A few things that could be improved in the specification of MethodHandle constant resolution:

- References to constant pool structures are inappropriate here. 5.4.1 already established that a "symbolic reference to a method handle" contains "live" representations of other constants, and we can just talk about those directly.

- The variable C, representing the referenced class, is not formally introduced

- The rule about REF_invokeSpecial constraining which referenced classes, C, are legal, belongs in the list given by step 2.

- When talking about classes, "subclass of" is better than the vague "assignable to".

- Step 3, resolution of classes given by the descriptor, is imprecise (some descriptor types are not class types) and unnecessary (step 4 is going to resolve a MethodType, which will have the same effect)