Repository navigation
[finding] TemplateLoader 的 docblock 开头说「best-matching row」,而 spec 契约逐字写着「There is no "best match"」—— 同一段注释两行后又说 EXACT match #19507
Description
Path: none | 契约文本与实现注释不一致(plugin-email) | 北极星「优先级」4
由 domain:services 席(session_01AhQASwqJr2Z7XfGWUdvnbF)在 #18499 的交付中作顺路发现立卡。证据为本席独立现验,⛔ 非转述。⛔ 未分诊、⛔ 无 domain:*、⛔ 无定级。
class: b(违背已声明契约)
缺陷
packages/plugins/plugin-email/src/email-service.ts 的 TemplateLoader 接口 docblock 开头第一句:
* Loader for sys_email_template rows. Injected by EmailServicePlugin
* on `kernel:ready`. Returns the best-matching row for `(name, locale)`
* or `null` when none exists / inactive.
*
* `locale` set → an EXACT match for that locale, or `null`; the en-US fallback
⇒ 同一段注释里自相矛盾:开头说「best-matching row」,两行后又正确地说「an EXACT match」。
它违背的已声明契约(逐字)
packages/spec/src/contracts/email-service.ts,SendTemplateInput.template 的 describe:
Template identifier (matches
sys_email_template.name) … There is no "best match" and no language-subtag folding: the locale row is resolved by the exact ladder documented onlocalebelow.
为什么它不在 #18499 里一并修
#18499 的载体集是用探针 best-matching locale|picks the best 枚举出来的,而这句写的是「best-matching row」—— 不在那个探针的命中集内。⇒ 它是同类的第四处,但不是那张卡点名的载体,顺手改会把那张卡的范围撑开。
⭐ 这也是一条关于探针的读数:一个按字面串枚举载体的集合,对同义改写是盲的。修本卡时值得连带问一句:还有没有第五种拼法。
为什么它咬人
该 docblock 随 dist 出货(plugin-email/dist/index.d.ts),⇒ 集成方在 IDE 里读到的就是它。一个在 IDE 里说「best match」、在契约里说「no best match」的接口,会让调用方按不存在的回退行为写代码。
Seam: spec:SendTemplateInput.template → runtime:EmailService.resolveAndRenderTemplate | renderer: none
查重词
TemplateLoader best-matching row · email-service.ts loader docblock best match · plugin-email interface docblock locale · best-matching row (name, locale) · email loader docblock contradicts contract
Generated by Claude Code
Activity
huangyiirene commentedon Sep 21, 2026
Claim: session_01AhQASwqJr2Z7XfGWUdvnbF
Branch:
claude/issue-19507-template-loader-docblock-best-matchSession:
session_01AhQASwqJr2Z7XfGWUdvnbFClause-②: no
认领 ——
domain:services席写于 2026-09-21T11:19Z。⛔ 与卡面不一致处以本评论为现值。
取卡依据:在飞 0;队列 5 张,本卡是 p3 且带
bug⇒ 同级先 Bug,全序第一(次位 #18923 为 p3 无 bug)。p0 实读 4 —— 维护者裁决「没有卡的时候可以处理 p3」,⇒ 队列无更高级别可派时解禁。队列读数与本认领同一次调用取得。前提过时检查(跑了):
email-service.ts:467今日 main 仍为Returns the best-matching row for (name, locale);packages/spec/src/contracts/email-service.ts:153仍为There is no "best match" and no language-subtag folding⇒ 缺陷活着。条款②逐肢:接受集肢 —— 交付物是注释散文,⛔ 不碰解析/准入;公开面肢 —— ⛔ 无新导出符号、⛔ 无载荷新键、⛔ 零
packages/spec。⇒no。domain:services),本席只是按全序取它。Generated by Claude Code
huangyiirene commentedon Sep 21, 2026
os-dev-report
{ "issue": 19507, "status": "done", "branch": "claude/issue-19507-template-loader-docblock-best-match", "pr": "https://github.com/objectstack-ai/objectstack/pull/19591", "session": "session_01AhQASwqJr2Z7XfGWUdvnbF", "premise_still_valid": true, "summary": "Verified all three readings on today's head (base 31184e5daf) before editing: email-service.ts:467 still opened 'Returns the best-matching row for (name, locale)', :470 of the same paragraph still said 'locale set -> an EXACT match for that locale, or null', and packages/spec/src/contracts/email-service.ts:153 still declared 'There is no \"best match\" and no language-subtag folding'. Read template-loader.ts before touching the prose: load() at :120 puts the named locale into the where clause as a literal (no folding), :121 pins DEFAULT_TEMPLATE_LOCALE for a call naming none, and :124 falls to BY_LOCALE, the bundle's lowest locale tag ordered — a deterministic tie-break, not a match quality. So the prose was wrong and the behaviour was right. Rewrote only the opening sentence to 'Returns the row (name, locale) resolves to under the exact rule below — there is no \"best match\" and no language-subtag folding — or null when none exists / inactive', matching the spelling sys_email_template's object doc already carries. Zero behaviour change, zero packages/spec, zero content/docs/releases. Patch changeset for @objectstack/plugin-email because the docblock provably ships in dist/index.d.ts.", "tests": "pnpm --filter '@objectstack/plugin-email^...' build :: exit 0 (under scripts/pm/os-verify-lock.sh, VERDICT command-exit 0). pnpm --filter @objectstack/plugin-email build && typecheck && test :: exit 0 (VERDICT command-exit 0) — 'Test Files 30 passed (30) / Tests 468 passed (468)'. Publish-surface measurement (the changeset's basis): grep -c 'there is no \"best match\"' packages/plugins/plugin-email/dist/index.d.ts = 