Dare’s Holy Grail vs. Rory’s “Huh?”

Dare and Rory and others have been conversing on a subject very close to my heart (and my work) — mapping XML code to objects.

In his post, Dare demonstrates how having dynamically typed properties would make it easier (and more elegant) for developers to write code vs. doing the same thing in a statically typed language. He then laments that there is no system today that would combine both strongly and dynamically typed features.

Although JavaScript is not a strongly typed language, it is a dynamically typed one. I thought it might be interesting for y’alls to continue exploring the building of a type dynamically. I make a weak attempt of strengthening the type system, but it is not something that should mislead you to think that JS type system is suddenly “cured“.

Consider the following code fragment (complete code and results of its execution can be seen here):

var entry = new entryType("Happy New Year!", "1/1/2004");
entry.subject = new typedXmlElement("Happy New Year, Yay!", "xs:string", "dc");
entry.requireModeration = true;
entry.subject.longDesc = new typedXmlElement("Common holiday with gifts and stuff", "xs:string", "myns");
entry.subject.longDesc.modifiedOn = new Date();
// write complete XML of this entry
var xml = entry.serialize();

In my humble opinion, the beauty of dynamic types is that it elevates a type instance to the level of a type and brings the sense of evolution to type. One can see it as bringing more “realism“ to the OOP: I am an instance of a Person, but, though sharing the same traits, a generic Person and I are different — I am a semantic singleton, a one-of-a-kind type-and-instance-in-one. As I evolve, I gain (or lose) new methods and properties, yet remain the same instance that I was born.

I may be wrong (or under-educated), but it seem that SOA, the “next big thing“ looming on the horizon, would benefit from  dynamically and strongly typed language of some kind.

DevDays 2004, Atlanta

Sittin’ in the hotel room, looking through the freshly acquired DevDays materials… Although the folks who organized it managed to put together a good collection of information and links to more information, I can’t shake off a feeling that I’ve watched a trailer rather than a full-length feature. It seems that the speakers, having only an hour to talk, sped up past the meat of the topic and just ran through the basics. I guess I was really expecting more from the threat modeling talk, but I understand how it can’t be covered within one hour.

On a positive note, I did have a good conversation with Jeff Prosise about the process of book writing. Also met Doug Turnure, who also recently started his blog and seems to be having the same “non-commental” issues. I can’t offer any good advice, except, to paraphrase Don Park’s post:

 “Keep feeding your Tamagotchi“.

Crashing Through Deadlocks

Although concurrency management is somewhat represented in C# with the lock statement, there’s plenty more to be found in System.Threading namespace. Starting from Monitor class, which is what the lock statement is based on, to the ReaderWriterLock and the Mutex class. There’s even an Interlocked class, which provides very nice facilities for atomic operations.

In debugging deadlocks, I found myself using the Monitor.TryEnter method in order to determine the actual place of a deadlock. It basically operates as your standard Enter, except after a specified time span, it times out and returns false as the return value. Otherwise, when the thread is acquired, it returns true.

Inspired by a neat hack by Peter Golde, I created a simple wrapper — the ImpatientLock — for the TryEnter method, which allows adding the deadlock debugging statement with minimal modification to the existing code:

// lock(obj)
using(ImpatientLock.Try(obj))
{
   // .. do stuff
}

Except in the case of ImpatientLock, the exception is thrown, documenting the stack trace and thus pointing out location of the deadlock. Here’s the complete source code for the ImpatientLock, just in case you find it useful.

using System.Threading;

public class ImpatientLock
{
  public static IDisposable Try(object obj)
  {
    return new DisposableLock(obj, DisposableLock.DefaultTimeSpan);
  }

  public static IDisposable Try(object obj, TimeSpan timeSpan)
  {
    return new DisposableLock(obj, timeSpan);
  }

  private class DisposableLock : IDisposable
  {
    private static readonly TimeSpan DefaultTimeSpan = TimeSpan.FromMinutes(5);
    private readonly object Obj;
    public DisposableLock(object obj, TimeSpan timeSpan)
    {
      Obj = obj;
      if (!Monitor.TryEnter(obj, timeSpan))
      { 
          throw new ImpatientLockException();
      }
    }

    public void Dispose()
    {
      Monitor.Exit(Obj);
    }
  }

  private class ImpatientLockException : Exception
  {
    public ImpatientLockException()
      :base("Your time's up. I am quittin'!")
    {
    }
  }
}

In case you are battling deadlock issues, there are a couple of really good articles on the subject — most notably this one by Dr. GUI, and this one by unknown (to me) MS folks. There are also some pretty intriguing applications that attempt detection of potential deadlocks by analyzing the call tree of your code, although I found those too simplistic to detect anything realistically serious.

