Re: Proposal: statefull objects

From: Gennadiy E. Rozental (rogeeff_at_[hidden])
Date: 2001-08-30 15:00:14

--- In boost_at_y..., Douglas Gregor <gregod_at_c...> wrote:
> On Thursday 30 August 2001 03:22, you wrote:
> > Hi, Fernando
> >
> > I spend some time overlooking this discussion and realized that 
what
> > really bother me is not the extern semantic. I agree with you that
> > providing an ability to the user to write s == value and *s == 
value
> > is not very good idea. It's confusing and error prone. So external
> > semantic should be pointer-like. BUT the implementation should be
> > value-based not pointer based, to eliminate dereferensing. Here 
what
> > I did. Compiled and tested on MSVC and sun workshop
> 
> > class optional {
> > public :
> >     typedef T value_type;
> >
> >     // Constructors does not allow convertion from type T. Use
> > *optional<T> = value; instead
> >     optional()
> >
> >     : m_initialized( false )                            {}
> [snip]
> >     // Data members
> >     T           m_v;
> >     bool        m_initialized;
> > };
> 
> This requires T to be DefaultConstructible, which is unnecessarily 
> restrictive.
> 
I realize that. This is tradeoff. It will be acceptable in most 
cases - at least in original examples.

If you can't live with that I would propose add something like this: 
anothey policy template parameter

template <typename T, class InstantiationPolicy = 
ValueIntantiationPolicy<T> >
class optional {
    void assign( T const& t ) { m_value = t; }
private:
    T& value_helper() { return m_value; }
    InstantiationPolicy m_value;
}

template<typename T>
class ValueIntantiationPolicy {
public:
    operator T&()   { return m_v; }
    void operator=( T const& arg ) { m_v = arg; }
private:
    T m_v;
}

And you could add as much other policies as you need.

Gennadiy.

> 	Doug