http://ayende.comAyende @ RahienCopyright (C) Ayende Rahien 2004 - 2021 (c) 202660If I took a stab at actually implementing this, i would probably follow the same sort of patterns that the Automatic Transaction Facility in Windsor uses, with the [Transactional]/[Transaction] attributes and the Interceptor to handle the transactions. I would basically do something like create a [NullObject] attribute that i can apply to a property and create an Interceptor that looks for the [NullObject] attribute and handle a null implementation through the interceptor. that being said - i actually have no idea how to implement the interceptor. i'm only talking from the theories in my head of how i understand the interceptors in Windsor to work. :) http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment8http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment8Thu, 28 Jun 2007 13:29:07 GMTIt can't work. There is no general-purpose rule available to a runtime code factory like DynamicProxy for it to understand the contract a given null object should really have. This is particularly true when return value or output params are met. The consumer of the null object might not crash on an NRE when it calls a method on the null object but then it might fail when it consumes the result. Moreover, as others have mentioned here, I think it's the component's responsibility to use a suitable default implementation (such as a null object) and to guard against nulls as needed. For example, in externally visible services, I am very careful to check for null parameter values and throw ArgumentNullException if I know the component can't handle them. If I don't have a suitable default implementation then the dependency becomes a required constructor parameter. http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment7http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment7Thu, 28 Jun 2007 04:53:05 GMTI tend to think that the use of a null object pattern for a dependency should be the decision of the class that holds the reference on the null object rather than the container. If the container is going to set the value of an optional dependency, then why would it set it to the null object implementation rather than a real implementation? I can't really come up with a tangible story that would drive the implementation... but that's probably a limitation of my own rather than a statement about objective reality. That said... it would be kinda cool to have a dynamic null object factory/facility that would allow us to get a null object implementation of an interface that is based on DyanamicProxy http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment6http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment6Thu, 28 Jun 2007 00:53:55 GMTI think that at least part of the difference is that scenarios like the NullLogger is not something that I would use an optional dependency for. I would inject a null logger in the ctor and be done with it. My use of optional dependencies is usually something that is truly optional, stuff like INHibernateInitializationAware, which lets you register into the container and affect other stuff. http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment5http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment5Thu, 28 Jun 2007 00:48:25 GMTI agree with hammet on this one... components with optional dependencies should be usable with or without the container, perhaps the class itself makes use of a static NullObjectFactory to create the appropriate null object, but I think having it created/injected by the container transparently implies that constructor can't be used without the container in the mix. http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment4http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment4Wed, 27 Jun 2007 23:30:25 GMTI disagree. I think classes should be coded in a way that they can used with/without a container. So, not relying on anything to work well. And I think that Castle.MicroKernel was coded to match exactly this goal. In other words, reversibility. Btw, can you show up on GTalk this weekend? We need to talk about the presentation.. http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment3http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment3Wed, 27 Jun 2007 22:24:06 GMTI would be freaked out if _you_ couldn't :-) I think that this is something that should be handled in the container level, freeing the application code from checking that, or supplying a null implementation explicitly. IUsageNotification is an optional dependency, for instance, and I may want to write it up for some controllers, and not for some, and I don't really care to maintain a NullImpl for that (rapidly changing interface, for one thing). http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment2http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment2Wed, 27 Jun 2007 22:19:32 GMTYes I can. But I dont quite agree with him. If the class exposes optional dependencies it should be capable of using default implementation of those. That's something I mention here http://www.castleproject.org/container/facilities/v1rc3/logging/index.html#BestPractice http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment1http://ayende.com/2603/challenge-windsor-null-object-dependency-facility#comment1Wed, 27 Jun 2007 22:14:36 GMT