Testing in Rails

One of the most fundamental aspects of programmer productivity is the feedback loop. "Scripting" languages like Ruby and Python are great for this because you can run your code immediately after writing it. Conversely, lower-level languages like C must be compiled before being run.

Two programmers swordfight from their office chairs, using "Compiling!" as an excuse

The Feedback Loop

  1. Decide what to do next
  2. Think of an approach
  3. Write some code
  4. Compile/run the code
  5. Observe the output
  • If it looks good, move on to Step 1 for the next task.
  • If there are problems, go back to Step 1 for this task.

Having a short feedback loop –– from brain to fingertips to running process –– lowers the resistance for trying new things and protects you from distractions that sneak in while you're waiting between steps.

Unfortunately, there's more than one way to distract a programmer!

The Rails Feedback Loop

  1. Decide what to do next
  2. Read every related Rails Guide at least twice
  3. Spend an hour poring over StackOverflow answers from two major versions ago
  4. Copy-paste some sketchy-looking code
  5. Run the code, which immediately screws up your database
  6. Run bundle exec rake db:reset
  7. Get a migraine because you forgot to put your painstakingly-designed seed data in db/seeds.rb, and now it's all gone
  8. Go for a walk
  9. Write some code
  10. Refresh browser
  11. Scratch your head over why it says there's a missing template
  12. Spend another hour tweaking your new code until you realize you made a typo in the filename and nothing you've been doing for the past three hours had a chance of working in the first place
  13. Take a break to read The Hitchhiker's Guide to Rails:

Rails is big. Really big. You just won't believe how vastly, hugely, mind-bogglingly big it is. I mean, you may think it's a long way down the road to the chemist, but that's just peanuts to Rails.

-- Douglas Adams

(Okay, so maybe Mr. Adams was talking about space, but he totally would have said it this way if he'd been writing about web development instead of interstellar travel.)

Rails ships with many features to save precious seconds in developer feedback loops, but there's no two ways about it: in anything but the most trivial app, it can be pretty complex to make sure your code is actually working correctly.

In this lesson, we'll learn to shorten our feedback loop with different flavors of Rails tests, combining some standard approaches suggested in the Guides themselves with some more advanced practices that require additional dependencies (namely Capybara).

Objectives

After this lesson, you should be able to...

  • List the different types of tests applicable to a Sinatra/Rails app.
  • Compare and contrast the different types of tests.
  • Describe the usage of Capybara in feature testing.

Three Test Types

We'll be covering three types of tests:

  • Models (RSpec)
  • Controllers (RSpec)
  • Features (RSpec/Capybara)

Features are the fanciest, so we'll leave them for last. They are preferred over regular Rails "View" tests.

RSpec

By default, Rails uses Test::Unit for testing, which keeps its tests in the mysteriously-named test/ folder.

If you're planning from the start to use RSpec instead, you can tell Rails to skip Test::Unit by passing the -T flag to rails new, like so:

rails new cool_app -T

Then, you will add the gem to your Gemfile:

gem 'rspec-rails'

And use the built-in generator to add a spec folder with the right boilerplate:

bundle install
bundle exec rails g rspec:install

This is the Rails equivalent of the usual rspec --init.

An Automobile Analogy

It's not enough to know how to test applications; we must also understand why it makes sense to test them in a certain way. To help with this, try thinking of Rails applications like cars:

  • Models are the basic parts that make cars useful, like the fuel tank, engine, and tires.

  • Views are the interactive parts that the driver (user) can see and touch, like the steering wheel, pedals, and gear shift.

  • Controllers are the rest of the connecting parts under the hood that connect the views to the models, like how the steering column (along with the rest of the steering assembly) connects the wheel to the tires. The users don't really see them, think of them, or even necessarily know they exist, but they're just as necessary.

Models are not too difficult to test because they have very specific purposes that can be easily separated from the rest of the application.

This holds true for cars as well: a good tire has a tread that grips the road in adverse conditions, doesn't puncture easily, and so on. You don't really need to think about the rest of the car when you're testing what makes a good wheel.

This gets a little trickier for views and controllers, but we'll talk more about that once we're done with models.

Model Tests

These go in spec/models, one file per model.

Model tests use the least amount of special features, since all you really need is the model class itself. The most common usage for model tests is to make sure you have set up your validations correctly.

