A code smell, a term coined by Kent Beck, refers to problems within the design and structure of the underlying code in software. However, code smells are explicitly not bugs or defects. Rather, they are problems with the internal quality of the software; that is, problems with maintainability, analyzability, complexity, testability, etc.
In this module, we will long as some common examples of “smelly code”, and look at solutions.
Methods that are overly long and complicated are difficult to read and understand, difficult to use, and most importantly difficult to maintain.
Functions, as a tool, are good at doing one thing and one thing only. If our functions are overly long, especially if they feature multiple loops and conditionals, most likely the function is doing several things. This makes the function harder to understand. If the function is harder to understand, it’s harder to test, re-use, modify, remove and replace, etc.
Consider the following example, which takes in a file of point coordinates, like:
1.0, 3.5
2.7, 1.2
0.8, -3.2
5.0, 4.2
…and calculates the total distance between these points, printing the total distance with 3 decimal places.
public class DistanceCalculator {
public static void main(String[] args) throws IOException {
String filename = args[0];
FileReader fileReader = new FileReader(filename);
BufferedReader bufferedReader = new BufferedReader(fileReader);
List<Point> pointList = new ArrayList<>();
for(String line = bufferedReader.readLine(); line != null; line = bufferedReader.readLine()) {
String[] splitLine = line.split(",");
double x = Double.parseDouble(splitLine[0].strip());
double y = Double.parseDouble(splitLine[1].strip());
Point point = new Point(x, y);
pointList.add(point);
}
double totalDistance = 0.0;
for (int index = 0; index < pointList.size() - 1; index++) {
Point first = pointList.get(index);
Point second = pointList.get(index + 1);
double differenceX = Math.abs(first.getX() - second.getX());
double differenceY = Math.abs(second.getX() - second.getY());
double distance = Math.sqrt(differenceX * differenceX + differenceY * differenceY);
totalDistance += distance;
}
System.out.printf("Total distance = %.3f", totalDistance);
}
}
class Point {
private final double x, y;
Point(double x, double y) {
this.x = x;
this.y = y;
}
public double getX() { return x; }
public double getY() { return y; }
}
Notice how large that main method is. I want to break this up into several methods. After breaking up main into several separate methods, I get the following:
public class DistanceCalculator {
public static void main(String[] args) throws IOException {
String filename = args[0];
BufferedReader bufferedReader = getBufferedReader(filename);
List<Point> pointList = getPointList(bufferedReader);
double totalDistance = getTotalDistance(pointList);
System.out.printf("Total distance = %.3f", totalDistance);
}
public static BufferedReader getBufferedReader(String filename) throws FileNotFoundException {
FileReader fileReader = new FileReader(filename);
return new BufferedReader(fileReader);
}
public static List<Point> getPointList(BufferedReader reader) throws IOException {
List<Point> pointList = new ArrayList<>();
for (String line = reader.readLine(); line != null; line = reader.readLine()) {
pointList.add(getPointFromLine(line));
}
return pointList;
}
public static Point getPointFromLine(String line) {
String[] splitLine = line.split(",");
double x = Double.parseDouble(splitLine[0].strip());
double y = Double.parseDouble(splitLine[1].strip());
return new Point(x, y);
}
public static double getTotalDistance(List<Point> pointList) {
double totalDistance = 0.0;
for (int index = 0; index < pointList.size() - 1; index++) {
Point first = pointList.get(index);
Point second = pointList.get(index + 1);
totalDistance += first.distanceTo(second);
}
return totalDistance;
}
}
class Point {
private final double x, y;
Point(double x, double y) {
this.x = x;
this.y = y;
}
public double getX() { return x; }
public double getY() { return y; }
public double distanceTo(Point point) {
double differenceX = Math.abs(this.getX() - point.getX());
double differenceY = Math.abs(this.getX() - point.getY());
return Math.sqrt(differenceX * differenceX + differenceY * differenceY);
}
}
Notice now that each method is only a few lines. Also, it becomes much easier to design a specific test-case for each method. In general, if I ask a question, like:
1) What are we using to read the file?
2) How do we read the file into a list of points?
3) How are we calculating the distance between two points?
etc.
That the answer to those questions are in a specific function whose name and purpose is clear, and each function does only one thing.
We will examine this case, as well as another case, in the “Extract Method” module.
