[eigen] Regression in LDLT

[Keir Mierle <mierle@xxxxxxxxx>]

It seems there is a regression in Eigen 2.0.15 and 2.0.16. The following code compiles and works fine with 2.0.12:

#include <Eigen/Core>

#include <Eigen/Array>

#include <Eigen/LU>

#include <Eigen/Cholesky>

#include <iostream>

using namespace Eigen;

typedef Matrix<double, Dynamic, Dynamic, RowMajor> Mat;

typedef Matrix<double, Dynamic, 1> Vec;

int main() {

  Mat A(5, 5);

  A..setRandom();

  double bs[] = {1, 2, 3, 4, 5};

  double xs[5];

  Map<Vec> b(bs, 5);

  Map<Vec> x(xs, 5);

  A.part<UpperTriangular>().ldlt().solve(b, &x);

  A.lu().solve(b, &x);

  std::cout << "solution:\n" << x << "\n";

}

However, with 2.0.15 or 2.0.16 I get the following error:

In file included from /usr/include/eigen2/Eigen/Cholesky:36,

                 from lu_regression.cc:7:

/usr/include/eigen2/Eigen/src/Cholesky/LDLT.h: In member function ‘const Eigen::LDLT<typename Eigen::MatrixBase<Derived>::PlainMatrixType> Eigen::MatrixBase<Derived>::ldlt() const [with Derived = Eigen::Part<Eigen::Matrix<double, 10000, 10000, 1, 10000, 10000>, 1024u>]’:

lu_regression.cc:25:   instantiated from here

/usr/include/eigen2/Eigen/src/Cholesky/LDLT.h:195: error: conversion from ‘const Eigen::Part<Eigen::Matrix<double, 10000, 10000, 1, 10000, 10000>, 1024u>’ to non-scalar type ‘const Eigen::LDLT<Eigen::Matrix<double, 10000, 10000, 3, 10000, 10000> >’ requested

Any idea what's going on here?

Thanks,

Keir