Developing Without a Build: Introduction
This article is part of a series on developing without a build:
- Introduction (this article)
- es-dev-server
- Testing (coming soon!)
In this article, we explore why and if we should do development without a build step, and we give an overview of the current and future browser APIs that make this possible. In the followup articles, we look into how es-dev-server can help us with that and how to handle testing.
Modern web development
In the early days of web development, all we needed was a simple file editor and a web server. It was easy for newcomers to understand the process and get started making their own web pages. Web development has changed a lot since then: the complexity of the tools we use for development has grown just as much as the complexity of the things that we're building on the web.
Imagine what it's like if you're coming in completely new to web development:
- You first need to learn a lot of different tools and understand how each of them is changing your code before it can actually run in the browser.
- Your IDE and linter likely do not understand the syntax of this framework that was recommended to you by a friend, so you need to find the right mix of plugins that makes it work.
- Source maps need to be configured properly for all of the tools in the chain if you want to have any chance of debugging your code in the browser. Getting them to work with your tests is a whole other story.
- You decided to keep things simple and not use typescript. You're following the tutorials and but can't get this decorators thing to work and the error messages are not helping. Turns out you didn't configure your babel plugins in the correct order...
It may sound exaggerated, and I know that there are very good starter projects and tutorials out there, but this experience is common to a lot of developers. You may have jumped through similar hoops yourself.
I think that's really a shame. One of the key selling points of the web is that it's an easy and open format. It should be easy to just get started right away without a lot of configuration and ceremony.
I'm not criticizing the build tools themselves, they all have a role and a purpose. And for a long time, using a build was the only real way to actually create complex applications on the web. Web standards and browser implementations were just not there to support modern web development. Build tools have really helped push web development forward.
But browsers have improved a lot in the past years, and there are many exciting things to come in the near future. I think that now is a good time to consider if we can do away with a big part of the tooling complexity, at least during development. Maybe not yet for all types of projects, but let's see how far we can go.
Loading modules in the browser
This is not a step by step tutorial, but you can follow along with any of the examples by using any web server. For example http-server from npm. Run it with -c-1 to disable time-based caching.
npx http-server -o -c-1
Loading modules
Modules can be loaded in the browser using regular script tags with a type="module" attribute. We can just write our module code directly inline:
<!DOCTYPE html>
<html>
<head></head>
<body>
<script type="module">
console.log('hello world!');
</script>
</body>
</html>
From here we can use static imports to load other modules:
<script type="module">
import './app.js';
console.log('hello world!');
</script>
Note that we need to use an explicit file extension, as the browser doesn't know which file to request otherwise.
The same thing works if we use the src attribute:
<script type="module" src="./app.js"></script>
Loading dependencies
We don't write our code in just one file. After importing the initial module, we can import other modules. For example, let's create two new files:
src/app.js:
import { message } from './message.js';
console.log(`The message is: ${message}`);
src/message.js:
export const message = 'hello world';
Place both files in a src directory and import app.js from your index.html:
<!DOCTYPE html>
<html>
<head></head>
<body>
<script type="module" src="./src/app.js"></script>
</body>
</html>
If you run this and check the network panel, you will see both modules being loaded. Because imports are resolved relatively, app.js can refer to message.js using a relative path:
This seems trivial, but it is extremely useful and something we did not have before with classic scripts. We no longer need to coordinate dependencies somewhere central or maintain a base URL. Modules can declare their own dependencies, and we can import any module without knowing what their dependencies are. The browser takes care of requesting the correct files.
Dynamic imports
When building any serious web application we are usually going to need to do some form of lazy loading for best performance. Static imports like we saw before cannot be used conditionally, they always need to exist at the top level.
For example, we cannot write:
if (someCondition) {
import './bar.js';
}
This is what dynamic imports are for. Dynamic imports can import a module at any time. It returns a Promise that resolves with the imported module.
For example let's update the app.js example we created above:
window.addEventListener('click', async () => {
const module = await import('./message.js');
console.log(`The message is: ${module.message}`);
});
Now we are not importing the message module right away, but delaying it until the user has clicked anywhere on the page. We can await the promise returned from the import and interact with the module that was returned. Any exported members are available on the module object.
Lazy evaluation
This is where developing without a bundler has a significant benefit. If you bundle your application before serving it to the browser, the bundler needs to evaluate all your dynamic imports to do code splitting and output separate chunks. For large applications with a lot of dynamic imports, this can add significant overhead as the entire application is built and bundled before you can see anything in the browser.
