More transitive-property support. - Implemented transitive properties in sort and group specs. Sort/group specs can now be lists of specs. - All regression tests except for one metakit backend test related to metakit having no representation of NULL pass - Fixed more PEP 8 whitespace peeves (and probably introduced some new ones :-) - Moved Proptree from support.py to hyperdb.py due to circular import - Moved some proptree-specific methods from Class to Proptree - Added a test for sorting by ids -> should be numeric sort (which now really works for all backends) - Added "required" attribute to all property classes in hyperdb (e.g., String, Link,...), see Feature Requests [SF#539081] -> factored common stuff to _Type. Note that I also converted to a new-style class when I was at it. Bad: The repr changes for new-style classes which made some SQL backends break (!) because the repr of Multilink is used in the schema storage. Fixed the repr to be independent of the class type. - Added get_required_props to Class. Todo: should also automagically make the key property required... - Add a sort_repr method to property classes. This defines the sort-order. Individual backends may use diffent routines if the outcome is the same. This one has a special case for id properties to make the sorting numeric. Using these methods isn't mandatory in backends as long as the sort-order is correct. - Multilink sorting takes orderprop into account. It used to sort by ids. You can restore the old behaviour by specifying id as the orderprop of the Multilink if you really need that. - If somebody specified a Link or Multilink as orderprop, we sort by labelprop of that class -- not transitively by orderprop. I've resited the tempation to implement recursive orderprop here: There could even be loops if several classes specify a Link or Multilink as the orderprop... - Fixed a bug in Metakit-Backend: When sorting by Links, the backend would do a natural join to the Link class. It would rename the "id" attribute before joining but *not* all the other attributes of the joined class. So in one test-case we had a name-clash with priority.name and status.name when sorting *and* grouping by these attributes. Depending on the order of joining this would produce a name-clash with broken sort-results (and broken display if the original class has an attribute that clashes). I'm now doing the sorting of Links in the generic filter method for the metakit backend. I've left the dead code in the metakit-backend since correctly implementing this in the backend will probably be more efficient. - updated doc/design.html with the new docstring of filter.