for both full and minimal bdb storages.

added while a pack was stuck between a mark and a sweep, the mark
would not have found the new objects as root reachable (because they
wouldn't have been stored yet), but the sweep will find then and
erroneously delete them.

_setupDBs(): Add a _packing flag which gets set whenever we're doing a
classic or auto pack

_docommit(): When the _packing flag is set, copy all oids in the _oids
table to the _packmark table, so any objects added between the mark
and sweep phases will not be gc'd.  This is fine because if they're
still not root reachable by the next gc pass, they'll get collected
then.

pack(): Set and reset the _packing flag around the calls to _dopack().

Also, get ZERO from the package, and add the `info' table for storage
metadata.

_docommit(): Fix a reference counting bug caused by not incref'ing the
objects referred to by the new pickle of a new object revision.

added while a pack was stuck between a mark and a sweep, the mark
would not have found the new objects as root reachable (because they
wouldn't have been stored yet), but the sweep will find then and
erroneously delete them.

_setupDBs(): Add a _packing flag which gets set whenever we're doing a
classic or auto pack

_docommit(): When the _packing flag is set, copy all oids in the _oids
table to the _packmark table, so any objects added between the mark
and sweep phases will not be gc'd.  This is fine because if they're
still not root reachable by the next gc pass, they'll get collected
then.

pack(), autopack(): Set and reset the _packing flag around the calls
to _dopack().

Small refactoring and bug caught in computation of "plast" which had the potential to cause slowness.

- make it easier for an admin to configure logging levels
  (by making the defaults more sane)
- refactor the tests a little, in preparation for more tests

- remove methods that aren't needed
- rename "resolved" to "instance" for readability

Remove __version__

Remove __version__

ZODB4 will have two new methods in the storage api:

    get_version() -> string
    set_version(string)

Prepare for this now by changing the packtime table into an info
table, which is just a key/value mapping for meta-information about
the storage.  For now, the only key we define is `packtime' although
we're reserving `version' for ZODB4.

when attempting to get a list of users, use the __allow_groups__ alias.
This doesn't break anything as it is one of the (unwritten) contracts
of a user folder to install itself as __allow_groups__ in its container
(it is canonized in the manage_beforeDelete and manage_afterAdd methods
of BasicUserFolder, and the traversal machinery refers to a user folder
by its __allow_groups__ alias exclusively).

It's actually likely that all code in Zope's security machinery that directly
refers to a user folder by its 'acl_users' alias in this way is wrong.

Referring to the user folder by its __allow_groups__ alias in the security
machinery exclusively gives us the ability to insert a "proxy" object as
can act as an intermediary for a "real" UserFolder object, allowing us to
massage the output from its interface methods as necessary to allow for
grouping and other tricks.

Some indexes were storing acquisition wrappers in the cyclic references
back to the ZCatalog instance.  Removed the acquisition wrappers and
cleaned up a bad use of apply().  The acquisition wrappers were revealed
by AdaptableStorage.

There's a second problem, in that indexes are unknowingly storing references
back to the ZCatalog in an unused attribute.  In the 2_6 branch it's
hard to fix without breaking things, but on the HEAD there's a better way
to fix this.  Coming soon, I hope.

run from Daemon.

of a catalog metadata was identical with the name of an acquired
object.
~

#740: _tzoffset handled positive numerical offsets incorrectly. This has been fixed. A large set of tests for different cases has been added, and for the purpose of those tests DateTime has a new feature: A tzoffset() method.

- add at least something of a test for the datatype

Merge with HEAD: Collector #763: There was no error when you had a sendmail-tag without specifying a mailhost or smpthost. Also added a missing import.

result in an Unauthorized error rather than AttributeError.  Added a test
to ensure the bug stays fixed.

Also, set the FullOpenCloseTest's level to 2 so
testCloseWithCheckpointingThread() isn't normally run, even if you
have BerkeleyDB installed.  This is the single longest test in the
--all suite.  This shaves about 21 seconds off the tests.

Also, set the suite's test level to 2 so these aren't normally run
(even if you have BerkeleyDB installed).  These tests take a long
time.

  contained 'None', which lead to 'None' as connection hook.

no other handlers are provided.  This is needed to avoid warnings on stderr
from the logging package.

Needs more work, but can deal with at least logging to files and STDERR.
Log rotation has been tested.

- add a way to control how logging is re-initialized

- Cache the translations from severity to level and string in dicts,
  so these are normally done with a single dict.get() call rather than
  a Python function call.

- Extract the PEP 282 level first, and check it against the effective
  level of the logger, so we can take an early exit if the logger
  won't handle the event.

- Got rid of textwrap.  It slows things down, can frustrate grepping
  programs, and wasn't effective unless you were using Python 2.3
  anyway.

- Got rid of the addHandler(NullHandler()) (seems like superstition).

- Got rid of the log_format class variable (YAGNI).

trunk.

       why security assertions should be placed

       PASSWORD() function add to lib/python/AccessControl/AuthEncoding.py.