The purpose of this calisthenics exercise is to familiarize you with problems from our business domain, while simultaneously getting you to think hard about how you’re programming and pushing you towards a better object-oriented style. First we’ll delve more deeply into the reasoning behind the rules you’ll be bound by during this exercise, and then explain the rules themselves. Finally, you’ll be given the business requirements for the exercise. You’re welcome to use any language you like, as long as it has object-oriented features. This includes, but isn’t limited to, languages such as Java, Javascript, Scala, Ruby, Groovy, Python, Smalltalk, C++, Self, and Common Lisp.
Following the rules given here with discipline will force you to come up with the harder answers that lead to a much richer understanding of object oriented programming. If you write a thousand lines that follow all these rules you will find that you have created something completely different than you expected. Follow the rules, and see where you end up. If it isn’t comfortable, back off and see what you can leverage that is comfortable. You might find that if you keep working at it, you’ll find your code naturally conforming to these rules. Most new things worth doing are difficult, allow yourself the chance to internalize them.
7 of these 9 rules are simply ways to visualize and implement the holy grail of object oriented programming – encapsulation of data. In addition, another drives the appropriate use of polymorphism (not using else and minimizing all conditional logic), and another is a naming strategy that encourages concise and straightforward naming standards – without inconsistently applied and hard to pronounce abbreviations. The entire thrust is to craft code that has no duplication in code or idea; code which concisely expresses simple and elegant abstractions for the incidental complexity we deal with all day long.
We should also point out that the more you practice applying the rules, the more the advantages come to fruition. Your first attempts to decompose problems in the style presented here will feel awkward and likely lead to little gain you can perceive. There is a skill to the application of the rules – this is the art of the programmer raised to another level.
In the long run, you will inevitably find that these rules contradict each other in some situations, or the application of the rules leads to degenerate results. For the purpose of the exercise, however, spend 20 hours and 1000 lines writing code that conforms 100% to these rules. You will find yourself having to break yourself of old habits and change rules that you may have lived with for your whole programming life. Each of the rules has been chosen such that if you follow it you will encounter situations that would typically have an easy answer that is not available to you.
Ever stare at a big old method wondering where to start? A giant method lacks cohesiveness. One guideline is to limit method length to 5 lines, but that kind of transition can be daunting if your code is littered with 500-line monsters. Instead, try to ensure that each method does exactly one thing – one control structure, or one block of statements, per method. If you’ve got nested control structures in a method, you’re working at multiple levels of abstraction, and that means you’re doing more than one thing.
As you work with methods that do exactly one thing, expressed within classes doing exactly one thing, your code begins to change. As each unit in your application becomes smaller, your level of re-use will start to rise exponentially. It can be difficult to spot opportunities for reuse within a method that has five responsibilities and is implemented in 100 lines. A three-line method that manages the state of a single object in a given context is usable in many different contexts.
Use the Extract Method feature of your IDE to pull out behaviors until your methods only have one level of indentation, like this:
before:
class Board
{
...
String board()
{
StringBuffer buf = new StringBuffer();
for (int i = 0; i < 10; i++)
{
for (int j = 0; j < 10; j++)
{
buf.append(data[i][j]);
}
buf.append("\n");
}
return buf.toString();
}
}after:
class Board
{
...
String board()
{
StringBuffer buf = new StringBuffer();
collectRows(buf);
return buf.toString();
}
void collectRows(StringBuffer buf)
{
for (int i = 0; i < 10; i++)
{
collectRow(buf, i);
}
}
void collectRow(StringBuffer buf, int row)
{
for (int i = 0; i < 10; i++)
{
buf.append(data[row][i]);
}
buf.append("\n");
}
}Notice that another effect has occurred with this refactoring. Each individual method has become virtually trivial to match its implementation to its name. Determining the existence of bugs in these much smaller snippets is frequently much easier.
See also:
Every programmer understands the if/else construct. It’s built into nearly every programming language, and simple conditional logic is easy for anyone to understand. Nearly every programmer has seen a nasty nested conditional that’s impossible to follow, or a case statement that goes on for pages. Even worse, it is all too easy to simply add another branch to an existing conditional rather than factoring to a better solution. Conditionals are also a frequent source of duplication. Status flags and state of residence are two examples which frequently lead to this kind of trouble:
if (status == DONE)
{
doSomething();
}
else
{
...
