Catching exceptions with "catch, when"

Catch blocks already allow you to filter on the type of the exception:

catch (SomeSpecificExceptionType e) {...}

The when clause allows you to extend this filter to generic expressions.

Thus, you use the when clause for cases where the type of the exception is not distinct enough to determine whether the exception should be handled here or not.

A common use case are exception types which are actually a wrapper for multiple, different kinds of errors.

Here's a case that I've actually used (in VB, which already has this feature for quite some time):

try
{
    SomeLegacyComOperation();
}
catch (COMException e) when (e.ErrorCode == 0x1234)
{
    // Handle the *specific* error I was expecting. 
}

Same for SqlException, which also has an ErrorCode property. The alternative would be something like that:

try
{
    SomeLegacyComOperation();
}
catch (COMException e)
{
    if (e.ErrorCode == 0x1234)
    {
        // Handle error
    }
    else
    {
        throw;
    }
}

which is arguably less elegant and slightly breaks the stack trace.

In addition, you can mention the same type of exception twice in the same try-catch-block:

try
{
    SomeLegacyComOperation();
}
catch (COMException e) when (e.ErrorCode == 0x1234)
{
    ...
}
catch (COMException e) when (e.ErrorCode == 0x5678)
{
    ...
}

which would not be possible without the when condition.


From Roslyn's wiki (emphasis mine):

Exception filters are preferable to catching and rethrowing because they leave the stack unharmed. If the exception later causes the stack to be dumped, you can see where it originally came from, rather than just the last place it was rethrown.

It is also a common and accepted form of “abuse” to use exception filters for side effects; e.g. logging. They can inspect an exception “flying by” without intercepting its course. In those cases, the filter will often be a call to a false-returning helper function which executes the side effects:

private static bool Log(Exception e) { /* log it */ ; return false; }

… try { … } catch (Exception e) when (Log(e)) { }

The first point is worth demonstrating.

static class Program
{
    static void Main(string[] args)
    {
        A(1);
    }

    private static void A(int i)
    {
        try
        {
            B(i + 1);
        }
        catch (Exception ex)
        {
            if (ex.Message != "!")
                Console.WriteLine(ex);
            else throw;
        }
    }

    private static void B(int i)
    {
        throw new Exception("!");
    }
}

If we run this in WinDbg until the exception is hit, and print the stack using !clrstack -i -a we'll see the just the frame of A:

003eef10 00a7050d [DEFAULT] Void App.Program.A(I4)

PARAMETERS:
  + int i  = 1

LOCALS:
  + System.Exception ex @ 0x23e3178
  + (Error 0x80004005 retrieving local variable 'local_1')

However, if we change the program to use when:

catch (Exception ex) when (ex.Message != "!")
{
    Console.WriteLine(ex);
}

We'll see the stack also contains B's frame:

001af2b4 01fb05aa [DEFAULT] Void App.Program.B(I4)

PARAMETERS:
  + int i  = 2

LOCALS: (none)

001af2c8 01fb04c1 [DEFAULT] Void App.Program.A(I4)

PARAMETERS:
  + int i  = 1

LOCALS:
  + System.Exception ex @ 0x2213178
  + (Error 0x80004005 retrieving local variable 'local_1')

That information can be very useful when debugging crash dumps.


When an exception is thrown, the first pass of exception handling identifies where the exception will get caught before unwinding the stack; if/when the "catch" location is identified, all "finally" blocks are run (note that if an exception escapes a "finally" block, processing of the earlier exception may be abandoned). Once that happens, code will resume execution at the "catch".

If there is a breakpoint within a function that's evaluated as part of a "when", that breakpoint will suspend execution before any stack unwinding occurs; by contrast, a breakpoint at a "catch" will only suspend execution after all finally handlers have run.

Finally, if lines 23 and 27 of foo call bar, and the call on line 23 throws an exception which is caught within foo and rethrown on line 57, then the stack trace will suggest that the exception occurred while calling bar from line 57 [location of the rethrow], destroying any information about whether the exception occurred in the line-23 or line-27 call. Using when to avoid catching an exception in the first place avoids such disturbance.

BTW, a useful pattern which is annoyingly awkward in both C# and VB.NET is to use a function call within a when clause to set a variable which can be used within a finally clause to determine whether the function completed normally, to handle cases where a function has no hope of "resolving" any exception that occurs but must nonetheless take action based upon it. For example, if an exception is thrown within a factory method which is supposed to return an object that encapsulates resources, any resources that were acquired will need to be released, but the underlying exception should percolate up to the caller. The cleanest way to handle that semantically (though not syntactically) is to have a finally block check whether an exception occurred and, if so, release all resources acquired on behalf of the object that is no longer going to be returned. Since cleanup code has no hope of resolving whatever condition caused the exception, it really shouldn't catch it, but merely needs to know what happened. Calling a function like:

bool CopySecondArgumentToFirstAndReturnFalse<T>(ref T first, T second)
{
  first = second;
  return false;
}

within a when clause will make it possible for the factory function to know that something happened.

Tags:

C#

Exception Handling

Recent Posts