ltl2tgba_fm.cc 23.4 KB
Newer Older
1
// Copyright (C) 2003, 2004  Laboratoire d'Informatique de Paris 6 (LIP6),
Alexandre Duret-Lutz's avatar
Alexandre Duret-Lutz committed
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
// dpartement Systmes Rpartis Coopratifs (SRC), Universit Pierre
// et Marie Curie.
//
// This file is part of Spot, a model checking library.
//
// Spot is free software; you can redistribute it and/or modify it
// under the terms of the GNU General Public License as published by
// the Free Software Foundation; either version 2 of the License, or
// (at your option) any later version.
//
// Spot is distributed in the hope that it will be useful, but WITHOUT
// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
// or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
// License for more details.
//
// You should have received a copy of the GNU General Public License
// along with Spot; see the file COPYING.  If not, write to the Free
// Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
// 02111-1307, USA.

22
#include "misc/hash.hh"
23
#include "misc/bddalloc.hh"
24
#include "misc/bddlt.hh"
25
#include "misc/minato.hh"
26
27
28
29
30
31
#include "ltlast/visitor.hh"
#include "ltlast/allnodes.hh"
#include "ltlvisit/lunabbrev.hh"
#include "ltlvisit/nenoform.hh"
#include "ltlvisit/destroy.hh"
#include "ltlvisit/tostring.hh"
32
#include "ltlvisit/postfix.hh"
33
#include "ltlvisit/apcollect.hh"
34
#include <cassert>
35
#include <memory>
36
37
38
39
40
41
42
43
44
45
#include "tgba/tgbabddconcretefactory.hh"
#include "ltl2tgba_fm.hh"

namespace spot
{
  using namespace ltl;

  namespace
  {

46
47
    // Helper dictionary.  We represent formulae using BDDs to
    // simplify them, and then translate BDDs back into formulae.
48
49
50
51
52
    //
    // The name of the variables are inspired from Couvreur's FM paper.
    //   "a" variables are promises (written "a" in the paper)
    //   "next" variables are X's operands (the "r_X" variables from the paper)
    //   "var" variables are atomic propositions.
53
    class translate_dict
54
55
56
    {
    public:

57
58
      translate_dict(bdd_dict* dict)
	: dict(dict),
59
60
61
62
63
64
65
66
67
68
	  a_set(bddtrue),
	  var_set(bddtrue),
	  next_set(bddtrue)
      {
      }

      ~translate_dict()
      {
	fv_map::iterator i;
	for (i = next_map.begin(); i != next_map.end(); ++i)
69
	  destroy(i->first);
70
	dict->unregister_all_my_variables(this);
71
72
      }

73
74
      bdd_dict* dict;

75
      /// Formula-to-BDD-variable maps.
76
77
      typedef Sgi::hash_map<const formula*, int,
			    ptr_hash<formula> > fv_map;
78
      /// BDD-variable-to-formula maps.
79
      typedef Sgi::hash_map<int, const formula*> vf_map;
80
81
82
83
84
85
86
87
88

      fv_map next_map;	       ///< Maps "Next" variables to BDD variables
      vf_map next_formula_map; ///< Maps BDD variables to "Next" variables

      bdd a_set;
      bdd var_set;
      bdd next_set;

      int
89
      register_proposition(const formula* f)
90
      {
91
	int num = dict->register_proposition(f, this);
92
93
94
95
96
	var_set &= bdd_ithvar(num);
	return num;
      }

      int
97
      register_a_variable(const formula* f)
98
      {
99
	int num = dict->register_acceptance_variable(f, this);
100
101
102
103
104
	a_set &= bdd_ithvar(num);
	return num;
      }

      int
105
      register_next_variable(const formula* f)
106
107
108
109
110
111
112
113
114
115
116
      {
	int num;
	// Do not build a Next variable that already exists.
	fv_map::iterator sii = next_map.find(f);
	if (sii != next_map.end())
	  {
	    num = sii->second;
	  }
	else
	  {
	    f = clone(f);
117
	    num = dict->register_anonymous_variables(1, this);
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
	    next_map[f] = num;
	    next_formula_map[num] = f;
	  }
	next_set &= bdd_ithvar(num);
	return num;
      }

      std::ostream&
      dump(std::ostream& os) const
      {
	fv_map::const_iterator fi;
	os << "Next Variables:" << std::endl;
	for (fi = next_map.begin(); fi != next_map.end(); ++fi)
	{
	  os << "  " << fi->second << ": Next[";
	  to_string(fi->first, os) << "]" << std::endl;
	}
135
136
	os << "Shared Dict:" << std::endl;
	dict->dump(os);
137
138
139
	return os;
      }

