Catching AggregateException

后端 未结 2 2048
情书的邮戳
情书的邮戳 2021-02-03 22:27

I am trying to throw and catch an AggregateException. I did not use exceptions very much on C#, but the behaviour I found is a little bit surprising.

My code is:

相关标签:
2条回答
  • 2021-02-03 22:57

    This is actually kind of interesting. I think the problem is that you're using AggregateException in an unexpected way, which is causing an error inside the PLINQ code.

    The entire point of AggregateException is to group together multiple exceptions that may occur simultaneously (or nearly so) in a parallel process. So AggregateException is expected to have at least one inner exception. But you're throwing new AggregateException("i"), which has no inner exceptions. The PLINQ code tries to examine the InnerExceptions property, hits some sort of error (probably a NullPointerException) and then it seems to go into a loop of some sort. This is arguably a bug in PLINQ, since you're using a valid constructor for AggregateException, even if it is an unusual one.

    As pointed out elsewhere, throwing ArgumentException would be more semantically correct. But you can get the behavior you're looking for by throwing a correctly-constructed AggregateException, for example by changing the IsEven function to something like this:

    private static bool IsEven(int i)
    {
        if (i % 10 == 0){
            //This is still weird
            //You shouldn't do this. Just throw the ArgumentException.
            throw new AggregateException(new ArgumentException("I hate multiples of 10"));
        }
        return i % 2 == 0;
    }
    

    I think the moral of the story is to not throw AggregateException unless you really know exactly what you're doing, particularly if you're already inside a parallel or Task-based operation of some kind.

    0 讨论(0)
  • 2021-02-03 23:12

    I agree with others: this is a bug in .Net and you should report it.

    The cause is in the method QueryEnd() in the internal class QueryTaskGroupState. Its decompiled (and slightly modified for clarity) code looks like this:

    try
    {
      this.m_rootTask.Wait();
    }
    catch (AggregateException ex)
    {
      AggregateException aggregateException = ex.Flatten();
      bool cacellation = true;
      for (int i = 0; i < aggregateException.InnerExceptions.Count; ++i)
      {
        var canceledException =
            aggregateException.InnerExceptions[i] as OperationCanceledException;
        if (IsCancellation(canceledException))
        {
          cacellation = false;
          break;
        }
      }
      if (!cacellation)
        throw aggregateException;
    }
    finally
    {
      this.m_rootTask.Dispose();
    }
    if (!this.m_cancellationState.MergedCancellationToken.IsCancellationRequested)
      return;
    if (!this.m_cancellationState.TopLevelDisposedFlag.Value)
      CancellationState.ThrowWithStandardMessageIfCanceled(
        this.m_cancellationState.ExternalCancellationToken);
    if (!userInitiatedDispose)
      throw new ObjectDisposedException(
        "enumerator", "The query enumerator has been disposed.");
    

    Basically, what this does is:

    • rethrow the flattened AggregateException if it contains any non-cancellation exceptions
    • throw new cancellation exception if cancellation was requested (or return without throwing, I don't really understand that part, but I don't think it's relevant here)
    • else throw ObjectDisposedException for some reason (assuming userInitiatedDispose is false, which it is)

    So, if you throw an AggregateException with no inner exceptions, ex will be an AggregateException containing your empty AggregateExcaption. Calling Flatten() will turn that into just an empty AggreateException, which means it doesn't contain any non-cancellation exception, so the first part of the code thinks this is cancellation and doesn't throw.

    But the second part of the code realizes this isn't cancellation, so it throws a completely bogus exception.

    0 讨论(0)
提交回复
热议问题