A smoke test
A single test

This weekend I was reminded how complex (and in some cases, convoluted) typical test setup is for Rails applications. The steps below were all necessary in my case, but please don't read into this list any criticisms of these tools' excellent open source maintainers. In this case, Jonas Nicklas, Hiro Asari (of Travis) & Mike Moore all took time out of the holiday weekend to help unblock me.

These are the things I either knew in advance or discovered in the process of setting up the solitary test you see running in the GIF above:

  1. For a new Rails project, the "minitest" gem is included even though it's absent from the generated Gemfile.
  2. In spite of minitest's aforementioned inclusion, the addition of the minitest-rails gem is apparently necessary to avoid some tests from raising errors
  3. The minitest-rails gem must be added to the :development group (in addition to :test) of the Gemfile, so that its rake tasks are available from the command line. That means that running an app in development unnecessarily pollutes it with all the class loading and monkey-patching of its test-scope dependencies
  4. Outside of Rails, Minitest::Spec is the test class used for describe...it style tests. In a Rails app, however, each test is an instance of ActiveSupport::TestCase with the Minitest::Spec::DSL merely mixed in (this required an update to minitest-given)
  5. If you don't add require "rails/test_unit/railtie" to your application.rb, then the server started by rake test will actually be initialized in development mode, even if your test helper explicitly sets ENV['RAILS_ENV'] = "test" (further blurring the line between production-scope and test-scope dependencies)
  6. Before any tests are defined (e.g. in a test/test_helper.rb file), one must first require File.expand_path("../../config/environment", __FILE__) and then require "rails/test_help" or errors will be raised
  7. To run full-stack tests with browser automation, one might add the minitest-rails-capybara gem and then must explicitly require "minitest/rails/capybara" (and to know that this both installs and require's Capybara itself)
  8. Rack::Test is used by Capybara's default driver, but it can't make web requests outside the Rack app (e.g. Github OAuth, Harvest integration). In this case, Selenium was added by way of the selenium-webdriver gem and the setting Capybara.default_driver = :selenium
  9. Without Firefox installed, one might need the chromedriver bin (via brew install chromedriver), as well as to override the Capybara driver registration with Capybara.register_driver :selenium { |app| Capybara::Selenium::Driver.new(app, :browser => :chrome) }
  10. minitest-rails-capybara expects its test files to be placed in test/features
  11. When using Capybara, the Minitest::Spec DSL changes from describe/it to feature/scenario, unless you use describe 'foo', :capybara do
  12. In order to extract one's use of Capybara's API into other modules or classes, they must extend or include the Capybara::DSL
  13. The test database in such an arrangement is, by default, persisted between tests and test runs; wiping and re-seeding the database must be handled elsewise (perhaps one might Rake::Task["db:setup"].invoke in a before hook)
  14. To enable continuous integration ("CI"), one might go to Travis, login, click "Accounts", click the repo owner, manually sync repos, then click "on" for the repo under test
  15. Rails tests will require non-default CI settings, necessitating a .travis.yml file with language: ruby, and a before_script entry of RAILS_ENV=test bundle exec rake db:setup
  16. Any secure environment variables needed by the application must be added with the travis CLI (acquired by gem install travis), then created with travis encrypt SOMEVAR="some val" --add env.global and committed
  17. To run browser tests that execute JavaScript in a headless CI environment, one might use the poltergeist gem, which implements a PhantomJS-based driver
  18. To juggle the two drivers, an environment variable might flag their use (Capybara.default_driver = ENV['HEADLESS'] ? :poltergeist : :selenium), with - HEADLESS=true in the env.global key of the .travis.yml
  19. To speed up CI, one should add cache: bundler to the .travis.yml file [Note: at this time I can't seem to get caching to work, despite this]
  20. Github OAuth applications require a static port in the OAuth callback URL, meaning the test must fix its port to a known value via Capybara.server_port = ENV['SOME_SET_PORT']
  21. Capybara makes web requests to 127.0.0.1, and any request to localhost will spawn a separate browser session. As a result, the OAuth success callback won't be able to be verified unless the application's callback URL is set explicitly to 127.0.0.1
  22. Github's login page throws uncaught JavaScript errors, so Poltergeist's Capybara driver must be overridden so as to specify that it should not re-raise any JS errors via Capybara.register_driver :poltergeist { |app| Capybara::Poltergeist::Driver.new(app, :js_errors => false) }

As for the task of actually writing any tests, I leave that as an exercise to the reader.

I don't have a solution in mind as to how to make this setup process simpler (and surely it would be, if only I'd acquiesce a bit more to the Omakase test suites), but it certainly feels like the requisite work to get up-and-running with full-stack testing presents too high a barrier of entry, especially given that Ruby's testing tools are, generally, the good ones. Also, while I typically use RSpec, I don't believe much, if any, of the work above would have been obviated by it (or Cucumber, for that matter).

If you enjoy this post, let us know by twitter or e-mail! If you'd like to discuss it, open an issue on our feedback repo! Or share it on HN!