Repository navigation
[Decision] Should the driver-sql suite REFUSE to report green when live dialect cells were skipped? — shape 2 of #18200, reserved to the maintainer as a new required gate #18653
Description
Filed by the domain:engine execution seat, session_01CqmCgU5RGDoJYhHUMVp2af, R1, out of #18200 / PR #18649. ⛔ No domain:* and no priority:* asserted.
The #18200 dispatch built shape 1 (make the skip loud) under a seat ruling, and explicitly reserved shape 2 to the maintainer's floor because a suite that refuses to report green is a new required gate. The dispatch invited the dev to make the case if the work produced one. It did, and the case rests on a number the card did not have.
维护者速读
driver-sql 的本地测试跑完会说「178 通过 | 11 跳过」。⭐ 那个 11 是假的 —— 真实情况是 189 个文件里有 67 个含被跳过的 live 单元格,其中 56 个报告为「通过」。也就是说,一次本地全绿看不见 688 个测试,占这个套件能跑的 21%。
这一轮里它已经咬了三次:#17469 带着四个红 CI 交付、今天 #17231 又花掉一整轮补丁、而我自己在诊断那个红时差点重复立卡席当初的同一个误判,只因为多拉了一次基线日志才没踩进去。
#18200 已经修好了「看得见」那一半(PR #18649):跑完会明说哪些方言没跑、设哪个环境变量能跑、67/56 的拆分,没有 PG 时还打印一份跑过才写下来的七行本地起库配方。
剩下的问题只有一个:要不要让它在 live 单元格被跳过时直接不给绿?
| 做法 | 代价 | |
|---|---|---|
| A(席位荐) | 就到 shape 1 为止,先看它够不够 | 0;但它要求读的人真的去读 |
| B | live 单元格被跳过且没传显式 opt-out 就非零退出 | 每个没有 PG/MySQL 的本地跑都要带 opt-out;而人人都传的 opt-out 等于没有门禁 |
席位推荐 A,但带一个具名的重新触发条件:shape 1 落地之后若再出现第四次同类事故(本地绿 → CI 在没人跑过的 live 单元格上红),那就是 B 的证据,直接照 B 办,⛔ 不必再讨论一次。
请裁:A / B?
四棱
① 实际业务需求 —— 有实测拉动,而且是重复的。 三次实例(#17469 四个红 CI;#17231 一整轮补丁;加上本席的近失误)。⛔ 不是投机面,代价是每次一整轮。
② 项目长远合理性 —— 中性偏 A。 「声明即强制」在这里的具体形态是:套件不该报告一个它没验证过的状态。shape 1 已经让它不再报告假状态 —— 它现在明说自己没跑什么。⇒ 长远形态的核心已经拿到;B 买的是「强迫读者注意」,那是人机工程,⛔ 不是契约正确性。
③ 防 AI 写错 —— 唯一明确指向 B 的一棱,且理由不弱。 本仓的 dev 是 AI。一个 grep 0 failed 的 agent 不会读那个块。 shape 1 帮的是会看的读者,B 帮的是不看的。
④ 创业阶段不扩散 —— 指向 A,而且理由是机械的。 B 让每一次没有 PG/MySQL 的本地跑都需要 opt-out。
⇒ ③ 指 B,④ 指 A,①② 中性偏 A。⭐ 分歧的真正来源是一个没人测过的量:shape 1 之后还会不会再发生。 所以席位推荐的不是「不做 B」,而是先让那个量可测,并预先说好什么读数会触发 B。
Governing text: 人工地板条款 —— 「新增必需门禁/hook/棘轮」恒交维护者(SKILL.md 代裁人工地板段)。以及 #18200 卡面自陈的失败形态:「it ran, it reported healthy, and nothing was delivered」。
读数
⭐ 来自 PR #18649 的两状态实测(dev 测,本席按下述方式佐证):
| 状态 | 结果 |
|---|---|
| A(无服务器) | 178 passed | 11 skipped (189) / 2627 passed | 168 skipped ⇒ 块报「3 个方言跑了 1 个」,168 个跳过散在 67/189 个文件里,11 个 vitest 报为 skipped,另外 56 个它报为 PASSED |
| B(真 PostgreSQL 16.13) | 186 passed | 3 skipped (189) / 3315 passed | 85 skipped ⇒ 「3 个方言跑了 2 个」 |
| 差额 | 688 个测试,占可跑总量的 21% |
⭐ no-op 对照:改动前在同一基线上跑,摘要逐字节相同 ⇒ 这个 reporter ⛔ 不动任何计数、不动退出码。这是 shape 1「纯声明」成立的证据。
本席的佐证(非重跑,如实标明):driver-sql 下 .test.ts 文件 189 个(卡片写的 188 已过期,#17231 今早加了一个);静态 grep 引用 live-cell 守卫的文件 76 个 —— 比 dev 实测的 67 大,方向正确(静态上界含不止 live 守卫的 skipIf)。必中对照:含 describe( 的文件 185;暗对照:伪造的环境变量名 0。
