Best practices for exception management in Java or

问题:

I\'m stuck deciding how to handle exceptions in my application.

Much if my issues with exceptions comes from 1) accessing data via a remote service or 2) deserializing a JSON object. Unfortunately I can\'t guarantee success for either of these tasks (cut network connection, malformed JSON object that is out of my control).

As a result, if I do encounter an exception I simply catch it within the function and return FALSE to the caller. My logic is that all the caller really cares about is if the task was successful, not why it is wasn\'t successful.

Here\'s some sample code (in JAVA) of a typical method)

public boolean doSomething(Object p_somthingToDoOn)
{
    boolean result = false;

    try{
        // if dirty object then clean
        doactualStuffOnObject(p_jsonObject);

        //assume success (no exception thrown)
        result = true;
    }
    catch(Exception Ex)
    {
        //don\'t care about exceptions
        Ex.printStackTrace();
    }
    return result;
}

I think this approach is fine, but I\'m really curious to know what the best practices are for managing exceptions (should I really bubble an exception all the way up a call stack?).

In summary of key questions:

1. Is it okay to just catch exceptions but not bubble them up or formally notifying the system (either via a log or a notification to the user)?
2. What best practices are there for exceptions that don\'t result in everything requiring a try/catch block?

Follow Up/Edit

Thanks for all the feedback, found some excellent sources on exception management online:

• Best Practices for Exception Handling | O\'Reilly Media
• Exception Handling Best Practices in .NET
• Best Practices: Exception Management (Article now points to archive.org copy)
• Exception-Handling Antipatterns

It seems that exception management is one of those things that vary based on context. But most importantly, one should be consistent in how they manage exceptions within a system.

Additionally watch out for code-rot via excessive try/catches or not giving a exception its respect (an exception is warning the system, what else needs to be warned?).

Also, this is a pretty choice comment from m3rLinEz.

I tend to agree with Anders Hejlsberg and you that the most callers only care if operation is successful or not.

From this comment it brings up some questions to think about when dealing with exceptions:

• What is the point this exception being thrown?
• How does it make sense to handle it?
• Does the caller really care about the exception or do they just care if the call was successful?
• Is forcing a caller to manage a potential exception graceful?
• Are you being respectful to the idoms of the language?
  • Do you really need to return a success flag like boolean? Returning boolean (or an int) is more of a C mindset than a Java (in Java you would just handle the exception) one.
  • Follow the error management constructs associated with the language :) !

回答1:

It seems odd to me that you want to catch exceptions and turn them into error codes. Why do you think the caller would prefer error codes over exceptions when the latter is the default in both Java and C#?

As for your questions:

1. You should only catch exceptions that you can actually handle. Just catching exceptions is not the right thing to do in most cases. There are a few exceptions (e.g. logging and marshalling exceptions between threads) but even for those cases you should generally rethrow the exceptions.
2. You should definitely not have a lot of try/catch statements in your code. Again, the idea is to only catch exceptions you can handle. You may include a topmost exception handler to turn any unhandled exceptions into something somewhat useful for the end user but otherwise you should not try to catch each and every exception in every possible place.

回答2:

This depends on the application and the situation. If your building a library component, you should bubble up exceptions, although they should be wrapped to be contextual with your component. For example if your building an Xml Database and let\'s say you are using the file system to store your data, and you are using file system permissions to secure the data. You wouldn\'t want to bubble up a FileIOAccessDenied exception as that leaks your implementation. Instead you would wrap the exception and throw an AccessDenied error. This is especially true if you distribute the component to third parties.

As for if it\'s okay to swallow exceptions. That depends on your system. If your application can handle the failure cases and there is no benefit from notifying the user why it failed then go ahead, although I highly recommend that your log the failure. I\'ve always found it frustating being called to help troubleshoot an issue and find they were swallowing the exception (or replacing it and throwing a new one instead without setting the inner exception).

In general I use the following rules:

1. In my components & libraries I only catch an exception if I intend to handle it or do something based on it. Or if I want to provide additional contextual information in an exception.
2. I use a general try catch at the application entry point, or the highest level possible. If an exception gets here I just log it and let it fail. Ideally exceptions should never get here.

I find the following code to be a smell:

try
{
    //do something
}
catch(Exception)
{
   throw;
}

Code like this serves no point and should not be included.

回答3:

I would like to recommend another good source on the topic. It\'s an interview with inventors of C# and Java, Anders Hejlsberg and James Gosling respectively, on the topic of Java\'s Checked Exception.

Failure and Exceptions

There are also great resources at the bottom of the page.

I tend to agree with Anders Hejlsberg and you that the most callers only care if operation is successful or not.

Bill Venners: You mentioned scalability and versioning concerns with respect to checked exceptions. Could you clarify what you mean by those two issues?

Anders Hejlsberg: Let\'s start with versioning, because the issues are pretty easy to see there. Let\'s say I create a method foo that declares it throws exceptions A, B, and C. In version two of foo, I want to add a bunch of features, and now foo might throw exception D. It is a breaking change for me to add D to the throws clause of that method, because existing caller of that method will almost certainly not handle that exception.