When serving unbundled modules, the entire process is lazy. The browser only does the necessary work to load the modules that were actually requested.
Dynamic imports are supported by the latest versions of Chrome, Safari, and Firefox. It's not supported in the current version of Edge but will be supported by the new Chromium-based Edge.
Read more about dynamic imports at MDN
Non-relative requests
Not all browser APIs resolve requests relative to the module's location. For example when using fetch or when rendering images on the page.
To handle these cases we can use import.meta.url to get information about the current module's location.
import.meta is a special object which contains metadata about the currently executing module. url is the first property that's exposed here, and works a lot like __dirname in NodeJS.
import.meta.url points to the url the module was imported with:
console.log(import.meta.url); // logs http://localhost:8080/path/to/my/file.js
We can use the URL API for easy URL building. For example to request a JSON file:
const lang = 'en-US';
// becomes http://localhost:8080/path/to/my/translations/en-US.json
const translationsPath = new URL(`./translations/${lang}.json`, import.meta.url);
const response = await fetch(translationsPath);
Read more about import.meta at MDN
Loading other packages
When building an application you will quickly run into having to include other packages from npm. This also works just fine in the browser. For example, let's install and use lodash:
npm i -P lodash-es
import kebabCase from '../node_modules/lodash-es/kebabCase.js';
console.log(kebabCase('camelCase'));
Lodash is a very modular library and the kebabCase function depends on a lot of other modules. These dependencies are taken care of automatically, the browser resolves and imports them for you:
Writing explicit paths to your node modules folder is a bit unusual. While it is valid and it can work, most people are used to writing what's called a bare import specifier:
import { kebabCase } from 'lodash-es';
import kebabCase from 'lodash-es/kebabCase.js';
This way you don't specifically say where a package is located, only what it's called. This is used a lot by NodeJS, whose resolver will walk the file system looking for node_modules folders and packages by that name. It reads the package.json to know which file to use.
The browser cannot afford to send a bunch of requests until it stops getting 404s, that would be way too expensive. Out of the box, the browser will just throw an error when it sees a bare import. There is a new browser API called import maps which lets you instruct the browser how to resolve these imports:
<script type="importmap">
{
"imports": {
"lodash-es": "./node_modules/lodash-es/lodash.js",
"lodash-es/": "./node_modules/lodash-es/"
}
}
</script>
It's currently implemented in chrome behind a flag, and it's easy to shim on other browsers with es-module-shims. Until we get broad browser support, that can be an interesting option during development.
It's still quite early for import maps, and for most people, they may still be a bit too bleeding edge. If you are interested in this workflow I recommend reading this article
Until import maps are properly supported, the recommended approach is to use a web server which rewrites the bare imports into explicit paths on the fly before serving modules to the browser. There are some servers available that do this. I recommend es-dev-server which we will explore in the next article.
Caching
Because we aren't bundling all of our code into just a few files, we don't have to set up any elaborate caching strategies. Your web server can use the file system's last modified timestamp to return a 304 if the file hasn't changed.
You can test this in your browser by turning off Disable cache and refreshing:
Non-js modules
So far we've only looked into javascript modules, and the story is looking pretty complete. It looks like we have most of the things we need to write javascript at scale. But on the web we're not just writing javascript, we need to deal with other languages as well.
The good news is that there are concrete proposals for HTML, CSS and JSON modules, and all major browser vendors seem to be supportive of them:
The bad news is that they're not available just yet, and it's not clear when they will be. We have to look for some solutions in the meantime.
JSON
In Node JS it's possible to import JSON files from javascript. These become available as javascript objects. In web projects, this is used frequently as well. There are many build tool plugins to make this possible.
Until browsers support JSON modules, we can either just use a javascript module which exports an object or we can use fetch to retrieve the JSON files. See the import.meta.url section for an example which uses fetch.
HTML
Over time web frameworks have solved HTML templating in different ways, for example by placing HTML inside javascript strings. JSX is a very popular format for embedding dynamic HTML inside javascript, but it is not going to run natively in the browser without some kind of transformation.
If you really want to author HTML in HTML files, until we get HTML modules, you could use fetch to download your HTML templates before using it with whatever rendering system you are using. I don't recommend this because it's hard to optimize for production. You want something that can be statically analyzed and optimized by a bundler so that you don't spawn a lot of requests in production.
Luckily there is a great option available. With es2015/es6 we can use tagged template string literals to embed HTML inside JS, and use it for doing efficient DOM updates. Because HTML templating often comes with a lot of dynamism, it is actually a great benefit that we can use javascript to express this instead of learning a whole new meta syntax. It runs natively in the browser, has a great developer experience and integrates with your module graph so it can be optimized for production.
