driver-sql: 37 more live-DDL it() blocks across 9 files inherit vitest's 5000ms default (same shape as #13688) #13902

Activity

1. claude

   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 Sep 1, 2026
2. claude

   assigned

   zhuangjianguo

   on Sep 1, 2026
3. claude commented on Sep 1, 2026

   

   claudeBot

   on Sep 1, 2026 – with Claude

   Contributor

   More actions

   Claim — domain:engine lane PM, session session_01F3jdziLbAPGeceVNmSox5L

   Branch: claude/issue-13902-live-ddl-test-budgets. Dispatching now. Clause-②: no — test files only, nothing under packages/spec/src/**, no published surface, no accept/reject behaviour. ⛔ 不预挂.

   ⭐ 分诊做了仪器自检，这一条我特别记名

   同一正则（^\s*\}, ?[0-9_]+\);）对 #13688 已落地的目标文件回 2 ⇒ 下面 9 个文件的 0 不是正则假阴。

   ⇒ 零读数带发火对照 —— 这正是本车道要求的形状，⛔ 不必再自证一遍表达式有效，但要在当前树上重跑（#13688 之后 main 已动过很多次）。

   裁定

   1. 预算用 60_000，⛔ 不要重新裁「用哪个数」。 父卡 driver-sql (live MySQL): the two #13056 orphan-shadow-cleanup tests run 3–4 connect→DDL→disconnect round trips under vitest's DEFAULT 5000ms, so Temporal Conformance reddens unrelated PRs #13688 的修复（e4dc299a / PR test(driver-sql): budget the #13056 live-MySQL orphan-shadow tests #13908）已经选定，并把理由写进代码注释：「Sized like this file's siblings (60_000 is 3 of the 4 sibling budgets); not an assertion that this test is normally anywhere near that slow.」 全包实测分布 60_000×4 / 40_000×1 / 120_000×1 ⇒ 照抄，并照抄那句免责注释。
   2. ⛔ 逐 it()，不逐文件、更不逐包。 全文件计 it( 有 87 条，卡的 37 是「嵌在 declareDialectCell(...) 回调里」的子集。⭐ 卡的口径是对的 —— 其余是本就该吃 5000ms 默认值的快单测，一刀切会把它们一起悄悄放宽。⚠️ 「某文件 0 条显式预算」只说明没有可抄的行，⛔ 不构成整文件改的依据。
   3. ⛔ 不得走包级 testTimeout，也不得援引 止血:packages/spec 的 vitest 缺 testTimeout —— 12 个跑 TS 编译器的用例配在 5s 默认值下 #4850。 止血:packages/spec 的 vitest 缺 testTimeout —— 12 个跑 TS 编译器的用例配在 5s 默认值下 #4850（packages/spec）确实是包级解，但两包用例构成不同：packages/spec 那批是齐的，driver-sql 是快单测与 live-DDL 混装，包级会静默重定 150+ 个。⇒ ⭐ 这一点必须写进 PR 正文，否则下一个人会照 止血:packages/spec 的 vitest 缺 testTimeout —— 12 个跑 TS 编译器的用例配在 5s 默认值下 #4850 改回去。
   4. sql-driver-backend-fault-envelope.test.ts 的 2 条默认排除。 它的 it() 只对 beforeAll 建好的连接发 1–2 次查询，⛔ 不是完整 connect/DDL/disconnect 周期，成本模型不同（分诊已复验 :160 与 :343）。⇒ 要纳入需单独给理由。
   5. ⭐⭐ 诚实边界，⛔ 不得越过：卡自陈这是结构扫描而非实测红（本环境无 live MySQL/PG）。⇒ ⛔ PR 正文不得把 37 条写成「37 条已知会超时」 —— 证据是形状 + driver-sql (live MySQL): the two #13056 orphan-shadow-cleanup tests run 3–4 connect→DDL→disconnect round trips under vitest's DEFAULT 5000ms, so Temporal Conformance reddens unrelated PRs #13688 这一个已证实例，不是 37 次观测。

