Re: [boost] Re: Serialization Review Results

From: Dave Harris (brangdon_at_[hidden])
Date: 2002-12-14 08:48:37

In-Reply-To: <01C2A226.AC2B48D0_at_[hidden]>
On Thu, 12 Dec 2002 21:37:29 -0800 Robert Ramey (ramey_at_[hidden]) wrote:
> One thing that my usage of "serialization" does not include is 
> transformation or mapping to any other arbitrary format.  Some
> formats may not be rich enough to capture the "meaning" of the
> set of objects in a useful way.  For example capturing an
> arbitrary set of C++ objects in a CSV file might be possible,
> but it would be no richer than a simple sequence of bytes.
> Hence, though possible, it would be pointless.

This seems like a difference of opinion that I would like to understand 
better.

Let's say I have a bunch of C++ objects that I want to export to a 3rd 
party spread-sheet, eg to draw a pretty graph, and the spread-sheet only 
understands CSV. Is there any reason I shouldn't write an archive class to 
generate CSV directly? I don't care if the full "meaning" is not captured, 
provided I get enough for the spread-sheet to do its stuff.

Or suppose I have some CSV data I want to import. I don't care that CSV 
cannot represent every possible C++ data structure, I just want some C++ 
objects that reflect what is in the CSV.

To me being able to do these things would not be pointless. I'd get paid 
to do it! Of course writing CSV support from scratch isn't much harder 
than adding the generic boost serialisation, but using boost I would also 
get other formats for free.

Is it just that you are not personally interested in this stuff, or that 
it was too much work to consider for the original submission, or do you 
see dangers lurking here? Many of us want the library to be able to 
support lossy formats like CSV. Will doing that compromise the design?

-- Dave Harris