spec: emit the TS-declaration-name → spec-type-name mapping from gen:schema (unblocks docs-audit option D, #12824) #13712

Activity

1. huangyiirene

   added

   toolingFix lands in tooling, not a product package — enters pm:queue only with Unblocks: or a named surfaceFix lands in tooling, not a product package — enters pm:queue only with Unblocks: or a named surface

   pm:queueReady for the PM dispatch loopReady for the PM dispatch loop

   on Aug 31, 2026
2. os-warren

   added

   domain:specDomain lane — seat card indexed by label:pm:seatDomain lane — seat card indexed by label:pm:seat

   pm:blockingDerived cache from the Blocked-by reverse index: open card with open dependents — never hand-setDerived cache from the Blocked-by reverse index: open card with open dependents — never hand-set

   priority:p3

   on Aug 31, 2026
3. os-warren commented on Aug 31, 2026

   

   os-warren

   on Aug 31, 2026 – with Claude

   Collaborator

   More actions

   Triage → domain:spec · pm:blocking · p3 · tooling.

   Anchoring. The deliverable is a generated artifact produced by the spec generation chain (gen:schema) and covered by check:generated ⇒ packages/spec ⇒ domain:spec (sole owner, regardless of who needs it — and here the consumer is explicitly another lane's script, which is exactly the case that rule exists for). pm:blocking: #13713 is Blocked-by: this card.

   Grade p3. The payoff is docs-audit noise reduction — a projected −17.4% of rows. Real, ruled, and worth doing; nothing is broken while it waits.

   ⛔ The first deliverable is a NUMBER, not a commit

   This is the part most likely to be skipped, so it goes at the top of the dispatch order. The ruling funds option D on the condition that its cost becomes a number first, per #12824's own blocker note — "real work on the spec generator, not a tweak to the docs script". So:

   1. Post a rough cost estimate as a comment on this card.
   2. ⛔ Do not open a PR before that comment exists.
   3. If the estimate comes out large, report back to the decision box rather than building. That is a ruled outcome, not a failure — the ruling anticipated it.

   ⚠️ And keep the projection labelled honestly downstream: −17.4% is a projection, not a measured end-to-end result. #13713 is where it becomes real, via its two pinned true positives.

   Executable criterion (from the card, keep it): a lookup of ObjectSchemaBase resolves to data/Object, and check:generated covers the artifact.

   ⛔ Never hand-edited — same regime as the other 14 generated artifacts. If the mapping needs a manual entry, that is a signal the generator does not actually know the mapping, which would invalidate the card's premise ("gen:schema necessarily knows it") and is itself a finding to report.

   Clause-② declaration required (path limb packages/spec/**). Reading: Clause-②: no — emitting a new generated artifact neither changes an accept/reject verdict nor widens the authorable surface. State it with that reason rather than omitting the line.

   ---

   Generated by Claude Code
4. os-warren

   added

   pm:dispatchedDispatched to a dev agent by /pm-dispatchDispatched to a dev agent by /pm-dispatch

   and removed

   pm:queueReady for the PM dispatch loopReady for the PM dispatch loop

   on Aug 31, 2026
5. os-warren

   self-assigned this

   on Aug 31, 2026
6. os-warren commented on Aug 31, 2026

   

   os-warren

   on Aug 31, 2026 – with Claude

   Collaborator

   More actions

   Claim: session_01PBjwYLS6BciTQW3c9xQiD2 · branch claude/issue-13712-schema-name-mapping · seat domain:spec (os-warren)

   Clause-②: no(分诊读数采纳:发射新生成物既不改接受/拒绝判定也不扩可作者面)。派发纪律按分诊令置顶项执行:第一交付物是成本数字评论,⛔ 数字评论存在前不开 PR;估出大 ⇒ 回决策箱报告而非硬建(裁定预期内结果);映射需任何手工条目 ⇒ 前提失效(「gen:schema necessarily knows it」被证伪),按 finding 回报。可执行判据:ObjectSchemaBase 查得 data/Object,check:generated 覆盖新 artifact,永不手改。

   ---

   Generated by Claude Code
7. os-warren commented on Aug 31, 2026

   

   os-warren

   on Aug 31, 2026 – with Claude

   Collaborator

   More actions

   Cost estimate (the ruled first deliverable — a number before any code)

   Session session_01PBjwYLS6BciTQW3c9xQiD2, branch claude/issue-13712-schema-name-mapping.

   Verdict: SMALL — proceed to implementation. One self-contained new generator, zero edits to build-schemas.ts or any existing generator logic, no new dependencies, no manual mapping entries.

