feat(spec)!: ObjectSchema 在 parse 路径上收紧，而不只是 create() —— #1535 的奠基例子一直是活的（#4001 批 3a） - #4522
Merged
Merged
Conversation
…e() (#4001) #1535's founding example was still reproducible on the path that matters. ObjectSchema.parse ({ …, workflows: ['x'] }) → silently discarded ObjectSchema.create({ …, workflows: ['x'] }) → rejected since #1535 #1535 built the guard as a hand-rolled check inside the `create()` factory, reasoning that authored `*.object.ts` modules call `create()`. They do — but they are not the only producer, and not the path most instances travel: `defineStack({ objects })`, `/api/v1/meta/types/object` and the Studio form all reach this schema through `parse()`/`safeParse()`. So object-level `workflows` — an author believing they had wired up automation and shipping dead metadata, the example this whole campaign was filed on — kept being eaten in silence for the entire time #1535 was considered fixed. The base shape is now `.strict()`, carrying the `UNKNOWN_KEY_GUIDANCE` tombstones plus the semantic renames the WARNING layer already knew (`capabilities`/`features` → `enable`). Graduating a surface from warn to reject must not cost the author a prescription — edit distance cannot reach those two, so they are named explicitly. `create()` is unaffected: its own check runs before parsing and throws a richer located Error. Safe on the read path for the reason the other closed registered types are: the ADR-0010 envelope is declared, and `stripReadDecorations` removes `_diagnostics`/`_draft` before any strict re-parse (cloud#971). Verified rather than assumed — every `ObjectSchema.create()` call across `platform-objects` and the three example apps uses only declared top-level keys. New tombstone: `namespace` (retired in ADR-0006 D4) had none, so it was stripped in silence. `{ namespace: 'sys', name: 'user' }` shipped as plain `user`, under a name its author never intended. The rejection now carries `name: "sys_user"`. AND a coverage regression this change would otherwise have introduced. The unknown-key warning layer gated each metadata collection on its ROOT schema's posture, so closing `object` at the root would have switched off the warnings for everything BENEATH it — its 71 nested strip-mode sites stopping in the same commit, with nothing anywhere to say so. Posture is a per-node property; the walk now treats it as one. A strict root stays silent at its own level (the parse owns that failure) and the descent continues. Nested `object.fields.*` warnings are unaffected by the graduation. That is the fourth instance in this campaign of one shape: an improvement in one place quietly removing coverage in another. Three tests that asserted the strip as correct behaviour are now rejection tests. The `compactLayout` one had pinned the author-hostile outcome in so many words: "the retired key is STRIPPED, not aliased — an old-key author gets no highlightFields rather than silently working." Verified: spec 283 files / 7159 tests, `tsc --noEmit` clean, all 8 generated artifacts current, all 15 `check:*` gates green. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WnqGjQFQMqd5k81LYV8SCY
|
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
github-actions
Bot
added
documentation
Contributor
📓 Docs Drift CheckThis PR changes 1 package(s): 107 hand-written doc(s) reference the affected code and may need an implementation-accuracy re-verification:
|
…authz-surface-s8vnok
os-zhuang
marked this pull request as ready for review
August 1, 2026 16:23
This was referenced Aug 1, 2026
akarma-synetal
pushed a commit
to akarma-synetal/framework
that referenced
this pull request
Aug 2, 2026
