public static void ConstructTransformFromMatrixWithDesiredScale(FMatrix aMatrix, FMatrix bMatrix,
FVector desiredScale, ref FTransform outTransform)
{
// the goal of using M is to get the correct orientation
// but for translation, we still need scale
var m = aMatrix * bMatrix;
m.RemoveScaling();
// apply negative scale back to axes
var signedScale = desiredScale.GetSignVector();
m.SetAxis(0, m.GetScaledAxis(EAxis.X) * signedScale.X);
m.SetAxis(1, m.GetScaledAxis(EAxis.Y) * signedScale.Y);
m.SetAxis(2, m.GetScaledAxis(EAxis.Z) * signedScale.Z);
// @note: if you have negative with 0 scale, this will return rotation that is identity
// since matrix loses that axes
var rotation = new FQuat(m);
rotation.Normalize();
// set values back to output
outTransform.Scale3D = desiredScale;
outTransform.Rotation = rotation;
// technically I could calculate this using FTransform but then it does more quat multiplication
// instead of using Scale in matrix multiplication
// it's a question of between RemoveScaling vs using FTransform to move translation
outTransform.Translation = m.GetOrigin();
}
