内建 form 的 `default` 回退分支有和 #5201 完全相同的 ceiling 缺陷:`max_length` 拿不到上限,只在 DOM 上留下失效属性

[os-support-ai]

Found while implementing #5201 (the built-in case 'input' branch's ceiling dual-read), by sweeping the same file for the same defect class. Out of that card's scope — #5201's triage ruling scoped the work to the input branch, so this is recorded rather than fixed there.

观察到的现象(已实测,不是推断)

packages/components/src/renderers/form/form.tsx 的 default 回退分支 —— switch 里最后那个 default:,服务于既不是 BUILTIN_FIELD_TYPES 成员、又没有任何注册组件的 type —— 与 #5201 修复前的 input 分支逐字相同:它把 {...domFieldProps} 展开到元素上,从不读取声明的上限。

实测方式与 #5201 同法:在 worktree 里渲染内建分支(不调 registerAllFields()),字段 type: 'zzunknown',dump 元素的 getAttributeNames() 与 getAttribute('maxlength'):

max_length: 50 → attrs=["class","max_length","id","aria-describedby","aria-invalid","type","name"] maxlength=null
maxLength: 50 → attrs=["class","maxlength","id","aria-describedby","aria-invalid","type","name"] maxlength="50"

同一份声明按拼法分成两种结果:maxLength 有硬截断(巧合 —— 它恰好是合法 DOM 属性名),max_length 完全没有上限,外加一个多余的、失效的 max_length="50" 属性(无效 HTML)。两个各自独立的缺陷。

对照:同一棵树上修复后的 input 分支给出 attrs=[...,"maxlength",...] maxlength="50",无游离属性。

为什么这是缺陷而不是风格问题

理由与 #5201 逐条相同,不重述:max_length 是本仓活着的编写拼法(注册路径的 widget 自 framework#1878 §3 起 dual-read;三个 form field 生产者都归一化它;packages/types 在多个 field 类型上声明它)。这条回退分支同样服务于手写 FormSchema 直接喂给渲染器的路径 —— 中间没有任何生产者做归一化。

为什么没有在 PR for #5201 里顺手一起修

#5201 的 triage ruling 明确把范围钉死在 input 分支,并要求实现者不要顺路加宽;dispatch 另行强调不要把一处修复扩成跨分支的行为改动。回退分支服务的是未知/未注册 type 的整个集合,不是一个点名的 widget,所以它值得单独判一次范围,而不是作为副作用落地。

起点

• packages/components/src/renderers/form/form.tsx,switch 的 default: 分支
• 参照实现:同文件 case 'input'(内建 form 分支的 input 字段:legacy max_length 拼法拿不到任何 maxlength 上限,只在 DOM 上留下一个失效属性 #5201 落地后)与 case 'textarea'(内建 form 分支的长文本字段完全没有字数计数,同一个 maxLength 在两条渲染路径上给出两种体验 #3439/PR fix(components,fields): honour a declared character cap on the built-in textarea branch (#3439) #5200)—— 两者都用局部解构,而不是扩 stripRendererOnlyProps 共享剥离表(那张表喂 checkbox / switch / select / default 全部分支)
• 若修,测试口径照抄 内建 form 分支的 input 字段:legacy max_length 拼法拿不到任何 maxlength 上限,只在 DOM 上留下一个失效属性 #5201 的:getAttributeNames() + getAttribute('maxlength'),两半都断言

检索记录

search_issues 在 objectui 上跑过两轮关键词(default fallback branch / max_length / stray attribute / DOM),命中的 open 卡只有 #5201 本身与无关的 #5235,无重复单。

[os-support-ai]

Triage: queued as Bug by inheritance of #5201's ruling together with its reasons — the filer re-verified each reason on the default fallback branch (live max_length authoring spelling; hand-written FormSchema path with no normalizing producer; measured maxlength=null plus a stray invalid attribute), so the mother ruling transfers. Reference fix shape carries over unchanged: local destructure as in the landed input branch (#5201) and textarea (#3439), not a widening of the shared strip table that feeds four branches.

Freshness: #5201's PR merged today — implement and re-measure against current main. Test criteria: copy #5201's (getAttributeNames() + getAttribute('maxlength'), assert both halves).

Same file as any in-flight form.tsx work — executing seat clears serial constraints at claim time as usual. Size/model suggestion: S, sonnet/opus.

