NDepend getting my juices flowing

I have been using NDepend to analyse the latest version of .NExtra. The code is spotless (this is where you should detect the irony), but the analysis is highlighting some really interesting design flaws that I will probably fix in the next major version.

First of all, almost all cases where method or type names are too long are where I have created facades or abstractions for native .NET entities, like Membership. Here, I simply add exceptions where NDepend will not raise any warnings, since the whole point with the facades are that they should mirror the native classes, to simplify switching them out with your own implementations, mocks etc.

Second, NDepend warns for naming conventions that I do not agree with, such as that method names should start with m_. Here, I simply remove the rules, since I do not find them valid. However, when I look at it after removing the rules, I could have kept them and renamed them and make them check that my own conventions are followed. I will probably do so later on.

Third, and the thing I learned the most from today, was that I turned out to have circular namespace dependencies, despite that I have put a lot of effort into avoiding circular namespace and assembly dependencies. The cause of the circular dependencies turned out to be between X and X/Abstractions namespaces.

Circular dependency graph

The base and abstraction namespaces depend on eachother

For instance, have a look at NExtra.Geo, which contains geo location utility classes. The namespace contains entities like Position and enums like DistanceUnits, as well as implementations of the interfaces in the Abstractions sub namespace.

Now, what happens is that the interfaces define methods that use and return types and entities in the base namespace. At the same time, the implementations in the base namespace implement the interfaces in the Abstractions namespace. And there you go, circular namespace dependencies.

Now, in this particular case, I do not think that it is too much of a deal, but it highlights something that has annoyed me a bit while working with the .NExtra library.

In the library’s unit tests, the test classes only knows about the interfaces, but the setup method either selects a concrete implementation or a mock for the various interfaces. For an example, look here. This force me to refer both the base namespace as well as the abstractions namespace. Once again, this is not that bad, but it raises the question…

In the next minor version of .NExtra, I will probably get rid of the abstraction namespaces, since they add nothing but complexity. Sure, they tuck away all interfaces, but why should they?