tut50.org 27.7 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
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
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
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
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
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
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
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
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
594
595
596
597
598
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
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
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
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
# -*- coding: utf-8 -*-
#+TITLE: Explicit vs. on-the-fly: two interfaces for exploring automata
#+DESCRIPTION: Explanation of the explicit and on-the-fly automata interfaces in Spot
#+SETUPFILE: setup.org
#+HTML_LINK_UP: tut.html


When exploring automata (i.e., following its transition structure),
there are two different interfaces that can be used:
  1. the *on-the-fly* =twa= interface, and
  2. the *explicit* =twa_graph= interface.

To demonstrate the difference between the two interfaces, we will
write an small depth-first search that prints all states accessible
from the initial state of an automaton.

* The explicit interface
   :PROPERTIES:
   :CUSTOM_ID: explicit-interface
   :END:

The explicit interface can only be used on =twa_graph= objects.  In
this interface, states and edges are referred to by numbers that are
indices into state and edge vectors.  This interface is lightweight,
and is the preferred interface for writing most automata algorithms in
Spot.

** How this interface works

The picture below gives a partial view of the classes involved:

#+BEGIN_SRC plantuml :file uml-explicit-classes.png
  package stl {
    class "Forward Iterator" <<Concept>>
    hide "Forward Iterator" members
    hide "Forward Iterator" circle
  }
  package spot {
    package internal {
      class "state_out<digraph<twa_graph_state, twa_graph_edge_data>>" {
         unsigned t_
         __
         edge_iterator<digraph<twa_graph_state, twa_graph_edge_data>> begin()
         edge_iterator<digraph<twa_graph_state, twa_graph_edge_data>> end()
      }

      class "edge_iterator<digraph<twa_graph_state, twa_graph_edge_data>>" {
        unsigned t_
        __
        bool operator==(edge_iterator)
        bool operator!=(edge_iterator)
        edge_storage<twa_graph_edge_data>& operator*(edge_iterator)
        edge_storage<twa_graph_edge_data>* operator->(edge_iterator)
        edge_iterator operator++()
        .. other methods hidden ..
      }

      "Forward Iterator" <|.. "edge_iterator<digraph<twa_graph_state, twa_graph_edge_data>>": model of

      class "edge_storage<twa_graph_edge_data>" {
        +unsigned src
        +unsigned dst
        +unsigned next_succ
      }

      class "distate_storage<twa_graph_state>" {
        +unsigned succ
        +unsigned succ_tail
      }

    }

    class twa_graph {
      +typedef digraph<twa_graph_state, twa_graph_edge_data> graph_t
      __
      +twa_graph(const bdd_dict_ptr&)
      +graph_t& get_graph()
      .. exploration ..
      +unsigned get_init_state_number()
      +internal::state_out<graph_t> out(unsigned src)
      .. other methods hidden ..
    }

    twa <|-- twa_graph
    abstract class twa
    hide twa members

    class "digraph<twa_graph_state, twa_graph_edge_data>" {
      +internal::edge_storage<twa_graph_edge_data> edge_storage(unsigned s)
      +internal::distate_storage<twa_graph_state> state_storage(unsigned s)
      +internal::state_out<graph_t> out(unsigned src)
      .. other details omitted ..
    }

    twa_graph *--> "1" "digraph<twa_graph_state, twa_graph_edge_data>"

    class twa_graph_edge_data {
     +bdd cond
     +acc_cond::mark_t acc
    }

    "digraph<twa_graph_state, twa_graph_edge_data>" *--> "*" "edge_storage<twa_graph_edge_data>"
    "digraph<twa_graph_state, twa_graph_edge_data>" *--> "*" "distate_storage<twa_graph_state>"

    twa_graph_edge_data <|-- "edge_storage<twa_graph_edge_data>"
    twa_graph_state <|-- "distate_storage<twa_graph_state>"
    hide twa_graph_state members

    class acc_cond
    hide acc_cond members
    twa *--> acc_cond

    "state_out<digraph<twa_graph_state, twa_graph_edge_data>>" o--> "digraph<twa_graph_state, twa_graph_edge_data>"
    "digraph<twa_graph_state, twa_graph_edge_data>" ...> "state_out<digraph<twa_graph_state, twa_graph_edge_data>>" : "create"
    "state_out<digraph<twa_graph_state, twa_graph_edge_data>>" ...> "edge_iterator<digraph<twa_graph_state, twa_graph_edge_data>>" : "create"
    "edge_iterator<digraph<twa_graph_state, twa_graph_edge_data>>" o--> "digraph<twa_graph_state, twa_graph_edge_data>"
  }

