Crashes when tarfile contains a symlink and unpack directory contain it too

[barry]

I added some comments.  I'd like to get a second opinion about the fix
implementation, but if it looks okay to someone else, I'll be happy to land
this.

https://bugs.python.org/review/23228/diff/13610/Lib/tarfile.py
File Lib/tarfile.py (right):

https://bugs.python.org/review/23228/diff/13610/Lib/tarfile.py#newcode2183
Lib/tarfile.py:2183: os.remove(targetpath)
I'm a little concerned about the security implications of this.  It means a
carefully crafted tarball could retarget existing symlinks.  But I think that's
the way tar(1) works if you don't give it -k (at least GNU tar).  I'd like to
get a second opinion.

https://bugs.python.org/review/23228/diff/13610/Lib/test/test_tarfile.py
File Lib/test/test_tarfile.py (right):

https://bugs.python.org/review/23228/diff/13610/Lib/test/test_tarfile.py#newc...
Lib/test/test_tarfile.py:2177: self.cwd = os.getcwd()
I'd probably move the chdir into a try/finally inside the test itself, just so
any weird failure causing tearDown not to be called wouldn't corrupt cwd for
other tests.  Also, putting the chdir closer to the os.mkdirs in the test makes
the latter a little more comprehensible at first glance.

[Martin Panter]

https://bugs.python.org/review/23228/diff/17198/Lib/tarfile.py
File Lib/tarfile.py (right):

https://bugs.python.org/review/23228/diff/17198/Lib/tarfile.py#newcode2205
Lib/tarfile.py:2205: if getattr(e, "winerror", None) is None:
What about AttributeError and NotImplementedError, which never have a “winerror”
attribute? Perhaps you should add a test for extracting symlinks that runs in
situations where these exceptions occur.