Looks like the Silverlight related posts I’ve announced would have to wait one more blog post due to the event which happened to me today and which made me thinking about a few of the architecture related things which at the end resulted with a few of (at least for me) interesting thought I felt sharing with the community.
So, a fellow blogger @RadenkoZec made a nice blog post about Facade design pattern in which comments we had a discussion if the example is appropriate or not, and if the implementation is Facade design pattern or some other pattern. I won’t repeat here in detail what we were discussed there (go to the blog post to read comments) but there were couple of things in that discussion I was thinking about during the evening…
Patterns should be explored in 3D
It is very interesting that the sites on the net (like the dotfactory.com) seems to be primarily focused on GoF patterns while I was not able to find sites covering at the same time PoEAA patterns andor DDD patterns.
That fact might look irrelevant but if you would check out the content of today’s discussion you would se that the same example code used in blog post
looked to Radenko as Facade (GoF)
and to me more like a Gateway (PoEEA)
If you just take a look at above pictures, you would find for sure at least some similarity and that’s why one should look at all of the patterns in whole and not just be exclusive in picking “The book”. Another example of this interpretation conflict can be found between the PoEAA and DDD where patterns such as repository, factory etc some times have different usages if not different meanings.
In other words, IMHO every developer should persist knowledge in three dimensions:
- X– DDD, PoEAA, GoF
- Y – specific patterns and their implementations, implications; anti patterns too.
- Z – time
For the ones of you probably asking what time has to do with this subject - here’s an answer. Software landscape changes significantly in last 20 years so some of the “scriptures” should be taken with a grain of salt.
Here are couple of examples illustrating my point:
- Observer pattern doesn’t have a lot of sense being implemented out of box with .NET events,
- Singletons (btw, would have a dedicated post showing how evil they are) are kind of obsolete with the usage of the IoC containers etc
Another aspect of the need for considering the time dimension in design patterns exploration is the fact that in last few years we are all witnessing both the rise of the dynamic languages and the enhancement of static languages where they are getting aspects not existing in the time when “pattern Bibles” were made (lambdas, generics, C# 4.0 dynamic etc). Every of the patterns therefore should be heretically challenged in the lieu of current state of technology.
Context is the king
By mare looking at above diagrams (which for my point can be both pure UML diagrams) it would be really hard to tell what is the difference..
In both cases we have multiple classes encapsulated in one class orchestrates their calls and expose simple API to be consumed.
But if you take a look at the context you can see based on the provided code sample that the packages in case of facade are interconnected, each one of their classes is dependable in it’s own way dependable on other classes and none of them can exist without others. I could imagine refactoring merging those 3 classes in one. In other words, even physically we have multiple classes caring their own implementation they are so connected that on architectural level there’s just one entity which is hiding behind the facade.
That’s why for me Facade feels like “1 – 1” relationship type pattern where a facade hides the complexity of the class (similar in a way to adapter but let’s not digress with that here)
The second case is with clearly different context where all of the packages behind the “facade” (Gateway) are quite independent. They do not depend on each other at all, they cooperate as partners.
The purpose of the PricingGateway is also to expose a simple API to pricing Package but this time when the API would be used Gateway would perform a role of a conductor, orchestrating the calls to the "hided" elements.
That’s why for me Gateway fells more like a “1 – n” relationship type design pattern
To summarize: In order to distinct patterns one would have to understand the problemimplementation context before picking the right flavor – appropriate pattern.
Patterns as a matter of feelings
As you have probably noticed in the text above I am using the “feels like to me” which I’m pretty sure is looking quite unrelated to explicit and scientific thing such are the design patterns so let me clarify that for you
As many other, I’ve stepped through couple of phases in my Pattern life
- Discovery of patterns (we hear about them from some of the cool kids and all we do at this stage is trying to pretend as best as we can to look like we know what they are talking about)
- Trying to get it (AKA “I foundread the GoF book”)
- Becoming a believer (including making the presentations to people in phase #1)
- Getting it for real (reading every patterns book, blog post etc you can find and memorizing the UML diagrams, use cases etc)
- Seeing patterns everywhere and doing them as much as we can
- Paying the price from highly complex code base .
- Starting to understand the price of patterns application and start using it only when a real design pain is identified which can not be solved on simpler (but still clean) solution
- Instead of patterns catalogs starting to focus on basic principles.
So, as far I can tell I am in that phase #8 where all of the patterns somehow melted with each other and all of them “look similar”. I forgot half of their UML diagrams and reference implementations. The only thing in my head left is the name of the pattern, notion of the use case when it is useful and (very blurry for most of them) how it works.
Reading my last statement someone could make a conclusion that I just got more stupid (possible option I admit ) and to lazy to remember things but I would disagree with him because every one of the patterns I faced is based on the same set of plain logic principles which in case you know and feel them you are good to go with doing “your own” solution which would somehow match some of the patterns.
Here are some of examples of the design principles I have on my mind: SOLID, open-close, DRY, KISS, orthogonal code, OOP principles (encapsulation, abstraction), TDD, dependency injection and the list could go on with pages
To summarize: I strongly believe that if the developer is feeling natural regarding the design principles (spits out the code following those principles without even thinking) the knowledge of the design pattern can be at much more abstract level (in a way like knowing a page index of a book) without memorizing every bit of their reference implementation (that’s why we have bing aren’t we?) and the created code would still comply to all of those patterns.
I hope all of my rumblings I shared with you my dear reader would help me now to make up the point of this blog post and that is to explain how come when Radenko pulled (completely rightfully) on his blog post couple of referent examples from sites and books of folks far smarter then I am proving that his example implementation is the same as the one they provided, the only thing I could tell him is that his example “doesn’t feel like a case for Facade to me”.
I was not claiming neither that the guy who wrote a book did it wrong nor that the dotfactory.com site got it wrong in their sample.
It was just a gut filling of a pragmatic “duct tape architect” which I would have about that code if it would be mine which I shared with him and the community.