Assembly wide multicast attributes. Are they evil?

Question

I am working on a project where we have several attributes in AssemblyInfo.cs, that are being multicast to methods of a particular class.

[assembly: Repeatable(
AspectPriority = 2,
AttributeTargetAssemblies = "MyNamespace",
AttributeTargetTypes = "MyNamespace.MyClass", 
AttributeTargetMemberAttributes = MulticastAttributes.Public,
AttributeTargetMembers = "*Impl", Prefix = "Cls")]

What I don't like about this, is that it puts a piece of logic into AssemblyInfo (Info, mind you!), which for starters should not contain any logic at all. The worst part of it, is that the actual MyClass.cs does not have the attribute anywhere in the file, and it is completely unclear that methods of this class might have them. From my perspective it greatly hurts readability of the code (not to mention that overuse of PostSharp can make debugging a nightmare).Especially when you have multiple multicast attributes.

What is the best practice here? Is anyone out there is using PostSharp attributes like this?

Answer 1

I'm sure this will be an unpopular answer but maybe I can get my peer pressure badge...

Your instincts are correct. Putting logic in metadata of any kind is a horrible, horrible sin for which one burns eternally in the hellfire of unmaintainability.

I mean no disrespect by this although I'm certain it will be interpreted otherwise.

The best practice would be to not use "aspect-oreinted programming" tools, which are crutches that enable the lameness of poor design and testing practices. Instead, look at your design and ask yourself "why."

Why did I feel the need to use this tool? What design problem was I trying to solve?

Once you have a firm grasp of the problem, go pick up Design Patterns Explained (Shalloway & Trott) or Head First Design Patterns (Freeman, Robson, Bates, & Sierra).

In the end, a pattern-oriented solution will be easier to understand, easier to test, and easier to change. The only additional cost will be the one-time fee of mastering design patterns in place of the recurring charge of trying to figure out where all these aspects are, how they fit together, and how they influence one another every time you make a change.

Answer 2

Let me first answer to Max: indeed, aspects are not an alternative to good OOP patterns. They are a complement. Any good AOP design starts with a good OOP design. But OOP patterns sometimes force you to write a lot of plumbing code manually. For these cases, aspects can be used to automate the implementation of OOP pattern, not to replace them.

When you use AOP intelligently, your solution can become easier to understand (business code is not mixed with maintenance code), to test (you can test the aspect independently from business code, i.e. you don't have to test that any business method traces properly), change (you just have to change the aspect when you want to change the pattern, instead of changing every implementation of the pattern). Now, if you abuse from AOP, if you use it as a hacking tool, if you do not think in terms of OOP patterns before, then your're going to get more costs than benefits from AOP. As any sharp tool, AOP should be used intelligently.

Back to the original question.

Who tells you should put aspects in AssemblyInfo.cs? You could create a new file called GlobalAspects.cs and put all assembly-level aspects there. You're right that AssemblyInfo.cs should just be for assembly-level metadata.

But like you, I don't like assembly-level aspects. I think there should be avoided. The principal problem with assemly-level aspects is that they rely on naming conventions, and this is evil. (This evil is called pointcut fragility in the academic AOSD community.) Indeed, when you rename a class or namespace, you change the set of methods to which the aspect applies, and this can quickly become a nightmare. That's why I never use aspects based on naming conventions for myself.

What about code readibility? To a great extent, I think readable code is short code. If I have a business method called CreateProduct, I probably want to see just the code creating the product. Most of the time, I am not interested in code that handles transactions, exceptions, or tracing. It's enough if I know that some aspects handle that for me.

And how do I know? With PostSharp, you have the Visual Studio Extension. With AspectJ, you have the AspectJ plug-in for Eclipse (AJDT). They show you, inside the IDE, which aspects are applied to the code you currently see. And if you really want to see details (but you seldom really want), you can use the debugger to step into aspects, or use Reflector to see produced code.

Summary:

1. Good AOP design always starts with a good OOP design.
2. Avoid relying on naming conventions to apply aspects.
3. Use PostSharp extension for Visual Studio or AJDT to visualize aspects in your code.