Clang-40 compillation errors

[yurivict]

In file included from /usr/local/include/boost/gil/gil_all.hpp:26:
/usr/local/include/boost/gil/channel_algorithm.hpp:54:85: error: non-type template argument evaluates to -1, which cannot be narrowed to type 'unsigned long' [-Wc++11-narrowing]
struct unsigned_integral_max_value : public mpl::integral_c<UnsignedIntegralChannel,-1> {};
                                                                                    ^
/usr/local/include/boost/gil/channel_algorithm.hpp:204:19: note: in instantiation of template class 'boost::gil::detail::unsigned_integral_max_value<unsigned long>' requested here
        if (src > unsigned_integral_max_value<uintmax_t>::value - div2)
                  ^
In file included from /usr/ports/graphics/k3d/work/k3d-k3d-0.8.0.6/k3dsdk/utility_gl.cpp:37:
/usr/ports/graphics/k3d/work/k3d-k3d-0.8.0.6/k3dsdk/utility_gl.h:107:22: error: non-constant-expression cannot be narrowed from type 'k3d::basic_rgb<double, k3d::color_traits<k3d::double_t> >::sample_type' (aka 'double') to 'GLfloat' (aka 'float') in initializer list [-Wc++11-narrowing]
        GLfloat color[] = { Color.red, Color.green, Color.blue, Alpha };
                            ^~~~~~~~~
/usr/ports/graphics/k3d/work/k3d-k3d-0.8.0.6/k3dsdk/utility_gl.h:107:22: note: insert an explicit cast to silence this issue
        GLfloat color[] = { Color.red, Color.green, Color.blue, Alpha };
                            ^~~~~~~~~
                            static_cast<GLfloat>( )
/usr/ports/graphics/k3d/work/k3d-k3d-0.8.0.6/k3dsdk/utility_gl.h:107:33: error: non-constant-expression cannot be narrowed from type 'k3d::basic_rgb<double, k3d::color_traits<k3d::double_t> >::sample_type' (aka 'double') to 'GLfloat' (aka 'float') in initializer list [-Wc++11-narrowing]
        GLfloat color[] = { Color.red, Color.green, Color.blue, Alpha };
                                       ^~~~~~~~~~~
/usr/ports/graphics/k3d/work/k3d-k3d-0.8.0.6/k3dsdk/utility_gl.h:107:33: note: insert an explicit cast to silence this issue
        GLfloat color[] = { Color.red, Color.green, Color.blue, Alpha };
                                       ^~~~~~~~~~~
                                       static_cast<GLfloat>( )
/usr/ports/graphics/k3d/work/k3d-k3d-0.8.0.6/k3dsdk/utility_gl.h:107:46: error: non-constant-expression cannot be narrowed from type 'k3d::basic_rgb<double, k3d::color_traits<k3d::double_t> >::sample_type' (aka 'double') to 'GLfloat' (aka 'float') in initializer list [-Wc++11-narrowing]
        GLfloat color[] = { Color.red, Color.green, Color.blue, Alpha };
                                                    ^~~~~~~~~~
/usr/ports/graphics/k3d/work/k3d-k3d-0.8.0.6/k3dsdk/utility_gl.h:107:46: note: insert an explicit cast to silence this issue
        GLfloat color[] = { Color.red, Color.green, Color.blue, Alpha };
                                                    ^~~~~~~~~~
                                                    static_cast<GLfloat>( )
/usr/ports/graphics/k3d/work/k3d-k3d-0.8.0.6/k3dsdk/utility_gl.h:107:58: error: non-constant-expression cannot be narrowed from type 'double' to 'GLfloat' (aka 'float') in initializer list [-Wc++11-narrowing]
        GLfloat color[] = { Color.red, Color.green, Color.blue, Alpha };
                                                                ^~~~~
/usr/ports/graphics/k3d/work/k3d-k3d-0.8.0.6/k3dsdk/utility_gl.h:107:58: note: insert an explicit cast to silence this issue
        GLfloat color[] = { Color.red, Color.green, Color.blue, Alpha };
                                                                ^~~~~
                                                                static_cast<GLfloat>( )

[yurivict]

@barche Jan, are you able to fix these? The FreeBSD port for k3d has been deleted some while ago. I am trying to restore it now. But it doesn't build with clang-40. Are you able to fix all clang-40 build errors?

Thanks, Yuri

[barche]

Hi Yuri,

Sorry for the slow reply, I haven't build K-3D using clang for some years. I don't have a freeBSD system to test this on, but I could try building on my Mac and see if I can reproduce this. I think in this case adding explicit casts is the solution, e.g.

GLfloat color[] = { static_cast<float>(Color.red), static_cast<float>(Color.green), static_cast<float>(Color.blue), static_cast<float>(Alpha) };

[yurivict]

Yes, this would have been great if you could fix clang issues on any system. The rest of the differences if any shouldn't be too hard to accommodate.

