finding(plugin-grid): `DATA_SOURCE_WIDGET_TYPES` 是「哪些 widget 要 DataSource」的第四份私有副本 —— 与表单规则不同集、零 gate

[yinlianghui]

发现于 #4790 的实施。同 #4770 / #4790 的失效模式,不在 #4790 范围内(该卡正文只点名两张表),单独立此单记录。

事实(实读 origin/main @ f2dc8fa)

packages/plugin-grid/src/components/bulkParamToField.ts:40-46:

const LOOKUP_WIDGET_TYPES = new Set(['lookup', 'master_detail']);
const USER_WIDGET_TYPES = new Set(['user', 'owner']);
const DATA_SOURCE_WIDGET_TYPES = new Set([...LOOKUP_WIDGET_TYPES, ...USER_WIDGET_TYPES]);

消费点两处:fieldNeedsDataSource()(→ BulkActionDialog.tsx:600 决定是否把网格的 DataSource 传给 param 控件)与 isLookupishParam()(→ 决定是否补 reference_to / display_field)。

它编码的正是对象表单那条规则的同一件事 ——「这个 widget 要查记录,所以得把 DataSource 递给它」—— 但成员是第三种集合:

	成员
@object-ui/core EXPANDABLE_FIELD_TYPES(schema 字段类型域)	lookup, master_detail, tree, user
表单 needsDataSourceWiring(widget 键域,#4790 后派生自上表)	lookup, master_detail, tree, user + object-ref, filter-condition, recipient-picker
plugin-grid DATA_SOURCE_WIDGET_TYPES(widget 键域)	lookup, master_detail, user, owner

即:多 owner,少 tree 与三个 widget-hint picker。没有任何 gate 能发现它继续漂。

为什么今天没有用户症状(据此判为 observation-class)

• 少 tree:批量动作 param 的 type 来自 spec 的 param 类型词表,tree 走 resolveFormWidgetType 后本就落到 lookup,所以这一格实际到不了。
• 少三个 picker:那三个是共享规则页面的 widget hint,批量对话框的 param 目前不产出它们。
• 多 owner:见 owner widget 拿不到表单下发的 dataSource,而 plugin-grid 认为它需要 —— 且 owner 根本不在 spec 的 FieldType 里(enforce-or-remove 待判) #4814,owner 根本不是 spec 的 FieldType,这一格同样是声明多于消费。

所以现状是「三份写法各自恰好够用」,不是线上缺口 —— 但这正是 #4770 与 #4790 两次都描述的前夜状态:注释声称同步、机制不保证,直到某一侧加键才炸。

修法方向(与 #4789 / #4790 同型)

单一定义 + 各消费面自己做归一化 + identity 钉(钉对象身份,成员相同的私有副本骗不过)。落点候选:core 的 EXPANDABLE_FIELD_TYPES(参考半)+ 各面自己的 widget-hint 扩展,与 #4790 给表单定的形状一致。owner 那一格取决于 #4814 的裁决,建议等 #4814 判完再动,否则会把一个待退役的别名钉进共享表。

Blocked-by: #4814

参考

• CASCADE option 允许表现存三份未导出的重复(app-shell / components / plugin-grid)—— 抬升到 @object-ui/fields 并让三处 import 同一常量 #4770 / PR refactor(core,fields): one exported cascade option allow-table, not three private copies (#4770) #4789(cascade 允许表定型)、DATA_SOURCE_FIELD_TYPES(form 层)与 core 的 EXPANDABLE_FIELD_TYPES 已知不同集且零 gate —— 同 #4770 的第二张允许表待收敛 #4790 / 本卡 PR(表单 data-source 规则派生自 core)
• packages/plugin-grid/src/components/bulkParamToField.ts:40-46,144-146、packages/plugin-grid/src/components/BulkActionDialog.tsx:600

[os-support-ai]

Triage first-touch grade (concentrated batch; triage seat, session session_01NYoiJGd1MxXdcvM4EGAMrS): promoted to pm:queue, type Task — not blocked: the body's Blocked-by: #4814 is already satisfied (#4814 closed 08-17 via merged PR #4915, ruling A′ retired the owner spelling). Premise re-verified on the merged ref, not the card text: origin/main:packages/plugin-grid/src/components/bulkParamToField.ts:51 now reads USER_WIDGET_TYPES = new Set(['user']) with a tombstone comment at :45 — so the owner cell of this card's table is done, and the remaining ask is exactly the #4790-shaped convergence: single definition (core EXPANDABLE_FIELD_TYPES as the reference half + per-surface widget-hint extension), per-consumer normalization, and an identity pin (assert same object, so a member-identical private copy can't fake it).

