<inhales> It has been more than 4 years since I started Web Dev and to this React was a very specific “ick” for me, everything about it felt… Off in some way, and this was before me discovering Preact, Solid, Vue, many more frameworks, and how cleaner their patterns were. Signals, SFCs, and many other things were one of the several things which set them apart as “React Alternatives”, and really good ones, so much so I could probably devolve this article into why you should stop using React as a whole, and maybe I will. As of Recent, but by recent I mean like, maybe 2-3~ years??? React has focused an emphasis on what they dub “React Server Components”, now, on first thought you may have the following responses:
- “Huh, I wonder how different this is from Server Rendering solutions from Next.js and Remix."
- "… Why is React, a client-side UI rendering library, looking into something that only occurs on a server?"
- "Do they enhance how server rendering works right now? What is this? Is this on top of existing server rendering?"
- "Is this a really out of place April Fools’ joke?”
If you thought any of that, and are sane, you’d very disappointed to hear that React just threw away everything that made it decent, and just took the worst ways of going about creating “Server Components”, and essentially, reinvented PHP but even more bare bones. What do I mean by this? To answer that question, it runs very deep. Today, I read an article from some idiot on why you should start using “Server Components” in all your projects and reading it really cemented in how asinine the crowd it is for, and the technology itself is. If you are interested in reading that original bit, it’s linked here. But I suppose we really need to get into the knitty-gritty of this thing from the start, well supposedly explaining where my hatred for a lot of things stem from.
A Brief History of React, JSX, and Server Components
It was 2011, Angular.js and Backbone were still relatively fresh at the time, but Facebook was cooking up their own alternative internally for their own usage, they probably didn’t predicate how their little project would impact a good chunk of web development history going forward. Don’t remember if React (as in the JavaScript Library) or React Native was the internal result of a hackathon, but one of them was. React was initially presented as a “lighter” approach of client-side reactive UI components. At the time, it really was, a lot of people using Angular.JS (not to be confused with Angular) really did say and confirm that React felt a lot faster. But there was a catch, no one really wanted to write nested function calls, no, that looked way too hideous to deal with, except some people thought the imperative function calls were “good” and dissented JSX like the plague.
JSX was introduced as an extension of the JavaScript syntax to help write React code better, it has since bloomed into a general grammar specification which is utilised by many in the JavaScript ecosystem, primarily by Solid and Preact, but Vue also presents JSX/TSX as an option to write Vue in (cool side note: Vue’s specific JSX transform will be updated according to Evan You himself to accomodate Vapor Mode once it launches), but internally, also is used by Svelte to do typechecking via TypeScript. I do personally love JSX very much as a templating format, it can turn into symbol soup but it’s more related to how JavaScript itself functions, but I suppose technically when you have a compiler you could allow for more. And Sure, “seperation of concerns” is a thing, but really, think about it, if it really was that big of an issue why do you think we are using Component Frameworks to isolate bits and pieces into one reusable chunk? SFC files for example are similar to JSX with the same goal but go about it in a very different way. Anywho, moving onto “Server Components”.
The term “Server Components is a broad term in web development itself but has been around for a very long time. When thinking of them, a somewhat predictive thought would be thinking about PHP and it’s frameworks, or even something Ruby on Rails (albeit it being comically a big example of how to not do things), but in context of JavaScript, people may think of Templating Engines like Pug, Liquid, EJS, Handlebars, so many more, but generally speaking anything that was a reusable template or component which only rendered to HTML on server and then sent to the client would quality as a server component. So, where did React lie on this spectrum? React could always render it’s components to static HTML ahead of time as far my memory goes, at least it could since 2013~. You’d simply call the renderToString()
(do note: this method call has been deprecated in favour of other APIs) method from "react-dom/server"
package or call the other methods in the package to render the HTML to a stream or anything else. You might be thinking, then what is the case for React’s own take on Server Components? To answer that question in one word: clownery. So what can they do that client components can’t? Simply put the only thing they are about are providing some sort of pre-rendered templates which are hydrophobic, not hydrophobic as in they water, hydrophobic in the sense they literally cannot “hydrate” by default compared to normal components rendered on server. RSCs also feature a major change, these components can be “async”, but normal components cannot be async (until the use
hook drops in stable), and this creates a clear divide which causes more friction rather than creating less friction.
