Test Doubles: in theory, in Minitest and in RSpec
Those of us that do Test Driven Development have heard about doubles, mocks, stubs, fakes and spies multiple times. Unfortunately there is a ton of confusion about all these words and their meaning. Let's see what each an every one of these really mean, where we should use them and how the popular testing frameworks for Ruby implement these.
So, first things first. One of the biggest misconceptions is that doubles are types of objects that are used in testing.
Dummies, mocks, stubs, fakes and spies are test doubles. Test double is the category of these test objects. I think that the confusion around this concept arises from the naming that the popular testing frameworks use for these objects.
So remember - dummies, mocks, stubs, fakes and spies are test doubles. Also, another thing to keep in mind is the word mock. The words mock and doubles were used interchangeably in the past and their definitions got skewed.
Now, that we got this right, what is a dummy?
The simplest type of test double is test dummy. Basically, the dummy is dumb, hence the naming. This means that the object is just a placeholder for another object, but it's interfaces (or methods) do not return anything.
Confusing? Let's see an example.
We have this class
Blog that has a
class Post attr_reader :published def publish! @published = true end end class Blog attr_reader :published_posts def initialize author @author = author @published_posts =  end def publish! post published_posts << post post.publish! end # Moar methods... end
Now, we need to test the
publish! method. How do we go about this? We
obviously need to create a
Post object first, then create a
object and pass the author as an argument to the initializer. Afterwards, we
need to assert that the
publish! method will add the post to the
Let's test our method with Minitest:
require 'minitest/autorun' class BlogTest < Minitest::Test def test_publish_adds_post_to_published_posts blog = Blog.new(author) post = Post.new blog.publish!(post) assert_includes(blog.published_posts, post) end end
If we run the test, it will fail. Why? Well, what is
author? As you can see
if we exclude author when creating the
Blog object it will complain with an
ArgumentError. But, in the code that we are testing,
no role to play. It basically does nothing. It is important to the class, but
it's not important to the method under test.
This is a great place to use a dummy. Let's add the author using a dummy. Minitest and RSpec do not have the concept of a pure dummy, although RSpec has a trick that you can use.
If you need a dummy in Minitest, you can just use an
nil. Or an
Hash. Or.. whatever.
require 'minitest/autorun' class BlogTest < Minitest::Test def test_publish_adds_post_to_published_posts author = Object.new blog = Blog.new(author) post = Post.new blog.publish!(post) assert_includes(blog.published_posts, post) end end
The same test, but using RSpec:
require 'rspec/autorun' describe Post do describe "#publish!" do it 'adds the post to the published_posts array' do author = Object.new blog = Blog.new(author) post = Post.new blog.publish!(post) expect(blog.published_posts).to include(post) end end end
Also, another trick in RSpec is to use a double. A double that doesn't to anything is a dummy. Remember that this only applies to RSpec:
require 'rspec/autorun' describe Post do describe "#publish!" do it 'adds the post to the published_posts array' do author = double blog = Blog.new(author) post = Post.new blog.publish!(post) expect(blog.published_posts).to include(post) end end end
Do not get confused by the
double keyword. For now, just pretend that it's
an object that does nothing. Just like
Stubs are dummies, with one notable difference. Stubs have methods that they respond to. But, their methods do not do anything except they return a predefined value. This means that their methods do not have an effect, they just return a value which will drive the test through the production code.
Let's see some examples. We have the
Blog class which has the
class Blog attr_reader :published_posts def initialize user @user = user end def publish! post published_posts << post if @user.author? end # Moar methods... end
How can we test the
publish! method? Due to the check that occurs in the
publish! method, we need a stub
User object, which will return true
author? method is called on it.
Creating a stub in Minitest is dead easy. Let's write the test for the publish method:
require 'minitest/autorun' class BlogTest < Minitest::Test def test_publish post = Post.new user = User.new blog = Blog.new(user) user.stub :author?, true do blog.publish!(post) assert_includes blog.published_posts, post end end end
Stubbing in Minitest is done by calling
.stub on the object/class that you
want to stub a method on, with three arguments: the method name, the return
value of the stub and a block. The block is basically the scope of the stub, or
in other words, the stub will work only in the provided block.
Another way to achieve the same, which is a bit weird, is to avoid creating a
User object, create a dummy and declare a stub method on it.
require 'minitest/autorun' class BlogTest < Minitest::Test def test_publish post = Post.new user = Object.new # Tha stub def user.author? true end blog = Blog.new(user) blog.publish!(post) assert_includes blog.published_posts, post end end
Using this way, we define the
author? method on the singleton class of the
user object, which will return
The same test, or spec if you will, in RSpec is quite easy to write as well.
require 'rspec/autorun' describe Blog do describe '#publish' do it 'adds the published post to the published_posts collection' do post = Post.new user = double(:author? => true) blog = Blog.new(user) blog.publish!(post) expect(blog.published_posts).to include(post) end end end
As you can see, the RSpec syntax is easy to understand as well. We create a
double, which is basically a dummy (makes sense, right?) and we declare a stub
on it. The stub is the
author? which will return a predefined value, or in
I hope you guessed it - a spy is a stub. It's just an object whose methods do nothing and return a predefined values to drive the test. However the spy remembers certain things about the way and occurance of it's methods being called. For example, sometimes you want to assert that a certain method has been called. Another time, you might want to test how many times a method has been called.
