Ok, so I've used Stef's Ginsu port, and with very little patching it ran
very nicely in 3.1.
Here's what I did:
1. implement a method AlignmentMorph>>inset: (just returns self)
2. changed UpdatingMenuItemMorph>>enablement: to ask the target and not
the wordingProvider
3. implemented ChooserMorph>>choose... with additional args
(buttons+values+lines), which are silently ignored
4. changed MethodDefinition>>classReferences and ...>>globalReferences
to use isVariableBinding instead of isKindOf: Association. BTW, does
anybody know why the various Associationlike subclasses of LookupKey are
not subclasses of Association?

Now I'm fooling around with the mechanism which moves definitions into
and out of the Unmanaged package. It seems to do the wrong thing for
clusters, because classes defined within different packages within the
same cluster can't see each other. Seems like at least one of
visibleClasses, allVisibleClasses, visibleClassNames and
allVisibleClassNames is broken :-( Without really understanding the
intended function of these messages, it looks to me that the overloaded
definitions of visibleClasses in Cluster and Package shadows the one
which seems most sensible to me, in Module...

Cheers,
Hans-Martin