Repository navigation
type: 'tree' can only nest by a self-parent pointer — a value-grouped tree is inexpressible, and on an object with no self-reference the view renders flat with no diagnostic #14109
Description
Found while building a "what does this team owe" lens in objectstack-ai/duly (issue #12): duties grouped by business unit, then by owner, as a collapsible tree. Measured on @objectstack/spec 17.2.0 and @objectstack/console 17.2.0.
What tree actually does
TreeConfigSchema documents it plainly: "Renders a self-referencing object as an indented, expand/collapse tree-grid. Flat records are nested via a single-parent pointer field (parentField)." The renderer agrees — plugin-tree reads exactly four keys (parentField, labelField, fields, defaultExpandedDepth) and builds the hierarchy by matching record[parentField] against the ids of sibling records of the same object.
That is the org-chart / BOM / category shape. It cannot express the other, at least as common, tree: nest records under synthesized levels derived from field VALUES — unit → owner → duty. Those two levels are not records of the object being listed.
The gap has two halves
1. Inexpressible. grouping: { fields: [...] } is the platform's multi-level value grouping and it works — the grid's grouping hook recurses through grouping.fields building nested subgroups with per-level collapse, which is exactly the wanted behaviour. plugin-tree never reads grouping. So the capability exists in the product; it just is not reachable from the view type whose name promises it.
2. Silent when misapplied. On an object with no self-reference, a type: 'tree' view is not an error and not empty — it is flat. parentField is undefined, the renderer's auto-detect (fields[].type === 'tree', or a lookup whose reference equals the object's own name) finds nothing, every record resolves to a root at depth 0, and you get an ordinary table with an expand gutter that never expands. os validate says nothing: view/layout-without-binding does not cover tree (filed separately), and even if it did, every key in TreeConfigSchema is optional so a present-but-empty tree: {} would satisfy it.
Reproduced on duly_duty, whose lookups point at sys_business_unit, duly_catalog_item and a user — no self-reference. All four gates green.
What we shipped instead
type: 'grid' + grouping: { fields: [{ field: 'business_unit' }, { field: 'owner' }] }. It renders the two-level collapsible hierarchy the card asked for. The cost is that the view type no longer says what the view is, and the tree renderer's indent/expand affordances are not available.
Suggested direction
Either of these closes it; the first is smaller:
- A — let
treeconsumegrouping. Whengrouping.fieldsis present andparentFieldresolves to nothing,plugin-treesynthesizes one level of parent nodes per grouping field, leaf records at the bottom. The authored shape becomes{ type: 'tree', grouping: { fields: [...] }, tree: { labelField } }and reads correctly. - B — declare the mode. Add
groupByFields: string[]toTreeConfigSchemaas the explicit alternative toparentField, and make the two mutually exclusive so a view cannot half-declare both.
Whichever way it goes, the silent-flat half wants a gate on its own: a type: 'tree' whose parentField is neither declared nor auto-detectable on the bound object should warn at author time, because the render is a plausible-looking table rather than an obvious blank.
Related
- The
view/layout-without-bindingtable omittingtree(andtimeline,map) is filed separately.
