Re:[boost] [serialization] write/read binary

From: Robert Ramey (ramey_at_[hidden])
Date: 2003-10-30 17:27:23

Vladimir Prus wrote:

>I have one more comment about the serialization library.

>I have a class with three member variables;

>        unsigned resource_count;
>        unsigned content_size;
>        int* content;

>I started writing 'serialize' method

>        void serialize(Archive& ar, unsigned int version)
>       {
>            ar & resource_count & content_size;
>	    ar.write_binary(content, content_size);
>        }

>and realized that it won't work for input archives -- so I need to split 
>serialization. This looks like overkill for such a simple class. Maybe, a 
>better solution would be to introduce 'array' method which
>- takes a pointer and the size of elements
>- is defined for both input and output archieves.

>        void serialize(Archive& ar, unsigned int version)
>        {
>            ar & resource_count & content_size;
>	    ar.array(content, content_size);
>        }

>and it will work for both input and output. (For input, the pointer need to be 
>created). Another solution, which is probably clearer, it to define a wrapper 
>class c_array, which takes pointer and a size and does the right thing. I'd 
>write:

>      ar & c_array(content, content_size);

>Surely, it's not 100% required, but would be nice.

There a several possibilities here that might suggested.

a) you idea above

b) arrays are already addressed so that the follownig might work
if count is known at compile time

ar & static_cast<int[size_t] &](*content)

c) override serialize for an int *

....

d) replace read_binary and write_binary with
serialize_binary(Archive ar, void * & t, unsigned int size)

and define an optional splitter for this - I like this idea the best

I'v put off any work with read/write archive for now as I want to 
get all the pending stuff absolutly done before I add any more
features.  As this is now only implemented for binary archives
its still a little up in the air as to its final form.  I was going
to do it sooner but I need some iteratator adaptros for 
changing data to base64 and escaped ascii etc.  Then I
wanted to convert my old home-made iterator adaptors
to the recently introdude MOAIA* . I finally got that about
done when the load_construct question arose, then the jam
build/test got "fixed" when I wasn't looking so the stack
got deeper faster than I could pop it off.  Anyway, I'm 
just getting back to a fixed point but a read_binary/write
binary are still unaddressed.

Robert Ramey

*Mother Of All Iterator Adaptors