Hi,

After following the tutorial code I tried the following example out.

Code: Select all

    static const int num_variables = 10;
    static const int num_constraints = 10;
    SparseMatrix<float> J;
    Matrix<float, num_constraints, 1> b;
    J.resize(num_constraints, num_variables);
    for (int r = 0; r < num_constraints; r++)
    {
        for (int c = 0; c < num_variables; c++)
        {
            J.insert(r, c) = Matrix<float, 1, 1>::Random()(0, 0);
        }
        b(r) = Matrix<float, 1, 1>::Random()(0, 0);
    }

    Eigen::SparseQR<SparseMatrix<float>, COLAMDOrdering<int> > solver;
    J.makeCompressed();
    solver.analyzePattern(J);
    solver.factorize(J);
    Matrix<float, Dynamic, 1> e = solver.solve(b);
    //
    // here we would be updating values in the J (with coeffRef) and b
    //
    //solver.analyzePattern(J); // this fixes the problem
    solver.factorize(J); // CRASH - here!!!
    e = solver.solve(b);


Basically it won't work as specified in the example when I try and run factorize without analyzePattern first. It is obviously not preferable to run COLAMD if the structure of the matrix has not changed.

Cheers,
David...

You're right, there was a bug. Fixed in default and 3.2 branch:
https://bitbucket.org/eigen/eigen/commits/7b6329e8b626/
https://bitbucket.org/eigen/eigen/commits/a1f5fed0d734/

Hi Gael,

Great - thanks for the fix, I will try it out now.

Cheers,
David...