[patch] paranoid checksum checking

[Nick Burrett]

Wayne Davison wrote:
> On Wed, Jul 27, 2005 at 11:05:17AM +0100, Nick Burrett wrote:
> 
>>The attached patch incorporates your suggestions, and calls the option
>>'--read-verify'.
> 
> 
> Note that your suggested option does not need to affect the --checksum
> code in any way -- that code sends pre-transfer checksums to determine
> if files need to be sent based on their checksum value rather than on
> their file attributes (e.g. size & modtime).

> Any transferred file is
> accompanied by a full-file checksum for verification purposes, so you
> don't want --read-verify to imply --checksum, as the two options are
> not related.

The updated patch that I supplied, removed this association.  My code
change only enables MD4 calculations of files for the case when
always_checksum == 0 && read_verify == 1.

> In fact, the file that rsync wrote out is already known
> to match the checksum of the sending side, so the only thing that your
> patch will do is check for disk-write errors (e.g. bad disk sectors).

.. or dodgy NFS write operations.  Though the more I think about it, the 
more inclined to think that the patch doesn't really achieve much.  If 
we were writing dodgy data to disk, then either rsync would be broken 
and the world would complain, or my computer would be broken and I'd 
have noticed a long time ago.

Nick.