May 28 2009

The Golden Boy

A friend of mine recently made a comment to me, and I’ve been thinking about it quite a bit since. He was commenting on how the management at his company was suffering from the “hero syndrome” where any employee that makes some sort of noteworthy accomplishment suddenly becomes the hero of the company. Once someone is the hero, all things go through that employee. Any new ideas are brought to the hero first. Any new interesting work is assigned to the hero. heroBasically, the hero is viewed as the most important employee in the company. We used to have a different name for this at a company I worked at years ago. We called it “the golden boy”. It turns out this is a very common mindset. I can think of more then a few companies that I’ve worked at that have had a similar mentality. Sometimes I’ve been the golden boy. Sometimes… not so much.

My wife pointed out that this attitude runs on a much deeper level then just the workplace. We love movies about the lone hero who is able to accomplish what no one else can. We like our sports team, but love the star player.

The really interesting part to me is that the golden boy torch seems to be passed quite frequently. Today’s hero may be replaced tomorrow when another employee meets a tight deadline or creates an impressive piece of technology. The real problem with this mentality is that it limits a companies potential. Tomorrows golden boy is certainly no less intelligent or less talented today, but by not giving him/her the same opportunities a company is artificially limiting what it can achieve.

My friend’s conclusion was that the real goal should be to cultivate a sense of team, and I think he’s spot on. It’s terribly cliché, but a team will always out perform an individual. The problem with individuals is that no matter how incredible they are, at some point they will fail. Humans are just not capable of perfection. If you ever read through the Old Testament, it’s full of stories about individuals. However, almost every one of them is near fatally flawed. The Jewish people still consider King David as one of their greatest heroes, but he killed a man in order to steal his wife (resulting in all sorts of dysfunction in his family). Multiple ethnic groups consider Abraham to be their patriarch, but he actually gave his wife away to other men on two separate occasions in order to save his own neck. I think the point is that great men are still just that… men. And people are defective at best. 

Teams are not perfect either, but with more then one voice the likely hood of major oversights or completely self serving decisions is put in check. One of the hardest jobs I ever had was where I was the only developer. I agonized over every technical decision because I didn’t have anyone to bounce my thoughts off of. I was always afraid that I was missing something obvious. I think every developer has had the experience of architecting a beautiful solution to some complex problem, only to have a co-worker point out that the same thing could have been done in two lines of code. This is the same beauty in the architecture of the American governing system. Decisions about the direction of the country are made by a team. You can argue how good of a team our government is, but it seems to me that it’s still highly preferable to a monarchy.

So the question then is how do we avoid the trap of making certain individuals into heroes and instead cultivate a sense of team? This is where you come in. I’d love to hear your thoughts. However, I think the first step is to stop the golden boy mentality. It seams to be the antithesis of team.

  • Share/Save/Bookmark

May 17 2009

End “Null Reference” Now

Stop Nulls

Here’s an interesting statistic, 87.35% of exceptions that occur in the wild are “null reference” exceptions. Here’s another interesting stat, 74.32% of all statistics are made up.

Ok, so I have no idea what the actual percentages are, but my experiance is that “null reference” is hands down the most common exception that I run in to. Well, I’d like to see that put to an end. It’s 2009. We’ve put a man on the moon. We can solve this.

Now I’m no expert when it comes to creating programming languages, but I’d love to see the compiler help out with this. Imagine something like this (notice the attribute in the parameter decleration)…

public void MyNullSafeFunction([NonNull]object value)
{
     // put code here with a smile
     // because you know that "value" can't be null.
}

And then later when some idiot (most likely me) tries to code this…

object value = null;
MyNullSafeFunction(value);

…the compiler would give a nice friendly reminder that I can’t pass a null value as the first argument to MyNullSafeFunction. Infact, there’s no reason it couldn’t even be smart enough to handle this…

object value = null;

if (someVariableThatMyOrMayNotEqualTrue)
{
     value = 4;
}

MyNullSafeFunction(value);

Again, the compiler could warn me that the variable “value” cannot be garanteed to be non-null in all cases. But, what about this case…

object value = FunctionFromLibraryIDoNotHaveSourceCodeFor();
MyNullSafeFunction(value);

