# Monday, December 05, 2005

I'm still on this anonymous delegates kick. Maybe I'm on the wrong team and should consider a move to the C# team??

I've seen references to the feature of anonymous delegates as both "anonymous delegates" and "anonymous methods". From a philosophical standpoint, both seem like bad names. It had been bothering me for a while, so I decided to ask my fellow CLR buddy, Joe, the question. Joe is a good person to go to since he's a lot smarter than me. Joe handles concurrency and threading for our team. Joe was doing such a good job at concurrency that they decided that he could handle another job ;)

Anyway, Joe agreed that both names are misleading. Here's what we came up with for what the features would be if you interpreted the names literally:

- anonymous delegates: the ability to define a delegate (the actual declaration) in a more dynamic manner, in which you do not need to statically define the signature/contract of the delegate up-front. Joel and I did a quick co-routines implementation on top of the framework last summer. My first thought was to use delegates. This failed pretty quickly given that you need to delegate signature to be something better than "public delegate Object[] CoRoutineDelegate(Object[])".

- anonymous methods: the ability to define and generate a method in a more dynamic manner, in which you do not need to statically define the signature/contact of the method up-front or care which class it sites on. All you really want here is some sort of handle/reference to that method that you can pass around and call through as needed.

The second definition here feels a lot more like what I've been calling anonymous delegates. In fact, that's more or less how this feature that we've been discussing actually works. I guess "anonymous methods" is really a better term.

Just for kicks, I decided to make a category on anonymous methods so that folks can link back to all the posts, since there are quite a few now. I'm hoping that this post was the last one on the subject, unless I get some interesting feedback from the C# team. With that, it's now time to get back to my factoring features in Whidbey series, which is in dire need of some attention.

Monday, December 05, 2005 7:07:09 PM (GMT Standard Time, UTC+00:00)  #    Disclaimer  |  Comments [0]  | 
# Saturday, December 03, 2005

I recently posted twice (one and two) on anonymous delegates. I guess they really caught my attention for some reason, even though I've known about them for a long time. The remaining question to close out this debate is whether anonymous delegates are the .Net Framework answer to closures. Another related topic is continuations. I'm certain that these are not that.

First, what is a closure?

Here is how wikipedia defines closures:  "Unlike garden-variety functions which retain no memory of what happened in previous calls, closures are capable of storing information across function calls."

Here is another explanation from guile, which is an open source scheme interpreter that I've never heard of before: "The concept of closure is the idea that a lambda expression "captures" the variable bindings that are in lexical scope at the point where the lambda expression occurs. The procedure created by the lambda expression can refer to and mutate the captured bindings, and the values of those bindings persist between procedure calls. "

My definition is (as of 5 mins ago): a closure is the outcome of capturing both the in-scope variables used by a block of code and the block of code itself to the end of preserving their relationship beyond the normal lifetime of those in-scope local variables. As a result, the local variables can be shared across all calls to the block of code (i.e. method). 

So, do anonymous delegates in C# v2.0 conform to the definition of lexical closures? That seems to be a matter of great debate. It is amazing when you start to look at a topic and then you realize that there has been a whole crowd to the party before you. Anonymous delegates in C# appear to be just such a topic. Brad appears to have been interested in the topic about a year ago. Brad takes the stance (at least at that time) that anonymous delegates are not closures. abhinaba also takes the same stance. Dan Muller takes the opposite opinion, believing that C# has implemented closures.

I've come to the conclusion that anonymous delegates are closures. If you look at my somewhat-in-depth explanation of anonymous delegates in my last post, you'll see that the way they work seems to conform to the three definitions above. In summary, the C# compiler generates a class on your behalf which contains both an instance method (which is your anonymous delegate) and fields that map directly to the in-scope variables that you access in the containing method. All access to these variables are wired up to these generated fields on the class. That generated class is in itself the closure. Being heap-allocated, it naturally can have a different lifetime than what the containing method has on the stack, given that stack-frames have a tendency to go way.

I have a mail out to the C# team asking them their opinion on this interesting subject. I'm curious to see how they'll respond.

Saturday, December 03, 2005 5:40:06 AM (GMT Standard Time, UTC+00:00)  #    Disclaimer  |  Comments [1]  | 

Here is a better example of using an anonymous delegate that proves, in my mind, that they are lexical closures. I sure hope my definition of closures is correct ;)

Here's the code. Please do guess what it does or better yet, compile and run it. The important aspect is to guess which values of "s" do or do not match.

using System;

using System.Threading;





namespace AnonDelegateTest2

{

class Program

{

delegate void FunkyDelegate();



static void Main(string[] args)

{

FunkyDelegate fd = GetDelegate();



for (Int32 i = 0; i < 10; i++)

{

fd();

}

}



static FunkyDelegate GetDelegate()

{

String s = DateTime.Now.Ticks.ToString();



FunkyDelegate fd = delegate()

