[boost] [ublas] equals(matrix<int>, matrix<int>) causes abort

From: David Cameron (dave.bc_at_[hidden])
Date: 2005-11-02 03:04:04

I've been writing some code to serialize and deserialize
boost::numeric::ublas::matrixs. While writing unit tests for it I
found out about the boost::numeric::ublas::equals function in
matrix_assign.hpp. I was relieved, because I had previously been using
a templated function to compare matrices.

Unfortunately, it doesn't work for me with matrix<int>s. Consider the
following cppunit test code:

  void testMatrixEqualsOperator() {
     using namespace boost::numeric::ublas;
     matrix<double> test;
     test.resize(3,3);
     CPPUNIT_ASSERT( equals(test,test) );

     matrix<int> test1;
     test.resize(4,4);
     CPPUNIT_ASSERT( equals(test1,test1) );
  }

The first four lines work. Adding the next three lines causes the unit
tests to fail with an Abort. This is not the standard way CppUnit
records a failed test, it must be something failing inside the call to
equals().

Previous messages referring to the equals() function argued that it
wasn't ready for use because there was rarely a need to compare
matrices in real applications. I would argue that it should be working
as intended, if only to allow for proper unit testing of code that
uses matrices.

Here is the function I have been using instead. It works well for me
with doubles and ints. Although, I imagine it is not particularly
fast. E is for Element.

  template<class E>
  bool areMatricesEqual(
    boost::numeric::ublas::matrix<E> lhs,
    boost::numeric::ublas::matrix<E> rhs)
  {
    using boost::numeric::ublas;

    if (&lhs == &rhs)
        return true;
    if (lhs.size1() != rhs.size1())
        return false;
    if (lhs.size2() != rhs.size2())
        return false;
    typename matrix<E>::iterator1 l(lhs.begin1());
    typename matrix<E>::iterator1 r(rhs.begin1());
    while (l != lhs.end1()) {
        if (*l != *r)
            return false;
        ++l;
        ++r;
    }
    return true;
  }

I am using gcc-3.3 on ubuntu-5.04 running on an AthlonXP. I'd be happy
to provide any more info to help track down the bug. Assuming it is
one.

--
David Cameron
University of British Columbia
http://bar.psych.ubc.ca/People/Dave.html