My “Whoa, I didn’t know that!” moments with Jest

Briwa
The Startup
Published in
6 min readMay 26, 2019
There is more to Jest than it meets the eye. (Image from https://jestjs.io)

Jest has always been my go-to unit testing tool. It is so robust I’m starting to think that I’ve always been underutilizing it. Although the tests are passing, over time I’ve refactored my test here and there because I didn’t know Jest can do that. It’s always a different code every time I checked back Jest docs.

So, I’m going to share some of my favorite tricks with Jest that some of you might already know because you didn’t skip reading the docs like me (shame on me), but I hope this helps those who did!

FWIW, I’m using Jest v24.8.0 as the reference, so be aware if certain things don’t work on the Jest version you’re currently using. Also, the examples do not represent the actual test code, it is just merely a demonstration.

#1. .toBe vs .toEqual

All these assertions looked good to me at first:

expect('foo').toEqual('foo')
expect(1).toEqual(1)
expect(['foo']).toEqual(['foo'])

Coming from using chai to do equality assertions ( to.equal), it's just natural. In fact, Jest wouldn't complain and these assertions are passing as usual.

However, Jest has .toBe and .toEqual. The former is used to assert equality using , while the latter is to assert deep equality on objects and arrays. Now, .toEqual has a fallback to use if it turns out that it doesn't need deep equality, such as asserting equalities on primitive values, which explains why the earlier example was passing just fine.

expect('foo').toBe('foo')
expect(1).toBe(1)
expect(['foo']).toEqual(['foo'])

So, you can skip all the if-elses in .toEqual by using .toBe if you already know what kind of values you're testing.

A common mistake is that you would be using .toBe to assert equality on non-primitive values.

expect(['foo']).toBe(['foo'])

If you look at the source code, when .toBe fails, it would try to see if you are indeed making that mistake by calling a function that is used by .toEqual. This could be a bottleneck when optimizing your test.

If you are sure that you are asserting primitive values, your code can be refactored as such, for optimization purpose:

expect(Object.is('foo', 'foo')).toBe(true)

Check out more details in the docs.

#2. More befitting matchers

Technically, you can use .toBe to assert any kind of values. With Jest, you can specifically use certain matchers that would make your test more readable (and in some cases, shorter).

For array length:

// 🤔
expect([1,2,3].length).toBe(3)

// 😎
expect([1,2,3]).toHaveLength(3)

Defined values:

const canBeUndefined = foo()

// 🤔
expect(typeof canBeUndefined !== 'undefined').toBe(true)

// 🤔
expect(typeof canBeUndefined).not.toBe('undefined')

// 🤔
expect(canBeUndefined).not.toBe(undefined)

// 😎
expect(canBeUndefined).toBeDefined()

And validating the instance:

class Foo {
constructor(param) {
this.param = param
}
}

// 🤔
expect(new Foo('bar') instanceof Foo).toBe(true)

// 😎
expect(new Foo('bar')).toBeInstanceOf(Foo)

These are just a few I picked from a long list of Jest matchers in the docs, you can check out the rest.

#3. Snapshot testing on a non-UI elements

You might have heard about snapshot testing in Jest, where it helps you monitor changes on your UI elements. But snapshot testing is not limited to that.

Consider this example:

const allEmployees = getEmployees()
const happyEmployees = giveIncrementByPosition(allEmployees)

expect(happyEmployees[0].nextMonthPaycheck).toBe(1000)
expect(happyEmployees[1].nextMonthPaycheck).toBe(5000)
expect(happyEmployees[2].nextMonthPaycheck).toBe(4000)
// ...etc

It would be tedious if you have to assert more and more employees. Also, if it turns out that there are more assertions to be done for each employee, multiple the number of the new assertions with the employee count and you get the idea.

With snapshot testing, all of these can be done simply as such:

const allEmployees = getEmployees()
const happyEmployees = giveIncrementByPosition(allEmployees)

expect(happyEmployees).toMatchSnapshot()

Whenever there are regressions, you would exactly know which tree in the node that doesn’t match the snapshot.

Now, this handiness comes with a price: it is more error-prone. There are chances that you wouldn’t know that the snapshot is in fact wrong and you would end up committing it anyway. So, double check your snapshot as if it is your own assertion code (because it is).

