git-annex/doc/todo/resuming_encrypted_uploads.mdwn

Resuming interrupted uploads to encrypted special remotes is not currently
possible, because gpg does not produce consistent output. Special remotes
that could support resuming include rsync and glacier.

Without consistent output, git-annex would need to locally cache the encrypted
file, and reuse that cache when resuming an upload. This would make
encrypted uploads more expensive in terms of both file IO and disk space
used. 

[It would be possible to write to the cache at the same time the special
remote is being fed data, and if the special remote upload fails, continue
writing the rest of the file. That would avoid half the overhead, since
the file would not need to be read from, just written to. (Although OS
caching may accomplish the same thing.)]

Also, `git annex unused` would need to show temp files for uploads,
the same as it currently shows temp files for downloads, and users would
sometimes need to manually dropunused old uploads, that never completed.

The question, then, is whether resuming uploads is useful enough to add
this overhead and user-visible complexity.
--[[Joey]] 

> The new-style chunking code chunks and then encrypts. This means that
> each individual chunk is a stand-alone file that can be decrypted later,
> and so resumes of uploads to encrypted, chunked remotes works now.
>
> I think that's better than the ideas discussed above, so [[done]]
> --[[Joey]]