greatroar

I don't make the decisions, but I think it would be sane to produce a tarball with missing entries and signal the fact that it's incomplete by restic dump's exit...

Instead of a side channel, we could also use internal/errors.Fatal inside internal/dump, but that would also be a pretty large change.

I find 1% acceptable, but we could round up to a constant block size instead, since storage systems will do that anyway. The common objections to doing so don't really...

@MichaelEischer, you wrote in https://github.com/restic/restic/issues/5291 that > Padding an individual chunk is currently only possible for compressed blobs (zstd supports skippable frames that could be used to store padding bytes)...

Or, as you guys discussed over at #5331, we can add a nominally zero-sized (but padded) blob to every pack and teach prune to not treat it as a duplicate,...

I hit a snag. While inserting a skippable frame this way doesn't break the format, it does break older restic implementations that assume a zero-sized blob is uncompressed. That is,...

> We could discuss if after repacking (in `prune`) such a padding blob is still needed or not... The way pruning/repacking works now, it will remove the padding blob, then...

This looks perfectly reasonable as a feature. Some preliminary implementation hints: From what I gather, you can only determine this attribute using a [statx(2)](https://man7.org/linux/man-pages/man2/statx.2.html) call. Go's x/sys/unix has this as...

A repo can have multiple passwords (restic key add), so multiple admins can keep their passwords secret from one another. What they implicitly share, though, is the master key for...

Oh, I hadn't checked the archiver tests. Making this a draft. Maybe wait for #5021 to be finished before we tackle this.