Suppose we're working with this model:

# app/models/monster.rb

class Monster < ActiveRecord::Base
  validates :name, presence: true
  validates :size, inclusion: { in: ["tiny", "average", "like, REALLY big"] }
  validates :taxonomy, format: { with: /\A[A-Z](\.|[a-z]+) [a-z]{2,}\z/,
    message: "must include genus and species, like 'Homo sapiens'" }
end

Testing for Validity

First, we'll make sure that it understands a valid Monster:

# spec/models/monster_spec.rb

describe Monster do
  let(:attributes) do
    {
      name: "Dustwing",
      size: "tiny",
      taxonomy: "Abradacus nonexistus"
    }
  end

  it "is considered valid" do
    expect(Monster.new(attributes)).to be_valid
  end
end

Some questions to answer first:

What is let?

If you haven't seen let before, it is a standard helper method that takes a symbol and a block. It runs the block once per example in which it is called and saves the return value in a local variable named according to the symbol. This means you get a fresh copy in every test case.

Why is let better than before :each?

It's more fine-grained, which means you have better control over your data. It can be used in combination with before statements to set up your test data just right before the examples are run.

Why did we use let to make an attribute hash?

We could have put the entire Monster.new call inside our let block, but using an attribute hash instead has some advantages:

  • If we want to tweak the data first, we can just pass attributes.merge(name: "Other") while preserving the rest of the attributes.
  • We can also refer to attributes when making assertions about what the actual object should look like.

It's a good balance between saving keystrokes and maintaining the flexibility of your test data.

Where does be_valid come from?

RSpec provides plenty of built-in matchers, which you can peruse in their API docs, but be_valid is conspicuously absent from the list.

This code uses a neat trick that RSpec refers to as "predicate matchers", and you'll see it a lot in Rails testing.

In Ruby, it's conventional for methods that return true or false to be named with a question mark at the end. These methods are called predicate methods, because "predicate" is an English grammar term for the part of a sentence that makes a statement about the subject.

As you learned earlier in this unit, Rails provides a valid? method that returns true or false depending on whether the model object in question passed its validations.

In RSpec, when you call a nonexistent matcher (such as be_valid), it strips off the be_ (valid), adds a question mark (valid?), and checks to see if the object responds to a method by that name (monster.valid?).

Testing for Validation Failure

Now, let's add some tests to make sure our validations are working in the opposite direction:

# spec/models/monster_spec.rb

  let(:missing_name) { attributes.except(:name) }
  let(:invalid_size) { attributes.merge(size: "not that big") }
  let(:missing_species) { attributes.merge(taxonomy: "Abradacus") }

  it "is invalid without a name" do
    expect(Monster.new(missing_name)).not_to be_valid
  end

  it "is invalid with an unusual size" do
    expect(Monster.new(invalid_size)).not_to be_valid
  end

  it "is invalid with a missing species" do
    expect(Monster.new(missing_species)).not_to be_valid
  end

Note that each of these let blocks rely on the first one, attributes, which contains all of our valid attributes. missing_name uses the Rails hash helper except to exclude the name key while the other two use the standard Ruby merge method to overwrite valid attributes with invalid ones.

I saw some search results using should. What is that?

should is an older RSpec syntax that has been deprecated in favor of expect.

Any other RSpec features to know about?

Several RSpec features have been moved over time into the rspec-collection_matchers gem, which can make some detailed assertions more readable.

Controller Tests

The biggest risk in writing controller tests is redundancy: controllers exist to connect views and models, so it's difficult to test them in isolation.

First, we'll go over how to write controller tests. Then, our discussion of the why will bring us into the final subject, "feature tests".

# spec/controllers/monsters_controller_spec.rb

describe MonstersController, type: :controller do
  let(:attributes) do
    {
      name: "Dustwing",
      size: "tiny",
      taxonomy: "Abradacus nonexistus"
    }
  end

  it "renders the show template" do
    monster = Monster.create!(attributes)
    get :show, id: monster.id
    expect(response).to render_template(:show)
  end

  describe "creation" do
    before { post :create, monster: attributes }
    let(:monster) { Monster.find_by(name: "Dustwing") }

    it "creates a new monster" do
      expect(monster).to_not be_nil
    end

    it "redirects to the monster's show page" do
      expect(response).to redirect_to(monster_path(monster))
    end
  end
end

You can use the get and post methods (along with patch and delete) to initiate test requests on the controller. A response object is available to set expectations on, such as render_template or redirect_to.

The tests above are great, especially while we're still getting used to how controllers are wired. However, almost these exact tests could be copied for any controller set up according to Rails' RESTish conventions. There's nothing inherently wrong with that, but the redundance, along with the need to test views, inspired the creation of a new type of test supported by Capybara known as a "Feature Test".

Feature Tests

If you were going to write tests for a car's steering wheel, what would you start with?

Here's one idea:

When the steering wheel is rotated to the left, the tires rotate to the left.

This makes sense, but it's testing much more than the steering wheel. This test relies on the view (steering wheel), the model (tires), and the controller (steering column)!

