A few months ago I wrote about writing a generic plugin loader/manager in C#, where I offered links to several articles and referenced a rather excellent book about C# which I had used to base my plugin loader on. I hadn’t really given it much thought since then, but according to Google Analytic’s, it’s one of the most hit posts on my blog. Recently I had a request to release the source code.

Now when I first wrote that post, I was hesitant to include any code, and I still am – not because I think there is something new and unique with what I’ve done – but, rather that what I have written is not terribly difficult to write. I do not mean to sound snobbish or arrogant at all, I’m just telling you how it is, all I’ve done is to do a bit of reflection on assemblies in a folder, and load instances of certain interfaces.

Anyway…

The scenario is that you want to provide a way for for 3rd parties to be able to add additional functionality to your application at run time. We need to provide a common way for 3rd parties to be able to register their new functionality into our application, in order that the user can take advantage of the exciting new feature being added to the application.

Straight away, you should be thinking to yourself: Interface!


public interface IPlugin
 {
 /// <summary>
 /// Does what ever It is.
 /// </summary>
 void Do(Action it);
 }

Anyone who now wants to create a plugin for our application must implement our IPlugin interface, as it defines the contract to which our application is bound to, in order to recognise and load plugins. Thus any assembly, that has a class which implements IPlugin is considered by our application to be a plugin which is capable of offering additional functionality.

We can now attempt to load our plugins. We need to have a class which can scan a folder for assemblies, scan those assemblies for types which implement IPlugin, and then create instances of them which our application can use. Loading the assemblies is easy:

public class PluginLoader<T>
{
 private IList<T> pluginsList = new List<T>();
 ...
}

...

public virtual IList<T> LoadPlugins()
 {
 foreach (string file in Directory.GetFiles(this.pluginFolderPath, "*.dll", SearchOption.AllDirectories))
 {
 Assembly assembly = Assembly.LoadFile(file);
 this.LoadObjects(assembly);
 }

 return this.pluginsList;
 }

We create a generic class, so we can use it with any type of plugin, and not just ones which implement IPlugin, then it’s just simple directory recursion to load all the files in the specified folder which have a file extension of .dll. The real magic happens in the LoadObjects() method.

var types = from t in assembly.GetTypes()
 where t.IsClass &&
 (t.GetInterface(typeof(T).Name) != null)
 select t;

 foreach (Type t in types)
 {
 T plugin = (T)assembly.CreateInstance(t.FullName, true);
 this.pluginsList.Add(plugin);
 }

Using LINQ, we extract all the types from the assembly which are classes, which implement the interface which is a typeof(T) – T being the type we specified when we instantiated the class. You could just as easily here specify the type should inherit from some other type, and you could also check to see if a class has some assembly level attribute.

Why would you want to do that? Well, we can use an assembly level attribute to decorate the plugin class with meta-data about the plugin, such as the author, a short description, the name of the plugin, and it’s version.

public virtual KeyValuePair<string, List<KeyValuePair<string, string>>> GetPluginInformation(Type type)
 {
 var attributeInfo = from pa in type.GetCustomAttributes(false)
 where (pa.GetType() == typeof(PluginAttribute))
 select pa;

 foreach (PluginAttribute p in attributeInfo)
 {
 data.Add(new KeyValuePair<string, string>("Author", p.Author));
 data.Add(new KeyValuePair<string, string>("Description", p.Description));
 data.Add(new KeyValuePair<string, string>("Name", p.Name));
 name = p.Name;
 data.Add(new KeyValuePair<string, string>("Type", p.Type.ToString()));
 data.Add(new KeyValuePair<string, string>("Version", p.Version));
 }

 this.attributeInformation = new KeyValuePair<string, List<KeyValuePair<string, string>>>(name, data);

 return this.attributeInformation;
 }

Or you could create a struct to use as a DTO for the plugin meta-data, it’s up to you.

I’ve removed all the comments and exception handling from the code I’ve posted purely just to save space, you’d really want to include that – especially the exception handling. But that’s really all there is to it, you might want to have a property to access the actual plugin list, or return it from a LoadPlugins method, it’s up to you.

You’ll notice I’ve made the methods virtual, you may want to use this class as a base class in another class. For instance I’ve got an additional class which inherits from my plugin loader class which does specific tasks for a particular type of plugin, and another which does different tasks for a different type of plugin.

I have recently started using psake to do some build automation at work, and I’ve found that there is not a great deal of information about how to write a build script using psake available on the internet. It isn’t all that amazingly difficult if truth be told, however there are a couple of ‘gotchas’, and I would like to share what I have learned in the hopes that it benefits someone.

If you have not heard of it, psake is a”…build automation tool written in PowerShell”, started by James Kovacs. With it, you can write build scripts, with which you can automate the build and deployment of your .NET project. Recently the version 1.00 and 2.0 0 version were released. Why two versions? Well, the 1.00 version is primarily for PowerShell 1.0, but psake 1.00 is being “retired”. This article assumes the reader is using psake 2.01 and PowerShell 2.0.

