Merge branch 'docs-writing-e2e-tests' into 'master'

First iteration of documentation for writing end-to-end tests

See merge request gitlab-org/gitlab-ce!27603

qa/README.md

@@ -49,8 +49,10 @@ will need to [modify your GDK setup](https://gitlab.com/gitlab-org/gitlab-qa/blo
 
 ### Writing tests
 
-1. [Using page objects](qa/page/README.md)
-2. [Style guide](STYLE_GUIDE.md)
+- [Writing tests from scratch tutorial](docs/WRITING_TESTS_FROM_SCRATCH.md)
+    - [Best practices](docs/BEST_PRACTICES.md)
+    - [Using page objects](qa/page/README.md)
+    - [Guidelines](docs/GUIDELINES.md)
 
 ### Running specific tests
 

qa/docs/BEST_PRACTICES.md

+# Best practices when writing end-to-end tests
+
+The majority of the end-to-end tests require some state to be built in the application for the tests to happen.
+
+A good example is a user being logged in as a pre-condition for testing the feature.
+
+But if the login feature is already covered with end-to-end tests through the GUI, there is no reason to perform such an expensive task to test the functionality of creating a project, or importing a repo, even if this features depend on a user being logged in. Let's see an example to make things clear.
+
+Let's say that, on average, the process to perform a successfull login through the GUI takes 2 seconds.
+
+Now, realize that almost all tests need the user to be logged in, and that we need every test to run in isolation, meaning that tests cannot interfere with each other. This would  mean that for every test the user needs to log in, and "waste 2 seconds".
+
+Now, multiply the number of tests per 2 seconds, and as your test suite grows, the time to run it grows with it, and this is not sustainable.
+
+An alternative to perform a login in a cheaper way would be having an endpoint (available only for testing) where we could pass the user's credentials as encrypted values as query strings, and then we would be redirected to the logged in home page if the credentials are valid. Let's say that, on average, this process takes only 200 miliseconds.
+
+You see the point right?
+
+Performing a login through the GUI for every test would cost a lot in terms of tests' execution.
+
+And there is another reason.
+
+Let's say that you don't follow the above suggestion, and depend on the GUI for the creation of every application state in order to test a specific feature. In this case we could be talking about the **Issues** feature, that depends on a project to exist, and the user to be logged in.
+
+What would happen if there was a bug in the project creation page, where the 'Create' button is disabled, not allowing for the creation of a project through the GUI, but the API logic is still working?
+
+In this case, instead of having only the project creation test failing, we would have many tests that depend on a project to be failing too.
+
+But, if we were following the best practices, only one test would be failing, and tests for other features that depend on a project to exist would continue to pass, since they could be creating the project behind the scenes interacting directly with the public APIs, ensuring a more reliable metric of test failure rate.
+
+Finally, interacting with the application only by its GUI generates a higher rate of test flakiness, and we want to avoid that at max.
+
+**The takeaways here are:**
+
+- Building state through the GUI is time consuming and it's not sustainable as the test suite grows.
+- When depending only on the GUI to create the application's state and tests fail due to front-end issues, we can't rely on the test failures rate, and we generates a higher rate of test flakiness.
+
+Now that we are aware of all of it, [let's go create some tests](./WRITING_TESTS_FROM_SCRATCH.md).

qa/STYLE_GUIDE.md → qa/docs/GUIDELINES.md

@@ -43,4 +43,4 @@ end
 
 Notice that in the above example, before clicking the `:operations_environments_link`, another element is hovered over.
 
-> We can create these methods as helpers to abstract multi-step navigation.
+> We can create these methods as helpers to abstract multi-step navigation.
\ No newline at end of file