#+END_SRC

#+RESULTS:
[[file:uml-explicit-classes.png]]


An ω-automaton can be defined as a labeled directed graph, plus an
initial state and an acceptance condition.  The =twa_graph= of Spot
stores exactly these three components: the transition structure is
stored as an instance of =digraph= (directed graph), the initial state
is just a number, and the acceptance condition is an instance of
=acc_cond= which is actually inherited from the =twa= parent.  You can
ignore the =twa= inheritance for now, we will discuss it when we talk
about [[#on-the-fly-interface][the on-the-fly interface in the next section]].

In this section we are discussing the "explicit interface", which is a
way of exploring the stored graph directly.

The =digraph= template class in Spot is parameterized by classes
representing additional data to store on state, and on edges.  In the
case of a =twa_graph=, these extra data are implemented as
=twa_graph_state= (but we won't be concerned about this type) and
=twa_graph_edge_data=.  The class =twa_graph_edge_data= has two
fields: =cond= is a BDD representing the label of the edge, and
=acc= represents the acceptance sets to which the edge belong.

The =digraph= stores a vector of states, and a vector of edges, but
both states and edges need to be equipped with field necessary to
represent the graph structure.  In particular, a state holds two edges
numbers representing the first (=succ=) and last (=succ_tail=) edges
exiting the state (that "last edge" is only useful to append new
transitions, it is not used for iteration), and each edge has three
additional fields: =src= (the source state), =dst= (the destination
state), and =next_succ= (the index of the next edge leaving =src=, in
the edge vector).  By way of template inheritance, these
=digraph=-fields are combined with the =twa_graph= specific fields, so
that all edges can be represented by an instance of
=std::vector<internal::edge_storage<twa_graph_edges_data>>=: each
element of this vector acts as a structure with 5 fields; likewise for
the state vector.

Calling =get_init_state_number()= will return a state number which is
just an index into the state vector of the underlying graph.

From a state number =s=, it is possible to iterate over all successors
by doing a =for= loop on =out(s)=, as in:

#+BEGIN_SRC C++ :results verbatim :exports both
  #include <iostream>
  #include <spot/twa/twagraph.hh>
  #include <spot/tl/parse.hh>
  #include <spot/twaalgos/translate.hh>

  void example(spot::const_twa_graph_ptr aut)
  {
    unsigned s = aut->get_init_state_number();
    for (auto& e: aut->out(s))
      std::cout << e.src << "->" << e.dst << '\n';
  }

  int main()
  {
    // Create a small example automaton
    spot::parsed_formula pf = spot::parse_infix_psl("FGa | FGb");
    if (pf.format_errors(std::cerr))
      return 1;
    example(spot::translator().run(pf.f));
  }
#+END_SRC

#+RESULTS:
: 2->0
: 2->1
: 2->2

In the above lines, =aut->out(s)= delegates to
=aut->get_graphs().out(s)= and returns a =state_out<graph_t>=
instance, which is a small temporary object masquerading as an STL
container with =begin()= and =end()= methods.  The ranged-for loop
syntax of C++ works exactly as if we had typed

#+BEGIN_SRC C++
  // You could write this, but why not let the compiler do it for you?
  // In any case, do not spell out the types of tmp and i, as those
  // should be considered internal details.
  void example(spot::const_twa_graph_ptr aut)
  {
    unsigned s = aut->get_init_state_number();
    auto tmp = aut->get_graph().out(s);
    for (auto i = tmp.begin(), end = tmp.end(); i != end; ++i)
      std::cout << i->src << "->" << i->dst << '\n';
  }
#+END_SRC

In the above =example()= function, the iterators =i= and =end= are
instance of the =internal::edge_iterator<spot::twa_graph::graph_t>=
class, which redefines enough operators to act like an STL Foward
Iterator over all outgoing edges of =s=.  Note that the =tmp= and =i=
objects hold a pointer to the graph, but it does not really matters
because the compiler will optimize this away.

