test(driver-turso): drop the dead as never on date-bucket-parity's create options gate (#6394)

[os-zhuang]

Fixes #6394

What changed

One cast, in packages/drivers/driver-turso/src/date-bucket-parity.test.ts:

         await driver.create(
           'bucket_storage_probe',
           { id: 'p1', at: new Date('2026-01-10T09:00:00Z') },
-          { bypassTenantAudit: true } as never,
+          { bypassTenantAudit: true },
         );

That is the entire diff — 1 file, +1/−1.

Premise verified before implementing

• The cast still exists on current origin/main, located by pattern (as never in date-bucket-parity.test.ts), not by line number. It was the file's only remaining as never: [finding][drivers] checkDateBucketParity 的每个调用点都写 driver as never，把 BucketableDriver 替身的编译期一致性整个抹掉 —— 且实测该 cast 并不必要 #6354 already removed the ones on the checkDateBucketParity(…) stand-in call sites (lines 43–45 and 128–130 are cast-free on main).
• No guard from [finding][devx] 没有任何门禁阻止 as never 被加回来 —— #6354 一次清掉 10 个死 cast，但清理本身没有护栏 #6399 conflicts with it. [finding][devx] 没有任何门禁阻止 as never 被加回来 —— #6354 一次清掉 10 个死 cast，但清理本身没有护栏 #6399's resolution landed as the check:verify-stand-in gate in lint.yml; it passes green both before and after this change. It guards the stand-in call sites, not driver.create's options parameter.
• bypassTenantAudit is a declared key, not a wild one — packages/spec/src/data/driver.zod.ts:99, on DriverOptionsSchema. Corroborating: the dogfood and turso parity tests both pass createOptions: { bypassTenantAudit: true } to checkDateBucketParity with no cast and always have.

Evidence

Dependency closure built before typechecking (pnpm --filter @objectstack/driver-turso... build), so the verdict is not read off stale dist/*.d.ts. Re-run after merging origin/main @ 64d764e.

> @objectstack/driver-turso@17.0.0-rc.5 typecheck
> tsc --noEmit
EXIT=0

> @objectstack/driver-turso@17.0.0-rc.5 test
> vitest run
Test Files  27 passed (27)
     Tests  836 passed (836)
EXIT=0

Zero output on typecheck, so the cast was dead: nothing depended on it.

I also confirmed the test file is genuinely in the compile graph rather than silently excluded — driver-turso/tsconfig.json has "include": ["src/**/*"], and an unrelated const x: number = "definitely-not-a-number" appended to this same file does red:

src/date-bucket-parity.test.ts(137,7): error TS2322: Type 'string' is not assignable to type 'number'.

That control was reverted; it is not in the diff.

Scope note — this restores no checking, and the issue's rationale on that point is wrong

The issue argues the cast's cost is that it "抹成 never，于是 options 里拼错的键、写错类型的值都不会红". I tried to measure that payoff and it does not exist at this gate. With the cast removed:

Injected at the call site	tsc result
{ bypassTenantAudditt: true } (misspelled key)	EXIT=0 — no error
{ bypassTenantAudit: 'not-a-boolean' } (wrong type)	EXIT=0 — no error

The reason: TursoDriver overrides create with options?: any (packages/drivers/driver-turso/src/turso-driver.ts:523), so the override — not SqlDriver's options?: DriverOptions — is what resolves here. The parameter was already any before the cast ever applied.

The change is still correct and still worth landing: the cast reads as a needed escape hatch, and a future reader narrowing TursoDriver's signatures would have to work out whether it was load-bearing. Removing dead noise is the honest description of the benefit — restoring type safety is not.

Filed rather than fixed here (out of scope, per the dispatch):

• [finding][drivers] TursoDriver 五个 override 的 options?: any —— #5181 收窄的是 query，options 这条轴还没人碰过 #6402 — I commented with a measurement correcting its inventory: the options?: any widening covers 16 overrides, not the 5 its table lists, and all 16 have a base declaring DriverOptions. create is among the 11 missing, which is precisely why this PR restores nothing.