Sure, the compiler could analyse the function to verify that it couldn’t possibly return null, but that’s just not very efficent. Especially if that functions return value is dependant on the return value from another function, which is dependant on the return value from another function, which is dependant on the return value from another function, which is dependant on the return value from another function, which is dependant on the return value from another function, which is dependant on the return value from another function. This is where things get harder. But, what if FunctionFromLibraryIDoNotHaveSourceCodeFor had a signature like this…

[NonNull]
public object FunctionFromLibraryIDoNotHaveSourceCodeFor()
{
     // really awesome code here...
}

Ah, problem solved. The compiler could once again verify that the function would not return null in an efficent manner.

However, this is where the problems start to come in. This would require the small task of retro fitting years worth of code with [NonNull] attributes. Also, it could easily end up like the c++ “const” keyword (i.e. really helpful… if everyone uses it). If half the development community adopted the practice and the other half didn’t it could end up being a real nightmare.

However, this isn’t reality land. This is blog-ality land where I can happily ignore the messy details. And so, this marks the begining of my offical campaign to end null references errors forever (I’ll leave working out the details as an excercise for the implementors).

  • Share/Save/Bookmark

May 16 2009

Dynamic… But Fast: The Tale of Three Monkeys, A Wolf and the DynamicMethod and ILGenerator Classes

 

NOTE: This is an article that I originally wrote for codeproject.com. You can view the original article here…

http://www.codeproject.com/KB/database/DynamicMethod_ILGenerator.aspx

 

Once upon a time, there were three little code monkeys. Monkey number one worked at the straw factory out on 7th and Penn. One day the first monkey’s boss, B. B. Wolf, brought him a new assignment. The HR department’s application was being upgraded to C#. Monkey number one’s job was to write the code to populate the Person class with data from the database. The monkey got straight to work and quickly produced code similar to the following (the actual code could not be used for legal reasons):

public class ManualBuilder
{
    public Person Build(SqlDataReader reader)
    {
        Person person = new Person();

        if (!reader.IsDBNull(0))
        {
            person.ID = (Guid)reader[0];
        }

        if (!reader.IsDBNull(1))
        {
            person.Name = (string)reader[1];
        }

        if (!reader.IsDBNull(2))
        {
            person.Kids = (int)reader[2];
        }

        if (!reader.IsDBNull(3))
        {
            person.Active = (bool)reader[3];
        }

        if (!reader.IsDBNull(4))
        {
            person.DateOfBirth = (DateTime)reader[4];
        }

        return person;
    }
}

This code worked well at first. It was clean and really fast. However, the head of the HR department decided that the new software should really have a few additional features. Every new feature seemed to either require adding fields to the Person table, creating a new table, or moving fields from the Person table to one of the new tables. Every time a new feature was introduced, the monkey had to either edit his code or write mapping code for the new table. It seemed that the monkey was constantly the bottleneck for any new feature being worked on. One day, Mr. Wolf called the monkey into his office. Mr. Wolf huffed, and puffed, and laid the monkey off.

Monkey Number Two
Later, Mr. Wolf was let go from the straw factory due to allegations of improper conduct with Mrs. Pig. He ended up taking a new job at the lumber yard, which just happened to be where monkey number two worked. Mr. Wolf was hired by the lumber yard specifically because of his experience with upgrading HR applications which, interestingly, was exactly the type of project that the lumber yard was about to start. Unsurprisingly, monkey number two was given the task of writing the code to populate the Person class with the data from the database. Mr. Wolf informed monkey number two of the first monkey’s fate and not-so-subtly implied that this would also be monkey number two’s fate if he did not come up with a more flexible solution. The monkey thought about it for a while and produced something similar to the following:


public class ReflectionBuilder
{
    private PropertyInfo[] properties;

    private ReflectionBuilder() { }

    public T Build(SqlDataReader reader)
    {
        T result = (T)Activator.CreateInstance(typeof(T));