In fact after operators are inlined and useless temporary variables
removed, the above loop compiles to something equivalent to this:

#+BEGIN_SRC C++
  // You could also write this lower-level version, and that sometimes
  // helps (e.g., if you want to pause the loop and then resume it, as
  // we will do later).
  void example(spot::const_twa_graph_ptr aut)
  {
    unsigned s = aut->get_init_state_number();
    auto& g = aut->get_graph();
    unsigned b = g.state_storage(s).succ; // first edge of state s
    while (b)
      {
        auto& e = g.edge_storage(b);
        std::cout << e.src << "->" << e.dst << '\n';
        b = e.next_succ;
      }
  }
#+END_SRC

Note that ~next_succ==0~ marks the last edge in a successor group;
this is why edge numbers start at 1.

Despite the various levels of abstractions, these three loops compile
to exactly the same machine code.

** Recursive DFS

Let us write a DFS using this interface.  A recursive version is easy:
we call =dfs_rec()= from the initial state, that function updates a
vector of visited states in order to not visit them twice, and recurse
on all successors of the given state.

#+BEGIN_SRC C++ :results verbatim :exports both
  #include <iostream>
  #include <spot/twa/twagraph.hh>
  #include <spot/tl/parse.hh>
  #include <spot/twaalgos/translate.hh>

  void dfs_rec(spot::const_twa_graph_ptr aut, unsigned state, std::vector<bool>& seen)
  {
    seen[state] = true;
    for (auto& e: aut->out(state))
      {
         std::cout << e.src << "->" << e.dst << '\n';
         if (!seen[e.dst])
           dfs_rec(aut, e.dst, seen);
      }
  }

  void dfs(spot::const_twa_graph_ptr aut)
  {
    std::vector<bool> seen(aut->num_states());
    dfs_rec(aut, aut->get_init_state_number(), seen);
  }

  int main()
  {
    // Create a small example automaton
    spot::parsed_formula pf = spot::parse_infix_psl("FGa | FGb");
    if (pf.format_errors(std::cerr))
      return 1;
    dfs(spot::translator().run(pf.f));
  }
#+END_SRC

#+RESULTS:
: 2->0
: 0->0
: 2->1
: 1->1
: 2->2

** Iterative DFS (two versions)

 Recursive graph algorithms are usually avoided, especially if large
 graphs should be processed.

 By maintaining a stack of states to process, we can visit all
 accessible transitions in a "DFS-ish" way, but without producing
 exactly the same output as above.

 #+BEGIN_SRC C++ :results verbatim :exports both
   #include <iostream>
   #include <stack>
   #include <spot/twa/twagraph.hh>
   #include <spot/tl/parse.hh>
   #include <spot/twaalgos/translate.hh>

   void almost_dfs(spot::const_twa_graph_ptr aut)
   {
     std::vector<bool> seen(aut->num_states());
     std::stack<unsigned> todo;
     auto push_state = [&](unsigned state)
       {
         todo.push(state);
         seen[state] = true;
       };
     push_state(aut->get_init_state_number());
     while (!todo.empty())
       {
         unsigned s = todo.top();
         todo.pop();
         for (auto& e: aut->out(s))
           {
              std::cout << e.src << "->" << e.dst << '\n';
              if (!seen[e.dst])
                push_state(e.dst);
           }
       }
   }

   int main()
   {
     // Create a small example automaton
     spot::parsed_formula pf = spot::parse_infix_psl("FGa | FGb");
     if (pf.format_errors(std::cerr))
       return 1;
     almost_dfs(spot::translator().run(pf.f));
   }
 #+END_SRC

 #+RESULTS:
 : 2->0
 : 2->1
 : 2->2
 : 1->1
 : 0->0

 So this still prints all accessible edges, but not in the same order
 as our recursive DFS.  This is because this version prints all the
 outgoing edges of one state before processing the successors.

