Is ITK_USE_FLOAT_SPACE_PRECISION obsolete? Build errors when = YES.

[seanm]

I've been building ITK with various build option permutations (to try and shake out the last -Wextra-semi warnings) and it seems to me that ITK does not build with ITK_USE_FLOAT_SPACE_PRECISION = YES.

ex:

Modules/Core/ImageFunction/test/itkVectorInterpolateImageFunctionTest.cxx:223:14: error: no matching conversion for functional-style cast from 'itk::SpacePrecisionType[3]' (aka 'float[3]') to 'ContinuousIndexType' (aka 'ContinuousIndex<double, Self::ImageDimension>')
    cindex = ContinuousIndexType(darray);
             ^~~~~~~~~~~~~~~~~~~~~~~~~~

and

Modules/Filtering/ImageGrid/test/itkBinShrinkImageFilterTest2.cxx:45:12: error: no matching function for call to 'FloatAlmostEqual'
      if (!itk::Math::FloatAlmostEqual(pt[i], it.Get()[i]))
           ^~~~~~~~~~~~~~~~~~~~~~~~~~~

and others too.

Seems like no one is using this if it doesn't even build?

[dzenanz]

ITK_USE_FLOAT_SPACE_PRECISION tells whether the default precision should be float instead of double. I am fine with removal of this option, as long as float precision can be achieved by providing it as a (template) parameter instead of the default double to relevant classes (interpolators, PointTypes, etc).

[dzenanz]

Thorough testing of float space precision is hard, and having a build option to make float the default is an excellent way to test it. That's why I would prefer if the build with this option gets fixed, instead of the option being removed.

[dzenanz]

How many compile errors are there when ITK_USE_FLOAT_SPACE_PRECISION = YES?

[seanm]

I count 7, but it aborts building at 74%.

[dzenanz]

I kicked off a build with ITK_USE_FLOAT_SPACE_PRECISION ON, and I plan to tackle compile errors over the following days.

[seanm]

If you decide to fix the errors, you might want to do it in the release branch too.

[dzenanz]

This broke again with recent changes. Here is an output for one project:

