[darcs-users] Clever idea for storing binaries, comments?

[Quag]

On Wed, 20 Oct 2004 18:20:57 -0500, Taral <taral at taral.net> wrote:
> On Wed, Oct 20, 2004 at 05:32:38PM -0400, Michael Conrad wrote:
> > I'm not sure if this has been thought of by someone else (what idea
> >  hasn't?) but I was just thinking it would be cool if the binaries
> > were able  to take advantage of diffing to save space.  This is a
> > naturally difficult problem, but what if we had a system
> > where you add a line break to the hex dump whenever you run into
> > a 0-byte?  That would give length-512 text lines (in hex) on
> > average, so maybe we'd want to beak after any byte matching 0x1F
> > (length-64 text lines) or something.
>
> Or we could use a binary diff algorithm like bsdiff.

I had a look at xdelta3 (http://xdelta.org/xdelta3.html) a little
while ago. There are a few limitations: 1) it is a C library (although
it is small), 2) It only creates a one-way diff src->target, so the
patch would have to store two diffs which would require both the old
and new binary files twice.

Jonathan.