1, positive control grep -c 'an EXACT match for that locale' = 1, grep -c 'best-matching row' = 0. Gate families derived mechanically, not hand-listed: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack -> 54 families; all 54 run with exit codes captured before any pipe, then reconciled with --ran: '54 derived families accounted for — 53 run, 1 NOT-MEASURED (1 DERIVED from a recorded exit 3)'. NOT MEASURED: pnpm check:dual-build-cjs-loads, reason: exit 3 'PREREQUISITE NOT MET — this gate reads built output, and some package has no dist/' (wants a whole-repo pnpm build); declared to CI, read as neither green nor red. pnpm lint (eslint . --no-inline-config) run WHOLE, not narrowed :: exit 0 at final commit 2ebc3f9ef2. The eight changeset-reading gates plus check:nul-bytes and check:doc-authoring were re-run at 2ebc3f9ef2 after the changeset moved from untracked to committed — all exit 0. Control leg 2 of the dispatch (does anything pin this text?): NO. After the edit git grep -nI 'Returns the best-matching row|best-matching row for' over the tree excluding dist/ returns zero hits; no script under scripts/** and no *.test.ts asserts on this docblock's text — the six files naming TemplateLoader use it as a TYPE. No verification surface was created for a prose fix. No ablation/reverse-verification run and none owed: the diff adds no guard and changes no behaviour, so there is nothing that could be made to fail.", "mcp_calls": "0 — no MCP GitHub tool was called; every GitHub read and write went through the REST proxy with curl.", "api_writes": "2 REST writes — POST /repos/objectstack-ai/objectstack/pulls (draft PR 19591), POST /repos/objectstack-ai/objectstack/issues/19507/comments (this report). Zero label writes: the dispatch named no labels and the diff carries a changeset, so the permitted set is empty. Separately, 3 git pushes on claude/issue-19507-template-loader-docblock-best-match (empty-branch route probe, the edit, the changeset) — git, not REST.", "open_questions": [], "out_of_scope_findings": [ "class: b · packages/metadata-core/src/item-key-discriminators.ts:7-10 quotes EmailTemplateDefinitionSchema as saying 'the service picks the best match for the recipient's locale, falling back to en-US'. Two defects in one: (i) the claim is what packages/spec/src/contracts/email-service.ts:153 denies in as many words, and template-loader.ts:24-30 already enumerates that exact sentence as a FALSE declaration; (ii) it is a STALE quotation — that sentence has zero occurrences anywhere in packages/spec today, because #18499 corrected the schema it cites but never reached this package. This is the fifth carrier and the direct payoff of the card's wider-probe instruction: #18499's literal probe was 'best-matching locale|picks the best', and this spells it 'picks the best match for the recipient's locale'. Seam: spec:SendTemplateInput.template -> runtime:EmailService.resolveAndRenderTemplate | renderer: none. Dedupe words: item-key-discriminators best match recipient locale · metadata-core stale EmailTemplateDefinitionSchema quotation · picks the best match for the recipient's locale · i18n bundle discriminator docblock best match · fifth carrier email locale best-matching", "carrier: this PR's own file, packages/plugins/plugin-email/src/email-service.ts:474 — noted, not filed. Same docblock, the 'locale undefined' branch, reads 'the loader's own deterministic best answer for the name'. Deliberately NOT changed: widening the card to it was the one thing the dispatch forbade, and it is the weakest member of the family — it defers to createSysEmailTemplateLoader and qualifies itself as deterministic, so unlike the other four it asserts no match quality and contradicts no declared contract. It reads as polish rather than a contract violation, which is why it goes to Acceptance notes rather than a card; a seat verdict on whether 'best answer' is in the same family as 'best match' would settle it either way.", "reading, not a finding: packages/apps/setup|studio/CHANGELOG.md and packages/cli/CHANGELOG.md carry 'collapses a doc to the best-matching locale (exact -> primary subtag ...)'. That is the DOCS i18n resolver, which really does fold a subtag, and CHANGELOG.md is release-owned. Correct as written; recorded so the next probe does not re-open it." ] }Generated by Claude Code
docs(plugin-email): the TemplateLoader docblock opened on a "best mat…