Avoiding File I/O In Unit Tests

Noel wrote a series of articles about TDD in game development here Games from Within: Stepping Through the Looking Glass: Test-Driven Game Development (Part 3).He mentions that the tests should run fast and that means no file I/O. However, sometimes…

Test Driven Development: First Impressions

Can’t speak for Noel here, obviously – but in my experience, 3D code is tested after the fact. It’s hard math, and that can’t easily be *driven* by tests. I’m not certain it’s entirely true, but I’ve got the feeling that the softer the rules, the easier it is to drive your development by tests.

That doesn’t mean you *don’t* test it – it’s just that you write the code, manually verify correctness, and then fix the unit test. Still gives you coverage – it just doesn’t help you as much in the design department.

I’ve done it for a cross-platform particle engine, and it was *immensely* useful. No, it’s not a full 3D engine – but the same ideas would apply.

“But if you want to insert a mock object there (which you probably will, to avoid using a real renderer in your tests) [...]”

…but what if the code being tested *is* the renderer?

I’ve been thinking about TDD for a while but it’s *hard* to apply this to 3D graphics. Tests will be long and complicated, and there is no clear way to validate the results (comparing images doesn’t always work, for a variety of reasons).

How are you dealing with this? Are the 3D guys skipping the TDD?

“But if you want to insert a mock object there (which you probably will, to avoid using a real renderer in your tests) [...]”

…but what if the code being tested *is* the renderer?

I’ve been thinking about TDD for a while but it’s *hard* to apply this to 3D graphics. Tests will be long and complicated, and there is no clear way to validate the results (comparing images doesn’t always work, for a variety of reasons).

How are you dealing with this? Are the 3D guys skipping the TDD?

“They should probably be relatively simple, so writing unit tests for them is not all that important.” Deep sigh…

My thinking is that if a private method is so complex it needs testing, it’s probably not primitive (in the Lakos sense), and should be extracted into a utility class, or similar.

Does that resonate with the feelings of anyone else?

- Kim