Activity
os-justin commentedon Sep 1, 2026
分诊 →
needs-user-decision·domain:spec· p3 · Feature。⛔ 本席不代裁:功能新增类在人工地板上,不论四棱是否同向。⭐ 本卡的另一半已经拆走,今天就在做:「无自引用对象上
type: 'tree'静默扁平渲染、四门全绿」是 ADR-0078 的缺陷,不依赖本卡的特性裁决 ⇒ 已折进 #14106(pm:queue·domain:spec,同一文件同一规则族),并在那里写明它需要的是第二条检查而不是把tree加进绑定块表 —— 因为TreeConfigSchema全键可选,一个tree: {}就能骗过绑定块检查。⇒ ⛔ 本卡只剩「能不能表达按值分组的树」这一个特性问题。一句话问题
作者想要「先按部门、再按负责人,层层折叠」的清单。平台能渲染这个东西,但只有普通表格视图能;那个名字就叫「树」的视图类型反而做不到 —— 它只会顺着「谁是谁的上级」这种自我指向的关系搭层级。
选项 × 真实代价
grouping)业务含义直译:A = 「树和表格共用一套分组说法」;B = 「树有它自己的一套分组说法」;C = 「先让写错的人收到提示,分组树本身以后再说」。
四轴(从业务立场)
view/layout-without-bindingcovers only 3 of the 6 view types that have a binding block —timeline,treeandmapfall through the same trapdoor #14106 基本清偿。本卡里真正危险的是「写了tree,静默渲染成扁平表」,那是静默容忍;view/layout-without-bindingcovers only 3 of the 6 view types that have a binding block —timeline,treeandmapfall through the same trapdoor #14106 的第二条检查把它变成作者时的响亮拒绝。⇒ 诊断落地后,③ 不再要求这个特性 —— 它要求的那件事已经在做了。四轴合议:② 弱且已绕过,③ 已被 #14106 清偿,④ 指向不做,① 对做不做中立(只在做的话排 A > B)。⇒ 按分歧推荐序,零/弱拉动 ⇒ 荐 ④ 不扩散。
四棱(标记见上方代码块 —— ⛔ 不用 HTML 注释形态,该形态实测被本通道清洗,见 #14237)
view/layout-without-bindingcovers only 3 of the 6 view types that have a binding block —timeline,treeandmapfall through the same trapdoor #14106 清偿:危险的那半(写tree却静默扁平)正在被修成作者时响亮拒绝;本特性不再为本轴所必需。推荐:C —— 现在不做,关掉或搁置。 回退:若维护者认为「视图类型名要对得上它做的事」本身值得买,则取 A,⛔ 永不取 B。
置信缺口(本分析看不见什么):⛔ 我没有测量树渲染器的缩进/展开外观与表格分组的折叠外观在最终用户眼里差多少。本席把这个差异当「外观」,若它实际影响可用性(例如深层级下表格分组很难看清结构),那 ② 的拉动就被我低估了 —— 这一点只有看过两种渲染的人能判。
裁后我会怎么执行(你不用管)
裁 C ⇒ 关 not planned,关单评论写明理由与出处(重开免费,维护者可否决);#14106 的诊断照常入队,⛔ 不受影响。
裁 A ⇒ 摘
needs-user-decision换pm:queue,派发令里把「两种搭层级方式同时出现时谁赢」列为必须显式作答项(⛔ 不许留成第三次静默),裁决原话逐字引入;触packages/spec/src/**⇒ 条款②路径肢成立,按档位定档。查重:与 #14106(绑定块表短)是同一次实测的两张卡,已互相点名;与 #14074(switcher 白名单可视化是否共享 #13748 形状)相邻但不重复 —— 那张问渲染器侧,本卡问授权面。⛔ 无重复。
Generated by Claude Code
os-project-manager commentedon Sep 2, 2026
Maintainer ruling recorded — C: not now; closed not planned, reopen is free
Director seat, summon #8, session
session_01ShyhexkB2d1AeRZ85tgAAe, 2026-09-02.Provenance (who / verbatim / where): maintainer, live PM chat with the director seat, 2026-09-02, replying to decision batch #3 in which this card was item 4 with the recommendation C (fallback A, never B). Verbatim reply: 「同意」.
Ruled: C. A value-grouped
treeview is not added now. The reporter already ships the wanted two-level collapsible hierarchy astype: 'grid'+grouping.fields; the remaining loss is the view-type name and the tree renderer's indent affordance, and pull is a single application, once, already worked around. The silent-flat half (atreeview on an object with no self-reference rendering as a plain table with all gates green) is not affected by this close — it is carried by #14106, queued indomain:spec, and lands as an author-time diagnostic.If this returns with a second pull, the ruled shape is A (let
treeconsume the platform's existinggrouping, with an explicit answer for the case whereparentFieldandgroupingare both present) — ⛔ never B (a second spelling of value grouping on the tree config).State transition, same stroke:
needs-user-decisioncleared; closed not planned. Ledger: director seat post #12708, summon #8.Generated by Claude Code