Re: [boost] RE: Serialization Library Review

From: Beman Dawes (bdawes_at_[hidden])
Date: 2002-11-17 19:39:46

At 12:51 PM 11/17/2002, Robert Ramey wrote:

 >From: "Peter Dimov" <pdimov_at_[hidden]>
 >
 >>> I'm really interested in the XDR format, not because I care about the
 >>> format itself, but because others seem to use it as some sort of 
litmus
 >>> test for serialization libraries. Thus knowing that Robert's library
 >>> handles XDR well is of interest.
 >
 >>FWIW, in my experience, XML is a better test for whether a serialization
 >>library handles custom formats well. Sequence-based, header-only formats 

 >are
 >>very similar, but a reasonable XML serializer creates a tree-like
 >>representation, with nested tags.
 >
 >After a cursory investigation as to what it would take to make an XML
 >archive, I've concluded that:
 >
 >a) I don't thing XML is rich enough to capture all possible C++ data
 >structures

Uh... I think one of us must be misunderstanding something about XML. If 
you can represent serialization of some objects as a character based file, 
can't you turn it into dumb XML by wrapping it in 
<my_dumb_tag>....</my_dumb_tag>?  Perhaps you mean you don't think it is 
possible to define a set of tags that themselves are rich enough to capture 
all possible C++ data structures?

But so what? You can just supply content directly for any C++ structures 
that don't represent well in the XML tagged style.  And we know from 
experience that many C++ data structures can be usefully represented in an 
XML style format.

 >b) This would imply the creation of a system of reflection for C++
 >    i) I not convinced this is a good idea
 >    ii) should be dealt with as an indepent project in any case
 >
 >c) XML has a central task to represent data in a platform/programming
 >language
 >independent way.  Serialization has the central task of saving/restore 
the
 >totality of the state of C++ data structures.  These tasks are not
 >identical
 >and any course of action based on the presumption that they are is
 >doomed to be frustrating an most likely a failure in my opinion.

That seems a rigid view of XML.  Just think of it as a way to markup the 
output you are already producing to make it a bit easier for other tools to 
work with. Some of the whitespace or other delimiters become tags. If the 
output you are already producing captures various C++ data structures, 
adding XML markup just enhances the already captured data. That's probably 
a dumb way to use the power of XML, but it may be sufficient for 
serialization needs.

On the input side, I don't think anyone was suggesting being able to parse 
the full generality of XML, but just to be able to take in the exact markup 
that was previously output.

--Beman