Bug in pretty printing

[doerwalter]

The code in IPython.lib.pretty.RepresentationPrinter.pretty has a bug: it fails to look in the class __dict__s of the classes in the MRO. Fixes this is pretty easy:

--- pretty.py.orig  2012-07-11 15:38:34.000000000 +0200
+++ pretty.py   2012-07-11 15:32:05.000000000 +0200
@@ -354,7 +354,7 @@
                         # Some objects automatically create any requested
                         # attribute. Try to ignore most of them by checking for
                         # callability.
-                        if '_repr_pretty_' in obj_class.__dict__:
+                        if '_repr_pretty_' in cls.__dict__:
                             meth = obj_class._repr_pretty_
                             if isinstance(meth, collections.Callable):
                                 return meth(obj, self, cycle)

[juliantaylor]

is it possible to add a somewhat more meaningful testcase for this?
Its not really clear to my why the behavior tested for in the testcase is the desired behavior compared to the pre fix situation.
Getting Dummy1() as representation for Dummy2 violates the rule (convention?) that representation should be unambigous.