Adding a new exception to a throws clause in a new version breaks client code. It\'s like adding a method to an interface. After you publish an interface, it is for all practical purposes immutable, because any implementation of it might have the methods that you want to add in the next version. So you\'ve got to create a new interface instead. Similarly with exceptions, you would either have to create a whole new method called foo2 that throws more exceptions, or you would have to catch exception D in the new foo, and transform the D into an A, B, or C.

Bill Venners: But aren\'t you breaking their code in that case anyway, even in a language without checked exceptions? If the new version of foo is going to throw a new exception that clients should think about handling, isn\'t their code broken just by the fact that they didn\'t expect that exception when they wrote the code?

Anders Hejlsberg: No, because in a lot of cases, people don\'t care. They\'re not going to handle any of these exceptions. There\'s a bottom level exception handler around their message loop. That handler is just going to bring up a dialog that says what went wrong and continue. The programmers protect their code by writing try finally\'s everywhere, so they\'ll back out correctly if an exception occurs, but they\'re not actually interested in handling the exceptions.

The throws clause, at least the way it\'s implemented in Java, doesn\'t necessarily force you to handle the exceptions, but if you don\'t handle them, it forces you to acknowledge precisely which exceptions might pass through. It requires you to either catch declared exceptions or put them in your own throws clause. To work around this requirement, people do ridiculous things. For example, they decorate every method with, \"throws Exception.\" That just completely defeats the feature, and you just made the programmer write more gobbledy gunk. That doesn\'t help anybody.

EDIT: Added more details on the converstaion

回答4:

Checked exceptions are a controversial issue in general, and in Java in particular (later on I\'ll try to find some examples for those in favor and opposed to them).

As rules of thumb, exception handling should be something around these guidelines, in no particular order:

• For the sake of maintainability, always log exceptions so that when you start seeing bugs, the log will assist in pointing you to the place your bug has likely started. Never leave printStackTrace() or the likes of it, chances are one of your users will get one of those stack traces eventually, and have exactly zero knowledge as to what to do with it.
• Catch exceptions you can handle, and only those, and handle them, don\'t just throw them up the stack.
• Always catch a specific exception class, and generally you should never catch type Exception, you are very likely to swallow otherwise important exceptions.
• Never (ever) catch Errors!!, meaning: Never catch Throwables as Errors are subclasses of the latter. Errors are problems you will most likely never be able to handle (e.g. OutOfMemory, or other JVM issues)

Regarding your specific case, make sure that any client calling your method will receive the proper return value. If something fails, a boolean-returning method might return false, but make sure the places you call that method are able to handle that.

回答5:

You should only catch the exceptions you can deal with. For example, if you\'re dealing with reading over a network and the connection times out and you get an exception you can try again. However if you\'re reading over a network and get a IndexOutOfBounds exception, you really can\'t handle that because you don\'t (well, in this case you wont) know what caused it. If you\'re going to return false or -1 or null, make sure it\'s for specific exceptions. I don\'t want a library I\'m using returning a false on a network read when the exception thrown is the heap is out of memory.

回答6:

Exceptions are errors that are not part of normal program execution. Depending on what your program does and its uses (i.e. a word processor vs. a heart monitor) you will want to do different things when you encounter an exception. I have worked with code that uses exceptions as part of normal execution and it is definitely a code smell.

Ex.

try
{
   sendMessage();

   if(message == success)
   {
       doStuff();
   }
   else if(message == failed)
   {
       throw;
   }
}
catch(Exception)
{
    logAndRecover();
}

This code makes me barf. IMO you should not recover from exceptions unless its a critical program. If your throwing exceptions then bad things are happening.

回答7:

All of the above seems reasonable, and often your workplace may have a policy. At our place we have defined to types of Exception: SystemException (unchecked) and ApplicationException (checked).

We have agreed that SystemExceptions are unlikely to be recoverable and will bve handled once at the top. To provide further context, our SystemExceptions are exteneded to indicate where they occurred, e.g. RepositoryException, ServiceEception, etc.

ApplicationExceptions could have business meaning like InsufficientFundsException and should be handled by client code.

Witohut a concrete example, it\'s difficult to comment on your implementation, but I would never use return codes, they\'re a maintenance issue. You might swallow an Exception, but you need to decide why, and always log the event and stacktrace. Lastly, as your method has no other processing it\'s fairly redundant (except for encapsulation?), so doactualStuffOnObject(p_jsonObject); could return a boolean!

回答8:

After some thought and looking at your code it seems to me that you are simply rethrowing the exception as a boolean. You could just let the method pass this exception through (you don\'t even have to catch it) and deal with it in the caller, since that\'s the place where it matters. If the exception will cause the caller to retry this function, the caller should be the one catching the exception.

It can at times happen that the exception you are encountering will not make sense to the caller (i.e. it\'s a network exception), in which case you should wrap it in a domain specific exception.

