Added option to output to a file to workaround M$ OSes' inability to pipe the standard error stream.

Maybe the tests shouldn't use Unix-specific code then ;-)

Fixed a bug in lesson twelve. Changed the link in reportAction to correctly link back to the sightings page. This bug was reported by many people including Vinod Kurup.

The changes simply emulate the setup and teardown as done in the other
testing module for PageTemplates, testHTMLTests. I am not 100% sure if the fix
is the correct thing to do but it makes the tests run on Mac OS X and
does not affect them on Linux.

The main reason behind the failures is most likely some other test that runs
before which fiddles with the security policies but never cleans up
afterwards. I did some poking but could not find anything.

now, but I don't want to forget about it, so print a warning to
stderr.

If you think this is too gross, feel free to comment out the body of
the method (but still override it so the test won't fail -- I don't
want to disable the entire Full.py test suite just because of this).

moment, i.e. the undo of an object creation.  Return None for the data
in that case.

_TransactionsIterator: Add a _closed flag which gets set in the
close() method, and checked in __getitem__().  Iterating over a closed
iterator raises an IOError.

Should strop some DeprecationWarnings. I left the rfind at line 166 since it wasn't obvious what to do, maybe someone else can fix it?

due to some internal problems (probably because of a wrong jumptable).
The unittest for the splitter has been temporarily disabled on Mac OSX.

The base setUp() initializes _storage to None.  A subclass that
extends it must call the superclass method before initializing
_storage.

Also, mixin MTStorage, since this storage out to be thread-safe.

ConflictError as a ConflictError.  Replace "is" with "issubclass".

Make SessionDataManagerTraverser inherit from Persistent.  This is in order to avoid a situation in which a nonpersistent instance of this class is inserted into the before traverse hook, causing Zope to not be able to start if the class' containing module can't be imported.  The existence of the bug stemmed from my lack of knowledge about how the traverse hook worked (I didn't know it actually stored an object reference).

(merge from branch)

due to changes in the security machinery

but one of the uses of string.atoi wasn't fixed.  There also wasn't a
test for this. :(

Fixed bug in handling numneric time-zone offsets. We need to subtract,
rather than add the offset, since the offset is the offset of the
local time from UT, not the other way around.

still shouldn't import Zope in unit tests, but most of the existing CMF tests
depend on importing Zope and there was a hidden exception occurring anyway.
(OFS.Application.initialize() was failing.)