The tree / picker-hint divergences remain measured-but-asymptomatic; the card's own analysis of why is the spec for which cells the shared table must and must not contain. Size/model suggestion: S–M, sonnet/opus.

---

Generated by Claude Code

[os-support-ai]

Claim: PM loop round 15

Session: session_01RV6yuVCxymHYE16PL9vQkE
Branch: claude/issue-4815-data-source-widget-types-convergence
Worktree: objectui-issue-4815
Domain: repo:objectui
File surface: packages/plugin-grid/src/** + packages/core/src/** (the shared reference table) + their tests (stop on breach; explain in the report)
Container & model: S–M, mode:subagent, model: opus (taking the triage seat's upper suggestion — the identity-pin design is the hard part, not the edit)
Clause-②: no — this converges private copies onto one definition. It changes no contract surface and widens nothing published. If your implementation turns out to change which widget types are treated as data-source-bearing anywhere, that is a behaviour change: stop and report it rather than absorbing it into a "convergence".
Serial constraints cleared: packages/plugin-grid and packages/core are both free. Open PRs hold disjoint surfaces, each measured by git diff --name-only origin/main...: PR #5290 apps/console, PR #5294 packages/plugin-view, PR #5295 .github/workflows + scripts/ + root manifests, PR #5297 packages/plugin-map, PR #5298 packages/components + packages/types, PR #5300 / #5303 packages/plugin-dashboard, PR #5302 packages/cli, PR #5310 packages/plugin-report (consumes @object-ui/core but does not modify it — verified on its diff), PR #5311 packages/plugin-form.

Not blocked — verified, not assumed

The body carries Blocked-by: #4814. That is satisfied: #4814 closed 2026-08-17 via merged PR #4915 (ruling A′ retired the owner spelling). The triage seat re-verified the premise on the merged ref rather than from the card text — origin/main:packages/plugin-grid/src/components/bulkParamToField.ts:51 now reads USER_WIDGET_TYPES = new Set(['user']) with a tombstone at :45. So the owner cell of the card's table is already done; do not re-do it.

⚠️ That reading is two days old. Re-derive on current origin/main before editing — the card's own table may have moved again. Treat every figure in the card as a previous round's reading needing re-verification, not as fact.

Scope — the #4790-shaped convergence

DATA_SOURCE_WIDGET_TYPES is the fourth private copy of "which widgets need a DataSource", with a different member set from the form rules and zero gate. Three deliverables:

1. Single definition — core's EXPANDABLE_FIELD_TYPES as the reference half, plus a per-surface widget-hint extension.
2. Per-consumer normalization.
3. An identity pin — ⚠️ assert the same object, not the same members. A member-identical private copy must fail the test. This is the whole point: a value-equality assertion would pass against exactly the defect this card exists to remove, which makes it worse than no test.

The card's own analysis of the tree / picker-hint divergences is the spec for which cells the shared table must and must not contain — those remain measured-but-asymptomatic and are not yours to resolve. Read that analysis before choosing membership.

Verification

• Counter-probe every zero with a known-present neighbouring term.
• Reverse-verify: revert with tests kept, predict exactly which assertions go red and name them BEFORE running, then report predicted-vs-observed. Any assertion green on both legs pins nothing — say so rather than counting it. In particular, prove the identity pin actually discriminates: a leg where a member-identical copy is substituted should go red.
• State whether a build artifact sits between your edit and the thing under test, per leg — this spans two packages, so check whether the tests resolve @object-ui/core via the root vitest.config.mts alias to src or via package exports to dist.

Changeset: patch. ⛔ NEVER declare major — objectui's major is pinned to @objectstack's major.
⛔ Do not edit content/docs/releases/, docs/adr/**, .claude/**, skills/**, AGENTS.md, CLAUDE.md.

---

Generated by Claude Code