Inversion of Control, Dependency Injection and Service Locators oh my!

This post is part of my “Software Development” series. The TOC for the entire series can be found here.

In this post, we’re going to look, at a higher level, at the reasons for and benefits of using a framework for Inversion of Control (IoC), Dependency Injection (DI) and Service Location. I’m not going to spend a large amount of time in this blog series describing the intimate details of what IoC, DI and Service Locators are.  Martin Fowler did a fantastic job of it and there’s no point in me just regurgitating his work.

Instead, I’ll give my vision of what they are and how they are relevant to how I architect and develop solutions.  It’s important to note that I’m starting out this series with high-level thoughts about architecture and design.  I’m not just laying down some code, leaving the decisions about how things will interact to some arbitrary, organic process.  This is how software development gets done. 

You’ve got to think about not just the next step, but visualize the end point several months, or even years out and try to connect the dots back to where you are today.  If I implement something in a certain way today, is it going to screw me 3 months down the road?  Spending an extra day now thinking about architecture is going to turn into saving weeks of the customer breathing down your neck and you pull out your hair and work late night trying to hammer a square peg into a round hole.

So let’s look at IoC, DI and Service Location and how they’re going to help us build better software.  Inversion of Control as a concept is something that’s really broad.  It’s really anything that turns the older procedural way of developing upside down (i.e. inverts program control, hence the name).  Moving from a console app to a windowed, event-driven app is technically inversion of control, so you’ve already been using IoC. 

To talk about IoC as a useful concept that’s actually going to be of use, we’ll be using some smaller units/principals of IoC.  The two we’re going to look at are Dependency Injection, or DI, and Service Location.  Neither one is complex or arcane, and you probably already use them to some degree.

Let’s assume we have two arbitrary classes, A and B, and that A uses an instance of B internally.  There are, essentially, 3 ways that A can get the instance of B. 

First, it can directly create it by calling new B().  While this is simple, it’s the least preferred, and generally worst, way to do it.  By directly creating B, A is now tightly coupled to B’s implementation.  B can’t be easily changed out to something else.  If B is in another assembly, A has to have a reference to it.  It’s also difficult to test A by itself. 

The second way that A can get B is that whoever created the instance of A could pass into it an instance of B.  Maybe through a constructor

var b = new B();
var a = new A(b);

or through a property

var b = new B();
var a = new A();
a.B = b;

This is called Dependency Injection. The former is constructor injection, the latter is property injection – not so complex, eh?  This is useful because, if B is an interface, it’s really easy to swap out the implementation passed to A and the assembly containing A doesn’t necessarily have to have a reference to the assembly holding the implementation of B – it only needs a reference to the interface.  This is really nice when we want to do mocking.

The third way that A can get B is to go get it from some central repository or “factory”.

B B { get; set; }

public A()
this.B = MyClassFactory.GetB();

This is service location.  A doesn’t really know what B is, it lets the service locator (MyClassFactory) resolve that.  This is really helpful for plug-in architectures as well as for lazy loading objects.

Generally speaking, your code should only be doing #2 (DI) and #3 (service location) for objects of any complexity.  This is not to say that you shouldn’t ever call the new operator in your code.  There’s obviously a grey line out there below which it would be pretty stupid to not just use new for.  If you need to use a simple entity class or structure, then creating one makes sense.  If you need access to the DAL and I see you creating a DAL object in your View, you’re fired.

There are plenty of framework out there that provide DI and service location –  Ninject, Unity, CastleWindsor, StructureMap, Autofac – the list goes on and on.  Instead of using one of them, however, I chose to roll my own.  Normally I wouldn’t recommend such a drastic action – after all reinventing the wheel isn’t usually wise – but in my case it was essentially required, so I’ll give you the short back story.

My team and I had been working on a large desktop application.  We had designed it from the start using Microsoft’s SCSF/CAB framework, which provide DI, service location and a load of other stuff.  It turned out that we were using maybe 10% of the framework, and that the framework’s Studio plug-ins were causing fits with some installations of Studio.  Spending time screwing with Studio plug-ins and trying to get things to compile is a terrible waste of manpower, so we were already getting close to jettisoning it when we decided that we wanted the core of our app logic to be Compact Framework compatible.

