This patch fix extraction of tarfile which override tarfile uid/gid 

tarfile.py extract function override uid/gid of a file if uid/gid is not
existant on the system.

When user is not root (uid 0), common behaviour of extracting is correct.
When user is root, python implementation try to preserve (like tar -p) ownership
from tarball, but if uid/gid doesn't exist it remplace by process uid/gid.
This leads to fake the expected behaviour of preservation!

By example, extracting tarball with a root filesystem will create incorrect
ownership for file where owner/group is not on the current filesystem.

Lars, any comment?

The patch is fine. Thank you very much for it, Sebastien. I think we have to go without a unit test.

Should this bug be fixed in 3.3, or 2.7+3.2+3.3?

Yes, it should be fixed in all affected branches.

The patch looks ok. Can you push it Lars?

Yes, I can do that as soon as I've managed to wrap my head around using Mercurial and the new way of developing Python. I have been away from Python programming for quite some time and haven't adapted my workflow yet.

You can get a lot of information on this guide:
http://docs.python.org/devguide/setup.html#getting-set-up

You can also ask on IRC (#python-dev on Freenode), or by email on the python-dev mailing list.

It was really hard for me to switch from Subversion to Mercurial. Not because of the merge and the push things, but because of the 4 different Python branches! So don't hesitate to ask me questions about how I use Mercurial ;-)

New changeset 2bc122347351 by Lars Gustäbel in branch '3.2':
Issue #12841: Fix tarfile extraction of non-existent uids/gids.
http://hg.python.org/cpython/rev/2bc122347351

New changeset da59abc0ce3b by Lars Gustäbel in branch 'default':
Merge with 3.2: Issue #12841: Fix tarfile extraction of non-existent uids/gids.
http://hg.python.org/cpython/rev/da59abc0ce3b

New changeset b64ef2951093 by Lars Gustäbel in branch '2.7':
Issue #12841: Fix tarfile extraction of non-existent uids/gids.
http://hg.python.org/cpython/rev/b64ef2951093

Close as fixed. Thanks all!

> I think we have to go without a unit test.
May I ask why?  I don’t know tarfile well, but I know that a lot can be done with unittest.

It's the low-level operating system aspects of tarfile that are very difficult to test, e.g. filesystem and operating system dependent features such as symbolic links, hard links, file permissions, ownership. It is not even possible to reliably determine the filesystem the testsuite currently runs on. Also, superuser privileges are needed for some operations to work, e.g. chown(). A testsuite is normally not run as root, so a test that depends on this will never get enough coverage.

I understand.  Unit tests can work with file permissions, ownership and links, but they can’t create different filesystems or ask for superuser rights.  test_shutil contains a few tests that purport to test cross-filesystems usage but may or may not actually do it (#9999).