A quick note here, I should point out the primary example I based my first build script on is Ayende Rahien’s build script for Rhino Mocks.

Our first psake build script

First, I have to make two assumptions:

  1. I have to assume that you have PowerShell installed, if you are running Vista/Windows 7 then it’s installed by default, if you are on XP, then it is a manual install.
  2. You have installed the psake module into Powershell – see the release announcement linked to above for details on how to do this.
Example solution

Example solution

With those assumptions out of the way, we are going to write a build script to automate the building of a simple C# solution, containing a Windows Forms application, and two class library assemblies.

Hopefully this should be simple enough to easily follow along with what is happening in the build script, but complex enough that you can see how sophisticated your build scripts can be.

You can see that I have already taken the liberty of adding an extra file to the solution, “build.ps1″, this is our build script, with which we can command psake to do great things for us.

With our solution setup, we can now write our first build script:

properties {
 $base_dir = Resolve-Path .
 $sln_file = "$base_dir\WindowsFormsApplication.sln"
}

Task default -depends Compile

Task Compile {
 msbuild "$sln_file"
}

Open a command prompt into the directory containing the .sln file, which is where your build.ps1 script
should be, and run this command: invoke-psake .\build.ps1 -taskList Compile

psake Compile task output

psake Compile task output

You should receive some output to the console window like the above.What the script does is to work out where on the file system the script is running, and gets the path to that folder, and builds the path to the specified solution file, and runs msbuild, passing the full path to the solution file as a parameter.

The script as it is has some drawbacks though. In it’s current form, msbuild will only build the default configuration of the solution. What if we want to build a Debug version, or a Release version, or some other configuration we’ve created? Additionally, it doesn’t let us specify a directory to put the built binaries, they just get put in the default locations as specified in the projects contained in the solution – what if we want to put them in a custom location?

If we modify our build script slightly, we can introduce this functionality. Firstly, we need to specify some additional properties:

properties {
 $base_dir = Resolve-Path .
 $build_dir = "$base_dir\build"
 $sln_file = "$base_dir\WindowsFormsApplication.sln"
 $debug_dir = "$build_dir\Debug\\"
 $release_dir = "$build_dir\Release\\";
}

Notice the double backslash, msbuild requires a trailing slash when paths are specified, and it seems to require the additional backslash as well, or else it doesn’t work. With those additional properties in place, we can introduce two new tasks to our build script:

Task Clean {
 remove-item -force -recurse $debug_dir -ErrorAction SilentlyContinue
 remove-item -force -recurse $release_dir -ErrorAction SilentlyContinue
}

Task Init -depends Clean {
 new-item $debug_dir -itemType directory
 new-item $release_dir -itemType directory
}

Powershell’s easy to read syntax should make it easy to follow with what is happening now. In the Clean task, we forcefully and recursively remove any files and the folder from the specified path, and if there are any errors then they are not displayed. Now that we know that those folders are going to be cleaned and created, we can create two further tasks, where we can create Debug and Release versions of our solution:

Task Debug -depends Init {
 msbuild $sln_file "/nologo" "/t:Rebuild" "/p:Configuration=Debug" "/p:OutDir=""$debug_dir"""
}

Task Release -depends Init {
 msbuild $sln_file "/nologo" "/t:Rebuild" "/p:Configuration=Release" "/p:OutDir=""$release_dir"""

Again, the syntax here is relatively straightforward to follow along with, we execute msbuild, passing it the solution file to build, specify not to show the logo (suppressing the output of “copyright microsoft msbuild etc), tell it to execute the rebuild target and to build the Debug configuration, copying the output to the specified debug directory. Notice that that there are no spaces in the paramters that we pass to msbuild. For example, if we passed the parameters like this: “/p:Configuration=Release /p:OutDir=”"$release_dir”"”, then it would fail and we would get a msbuild parse error saying it was invalid.

Build script output for Debug task

Build script output for Debug task

Summary

In a relatively short amount of code, less than 30 lines, we have accomplished quite a lot. We can now issue the command: invoke-psake .\build.ps1 -taskList Debug, and psake will automatically clean the output folders, do a full rebuild of the Debug configuration and copy the output to a custom location on our filesystem. What’s more, the build script we have written is small, compact, easy to read, easy to maintain and easy to build/extend upon in the future.

As this is getting a bit long already, I’m going to cut things short here, however, there are some additional things that you can do as part of the build script that are very nice, such as automatically versioning the assembly before you do the full build. If you take a look at Ayende Rahien’s example from Rhino Mocks, that is covered there.

Also in the new version of psake, there are pre and post conditions and actions that you can add onto your tasks, although I haven’t had the opportunity to use them yet. I’ll try to cover those in a future blog post though.

Nearly every application I have ever used has produced some sort of log output, usually to record details of exactly what the application was doing when an error was encountered. To be perfectly truthful, it is not something that I have ever really given very much consideration to in either my own or work applications, there have always been more pressing things to attend to, such as meeting (unrealistic) deadlines, or implementing some interesting new feature. I’ve used log4j before in a work application, mostly because someone else had already done all the configuration.