   ZONE 2 — 我的假设，⭐ 推翻比修好更值钱

   • A2.1 在当前树上重跑分诊的正则并带发火对照（对 driver-sql (live MySQL): the two #13056 orphan-shadow-cleanup tests run 3–4 connect→DDL→disconnect round trips under vitest's DEFAULT 5000ms, so Temporal Conformance reddens unrelated PRs #13688 已落地文件应回 2）。9 个文件 / 37 条还成立吗？有没有新增的？
   • A2.2 那 37 条真的都是完整 connect/DDL/disconnect 周期吗？分诊只 spot-check 了 sql-driver-11627-hash-shadow-key.test.ts:135 一处作代表。⚠️ 若某条其实是 beforeAll 复用型（像 backend-fault-envelope），它属于被排除的那一类 —— 逐条核，⛔ 不要照单全收 37。
   • A2.3 declareDialectCell(...) 之外还有没有同形的 live-DDL it()？（卡的口径可能漏掉另一种包装。）⛔ 折入前先报。
   • A2.4 ⚠️ 加 60_000 会不会让某条本该红的用例变成慢慢地绿？即：有没有哪条的失败模式是「挂住」而不是「断言不符」？⛔ 这是我最担心的一点 —— 放宽预算的代价是把真超时藏起来。

   STOP

   • A2.4 找到任何一条的既有红是超时本身（而非断言）⇒ 停手报告，⛔ 别用预算把它盖掉。
   • 需要改 vitest.config.ts、turbo 配置或 scripts/ ⇒ 停手（那是包级解，已被裁掉）。
   • 门要你删台账行、或 shrink-only ratchet 计数上升 ⇒ 停手。

   ⛔ 不得跳过/禁用/隔离任何测试；⛔ 不得空提交或关闭重开 PR 来踢 CI。PR：draft，英文，Fixes #13902。⛔ 不 flip、不入队、不挂 auto-merge、不自审 —— 放行是 PM 的动作。

   ---

   Generated by Claude Code
4. zhuangjianguo

   added a commit that references this issue on Sep 1, 2026

   test(driver-sql): budget the 35 inline live-DDL it() blocks (#13902)

   ...

   5abe7ca
5. claude commented on Sep 1, 2026

   

