webcodr

Snapshot Tests With Jest

Writing tests can sometimes be a tedious task. Mocks and assertions can be a pain in the ass. The latter is especially nasty when HTML is involved. Give me the second p element from the 30th div within an article in aside etc. – no thanks.

The creators of Jest (Facebook) have found a better way: Snapshot tests!

How does it work?

Take a look the following assertion:

it('should create a foo bar object', () => {
  const result = foo.bar()
  expect(result).toMatchSnapshot()
})

toMatchSnapshot() takes what ever you give to expect(), serializes it and saves it into a file. The next test run will compare the expected value to the stored snapshot and will fail if they don’t match. Jest shows a nicely formatted error message and diff view on failed tests.

This is really useful with generated HTML and/or testing UI behaviour. Just call the method and let it compare to the snapshot.

Updating snapshots

You added something to your code and the snapshot has to be updated? No problem:

jest --updateSnapshot

If you’re using the Jest watcher it’s even simpler. Just press u to update all snapshots or press i to update the snapshots interactively.

What about objects with generated values?

Here’s an example with an randomized id:

it('should fail every time', () => {
  const ship = {
    id: Math.floor(Math.random() * 20),
    name: 'USS Defiant'
  }

  expect(ship).toMatchSnapshot()
})

The id will change on every test run, so this test will fail every time. Well, shit? Nope. Jest got you covered:

it('should create a ship', () => {
  const ship = {
    id: Math.floor(Math.random() * 20),
    name: 'USS Defiant'
  }

  expect(ship).toMatchSnapshot({
    id: expect.any(Number)
  })
})

Jest will now only compare the type of the id and the test will pass.

For certain objects like a date, there is another possibility:

Date.now = jest.fn(() => 1528902424828)

A call of Date.now() will call the mock method and always return the same value.

Some advice

  1. Always commit your snapshots! If they are missing,CI systems will always create new snapshots and the tests will become useless.

  2. Snapshot tests are an awesome tool, but don’t be too lazy. They are no replacement for other assertion types, especially if you’re working test-driven. Rather use them alongside with your other tests.

  3. Write meaningful test names. Well, you heard that one before, didn’t you? Really, it helps a a lot when tests fail or you have to look inside a snapshot file. Jest takes a test name as an id inside a snapshot file. That’s why you have to update a snapshot after changing the name.

Introducing DeliveryGuy

I like the Fetch API. It’s supported by all modern browsers, easy to use and has some really good polyfills for older devices. But Fetch has one major flaw: it will only throw errors if there is a network problem.

That’s a really odd decision from my point of view. Nearly any HTTP library out there throws errors or rejects the promise in case of a HTTP error.

A Fetch response object has the property ok to determine if the server responded with an error, but that’s not very comfortable to use.

Since my team and I have decided to use Fetch in a Vue-based web app, I decided to create a little wrapper for much more convenience. Say hello to DeliveryGuy.

Usage

Well, surprise, it’s a Node module, so just use any package manager you like. My personal choice is yarn.

yarn install delivery-guy

Example

import { deliverJson } from 'delivery-guy'

const getItems = async () => {
  try {
    const items = await deliverJson('/api/items')
    console.log(items)
  } catch (e) {
    console.error(e.message)
    console.log('HTTP Status', e.response.status)
    console.log('Response Body'. e.responseBody)
  }
}

What’s going on here?

DeliveryGuy exports two main functions:

  • deliver() will return a response promise like fetch() does.
  • deliverJson() presumes your response body contains JSON. It’s basically a shortcut and returns the promise of Response.json().

Both will accept the same two parameters as fetch() does and pass them along.

If the server responds with a HTTP error, DeliveryGuy will throw an error.

Due to the inheritance limitations of built-in classes with ES5 I mentioned in my last post, it’s only possible to set custom properties of a custom error class.

DeliveryGuy provides additional two properties on an error object:

  • response has the original response object of a Fetch call.
  • responseBody contains the response body and will try to parse it as JSON. If JSON.parse fails, it will return the response body in its original state.