The Next.js Incident
A while back, Next.js introduced support for RSCs in v13 and in all honesty, it went downhill from there. It introduced the App Router, which had a new way of File System Routing attached with it. Rather than relying on <Suspense />
directly in a route to stream the UI, a file named loading.js
must be created, this not only convoluted the whole thing, but it is an unnecessary system all together when compared to other solutions Next.js “yoinked” this from. Although it is indeed possible to use <Suspense />
to stream the UI, Next.js opts to promote their new File System Routing to achieve the same effect albeit having a lot of negatives attached to said approach of handling of streaming UI over the wire. So that implies you need to do this:
export default function Loading() {
return <LoadingSkeleton />;
}
Over something simpler like this:
export default function Route() {
return (
<Suspense fallback={<LoadingSkeleton />}>
<UI />
</Suspense>
);
}
Another thing that should be mentioned is that Next.js actively allows and borderline encourages “Component Level Data Access” despite it being not only a bad practice, but a security risk too, as to why; it will become apparent in a bit. To counter that, Sebastian Markbàge, a former Meta Developer who now works at Vercel, has written a blog post when it comes to handling security within the context of a Next.js Application.
Server and Client directives introduce noticable overhead for the developer mentally, but also introduce friction between interactions, it is one thing to be simpler and efficient, it is another to badly rip-off something else. At the end of the day, this does feel like a wanna-be PHP clone with the usage of the "use server"
directive:
export function Bookmark(slug) {
return (
<button
formAction={async () => {
"use server";
await sql`INSERT INTO Bookmarks (slug) VALUES (${slug})`;
}}
>
<BookmarkIcon />
</button>
);
}
Next.js despite advertising to be “modern” and “friendly” likes to introduce complicated and heinous practices which are the polar opposite, including things like security as mentioned above. A tweet by Tom Sherman points out how Next.js RSCs allow you to easily leak secrets to the client instead of them staying on the server as you’d expect, again; bad design leads to bad practices.
https://twitter.com/tomus_sherman/status/1654211227172716545
Pardon me, but I am about to go off on an unrelated tangent. Over the past few months, or years even, Next.js has focused on releasing fast with more and more breaking changes, instead of slowing down with more quality changes and refinement over time, the buggy-ness of the framework has gone quite up, and the general quality took a nose dive to hell, what used to take a few hundred milliseconds for HMR now takes a solid 1-2 seconds to be accomplished. Next.js also seems to be against the JavaScript community in a way, trying to split apart and do their own thing for no reason other than “just because”, unlike with others finally shaking hands and collaborating with each other and using more and more shared tooling, a popular example would be Remix and Angular now using Vite internally, another would be Solid introducing Signals to the JavaScript World properly, there’s a lot more examples tho, including UnoCSS by one of the Vue Developers.
Another thing that should be accounted for is, the existence of Turbopack, my only question is: Why?
Turbopack has not only done a lot of false advertising by misleading benchmarks, but it has also more or less failed to meet the goal of a “Rusty Webpack Implementation”, the community already had a lot of great and really good interconnecting tools, we had reached a point of refinement with our current generation of build systems, a lot of old build systems have really good almost drop-in in replacements, such as:
- Parcel -> Vite (Not Really but Vite is the closest successor to Parcel 1.x besides Parcel 2)
- Rollup -> Rolldown (A Rusty Rewrite of Rollup)
- Webpack -> Rspack (A Rusty Rewrite of Webpack)
https://twitter.com/youyuxi/status/1587279357885657089
Checkout Evan You’s benchmark where he debunks these claims made by Vercel.
