[Insight-developers] LightObject Destructor and Exception Safety

[Bill Lorensen]

Brad,

Has this exception ever been thrown? If not, then we can:

1) remove the test
2) issue a warning
3) leave it there with a comment

Bill

On 4/4/07, Brad King <brad.king at kitware.com> wrote:
>
> Hello All,
>
> The LightObject::~LightObject() destructor contains this code:
>
>   /**
>    * warn user if reference counting is on and the object is being
>    * referenced by another object
>    */
>   if ( m_ReferenceCount > 0)
>     {
>     itkExceptionMacro(
>       << "Trying to delete object with non-zero reference count.");
>     }
>
> A basic rule for exception safety is that destructors should not throw
> exceptions.  I can explain why in more detail if anyone wants.  The
> following problem we've encountered is one reason.
>
> Consider a subclass of LightObject whose constructor throws an exception
> because it fails to allocate some resource.  This is what happens:
>
> 1.) The LightObject constructor initializes m_ReferenceCount to 1.
>     In the case that the whole object is constructed the NewMacro
>     removes this extra reference before returning the smart pointer
>     holding the object.
> 2.) The subclass constructor throws an exception.
>     The partially-constructed object must be destroyed.
> 3.) The destructor of LightObject is called to destroy the piece
>     of the object that was constructed before the exception was thrown.
> 4.) The above code throws another exception.
> 5.) In C++ when an exception is thrown during the handling of another
>     exception the program terminates.
>
> The real problem is that the destructor should not be throwing an
> exception.  The symptom (#5) can be fixed by using
> std::uncaught_exception to detect if an exception is already being
> handled:
>
>   if(m_ReferenceCount > 0 && !std::uncaught_exception())
>     {
>     itkExceptionMacro(
>       << "Trying to delete object with non-zero reference count.");
>     }
>
> However, I think we should come up with an alternative means of
> reporting the early destruction of an object under normal circumstances.
>
> Ideas?
> -Brad
> _______________________________________________
> Insight-developers mailing list
> Insight-developers at itk.org
> http://www.itk.org/mailman/listinfo/insight-developers
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.itk.org/mailman/private/insight-developers/attachments/20070404/3f4578a0/attachment.htm