> > I believe that svn_tdelta__winfo_t's data_offset is always equal to
> > the window's new->len. If you eliminate that variable, you can get
> > rid of svn_tdelta__winfo_t altogether. That would simplify your code
>
> Indeed it is, but that would mean exposing the implementation details
> of svn_string_t. If that's O.K., I'll do that an remove svn_tdelta__winfo_t.
svn_string_t isn't supposed to be opaque --- it's transparent. You're
supposed to be able to go into a string and tweak things arbitrarily.
> In fact, I already _do_ have hash tables and other state ... I've
> just now finished the core vdelta driver, along with the dumb
> windowing method. I need to clean up the code a bit, so I'll have a
> patch ready sometime tomorrow.
Great!
Experimenting with windowing algorithms sounds like a lot of fun.
Obviously, one wants to have the two windows slide along so as to have
the greatest degree of resemblance between the source and target.
Deltas for huge files with small local changes should still be small,
but if your windowing is not ideal, they could still end up being
proportional to the overall size of the file.
Received on Sat Oct 21 14:36:06 2006