I've been working on a request
to the West Wind Application Configuration library
to add JSON support. The config library is a very easy
to use code-first approach
to configuration: You
create a class that holds the configuration data that inherits from a base configuration class, and then assign a persistence provider at runtime that determines where and
how the configuration data is store. Currently the library supports .NET Configuration stores (web.config/app.config), XML files, SQL records and string storage.About once a week somebody asks me about JSON support and I've deflected this question for the longest time because frankly I think that JSON as a configuration store doesn't really buy a heck of a lot over XML. Both formats require the user
to perform some fixup of the plain configuration data - in XML into XML tags, with JSON using JSON delimiters for properties and property formatting rules. Sure JSON is a little less verbose and maybe a little easier
to read if you have hierarchical data, but overall the differences are pretty minor in my opinion. And yet - the requests keep rolling in.Hard Link Issues in a Component LibraryAnother reason I've been hesitant is that I really didn't want
to pull in a dependency on
an external JSON library - in this case JSON.NET - into the core library. If you're not using JSON.NET elsewhere I don't want a user
to have
to require a hard dependency on JSON.NET unless they want
to use the JSON feature. JSON.NET is also sensitive
to versions and doesn't play nice with multiple versions when hard linked. For example, when you have a reference
to V4.4 in your project but the host application has a reference
to version 4.5 you can run into assembly load problems. NuGet's Update-Package can solve some of this *if* you can recompile, but that's not ideal for a component that's supposed
to be just plug and play. This is no criticism of JSON.NET - this really applies
to any dependency that might change. So hard linking the DLL can be problematic for a number reasons, but the primary reason is
to not force loading of JSON.NET unless you actually need it when you use the JSON configuration features of the library.Enter Dynamic LoadingSo rather than adding
an assembly reference
to the project, I decided that it would be better
to dynamically load the DLL at runtime and then use dynamic typing
to access various classes. This allows me
to run without a hard assembly reference and allows more flexibility with version number differences now and in the future.But there are also a couple of downsides:No assembly reference means only dynamic
access - no compiler type checking or IntellisenseRequirement for the host application
to have reference
to JSON.NET or else get runtime errorsThe former is minor, but the latter can be problematic. Runtime errors are always painful, but in this case I'm willing
to live with this. If you want
to use JSON configuration settings JSON.NET needs
to be loaded in the project. If this is a Web project, it'll likely be there already.So there are a few things that are needed
to make this work:Dynamically
create an instance and optionally attempt
to load
an Assembly (if not loaded)Load types into dynamic variablesUse Reflection for a few tasks like statics/enumsThe dynamic keyword in C# makes the formerly most difficult Reflection part - method calls and property assignments - fairly painless. But as cool as dynamic is it doesn't handle all aspects of Reflection. Specifically it doesn't deal with object activation, truly dynamic (string based) member activation or accessing of non instance members, so there's still a little bit of work left
to do with Reflection.Dynamic Object InstantiationThe first step in getting the process rolling is
to instantiate the type you need
to work with. This might be a two step process - loading the instance from a string value, since we don't have a hard type reference and potentially having
to load the assembly. Although the host project might have a reference
to JSON.NET, that instance might have not been loaded yet since it hasn't been accessed yet. In ASP.NET this won't be a problem, since ASP.NET preloads all referenced assemblies on AppDomain startup, but in other executable project, assemblies are just in time loaded only when they are accessed.Instantiating a type is a two step process: Finding the type reference and then activating it. Here's the generic code out of my ReflectionUtils library I use for this:/// <summary>
/// Creates
an instance of a type based on a string. Assumes that the type's
/// </summary>
/// <param name="typeName">Common name of the type</param>
/// <param name="args">Any constructor parameters</param>
/// <returns></returns>
public static object CreateInstanceFromString(string typeName, params object[] args)
{
object instance = null;
Type type = null;
try
{
type = GetTypeFromName(typeName);
if (type == null)
return null;
instance = Activator.CreateInstance(type, args);
}
catch
{
return null;
}
return instance;
}
/// <summary>
/// Helper routine that looks up a type name and tries
to retrieve the
/// full type reference in the actively executing assemblies.
