Golden testing for some requests

[chelmertz]

The logic is getting a bit hairy to properly handle giving out points/dealing with comments etc., trying to keep it all in one's head.

Maybe use json tags for the specific fields we need to shred.

[chelmertz]

A good flow, for point related bugs, could be:

• There's a PR shown in the GUI that is either above 0 points but uninteresting, or it has negative points, but is interesting
• Click somewhere in the GUI to turn that part of the graphql response into a test
  • When I develop, I start the elly process in the foreground. Maybe only render this button (and only listen to a specific endpoint) when a -golden flag is provided
• Either turn the response into a unit test
  • "Zoom in" by changing the expected outcome
• Or, prepared with a fuzzying boilerplate setup, seeded with example data from the graphql response
  • Having properties (mixin PBT and fuzzying here) is the end goal. Based on re-reading the existing fuzzy-tests, it might be hard to create a relevant test based on a single bad PR point assignment.
  • Let's start with unit tests.

[chelmertz]

Need to either:

• Persist the raw (or deserialized) version of a PR, so that we can use it for tests
  • Wasteful and possibly impacting the storage layer 👎
• Or: hit the Github API again, either using the big graphql query or a custom smaller one, for less impact
  • Using the real one diverges as little as possible 👍