Test which reminded me why I don’t really like RSpec

Recently, our friend from a different software company asked us for some help with mutant setup. We asked to share a sample test to discover what can be wrong. When I read the snippet on a slack channel, I had immediately written:

— Oh man, this example reminds me why I don’t like RSpec.

— Feel free to review it, it’s written by one of our juniors and I need to give him some feedback soon. — he responded.

Just to be clear: I have no problem with the RSpec library itself. It’s a great tool, I’ve used it in several projects, but I don’t like how a lot of people utilise it. Let’s have a look at it.

Let’s have a look at the original example:

# frozen_string_literal: true

require "rails_helper"

RSpec.describe Api::Students::Update do
  let(:user) { create(:user) }
  let(:params) { { last_name: "something" } }
  subject(:result) { described_class.call(current_user: user, params: params) }

  context "when user is a student" do
    it { is_expected.to be_success }
  end

  context "when user is teacher" do
    let(:user) { create(:user, :teacher) }
    it { is_expected.to be_failure }
  end

  describe "checking Address update" do
    let(:new_zip_code) { Faker::Address.zip_code }
    before { params[:zip_code] = new_zip_code }

    context "when user has address and want to change something in their address" do
      let!(:address) { create(:address, owner: user) }

      it "will succeed" do
        expect(result.success?).to eq(true)
        expect(address.reload.zip_code).to eq(new_zip_code)
      end
    end

    context "when user has not have any address and want to change something in their address" do
      let(:new_zip_code) { Faker::Address.zip_code }
      before { params[:zip_code] = new_zip_code }
      it { is_expected.to be_success }
      it "create address with given params" do
        expect { result }.to change(Address.all, :count).from(0).to(1)
        expect(Address.last.zip_code).to eq(new_zip_code)
        expect(user.address).to eq(Address.last)
      end
    end
  end

  describe "checking parent update" do
    let(:parent_first_name) { Faker::Name.female_first_name }
    let(:params) { { parent: { first_name: parent_first_name } } }

    context "when user has parent and want to change something" do
      let(:parent) { create(:parent) }
      let(:user) { create(:user, profession: student) }
      let(:student) { create(:student, parent: parent) }

      it { is_expected.to be_success }
      it "will succeed" do
        result
        expect(parent.reload.first_name).to eq(parent_first_name)
      end
    end

    context "when user has not have any parent and want to change something in their parent" do
      it { is_expected.to be_success }
      it "will create parent with given params" do
        expect { result }.to change(Parent.all, :count).from(0).to(1)
        expect(Parent.last.reload.first_name).to eq(parent_first_name)
        expect(user.profession.parent).to eq(Parent.last)
      end
    end
  end
end

Here’s a list of my thoughts after reading test above:

RSpec specific syntax sugar to actual test code ratio is too d*mn high
Tests should be verbose about their subject, not about the plumbing around.
I don’t get what the test is about, it’s unreadable because of the nested it in context

— hey, but we have IDEs which can fold the blocks of code…

— Cool, but not here on Slack, nor on GitHub where you usually make your code reviews. I simply don’t want to jump around the file to see what’s the input to a service call.

Setup is done through FactoryBot which sets some artificial database state, often not following the business rules (if your business rules live in ActiveRecord models — I’m sorry, we’re past that since years). It’s better to use domain services to setup the initial state. I’ve seen a lot of codebases struggling with gigantic test execution time because of too many database object being created because of how FactoryBot was used.
context is only used to overwrite let, so there’s different setup in different examples. Why not keep the structure flat and do the setup explicit in every example? If you need something different that declared in let, just use local variable in the example. let is great for specifying dependencies and things that don’t change per each test case.
The main input to this class giving different outcomes are params — this should be clearly visible how they differ in the input and what output they give, why not pass them explicitly to call?
Scope the test expectations, making assertion on Parent.all won’t give any guarantee that the service assigned data to a desired Parent object.

Talk is cheap, so I did a 5 minutes refactoring resulting in this:

# frozen_string_literal: true

require "rails_helper"

RSpec.describe Api::Students::Update do
  specify "when user is a student" do
    expect(Api::Students::Update.call(current_user: user, params: { last_name: "something" })).to be_success
  end

  specify "when user is a teacher" do
    expect(Api::Students::Update.call(current_user: teacher, params: { last_name: "something" })).to be_failure
  end

  specify "when user has address and want to change something in their address will succeed" do
    address = create(:address, owner: user)

    result = Api::Students::Update.call(current_user: user, params: { last_name: "something", zip_code: new_zip_code })

    expect(result.success?).to eq(true)
    expect(address.reload.zip_code).to eq(new_zip_code)
  end

  specify "when user has not have any address and want to change something in their address create address with given params" do
    result =
      expect {
        Api::Students::Update.call(current_user: user, params: { last_name: "something", zip_code: new_zip_code })
      }.to change(Address.all, :count).from(0).to(1)

    expect(result.success?).to eq(true)
    expect(Address.last.zip_code).to eq(new_zip_code)
    expect(user.address).to eq(Address.last)
  end

  specify "when user has parent and want to change something" do
    user = create(:user, profession: student)

    result = Api::Students::Update.call(current_user: user, params: { parent: { first_name: parent_first_name } })

    expect(result).to be_success
    expect(parent.reload.first_name).to eq(parent_first_name)
  end

  specify "when user has not have any parent and want to change something in their parent" do
    user = create(:user, profession: student)

    result =
      expect {
        Api::Students::Update.call(current_user: user, params: { parent: { first_name: parent_first_name } })
      }.to change(Parent.all, :count).from(0).to(1)

    expect(result).to be_success
    expect(Parent.last.reload.first_name).to eq(parent_first_name)
    expect(user.profession.parent).to eq(Parent.last)
  end

  let(:user) { create(:user) }
  let(:teacher) { create(:user, :teacher) }
  let(:student) { create(:student, parent: parent) }
  let(:parent) { create(:parent) }
  let(:new_zip_code) { Faker::Address.zip_code }
  let(:parent_first_name) { Fake::Name.female_first_name }
end

What are your thoughts? Which version is more readable to you?

Test which reminded me why I don't really like RSpec

Test which reminded me why I don’t really like RSpec

You might also like