Snowpack 3.0 Release Candidate is here! Read the announcement post →

Creating Your Own Plugin

Who is this page for?

  • Anyone interested in writing a custom plugin for Snowpack.
  • Anyone extending Snowpack to support new file types.
  • Anyone extending Snowpack with a custom bundler or production optimization.
  • Anyone adding Fast Refresh or automatic HMR for a framework.

Overview

A Snowpack Plugin is an object interface that lets you customize Snowpack’s behavior. Snowpack provides different hooks for your plugin to connect to. For example, you can add a plugin to handle Svelte files, optimize CSS, convert SVGs to React components, run TypeScript during development, and much more.

Snowpack’s plugin interface is inspired by Rollup. If you’ve ever written a Rollup plugin before, then hopefully these concepts and terms feel familiar.

Plugin Use-Cases

Snowpack uses an internal Build Pipeline to build files in your application for development and production. Every source file passes through the build pipeline, which means that Snowpack can build more than just JavaScript. Images, CSS, SVGs and more can all be built by Snowpack.

Build Plugins

Snowpack finds the first plugin that claims to resolve the given file. It then calls that plugin’s load() method to load the file into your application. This is where compiled languages (TypeScript, Sass, JSX, etc.) are loaded and compiled to something that can run on the web (JS, CSS, etc).

Transform Plugins

Once loaded, every file passes through the build pipeline again to run through matching transform() methods of all plugins that offer the method. Plugins can transform a file to modify its contents before finishing the file build.

Dev Tooling Plugins

Snowpack plugins support a run() method which lets you run any CLI tool and connect its output into Snowpack. You can use this to run your favorite dev tools (linters, TypeScript, etc.) with Snowpack and automatically report their output back through the Snowpack developer console. If the command fails, you can optionally fail your production build.

Bundler Plugins

Snowpack builds you a runnable, unbundled website by default, but you can optimize this final build with your favorite bundler (webpack, Rollup, Parcel, etc.) through the plugin optimize() method. When a bundler plugin is used, Snowpack will run the bundler on your build automatically to optimize it.

See our official @snowpack/plugin-webpack bundler plugin for an example of using the current interface.

Example: Getting Started

To create a Snowpack plugin, you can start with the following file template:

// my-snowpack-plugin.js
module.exports = function (snowpackConfig, pluginOptions) {
return {
name: 'my-snowpack-plugin',
// ...
};
};
// snowpack.config.json
{
"plugins": [
["./my-snowpack-plugin.js", { "optionA": "foo", "optionB": "bar" }]
]
}

A Snowpack plugin should be distributed as a function that can be called with plugin-specific options to return a plugin object. Snowpack will automatically call this function to load your plugin. That function accepts 2 parameters, in this order:

  1. the Snowpack configuration object (snowpackConfig)
  2. (optional) user-provided config options (pluginOptions)

Example: Transform a File

For our first example, we’ll look at transforming a file.

module.exports = function (snowpackConfig, pluginOptions) {
return {
name: 'my-commenter-plugin',
async transform({ id, contents, isDev, fileExt }) {
if (fileExt === '.js') {
return `/* I’m a comment! */ ${contents}`;
}
},
};
};

The object returned by this function is a Snowpack Plugin. A plugin consists of a name property and some hooks into the Snowpack lifecycle to customizes your build pipeline or dev environment. In the example above we have:

  • The name property: The name of your plugin. This is usually the same as your package name if published to npm.
  • The transform method: A function that allows you to transform & modify built files. In this case, we add a simple comment (/* I’m a comment */) to the beginning of every JS file in your build.

This covers the basics of single-file transformations. In our next example, we’ll show how to compile a source file and change the file extension in the process.

Example: Build From Source

When you build files from source, you also have the ability to transform the file type from source code to web code. In this example, we’ll use Babel to load several types of files as input and output JavaScript in the final build:

const babel = require('@babel/core');

module.exports = function (snowpackConfig, pluginOptions) {
return {
name: 'my-babel-plugin',
resolve: {
input: ['.js', '.jsx', '.ts', '.tsx', '.mjs'],
output: ['.js'],
},
async load({ filePath }) {
const result = await babel.transformFileAsync(filePath);
return result.code;
},
};
};

This is a simplified version of the official Snowpack Babel plugin, which builds all JavaScript, TypeScript, and JSX files in your application with the load() method.

The load() method is responsible for loading and build files from disk while the resolve property tells Snowpack which files the plugin can load and what to expect as output. In this case, the plugin claims responsibility for files matching any of the file extensions found in resolve.input, and outputs .js JavaScript (declared via resolve.output).

See it in action: Let’s say that we have a source file at src/components/App.jsx. Because the .jsx file extension matches an extension in our plugin’s resolve.input array, Snowpack lets this plugin claim responsibility for loading this file. load() executes, Babel builds the JSX input file from disk, and JavaScript is returned to the final build.

Example: Multi-File Building

For a more complicated example, we’ll take one input file (.svelte) and use it to generate 2 output files (.js and .css).

const fs = require('fs').promises;
const svelte = require('svelte/compiler');

module.exports = function (snowpackConfig, pluginOptions) {
return {
name: 'my-svelte-plugin',
resolve: {
input: ['.svelte'],
output: ['.js', '.css'],
},
async load({ filePath }) {
const fileContents = await fs.readFile(filePath, 'utf-8');
const { js, css } = svelte.compile(fileContents, { filename: filePath });
return {
'.js': js && js.code,
'.css': css && css.code,
};
},
};
};

This is a simplified version of the official Snowpack Svelte plugin. Don’t worry if you’re not familiar with Svelte, just know that building a Svelte file (.svelte) generates both JS & CSS for our final build.