UPDATE: Ian Griffith has more information about this technique over at his blog, enhanced and updated with the help of Eric Gunnerson.

Regrettable Software, Tossed Salad, and Web Development

I’ve done some reading (and posting) on the excellent ASP.NET forum and, by far it seems, the most frequent cause of the issues that people have with their code is making the distinction between server side and client side of a Web application. I am not exactly sure why. Maybe the problem stems from the fact that most of today’s Web development platforms allow (and encourage) using the “tossed salad” declaration of server side and client side: your server-side code is mixed together with your client-side code.

ASP, ASP.NET, ColdFusion, JSP, PHP, and other scripting language platforms all attempt to marry DHTML and server-side logic in a single file. I am not saying that they are doing it wrong — there are far too many more of them than just me. But looking at posts full of confusion and hacking burn-out makes me wonder if the cow of architecture and strategic thinking was sacrificed to the gods of convenience.

It stands to reason whether it would be better to engineer your development platform in a way that would encourage architectural thinking rather than aid the hapless member of a cargo cult in quickly hacking something together. Something that inevitably becomes a piece of “regrettable software” — software that is painful to build, painful to use, and painful to maintain.

In the meantime, there’s definitely a market for a “know thy (sever|client) side” training. Speaking of which, trying to learn by coding in one of the “tossed salad” platforms may lead to a Gordian Knot of concepts in the students’ head. So, if you are trying to learn Web development platform from scratch, be that ASP.NET or ColdFusion MX, make sure to keep the layers separate:

  • Start from HTML
  • Grok CSS
  • Learn JavaScript
  • Move on to the server side language

Just not all at once.

Representing xsd:choice element in C#

The issue of correctly representing semantics of an xsd:choice element seemed like a good challenge, so I decided to tackle it.

Per XML Schema reference, xsd:choice is a group element, which allows one and only one of the elements contained in the group to be present within the containing element. If you try to translate this statement into the rough CLR terms, the parent instance of a type may contain only one instance out of a defined group of types. For example, suppose we have an instance of a Boat type, which can hold only a instance of a Cabbage type, an instance of a Goat type, or an instance of a Wolf type, but not any combination of those instances. What OOP concept does this sound like? That’s right, polymorphism. Let’s imagine an abstract class Load, from which Cabbage, Goat, and Wolf all inherit. Then, our declaration will look like so:

class Boat { public Load load; }

class Load { /* ... */ }

class Cabbage : Load { /* specific features of Cabbage */ }

class Goat : Load { /* specific features of Goat */ }

class Wolf : Load { /* specific features of Wolf */ }

Simple enough? Polyphormism-shmolymorphism, and we got ourselves a nice and semantically correct representation of a choice element. Unfortunately, as it so often happens in reality, all three of our instances competing for the coveted place on the Boat, must also inherit some of their traits from other base classes. For instance, Cabbage is also a Vegetable, Goat is a Herbivore, and Wolf is a Carnivore:

class Cabbage : Load, Vegetable { /* ... */ }

class Goat : Load, Herbivore { /* ... */ }

class Wolf : Load, Carnivore { /* ... */ }

As Steve Saxon pointed out in a comment to my previous post, you can’t do that in CLR — a type may inherit from only one base type. In these situations, you can simulate multiple inheritance using interfaces. Let’s redefine our Load class type as an interface:

interface ILoad {}

This interface is not doing much — it is there only to indicate that an instance of the type that implements this interface may be a load on the boat. Now, Cabbage, Goat, and Wolf can all both inherit their traits from base classes and also be marked as potential suitors for the boat ride:

class Cabbage : ILoad, Vegetable { /* ... */ }

class Goat : ILoad, Herbivore { /* ... */ }

class Wolf : ILoad, Carnivore { /* ... */ }

class Boat { public ILoad load; }

Now, let’s look at the definition of the Boat type. Although it can clearly carry one instance of ILoad, there is no indication of that exactly the load is — as the instances are cast into ILoad, all of their specifics are hidden by the marker interface. The type information is still available — all you need to do is to attempt to cast the instance of the load to one of the three types we’ve defined, and one of them is bound to be the answer, unless the boat is empty. Trying to be good programmers, we don’t want to leave this task entirely up to the developers who come behind us. So, with a few modifications, we can add management of boat loading and unloading. Here’s what we’ll do:

Inside Boat type, create a nested class called LoadManager:

public class Boat
{
/* ... */
public class LoadManager {}
}

This new type will contain one private field of type ILoad:

public class LoadManager
{
private ILoad _load;
}

Create accessor properties for each of our riverside travelers:

public class LoadManager
{
private ILoad _load;
public Cabbage Cabbage { get return _load as Cabbage; set _load = value; }
public Goat Goat { get return _load as Goat; set _load = value;}
public Wolf Wolf { get return _load as Wolf; set _load = value; }
}

As you can see, the logic of the LoadManager is slightly different from the plain ILoad implementation — now you have a 3 doors of different shapes leading to the same room instead of 1 door that changes shape depending on the entrant, yet they represent the same structure semantically.

To complete our journey, let’s modify the Boat type to have a property of LoadManager type rather than a field of ILoad type (we’ll also throw in lazy initialization just for kicks):

public class Boat
{
/* ... */
private LoadManager _loadManager;
public LoadManager Load { get if (_loadManager == null) _loadManager = new LoadManager(); return _loadManager; }
}

Whew! That was a long trip. However, at the end of the road, we’ve together some pretty code (full listing can be found here):

Boat boat = new Boat();
Console.WriteLine("Loading Cabbage...");
boat.Load.Cabbage = new Cabbage();
Console.WriteLine("Loading Goat...");
boat.Load.Goat = new Goat();
Console.WriteLine("Load is Cabbage: " + boat.Load.IsCabbage); // not likely
Console.WriteLine("Load is Goat: " + boat.Load.IsGoat);

What’s next? Well, how about figuring out how to implement a CLR representation of minOccurs and maxOccurs attributes of the same xsd:choice element?…

Generating C# code from XSD

Steve Saxon had posted a few times on the XSD -> C# code generator that he’s building. Looks pretty cool. He also mentions that ideally some of the code should be generated as nested classes. It is true, some constructs such as “xsd:choice” are not easy to translate into an elegant C# equivalent. So, here’s an idea: what if we represent “xwd:choice” as its own class? Possibly a sub-class of some abstract XsdChoice implementation, this class provides solid implementation of properties for each element, but uses the same underpinnings to check the facets and provide the actual “choice” framework. Same for the simple types with restrictions. Then, the public definition of Steve’s ProductType class may go something like this :

public class ProductType : XsdComplexType
{
   public int number {get;set;}
   public howBigChoice howBig {get;}
   public howBig2Choice howBig2 {get;}
   public zipCodeSimpleType {get;}
   public DateTime effDate {get;}
}

public class howBigChoice : XsdChoice
{
   public sizeType size {get;}
   public int height {get;set;}
}

public class howBig2Choice : XsdChoice
{
// same here
}

public class sizeType : XsdType
{
  public bool isValid(int size);
  public int Value {get;set;}
}

public class zipCodeType : XsdType
{
  public bool isValid(string zipCode);
  public string Value {get;set;}
}

What do you think?

TrackBack

HTML, CSS and Other Curious Stuff That You May Find Hard-coded In Your Web Application

Design View Dude is asking interesting questions about composing HTML/CSS mark-up in the process of building a .NET Web application.

Ok, I may be a little too radical on this, but IMHO the controls should emit as little HTML mark-up as possible. Ideally, none. I am puzzled at how such great concepts as XSLT and XML-driven rendering are completely (well, aside from the Xml control, which doesn’t really count) ignored in the .NET strategy of building Web content. In proper development environment, the producer of HTML code (the one with visual design skills), and software developer (the one with application architecture skills) are rarely the same person. It would be only logical to keep their activities separate.

So, if you ask me, I would like to see framework built in such a way that content (this includes UI) is emitted in XML, which then can be freely shaped into desired context using XSLT.

That way, we don’t have to worry about script being injected in the wrong place or mark-up being non-XHTML-compliant.

Similarly, we don’t have worry about application developers designing the presentation UI and HTML being hard-coded into the logic of the software.

It will also ease your job in developing visualizers for the development environment — all you need is a test harness stylesheet, which takes emitted XML and creates a usable Web application out of it. 

Think about it — skinning and tweaking the look and feel would no longer be connected to the application development cycle.