I’m excited to announce the release of @std/esm
(standard/esm), an opt-in, spec-compliant, ECMAScript (ES) module loader that enables a smooth transition between Node and ES module formats with near built-in performance! This fast, small, zero dependency package is all you need to enable ES modules in Node 4+ today!
A tale of two module formats
With ESM landing in browsers, attention is turning to Node’s future ESM support. Unlike browsers, which have an out-of-band parse goal signal and no prior module format, support for ESM in Node is a bit more…prickly. Node’s legacy module format, a CommonJS (CJS) variant, is a big reason for Node’s popularity, but CJS also complicates Node’s future ESM support. As a refresher, let’s look at an example of both module syntaxes.
CJS:
[code language=”javascript”]const a = require("./a")
module.exports = { a, b: 2 }[/code]
ESM:
[code language=”javascript”]import a from "./a"
export default { a, b: 2 }[/code]
Note: For more in-depth comparisons see Nicolás Bevacqua’s excellent post.
Because CJS is not compatible with ESM, a distinction must be made. After much discussion, Node has settled on using the “.mjs” (modular JavaScript) file extension to signal the “module” parse goal. Node has a history of processing resources by file extension. For example, if you require
a .json
file, Node will happily load and JSON.parse
the result.
ESM support is slated to land, unflagged, in Node v10 anytime between October 2018 and April 2020. This puts developers, esp. package authors, in a tough spot. They could choose to:
- Go all in, shipping only ESM, and alienate users of older Node versions
- Wait until Jan 1, 2020, the day after Node 8 support ends, to go all in
- Ship both transpiled CJS and ESM sources, inflating package size and shouldering the responsibility for ensuring 1:1 behavior
None of those choices seem super appealing. The ecosystem needs something that meets it where it is to span the CJS to ESM gap.
Bridge building
Enter the @std/esm
loader, a user-land package designed to bridge the module gap. Since Node now supports most ES2015 features, @std/esm
is free to focus solely on enabling ESM.
The loader stays out of your way and tries to be a good neighbor by:
- Not polluting stack traces
- Working with your existing tools like Babel and webpack.
- Playing well with other loaders like babel-register
(using .babelrc"modules":false
) - Only processing files of packages that explicitly opt-in with a
@std/esm
configuration object or having@std/esm
as a dependency, dev dependency, or peer dependency - Supporting versioning
(i.e. package “A” can depend on one version of@std/esm
and package “B” on another)
Unlike existing ESM solutions which require shipping transpiled CJS, @std/esm
performs minimal source transformations on demand, processing and caching files at runtime. Processing files at runtime has a number of advantages.
- Only process what is used, when it’s used
- The same code is executed in all Node versions
- Features are configurable by module consumers
(e.g. module “A” consumes module “C” with the default@std/esm
config while module “B” consumes module “C” withcjs
compat rules enabled) - More spec-compliance opportunities
(i.e.@std/esm
can enforce Node’s ESM rules for environment variables, error codes, path protocol and resolution, etc.)
Standard features
Defaults are important. The @std/esm
loader strives to be as spec-compliant as possible while following Node’s planned built-in behaviors. This means, by default, ESM requires the use of the .mjs
extension.
Out of the box, @std/esm
just works, no configuration necessary, and supports:
Unlockables
Developers have strong opinions on just about everything. To accommodate, @std/esm
allows unlocking extra features with the "@std/esm"
package.json field. Options include:
- Enabling unambiguous module support (i.e. files with at least an
import
,export
, or"use module"
pragma are treated as ESM) - Supporting named exports of CJS modules
- Top-level
await
in main modules - Loading gzipped modules
Performance
Before I continue, let me qualify the following section:
It’s still super early, mileage may vary, and results may be hand wavey!
Testing was done using Node 9 compiled from PR #14369, which enables built-in ESM support. I measured the time taken to load the 643 modules of lodash-es, converted to .mjs
, against a baseline run loading nothing. Keep in mind the @std/esm
cache is good for the lifetime of the unmodified file. Ideally, that means you’ll only have a single non-cached load in production.
- Loading CJS equivs was ~0.28 milliseconds per module
- Loading built-in ESM was ~0.51 milliseconds per module
- First
@std/esm
no cache run was ~1.6 milliseconds per module - Secondary
@std/esm
cached runs were ~0.54 milliseconds per module
Initial results look very promising, with cached @std/esm
loads achieving near built-in performance! I’m sure, with your help, parse and runtime performance will continue to improve.
Getting started
- Run
npm i --save @std/esm
in your app or package directory. - Call
require("@std/esm")
before importing ES modules.
index.js:
require("@std/esm") module.exports = require("./main.mjs").default
For package authors with sub modules:
// Have "foo" require only "@std/esm".
require("foo") // Sub modules work! const bar = require("foo/bar").default
Enable ESM in the Node CLI by loading @std/esm
with the -r
option:
node -r @std/esm file.mjs
Enable ESM in the Node REPL by loading @std/esm
upon entering:
$ node > require("@std/esm") @std/esm enabled > import p from "path" undefined > p.join("hello", "world") 'hello/world'
Meteor’s might
The @std/esm
loader wouldn’t exist without Ben Newman, creator of the Reify compiler from which @std/esm
is forked. He’s proven the loader implementation in production at Meteor, since May 2016, in tens of thousands of Meteor apps!
All green thumbs
Even though @std/esm
has just been released, it’s already had a positive impact on several related projects:
- Fixing Acorn’s strict mode pragma detection and aligning parser APIs
- Improving dynamic import support of Babel and Acorn plugin
(the dynamic import Acorn plugin is used by webpack for code splitting) - Improving property iteration order in ChakraCore
- Improving the parse, load time, and spec compliance of Reify
- Inspiring a fast top-level parser proof of concept
- Spurred championing of
export * as ns from "mod"
andexport default from "mod"
proposals
What’s next
Like many developers, I want ES modules yesterday. I plan to use @std/esm
in Lodash v5 to not only transition to ESM but also leverage features like gzip module support to greatly reduce its package size.
The @std/esm
loader is available on GitHub. It’s my hope that others are as excited and as energized as I am. ES modules are here! This is just the start. What’s next is up to you. I look forward to seeing where you take it.
Final Thought
While this is not a Microsoft release, we’re proud to have a growing number of core contributors to fundamental JavaScript frameworks, libraries, and utilities at Microsoft. Contributors like Maggie Pint of Moment.js, Matthew Podwysocki of ReactiveX, Nolan Lawson of PouchDB, Patrick Kettner of Modernizr, Rob Eisenberg of Aurelia, Sean Larkin of webpack, and Tom Dale of Ember, to name a few, who in addition to their roles at Microsoft, are helping shape the future of JavaScript and the web at large through standards engagement and ecosystem outreach. I’m happy to share this news on the Microsoft Edge blog to share our enthusiasm with the community!
― John-David Dalton, Program Manager, Microsoft Edge