/// </summary>
/// <param name="typeName"></param>
/// <returns></returns>
public static Type GetTypeFromName(string typeName)
{
Type type = null;
// Let default name binding find it
type = Type.GetType(typeName, false);
if (type != null)
return type;
// look through assembly list
var assemblies = AppDomain.CurrentDomain.GetAssemblies();
// try
to find manually
foreach (Assembly asm in assemblies)
{
type = asm.GetType(typeName, false);
if (type != null)
break;
}
return type;
}
To use this for loading JSON.NET I have a small factory function that instantiates JSON.NET and sets a bunch of configuration settings on the generated object. The startup code also looks for failure and tries loading up the assembly when it fails since that's the main reason the load would fail. Finally it also caches the loaded instance for reuse (according
to James the JSON.NET instance is thread safe and quite a bit faster when cached). Here's what the factory function looks like in JsonSerializationUtils:/// <summary>
/// Dynamically creates
an instance of JSON.NET
/// </summary>
/// <param name="throwExceptions">If true throws exceptions otherwise returns null</param>
/// <returns>Dynamic JsonSerializer instance</returns>
public static dynamic CreateJsonNet(bool throwExceptions = true)
{
if (JsonNet != null)
return JsonNet;
lock (SyncLock)
{
if (JsonNet != null)
return JsonNet;
// Try
to create instance
dynamic json = ReflectionUtils.CreateInstanceFromString("Newtonsoft.Json.JsonSerializer");
if (json == null)
{
try
{
var ass = AppDomain.CurrentDomain.Load("Newtonsoft.Json");
json = ReflectionUtils.CreateInstanceFromString("Newtonsoft.Json.JsonSerializer");
}
catch (Exception ex)
{
if (throwExceptions)
throw;
return null;
}
}
if (json == null)
return null;
json.ReferenceLoopHandling =
(dynamic) ReflectionUtils.GetStaticProperty("Newtonsoft.Json.ReferenceLoopHandling", "Ignore");
// Enums as strings in JSON
dynamic enumConverter = ReflectionUtils.CreateInstanceFromString("Newtonsoft.Json.Converters.StringEnumConverter");
json.Converters.Add(enumConverter);
JsonNet = json;
}
return JsonNet;
}This code's purpose is
to return a fully configured JsonSerializer instance. As you can see the code tries
to create an instance and when it fails tries
to load the assembly, and then re-tries loading.Once the instance is loaded some configuration occurs on it. Specifically I set the ReferenceLoopHandling option
to not blow up immediately when circular references are encountered. There are a host of other small config setting that might be useful
to set, but the default seem
to be good enough in recent versions. Note that I'm setting ReferenceLoopHandling which requires
an Enum value
to be set. There's no real easy way (short of using the cardinal numeric value)
to set a property or pass parameters from static values or enums. This means I still need
to use Reflection
to make this work. I'm using the same ReflectionUtils class I previously used
to handle this for me. The function looks up the type and then uses Type.InvokeMember()
to read the static property.Another feature I need is have Enum values serialized as strings rather than numeric values which is the default.
To do this I can use the StringEnumConverter
to convert enums
to strings by adding it
to the Converters collection.As you can see there's still a bit of Reflection
to be done even in C# 4+ with dynamic, but with a few helpers this process is relatively painless.Doing the actual JSON ConversionFinally I need
to actually do my JSON conversions. For the Utility class I need serialization that works for both strings and files so I created four methods that handle these tasks two each for serialization and deserialization for string and file.Here's what the File Serialization looks like:/// <summary>
/// Serializes
an object instance
to a JSON file.