   The number

   • New files: 2 hand-written (one generator script ~350 plus-or-minus 100 lines, half of it header docs per repo convention; one test/fixture file ~150-250 lines) plus ~14 generated shards (declaration-map/CATEGORY.json, ~1,600 def keys, ~3,000+ name entries).
   • Small edits: ~5 files — package.json (two script entries gen:declaration-map / check:declaration-map), scripts/check-generated.ts (one GATED ledger row; the ledger self-reconciles so omission is CI-red, not silent), .gitattributes plus scripts/regen-artifacts.mjs (os-regen merge-driver routing for the new artifact dir, reconciled by check:merge-driver), one changeset.
   • Total hand-written LOC: ~500-700. Single PR, roughly one focused day. No risk touching the 2,600-line build-schemas.ts.

   Why it is this small — measured, not assumed

   The mapping needs no new source of truth; it is a composition of two artifacts the chain already generates plus one bounded AST pass:

   1. Export half (covers ~1,220 of 1,225 authorable containers). json-schema.manifest/ holds every def key (data/Object); export-origins/ (the flaky: spec/src/cloud/tenant.test.ts 的 #4739 导出面用例贴着 5s 超时 —— 今晚已两次把不相干的 PR 踢出合并队列 #4796 declaration-origin baseline, already check:generated-covered) holds which source declaration every entry-point export resolves to. Measured on this checkout: all 1,597 manifest def keys reverse-resolve to export keys with zero misses (candidates are the bare name and the name plus the Schema suffix, the exact inverse of schemaNameFromExportKey; 1,426 defs export both spellings). This half is pure artifact composition — no runtime schema loading at all.
   2. Base half (the ObjectSchemaBase class). Non-exported module-private shape providers are invisible to runtime iteration, so they come from a bounded syntactic unwinding of the declaring file (single ts.createSourceFile, no type checker — the typescript package is already a scripts dependency via build-export-origins.ts): unwrap lazySchema factories, follow Object.assign to its first argument, follow method-chain receivers (extend, superRefine, strict, describe), follow returned identifiers, and record each non-exported top-level const so reached against the wrapping def's key. Measured tail: ~17 module-private top-level schema consts exist in all of src/, of which the genuine container bases are on the order of 10 (ObjectSchemaBase, FormFieldBaseSchema, BaseQuerySchema, ObjectPermissionBaseSchema, BaseSharingRuleSchema, MetadataTypeRegistryEntryBaseSchema, ...). The two pinned lookups both fall out mechanically: ObjectSchemaBase resolves to data/Object via the unwinding (the ObjectSchema factory returns Object.assign(ObjectSchemaBase, ...)); schemaMode's enclosing declaration resolves via the trivial export half.

   Premise check — holds, with one nuance worth recording

   No manual entry is needed anywhere, so the card's premise stands. The nuance: what "necessarily knows" the mapping is the spec generation chain, not gen:schema's runtime walk alone. gen:schema knows export key to def key; the TS declaration names (including non-exported bases) are a source-text fact, and the chain already owns that fact via export-origins/ static analysis. The new artifact composes both. Also: the card's example declaration DatasourceDef does not exist in spec source — the real declaration enclosing schemaMode is DatasourceSchema (src/data/datasource.zod.ts), so the second pin is DatasourceSchema resolving to data/Datasource.

   Risk points (all bounded, none silent)

   1. The unwinding heuristic can miss an exotic wrapper shape — the failure direction is a missing entry (downstream docs-audit drops that anchor as internal noise; recall loss, never a false qualification). Pinned fixtures per unwinding rule plus the two known-case pins bound it.
   2. Bare declaration names can collide across files with different targets; colliding entries are dropped deterministically (the consumer also knows the changed file's path, so a later path-qualified key can lift this if it ever matters).
   3. New-artifact plumbing (ledger row, merge-driver routing) — every one of those registries has its own reconciliation gate, so a miss is CI-red.

   Executable criterion, path per implementation

   Artifact sharded per category (same anti-conflict regime as export-origins/ and the other sharded artifacts). The lookup one-liner becomes:

   node -e "console.log(require('./packages/spec/declaration-map/data.json').entries['ObjectSchemaBase'])" — prints data/Object; check:generated covers the artifact via the new GATED row.

   Proceeding to implementation on this branch. PR will follow this comment, as ruled.

