Hacktoberfest 2026:维护者为十月标记出来的 issue,仍然开放、适合新手。 浏览 Hacktoberfest issue

SSR + strict CSP for cached pages: let hydration data be inert so inline-script hashes are build-time constants (and streaming works under a hash-based CSP)

未关闭
#4,028 0 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

还没有人认领这个 Issue。

评估

难度
5/5
预计耗时
一周以上
新手友好度
8/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
活跃
技术栈
html, javascript, typescript
领域
frontend, security

调研方向

This is a design proposal with no maintainer response yet, so decide scope with maintainers before coding. Start by reading the inline-script constants (REPLACE_SCRIPT, HEAD_SCRIPT, HydrationScript) and the renderToStream / renderToString entry points the issue names. Done means a failing test that shows which inline scripts change between renders, as the issue's first deliverable.

由索引模型根据 Issue 内容生成。

描述

Problem. We run a Solid 2 storefront (rc.13, @solidjs/vite-plugin start mode) behind a
shared CDN cache, with a CSP that has no 'unsafe-inline'. Three facts collide:

  1. A nonce is unusable on a cached page. The stored response publishes its CSP header and its
    <script nonce> attributes together, to every later visitor, for the whole cache lifetime. The
    published constant becomes the grant. SvelteKit forbids nonces on prerendered pages for the
    same reason, and Next.js documents that nonces force dynamic rendering and disable static, ISR
    and CDN caching.
  2. Hashes are the remaining mechanism, but Solid's inline scripts cannot be hashed ahead of
    time.
    • Every script that carries hydration data is per render: the seroval cross-reference header,
      the stub batch, the resolutions, $df("k"), head ops, window.location=.
    • The constant runtimes (REPLACE_SCRIPT, HEAD_SCRIPT, seroval's promise-resolver factory)
      are concatenated into the same element as that data.
    • Only HydrationScript is a standalone constant.
      So a host must render, re-read the finished body, and hash the scripts. That only works with
      the async render mode, where the head is sent after the body.
  3. Streaming is impossible under a hash-based CSP. The header is gone before the late scripts
    exist.

A side effect we measured: in async mode, the order of the resolution chunks (and their $R ids)
follows the order in which async sources settle. Identical data with different API latency gives
different bytes (12 renders, 2 variants), so a body-digest ETag misses revalidations that should
be 304s.

Proposals, ranked.

  1. Inert hydration data plus constant executable code (opt-in, for example
    renderToStream(…, { dataMode: 'json' })):
    • Write every serialized record as <script type="application/json" data-hy> using the JSON
      codec @solidjs/web/serialization already ships (createJSONSerializer /
      createJSONDataTable).
    • Keep all executable bytes constant per build: the runtime in its own <script>, or an
      external module, and a fixed pump such as <script>_$HY.p()</script> after each flush.
    • Make the late redirect a <meta http-equiv="refresh"> or a data record.
    • Two details to handle: the decoder must stamp p.s / p.v on settled promises, as
      PROMISE_SUCCESS does today, since hydration reads them synchronously; and seroval's
      cross-JSON already resolves out-of-order promise nodes by id.
      Result: CSP hashes are build-time constants (a static header, no body re-parse), and streaming
      SSR works under a strict, hash-only CSP
      . The cost from our measurement is about 5% larger
      payloads after compression and slower decode (0.7 vs 0.1 ms for a 120-product payload on a
      laptop).
  2. Interim, small: the renderer records the SHA-256 of every inline script it writes and
    exposes the set on the settled result (or the request event), for hosts that render in async
    mode or with renderToString. Authored <script> in JSX stays the app's job, since the
    compiler bakes it into the template.
  3. Determinism: in async mode, emit resolutions in a stable order (for example, by
    serialization key at flush) rather than in settle order, so identical data gives identical
    bytes.

Prior art.

  • SvelteKit: csp.mode: 'hash' | 'nonce' | 'auto'. The framework adds hashes or nonces for
    the inline scripts and styles it generates. auto uses hashes for prerendered pages and nonces
    for dynamic ones, and "using nonces with prerendered pages is insecure and therefore forbidden".
  • Astro 6: security.csp computes SHA-256 hashes of the page's scripts and styles and emits
    them in a <meta http-equiv> policy.
  • Next.js: nonces only for inline scripts, which require dynamic rendering. Experimental SRI
    covers external files only.
  • React Router / Remix and Qwik: nonce props only.
  • SolidStart v2 already moved server-function payloads to seroval's JSON mode by default
    (solid-start#2042, after #1825).

What we would contribute.

  • A failing test for each claim above: the per-render inventory, and the settle-order variance.
  • A prototype of (1) behind an option: the serializer switch, the pump and the decoder wrapper
    that stamps s / v.
  • Size and decode benchmarks on a real catalogue page.
  • The docs section on running Solid behind a shared cache with a strict CSP.

We are happy to start with (2) or (3) if (1) is too large for 2.0.

主要语言
TypeScript
星标
36.1k
派生
1.1k
平均合并
10 小时 41 分钟
30 天内合并 PR
363

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。

solidjs/solid 的其他 Issue

查看 solidjs/solid 的全部 Issue

相似的 Issue

更多 TypeScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。