error in TensorFunctions.h during make on macOS with M2 chip

[x-avent]

When did the error occur?

[ ] CMake configuration (cmake / ccmake)
[X] Compilation (make)
[ ] Installation (make install)

Build environment

• OS: MacOS
• OS version: 14.1
• Type of system: macbook pro M2

ANTs version

git clone https://github.com/ANTsX/ANTs.git on Jan 16, 2024

Build configuration and logs

[build.log](https://github.com/ANTsX/ANTs/files/13956078/build.l
CMakeOutput.log
og)
CMakeError.log

CMakeCache.txt

Additional context

[cookpa]

Corrected link to build.log: https://github.com/ANTsX/ANTs/files/13956078/build.log

[cookpa]

Error message:

In file included from /Users/xeniagrande/ANTs/Examples/antsRegistration2DDouble.cxx:1:
In file included from /Users/xeniagrande/ANTs/Examples/antsRegistrationTemplateHeader.h:14:
In file included from /Users/xeniagrande/ANTs/Examples/itkantsRegistrationHelper.h:21:
In file included from /Users/xeniagrande/ANTs/Utilities/ReadWriteData.h:26:
In file included from /Users/xeniagrande/ANTs/Tensor/itkLogTensorImageFilter.hIn file included from :/Users/xeniagrande/ANTs/Examples/antsRegistrationTemplateHeader.cxx20::
1:
/Users/xeniagrande/ANTs/Tensor/TensorFunctions.hIn file included from :/Users/xeniagrande/ANTs/Examples/antsRegistrationTemplateHeader.h441::1414:
:In file included from  /Users/xeniagrande/ANTs/Examples/itkantsRegistrationHelper.h:error: no viable conversion from 'vnl_matrix_inverse<double>' to 'MatrixType' (aka 'vnl_matrix<double>')
21:
In file included from /Users/xeniagrande/ANTs/Utilities/ReadWriteData.h:26:
In file included from /Users/xeniagrande/ANTs/Tensor/itkLogTensorImageFilter.h:20:
/Users/xeniagrande/ANTs/Tensor/TensorFunctions.h:441  MatrixType inv = vnl_matrix_inverse<double>(DT);:
14             ^     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~:
 error: no viable conversion from 'vnl_matrix_inverse<double>' to 'MatrixType' (aka 'vnl_matrix<double>')
  MatrixType inv = vnl_matrix_inverse<double>(DT);
             ^     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/vnl_matrix.h:141:3: note: candidate constructor not viable: no known conversion from 'vnl_matrix_inverse<double>' to 'const vnl_matrix<double> &' for 1st argument
  vnl_matrix(vnl_matrix<T> const&);    // from another matrix.
  ^
/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/vnl_matrix.h:141:3: note: candidate constructor not viable: no known conversion from 'vnl_matrix_inverse<double>' to 'const vnl_matrix<double> &' for 1st argument
/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/vnl_matrix.h:148:3: note: candidate constructor not viable: no known conversion from 'vnl_matrix_inverse<double>' to 'vnl_matrix<double> &&' for 1st argument
  vnl_matrix(vnl_matrix<T> const&);    // from another matrix.
  ^  vnl_matrix(vnl_matrix<T> &&);