[barche]

OK, so it builds using clang 5 on Linux now, hopefully this also gets you further on FreeBSD!

[yurivict]

It still breaks for me here:

In file included from /usr/local/include/boost/gil/gil_all.hpp:26:
/usr/local/include/boost/gil/channel_algorithm.hpp:54:85: error: non-type template argument evaluates to -1, which cannot be narrowed to type 'unsigned long' [-Wc++11-narrowing]
struct unsigned_integral_max_value : public mpl::integral_c<UnsignedIntegralChannel,-1> {};
                                                                                    ^
/usr/local/include/boost/gil/channel_algorithm.hpp:204:19: note: in instantiation of template class 'boost::gil::detail::unsigned_integral_max_value<unsigned long>' requested here
        if (src > unsigned_integral_max_value<uintmax_t>::value - div2)
                  ^

Also, you should remove all register keywords. Latest developments in compiles rendered it useless, and it only causes warnings now.

clang-5.0.0

[barche]

Is that the complete error? It is situated completely in boost, on my system it selects the gil library included in K-3D and then it compiles.

[yurivict]

This line https://github.com/K-3D/k3d/blob/master/k3dsdk/bitmap.h#L36 includes gil_all.hpp that isn't in the project.

It can't use the include from the pre-installed boost, but override only the gil package. This is just too convoluted. :-)

[barche]

Actually, what is added is an extension to gil, so files from the system boost are still needed. I can't reproduce this error on my linux system, can you please post the full error, including the location in the K-3D code that triggers this? If there is no instantiation in K-3D, I guess this is really a bug in boost, I'm using version 1.66 but the referred lines are the same.

[yurivict]

Here is one full instance of this error:

--- k3dsdk/CMakeFiles/k3dsdk.dir/utility_gl.cpp.o ---
In file included from /usr/ports/graphics/k3d/work/k3d-7111d0e/k3dsdk/utility_gl.cpp:37:
In file included from /usr/ports/graphics/k3d/work/k3d-7111d0e/k3dsdk/utility_gl.h:28:
In file included from /usr/ports/graphics/k3d/work/k3d-7111d0e/k3dsdk/bitmap.h:36:
In file included from /usr/local/include/boost/gil/gil_all.hpp:26:
/usr/local/include/boost/gil/channel_algorithm.hpp:54:85: error: non-type template argument evaluates to -1, which cannot be narrowed to type 'unsigned long' [-Wc++11-narrowing]
struct unsigned_integral_max_value : public mpl::integral_c<UnsignedIntegralChannel,-1> {};
                                                                                    ^
/usr/local/include/boost/gil/channel_algorithm.hpp:204:19: note: in instantiation of template class 'boost::gil::detail::unsigned_integral_max_value<unsigned long>' requested here
        if (src > unsigned_integral_max_value<uintmax_t>::value - div2)
                  ^

[barche]

OK, so it appears there is no instantiation happening from K-3D. Can you try compiling and posting the output of the following test program:

#include <iostream>
#include <typeinfo>
#include <boost/gil/gil_all.hpp>
int main()
{
  std::cout << typeid(boost::uintmax_t).name() << std::endl;
  std::cout << typeid(uint64_t).name() << std::endl;
  return 0;
}

I compiled it with

clang++ -o test test.cpp

[yurivict]

$ ./test 
m
m

[barche]

Can you try adding the following in bitmap.h, just before the existing channel_converter_unsigned:

template <> struct channel_converter_unsigned<bits32f,uintmax_t> : public std::unary_function<bits32f,bits32f>
{
    uintmax_t operator()(bits32f   x) const { return uintmax_t(static_cast<float>(x)); }
};

[yurivict]

With channel_converter_unsigned<bits32f,uintmax_t> it breaks the same way.

[barche]

Well then, if the error has absolutely no indication of what part of K-3D is causing this, then I'm out of ideas...

[yurivict]

To me, it looks like boost tries to instantiate its own template with the wrong argument: for mpl::integral_c, the base of unsigned_integral_max_value, it supplies -1 when a type is expected. But since they both are in the boost headers, it looks like an internal boost error.

[barche]

Yes, you could try replacing the -1 with static_cast<UnsignedIntegralChannel>(-1) in your boost header, see how far that gets us. If it works that way, it's probably best to submit a bug report to Boost

[yurivict]

Substituting with static_cast<UnsignedIntegralChannel>(-1) indeed helped.

I submitted the ticket to boost: https://svn.boost.org/trac10/ticket/13397

[mloskot]

FYI, this issue has been fixed in GIL released in Boost 1.68 ​https://github.com/boostorg/gil/commit/18eacb424baf69852bb24858e1b3a5c86e1f9033 So, the https://svn.boost.org/trac10/ticket/13397 has just been closed.

Thank you for reporting the issue anyway.