…ctstack-ai#4001 批 5） (objectstack-ai#4529) * feat(spec)!: translation closes at both doors, and objectstack-ai#3778's ten-key guard retires into the message (objectstack-ai#4001) A translation that resolves to nothing is indistinguishable from a translation nobody wrote — no wrong string appears, just the source language, forever. So this type had the most literal version of the silent-strip failure in the spec. objectstack-ai#3778 already knew that, and fixed it for ten keys: a `z.preprocess` scanning for the retired object-first dialect. It had the shape every workaround for `.strip` has — it caught only the mistakes someone had already thought of, and it ran on the item door only, so the same ten keys in a file-authored bundle were dropped in silence. Same asymmetry objectstack-ai#4522 found in objectstack-ai#1535's object guard. The guard is now redundant and gone; its ten prescriptions ride the rejection as `guidance`. What was worth keeping was never the detection — detection generalizes for free once the default flips — it was the prose. Closed across every authorable group (objects/fields/views/actions/sections, apps, dashboards, pages, settings, metadata forms) and the i18n config, whose four objectstack-ai#3494-removed knobs get tombstones. Two gates were found doing half their job: - `translation` came off the ADR-0010 envelope debt list (down to four). - `metadata-create-seeds.test.ts` — the canonical "create shape ≠ spec" guard — asserts every seed parses. The `translation` seed ships `{ name, label, locale, objects }` and the type declared neither `name` nor `label`, so two thirds of the authoritative create shape was stripped while the gate reported green. A gate on a `.strip` schema catches a missing required key and never an extra undeclared one. Registered types closed at the top level: 17 of 25. Verified: 284 files / 7208 tests, tsc clean, 8 generated artifacts current, 15 check gates green, and the real bundles in examples/app-crm, app-todo and platform-objects all parse at module load. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WnqGjQFQMqd5k81LYV8SCY * docs(i18n): the retired-dialect rejection now covers both doors, not just the metadata one (objectstack-ai#4001) Three passages described objectstack-ai#3778's guard accurately for what it was — ten keys, rejected at the metadata door — and objectstack-ai#4001 changed both halves of that: any undeclared key is rejected, in a runtime item and in a file-authored bundle. - ui/translations.mdx: "only the groups on this page are accepted" was aspirational for bundles; it is now literally true. Says why this surface cares more than most — a translation that resolves to nothing looks exactly like one nobody has written yet. - i18n-standard.mdx: same correction on the retired-dialect callout. - i18n-standard.mdx: the `translationService` design-intent callout already warned the key is unrecognized; copying that snippet is now a build-time rejection rather than a silent drop, which is the part a reader acts on. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WnqGjQFQMqd5k81LYV8SCY --------- Co-authored-by: Claude <noreply@anthropic.com>
akarma-synetal
pushed a commit
to akarma-synetal/framework
that referenced
this pull request
Aug 4, 2026
…e 关 (objectstack-ai#4721) (objectstack-ai#4922) * feat(spec,metadata-protocol)!: reject a sort node spelling its direction `direction` (objectstack-ai#4721) `SortNodeSchema` was a plain `z.object`, so zod's `.strip` default applied: SortNodeSchema.parse({ field: 'updated_at', direction: 'desc' }) → { field: 'updated_at', order: 'asc' } The foreign key was discarded, `order` fell back to `asc`, and the sort ran the OPPOSITE way under an ordinary success — with `limit`, a different set of rows, with no signal anywhere in the response. `direction` is not a typo: it is `IReportService.orderBy`'s live vocabulary, which plugin-auth's objectql adapter already translates by hand. A translation known to be necessary and enforced nowhere is the ADR-0049 shape. Both doors onto that shape are closed here, in one change: - `SortNodeSchema` (spec/src/data/query.zod.ts) → `strictObject` with `aliases: { direction: 'order' }`, so the rejection carries the translation. Edit distance can never bridge `direction` → `order`, so a bare "unrecognized key" would leave the caller where the silent strip did. - `normalizeSortNodes` (metadata-protocol/src/protocol.ts) — the ingress every external `orderBy` funnels through — refuses `{ field, direction }` with 400 INVALID_SORT naming `order` and quoting the corrected node. Closing only the schema would repeat the objectstack-ai#1535/objectstack-ai#4522 door asymmetry: `SortNodeSchema.parse` is reachable by three paths the REST normalizer never sees, and the normalizer runs ahead of any QueryAST parse. Deliberately NOT in scope: `QuerySchema`'s top level stays non-strict (`QuerySchema.safeParse({object:'sales', nonsenseKey:1}).success === true`) — tracked in the objectstack-ai#4001 campaign map for its own batch. The `{field: direction}` map form is untouched: there `direction` is an ordinary column name, and refusing it would be the mirror-image bug. Strictness ledger: `query.zod.ts` keeps its `open` class for the four dialect sites; `SortNodeSchema` is carved out as authorable (4 strip of 5), which also resolves the recorded classification conflict — the FILE was the wrong unit. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ehu85kbvMcrNTUJjwxvLJ9 * docs(spec): keep SortNodeSchema's page description short; rationale moves to line comments `build-docs.ts` takes the FIRST `/** */` block in a `.zod.ts` file as the reference page's description, and joins every line of it with a blank line. The long objectstack-ai#4721 rationale therefore rendered as a 56-line wall at the top of `content/docs/references/data/query.mdx`, where a customer reads what a sort node IS — not why one schema in the file is strict. The prose is unchanged, it is just `//` instead of `/** */` so the generator cannot pick it up, with a note at the top saying why it must stay that way. Regenerated: query.mdx (two lines), and the two skill reference indexes, which grew transitive entries because query.zod.ts now imports shared/strict-object. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ehu85kbvMcrNTUJjwxvLJ9 --------- Co-authored-by: Claude <noreply@anthropic.com>
akarma-synetal
pushed a commit
to akarma-synetal/framework
that referenced
this pull request
Aug 6, 2026
批 19/20) (objectstack-ai#5250) * feat(spec)!: object 内层块拒绝未知键 —— 13 个站点收紧,IndexSchema 依据实测保留 (objectstack-ai#4001 批 19/20) `data/object.zod.ts` 的顶层自 objectstack-ai#1535/objectstack-ai#4519/objectstack-ai#4522 起就会拒绝未知键,而它下面 的内层块一直是 zod 默认的 `.strip`。这个不对称正是本批的价值所在:一个已经 见过顶层拒绝拼写错误的作者,完全有理由把 `lifecycle: { maxAge: '30d' }` 的 干净解析读作"接受"。object 是本仓作者写入量最大的元数据类型。 批 20 —— 14 个站点关掉 13 个: - `access`(ADR-0066 D2)、`lifecycle` 及其四个子块(retention/ttl/storage/ archive)、`fieldGroups[]`(ADR-0085)、`external`(ADR-0015)、 `userActions`、`systemFields`、`activityMilestones[]`、`publicSharing`、 `objectExtensions[]`。 - 可达性是实测而非假定:从 24 个元数据类型根加 `ObjectStackSchema` 做 BFS (4810 节点 / 25 根),每一个都按**身份**判定 `direct`,没有一个依赖 objectstack-ai#5056 发现的 `derived-clone` 桥;同一次运行里 `ObjectSchema`/`PageSchema` 为正控制,批 14 的 `EmbedConfigSchema` 为负控制(unreachable,重叠 0.00)。 - parse 门单独探针、逐路径验证:今天顶层 REJECT、13 条嵌套路径全部静默 STRIP,所以每一次收紧转化的都是实测到的静默丢弃,而不是猜测。⚠️ `IndexSchema` 有意保留,这个"不动"才是本批的发现(objectstack-ai#5114 类,这次是在 发布**之前**抓到而不是之后)。console 为这个形状自带了一份手抄 JSON-Schema (objectui `metadata-admin/EmbeddedItemEditor.tsx`)—— 因为 `index` 是框架 不发布 schema 的内嵌子类型 —— 而这份抄件已经漂移:它用 `where` 表示部分索引 谓词,spec 声明的是 `partial`。编辑器把表单结果拼回 `object.indexes[]` 后 PUT 整个对象,`saveMetaItem` 逐字保留 body 并做校验,所以单独关掉这个形状会 让 console 自己渲染的控件 422。该能力两个方向上其实都已经死了 —— `driver-sql` 的 `syncDeclaredIndexes` 只读 `name`/`fields`/`unique` —— 这恰恰是必须先修生产者、并给 `type`/`partial` 一个 ADR-0049 结论的原因: 今天把作者指向 `partial`,是一条声称超出平台实际交付的 guidance(finding 18)。 批 19 —— `ui/app.zod.ts` 的 `BaseNavItemSchema`:`verify` 检查结论为否, 不改姿态。台账假定成员 `.extend()` 该基底(finding 16 的陷阱),实际它们 spread `...shape`;两者机制不同,并且双向实测过。