PowerShell, Tests and CI/D

Maybe you have an ASP.NET web site, and maybe you have a collection of PowerShell modules used for various automation tasks, like moving files between environments, kicking off batch processes, moving assets to the your CDN or similar things. If so these bits of PowerShell contain important bits of logic; if they don't work your site doesn't work properly.

Naturally you want to apply the same discipline to the PowerShell code as to any other production code, which includes TDD and Continuous Integration/Delivery/Deployment (CI/D). Furthermore I'll assume that you already have tests around your C# code, and that your CI/D setup runs these tests and reports the result.

In this situation you can TDD by writing test functions in the PowerShell modules themselves and call them from the command line. This, of course, becomes tiresome quickly as the number of tests rise, so you'll soon find yourself  writing ad hoc test runners for each PowerShell module. This works. Sort of. But it's still tiresome to run the tests selectively, and just as importantly, this doesn't fit into your CI/D setup which is geared towards your xUnit (or similar) test suites.

One quick and simple way to get around this issue is to write the tests in C# just as any other tests and simply call the PowerShell from there. Like so:



This isn't perfect:

• It's somewhat slow. 
• The call out to the PowerShell under test is tugged away inside a string.Format call
• The tests and the code under test are written in different languages, which doesn't sit well with me. It introduces an odd asymmetry and it creates an artificial barrier between the tests and the code under test.

The big win on the other hand is that it fits right into your existing TDD and CI/D setup.

Hello Nancy World in F#

As you may noticed (if you follow this blog or go to the same conferences as me) I really like the Nancy web framework. So naturally, since I've decided to finally learn F# properly, I threw together a hello world Nancy app in F#. It's very simple; the complete code is listed below. All I did was create an F# console application, install the Nancy and Nancy.Hosting.Self NuGet packages and type in this code, which will return the string "Hello" if you point your browser to "http://localhost:8888/nancy/":



So what's going in that code? Fist thing to notice is the type HelloModule which inherits NancyModule. On start up Nancy will pick up all NancyModules and run their constructor. In it's constructor HelloModule sets up a single route, "/", and tells Nancy what to do on a HTTP GET to that route, namely to execute a lambda that always returns "Hello". Lastly, since this is self hosted, there is a bit of setup in the "main" function: A Nancy host is instantiated, told where to listen for requests and started. This will make the Nancy self host start up Nancy and start listening for requests. That's it. Short and sweet.

More OO Around That Island of Functional

In my last post I showed how C# seamlesly supports having little functional islands inside an otherwise OO code base. I did that using a simple object model, in which one of the objects was implemented in a functional manner. The other objects, though, were just data containers, really.In that post I made the off hand remark that these could be better encapsulated. In this post I'll address this, and also show that the "other" objects could easily have more behavior as well. All-in-all this brings the example closer to the original intention: To show how the overall design can be OO, but the insides of individual classes can be functional.

The classes involved were CourseTemplate, Course, Registrant and CoursePlan. The example is to create a course plan, where registrants are placed in in courses bsed on which course they wish to take.

Slightly Better Encapsulation

To get to a more object oriented design the first step is to move away from simple auto properties with public getters and public setters to a least have private setters. Using CourseTemplate as an example this simply means moving to this:

Adding Basic Domain Operations

Nest step is to start adding basic domain operations that make the objects have just a little bit of actual behavior and some more encapsulation. To illustrate lets add a method for wishing courses to the Registrant class, and lets also add an override of equals that makes more sense from a domain perspective:

Adding other Behaviors

We're getting the hang of this. Let's add the same type of basic domain methods along with a few other (made up) methods to the Course class:

By adding Register, IsFull and IsEmpty some responsibility is being moved from the CoursePlan version in the last post to the Course - which makes much more sense. By adding Start and AwardDiplomas I point toward the fact that these classes can have more behavior if we need it.

The Functional Island is Still Functional

Despite these changes the CoursePlan doesn't change all that much. The places the implementation has changed is where the new methods on Course are used. The implementation is still just as functional:

Conclusion

The functional islands I talked about in the last post can easily exist in a truly object oriented code base.