Hey, isn’t caching in Next.js good like other advertised features?
Contrary to popular belief as usual, no, it is not; rather it is far more than what it should have been. Next.js by default caches everything, and to not cache means you need to explicitly opt yourself out of caching things, which makes it really easy to shoot yourself in the foot in a lot of cases, and if I really wanted to shoot myself in the foot, why not use C++ for that instead?
When you’re aiming to bypass caching for individual data fetches, setting the cache
option to no-store
does the trick:
fetch(`https://...`, { cache: 'no-store' })
Now, to circumvent caching at the route level, delve into the Route Segment Config options. Here’s how you can handle it:
export const dynamic = 'force-dynamic'
However, there’s another option using the unstable API called unstable_noStore to opt out of caching at the component level. They recommend folks to utilize this over I get where you’re coming from. Sometimes, all you want is to fetch dynamic data without worrying about caching. It seems like these options make things a bit more convoluted than necessary, doesn’t it? It’s very easy to serve stale data to the user or even share private data by mistake because it’s shared and persistent if you are not careful. This tweet by Flavio explains why it’s so bad, another thing to note is that there is a GitHub Discussion which really concerns me on why this is what to is; Sad reality, honestly.
https://twitter.com/flaviocopes/status/1736317822609887362
Remix Loaders and Data Fetching
About three years ago, a framework named Remix started catching up hype, and rightfully so for a lot of reasons, it had been around for way longer but closed source, they went open source around that time ago and with some really good clear-cut branding that aligned with their goals; Using the existing Web itself to push it forward. Naturally, they introduced web standards based solutions including their built in components and re-exports of React Router. One of the many good things about Remix were Data Loaders, a way to load server data without any leakage to the client, as a React Hook. This is a neat way to load data without worrying about separation of concerns. Here’s how you are going to load data in Remix.
export async function loader() {
const user = await getUser();
//loading data in parallel without creating waterfalls
const [notifications, comments, posts] = await Promise.all([
getNotifications(user.id),
getComments(user.id),
getPosts(user.id)
]);
return json({
notifications,
comments,
posts
});
}
export async function clientLoader({ request, params, serverLoader }) {
const [serverData, clientData] = await Promise.all([
serverLoader(),
getDataFromClient()
]);
return {
...serverData,
...clientData,
};
}
export default function Route() {
const data = useLoaderData<typeof loader>();
// render your data
// ...
}
This not only is simpler to understand and ergonomic in nature, but it also makes sure you aren’t leaking anything easily, if at all to the client. You can read more about how Remix does Data Fetching here. Remix always keeps your data in sync with your UI. Additionally, instead of putting server-side code in your component to compute form data, you can simply use an API called Data Actions, which are both highly secure and really ergonomic to work with.
export async function action({ request }) {
const body = await request.formData();
const bookmark = await createBookmark(body);
return redirect(`/bookmark/${bookmark.id}`);
}
export function CreateBookmark() {
const actionData = useActionData<typeof action>();
return (
<Form method="post" action="/bookmarks/new">
<p>
<label>
Name:{" "}
<input
name="name"
type="text"
defaultValue={actionData?.values.name}
/>
</label>
</p>
{actionData?.errors.name ? (
<p style={{ color: "red" }}>
{actionData.errors.name}
</p>
) : null}
<p>
<label>
Description:
<br />
<textarea
name="description"
defaultValue={actionData?.values.description}
/>
</label>
</p>
{actionData?.errors.description ? (
<p style={{ color: "red" }}>
{actionData.errors.description}
</p>
) : null}
<p>
<button type="submit">Create</button>
</p>
</Form>
);
}
Remix’s approach to Data Modelling and providing APIs for such is far more ergonomic and cleaner than say, Next.js’ approach to Data Modelling in context of RSCs.
Common Misconceptions and Fanboyism
”React Server Components have zero-bundle cost.”