}Object-oriented languages give us a powerful tool, polymorphism, for handling complex conditional cases. Designs that use polymorphism can be easier to read and maintain, and express their intent more clearly. But it’s not always easy to make the transition, especially when we have ELSE in our back pocket. So as part of this exercise, you’re not allowed to use ELSE. Try the Null Object pattern; it may help in some situations. There are other tools that can help you rid yourself of the else as well. See how many alternatives you can come up with.
See also:
In the Java language, int is a primitive, not a real object, so it obeys different rules than objects. It is used with a syntax that isn’t object-oriented. More importantly, an int on its own is just a scalar, so it has no meaning. When a method takes an int as a parameter, the method name needs to do all of the work of expressing the intent. If the same method takes an Hour as a parameter, it’s much easier to see what’s going on. Small objects like this can make programs more maintainable, since it isn’t possible to pass a Year to a method that takes an Hour parameter. With a primitive variable the compiler can’t help you write semantically correct programs. With an object, even a small one, you are giving both the compiler and the programmer additional info about what the value is and why it is being used.
Small objects like Hour or Money also give us an obvious place to put behavior that would otherwise have been littered around other classes. This becomes especially true when you apply Rule 9, and only the small object can access the value. Note that this does not mean using object wrappers that are available in languages like Java. Using an Integer instead of an int confers no additional advantages in terms of expressing intent, whereas using a wrapper that expresses meaning within the problem domain both clarifies its usage and makes intent evident.
See also:
If your language of choice doesn’t support higher order functions or blocks (e.g. Java) then application of this rule is difficult. Regardless of such support, this rule can be summed up thusly: any class that contains a collection should contain no other member variables. Each collection gets wrapped in its own class, so now behaviors related to the collection have a home. You may find that filters become a part of this new class. Also, your new class can handle activities like joining two groups together or applying a rule to each element of the group.
If your language of choice supports higher order functions or blocks then the application of this quite a bit simpler. The goal of this rule is to prevent the exposure of the implementation of a domain specific collection outside of that collection. So that, for example, we can switch internally from using a Dictionary to a List without changing our callers. In a language such as Ruby, you can comply with this rule by providing the standard enumerable methods that apply to your domain on your domain object itself. For example:
class Employees
...
def reportingTo(manager)
employees.select { | employee | employee.reportsTo(manager) }
end
...
endSometimes it’s hard to know which object should take responsibility for an activity. If you start looking for lines of code with multiple dots, you’ll start to find many misplaced responsibilities. If you’ve got more than one dot on any given line of code, the activity is happening in the wrong place. Maybe your object is dealing with two other objects at once. If this is the case, your object is a middleman; it knows too much about too many people. Consider moving the activity into one of the other objects.
If all those dots are connected, your object is digging deeply into another object. These multiple dots indicate that you’re violating encapsulation. Try asking that object to do something for you, rather than poking around its insides. A major part of encapsulation is not reaching across class boundaries into types that you shouldn’t know about.
The Law of Demeter ("Only talk to your friends") is a good place to start, but think about it this way: You can play with your friends, you can play with your privates, but you can’t play with your friends’ privates:
a.getFoo().getBar().getBaaz().doSomething(); // badNote that this rule doesn’t forbid the following:
a.foo(b.foo()); // okayRather, it’s a simplified way of stating the Law of Demeter.
It’s worth noting that there are specific places where multiple dots make sense, but these are usually within the context of building a DSL or a design pattern, such as a Builder:
builder.property1("value").property2("value").build();In this case it doesn’t violate the spirit of the rule since it doesn’t return any internal state or other classes.
See also:
It’s often tempting to abbreviate in the names of classes, methods, or variables. Resist the temptation – abbreviations can be confusing, and they tend to hide larger problems.
Think about why you want to abbreviate. Is it because you’re typing the same word over and over again? If that’s the case, perhaps your method is used too heavily and you are missing opportunities to remove duplication. Is it because your method names are getting long? This might be a sign of a misplaced responsibility, or a missing class.
Try to keep class and method names to 1-2 words, and avoid names that duplicate the context. If the class is an Order, the method doesn’t need to be called shipOrder(). Simply name the method ship() so that clients call order.ship() – a simple and clear representation of what’s going on.
This means no class over 50 lines and no package over 10 files.
Classes over 50 lines usually do more than one thing, which makes them harder to understand and harder to reuse. 50-line classes have the added benefit of being visible on one screen without scrolling, which makes them easier to grasp quickly.
