home

Weekend .NET Pains - Why I Prefer Ruby

Oct 18, 2010

It's been about 4 months since I've done any .NET programming, in that time I've been largely doing Ruby and Java. I want to contrast the .NET code and Ruby code I came up with to do something quite basic. First though, a honest warning:

The .NET solution is likely over-engineered and the Ruby one likely over-simplified. I'll be the first to admit that my .NET approach doesn't feel right. So this is as much to compare the two as to fix my messed up code.

So what was I trying to do? Read configuration data from a file. Really, that simple. You might wonder why I didn't hook directly into .NET's support for configuration (app/web.config). Personally, I've always found it quite inadequate, as are the mechanism for enhancing it. (for some insight on that, 6 years ago I wrote about writing custom configuration and 10 months ago wrote about a world beyond web.config).

For the purpose of this post, forget about reloading on changes - all I really want is a configuration class, driven by a file using a non-shitty format, and for the whole thing to easily allow me to test. Here's my configuration file:

{
  "LdapConfiguration":
   {
      "Url": "LDAP://someserver",
      "UserName": "some user",
      "Password": "some password"
   },

   "DataStoreConfiguration":
   {
      "ConnectionString": "Data Source=SERVER;Initial Catalog=DB;User Id=USER;Password=PASSWORD;"
   }
}

Now, remember, one of the points is that I want to be able to change these configuration settings on the fly for testing. For example, I want to test that my code properly handles a failure connecting to the LDAP server (and not in some half-assed mock way either). So I decide on the following approach for my configuration (and this may very well be where I made a fundamental design mistake):

public class Configuration
{
   private static Data _data = return new JsonConverter().FromFile<Data>(Runtime.ApplicationRoot + "settings.json");;

   private class Data
   {
      public ILdapConfiguration LdapConfiguration { get; set; }
      public IDataStoreConfiguration DataStoreConfiguration { get; set; }
   }

   public static ILdapConfiguration LdapConfiguration
   {
      get { return _data.LdapConfiguration; }
   }

   public static IDataStoreConfiguration DataStoreConfiguration
   {
      get { return _data.DataStoreConfiguration; }
   }

   private Configuration(){}
}

Essentially, my singleton Configuration class will be composed of a configuration object for each service. Here's what a specific configuration looks like:

public interface ILdapConfiguration
{
   string Url { get; }
   string UserName { get; }
   string Password { get; }
}

public class LdapConfiguration : ILdapConfiguration
{
   public string Url { get; internal set; }
   public string UserName { get; internal set; }
   public string Password { get; internal set; }
}

Why do all of that? Because it lets me inject the appropriate configuration settings into code using a DI framework:

Bind<LdapConfiguration>().ToConstant(Configuration.LdapConfiguration); //ninject binding

So, if that's all there was to it, I'd probably sigh and be on my way. But it isn't. My JsonConverter is a facade over JSON.NET. Guess what? It wouldn't deserialize my json into the appropriate object. This is the part that got frustrating.

AN IMPORTANT NOTE. I'm not complaining about JSON.NET - even if it sounds like I am. I've written 4 serializers for .NET and I've ran into all of these fundamental issues with .NET and haven't solved any of them as well as JSON.NET has.

Here's what the code started as:

public T FromFile<T>(string path)
{
   return FromJson<T>(System.IO.File.ReadAllText(path));
}

public T FromJson<T>(string json)
{
   return JsonConvert.DeserializeObject<T>(json);
}

The first problem? JSON.NET doesn't like my private constructor. I can't say I agree with that decision, but the JSON.NET library has to make a decision and one way or another people won't be happy with the choice. Thankfully its pretty easy to override the default behavior:

public T FromJson<T>(string json)
{
   var settings = new JsonSerializerSettings
   {
      ConstructorHandling = ConstructorHandling.AllowNonPublicDefaultConstructor,
   };
   return JsonConvert.DeserializeObject<T>(json, settings)
}

Next? It doesn't like the internal settings on my properties. Another setting:

var settings = new JsonSerializerSettings
{
   ConstructorHandling = ConstructorHandling.AllowNonPublicDefaultConstructor,
   ContractResolver = new DefaultContractResolver { DefaultMembersSearchFlags = BindingFlags.Instance | BindingFlags.Public | BindingFlags.NonPublic | BindingFlags.FlattenHierarchy },
};

I'd specifically like to point out that, despite having used BindingFlags hundreds of times, they are still obscure and confusing to me...I generally take 5 tries before I get something that works.

With those problems solved, its now having a hard time deserializing my interfaces. I don't blame it..how the heck is it supposed to know that LdapConfiguration should be used whenever ILdapConfiguration is specified? So we have to add some converters. I came up with a simple utility class to do this:

public T FromJson<T>(string json)
{
   var settings = new JsonSerializerSettings
   {
      ConstructorHandling = ConstructorHandling.AllowNonPublicDefaultConstructor,
      ContractResolver = new DefaultContractResolver { DefaultMembersSearchFlags = BindingFlags.Instance | BindingFlags.Public | BindingFlags.NonPublic | BindingFlags.FlattenHierarchy },
   };
   settings.Converters.Add(new InterfaceConverter<ILdapConfiguration, LdapConfiguration>());
   settings.Converters.Add(new InterfaceConverter<IDataStoreConfiguration, DataStoreConfiguration>());
   return JsonConvert.DeserializeObject<T>(json, settings)
}

internal class InterfaceConverter<To, From> : Newtonsoft.Json.JsonConverter
{
   public override bool CanConvert(Type type)
   {
      return type == typeof(To);
   }
   public override void WriteJson(JsonWriter writer, object value, JsonSerializer serializer)
   {
      serializer.Serialize(writer, value);
   }

   public override object ReadJson(JsonReader reader, Type objectType, object existingValue, JsonSerializer serializer)
   {
      return serializer.Deserialize<From>(reader);
   }
}

And finally, with that, I've managed to walk through the verbosity and rigidness of both C# and the .NET framework.

Do you wanna see my Ruby version?

class Settings
  @@settings = YAML::load_file(Rails.root + 'config/config.yml')[Rails.env]

  class MissingSettingOptionError < StandardError;
  end

  def self.method_missing(key)
    raise MissingSettingOptionError, "#{key.to_s} is not in the config file" unless @@settings.include?(key.to_s)
    @@settings[key.to_s]
  end

end

That's it. Sure I'm leaning a little heavily on method_missing, but there's nothing stopping you from creating a more explicit interface:

class Settings
  ...
  def self.LdapUrl
    @@settings['ldap']['url']
  end
end

And you know what? Its just as easy to test, if not more so, than the C# version. Its also completely reusable from project to project. And it uses YAML which is such a nicer configuration format (which I didn't use in .NET 'cuz I didn't want yet another serialization library referenced).

So, before you criticize my code too harshly, I ask that you at least recognized that I, surprisingly, didn't bash anything. And I readily admit my .NET version might be over-engineered and that I'm even a little rusty.