Ladle: The Story About Modules and Performance

Hello everyone, my name is Wojta Mikšu. I work at Uber as a web infrastructure engineer, and today I will tell you about a new open tool that supercharges developing and testing your React components.

This talk has 4 sections. The first is a short history lesson about Javascript features we've been sorely missing for a long time. Then we talk about Javascript modules, also known as ES6 modules or ESM. The third section introduces Vite, a new type of bundler, followed by an introduction of Ladle, a tool we built on top of Vite. And finally, we will wrap this up with some future predictions.

Do you remember what 1995 browsers looked like? I don't, but they were very different for sure. It's not surprising that JavaScript was missing some features when it was introduced. The one feature that's relevant to this talk is the way how JavaScript is loaded into browsers. You have to use the script tag. The code itself can be inlined or point to a file. This was fine 15, 10 years ago when JavaScript was used to add a bit of interactivity. But when we build modern frontend applications, it causes some serious issues. The files are loaded and executed synchronously and their order matters. Top level variables end up in the global scope, so it's easy for two unrelated libraries to cause naming collisions. Every time you create a new file, you have to load it through an additional script tag and bind it through the global scope. There's no easy way to eliminate unused code either.

JavaScript was missing a concept of modularization. This was especially a big problem for server-side JavaScript known as Node.js. Some better system for splitting and encapsulating code was badly needed. So, Node.js adopted CommonJS. This syntax should be very familiar to anyone that touched JavaScript in recent years. It addresses a lot of issues, but not all of them. It introduces a concept of code providers and consumers. It's also a whole philosophy and it was used to create the biggest package registry for code sharing called npm. It enables dynamic code loading. However, there are some major weaknesses that make it unusable for browsers. For example, there are file system references like the usage of dirname. But the biggest issue is that module resolution and loading needs to be done synchronously.

That would block rendering in browsers and lead to really bad experiences. CommonJS was developed independently from the ECMAStandards party, so it never became the core part of JavaScript. However, it was still a better system than JavaScript ever had, and developers wanted to use server code in their frontend applications. So the bundlers like Webpack were introduced. Webpack can analyze and resolve all CommonJS modules before creating a single bundled file that serves two browsers. This was a big win-win. Developers can use CommonJS to split code into modules and don't have to worry about browser support in it.

But we were still missing an official standardized way that would work in browsers and other runtimes without additional tooling. JavaScript modules were introduced. Today, they are supported by all modern browsers and runtimes. You can load them by setting the type attribute to module. And the best part is that browsers understand import-export syntax and load additional modules when needed. Modules work asynchronously and don't prevent pages being interactive. There's a promise API to import modules dynamically. The code can be easily tree shaken. Only the bits you use need to be loaded. It works everywhere across all browsers and environments. And it is an official standard now.

Don't get me wrong. There are still some issues when it comes to compatibility with common JS and existing ecosystem, so the adaption can be sometimes cumbersome. However, it's clear that this is the future of JavaScript.

So what is Vite? It's a new development environment built on the fact that JavaScript modules don't need to be bundled before being served to browsers. It was a major and very slow part of Webpack. Therefore, the dev server startup can be instant. Some bundling and compilation is still needed. A lot of existing libraries need to be converted from common JS. We also need to remove things like JSX and TypeScript. Vite uses ESbuild and Rollup plugins for that. Let's compare Webpack and Vite startup process side-by-side. This is the old Webpack approach.

The code base needs to be all parsed and all modules resolved before creating a bundle. Only then the server can be started and the bundle served. The bigger your application gets, the longer the startup time is.

On the other hand, Vite starts the server immediately and serves the entry module to the browser. Since the browser understands following imports, it loads additional modules only when needed. Vite only needs to do a light preprocessing, like stripping typescript types, before sending these files back to the browser. However, it doesn't matter how big your code base is, the server always starts instantly and incremental rebuilds also stay very, very fast.

We love these new performance benefits, so it made sense to put Vite to a test and use it as a foundation for an alternative solution to storybook. We built an open-sourced tool called Ladle. A tool for developing and testing React components through stories, built from ground up on top of Vite.

Let's do a demo. We are starting with an empty folder, so let's initialize our project. We also need to add some dependencies as react, react-dom and ladle. By default Ladle is looking for the folder source. And specifically looks for files ending with .stories.dsx. So let's create file welcome-stories.dsx.

Stories are also just React components, but they need to be exported. So let's export story name first, that exports a heading. Of course this could be like any React component, for example a component from your design system. And now we can start Ladle with command ladleserve. This immediately starts a v8 dev server and this is the Ladle's UI. On the left side you can find your component and on the right side there's a navigation listing all discovered stories. By default the file names and the export names are used to create this navigation. But this can be also changed through the code through parameters story name and export default title.

Let's add some additional stories to demonstrate other features. So this story has a button that's not very accessible. As you can see the call contrast is not great. You can write accessibility, you can run accessibility check with X by clicking on this icon. And now X tells us there is one violation and it gives us details so we can debug it and fix it. Another feature is useful when testing responsiveness and responsive design.

It allows you to set different viewports and test different media queries. Through this add-on, you can change the viewport and test different sizes. You can also set a default viewport size through the parameter .meta and width.

Another feature is useful when you want to display and test different variations of a single component. In this case, we are changing the value of size. This feature is called controls and right now we can change this one variable called size and we can pick from three different values. As you can see, even without touching the code, we can see different variations. And this is how you set it up in the code through the API called archetypes. You define what variables you want to make dynamic, you define the options and then this gets passed as a prop into the story itself when it can be used inside of your component. There are other control types you can use like booleans, strings, radios. So this is very flexible for things like design systems when you want to display different variations of a single component, but you don't want to create multiple stories.

Another feature is great for logging event handlers. So, for example, this button is using action handler and once we start clicking on it, you get this notification and it shows you what event was fired and its details. The action is a function as exploited from Nadl and can be just simple pass to your handlers. And there are some other features you can switch between light and dark theme. You can switch to full screen. You can switch between left to right and right to left. You can also preview the source code of the story. It highlights the specific story and lines of code. It also shows the location of the story. And finally, Ladle exposes endpoint meta.json. This gives you a structured information about what stories are in your instance. This can be extremely useful for further automation and testing. For example, we use this file to run a playwright script that opens each URL and takes a snapshot. Then we can use it for visual snapshot testing. As you can see, this metadata also exports information like the width. When you set the default viewport, not only does your story get displayed in this viewport, but it also gets exported with it, together in this file. This can be very useful for further testing automation.

What is ladle? Ladle is weed plugin, ReactUI and CLI in one single package. It's a single dependency, so you will never run into versioning issues between multiple packages.

It requires no initial configuration, but you can still customize things like fonts, add additional stylesheets or fully access weed's configuration. It fully supports component story format and works as a drop-in replacement of storybook.

It also supports different monorepo setups and has a great programmatic API, so it can be re-bundled as a part of your library. It was released earlier this year and so far it got great feedback from the community. We've been using it for many internal and public projects and saw sometimes even 10x improvements when it comes to starting dev servers and hot module replacement.

So where is the frontend tooling heading now? It's pretty clear that modules are not going anywhere and over time they will rule everything. This is great for our tooling. We can skip some heavy processing before sending our code to browsers. Browsers can do the module resolution for us. They're also much better when it comes to caching. There's also a trend to use other languages like Rust and Go when compiling JavaScript. That's another great source of speed improvements.