[Insight-developers] ExceptionObject assignment crash on MSVC

[Luis Ibanez]

Hi Niels,

Here is what I have found so far:


In Visual Studio 7.1, the superclass of itk::ExceptionObject,
the std::exception, is not initializing its "what" component.

When running in Debug mode, it crashes in the expressions

         std::cout <<  E << std::endl;

the "what" component of std::exception is only set at construction
time, and we are not explicitly setting it up.

This can be fixed in the three constructors of itk::ExceptionObject
with the following changes:

-------------------------------------------------------------
$ cvs -q diff
Index: itkExceptionObject.cxx
===================================================================
RCS file: /cvsroot/Insight/Insight/Code/Common/itkExceptionObject.cxx,v
retrieving revision 1.14
diff -r1.14 itkExceptionObject.cxx
134c134
< ExceptionObject::ExceptionObject()
---
 > ExceptionObject::ExceptionObject():Superclass("")
144c144
< :
---
 > :Superclass(desc),
154c154
< :
---
 > :Superclass(desc.c_str()),
161c161
< Superclass(orig),
---
 > Superclass(orig.what()),
------------------------------------------------------

This bring us up to the next problem:


Apparently in the triple assignment:

         E = F = G;

Something goes wrong in the reference counting of E,F,G
is done improperly, and when we get to the next line:

          std::cout << F << std::endl;

it attempts to use an invalid object for invoking
its Print() method.


      This requires a closer look....


...and still is unclear why is that it is a
problem only in VS71 (and maybe the same in VS6 ?).



     Luis



------------------
Niels Dekker wrote:
> Luis Ibanez wrote:
> 
>> Removing the lines from the test may not be the right way to
>> address the problem. That will just give us a deceptively
>> green Dashboard and a false sense of security.
> 
> 
> Sorry, typo, I meant to say: feel free to remove those two lines of code 
> from itkExceptionObject.cxx:
> 
>  // Assign its superclass:
>  static_cast<Superclass &>(*this) = orig;
> 
> I didn't mean to suggest changing the test.
> 
> 
>> The memory leak reported for std::exception shouldn't result
>> in a crash, isn't it? ... it should just ... leak memory.
> 
> 
> Good point. But still, the bug report on the Microsoft website gives me 
> a bad feeling about their implementation of std::exception::operator=. 
> Hope you can prove me wrong.
> 
>> I'll try running this in Debug mode...
> 
> 
> Please do. Note that the previous version of itk::ExceptionObject 
> (before I did anything to the class) didn't call 
> std::exception::operator=.  And it didn't crash!
> 
> 
> Now I'm really going home!  Sorry to leave you with the crash (at least 
> for today).
> 
> 
> Kind regards,
> 
> Niels
> 
> 
> 
> 
> 
>