David Eisinger

Notes > Good Tests

(Notes for a Viget article)

• Most importantly: give you confidence to make changes
  • This gets more and more important over time
• Secondarily:
  • Tells you it works during development
  • Help your code reviewers
  • Serves as a kind of documentation (though not a very concise one)
• Focus on two kinds of tests: unit and integration
  • Unit: test your objects/functions directly
  • Integration: simulated browser interactions
  • If you’re building an API, you might also have request specs
    • But ideally you’re testing the full integration of UI + API
• Unit tests
  • Put complex logic into easily testable objects/functions
  • This is where TDD can come into play
  • Avoid over-stubbing/mocking – what are you even testing
    • It is OK to go down the stack in your unit tests
• Integration tests
  • You need proper end-to-end testing
  • Set up your data (fresh per test)
  • Visit a page
  • Interact with it
  • Make assertions about the results
  • Generally folder per controller, file per action (e.g. spec/features/posts/create_spec.rb)
• Coverage
  • We shoot for 100% in SimpleCov (So all the Ruby is tested)
  • Some consider this too high or too burdensome – I don’t
  • If it’s 100%, you instantly know if you have any untested code
    • If it’s, say, 94%, and you add 100 lines, six of those can be untested – hope they’re perfect!
    • In other words, at less than 100% coverage, you don’t know if your new feature is fully covered or not
  • Occasionally you have to ignore some code – e.g. something that only runs in production
  • It’s OK if you’re not at 100% right now – set the threshold to your current level, and increase it as you add tests and new well-tested features
  • Already covered here
• Third-party/network calls
  • Major libraries often have mock services (e.g. stripe-mock)
  • VCR is … OK but can become a maintenance problem
    • Blocking access to the web is good though – webmock
  • A better approach
    • Move your integration code into a module
    • Create a second stub module with the same API
    • Use JSON Schema to ensure stub stays in sync (i.e. both the real client and the stub client validate against the schema)
    • This will lead to more reliable tests and also more robust code
• Flaky tests are bad
  • They eat up a lot of development time (esp. as build times increase)
  • Try to stay on top of them and squash them as they arise
  • Some frameworks have retry options/libraries that can help (bandage not cure)
    • rspec-retry
  • In general, though, flaky tests suck and generally indicate lack of quality with either your code or your tools
    • So write better code or pick better tools
• Tests are code, but they’re not application code
  • And the way you approach them should be slightly different
  • Some (or even a lot of) repetition is OK; don’t be too quick to refactor
  • Ideally someone can get a sense of what a test is doing by looking at a single screen of code
  • As opposed to jumping around between early setup, shared examples, complex factories w/ side-effects, etc.
  • Think of it as half programming, half writing