`view/layout-without-binding` covers only 3 of the 6 view types that have a binding block — `timeline`, `tree` and `map` fall through the same trapdoor

[os-warren]

Measured on @objectstack/cli 17.2.0 / @objectstack/spec 17.2.0, from a real app (objectstack-ai/duly), while adding kanban / gantt / timeline lenses over one object.

The rule

checkViewCompleteness (spec/dist/kernel — functional-completeness.ts) drives the view/layout-without-binding warning off this table:

var VIEW_BINDING_BLOCKS = {
 kanban: "kanban",
 calendar: "calendar",
 gantt: "gantt"
};

ListViewSchema.type has nine members, and six of them carry a type-specific binding block: kanban, calendar, gantt, timeline, map, tree. The table names three.

Why the other three are the same defect, not a lesser one

The rule's own message is "the renderer falls back to literal default field names … the view renders empty while authoring reports success." That is literally what the uncovered three do. From the console's ListView adapter (plugins-views-*.js, 17.2.0), the fallbacks are hard-coded strings:

view type	renderer fallback when the block is absent	gate
calendar	startDateField || 'start_date', endDateField || 'end_date'	✅ warns
gantt	startDateField || 'start_date', endDateField || 'end_date', progressField || 'progress', dependenciesField || 'dependencies'	✅ warns
kanban	groupBy = groupByField || groupField || <inferred>	✅ warns
timeline	startDateField || 'created_at', titleField || 'name'	❌ silent
map	locationField || 'location'	❌ silent
tree	labelField || titleField || 'name'	❌ silent

timeline is the sharpest of the three: 'created_at' is a plausible-looking name that many objects do not declare, and the timeline renderer drops every row whose start date does not parse — so the view is blank, not merely mis-titled.

Reproduction (measured, both directions)

In a stack with a duly_task object that declares neither created_at nor start_date:

1. listViews.recent = { type: 'timeline', … } with its timeline: { startDateField: 'last_update_at', titleField: 'subject', … } block → os validate --json → warnings: [].
2. Delete the timeline block (confirmed on disk: startDateField occurrences 1 → 0, marker present) → os validate --json → warnings: [] again, valid: true. No diagnostic at all.
3. Same ablation against the sibling gantt view → view/layout-without-binding warning fires as designed.

So the gate is working; the table it reads is short.

Suggested fix

Extend VIEW_BINDING_BLOCKS to the full set and give each the same shaped fix hint the existing three carry:

var VIEW_BINDING_BLOCKS = {
 kanban: "kanban", calendar: "calendar", gantt: "gantt",
 timeline: "timeline", map: "map", tree: "tree",
};

Hints, matching the required keys in each block's schema:

• timeline: { startDateField: '<date_field>', titleField: '<text_field>' } (both required by TimelineConfigSchema)
• map: { latitudeField: '<number_field>', longitudeField: '<number_field>' } or { locationField: '<field>' }
• tree: { labelField: '<text_field>', parentField: '<self_lookup_field>' }

tree is the one worth a second look: TreeConfigSchema's keys are all optional, so "has a tree block" is a weaker assertion there than for the others — see the sibling issue on tree having no value-grouped mode.

Related

• spec: measure whether timeline / gantt / map / other switcher visualizations share the #13748 shape — whitelisted in appearance.allowedVisualizations with no binding block, renderer behavior unmeasured #14074 asks whether the switcher-whitelisted visualizations share the 视图声明 calendar 可视化但缺 calendar 配置块时，平台不报错——直接渲染成一屏错位画面（全部记录堆在今天） #13748 shape. This is the authored-type half of the same root cause; I have posted the renderer measurements there.

[os-justin]

Triage → domain:spec · p2 · Bug. ⚠️ Note the lane split from its four family siblings — this is the one card of the five that does not land in packages/lint.

Anchoring — measured on origin/main (66ecc50a). VIEW_BINDING_BLOCKS and the layout-without-binding rule live in packages/spec/src/kernel/functional-completeness.ts (confirmed: 2 files carry the symbol, both under packages/spec/src). The card's own citation of "spec/dist/kernel" is the built artifact; the source is the above. ⇒ domain:spec.

⇒ ⚠️ Its four siblings (#14105 · #14107 · #14108 · #14148) all land in packages/lint ⇒ domain:devx. Same defect class, two lanes. ⛔ They cannot be folded into one PR across that boundary.

Type = Bug. ADR-0078 (no silently inert metadata) is the declared invariant; the rule exists and enforces it for 3 of the 6 view types that carry a binding block. Declared ≠ enforced on the uncovered three ⇒ Bug. The change narrows the accept set, so it is ⛔ not a Feature.

