C++ Eigen: recursive functions accepting any matrix class

Question

I want to have a recursive function

template <typename Derived>
void f(Eigen::MatrixBase<Derived>& m)
{
    size_t blockRows = ...
    size_t blockCols = ...
    ....
    f(m.block(0, 0, blockRows, blockCols));
}

This unfortunately results in an infinite compile time recursion.

The first call would be to

f<Eigen::MatrixBase<Derived> >

The second would be to

f<Eigen::Block<Eigen::MatrixBase<Derived>, ... > >

The third call would be to

f<Eigen::Block<Eigen::Block<Eigen::MatrixBase<Derived>, ... >, ... > >

Every time a block of block is requested.

What is the best practice to implement recursive functions in Eigen, which still work on any Eigen matrix type?

I think, I should use some type, that still wraps the same piece of memory, but is not an expression template and it is evaluated.

Answer 1

You can cast your recursive block to an Eigen::Ref to avoid the infinite type instantiation:

Ref<MatrixXd> bl = m.block(0, 0, blockRows, blockCols);
f(bl);

To stay fully generic, you can replace MatrixXd by typename Derived::PlainObject.