If on the other hand, the exception signals an unrecoverable error in your program (i.e. the eventual result of this exception will be program termination) I personally like to make that explicit by catching it and throwing a runtime exception.

回答9:

If you are going to use the code pattern in your example, call it TryDoSomething, and catch only specific exceptions.

Also consider using an Exception Filter when logging exceptions for diagnostic purposes. VB has language support for Exception filters. The link to Greggm\'s blog has an implementation that can be used from C#. Exception filters have better properties for debuggability over catch and rethrow. Specifically you can log the problem in the filter and let the exception continue to propagate. That method allows an attaching a JIT (Just in Time) debugger to have the full original stack. A rethrow cuts the stack off at the point it was rethrown.

The cases where TryXXXX makes sense are when you are wrapping a third party function that throws in cases that are not truly exceptional, or are simple difficult to test without calling the function. An example would be something like:

// throws NumberNotHexidecimalException
int ParseHexidecimal(string numberToParse); 

bool TryParseHexidecimal(string numberToParse, out int parsedInt)
{
     try
     {
         parsedInt = ParseHexidecimal(numberToParse);
         return true;
     }
     catch(NumberNotHexidecimalException ex)
     {
         parsedInt = 0;
         return false;
     }
     catch(Exception ex)
     {
         // Implement the error policy for unexpected exceptions:
         // log a callstack, assert if a debugger is attached etc.
         LogRetailAssert(ex);
         // rethrow the exception
         // The downside is that a JIT debugger will have the next
         // line as the place that threw the exception, rather than
         // the original location further down the stack.
         throw;
         // A better practice is to use an exception filter here.
         // see the link to Exception Filter Inject above
         // http://code.msdn.microsoft.com/ExceptionFilterInjct
     }
}

Whether you use a pattern like TryXXX or not is more of a style question. The question of catching all exceptions and swallowing them is not a style issue. Make sure unexpected exceptions are allowed to propagate!

回答10:

I suggest taking your cues from the standard library for the language you\'re using. I can\'t speak for C#, but let\'s look at Java.

For example java.lang.reflect.Array has a static set method:

static void set(Object array, int index, Object value);

The C way would be

static int set(Object array, int index, Object value);

... with the return value being a success indicator. But you\'re not in C world any more.

Once you embrace exceptions, you should find that it makes your code simpler and clearer, by moving your error handling code away from your core logic. Aim to have lots of statements in a single try block.

As others have noted - you should be as specific as possible in the kind of exception you catch.

回答11:

If you\'re going to catch an Exception and return false, it should be a very specific exception. You\'re not doing that, you\'re catching all of them and returning false. If I get a MyCarIsOnFireException I want to know about it right away! The rest of the Exceptions I might not care about. So you should have a stack of Exception handlers that say \"whoa whoa something is wrong here\" for some exceptions (rethrow, or catch and rethrow a new exception that explains better what happened) and just return false for others.

If this is a product that you\'ll be launching you should be logging those exceptions somewhere, it will help you tune things up in the future.

Edit: As to the question of wrapping everything in a try/catch, I think the answer is yes. Exceptions should be so rare in your code that the code in the catch block executes so rarely that it doesn\'t hit performance at all. An exception should be a state where your state machine broke and doesn\'t know what to do. At least rethrow an exception that explains what was happening at the time and has the caught exception inside of it. \"Exception in method doSomeStuff()\" isn\'t very helpful for anyone who has to figure out why it broke while you\'re on vacation (or at a new job).

回答12:

My strategy:

If the original function returned void I change it to return bool. If exception/error occurred return false, if everything was fine return true.

If the function should return something then when exception/error occurred return null, otherwise the returnable item.

Instead of bool a string could be returned containing the description of the error.

In every case before returning anything log the error.

回答13:

Some excellent answers here. I would like to add, that if you do end up with something like you posted, at least print more than the stack trace. Say what you were doing at the time, and Ex.getMessage(), to give the developer a fighting chance.

回答14:

try/catch blocks form a second set of logic embedded over the first (main) set, as such they are a great way to pound out unreadable, hard to debug spaghetti code.

Still, used reasonably they work wonders in readability, but you should just follow two simple rules:

• use them (sparingly) at the low-level to catch library handling issues, and stream them back into the main logical flow. Most of the error handling we want, should be coming from the code itself, as part of the data itself. Why make special conditions, if the returning data isn\'t special?

• use one big handler at the higher-level to manage any or all of the weird conditions arising in the code that aren\'t caught at a low-level. Do something useful with the errors (logs, restarts, recoveries, etc).

Other than these two types of error handling, all of the rest of the code in the middle should be free and clear of try/catch code and error objects. That way, it works simply and as expected no matter where you use it, or what you do with it.

Paul.

回答15:

I may be a little late with the answer but error handling is something that we can always change and evolve along time. If you want to read something more about this subject I wrote a post in my new blog about it. http://taoofdevelopment.wordpress.com

Happy coding.