opinions? suggestions? code review?

Thomas, so you have a single stdin path? I would suggest at least a --stdin-name, so you can pipe different files into attic. A more advanced option would be, when the stream is a tar archive, to pick up names from the tar metadata

it is currently using the convention that "-" instead of a filename means "read from stdin".
piping any file into attic would then work like "attic create repo - < any_file".

so why would we need --stdin-name (or --read-from like I would call it)?

That's not what I meant. When you archive a directory, attic stores metadeta from the filesystem (i.e. it knows that you've archived /etc/hostname, /etc/passwd etc). When you archive from "-", the path metadata is simply "stdin" (AFAICT from your commits). So --stdin-name would provide the missing filename metadata for the piped-in stream.

More interestingly, if a tar archive is piped in, attic could pick up metadata from the tar format. This is what a --stdin-format would do (leaving room for --stdin-format=plain, stdin-format=tar, --stdin-format=zip, so that you can treat a piped-in tarfile either as a single file OR as a collection of files)

@danmbox please give specific use cases / examples where that metadata would be interesting. Also, how would they be used (e.g. at extract time)? E.g. if i make a disk image by reading from /dev/sda, that filename (and other metadata of that file) would be rather uninteresting. Really interesting is rather image contents, like "win 7 pro 64bit std install, machine type x", but I can't get that info from /dev/sda's metadata, but would have to put it manually in the archive name.

Also, I have no idea why you are referring to piped tar archives here. I'ld suggest you split that off to a separate ticket.

What's wrong with using the archive name for that? Since backing up stdin will always just be an archive containing a single (unnamed) object, you could simply indicate the name of that object using the name of the archive.

And I don't see why it should be attic's job to extract any information from a tar stream, or any other specific file format, for that matter. There's already a tool for that task and it's called tar.

@dnnr, the archive name is one piece of metadata, the filenames are a different piece of metadata. Let's not mix them up. And unless you have something like tar parsing, it's impossible to create a snapshot containing multiple files without saving the stream to a temporary disk file (which kills performance, and also requires extra disk space, proportional to the uncompressed size).

In the meantime I've tried pg_dump -CFt mydb | /opt/attic/bin/attic create /tmp/attic::pg_mydb several times and I can attest that it works. My comments regarding filename metadata could be saved for later as an enhancement request, if this pull request gets merged.

The tar format idea is stolen from zbackup, and is one way to allow piping in a collection of files (not everyone wants to pipe in just virtual images -- ). E.g. pg_dump -Ft generates multiple filenames inside its tar archive stream, and they are useful to see when listing the snapshot.

closing this pull request, seems unwanted.