 For many algorithms, this different ordering makes no difference, and
 this order should even be preferred: groups of transitions leaving
 the same state are usually stored consecutively in memory, so they
 are better processed in chain, rather than trying to follow exactly
 the order we would get from a recursive DFS, which will jump at
 random places in the edge vector.

 In fact writing the iterative equivalent of the recursive =dfs()= is
 a bit challenging if we do not want to be wasteful.  Clearly, we
 can no longer use the ranged-for loop, because we need to process
 one edge, save the current iterator on a stack to process the
 successor, and finally advance the iterator once we pop back to it.

Given the above data structure, it is tempting to use a
=std::stack<spot::internal::edge_iterator<spot::twa_graph::graph_t>>=,
but that is a bad idea.  Remember that those =internal::edge_iterator=
are meant to be short-lived temporary objects, and they all store a
pointer to graph.  We do not want to store multiple copies of this
pointer on our stack.  Besides, *you* do not want to ever write
=spot::internal= in your code.

So a better implementation (better than the
=std::stack<spot::internal::edge_iterator<...>>= suggestion) would be
to maintain a stack of edge numbers.  Indeed, each edge
stores the number of the next edge leaving the same source
state, so this is enough to remember where we are.

 #+BEGIN_SRC C++ :results verbatim :exports both
   #include <iostream>
   #include <stack>
   #include <spot/twa/twagraph.hh>
   #include <spot/tl/parse.hh>
   #include <spot/twaalgos/translate.hh>

   void dfs(spot::const_twa_graph_ptr aut)
   {
     std::vector<bool> seen(aut->num_states());
     std::stack<unsigned> todo;    // Now storing edges numbers
     auto& gr = aut->get_graph();
     auto push_state = [&](unsigned state)
       {
         todo.push(gr.state_storage(state).succ);
         seen[state] = true;
       };
     push_state(aut->get_init_state_number());
     while (!todo.empty())
       {
         unsigned edge = todo.top();
         todo.pop();
         if (edge == 0U)           // No more outgoing edge
           continue;
         auto& e = gr.edge_storage(edge);
         todo.push(e.next_succ); // Prepare next sibling edge.
         if (!seen[e.dst])
            push_state(e.dst);
         std::cout << e.src << "->" << e.dst << '\n';
       }
   }

   int main()
   {
     // Create a small example automaton
     spot::parsed_formula pf = spot::parse_infix_psl("FGa | FGb");
     if (pf.format_errors(std::cerr))
       return 1;
     dfs(spot::translator().run(pf.f));
   }
 #+END_SRC

 #+RESULTS:
 : 2->0
 : 0->0
 : 2->1
 : 1->1
 : 2->2

This version is functionally equivalent to the recursive one, but its
implementation requires more knowledge of the graph data structure
than both the recursive and the =almost_dfs()= version.

* The on-the-fly =twa= interface
   :PROPERTIES:
   :CUSTOM_ID: on-the-fly-interface
   :END:

The =twa= class defines an abstract interface suitable for on-the-fly
exploration of an automaton.  Subclasses of =twa= need not represent
the entire automaton in memory; if they prefer, they can compute it as
it is explored.

Naturally =twa_graph=, even if they store the automaton graph
explicitly, are subclasses of =twa=, so they also implement the
on-the-fly interface, even if they do not have to compute anything.

** How this interface works

The following class diagram has two classes in common with the
previous one: =twa= and =twa_graph=, but this time the focus is on the
abstract interface defined in =twa=, not in the explicit interface
defined in =twa_graph=.

#+BEGIN_SRC plantuml :file uml-otf-classes.png
package spot {
  package internal {
    class succ_iterable {
      +internal::succ_iterator begin()
      +internal::succ_iterator end()
    }

    class succ_iterator {
      succ_iterator(twa_succ_iterator*)
      bool operator==(succ_iterator) const
      bool operator!=(succ_iterator) const
      const twa_succ_iterator* operator*() const
      void operator++()
    }
  }

