[Insight-developers] itkVTKImageIO testing question

[Bradley Lowekamp]

Hello Gabe,

This is the block of code you are referring too:


      int size = this->GetComponentSize();
      switch( size )
        {
        case 1:
          break;
        case 2:
          ByteSwapper<uint16_t>::SwapRangeFromSystemToBigEndian((uint16_t *)buffer, this->GetImageSizeInComponents() );
          break;
        case 4:
          ByteSwapper<uint32_t>::SwapRangeFromSystemToBigEndian((uint32_t *)buffer, this->GetImageSizeInComponents() );
          break;
#ifdef ITK_HAS_INT_64
        case 8:
          ByteSwapper<uint64_t>::SwapRangeFromSystemToBigEndian((uint64_t *)buffer, this->GetImageSizeInComponents() );
          break;
#endif
        default:
          itkExceptionMacro(<< "Unknown component size");
        }


For all systems that I am aware which ITK supports, the byte order is dependent only on the size (number of bytes) not the specific type. I think that double is 8 bytes on all systems... so likely uint64_t could be changed to double. I just thought that fixed width integer conveyed the size switching better.


I am trying to add ImageIOFactories for these new ImageIO classes and enable their automatic registration in ImageIOFactory. This would switch the current VTK tests to utilize the VTKImageIO2, and enable the reuse of the existing streaming IO related tests. However I have run into a road block. The library ITKIOReview depends on ITKIO, but adding the factories to ITKIO would add a circular dependency, and thankfully CMake was smart enough to know this was a problem before I did. So this is a significant problem to automatically using these ImageIOs, and it's unclear what the solution is.

Brad


On Jun 18, 2010, at 4:30 PM, Gabe Hart wrote:

> Hi Brad,
> 
> As one of my first projects after joining Kitware, I've been working on 
> importing the tests for itkMRCImageIO and itkVTKImageIO.  In doing so, 
> I've run across an issue that Luis and I are confused on and wanted your 
> input.  Following the format of your original test, my test for 
> itkVTKImageIO uses a templated class to execute write and read tests for 
> all supported pixel types (and ensured that IO fails on some of the 
> unsupported ones).
> 
> The test (itkVTKImageIO2Test) executes nicely on my machine (Ubuntu 
> 10.04 x64), but is failing in the continuous builds on BillsLaptop 
> (Win32 bcc5.5).  In particular, it fails to successfully read an image 
> whose pixel type is itk::Vector<double,3> when reading in binary mode.  
> I think it would also fail for a pixel type of double too.  The failure 
> seems to be coming from line 452 where an exception is thrown saying 
> "Unknown component size".  When adding a print statement to look at the 
> size variable for the switch, it seems that indeed size = 8.  This would 
> explain the success on my machine (64-bit) but the failure on 
> BillsLaptop (32-bit).
> 
> The question this then raises is whether this is the correct behavior 
> for this situation.  The switch statement seems to only consider 
> different sizes of ints and does not seem to handle the options for 
> floats or doubles.  In the present state, it seems to confuse a double 
> with a 64 bit integer and thus is unable to read it on a 32 bit 
> machine.  Is this something that should be fixed, or is there something 
> tricky going on behind the scenes?
> 
> Thanks for your help,
> -Gabe

========================================================
Bradley Lowekamp  
Lockheed Martin Contractor for
Office of High Performance Computing and Communications
National Library of Medicine 
blowekamp at mail.nih.gov


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/mailman/private/insight-developers/attachments/20100618/146ca879/attachment.htm>