tar --listed-incremental includes all directories in each increment?

Question

Given a tree structure with files in it, say:

source
|-- bar
|   +-- three.txt
|-- baz
|   +-- four.txt
+-- foo
    |-- one.txt
    +-- two.txt

If I do

tar czvf dest/archive-1.tgz --listed-incremental dest/archive.snar source/*

That creates a .tgz file containing all the current files and folders, just what I'd expect:

source/bar/
source/baz/
source/foo/
source/bar/three.txt
source/baz/four.txt
source/foo/one.txt
source/foo/two.txt

If I immediately do:

tar czvf dest/archive-2.tgz --listed-incremental dest/archive.snar source/*

That seems to create .tgz file containing all the subfolders of the current directory, with no files:

source/bar/
source/baz/
source/foo/

This isn't what I want. On a toy example like this one, it's not so bad, but on a large folder structure it makes it much harder to tell that the .tgz is essentially empty. It's a lot of noise. Ideally, tar would tell me that there are no updated files and exit with an error code, but even if it made an empty tar, that'd be better (for me, in this case) than an archive with lots of empty folders.

Is there any way to make tar --listed-incremental behave the way that I would like? I've done a few searches to find solutions, but I haven't found much.

I may end up just stop using --listed-incremental and doing some work with find to assemble a file list instead, but that's a lot more work if there's a way to make --listed-incremental do what I want. Any suggestions?

Answer 1

The manual page mentions:

When extracting from the incremental backup GNU tar attempts to restore the exact state the file system had when the archive was created. In particular, it will delete those files in the file system that did not exist in their directories when the archive was created.

To implement this they would either need a special file within the archive containing a list of all deleted files, or a copy of all directory inodes so that the relevant files can be removed. To keep in-line with the archive format it was probably cleaner for them just to include the directories rather than complicate things with an additional file.

Edited in an attempt to clarify the answer:

Example 1: Extending the original example...

Step 1: Remove one of the existing files, and create an incremental archive

$ rm source/foo/two.txt
rm: remove regular empty file `source/foo/two.txt'? y

$ tar czvf dest/archive-3.tgz --listed-incremental dest/archive.snar source/*
source/bar/
source/baz/
source/foo/

Step 2: Recreate the previous file within foo

$ touch source/foo/two.txt

Step 3: Extract the incremental archive, which will remove the file we just created (it knows that this file was missing from the original archive as the directory entry, source/foo/, included in archive has no mention of it in its "table-of-contents")

$ tar xvfz dest/archive-3.tgz --listed-incremental dest/archive.snar
source/bar/
source/baz/
source/foo/
tar: Deleting `source/foo/two.txt'

Example 2: Including the top-level directory, where the foo directory (and it's files) will be deleted if missing from the incremental archive.

Step 1: Create the initial archive using "source" rather than "source/*"

$ tar czvf dest/archive-1.tgz --listed-incremental dest/archive.snar source
source/
source/bar/
source/baz/
source/foo/
source/bar/three.txt
source/baz/four.txt
source/foo/one.txt
source/foo/two.txt

Step 2: Remove the foo directory and create an incremental archive that does not contain a reference to foo.

$ rm -rf source/foo
$ tar czvf dest/archive-2.tgz --listed-incremental dest/archive.snar source
source/
source/bar/
source/baz/

Step 3: Recreate the foo directory

$ mkdir source/foo
$ touch source/foo/one.txt
$ touch source/foo/two.txt

Step 4: Extract the archive (at which point foo will be deleted as it does not appear within the "table-of-contents" for the "source/" entry.

$ tar xzvf dest/archive-2.tgz --listed-incremental dest/archive.snar source
source/
tar: Deleting `source/foo'
source/bar/
source/baz/