One of the most common mistakes I see in newer programmers is putting their entire program inside of a single Main.java class. This is a completely understandable mistake:
1) Newer programmers may not be as comfortable knowing when to create multiple classes 2) Newer programmers may not be comfortable with Object-Oriented programming
However, a programmer looking at a class can only keep so many ideas inside their head at one time. A class can therefore become hard to understand if:
There is no ideal class length. Some classes, like “Plain Old Java Objects” (POJOs) will inherently be short: just some fields, a constructor, and some getters and setters. An example of a POJO may be:
public class Student implements Comparable<Student> {
private final int idNumber;
private String firstName, lastName;
private String email;
public static final String EMAIL_DOMAIN = "virginia.edu";
public Student(int idNumber, String firstName, String lastName, String email) {
if (firstName == null || lastName == null) {
throw new IllegalArgumentException("Students names cannot be null when instantiated");
}
this.idNumber = idNumber;
this.firstName = firstName;
this.lastName = lastName;
this.email = email;
}
public int getIdNumber() {
return idNumber;
}
public String getFirstName() {
return firstName;
}
public void setFirstName(String firstName) {
this.firstName = firstName;
}
public String getLastName() {
return lastName;
}
public void setLastName(String lastName) {
this.lastName = lastName;
}
public String getEmail() {
return email;
}
public void setEmail(String email) {
this.email = email;
}
@Override
public String toString() {
return "example.Student{" +
"idNumber=" + idNumber +
", firstName='" + firstName + '\'' +
", lastName='" + lastName + '\'' +
", email='" + email + '\'' +
'}';
}
@Override
public int compareTo(Student other) {
//sort students in ascending order by id number
return idNumber - other.idNumber;
}
}
However, many of our classes will be more complicated. For example, imagine we had a class called Registrar that handled:
All of these things in one single class can make the class difficult to understand. Instead, we want to decompose all of these different needs into several smaller classes. Each class individually exposes its behavior through a smaller interface that is easier to understand.
To reference Clean Code once again:
“The ideal number of arguments for a function is zero (niladic). Next comes one (monadic), followed closely by two (dyadic). Three arguments (triadic) should be avoided where possible. More than three (polyadic) requires very special justification - and then should be used anyway.”
Clean Code Chapter 3, Page 40 - Bob Martin
Let’s go back to our example in the Test-Driven-Development unit.
In this case, we wrote the function like:
public class StudentFinancialRecord {
private Student student;
private double overdue;
private boolean isExempt;
public StudentFinancialRecord(Student student) {
this.student = student;
this.overdue = 0;
this.isExempt = false;
}
public double calculateBill() {
int courseCount = student.getCourseList().size();
return 0.0; //TODO: Stub
}
public void setOverdue(double overdue) {
this.overdue = overdue;
}
public void setExempt(boolean isExempt) {
this.isExempt = isExempt;
}
}
Note that I shortened the class above to remove any unnecessary code. This is an example of a monadic (1 argument function). However, during lecture, I used a slightly different version
public class StudentFinancialRecord {
public double calculateBill(int courseCount, double overdue, boolean exempt) {
return 0.0; //TODO: Stub
}
}
This is an example of a triadic (3 argument function).
Which is better? Well, from a standpoint of make the code the easiest to understand, let’s consider how these are use:
To call calculateBill on the monadic version might look something like:
StudentFinancialRecord record = new StudentFinancialRecord(myStudent);
record.setOverdue(2000);
record.setExempt(true);
var balanceDue = record.calculateBill();
To call calculateBill on the triadic version might look something like:
StudentFinancialRecord record = new StudentFinancialRecord(myStudent.getID());
var balanceDue = record.calculateBill(myStudent.getCourseList().size(), 2000, true);
Which is easier to understand? Generally, the first example is. This is because we can think of the StudentFinancialRecord as an object (or record of information) that is tied to a Student, and having some amount overdue, and being either exempt or not-exempt from interest. These describe the state of the FinancialRecord.
It’s also much easier to remember the order of arguments in a monadic. That’s because there’s only 1 order the arguments can go in. Without scrolling back up, can you remember the order of the arguments in the triadic? If you can’t, then you can try to guess, but there are 6 possible ways to arrange the arguments.
Now consider if we had 4 arguments: 24 ways they could be organized! Or 5 arguments! 120 ways!
In general, it’s easier to make multiple monadic calls, where each method serves a clear individual purpose, then make one triadic call that combines all three purposes into one.
Now, modern IDEs help with this: they will show you the method signature as you are writing your method call, so you are less likely to get the order wrong. However, if your function is taking in and using so many inputs, it might be worth considering turning the function into a class like we did with StudentFinancialRecord. This allows to more easily isolate the complicated function that requires so much information, and test it independent of any tangentially related functionalities.
There is a trade-off here. As it stands, we have what is called temporal coupling, where you have to “Set up” the StudentFinancialRecord object before calling “calculateBill”. Just be aware of that trade-off when designing classes. As a general rule, your class should be built ready to be used. So it may make sense to move the setters to the Constructor. In fact, we could create overloaded constructors to help with this:
public class StudentFinancialRecord {
private Student student;
private double overdue;
private boolean isExempt;
public StudentFinancialRecord(Student student, double overdue, boolean isExempt) {
this.student = student;
this.overdue = overdue;
this.isExempt = isExempt;
}
public StudentFinancialRecord(Student student) {
this(student, 0.0, false);
}
}
This means the client code (the code using this class) has the choice to set up everything with the object at once or use the simpler constructor and the setter methods.
