Alternative tool backends

[Michael Terry <mike mterry name>]

Our code base has always paid lip service to the idea that duplicity isn't the only possible tool backend we might support. And that concept was one of the reasons I deprecated the non-gvfs-based cloud backends a while back, to make it easier to support a wider set of tools.

Duplicity has been a good tool backend for the past ten years. But realistically, it still creates problems for us. (A) it is complicated enough that a lot of the bugs we receive are actually due to it (or our interactions with it anyway). And (B) it's not clear whether it will be ported to python3 by 2020 when python2 goes EOL. Which will make packaging and supporting it harder for distributions.

I think it would be wise to at least examine the field and consider supporting another tool backend. Even if we don't do anything about it, it would be a useful exercise to understand our options.

Requirements

Here are my initial thoughts on must-haves and nice-to-haves. Any thoughts on these lists?

Must-haves:

- Encryption

- Incremental backups

- Supports POSIX file permissions, even when backing up to locations that don't

Nice-to-haves:

- Baked in support for gvfs (we could always use gvfsd-fuse to expose a location as local directories, but that was a little brittle the last time I tried it years ago)

- Backup files are as non-custom/opaque/unique as possible. Platonic ideal would be a user being able to browse files using normal tools (think an rsync backup). But with all our desired features, that's not likely. Still, something to think about.

- Compression

- Well maintained / well reviewed / considered stable / well translated

Default Tool Migration

If we did add support for a new tool backend, and we eventually liked it as a better default backend, migration to it would need careful consideration.

We do a full backup every few months. So once it was time for a new fresh backup, we could make it using the new tool. And then support restoring with the correct tool depending on the date chosen. Obviously it's not as easy as that. But it *would* be possible to quietly migrate folks.

But anyway, that's a whiles away. I'm not even sure there is an obvious good candidate that we'd like more than duplicity, yet.

Next Steps

I'll start looking at the field. And update a wiki page with the results:

https://wiki.gnome.org/Apps/DejaDup/Design/Tools

I welcome thoughts! And pointers to potentially suitable tools. I know there's an open bug about supporting the Restic tool (LP: #1800996). I guess I could look into Restic.