fix(lint,metadata): 撤销 http.server 豁免 —— 它的理由本身是错的 (#4251) - #4382
Merged
Conversation
#4321 put `http.server` in UNCONTRACTED_SLOTS on the ground that "no IHttpServer contract exists". The contract does exist (packages/spec/src/contracts/http-server.ts) and eight call sites were already resolving the slot as getService<IHttpServer>(...) when the exemption was written. An exemption is a claim like any other, and this one rested on a premise nobody checked -- the same shape as the gaps the rule exists to find. Revoked. That surfaced 9 erasures the exemption had been hiding: 7 in files never grandfathered, 2 as count growth inside grandfathered ones, none of which the baseline could legally absorb. All typed to IHttpServer; packages/metadata/src/plugin.ts came out clean entirely, so the baseline ratchets DOWN to 168 sites in 36 files and loses a file. Confirmed and reported, not changed: `http.server` and `http-server` are the SAME instance under two names (plugin-hono-server and qa/node-plugin each register it twice, two lines apart; runtime's config.server path registers only the dotted one) -- removing a registration is a runtime change and belongs with whoever picks the canonical name. And `IHttpServer` is defined TWICE and has already diverged: spec's (15 importers) declares write?()/end?(), core's (8 importers) does not. Spec's is the superset and the one the ledger points at; core's is a stale near-copy that should re-export it, in its own change. Verified: ratchet holds (168/36, none new, no files added vs d92c72d); eslint clean; metadata dts build succeeds; runtime 995/68, client 204/15, metadata 281/13, http-conformance 46/2. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
github-actions
Bot
added
documentation
Contributor
📓 Docs Drift CheckThis PR changes 1 package(s): 7 hand-written doc(s) reference the affected code and may need an implementation-accuracy re-verification:
|
os-zhuang
added a commit
that referenced
this pull request
Jul 31, 2026
…erver 定为 canonical 并入账 ledger (#4251) (#4393) * fix(spec,core,cloud-connection,metadata): one HTTP contract, one canonical slot name (#4251) packages/core/src/contracts/ was a dead near-copy of the real contracts -- zero importers (no relative import, no subpath export, not a tsup entry; core's barrel has re-exported the spec versions all along) -- and it had already DIVERGED (spec's IHttpResponse grew write?/end?, IHttpRequest grew rawBody?; the copy never did). Anyone who grepped into it read a stale contract nothing enforces -- the both-humans-and-AI failure mode behind the false http.server exemption (#4382). Deleted; zero-risk by construction. `http.server` is the canonical slot name and the ledger now says so: ServiceSlotContracts gains 'http.server': IHttpServer plus the deprecated 'http-server' alias entry (same instance -- hono-plugin, qa node-plugin and cloud's two server entrypoints all register both, alias commented "backward compatibility"). Canonical is the only name on EVERY provider path -- runtime's config.server path registers no alias, so the three cloud-connection plugins reading the alias alone found an empty slot there. All readers now go canonical-first with the alias as a fallback that dies with the alias registrations; registrations untouched this release, both sites carry the deprecation note. getRawApp?(): any joins IHttpServer -- the deliberate framework-handle escape, declared once with the rationale; four consumers declared it locally before (cloud-connection x2, metadata HMR, cloud node-server), and the local RawAppHost / HttpServerWithRawApp types are deleted. IMetadataService.bulkRegister/bulkUnregister declare the write options their implementation always accepted (bulkRegister's contract dropped the MetadataWriteOptions half it intersects in; bulkUnregister declared no options at all). Same shape as B2's IDataEngine read-methods gap. Baseline 168 -> 167 (marketplace-install-local's lookup typed while touched). Ledger test pins both slot entries and the alias equality. Verified: spec build (dts) + 7192 tests / 281 files; core, metadata, cloud-connection, plugin-hono-server dts builds; runtime 1001/69, rest 539/36, plugin-auth 579/26, plugin-sharing 226/11, service-settings 196/14, metadata 281/13, hono-server 135/12, http-conformance 46/2; ratchet holds 167/36 none new; eslint clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * fix(cloud-connection,metadata): the canonical-first read needs a per-name try -- getService throws on an empty slot (#4251) CI caught the shape my local run could not (I built cloud-connection but ran only metadata's tests -- cloud-connection's own suites mock a kernel that registers ONLY the alias): `getService` THROWS for an unregistered slot, so try { a = getService('http.server') ?? getService('http-server'); } catch {...} never reaches the alias -- the first name's throw exits the whole try. Split into a per-name try (readServer helper) in all four readers. Worth recording: the pre-existing alias-first read in metadata/plugin.ts had the SAME shape, so its `?? getService('http.server')` fallback never once fired either -- a decorative fallback, the declared-vs-actual gap this work line keeps finding, now actually implemented in both directions. Verified serially (the earlier 6-file FAIL was local vitest concurrency noise while spec's dts build ran in parallel -- single-file and serial reruns green): cloud-connection 64/12, metadata 281/13, both dts builds, eslint clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> --------- Co-authored-by: Jack Zhuang <277994282+os-zhuang@users.noreply.github.com> Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
清单里的 ①。
那条豁免的理由是错的
#4321 把
http.server写进UNCONTRACTED_SLOTS,理由写的是"不存在IHttpServer契约"。该契约存在 ——packages/spec/src/contracts/http-server.ts—— 而且写下这条豁免时,已经有 8 个调用点在用getService<IHttpServer>(…)。豁免和别的断言一样是一个 claim,而这一条建立在没人核过的前提上 —— 正是这条规则本身要抓的那种形状。
撤销之后露出 9 处
两种基线都不能合法吸收。全部类型化到
IHttpServer;packages/metadata/src/plugin.ts因此整个清空,基线下降到 168 站点 / 36 文件,并少掉一个文件。8 个测试站点调
getPort()(契约上是可选成员),按契约写成getPort!()—— 它们刚 bootstrap 完自己的 server,这个断言是准确的。另外两点:已确认,只报告不改
http.server与http-server是同一个实例的两个注册名。 plugin-hono-server 和 qa 的 node-plugin 各自并排注册两行;runtime 的config.server路径只注册带点的那个。metadata/src/plugin.ts用??同时读两个 —— 这就是它一直没暴露的原因。本 PR 不删任何注册 —— 那是运行时行为变更,该由决定规范名的人来做。IHttpServer有两份定义,而且已经分叉。packages/spec/src/contracts/http-server.ts(15 个导入方)声明了write?()和end?();packages/core/src/contracts/http-server.ts(8 个导入方)没有。spec 那份是超集,也是 ledger 指向的那份,所以它是源;core 那份是过期的近似复制,应当改为 re-export。单独一个变更 —— 合并一份重复契约不是 lint 修复。给写更宽 HTTP 契约的人留一条证据
getRawApp()现在有了第三个独立消费方(metadata 的 HMR 路由,加上 cloud-connection 的两个)。它刻意不在IHttpServer上 —— 契约框架无关,raw app 是框架自己的句柄 —— 所以每个消费方各自具名声明。三个足够用来判断这是不是仍然是对的答案了。验证
pnpm check:slot-lookup强制模式:✓ 168 unswept site(s) in 36 file(s), none new;no files added(对 d92c72d)eslint --no-inline-config全部改动文件 —— 干净metadatadts build 成功(该包在 type-check DEBT 名单,build 才是类型门 —— 这是上个 PR 的教训)runtime995 / 68、client204 / 15、metadata281 / 13、http-conformance46 / 2 —— 全绿🤖 Generated with Claude Code