TL;DR

DeliveryGuy allows you comfortably call the Fetch API without a hassle on HTTP errors. Just use try/catch and you’re done.

Please let me know on GitHub if you have feedback, a feature request or found a bug. Thank you!

Why custom errors in JavaScript with Babel are broken

Have you ever tried to write a custom error class in JavaScript? Well, it does work to a certain extend. But if you want to add custom methods or call instanceof to determine the error type it will not work properly.

Here is a little example of a custom error class:

class MyError extends Error {
  constructor(foo = 'bar', ...params) {
    super(...params)
    
    if (Error.captureStackTrace) {
      Error.captureStackTrace(this, MyError)
    }
    
    this.foo = foo
  }
  
  getFoo() {
    return this.foo
  }
}

try {
  throw new MyError('myBar')
} catch(e) {
  console.log(e instanceof MyError) // -> false
  console.log(e.getFoo()) // -> Uncaught TypeError: e.getFoo is not a function
}

Works fine in any browser with ES6/ES2015 support, but if you transpile the example with Babel to ES5 and execute the code, you will get the results shown in the comments.

Why

Due to limitations of ES5 it’s not possible to inherit from built-in classes like Error, see the Babel docs.

Possible solution

The docs mention a plug-in called babel-plugin-transform-builtin-extend to resolve this issue, but if you have to support older browsers it may not help. In order to work the plug-in needs support for __proto__. Take a guess which browser does not support __proto__ … and of course, it’s the web developers best friend aka Internet Explorer. Thankfully it affects only version 10 and below.

Workaround

If it’s not feasable to use the plug-in, you can at least access properties set in the constructor. A call to e.foo in the example is possible, but e instanceof MyError will return false, since you will always get an instance of Error.

Conclusion

Nothing of this ideal. We have to wait until it’s possible to use ES6/ES2015 directly. Yes, we all could set our transpile targets to ES6/ES2015 today, but our clients usually won’t allow it. Some customer is always browsing the web with an ancient device/browser.

Awesome tests with Vue and Jest

Jest is a very neat JavaScript testing library from Facebook. It’s mostly syntax-compatible with Jasmine and needs zero or very less configuration. Code coverage reports are there out-of-the-box and with sandboxed tests and snapshot testing it has some unique features.

Set-up Jest

Vue CLI

You are using Vue CLI? Consider yourself lucky, the set-up of Jest could not be simpler:

yarn add --dev jest @vue/cli-plugin-unit-jest
vue invoke unit-jest

Vue CLI will do the rest and also create an example spec for the HelloWorld component.

DIY

Install all necessary dependencies:

yarn add --dev @vue/test-utils babel-jest jest jest-serializer-vue vue-jest

Create jest.config.js in your project root directory:

module.exports = {
  moduleFileExtensions: ['js', 'json', 'vue'],
  transform: {
    '^.+\\.vue$': 'vue-jest',
    '^.+\\.js?$': 'babel-jest'
  },
  moduleNameMapper: {
    '^@/(.*)$': '<rootDir>/src/$1'
  },
  snapshotSerializers: ['jest-serializer-vue'],
  testMatch: ['<rootDir>/src/tests/**/*.spec.js']
}

Please adjust the paths in moduleNameMapper and testMatch to your project.

You should also add a modern JavaScript preset to your .babelrc file:

{
  "presets": ["es2015"]
}

Optional step

Add the following line to your .gitignore file:

/coverage

The set-up is now complete. Let’s write a test file!

Writing tests

Here is a litte example Vue component:

<template>
  <ul class="demo">
    <li class="demo__item" v-for="value in values" :key="value">{{ value }}</li>
  </ul>
</template>

<script>
export default {
  name: 'demo',
  data () {
    return {
      values: []
    }
  },
  created() {
    this.fetchValues()
  },
  methods: {
    async fetchValues() {
      const response = await fetch('/api/demo/values')
      this.values = await response.json()
    }
  }
}
</script>

The component Demo will fetch some values after its creation and display them in an unordered list. This is example is quite simple, but testing is a little more complex due to the usage of fetch, async and await.

