Edit on GitHub
Test guidelines for every pull request
We write tests to be confident that our product works as expected when used by our customers. As developers, we write tests as a gift to our future selves or to be confident that changes will not cause regressions or unintended behaviors.
We value test-writing as much as a feature or a bug fix, and it’s integrated in our core development workflows. It’s not an afterthought and definitely not a follow-up action.
This document was written to stress the importance of tests for every pull request being submitted. This is not to meet higher code coverage but rather to write effective and well-planned use cases depending on the changes being made.
This document aims to serve as the foundation of our test guideline, and serves as a reference on why we don’t merge code without tests. But of course, there’s always an exception. If it isn’t possible for some reason, let the reviewers know by writing a description to start a discussion and to fully understand the situation you are facing.
- Unit test - verify that individual, isolated parts work as expected.
- Integration test - verify that several units work together in harmony.
- End-to-end test - exercises most of the parts of a large application.
Not all test types are required in a single pull request. Only write whichever test types are most effective and appropriate.
In general, when to have these tests?
- Unexported functions or methods, which have low or no test coverage from the parent exported function/method, that affects critical functionality or behavior of the application.
- New features and bug fixes, especially customer and community bugs.
And when is it fine not to have these tests?
- For implementation details for standard libraries or external packages that are implicitly covered by the standard library or external package itself.
- May require external services running to effectively test the functionality, such as dependencies on feature flags via https://split.io, OAuth with third-party providers like Google, etc.
- Tests should be made at the most effective and lowest possible level, but if it requires too much effort or complicated setup to accomplish at a unit test level, it would be best to skip and assess feasibility on the next level such as integration or end-to-end testing.
- Mocks and test helpers.
- Types only.
- Interfaces only or interfaces to other repos, such as with private Enterprise via “einterfaces”.
- End-to-end tests codebase.
- Automatically generated code for database migrations, store layers, etc.
- External dependencies, modules, imports, or vendors.