   claudeBot

   on Sep 1, 2026 – with Claude

   Contributor

   More actions

   os-dev-report

   {
     "issue": 13902,
     "status": "done",
     "branch": "claude/issue-13902-live-ddl-test-budgets",
     "pr": "https://github.com/objectstack-ai/objectstack/pull/14098",
     "premise_still_valid": true,
     "clause_2": "Clause-②: no",
     "summary": "Added an explicit 60_000 budget to 35 it() blocks across 8 files in packages/drivers/driver-sql — every in-cell, unbudgeted it() that constructs a FRESH new SqlDriver(...) against a live MySQL/PG cell in its own body — plus one rationale comment per file carrying e4dc299a's disclaimer verbatim in spirit ('not an assertion that these tests are normally anywhere near that slow'). Scope held to per-it(): no vitest.config.ts, no turbo config, no scripts/, and the PR body states explicitly why #4850's package-level route is not a precedent here, so the next reader does not revert toward it. The card's 37 resolved to 35 in-scope + 2 already-excluded; A2.1 additionally found 3 blocks the card's instrument missed, and A2.2 put all 3 into the card's own excluded class, so the in-scope set is 35 either way. The honest boundary is kept: the PR body does NOT claim 37 tests are known to time out.",
     "tests": "All at HEAD 5abe7caa02, via scripts/pm/os-verify-lock.sh (shared-box seconds). (1) Dependency closure: pnpm --workspace-concurrency=2 --filter '@objectstack/driver-sql^...' build — VERDICT command-exit 0. (2) pnpm --filter @objectstack/driver-sql typecheck && pnpm --filter ... exec vitest run --maxWorkers=2 — VERDICT command-exit 0 (one && chain, so the verdict certifies both); vitest: 148 passed | 9 skipped test files (157), 2260 passed | 134 skipped tests (2394). Typecheck coverage checked rather than assumed: tsc --noEmit --listFiles names all 8 edited files (157 *.test.ts in the program), so this is not a green over source nothing read. (3) NOT MEASURED, stated plainly: the 35 edited blocks did not execute here — OS_TEST_MYSQL_URL / OS_TEST_POSTGRES_URL / OS_EXPECT_LIVE_DIALECT_MATRIX are all unset, so declareDialectCell collapses each live cell to one named skip ('is provisioned — set OS_TEST_MYSQL_URL to run this cell of the D-A3 driver axis'). (4) Gate union re-derived AFTER the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack (script read the change set itself from merge base dda969cd7: 8 paths): 22 path-matched + 5 convention-triggered; 23 ran, ALL exit 0, each captured before any pipe. Quoting the gates' own verdict lines: check-cross-package-test-inputs 'OK: 24 package(s) read outside themselves, all declared'; check:where-matcher '323 matcher(s) discovered ... 0 silently-wrong ... baseline key set verified against dda969c: no files added' (ratchet did not move); check-nul-bytes 'OK (scanned 7702 text file(s) ... no raw ASCII control bytes)'; check-driver-conformance 'OK — 50 covered cell(s), 0 in the DEBT ledger, 0 exempt'. NOT MEASURED (never counted as passes): check:dual-build-cjs-loads and check:type-check-debt need the whole workspace built (CI's Build Core scope); check:test-completeness exits PREREQUISITE NOT MET without a saved turbo run test log. (5) FULL repo lint, not a narrowing: node --stack-size=4000 node_modules/eslint/bin/eslint.js . --no-inline-config — VERDICT command-exit 0, zero output lines, 61s. ABLATION, two parts. NO REBUILD LEG APPLIES and none was fabricated: the mutated subject is a test file vitest reads from disk, and these files import the driver by relative specifier ('./sql-driver.js'), never through the package exports map, so nothing is served from dist/. Part 1 (literal, on the COMMITTED implementation), direction predicted before running as NO CHANGE: removed all 3 budgets from sql-driver-string-maxlength-varchar.test.ts; mutation confirmed ON DISK by blob hash 646b0050a5 -> 9450642b90 and by count 3 -> 0 budgets, never by the editor's exit code; both legs read identically (Test Files 1 passed (1); Tests 4 passed | 2 skipped (6)) with the two skips being the named provisioning guards — the measurement that establishes VACUITY of any local run over these blocks, not evidence the budget does nothing. Restored via git checkout HEAD -- ABSOLUTE_PATH under trap ... EXIT INT TERM; restore PROVEN by the blob returning to 646b0050a5 and git diff HEAD being empty for the path. A first attempt exited 1 in BOTH legs with 'No test files found' (repo-relative path handed to a vitest whose cwd is the package) — recorded as an invocation error, re-run correctly, and NOT reported as a red leg. Part 2 (mechanism probe, the leg that CAN run here), direction predicted before running: leg A reds at 5000ms, leg B greens. A temporary 6500ms-sleep probe under this package's real vitest config: leg A (no third argument) '× ... 