Full Text Search: The Key to Better Natural Language Queries for NoSQL in Node.js
In 1990, I read my first C++ programming book. I think it was by Herb Schildt. Because I had only programmed in non-OOP languages before then, it took me a while to get includes and understand how cout << "Hello, World!" << endl; worked exactly. After having those first few ah-ha! moments (as Grady Booch calls them), I was hooked. I got it. C++ and OOP were about combining data and behaviors, and in the real world this is a very good analogous model because many, many things have both state and behaviors.
The same exuberance I felt about C++ and OOP I now have for all things architectural and qualitative, especially modeling, patterns, anti-patterns, and refactoring. Modeling is simply using pictures as a means of convey a lot of easy to produce meaning in a tangible yet changeable way. Refactoring is a bit like mathematical factoring; refactoring describes how to simplify and improve code in a constrained, predictable manner. Finally, there are patterns and anti-patterns. Patterns are almost exactly what the word means. A pattern is an existing solution that can be reused to solve a problem. The idea is that software development is comprised of many common kinds of problems and a pattern is a good general solution. Patterns, then, refers to more than one pattern. Anti-patterns are patterns that have been demonstrated not to work.
If any or all of these things—patterns, anti-patterns, refactoring, or modeling—are new to you, I understand. In this article, I want to introduce just one pattern that is relatively easy to use and has a pretty high cool factor. This pattern is the State behavior pattern.
State As Behavior
The State pattern permits an object to change its behavior dynamically by changing its internal state, creating the illusion that the object has changed. What I like about the State pattern is that it is a flag killer.
To better understand how the State pattern is used, let's begin with the kind of code it can replace.
A flag is typically an enumerated type, integer, or Boolean that conveys meaning about the mode or state of an object or is used as a pivotal point to orchestrate logic. Generally, an enumerated value conveys more meaning than a plain old integer and would be considered the best of this weakest technique.
In Listing 1, there is a class named CounterWithFlag that prints some text in German or English. Change the value of the English property (the flag) to change the behavior of the method Count.
Listing 1: Using a flag to change behavior.
Public Class CounterWithFlag Private FEnglish As Boolean = True Public Property English() As Boolean Get Return FEnglish End Get Set(ByVal Value As Boolean) FEnglish = Value End Set End Property Public Sub Count() If (FEnglish) Then DoCountInEnglish() Else DoCountInGerman() End If End Sub Private Sub DoCountInGerman() Console.WriteLine("Ein") Console.WriteLine("Zwei") Console.WriteLine("Drei") End Sub Private Sub DoCountInEnglish() Console.WriteLine("One") Console.WriteLine("Two") Console.WriteLine("Three") End Sub End Class
There is a lot of code in the world written just like this. The flags become increasingly difficult to maintain as the problem and flags grow in complexity and number. Use this approach with three or four flags and you will have to master predicate calculus reduce nested conditional statements and repeated logic. Of the three solutions I will show you, this is the weakest and should only be used for very simple problems.
Polymorphism is Better
Of the possible ways to solve the problem presented in Listing 1—simulating perhaps a multi-language implementation—polymorphism is another possible solution. With polymorphism, we can implement an abstract base class or interface and then change the instance of the class or refer to alternate interfaces depending on the desired behavior. Listing 2 shows the polymorphic implementation.
Listing 2: Bi-lingual polymorphic counting.
Public MustInherit Class Counter Public MustOverride Sub Count() End Class Public Class EnglishCounter Inherits Counter Public Overrides Sub Count() Console.WriteLine("One") Console.WriteLine("Two") Console.WriteLine("Three") End Sub End Class Public Class GermanCounter Inherits Counter Public Overrides Sub Count() Console.WriteLine("Ein") Console.WriteLine("Zwei") Console.WriteLine("Drei") End Sub End Class
In Listing 2, we define an abstract base class, Counter, and implement two variants of Counter, EnglishCounter and GermanCounter. Listing 2 demonstrates how we can print in English or German by using the flag-version first and the polymorphic version second.
Listing 3: Flags versus polymorphism.
' The first example uses a loathsome flag Dim a As CounterWithFlag = New CounterWithFlag a.Count() Console.ReadLine() a.English = False a.Count() Console.ReadLine() ' The second example uses polymorphism Dim b As Counter = New EnglishCounter b.Count() Console.ReadLine() b = New GermanCounter b.Count() Console.ReadLine()
In the first half of the listing, we have to change and keep track of the flag, which controls the behavior of count. Flags can result in extremely confusing code. In the second half, we literally have to change the instance of the object, but, once changed, the code is not predicated on a flag and will be easier to maintain and always perform correctly. The second, polymorphic, approach is superior in most circumstances to the flags approach. The third solution employs the State Pattern.