Same-shape scan of other drivers (read-only, per dispatch)

Filed as #6754, unassigned, not fixed here. 27 dead { bypassTenantAudit: true } as any options-gate casts — 17 in driver-sqlite-wasm, 10 in driver-sql, across 12 files. Stripping all 27 leaves both packages at EXIT=0.

Those are the inverse trade to this PR: they sit on a real typed gate (SqlDriver.create declares options?: DriverOptions and SqliteWasmDriver does not override it), so removing them genuinely restores checking — measured by misspelling a key after the strip:

src/sql-driver-runtime-token-default.test.ts(111,73): error TS2561: Object literal may only
specify known properties, but 'bypassTenantAuditt' does not exist in type '{ transaction?:
unknown; … bypassTenantAudit?: boolean | undefined; }'. Did you mean to write 'bypassTenantAudit'?

All exploratory edits for that measurement were reverted; git diff against the commit is empty.

Changeset

None — this is a tests-only change. No runtime or published surface is touched, so there is nothing for a changeset to describe. skip-changeset to be applied at acceptance.

Gates

Enumerated from .github/workflows/lint.yml, every check:* step run one by one — 47 gates, all green locally.

ESLint job — 33 gates

pnpm lint, then check: — slot-lookup, query-options-erasure, verify-stand-in, nul-bytes, doc-authoring, docs-audit-scope, role-word, quick-reference-counts, adr-anchors, org-identifier, authz-resolver, service-providers, route-envelope, error-code-casing, wildcard-fallthrough, meta-type-normalized, init-service-contract, durability-log-level, startup-registry-verdict, objectui-changeset, release-notes, release-body, node-version, workflow-status-functions, shard-attestation, published-files, engine-double-contract, kernel-hook-pairs, resume-authority-declared, driver-memory-census, merge-driver, spec-parsed-alias

TypeScript Type Check job — 14 gates

check:type-check-coverage, check:driver-conformance, check:stall-guard, spec tsc --noEmit, spec check:generated --reconcile-only, spec check:skill-docs, spec check:spec-changes, spec check:upgrade-guide, spec check:authorable-surface, spec check:docs, spec check:skill-refs, check:skill-frame-sync, check:skill-compatibility, spec check:react-blocks, turbo build (66 tasks), turbo typecheck (120 tasks), turbo build packages/*/*, check:type-check-debt, examples typecheck, downstream-contract typecheck, spec check:api-surface, spec check:exported-any, spec check:dual-source-exports, spec check:skill-examples, lint check:doc-formula-expressions, check:i18n, check:i18n-coverage, check:app-nav-i18n

origin/main @ 64d764e merged in (PR #6706's remote-transport.ts / text-predicate work does not touch this file; the merge left driver-turso, driver.zod.ts and verify untouched). Suite re-run green after the merge.

CI

Converged on 5e4e828 — 24 checks: 21 success, 2 skipped, 1 failure.

The single red is Check Changeset, and it is the expected label gate, not a defect in this PR. Its log:

##[error]This PR adds no changeset. If it releases nothing …, apply the 'skip-changeset' label; otherwise run 'pnpm changeset' and name the packages.

I confirmed from the job log that it failed on the changeset count, not on an unusable diff base. The gate offers exactly two routes, and an empty-frontmatter changeset is explicitly rejected as a third (#5471 / #4898). A tests-only change releases nothing, so the label is the correct route and fabricating a changeset would be the wrong one — left for skip-changeset at acceptance.

Everything else is green, including ESLint, TypeScript Type Check, Test Core (3 shards), Build Core, Dogfood Regression Gate (3 shards), Dogfood Verify CLI, and Temporal Conformance (live PG + MySQL).

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
objectstack	Ignored		Aug 8, 2026 2:22pm

[github-actions]

📓 Docs Drift Check

No hand-written docs reference the 0 changed package(s). ✅