5007ms → Test timed out in 5000ms.' exit 1; leg B (with , 60_000) '✓ ... 6505ms' exit 0. Both matched the prediction — this proves the inherited default really is 5000ms here and that the third argument this PR adds is honoured. Probe deleted; git status --porcelain verified empty.",
     "mcp_calls": "4 — issue_read get, issue_read get_comments, pull_request_read (#13908), search_issues (returned 'API rate limit already exceeded', no data). Everything after that ran on the repo-scoped REST channel (probe: GET issues/13902 → HTTP 200), which I switched to and am declaring here: /search/issues is refused for this seat ('sessions are bound to their configured repositories', HTTP 403), so dedup used the REST list endpoint on a narrow slice (open issues labelled 'tests', 20 rows) plus a local grep, with #13902 itself as the firing positive control. PR creation, the skip-changeset label, the finding issue and this comment all went through REST.",
     "a2_answers": {
       "A2.1": "HOLDS, with one addition. Control fires: the triage expression ^\\s*\\}, ?[0-9_]+\\); still returns 2 against #13688's landed file, so the zeros are readings. Re-run structurally (TypeScript AST, not grep) on the current tree at dda969cd71: the card's 9 files / 37 blocks hold EXACTLY, per-file 9/6/5/4/4/3/3/2/1. NEW: sql-driver-diagnostic-value-probe.test.ts carries 3 more in-cell unbudgeted it() blocks — 40 total, not 37. Probable cause: the card's script pre-filtered to files with ZERO explicit budget lines, and this file has one — on its beforeAll at :335. It is the only such leak in the package (checked: every other file with budgets has no unbudgeted in-cell it()). Net scope effect: zero, because A2.2 puts all 3 in the excluded class.",
       "A2.2": "FALSIFIED IN PART, in the direction you suspected. The 37 must NOT be taken on faith: of the 40 in-cell unbudgeted blocks, 35 construct a fresh new SqlDriver(...) in their own body and 5 do NOT. The 5 are the beforeAll-reuse class: sql-driver-backend-fault-envelope.test.ts:360/:379 (the card's own 2 — re-verified, beforeAll at :160 and :343 builds the driver) plus the 3 newly-found sql-driver-diagnostic-value-probe.test.ts:366/:385/:431, which are STRONGER exclusions still — they are SYNCHRONOUS it()s reading a Map populated in beforeAll and touch no server at all. So the card's in-scope 35 (its 37 minus backend-fault-envelope's 2) is exactly right. Three of the 35 do their DDL via driver.execute('create table ...') or an internal probe rather than initObjects(...) (11389:439, 11565:201, keyed-text-mysql:285); all three still open a fresh live connection per test. 11565:201 is the cheapest — connect plus one metadata query, no DDL — included on the connect cost alone, flagged here so you can overrule it.",
       "A2.3": "YES — 4, reported and NOT folded in. Live-PG it() blocks outside declareDialectCell, unbudgeted: live-dialect-matrix.isolation.test.ts:281 and :314 (it.skipIf(!PG_CELL.available), PG_CELL.config(), both call initObjects — the full #13688 shape); sql-driver-autonumber-cold-race.test.ts:159 (it.skipIf(!pgCell.available), pgCell.config()); sql-driver-json-binding-without-ddl.test.ts:173 ({ ...PG_CELL.config(), schemaMode: 'validate-only' }). Discriminated by reading each new SqlDriver argument, not by inferring from the file: the other 42 outside-cell driver-constructing it()s are embedded SQLite (dialectCell('sqlite').config() or a better-sqlite3 :memory: literal) and are correctly on the fast default. Filed as #14100 together with a second adjacent population.",
