[Debian-med-packaging] Bug#724711: some notes

[Gert Wollny]

The last few days I did test the builds of ITK 4.8.2 on powerpc and
armhf - both without python bindings). Even so compiling took more than
24 hours, which means this was an experiment I'm not going to repeat
soon.

On armhf I have only a Ubuntu 15.10 installation (g++ 4.8), it builds
out of the box but the following tests fail:  

66:itkPatchBasedDenoisingImageFilterTest0
467:itkPatchBasedDenoisingImageFilterTestGaussian
468:itkPatchBasedDenoisingImageFilterTestRician
469:itkPatchBasedDenoisingImageFilterTestPoisson
789:itkHDF5ImageIOTest
983:itkReadWriteSpatialObjectTest
984:itkReadWriteSpatialObjectTest1
985:itkReadWriteSpatialObjectTest2
1563:itkSingleLevelSetsv4DenseImage2DTest
1565:itkSingleLevelSetsv4WhitakerImage2DTest
1566:itkSingleLevelSetsv4WhitakerImage2DTestThreads
1568:itkSingleLevelSetsv4ShiImage2DTest
1569:itkSingleLevelSetsv4WhitakerImage2DWithCurvatureTest
1573:itkLevelSetsv4EquationPropagationTermTest
1574:itkLevelSetsv4EquationLaplacianTermTest
1586:itkMultiLevelSetsv4WhitakerImageSubset2DTest

On powerpc the system was Debian/sid as of Saturday 2015/12/05. 

There was a problem with ITK_HAVE_SYNC_BUILTINS, I had to disable it,
otherwise some atomic operations were not properly linked. This may be
a problem with gcc. 

With this it compiled and the following tests failed: 

531:itkDicomDifficultFile
663:itkFFTWF_FFTTest
664:itkFFTWF_RealFFTTest
665:itkVnlFFTWF_FFTTest
666:itkVnlFFTWF_RealFFTTest
757:itkGDCMImageIOTest1
760:itkGDCMImageIOTest4
770:itkGDCMImageIONoCrashTest
789:itkHDF5ImageIOTest
1037:itkIOTransformMatlabTest
1563:itkSingleLevelSetsv4DenseImage2DTest
1565:itkSingleLevelSetsv4WhitakerImage2DTest
1566:itkSingleLevelSetsv4WhitakerImage2DTestThreads
1568:itkSingleLevelSetsv4ShiImage2DTest
1569:itkSingleLevelSetsv4WhitakerImage2DWithCurvatureTest
1573:itkLevelSetsv4EquationPropagationTermTest
1574:itkLevelSetsv4EquationLaplacianTermTest
1586:itkMultiLevelSetsv4WhitakerImageSubset2DTest

>From skimming over the logs I can say that there are all types of
errors - some are about floating point accuracy, but others are real
failures, like GDCM not understanding the input file, fftw not creating
a plan etc.