So this is a post I've been meaning to write for a while.

I've spoken briefly about this topic with friends and colleagues that I'm in the belief that if you identify purely as a frontend developer, I'm going to assume 99% of the time that you do not write tests.

If I'm right, this is not to say you are a bad developer. Far from it.

For a lot of developers out there, refreshing the browser and looking for console errors is about as much testing as they need (and as about as much as they can stomach!).

But this is a haphazard approach which relies on excessive cognitive load (to remember all the bits that need testing), experience with browser quirks (to have confidence across different OS/browser environments) and a human (who is by default error-prone and most probably stressed out with that looming deadline fast approaching).

Me: What if you have over 100 UI components?
You: I'll just fire up my browser and run through all the pages!
Me: ...

It also wouldn't give much confidence going into a release - where it's likely there's a christmas-naughty-list length of changes - you have no way to verify what was working before, does indeed work as expected now.

Breaking the cycle

I'm sure we've all experienced embarassing moments where we've had seemingly innocent refactoring break core functionality in some way later on.

This post isn't to re-live those awkward moments but rather, to reduce occurrances of them going forward - whether it's in your current or future projects.

I speak from experience when I say frontend developers get into a habit of not writing tests. I've heard all the excuses - I've made them myself on many occasions. Writing comprehensive tests is always a daunting task and the more you think about it, the less likely you'll get started.

I broke out of that cycle by asking myself, what is the easiest test I could write right now?

Smoke tests

Ok, so you're probably asking: what does "easy" mean exactly?
Well for a start, I wanted to do the basic minimum with the idea of revisiting the tests once I had more time/motivation. This had to cover 3 basic criteria:

  1. I should be write the least amount of code needed.
  2. I can write it in the shortest amount of time possible.
  3. I should actually be testing something meaningful.

After not much research, I found this pretty much describes smoke testing.

The idea is simple. Given that the worst thing that can happen in your web app is to have something throw an exception at startup, we would write unit tests that check each and every javascript component and widget we use, is able to initialize.

Here's an example for a single component named "MyComponent"

// mycomponent.test.js

import MyComponent from 'src/MyComponent';

describe('myComponent', () => {
    it('initializes', () => {
        new MyComponent();

And that's pretty much all there is to it.

Now do this for 100 components. The results are amazing.

Not only would you have written 100 tests in about 5 - 10 minutes, but you will have assured yourself that if something was to fall over, you'll know about it before anyone else does and most importantly, before it goes out to your customers.

Doubling up

Here's a tip. If your components have "render" methods, you could (and probably should) double up on your tests by ensuring these components are also able to render as well.

Extending our MyComponent example from above

// mycomponent.test.js

import MyComponent from 'src/MyComponent';

describe('myComponent', () => {
    it('initializes', () => {
        new MyComponent();

    it('renders', () => {
        var component = new MyComponent();

Boom! An easy 200 tests in about the same time it takes setup your webpack config.

Only the beginning

Once you start seeing the benefits and get into a habit of writing smoke tests (and unit tests in general), it becomes a starting point for more interesting tests. I find that once you're confident about the things that do work, you can focus more on writing tests for the things that don't.

If you've been holding off writing tests, give this a go! Hopefully I've been able to convince some of you for the better but I would be interested to know if this has worked for anyone else.