In that case, the resolve property takes only a single input file type (['.svelte']) but two output file types (['.js', '.css']). This matches the result of Svelte’s build process and the returned entries of our load() method.

See it in action: Let’s say that we have a source file at src/components/App.svelte. Because the .svelte file extension matches an extension in our plugin’s resolve.input array, Snowpack lets this plugin claim responsibility for loading this file. load() executes, Svelte builds the file from disk, and both JavaScript & CSS are returned to the final build.

Notice that .svelte is missing from resolve.output and isn’t returned by load(). Only the files returned by the load() method are included in the final build. If you wanted your plugin to keep the original source file in your final build, you could add { '.svelte': contents } to the return object.

Example: Server-Side Rendering (SSR)

Plugins can produce server-optimized code for SSR via the load() plugin hook. The isSSR flag tells the plugin that Snowpack is requesting your file for the server, and that it will expect a response that will run on the server.

Some frameworks/languages (like React) run the same code on both the browser and the server. Others (like Svelte) will create different output for the server than the browser. In the example below, we use the isSSR flag to tell the Svelte compiler to generate server-optimized code when requested by Snowpack.

const svelte = require('svelte/compiler');
const fs = require('fs');

module.exports = function (snowpackConfig, pluginOptions) {
return {
name: 'basic-svelte-plugin',
resolve: {
input: ['.svelte'],
output: ['.js', '.css'],
},
async load({ filePath, isSSR }) {
const svelteOptions = {
/* ... */
};
const codeToCompile = fs.readFileSync(filePath, 'utf-8');
const result = svelte.compile(codeToCompile, {
...svelteOptions,
ssr: isSSR,
});
// ...
},
};
};

If you’re not sure if your plugin needs special SSR support, you are probably fine to skip this and ignore the isSSR flag in your plugin. Many languages won’t need this, and SSR is always an intentional opt-in by the user.

Example: Optimizing & Bundling

Snowpack supports pluggable bundlers and other build optimizations via the optimize() hook. This method runs after the build and gives plugins a chance to optimize the final build directory. Webpack, Rollup, and other build-only optimizations should use this hook.

module.exports = function(snowpackConfig, pluginOptions) {
return {
name: 'my-custom-webpack-plugin',
async optimize({ buildDirectory }) {
await webpack.run({...});
}
};
};

This is an (obviously) simplified version of the @snowpack/plugin-webpack plugin. When the build command has finished building your application, this plugin hook is called with the buildDirectory path as an argument. It’s up to the plugin to read build files from this directory and write any changes back to the directory. Changes should be made in place, so write files only at the end and be sure to clean up after yourself (if a file is no longer needed after optimizing/bundling, it is safe to remove).

Testing

To develop and test a Snowpack plugin, the strategy is the same as with other npm packages:

  • Create your new plugin project (either with npm init or yarn init) with, for example, npm name: my-snowpack-plugin and paste in it the above-mentioned code snipped
  • Run npm link in your plugin’s project folder to expose the plugin globally (in regard to your development machine).
  • Create a new, example Snowpack project in a different location for testing
  • In your example Snowpack project, run npm install && npm link my-snowpack-plugin (use the name from your plugin’s package.json).
    • Be aware that npm install will remove your linked plugin, so on any install, you will need to redo the npm link my-snowpack-plugin.
    • (The alternative would be to use npm install --save-dev <folder_to_your_plugin_project>, which would create the “symlink-like” entry in your example Snowpack project’s package.json)

In your example Snowpack project, add your plugin to the snowpack.config.json along with any plugin options you’d like to test:

{
"plugins": [
["my-snowpack-plugin", { "option-1": "testing", "another-option": false }]
"
}

Publishing a Plugin

To share a plugin with the world, you can publish it to npm. For example, take a look at snowpack-plugin-starter-template which can get you up-and-running quickly. You can either copy this outright or simply take what you need.

In general, make sure to mind the following checklist:

  • ✔️ Your package.json file has a main entry pointing to the final build
  • ✔️ Your code is compiled to run on Node >= 10
  • ✔️ Your package README contains a list of custom options, if your plugin is configurable

Tips / Gotchas

  • Remember: A source file will always be loaded by the first load() plugin to claim it, but the build result will be run through every transform function.
  • Snowpack will always keep the original file name (App) and only ever change the extension in the build.
  • Extensions in Snowpack always have a leading . character (e.g. .js, .ts). This is to match Node’s path.extname() behavior, as well as make sure we’re not matching extension substrings (e.g. if we matched js at the end of a file, we also don’t want to match .mjs files by accident; we want to be explicit there).
  • The resolve.input and resolve.output file extension arrays are vital to how Snowpack understands your build pipeline, and are always required for load() to run correctly.
  • If load() doesn’t return anything, the file isn’t loaded and the load() of the next suitable plugin is called.
  • If transform() doesn’t return anything, the file isn’t transformed.
  • If you want to build a plugin that runs some code only on initialization (such as @snowpack/plugin-dotenv), put your side-effect code inside the function that returns your plugin. But be sure to still return a plugin object. A simple { name } object will do.