Fix is_floating_point symbol is not a class template

[sdebionne]

Closes #760

Tasklist

• Add test case(s)
• Ensure all CI builds pass
• Review and approve

[mloskot]

Thank you!

[striezel]

Are you sure this is the correct fix? Some of the GCC and Clang builds now fail with errors like

./boost/gil/io/typedefs.hpp:35:51: error: template argument 1 is invalid
   35 | template<> struct is_floating_point<gil::float64_t> : std::true_type {};
      |                                                   ^

or

./boost/gil/io/typedefs.hpp:34:37: error: use of undeclared identifier 'gil'; did you mean 'boost::gil'?
template<> struct is_floating_point<gil::float32_t> : std::true_type {};
                                    ^~~
                                    boost::gil

[sdebionne]

This fix might have uncovered other issues 😅 . I'll have a look !

[sdebionne]

TBH I did not know that there was an implementation of is_floating_point in Boost.GIL. The file io/typedefs.h include <type_traits> though.

[striezel]

The only place where is_floating_point is used within GIL seems to be the TIFF I/O extension, and there it's only an implementation detail: https://github.com/boostorg/gil/blob/688acf78f7fc4ecb5b1c6f1e4a8f4a5939d28e9b/include/boost/gil/extension/io/tiff/detail/is_allowed.hpp#L101C13-L101C30

Furthermore I'm a bit worried, because cppreference.com states:

If the program adds specializations for std::is_floating_point or std::is_floating_point_v, the behavior is undefined.

Doesn't sound like this is something one should do. As long as those lines were in the boost namespace (before the PR), that was not a specialization of std::is_floating_point but boost::is_floating_point, a different thing - although probably with similar intentions. But with the move to the std namespace it now is a specialization.

[sdebionne]

Ok let's revert my fix then and reopen the MR.