        for (int i = 0; i < reader.FieldCount; i++)
        {
            if (properties[i] != null && !reader.IsDBNull(i))
            {
                properties[i].SetValue(result, reader[i], null);
            }
        }

        return result;
    }

    public static ReflectionBuilder CreateBuilder
                                           (SqlDataReader reader)
    {
        ReflectionBuilder result = new ReflectionBuilder();

        result.properties = new PropertyInfo[reader.FieldCount];
        for (int i = 0; i < reader.FieldCount; i++)
        {
            result.properties[i] =
                    typeof(T).GetProperty(reader.GetName(i));
        }

        return result;
    }
}

This solution worked much better than the first monkey’s solution. As you might have guessed, the requirements for the HR “upgrade” were constantly changing. “Add this feature,” “remove that feature,” “move this here,” “move that there.” None of this seemed to matter. Monkey number two’s use of reflection meant that his code could automatically recognize the changes. Better yet, when new tables and objects were created, the same code could be used with no additional changes. All was going extremely well. Monkey number two felt certain that he was in line for a major promotion.

But then the unthinkable happened… the application went live. Suddenly Mr. Wolf was inundated with calls from unhappy HR employees complaining about how slooooow the new application was. After a few weeks, Mr. Wolf was let go for his incompetence and the lumber yard went back to using their old HR software. However, Mr. Wolf did manage to terminate monkey number two on his way out.

Monkey Number Three
Despite his apparent ineptitude as a project manager, Mr. Wolf quickly landed a new job down at the brick yard. The brick yard was in the process of migrating their old HR software and felt they could benefit from Mr. Wolf’s “expertise.” Coincidentally, the brick yard was also the employer of monkey number three. Once again, Mr. Wolf assigned the task of loading the Person class with data from the database to monkey number three and again he implied that the monkey’s continued employment depended on not creating either of the issues that monkeys number one and two had created. Monkey number three did some research and stumbled upon the DynamicMethod and ILGenerator classes in .NET 2.0. These classes would allow the monkey to dynamically create and compile code at runtime. This would give him the best of both worlds. His code could be dynamic like monkey number two’s, but since it was actually compiled, it would be as fast as monkey number one’s.

He did some experimenting. The downside was that the dynamic code needed to be written using IL (intermediate language) instead of C#. However, with a small amount of Googling, some code decompiling using ildasm.exe from the .NET SDK, and some good old trial and error, the monkey was able to create code similar to the following:


public class DynamicBuilder
{
    private static readonly MethodInfo getValueMethod =
        typeof(IDataRecord).GetMethod("get_Item",
             new Type[] { typeof(int) });

    private static readonly MethodInfo isDBNullMethod =
        typeof(IDataRecord).GetMethod("IsDBNull",
            new Type[] { typeof(int) });

    private delegate T Load(IDataRecord dataRecord);
    private Load handler;

    private DynamicBuilder() { }

    public T Build(IDataRecord dataRecord)
    {
        return handler(dataRecord);
    }

    public static DynamicBuilder CreateBuilder
                                         (IDataRecord dataRecord)
    {
        DynamicBuilder dynamicBuilder = new DynamicBuilder();

        DynamicMethod method = new DynamicMethod("DynamicCreate",
                typeof(T), new Type[] { typeof(IDataRecord) },
                typeof(T), true);

        ILGenerator generator = method.GetILGenerator();

        LocalBuilder result = generator.DeclareLocal(typeof(T));
        generator.Emit(OpCodes.Newobj,
                typeof(T).GetConstructor(Type.EmptyTypes));

        generator.Emit(OpCodes.Stloc, result);

        for (int i = 0; i < dataRecord.FieldCount; i++)
        {
            PropertyInfo propertyInfo =
                    typeof(T).GetProperty(dataRecord.GetName(i));

