Java Exception Handling Best Practices

Background

As a professional Java Developer, over the past decade or so, I've worked with all sorts of Java developers and I have had the opportunity to maintain thousands of lines of second-hand Java code. One thing that continues to pop up, even to this day, is bad exception handling. It is almost as if exception handling is a an afterthought to a lot of Java developers. A lot of times, I see exceptions being treated as a nuisance, an inconvenience, a burden, akin to child-proof pill bottles (which are proven to be more adult-proof than child-proof), and flashing VCR clocks (you youngsters out there surely have no idea what I'm talking about).

Add a little tequila...     
...to your Java

Overview

In this article, I'll discuss three common exception handling mistakes, why they are mistakes, and the correct approach that should be taken in each situation.

Common Mistake #1 - Using Exceptions as Conditionals

This mistake occurs when a developer uses a try/catch in a situation where a simple if/else statement would suffice. For example, handling null values:

Object somevalue = ...
try
{
     somevalue.doSomething(); //do the non-null condition
}
catch(NullPointerException npe)
{
     ...//do the null condition
}

Best Practice

For both performance and asthetics, the example above is better coded as:

Object somevalue = ...
if(somevalue != null)
{
     somevalue.doSomething(); //do the non-null condition
}
else
{
     ...//do the null condition
}

It is important to note, though, that in some situations it is acceptable to use a try/catch for conditional coding, but it should be limited to situations where the cost of determining the conditional state is greater than or equal to the cost of performing the "normal-state" operation. For an example of this, see the integer parsing code in the "Best Practice" to Common Mistake #2 below.

Common Mistake #2 - Exception Dropping

No, this isn't what you do at a party to brag about all of the "famous" exceptions you know (think Name Dropping). Instead, this is the practice where a developer makes an (usually wrong) assumption that an exception in a specific situation is irrelevant or unimportant, therefore choosing to catch the exception and do absolutely nothing with it. For example:

try
{
     //... do something that potentially throws an IOException
}
catch(IOException ioe)
{
}

In this example, should an IOException be thrown in the try clause, the catch clause traps the exception and then drops it. Should something bad happen, the developer and/or user won't know about it, and the program will continue after the catch as if nothing has happened.

Best Practice

There are actually two different solutions to this problem, but only one of them is correct in any given situation. The developer must analyze the situation and choose the most correct approach. First, the developer should determine if it is really necessary to catch the exception in the first place. It may actually be more reasonable to allow the exception to propogate up the stack. A few questions to ask are, "Can I recover from this exception?", "Do I have an alternate course of action should this exception occur?", and "Is this exception purely a local concern, or would the caller of my method be interested in this exception?" The answer to these questions, may lead the developer to determine that the exception should be propogated up the stack.

In the second situation, a developer needs to ask herself if it is genuinely reasonable for the exception to be dropped. There are actually a few situations where it is perfectly reasonable to drop an exception. The following example illustrates a situation where dropping an exception would be considered reasonable:

int intValue = 1; //if someValuePassedIn is invalid, default to 1
try
{
     intValue = Integer.parseInt(someValuePassedIn);
}
catch(NumberFormatExcepion nfe)
{
     //ignore NumberFormatException because we want to default to 1 if someValuePassedIn is invalid
}

In this example, the exception is dropped, but a comment is placed in the code to illustrate that the developer has at least thought about the consequences of dropping the exception and has accepted the responsibility for doing so. It should also be noted that this example seems to violate "Common Mistake #1" above. However, because it would be as costly, or more costly to try to determine if the "someValuePassedIn" is a valid integer, it is better here to determine its validity as we parse it.

An alternative to this solution may be to log the exception stack trace for later use should the same type of exception be unexpectedly thrown by another line of code within the try clause.