  class acc_cond
  hide acc_cond members

  together {
  abstract class twa {
    #twa_succ_iterator* iter_cache_
    #bdd_dict_ptr dict_
    __
    #twa(const bdd_dict_ptr&)
    .. exploration ..
    +{abstract}state* get_init_state()
    +{abstract}twa_succ_iterator* succ_iter(state*)
    +internal::succ_iterable succ(const state*)
    +void release_iter(twa_succ_iterator*)
    .. state manipulation ..
    +{abstract} std::string format_state(const state*)
    +state* project_state(const state*, const const_twa_ptr&)
    .. other methods not shown..
  }
  abstract class twa_succ_iterator {
    .. iteration ..
    {abstract}+bool first()
    {abstract}+bool next()
    {abstract}+bool done()
    .. inspection ..
    {abstract}+const state* dst()
    {abstract}+bdd cond()
    {abstract}+acc_cond::mark_t acc()
  }

  abstract class state {
    +{abstract}int compare(const state*) const
    +{abstract}size_t hash() const
    +{abstract}state* clone() const
    +void destroy() const
    #~state()
  }
  }


  class twa_graph

  twa <|-- twa_graph
  twa *--> acc_cond

  class twa_graph_state

  hide twa_graph members
  hide twa_graph_state members
  hide twa_graph_succ_iterator members

  twa_succ_iterator <|-- twa_graph_succ_iterator
  succ_iterable o--> twa
  twa_succ_iterator <--* succ_iterable
  twa_succ_iterator <--o succ_iterator
  twa ...> succ_iterable : "create"
  succ_iterable ...> succ_iterator : "create"
  state <|-- twa_graph_state
}
#+END_SRC

#+RESULTS:
[[file:uml-otf-classes.png]]


To explore a =twa=, one would first call =twa::get_init_state()=,
which returns a pointer to a =state=.  Then, calling
=twa::succ_iter()= on this =state= will return a =twa_succ_iterator=
that allows iterating over all successors.

Different subclasses of =twa= will instantiate different subclasses of
=state= and =twa_succ_iterator= .  In the case of =twa_graph=, the
subclasses used are =twa_graph_succ_iterator= and =twa_graph_state=,
but you can ignore that until you have to write your own =twa=
subclass.

The interface puts few requirement on memory management: we want to be
able to write automata that can forget about their states (and
recompute them), so there is no guarantee that reaching twice the same
state will give return the same pointer.  Even calling
=get_init_state()= twice could return different pointers.  The only
way to decide whether two =state*= =s1= and =s2= represent the same
state is to check that ~s1->compare(s2) == 0~.

As far as memory management goes, there are roughly two types of =twa=
subclasses: those that always create new =state= instances, and those
that reuse =state= instances (either because they have a cache, or
because, as in the case of =twa_graph=, they know the entire graph).

From the user's perspective, =state= should never be passed to =delete=
(their protected destructor will prevent that).  Instead, we should
call =state::destroy()=.  Doing so allows each subclass to override
the default behavior of =destroy()= (which is to call =delete=).  States
can be cloned using the =state::clone()= methode, in which case each
copy has to be destroyed.

=twa_succ_iterator= instances are allocated and should be deleted once
done, but to save some work, they can also be returned to the
automaton with =twa::release_iter=.  By default, this method stores the
last iterator received to recycle it in the next call to =succ_iter()=,
saving a =delete= and =new= pair.

To summarize, here is a crude loop over the successors of the initial
state:

#+BEGIN_SRC C++ :results verbatim :exports both
  #include <iostream>
  #include <spot/twa/twa.hh>
  #include <spot/tl/parse.hh>
  #include <spot/twaalgos/translate.hh>

  void example(spot::const_twa_ptr aut)
  {
    const spot::state* s = aut->get_init_state();
    spot::twa_succ_iterator* i = aut->succ_iter(s);
    for (i->first(); !i->done(); i->next())
      {
         const spot::state* dst = i->dst();
         std::cout << aut->format_state(s) << "->"
                   << aut->format_state(dst) << '\n';
         dst->destroy();
      }
    aut->release_iter(i); // "delete i;" is OK too, but inferior
    s->destroy();
  }

  int main()
  {
    // Create a small example automaton
    spot::parsed_formula pf = spot::parse_infix_psl("FGa | FGb");
    if (pf.format_errors(std::cerr))
      return 1;
    example(spot::translator().run(pf.f));
  }
#+END_SRC

#+RESULTS:
: 2->0
: 2->1
: 2->2

Notice that a =twa_succ_iterator= allows iterating over outgoing
edges, but only offers access to =dst()=, =acc()=, and =cond()= for
this edge.  The source state is not available from the iterator.  This
is usually not a problem: since the iterator was created from this
state, it is /usually/ known.

