[finding][drivers] 五个驱动的 `find/count/…` 仍声明 `query: QueryAST`，而调用方已可省略 `object` —— 双变让它编译，但声明开始说谎

[qq9340100]

Blocked-by: #5181

#5181 把 IDataDriver 的 query 参数收窄为 DriverQuery（即 Omit< QueryAST, 'object' >）之后留下的机械收尾。#5181 的派单明确 ⛔ 不改 driver 代码，故单独记录。

现状（实测）

契约侧六个方法（find / findOne / count / updateMany / deleteMany / explain）的 query 参数已经是 DriverQuery；五个驱动的实现仍旧写 query: QueryAST：

• packages/drivers/driver-memory/src/memory-driver.ts:283,342,444,469,502
• packages/drivers/driver-mongodb/src/mongodb-driver.ts:236,248,366,437,455,544
• packages/drivers/driver-sql/src/sql-driver.ts:2091,2251,2825,2837,2876
• packages/drivers/driver-sqlite-wasm（继承 sql 侧）
• packages/drivers/driver-turso/src/turso-driver.ts:499,504,535,863,870（这些是 query: any，另一种形态的同一问题）

这不是红的，也不会红：方法参数按双变比较，所以实现声明得比契约宽照样满足契约。全仓 pnpm typecheck 125/125 绿，就是在这个前提下拿到的。

为什么仍是一笔账

调用方现在有权省略 object（这正是 #5181 的目的）。于是一个声明 query: QueryAST 的实现，其类型说 query.object 是 string，而运行时它可能是 undefined。今天没有人踩到，因为实测五个驱动没有一个读 query.object：

grep -rn "query\.object" packages/drivers/*/src --include=*.ts   # 零命中

（sql-driver.ts:2414 的 r.object 是 sys_sequence 的行列，turso/remote-transport.ts 的 s.object 是 schema 批同步的入参，都不是 query AST。）

所以这是休眠的类型谎言，不是活体缺陷：下一个在驱动里写 query.object 的人会拿到一个编译期看着是 string、运行期是 undefined 的值，而没有任何门会拦他。

建议范围

把五个驱动的这些签名从 QueryAST / any 换成 DriverQuery（从 @objectstack/spec/contracts 导入），一轮翻完。翻完之后「驱动读 query.object」直接变成编译错误，谎言就没了藏身处。纯机械，无运行时变化，turso 那几处 any 顺手一并收紧还能多拿回一批检查。

按观察类记录、不挂 pm:queue，请分诊轮定级。

会话：session_011M7UwH25Unfi73UHim7ajY（#5181 实施期间发现，未认领）

---

Generated by Claude Code

[claude]

发现分诊轮(#4949 纪律):持有(留 finding),域维持 domain:drivers。

过时前提检查(origin/main @ 9e3709a)——一条硬前提尚未成立

• ⛔ DriverQuery 在 origin/main 零命中(阳性对照:packages/spec/src/contracts/data-driver.ts 与 data-driver.test.ts 都在)。正文的「契约侧六个方法的 query 参数已经是 DriverQuery」只在 PR refactor(spec)!: IDataDriver 的 query 参数改为 DriverQuery，对象名只写一遍 (#5181) #6076([spec] IDataDriver 的 query 参数要求 QueryAST.object 与第一实参重复 —— 下游被迫 as any(20 处实测),提议 Omit/optional 化 #5181,draft)的分支上成立 ⇒ 本单在 refactor(spec)!: IDataDriver 的 query 参数改为 DriverQuery，对象名只写一遍 (#5181) #6076 合入前无法实施:要导入的类型还不存在。已在正文补 Blocked-by: #5181 并写后回读确认。
• 驱动侧签名形状成立(单行 grep 命中):memory :283/:342/:469/:502、mongodb :236/:248/:437/:455/:468/:544、sql :2156/:2316/:2890/:2902(另 :3109/:3265/:3269 是 query: any)、turso :506/:511/:547/:972/:979(query: any)。行号与正文略有出入(正文取的是在飞分支或跨行签名的行号),形状结论一致。
• 休眠性证据成立:git grep 'query\.object' origin/main -- 'packages/drivers/*/src' 零命中 ⇒ 今天没有活体缺陷,是「休眠的类型谎言」。

冻结面裁定(#5499)

本单五个驱动跨冻结线:driver-memory / driver-mongodb 属维护者 2026-08-05 冻结族,driver-sql / driver-sqlite-wasm / driver-turso 明确不在冻结范围。#5499 的边界条款对这种跨面单已有既定处置 ——「跨面分歧单只冻 driver-memory 半边,保留 finding,下次发现分诊轮按不冻结侧重新框定范围」(#5298 / #5299 / #5369 即此形)。本单照此办:保留 finding,不挂 pm:on-hold(它本就不在队列里,再挂一个 hold 标签只是给同一件事两个状态)。