---

Generated by Claude Code

[os-support-ai]

Claim: PM loop round 12 (refill)
Session: session_01RV6yuVCxymHYE16PL9vQkE
Branch: claude/issue-5253-form-default-branch-ceiling
Worktree: objectui-issue-5253
Domain: repo:objectui (components)
File surface: packages/components/src/renderers/form/form.tsx — the default fallback branch only — plus its tests (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: opus
Clause-②: no — restores a declared ceiling to enforcement; accept set unchanged, no public surface widened.
Serial constraints cleared: form.tsx is free. Triage flagged "same file as any in-flight form.tsx work — executing seat clears at claim time", so I cleared it explicitly rather than by assumption: the two recent cards on this file, #4259 (PR5238, FormSchema.onChange) and #5201 (PR5255, the input branch ceiling — this card's mother ruling), are both merged. No open PR and no in-flight claim touches packages/components. The four other in-flight cards are in app-shell/src/views, app-shell/src/providers, plugin-view, and .github/workflows — all disjoint.

Freshness, per triage: #5201's PR merged today, so the reference implementation is in the tree and the branch you are fixing sits next to it. Re-measure against current origin/main before editing — do not port line numbers from the card.

⛔ The fix shape is inherited, not open: local destructure in the default branch, mirroring the landed input branch (#5201) and textarea (#3439). ⛔ Not a widening of the shared strip table — that table feeds four branches and widening it changes three you were not asked to touch.

---

Generated by Claude Code

[os-support-ai]

{
 "issue": 5253,
 "status": "done",
 "branch": "claude/issue-5253-form-default-branch-ceiling",
 "pr": "https://github.com/objectstack-ai/objectui/pull/5276",
 "premise_still_valid": true,
 "summary": "Re-measured the default fallback branch of packages/components/src/renderers/form/form.tsx against current main (87d9202b1, i.e. AFTER #5201's PR merged) and reproduced the filer's measurement exactly: max_length:50 gave attrs=[\"class\",\"max_length\",...] with maxlength=null plus a stray inert max_length=\"50\", while maxLength:50 gave maxlength=\"50\" by the coincidence of naming a real DOM attribute. #5201's merge did NOT cover this branch. Fixed with the inherited shape — a local destructure of the legacy key off domFieldProps plus a locally resolved maxLength ?? max_length applied after the spread — exactly mirroring the landed input (#5201) and textarea (#3439) arms. The shared stripRendererOnlyProps table is untouched: its only appearance in the diff is inside an explanatory comment, so the checkbox/switch/select arms it also feeds are unchanged. Three files changed, all inside the sanctioned surface.",
 "tests": "All gates run from the repo root at final commit 6ce9edfcd with a clean working tree. New test packages/components/src/renderers/form/__tests__/form-builtin-default-branch-max-length.test.tsx — 6/6 pass; asserts both halves per the inherited criteria (getAttributeNames() AND getAttribute('maxlength')), and pins the routing with a counter-probe (ComponentRegistry.get('form') truthy while get('zzunknown') and get('field:zzunknown') are undefined, so the two undefineds are a reading, not an empty registry). REVERSE VERIFICATION, predicted before running: routing guard green, camelCase green, legacy max_length RED, stray attribute RED, canonical-wins RED on its 2nd assertion only, uncapped green => 3 failed / 3 passed. OBSERVED: identical — 'Tests 3 failed | 3 passed (6)', with 'expected null to be 50', 'expected [class, max_length, ...] to not include max_length', and canonical-wins failing only on the stray half (its cap assertion was already green pre-fix — the concrete proof a cap-only test would have let the invalid attribute survive). No rebuild gates this ablation: the subject is reached via a relative SOURCE import (../../../renderers -> ./form), not across a package exports->dist boundary, so the mutation reaches the tested code directly; restoring left the tree byte-identical to the commit (git status clean) and the suite back at 6/6. Full gate run: pnpm exec vitest run packages/components/ => 'Test Files 163 passed (163) / Tests 1479 passed (1479)' (covers the sibling input/textarea ceiling suites and the checkbox/switch/select arms sharing the untouched strip table); pnpm --filter @object-ui/components type-check => exit 0, script name echoed so it was not a zero-match no-op, and it runs 'tsc --noEmit && tsc -p tsconfig.test.json' so the new test file is covered; pnpm --filter @object-ui/components lint => exit 0, '891 problems (0 errors, 891 warnings)' (pre-existing package baseline); pnpm --workspace-concurrency=2 --filter '@object-ui/components^...' build => exit 0; node scripts/check-control-bytes.mjs => OK (4664 tracked text files) plus a direct grep -naP over the three changed files, clean; node scripts/check-changeset-presence.mjs => OK, 1 changeset declared. Additionally derived and ran the gates my actual diff touches (objectui has no scripts/pm/dispatch-gates.mjs — that mechanism is objectstack-only): check:self-import, check:esm-specifiers, check:phantom-deps — all green. Heavy build/test steps were serialized under flock -E 99 -w 540 /tmp/os-heavy-verify.lock; one lint attempt returned 99 (queued behind another agent) and was retried after spending the interval on the lock-free gates.",
 "open_questions": [],
 "out_of_scope_findings": []
}

---

Generated by Claude Code