Hey,

I am faced with the following problem:
I have to assemble one gigantic affine linear transform and have to do in the middle a non standard operation.
At this point this code works, but is very non-Eigen eye candy:

Matrix4f mE;
Affine3f t;
t = Affine3f(Translation3f(pos[ 0 ], pos[ 1 ], pos[ 2 ]) *
Translation3f(cen[ 0 ], cen[ 1 ], cen[ 2 ]) *
Translation3f(centerX, centerY, 0.0f) *
AngleAxisf(rot[0], Vector3f::UnitX()) *
AngleAxisf(rot[1], Vector3f::UnitY()) *
AngleAxisf(rot[2], Vector3f::UnitZ()) *
Scaling(scl[0], scl[1], scl[2])).matrix() *
mE *
Affine3f( Translation3f(-centerX, -centerY, 0) *
Translation3f(-cen[0], -cen[1], -cen[2]) ).matrix();

Is there a way to avoid the parenthesization and conversion into affine3f and matrices?

Using Affine3f(mE) is simple and probably cleaner. Another solution is do something like:


t = .... * (mE * ....);

and declare mE as a 3x4 matrix such that (mE * ....) returns an Affine3f and not a Projective3f. You can also use mE.topRows<3>(). The parenthesis is important, because here (.... * mE) would mean applying the transformation to the set of homogeneous vectors stored in mE and so you would get a Matrix4f.

Thanks for the reply.
I did wind up using Affine3f(mE) in the expression (more like mE is directly an affine)