A common misconception that is seen across Vercel Glazers is that they think they are getting free performance by using Server Components, this however, is not true, quite the contrary in reality. RSCs render on the server, and still need to be sent over wire with their own bundle cost, in case of Next.js, it has shown to often remain the same or slightly decreased bundle size rather than decrease or be zero as advertised, another is runtime performance; RSCs do “hydrate” if you put a client-directive in them, which defeats the point of having RSCs in the first place, to have fully server-only rendered bits of UI. The popular form example where RSCs were first demo’d with the server-directive do actually hydrate and attach dynamic behaviour to the client-side DOM.
What about Redwood.js?
Same shit; Different Disc. Redwood copies Next.js RSCs word for word; bar for bar. Same plague, different realm. It’s an understandably safe play but a disappointing one none-the-less.
What about Vercel’s design? Why do you hate it to so much?
Like I said, it introduces more friction and fucks with both UX and DX, which is why I hate it, and as I talk about how others handle it better in the next section, it will start to become far more clear.
Just because Vercel is very oriented on vendor lock-in and making money off of it, and once bought a renaissance to the server-rendering space, doesn’t mean all their decisions are good or valid, they’re driven by money like most other companies rather than building a good product at this point and it has been reflected in their actions for the last few years now at this point.
Why most Frameworks handle this far better
You might think only Remix is the only one handling the task of doing one of RSC’s jobs better, but it isn’t the only one doing that, every framework and their “meta-framework” offer a very good alternative solution to the problem. In my eyes Qwik does an excellent job because it had the end goal with server-rendering being a first party thing in mind. The @builder.io/qwik-city
package exports 2 things which helps with this:
routeLoader$
routeAction$
On first glance, these seem like a match made in heaven, but that’s because they really are and I cannot stress THAT enough. For example, this would create the same effect as a RSC would do but far cleaner:
Data can be written to the server using something like:
export const useAddUser = routeAction$(async (data, requestEvent) => {
const userID = await db.users.add({
firstName: data.firstName,
lastName: data.lastName,
});
return {
success: true,
userID,
};
});
export default component$(() => {
const action = useAddUser();
return (
<>
<Form action={action}>
<input name="firstName" />
<input name="lastName" />
<button type="submit">Add user</button>
</Form>
{action.value?.success && (
<p>User {action.value.userID} added successfully</p>
)}
</>
);
});
And you can load Data similarly, like this:
export const useProductDetails = routeLoader$(async (requestEvent) => {
const res = await fetch(`https://.../products/${requestEvent.params.productId}`);
const product = await res.json();
return product as Product;
});
export default component$(() => {
const signal = useProductDetails();
return <p>Product name: {signal.value.product.name}</p>;
});
Qwik’s approach to Data Writing and Data Fetching not only is ergonomic, it has far less chances of leaking and causing security issues, and is paired with automatic type inference for better DX too, very seamless; very clean.
A Sad but Predictable Conclusion
Overall, Next.js’ unhealthy obsession with RSCs as a marketing strategy to “improve user-experience” is ironic because it ends up damaging it, but also makes it harder for developers for no reason when other approaches have proven to achieve the same end result with an arguably far better yield. Bottom line is, Next.js’ implementation of RSCs make it harder to justify them in the long run, I am very much looking forward to the approach by Remix in the upcoming year, as always, I’m confident in Remix team when it comes to delivering. In the end, shoutout to all major frameworks which do it nice and right; Remix, Qwik, and even SvelteKit. However, again this is not to say RSCs are an entirely bad idea, they just have their own set of use cases, such as:
- Better Composition, and DX such as Type Inference
- No More Painful work with React Contexts.
In all honesty, I’d love RSCs, if they were implemented better, similar to Qwik for example. Ultimately, RSCs being good or not will come down to framework-specific implementation, and for that, I’ll count on Remix, and with that; I’ll sign myself out for now.