Of course, if the Constructor ends up holding every field, and our class necessarily has a complicated structure, it can end up with a large argument list. This is where something like a Factory or Builder class could come in handy, which we will discuss in design patterns.
Speaking of our previous example, let’s consider the function:
public double calculateBill(List<Integer> registeredCourseNumbers, double overdue, boolean exempt) {
return 0.0; //TODO: Stub
}
I know for certain this function does more than one thing because of the boolean argument exempt. In fact, the specification tells me this:
3) Increase the value of the field overdue amount by 10% if exempt is false (this is done AFTER step 2)
- if exempt is true, this penalty is waived, and overdue does not change.
This step alone means we have two different functions here:
As such, we can separate this into two different functions:
public double calculateBillWithInterest(List<Integer> registeredCourseNumbers, double overdue) {
return 0.0; //TODO: Stub - exempt is false
}
public double calculateInterestExemptBill(List<Integer> registeredCourseNumbers, double overdue) {
return 0.0; //TODO: Stub - exempt is true
}
The reason we want to remove boolean arguments is that it requires whoever is calling the function to know at least some information about how the function works. This means understanding the function requires more knowledge. Instead, we want our functions to be simple to use as possible. So instead of:
public void orderHamburger(boolean hasCheese);
…do…
public void orderHamburger();
public void orderCheeseburger();
While the latter has more functions, both functions are clearly communicating their intent and are niladics, and therefore cannot be called with incorrect arguments!
On top of removing the boolean, we may want to consider whether a student who is interest exempt should be handled by another class entirely. For example:
public class StudentFinancialRecord {
protected double overdue;
public double calculateBill(List<Integer> registeredCourseNumbers, BillCalculator billCalculator) {
return billCalculator.calculate(registeredCourseNumbers);
}
}
…and…
public interface BillCalculator {
public double calculate(List<Integer> registeredCourseNumbers);
}
public class TaxExemptBillCalculator {
public double calculate(List<Integer> registeredCourseNumbers) {
// implement tax exempt approach
}
}
public class NonExemptBillCalculator {
public double calculate(List<Integer> registeredCourseNumbers) {
// implement tax exempt approach
}
}
Now, we calculate at run time the correct amount due based on whether the Student is exempt. This has the advantage of moving the logic of calculating the bill outside of the class that simply keeps a record of the information.
This is called a Strategy Pattern, and is a very popular design pattern that lets us separate data from how the data is used.
Consider the following class:
public class Building {
private String name;
private String streetName;
private int streetNumber;
private int zipCode;
private String city;
private String state;
private double latitude;
private double longitude;
}
This class has 8 different fields! Imagine if each of these fields had getters/setters. That’s 16 different methods!
But consider that five of the fields are simply related to the address of the building. Two of the fields relate to the geographic coordinates of the building. In effect, we actually have three different ideas here:
1) The idea of a building with a name, address, and geolocation 2) The address of the building 3) The geolocation of the building
This is an example of primitive obsession (for the sake of design, we typically consider the Java String class to be in effect a primitive, even though it is technically a class). We are modeling complex, unrelated data in one class directly with primitive datatypes. This means the Building class is more complicated than it needs to be.
Instead, we can combine the primitives relevant to specific behaviors together:
public class Building {
private String name;
private Address address;
private Coordinate coordinates;
}
public class Address {
private String streetName;
private int streetNumber;
private int zipCode;
private String city;
private String state;
}
private class Coordinates {
private double latitude;
private double longitude;
}
Now, we can shorten our 16 getters and setters in Building to just 6:
getName/setNamegetAddress/setAddressgetCoordinate/setCoordinateAdditionally, because we can think about an Address of a building as just an Address and the Coordinate as just a Coordinate, it makes our code more modular and easier to understand and describe.
Consider the following code from a SudokuSolving program I wrote. While there are other functions called by this method, for the sake of space, I’m only highlighting this function.
public class BoxLineReductionSolver implements OneStepSolver {
@Override
public String solveOneStep(SudokuGrid g) {
for (RowContainer row : g.getRows()) {
for (int missingValue : row.getMissing()) {
List<Cell> cells = getCellsWithPossibility(row, missingValue);
if (inSameBox(cells)) {
int boxNumber = cells.get(0).getBoxNumber();
if (removePossibilitiesExcludingRow(missingValue, row.rowNumber, g.getBoxes().get(boxNumber))) {
return "BoxLineReduction: " + missingValue + " in Box " + boxNumber + " must be in Row " + row.rowNumber + "; removed " +
missingValue + " from all other cells in the Box.";
}
}
}
}
for (ColumnContainer col : g.getColumns()) {
for (int missingValue : col.getMissing()) {
List<Cell> cells = getCellsWithPossibility(col, missingValue);
if (inSameBox(cells)) {
int boxNumber = cells.get(0).getBoxNumber();
if (removePossibilitiesExcludingColumn(missingValue, col.colNumber, g.getBoxes().get(boxNumber))) {
return "BoxLineReduction: " + missingValue + " in Box " + boxNumber + " must be in Column " + col.colNumber + "; removed " +
missingValue + " from all other cells in the Box.";
}
}
}
}
return null;
}
}
This function has a number of problems, and violates many rules of design and code quality that we have and will discuss. However, the biggest problem here is duplicate code.