140
      formula*
141
142
143
144
      var_to_formula(int var) const
      {
	vf_map::const_iterator isi = next_formula_map.find(var);
	if (isi != next_formula_map.end())
145
	  return clone(isi->second);
146
147
	isi = dict->acc_formula_map.find(var);
	if (isi != dict->acc_formula_map.end())
148
	  return clone(isi->second);
149
150
	isi = dict->var_formula_map.find(var);
	if (isi != dict->var_formula_map.end())
151
	  return clone(isi->second);
152
	assert(0);
153
154
155
	// Never reached, but some GCC versions complain about
	// a missing return otherwise.
	return 0;
156
157
      }

158
      formula*
159
160
161
      conj_bdd_to_formula(bdd b)
      {
	if (b == bddfalse)
162
163
	  return constant::false_instance();
	multop::vec* v = new multop::vec;
164
165
166
	while (b != bddtrue)
	  {
	    int var = bdd_var(b);
167
	    formula* res = var_to_formula(var);
168
169
170
	    bdd high = bdd_high(b);
	    if (high == bddfalse)
	      {
171
		res = unop::instance(unop::Not, res);
172
173
174
175
		b = bdd_low(b);
	      }
	    else
	      {
176
		assert(bdd_low(b) == bddfalse);
177
178
179
180
181
		b = high;
	      }
	    assert(b != bddfalse);
	    v->push_back(res);
	  }
182
	return multop::instance(multop::And, v);
183
184
      }

185
186
      const formula*
      bdd_to_formula(bdd f)
187
      {
188
	if (f == bddfalse)
189
	  return constant::false_instance();
190

191
192
193
194
195
196
197
198
199
	multop::vec* v = new multop::vec;

	minato_isop isop(f);
	bdd cube;
	while ((cube = isop.next()) != bddfalse)
	  v->push_back(conj_bdd_to_formula(cube));

	return multop::instance(multop::Or, v);
      }
200
201
202
203
204
205
206
207
208
209
210

      void
      conj_bdd_to_acc(tgba_explicit* a, bdd b, tgba_explicit::transition* t)
      {
	assert(b != bddfalse);
	while (b != bddtrue)
	  {
	    int var = bdd_var(b);
	    bdd high = bdd_high(b);
	    if (high == bddfalse)
	      {
211
		// Simply ignore negated acceptance variables.
212
213
214
215
		b = bdd_low(b);
	      }
	    else
	      {
216
		formula* ac = var_to_formula(var);
217

218
		if (!a->has_acceptance_condition(ac))
219
220
		  a->declare_acceptance_condition(clone(ac));
		a->add_acceptance_condition(t, ac);
221
222
223
224
225
226
227
228
		b = high;
	      }
	    assert(b != bddfalse);
	  }
      }
    };


229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272

    // Gather all promises of a formula.  These are the
    // right-hand sides of U or F operators.
    class ltl_promise_visitor: public postfix_visitor
    {
    public:
      ltl_promise_visitor(translate_dict& dict)
	: dict_(dict), res_(bddtrue)
      {
      }

      virtual
      ~ltl_promise_visitor()
      {
      }

      bdd
      result() const
      {
	return res_;
      }

      using postfix_visitor::doit;

      virtual void
      doit(unop* node)
      {
	if (node->op() == unop::F)
	  res_ &= bdd_ithvar(dict_.register_a_variable(node->child()));
      }

      virtual void
      doit(binop* node)
      {
	if (node->op() == binop::U)
	  res_ &= bdd_ithvar(dict_.register_a_variable(node->second()));
      }

    private:
      translate_dict& dict_;
      bdd res_;
    };


