On Friday 25 June 2010 08:19:54 pm Michal Ludvig wrote:
> Hi Aaron,
> such a feature would indeed be very welcome. However Amazon S3 has no
> concept of symbolic links - you will have to store them as objects,
> however in some way where it's easy to tell what the object represents
> (file, symlink, ...). Your approach should also be generic enough to
> allow for storing device nodes (minor and major numbers) and directories
> (ownership, mode).
> Are you keen to work on that?
In part. I'd need to focus on scratching my immediate itch, which would
basically be an option to treat locally symlinked files as if they
were "real" files, when transmitting/syncing from local to S3 remote. 
So the first part of what you just said, about object representation, would be
handled naively; The second part, about genericity, I could accomodate
Above all, I'm certainly keen on working with you to implement in a way that
provides a good foundation for someone to extend - adding any
features/functionality I don't complete myself. (I can't not code that
Doing everything you outline sounds very interesting. The available cycles I
have is finite, so I'd need prioritize in my contribution - finishing the
portion described above first, then re-evaluate.
Would that be worthwhile for you, Michal?
 In other words: suppose within local directory /src, there is a
symlink /src/A pointing to /something/A. Then "s3cmd sync --some-flag /src/
s3://bucketname" would effectively behave as if you had first executed:
cp -a /something/A /src/A
Obviously, this would not be reversible across the corresponding "s3cmd sync
s3://bucketname /some/local/path" command.