/// </summary>
/// <param name="value">the value
to serialize</param>
/// <param name="fileName">Full path
to the file
to write out with JSON.</param>
/// <param name="throwExceptions">Determines whether exceptions are thrown or false is returned</param>
/// <param name="formatJsonOutput">if true pretty-formats the JSON with line breaks</param>
/// <returns>true or false</returns>
public static bool SerializeToFile(object value, string fileName, bool throwExceptions = false, bool formatJsonOutput = false)
{
dynamic writer = null;
FileStream fs = null;
try
{
Type type = value.GetType();
var json = CreateJsonNet(throwExceptions);
if (json == null)
return false;
fs = new FileStream(fileName, FileMode.Create);
var sw = new StreamWriter(fs, Encoding.UTF8);
writer = Activator.CreateInstance(JsonTextWriterType, sw);
if (formatJsonOutput)
writer.Formatting = (dynamic)Enum.Parse(FormattingType, "Indented");
writer.QuoteChar = '"';
json.Serialize(writer, value);
}
catch (Exception ex)
{
Debug.WriteLine("JsonSerializer Serialize error: " + ex.Message);
if (throwExceptions)
throw;
return false;
}
finally
{
if (writer != null)
writer.Close();
if (fs != null)
fs.Close();
}
return true;
}You can see more of the dynamic invocation in this code. First I grab the dynamic JsonSerializer instance using the CreateJsonNet() method shown earlier which returns a dynamic. I then
create a JsonTextWriter and configure a couple of enum settings on it, and then call Serialize() on the serializer instance with the JsonTextWriter that writes the output
to disk. Although this code is dynamic it's still fairly short and readable.For full circle operation here's the DeserializeFromFile() version:/// <summary>
/// Deserializes
an object from file and returns a reference.
/// </summary>
/// <param name="fileName">name of the file
to serialize to</param>
/// <param name="objectType">The Type of the object. Use typeof(yourobject class)</param>
/// <param name="binarySerialization">determines whether we use Xml or Binary serialization</param>
/// <param name="throwExceptions">determines whether failure will throw rather than return null on failure</param>
/// <returns>Instance of the deserialized object or null. Must be cast
to your object type</returns>
public static object DeserializeFromFile(string fileName, Type objectType, bool throwExceptions = false)
{
dynamic json = CreateJsonNet(throwExceptions);
if (json == null)
return null;
object result = null;
dynamic reader = null;
FileStream fs = null;
try
{
fs = new FileStream(fileName, FileMode.Open, FileAccess.Read);
var sr = new StreamReader(fs, Encoding.UTF8);
reader = Activator.CreateInstance(JsonTextReaderType, sr);
result = json.Deserialize(reader, objectType);
reader.Close();
}
catch (Exception ex)
{
Debug.WriteLine("JsonNetSerialization Deserialization Error: " + ex.Message);
if (throwExceptions)
throw;
return null;
}
finally
{
if (reader != null)
reader.Close();
if (fs != null)
fs.Close();
}
return result;
}This code is a little more compact since there are no prettifying options
to set. Here JsonTextReader is created dynamically and it receives the output from the Deserialize() operation on the serializer.You can take a look at the full JsonSerializationUtils.cs file on GitHub
to see the rest of the operations, but the string operations are very similar - the code is fairly repetitive.These generic serialization utilities isolate the dynamic serialization logic that has
to deal with the dynamic nature of JSON.NET, and any code that uses these functions is none the wiser that JSON.NET is dynamically loaded.Using the JsonSerializationUtils WrapperThe final consumer of the SerializationUtils wrapper is
an actual ConfigurationProvider, that is responsible for handling reading and writing JSON values
to and from files. The provider is simple a small wrapper around the SerializationUtils component and there's very little code
to make this work now:The whole provider looks like this:/// <summary>
/// Reads and Writes configuration settings in .NET config files and
/// sections. Allows reading and writing
to default or external files
/// and specification of the configuration section that settings are
/// applied
to.
/// </summary>
public class JsonFileConfigurationProvider<TAppConfiguration> : ConfigurationProviderBase<TAppConfiguration>
where TAppConfiguration: AppConfiguration, new()
{
/// <summary>
/// Optional - the Configuration file where configuration settings are
/// stored in. If not specified uses the default Configuration Manager
/// and its default store.
