Fixed in r5005. I didn't test this thoroughly, but it looks reasonably
straight-forward.

Yes, this is definitely legal (with or without ":equality").

I did think about cases like that a lot when doing the grounding, but apparently
I still messed up something.

Bug from Blai:

The translator fails when parsing the following very simple
domain and problem pddls:

 (define (domain test)
     (:requirements :equality)
     (:predicates (fluent ?x ?y))
     (:action solve
         :parameters (?x ?y)
         :precondition (fluent ?x ?x)
         :effect (fluent ?x ?y)
     )
 )

 (define (problem p1)
     (:domain test)
     (:objects obj1 obj2)
     (:init (fluent obj1 obj1))
     (:goal (fluent obj1 obj2))
 )

 The guilty line is the precondition where fluent receives ?x twice.
 But, this should be legal since :equality is turned on, right? Indeed,
 FF accepts it. Any suggestion or patch?