Server polyfills are not user-controllable: no way to opt out of `@angular/platform-server/init` or add server-only polyfills
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 48/100
- Issue 类型
- 功能
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
- 技术栈
- angular, typescript
- 领域
- build-system
调研方向
Start with application-code-bundle.ts at createServerPolyfillBundleOptions and trace how options.polyfills is consumed in options.ts, execute-build.ts, dev-server/vite/index.ts, and browser-esbuild. The issue proposes a normalization step and separate browser and server lists; check the existing normalization patterns and add the server polyfills spec mentioned under “Anything else relevant?”. Done means the object form works while array form preserves current behavior, with tests covering the server bundle.
由索引模型根据 Issue 内容生成。
描述
Command
build, serve
Description
@angular/build:application derives the server polyfills list entirely on its own and offers no option to change it. In createServerPolyfillBundleOptions (application-code-bundle.ts#L164-L184):
zone.js/node(orzone.jsonplatform: 'neutral') is added when the browserpolyfillslist containszone.js,@angular/localize/initis added when the browser list contains it,@angular/platform-server/initis always added.
So the browser polyfills list is fully explicit, and the server polyfills list is fully implicit. Two concrete gaps follow from that:
1. There is no opt-out of the init global patch
@angular/platform-server/init is a side-effectful module whose only job is Object.assign(globalThis, domino.impl) plus globalThis.KeyboardEvent = domino.impl.Event. The framework introduced it as something an app may import before bootstrap for code that touches DOM globals at module-evaluation time (angular/angular#40559). Most apps do not need it: platform-server runs the identical shim itself in DominoAdapter.makeCurrent() → setDomTypes() on every platform init, so for everything that runs after bootstrap the init patch is redundant and is overwritten on the first render anyway.
Since v15 the CLI adds it to every server build with no way to say no. Evidence that people need the opt-out:
- The framework's own SSR benchmarks patch
@angular/buildto strip theplatform-server/initimport frompolyfills.server.mjs(modules/ssr-benchmarks/README.mdin angular/angular), and themain.server.tsthere notes thatɵENABLE_DOM_EMULATION"won't prevent the monkey patching of DOM apis as this is introduced by the CLI". - Anyone running the server bundle outside Node with a real DOM (the benchmark case, or
ENABLE_DOM_EMULATION: false) gets Domino's classes forced ontoglobalThisregardless. - Apps that want to keep
globalThisclean on the server (noHTMLElement,Node,Event, … leaking into third-party code that feature-detects the environment withtypeof window/typeof HTMLElement) have no option either.
Measured on a default ng new --ssr app (details in #34514): the init module costs ~235 kB on disk, ~1.1 MB of retained heap for the life of the process and ~7 ms of startup, while the Object.assign itself is negligible (~7 KB, 0.03 ms). Prerendered and SSR output are byte-identical with and without it.
2. There is nowhere to put a server-only polyfill
The ssr.platform option's own description says "Please note that this feature does not provide polyfills for Node.js modules." With platform: 'neutral' (Cloudflare Workers and similar), users are expected to bring polyfills themselves, yet the only polyfills option feeds the browser bundle. Today a server-only polyfill has to be imported manually at the top of main.server.ts, where it runs after zone.js and init and after any shared chunks.
Describe the solution you'd like
Let polyfills take either the current array or a browser/server object, mirroring the existing outputPath shape (string | { base, browser, server, media }):
"polyfills": ["zone.js"] // today, unchanged
"polyfills": {
"browser": ["zone.js"],
"server": ["zone.js/node", "@angular/platform-server/init"]
}
Semantics:
- Array form keeps today's behaviour exactly:
browser= the array,server= derived as now (zone.js/node, localize,init). Non-breaking; no migration needed. - An explicit
serverlist is the server polyfills list, the same way the browser list already is: the builder adds nothing on top. To drop theinitglobal patch, omit it. To keep today's output, list it. Theng new/ssrschematics can emit the explicit form so new projects see what they get. isZonelessAppis evaluated per list (browser list for the browser bundle, server list for the server bundle), sozone.jsvszone.js/nodeand the async/await support flags fall out naturally.getEsBuildCommonPolyfillsOptionsand thepolyfills.serverbundle are unchanged; only the input list is. Normalization follows the existingnormalizeOptimization/normalizeSourceMapspattern (a smallnormalizePolyfillsinoptions.ts, folding in the string-to-array shim thatbrowser-esbuildstill relies on).- Consumers that read
options.polyfillsand would switch to.browser/.server:application-code-bundle.ts(browser bundle, server bundle,isZonelessAppchecks),execute-build.ts(top-level-await hint),dev-server/vite/index.ts(es2016target), andbrowser-esbuild's string handling.
