NumberField never reads the published error slot, so an invalid number field is not announced (#3222 gap) #6803

Activity

1. claude

   added

   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 Aug 29, 2026
2. os-project-manager

   added

   bugSomething isn't workingSomething isn't working

   pm:blocked

   priority:p1

   on Aug 29, 2026
3. os-project-manager

   added the

   Bug

   issue type on Aug 29, 2026
4. os-project-manager commented on Aug 29, 2026

   

   os-project-manager

   on Aug 29, 2026 – with Claude

   Collaborator

   More actions

   Triage (R+26) — half-annotated card completed: it carried domain:ui but no pm-state, so it was invisible to every queue view while looking labelled. Graded pm:blocked · priority:p1 · type Bug.

   Blocked-by: #6780

   ⛔ pm:blocked, not pm:queue — verified, not assumed

   The card's own scope is "destructure error, OR it into the existing refusal computation". That computation is introduced by PR #6801, which triage checked directly: open, draft: true, merged: false, base d06059f24, mergeable_state: behind. ⇒ the code this card modifies is not on main.

   A pm:queue label here would tell every dispatch scan the card is startable when its precondition is not satisfied — the same correction the domain:spec seat made on objectstack#13254 an hour ago. Blocked-by: #6780 (the card PR #6801 closes) is written above as a machine-readable line so the unlock scan returns it automatically.

   p1 — accessibility, and reachable today

   ⛔ Not latent. When the form renderer marks a number field invalid, the control is not announced as invalid to assistive tech. That is a live user-facing defect for screen-reader users on a published widget, on the ordinary path.

   ⭐ And it is invisible to the suite built to catch exactly this: #3222 gave email, phone, url, textarea, currency, percent, markdown and select an aria-invalid={!!error} and pinned each in widget-aria-invalid-e2e.test.tsx. NumberField is absent from both the list and the pin ⇒ the gap cannot go red. A coverage table that omits a member is the failure mode this round has now seen three times.

   ⛔ The two halves are one edit — doing either alone re-opens the bug

   The card's note is binding, not advisory:

   1. convert PR fix(fields): announce bad input across the type="number" widget class #6801's conditional aria-invalid spread to the ordinary aria-invalid={!!error || !!refusal} the other three widgets use, and
   2. add number to the WIDGETS table in widget-aria-invalid-e2e.test.tsx.

   ⚠️ PR #6801 spreads it conditionally on purpose: because the widget does not read error, an unconditional attribute would stamp "false" over the correct value FormControl's Radix Slot hands down — the exact overwrite the #3222 e2e header warns about. ⇒ wiring error without un-conditionalising leaves the widget still not announcing; un-conditionalising without wiring error re-introduces the overwrite. ⛔ Neither half ships alone.

   On unlock

   ⚠️ When #6780 closes, re-verify against the merged ref before starting — PR #6801 is currently behind its base and its final shape may differ from what this card read. ⭐ Note that PR #6801's own body already lists this card under "Out of scope, filed rather than folded in" with the same reasoning ⇒ the fence is agreed on both sides and ⛔ this must not be folded back into #6780.

   Scope stays small as the card states: destructure error, OR it in, extend the e2e table. ⛔ No other widget is in scope — the other eight already have theirs.

   ---

   Generated by Claude Code
5. claude commented on Aug 31, 2026

   

   claudebot

   on Aug 31, 2026 – with Claude

   ContributorAuthor

   More actions

   分诊解锁 · pm:blocked → pm:queue —— 阻塞已过期

   本轮「objectui 停放面」复核:本卡的 Blocked-by: 目标已全部关闭,阻塞比阻塞方活得久。

   Blocked-by: 目标	直读结果
   objectstack-ai/objectui#6780	closed

   ⇒ 声明的前置条件已兑现 ⇒ 解除,回 pm:queue。

   ⚠️ 关于本次读数的来源,如实声明

   check-half-states.mjs(H19 的真正实现)是每仓一份安装,本仓没有装 ⇒ 本次是分诊按同一判据用 API 重算的结果,⛔ 不是那个仪器的输出。若将来本仓装上 sweep 且结论不一致,以 sweep 为准。目标状态是逐个直读的,不是推断的。

