finding(app-shell): `view-column-io`'s colLabel/colFieldName still read the retired `accessorKey`/`header` aliases, so the column LIST re-shows an identity the column inspector just stopped showing

[os-zhuang]

Found while implementing #5344 (retiring the READ tolerance in ViewColumnInspector). Filed unassigned, not claiming. Out of that card's granted file surface (ViewColumnInspector.tsx + its tests only), so recorded rather than fixed.

The fact

packages/app-shell/src/views/metadata-admin/previews/view-column-io.ts carries the same undeclared-alias read that #5344 just retired one file over:

// :29 — colLabel()
return String(o.label ?? o.header ?? o.field ?? o.accessorKey ?? `col ${i + 1}`);
// :39 — colFieldName()
const v = o.field ?? o.accessorKey;

Its sole importer is previews/FieldsListEditor.tsx, which supplies the row labels for the column manager — and FieldsListEditor renders inside ViewColumnInspector's own panel (it is rendered by both ViewVariantInspector and ViewColumnInspector; see that file's header docblock).

Why it is worth recording now

After #5344 the two surfaces disagree within one panel. Measured on the #5344 branch, mounting the real component with a stored column {accessorKey: 'name', header: 'Name'}:

• the inspector's Field key control shows '' and its Header control shows '' — the ruled behaviour, since ListColumn refuses both keys by name (unrecognized_keys ["accessorKey","header"]);
• the column list row one line above still reads Name, because colLabel consults o.header.

So the ruling's observable requirement — "the inspector stops presenting a spec-refused spelling as a valid column identity" — is met by the identity controls and not by the list rendered beside them. A pin for exactly this is committed in ViewColumnInspector.identityRead.test.tsx (the test named "records the residue this card is fenced out of"), which asserts getAllByText('Name') has length 1 and carries a comment saying the expectation flips when this is fixed.

Two details worth having before triage

1. colLabel's precedence is not merely tolerant, it is inverted. The chain is label ?? header ?? field ?? accessorKey — header is preferred over field. A perfectly canonical column that also happens to carry a stray header key displays header's value rather than its declared label/field. That is a different (and stronger) defect than the plain ?? accessorKey alias.
2. colFieldName feeds more than a label. It backs usedFieldNames(), which is what the Add-field picker uses to decide which object fields are already taken. Reading accessorKey there means a spec-refused column still reserves a field name in the picker.

Scope note

Same family as #5068 (renderer), #5120 (packages/components / plugin-dashboard), #5340 (bridgeListView), and #5344 (the column inspector, now fixed). This is the sixth site and the second editor-side one. Whether it should be retired the same way #5344 was, or whether the list deliberately wants a best-effort human label for a column it cannot otherwise name, is a judgement this finding does not take — note the two surfaces are not symmetric: an empty field-key box invites re-authoring, an empty list row would leave the author with nothing to click.

Reproduce

sed -n '1,45p' packages/app-shell/src/views/metadata-admin/previews/view-column-io.ts
grep -rn "view-column-io" packages/app-shell/src --include=*.tsx # sole importer

[claude]

Finding first-touch (triage daily round, session session_01Kktexqp6uVuFMztvvTMf3V, 2026-08-24): promoted to pm:queue (domain:ui already present), Bug — same undeclared-alias read #5344 retired, one file over, and the two surfaces now disagree inside one panel. Same retirement shape; re-verify the #5344 landing on current main first.

---

Generated by Claude Code

[yinlianghui]

Claim: PM loop round 38 (refill slot)
Session: session_01CSoz9uGhaaSgiq3hshtN7L
Branch: claude/issue-5725-view-column-io-alias-reads
Worktree: objectui-issue-5725
Domain: domain:ui
File surface: packages/app-shell/src/views/metadata-admin/previews/view-column-io.ts + its tests + the pin in ViewColumnInspector.identityRead.test.tsx + a changeset. FieldsListEditor.tsx (the sole importer) is read-only unless the fix demands otherwise — say so if it does.
Container & model: M, mode:subagent, model: opus
Clause-②: no — an internal editor helper stops reading two undeclared aliases. No published declaration moves; ListColumn already refuses both keys by name.
Serial constraints cleared: packages/app-shell/src/views/metadata-admin/** — no in-flight claim (#5565 and #5835 both landed earlier this round and neither touched this subtree; #5835's fence explicitly kept it out). Concurrent siblings: #6010 (packages/components/src/renderers/form/form.tsx), #5808 (packages/plugin-detail/src/renderers/__tests__/), #5903 (packages/types + packages/plugin-gantt), #5872 (packages/test-support + ~10 parity test files). Disjoint by full path.

裁决 — triage's grade

Triage seat, 2026-08-24, promoted → pm:queue, Bug. Verbatim: "same undeclared-alias read #5344 retired, one file over, and the two surfaces now disagree inside one panel. Same retirement shape; re-verify the #5344 landing on current main first."

Read #5344's PR before writing anything — the fence says to reuse its retirement shape, and this is the sixth site in that family (#5068 renderer, #5120 packages/components/plugin-dashboard, #5340 bridgeListView, #5344 the column inspector) and the second editor-side one.

