Declarative data requirements

In our front-ends we'll refer to server-side rendering to increase performance, please crawlers, and making the app work without JavaScript enabled on the client. To achieve this we go through quite a bit of trouble. frameworks like Next, Gatsby,... aim to alleviate this concern from us and don't get me wrong, they do a great job, but what makes all of this so hard that we need these?

Let's look at how we approach this on the client and then move on to the issues we have on the server.

Client-side

When we are on the client we'll often declare our data-requirements in componentDidMount or a useEffect (pick your poison), we'll await the Promise and set it in our state, that's the simplest case we have to visualize data located in an API inside of our API. Notice how all the rendering just continues, a placeholder is where that data is supposed to be and the world just seemingly stands still until the API response has been delivered. When the data arrives we call setState and our Virtual-DOM library will know that it's time to show some new things on screen, these new things could in turn have their own data-requirements and we could be creating a waterfall of spinners.

Not only magicians create suspense...

The Suspense was killing me too waiting for this part. For those who don't know the concept of Suspense it is essentially a way to abstract the loading states that we enter in the above. We call our Promise and store it in state, rather than checking whether we are still fetching inside of our render function we use the throw keyword from JS and rather than throwing an error we throw that Promise up, when a Suspense boundary catches this it will render a loading state.

let globData
const getData = () => {
  if (globData) return globData
  return new Promise((res) => {
    setTimeout(() => {
      res((globData = [{ id: 1, name: 'Jovi De Croock' }]))
    }, 500)
  })
}

const App = () => {
  const [data, setData] = useState(getData)

  if (data.then) {
    throw data
  }

  return (
    <div>
      {data.map((person) => (
        <span key={person.id}>{person.name}</span>
      ))}
    </div>
  )
}

render(
  <Suspense fallback={<p>Loading...</p>}>
    <App />
  </Suspense>
)

In the above we can see that we fetch data and when it resolves the fallback UI dissapears and we render the intended display of todo's. This abstraction however does more, when we have a waterfall, we can stay in the fallback state without any problem as the subsequent loading state will throw up and it can be caught in a new boundary or the same one making the loading state consistent until all data loads in.

Server-side

When we approach the concept of server-side rendering we'll often refer to stringifying the HTML and sending it to the client, rendering to string as it stands is a synchronous process where we don't invoke lifecycles and just call render until we have passed through the whole tree. It's important to take this single pass in consideration when we consider the 'world standing still' while we resolve data on the client. A server-side render does not support doing multiple passes through the data, which means we'll be rendering the loading spinners rather than our data as while the world stands still we'll take the string-snapshot and sent it to the client.

import { useState, useEffect } from 'preact/hooks'
import renderToString from 'preact-render-to-string'

const getData = () =>
  new Promise((res) => {
    setTimeout(() => {
      res([{ id: 1, name: 'Jovi De Croock' }])
    }, 500)
  })

class App extends Component {
  constructor(props) {
    super(props)
    this.state = {
      data: [],
      isLoading: true,
    }
  }

  componentDidMount() {
    getData().then((data) => {
      this.setState({ isLoading: false, data })
    })
  }

  render(props, { isLoading, data }) {
    if (isLoading) {
      return <p>loading...</p>
    }

    return (
      <div>
        {data.map((person) => (
          <span key={person.id}>{person.name}</span>
        ))}
      </div>
    )
  }
}

const stringifiedHTML = renderToString(<App />)
The above stringified HTML

We don't invoke lifecycles, effects, ... during a render-to-string as we won't react to state-updates anyway, also how would the render to string process even know when all data is resolved? Popular frameworks have introduced concepts like getServerSideProps which allows us to run a function when the page renders and resolve the data-requirements, this gives us observability into the data actually finishing the loading state and is actually great as we execute 1 Promise and then 1 render-to-string, no waterfalls, ... This concept however gives us 1 more thing to think about, how great would it be if we could just leverage the virtual-dom to define our data requirements and it would automatically work? Well it's possible, a few years ago Formidable came with a library named react-ssr-prepass which essentially leverages the concept of Suspense throwing Promises up the tree to resolve data.

This means that we would invoke prepass and then go on to render-to-string, as long as we don't reach the end of the tree (due to Promises being thrown) prepass would know it has not resolved all the data requirements. This brings one important pitfall to be aware of, the waterfalls that could be acceptable on the client will be causing a huge delay in the HTML skeleton arriving. Another thing to consider is that we'll need a client/... that supports caching the responses we get during the prepass, because when we follow the prepass with our renderToString we'll need to be able to synchronously retrieve the data. We will also need to be able to send the data back to the client so the data-cache can be hydrated and the dom-hydration can complete elegantly.

At the time of writing React has given us React 18 which brings support for asynchronous renderToString, do note that this async renderToString won't resolve data either, it will render the Suspense fallback instead.

Concluding

I do hope this gives some insight into the trouble front-end frameworks have gone through when we reason about server-side renders and data, this post deliberately leaves out client-side data-hydration as there is also a lot to write about that as well.