273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
    // The rewrite rules used here are adapted from Jean-Michel
    // Couvreur's FM paper.
    class ltl_trad_visitor: public const_visitor
    {
    public:
      ltl_trad_visitor(translate_dict& dict)
	: dict_(dict)
      {
      }

      virtual
      ~ltl_trad_visitor()
      {
      }

288
289
      bdd
      result() const
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
      {
	return res_;
      }

      void
      visit(const atomic_prop* node)
      {
	res_ = bdd_ithvar(dict_.register_proposition(node));
      }

      void
      visit(const constant* node)
      {
	switch (node->val())
	  {
	  case constant::True:
	    res_ = bddtrue;
	    return;
	  case constant::False:
	    res_ = bddfalse;
	    return;
	  }
	/* Unreachable code.  */
	assert(0);
      }

      void
      visit(const unop* node)
      {
	switch (node->op())
	  {
	  case unop::F:
	    {
	      // r(Fy) = r(y) + a(y)r(XFy)
324
325
326
	      const formula* child = node->child();
	      bdd y = recurse(child);
	      int a = dict_.register_a_variable(child);
327
328
329
330
331
332
	      int x = dict_.register_next_variable(node);
	      res_ = y | (bdd_ithvar(a) & bdd_ithvar(x));
	      return;
	    }
	  case unop::G:
	    {
333
334
335
336
337
338
339
340
341
342
343
344
	      // The paper suggests that we optimize GFy
	      // as
	      //   r(GFy) = (r(y) + a(y))r(XGFy)
	      // instead of
	      //   r(GFy) = (r(y) + a(y)r(XFy)).r(XGFy)
	      // but this is just a particular case
	      // of the "merge all states with the same
	      // symbolic rewriting" optimization we do later.
	      // (r(Fy).r(GFy) and r(GFy) have the same symbolic
	      // rewriting.)  Let's keep things simple here.

	      // r(Gy) = r(y)r(XGy)
345
	      const formula* child = node->child();
346
	      int x = dict_.register_next_variable(node);
347
348
	      bdd y = recurse(child);
	      res_ = y & bdd_ithvar(x);
349
350
351
352
	      return;
	    }
	  case unop::Not:
	    {
353
	      // r(!y) = !r(y)
354
355
356
357
358
	      res_ = bdd_not(recurse(node->child()));
	      return;
	    }
	  case unop::X:
	    {
359
	      // r(Xy) = Next[y]
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
	      int x = dict_.register_next_variable(node->child());
	      res_ = bdd_ithvar(x);
	      return;
	    }
	  }
	/* Unreachable code.  */
	assert(0);
      }

      void
      visit(const binop* node)
      {
	bdd f1 = recurse(node->first());
	bdd f2 = recurse(node->second());

	switch (node->op())
	  {
377
	    // r(f1 logical-op f2) = r(f1) logical-op r(f2)
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
	  case binop::Xor:
	    res_ = bdd_apply(f1, f2, bddop_xor);
	    return;
	  case binop::Implies:
	    res_ = bdd_apply(f1, f2, bddop_imp);
	    return;
	  case binop::Equiv:
	    res_ = bdd_apply(f1, f2, bddop_biimp);
	    return;
	  case binop::U:
	    {
	      // r(f1 U f2) = r(f2) + a(f2)r(f1)r(X(f1 U f2))
	      int a = dict_.register_a_variable(node->second());
	      int x = dict_.register_next_variable(node);
	      res_ = f2 | (bdd_ithvar(a) & f1 & bdd_ithvar(x));
	      return;
	    }
	  case binop::R:
	    {
	      // r(f1 R f2) = r(f1)r(f2) + r(f2)r(X(f1 U f2))
	      int x = dict_.register_next_variable(node);
	      res_ = (f1 & f2) | (f2 & bdd_ithvar(x));
	      return;
	    }
	  }
	/* Unreachable code.  */
	assert(0);
      }

      void
      visit(const multop* node)
      {
	int op = -1;
	switch (node->op())
	  {
	  case multop::And:
	    op = bddop_and;
	    res_ = bddtrue;
	    break;
	  case multop::Or:
	    op = bddop_or;
	    res_ = bddfalse;
	    break;
	  }
	assert(op != -1);
	unsigned s = node->size();
	for (unsigned n = 0; n < s; ++n)
	  {
	    res_ = bdd_apply(res_, recurse(node->nth(n)), op);
	  }
      }

      bdd
      recurse(const formula* f)
      {
	ltl_trad_visitor v(dict_);
	f->accept(v);
	return v.result();
      }


    private:
      translate_dict& dict_;
      bdd res_;
    };