There are some really good production ready and feature complete libraries that can be used for this:
- htm, JSX using template literals. Works with libraries that use JSX, such as react
- lit-html, a HTML templating library
- lit-element, integrates lit-html with web components
- haunted, a functional web components library with react-like hooks
- hybrids, another functional web component library
- hyperHTML, a HTML templating library
For syntax highlighting you might need to configure your IDE or install a plugin.
CSS
For HTML and JSON there are sufficient alternatives. Unfortunately, with CSS it is more complicated. By itself, CSS is not modular as it affects the entire page. A common complaint is that this is what makes CSS so difficult to scale.
There a lot of different ways to write CSS, it's beyond the scope of this article to look into all of them. Regular stylesheets will work just fine if you load them in your index.html. If you are using some kind of CSS preprocessor you can run it before running your web server and just load the CSS output.
Many CSS in JS solutions should also work if the library publishes an es module format that you can import.
Shadow dom
For truly modular CSS I recommend looking into Shadow dom, it fixes many of the scoping and encapsulation problems of CSS. I've used it with success in many different types of projects but it's good to mention that it's not yet a complete story. There are still missing features that are being worked out in the standard so it may not yet be the right solution in all scenarios.
Good to mention here is the lit-element library, which offers a great developer experience when authoring modular CSS without a build step. lit-element does most of the heavy lifting for you. You author CSS using tagged template literals, which is just syntax sugar for creating a Constructable Stylesheet. This way you can write and share CSS between your components.
This system will also integrate well with CSS modules when they are shipped. We could emulate CSS modules by using fetch, but as we saw with HTML it's hard to optimize this for production use. I'm not a fan of CSS in JS, but lit-element's solution is different and very elegant. You're writing CSS in a JS file, but it's still valid CSS syntax. If you like to keep things separated, you can just create a my-styles.css.js file and use a default export of just a stylesheet.
Library support
Luckily the amount of libraries shipping es module format is growing steadily. But there are still popular libraries which only ship UMD or CommonJS. These don't work without some kind of code transformation. The best thing we can do is open issues on these projects to give them an indication of how many people are interested in supporting the native module syntax.
I think this is a problem that will disappear relatively quickly, especially after Node JS finishes their es modules implementation. Many projects already use es modules as their authoring format, and I don't think anyone really likes having to ship multiple imperfect module formats.
Final thoughts
The goal of this article is to explore workflows where we don't need to do any building for development, and I think we've shown that there are real possibilities. For a lot of use cases, I think we can do away with most of the tooling for development. In other cases, I think they can still be useful. But I think our starting point should be reversed. Instead of trying to make our production builds work during development, we should be writing standard code that runs in the browser as is and only perform light transformations if we think it's necessary.
It's important to reiterate that I don't think that build tools are evil, and I am not saying that this is the right approach for every project. That is a choice that each team should make for themselves based on their own requirements.
es-dev-server
You can do almost everything described in this article with any regular web server. That being said, there are still web server features which can really help with the development experience. Especially if we want to run our applications on older browsers, we might need some help.
At open-wc we created es-dev-server, a composable web server that focuses on developer productivity when developing without a build step.
Check out our next article to see how we can set it up!
Getting started
To get started with developing without any build tools, you can use the open-wc project scaffolding to set up the basics:
npm init @open-wc
It sets up the project with lit-element, a web component library. You can swap this for any library of your choosing, the setup is not specific to web components.
Top comments (4)
Great article!
I know that you mention it’s out of scope to go through all CSS solutions, but is there any one that you’d recommend/endorse for someone just getting started? I know I’ve personally had success with github.com/lukejacksonn/csz, but curious if there are others that aren’t webpack-dependent.
Also, shameless plug for @pika/web to install npm dependencies that don’t need bare module imports or a special dynamic file server: github.com/pikapkg/web
Thanks!
I don't have that much experience with build-less css-in-js solutions yet. That looks like a pretty interesting library though. I think we can collect some different libraries, and dive into an investigation in a followup article :)
Pika is really interest too!
Should we be concerned about the rabbit hole of HTTP requests that could have it's own negative impacts... if there are any? I'm still not leveraging the benefits of HTTP/2's request and response multiplexing; so I'm wondering if bundling and all it's "tree shaking" refinements for a single file request may be more performant? 🤷♂️
I see es-dev-server has the word "dev" in it. Is this only recommended for development, not production? Is there not yet any HTTP/2 server for production that is better than bundling?