[boost] serialization - fast wire

From: Robert Ramey (ramey_at_[hidden])
Date: 2003-12-16 12:20:03

Hurd, Matthew wrote:

>Any thoughts about including a so called "fast wire" approach in
>serialisation.  That is, using a block of data that is the same as the
>internal representation, including alignment padding, and being able to
>blurt that out, perhaps with an optional native data representation
>(NDR) suffix or prefix that describes the block.

this is addressed in the current draft (#13) under the names read_binary
and write_binary which basically just copies raw binary data to/from 
the archive.  Currently this is implemented only native binary archives.
In the next draft it is implemented in all archives.  Implementation
depends upon the archvive.  Native Binary archives use fast raw
data copy.  Text/XML archives use slow portable base64 encoding
for binary data.

>I understand that you're concentrating on accuracy and "beauty" rather
>than performance, as you quite rightly should, but there might be some
>interesting thoughts here for you and boost::serialization, so I thought
>I'd pass this on...

My goal is to meet all the requirments "agreed" upon during last
years review of the previous version of the serialization library.  Many
of the requirements would seem to conflict.  In fact, they can
be reconciled by sufficiently fine grained factorng of the library.

I do not believe that performance will be compromised by any
of the requirements.

Robert Ramey