Better Error Handling with the New Exception Filters in C# 6
Like this article? We recommend
Like this article? We recommend
Like this article? We recommend
Exception filters have been added to C# 6 as part of the "co-evolution" strategy for C# and VB.NET. Exception filters have been present in VB.NET for several releases, and have also been available in F#. They're new for C# in version 6.
New Behaviors for Exception Filters
Exception filters now provide more general control over when an exception should be handled. In previous releases, the only way you could indicate to handle an exception was based on the type of exception. You caught all exceptions of a given type, including all exception classes derived from that given type. That syntax might not always cause limitations, but in several scenarios it resulted in more cumbersome error-handling code. Those scenarios are addressed by the exception filters in C# 6.
Using the new syntax for exception handlers is easy: Just add a when clause to a catch handler. If the expression following the when keyword evaluates to true, the catch block runs. If the expression following the when keyword evaluates to false, the exception continues to propagate:
try
{
DoSomeWork();
} catch (SomeException e) when (passesFilter(e))
{
// Handle errors....
}The syntax for exception filters was revised as this release was being developed. The original syntax used if rather than when:
// Original syntax used 'if' instead of 'when'
try
{
DoSomeWork();
} catch (SomeException e) if (passesFilter(e))
{
// Handle errors....
}This syntax was rejected because a little reformatting could make it look like a very different construct:
// Original syntax caused confusion
try
{
DoSomeWork();
} catch (SomeException e)
if (passesFilter(e))
{
// Handle errors....
}With that reformatting, developers could easily misread an exception filter as an if clause inside a catch block. To avoid that confusion, the keyword was changed from if to when.
Exceptions and Call Stacks
The syntax for exception filters doesn't indicate that anything new and interesting is happening, but there's quite a bit of difference in the underlying implementation. In terms of the logic, you could have achieved the same external behavior by using a catch clause containing the conditional logic of the exception filter. Whenever the exception filter returned false, you would rethrow the exception:
// Same logic, old-school C#:
try
{
DoSomeWork();
}
catch (SomeException e)
{
if (passesFilter(e))
{
// Handle errors....
}
else
{
throw;
}
}There are several advantages to the newer syntax. Using an exception filter, the exception is not caught and rethrown when the exception filter returns false. With this behavior, the original exception and all the call stack information will be preserved when your exception filters return false. If you used the classic syntax, the catch and rethrow interrupt that flow. When you use the throw statement with no arguments, the same exception is thrown, but the original information in the call stack has been modified. The catch clause has been executed, and then the exception is rethrown.
In a large production system, the older syntax can make finding and diagnosing errors much more difficult because you need to dig through more layers of exceptions to find the root cause of the problem. The newer exception filters are preferred in scenarios where an exception can be handled only based on other program states or conditions.
Scenarios for Exception Filters
The examples I've shown to this point provide a filter that returns true or false based on the properties of an exception. That's one of the core scenarios for exception filters. However, filters also can determine whether an exception should be processed based on any program state. For example, one use might be to retry a web request up to 10 times in case of a timeout failure:
var done = false;
do
{
int failures = 0;
try {
var data = SimulatedWebRequest();
workWith(data);
done = true;
} catch(TimeoutException e) when (failures++ < 10)
{
WriteLine("Timeout error: trying again");
}
} while (!done);You can expand on the idea above to create filters for any logical condition you need. The only restriction on the expression for an exception filter is that it must be synchronous: You cannot await an asynchronous task as part of an exception filter. Ideally, exception filter clauses also shouldn't be long-running (blocking) operations.
One common scenario for exception filters is to implement logging as a side-effect of an exception filter that always returns false. Let's update the first sample above to include logging:
try
{
DoSomeWork();
}
catch (Exception e) when (logException(e)) { }
catch (SomeException e) when (passesFilter(e))
{
// Handle errors....
}There are three important points to note here:
- The initial catch clause has an empty body. It never handles the exception.
- The logException() method must always return false. Otherwise, your logging method would cause the empty catch clause to swallow the exception.
- The catch clause for logging is above any catch clauses that might actually process the exception. This is true even though the logging catch clause is set to catch the most general exception class, System.Exception.
It's still true in C# 6 that once an exception has been handled by a catch clause, no catch clauses that follow will be executed. In previous versions of C#, we had to arrange our catch clauses carefully from the most specific exception classes to the most general. With the new exception filters in C# 6, we can have multiple catch clauses for the same exception class, with different conditions on the exception filters. We can also have scenarios like the one above, where the first catch clause catches a more general exception than a later clause would do. Because logException() would always return false, the other catch clauses are still examined in order. Care is still required; because the first catch clause whose exception filter returns true will process the exception, no catch clauses declared later will be entered.
This added flexibility can make it easier to write catch clauses for exceptions where some property or program variable determines the correct action.
There's one final scenario where I've found exception filters to be very valuable in debugging. If I can't track down the original cause of an exception because of multiple layers of catching, wrapping, and throwing, I still can ensure that an exception is not processed when I'm running an application with a debugger attached. If I add an exception filter to prevent catching exceptions when the debugger is attached, it's easier to find the root cause of an issue:
try
{
DoSomeWork();
}
catch (Exception e) when (logException(e)) { }
catch (SomeException e) when (passesFilter(e) && !Debugger.IsAttached)
{
// Handle errors....
}The code above includes one more exception filter that tests for the presence of a debugger attached to the current process. If a debugger is present, the catch clause is not entered, and the exception continues to propagate. If I add this filter in enough places, eventually every exception causes the debugger to stop on an unhandled exception error. It's a rather large hammer to find issues, but it's a very effective hammer.
Some Initial Guidance on Exception Filters
I use exception filters less often than other new features in C# 6. However, when I need this feature, I truly appreciate it. Exception filters are most helpful when I'm trying to diagnose errors in a large system, where exception handling is difficult, and we have limited flexibility in writing the code that handles, processes, and reports errors.
Furthermore, we don't typically write exception handling everywhere in our code. One of the advantages of using exceptions to communicate errors is that not every method must handle all possible errors. Exceptions continue to propagate up the call stack until a handler is found. We'll only add new exception filters in those methods that are already handling exceptions.
The enhanced syntax in C# 6 exception filters provides greater functionality and flexibility in methods that report or process exceptions. In new projects, I make use of the logging idiom I showed earlier. I also use the debugger-attached trick when I just can't find the root cause of an exception. Overall, I believe the revised exception filters will be used less often than some of the other features I've profiled in this series of articles. However, I do believe that it will be very well appreciated when it is used.