       "A2.4": "NO — no STOP condition triggered. Every detector was run with a firing positive control from this same package, so each zero is a measurement, not a broken expression. Inside the 35: clock reads (Date.now / performance.now) 0, control 9; duration assertions 0, control 1 (sql-driver-connect-bound.test.ts:196, whose 40_000 budget IS load-bearing for elapsed > 8_000 and < 20_000 — and which is not in the edited set); explicit sleeps/timers 0, control 4; unbounded loops 0, expression proved on live-dialect-matrix.testkit.ts:190 (for (;;)) since no *.test.ts has one; lock/deadlock subjects 0, control 7; Promise.all concurrency 0, control 7; retry/poll loops 0, control 8; prose naming a timeout as the expected red — 3 raw hits, ALL falsified as substring false positives (Shanghai, change, unchanged). Every one of the 35 asserts on information_schema contents, an error envelope, or row values, so a real regression here surfaces as an ASSERTION MISMATCH, which a larger budget does not hide. Stated cost rather than buried: a FUTURE regression manifesting as unbounded slowness now takes 60s to surface instead of 5s — the same cost #13688 accepted, and what the per-file disclaimer comment keeps visible."
     },
     "open_questions": [
       {
         "question": "The package's budget distribution is not quite what the dispatch order recorded, and it changes one sentence in the PR body (not the decision). Measured on this tree: 9 explicit numeric budgets — but only 3 are on it() calls (60_000 x2, both from #13688; 40_000 x1 on connect-bound), while 6 are on beforeAll hooks (60_000 x5, 120_000 x1). The order's '60_000 is 3 of the 4 sibling budgets' was counting hook budgets as it() siblings. Does the decree still stand?",
         "options": [
           "A — stands, unchanged: 60_000 is 7 of the 9 explicit budgets regardless of position, and it is the value #13688 chose for exactly this shape. The 40_000 is a duration-assertion budget (different purpose) and the 120_000 a single heavy beforeAll. I wrote the PR body with the accurate count.",
           "B — revisit the number now that the it()-level precedent is thinner than it looked (only #13688's own 2).",
           "C — treat the hook-vs-it() split as a signal that these 8 files should hoist their live work into beforeAll hooks instead of being budgeted per test."
         ],
         "recommendation": "A. The measurement strengthens the decree rather than weakening it, and I did not re-litigate — I only corrected the sentence. C is a real architectural observation (the package's DOMINANT pattern is to hoist live work into a budgeted beforeAll, which is exactly why the excluded files look different) but it rewrites 35 tests' structure and is far outside this card; it belongs in triage on #14100 if anywhere."
       },
       {
         "question": "The card's sweep instrument had a file-level pre-filter that hid one file (diagnostic-value-probe). It cost nothing here because those 3 blocks were excludable anyway — but the same filter would hide any future file that has SOME budgets and unbudgeted in-cell it()s. Worth recording anywhere durable?",
         "options": [
           "A — no: the sweep was a one-off, this PR's body records the lesson, and no committed artefact carries the expression.",
           "B — yes, as a note on #14100, so the next person sweeping this package uses a per-it() AST walk rather than a file-level grep.",
           "C — commit the AST walk as a gate script, so 'a live-DDL it() with no explicit budget' cannot reappear silently."
         ],
         "recommendation": "B for now. C is genuinely tempting — it is the 'extend one guard to close the whole class' shape — but it would add a new gate family to a repo that already runs 197, on a defect class with one measured instance; that is a scope and gate-budget call that is yours, not mine."
       }
     ],
     "out_of_scope_findings": [
       "filed as #14100: driver-sql live-DDL work still on vitest's 5000ms default outside #13902's scoping — Population A, the 4 live-PG it() blocks beyond declareDialectCell from A2.3; Population B, sql-driver-backend-fault-envelope.test.ts's two beforeAll hooks (:160, :343) that each open a live driver, run initObjects(...) and insert rows with NO budget. B is the sting in the exclusion: #13902 correctly excluded that file's two it() blocks because the live cost is in the hook — which left that cost unbudgeted. The package already answers this shape at sql-driver-diagnostic-value-probe.test.ts:335 (a beforeAll carrying }, 60_000). Unassigned and unlabelled for your triage. Dedup: REST list of the 20 open 'tests'-labelled issues + local grep, with #13902 itself as the firing positive control; no duplicate."
     ]
   }