qa/docs/WRITING_TESTS_FROM_SCRATCH.md

+# Writing end-to-end tests step-by-step
+
+In this tutorial, you will find different examples, and the steps involved, in the creation of end-to-end (_e2e_) tests for GitLab CE and GitLab EE, using GitLab QA.
+
+> When referring to end-to-end tests in this document, this means testing a specific feature end-to-end, such as a user logging in, the creation of a project, the management of labels, breaking down epics into sub-epics and issues, etc.
+
+## Important information before we start writing tests
+
+It's important to understand that end-to-end tests of isolated features, such as the ones described in the above note, doesn't mean that everything needs to happen through the GUI.
+
+If you don't exactly understand what we mean by **not everything needs to happen through the GUI,** please make sure you've read the [best practices](./BEST_PRACTICES.md) before moving on.
+
+## This document covers the following items:
+
+0. Identifying if end-to-end tests are really needed
+1. Identifying the [DevOps stage](https://about.gitlab.com/stages-devops-lifecycle/) of the feature that you are going to cover with end-to-end tests
+2. Creating the skeleton of the test file (`*_spec.rb`)
+3. The [MVC](https://about.gitlab.com/handbook/values/#minimum-viable-change-mvc) of the test cases logic
+4. Extracting duplicated code into methods
+5. Tests' pre-conditions (`before :all` and `before`) using resources and [Page Objects](./qa/page/README.md)
+6. Optimizing the test suite
+7. Using and implementing resources
+8. Moving elements definitions and its methods to [Page Objects](./qa/page/README.md)
+    - Adding testability to the application
+
+### 0. Are end-to-end tests needed?
+
+At GitLab we respect the [test pyramid](https://gitlab.com/gitlab-org/gitlab-ce/blob/master/doc/development/testing_guide/testing_levels.md), and so, we recommend to check the code coverage of a specific feature before writing end-to-end tests.
+
+Sometimes you may notice that there is already a good coverage in other test levels, and we can stay confident that if we break a feature, we will still have quick feedback about it, even without having end-to-end tests.
+
+If after this analysis you still think that end-to-end tests are needed, keep reading.
+
+### 1. Identifying the DevOps stage
+
+The GitLab QA end-to-end tests are organized by the different [stages in the DevOps lifecycle](https://gitlab.com/gitlab-org/gitlab-ce/tree/master/qa/qa/specs/features/browser_ui), and so, if you are creating tests for issue creation, for instance, you would locate the spec files under the `qa/qa/specs/features/browser_ui/2_plan/` directory since issue creation is part of the Plan stage.
+
+ In another case of a test for listing merged merge requests (MRs), the test should go under the `qa/qa/specs/features/browser_ui/3_create/` directory since merge request is a feature from the Create stage.
+
+> There may be sub-directories inside the stages directories, for different features. For example: `.../browser_ui/2_plan/ee_epics/` and `.../browser_ui/2_plan/issues/`.
+
+Now, let's say we want to create tests for the [scoped labels](https://about.gitlab.com/2019/04/22/gitlab-11-10-released/#scoped-labels) feature, available on GitLab EE Premium (this feature is part of the Plan stage.)
+
+> Because these tests are for a feature available only on GitLab EE, we need to create them in the [EE repository](https://gitlab.com/gitlab-org/gitlab-ee).
+
+Since [there is no specific directory for this feature](https://gitlab.com/gitlab-org/gitlab-ee/tree/master/qa/qa/specs/features/browser_ui/2_plan), we should create a sub-directory for it.
+
+Under `.../browser_ui/2_plan/`, let's create a sub-directory called `ee_scoped_labels/`.
+
+> Notice that since this feature is only available for GitLab EE we prefix the sub-directory with `ee_`.
+
+### 2. Test skeleton
+
+Inside the newly created sub-directory, let's create a file describing the test suite (e.g. `editing_scoped_labels_spec.rb`.)
+
+#### The `context` and `describe` blocks
+
+Specs have an outer `context` that indicates the DevOps stage. The next level is the `describe` block, that briefly states the subject of the test suite. See the following example:
+
+```ruby
+module QA
+  context 'Plan' do
+    describe 'Editing scoped labels properties on issues' do
+    end
+  end
+end
+```
+
+#### The `it` blocks
+
+Every test suite is composed by at least one `it` block, and a good way to start writing end-to-end tests is by typing test cases descriptions as `it` blocks. Take a look at the following example:
+
+```ruby
+module QA
+  context 'Plan' do
+    describe 'Editing scoped labels properties on issues' do
+      it 'replaces an existing label if it has the same key' do
+      end
+
+      it 'keeps both scoped labels when adding a label with a different key' do
