Remove floating point support for boost 1.58.0. As currently implemented, return by value operations are not reliable because normalization and other bit pattern changes can occur. Docs not updated as yet. Floating point support will be corrected and reinstated for boost 1.59.0, based on in-place customization and inplace and/or copy signatures to replace the unsafe (i.e. native_to_big/little ) by value returns. The big/little_to_native return by value functions are safe, but require reimplementation.

test/endian_operations_test.cpp

@@ -57,9 +57,6 @@ template<> struct value_type<long>  { typedef long type; };
 template<> struct value_type<unsigned long>  { typedef unsigned long type; };
 template<> struct value_type<long long>  { typedef long long type; };
 template<> struct value_type<unsigned long long>  { typedef unsigned long long type; };
-template<> struct value_type<float>  { typedef float type; };
-template<> struct value_type<double>  { typedef double type; };
-template<> struct value_type<long double>  { typedef long double type; };
 
 template <class T1,  class T2>
 struct default_construct