When an engineer makes a change to source code or tests, there are multiple steps before that change finally makes it into production. In the most basic form, these steps can merely consist of a developer compiling code on their development machine and copying the output to the production environment. Don't do this. Just don't.

This introduces innumerable opportunities for human error. In order to truly provide reliable deliveries, this process should be automated from end to end. To take it a step further, multiple stages of testing should be incorporated, along the way, to ensure the quality of the release.

Figure 1: Delivery Pipeline

To begin eliminating human error, some automated workflow needs to be put in place. This consists of a gated build, a continuous integration build, and one or more release definitions. Each build provides another layer of confidence in the quality of the delivery.

It all starts when a software engineer or quality engineer makes a change to source or test code. In order to ensure that the change passes basic inspection, we set up a gated build. The gated build is triggered automatically when the developer attempts to check in a change. It builds the code and executes the unit tests. Only if the code compiles and all unit tests pass, will the change actually be committed to source control.

Figure 2: Gated Build

This ensures the health of the code-base in two ways. First, the change actually compiles, thereby preventing blockers for other developers. It is very costly for development to halt, because somebody checked in something that doesn't compile. Second the unit tests ensure that each individual part of the code works as designed. This is the first layer of protection from costly regression errors.

Once the change hits source control, another process is set into motion. The change is retrieved from source control and built, similar to the gated build. This time, however, instead of running unit tests, the build executes functional, integration, and system tests.

Figure 3: Continuous Integration Build

Each type of testing provides an additional level of confidence in the final product. Functional testing verifies that the individual components behaves according to the requirements. Integration testing ensures that the components work together, and finally, system testing validates the entire system as a whole.

This type of build is known as a continuous integration build, because every change is automatically integrated into the system. Then it is tested against the complete system.

If any test fails along the way, then the entire build fails. It's only if all tests pass that the build artifacts get turned into a release. The entire process is depicted in the following flowchart.

Figure 4: Delivery Workflow

Each step along the way can be automated. For safety and accountability, it's a good idea to add some human control along the way. This is provided at two stops in the chain. The first is creating the release from the build artifacts. Just because you check something in doesn't mean you want to turn it into a release. So this part is left up to a human.

Second, after creating a release, it is up to the humans to decide what to do with it. In an ideal situation, there should be at least three release environments: quality assurance (QA), pre-production, and production. Once a release has been made, it can be deployed to any of these environments.

Figure 5: End-to-End Delivery Pipeline

The release would first be deployed to a QA environment, where it would be tested throughout development. When the change is ready to be hardened for an upcoming delivery, the same release can then be deployed to pre-production. Once the team and client are confident in the final quality of the release, it can be deployed to production.

Automating the delivery pipeline takes the guesswork out of providing quality releases. It frees the engineers to focus their time and energy where it matters, rather than on the overhead of integrating changes and running tests manually. If you want to take your process to the next level, automating your delivery pipeline is an investment that will pay for itself.

Do you have your own stories or ideas about automating the delivery pipeline? I'd love to hear from you in the comments!

Mocking is an essential component to any good unit testing strategy. It allows us to replace the concrete implementation of a class with a test version whose behavior you control. When designing a system from scratch, we can use dependency inversion to make sure everything is mockable. However, when working with an existing system, there may be hidden dependencies that require some creative thinking.

I recently came across a situation where I needed to unit test some code that had dependencies on Microsoft's OWIN extension methods. The methods in question belonged to the static class OwinContextExtensions.

OwinContextExtensions

I was using Moq as my mocking framework, which doesn't provide a mechanism for mocking static extension methods.

To overcome this hurdle, I created a wrapper class which exposed the same methods, but as instance methods. I then extracted an interface from my new wrapper class, so that I could mock it up. Here is the new interface for the wrapper class. Notice that it has the same methods as the original extension class, but as regular methods.

IOwinContextExtensionsWrapper

Finally, here's the wrapper class itself. As you can see, it still makes use of the original extension methods. However, this implementation is abstracted from the user, in this case the unit test.

OwinContextExtensionsWrapper

To bring it all together, once the wrapper class and interface were in place, it was time to wire everything up. First I had to create a number of mocks on which there were dependencies. In this case, the wrapper is provided to the controller by means of a property, which it later uses to get the UserManager.

In conclusion, if you encounter static extension methods when you're unit testing, don't panic. Just whip out a wrapper class and interface to break that hard dependency. This is just one example of overcoming a dependency in existing code. However, when designing for new functionality, always try to think of how you're going to mock the classes you create.