[MERGE] Handle references to line data in _patiencediff_c.c properly

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Lukáš Lalinský wrote:
> This code is called in bzrlib only with lists or tuples of strings,
> which works fine, because PySequence_Fast always returns the same object
> and so the items from PySequence_Fast_GET_ITEM will not be deleted until
> the original sequence is deleted. But if it's called with a sequence
> than needs to be converted to list (e.g. PyUnicode), the items will be
> garbage collected in the contructor and still used later in code (which
> leads to segfaults).
> 
> I really don't how to write a failing test for something like this, so
> there are no tests in the patch. :(
> 
> Lukas

So... just write a test that we can do PatienceDiff on a Unicode sequence.

matcher = PatienceSequenceMatcher(u'Ab\xb5\u1234', u'....')
matcher.get_matching_blocks()

However, I'm also noticing that you switched "free(b)' => 'delete_lines(b,
bsize)' but that delete_lines isn't freeing the b pointer.

BB:tweak

If it were up to me, I would just have hung onto the PySequenceFast object for
the lifetime of the 'lines' variables, but both are valid. (Just a lot fewer
INCREF/DECREF to keep track of.)

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFIqIAYJdeBCYSNAAMRAp1aAJ48j7rVEccZbJVoY89pT/BGZ0N8iACgzHt7
/iVLxKQct9cbKJmakbG7yGU=
=lm5j
-----END PGP SIGNATURE-----