fix: resolve pickling and deepcopying bug with dynamically added traits (closes #2 and #16)

[satra]

thanks to a pointer from @rkern, this might actually resolve the bugs listed in #2 and #16. however somebody with knowledge of traits should see if making this instance related changes creates other problems.

i didn't know where to add tests but the following test would fail currently and would get addressed by the changes in the pr.

def test_hastraits_pickle_deepcopy():
    from cPickle import dumps, loads
    from copy import deepcopy
    a = HasTraits()
    a.add_trait('foo', Int)
    a.foo = 1
    assign_a = lambda : setattr(a, 'foo', 'a')
    yield assert_raises, Exception, assign_a
    pkld_a = dumps(a)
    unpkld_a = loads(pkld_a)
    assign_a_again = lambda : setattr(unpkld_a, 'foo', 'a')
    yield assert_raises, Exception, assign_a_again
    copied_a = deepcopy(a)
    assign_a_again = lambda : setattr(copied_a, 'foo', 'a')
    yield assert_raises, Exception, assign_a_again

[dpinte]

could you add your test to the test suite (e.g. in traits/tests/test_regression.py) ?

[satra]

@dpinte - i'm not sure this is a clean fix - somebody with real knowledge of traits should take a look. this actually didn't work fully for me - it broke other things that i didn't have time track down.

[satra]

@dpinte - this is ancient but this bug still exists, adding a test.

[mdickinson]

Thank you for this contribution! Your fix was slightly reworked in #373, which is now merged.

[satra]

@mdickinson - thank you very much.

[mdickinson]

Unfortunately, #373 had to be reverted (see the PR description for details). I'll re-open the relevant issues.