This is called an acceptance test because it is phrased in terms of features that provide value to the user. (It could also be called an integration test because it tests more than one piece of the system at once.)

Can we isolate the steering wheel while still testing its functionality?

Not really –– the whole point of the steering wheel is to control the tires. We could talk about how it looks or what it's made of, but the functionality is inherently tied to the underlying system, just like the views in a Rails app. All of those forms and templates are meaningless without controllers and models to populate them.

In the last section, we did our best to isolate the controller, and, as a result, we wrote many of our tests in terms of the controller's internal parts (such as redirects and request methods). We don't care what the HTML looks like, what button the user pressed, or how the models are behaving.

This is called a unit test, because it tests a single unit of functionality.

For a car, it might look like this:

When the steering column's flange rotates, the steering shaft transmits the rotation to the steering box.

Phew. Good thing they covered steering assemblies on the last episode of Car Talk, or I wouldn't have known where to start!

By now you're probably realizing that if you say "steering" enough times, it stops sounding like a real word.

But, more importantly, it can be difficult to write isolated unit tests, and it's not always clear whether they're useful. Compare the jargon-heavy, extremely specific unit test, above, to this test covering the steering wheel (view) and steering column (controller):

When the steering wheel is rotated to the left, the steering column transmits the rotation to the steering box.

This is a feature test.

The acceptance test at the top of this section covers too much ground, making it brittle and difficult to maintain.

The unit test in the middle of this section is so specific that it almost feels like we just rewrote the controller code with different phrasing.

The feature test, on the other hand, is Just Right. It lets us think like a user (in terms of the steering wheel, or view) while still making intelligent assertions about how the underlying system should respond to input (in terms of the steering column, or controller).

Now, on to the how.

Capybara

When you see key words like visit, fill_in, and page, you know you're looking at a Capybara test.

To set up Capybara, one must first add the gem to the Gemfile:

gem 'capybara'

Then set up Capybara-Rails integration in spec/rails_helper.rb:

require 'capybara/rails'

Then set up Capybara-RSpec integration in spec/spec_helper.rb:

require 'capybara/rspec'

Feature tests are traditionally located in spec/features, but you can put them anywhere if you pass the :type => :feature option to your describe call.

To test our monster manager with Capybara, we'll start by setting up a GET request and then use Capybara's convenient helper functions to interact with the page just like a user would:

# spec/features/monster_creation.rb

describe "monster creation", type: :feature do
  before do
    visit new_monster_path
    fill_in "Name", with: "Dustwing"
    select "tiny", from: "monster_size"
    fill_in "Taxonomy", with: "Abradacus nonexistus"
    click_button "Create Monster"
  end

When click_button is called, this will trigger the POST request to the controller's create action, just as if a user had clicked it in their browser.

Now, we can write our original controller tests like usual:

  let(:monster) { Monster.find_by(name: "Dustwing") }

  it "creates a monster" do
    expect(monster).to_not be_nil
  end

  it "redirects to the new monster's page" do
    expect(current_path).to eq(monster_path(monster))
  end

And because we're in Capybara land, we also have a very convenient way of making assertions about the final GET request:

  it "displays the monster's name" do
    within "h1" do
      expect(page).to have_content(monster.name)
    end
  end

within sets the context for our next expectation, restricting it to the first <h1> tag encountered on the page. This way, our expect call will only pass if the specified content ("Dustwing") appears inside that first heading.

One interesting thing about this approach is that we're being much less explicit about certain expectations. For example, we're testing the redirect not with the initial 302 response but instead by examining the current path in Capybara's virtual "browser session". This is much more powerful and intuitive, and it doesn't sacrifice much in the way of expressivity.

Summary

The hardest part about testing usually ends up being the "why" and not the "how". Why write the test this way and not that way?

You will probably see Capybara feature tests in wider usage than direct controller and view tests, but they're not universal. If you're doing something unusual, like a series of complex redirects that change based on authorization level, it makes sense to write a more isolated controller test. But for standard CRUD functionality, Capybara is designed to save you a lot of time and mental effort.

These can serve as fairly reliable guidelines:

  • Models should always be thoroughly unit tested.
  • Controllers should be as thin as possible to keep your feature tests simple.
  • If you can't avoid making a controller complex, it deserves its own isolated test.
  • Capybara's syntax is much more powerful than Rails's built-in functionality for view tests, so stick with it whenever possible.
  • Don't get carried away with the details when testing views: you just need to make sure the information is in the right place. If your tests are too strict, it will be impossible to make even simple tweaks to your templates without breaking the build.

View Testing in Rails on Learn.co and start learning to code for free.