+      end
+    end
+  end
+end
+```
+
+### 3. Test cases MVC
+
+For the [MVC](https://about.gitlab.com/handbook/values/#minimum-viable-change-mvc) of our test cases, let's say that we already have the application in the state needed for the tests, and then let's focus on the logic of the test cases only.
+
+To evolve the test cases drafted on step 2, let's imagine that the user is already logged in a GitLab EE instance, they already have at least a Premium license in use, there is already a project created, there is already an issue opened in the project, the issue already has a scoped label (e.g. `foo::bar`), there are other scoped labels (for the same scope and for a different scope, e.g. `foo::baz` and `bar::bah`), and finally, the user is already on the issue's page. Let's also suppose that for every test case the application is in a clean state, meaning that one test case won't affect another.
+
+> Note: there are different approaches to create an application state for end-to-end tests. Some of them are very time consuming and subject to failures, such as when using the GUI for all the pre-conditions of the tests. On the other hand, other approaches are more efficient, such as using the public APIs. The latter is more efficient since it doesn't depend on the GUI. We won't focus on this part yet, but it's good to keep it in mind.
+
+Let's now focus on the first test case.
+
+```ruby
+it 'keeps the latest scoped label when adding a label with the same key of an existing one, but with a different value' do
+  # This implementation is only for tutorial purposes. We normally encapsulate elements in Page Objects.
+  page.find('.block.labels .edit-link').click
+  page.find('.dropdown-menu-labels .dropdown-input-field').send_keys ['foo::baz', :enter]
+  page.find('#content-body').click
+  page.refresh
+
+  scoped_label = page.find('.qa-labels-block .scoped-label-wrapper')
+
+  expect(scoped_label).to have_content('foo::baz')
+  expect(scoped_label).not_to have_content('foo::bar')
+  expect(page).to have_content('added foo::baz label and removed foo::bar')
+end
+```
+
+> Notice that the test itself is simple. The most challenging part is the creation of the application state, which will be covered later.
+
+> The exemplified test cases' MVC is not enough for the change to be submitted in an MR, but they help on building up the test logic. The reason is that we do not want to use locators directly in the tests, and tests **must** use [Page Objects](./qa/page/README.md) before they can be merged.
+
+Below are the steps that the test covers:
+
+1. The test finds the 'Edit' link for the labels and clicks on it
+2. Then it fills in the 'Assign labels' input field with the value 'foo::baz' and press enter
+3. Then it clicks in the content body to apply the label and refreshes the page
+4. Finally the expectation that the previous scoped label was removed and that the new one was added happens
+
+Let's now see how the second test case would look like.
+
+```ruby
+it 'keeps both scoped labels when adding a label with a different key' do
+  # This implementation is only for tutorial purposes. We normally encapsulate elements in Page Objects.
+  page.find('.block.labels .edit-link').click
+  page.find('.dropdown-menu-labels .dropdown-input-field').send_keys ['bar::bah', :enter]
+  page.find('#content-body').click
+  page.refresh
+
+  scoped_labels = page.all('.qa-labels-block .scoped-label-wrapper')
+
+  expect(scoped_labels.first).to have_content('bar::bah')
+  expect(scoped_labels.last).to have_content('foo::ba')
+  expect(page).to have_content('added bar::bah')
+  expect(page).to have_content('added foo::ba')
+end
+```
+
+> Note that elements are always located using CSS selectors, and a good practice is to add test specific attribute:value for elements (this is called adding testability to the application and we will talk more about it later.)
+
+Below are the steps that the test covers:
+
+1. The test finds the 'Edit' link for the labels and clicks on it
+2. Then it fills in the 'Assign labels' input field with the value 'bar::bah' and press enter
+3. Then it clicks in the content body to apply the label and refreshes the page
+4. Finally the expectation that the both scoped labels are present happens
+
+> Similar to the previous test, this one is also very straight forward, but there is some code duplication. Let's address it.
+
+### 4. Extracting duplicated code
+
+If we refactor the tests created on step 3 we could come up with something like this:
+
+```ruby
+it 'keeps the latest scoped label when adding a label with the same key of an existing one, but with a different value' do
+  select_label_and_refresh 'foo::baz'
+
+  expect(page).to have_content('added foo::baz')
+  expect(page).to have_content('and removed foo::bar')
+
