Re: [boost] [variant] Feature request

From: Larry Evans (cppljevans_at_[hidden])
Date: 2006-01-08 10:19:25

On 01/08/2006 04:37 AM, Ion Gaztañaga wrote:
[snip]
> types to call that constructor. What I would propose is the possibility 
> of selecting via integer value which member of boost::variant will be 
> constructed, just like which() returns the index of constructed member. 
> This way, we can efficiently compact members of classes.
> 
> class MyRWLock
> {
>     public:
>     enum { r_prioriy, w_prioriy, a_prioriy };
> 
>     MyRWLock(int run_time_policy);
> 
>     private:
>     struct r_prioriry_members
>     {
>      //...
>     };
>     struct w_prioriry_members
>     {
>      //...
>     };
>     struct a_prioriry_members
>     {
>      //...
>     };
> 
>     //Union of needed members
>     boost::variant<r_prioriry_members,
>                    w_prioriry_members,
>                    a_prioriry_members> m_data;
> };
> 
> inline MyRWLock::MyRWLock(int run_time_policy)
> : m_data(/*what do I have to do to select a constructor?*/)
> {
> 
> }
> 
[snip]

Wizofaus had a somewhat similar problem which he posted here:

   http://tinyurl.com/dq732

In summary, he wanted a run time integer index to select a
constructor to create some element of a variant.  A solution
posted in http://boost-consulting.com/vault/ under:

   Template Metaprogramming/little.funvec_.zip(little.funvec_foreach.cpp)

uses a static constant vector of function pointers:

     setter_rec<RecId>::prp_vec_ctor const
   setter_rec<RecId>::our_prp_vec;

which was used in:

   setter_rec<RecId>::set(void*in,int prop,int value)
   { our_prp_vec[prop](in,value); }

The above set corresponds to your runtime selection of
the ctor from a vector of "pseudo" ctor's stored in
our_prp_vec. The void*in corresponds to your &m_data.
The above value argument would not be needed since the
"pseudo" ctor's in our_prp_vec would supply the
required arguments while invoking a placement new
on &m_data.

I've got something almost like this on my local machine.
It's a successor to the boost sandbox boost/indexed_types.
Let me know if you're interested.