Here's how it works:

- On every store(), we write an entry to a objrev table containing the
  tuple of information (newserial, oid, oldserial).  We don't write
  this entry if the store is the first revision of an object on a new
  version.

  We do basically the same thing on restore() and transactionalUndo().

- On an abortVersion(), we write two entries to the objrev table, one
  that has (newserial, oid, oldserial) -- which points to the old
  serial in the version, and (newserial, oid, nvserial) -- which
  points to the non-version revision of the version revision.

- On commitVersion(), we do the same as abortVersion() except that we
  don't write the non-version data if we're committing to a different
  version.

- Now, when we pack, all we need to do is cruise from the beginning of
  the objrev table until we find an entry with a newserial > packtime.
  If the oldserial is ZERO, it's an object creation event which we
  don't need to worry about because there's no previous revision.  But
  otherwise, we can delete the oid+oldserial revision because we know
  it's not current.  We do this, updating pickle refcounts and then
  collecting any objects that are left unreferenced.

The cute thing is that autopacking will use the same algorithm.  The
main difference between autopack and classic pack, is that the latter
does a mark and sweep garbage collection phase after the normal objrev
collection phase.  Also, this algorithm means autopack needs only
three pieces of information:

- How often the thread should run (e.g. once per hour)

- How far in the past it should pack (e.g. pack to 4 hours ago).  We
  don't need a start time for the autopack window, because we'll
  always just start at the beginning of the objrev table.

- How often should autopack also do a classic pack (e.g. do a classic
  pack once per day).

Autopack isn't implemented in this checkin, but I believe it will be
nearly trivial to add.  That comes next.

checkTransactionalUndoAfterPackWithObjectUnlinkFromRoot()

for ZODB 3.2.

(__init__.py still makes format_exception available; is this needed?)

must've used this to disable the exception handling temporarily to
debug the exception formatting machinery, and forgotten to re-enable
it before checking in.

Unfortunately, this made it into the ZODB 3.1 release, making this a
3.1.1 backport candidate.

Fixes #651

Remade the test code for rfc822() to provide more stable tests.

Make TOC items and values sloppier.  These are utility methods used for custom introspection on the TOC.  Under concurrent use, they tended to cause a KeyError
to be raised out of __getitem__ due to a synchronization problem.  Making them more tolerant of desynchronization between _items and _data makes them useful.

Delayed Zope startup based on changes from Zope-2_7-development-branch
(with corrections), with the intent of merging this into the 2_6
branch and the head.

Until now, "import Zope" always opened the database automatically.
Unfortunately, that strategy caused the Python import lock to be held
by the main thread during database initialization, which led to a
deadlock if other threads required something to be imported before
completing initialization.  This can be a big problem for ZEO.

The only foreseen risk is that external scripts which "import Zope"
may depend in some way on the database being opened immediately.  Most
scripts just use "Zope.app()", which still works well.  If you have a
script that breaks with these changes, either add a call to the new
Zope.startup() function or set the ZOPE_COMPATIBLE_STARTUP environment
variable to a non-empty value.

- fixed broken except statement

storage speak, lrevid pointers to shared pickle data in earlier
transactions.

Specifically,

The _Record object now has a data_txn attribute that is either None or
the id of the transaction that contains the data used by the current
record.  Example: When transactionalUndo() modifies an object, it
typical creates a new data record that points at the transaction
before the undo.  The new record contains the same logical data as the
record it refers to.  (For consistency purposes, this is a stronger
claim than that the pickles in two different data records are the
same.)

_Record.__init__(): Gets data_txn passed in from
_RecordIterator.__getitem__(), which in turn getes the lrevid from
Full._loadSerialEx().

_loadSerialEx(): Now returns the backpointer, aka lrevid but only if
it is not the same as the serial argument.  They will always be the
same (and we'll always return None) except in the face of
transactionalUndo, commitVersion, and abortVersion.

input field (collector #117).

- determine the directory containing the supplemental data file in the
  normal way
- don't use assert statements; use the appropriate self.assert*()
  methods consistently
- organize imports in Python Normal Form
- whitespace cleanup

getId.  This is a change designed to make it possible to disambiguate user names and user ids in subclasses of user folders, while still doing the "right thing" with respect to local data structures that keep pointers to user ids (eg. local roles, etc.)

Zope tree every time the DAV Lock Manager control panel is visited, it
gives the user a chance to specify a path to start from, shortening
query times (especially in large databases).

PageTemplateFiles were previously owned by whatever object contained
them.  This resulted in very hard bugs if the user who owned the
container was removed.  Since PageTemplateFiles come from the
filesystem, they are now "unowned", similar to DTMLFiles.  Security is
still applied, but now it is applied correctly.

Ahem, *cough*: "Please run the test suite before checking in a change! At
the very least run the tests for the package you changed."

Date is a required header in RFC822 and RFC2822, so MailHost now adds it if it doesn't exist.