[MERGE] Add PreviewTree to tree_implementation tests

[Aaron Bentley]

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

Andrew Bennetts wrote:
> More fundamentally, I don't understand from your description why the current
> test_inv can't simply be replaced by your test_inv_alternatives — if the
> interface doesn't require an inventory attribute to exist on tree
> implementations, and we can do all the tests we want to do without one, why do
> we need seperate tests that use an inventory attribute?

We can't deprecate inventory attributes yet, but I was handling these
inventory tests the same way we handle tests of deprecated APIs.
Preserve them so that existing functionality doesn't change.  Add new
tests of new functionality.

Now admittedly, most of the code is the same.  But the test for symlink
targets is different, and that did cause me to tweak the WorkingTree
implementation.

So, I dunno.

I guess we can just test that Tree.inventory[] works, and just switch
the rest of the tests to use Tree.iter_entries_by_dir() instead.

But the new symlink test tests different interfaces, so I would keep that.

> I don't much like the duplication of test code here.  It's a big
maintenance
> problem.  At a minimum this sort of thing needs big, unmissable
comments in the
> docstrings of the two modules saying there's another module that MUST
be kept in
> sync.

I didn't intend for them to be kept in sync.  I intended for the
inventory tests to be preserved as is until such time as the .inventory
attribute was removed.

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

iD8DBQFIA2Wg0F+nu1YWqI0RAvKPAKCDK3a3YzLY02iYjKdXMbf52FWHqQCeI6Qj
EnZBx5PVrLG41w0t2dqxRmk=
=/Puv
-----END PGP SIGNATURE-----