Re: [boost] Re: [serialization] problems with writing binary

From: Vladimir Prus (ghost_at_[hidden])
Date: 2004-01-13 02:21:07

Robert Ramey wrote:

> >> struct A
> >>   {
> >>             int* content;
> >>             int size;
> >>
> >>             template<class A>
> >>             void serialize(A& a, unsigned version)
> >>             {
> >>                 a & size;
> >> a & make_binary_object(content, size*sizeof(int));
> >>             }
> >>   };
> >

> This is a question of motivation, intent and utility.
>
> a) motivation - I needed a method for save/loading raw bnary data.
> the save_binary/load_binary fill this need but I found I had two
> annoying features.
> 1) The above would have to use split

Right, that's annoying.

> 2) There was no way to specify an nvp wrapper.

That's no good, either. However, the above has problems which make it 
half-a-solution, IMO.

First, the above code won't compile with gcc. One needs:

                 boost::serialization::binary_object bo =
                 boost::serialization::make_binary_object(content, 
                                                          sizeof(content));
                 a & bo;

That's not convenient, but I think some solution can be found. But, second, 
the above code crashes. I attach a complete example. That's my primary 
problem with make_binary_object. An example which looks nice (and probably 
look nice even to you), crashes.

It won't crash if you either
1. Allocate the memory yourself (in which case you need split)
2. Restore a member variable.

If make_binary_object can be reliable used only for the second, then probably 
it should be called 'make_pod_object'? And should not take the size at all?

- Volodya