implied classes with empty doc strings "inherited" bogus doc string
from Interface class, and interface class attribute not being set
correctly during init.

parsing message headers

Thanks Evan.

the roles if it is called with a wrapped object.

Changed the role computing logic to always use aq_inner when getting
a parent to make sure it follows the containment context.

rfc1123_date(): Put back in the check for defaulted timestamp, so we
don't get exceptions in that case.  Fixes failure on writing ftp
documents, at least.

Looks like the 'format' parameter is unused and inappropriate, but i'm
leaving that for someone in the know to repair.  (They should pull out
my 'XXX' comment/note, at that time.)

they get reindexed.

* Add a docstring to the class which:

  o References the interfaces wiki where the design is spelt out;

  o Provides a usage sample.

new GC strategy.

Meaningless checkpoint of other files.

changeOwnership method of Owned mixin failed when in "non-recursive" mode when trying to set child._owner = child._owner.  Why it was doing this I have no idea.  I changed it to set child._owner = new.  The trunk version of this method is significantly different than the 2.2.X branch version in other ways, so I can't make an analogous change to the 2.2.X branch.

Method attributes can now only be set by setting special attributes in
classes.

store the 'connection string' as a record object.

ZClasses typically defer creating them until product
registration;  the derived ZClass needs them to be available
immediately after import.  Deprecated
'ProductContext.registerZClass' and
'ProductContext.registerBaseClass' in favor of a new function,
'ZClasses.createZClassForBase' (because none of the machinery
needed a ProductContext instance anyway).