   ---

   Generated by Claude Code
6. claude commented on Sep 1, 2026

   

   claudeBot

   on Sep 1, 2026 – with Claude

   Contributor

   More actions

   PM 复核 —— PR #14098。⛔ 我在派发词里引的预算分布是错的，先更正；⭐ 然后是本轮最漂亮的一次消融

   domain:engine 车道 PM，session session_01F3jdziLbAPGeceVNmSox5L。

   ⛔ 先认错：「60_000 是四个兄弟预算里的三个」—— 我把 hook 预算当成了 it() 兄弟

   席位实测本包 9 个显式数值预算，但按位置分：

   • it() 上只有 3 个 —— 60_000 ×2（两个都来自 driver-sql (live MySQL): the two #13056 orphan-shadow-cleanup tests run 3–4 connect→DDL→disconnect round trips under vitest's DEFAULT 5000ms, so Temporal Conformance reddens unrelated PRs #13688 自己）+ 40_000 ×1（connect-bound）
   • beforeAll 上有 6 个 —— 60_000 ×5、120_000 ×1

   ⇒ ⛔ 我那句话把 hook 上的预算算进了「it() 兄弟」。⭐ 席位没有借此重开裁定，只更正了 PR 正文里的那一句 —— 处置正确。

   裁定维持（选项 A）：60_000 在 9 个显式预算里占 7 个，与位置无关；40_000 是服务于时长断言的预算（用途不同），120_000 是单个重型 beforeAll。⇒ 决定不变，理由更正。
   ⛔ 不取 C（把 35 个测试的 live 工作上提到 beforeAll）—— 那是真实的架构观察，但它重写 35 个测试的结构，远超本卡；若要做，去 #14100 分诊。

   ⭐ A2.1：卡的口径成立，但它的仪器有一个文件级盲点

   控制项发火（同一正则对 #13688 已落地文件回 2）。用 TypeScript AST（⛔ 不是 grep）在当前树重扫：卡的 9 文件 / 37 条逐文件精确成立（9/6/5/4/4/3/3/2/1）。

   ⭐ 新增第 10 个文件：sql-driver-diagnostic-value-probe.test.ts 还有 3 条 in-cell 无预算 it() ⇒ 总数是 40，不是 37。
   ⚠️ 原因值得记：卡的脚本预先过滤掉了「有任何显式预算行」的文件，而这个文件在 beforeAll:335 上有一个 ⇒ 整个文件被滤掉了。⇒ ⭐ 一个文件级前置过滤，把一个真阳性藏了起来 —— 同族于本车道反复记的「看起来是读数，其实不是」。

   ⭐⭐ A2.2：按我的要求逐条核，结果是部分证伪，且方向正是我担心的那个

   我写的是「⛔ 不要照单全收 37」。实测 40 条里：35 条在自己体内构造全新 new SqlDriver(...)，5 条不是。

   那 5 条全是 beforeAll 复用类：卡自己排除的 2 条（backend-fault-envelope，beforeAll 在 :160/:343，已复验）加上新发现的 3 条 —— ⭐ 而后者是更强的排除：它们是同步 it()，只读 beforeAll 填好的 Map，完全不碰服务器。

   ⇒ 卡的在范围内 35 条恰好正确（它的 37 减去 backend-fault-envelope 的 2）。⭐ 两条不同的路径得到同一个数，这比直接接受 37 强得多。

   ⚠️ 席位主动交回一条让我裁的：11565:201 是 35 条里最便宜的（连接 + 一次元数据查询，无 DDL），仅凭「连接成本」纳入。⇒ 本席维持纳入 —— 判据是「每个测试开一次全新的 live 连接」，而不是「跑了多少 DDL」；连接建立本身就是 #13688 那条超时的主要成本。⭐ 它把这条标出来让我可推翻，做得对。

   ⭐ A2.3：4 条 declareDialectCell 之外的同形块，报告而⛔未折入

   live-dialect-matrix.isolation.test.ts:281/:314 · autonumber-cold-race.test.ts:159 · json-binding-without-ddl.test.ts:173。
   ⭐ 判别方式对：逐个读 new SqlDriver 的实参，⛔ 不从文件名推断 —— 其余 42 条 cell 外构造驱动的 it() 都是内嵌 SQLite，本就该吃快速默认值。已立 #14100。

   ⭐⭐ A2.4：我最担心的一问，被 8 个各自带发火对照的探测器回答了

   我问的是「加预算会不会把本该红的用例变成慢慢地绿」。答案：不会，而且证明方式是逐类排除：

   探测	35 条内	对照
   时钟读取（Date.now/performance.now）	0	9 ✓
   时长断言	0	1 ✓
   显式 sleep / timer	0	4 ✓
   无界循环	0	在 testkit 上证明表达式有效 ✓
   锁 / 死锁主题	0	7 ✓
   Promise.all 并发	0	7 ✓
   重试 / 轮询	0	8 ✓
   散文点名「超时」为预期红	3 → 全部证伪为子串误命中（Shanghai / change / unchanged）	—

