Re: [boost] offset_cast proposal

From: Terje Slettebø (tslettebo_at_[hidden])
Date: 2002-11-25 10:52:35

>From: "Pavol Droba" <droba_at_[hidden]>

> On Mon, Nov 25, 2002 at 02:47:54PM +0100, Terje Sletteb? wrote:
>
> > How did you do it on the ARM? As I understand, it requires
word-alignment
> > for words (and half-word alignment for half-words, if that is supported
on
> > the given ARM version).
> >
> Well it is probably a conjuntion of luck and the kind of data we were
working
> with. As I have mentioned before, offset_cast was used mostly for network
> packet analysis. And fortunately almost all network protocol structures
> have correct aligmnent. So we were casting only to correct offsets :)

Ah, yes, I was thinking of that, too. That's probably designed deliberately
that way.

> > > I have mentioned before, that binary analysis could be done also using
a
> > > more
> > > complicated framework. Would it make sense to create one?
> > >
> > What does it do?
>
> The framework I'm proposing should allow user to specify a structure
layout
> ( like what field starts where and what type if have ). Then there would
> be a kind of possibly hidden management which would do nescesary copy-in
and
> copy-out operations.
>
> I can imagine it should be something like this:
>
> void modify_binary_header( void* pData )
> {
> binary_struct b;
> b.add_field<int>( 0, "field1" );
> b.add_field<char>( 4, "field2" );
> b.add_field<int>( 5, "field3" );
>
> b.map_to_binary( pData, offset );
>
> b["field1"]=b["field2"]*2;
> }
>
> Please consider this as very preliminary example, just to show a desired
functionality.
> Construction of  tructure can be probably done in some better way using
static templates.
>
> Idea is that binary_struct would be responsible to some sort of safe
copying ( possibly byte-by-byte )
> of binary data into/out from the internal field representation. It can
provide also some
> other functionality like byte order corretion or handling of sub-byte
fields ( something what
> is not aligned to byte boundary )
>
> Well, this is my idea how to resolve such a problem where simple mapping
using offset_cast
> is not sufficient.

Aha, I understand. It's an interesting proposal. I guess it would still
require some way of detecting, or specifying, alignment requirements (to
avoid doing it completely pessimistic, by only reading a byte at a time).

Regards,

Terje