Hi Eigen users,

I am starting to get familiar with Eigen and I now use it in my research for many calculations. It is an awesome tool. Many thanks to all the developers!

I am starting to write template classes (simple functors) that work with Eigen types. By using MatrixBase<Derived> as template parameters, as suggested by the documentation, I was able to successfully specialise the class for plain matrices and the Block type, but I had no luck with the VectorBlock type. I cannot seem to find the right expression! I have no problem with template functions that are implicitly specialised.

Here is a trivial example illustrating this:

Code: Select all

#include <Eigen/Dense>
#include <iostream>

using namespace Eigen;
using namespace std;

template <typename Derived>
typename MatrixBase<Derived>::Scalar f(MatrixBase<Derived> const& v)
{
   return v.sum();
}

template <class Derived>
class g_func
{
private:
   MatrixBase<Derived> const& m_v;
   
public:
   typedef typename MatrixBase<Derived>::Scalar Scalar;
   
   g_func(MatrixBase<Derived> const& v): m_v(v) { }
   
   Scalar operator()(int x)
   {
      return m_v.head(x).sum();
   }
};

int main(int argc, char* argv[])
{
   VectorXd v = VectorXd::Random(10);
   // Specialising template functions is no problem
   cout << f(v) << endl;
   cout << f(v.block(0, 0, 5, 1)) << endl;
   cout << f(v.segment(0, 5)) << endl;

   // Specialising template classes works with VectorXd and
   // Block<VectorXd> but not VectorBlock<VectorXd>
   g_func<VectorXd> g1(v);
   cout << g1(2) << endl;
   g_func< Block<VectorXd> > g2(v.block(0, 0, 5, 1));
   cout << g2(2) << endl;
   g_func< VectorBlock<VectorXd> > g3(v.segment(0, 5)); // BUG!
   cout << g3(0) << endl;
}


I use Eigen 3.0 beta 2 and GCC 4.4.4 on Ubuntu. (I also tried on other Linux variants and with other GCC versions.)

Thanks in advance to anyone who can help!

Olivier

the VectorBlock<> template class is just a convenient class to declare and construct vector block expressions, but it is essentially a Block<>, so in your case you should do:

g_func< Block<VectorXd> > g3(v.segment(0, 5));

Thanks for the suggestion. I tried it before, but it does not work. I get:

Code: Select all

eigen-test.cpp:41: error: no matching function for call to ‘g_func<Eigen::Block<Eigen::Matrix<double, -0x00000000000000001, 1, 0, -0x00000000000000001, 1>, -0x00000000000000001, -0x00000000000000001, false, true> >::g_func(Eigen::VectorBlock<Eigen::Matrix<double, -0x00000000000000001, 1, 0, -0x00000000000000001, 1>, -0x00000000000000001>)’
eigen-test.cpp:22: note: candidates are: g_func<Derived>::g_func(const Eigen::MatrixBase<Derived>&) [with Derived = Eigen::Block<Eigen::Matrix<double, -0x00000000000000001, 1, 0, -0x00000000000000001, 1>, -0x00000000000000001, -0x00000000000000001, false, true>]
eigen-test.cpp:15: note:                 g_func<Eigen::Block<Eigen::Matrix<double, -0x00000000000000001, 1, 0, -0x00000000000000001, 1>, -0x00000000000000001, -0x00000000000000001, false, true> >::g_func(const g_func<Eigen::Block<Eigen::Matrix<double, -0x00000000000000001, 1, 0, -0x00000000000000001, 1>, -0x00000000000000001, -0x00000000000000001, false, true> >&)


oops sorry, it is:

g_func< Block<VectorXd,Dynamic,1> > g3(v.segment(0, 5));

Another solution is to take:

g_func< VectorBlock<VectorXd>::Derived > g3(v.segment(0, 5));

I agree that's not obvious.

Thank you!

I will make note of all my related beginner difficulties and then write a patch for the documentation.

I would note that interfacing directly with our expression templates is NOT a "beginner" issue!