Let us improve the above loop.  In the previous example, each of
=first()=, =done()=, =next()= is a virtual method call.  So if there
are $n$ successors, there will be $1$ call to =first()=, $n$ calls to
=next()=, and $n+1$ call to =done()=, so a total of $2n$ virtual
method calls.

However =first()= and =next()= also return a Boolean stating whether
the loop could continue.  This allows rewriting the above code as
follows:

#+BEGIN_SRC C++
  void example(spot::const_twa_ptr aut)
  {
    const spot::state* s = aut->get_init_state();
    spot::twa_succ_iterator* i = aut->succ_iter(s);
    if (i->first())
      do
        {
           const spot::state* dst = i->dst();
           std::cout << aut->format_state(s) << "->"
                     << aut->format_state(dst) << '\n';
           dst->destroy();
        }
      while(i->next());
    aut->release_iter(i);
    s->destroy();
  }
#+END_SRC

Now we have only $1$ call to =first()= and $n$ calls to =next()=,
so we almost halved to number of virtual calls.

Using C++11's ranged =for= loop, this example can be reduced to the
following equivalent code:

#+BEGIN_SRC C++ :results verbatim :exports both
  #include <iostream>
  #include <spot/twa/twa.hh>
  #include <spot/tl/parse.hh>
  #include <spot/twaalgos/translate.hh>

  void example(spot::const_twa_ptr aut)
  {
    const spot::state* s = aut->get_init_state();
    for (auto i: aut->succ(s))
      {
         const spot::state* dst = i->dst();
         std::cout << aut->format_state(s) << "->"
                   << aut->format_state(dst) << '\n';
         dst->destroy();
      }
    s->destroy();
  }

  int main()
  {
    // Create a small example automaton
    spot::parsed_formula pf = spot::parse_infix_psl("FGa | FGb");
    if (pf.format_errors(std::cerr))
      return 1;
    example(spot::translator().run(pf.f));
  }
#+END_SRC

#+RESULTS:
: 2->0
: 2->1
: 2->2

This works in a similar way as =out(s)= in the explicit interface.
Calling =aut->succ(s)= creates a fake container
(=internal::succ_iterable=) with =begin()= and =end()= methods that
return STL-like iterators (=internal::succ_iterator=).  Incrementing
the =internal::succ_iterator= will actually increment the
=twa_succ_iterator= they hold.  Upon completion of the loop, the
temporary =internal::succ_iterable= is destroyed and its destructor
passes the iterator back to =aut->release_iter()= for recycling.

** Recursive DFS (v1)

We can now write a recursive DFS easily.  The only pain is to keep
track of the state to =destroy()= them only after we do not need them
anymore.  This tracking can be done using the data structure we use to
remember what states we have already seen.

#+BEGIN_SRC C++ :results verbatim :exports both
  #include <iostream>
  #include <unordered_set>
  #include <spot/twa/twa.hh>
  #include <spot/tl/parse.hh>
  #include <spot/twa/twaproduct.hh>
  #include <spot/twaalgos/translate.hh>

  typedef std::unordered_set<const spot::state*,
                             spot::state_ptr_hash,
                             spot::state_ptr_equal> seen_t;

  void dfs_rec(spot::const_twa_ptr aut, const spot::state* s, seen_t& seen)
  {
    if (seen.insert(s).second == false)
      {
        s->destroy();
        return;
      }
    for (auto i: aut->succ(s))
      {
         const spot::state* dst = i->dst();
         std::cout << aut->format_state(s) << "->"
                   << aut->format_state(dst) << '\n';
         dfs_rec(aut, dst, seen);
         // Do not destroy dst, as it is either destroyed by dfs_rec()
         // or stored in seen.
      }
  }

  void dfs(spot::const_twa_ptr aut)
  {
    seen_t seen;
    dfs_rec(aut, aut->get_init_state(), seen);

