CWG727 allowed us to explicitly specialize inside the class scope itself as discussed in this post. But when I tried to explicitly specialize a conversion function, the specialization is rejected by three major compilers(gcc, clang and msvc). It's been 9+ years since this was changed and still the three major compilers reject this while only EDG accept it. Demo

struct S {
   
    template
    constexpr operator T()     
    {
        return 10;
    }
    //rejected in all compilers except EDG
    template<>
    constexpr operator int()
    {
        return 4;
    }  
};
/*accepted in all compilers
template<>
constexpr S::operator int()
{
    return 4;
}
*/
   
int main()
{   
    S s;
    int i = s;
    i++;
   
}

We have temp.spec:

An explicit specialization may be declared in any scope in which the corresponding primary template may be defined ([dcl.meaning], [class.mem], [temp.mem]).

So, is this just the implementations being too slow to implement this or is this actually ill-formed as per the latest c++ standard.