This is a pretty risky thing to do.
While it's true that you can serialize and deserialize a delegate just like any other object, the delegate is a pointer to a method inside the program that serialized it. If you deserialize the object in another program, you'll get a SerializationException
- if you're lucky.
For instance, let's modify darin's program a bit:
class Program
{
[Serializable]
public class Foo
{
public Func<string> Del;
}
static void Main(string[] args)
{
Func<string> a = (() => "a");
Func<string> b = (() => "b");
Foo foo = new Foo();
foo.Del = a;
WriteFoo(foo);
Foo bar = ReadFoo();
Console.WriteLine(bar.Del());
Console.ReadKey();
}
public static void WriteFoo(Foo foo)
{
BinaryFormatter formatter = new BinaryFormatter();
using (var stream = new FileStream("test.bin", FileMode.Create, FileAccess.Write, FileShare.None))
{
formatter.Serialize(stream, foo);
}
}
public static Foo ReadFoo()
{
Foo foo;
BinaryFormatter formatter = new BinaryFormatter();
using (var stream = new FileStream("test.bin", FileMode.Open, FileAccess.Read, FileShare.Read))
{
foo = (Foo)formatter.Deserialize(stream);
}
return foo;
}
}
Run it, and you'll see that it creates the object, serializes it, deserializes it into a new object, and when you call Del
on the new object it returns "a". Excellent. Okay, now comment out the call to WriteFoo
, so that the program it's just deserializing the object. Run the program again and you get the same result.
Now swap the declaration of a and b and run the program. Yikes. Now the deserialized object is returning "b".
This is happening because what's actually being serialized is the name that the compiler is assigning to the lambda expression. And the compiler assigns names to lambda expressions in the order it finds them.
And that's what's risky about this: you're not serializing the delegate, you're serializing a symbol. It's the value of the symbol, and not what the symbol represents, that gets serialized. The behavior of the deserialized object depends on what the value of that symbol represents in the program that's deserializing it.
To a certain extent, this is true with all serialization. Deserialize an object into a program that implements the object's class differently than the serializing program did, and the fun begins. But serializing delegates couples the serialized object to the symbol table of the program that serialized it, not to the implementation of the object's class.
If it were me, I'd consider making this coupling explicit. I'd create a static property of Foo
that was a Dictionary<string, Func<string>>
, populate this with keys and functions, and store the key in each instance rather than the function. This makes the deserializing program responsible for populating the dictionary before it starts deserializing Foo
objects. To an extent, this is exactly the same thing that using the BinaryFormatter
to serialize a delegate is doing; the difference is that this approach makes the deserializing program's responsibility for assigning functions to the symbols a lot more apparent.