We need to be able to set user transform/matrix directly. Requiring setting the orientation using Euler angles would be very complicated and fragile when you already have the position+orientation+scaling as a matrix.
I asked for the exact requirements on discourse and submitted the issue in gitlab to sort out this issue proactively, instead of having to push back on a pull request that breaks/removes the essential direct matrix setting API.
@ken-martin could you describe what are the user transform/matrix requirements that must be fulfilled for correct behavior?
For example we could require the user matrix to be a homogeneous transformation matrix composed of rigid transformation + scaling with positive factors, resulting in a:
- 4x4 matrix with bottom row equal to (0, 0, 0, 1)
- topeft 3x3 submatrix is orthogonal and has positive determinant
We may also define some soft requirements for accuracy and numerical stability. For example, maximum ratio between smallest and largest absolute value of scaling factor is recommended to be below 1e3 (or 1e6 or 1e12?). It might be also useful to specify a tolerance for the orthogonality of the orientation submatrix and for the (0,0,0,1) values in the last row.
Fully agree. We have been always using SetUserMatrix in 3D Slicer. Require using Euler angles for setting the actor orientation would be completely unacceptable. Setting the orientation using a quaternion would be sufficient and make sense in theory, but it would be impractical.