Incorrect solver detection in HexahedronCompositeFEMMapping

[ScheiklP]

Hi all,
HexahedronCompositeFEMMapping prints a warning, that the node is using an incorrect solver, even with a direct solver.

[WARNING] [HexahedronCompositeFEMMapping(HexahedronCompositeFEMMapping1)] This object only support Direct Solving but an Indirect Solver in the scene is c
 not implemented. This will produce un-expected behavior.

In the component example, no matter if you use the direct or indirect solver, the same warning is printed.

https://github.com/sofa-framework/sofa/blob/master/examples/Component/SolidMechanics/FEM/HexahedronCompositeFEMForceFieldAndMass.scn#L38

The warning comes from here, but I do not know enough about the solvers to understand the two variants of applyJT.

https://github.com/sofa-framework/sofa/blob/master/Sofa/Component/SolidMechanics/FEM/NonUniform/src/sofa/component/solidmechanics/fem/nonuniform/HexahedronCompositeFEMMapping.h#L103-L106

   void applyJT( const sofa::core::ConstraintParams* /*cparams*/, InDataMatrixDeriv& /*out*/, const OutDataMatrixDeriv& /*in*/) override
   {
       msg_warning() << "This object only support Direct Solving but an Indirect Solver in the scene is calling method applyJT(constraint) which is not implemented. This will produce un-expected behavior.";
   }

Best,
Paul

[alxbilger]

I am not sure the message is appropriate. I agree with you that it is not related to the solver. This method is used to project the constraint matrix from a subNode to its parent. If you have constraints in a subNode, it won't work.

The code of this mapping has poor quality. I refresh would be beneficial. But I don't know what it does...

[bakpaul]

Given the author of the component I would guess that it works... This line has certainly be added through a big refactoring, it is indeed misleading.

From what I understood the mapping is kind of the equivalent to a barycentric mapping but for Hexahedron. I know that it is more difficult to compute them because it is an under constrained problem, there exists an infinit number of way to express a coordinate in term of the 8 points... This does it properly.

Anyway it is quite old, and has a lot of commented code, so while I believe the guy who wrote it knew what he was doing, I cannot say the same about all the cleaning/refactoring done throughout the 16 year after that... To be used with caution.