   ⇒ 35 条全部断言在 information_schema 内容、错误信封或行值上 ⇒ 真实回归表现为断言不符，⛔ 更大的预算藏不住它。
   ⭐ 且 connect-bound.test.ts:196 的 40_000 是承重的（elapsed > 8_000 && < 20_000）—— 它不在被编辑集合里。
   ⭐ 成本被写出来而不是埋掉：一个未来「无界变慢」型回归现在要 60 秒而不是 5 秒才浮现 —— 与 #13688 接受的成本相同，并由每文件的免责注释保持可见。

   ⭐⭐ 消融：本轮最好的一次，因为它承认了 part 1 的空泛并另造了一条能跑的腿

   • Part 1（还原 3 个预算）预测为「无变化」，实测两腿读数相同 —— ⭐ 席位把这明确记为「本地对这些块的任何运行都是空泛的」的测量，⛔ 没有把两腿相同当成「预算无作用」的证据。因为本环境 OS_TEST_MYSQL_URL / OS_TEST_POSTGRES_URL 未设，declareDialectCell 把每个 live cell 折成一条具名 skip。
   • ⭐ Part 2 —— 机制探针，这条能跑：在本包真实 vitest 配置下放一个 6500ms sleep。
     • 腿 A（不给第三参数）→ × … 5007ms → Test timed out in 5000ms.，exit 1
     • 腿 B（给 , 60_000）→ ✓ … 6505ms，exit 0
       ⇒ 同时证明了两件事：继承的默认值在这里确实是 5000ms，且本 PR 加的第三参数确实被采纳。 探针已删，git status --porcelain 验空。
   • ⭐ 第一次尝试两腿都 exit 1 报 No test files found（把仓库相对路径交给了 cwd 在包内的 vitest）⇒ 记为调用错误、重跑，⛔ 没有报成红腿。

   门与边界

   23 个 gate 全 exit 0（管道前捕获），引了各自的判决行；check:where-matcher 的 ratchet 未移动。三个 NOT MEASURED（⛔ 从不当作通过）。全仓 lint 未声明窄化。typecheck 覆盖是验过的：tsc --listFiles 点名全部 8 个被编辑文件。
   ⭐ 诚实边界守住了：PR 正文没有声称「37 条已知会超时」，也没有声称那 35 条在本地跑过。

   裁定开放问题 ②：取 B，⛔ 不取 C

   席位问「要不要把那个 AST walk 提交成一道门」，推荐 B（记在 #14100），⛔ 不推荐 C，理由是「本仓已经跑 197 道门，而这个缺陷类只有一个实测实例；那是你的范围与门预算裁量，不是我的」。
   ⇒ ⭐ 这份克制正确，本席采纳 B。 ⛔ 不建门：一个只有一次实测实例的类，加一道常设门是用永久成本买一次性收益；#14100 上的一条笔记足以让下一个扫这个包的人用 per-it() 的 AST 走查而不是文件级 grep。⚠️ 若 #14100 的两个 population 落地后又出现第三次，届时再谈建门 —— 那时它就是三个实例而不是一个。

   ---

   Generated by Claude Code
7. zhuangjianguo

   closed this as completedin #14098on Sep 1, 2026
8. claude

   removed

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

   on Sep 1, 2026
9. akarma-synetal

   added a commit that references this issue on Sep 1, 2026

   test(driver-sql): budget the 35 inline live-DDL it() blocks (objectst…

   ...

   16c6e1c
10. os-support-ai

    added a commit that references this issue on Sep 1, 2026

    test(driver-sql): budget the live-DDL work outside #13902's scoping

    ...

    69424a1
11. akarma-synetal

    added a commit that references this issue on Sep 9, 2026

    test(driver-sql): give the live dialect cells a derived per-test budg…

    ...

    b4abb0a