+  scoped_label = page.find('.qa-labels-block .scoped-label-wrapper')
+
+  expect(scoped_label).to have_content('foo::baz')
+  expect(scoped_label).not_to have_content('foo::bar')
+end
+
+it 'keeps both scoped label when adding a label with a different key' do
+  select_label_and_refresh 'bar::bah'
+
+  expect(page).to have_content('added bar::bah')
+  expect(page).to have_content('added foo::ba')
+
+  scoped_labels = page.all('.qa-labels-block .scoped-label-wrapper')
+
+  expect(scoped_labels.first).to have_content('bar::bah')
+  expect(scoped_labels.last).to have_content('foo::ba')
+end
+
+def select_label_and_refresh(label)
+  page.find('.block.labels .edit-link').click
+  page.find('.dropdown-menu-labels .dropdown-input-field').send_keys [label, :enter]
+  page.find('#content-body').click
+  page.refresh
+end
+```
+
+By creating a reusable `select_label_and_refresh` method we remove the code duplication, and later we can move this method to a Page Object class that will be created for easier maintenance purposes.
+
+> Notice that the reusable method is created in the bottom of the file. The reason for that is that reading the code should be similar to reading a newspaper, where high-level information is at the top, like the title and summary of the news, while low level, or more specific information, is at the bottom.
+
+### 5. Tests' pre-conditions using resources and Page Objects
+
+In this section, we will address the previously mentioned subject of creating the application state for the tests, using the `before :all` and `before` blocks, together with resources and Page Objects.
+
+#### `before :all`
+
+A pre-condition for the entire test suite is defined in the `before :all` block.
+
+For our test suite example, some things that could happen before the entire test suite starts are:
+
+- The user logging in;
+- A premium license already being set up;
+- A project being created with an issue and labels already setup.
+
+> In case of a test suite with only one `it` block it's ok to use only the `before` block (see below) with all the test's pre-conditions.
+
+#### `before`
+
+A pre-condition for each test case is defined in the `before` block.
+
+For our test cases samples, what we need is that for every test the issue page is opened, and there is only one scoped label applied to it.
+
+#### Implementation
+
+In the following code we will focus on the test suite and the test cases' pre-conditions only:
+
+```ruby
+module QA
+  context 'Plan' do
+    describe 'Editing scoped labels properties on issues' do
+      before :all do
+        project = Resource::Project.fabricate_via_api! do |resource|
+          resource.name = 'scoped-labels-project'
+        end
+
+        @foo_bar_scoped_label = 'foo::bar'
+
+        @issue = Resource::Issue.fabricate_via_api! do |issue|
+          issue.project = project
+          issue.title = 'Issue to test the scoped labels'
+          issue.labels = @foo_bar_scoped_label
+        end
+
+        @labels = ['foo::baz', 'bar::bah']
+        @labels.each do |label|
+          Resource::Label.fabricate_via_api! do |l|
+            l.project = project.id
+            l.title = label
+          end
+        end
+
+        Runtime::Browser.visit(:gitlab, Page::Main::Login)
+        Page::Main::Login.perform(&:sign_in_using_credentials)
+      end
+
+      before do
+        Page::Project::Issue::Show.perform do |issue_page|
+          @issue.visit!
+        end
+      end
+
+      it 'keeps the latest scoped label when adding a label with the same key of an existing one, but with a different value' do
+        ...
+      end
+
+      it 'keeps both scoped labels when adding a label with a different key' do
+        ...
+      end
+
+      def select_label_and_refresh(label)
+        ...
+      end
+    end
+  end
+end
+```
+
+In the `before :all` block we create all the application state needed for the tests to run. We do that by fabricating resources via APIs (`project`, `@issue`, and `@labels`), by using the `Runtime::Browser.visit` method to go to the login page, and by performing a `sign_in_using_credentials` from the `Login` Page Object.
+
+> When creating the resources, notice that when calling the `fabricate_via_api` method, we pass some attribute:values, like `name` for the `project` resoruce, `project`, `title`, and `labels` for the the issue resource, and `project`, and `title` for `label` resources.
+
+> What's important to understand here is that by creating the application state mostly using the public APIs we save a lot of time in the test suite setup stage.
+
+> Soon we will cover the use of the already existing resources' methods and the creation of your own `fabricate_via_api` methods for resources where this is still not available, but first, let's optimize our implementation.
+
+### 6. Optimization
+
