mixedCase

Search articles

Our front end stack one year later

One year ago I wrote my third article in a series where I researched front end tools, and with all that I learned I began to improve our front end stack at Sling. It’s the big Angular app that I wrote about, on which I’ve worked for 2.5 years now. Back then we used Gulp, Bower and Less. No modules, no ES6, barely any unit tests. We still use Angular 1.x, but a lot has changed in the last 12 months!

First of all, we’re now using CommonJS modules and ES6 syntax throughout the app: fat arrows, destructuring, object literal shorthand, template strings, default function parameters, etc. We’re using Babel 6 and webpack to compile all this to ES5. More specifically, we use ng-annotate-loader to automatically add injection annotation to our Angular code, and html-loader so that we can require our HTML templates right into the UI Router states.

We’re no longer using Bower, instead 100% of our dependancies come from NPM. And we also exclusively use NPM scripts for running our local dev server, compiling the builds and running the tests and linter. So no more Gulp either.

We still use Less, and have now combined that with PostCSS for the autoprefixer plugin.

Our code is linted using ESLint, using a slightly modified AirBnb code style. Mainly we enabled more rules. And we’ve added a ton of unit tests! We now run our tests in pure Node 6.1.0 using Mocha, Chai and JSDom: no browser is needed, and the code doesn’t need to go through Babel or webpack. It’s super fast. In my second font end tools article I voiced some concerns about locking ourself in with webpack and making it harder to test our code when we’re using webpack-only features, but we solved that by simply using a __TEST__ global and checking for that before requiring any non-JS files.

We’ve added Istanbul to get code coverage as well. We’re currently almost at 50% so definitely still have a way to go, but considering that a year ago we had almost nothing, I am very proud of the team - myself included :)

The latest addition to our stack is Semaphore: a hosted continuous integration / delivery server. Every commit is automatically tested, we get test results right in our pull requests, and every successfully tested commit on the develop branch goes to staging, and the same for master to production. We’re now testing and deploying way more often, catching more bugs in an earlier state, and we save a lot of time and prevent errors with manual deploys. I can’t recommend this enough to everyone, and it was super simple to setup too.

Have feedback? Let me know on Twitter.