The new beta of VisualStudio 2005 has a built-in testing framework similar to NUnit. It also has a feature that automatically generates test stubs for you based on the code you've written. It creates a test for each method and fills in code to create the necessary objects, call the method, and check the results. In most cases, it doesn't really know what to fill in, so it just puts stub values in for the developer to replace later.

Although this may seem like a very beneficial feature, I fear that it will lead developers back to bad habits. Any time a tool makes it easy to do something the wrong way, people will follow the easy route and do it the wrong way. In this case, the problem is that easiest way to create your testcases is to wait until your class is finished then have the tool generate the tests for you. This is the exact opposite of Test First Design. It's actually Test Last Design. Writing the tests is the very last thing you do.

In fact, if I create tests for a class then add an additional method and try to create a test for the new method, I get a critical error - it doesn't let me do it. This may just be a problem with the beta (I'll be reporting it), but I don't believe it's intended to work this way.

The final problem I have with automatically generating testcases is that the testscases tend to test the wrong things. Developers will start to think that you have one testcase per target method and that will adequately test the method. They won't consider exceptional values, boundary conditions, etc. If you are given the method stubs and told to just modify them to turn them into the proper tests, you aren't in the right mindset for thinking - hey, I need a whole new test to check the boundary conditions.

Generating tests from code leads to poor practices. I suppose it's better than not having tests at all, but it leads developers in the wrong direction. This is a feature I expect to see badly abused in the future.