九个分支各自已经 `.strict()`, 基底是模块私有且从不 parse,关掉它是保证的 no-op —— 而 objectstack-ai#4583 明确指出 no-op 收紧并非中性。`Class` 单元格未改动:它是机读的,而八个既有判定词 没有一个对"既不是门也不是死物"的形状是诚实的,加第九个属于改机读契约。 每条 alias/guidance 都是对 schema 的断言(finding 18),因此逐条锚定到具名 的兄弟契约并在测试中断言;`fieldGroups[]` 三个已弃用的 collapse 别名保持 **接受** —— 收紧一个形状不该把有文档的弃用变成拒绝。 已知并如实钉住的一处:`systemFields` 是 `false | {…}` 联合,拒绝信息是 `invalid_union` 的裸 "Invalid input"(objectstack-ai#5014 摊平);批 18 的 `discriminatedUnion` 解法在这里不可用(一个分支是字面量,没有判别键)。 三处记录齐备:schema JSDoc + `object-strictness-batch20.test.ts` / `app-strictness-batch19.test.ts` + 台账散文行;计数由 os-regen 重新生成。 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ehu85kbvMcrNTUJjwxvLJ9 * build(spec): 合并 origin/main 后重算 strictness 计数(os-regen) merge 带入 objectstack-ai#5221 的 chart 站点(482 → 483),计数文件按 merge=os-regen 的 约定从合并后的树重新生成,而不是任何一边的算术。authorable 仍为 16。 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ehu85kbvMcrNTUJjwxvLJ9 * docs(spec): 把批 19/20 的两个保留判定接到已归档的 issue 上(objectstack-ai#5247/objectstack-ai#5248/objectstack-ai#5249) IndexSchema 的保留(生产者漂移 + ADR-0049)与 BaseNavItemSchema 的 Class 词汇缺口都已单独归档,台账行与 schema JSDoc 现在指向 issue,而不是只说 '交给维护者'——一条指不出去的判定,下一个人只能重新测一遍。 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ehu85kbvMcrNTUJjwxvLJ9 * build(spec): 合并 objectstack-ai#5237 后重算 strictness 计数(os-regen)—— authorable 16 + (-1) → 15 合并队列以 MERGE_CONFLICT 踢出后重新同步 origin/main(merge,非 rebase)。 objectstack-ai#5237(objectstack-ai#5073 allowAddTab)与本批从两个不同目录改这份台账,散文行不重叠, git 干净合并;数字零手改,counts.md 按 merge=os-regen 从合并后的树整体重生成。 重算结果 15 = `ui/` 7 → 6(objectstack-ai#5237)+ `data/` 22 → 9(批 20),两边分支都没写过 这个数——正是 objectstack-ai#5107 拆分要达到的效果。这一条也补进 data/ 段落:它是第十二次 实例,也是第一次没让任何人付出代价。 objectstack-ai#5237 对 `ui/app.zod.ts` 的改动(`apis` 墓碑文案)与本批的 BaseNavItemSchema 判定互不相干,两侧都在;它对 scripts/strictness-ledger.test.ts 的 view 计数 5 也是它自己的领地。 门:spec 308 文件 / 7929 测试通过;typecheck 通过;check:generated 9/9; check:strictness-ledger 通过(484 站点,authorable 剩 15)。 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ehu85kbvMcrNTUJjwxvLJ9 --------- Co-authored-by: Claude <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
#4001 的批 3 前半（清单）。
#1535 的守卫没盖住真正要紧的那条路
#1535 把未知键守卫做成了
create()工厂里的手写检查，理由写在代码里：「作者写的*.object.ts模块都调create()」。这话没错，但它们不是唯一的生产者，也不是大多数实例走的路：
defineStack({ objects })、/api/v1/meta/types/object、Studio 表单，三条全部经由parse()/safeParse()。于是对象级
workflows: [...]——作者以为接好了自动化、实际发布了一份死元数据，本 issue 正文举的那个例子——在 #1535 被认为已修复的整段时间里，一直在主路径上被静默吃掉。现在 base shape 是
.strict()，带上UNKNOWN_KEY_GUIDANCE的墓碑，以及告警层早就知道的语义改名（capabilities/features→enable）。毕业不该让作者少拿一条处方——这两个编辑距离够不着，只能显式命名。create()不受影响：它自己的检查在 parse 之前跑，抛的是更详细的定位错误。读取路径安全，理由和其他已关闭的注册类型一样：ADR-0010 信封已声明，且
stripReadDecorations会在任何严格重解析前剥掉_diagnostics/_draft（cloud#971）。实测而非推断——platform-objects加三个示例应用里所有ObjectSchema.create()调用，顶层键全部已声明。新墓碑：
namespace它此前没有墓碑，所以被静默剥离。
{ namespace: 'sys', name: 'user' }会以user发布——作者以为自己声明了命名空间，实际对象名不是他要的那个。拒绝信息现在直接给出name: "sys_user"。顺带：修掉一个本改动本会引入的覆盖回归
未知键告警层是按根 schema 的姿态决定整个集合走不走的。所以
object顶层一收紧，整个objects集合被排除——它下面 71 个仍是 strip 的嵌套站点会在同一个 commit 里停止告警，而没有任何地方会说这件事。姿态是每个节点的属性，走查现在按节点处理：strict 的根在自己那层沉默（解析已经在报，再警告就是重复），下探照常。
object.fields.*的嵌套告警不受这次毕业影响。这是本战役第四次撞到同一个形状：一处改进悄悄拿掉另一处的覆盖。
三个「把静默剥离钉成预期」的测试改写了
namespace/ 退役的compactLayout别名 / 移除的detail块。其中compactLayout那个把对作者不利的结局写得很直白：和 step 2 的
position.test.ts（断言一个不存在的层级）是同一类：测试把 strip 时代的假象固化成了正确行为。验证
@objectstack/spec：283 文件 / 7159 用例通过，tsc --noEmit干净check:*全绿object收紧后objects.acct.fields.amt.lenght仍然报告；根层不再重复报告参考
workflows: [...](and any unknown ObjectSchema key) is silently stripped at build — no error/warning (ADR-0032 'no silent failure', metadata layer) #1535（本 PR 补完的那个修复）、排查「手抄 spec 清单 + "keep in sync" 注释」模式:一天内确认三例,全部曾静默漂移 #3786 / feat(spec): 未知键 lint 下探到嵌套元数据（补上 227 个此前无声的站点） #4218（告警层）namespace退役）、ADR-0010（保护信封）、ADR-0085（compactLayout/detail）🤖 Generated with Claude Code
https://claude.ai/code/session_01WnqGjQFQMqd5k81LYV8SCY
Generated by Claude Code