specialist-publisher: # Automated tests
We wrote a lot of automated tests while migrating Specialist Publisher to the new platform. The general approach for these is:
- Log in as an authenticated user
- Stub API calls
- Interact with the app
- Assert that the right API calls are made
- Assert that the right elements appear on the page
We found that we struggled to test user journeys. Normally, in a Rails app, you'd mutate data in the database as you interact with the application. In our case, all of the data for the app is persisted externally. By stubbing the data persistance layer, we found we struggled to test user-journeys through the app.
To this day, this is something we haven't figured out completely. Instead, we took a (lean?) approach and decided to plug this hole with a QA process instead that would essentially test these user journeys before rolling out new formats to production. In retrospect, this has become a time-consuming, onerous process and we'd have probably taken a different approach from the beginning given the luxury of some time to work on a better solution.
There is an RFC about this here.
Here is a specific example of where we struggled with stubs.
In this example, we are testing the behaviour on publish. We are having to resort to stubbing a sequence of responses from the Publishing API so that the stubs are representative of what happens when you actually publish something, i.e. its state changes to 'published' and it gets a 'first_published_at' timestamp.
This is far from intuitive and it is difficult to understand what the test is
doing. We had to add the
stub_publishing_api_has_item_in_sequence helper to
specifically address the need to stub a sequence of responses rather than a
single canned response.