What’s challenging about creating such small classes is that there are often groups of behaviors that make logical sense together. This is where we need to leverage packages. As your classes become smaller and have fewer responsibilities, and as you limit package size, you’ll start to see that packages represent clusters of related classes that work together to achieve a goal. Packages, like classes, should be cohesive and have a purpose. Keeping those packages small forces them to have a real identity. If the real identity comes out to more than 50 lines, that is ok. This is software engineering; there is no black and white- as Uncle Bob regularly stresses, our craft is about trade-offs.
Most classes should simply be responsible for handling a single state variable, but there are a few that will require two. Adding a new instance variable to a class immediately decreases the cohesion of that class. In general, while programming under these rules, you’ll find that there are two kinds of classes, those that maintain the state of a single instance variable, and those that coordinate two separate variables. In general, don’t mix the two kinds of responsibilities.
The discerning reader might have noticed that rules 3 and 4 can be considered to be isomorphic. In a more general sense, there are few cases where a cohesive single job description can be created for a class with many instance variables. As an example of the kind of dissection we are asking you to engage in:
Note that we could simply encode CustomerId, First Name and Last Name on the Customer object detailed in the image above. The Name object here contains a list of names, allowing the new model to absorb people with first and last name (as well other possibilities such as middle & full) middle, and other given names. Frequently, decomposition of instance variables leads to an understanding of commonality of several related instance variables. Sometimes several related instance variables actually have a related life in a first class collection.
Indeed, it is the authors' experience that decomposing objects from a set of attributes into a hierarchy of collaborating objects, leads much more directly to an effective object model. Prior to understanding this rule, we spent many hours trying to follow data flows through large objects. It was possible to tweeze out an object model, but it was a painstaking process to understand the related groups of behavior and see the result. In contrast, the recursive application of this rule has lead to very quick decomposition of complex large objects into much simpler models. Behavior naturally follows the instance variables into the appropriate place.
The last sentence of the previous rule leads almost directly to this rule. If your objects are now encapsulating the appropriate set of instance variables but the design is still awkward, it is time to examine some more direct violations of encapsulation. The behavior will not follow the instance variable if it can simply ask for the value in its current location. The idea behind strong encapsulation boundaries is to force programmers working on the code after you leave it to look for and place behavior into a single place in the object model. This has many beneficial downstream effects, such as a dramatic reduction in duplication errors and a better localization of changes to implement new features. This does not mean that variables of the same class cannot interact with one another in methods such as ‘equals’; because access to instance variables doesn’t leave the class, encapsulation is not violated.
Another way this rule is commonly stated is "Tell, don’t ask"
See also:
You have been tasked with writing a new Job Application domain for TheLadders. Requirements for this system are below:
We have 5 main entities in this domain:
- Jobs
- Jobseekers
- Employers
- Resumes
- Job Applications
Here are the interactions:
- Employers can post jobs.
- Employers should be able to see a listing of the jobs they have posted.
- Jobseekers can save jobs onsite for later viewing.
- Jobseekers can apply to jobs posted by employers.
- There are 2 different kinds of Jobs posted by employers: JReq and ATS.
- JReq jobs require a resume to apply to them.
- ATS jobs do not.
- Jobseekers can not apply to a job with someone else’s resume.
- Jobseekers should be able to apply to different jobs with different resumes.
- Jobseekers should be able to see a listing of jobs they have saved for later viewing.
- Jobseekers should be able to see a listing of the jobs for which they have applied.
- Employers should be able to see jobseekers who have applied to their jobs by both job and day. If possible, we would like to be able to combine the 2 and see jobseekers who have applied to a given job on a given day.
- TheLadders should be able to get a report of which jobseekers have applied to jobs on any given day.
- TheLadders should be able to get the job application report in either csv or html format.
- TheLadders should be able to ascertain jobseeker, job, employer and job application date from the job application report.
- TheLadders should be able to see aggregate job application numbers by job and employer.
- TheLadders should be able to see how many job applications failed and how many succeeded aggregated by job and employer.
- Jobseekers, when displayed, should be displayed by their name.
- Employers, when displayed, should be displayed by their name.
- Jobs, when displayed, should be displayed with a title and the name of the employer who posted it.
- TheLadders wants the system to be able to handle more than one job with the same title.
- TheLadders wants the system to be able to handle more than one jobseeker with the same name.
- TheLadders wants the system to be able to handle more than one employer with the same name.
Good luck! Remember, the most important part of being agile is early feedback, so be sure to consult with whoever is guiding you through this exercise often. There are no stupid questions if asking them leads to better results.
Based on "Jeff Bay’s 9 steps to better software design today" => http://www.cs.helsinki.fi/u/luontola/tdd-2009/ext/ObjectCalisthenics.pdf