判级:持有 —— 重启条件(可退出,有明确触发点)

PR #6076 合入 ⇒ 立即按非冻结侧重新框定范围并晋级 pm:queue:范围 = driver-sql / driver-sqlite-wasm / driver-turso 的签名换成 DriverQuery(turso 那几处 any 顺带收紧),⛔ driver-memory / driver-mongodb 的对应行留在冻结面内,随 #5499 自己的重启条件处理(mongodb 待真实客户需求解冻;memory 随 sqlite :memory: 替代完成后整批关闭 not planned)。

晋级时的在飞注记(现在测的,晋级时要重测):PR #6047 动 driver-turso/src/remote-transport.ts 与四个测试,不含 turso-driver.ts ⇒ 无文件冲突,但同包,建议排其后。

本评论来自分诊座位 Routine(#5474 试点),不构成认领。

---

Generated by Claude Code

[claude]

Findings triage (#4949 discipline): promote — finding → pm:queue. domain:drivers retained. The restart condition fired, and re-pricing narrowed the scope materially — read the scope section before dispatching, do not reuse the estimate in the previous verdict.

Restart condition met. The 2026-08-06 17:57Z hold said, verbatim: 「PR #6076 合入 ⇒ 立即按非冻结侧重新框定范围并晋级 pm:queue」. #6076 merged as 6513c17 ("refactor(spec)!: IDataDriver 的 query 参数改为 DriverQuery，对象名只写一遍 (#5181)"). Promoting per that pre-authorization.

Stale-premise check (origin/main @ 42af12f, read from the tree):

• packages/spec/src/contracts/data-driver.ts:40 now exports type DriverQuery = Omit<QueryAST, 'object'>, and the interface declares it at :141 (find), :167 (findOne), :195 (count), :211 (updateMany), :214 (deleteMany), :345 (explain).
• The contract's own doc at :33 says implementations "that still declare query: QueryAST keep compiling". That is the finding, confirmed from the authoritative side: this is a declared-but-not-followed-through divergence, not a break and not a type error. Priority is accordingly low — it buys signature honesty, not behaviour.

Scope — re-measured, and smaller than the previous verdict framed it. That verdict scoped the work as 「driver-sql / driver-sqlite-wasm / driver-turso 的签名换成 DriverQuery」. Measured against main, two of those three need nothing:

package	own query: QueryAST sites	note
driver-sql/src/sql-driver.ts	10	the whole of the work — :2296, :2329, :2456, :3030, :3042, :6993, …
driver-sqlite-wasm	0	SqliteWasmDriver extends SqlDriver (:67) — inherits, nothing of its own
driver-turso	0	TursoDriver extends SqlDriver (:173) — inherits. remote-transport.ts's five QueryAST hits (:1207, :1343, :2268, :2276, :2280) are doc comments and error-message text naming QueryASTSchema, which is still the correct spec name. ⛔ Do not "fix" those strings.

So the non-frozen half is effectively one file. The turso any-tightening the old verdict bundled in ("turso 那几处 any 顺带收紧") is not part of this — it has no DriverQuery signature to hang off, and a drive-by tightening in a package whose driver inherits everything is a separate judgement. File it separately if it is still wanted.

Freeze compliance (#5499, maintainer 2026-08-05, reaffirmed 2026-08-06 07:18Z ruling A). driver-memory (4 sites: :283, :342, :469, :502) and driver-mongodb (6 sites: :201, :236, :248, :437, :455, :468) carry the same stale signature and are ⛔ out of scope — they stay on the frozen side and follow #5499's own restart conditions (mongodb thaws on real customer demand; memory closes not-planned with the sqlite :memory: replacement). The acceptance criterion is therefore "every non-frozen implementation matches the contract", not "no QueryAST remains in packages/drivers" — a dev who greps the whole directory will otherwise report the job unfinished, or worse, finish it.

Serialization: clear, re-measured this round. The old note ("PR #6047 … 建议排其后") is spent — #6047 is merged. All 6 currently-open PRs were enumerated and their file lists read; the only one touching packages/drivers/** is #4935 (chore: version packages (rc)), which touches CHANGELOG.md / package.json only. No conflict with sql-driver.ts. No Blocked-by: needed.

Dedup: three-repo scan. #5181 / PR #6076 is the parent and is already cross-linked; #6074 (the FilterCondition index-signature finding, held) is adjacent but orthogonal — it is about unknown $ operators in the type layer, not the object key. No duplicate.

本评论来自分诊座位 Routine(#5474 试点),不构成认领。

---

Generated by Claude Code