public virtual MemoryStream

When Java was first created, one of it's main strengths was "Write Once, Run Anywhere", or WORA for short. It didn't take long before developers realized this was an overstatement and WOTA (Write Once, Test Anywhere) was a more accurate description of this feature. This was true back then for AWT applications and still is today for J2EE apps.

Unfortunately, I have realized that .NET isn't much better in this regard. And I'm not talking about MONO here, I'm talking about the ability to run a .NET application on any Windows platform; something which I assumed would naturally hold true. Well we all know what they say about assuming things, and I guess I did make an ass out of myself when it turned out SharpReader was unable to run on Windows 98 & ME.

I finally figured out last night what caused this problem, and it turned out to be a library that has it's roots in the Java world: log4net. Not that this makes it Java's fault of course - log4j will happily run on any Windows or non-Windows platform out there. Unfortunately, log4net does not.

To make things worse, because I was using static initializers for the loggers, the application never even reached the first line of my code and instead crashed with some abstract unhelpful error-dialog, which surely did not help in figuring out what was causing this failure. So in an attempt to get some more helpful feedback, I created this little application-loader to start SharpReader:

static void Main(string[] args)
{
    try
    {
        string appName = args[0];
        Assembly assembly = Assembly.LoadFrom(appName);
        assembly.EntryPoint.Invoke(null, null);
    }
    catch (Exception e)
    {
        Console.WriteLine(e);
    }
}

And bang! There was my stacktrace and my cause. From there on the solution was pretty simple - I just created a factory method that wraps the log4j.LogManager ILog-factory, which catches the PlatformNotSupportedException. If the user is running Windows NT/2K/XP, a standard log4net logger is returned, otherwise a simple file-logger is returned that does not use log4net. This file-logger still implements log4net.ILog though so the clients won't notice any difference and don't need any changes apart from where they create the logger.

Now since all I had was a P166 with a whopping 32Mb of memory to test this, things did run very slow. I know SharpReader now starts and can retrieve feeds on 98/ME, but because of this excruciating slowness, I have not done any extensive testing on it. Please let me know if you run into any further problems on these platforms.