Defensive Programming
I like to introduce the idea of defensive programming by talking about defensive driving.
When you were learning to drive, you were likely told that you should drive defensively. But what does that actually mean? Are we saying you should drive ready to protect yourself, with violence if needed, like some Mad Max action sequence?
As fun as that sounds, no. Defensive driving is the idea that, when driving, you should be prepared for people around you to do incorrect things. For example, the car you are passing in the right lane may try to switch lanes for no clear reason while you are passing, so you should pay attention to that car as you pass. When the traffic light turns green, you should still look before you go to ensure the intersection is clear or no one is running the red light. As you go around a blind corner, you should be mentally prepared in case someone coming the other direction strays out of their lane and into yours.
Driving defensively is driving so that if other drivers make a mistake, you won’t get caught up in it. You cannot control how other people drive. You can only drive in a way to maximize your own safety.
Understanding Defensiveness
Considered writing a command line program that takes in the name of, say, an Excel file, and performs some data manipulation or summarization based on the contents of the file. Consider all the ways the program could fail that are not the fault of the developer of that program:
1) The user may not include any command line arguments
2) The user may link to an invalid file
3) The user may have errors in their input file, such as missing columns, or bad format.
4) The user may have the file open in Excel, which can result in the file not being openable within your program
In general, all of these exceptions are not the fault of the program, but rather how the program is being used.
These exceptions are fundamentally different from programmer errors, like NullPointerExceptions caused by not correctly initiating your data structures. In essence, we are drawing a line between programmer error and user error.
In general:
1) A user should never see a programmer error: the program as released to the user should be stable and function correctly.
2) A user should be informed when they make a user error. The error message should include a clear description of what the user did incorrectly in their interaction with the software
Client Class
When developing our own classes, we have to consider clients. A client class is any class that uses the code we are writing. This can be another class in the project, or this could be someone using our class like a library (such as the spreadsheet libraries in poi). In this case, we want to design our class so that:
1) The client uses our class correctly and as intended 2) If the client uses our class incorrectly, they are made aware of this 3) Our class cannot be used incorrectly by the client to produce an erroneous state.
Example: BankAccount
Consider the following class BankAccount:
public class BankAccount {
private final int id;
private double balance;
public BankAccount(int id, double startingBalance) {
this.id = id;
this.balance = startingBalance;
}
public BankAccount(int id) {
this(id, 0);
}
}
For now, we simply have a constructor. Let’s say that we are implementing this BankAccount for a new bank. The bank tells you that they want to ensure that no BankAccount can never have a negative balance. You start writing this constructor, and already, you notice a problem.
Using this constructor, a client class could violate this rule:
BankAccount negativeAccount = new BankAccount(23145, -2000);
The above line creates a bank account with a negative balance. Because this is a constructor method, we can’t “not return” an object. However, we also can’t do something like “just set the balance the zero”, because whatever client instantiates a BankAccount object this way is clearly confused about this “no negative balance rule.” So, instead, we can throw an exception!
Creating our Exception
public class NegativeBalanceException extends IllegalArgumentException{
public NegativeBalanceException(String message) {
super(message);
}
}
I created an exception with its own error message that indicates why the constructor was used wrong when trying to create an account with the negative balance. Because this exception relates to illegal arguments in the constructor, I decided that NegativeBalanceException should extend IllegalArgumentException.
Now, we update our constructor:
public BankAccount(int id, double startingBalance) {
this.id = id;
if (startingBalance < 0) {
throw new NegativeBalanceException(getNegativeBalanceMessage());
}
this.balance = startingBalance;
}
private String getNegativeBalanceMessage() {
return "Error: attempted to create account #" + id + " with a negative balance.";
}
withdraw and deposit
Consider the following simple functions for withdraw and deposit for our BankAccount class.
public void withdraw(double amount) {
balance -= amount;
}
public void deposit(double amount) {
balance += amount;
}
These function perform the basic functionality we intend. But now, consider the withdraw function. Could this function
be used to violate our “no negative balances” rule? Yes! We need to ensure no one can withdraw more money than their balance.