    // Do not forget to destroy all states in seen.
    for (auto s: seen)
      s->destroy();
  }

  int main()
  {
    // Create a small example automaton
    spot::parsed_formula pf = spot::parse_infix_psl("FGa | FGb");
    if (pf.format_errors(std::cerr))
      return 1;
    dfs(spot::translator().run(pf.f));
  }
#+END_SRC

#+RESULTS:
: 2->0
: 0->0
: 2->1
: 1->1
: 2->2

** Recursive DFS (v2)

Using a hash map to keep a unique pointer to each state is quite
common.  The class =spot::state_unicity_table= can be used for this
purpose.  =spot::state_unicity_table::operator()= inputs a =state*=,
and returns either the same state, or the first equal state seen
previously (in that case the passed state is destroyed).  The
=spot::state_unicity_table::is_new()= behaves similarly, but returns
=nullptr= for states that already exist.

With this class, the recursive code can be simplified to this:

#+BEGIN_SRC C++ :results verbatim :exports both
  #include <iostream>
  #include <spot/twa/twa.hh>
  #include <spot/tl/parse.hh>
  #include <spot/twa/twaproduct.hh>
  #include <spot/twaalgos/translate.hh>

  void dfs_rec(spot::const_twa_ptr aut, const spot::state* s,
               spot::state_unicity_table& seen)
  {
    if (seen.is_new(s))
      for (auto i: aut->succ(s))
        {
          const spot::state* dst = i->dst();
          std::cout << aut->format_state(s) << "->"
                    << aut->format_state(dst) << '\n';
          dfs_rec(aut, dst, seen);
        }
  }

  void dfs(spot::const_twa_ptr aut)
  {
    spot::state_unicity_table seen;
    dfs_rec(aut, aut->get_init_state(), seen);
  }

  int main()
  {
    // Create a small example automaton
    spot::parsed_formula pf = spot::parse_infix_psl("FGa | FGb");
    if (pf.format_errors(std::cerr))
      return 1;
    dfs(spot::translator().run(pf.f));
  }
#+END_SRC

#+RESULTS:
: 2->0
: 0->0
: 2->1
: 1->1
: 2->2

Note how this completely hides all the calls to =state::destroy()=.
They are performed in =state_unicity_table::is_new()= and in
=state_unicity_table::~state_unicity_table()=.

** Iterative DFS

For a non-recursive version, let us use a stack of
=twa_succ_iterator=.  However these iterators do not know their
source, so we better store that in the stack as well if we want to
print it.

#+BEGIN_SRC C++ :results verbatim :exports both
  #include <iostream>
  #include <stack>
  #include <spot/twa/twa.hh>
  #include <spot/tl/parse.hh>
  #include <spot/twa/twaproduct.hh>
  #include <spot/twaalgos/translate.hh>

  void dfs(spot::const_twa_ptr aut)
  {
    spot::state_unicity_table seen;
    std::stack<std::pair<const spot::state*,
                         spot::twa_succ_iterator*>> todo;

    // push receives a newly-allocated state and immediately store it in
    // seen.  Therefore any state on todo is already in seen and does
    // not need to be destroyed.
    auto push = [&](const spot::state* s)
      {
         if (seen.is_new(s))
           {
             spot::twa_succ_iterator* it = aut->succ_iter(s);
             if (it->first())
               todo.emplace(s, it);
             else                 // No successor for s
               aut->release_iter(it);
           }
      };
    push(aut->get_init_state());
    while (!todo.empty())
      {
         const spot::state* src = todo.top().first;
         spot::twa_succ_iterator* srcit = todo.top().second;
         const spot::state* dst = srcit->dst();
         std::cout << aut->format_state(src) << "->"
                   << aut->format_state(dst) << '\n';
         // Advance the iterator, and maybe release it.
         if (!srcit->next())
           {
              aut->release_iter(srcit);
              todo.pop();
           }
         push(dst);
      }
  }

  int main()
  {
    // Create a small example automaton
    spot::parsed_formula pf = spot::parse_infix_psl("FGa | FGb");
    if (pf.format_errors(std::cerr))
      return 1;
    dfs(spot::translator().run(pf.f));
  }
#+END_SRC

#+RESULTS:
: 2->0
: 0->0
: 2->1
: 1->1
: 2->2