Introduced in PR #20, they get tested via the 'py27-pure' tox environment.

build os x wheels and deploy to pypi

Conflicts:
	.travis.yml

Drop support for Python 2.6 and 3.2.

- 2.6 is long out-of-maintenance, and a security quagmire.

- 3.2 cannot be tested on Travis.

Make Python objects pickle like C; unpickling empty BTrees in Python works. Fixes #19

On it's own, this breaks on protocols >= 2 (including Python 3 by
default) because the pickled object's class is required to match the
reduce output. One fix would be to change the pickle format to use a
custom constructor function, but then we'd have different pickle formats
in C and Python.

Rather than do that, we instead create a __class__ property for the
python implementation. This bypasses the protocol limitations. The
__class__ property wasn't assignable in the C implementation, so the
compatibility risk should be very small.

All the special pickle machinery is only defined in Python if needed.

clean redundant blank lines and indent some lines

Ensure that we don't overlook errors in first PyObject_RichCompareBool call

Python 3.5 turns such cases into SystemErrors.

See: https://bugs.python.org/issue23571

Fixes #15.

Monte Carlo tests suck for reproducibility.

Human-readable __repr__ for pure-Python Bucket and Set objects.

Ensure that the changelog entry uses an imperative verb. You'd think I'd remember that by now. [skip ci]

Fix setting _p_changed when removing from small Python BTrees/TreeSets.

Pure Python Set objects were also not marking themselves changed on addition. This impacted things like zope.index.topic.filter

Make all the tests in the w_zodb Tox environment run.

Fix setting _p_changed for small pure-Python BTrees.