------ Build started: Project: ITKImageGridGTestDriver, Configuration: Debug x64 ------
itkResampleImageFilterGTest.cxx
C:\Dev\ITK-git\Modules\Core\Common\include\itkImageAlgorithm.hxx(244,28): error C2664: 'itk::Point<itk::ImageBase<2>::PointValueType,2> itk::ImageBase<2>::TransformContinuousIndexToPhysicalPoint<itk::SpacePrecisionType,itk::SpacePrecisionType>(const itk::ContinuousIndex<TCoordinate,2> &) const': cannot convert argument 1 from 'itk::ImageAlgorithm::EnlargeRegionOverBox::ContinuousInputIndexType' to 'const itk::ContinuousIndex<TCoordinate,2> &'
C:\Dev\ITK-git\Modules\Core\Common\include\itkImageAlgorithm.hxx(244,28): error C2664:         with
C:\Dev\ITK-git\Modules\Core\Common\include\itkImageAlgorithm.hxx(244,28): error C2664:         [
C:\Dev\ITK-git\Modules\Core\Common\include\itkImageAlgorithm.hxx(244,28): error C2664:             TCoordinate=float
C:\Dev\ITK-git\Modules\Core\Common\include\itkImageAlgorithm.hxx(244,28): error C2664:         ]
(compiling source file '../../../../../ITK-git/Modules/Filtering/ImageGrid/test/itkResampleImageFilterGTest.cxx')
    C:\Dev\ITK-git\Modules\Core\Common\include\itkImageAlgorithm.hxx(245,9):
    Reason: cannot convert from 'itk::ImageAlgorithm::EnlargeRegionOverBox::ContinuousInputIndexType' to 'const itk::ContinuousIndex<TCoordinate,2>'
        with
        [
            TCoordinate=float
        ]
    C:\Dev\ITK-git\Modules\Core\Common\include\itkImageAlgorithm.hxx(245,9):
    No user-defined-conversion operator available that can perform this conversion, or the operator cannot be called
    C:\Dev\ITK-git\Modules\Core\Common\include\itkImageBase.h(575,3):
    see declaration of 'itk::ImageBase<2>::TransformContinuousIndexToPhysicalPoint'
    C:\Dev\ITK-git\Modules\Core\Common\include\itkImageBase.h(555,3):
    could be 'void itk::ImageBase<2>::TransformContinuousIndexToPhysicalPoint(const itk::ContinuousIndex<TIndexRep,2> &,itk::Point<TCoordinate,2> &) const'
        C:\Dev\ITK-git\Modules\Core\Common\include\itkImageAlgorithm.hxx(244,28):
        'void itk::ImageBase<2>::TransformContinuousIndexToPhysicalPoint(const itk::ContinuousIndex<TIndexRep,2> &,itk::Point<TCoordinate,2> &) const': expects 2 arguments - 1 provided
    C:\Dev\ITK-git\Modules\Core\Common\include\itkImageAlgorithm.hxx(244,28):
    while trying to match the argument list '(itk::ImageAlgorithm::EnlargeRegionOverBox::ContinuousInputIndexType)'
    C:\Dev\ITK-git\Modules\Core\Common\include\itkImageAlgorithm.hxx(244,28):
    the template instantiation context (the oldest one first) is
        C:\Dev\ITK-git\Modules\Filtering\ImageGrid\test\itkResampleImageFilterGTest.cxx(117,88):
        see reference to class template instantiation 'itk::ResampleImageFilter<itk::Image<int,2>,itk::Image<int,2>,float,float>' being compiled
        C:\Dev\ITK-git\Modules\Filtering\ImageGrid\include\itkResampleImageFilter.hxx(527,1):
        while compiling class template member function 'void itk::ResampleImageFilter<itk::Image<int,2>,itk::Image<int,2>,float,float>::GenerateInputRequestedRegion(void)'
        C:\Dev\ITK-git\Modules\Filtering\ImageGrid\include\itkResampleImageFilter.hxx(559,23):
        see reference to function template instantiation 'itk::ImageRegion<2> itk::ImageAlgorithm::EnlargeRegionOverBox<itk::Image<int,2>,itk::Image<int,2>,itk::Transform<TTransformPrecisionType,2,2>>(const itk::ImageRegion<2> &,const InputImageType *,const OutputImageType *,const TransformType *)' being compiled
        with
        [
            TTransformPrecisionType=float,
            InputImageType=itk::Image<int,2>,
            OutputImageType=itk::Image<int,2>,
            TransformType=itk::Transform<float,2,2>
        ]
itkTileImageFilterGTest.cxx
Generating Code...
Done building project "ITKImageGridGTestDriver.vcxproj" -- FAILED.

[dzenanz]

I now turned on ITK_USE_FLOAT_SPACE_PRECISION in Windows11-VS22x64-RelWithDebInfo-Favorite-Remotes nightly test build.

[seanm]

I now turned on ITK_USE_FLOAT_SPACE_PRECISION in Windows11-VS22x64-RelWithDebInfo-Favorite-Remotes nightly test build.

I hear Bill Lorensen saying 'untested code is broken code' in my head :)

[seanm]

a thought: Is this broken in the 5.4.1 tag?

[dzenanz]

Probably not. This got broken in master.

[hjmjohnson]

Can you give some context for why this is needed? I'm trying to think of a usecase where this has benefit.

Is there someone who uses this feature? v6.0.0 could be a good time to remove the feature.

PS: I'm building with this flag on with wrapping turned on.

[hjmjohnson]

addressed in

commit c9b342eefbd95a2506c8e816e33c11976ea0de62 (remove-alias)
Author: Hans Johnson <[email protected]>
Date:   2024-12-17 20:11:43 -0600

    COMP: Fix building with ITK_USE_FLOAT_SPACE_PRECISION=ON
    
    Need to match precision for types used to allow building.
    In a few cases double precision was assumed for values
    related to spacing precision.

commit cad42971d5d65d17b56359b06ab9924696cfdce1
Author: Hans Johnson <[email protected]>
Date:   2024-12-17 20:09:50 -0600

    COMP: Prohibit both ITK_USE_FLOAT_SPACE_PRECISION & ITK_WRAP_PYTHON
    
    Using float space precision is not supported with wrapping python.
    The python wrapping assumes a limited number of types, and those
    types would need to be aliased for supporting float space
    continuous indexes (and other types)

[dzenanz]

I also turned on ITK_USE_FLOAT_SPACE_PRECISION in nightly build named Windows11-VS22x64-RelWithDebInfo-Favorite-Remotes. If some change breaks it, we should know soon thereafter.