pnpm --filter @object-ui/react run type-check` TYPECHECK_EXIT=0 (echoed `tsc --noEmit && tsc -p tsconfig.test.json`; first attempt was TS2307 module-not-found until `pnpm --filter '@object-ui/react^...' build` BUILD_EXIT=0 built the dependency closure). NOT-MEASURED check: `tsc -p tsconfig.test.json --listFiles` lists BOTH new files, so typecheck really covers them. eslint on the 5 changed files ESLINT_EXIT=0, 5 files linted, 0 errors, 16 warnings — all pre-existing no-explicit-any in SchemaRenderer.tsx at untouched lines, 0 messages on the new files. check-changeset-presence EXIT=0, check-changeset-no-major EXIT=0, check-control-bytes EXIT=0 (5629 files), check-type-check-coverage EXIT=0, check-lint-coverage EXIT=0 (46/46). Declared narrowing: repo-wide `pnpm lint` not run locally — population read from eslint's own config (files '**/*.{ts,tsx}' plus its ignores), count read from --format json (5), and the config declares no parserOptions.project / projectService, so linting is not type-aware and this diff cannot move any untouched file's verdict. ABLATIONS, nine call sites, one at a time: implementation committed FIRST, then one call site replaced with bare truthiness, mutation proven on disk (anchor count 1 to 0 and mutation count 0 to 1, plus git diff --numstat), targeted 7-file suite run, restore via `git checkout HEAD -- ABSOLUTE_PATH` proven by `git diff HEAD` empty, driver in try/finally so a kill mid-mutation still restores. No build leg is needed and this is proven, not assumed: vitest.config.mts aliases every @object-ui/* to that package's src/, and five of nine ablations going RED is the live evidence a source mutation reaches the run. Baseline 7 passed / 123 passed, exit 0. RED: properties evaluation guard 1 failed/122 (degeneratePropsBag); props evaluation guard 3 failed/120; propsWithoutCanonicalKeys props bag 2 failed/121; collectDroppedPropsKeys authored bag 2 failed/121 (propsBagDiagnostic 'returns null for a string / an array authored bag'); scanBag early return 1 failed/122 (unevaluatedExpressionDiagnostic 'ignores non-string values and degenerate bags'). GREEN (no test covers them) — measured with a direct probe through the real renderer / the exported functions, guarded vs ablated, instead of assumed: (a) propsWithoutCanonicalKeys' properties guard IS load-bearing — props {'0':'ZERO', title:'T'} with properties 'ab' gives prop 0='ZERO' guarded and 0='a' ablated, with properties ['X'] gives 0='ZERO' vs 0='X' (a degenerate bag was subtracting the author's own key); (b) winningVisibilityKey's guard shows NO observable difference across five shapes (string, array, number, true, real bag) — same render outcome, byte-identical console.warn — it is dominated by hasOwnProperty being unable to hold a visibility-key name on a primitive/array AND by its only consumer already being gated on rawPropertiesBag; (c) collectDroppedPropsKeys' outgoing guard IS load-bearing for a direct caller (null vs ['0'..'8'] for a string, null vs ['0','1'] for an array) but unreachable from SchemaRenderer, since propsWithoutCanonicalKeys has returned an object at every exit since #6752; (d) the uncounted sixth (`hoisted`) IS load-bearing — collectUnevaluatedExpressions({'0':'${data.x}'}, ['${data.x}']) reports the residual expression on the schema channel guarded and returns NO findings ablated, so ablating the spelling the card did not count silences a diagnostic. Probe file was temporary and is deleted; final tree clean.",
  "mcp_calls": "6 — issue get, issue get_comments, create_pull_request, pull_request_read (body read-back), add_issue_comment, and one comment read-back. gh CLI is absent on this seat; no REST channel was needed and no issue search was needed (nothing to file).",
  "open_questions": [],
  "out_of_scope_findings": []
}

---

Generated by Claude Code