Question

I want to represent a 2D shape in such a way that it can be interacted with as if it were a vector of points, in particular I want to be able to call operator[] and at() on it and return references to things that act like 2D points. Currently I just use a class whose only member variable is a vector of points and that has various arithmetic and geometric operations defined pointwise on its elements.

However, in other parts of my code I need to treat a vector of n points as an element of 2n dimensional space and perform basic linear algebra on it (e.g. projecting the vector onto a given subspace of R^2n). Currently I’m creating an Eigen::VectorXd object every time I want to do this, and then converting back after performing these operations. I don’t want to do this, as I make the conversion often enough that all the copying is a noticeable source of inefficiency.

If I was storing the data as a flat array of doubles/floats/ints, I could cast a pointer to its nth element to a pointer to a Point (whose members would just be a pair of doubles/floats/ints). However, as I don’t know the internal representation that Eigen uses for vectors (and it may well change), this isn’t possible.

Is there a sensible way of solving this? I could just use Eigen::Vectors everywhere, but I really want most of the code to be able to pretend that it is dealing with a set of points.

Answer 1

However, as I don’t know the internal representation that Eigen uses for vectors (and it may well change), this isn’t possible.

Eigen offers the Map classes that allow mapping plain arrays to Eigen structures. For example:

double numbers[2];
Eigen::Vector2f::Map( numbers ).dot( Eigen::Vector2f::Constant(1) );