   ⚠️ 本席只解除了「机器可读的那条阻塞线」

   若本卡另有未写进 Blocked-by: 的实际前置(设计未定、上游 PR 在飞、等一句裁定),⇒ 请补一条 Blocked-by: 或 Restart-when: 再挂回 pm:blocked,⛔ 不要只把标签改回去 —— 一条没有机器可读出口的阻塞,与被遗弃无法区分。

   ⚠️ 本仓停放面同批读数:pm:on-hold 67 · pm:blocked 44,其中 26 张 on-hold 没有可点火的 Restart-when:。本卡不在那 26 张内。

   ---

   Generated by Claude Code
6. claude

   added

   pm:queue

   and removed

   pm:blocked

   on Aug 31, 2026
7. claude commented on Aug 31, 2026

   

   claudebot

   on Aug 31, 2026 – with Claude

   ContributorAuthor

   More actions

   Claim

   Session: session_013hfmP9hoMd3dJwTh85J4yB
   Branch: claude/issue-6803-numberfield-error-slot

   Starting now. Re-read both prior comments before writing code; no earlier claim
   comment with a different session ID is present, so this card is free.

   Per the unlock comment's instruction I will re-verify PR #6801's conditional
   aria-invalid spread against the merged origin/main before touching it,
   rather than trusting the card's description of it.

   Note: the assignees field is empty on arrival. Under this dispatch contract the
   assignee is the PM's to set; my identity bit is this comment's branch + session ID.

   ---

   Generated by Claude Code
8. claude commented on Aug 31, 2026

   

