Proper strategy for writing gem integration&unit tests with mocking?

I am writing gem that parses a specific format. It is intended to be semi-private gem, not widespread-use gem.

There are three main classes in the gem - Parser, Node and Stack. They are heavily depending on each other. And I wonder, which testing strategy works best in this case:

1. Write a test for each class, mocking/stubbing the other two, and write an integration test, covering real parsing examples?
2. Write a test for each class, using real objects instead of mocks, and write an integration test, covering real parsing examples?
3. Write a test for each class, mocking/stubbing the other two, and write another test for each class, using real objects this time?

My classes are not very big, but I want to learn best practices of code testing, so argument "write as you want, this is not important" is, IMHO, invalid ;-)