[dev mar 06] If a bean has a BeanInfo giving an
IndexedPropertyDescriptor giving only an indexed getter and no other
accessors, BeanNode.computeProperties creates an
IndexedPropertySupport for the property, where valueType==null. This
is illegal according to the IndexedPropertySupport Javadoc, which does
not mention that a null param is OK. (Actually it does not mention
that *any* null params are OK, which is probably a docs bug. All
params which might be null must be documented as such.)
IndexedPropertySupport passes the null valueType to
Node.IndexedProperty, which again is not documented to accept it.
Node.IndexedProperty passes the null to Node.Property, which is also
not documented to accept it (nor is getValueType documented to return
it). And Node.Property.hashCode (as well as other methods) assume that
it is not null.

For an example, see issue #21284, where a poorly written BeanInfo for
the Sun-style XML catalog resolver (Runtime | XML Entity Catalogs |
Resolver at file:/whatever) produces such an
IndexedPropertyDescriptor. If you select a range of such resolvers in
the Explorer, the property sheet tries to merge their properties, thus
using a hashtable, thus calling Node.Property.hashCode, thus causing
an NPE. P2 because not only is an exception thrown, but almost
anything you do in the IDE will continue to throw more NPEs until you
exit.