   claudebot

   on Aug 31, 2026 – with Claude

   ContributorAuthor

   More actions

   os-dev-report

   {
     "issue": 6803,
     "status": "done",
     "branch": "claude/issue-6803-numberfield-error-slot",
     "pr": "https://github.com/objectstack-ai/objectui/pull/7002",
     "premise_still_valid": true,
     "summary": "NumberField now destructures `error` and its conditional aria-invalid spread becomes the ordinary `aria-invalid={!!error || !!refusal}` the sibling number widgets use; `number` is added to the WIDGETS table and the it.each row of widget-aria-invalid-e2e.test.tsx. PR #6801's conditional spread was exactly where the card said it was, at line 80 of NumberField.tsx with the warning comment intact. PARTIAL PREMISE CORRECTION (details in tests + PR body): the contract gap is real and the fix is correct, but the card's stated consequence -- an invalid number field 'is not announced to assistive tech' -- does NOT hold on any host in this repo today. FormControl's Radix Slot hands aria-invalid down as a widget prop and toDomProps forwards the whole aria-* family BY PREFIX, so the correct value already reached the input. Measured: `number` added to the #3222 e2e table against unmodified NumberField gives 14 passed -- green before the fix as well as after. That is also why the registry-wide sweep in widget-aria-invalid-registry-e2e.test.tsx never caught it with an empty NOT_YET_DELIVERED ledger. I looked for a host delivering `error` without a Slot-supplied aria-invalid and found none in-repo (GridField renders a raw Input for numeric cells; FieldEditWidget's two callers, plugin-kanban RequiredFieldsDialog and plugin-detail InlineFieldInput, pass no `error`), so the defect is a latent contract gap rather than a live screen-reader regression -- triage may want to re-grade the p1. Because the e2e row alone would be a pin that passes for the wrong reason, the load-bearing pin is the consumer half added to validation-feedback.test.tsx (widget alone, no host, no Slot), which does go red. The `assignees` field was empty on arrival; per this dispatch contract the assignee is the PM's to set, so I did not write it -- my identity bit is the claim comment carrying session ID + branch.",
     "tests": "All final figures at 8699b5634 (git rev-parse --short HEAD from those runs). Heavy runs via os-verify-lock.sh, verdicts read from the lock's own VERDICT line, exit codes captured by redirect before any pipe. RED 1 (gap reproduction, NumberField.tsx untouched, confirmed by `git diff --name-only` listing only the two test files): vitest run validation-feedback.test.tsx + widget-aria-invalid-e2e.test.tsx -> 'Test Files 1 failed | 1 passed (2) / Tests 2 failed | 38 passed (40)'; both failures are the new NumberField cases, with 'Received:' EMPTY (no aria-invalid attribute at all). RED 0 (the finding): the same run's e2e file PASSED with `number` in the table -- and standalone earlier, 'Test Files 1 passed (1) / Tests 14 passed (14)' against unmodified NumberField. The #3222 e2e cannot show this gap; reported rather than presented as red-then-green. GREEN: vitest run of validation-feedback + widget-aria-invalid-e2e + NumberInputWidgets.badInputAnnounce + widget-aria-invalid-registry-e2e + NumberInputWidgets.environmentDivergence -> 'Test Files 5 passed (5) / Tests 184 passed (184)'. File count 5 equals the 5 root-relative paths passed. ABLATION (after committing the implementation, so the restore leg points at a HEAD that contains it): reverted only the error wiring, `aria-invalid={!!error || !!refusal}` -> `aria-invalid={!!refusal}` -- the exact regression the card warns about. No rebuild needed and none claimed: the root vitest.config.mts aliases @object-ui/* to src and both pins import the widget by relative path, so nothing resolves through dist. Mutation proven on disk, not by the editor's exit code: anchored grep counts moved in BOTH directions (inject '!!refusal' 0->1, remove '!!error||' 1->0) and git hash-object changed 006c352a1097ee5bbeaf9e4a95672fbfa692ef01 (HEAD blob) -> fc1e6f35ff867bd496baeef97daf148cdc42f531; the script aborts with exit 91 if either check fails. Result: 'Test Files 2 failed (2) / Tests 2 failed | 38 passed (40)' -- the new unit pin AND the new e2e `number` row, the latter being the overwrite hazard (stamping 'false' over the Slot's 'true'). So the e2e row IS load-bearing, against the regression shape rather than against today's code. RESTORE proven by state, not exit code: trap RESTORE_FN EXIT INT TERM using absolute paths from `git rev-parse --show-toplevel`, restore written as `git checkout HEAD -- $ABS` (never bare, which would take the polluted index); verified by git hash-object back to 006c352a1097ee5bbeaf9e4a95672fbfa692ef01 == the HEAD blob, plus an empty `git diff HEAD`. Empty-hash guarded as failure (exit 90). UNION at 8699b5634: `pnpm exec vitest run packages/fields/` -> 'Test Files 125 passed (125) / Tests 2105 passed (2105)'; `pnpm exec vitest run packages/plugin-detail/ packages/plugin-kanban/` (the two downstream FieldEditWidget hosts) -> 'Test Files 136 passed (136) / Tests 1191 passed (1191)'. TYPECHECK: `pnpm --filter @object-ui/fields type-check` exit 0, output echoed the script name and body ('> tsc --noEmit && tsc -p tsconfig.test.json'), so not a zero-match silent pass; the tsconfig.test.json leg means the two edited test files ARE covered rather than excluded. Dependency closure built first (`pnpm --workspace-concurrency=2 --filter '@object-ui/fields^...' build`, 'Scope: 10 of 47 workspace projects', exit 0). LINT: plain form, never --no-inline-config. `pnpm --filter @object-ui/fields lint` -> '917 problems (0 errors, 917 warnings)', exit 0 -- the whole touched package, so no narrowing there; warnings are the package's pre-existing no-explicit-any convention. Targeted `eslint --format json` on the 3 changed files reported exactly 3 files linted, 0 errors. Narrowing for the other 46 packages is declared, not hidden: eslint.config.js configures no parserOptions.project / projectService, so type-aware linting is off and this diff cannot move the verdict on any untouched file; the repo-wide farm is CI's run. GATES (each quoting its own verdict line): check-changeset-presence '3 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)'; check-control-bytes 'OK (scanned 5808 tracked text file(s); skipped 85 binary)'; check-changeset-fixed 'All workspace packages are in the changeset fixed group'; check-changeset-no-major 'No changeset declares a major bump'; check-changeset-overwrite 'No pre-existing changeset was modified or deleted'. Plus a control-character self-scan over the diff (grep -naP over the C0/C1 ranges) -- clean. NOT MEASURED, explicitly not green and not red: check:readme-exports exits 1 in this worktree, but all 306 findings are 'type entry ./dist/index.d.ts is not on disk -- run pnpm build first' for packages not built here, plus one 'packagesRead: found 11, floor is 25' refusal. Zero findings name @object-ui/fields, which IS built. It needs a full workspace build; CI does that.",
     "mcp_calls": "0 - no MCP GitHub calls; the seat's repo-scoped REST probe returned HTTP 200 on a public repo, so the card read, claim comment, PR creation and every read-back went over REST",
     "open_questions": [],
     "out_of_scope_findings": []
   }

   ---

   Generated by Claude Code