Hacker News Reader

You’re wrong about the historical aspects and motivations but I don’t have the energy to argue about it now and will save it for another post. (Vercel isn’t setting React’s direction; rather, they’re the ones who funded person-decades of work under the direction set by the React team.)

I’ll just correct the allegation about the Vite — it’s being worked on but the ball is largely in the Vite team’s court because it can’t work well without bundling in DEV (and the Vite team knows it and will be fixing that). The latest work in progress is here: https://github.com/facebook/react/pull/33152.

Re: people not “getting” it — you’re kind of making a circular argument. To refute it I would have to shut up. But I like writing and I want to write about the topics I find interesting! I think even if you dislike RSC, there’s enough interesting stuff there to be picked into other technologies. That’s really all I want at this point. I don’t care to convince you about anything but I want people to also think about these problems and to steal the parts of the solution that they like. Seems like the crowd here doesn’t mind that.

I also appreciate that you’re doing these explainers so that people don’t have to go the long way round understand what problems exists that call for certain shapes of solutions — especially when those solutions can feel contrived or complicated.

As someone who’s been building web UI for nearly 30 years (scary…), I’ve generally been fortunate enough that when some framework I use introduces a new feature or pattern, I know what they’re trying to do. But the only reason I know what they’re trying to do is because I’ve spent some amount of time running into the problems they’re solving. The first time I saw GraphQL back in 2015, I “got” it; 10 years later most people using GraphQL don’t really get it because they’ve had it forced upon them or chose it because it was the new shiny thing. Same was true of Suspense, server functions, etc.

You can of course still just export a static site and host it on a basic CDN, as you say. And you can self host Next.js in the default "dynamic" mode, you just need to be able to run an Express server, which hardly locks you into any particular vendor.

Where it gets a little more controversial is if you want to run Next.js in full fat mode, with serverless functions for render paths that can operate on a stale-while-revalidate basis. Currently it is very hard for anyone other than Vercel to properly implement that (see the opennextjs project for examples), due to undocumented "magic". But thankfully Next.js / Vercel have proposed to implement (and dogfood) adapters that allow this functionality to be implemented on different platforms with a consistent API:

https://github.com/vercel/next.js/discussions/77740

I don't think the push for RSC is at all motivated by the shady reasons you're suggesting. I think it is more about the realisation that there were many good things about the way we used to build websites before SPA frameworks began to dominate. Mostly rendering things on the server, with a little progressive enhancement on the client, is a pattern with a lot of benefits. But even with SSR, you still end up pushing a lot of logic to the client that doesn't necessarily belong there.

> thankfully Next.js / Vercel have proposed to implement (and dogfood) adapters that allow this functionality to be implemented on different platforms with a consistent API:

Seeing efforts like this (started by the main dev of Next.js working at Vercel) convinces me that the Vercel team is honestly trying to be a good steward with their influence on the React ecosystem, and in general being a beneficial community player. Of course as a VC-funded company its purpose is self-serving, but I think they're playing it pretty respectably.

That said, there's no way I'm going to run Next.js as part of a server in production. It's way too fat and complicated. I'll stick with using it as a static site generator, until I replace it with something simpler like Vite and friends.

> IMO if one has to work this hard to sell/explain a tech there's 2 possibilities 1/ there is no real need of tech 2/ it's a flawed abstraction

There’s of course a third option: the solution justifies the complexity. Some problems are hard to solve, and the solutions require new intuition.

It’s easy to say that, but it’s also easy to say it should be easier to understand.

I’m waiting to see how this plays out.

While RSC as technology is interesting, I don't think it makes much sense in practice.

I don't want to have a fleet of Node/Bun backend servers that have to render complex components. I'd rather have static pages and/or React SPA with Go API server.

You get similar result with much smaller resources.

It's convenient for integrating with backends. You can use async/await on the server, no need for hooks (callbacks) for data loading.

It allows for dynamism (user only sees the menus that they have permissions for), you can already show those parts that are already loaded while other parts are still loading.

(And while I prefer the elegance and clean separation of concerns that come with a good REST API, it's definitely more work to maintain both the frontend and the backend for it. Especially in caes where the backend-for-frontend integrates with more backends.)

So it's the new PHP (with ob_flush), good for dashboards and big complex high-traffic webshop-like sites, where you want to spare no effort to be able to present the best options to the dear customer as soon as possible. (And also it should be crawlable, and it should work on even the lowest powered devices.)

RSCs work just fine with static deployments and SPAs. (All Next sites are SPAs.)

How do you avoid having your users stare at spinners while their browser makes api calls (some of them depending on each other) in order to render the page?

That's fine for you, but not all React users are you. It makes much sense in practice for me.

Every use case has its optimal stack. Isomorphic rendering (NextJS, Nuxt, Sveltekit with "non-static" adapters, ...) is a good fit for very few use cases only.

Many "thought leaders" still don't get the math right. At first visit, your Next app can't serve individual content. So you need two round trips. Both are slow as they are typically served from a Node server.

Serving the app (properly built and bundled, e.g. using Astro or a small and fast SPA like Solid or Svelte) from a CDN and the data from an API is faster at first visit.

At consecutive visits the Next app can serve the rendered page with individual content. Nice and fast. But the CDN hosted app is still in the browser cache. It's even faster! So also just one request to the backend for the individual data is needed.

Regarding SEO the arguments for isomorphic rendering are also flawed. If you care about SEO, just create static html (again, Astro makes it very easy) and put it on a CDN. Why should a crawler care about individual content, that isomorphic frameworks can provide? For SEO the response of a anonymous request matters. So just the static content.

IMO 99% of the use cases are better solved with traditional server rendered MPAs (e.g. Django or ASP.NET MVC), fast SPAs (not React, but Solid, Svelte or Vue) and if SEO and first paint really matter static sites (e.g. Astro).

I think there's a world where you would use the code structuring of RSCs to compile a static page that's broken down into small chunks of html, css, js.

Basically: If you replace the "$1" placeholders from the article with URIs you wouldn't need a server.

(In most cases you don't need fully dynamic SSR)

The big downside is that you'd need a good pipeline to also have fast builds/updates in case of content changes: Partial streaming of the compiled static site to S3.

(Let's say you have a newspaper with thousands of prerendered articles: You'd want to only recompile a single article in case one of your authors edits the content in the CMS. But this means the pipeline would need to smartly handle some form of content diff)

RSC is perfectly capable of being run at the build-time, which is the default. So that’s not too far from what you’re describing.

I find your analysis very good and agree on why companies like Vercel are pushing hard on RSC.

[flagged]

Please don't do this here. If a comment seems unfit for HN, please flag it and email us at hn@ycombinator.com so we can have a look.

Sorry, what? Is it just my phrasing or my rant on VC-backed entities pushing things to gain advantage?

Tangent: next.js is pretty amazing but it’s still surprising to me that it’s become to default way to write react. I just don’t enjoy writing next.js apps even though typescript is my absolute favorite language, and I generally love react as well.

for what it's worth I am a NextJS developer and everyone on my team had a pretty easy time getting used to client/server components.

Do I wish that it were something like some kind of Haskell-style monad (probably doable in TypeScript!) or a taint or something, rather than a magic string comment at the top of the file? Sure, but it still doesn't seem to be a big deal, at least on my team.