Best Practices to Write Unit Tests the Right Way

There are countless articles on the web about unit tests: TDD approach, beginner's guides, mocking frameworks, test coverage tools, and more. However, the vast majority of these articles are either too "Hello World"-like or focus more on tools, while missing the key point — how to write unit tests that are useful and deliver the most value?

In this article, I'll share some best practices, which I believe will get your unit tests to the next level. This will be an example-driven article as in my view, this is the best way to explain things in programming.

We're going to start with a code example (a C# class plus a few unit tests for this class) and highlight some important points related to what we want to test and how we're going to do that. Then we'll review one of the many ways (my favorite one) to structure unit tests. Finally, I'll share a few awesome libraries that I use in all my test projects.

Ready?

Lets' get started🍿

Example 1

UserManager

public class UserManager : IUserManager
{
    private readonly IUserRepository _userRepository;
    private readonly IUserGuard _userGuard;

    public UserManager(IUserRepository userRepository, IUserGuard userGuard)
    {
        _userRepository = userRepository;
        _userGuard = userGuard;
    }

    public async Task ChangeEmail(int userId, string newEmail, CancellationToken ct = default)
    {
        var user = await _repository.GetUser(userId, ct);

        _guard.UserExists(user);
        _guard.UserIsAllowedToChangeEmail(user!);
        _guard.EmailIsValid(newEmail);
        _guard.EmailIsNew(user!, newEmail);

        user!.Email = newEmail;
        await _repository.Save(user, ct);
    }
}

Test strategy

1. We want to verify that all calls are to IUserGuard happen in the correct order and, most importantly, come before _repository.Save. In other words, validation has to go before the data is saved to the database.
2. Make sure we pass the correct parameters to all methods that we call here.
3. Our ChangeEmail method calls _repository.Save.
4. We will save the user object with the new email address.

Tests

public class UserManagerTests
{
    public class ChangeEmail : UserManagerTestsBase
    {
        [Theory]
        [AutoData]
        public async Task Should_change_email(int userId,
                                              string newEmail,
                                              User user,
                                              CancellationToken ct)
        {
            // arrange
            Repository.GetUser(userId, ct).Returns(user);

            // act
            await Manager.ChangeEmail(userId, newEmail, ct);

            // assert
            Received.InOrder(() =>
                             {
                                 Guard.Received(1).UserExists(user);
                                 Guard.Received(1).UserIsAllowedToChangeEmail(user);
                                 Guard.Received(1).EmailIsValid(newEmail);
                                 Guard.Received(1).EmailIsNew(user, newEmail);

                                 Repository.Received(1).Save(Arg.Is<User>(x => x == user &&
                                                                               x.Email == newEmail),
                                                             ct);
                             });
        }
    }

    public abstract class UserManagerTestsBase
    {
        protected readonly UserManager Manager;
        protected readonly IUserRepository Repository;
        protected readonly IUserGuard Guard;

        protected UserManagerTestsBase()
        {
            Repository = Substitute.For<IUserRepository>();
            Guard = Substitute.For<IUserGuard>();
            Manager = new UserManager(Repository, Guard);
        }
    }
}

Notes

1. Instead of hardcoding test data, we generated random values using [AutoData] (AutoFixture.Xunit2 library).
2. To validate the order of method calls, we used Received.InOrder(...) (NSubstitute — the best mocking library for .NET).
3. We used Arg.Is<User>(x => x.Email == newEmail) in order to make sure we changed the email address user before saving this object to the database.

Now we're going to add one more method to the UserManager class and unit test it.

Example 2

public class UserManager : IUserManager
{
    private readonly IUserRepository _repository;
    private readonly IUserGuard _guard;

    public UserManager(IUserRepository repository, IUserGuard guard)
    {
        _repository = repository;
        _guard = guard;
    }

    // public async Task ChangeEmail(int userId, string newEmail, CancellationToken ct = default)
    // {...}

    public async Task ChangePassword(int userId, string newPassword, CancellationToken ct = default)
    {
        var user = await _repository.GetUser(userId, ct);

        if (user == null)
        {
            throw new ApplicationException($"User {userId} not found");
        }

        user.Password = newPassword;
        await _repository.Save(user, ct);
    }
}

Test strategy

1. This method has a if statement, so we want to make sure that the correct branch of code runs depending on the condition in the if statement.
2. Instead of ignoring the exception, we want to validate the type of thrown exception along with its message.
3. Same as in the previous example, we want to verify that we passed the user object with an updated password to the _repository.Save method.
4. If an exception is thrown, then the _repository.Save the method must not get called.