Well Microsoft’s “port” of the SCSF to the Compact Framework, called the MCSF, turns out to be a bloated, unbelievably slow, steaming pile of dog crap.  Side note: just because something compiles under the Compact Framework does not mean it should be used in the Compact Framework.  Well, I didn’t want to refactor everything we’d done to some other IoC framework, plus I couldn’t find one at the time that actually worked for the Compact Framework, so I decided to create one. 

I took the tack that I’d create it for the CF first, with a high emphasis on small footprint, minimal resource usage and speed and I simply matched the SCSF object model where our code base interfaced with it and where it made sense – in some places I didn’t like how the SCSF had done things, so I “fixed” their mistakes.  The end result was the OpenNETCF.IoC framework, which turns out to stack up quite well against the other frameworks.  It also has the huge benefit of working on the Full Framework, Windows Phone 7 and MonoTouch (probably MonoDroid as well, though I’ve not tested that). We’ll be looking at OpenNETCF.IoC in a lot more depth in this series.

On Software Development

This post is part of my “Software Development” series.  The TOC for the entire series can be found here.

Developing good software is hard.  Really hard.  Sure, anyone can buy a book on writing software or pull up some code samples and get something that compiles and runs, but that’s not’s really developing software.  A lot of code in the wild – I’d bet a vast majority of it – just plain sucks.

It’s hard to point out where the blame lies.  It seems that most developers are environmentally or institutionally destined to write bad code. Schools teach how to write code, but not how to architect it or to follow reasonable design practices.  In the zeal for clarity, publishers churn out books, blogs and samples that show bad practices (when is it ever a good idea to access a data model from your UI event handler?).  Managers and customers alike push hard to get things done now, not necessarily done right – only to find months or years later that doing it right would have saved a boatload of time and money.  And let’s face it – many developers are simply showing up to work to pull a paycheck.  You know who they are.  You’ve no doubt worked with them in the past.  You’re probably working with them now.

I was watching Gordon Ramsay the other day and it occurred to me that he and I are alike in our own peculiar way.  I’m not saying that I see myself as the “Gordon Ramsay of Software Development” – hardly –   but we share a common trait.  Just as Gordon gets angry and starts spewing colorful language when he walks into a crap kitchen, it bothers the hell out of me to see complete idiots in my chosen field out there just making a mess of things.  When I see bad code – not necessarily minor errors, or code that could be refactored and made better – but just outright shit code that should not have occurred to a developer in the first place it pisses me off.  By the nature of my work, often getting called in only when the project is off the rails, I see it all the time. Code that, on review, a peer or mentor should have seen and said “Whoa!  There’s no way that’s going into our code base”.  Code that just makes it harder for the next person to do their job.

In an effort to simplify things for my own code, for my customers’ code as well as anyone who is willing to listen to my ravings, I’ve spent a lot of time building, testing, fixing and extending tools and frameworks that many of which I turn around and give away.  This isn’t out of altruism, no, it’s largely because I’m a lazy developer.  I hate writing the same thing twice.  When I start a project, I don’t want to spend large amounts of time building up the same infrastructure that every project needs. Building up a framework for handling UI navigation isn’t what I’d call interesting, but just about every project needs it.  Handling object dependencies and events is common.  Writing a DAL for serializing and deserializing entities is not just drudgery, I find it’s highly susceptible to errors because you end up doing a lot of copy and paste.

I have these cool tools and frameworks that I use in literally every project I work on now.  That’s great for me, but it doesn’t really help others, right?  Without reasonable documentation or explanation, only a small handful of people are going to go through the effort of getting the tools and trying to understand them – even if they are deceptively simple and could potentially save you weeks of effort. 

So I’ve decided to put together a series of blogs over the coming weeks and months that explain, hopefully in simple terms, what these frameworks do, how to use them, and most importantly, why they are worth using.  There’s nothing groundbreaking here.  I didn’t invent some new way to do things.  I’ve simply appropriated other peoples’ ideas and extended them to work in the environments that I work.

Generally I’ll be covering the following topics and frameworks:

  • Dependency Injection and Inversion of Control (using OpenNETCF IoC)
  • Event Aggregation (using OpenNETCF IoC)
  • Plug-in Architectures and interface-based programming (using OpenNETCF IoC)
  • Software features as services (using OpenNETCF IoC)
  • Data Access through an ORM (using OpenNETCF ORM)
  • Parameter Checking (using OpenNETCF Extensions)
  • Exposing data services over HTTP (using Padarn)
  • Whatever else I think of

If there’s a topic you’d like me to talk about, feel free to send me an email.  I may turn on comments here and let you post ideas, but I find that when I enable comments on my blog, I start getting more comment spam than I really want to deal with, so if comments are turned off just drop me a line.

