Monday, January 08, 2007

Castle project - MicroKernel

In the "Team Work” post I have been talking about an architecture based on contract, in other words – an interface. The presenter for example was using the ILocalDataAccessService and the IOnlineService services.
What happens if the current presenter needs another service? We should change the presenter constructor, and all initialization. We could have the same problem as the one I talked about in the previous post, with the Object Mother – we could have to change the code in a lot of places and maybe hundreds of tests. This is one of the many problems which could appear because of a tight coupled architecture. The dependency using constructor injection isn’t enough sometimes, for example a presenter can use a service only in a specific moment, or it can use a service which isn’t available all the time, so the service shouldn’t necessarily be given as a constructor parameter.
The ideal way is to have an infrastructure, a framework which solves all these problems automatically for us, so we shouldn’t worry about a complex service dependency, or service creation.

The first idea is to have a collection of services, which can be retrieved at any time using their names, and/or interface they implement. These would solve our problem with the service we don’t use all the time, or they could be unavailable at a specific time.

IService service = (IService) Kernel[“My service”];

The second idea is to automate the construction of services, if a service A needs to be constructed and it depends on another service - B, the framework should automatically retrieve the service B from the collection and construct the service A. So the services should be instantiated automatically without having to call explicitly the constructors.
Kernel.AddService(“Service A”, typeof(AService));
Kernel.AddService(“Service B”, typeof(BService));

If the BService class has a constructor with a AService parameter, the framework should automatically call the constructor with the service “A” added in the first step.

Another problem would be the service “style”: is the service a singleton? Or should a different instance be created for every call of Kernel[“My Service”] ? Or maybe create a service for each thread? These details should also be handled by the framework.

So, we shouldn’t be worried when changing the constructor of a service. The infrastructure should have to worry now. We could also use this framework for a plugin architecture: until now, I have worked at a few applications which used plug-ins, and most of the time, a plugin defined in fact a service. I had to do every time my own manager for these plugins. Some plugins were singletons, some would had to be instantiated for every call, but I had to think how to do all this by myself. I didn’t use a framework which could do all the work for me.

Some days ago I have searched for such a framework and I found exactly what I was looking for. It is called the Castle Project – MicroKernel. I started using this framework, and – until now, I am very pleased with the results.

Thursday, January 04, 2007

TDD problems - Object Mother

I started working on a new project some time ago, and decided to work in a TDD style as much as I could. The idea was that there were very vague specifications for the project, resulting in a very “agile” project. So what methodology other than XP Programming and TDD should I adopt, I thought.(If you don't know about TDD (Test Driven Development) read about it on wiki.)

I have had some experience with TDD before, but I usually stopped writing tests when I started to develop the user interface. Although I used the MVP design pattern, so I should have been able to test the presenter using a mock as a view, I think this is a very time consuming process, and sometimes maybe not very useful, especially when the presenter logic is not very complex – it takes data from the object and passes it to the view.

The project I was talking about is a monitoring engine. Used to monitor computers, software applications and other stuff… The business was pretty complex, and there was no user interface, so the TDD was almost a must. I started to work in almost in a perfect TDD style, but after some time I found it very difficult to maintain the unit tests.

The problem was that a small change in the business code determined a lot of changes in the unit tests. It was pretty upsetting when seeing 100 compile errors when changing just the number of parameters of a constructor. And very time consuming. At some point, 50% of my work consisted in maintaining the unit tests. Most of my tests were constructing business objects in order to test them. A small change in the constructor or in the manner the business object was constructed determined as I said a lot of errors in the tests. So my idea was to separate the object construction from the unit tests, and make some kind of object factory, but a smart factory, which could create very complex objects with just a function call, or let me customize the object creation.

I found that this method is in fact a pattern, called “Object Mother”:

“Through a handful of simple method calls, this utility provided a complete, valid, and customizable structure of business objects (think of an invoice: its lines, all related charges, remit to, bill to).

Object Mother starts with the factory pattern, by delivering prefabricated test-ready objects via a simple method call. It moves beyond the realm of the factory by

  1. Facilitating the customization of created objects,
  2. Providing methods to update the objects during the tests, and
  3. If necessary, deleting the object from the database at the completion of the test.

Now, when I change the number of parameters in a constructor, I have to change only the object mother not all my 1000 test units, so I consider this method very helpful.