Re: [Boost-users] [serialization] Serializing classes with nodefault constructors

From: Hugh Hoover (hugh_at_[hidden])
Date: 2006-05-20 19:45:12

On May 20, 2006, at 14:49, Robert Ramey wrote:

> Hugh Hoover wrote:
>> Robert Ramey has mentioned in other messages that default
>> construction with de-serializing is tantamount to "proper"
>> construction anyway, and I agree with that - you need to consider the
>> de-serialization as a form of construction rather  than something you
>> do the object after construction.
>
> Hmmm - I don't remember saying that.  In any case I feel quite

In the discussion:
http://lists.boost.org/boost-users/2005/09/13827.php

You don't say exactly that - but that's what I inferred :)

> confident that the load_constract_data method is functional,
> convenient, and even necessary in some cases.  I have in
> reason to believe that it doesn't work with abstract base
> classes either.  The serialization library keeps track of the
> types of derived pointers.

as near as I can tell, it DOES work for non-abstract derived classes  
in my tests (and the examples), but if you try to deserialize like:

Base * pb = NULL;
ar >> pb;

Then you end up in load_construct_data_adl<class Archive, class T> 
(Archive& ar, T*t,version) - which calls  
load_construct_data<Archive,T> - which gets the default version  
because there is no definition of load_construct_data for the base  
class (in my case, anyway).  I've not managed to get it to work in  
this case - and I didn't need it to, since I could write a serialize  
function with a private default constructor.