Use tarfile instead of GNU tar

[dagss]

Currently GNU tar with the --strip-components option is needed, it would be better to use the builtin tarfile module.

There's prior art in "distlib" (in util.py) for supporting extraction of different extraction schemes:

https://bitbucket.org/vinay.sajip/distlib/src/393ccd9db71b2c77ed3973fc33bd5d28b090174b/distlib/util.py?at=default

Note that we support a "strip=1" feature so that one does not need to go "cd zlib-1.2.3" after extraction. To support this, the code would likely look something like this (this seems to be OK looking briefly at Python/Lib/tarfile.py):

def remove_prefix(tarinfo):
    tarinfo.path = pjoin(os.path.split(tarinfo.path)[strip:]) # but raise error if there are fewer components

tar = tarfile.open("sample.tar.gz")
tar.extractall(members=(remove_prefix(tarinfo) for tarinfo in tar))
tar.close()

Note: in test_source_cache.py, make_mock_archive/make_temporary_tarball should be modified to put items into a sub-directory to test the strip feature.

[dagss]

Also: GitSourceCache uses tar to extract output from "git archive", so instead one must read stdout and feed it to tarfile.

[certik]

This is not very high priority, as tar works well for now.