Of course, there are some tools to help us:

yarn add --dev fetch-mock flush-promises

Now, let’s write a test:

import { shallow } from '@vue/test-utils'
import fetchMock from 'fetch-mock'
import flushPromises from 'flush-promises'
import Demo from '@/components/Demo.vue'

const values = [
  'foo',
  'bar'
]

describe('Demo.vue', () => {
  beforeEach(() => {
    fetchMock.get('/api/demo/values', values)
  })

  it('renders component', async () => {
    const wrapper = shallow(Demo)
    await flushPromises()

    expect(wrapper.vm.values).toEqual(values)
  })

  afterEach(() => {
    fetchMock.restore()
  })
})

What’s going on?

  1. shallow from Vue Test Utils creates a wrapper of the rendered and mounted component, any child components will be stubs. If you need child components in your test, please use mount instead of shallow.

  2. fetchMock will create a mocked version of the Fetch API. In this case it will return the defined values for a GET request to /api/demo/values. If you send a request that’s not defined in fetchMock, it will throw an exception and break your tests.

  3. The test itself is defined as async to use await for flushPromises(). It will wait until the mocked request is finished and the values are stored in the component’s data.

  4. You can now access the data property values and compare the content to the response of the mocked HTTP request.

Conclusion

Setting up Jest for Vue is easy, even if you have to do it manually.

A little warning: setting up Jest for an existing app can be tedious. The current AngularJS app of our customer can’t be tested with Jest, at least for now. The AngularJS HTTP mock does not work and I haven’t figured out the problem yet.

But enough of Angular: the real deal comes with the testing itself. Async/await is a nice and simple way for testing asynchronous behaviour. I don’t think this could be easier and it’s a reliable method with the power of modern JavaScript. Try to imagine what the demo test would look like in ES5 …

Vue Loader Setup in Webpack

Since it’s no option to use Vue CLI in my current project, I had to manually add vue-loader to the Webpack config. Well, I was positively surprised how simple it was. Especially compared to a manual set-up of Angular 2 shortly after its launch in late 2016, while Angular CLI was an alpha version and buggy as hell.

Shall we begin?

Modules

Let’s add the necessary modules to the package:

yarn add --dev vue-loader vue-template-compiler
yarn add vue

If you want to use a template engine like Pug or prefer TypeScript over JavaScript, you can add the respective Webpack loader package, pug-loader for example. Webpack will also tell you in detail, if modules are missing.

Webpack

Just add the following rule to your Webpack config:

{
  test: /\.vue$/,
  use: [
    {
      loader: 'vue-loader'
    }
  ]
}

Webpack will now be able to use import statements with Vue single file components.

If you want to have a separate JavaScript file with your Vue application, you can add a new entry point:

{
  entry: {,
    'current-application': [
      path.resolve(__dirname, 'js/current-application.js')
    ],
    'vue-application': [
      path.resolve(__dirname, 'js/vue-application.js')
    ]
  }
}

Not fancy enough? How about vendor chunks?

{
  optimization: {
    splitChunks: {
      chunks: 'all',
      cacheGroups: {
        vendor: {
          chunks: 'initial',
          test: /node_modules\/(?!(vue|vue-resource)\/).*/,
          name: 'vendor',
          enforce: true
        },
        'vue-vendor': {
          chunks: 'initial',
          test: /node_modules\/(vue|vue-resource)\/.*/,
          name: 'vue-vendor',
          enforce: true
        }
      }
    }
  }
}

This will separate your vendor files from node_modules into vendor.js and vue-vendor.js. The property test contains a regex to determine which modules should go into the vendor chunks.

Of course, this comes not even close to what Vue CLI can do. I highly recommend to use Vue CLI when it’s feasible. It’s quite easy to configure while not being dumbed down, has an excellent documentation and is very well maintained. At least for our customer’s set-up it would be difficult to use Vue CLI at the moment, but we are eager to migrate to Vue CLI asap.