444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527

    // Check whether a formula has a R or G operator at its top-level
    // (preceding logical operators do not count).
    class ltl_possible_fair_loop_visitor: public const_visitor
    {
    public:
      ltl_possible_fair_loop_visitor()
	: res_(false)
      {
      }

      virtual
      ~ltl_possible_fair_loop_visitor()
      {
      }

      bool
      result() const
      {
	return res_;
      }

      void
      visit(const atomic_prop*)
      {
      }

      void
      visit(const constant*)
      {
      }

      void
      visit(const unop* node)
      {
	if (node->op() == unop::G)
	  res_ = true;
      }

      void
      visit(const binop* node)
      {
	switch (node->op())
	  {
	    // r(f1 logical-op f2) = r(f1) logical-op r(f2)
	  case binop::Xor:
	  case binop::Implies:
	  case binop::Equiv:
	    node->first()->accept(*this);
	    if (!res_)
	      node->second()->accept(*this);
	    return;
	  case binop::U:
	    return;
	  case binop::R:
	    res_ = true;
	    return;
	  }
	/* Unreachable code.  */
	assert(0);
      }

      void
      visit(const multop* node)
      {
	unsigned s = node->size();
	for (unsigned n = 0; n < s && !res_; ++n)
	  {
	    node->nth(n)->accept(*this);
	  }
      }

    private:
      bool res_;
    };

    // Check whether a formula can be part of a fair loop.
    // Cache the result for efficiency.
    class possible_fair_loop_checker
    {
    public:
      bool
      check(const formula* f)
      {
528
529
	pfl_map::const_iterator i = pfl_.find(f);
	if (i != pfl_.end())
530
531
532
533
	  return i->second;
	ltl_possible_fair_loop_visitor v;
	f->accept(v);
	bool rel = v.result();
534
	pfl_[f] = rel;
535
536
537
538
539
	return rel;
      }

    private:
      typedef Sgi::hash_map<const formula*, bool, ptr_hash<formula> > pfl_map;
540
      pfl_map pfl_;
541
542
    };

543
544
545
546
547
548
549
550
551
552
    class formula_canonizer
    {
    public:
      formula_canonizer(translate_dict& d)
	: v_(d)
      {
	// For cosmetics, register 1 initially, so the algorithm will
	// not register an equivalent formula first.
	b2f_[bddtrue] = constant::true_instance();
      }
553

554
555
      ~formula_canonizer()
      {
556
	while (!f2b_.empty())
557
	  {
558
559
560
561
	    formula_to_bdd_map::iterator i = f2b_.begin();
	    const formula* f = i->first;
	    f2b_.erase(i);
	    destroy(f);
562
	  }
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
      }

      bdd
      translate(const formula* f)
      {
	// Use the cached result if available.
	formula_to_bdd_map::const_iterator i = f2b_.find(f);
	if (i != f2b_.end())
	  return i->second;

	// Perform the actual translation.
	f->accept(v_);
	bdd res = v_.result();
	f2b_[clone(f)] = res;

	// Register the reverse mapping if it is not already done.
	if (b2f_.find(res) == b2f_.end())
	  b2f_[res] = f;
	return res;
      }

      const formula*
      canonize(const formula* f)
      {
	bdd b = translate(f);

	bdd_to_formula_map::iterator i = b2f_.find(b);
	// Since we have just translated the formula, it is necessary in b2f_.
	assert(i != b2f_.end());

	if (i->second != f)
594
	  {
595
596
	    destroy(f);
	    f = clone(i->second);
597
	  }
598
	return f;
599
600
      }

