What is the best practice in C# to propagate an exception thrown in a finally block without losing an exception from a catch block?

Question

When an exception is possible to be thrown in a finally block how to propagate both exceptions - from catch and from finally?

As a possible solution - using an AggregateException:

internal class MyClass
{
    public void Do()
    {
        Exception exception = null;
        try
        {
            //example of an error occured in main logic
            throw new InvalidOperationException();
        }
        catch (Exception e)
        {
            exception = e;
            throw;
        }
        finally
        {
            try
            {
                //example of an error occured in finally
                throw new AccessViolationException();
            }
            catch (Exception e)
            {
                if (exception != null)
                    throw new AggregateException(exception, e);
                throw;
            }
        }
    }
}

These exceptions can be handled like in following snippet:

private static void Main(string[] args)
{
    try
    {
        new MyClass().Do();
    }
    catch (AggregateException e)
    {
        foreach (var innerException in e.InnerExceptions)
            Console.Out.WriteLine("---- Error: {0}", innerException);
    }
    catch (Exception e)
    {
        Console.Out.WriteLine("---- Error: {0}", e);
    }

    Console.ReadKey();
}

Answer 1

I regularly come into the same situation and have not found a better solution yet. But I think the solution suggested by the OP is eligible.

Here's a slight modification of the original example:

internal class MyClass
{
    public void Do()
    {
        bool success = false;
        Exception exception = null;
        try
        {
            //calling a service that can throw an exception
            service.Call();
            success = true;
        }
        catch (Exception e)
        {
            exception = e;
            throw;
        }
        finally
        {
            try
            {
                //reporting the result to another service that also can throw an exception
                reportingService.Call(success);
            }
            catch (Exception e)
            {
                if (exception != null)
                    throw new AggregateException(exception, e);
                throw;
            }
        }
    }
}

IMHO it will be fatal to ignore one or the other exception here.

Another example: Imagin a test system that calibrates a device (DUT) and therefore has to control another device that sends signals to the DUT.

internal class MyClass
{
    public void Do()
    {
        Exception exception = null;
        try
        {
            //perform a measurement on the DUT
            signalSource.SetOutput(on);
            DUT.RunMeasurement();
        }
        catch (Exception e)
        {
            exception = e;
            throw;
        }
        finally
        {
            try
            {
                //both devices have to be set to a valid state at end of the procedure, independent of if any exception occurred
                signalSource.SetOutput(off);
                DUT.Reset();
            }
            catch (Exception e)
            {
                if (exception != null)
                    throw new AggregateException(exception, e);
                throw;
            }
        }
    }
}

In this example, it is important that all devices are set to a valid state after the procedure. But both devices also can throw exceptions in the finally block that must not get lost or ignored.

Regarding the complexity in the caller, I do not see any problem there either. When using System.Threading.Tasks the WaitAll() method, for example, can also throw AgregateExceptions that have to be handled in the same way.

One more note regarding @damien's comment: The exception is only caught to wrap it into the AggregateException, in case that the finally block throws. Nothing else is done with the exception nor is it handled in any way.

For those who want to go this way you can use a little helper class I created recently:

public static class SafeExecute
{
    public static void Invoke(Action tryBlock, Action finallyBlock, Action onSuccess = null, Action<Exception> onError = null)
    {
        Exception tryBlockException = null;

        try
        {
            tryBlock?.Invoke();
        }
        catch (Exception ex)
        {
            tryBlockException = ex;
            throw;
        }
        finally
        {
            try
            {
                finallyBlock?.Invoke();
                onSuccess?.Invoke();
            }
            catch (Exception finallyBlockException)
            {
                onError?.Invoke(finallyBlockException);

                // don't override the original exception! Thus throwing a new AggregateException containing both exceptions.
                if (tryBlockException != null)
                    throw new AggregateException(tryBlockException, finallyBlockException);

                // otherwise re-throw the exception from the finally block.
                throw;
            }
        }
    }
}

and use it like this:

public void ExecuteMeasurement(CancellationToken cancelToken)
{
    SafeExecute.Invoke(
        () => DUT.ExecuteMeasurement(cancelToken),
        () =>
        {
            Logger.Write(TraceEventType.Verbose, "Save measurement results to database...");
            _Db.SaveChanges();
        },
        () => TraceLog.Write(TraceEventType.Verbose, "Done"));
}

Answer 2

As the comments have suggested this may indicate "unfortunately" structured code. For example if you find yourself in this situation often it might indicate that you are trying to do too much within your method. You only want to throw and exception if there is nothing else you can do (your code is 'stuck' with a problem you can't program around. You only want to catch an exception if there is a reasonable expectation you can do something useful. There is an OutOfMemoryException in the framework but you will seldom see people trying to catch it, because for the most part it means you're boned :-)

If the exception in the finally block is a direct result of the exception in the try block, returning that exception just complicates or obscures the real problem, making it harder to resolve. In the rare case where there is a validate reason for returning such as exception then using the AggregateException would be the way to do it. But before taking that approach ask yourself if it's possible to separate the exceptions into separate methods where a single exception can be returned and handled (separately).