This solves both gaps with one option and no new concept: server polyfills become as explicit as browser polyfills.
I am happy to implement this and open the PR once the shape is agreed; the single-copy fix in #34516 is independent of it and already covers the duplicate.
Describe alternatives you've considered
A boolean under ssr, next to ssr.platform, which already shapes the server polyfill list (zone.js vs zone.js/node), the createRequire banner and the xhr2 external:
"ssr": { "entry": "src/server.ts", "platform": "node", "domShims": false }
One if around line 184. It only addresses gap 1, and the name needs care so it is not confused with the framework's private ɵENABLE_DOM_EMULATION, which controls the adapter and does not affect the global shims.
Leaving it manual: users who want init could import it in main.server.ts as the v12 schematic did. That would be a breaking change for apps that rely on the implicit import, and it still leaves gap 2.
Not an alternative, but related: the same unconditional init import is also why the server build currently ships two copies of Domino (#34514). That has a bundling-level fix that does not need this option; this request is about control, not about the duplicate.
Anything else relevant?
- No spec or e2e currently asserts anything about
polyfills.server.mjs,zone.js/nodeor@angular/platform-server/init;tests/options/polyfills_spec.tsonly coversdist/browser/polyfills.js. Either solution should add a server polyfills spec. - The webpack server builder only adds
initwhen@angular/platform-serveris installed (isPackageInstalledinwebpack/configs/common.ts); the esbuild path has no equivalent guard. - Repro and measurements: https://github.com/ChristopherPHolder/ng-double-domino
- 主要语言
- TypeScript
- 星标
- 27k
- 派生
- 11.8k
- 平均合并
- 1 天 2 小时
- 30 天内合并 PR
- 241
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
angular/angular-cli 的其他 Issue
-
area: @angular/build gemini-triaged
难度 2/5 1-3 小时 新手友好度 74/100
angular/angular-cli#33955 ·
维护者通常 1 天内回复
-
area: @angular/cli gemini-triaged
难度 2/5 1-3 小时 新手友好度 72/100
angular/angular-cli#33055 · 1 条评论 · 3 个 reaction ·
维护者通常 1 天内回复
-
area: @angular/build
难度 3/5 1-2 天 新手友好度 55/100
angular/angular-cli#34524 · 2 个 reaction ·
维护者通常 1 天内回复
-
`ng generate app-shell` and `ng generate module --route` produce broken imports when a component `type` is configured可能已有人在做 @MeAkib 于 5 天前认领。 未关闭area: @schematics/angular
难度 3/5 1-2 天 新手友好度 35/100
angular/angular-cli#34449 ·
维护者通常 1 天内回复
-
Tailwind CSS v4 classes in new or moved files are not picked up by ng serve until another rebuild可能已有人在做 @MeAkib 于 6 天前认领。 未关闭area: @angular/build
难度 4/5 3-5 天 新手友好度 35/100
angular/angular-cli#34446 ·
维护者通常 1 天内回复
查看 angular/angular-cli 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 72/100
VictoriaMetrics/victorialogs-datasource#747 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 79/100
维护者通常 1 天内回复
-
bug from-studio
难度 2/5 1-3 小时 新手友好度 62/100
esengine/DeepSeek-Reasonix#14608 ·
维护者通常 1 天内回复
-
triage
难度 2/5 1-3 小时 新手友好度 75/100
getarcaneapp/arcane#4590 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 77/100
Clouditera/VulnHunter#57 ·