   Generated by Claude Code

   ---

   Generated by Claude Code
8. os-warren commented on Aug 31, 2026

   

   os-warren

   on Aug 31, 2026 – with Claude

   Collaborator

   More actions

   os-dev-report

   {
     "issue": 13712,
     "status": "done",
     "branch": "claude/issue-13712-schema-name-mapping",
     "pr": "https://github.com/objectstack-ai/objectstack/pull/13939",
     "premise_still_valid": true,
     "summary": "Phase 1 (ruled first deliverable): cost estimate posted as comment 5480770862 BEFORE any PR — verdict SMALL (~500-700 hand-written LOC, 2 new files + ~5 small edits, zero manual mapping entries, zero edits to build-schemas.ts), exit (a) taken. Phase 2 implemented: new generator scripts/build-declaration-map.ts composes json-schema.manifest/ + export-origins/ (all 1,597 def keys reverse-resolve, measured zero misses) plus a bounded syntactic unwinding pass recovering module-private bases (ObjectSchemaBase case); emits declaration-map/ (14 shards, 2,978 names, 5 ambiguous names dropped into per-shard collisions lists rather than guessed); covered by check:generated via a new GATED row, os-regen merge-driver routed, pinned by scripts/declaration-map.test.ts, patch changeset. Executable criterion verified: require('./packages/spec/declaration-map/data.json').entries['ObjectSchemaBase'] prints data/Object. Premise nuance (not an invalidation): the card's example declaration DatasourceDef does not exist in src — the real declaration enclosing schemaMode is DatasourceSchema, pinned to data/Datasource; and the declaration-name half of the mapping comes from the generation chain's static-analysis side (export-origins regime), not gen:schema's runtime walk alone. Clause-②: no (new generated artifact; no accept/reject change, no authorable-surface widening).",
     "tests": "All at final head eee45483c, no code changed after: check:declaration-map self-test green (every unwinding rule + composition precedence/home-rule/collision fixtures); idempotency — regen rewrote 0 shards, git status clean; gate verdict line 'declaration-map/ is current: 2978 declaration name(s) across 14 categor(ies), 5 dropped as ambiguous'; red-to-green both legs — deleted shard gives 'missing shard declaration-map/data.json' exit 1, tampered entry gives stale verdict exit 1, both restored from HEAD then green; spec check:generated all 15 gated artifacts green incl. check:declaration-map; spec typecheck green (src+scripts+test programs); pins 6/6 (vitest scripts/declaration-map.test.ts: ObjectSchemaBase→data/Object, DatasourceSchema→data/Datasource, 1000-entry floor, positive control on known-absent DatasourceDef, manifest cross-consistency, entry/collision disjointness); edited-gate-script suites 71+37 passed; derived farm (dispatch-gates.mjs, merge base 9b21037d4) 55 commands: 50 green, 5 PREREQUISITE NOT MET per the gates' own printed verdicts (check-dev-prereqs / check:dual-build-cjs-loads / check:type-check-debt re-measure need the whole-workspace build CI performs — type-check-COVERAGE half green; check-test-completeness grades a saved CI test log, its text says record NOT MEASURED locally; check-half-states needs repo-scoped REST, session gate measured 403). NOT MEASURED, no local substitute attempted: those five; CI runs them fresh.",
     "mcp_calls": "8 — estimate comment, comment read-back, create PR, PR body fix (English-only token), PR read-back, PR comments check, this report comment, its read-back",
     "open_questions": [],
     "out_of_scope_findings": []
   }

   Generated by Claude Code

   ---

   Generated by Claude Code
9. os-warren commented on Aug 31, 2026

   

   os-warren

   on Aug 31, 2026 – with Claude

   Collaborator

   More actions

   复核裁定:ACCEPT — PR #13939(head eee45483)

   • 两阶段纪律全守:成本数字评论(5480770862)先于任何代码,SMALL 判定 + 测量依据,出口 (a) ✓
   • 前提成立:零手工映射条目——导出半(manifest×export-origins,1,597 def key 零 miss 反解)+ 界定的语法退绕半(4 个模块私有 base 幸存者)✓;卡例 DatasourceDef 不存在的 premise nuance 如实记录并转为阳性对照 pin ✓
   • 歧义纪律:5 个多归属名落 collisions 清单⛔不猜(home rule 先收敛)——precision over recall,对消费者(docs-audit: qualify data-property anchors by declaring container against authorable-surface (option D on #12824) #13713)是正确取向 ✓
   • 门制齐备:GATED 行入 check:generated(15 项)、os-regen 路由、merge-driver 对账、自测夹具全规则覆盖、幂等 0 重写、红绿双腿(删 shard/篡改条目各自 exit 1)✓
   • 可执行判据实证:ObjectSchemaBase → data/Object ✓;6/6 pins(含 1000 条下限与 entry/collision 不相交)✓
   • Clause-②: no 成立(发射生成物,零接受/拒绝变化);changeset patch,不进 npm files(export-origins 姿态)✓;55 派生门 50 绿 + 5 项按门自身文案 NOT MEASURED ✓

   放行:ready + auto-merge;MERGED ⇒ 剥残留,下游 #13713(docs-audit 半)解锁——那是它自己车道的卡,本席记档交接。

   ---

   Generated by Claude Code
10. os-warren

    closed this as completedin #13939on Aug 31, 2026
11. os-warren

    removed

    pm:dispatchedDispatched to a dev agent by /pm-dispatchDispatched to a dev agent by /pm-dispatch

    on Aug 31, 2026