{

Console.WriteLine("Entering fd");

Console.WriteLine(s);

s = DateTime.Now.Ticks.ToString();

Thread.Sleep(100);

Console.WriteLine(s);

Console.WriteLine("Leaving fd");

Console.WriteLine();

};



return fd;

}

}



}
Saturday, December 03, 2005 2:06:27 AM (GMT Standard Time, UTC+00:00)  #    Disclaimer  |  Comments [0]  | 
# Wednesday, November 30, 2005

I posted an interesting bit of C# a couple weeks ago relating to anonymous delegates. I asked in that post what the result of the following code is. Here is the code and then the result

Code:
using System;
using System.Threading;

namespace AnonMethods
{
    class Program
    {
        static void Main(string[] args)
        {
            for (Int32 i = 0; i < 20; i++)
            {
                Thread t = new Thread(delegate() { Thread.Sleep(10); Console.WriteLine(i); });
                t.Start();
            }
            Thread.Sleep(5000);
        }
    }
}

Result:
6
6
6
6
6
9
12
12
12
13
13
13
15
15
19
20
20
20
20
20

Why?
That's the strangest result from a for loop that I've ever seen. First, I'd suggest that you compile the code for yourself (remember, C# Express is free) and then take and then decompile it in either ildasm or Reflector. I find the C# in Reflector a whole lot easier to read than the IL, so I'd recommend that you start with it. I imagine that you'll get a real kick out of what you see.

First, I'd like to caveat my explanation by saying that this is a CLR guy's explanation of what the C# compiler is doing. I haven't talked to the C# team, but just took a look at the code emitted by the compiler, which both you and I are free to ponder. All that being said, here's the basic idea of what I think is going  on without going into great detail of the mechanics of anonymous delegates.

Once the C# source code is compiled, the anonymous delegate goes away and is replaced by the mechanism that I'm about to describe. Put another way, anonymous delegates are a C# source-code-only feature. For every anonymous delegate, there is a strangely, but probably predictably, named nested class added to the class that indirectly contained (within one of the methods within the class) the anonymous delegate. The class, as you might expect, exposes some surface area that is accessed from the parent class. The class is actually private, which means it cannot be accessed from anywhere but the parent class. There is a default constructor, which does nothing. There is a method that is the anonymous delegate transformed into a regular named method. Remember, anonymous delegates are a C#, not a CLR feature, so the CLR does need a named method to call. Lastly, there are a set of fields exposed on this compiler-generated class -- and this is the part that gets interesting -- that are the local variables that are accessed from both the containing method and the anonymous delegate. Variables that are defined in the containing method but not accessed from within the anonymous delegate or that are defined and used within the anonymous delegate do not get this special treatment. As you might guess, the C# compiler wires up the method that takes the delegate and the generated named method. In addition, all accesses to the shared local variables are wired up through the public fields on the generated class instead of through the actual locals. Very interesting.

OK, so that's the basic explananation of how anonymous delegates work. There are some other subtleties that I noticed, but they are not important for this explanation. I'm going to get further insight on those from the C# team and will post on those when I learn more about them.

So, then, how does all that lead to the strange and unexpected results from the for loop. Well, since the "i" variable is accessed by both the containing method and the anonymous delegate, the C# compiler rewires all accesses to "i" to the "i" field on the generated class. And since each call to the anonymous delegate in our example does its work on a separate thread, then "i" is updated by the for loop on a different schedule than it is written by Console.WriteLine in the anonymous delegate. That's why the value printed out by each call to the anonymous delegate is essentially a race condition between the for loop continually iterating and threads being created and getting time on the processor(s).

Another aspect of this is that shared locals with anonymous delegates essentially turn value types into reference types. That's a really strange behaviour and realization.

Like I said in the earlier post, the trick is determining a scenario where this behaviour is useful. I haven't found that yet, although I'm sure it will be very interesting once I find it.

Wednesday, November 30, 2005 3:51:11 PM (GMT Standard Time, UTC+00:00)  #    Disclaimer  |  Comments [5]  | 
# Friday, November 11, 2005

What does the following C# code do? Guess and then run it.

using System;
using System.Threading;

namespace AnonMethods
{
class Program
{
static void Main(string[] args)
{
for (Int32 i = 0; i < 20; i++)
{
Thread t = new Thread(delegate() { Thread.Sleep(10); Console.WriteLine(i); });
t.Start();
}
Thread.Sleep(5000);
}
}
}

We were playing with this and similar code in the experts area at the Ottawa VS launch yesterday. Fun, fun.

My task ahead is to find a scenario that makes the subtleties of anonymous delegates useful. Anonymous delegates are certainly useful. It is the trick above with "i" that I'm talking about. I'll post in a couple days with more data on what is going on.

Friday, November 11, 2005 2:51:50 PM (GMT Standard Time, UTC+00:00)  #    Disclaimer  |  Comments [1]  |