601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
    private:
      ltl_trad_visitor v_;
      // Map each formula to its associated bdd.  This speed things up when
      // the same formula is translated several times, which especially
      // occurs when canonize() is called repeatedly inside exprop.
      typedef std::map<bdd, const formula*, bdd_less_than> bdd_to_formula_map;
      bdd_to_formula_map b2f_;
      // Map a representation of successors to a canonical formula.
      // We do this because many formulae (such as `aR(bRc)' and
      // `aR(bRc).(bRc)') are equivalent, and are trivially identified
      // by looking at the set of successors.
      typedef std::map<const formula*, bdd> formula_to_bdd_map;
      formula_to_bdd_map f2b_;
    };

  }

  typedef std::map<bdd, bdd, bdd_less_than> prom_map;
  typedef Sgi::hash_map<const formula*, prom_map, ptr_hash<formula> > dest_map;

  static void
  fill_dests(translate_dict& d, dest_map& dests, bdd label, const formula* dest)
  {
624
    bdd conds = bdd_existcomp(label, d.var_set);
625
626
    bdd promises = bdd_existcomp(label, d.a_set);

627
628
629
630
631
632
633
634
635
636
637
638
639
    dest_map::iterator i = dests.find(dest);
    if (i == dests.end())
      {
	dests[dest][promises] = conds;
      }
    else
      {
	i->second[promises] |= conds;
	destroy(dest);
      }
  }


640
  tgba_explicit*
641
  ltl_to_tgba_fm(const formula* f, bdd_dict* dict,
642
		 bool exprop, bool symb_merge, bool branching_postponement,
643
		 bool fair_loop_approx, const ltl::atomic_prop_set* unobs)
644
  {
645
646
    possible_fair_loop_checker pflc;

647
648
649
650
    // Normalize the formula.  We want all the negations on
    // the atomic propositions.  We also suppress logic
    // abbreviations such as <=>, =>, or XOR, since they
    // would involve negations at the BDD level.
651
652
653
    formula* f1 = unabbreviate_logic(f);
    formula* f2 = negative_normal_form(f1);
    destroy(f1);
654

655
656
    std::set<const formula*> formulae_seen;
    std::set<const formula*> formulae_to_translate;
657

658
    translate_dict d(dict);
659
    formula_canonizer fc(d);
660

661
662
    // Compute the set of all promises occurring inside the formula.
    bdd all_promises = bddtrue;
663
    if (fair_loop_approx || unobs)
664
665
666
667
668
669
      {
	ltl_promise_visitor pv(d);
	f2->accept(pv);
	all_promises = pv.result();
      }

670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
    // These are used when atomic propositions are interpreted as
    // events.  There are two kinds of events: observable events are
    // those used in the formula, and unobservable events or other
    // events that can occur at anytime.  All events exclude each
    // other.
    bdd observable_events = bddfalse;
    bdd unobservable_events = bddfalse;
    if (unobs)
      {
	bdd neg_events = bddtrue;
	std::auto_ptr<ltl::atomic_prop_set> aps(ltl::atomic_prop_collect(f));
	for (ltl::atomic_prop_set::const_iterator i = aps->begin();
	     i != aps->end(); ++i)
	  {
	    int p = d.register_proposition(*i);
	    bdd pos = bdd_ithvar(p);
	    bdd neg = bdd_nithvar(p);
	    observable_events = (observable_events & neg) | (neg_events & pos);
	    neg_events &= neg;
	  }
	for (ltl::atomic_prop_set::const_iterator i = unobs->begin();
	     i != unobs->end(); ++i)
	  {
	    int p = d.register_proposition(*i);
	    bdd pos = bdd_ithvar(p);
	    bdd neg = bdd_nithvar(p);
	    unobservable_events = ((unobservable_events & neg)
				   | (neg_events & pos));
	    observable_events &= neg;
	    neg_events &= neg;
	  }
      }
    bdd all_events = observable_events | unobservable_events;

704
705
706
707
708
709
710
711
712
713
    formulae_seen.insert(f2);
    formulae_to_translate.insert(f2);

    tgba_explicit* a = new tgba_explicit(dict);

    a->set_init_state(to_string(f2));

    while (!formulae_to_translate.empty())
      {
	// Pick one formula.
714
	const formula* f = *formulae_to_translate.begin();
715
716
717
	formulae_to_translate.erase(formulae_to_translate.begin());

