Compilation error with g++

[thk686]

I am getting:

/usr/local/lib/R/site-library/BH/include/boost/mpl/assert.hpp:429:42: error: no 
matching function for call to ‘assertion_failed(mpl_::failed************ (boost:
:geometry::index::detail::indexable<std::pair<Eigen::Matrix<double, 1, 1, 0, 1, 
1>, double>, false>::NOT_VALID_INDEXABLE_TYPE::************)(Eigen::Matrix<doubl
e, 1, 1, 0, 1, 1>))’
     , BOOST_PP_CAT(mpl_assertion_in_line_,counter) = sizeof( \
                                                            ~~~
         boost::mpl::assertion_failed<(c)>( BOOST_PP_CAT(mpl_assert_arg,counter)
::assert_arg() ) \
         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~
         ) \
         ~                                 

with g++ (6.3). This compiles fine with Clang++ (8.1). I am registering a matrix from the Eigen library as the coordinate type. The header look like:

// Boost.Geometry (aka GGL, Generic Geometry Library)

// Copyright (c) 2010 Alfredo Correa
// Copyright (c) 2010-2012 Barend Gehrels, Amsterdam, the Netherlands.
// Copyright (c) 2016 Norbert Wenzel

// Use, modification and distribution is subject to the Boost Software License,
// Version 1.0. (See accompanying file LICENSE_1_0.txt or copy at
// http://www.boost.org/LICENSE_1_0.txt)

// Modified from the orignal code by Tim Keitt 4/7/17
// Changes Copyright (c) 2017 Tim Keitt

#ifndef BOOST_GEOMETRY_GEOMETRIES_ADAPTED_EIGEN_MATRIX_HPP
#define BOOST_GEOMETRY_GEOMETRIES_ADAPTED_EIGEN_MATRIX_HPP

#define BOOST_GEOMETRY_ADAPTED_EIGEN_MATRIX_TAG_DEFINED

#include <cstddef>

#include <boost/type_traits/is_arithmetic.hpp>

#include <boost/geometry/core/access.hpp>
#include <boost/geometry/core/cs.hpp>
#include <boost/geometry/core/coordinate_dimension.hpp>
#include <boost/geometry/core/coordinate_type.hpp>
#include <boost/geometry/core/tags.hpp>

#include <Eigen/Dense>

namespace boost { namespace geometry
{


#ifndef DOXYGEN_NO_TRAITS_SPECIALIZATIONS
namespace traits
{


#ifndef DOXYGEN_NO_DETAIL
namespace detail
{


// Create class and specialization to indicate the tag
// for normal cases and the case that the type of the std-array is arithmetic
template <bool>
struct eigen_matrix_tag
{
  typedef geometry_not_recognized_tag type;
};


template <>
struct eigen_matrix_tag<true>
{
  typedef point_tag type;
};


} // namespace detail
#endif // DOXYGEN_NO_DETAIL


// Assign the point-tag, preventing arrays of points getting a point-tag
template <typename CoordinateType, std::size_t DimensionCount>
struct tag<Eigen::Matrix<CoordinateType, DimensionCount, 1>>
  : detail::eigen_matrix_tag<boost::is_arithmetic<CoordinateType>::value> {};


template <typename CoordinateType, std::size_t DimensionCount>
struct coordinate_type<Eigen::Matrix<CoordinateType, DimensionCount, 1>>
{
  typedef CoordinateType type;
};


template <typename CoordinateType, std::size_t DimensionCount>
struct dimension<Eigen::Matrix<CoordinateType, DimensionCount, 1>>: boost::mpl::int_<DimensionCount> {};


template <typename CoordinateType, std::size_t DimensionCount, std::size_t Dimension>
struct access<Eigen::Matrix<CoordinateType, DimensionCount, 1>, Dimension>
{
  static inline CoordinateType get(Eigen::Matrix<CoordinateType, DimensionCount, 1> const& a)
  {
    return a(Dimension);
  }

  static inline void set(Eigen::Matrix<CoordinateType, DimensionCount, 1>& a,
                         CoordinateType const& value)
  {
    a(Dimension) = value;
  }
};


} // namespace traits
#endif // DOXYGEN_NO_TRAITS_SPECIALIZATIONS


}} // namespace boost::geometry


#define BOOST_GEOMETRY_REGISTER_EIGEN_MATRIX_CS(CoordinateSystem) \
namespace boost { namespace geometry { namespace traits {      \
template <class T, std::size_t N>                              \
struct coordinate_system<Eigen::Matrix<T, N, 1>>               \
{                                                              \
  typedef CoordinateSystem type;                               \
};                                                             \
}}}


#endif // BOOST_GEOMETRY_GEOMETRIES_ADAPTED_EIGEN_MATRIX_HPP

The BH R package uses Boost 1.62.

THK

[awulkiew]

Ok, so you adapt Eigen::Matrix to Boost.Geometry Point concept. The code above looks ok.

What's the program you're compiling? I see that you're creating R-tree and passing std::pair<Eigen::Matrix<...>, double> as ValueType. Is that correct? Or are you calling some algorithm?

The NOT_VALID_INDEXABLE_TYPE MPL assertion occurs when tag<ValueType>::type is neither point_tag, box_tag nor segment_tag. Could you check in your code if the tag is set properly? E.g.: std::cout << boost::is_same<bg::tag<Eigen::Matrix<...>>::type, point_tag>::value;.

How are you including Boost.Geometry in your code and what is the order of includes? It's possible that the order of #includes is important for GCC. This wouldn't be the first time something like this happen with this compiler.

[thk686]

// [[Rcpp::export]]
void test_tag()
{
  Rcout << boost::is_same<bg::tag<Eigen::Matrix<double, 2, 1>>::type, bg::point_tag>::value << std::endl;
}

Returns 1.

[thk686]

Tried moving header to first and last in source file without luck.

[awulkiew]

This may be a problem with wrong traits specializations. If Eigen::Matrix rows, cols, options etc. are int, not std::size_t then DimensionCount in specializations should be of type int too.

Btw, I see that Eigen::Matrix is defined as:

Matrix<typename Scalar,
   int RowsAtCompileTime,
   int ColsAtCompileTime,
   int Options = 0,
   int MaxRowsAtCompileTime = RowsAtCompileTime,
   int MaxColsAtCompileTime = ColsAtCompileTime>

Maybe this particular compiler would prefer to have all 6 template parameters in specializations? You pass only 3.

Ok, so bg::tag seems to work. How about some simple algorithm? Does distance(p1, p2) or disjoint(p1, p2) compile and generate correct result?

[thk686]

It looks like changing the Eigen template types from size_t to int fixes the problem. THK

…

On Tue, Dec 5, 2017 at 7:09 AM, Adam Wulkiewicz ***@***.***> wrote: This may be a problem with wrong traits specializations. If Eigen::Matrix rows, cols, options etc. are int, not std::size_t then DimensionCount in specializations should be of type int too. Btw, I see that Eigen::Matrix is defined as: Matrix<typename Scalar, int RowsAtCompileTime, int ColsAtCompileTime, int Options = 0, int MaxRowsAtCompileTime = RowsAtCompileTime, int MaxColsAtCompileTime = ColsAtCompileTime> Maybe this particular compiler would prefer to have all 6 template parameters in specializations? You pass only 3. Ok, so bg::tag seems to work. How about some simple algorithm? Does distance(p1, p2) or disjoint(p1, p2) compile and generate correct result? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#416 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAkLyXcS9j9GpitNVupwsbbOlMQqH_PAks5s9UBsgaJpZM4O67mU> .

[awulkiew]

Ok, great.