Grade p2, and flag as the family's cheapest fix. The remedy is extending a three-entry table to six plus three fix hints; the card supplies both directions of the measurement (ablate timeline → silent; ablate the sibling gantt → warns as designed), so the gate is proven working and the table proven short. timeline is the sharpest member — 'created_at' is a plausible-looking default that many objects do not declare, and the renderer drops every row whose start date fails to parse ⇒ blank, not merely mis-titled.

⚠️ Dispatch-side note for the domain:spec seat — Clause ② path limb

packages/spec/src/** is the Clause-② path limb. This card's diff lands there by construction ⇒ the path limb trips regardless of how narrow the change is, and the dispatch tier must be priced accordingly (tier single-source = CONTRACT_REVIEW_TIER, ⛔ not a model name written here).

⛔ This is not me pre-hanging needs:contract-review — forward pre-hanging was retired by the 2026-08-28 ruling; the label goes on with a reviewable contract increment at PR time. Recording the tier consequence only so the card is not priced as a trivial table edit.

The tree caveat is real and should survive into implementation. The card flags that TreeConfigSchema's keys are all optional, so "has a tree block" is a weaker assertion there than for the others. ⇒ adding tree to the table buys less than adding timeline/map, and the sibling card #14109 (a tree cannot express value-grouping at all) is the reason why. ⛔ Do not let the weaker tree case stall the strong timeline/map half.

Dedup: related-but-distinct from #14074 (whether switcher-whitelisted visualizations share the #13748 shape) — that is the renderer half, this is the authored-type half; the reporter has already cross-posted measurements. ⛔ No duplicate.

---

Generated by Claude Code

[os-justin]

范围补充 —— 把 #14109 的「静默扁平」半边折进本卡

分诊席补记。本卡入队时定级/路由不变(domain:spec · p2 · Bug),这条只加一项范围。

#14109 有两个半边:① tree 表达不了「按字段值分组的树」(能力扩张 ⇒ 已入决策箱);② 在没有自引用的对象上,type: 'tree' 视图既不报错也不空 —— 它扁平渲染,带一个永远展不开的展开槽,四道门禁全绿。

半边 ② 是 ADR-0078 的同一条不变量,与本卡同族、同文件,且不依赖那个特性裁决 ⇒ 折进本卡,⛔ 不另立卡、⛔ 不等裁。

⚠️ 但它不是「把 tree 加进 VIEW_BINDING_BLOCKS」就能解决的 —— 本卡正文已经预告了这点

本卡自己写着:TreeConfigSchema 的键全部可选,所以「有没有 tree 块」对 tree 而言是比对其它五种更弱的断言 —— 一个 tree: {} 就能满足绑定块检查。#14109 独立测到了同一件事的另一面:parentField 缺席时渲染器会去自动探测(fields[].type === 'tree',或一个指向对象自身的 lookup),探测不到就每条记录都是深度 0 的根。

⇒ ⭐ 两条检查,⛔ 不是一条:

1. 绑定块在不在 —— 本卡主体,把表从 3 扩到 6。对 timeline / map 足够(它们的必填键是真必填)。
2. tree 的父指针解不解析得出来 —— parentField 既未声明、也无法在绑定对象上自动探测 ⇒ 作者时告警。这一条是 tree 专属的,因为只有 tree 的块可以整体为空还「合法」。

⛔ 只做 1 而宣称覆盖了 tree,会留下一个通过了新门禁却依旧静默扁平的视图 —— 那比不加检查更糟,因为门禁会为它背书。

验收补一条:一个绑定在无自引用对象上的 type: 'tree' 视图(tree 块存在但为空,或整个缺席)必须在 os validate 上产生诊断;夹具覆盖失败形与干净形两向。

为什么折而不是串行

同一文件(packages/spec/src/kernel/functional-completeness.ts)、同一规则族、同一 ADR、同一位报告者、同一次实测会话 ⇒ 家族派发五门里前四门直接成立;排除清单只有一条:#14109 的能力半边 ⛔ 不在本卡范围内(它在决策箱等裁,⛔ 实施者不得顺手实现)。

⚠️ 若 spec 席判断这条把本卡从「扩个表」变成了两种不同逻辑因而不该折,那是该席的 fold-or-serial 裁量 —— 请显式作答并回帖,⛔ 不要默认串行(未作答的默认串行是漏答,不是安全选项)。

---

Generated by Claude Code