/// </summary>
public string JsonConfigurationFile
{
get { return _JsonConfigurationFile; }
set { _JsonConfigurationFile = value; }
}
private string _JsonConfigurationFile = string.Empty;
public override bool Read(AppConfiguration config)
{
var newConfig = JsonSerializationUtils.DeserializeFromFile(JsonConfigurationFile, typeof(TAppConfiguration)) as TAppConfiguration;
if (newConfig == null)
{
if(Write(config))
return true;
return false;
}
DecryptFields(newConfig);
DataUtils.CopyObjectData(newConfig, config, "Provider,ErrorMessage");
return true;
}
/// <summary>
/// Return
/// </summary>
/// <typeparam name="TAppConfig"></typeparam>
/// <returns></returns>
public override TAppConfig Read<TAppConfig>()
{
var result = JsonSerializationUtils.DeserializeFromFile(JsonConfigurationFile, typeof(TAppConfig)) as TAppConfig;
if (result != null)
DecryptFields(result);
return result;
}
/// <summary>
/// Write configuration
to XmlConfigurationFile location
/// </summary>
/// <param name="config"></param>
/// <returns></returns>
public override bool Write(AppConfiguration config)
{
EncryptFields(config);
bool result = JsonSerializationUtils.SerializeToFile(config, JsonConfigurationFile,false,true);
// Have
to decrypt again
to make sure the properties are readable afterwards
DecryptFields(config);
return result;
}
}This incidentally demonstrates
how easy it is
to create a new provider for the West Wind Application Configuration component. Simply implementing 3 methods will do in most cases.Note this code doesn't have any dynamic dependencies - all that's abstracted away in the JsonSerializationUtils(). From here on, serializing JSON is just a matter of calling the static methods on the SerializationUtils class.Already, there are several other places in some other tools where I use JSON serialization this is coming in very handy. With a couple of lines of code I was able
to add JSON.NET support
to an older AJAX library that I use replacing quite a bit of code that was previously in use. And for any other manual JSON operations (in a couple of apps I use JSON Serialization for 'blob' like document storage) this is also going
to be handy.Performance?Some of you might be thinking that using dynamic and Reflection can't be good for performance. And you'
d be right… In performing some informal testing it looks like the performance of the native code is nearly twice as fast as the dynamic code. Most of the slowness is attributable
to type lookups.
To test I created a native class that uses
an actual reference
to JSON.NET and performance was consistently around 85-90% faster with the referenced code. This will change though depending on the size of objects serialized - the larger the object the more processing time is spent inside the actual dynamically activated components and the less difference there will be. Dynamic code is always slower, but
how much it really affects your application primarily depends on
how frequently the dynamic code is called in relation
to the non-dynamic code executing. In most situations where dynamic code is used '
to get the process rolling' as I do here the overhead is small enough
to not matter.All that being said though - I serialized 10,000 objects in 80ms vs. 45ms so this is hardly slouchy performance. For the configuration component speed is not that important because both read and write operations typically happen once on first
access and then every once in a while. But for other operations - say a serializer trying
to handle AJAX requests on a Web Server one would be well served
to create a hard dependency.Dynamic Loading - Worth it?Dynamic loading is not something you need
to worry about but on occasion dynamic loading makes sense. But there's a price
to be paid in added code and a performance hit which depends on
how frequently the dynamic code is accessed. But for some operations that are not pivotal
to a component or application and are only used under certain circumstances dynamic loading can be beneficial
to avoid having
to ship extra files adding dependencies and loading down distributions. These days when you
create new projects in Visual Studio with 30 assemblies before you even add your own code, trying
to keep file counts under control seems like a good idea. It's not the kind of thing you do on a regular basis, but when needed it can be a useful option in your toolset… © Rick Strahl, West Wind Technologies, 2005-2013Posted in .NET C#
Tweet
!function(d,s,id){var js,fjs=d.getElementsByTagName(s)[0];if(!d.getElementById(id)){js=d.createElement(s);js.id=id;js.src="//platform.twitter.com/widgets.js";fjs.parentNode.insertBefore(js,fjs);}}(document,"script","twitter-wjs");
(function() {
var po = document.createElement('script'); po.type = 'text/javascript'; po.async = true;
po.src = 'https://apis.google.com/js/plusone.js';
var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(po, s);
})();
