Reimplement basic_reduce()'s rules in ltl_simplifier.

So far I have only checked these rewritings with reduccmp.test.
There are probably a few kinks to iron out.

* src/ltlvisit/simplify.cc: Reimplement most of the basic
rewriting rules, leaving some FIXME comments for dubious ones.
* src/ltlast/multop.cc, src/ltlast/multop.hh: Ignore NULL
pointers in the vector.
* src/ltlvisit/reduce.cc (reduce): Do not call basic_reduce().
* src/ltltest/reduccmp.test: Adjust tests.

src/ltlast/multop.cc

@@ -163,7 +163,7 @@ namespace spot
 
     // We match equivalent formulae modulo "ACI rules"
     // (i.e. associativity, commutativity and idempotence of the
-    // operator).  For instance If `+' designate the OR operator and
+    // operator).  For instance if `+' designates the OR operator and
     // `0' is false (the neutral element for `+') , then `f+f+0' is
     // equivalent to `f'.
     formula*
@@ -178,6 +178,16 @@ namespace spot
 	vec::iterator i = v->begin();
 	while (i != v->end())
 	  {
+	    // Some simplification routines erase terms using null
+	    // pointers that we must ignore.
+	    if ((*i) == 0)
+	      {
+		// FIXME: We should replace the pointer by the
+		// first non-null value at the end of the array
+		// instead of calling erase.
+		i = v->erase(i);
+		continue;
+	      }
 	    if ((*i)->kind() == MultOp)
 	      {
 		multop* p = static_cast<multop*>(*i);
@@ -187,6 +197,7 @@ namespace spot
 		    for (unsigned n = 0; n < ps; ++n)
 		      inlined.push_back(p->nth(n)->clone());
 		    (*i)->destroy();
+		    // FIXME: Do not use erase.  See previous FIXME.
 		    i = v->erase(i);
 		    continue;
 		  }

src/ltlast/multop.hh

@@ -65,7 +65,8 @@ namespace spot
       /// formulae as argument.  This vector is acquired by the
       /// spot::ltl::multop class, the caller should allocate it with
       /// \c new, but not use it (especially not destroy it) after it
-      /// has been passed to spot::ltl::multop.
+      /// has been passed to spot::ltl::multop.  Inside the vector,
+      /// null pointers are ignored.
       ///
       /// All operators (Or, And, Concat) are associative, and are
       /// automatically inlined.  Or and And are commutative, so their

src/ltltest/reduccmp.test

@@ -100,8 +100,10 @@ for x in ../reduccmp ../reductaustr; do
      run 0 $x '(a U b) | (a U c)' 'a U (b | c)'
      run 0 $x '(a R b) | (c R b)' '(a | c) R b'
 
-     run 0 $x 'X(a & GFb)' 'Xa & GFb'
+     run 0 $x 'Xa & FGb' 'X(a & FGb)'
-     run 0 $x 'X(a | GFb)' 'Xa | GFb'
+     run 0 $x 'Xa | FGb' 'Xa | FGb' # We'd prefer 'X(a | FGb)'
+     run 0 $x 'Xa & GFb' 'Xa & GFb' #             'X(a & GFb)'
+     run 0 $x 'Xa | GFb' 'X(a | GFb)'
      # The following is not reduced to F(a) & GFb.  because
      # (1) is does not help the translate the formula into a
      #     smaller automaton, and ...
@@ -109,10 +111,9 @@ for x in ../reduccmp ../reductaustr; do
      # (2) ... it would hinder this useful reduction (that helps to
      #     produce a smaller automaton)
      run 0 $x 'F(f1 & GF(f2)) | F(a & GF(b))' 'F((f1&GFf2)|(a&GFb))'
+     # FIXME: Don't we want the opposite rewriting
+     run 0 $x 'Fa & GFb' 'Fa & GFb' # We'd prefer 'F(a & GFb)'
      run 0 $x 'G(a | GFb)' 'Ga | GFb'
-
-     run 0 $x 'X(a & GFb & c)' 'X(a & c) & GFb'
-     run 0 $x 'X(a | GFb | c)' 'X(a | c) | GFb'
      # The following is not reduced to F(a & c) & GF(b) for the same
      # reason as above.
      run 0 $x 'F(a & GFb & c)' 'F(a & GFb & c)'

src/ltlvisit/reduce.cc

@@ -22,7 +22,6 @@
 // 02111-1307, USA.
 
 #include "reduce.hh"
-#include "basicreduce.hh"
 #include "ltlast/allnodes.hh"
 #include <cassert>
 
@@ -74,13 +73,6 @@ namespace spot
 	  f2->destroy();
 	  f2 = f1;
 
-	  if (opt & Reduce_Basics)
-	    {
-	      f1 = basic_reduce(f2);
-	      f2->destroy();
-	      f2 = f1;
-	    }
-
 	  f1 = simplifier.simplify(f2);
 	  f2->destroy();
 	  f2 = f1;