Re: [boost] [Serialization] How to force an archive to store true_type even if it is this_type?

Subject: Re: [boost] [Serialization] How to force an archive to store true_type even if it is this_type?
From: Robert Ramey (ramey_at_[hidden])
Date: 2012-04-18 19:10:55

nico wrote:
> Hello.
>
> When we serialize a polymorphic object, an archive stores the
> information about true_type (i.e. the concrete type of an object)
> only if it is not
> equal to this_type (i.e. the type of the serialized object).
>
> I.e. : the type of the serialized object is stored only if it is
> serialized from a base class.
>
> How can I force the archive to always store the type?

replace
ar & t;
with
ar & static_cast<base_type &>(t);

Robert Ramey

> PS : a word about why I'm trying to do that.
> When the true_type is "registered" (I'm not sure register is the right
> word), we can deserialize the object to any base class.
> Exemple:
>
>    SERIALIZATION:
>        Leaf* leaf = new Leaf();
>        ar & leaf;
>
>    SERIALIZATION:
>        Base* b = 0;
>        ar & b; // would work only if Leaf has been saved in the
> archive as true_type

In general, you should be sure that that type "saved" is always
the same as the type "loaded".

> For tools in a multi-team project, it' powerfull : data for tests has
> been serialized by a team, and the other team can deserialize, let's
> say an
> image, to the interface class IImage without asking: what was the
> true_tyep
>> an Image3D, an AImage2D, an Image2DRGB, an Image2DRAW, and so on.

To get the most leverage from your design, you should make modules which
use polymorphic types ONLY use the base class (reference or pointer) and
avoid using the derived type at all.  For example:

instead of using

derived_type d(initialization...)
derived_type *dptr = new derived_type(initialization...)

use

base_type & b = d(initialization...)
base_type *bptr = new derived_type(initialization...);

Better yet use

base_type &b = factory(initilization....) where factory returns some a 
base_type &

etc. ....

Robert Ramey

>
> _______________________________________________
> Unsubscribe & other changes:
> http://lists.boost.org/mailman/listinfo.cgi/boost