Two defects that are NOT in dispute — fix these regardless of how the third resolves

The card names three things, and only one of them is a judgement call. ⛔ Do not let the open question stall the two that are not.

1. colLabel's precedence is INVERTED, not merely tolerant. The chain at :29 is label ?? header ?? field ?? accessorKey — header is preferred over field. So a perfectly canonical column that also carries a stray header key displays header's value instead of its declared label/field. ⚠️ That is a bug under every reading of the third question: even if the list keeps a best-effort fallback, a declared identity must outrank an undeclared alias.

2. colFieldName backs more than a label. At :39 it is o.field ?? o.accessorKey, and it feeds usedFieldNames() — which the Add-field picker uses to decide which object fields are already taken. So a spec-refused column still reserves a field name in the picker, hiding a field the author is entitled to add. ⚠️ Also a bug under every reading: whatever the display fallback should be, a refused spelling must not consume a name.

PM建议路线 on the third question — ⛔ a lean, NOT a ruling

The open question is whether colLabel should drop its alias fallback entirely (matching #5344) or keep a best-effort human label. The card declines to take it, and names the asymmetry: "an empty field-key box invites re-authoring, an empty list row would leave the author with nothing to click."

Lean: follow triage's "same retirement shape" — drop the undeclared aliases — but ⛔ do NOT let a row render nameless. A column the helper cannot otherwise name should fall through to the existing positional fallback `col ${i + 1}`, which already exists at the end of that chain and is exactly the "something to click" the asymmetry worries about. That resolves the tension without inventing a new affordance.

⛔ If your measurement shows that fallback does not actually produce a clickable row — the list keys on the label, the empty case collapses the row, anything of that shape — ⛔ stop and report. That is a UX call for the maintainer, not a dev call and not mine.

PM mechanism assumptions — measure, do not assume

1. Re-verify the finding(app-shell): the designer's ViewColumnInspector writes back whichever column spelling it found, so editing a stored accessorKey column re-saves a spelling object-grid no longer renders #5344 landing on current main first, as triage instructs. ⚠️ packages/app-shell moved twice today (Approvals drawer summary card can render a field the object declares hidden: true #5565's console-side seam, [finding] RecordDetailView carries a second confirm runtime whose options bag no feeder ever fills #5835's RecordDetailView), and this seat merged eight PRs. Re-derive view-column-io.ts:29 and :39 and the sole-importer claim (grep -rn "view-column-io" packages/app-shell/src --include=*.tsx) on your merge-base, and report any delta.
2. ⚠️ There is an existing pin that ASSERTS THE BUG and is documented to flip. ViewColumnInspector.identityRead.test.tsx, the test named "records the residue this card is fenced out of", asserts getAllByText('Name') has length 1 and carries a comment saying the expectation flips when this is fixed. ⛔ Do not read its failure as a regression, and ⛔ do not delete it — flip it, and say in the report that you did. A pin written to be flipped is the previous card handing you its receipt.
3. usedFieldNames() is the reachability claim for defect 2. Verify that colFieldName really feeds it and that it really drives the Add-field picker, in those functions' own source. If the chain is broken somewhere, defect 2 is smaller than the card prices and you should say so.

Verification

Pin the panel-level disagreement, because that is the actual defect: one stored column {accessorKey: 'name', header: 'Name'}, one mount, asserting the inspector's identity controls and the list row beside them agree. That is the observable the #5344 ruling required and did not get.

Direction predicted before running: revert each fix and show its own assertion goes red — the inverted-precedence case and the picker-reservation case are separate and need separate legs. Counter-probe required: a canonical column (declared field + label, no aliases) must still show its label in the list and still reserve its name in the picker — otherwise "the aliases are gone" is satisfiable by breaking the list and the picker outright.

Prove every mutation on disk by grepping the injected text and separately the removed text; ⚠️ print the landing site rather than assuming it, and ⚠️ assert anchor uniqueness before writing — both traps were measured in this round (#4730's mis-anchored injection that still moved the number in the predicted direction; #5927's anchor hitting three sites on byte-identical lines). Restore under trap … EXIT INT TERM; git diff HEAD --stat empty afterwards.

⚠️ Run app-shell path-filtered — whole-package vitest there is ~784 s and holds the shared lock; ⛔ do not run it. Gates by name with exit codes: pnpm --filter @object-ui/app-shell type-check (⚠️ hyphenated; echo the script name so a zero-match silent pass cannot read as green — and note a first RED reading Cannot find module @object-ui/... is the unbuilt-closure signature while Cannot find module on a bare package name in a fresh worktree is the missing-node_modules signature; they need different fixes and #5835 hit the second one today), the path-filtered vitest, eslint on changed files with merge-base delta. ⚠️ Plain eslint . per package — ⛔ no --no-inline-config; it manufactures phantom errors (measured twice today).

Run node scripts/check-changeset-presence.mjs rather than guessing; empty frontmatter is this repo's "releases nothing" declaration. Draft PR, Fixes #5725. ⛔ Do not mark it ready, ⛔ do not enqueue, ⛔ do not enable auto-merge — the PM lands it. Post the terminal report here (plain-text first-line marker os-dev-report; ⛔ avoid short angle-bracket placeholders in prose — this repo's sanitizer eats them) and return it.

---

Generated by Claude Code