Benchmarking OpenNETCF’s IoC Framework

Last night I was browsing around the Net on a back tracking a visit to the OpenNETCF Inversion Of Control (IoC) project.  I came upon a page where someone had done a quick set of benchmarks for many of the popular IoC engines out there [].  This got me wondering.  How would OpenNETCF IoC stack up against these, after all I’ve never done any performance comparisons against any other framework.

Well I downloaded the benchmarking code the original author had put together and added added an IoC version of the test (which was really easy to do).  I then ran the singleton test, which essentially registers a set of dependent classes by interface name (in theRootWorkItem.Services collection) and then uses the IoC engine to extract the from the store 1,000,000 times.  It then reports how long it took to do the extraction.  Simple enough.

Here are the results:

The hard numbers are not really important – the actual number is going to depend on your hardware – what is interesting is how they look relative to one anotehr.  You can see that if you exclude the new operator (which isn’t IoC, it’s just there for reference), then the Dynamo engine (which is the test author’s engine) preforms the best.  Now I know nothing about that engine, so if you’re interested, I’ll let you investigate that further.

What was interesting to me is that IoC is three times faster than the next fastest IoC engine (AutoFac), five times faster than Structure Map, roughly 6 times faster than Castle Windsor or Unity and a whopping 25 times faster than the ever-popular Ninject engine.  Evidently my focus on performance when developign the engine paid off.

How about the time it takes to register all of the types that get created?  Are we paying the piper at creation time in order to save it at resolution?  Here are the time results for setup (instantiation of singletons):

Again, IoC is at near front of the pack.

Binary size isn’t overly relevent any longer, I mean if you’re using managed code then you’ve already made the decision that file sizes aren’t critical, but it’s still interesting to look at.  Here’s a breakdown of the sizes of each of the libraries.  The libraries marked with an asterisk also have configuration files that are *not* included in the resported size (so deployment would actually be larger than what the graph shows).

All in all, I’m pretty pleased with the results.  We dogfood the OpenNETCF IoC framework for all of our projects, both desktop and Compact Framework, so it gets a fair amound of testing and beating and you can be assured that if there are bugs I’ll do my best to resolve them.  If you’ve not taken a look at the project, give it a try.


IoC for Silverlight Mobile?

Nick Randolph and I have been kicking around the idea of adapting the OpenNETCF.IoC project to Silverlight Mobile on Windows Phone 7 (and by “discussing” I mean I’m trying to convince him to do it for me).  I just noticed over on the IoC stats page that we’ve had some referrals from someone who apparently has either already done it or is in the process of adapting it for PRISM.  I’ve contacted him to see if we might collaborate, but this is exciting news. Having a dependency injection framework right out of the box for phone development can only help the community provide solid applications.

Updates to OpenNETCF.IoC Framework

We recently ported a desktop SCSF/CAB application over to use the OpenNETCF.IoC framework.  The primary thinking here was that the SCSF has way, way more stuff than we really need, want or use and the SCSF wizards had done wacky things to a few Studio installations, so we wanted to get rid of all of that as well.

I’ve checked in the fruits of the labors into the Codeplex project, including new desktop project and solution files.

Now you can be even closer to one codebase for both desktop and device projects.  If nothing else, it allows you to not have to remember two different injection frameworks if, like me, you end up doing a lot of work on both platforms.

Revisiting the Form Stack

Long, long ago I wrote an article for MSDN on creating a multi-Form CF application that used a Form Stack.  Since we all tend to grow and learn as developers, we find different ways to do things (and generally we scoff at code we wrote years before as inferior crap). 

Well Peter Nowak is giving a presentation of the OpenNETCF.IoC framework next week and I decided that a sample of using the UI elements of the library (Workspaces, SmartParts, etc) might be handy and so I decided to rewrite the Form Stack following my latest thinking.  Basically the idea is to allow a user to navigate through Forms (actually Views – you are separating your Views from the Model, right?) like you would on a browser.  You can move forward and back as well as “adding” to the end or top of the stack.  We also don’t want to be constantly creating new instances of the View classes because we like applications to perform well.