The third solution employs the State pattern. In short, the state pattern is an internalization of the polymorphic solution. By employing the second solution—see Listing 2—we are making the consumer responsible for changing the instance of the object when conditions have changed. By internalizing the polymorphic, or dynamic aspects, we can permit the class to change its behavior on-the-fly while still permitting the consumer to do so. Listing 4 demonstrates the State pattern.
Listing 4: State as Behavior.
Public Enum States English German End Enum Public Class CounterWithState Private current As State = New EnglishState(Me) Public Sub Count() current.Count() End Sub Public WriteOnly Property State() As States Set(ByVal Value As States) If (Value = States.English) Then current = New EnglishState(Me) Else current = New GermanState(Me) End If End Set End Property End Class Public MustInherit Class State Private owner As CounterWithState = Nothing Public Sub New(ByVal owner As CounterWithState) Me.owner = owner End Sub Public MustOverride Sub Count() End Class Public Class GermanState Inherits State Public Sub New(ByVal owner As CounterWithState) MyBase.New(owner) End Sub Public Overrides Sub Count() Console.WriteLine("Ein") Console.WriteLine("Zwei") Console.WriteLine("Drei") End Sub End Class Public Class EnglishState Inherits State Public Sub New(ByVal owner As CounterWithState) MyBase.New(owner) End Sub Public Overrides Sub Count() Console.WriteLine("One") Console.WriteLine("Two") Console.WriteLine("Three") End Sub End Class
In listing 4, the class for general consumption is CounterWithState. The consumer creates an instance of CounterwithState and calls Count. Now, based on an internal condition, an external condition, or a combination of both, we can change the internal instance of the state field (shown bold and underlined) from EnglishState to GermanState. The end result is that CounterWithState.Count will invoke state.Count, and the internal state object determines which behavior is invoked.
It is important to note that we didn't illustrate the use of the reference to the containing object. The purpose of the state objects having a reference to its owning object is that each state object is working on behalf of its owning object. For example, if we added a property and field pair, the CounterWithState property getter and setter would be implemented in terms of a matching State getter and setter. In addition to the state object getting or setting its owning object's field, each state class' getter and setter could include different kinds of logic.
Knowing When to Use the State Pattern
When you have mastered the State Pattern, you might be inclined to use it all of the time. This is to be expected, but as you can see from Listing 4, you will incur some extra labor upfront in writing the code. The key is to find a balance.
Flags are the weakest solution but are easy to implement and work okay as long as there aren't too many flags. Polymorphism always works, but this can mean creating and keeping track of a much more complicated genealogy. Use Polymorphism to remove flags for moderately complex problems where you have semantically similar behaviors. Finally, the State Pattern is very powerful but requires the greatest understanding of OOP. You can nest the state-behavior classes to conceal them from consumers, and this pattern will permit you—the producer—as well as the consumer to dynamically change the behavior of a class. Use the State Pattern to divide complex behaviors into classes and eliminate flags altogether; however, be prepared to pay an upfront cost that may be substantially or initially greater than using flags but ultimately more maintainable and less expensive over time.
Marvin Hamlisch was playing at the Wharton Center at Michigan State University a couple of years ago. A parent asked Mr. Hamlisch to play happy birthday for a child. Rather than get upset, Mr. Hamlisch pondered whether the masters—Beethoven, Mozart, or Bach—had ever been asked this question. In response, he played happy birthday to the child by emulating the style of each of the masters. These renditions were so brilliant that for a moment the personage and music of each master composer came to life. In that moment, Marvin Hamlisch clearly demonstrated to me that he clearly had surpassed simple mechanical mastery of his craft and was able to adroitly and playfully elevate his performances to sublime art.
Knowing the difference between using flags, inheritance, and polymorphism, or employing a pattern and making a considered decision about which to use represents mastery. Choosing correctly is art.
Copyright © 2004 by Paul Kimmel. All Rights Reserved.
Paul Kimmel is the VB Today columnist, has written several books on .NET programming, and is a software architect. Paul enjoys flying planes, playing video games, shooting his Glock 21, and inventing new words—such as gobject, as in gobject-oriented programming, meaning a monolithic GUI object that encompasses controls, business rules, and everything else—for the human lexicon. You may contact him at email@example.com if you need assistance or are interested in joining the Lansing Area .NET Users Group (glugnet.org).