diff --git a/ChangeLog b/ChangeLog
index 7dc7d4dab..ab420218e 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,9 @@
+2005-06-06  Alexandre Duret-Lutz  <adl@src.lip6.fr>
+
+	* src/tgbaalgos/reductgba_sim_del.cc
+	(parity_game_graph_delayed::get_relation): Disable for generalized
+	automata, it's wrong.
+
 2005-05-25  Alexandre Duret-Lutz  <adl@src.lip6.fr>
 
 	* src/tgbaalgos/reductgba_sim_del.cc
diff --git a/src/tgbaalgos/reductgba_sim.hh b/src/tgbaalgos/reductgba_sim.hh
index ce166a2e0..bb33e70aa 100644
--- a/src/tgbaalgos/reductgba_sim.hh
+++ b/src/tgbaalgos/reductgba_sim.hh
@@ -69,8 +69,7 @@ namespace spot
 							     int opt = -1);
 
   /// Compute a delayed simulation relation on state of tgba \a f.
-  // FIXME: This method is correct but it builds sometime (when there are more
-  // than one acceptance condition) only a part of the simulation relation.
+  /// \bug Does not work for generalized automata.
   delayed_simulation_relation* get_delayed_relation_simulation(const tgba* a,
 							       std::ostream& os,
 							       int opt = -1);
diff --git a/src/tgbaalgos/reductgba_sim_del.cc b/src/tgbaalgos/reductgba_sim_del.cc
index 3890ca387..88e03e489 100644
--- a/src/tgbaalgos/reductgba_sim_del.cc
+++ b/src/tgbaalgos/reductgba_sim_del.cc
@@ -28,6 +28,7 @@ namespace spot
   /// The one priority is represent by a \a acceptance_condition_visited_
   /// which differ of bddfalse.
   /// This spoiler node are looser for the duplicator.
+  /// FIXME: get rid of these ugly globals
   static int nb_spoiler_loose_;
 
   static int nb_spoiler;
@@ -549,48 +550,7 @@ namespace spot
   void
   parity_game_graph_delayed::lift()
   {
-    // Before the lift we compute each vertices
-    // to know if he belong to a all accepting cycle
-    // of the graph.
-
-    if (this->nb_set_acc_cond() > 1)
-      {
-	for (std::vector<duplicator_node*>::iterator i
-	       = duplicator_vertice_.begin();
-	     i != duplicator_vertice_.end(); ++i)
-	  {
-	    /*
-	    for (std::vector<duplicator_node*>::iterator i2
-		   = duplicator_vertice_.begin();
-		 i2 != duplicator_vertice_.end(); ++i2)
-	      dynamic_cast<duplicator_node_delayed*>(*i2)->seen_ = false;
-	    for (std::vector<spoiler_node*>::iterator i3
-		   = spoiler_vertice_.begin();
-		 i3 != spoiler_vertice_.end(); ++i3)
-	      dynamic_cast<spoiler_node_delayed*>(*i3)->seen_ = false;
-	    */
-	    dynamic_cast<duplicator_node_delayed*>(*i)->set_lead_2_acc_all();
-	  }
-	for (std::vector<spoiler_node*>::iterator i
-	       = spoiler_vertice_.begin();
-	     i != spoiler_vertice_.end(); ++i)
-	  {
-	    /*
-	    for (std::vector<duplicator_node*>::iterator i2
-		   = duplicator_vertice_.begin();
-		 i2 != duplicator_vertice_.end(); ++i2)
-	      dynamic_cast<duplicator_node_delayed*>(*i2)->seen_ = false;
-	    for (std::vector<spoiler_node*>::iterator i3
-		   = spoiler_vertice_.begin();
-		 i3 != spoiler_vertice_.end(); ++i3)
-	      dynamic_cast<spoiler_node_delayed*>(*i3)->seen_ = false;
-	    */
-	    dynamic_cast<spoiler_node_delayed*>(*i)->set_lead_2_acc_all();
-	  }
-      }
-
     // Jurdzinski's algorithm
-    //int iter = 0;
     bool change = true;
 
     while (change)
@@ -620,10 +580,11 @@ namespace spot
     state_couple* p = 0;
     seen_map::iterator j;
 
-    /*
-      if (this->nb_set_acc_cond() > 1)
+    // This does not work for generalized automata.  A tarjan-like
+    // algorithm is required to tell whether a state can lead to an
+    // acceptance cycle.
+    if (this->nb_set_acc_cond() > 1)
       return rel;
-    */
 
     for (std::vector<spoiler_node*>::iterator i
 	   = spoiler_vertice_.begin();