0.4

  Change the contract of 'handle_exception'.  As before, if this
  method exists on a helper, it is called by the publisher when any
  other helper step raises an exception.  However, unlike its contract
  in 0.3, it *must* either raise an exception *or* return a
  three-tuple representing the (status, headers, and body_iter) that
  will be used to send to our WSGI upstream caller.  In particular, it
  is no longer legal to return None from 'handle_exception'.

  'handle_exception' is now a required method on helpers.

0.3

  Allow helpers to have a 'handle_exception' method.  If this method
  exists on a helper, it is called by the publisher when any other
  helper step raises an exception.   It is called with a single
  argument, which will be a three-tuple, representing the type, value,
  and traceback of the exception that occurred (the result of
  sys.exc_info()).  The 'handle_exception' method should either
  reraise the exception wholesale (e.g. 'raise t, v, tb') or raise a
  different exception.  The return value of 'handle_exception' is
  ignored.  If the helper's 'handle_exception' returns without
  raising, or the helper has no 'handle_exception' method, the
  original exception is re-raised by obob.

0.2


  The traversal algorithm implemented in ObobPublisher.__call__ used
  to be:::

     current = root
     for name in helper.path_elements():
         helper.before_traverse(current, name)
         current = helper.traverse(current, name)

  This presumed that the helper knows about all the path elements
  before traversing.  This is sometimes not the case.  To account for
  this, we now assume the helper provides a method named 'next_name'
  in order to allow the helper to compose path elements both before
  and during traversal.  The resulting traversal algorithm is::

     current = root
     while 1:
         helper.before_traverse(current)
         name = helper.next_name()
         if name is None:
             break
         current = helper.traverse(current, name)

  Note that the helper 'before_traverse' method is no longer passed
  the name (because we don't know any name yet).

  The change to the algorithm is primarily to allow hacks like
  CMFDynamicType's __before_publishing_traverse__, which modifies the
  traversal request name stack to find "alias" names.  This fixes
  breakages evident when publishing Plone through repoze.zope2.

0.1

  Initial release.