ifedorenko

FWIW, I was not able to reproduce the problem on Mac OS. I've used _slightly_ different setup as I didn't want to mess my home directory: export RESTIC_REPOSITORY=local:/tmp/restic-test-repo export RESTIC_PASSWORD=test123...

The behaviour you observe appears to be triggered by files with multiple links to them (i.e. `.build/src/github.com/restic/restic` directory in your example). Given that incremental restore behaviour is currently undefined, I...

I propose to fix this by failing restore (with error message and none-zero exit code) if target directory exists and is not empty.

There are seems to be two independent issues/ideas here: a) use b2-specific file checksum and b) retry failed checksum validation. As I mentioned in #2303, I am not in favour...

I believe many storage providers do not guarantee consistent file reads and at least some do not guarantee consistent directory list. For example here is a quote from [Amazon S3...

It varies backend to backend, and unfortunately not all backends document their concurrency models. I do not believe S3 alone can be used to achieve bullet-proof reliable prune by itself,...

I have yet another proposal, which boils down to * Introduce new lock types to lock individual blobs and snapshots * Use double-check locking to detect and retry conflicting locks...

> purge just writes the objects it wants to delete to a file This file tells backup, check, etc, what objects to avoid. I called this file "object lock file",...

I think there are many ways to skin this cat, but in the end we will need to _prove_ the new implementation guarantees snapshots can't reference deleted blobs under all...

I did read, multiple time, honestly ;-) When `prune` selects blobs it wants to delete, `backup` must not reuse to-be-deleted blobs. Likewise, when `backup` selects blobs it wants to reuse,...