coordinate_matrix sort() fails on to std::swap

[anonymous]

Orginally posted here:

​http://stackoverflow.com/questions/24228772/c11-compatibility-of-sparse-matrix-implementations

The following program does not compile with gcc 4.8 or clang 3.4 when --std=c++11 is set:

#include <boost/numeric/ublas/io.hpp>
#include <boost/numeric/ublas/matrix_sparse.hpp>

using namespace boost::numeric::ublas;

int main(int argc, char** argv) {
  coordinate_matrix<int> m1(100, 100, 100);

  for (int i = 0; i < 100; i++)
    m1.insert_element(i,i,i);

  compressed_matrix<int> m2(m1, 100);
}

We can solve the issue by providing a swap() routine:

#include <boost/numeric/ublas/io.hpp>
#include <boost/numeric/ublas/matrix_sparse.hpp>

using namespace boost::numeric::ublas;

namespace std {
    template<class M>
    inline
    void swap (boost::numeric::ublas::index_triple<M> i1, boost::numeric::ublas::index_triple<M> i2) {
        i1.swap (i2);
    }
}


int main(int argc, char** argv) {
  coordinate_matrix<int> m1(100, 100, 100);

  for (int i = 0; i < 100; i++)
    m1.insert_element(i,i,i);
  
  compressed_matrix<int> m2(m1, 100);
}

I am not an expert C++ template programmer, so I am unable to decide what is the cause here, but the C++ reference on std::sort explicitly mentions a swap() method.

[Casey Carter <Casey@…>]

Test case can be minimized to:

#include <boost/numeric/ublas/matrix_sparse.hpp>

int main() {
  boost::numeric::ublas::coordinate_matrix<int> m1(1, 1, 1);
  m1.insert_element(0, 0, 0);
}

The problem seems to be that when BOOST_UBLAS_STRICT_MATRIX_SPARSE is defined, coordinate_matrix and its iterator types use a proxy class as their reference type:

#ifndef BOOST_UBLAS_STRICT_MATRIX_SPARSE
        typedef T &reference;
#else
        typedef sparse_matrix_element<self_type> reference;
#endif

which violates the C++ standard's requirement that the reference type for forward iterators is a true reference. C++11 [forward.iterators]/1 states: "A class or pointer type X satisfies the requirements of a forward iterator if ... if X is a mutable iterator, reference is a reference to T; if X is a const iterator, reference is a reference to const T, ...".

[Andrew Medlin <amedlin@…>]

This also affects version 1.55.0 sparse vectors, which breaks the LU substitution.

For example, the following code produces incorrect output. It should swap the 2nd and 3rd elements, producing

1.000000, 3.000000, 2.000000

but instead it produces

1.000000, 3.000000, 3.000000

It is fixed by uncommenting the #define BOOST_UBLAS_NO_ELEMENT_PROXIES

//#define BOOST_UBLAS_NO_ELEMENT_PROXIES

#include <boost/numeric/ublas/vector_sparse.hpp>
#include <boost/numeric/ublas/lu.hpp>

using namespace boost::numeric::ublas;

int main()
{
    mapped_vector<float> y;
    y(0) = 1.0f;
    y(1) = 2.0f;
    y(2) = 3.0f;
    LOG("Initial vector is: " << y);
    permutation_matrix<std::size_t> perm(y.size());
    perm(0) = 0;
    perm(1) = 2;
    perm(2) = 2;
    swap_rows(perm, y);
    LOG("Swapped vector is: " << y);
}

[eg@…]

Hi, any chance to get this ticket solved in the next release ? version 1.57 doesn't seem to offer a fix for this issue, as far as I know.

[Andrew Medlin <amedlin@…>]

I think this bug represents a fundamental break in core functionality, so I am raising the severity. I believe this should be fixed with priority.