Common Mistake #3 - Exception Generalization

This is probably the most common and over-abused exception handling mistake I encounter. This is the practice where a developer will declare a method's throws clause or a try's catch clause to throw or catch a java.lang.Exception or even worse, a java.lang.Throwable. For example:

public void myMethod() throws Throwable
{
     try
     {
          ...//do something
     }
     catch(Exception e)
     {
          ...//do something
     }
}

This practice can have all sorts of nasty side effects. First of all, catching overgeneralized exceptions has two negative consequences. One, it can cause the code to inappropriately handle an exception that the developer had no expectation of the code throwing in the first place. RuntimeException is a perfect example of this phenomenon. By catching java.lang.Exception, due to the inheritance model, a developer will also catch all RuntimeExceptions which are typically thrown when an unexpected logic condition has occurred, usually as the result of a bug and/or coding issue. In the majority of the situations where a RuntimeException is thrown, the correct approach is to log the exception and let the system die gracefully as to allow the developer to fix whatever logic condition caused the problem, but by catching java.lang.Exception, depending on how it is handled, the developer may never know this logic condition ever occurred.

The second negative consequence of catching an overgeneralized exception is that it can prevent the code calling a method from ever realizing an exceptional state that it had an expectation or duty of handling. Catching java.lang.Throwable illustrates this point very well. By catching Throwable, the catch clause will, by definition, also catch all java.lang.Error objects that are thrown. According to Sun's documentation and by definition, "An Error is a subclass of Throwable that indicates serious problems that a reasonable application should not try to catch." (see java.lang.Error). So by catching Throwable, a developer is violating this convention and potentially catching Errors that in reality should be propogated up to the JVM so that the JVM can die gracefully. Instead, by catching the Error, the JVM may enter an unpredictable state and become very unstable. A good example of this phenomenon would be the java.lang.OutOfMemoryError. In earlier versions of Java, catching an OutOfMemoryError would cause the VM to attempt to push a frame onto the stack which would in turn cause another OutOfMemoryError to be thrown, which at best would cause the potential for stack trace information being lost (if any were available to begin with) and at worst, depending on the code structure, could cause code to get caught in an infinite loop of catching and throwing OutOfMemoryError.

Secondly, throwing overgeneralized exceptions has its own problems. The biggest problem of this practice is that subsequent developers wishing to make calls to these methods will have no reasonable expectation of what exceptions may be thrown. This prevents developers from being able to write defensive code, or write code that has any chance of reasonably dealing with any exception that occurs as a result of invoking such a method.

Best Practice

Methods should only be declared to throw whatever sepcialized exceptions that can logically be expected to be thrown. Likewise, a developer should only catch specialized exceptions that she has a reasonable expectation of handling. Also, even though it may seem overly verbose and overly complicated it is better to have many specialized catch statements than a single generalized catch statement, even if all of the specialized exceptions will be handled the same way. For example:

try
{
     ... //do something here that might cause several different types of exceptions to occur
}
catch(SomeExceptionA sea)
{
     handle(sea);
}
catch(SomeExceptionB seb)
{
     handle(seb);
}
catch(SomeExceptionC sec)
{
     handle(sec);
}
catch(SomeExceptionD sed)
{
     handle(sed);
}

private void handle(Exception e)
{
     ... //generically handle any Exception e
}

is preferrable over:

try
{
     ... //do something here that might cause several different types of exceptions to occur
}
catch(Exception e)
{
     ... //generically handle any Exception e
}

Even though it is more verbose and takes more effort to code, the first example eliminates any possibility of inadvertantly catching and handling any excpetion that wasn't intended to be caught.

Summary

By avoiding the common mistakes illustrated in this article, and instead following the suggested best practices, a developer can make their code more reliable, and execute more predictably, resulting in more efficient, less buggy, and more easily maintained code.

Reader Comments:

02/19/2009 2:01 PM - Felipe wrote:
Excellent article. And you're right bad exception handling is the most common error that we have to face, or even make most often. But exceptions are really powerful if you learn how to use it.

Post a Comment:

Your Name:
Your Comment:
 Last Updated: Tuesday June 10, 2014 Copyright © 2024 Wade Wassenberg & AgaveBlue.org