TestCase 携带 meta {} 且没有 skip 原因,唯一的逐测试通道是测试名;本包有 9 个文件用手写 skipIf 守 live 单元格,名字里没有 testkit 写的标记 ⇒ 按名字归因会少报。所以 shape 1 报的是文件级普查而非逐测试归因,这是测出来的边界,⛔ 不是偷懒。
查重词
driver-sql live cell skipped green · OS_EXPECT_LIVE_DIALECT_MATRIX opt-out · live-dialect-coverage reporter · refuse green on skipped dialect · 67 of 189 skipped live cell
相关
- [finding] a local green on driver-sql is blind to 11 of its 188 files — the live PG/MySQL cells report as SKIPPED, and that is what let #17469 round 1 ship four red CI jobs #18200 — shape 1,由 PR test(driver-sql): name the dialects a local run did not exercise, under the counts #18649 落地。本卡是它明确留给维护者的那一半。
- [finding]
check:cross-package-test-inputsexits 0 or 1 on the same commit depending on whether adist/exists — and lint.yml runs it with no build ahead of it, so CI only ever sees the blind leg #18621 — 另一条「门禁在它本该看见的状态下瞎掉」的卡,本轮三次独立测到。⚠️ 同族但不同缺陷:那张是门禁看不见,这张是套件看得见却不拦。分诊可判是否合并考虑。
Generated by Claude Code
Activity
hotlong commentedon Sep 18, 2026
Ruling: batch #158 item 3 · letter A (shape 1 — the loud, honest skip report PR #18649 landed — is the whole remedy; ⛔ no refuse-to-report-green gate; a named re-trigger is recorded) · maintainer 「同意」 2026-09-18T11:14Z
Director seat, summon #24,
session_01Wj1HUjzyeiBQ8atRf1ZhaL. Presented in detail with the recommendation A; the maintainer agreed. Facts (this card): the localdriver-sqlrun's 「178 passed | 11 skipped」 hid 688 tests (21% of the runnable suite) because 56 of the 67 files with skipped live dialect cells report PASSED; three incidents preceded shape 1 (#17469, #17231, the seat's own near miss). #18200 / PR #18649 now prints which dialects did not run, how to enable them, the 67/56 split and a tested seven-line local database recipe. Shape 2 — non-zero exit when live cells are skipped without an explicit opt-out — is a new required gate, reserved to the maintainer.Ruling — A
Shape 1 stands as the remedy. ⛔ B: an opt-out every local run without PostgreSQL / MySQL must pass is a ritual everyone learns to add, and a gate everyone bypasses is not a gate (the maintainer's standing word: no more gate development).
Named re-trigger (⛔ not a ladder): a fourth incident of the same shape after shape 1 — a local green run followed by CI red on live cells nobody ran — is the evidence for B; when it occurs the seat files the B card with the incident named and it is dispatched without re-litigating this ruling.
Four-facet reading: ① all three incidents predate shape 1, its effect is unmeasured; ② neutral; ③ this is the fleet's
grep '0 failed'habit, not the platform user's axis; ④ B adds a ritual, A adds nothing.Execution
needs-user-decisionstripped; closed not_planned (nothing to build; the re-trigger lives on this card).Generated by Claude Code