There is an ongoing discussion about TDD and Rails. It was recently heated by some some of the DHH statements in his RailsConf keynote and in the blog post: http://david.heinemeierhansson.com/2014/tdd-is-dead-long-live-testing.html

One aspect of this discussion is the confusion about - What makes a good Unit?

In the Rails community, I've seen people overusing mocks and stubs - you can detect it by looking at all the "should_receive" calls in the codebase. They're not always bad, but they might be potential code smells.

The reason we use should_receive is in a way to draw the boundary between what's important to test right now and what's outside of the test scope.

Unit example

Let's take an example - you've got an Order, which can have many OrderLines, a ShippingAddress and a Customer.

Do we have 4 units here, representing each class? It depends, but most likely it may be easier to treat the whole thing as a Unit. You can write a test which test the whole thing through the Order object. The test is never aware of the existence of the ShippingAddress. It's an internal implementation detail of the Order unit.

A class doesn't usually make a good Unit, it's usually a collection of classes that is interesting.

This way of defining a Unit gives you tests, that don't need to change whenever the internals change - that's a good thing. You don't want to change tests on every refactoring - in fact it's a smell.

Unit-based architectures
Some time ago, I wrote a surprisingly popular post: The four architectures that will inspire your programming in which I listed:

• Hexagonal Architecture
• Clean Architecture
• DDD
• DCI