Resolver: Read `exports` from the package root, not the closest `package.json`

未关闭
  • 哪里不对？
#2,153 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

@robhogan 已经在做这个了。

开始于 2026年11月4日。

  • #2174 来自 @robhogan —— 未关闭

评估

难度
4/5
预计耗时
3-5 天
新手友好度
22/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
活跃
技术栈
javascript, react-native
领域
build-system

调研方向

Start in packages/metro-resolver/src/resolve.js around the line where the package is looked up once (about L420), and trace how resolveFromNodeModulesPath, extraNodeModules and Haste pass the package root down. The change adds an opt-in resolver.unstable_resolveExportsFromPackageRoot flag that reads "exports" via getPackageForModule(packageRoot) when the subpath is empty. Done means the flag gates the new behaviour, existing tests pass with it off, and new tests cover nested package.json with root exports.

由索引模型根据 Issue 内容生成。

描述

Metro reads "exports" from the package.json closest to the requested subpath, not the one at the package root. So a nested {"type": "commonjs"} or a legacy shim package.json hides the root's "exports", which is different to Node's PACKAGE_RESOLVE, webpack and Vite. Previously reported in #1517 and #1582, this is to track the fix.

Examples I've seen in real apps (AI-driven comparison): @floating-ui/utils/dom on Bluesky web gets the legacy shim's build instead of dist/floating-ui.utils.dom.mjs, and on Mattermost, ESM imports of rxjs/operators get dist/cjs while rxjs gets dist/esm5. @kitten and @Nezz have seen the same thing break swr.

The fix should be small now that #2134 looks the package up once, here:

https://github.com/react/metro/blob/9e61545565438f265ffbced0322a97aea9908888/packages/metro-resolver/src/resolve.js#L420

Metro converts specifiers to absolute candidate paths early, so the caller has to pass the package root down. resolveFromNodeModulesPath has parsedSpecifier.packageName, extraNodeModules has the mapped directory and Haste has the directory of its package.json.

  • For "exports", use getPackageForModule(packageRoot). It gives a real rootPath, so pnpm symlinks work, and packageRelativePath comes from the specifier's subpath.
  • If packageRelativePath isn't empty, there's no package.json at the root, so keep using the closest to the target.
  • "browser" redirections, "imports" and self-references all keep using the closest package.json, which is correct for those.

This should go behind resolver.unstable_resolveExportsFromPackageRoot, off by default, so that we can get it out in 0.87.x without changing what anyone resolves, and then enable it by default for 0.88.

It's breaking because some packages only work on native through a nested "browser" field that this bypasses. As @Om-singhaI found, @mswjs/interceptors/fetch goes from its browser build to its Node build, which needs node:zlib. nanoid/async (nanoid 3) also moves, and needs unstable_applyBrowserSpecAfterExports on as well to avoid importing crypto. Subpaths with a nested package.json that aren't in the root's "exports" will now warn, then fall back as they do today.

主要语言
JavaScript
星标
5.7k
派生
711
平均合并
3 天 14 小时
30 天内合并 PR
77

环境准备

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

react/metro 的其他 Issue

查看 react/metro 的全部 Issue

相似的 Issue

更多 JavaScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。