That's where spies are used.
Again, let's see some code. Take these classes and methods for example:
class Blog attr_reader :published_posts def initialize user @user = user end def publish! post if @user.author? published_posts << post NotificationService.notify_subscribers(post) end end end
NotificationService.notify_subscribers method will send a
notification email to all the subscribers of the blog when a new post is
When testing this method, we are interested in two things:
- The post will be added to the
- The subscribers will be notified of the new post.
Since we already tested the first case in the example before, we'll just focus on no. 2. Let's test this method.
Minitest does not support spies by default, but as always, we can use another gem that plays nice with Minitest.
Spy is a really simple gem that does exactly that - spies.
Let's add some spies using Spy.
class BlogTest < Minitest::Test def test_notification_is_sent_when_publishing notification_service_spy = Spy.on(NotificationService, :notify_subscribers) post = Post.new user = User.new blog = Blog.new(user) blog.publish!(post) assert notificaion_service_spy.has_been_called? end end
Quite self-explanatory, isn't it?
The RSpec syntactic sugar is really nice and human. This is how we can create spies in RSpec and use them appropriately.
describe Blog do describe '#publish!' do it "sends notification when new post is published" do notification_spy = class_spy("Invitation") post = Post.new user = User.new blog = Blog.new(user) # If you want to set the expectation before the action is done: expect(notification_spy).to receive(:notify_subscribers).with(post) blog.publish!(post) # If you want to assert after the action is done: expect(NotificationService).to have_received(:notify_subscribers).with(post) end end end
Also, keep in mind that since spy has the functionality of a stub, it can
return fixed/predefined values. If you need your spy to return a value, you can
do this in RSpec very easily, by chaining the
and_return method to the spy:
describe Blog do describe '#publish!' do it "sends notification when new post is published" do notification_spy = class_spy("Invitation") post = Post.new user = User.new blog = Blog.new(user) expect(notification_spy).to receive(:notify_subscribers).with(post).and_return(true) blog.publish!(post) end end end
Before we go over the true mock, we have to make sure we understand the difference between “mocking”, “the RSpec mock” and a “True Mock”.
The whole issue behind the word mock is that in the beginning people started using it for simillar (but not the same) things and the word got left hanging in the air without a proper defnition.
So, “mocking” is usually used when thinking about creating objects that simulate the behavior of real objects or units.
On the other hand, RSpec also has the concept of a mock. From RSpec's documentation:
As you can see, there's some nomenclature overlapping and/or disagreement.
Now, throw away any knowledge you have about the word mock. The True Mock is a very interesting type of test object. Of course, it has all the functionalities that a dummy, a stub and a spy have. And, a bit more.
The True Mock is an object that knows the flow of the method under test, and the test can ask the mock if everything went well, usually called “verifying”. This means that the test will rely on the True Mock to track what are the effects of the method under test and what methods have been called.
Let's see an example. The code under test:
class Blog attr_reader :published_posts def initialize user @user = user end def publish! post published_posts << post if @user.can?(:publish) end def delete post post.delete! if @user.can?(:delete) end end
Minitest has this type of testing object - it's called (believe it or not)
class BlogTest < Minitest::Test def test_post_deletion user = Minitest::Mock.new post = Minitest::Mock.new blog = Blog.new(user) user.expect :can?, true post.expect :delete!, true blog.delete(post) user.verify post.verify end end
We create new
Minitest::Mock objects and we set an expectation that it should
delete! method, respectively. These will return
true. Although this seems like a stub (setting a return value), the last line
is what it differentiates it from a stub. The test asks the mocks to verify
that the expected method was called, which will result in the test
With RSpec, it's a bit different. RSpec does not have a
Mock object, but it's
double can do the same trick.
describe Blog do it 'can remove a post' do user = double post = double blog = Blog.new(user) expect(user).to receive(:can?).with(:publish) expect(user).to receive(:can?).with(:delete) expect(post).to receive(:delete!) blog.publish(post) blog.delete(post) end end
This is a bit ambigous, since it's half-way between spies and a true mock
objects. But if you look in all the examples above, RSpec's
double can morph
into anything. It all depends on the programmer what (s)he want's to use it as.
I guess that's why it's called
double, because it can be any type of a test
double you want.
Last but not least, the fake. The fake is basically a half-baked, half-working implementation of the test object. It is not production ready, but, it's enough functional for the test to use it.
The popular testing frameworks do not have the concept of a fake. This is due to that writing the (needed) implementation for the fake is left for the programmer to do.
You should avoid fakes because fakes are standalone objects which are not used in the production code, but, they will grow linearly as the system grows. This means that in the long run they can turn into code that will be very hard to maintain. Also, pretty much every test can be written with one of the aforementioned testing objects. So, if you are ever using fakes, make sure you know what you are getting into.
Now, that we got to the end, I hope that this article helped you understand the theory behind test doubles and their implementations in Ruby. Be careful where and how you use these test objects. If you go overboard you can create isolation between your tests and your production code, which can render your test suite useless.
Wow, you got to the very end. Thanks so much for reading!