            Label endIfLabel = generator.DefineLabel();

            if (propertyInfo != null &&
                propertyInfo.GetSetMethod() != null)
            {
                generator.Emit(OpCodes.Ldarg_0);
                generator.Emit(OpCodes.Ldc_I4, i);
                generator.Emit(OpCodes.Callvirt, isDBNullMethod);
                generator.Emit(OpCodes.Brtrue, endIfLabel);

                generator.Emit(OpCodes.Ldloc, result);
                generator.Emit(OpCodes.Ldarg_0);
                generator.Emit(OpCodes.Ldc_I4, i);
                generator.Emit(OpCodes.Callvirt, getValueMethod);
                generator.Emit(OpCodes.Unbox_Any,
                       dataRecord.GetFieldType(i));
                generator.Emit(OpCodes.Callvirt,
                        propertyInfo.GetSetMethod());

                generator.MarkLabel(endIfLabel);
            }
        }

        generator.Emit(OpCodes.Ldloc, result);
        generator.Emit(OpCodes.Ret);

        dynamicBuilder.handler =
                (Load)method.CreateDelegate(typeof(Load));

        return dynamicBuilder;
    }
}

Mr. Wolf was skeptical, so monkey number three did his best to explain what was going on.

The first few lines of CreateBuilder instantiate the DynamicMethod and ILGenerator classes. In short, it’s creating a new static method called DynamicCreate and adding that method to the object type that was passed in, i.e. the Person class in this example. The method takes SqlDataReader and returns an instance of the correct object. If this were non-dynamic code, you might call it like this:


Person myPerson = Person.DynamicCreate(mySqlDataReader);

The next line of code generates a variable of the generic type. So this,


LocalBuilder result = generator.DeclareLocal(typeof(T));

in non-dynamic code would be this:
Person myPerson;

The next piece of code instantiates the requested type of object and stores it in the local variable.


generator.Emit(OpCodes.Newobj,
                    typeof(T).GetConstructor(Type.EmptyTypes));
generator.Emit(OpCodes.Stloc, result);

In non-dynamic code, it would be this:
myPerson = new Person();

The code then loops through the fields in the data reader, finding matching properties on the type passed in. When a match is found, the code checks to see if the value from the data reader is null.

generator.Emit(OpCodes.Ldarg_0);
generator.Emit(OpCodes.Ldc_I4, i);
generator.Emit(OpCodes.Callvirt, isDBNullMethod);
generator.Emit(OpCodes.Brtrue, endIfLabel);

...

generator.MarkLabel(endIfLabel);

or

if (!mySqlDataReader.IsDBNull(1))
{
    ...
}

If the value in the data reader is not null, the code sets the value on the object.

generator.Emit(OpCodes.Ldloc, result);
generator.Emit(OpCodes.Ldarg_0);
generator.Emit(OpCodes.Ldc_I4, i);
generator.Emit(OpCodes.Callvirt, getValueMethod);
generator.Emit(OpCodes.Unbox_Any, dataRecord.GetFieldType(i));
generator.Emit(OpCodes.Callvirt, propertyInfo.GetSetMethod());

Again, in non-dynamic code, it would be this:
myPerson.Name = (string)mySqlDataReader[1];
The last part of the code returns the value of the local variable:

generator.Emit(OpCodes.Ldloc_0);
generator.Emit(OpCodes.Ret);

or

return myPerson;

The code then returns a handler to a delegate. When this handler is invoked, it calls the dynamically generated code, which can be seen in this code:

public T Build(SqlDataReader reader)
{
    return handler(reader);
}

Mr. Wolf had no clue what any of this meant, but not wanting to look dumb, he said, “Sounds promising, but let’s get some benchmarks before we move forward.” Monkey number three quickly threw together a test to use all three approaches. Each sample would load three million rows out of the Person table. Here were the results:

dynamicmappingspike image

Based on these results, Mr. Wolf had monkey number three implement his solution. Development went great. The live release went even better. The project was a huge success. It performed well and came in close to budget. Mr. Wolf received a huge bonus, retired early, and moved to a small private island. Monkey number three was later downsized and is currently unemployed.

Keep It Simple, Monkey
NOTE: This article is an extreme over-simplification. The code is intended to be an introduction to dynamic runtime code generation, not a full-blown solution. That being said, if you carefully and judiciously apply the ideas presented here, you should be able to be just as successful as monkey number three.

  • Share/Save/Bookmark