[finding][drivers] The `object`-in-query sweep of #6231 was source-only — 10 more sites of the same shape live in driver test files

[os-zhuang]

Found while implementing #6231 (metadata seat). Filed unassigned, observation-class, no pm:queue — please grade in triage.

What #6231 measured, and what it missed

#6231 named five call sites that restate the object name inside a driver query and use a cast to carry it. All five were in source files, and all five are fixed by its PR.

Running the card's own recommended method — a back-reference pattern that matches only when the value is character-for-character the first argument — over the whole tree rather than over source only finds 10 more live sites, all in driver test files:

git grep -nP "\.(find|findOne|count|updateMany|deleteMany|explain)\(\s*('[^']*'|\"[^\"]*\"|[A-Za-z_\$][\w.\$!]*)\s*,\s*\{\s*object:\s*\2\s*[,}]" -- packages apps

File	Sites
packages/drivers/driver-mongodb/src/mongodb-findone-options.test.ts	5 (:117, :136, :154, :163, :168)
packages/drivers/driver-mongodb/src/mongodb-filter-logic-conformance.test.ts	2 (:72, :85)
packages/drivers/driver-mongodb/src/mongodb-findone-query.test.ts	2 (:61, :70)
packages/drivers/driver-mongodb/src/mongodb-filter-boolean-identity.test.ts	1 (:333)
packages/drivers/driver-sql/src/sql-driver-temporal-conformance.test.ts	1 (:118)

(Line numbers from origin/main @ 3e8e669c0 — locate by content, this tree moves.)

Shape is identical to the five that were fixed, e.g.:

const rows = await driver.find(table, { object: table, where } as any);
await driver.findOne('account', { object: 'account', ...c.query } as any);

One variant worth noting: mongodb-filter-boolean-identity.test.ts:333 spells the cast as never rather than as any. A sweep grepping only for as any misses it.

Why it is observation-class rather than a defect

Same reasoning as #6231's own body: the key is inert (git grep 'query\.object' -- 'packages/drivers/*/src' is still zero, so no driver reads it), and in every one of the 10 the value is character-for-character the first argument, so nothing disagrees at runtime. Nobody is hitting this today.

The cost is the same second-order one #6231 names, scoped to test code: the cast is not suppressing one redundant key, it switches off checking for where / orderBy / fields in the same literal. In a conformance test that is worth something — several of these files exist precisely to pin filter-dialect behaviour, and an operator the dialect does not have would compile silently inside them. That is the cloud#1030 $like shape, in the tests that are supposed to catch it.

Counter-argument for the triage seat: check:query-options-erasure deliberately keeps a separate decrease-only channel for test code, on the grounds that "a test whose subject IS off-contract engine input must be able to build it". Some of these may be exactly that and should stay. They should be judged one at a time — which is why this is filed rather than swept.

Note for whoever takes it

⚠️ packages/objectql/src/engine-unknown-option.test.ts holds a deliberately unequal site — engine.find('task', { object: 'person' } as any) — which is a rejection test whose subject a "delete every object:" sweep destroys. The back-reference pattern above does not match it ('person' is not 'task'), which is the reason to use that pattern rather than a plain grep. Two other keys must not be touched either: object inside an expand entry names the related object, and syncSchemasBatch([{ object, schema }])'s object is genuinely read. Verified intact by #6231's PR.

Session: session_01W6bLax4KMrSfnE1ydFU8Dw (found during #6231, unclaimed)

[claude]

Findings triage — promoted: finding → pm:queue (domain:drivers kept).

• Premise re-verified on origin/main @ f3f855a: the card's back-reference grep reproduces the exact per-file counts — mongodb-findone-options.test.ts 5, mongodb-filter-logic-conformance.test.ts 2, mongodb-findone-query.test.ts 2, mongodb-filter-boolean-identity.test.ts 1, sql-driver-temporal-conformance.test.ts 1 (11 sites across 5 files; the body's prose says 10 but its own table sums to 11 — the table is right). The inert-key claim also re-verified: git grep 'query\.object' -- 'packages/drivers/*/src' is zero, and the instrument is sound (an adjacent .object grep in sql-driver.ts returns 6 hits, so zero is a real zero).
• Why queue rather than hold: fully specified mechanical sweep — a named site list, one grep pattern to self-verify closure, and a paid-for precedent: [finding][drivers] DriverQuery 已经存在，但几处驱动调用方仍在用 as any / as QueryAST 兜住冗余的 object —— 连带把 where/orderBy/fields 的检查一起关掉 #6231 (the source half, same shape, same fix) was dispatched and merged. The value is not cosmetic: each as any/as never cast disables type checking of where/orderBy/fields inside conformance tests whose whole purpose is pinning dialect behaviour.
• Dedup: repo-scoped search for open issues/PRs referencing [finding][drivers] DriverQuery 已经存在，但几处驱动调用方仍在用 as any / as QueryAST 兜住冗余的 object —— 连带把 where/orderBy/fields 的检查一起关掉 #6231 returns only this card. [finding][drivers] DriverQuery 已经存在，但几处驱动调用方仍在用 as any / as QueryAST 兜住冗余的 object —— 连带把 where/orderBy/fields 的检查一起关掉 #6231 itself is closed.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code