This was already discussed twice previously: original filename on s3 and Facilitate public pretty S3 URLs. Yet I am still constantly having problems with this when trying to use git-annex to do some particular stuff.

My latest example is a friend that wanted to sync files to a remote Webdav server. While I know he can use webdav support, the resulting filenames on the webdav server will be basically garbled beyond recognition for any user that will use the Webdav server without git-annex.

I understand the rationale behind the existing backends, but wouldn't it be possible to implement a new backend that would just use the filename as a key?

I know this would have several downsides:

  • lack of deduplication
  • much more exposed to corruption (no checksum to check against recorded? or can this be put somewhere else?)

The main advantage, for me, is much better interoperability: any remote becomes usable by other non-git-annex clients... It would also be great as it would allow me to store only a part of my git-annex files on a remote without having a forest of empty files (on broken filesystems) or symlinks (on real filesystems) for files that are missing, something that is a massive source of confusion for users I work with. It could, for example, allow me to create thumb drives that would solve the hide missing files problem. -- anarcat

done; the new git-annex export feature allows you to export a tree to a special remote. --Joey