Add a limited mechanism to capture warnings instead of having them
printed to stderr.

Tests work in my local ZODB sandbox.  Need to check whether any Zope 2
or Zope 3 tests call the deprecated methods, which could affect the
test results.

Add a few comments.

Move some of the test framework code from .py to .txt, and add a more
detailed comment at the top about the contract between .txt and the
.py file that uses it.

Move the interface test into .txt.  I guess it's not so bad.

Change some tests that displayed the contents of __dict__.  Other
persistent implementations may have extra state in __dict__.

Remove B, it is unused.

Change DocFileSuite() to accept an explicit set of globals.

Remove the pickling/P2 test, because it was too hard to make work with
the framework.  Include comment explaining how such a test should be
added.

by src/persistent/tests/test_persistent.

I've included some code to convert a text file containing doctest
tests into a unittest.TestCase.  This code wants to be generic, but
I'm not sure if it's simple enough yet.

This shouldn't affect any user code, because False == 0 and True == 1.

Remove unittest boilerplate.

Add docstrings for public cache methods.

If there is a jar and no oid, then the user assigned _p_jar directly.
All the Connection that assigns _p_jar also assigns an oid.

Delete importFile() and exportFile().  They always raised an
    exception, and there are no subclasses that provide an
    implementation.  I suspect they were decoys from the days
    before export/import was implemented on the Connection.
Remove UndoLogCompatible base class from DB, because it doesn't
    provide the undoLog() method needed to make it at all useful.
Deprecate all the cache_deactivate_after methods.
Replace list2dict with new dict.fromkeys() constructor.
Make imports global instead of local.
Add epydoc markup to DB class docstring.

It's harder to write serious tests, because some of the critical
correctness issues relate to concurrency.  We'll have to depend on the
various concurrent updates and NZODBThreads tests to handle these.

Import IPersistent so docstrings can refer to it.
Move $Id$ tag about @group tags so epydoc doesn't complain.
Don't return anything from _setDB().  The caller always has the object
    anyway.
Rename local variable in register() from object to obj to avoid
    shadowing builtin name.

Please do not only run the unit tests after changes; also test whether
Z3 starts up...

Add get() as preferred synonym for __getitem__.  Change several uses
    of subscripting to get().
Change add() to register new object with transaction.
Change several methods to raise RuntimeError if they are called when
    the Connection is closed.
Add experimental epydoc markup to doc strings.
Extend doc strings and write new tests for "User Methods."

However, fixing it so far *appears* to fix a bunch of shy FileStorage
pack[Now]WhileWriting test failures whose connection to loadBefore()
isn't obvious.  It's plausible that it's a real fix for those, just
not (yet) compelling.  It seems very likely to be a real fix for very
rare failures that have shown up only in ChrisM's overnight testrunner
reports (because those did have stuff related to loadBefore() in their
tracebacks).