+As already mentioned in the [best practices](./BEST_PRACTICES.md) document, end-to-end tests are very costly in terms of execution time, and it's our responsibility as software engineers to ensure that we optimize them as max as possible.
+
+> Differently than unit tests, that exercise every little piece of the application in isolation, usually having only one assertion per test, and being very fast to run, end-to-end tests can have more actions and assertions in a single test to help on speeding up the test's feedback since they are much slower when comparing to unit tests.
+
+Some improvements that we could make in our test suite to optimize its time to run are:
+
+1. Having a single test case (an `it` block) that exercise both scenarios to avoid "wasting" time in the tests' pre-conditions, instead of having two different test cases.
+2. Moving all the pre-conditions to the `before` block since there will be only one `it` block.
+3. Making the selection of labels more performant by allowing for the selection of more than one label in the same reusable method.
+
+Let's look at a suggestion that addresses the above points, one by one:
+
+```ruby
+module QA
+    context 'Plan' do
+      describe 'Editing scoped labels properties on issues' do
+        before do
+          project = Resource::Project.fabricate_via_api! do |resource|
+            resource.name = 'scoped-labels-project'
+          end
+
+          @foo_bar_scoped_label = 'foo::bar'
+
+          @issue = Resource::Issue.fabricate_via_api! do |issue|
+            issue.project = project
+            issue.title = 'Issue to test the scoped labels'
+            issue.labels = @foo_bar_scoped_label
+          end
+
+          @labels = ['foo::baz', 'bar::bah']
+          @labels.each do |label|
+            Resource::Label.fabricate_via_api! do |l|
+              l.project = project.id
+              l.title = label
+            end
+          end
+
+          Runtime::Browser.visit(:gitlab, Page::Main::Login)
+          Page::Main::Login.perform(&:sign_in_using_credentials)
+          Page::Project::Issue::Show.perform do |issue_page|
+            @issue.visit!
+          end
+        end
+
+        it 'correctly applies the scoped labels depending if they are from the same or a different scope' do
+          select_labels_and_refresh @labels
+
+          scoped_labels = page.all('.qa-labels-block .scoped-label-wrapper')
+
+          expect(page).to have_content("added #{@foo_bar_scoped_label}")
+          expect(page).to have_content("added #{@labels[1]} #{@labels[0]} labels and removed #{@foo_bar_scoped_label}")
+          expect(scoped_labels.count).to eq(2)
+          expect(scoped_labels.first).to have_content(@labels[1])
+          expect(scoped_labels.last).to have_content(@labels[0])
+        end
+
+        def select_labels_and_refresh(labels)
+          find('.block.labels .edit-link').click
+          labels.each do |label|
+            find('.dropdown-menu-labels .dropdown-input-field').send_keys [label, :enter]
+          end
+          find('#content-body').click
+          refresh
+        end
+      end
+    end
+  end
+```
+
+As you can see, now all the pre-conditions from the `before :all` block were moved to the `before` block, addressing point 2.
+
+To address point 1, we changed the test implementation from two `it` blocks into a single one that exercises both scenarios. Now the new test description is: `'correctly applies the scoped labels depending if they are from the same or a different scope'`. It's a long description, but it describes well what the test does.
+
+> Notice that the implementation of the new and unique `it` block had to change a little bit. Below we describe in details what it does.
+
+1. At the same time, it selects two scoped labels, one from the same scope of the one already applied in the issue during the setup phase (in the `before` block), and another one from a different scope.
+2. It runs the assertions that the labels where correctly added and removed; that only two labels are applied; and that those are the correct ones, and that they are shown in the right order.
+
+Finally, the `select_label_and_refresh` method is changed to `select_labels_and_refresh`, which accepts an array of labels instead of a single label, and it iterates on them for faster label selection (this is what is used in step 1 explained above.)
+
+### 7. Resources
+
+TBD.
+
+### 8. Page Objects
+
+> Page Objects are auto-loaded in the `qa/qa.rb` file and available in all the  test files (`*_spec.rb`).
+
+Page Objects are used in end-to-end tests for maintenance reasons, where page's elements and methods are defined to be reused in any test.
+
+Take a look at [this document that specifically details the usage of Page Objects](./qa/page/README.md).
+
+Now, let's go back to our examples.
+
+...
+
+#### Adding testability
+
+TBD.
\ No newline at end of file