Store and restore permissions, owner, group

[sourcefrog]

No description provided.

[rbtcollins]

Were you thinking to store these numerically, or symbollically, or both?

What about extended ACLs, or is that a separate ticket?

[sourcefrog]

I was going to start with just unix perms, numerically - the easy case. For user and group, it seemed more like to be robust across restoring on different machines to store the names, perhaps avoiding repeatedly serializing the same names. Perhaps it's pretty common to have all (or almost all) files be owned by one user. Extended ACLs, I would probably handle separately. I don't have many opinions on them.

…

[believeinlain]

Just thought I'd mention here that I'm working on this feature currently, in my fork here: https://github.com/believeinlain/conserve/tree/add-unix-permissions

I have unix permissions working and cargo test -- --include-ignored passes (with tests adapted to verify permissions are being restored correctly). I don't really plan on implementing Windows functionality, as I have no use for it, so the permissions are stored numerically as a u32 in the archive for each entry.

I want to take a look at user and group id before submitting a PR.

For my purposes, I really need it to be able to restore permissions, user, and group properly, but I don't need cross-platform or cross-machine compatibility, so just storing user and group id should be fine.

Trying to make archives work across machines is going to be tricky regardless, since the user id and username are not always consistent. Probably best to introduce user and group as an optional feature.

[sourcefrog]

@believeinlain great!

I think for more general usage I might want to store usernames rather than UIDs: even restoring to a different Linux machine they can end up accidentally different, even when the same usernames exist... Naively storing the username/group in the index would make the uncompressed index larger, though it should get mopped up by compression.

Just storing unix perms as u32 seems fine.

Maybe there could be an option to store ids vs names...

[believeinlain]

I created pull request #193 which stores and restores unix mode bits, based on the code I wrote earlier but cleaned up a bit. It passes all tests on my system but seems to be failing in CI due to the group write permission bit being different than expected - not yet sure how to address that.

I haven't got to storing user or group owner yet but I feel like that's a separate feature from just the mode bits, since they are stored separately.

[believeinlain]

PR #194 should address every part of this issue on Unix. Went with storing user and group names as strings for simplicity's sake.