  ^
/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/algo/vnl_matrix_inverse.h/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/vnl_matrix.h:148:3: note: candidate constructor not viable: no known conversion from 'vnl_matrix_inverse<double>' to 'vnl_matrix<double> &&' for 1st argument
:40:  vnl_matrix(vnl_matrix<T> &&);12
:  ^ 
note: explicit conversion function is not a candidate
/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/algo/vnl_matrix_inverse.h:40:12: note: explicit conversion function is not a candidate
  explicit operator vnl_matrix<T>() const { return this->inverse(); }
           ^
  explicit operator vnl_matrix<T>() const { return this->inverse(); }
           ^
In file included from /Users/xeniagrande/ANTs/Examples/antsRegistration2DDouble.cxx:1:
In file included from /Users/xeniagrande/ANTs/Examples/antsRegistrationTemplateHeader.h:14:
In file included from /Users/xeniagrande/ANTs/Examples/itkantsRegistrationHelper.h:21:
In file included from /Users/xeniagrande/ANTs/Utilities/ReadWriteData.h:26:
In file included from /Users/xeniagrande/ANTs/Tensor/itkLogTensorImageFilter.h:20:
/Users/xeniagrande/ANTs/Tensor/TensorFunctions.h:840:14: error: no viable conversion from 'vnl_matrix_inverse<double>' to 'MatrixType' (aka 'vnl_matrix<double>')
  MatrixType inv = vnl_matrix_inverse<double>(DT);
             ^     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/vnl_matrix.h:141:3: note: candidate constructor not viable: no known conversion from 'vnl_matrix_inverse<double>' to 'const vnl_matrix<double> &' for 1st argument
  vnl_matrix(vnl_matrix<T> const&);    // from another matrix.
  ^
/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/vnl_matrix.h:148:3: note: candidate constructor not viable: no known conversion from 'vnl_matrix_inverse<double>' to 'vnl_matrix<double> &&' for 1st argument
  vnl_matrix(vnl_matrix<T> &&);
  ^
/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/algo/vnl_matrix_inverse.h:40:12: note: explicit conversion function is not a candidate
  explicit operator vnl_matrix<T>() const { return this->inverse(); }
           ^
In file included from /Users/xeniagrande/ANTs/Examples/antsRegistrationTemplateHeader.cxx:1:
In file included from /Users/xeniagrande/ANTs/Examples/antsRegistrationTemplateHeader.h:14:
In file included from /Users/xeniagrande/ANTs/Examples/itkantsRegistrationHelper.h:21:
In file included from /Users/xeniagrande/ANTs/Utilities/ReadWriteData.h:26:
In file included from /Users/xeniagrande/ANTs/Tensor/itkLogTensorImageFilter.h:20:
/Users/xeniagrande/ANTs/Tensor/TensorFunctions.h:840:14: error: no viable conversion from 'vnl_matrix_inverse<double>' to 'MatrixType' (aka 'vnl_matrix<double>')
  MatrixType inv = vnl_matrix_inverse<double>(DT);
             ^     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/vnl_matrix.h:141:3: note: candidate constructor not viable: no known conversion from 'vnl_matrix_inverse<double>' to 'const vnl_matrix<double> &' for 1st argument
  vnl_matrix(vnl_matrix<T> const&);    // from another matrix.
  ^
/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/vnl_matrix.h:148:3: note: candidate constructor not viable: no known conversion from 'vnl_matrix_inverse<double>' to 'vnl_matrix<double> &&' for 1st argument
  vnl_matrix(vnl_matrix<T> &&);
  ^
/Users/xeniagrande/build/staging/include/ITK-5.4/vnl/algo/vnl_matrix_inverse.h:40:12: note: explicit conversion function is not a candidate
  explicit operator vnl_matrix<T>() const { return this->inverse(); }
           ^

[cookpa]

Reproduced on my Intel Mac. Result of recent ITK changes perhaps. I will find where the problem begins

[cookpa]

ITK bump in #1651 causes this error on Mac but not Linux

[cookpa]

Compiles fine on Mac with previous ITK (fabc102c520bcbd21ea35e7303e3756c223e10d7), from June 2023. I don't see any changes to the ITK matrix classes that would have caused this.

[cookpa]

OK so the problem looks to be conversion between matrix types in a function added a long time ago by @stnava

ANTs/Tensor/TensorFunctions.h

Line 441 in 62550fc

MatrixType inv = vnl_matrix_inverse<double>(DT);

I don't get why it stopped working now, but the compile error goes away if I explicitly get the inverse matrix with

MatrixType inv = vnl_matrix_inverse<double>(DT).inverse();

It doesn't look like the enclosing function GetMetricTensorCost is used anywhere, so I can't test it further.

[cookpa]

@x-avent could you try building #1659? It works on my Intel Mac

[x-avent]

Many thanks for the quick help! It works fine now.