Tests

public class UserManagerTests
{
    // public class ChangeEmail : UserManagerTestsBase
    // {...}

    public class ChangePassword : UserManagerTestsBase
    {
        [Theory, AutoData]
        public async Task Should_throw_ApplicationException_when_user_not_found(int userId,
                                                                                string newPassword,
                                                                                CancellationToken ct)
        {
            // act
            var action = () => Manager.ChangePassword(userId, newPassword, ct);

            // assert
            await action.Should()
                        .ThrowAsync<ApplicationException>()
                        .WithMessage($"User {userId} not found");

            await Repository.DidNotReceiveWithAnyArgs().Save(Arg.Any<User>(), Arg.Any<CancellationToken>());
        }

        [Theory, AutoData]
        public async Task Should_change_password_and_save(int userId,
                                                          string newPassword,
                                                          User user,
                                                          CancellationToken ct)
        {
            // arrange
            Repository.GetUser(userId, ct).Returns(user);

            // act
            await Manager.ChangePassword(userId, newPassword, ct);

            // assert
            await Repository.Received(1).Save(Arg.Is<User>(x => x == user &&
                                                                x.Password == newPassword),
                                              ct);
        }
    }

    public abstract class UserManagerTestsBase
    {
        protected readonly UserManager Manager;
        protected readonly IUserRepository Repository;
        protected readonly IUserGuard Guard;

        protected UserManagerTestsBase()
        {
            Repository = Substitute.For<IUserRepository>();
            Guard = Substitute.For<IUserGuard>();
            Manager = new UserManager(Repository, Guard);
        }
    }
}

Notes

1. For the ChangePassword the method we have two tests: the first test to validate that an exception gets thrown when a user is not found, and the second one to check that we are making a call to _repository.Save.
2. Note how we test exceptions: instead of try-catch we create a delegate to the test method and then do action.Should().ThrowAsync<ApplicationException>().

Our tests are done now, let's take a look at the unit test coverage in order to make sure we don't miss anything and there is no untested code left.

Test Structure

To a large extent, the layout of a test project is a matter of taste. Here is what I prefer to do. For classes with more than one method (managers, services, etc) the structure of the related class with unit tests would be as follows:

public class MyClassTests
{
    public class Method1 : MyClassTestsBase
    {
        [Theory]
        [AutoData]
        public async Task Should_return_A_when_X(parameters)
        {
            // arrange
            // ...

            // act
            // ...

            // assert
            // ...
        }

        [Theory]
        [AutoData]
        public async Task Should_throw_B_when_Y(parameters)
        {
            // arrange
            // ...

            // act
            // ...

            // assert
            // ...
        }
    }

    public class Method2 : MyClassTestsBase
    {
        [Fact]
        public async Task Should_return_A_when_X()
        {
            // arrange

            // act

            // assert
        }

        [Fact]
        public async Task Should_throw_B_when_Y()
        {
            // arrange

            // act

            // assert
        }
    }

    public abstract class MyClassTestsBase
    {
        protected readonly MyClass Instance;
        protected readonly IDependency1 Dependency1;
        protected readonly IDependency2 Dependency2;

        protected UserManagerTestsBase()
        {
            Dependency1 = Substitute.For<IDependency1>();
            Dependency2 = Substitute.For<IDependency2>();
            Instance = new MyClass(IDependency1, IDependency2);
        }
    }
}

This is what it looks like for the test runner:

Unit test structure for classes with a single method (message handlers, factories etc):

public class MyHandlerTests
{
    private readonly MyHandler _handler;

    public MyHandlerTests()
    {
        _handler = new MyHandler();
    }

    [Fact]
    public void Should_do_A_when_X()
    {
        // arrange

        // act

        // assert
    }
}

Alright, so we have an understanding of the key concepts of unit testing and learned the best way to structure our tests. Now let's check out a few brilliant libraries which, I'm absolutely sure, you'll love.

Libraries and frameworks

xUnit

It’s usually a choice between xUnit, nUnit, and MSTest, and my personal preference goes for the former. A while ago when Microsoft started using xUnit in its projects, this framework became a default option.

Autofixture + Cheat Sheet

Instead of using hard-coded values for tests, Autofixture can generate random data for us.

var fixture = new Fixture();
var firstName = fixture.Create<string>();
var numOfUsers = fixture.Create<int>();
var employees = fixture.CreateMany<Employee>();

Autofixture.Xunit2 nuget for [AutoData] attribute (the one we used in our examples).

NSubstitute

There are a few mocking libraries out there, but I find this one the most explicit, natural, and clean. A small bonus: NSubstitute.Analyzers — Roslyn analyzers to detect (during compilation) possible errors using NSubstitute.

FluentAssertions

Fantastic assertion framework with very detailed and well-structured documentation.

A couple of examples taken from their docs:

string actual = "ABCDEFGHI";
actual.Should().StartWith("AB").And.EndWith("HI").And.Contain("EF").And.HaveLength(9);

IEnumerable<int> numbers = new[] { 1, 2, 3 };

numbers.Should().OnlyContain(n => n > 0);
numbers.Should().HaveCount(4, "because we thought we put four items in the collection");

Isn't it beautiful?

Stryker.NET

Stryker offers mutation testing for your .NET Core and .NET Framework projects. It allows you to test your tests by temporarily inserting bugs in your source code.

It's a very interesting tool that modifies the source code and expects tests to fail. If they don't, it complains about incomplete unit tests or poor test coverage.

Summary

We've analyzed two simple, yet important code examples that cover the most popular scenarios for unit testing. Learned one of the ways to structure and group tests, and make a list of essential libraries that will make your tests clean, explicit, and more effective.

Cheers!