[boost] Re: Re: Re: Re: Re: relational table as anSTLcontainer--anyinterest?

From: Vertleyb (vertleyb_at_[hidden])
Date: 2002-11-02 22:17:52

> How 'bout this...
>
> #include <cassert>
>
> class some_class
> {
>     public:
>     some_class() : number(0) { }
>     int number;
> };
>
> class another_class
> {
>     public:
>
>     template <typename some_type>
>     struct some_type_defer { typedef some_type t_; };
>
>     template <typename some_type>
>     typename some_type_defer<some_type>::t_ & field()
>     {
>         static some_type test_object;
>         return test_object;
>     }
> };
>
> int main(int argc, char** argv)
> {
>     some_class c_;
>     another_class x_;
>
>     c_.number = 1;
>     c_ = x_.field<some_class>();
>     assert(c_.number == 0);
>
>     c_.number = 2;
>     x_.field<some_class>() = c_;
>     assert(x_.field<some_class>().number == 2);
> }
>
> ...It works with g++ 2.96(RedHat) and 3.2.

Do I understand correctly that if you use

template <typename some_type>
some_type & field()
{
        static some_type test_object;
        return test_object;
}

this  does not work, but if you change it to be:

template <typename some_type>
typename some_type_defer<some_type>::t_ & field()
{
        static some_type test_object;
        return test_object;
}

it works fine?  This is scary, isn't it? :o)

> Aren't templates fun... and twisted ;-)

They sure are.  Reminds me of science fiction movies, where robots start
leaving their own lives, independent of their creators :o).  Template
meta-programming was DISCOVERED (!!!)  Anybody knows who discovered it?

One more option would be to use:

tuple.field(column()); // create a temp object and pass by reference

Would compilers optimize away unnecessary temporary object creation?

Arkadiy