What is the difference between the classical (Detroit) and mockist (London) styles of unit testing? Which would you use for NestJS services?

The two styles differ in what a "unit" is and in what the test checks. For NestJS services I lean towards the classical style.

Classical (Detroit) style: a unit is a behaviour. One behaviour may use several real objects.

• You use real collaborators (the other objects your code works with) when they are fast and deterministic. Deterministic means they give the same result every time.
• You replace only the awkward ones, such as the database, the clock or an external API.
• Tests check state: the returned value, or what ended up in a fake repository.

Mockist (London) style: a unit is one class, and every collaborator is a mock.

• Tests check interactions: which methods were called, with which arguments.
• It supports outside-in design, where you start from the outer layer and design inwards.
• But it ties tests to how the class talks to its collaborators. So refactoring breaks the tests, even when the behaviour stays the same.

For Nest services I use:

• real domain helpers;
• test doubles only at process boundaries, such as repositories, HTTP clients and queues;
• assertions on outcomes.

I check an interaction only when the interaction is the outcome, for example "an email was sent".