Missing include: boost/qvm/detail/mat_assign.hpp not included in boost/qvm/detail/transp_impl.hpp

[KABoissonneault]

If boost::qvm::transposed is used on its own, we get a compilation error.

#include <boost/qvm/mat_traits.hpp>
#include <boost/qvm/mat_traits_defaults.hpp>

#include <boost/qvm/map_mat_mat.hpp>

struct float33
{
    float elements[9];
};

template<>
struct boost::qvm::is_mat<float33>
{
    static bool const value = true;
};

template<>
struct boost::qvm::mat_traits<float33> : boost::qvm::mat_traits_defaults<float33, float, 3, 3>
{
    template<int R, int C>
    static float& write_element(float33& m)
    {
        return m.elements[R*3+C];
    }
};

float33 transpose(float33 const& m)
{
    return boost::qvm::transposed(m);
}

The result is

In file included from /opt/compiler-explorer/libs/boost_1_74_0/boost/qvm/map_mat_mat.hpp:13,

                 from <source>:4:

/opt/compiler-explorer/libs/boost_1_74_0/boost/qvm/detail/transp_impl.hpp: In instantiation of 'boost::qvm::qvm_detail::transposed_<OriginalMatrix>::operator R() const [with R = float33; OriginalMatrix = float33]':

<source>:29:36:   required from here

/opt/compiler-explorer/libs/boost_1_74_0/boost/qvm/detail/transp_impl.hpp:46:27: error: 'assign' was not declared in this scope

   46 |                     assign(r,*this);

      |                     ~~~~~~^~~~~~~~~

(URL for convenience: https://gcc.godbolt.org/z/7918sb)

boost/qvm/detail/transp_impl.hpp should probably include boost/qvm/detail/mat_assign.hpp, the same way mat_operations.hpp does for identity_mat_

[zajo]

All QVM headers (that aren't in the detail directory) are designed to not cause compilation errors if included on their own (e.g. if they're the first #include in a compilation unit). This does not mean that you can instantiate every single template they define; this may require including additional headers, as is the case here. Consider that the call to assign is not qualified, which means that depending on the types used, a different overload may be needed -- and the correct overload may require including a user header.

In this case:

• If you want to minimize your header dependencies, you should #include <boost/qvm/mat_operations3.hpp> (which won't introduce 2D or 4D overloads): https://gcc.godbolt.org/z/8Ed9j3.

• If you don't care about header dependencies, you can just #include <boost/qvm/all.hpp>.