Let's dissect your code sample:
filenames.SelectMany(f =>
Assembly.LoadFrom(f).GetCustomAttributes(typeof(PluginClassAttribute), true)
.Cast<PluginClassAttribute>()
.Select(a => a.PluginType)
).ToList();
So, we start off with a string[] called filenames. We invoke the SelectMany extension method on the array, and then we invoke ToList on the result:
filenames.SelectMany(
...
).ToList();
SelectMany takes a delegate as parameter, in this case the delegate must take one parameter of the type string as input, and return an IEnumerable<T> (Where the type of T is inferred). This is where lambdas enter the stage:
filenames.SelectMany(f =>
Assembly.LoadFrom(f).GetCustomAttributes(typeof(PluginClassAttribute), true)
).ToList()
What will happen here is that for each element in the filenames array, the delegate will be invoked. f is the input parameter, and whatever comes to the right of => is the method body that the delegate refers to. In this case, Assembly.LoadFrom will be invoked for filename in the array, passing he filename into the LoadFrom method using the f argument. On the AssemblyInstance that is returned, GetCustomAttributes(typeof(PluginClassAttribute), true) will be invoked, which returns an array of Attribute instances. So the compiler can not infer that the type of T mentioned earlier is Assembly.
On the IEnumerable<Attribute> that is returned, Cast<PluginClassAttribute>() will be invoked, returning an IEnumerable<PluginClassAttribute>.
So now we have an IEnumerable<PluginClassAttribute>, and we invoke Select on it. The Select method is similar to SelectMany, but returns a single instance of type T (which is inferred by the compiler) instead of an IEnumerable<T>. The setup is identical; for each element in the IEnumerable<PluginClassAttribute> it will invoke the defined delegate, passing the current element value into it:
.Select(a => a.PluginType)
Again, a is the input parameter, a.PluginType is the method body. So, for each PluginClassAttribute instance in the list, it will return the value of the PluginType property (I will assume this property is of the type Type).
Executive Summary
If we glue those bits and pieces together:
// process all strings in the filenames array
filenames.SelectMany(f =>
// get all Attributes of the type PluginClassAttribute from the assembly
// with the given file name
Assembly.LoadFrom(f).GetCustomAttributes(typeof(PluginClassAttribute), true)
// cast the returned instances to PluginClassAttribute
.Cast<PluginClassAttribute>()
// return the PluginType property from each PluginClassAttribute instance
.Select(a => a.PluginType)
).ToList();
Lambdas vs. Delegates
Let's finish this off by comparing lambdas to delegates. Take the following list:
List<string> strings = new List<string> { "one", "two", "three" };
Say we want to filter out those that starts with the letter "t":
var result = strings.Where(s => s.StartsWith("t"));
This is the most common approach; set it up using a lambda expression. But there are alternatives:
Func<string,bool> func = delegate(string s) { return s.StartsWith("t");};
result = strings.Where(func);
This is essentially the same thing: first we create a delegate of the type Func<string, bool> (that means that it takes a string as input parameter, and returns a bool). Then we pass that delegate as parameter to the Where method. This is what the compiler did for us behind the scenes in the first sample (strings.Where(s => s.StartsWith("t"));).
One third option is to simply pass a delegate to a non-anonymous method:
private bool StringsStartingWithT(string s)
{
return s.StartsWith("t");
}
// somewhere else in the code:
result = strings.Where(StringsStartingWithT);
So, in the case that we are looking at here, the lambda expression is a rather compact way of defining a delegate, typically referring an anonymous method.
And if you had the energy read all the way here, well, thanks for your time :)