Of course there is more to it on snapshot testing. Check out the full docs.

#4. describe.each and test.each

Have you written some test that is somewhat similar to this?

describe('When I am a supervisor', () => {
test('I should have a supervisor badge', () => {
const employee = new Employee({ level: 'supervisor' })

expect(employee.badges).toContain('badge-supervisor')
})

test('I should have a supervisor level', () => {
const employee = new Employee({ level: 'supervisor' })

expect(employee.level).toBe('supervisor')
})
})

describe('When I am a manager', () => {
test('I should have a manager badge', () => {
const employee = new Employee({ level: 'manager' })

expect(employee.badges).toContain('badge-manager')
})

test('I should have a manager level', () => {
const employee = new Employee({ level: 'manager' })

expect(employee.level).toBe('manager')
})
})

That is painstakingly repetitive, right? Imagine doing it with more cases.

With describe.each and test.each, you could condense the code as such:

const levels = [['manager'], ['supervisor']]
const privileges = [['badges', 'toContain', 'badge-'], ['level', 'toBe', '']]

describe.each(levels)('When I am a %s', (level) => {
test.each(privileges)(`I should have a ${level} %s`, (kind, assert, prefix) => {
const employee = new Employee({ level })

expect(employee[kind])[assert](`${prefix}${level}`)
})
})

However, I have yet to actually use this in my own test, since I prefer my test to be verbose, but I just thought this was an interesting trick.

Check out the docs for more details on the arguments (spoiler: the table syntax is really cool).

#5. Mocking global functions once

At some point you would have to test something that depends on global functions on a particular test case. For example, a function that gets the info of the current date using Javascript object Date, or a library that relies on it. The tricky part is that if it's about the current date, you can never get the assertion right.

function foo () {
return Date.now()
}

expect(foo()).toBe(Date.now())
// ❌ This would throw occasionally:
// expect(received).toBe(expected) // Object.is equality
//
// Expected: 1558881400838
// Received: 1558881400837

Eventually, you had to override Date global object so that it is consistent and controllable:

function foo () {
return Date.now()
}

Date.now = () => 1234567890123

expect(foo()).toBe(1234567890123) // ✅

However, this is considered a bad practice because the override persists in between tests. You won’t notice it if there’s no other test that relies on Date.now, but it is leaking.

test('First test', () => {
function foo () {
return Date.now()
}

Date.now = () => 1234567890123

expect(foo()).toBe(1234567890123) // ✅
})

test('Second test', () => {
function foo () {
return Date.now()
}

expect(foo()).not.toBe(1234567890123) // ❌ ???
})

I used to ‘hack’ it in a way that it won’t leak:

test('First test', () => {
function foo () {
return Date.now()
}

const oriDateNow = Date.now
Date.now = () => 1234567890123

expect(foo()).toBe(1234567890123) // ✅
Date.now = oriDateNow
})

test('Second test', () => {
function foo () {
return Date.now()
}

expect(foo()).not.toBe(1234567890123) // ✅ as expected
})

However, there’s a much better, less hacky way to do it:

test('First test', () => {
function foo () {
return Date.now()
}

jest.spyOn(Date, 'now').mockImplementationOnce(() => 1234567890123)

expect(foo()).toBe(1234567890123) // ✅
})

test('Second test', () => {
function foo () {
return Date.now()
}

expect(foo()).not.toBe(1234567890123) // ✅ as expected
})

In summary, jest.spyOn spies on the global Date object and mock the implementation for now function just for one call. This would in turn keep Date.now untouched for the rest of the tests.

There is definitely more to it on the topic of mocking in Jest. Do check out the full docs for more details.

This article is getting longer, so I guess that’s it for now. These are barely scratching the surface of Jest’s capabilities, I was just highlighting my favorites. If you have other interesting facts, let me know as well.

And also, if you used Jest a lot, check out Majestic which is a zero-config GUI for Jest, a really good escape from the boring terminal output. I’m not sure if the author is in dev.to, but shout out to the person.

As always, thanks for reading my post!

Originally published at https://dev.to on May 26, 2019.

--

--