	// Translate it into a BDD to simplify it.
718
	bdd res = fc.translate(f);
719

720
721
722
723
724
725
726
727
	// Handle exclusive events.
	if (unobs)
	  {
	    res &= observable_events;
	    int n = d.register_next_variable(f);
	    res |= unobservable_events & bdd_ithvar(n) & all_promises;
	  }

728
729
	std::string now = to_string(f);

730
731
732
733
734
	// When branching_postponement is used, we must assume that
	// the source state is in a fair loop.
	bool pflc_from =
	  (fair_loop_approx && !branching_postponement) ? pflc.check(f) : true;

735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
	// We used to factor only Next and A variables while computing
	// prime implicants, with
	//    minato_isop isop(res, d.next_set & d.a_set);
	// in order to obtain transitions with formulae of atomic
	// proposition directly, but unfortunately this led to strange
	// factorizations.  For instance f U g was translated as
	//     r(f U g) = g + a(g).r(X(f U g)).(f + g)
	// instead of just
	//     r(f U g) = g + a(g).r(X(f U g)).f
	// Of course both formulae are logically equivalent, but the
	// latter is "more deterministic" than the former, so it should
	// be preferred.
	//
	// Therefore we now factor all variables.  This may lead to more
	// transitions than necessary (e.g.,  r(f + g) = f + g  will be
750
751
752
	// coded as two transitions), but we later merge all transitions
	// with same source/destination and acceptance conditions.  This
	// is the goal of the `dests' hash.
753
	//
754
	// Note that this is still not optimal.  For instance it is
755
	// better to encode `f U g' as
756
	//     r(f U g) = g + a(g).r(X(f U g)).f.!g
757
758
759
760
	// because that leads to a deterministic automaton.  In order
	// to handle this, we take the conditions of any transition
	// going to true (it's `g' here), and remove it from the other
	// transitions.
761
762
763
	//
	// In `exprop' mode, considering all possible combinations of
	// outgoing propositions generalizes the above trick.
764
	dest_map dests;
765

766
	// Compute all outgoing arcs.
767
768
769
770
771

	// If EXPROP is set, we will refine the symbolic
	// representation of the successors for all combinations of
	// the atomic properties involved in the formula.
	// VAR_SET is the set of these properties.
772
	bdd var_set = bdd_existcomp(bdd_support(res), d.var_set);
773
774
775
776
777
	// ALL_PROPS is the combinations we have yet to consider.
	// We used to start with `all_props = bddtrue', but it is
	// more efficient to start with the set of all satisfiable
	// variables combinations.
	bdd all_props = bdd_existcomp(res, d.var_set);
778
	while (all_props != bddfalse)
779
	  {
780
781
782
	    bdd one_prop_set =
	      exprop ? bdd_satoneset(all_props, var_set, bddtrue) : bddtrue;
	    all_props -= one_prop_set;
783

784
	    typedef std::map<bdd, const formula*, bdd_less_than> succ_map;
785
786
	    succ_map succs;

787
788
789
	    minato_isop isop(res & one_prop_set);
	    bdd cube;
	    while ((cube = isop.next()) != bddfalse)
790
	      {
791
		bdd label = bdd_exist(cube, d.next_set);
792
		bdd dest_bdd = bdd_existcomp(cube, d.next_set);
793
794
795
796
797
798
		const formula* dest = d.conj_bdd_to_formula(dest_bdd);

		// If we already know a state with the same
		// successors, use it in lieu of the current one.
		if (symb_merge)
		  dest = fc.canonize(dest);
799
800
801
802

		// If the destination cannot possibly be part of a fair
		// loop, make all possible promises.
		if (fair_loop_approx
803
804
		    && !(dest == constant::true_instance()
			 || (pflc_from && pflc.check(dest))))
805
		  label &= all_promises;
806

