plain source -> encrypted destination: rsync + gpg (--dest-filter patch)

[jw schultz]

On Thu, Jun 26, 2003 at 04:31:29PM +0200, addady at active.co.il wrote:
> > If i recall correctly (i've not gone back to review that
> > patch) that patch required whole-file transfers two avoid
s/two/to/  #ick
> > doing the operation twice per file.  If the filter is
> > deterministic and particularly if changes have minimal
> > cascade effect you may be able to still take advantage of
> > the rsync algorithm.
> 
> You recall correctly.
> It is not clear to me how using "whole-file" and still take advantage of
> the rsync algorithm.

You meant "how to use"?  The --whole-file option disables
the rsync algorithm.

> The idea of this patch seem to me very useful.
> I'm plan to use it with the "rsyncable" gzip version
> (http://svana.org/kleptog/rgzip.html).
> The "whole-file" is not necessary is this case.  I think that the patch need
> to be modifiy so that
> whole-file will be a option.

If the receiver does the filtering whole-file is necessary.
To show why i will describe what it takes if the receiver
does the filtering without whole-file assuming gzip is the
filter.

	gunzip to temp file to generate block checksums.*

	delete temp file.

	---- interval when other files may be processed ----

	gunzip to temp file* again and merge file with blocks
	from sender.

	delete temp file.

	gzip resultant file.

Notice that this also requires a reverse filter.

If the filtering is done on the sender.

	gzip file to temp.*

	match blocks or send whole gziped file to receiver.

	delete temp file.

Observe the sender-side only requires a forward filter.

*The temp file is needed because match and block
checksumming use memory mappings of the file so i don't
think it is capable of reading from a pipe.  Merging even
does seeks because matched blocks may be out of order.

Whether the rsync algorithm is useful depends on the
determinism and change-cascade effects of the filter.
A ignore-size option would have to be created to deal with
filters that alter the file length.  The --checksum option
would be best not used as it would require every file be
filtered during flist generation and then changed files
filtered again for sending.
	



-- 
________________________________________________________________
	J.W. Schultz            Pegasystems Technologies
	email address:		jw at pegasys.ws

		Remember Cernan and Schmitt