Look at the two for loops that start with:
for (RowContainer row : g.getRows()) {for (ColumnContainer col : g.getColumns()) {These two for-loops are nearly completely identical! In fact, the only difference is in the return String at the end. This is particularly bad because RowContainer and BoxContainer both extend the same class, SudokuContainer, so I’ve already done the design work to avoid this repetition.
The problem with this duplication is that this is a bug waiting to happen. Imagine if I need to update this function, for example to remove this violation of abstraction (we will discuss this later)
int boxNumber = cells.get(0).getBoxNumber();
Well, because that line of code is in two different places, I have to change it twice. If I change it in one place, but not the other, I now have two pieces of code that in theory should do the same thing, but are implemented differently.
We always want to be DRY (Don’t Repeat Yourself), and we never want to be WET (Write Everything Twice). To fix this, I can use my polymorphic structure:
public String solveOneStep(SudokuGrid g) {
String result = checkContainer(g, g.getRows());
if (result != null) return result;
result = checkContainer(g, g.getColumns());
if (result != null) return result;
return null;
}
private String checkContainer(SudokuGrid g, List<? extends SudokuContainer> containers) {
for (SudokuContainer container : containers) {
String result = checkSudokuContainerForBoxLineReduction(g, container);
if (result != null) {
return result;
}
}
return null;
}
private String checkSudokuContainerForBoxLineReduction(SudokuGrid g, SudokuContainer container) {
for (int missingValue : container.getMissing()) {
List<Cell> cells = getCellsWithPossibility(container, missingValue);
if (inSameBox(cells)) {
int boxNumber = cells.get(0).getBoxNumber();
if (removePossibilitiesExcludingColumn(missingValue, container.getContainerID(), g.getBoxes().get(boxNumber))) {
return "BoxLineReduction: " + missingValue + " in Box " + boxNumber + " must be in Column " + container.getContainerID() + "; removed " +
missingValue + " from all other cells in the Box.";
}
}
}
return null;
}
This is still not perfect (the functions are too long and complicated), but it is an improvement in that our code is now less repetitive.
It’s important to note that fewer lines of code isn’t inherently better. That is, doing things to shorten the number of lines of code without actually changing the code complexity isn’t beneficial.
Consider the following code:
public String getFirstValueFromCSVFileAsUppercase() {
return new BufferedReader(new FileReader(new File("filename.csv")))).readline().strip().split(",")[0].strip().toUpperCase();
}
This one line of code is completely unreadable. However, instead, consider the following:
public String getFirstValueFromCSVFileAsUppercase() {
File myCSVFile = new File("filename.csv");
FileReader fileReader = new FileReader(myCSVFile);
BufferedReader reader = new BufferedReader(fileReader);
String firstLine = reader.readLine();
firstLine = firstLine.strip();
String[] firstLineSplit = firstLine.split(",");
String firstValue = firstLineSplit[0].strip();
return firstValue.toUpperCase();
}
These two pieces of code do the same thing, but the second is more readable and maintainable, because it’s much more clear what the purpose of each step is in our code.
Additionally, we can now extract these steps into individual methods:
public String getFirstValueFromCSVFileUppercase() {
BufferedReader reader = getBufferedReaderFromFilename("filename.csv");
String[] firstRowArray = reader.getNextRowAsStringArray(reader);
return getStrippedStringAsUppercase(firstRowArray[0]);
}
public BufferedReader getBufferedReaderFromFilename(String filename) {
File myCSVFile = new File("filename.csv");
FileReader fileReader = new FileReader(myCSVFile);
return new BufferedReader(fileReader);
}
public String[] reader.getNextRow(reader) {
String firstLine = reader.readLine();
firstLine = firstLine.strip();
return firstLine.split(",");
}
public String toStrippedUppercase(String string) {
String strippedString = strip.strip();
return strippedString.toUppercase();
}
Does this take up more space? Yes. However, now the code is easier to understand, as each individual function has a single clear purpose. Additionally, each function may be more re-usable, meaning we won’t have to rewrite repeated functionality later as needed. Each function is also individually testable.
The above was just a sampling of Code Smells. You can find a more thorough and larger list at Refactoring Guru