Here’s a look at the new application (yes, I know it’s “developer ugly” but this is about architecture, not aesthetics):

You can see the stack in the list, and our current position is noted by the asterisk.  “Fwd” will move down to View B, “Back” will move up to Form A, or you can push a new A or B onto the stack at the current location, which will truncate everything currently above (after) the current position.  Again, think of how your browser works.  It’s important to know, also, that there are only 3 total View instances created at this point (one of each specific type).

The code for this application is in source control over at the OpenNETCF.IoC Framework Codeplex site.  You’ll notice it’s called FormStackCS, hinting that there may be a FormStackVB coming.  If you’d like to volunteer to do that port, by all means let me know (meaning don’t hold your breath waiting for me to do it).

IDisposable and Dependency Injection Containers

Recently I got feedback from Nick Randolph, another device developer MVP, regarding the OpenNETCF.IoC framework and handling the lifecycle of IDisposable objects (well we taked about a few things, but this is the one I tackled first).  The problem is that if you add an IDisposable object to any DI container (be it OpenNETCF.IoC, Unity, Ninject or whatever) the container knows nothing about the object’s Disposed state.  Unfortunately the IDisposable interface doesn’t expose an IsDisposed property or OnDisposed event. 

SO what, exactly, is the problem?  Well, let’s assume we create an IDisposable object and add it to the DI container.  AT some point later, we’re done with the object, so we call Dispose.  Well that internally releases the object’s native resources, but the object itself still has a root (the reference from the container) and so the GC will never actully collect the object and its managed resources.  On an embedded device this could be a problem.

The solution that we came up with is to create a wrapper for IDisposable objects and a new set of methods for the container for adding.  So now you can do something like this:

var instance = RootWorkItem.Items.AddNewDisposable<MyDisposableType>();

Now instead of this returning an instance of MyDisposableType, it returns a container object that holds it.  You access the MyDisposableType object itself via the Instance property of the container, which requires some small code work on the consumer’s part, but the nice thing is that when you call Dispose on that container, it in turn calls Dispose on the contained instance *and* it removes the object from the DI container so the GC can do it’s work on the object.

I’ve not rolled a new framework release with these changes because they really are just a proposed solution for now.  If you’re interested, go ahead and pull the latest code from the source code page (change set 32718) on the project site and give it a try.  There are a load of new unit tests that show the general idea on usage.

Added CommonServiceLocator support to OpenNETCF.IoC

It’s been a while since I did any check-ins to the OpenNETCF.IoC framework.  It’s not due to lack of interest or lack of thought on my part.  I’ve just been very, very busy lately.  We’ve now using the IoC project in a few commercial applications with very good results, so there isn’t a whole lot more to add (though I do want to add some sort of lifecycle management capabilities).

In the hope of getting a little more adoption of the framework, I’ve added an adapter to OpenNETCF.IoC for the Microsoft Patterns and Practices CommonServiceLocator (CSL).  You can use CSL code to extract object instances from the Items collection.  My hope is that this makes it easier for development teams using CSL to integrate OpenNETCF.IoC into their mobile and embedded projects.

If you have any other ideas or feature requests for OpenNETCF.IoC, by all means submit a request or start a discussion on the project site.  And if you’re using it, I’m always interested to see how it’s working out.

Improving dependency injection speed in the IoC Framework

I’ve been using the OpenNETCF IoC framework for several new projects lately, and I really like it.  I think my favorite feature is the fact it gives you the ability to subscribe to events from other objects without requiring an instance of the event publisher.  In fact you don’t even need to know the publishers type or even of its existence – you simply need to know the name of the event that might get raised.

The only downside I’ve hit lately is that the performance for creating objects is not fantastic.  Of course that’s to be expected.  To achieve all of this really loosely coupled architecture, we have to rely heavily on Reflection.  Not only do we have to do object creation through Reflection, but we have to do a lot of complex querying looking for type names and attributes, and this can take a while.

In an effort to improve things a bit, I’m working on updating the internal ObjectFactory class that is responsible for all of this heavy lifting.  I’ve added caches for most of the items that get queried through Reflection.  For example now when we want to find all of the published events from a specific Type, we use reflection the first time, but every subsequent time we pull the list from a cache.

A unit test for creating a single object that both publishes and subscribes to events yields nearly a 5,000% improvement in creating speed for the second object of the same type.  Yes, you read that right – it is 50 times faster creating the exact same object type the second time around.  There’s no avoiding the pain the first time we have to inspect a Type, but I’ve already noticed a marked improvement in app startup time using the new changes.