Re: [boost] RE: Serialization Library Review

From: Dave Harris (brangdon_at_[hidden])
Date: 2002-11-17 16:42:34

In-Reply-To: <01C28E26.51606B10_at_[hidden]>
On Sun, 17 Nov 2002 10:44:33 -0800 Robert Ramey (ramey_at_[hidden]) wrote:
> they are arbitrary. but unimaginably high in my view.

I agree if we must have limits these are fairly reasonable. I don't see 
the need for limits at all. It seems like premature optimisation to even 
think about this.

I would expect that a serialisation library adopted by boost, and possibly 
a future C++ standard, would be used very widely and over an extended 
period of time. Also I am quite sensitive about issues relating to 
customer's data files. The library could still be in use in 20 or 30 years 
time, and we could want to load data serialised today. I would seek to 
avoid any unnecessary restrictions.

This is a relatively minor point, in that it is easy to change one way or 
another. Not a show-stopper.

> I have gone to great lengths to NOT include ANY non fundemental types as
> primitives.

I appreciate that. I've changed my opinion a bit; I now think most of what 
I want here can be done by writing archive subclasses.

However, I'm not quite 100% convinced that the difference between 
fixed-length data and variable-length data shouldn't be made available 
through the archive's API. Assuming variable-length for operator>>() and 
fixed-length for binary_write() almost works, but wouldn't produce very 
readable text files. Maybe this is put-uppable with, though.

I do think that variable-length binary archive subclasses should supplant 
boarchive as the standard binary archive, because of their platform 
neutrality. I suppose that's unlikely to happen unless someone actually 
writes and submits them. Anyway, this issue isn't a show-stopper for me 
any more.

-- Dave Harris