Re: [boost] Serialization Review Results

From: Dave Harris (brangdon_at_[hidden])
Date: 2002-12-13 10:06:37

In-Reply-To: <20021213015852.3107.qmail_at_[hidden]>
On Thu, 12 Dec 2002 17:58:52 -0800 (PST) Augustus Saunders 
(infinite_8_monkey_at_[hidden]) wrote:
> I'm still not sure what UTD refers to.

"User defined type". The payload, the objects which get serialised or 
persisted. 

> What should a serialization library provide that iostreams doesn't?

(1) Archiving of polymorphic pointers. Loading the archive needs to create 
objects of the dynamic type rather than the static type. (I don't believe 
we have a good solution to this yet?)

(2) Archiving of shared objects. If two pointers point to the same object 
on saving, they should still point to the same object on loading.

(3) A canonical scheme for managing versioning.

(4) Space efficiency. Eg metadata such as version information should be 
stored once per class rather than once per object.

(5) Support for a variety of formats, with different trade-offs. Including 
(a) efficient formats; (b) human-readable formats; (c) platform 
independent formats; (d) standard formats such as XML. I want to be able 
to switch between formats without changing the code in my user-defined 
types.

[Later...]

I've now read your own answer to your question and we seem to have a lot 
of agreement.

-- Dave Harris