However, the application that is currently occupying the main focus of my attention at work had (almost) no logging in it what-s0-ever, but given the nature of the application, it can be very difficult to debug exactly what is happening internally using breakpoints. Again, the nature of the application doesn’t really lend itself well to showing MessageBox’s all the over the place.

MessageBox.Show("Log message");
 
// or..
 
Console.WriteLine("Log message");

How many times have you done that, or the equivalent, in an application? If you are like me, then you’ll have done it hundreds of times, probably in the same application. It’s really quick and easy to do, and can be very useful. You just have to remember to take it out of the Release build. What a pain.

A better way

Using a proper logging library (as I have now discovered) allows you much greater freedom. Freedom to not have to (overly) worry about leaving in Log.Debug(“Some message”) all over the code. Freedom to be able to configure what is logged, where it is logged to and how it is displayed. Freedom to receive concrete data on how your application is behaving in the real world, which will enable you to find and fix bugs faster than you would have thought possible.

There are a multitude of logging libraries available:

Now, while it would probably be an interesting exercise to write a small logging library, the world doesn’t need another one, so I immediately discounted that option.

I’ve used System.Diagnostics here and there, and an ex-colleague wrote a small logging library based around using Trace and Debug, but that is getting dangerously close to writing your own logging library, so again that’s out.

The Logging application block from the Enterprise Library just seems too weighty, and too much of a pain to configure (I can probably be proved wrong though), so I discounted that as well.

The object guy’s logging library seems pretty good, but the documentation and examples kind of suck, and it doesn’t look like it’s been updated in a while, so after playing with it a little bit (and not really liking it all that much to be perfectly honest), I discounted that as well, although it does seem to have a lot of fans on Stack Overflow.

NLog has plenty of documentation, and seems pretty straightforward to use, and it too has a lot of fans on Stack Overflow, but again, it’s not been updated in a while, so I discounted that too.

Log4Net to the rescue

Log4Net is part of the Apache Logging Services project, along with Log4j, and Log4Cxx. It is a C# port of Log4j, and works amazingly well. I am not going to write about how to set it up and configure it, as others have done that already, and the documentation is excellent.

I’ve read here and there on some blogs, a few message boards and some email lists, that log4net is difficult and time consuming to set-up, and that it is hard to use. Which is nonsense, because after ten minutes (which included downloading the latest version) I had my first small test application up and running and outputting debug messages to the Output window in Visual Studio. Pretty swish indeed.  I did spend some time reading about the best ways to use log4net, and I still do, but I didn’t find the initial set-up to be inordinately difficult at all.

On the whole, I am impressed with it’s ease of use, I’m now starting to use it in my own personal projects, and it is proving to be invaluable at work.

One of my current long term hobby project is a map/world editor that I’m writing for Andy’s (work in progress) game engine. It is still in a quite rudimentary stage, however, I have gotten an important aspect of the application into a fairly mature state already, and that is the plugin system.

I don’t want to get into too much detail over things that are available via some simple searches on google, so don’t expect too much code. There are some good articles here, here and here, which delve quite deeply into this, and how to develop a plugin framework in general. I developed my plugin framework based on the code in this book, which incidentally is a very good book.

Also, in my own google searches on this subject, a lot of the articles I found are getting to be four years old, in a few instances even older. Not that this automatically makes them invalid as sources of information, it’s just that they may not reflect best practices in modern C#. (Not that I am claiming to be a Jon Skeet-like C# guru). Also, quite a few of the ones I found make use of an XML configuration file to hold a reference to the plugins. I’m not a big fan of this type of thing, I much prefer convention over configuration.

Why Generic?

I’ve written a plugin system for the Editor because I want to provide a way to make it easily extendible in the future, without having to change the core application, to enable it to perform actions that are above and beyond it’s core capabilities. Byond that basic requirement, there are a couple of other reasons to make it a generic plugin framework.

Firstly, I’ve made it generic because I wanted to make a distinction between different types of plugins, such as a type of plugin for resources, and a type of plugin which adds functionality into the application (such as tools, UI widgets etc).

Secondly, and perhaps most important (to me, anyway), is that I’ll never need to write a C# plugin framework again, for any of my personal projects. I’ve had a go at writing a plugin framework before, a few times for none-work projects and once for a project at work, and I finally realised what my problem was – reinventing the wheel each time.

The plugin framework I’ve written ensures that I don’t have to worry about this the next time one of my pet projects needs a plugin system. The Editor will in fact have two types of plugins available, one to extend the types of data sources it can load in resources from, and another to provide additional functionality, like tools.

I’m fairly pleased with the way that it has turned out, and it works quite well in practice.

Update:

I have written another post on this subject with some code examples here: http://temporalcohesion.co.uk/2009/11/02/more-on-the-generic-plugin-manager/