fix: rework browser optimizeDeps.entries

[hi-ogawa]

• closes rework browser optimizeDeps #562

I thought having this is safe, but in practice, it looks like this would cause pre-bundling error (or just warning?) since it goes through a lot of server only dependencies whose resolutions somehow fail on browser resolver.

One example is when crawling @node-rs/argon2 hi-ogawa/lucia-auth-examples#1 (comment)

Not having this is okay since this is only to prevent dev full reload on late dep discovery (also users still have a way to configure it on their own), but there might be still a cheap attempt to mitigate this.

---

Some ideas:

• remove automatic optimizeDeps.entries and let users configure it (with optimizeDeps.exclude to counter against if they need)
• custom esbuild plugin to crawl only "use client" files?
  • we don't probably have to go deep and we can only check direct dependency of all routes/**/* files
• can we push to optimizeDeps.entries after we collected clientReferenceMap by loading server routes

[hi-ogawa]

• fix: rework browser optimizeDeps.entries #560 👈
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

Join @hi-ogawa and the rest of your teammates on Graphite

[pkg-pr-new]

Run & review this pull request in StackBlitz Codeflow.

commit: d4ea790

@hiogawa/react-server

npm i https://pkg.pr.new/hi-ogawa/vite-plugins/@hiogawa/react-server@560

@hiogawa/react-server-next

npm i https://pkg.pr.new/hi-ogawa/vite-plugins/@hiogawa/react-server-next@560

@hiogawa/transforms

npm i https://pkg.pr.new/hi-ogawa/vite-plugins/@hiogawa/transforms@560

@hiogawa/vite-plugin-ssr-middleware

npm i https://pkg.pr.new/hi-ogawa/vite-plugins/@hiogawa/vite-plugin-ssr-middleware@560

---

templates

• default

[hi-ogawa]

Hmm, it doesn't look really possible to precisely detect dependencies like @node-rs/argon2. Probably hard-coding them like Next.js https://github.com/vercel/next.js/blob/canary/packages/next/src/lib/server-external-packages.json is the best bet... (cf. #545)

Closing for now.