Friday, December 23, 2005

Handling InvocationTargetException

A quick hint on handling java.lang.reflect.InvocationTargetException.

This sucker is thrown whenever you invoke a method via reflection (typically via Method.invoke), and that method throws any sort of exception. So it doesn't indicate a problem with the reflection mechanism, it is simply how the reflection mechanism notifies you of exceptions thrown by the method you called.

Whenever you catch this exception (and you will have to catch it at some level), the first thing you should do is call getCause(). This returns the exception thrown from the method that was invoked via reflection, and that's what you care about. So whenever you need to log or report an occurrence of InvocationTargetException, always use the cause object rather than the InvocationTargetException itself.

Thursday, December 22, 2005

How not to catch Exceptions

When it comes to bad Java practices, top of my list would have to be catching exceptions like this:

try
{
   // stuff - anything really
}
catch (Exception e)
{
   // maybe log it if you're lucky
}

This is usually done because the "stuff" throws more than one kind of checked exception, and the code to handle each case is the same. So what's wrong with it? The biggest problem is that this not only catches the checked exceptions you're interested in, but also all RuntimeExceptions. The vast majority of RuntimeExceptions are only thrown from buggy code, and catching them is rarely a good idea. The above catch handles all kinds of exceptions the same way, regardless of whether it was caused by a bug or an ordinary failure that can happen from time to time when a program is running (e.g. a network failure). This only obscures bugs.

Consider the case where a bug causes a NullPointerException to be thrown - perhaps the most common type of bug in Java. If the catch clause simply logs the exception message (not uncommon), then you have to check the log to have any idea why your app is misbehaving. And if you don't log the stack trace (again, not uncommon), you still won't have a clue what caused it. A much better option is to simply let RuntimeExceptions go, which produces a screeching halt in your app and, in most cases, an immediate stack trace dump telling you exactly what happened. The details of what happens when you don't catch a RuntimeException at all will vary depending on the environment you're running in, but it's always safe to do so. For example, servlet containers usually output the stack trace in the HTTP response when a servlet throws a RuntimeException.

A good principle of catching exceptions is to only catch the those specific exceptions that you can actually deal with. E.g.

try
{
   // stuff that throws AException
   // stuff that throws BException
}
catch (AException e)
{
    // deal
}
catch (BException e)
{
    // deal
}

This way just deals with the exceptions that you can sensibly recover from and lets any RuntimeExceptions that indicate bugs to be thrown to the caller. If the code to handle both AException and BException is the same, you could always move it to a method.

"But what if the code throws, like, 10 different kinds of checked exceptions?". I've found it's actually pretty rare to have a block of code that throws more than about 4 unrelated checked exceptions. Some of the reflection APIs are pretty bad with this, however. It can be a pain to enter basically identical catch clauses for every one. A good alternative is:

try
{
   // stuff that throws AException
   // stuff that throws BException
   // stuff that throws CException
   // stuff that throws DException
}
catch (RuntimeException e)
{
   throw e;
}
catch (Exception e)
{
   // deal with all checked exceptions in the same way
}

This is actually not a bad way to get the best of both worlds (although I usually find it clearer to just handle each checked exception individually).

I think in hindsight, the inheritance hierarchy of exceptions is backwards. Instead of the RuntimeException extends Exception extends Throwable hierarchy, it should have been CheckedException extends UncheckedException extends Throwable. That would allow catching all checked exceptions in a single catch clause, while not obscuring unchecked exceptions thrown by buggy code.

In a similar vein, you almost never want to declare a method with throws Exception. This is basically throwing away the whole checked exception system and saying "this method can throw any kind of exception". This tends to make exception handling in the caller painful, and may give the caller no good option but to declare throws Exception itself. Before long, you end up with all your methods declared as throwing Exception and you have little indication of what exceptions you might actually have to deal with at runtime. (There is an argument to made that checked exceptions are a bad idea altogether, but that subject will have to wait for another day.)

There are a few cases where you might have a legitimate reason to catch Exception, or Error, or Throwable. Here are some of them:

  • When implementing Runnable. Reporting RuntimeExceptions thrown from Runnable.run is a bit of a pain in Java versions 1.4 and earlier. It's often easier just to catch them yourself and log the entire stack trace somewhere it can't be missed. Java 5 has a nicer mechanism for doing this, however.
  • When calling some sort of plug-in module. In this case, a bug in the plug-in doesn't mean the whole application needs to abort. You can catch the exception, notify the user, unload the plug-in and move on.
  • When dealing with poorly designed APIs that throw Exception.