Error codes
One way we could handle this is changing withdraw to a boolean function. For example:
public boolean withdraw(double amount) {
if (amount > balance) {
return false;
}
balance -= amount;
return true;
}
This approach falls under the example of “error codes”. That is, return a value that indicates the function was used incorrectly. In this case, true means the transaction was valid and accepted, and false means it is not. There are cases where this approach is used. For example, the Java built-in Set.add(E e) function returns true if the object was added to the set, and false if it is not. However, one downside of error codes is that they can be ignored. The client calls the illegal withdraw call, and then proceeds asusming the call worked correctly, never checking the return value. To avoid this, we instead can use an exception:
public void withdraw(double amount) {
if (amount > balance) {
throw new InsufficientFundsException(getInsufficientFundsMessage(amount));
}
balance -= amount;
}
private String getInsufficientFundsMessage(double amount) {
return "Error: insufficient funds in account #" + id + " - balance: " + balance +
" for transaction amount: " + amount;
}
With our InsufficientFundsException class:
public class InsufficientFundsException extends RuntimeException {
public InsufficientFundsException(String message) {
super(message);
}
}
Order of statements
Note that we check the pre-condition (amount shouldn’t be greater than balance) before we perform the actual transaction. There is a reason for this. Say we do this in the opposite order:
public void withdraw(double amount) {
balance -= amount;
if (amount > balance) {
throw new InsufficientFundsException(getInsufficientFundsMessage(amount));
}
}
You might think this makes no difference, because after all, an Exception is being thrown. However, let’s consider the programmer of the client class knows just enough about Exceptions to be dangerous, but not enough to actually work with the correctly. They might write code like the following:
BankAccount account = new BankAccount(8675309, 200);
account.withdraw(300);
System.out.println(account.getId() + " - " + account.getBalance())
Then, when the developer gets an InsufficientFundsException on line 2, they decide to “handle” the exception by adding a try-catch block, but they don’t actually do anything:
BankAccount account = new BankAccount(8675309, 200);
try {
account.withdraw(300);
} catch (Exception e) {
//make the error go away
}
System.out.println(account.getId() + " - " + account.getBalance())
Now, obviously the above code is silly. But remember, we are practicing defensive programming. If, in our withdraw method, we change the balance before throwing, like we do in the most recent version of withdraw, this can result in incorrect behavior:
8675309 - -100.0
Our goal is to make our code impossible to use incorrectly. But the above breaks that rule. On the other hand, if we change our implementation of withdraw back to checking the pre-condition and, if necessary, throwing an exception first:
public void withdraw(double amount) {
if (amount > balance) {
throw new InsufficientFundsException(getInsufficientFundsMessage(amount));
}
balance -= amount;
}
…then running the above silly client code will print:
8675309 - 200.0
That is, the transaction never “goes through”. We have ensured correct behavior, even if the client tries to get around our exceptions.
Another Example
Consider our transfer function below:
public void transfer(BankAccount recipient, double amount) {
balance -= amount;
recipient.balance += amount;
}
Now imagine our bad client developer writes:
BankAccount a = new BankAccount(1234, 400);
try {
a.transfer(null, 200);
} catch (Exception e) {
}
System.out.println(a.getId() + " - " + a.getBalance());
In this case, our transfer function would throw a NullPointerException when we try to access recipient.balance. However, because of the try-catch handling, we still get erroneous behavior. The above code prints:
1234 - 200.0
…indicating that the first half the transfer (deducting from the BankAccount a) still occurred! We could fix this in two ways, either add a null check:
public void transfer(BankAccount recipient, double amount) {
if (recipient == null) {
throw new NullPointerException();
}
balance -= amount;
recipient.balance += amount;
}
…or we could simply reverse the order of our operations so a NullPointerException would be thrown before we adjust any values:
public void transfer(BankAccount recipient, double amount) {
recipient.balance += amount;
balance -= amount;
}
Rollback and Throw
Sometimes while executing a function, it may not be obvious that a given request will violate our pre- or post- conditions of our functions until after we have completed the transactions. In this case, if an erroneous state is detected, then it’s important to rollback any changes the method made before throwing an exception, for the same reasons as in our withdraw case. That is, to make sure our code cannot be used to create an erroneous state.
For instance, imagine in our transfer function, we checked the value of the balances of each account at the end and realize that the sending account now has a negative balance! Well, we can’t let that stay, so we rollback the transaction, restoring both accounts to their starting balances, and then throw. However, the rollback must always occur before the throw.
public void transfer(BankAccount recipient, double amount) {
recipient.balance += amount;
balance -= amount;
if (balance < 0) {
//rollback
recipient.balance -= amount;
balance += amount;
//and throw
throw new InsufficientFundsException(getInsufficientFundsMessage(amount));
}
}
It is worth noting in this case, we could have checked the transaction at the start of the function to ensure that amount was greater than the sending account’s balance. However, I checked the post-condition status instead simply for the purposes of illustrating rollback and throw. The point is that there may be a situation where checking the post-conditions of a given operation may not be feasible before executing that operation. If you ever find yourself in that situation, remember to always rollback before throwing an exception.
assert keyword
One tool that can help during development (but not when you deploy) is the assert keyword. Similar to what assertTrue function does in JUnit, the assert statement takes the following format:
assert [boolean_statement]
If the asserted statement is false, then this results in an AssertionError, which is not the same as an exception, and cannot be caught with a try-catch for Exception or any sub-types.
public void withdraw(double amount) {
assert amount > 0 && amount <= balance;
balance -= amount;
}
However, by default in Java, AssertionErrors are actually disabled. To enable AssertionErrors to be thrown, you must add -ea to the VM Arguments (not command line arguments) in the run-configurations window:
Note that by default, VM arguments are hidden in IntelliJ. You can find them by clicking “modify options” (up and left of the class name you are running, in the above picture BankAccount.main). From there, click “add VM Options”. They can also be enabled with the hotkey Alt + V in the run configurations window.
However, assert statements should not be replied upon for deployed software! Instead, if you need an assert to enforce correct behavior, especially if the class is a library that will be used by client programs, then replace assert statements with Exceptions.