807
808
809
810
811
812
		// If we are not postponing the branching, we can
		// declare the outgoing transitions immediately.
		// Otherwise, we merge transitions with identical
		// label, and declare the outgoing transitions in a
		// second loop.
		if (!branching_postponement)
813
		  {
814
		    fill_dests(d, dests, label, dest);
815
816
817
		  }
		else
		  {
818
819
820
821
		    succ_map::iterator si = succs.find(label);
		    if (si == succs.end())
		      succs[label] = dest;
		    else
822
823
824
825
		      si->second =
			multop::instance(multop::Or,
					 const_cast<formula*>(si->second),
					 const_cast<formula*>(dest));
826
827
		  }
	      }
828
829
830
	    if (branching_postponement)
	      for (succ_map::const_iterator si = succs.begin();
		   si != succs.end(); ++si)
831
		fill_dests(d, dests, si->first, si->second);
832
	  }
833

834
	// Check for an arc going to 1 (True).  Register it first, that
835
836
837
	// way it will be explored before the other during the model
	// checking.
	dest_map::const_iterator i = dests.find(constant::true_instance());
838
839
840
841
	// COND_FOR_TRUE is the conditions of the True arc, so when
	// can remove them from all other arcs.  It might sounds that
	// this is not needed when exprop is used, but in fact it is
	// complementary.
842
843
	//
	// Consider
844
	//   f = r(X(1) R p) = p.(1 + r(X(1) R p))
845
	// with exprop the two outgoing arcs would be
846
847
        //         p                  p
	//     f ----> 1       f ----------> 1
848
849
	//
	// where in fact we could output
850
851
        //         p
	//     f ----> 1
852
	//
853
	// because there is no point in looping on f if we can go to 1.
854
	bdd cond_for_true = bddfalse;
855
856
	if (i != dests.end())
	  {
857
	    // When translating LTL for an event-based logic with
858
859
	    // unobservable events, the 1 state should accept all events,
	    // even unobservable events.
860
861
	    if (unobs && f == constant::true_instance())
	      cond_for_true = all_events;
862
863
864
865
866
867
868
869
870
	    else
	      {
		// There should be only one transition going to 1 (true) ...
		assert(i->second.size() == 1);
		prom_map::const_iterator j = i->second.begin();
		// ... and it is not expected to make any promises.
		assert(j->first == bddtrue);
		cond_for_true = j->second;
	      }
871
872
	    tgba_explicit::transition* t =
	      a->create_transition(now, constant::true_instance()->val_name());
873
	    a->add_condition(t, d.bdd_to_formula(cond_for_true));
874
875
876
877
878
	  }
	// Register other transitions.
	for (i = dests.begin(); i != dests.end(); ++i)
	  {
	    const formula* dest = i->first;
879
880
881
882
	    // The cond_for_true optimization can cause some
	    // transitions to be removed.  So we have to remember
	    // whether a formula is actually reachable.
	    bool reachable = false;
883

884
885
886
887
888
889
	    if (dest != constant::true_instance())
	      {
		std::string next = to_string(dest);
		for (prom_map::const_iterator j = i->second.begin();
		     j != i->second.end(); ++j)
		  {
890
891
892
		    bdd cond = j->second - cond_for_true;
		    if (cond == bddfalse) // Skip false transitions.
		      continue;
893
894
		    tgba_explicit::transition* t =
		      a->create_transition(now, next);
895
		    a->add_condition(t, d.bdd_to_formula(cond));
896
		    d.conj_bdd_to_acc(a, j->first, t);
897
		    reachable = true;
898
899
		  }
	      }
900
901
902
903
904
905
906
	    else
	      {
		// "1" is reachable.
		reachable = true;
	      }
	    if (reachable
		&& formulae_seen.find(dest) == formulae_seen.end())
907
908
909
910
911
912
	      {
		formulae_seen.insert(dest);
		formulae_to_translate.insert(dest);
	      }
	    else
	      {
913
		destroy(dest);
914
915
916
917
918
	      }
	  }
      }

    // Free all formulae.
919
    for (std::set<const formula*>::iterator i = formulae_seen.begin();
920
	 i != formulae_seen.end(); ++i)
921
      destroy(*i);
922

923
924
    // Turn all promises into real acceptance conditions.
    a->complement_all_acceptance_conditions();
925
926
927
928
    return a;
  }

}