Qt3DCore::QTransform should not use sub transforms to build the matrix passed to the backend. The problem is that with this approach if a later backend (e.g. animation or physics) sends a transform update to the frontend, there is no single well defined way to decompose this into an arbitrary set of input transformations.

It will be better to have the QTransform have properties for translation (QVector3D) , orientaiton (QQuaternion) and scale (QVector3D). Any affine transformation matrix the frontend receives can then be decomposed back into these 3 properties.