Karl Fogel wrote:
>Branko, what is the status of the delta combiner you were working on?
>
It's killing me. :-(
I spent less time on this in the last few weeks than I should have, and
the bloody thing is turning out to be a real bore. The theory is sound,
and the algorithm I wrote in fs-improvements also, but fiddling with all
those offsets is not something I'd recommend to anyone.
(Yes, I should have posted an update about this earlier; sorry about that.)
Anyway, I was too optimistic last time when I said it's just a few days
away.
>Right now, I'm pondering the question of how delta composition might
>be affected by our desire to operate on files "streamily" (that is, to
>avoid holding a file's entire contents in memory at once).
>
The composer will operate on delta windows, and has the same
requirements as the delta applicator: i.e., as long as we read the file
sequentially (without backtracking), we'll only have to keep one
window's worth of data in memory, plus the overhead needed for the delta
windows themselves.
>The public svn_fs.h interfaces are in good shape:
>
> svn_fs_file_contents()
> svn_fs_apply_textdelta()
> svn_fs_get_file_delta_stream()
>
>but under the hood, they turn out to be a shameless sham (or shamless
>shame, if you prefer):
>
> txn_body_get_mutable_source_stream()
> write_to_string()
>
>and for those that aren't a sham (svn_fs_file_contents), it's still
>not completely clear what the best way to handle deltification is.
>I've scattered some comments throughout the code, and am taking home
>fs-improvements.txt to reread your notes tonight. It seems reasonable
>to me that if one can compose file deltas at all, one can compose them
>for particular ranges of the target data w.l.o.g.;
>
Yes.
> however, I haven't
>actually got code to do it yet.
>
If we really need random-access deltified streams, we should take
another look at Jim's idea for indexing the delta windows in the delta
stream.
>Anyway, this is just a heads up that I'm 100% on this problem now. If
>you already have a Grand Plan, go ahead and start committing; or if
>not, let's come up with a Grand Plan asap.
>
O.K. But I can't think straight right now; got to get some sleep.
Brane
--
Brane �ibej
home: <brane_at_xbc.nu> http://www.xbc.nu/brane/
work: <branko.cibej_at_hermes.si> http://www.hermes-softlab.com/
ACM: <brane_at_acm.org> http://www.acm.org/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Sat Oct 21 14:36:32 2006