Wednesday, February 04, 2015

From "Golden Master" to Unit Tests

So I'm currently working on some code which has been written but needs refactoring to add some additional business requirements - there are very few tests.  So I started by wrapping the code that I want to work with a "golden master"/characterisation test below is the pseudo code for this:

[Test, Ignore (Run this once at the begining)]
void GenerateGoldenMaster()
{
Call service
Serialize result to xml
Save to file
}

[Test]
void CompareGoldenMaster()
{
Call service
Serialize result to xml
Save file
Compare with master & assert correctness
}

I save the file so that I can see the results side by side - sometimes this is helpful rather than assertion failure in the test runner.

If you've never heard of the "golden master"/characterisation test then it's a testing technique to ensure that as you refactor some legacy code the output is the same.  Anyhow, as I started the refactor the code to add the new functionality, I noticed a couple of bugs.  Now while you've got the "golden master" I would advise against fixing these bugs until a little later in the refatoring stage, I prefer to highlight and fix these bugs by using tests.  This is where you should be going - as I don't tend to like to run comparison tests in a CI environment.

A couple of things happened while running through the refactoring - the "golden master" didn't have enough test combinations in.  I therefore had to reconfigure the data in the database to give this additional test combinations.  I only did this once I had everything passing with the original "golden master" that I had created.  This gave me the confidence that the changes to the data would generate the necessary failing assertion.

Once I'd got to a point where I was happy that "most" of the combinations had been covered.  I then started to add the tests which will run in the CI environment.  Essentially, as this is a service with a db, I start by creating an in memory database, adding some data and asserting that the service response contains these bits of data.  I also use random data (guids) generated each test to add some testing invariance...

Now is when I want to highlight the bugs that I've found before as I now feel confident that the current refactored code is the same as it was before, but now I can add the tests to fix these bugs.  Insert the data as required then check that the output is incorrect - this will obviously create a red test which can now be fixed with confidence.

As the tests grow the "golden master" can be removed and deleted from the code base.

I've used the "golden master" a number of times within production code with no tests, if you find yourself in this position, then see if you can wrap the original code with a "golden master" and then start refactoring.  I think you'll be surprised how effect they are...

No comments: