dep-selector-libgecode 1.0.0.alpha.0

data/ext/libgecode3/vendor/gecode-3.7.3/gecode/int/extensional/base.hpp

@@ -0,0 +1,163 @@
+/* -*- mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- */
+/*
+ *  Main authors:
+ *     Mikael Lagerkvist <lagerkvist@gecode.org>
+ *
+ *  Contributing authors:
+ *     Christian Schulte <schulte@gecode.org>
+ *
+ *  Copyright:
+ *     Mikael Lagerkvist, 2007
+ *     Christian Schulte, 2008
+ *
+ *  Last modified:
+ *     $Date: 2010-07-15 01:46:18 +1000 (Thu, 15 Jul 2010) $ by $Author: schulte $
+ *     $Revision: 11192 $
+ *
+ *  This file is part of Gecode, the generic constraint
+ *  development environment:
+ *     http://www.gecode.org
+ *
+ *  Permission is hereby granted, free of charge, to any person obtaining
+ *  a copy of this software and associated documentation files (the
+ *  "Software"), to deal in the Software without restriction, including
+ *  without limitation the rights to use, copy, modify, merge, publish,
+ *  distribute, sublicense, and/or sell copies of the Software, and to
+ *  permit persons to whom the Software is furnished to do so, subject to
+ *  the following conditions:
+ *
+ *  The above copyright notice and this permission notice shall be
+ *  included in all copies or substantial portions of the Software.
+ *
+ *  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ *  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ *  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ *  NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ *  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ *  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ *  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+namespace Gecode { namespace Int { namespace Extensional {
+  /*
+   * The propagator proper
+   *
+   */
+
+  template<class View, bool subscribe>
+  forceinline
+  Base<View,subscribe>::Base(Home home, ViewArray<View>& x0,
+                             const TupleSet& t)
+    : Propagator(home), x(x0), tupleSet(t), last_data(NULL) {
+    if (subscribe)
+      x.subscribe(home, *this, PC_INT_DOM);
+
+    assert(ts()->finalized());
+
+    init_last(home, ts()->last);
+
+    home.notice(*this,AP_DISPOSE);
+  }
+
+  template<class View, bool subscribe>
+  forceinline
+  Base<View,subscribe>::Base(Space& home, bool share, Base<View,subscribe>& p)
+    : Propagator(home,share,p), last_data(NULL) {
+    x.update(home, share, p.x);
+    tupleSet.update(home, share, p.tupleSet);
+
+    init_last(home, p.last_data);
+  }
+
+  template<class View, bool subscribe>
+  forceinline void
+  Base<View,subscribe>::init_last(Space& home, Tuple** source) {
+    if (last_data == NULL) {
+      int literals = static_cast<int>(ts()->domsize*x.size());
+      last_data = home.alloc<Tuple*>(literals);
+      for (int i = literals; i--; )
+        last_data[i] = source[i];
+    }
+  }
+
+  template<class View, bool subscribe>
+  forceinline TupleSet::TupleSetI*
+  Base<View,subscribe>::ts(void) {
+    return tupleSet.implementation();
+  }
+
+  template<class View, bool subscribe>
+  PropCost
+  Base<View,subscribe>::cost(const Space&, const ModEventDelta&) const {
+    return PropCost::quadratic(PropCost::HI,x.size());
+  }
+
+#define GECODE_LAST_TUPLE(l) (*(l))
+
+  template<class View, bool subscribe>
+  forceinline Tuple
+  Base<View,subscribe>::last(int i, int n) {
+    return GECODE_LAST_TUPLE(last_data[(i*ts()->domsize) + n]);
+  }
+
+  template<class View, bool subscribe>
+  forceinline Tuple
+  Base<View,subscribe>::last_next(int i, int n) {
+    assert(last(i,n) != NULL);
+    assert(last(i,n)[i] == n+ts()->min);
+    int pos = (i*static_cast<int>(ts()->domsize)) + n;
+    ++(last_data[pos]);
+    if (last(i,n)[i] != (n+ts()->min))
+      last_data[pos] = ts()->nullpointer;
+    return last(i,n);
+  }
+
+
+  template<class View, bool subscribe>
+  forceinline void
+  Base<View,subscribe>::init_dom(Space& home, Domain dom) {
+    unsigned int domsize = ts()->domsize;
+    for (int i = x.size(); i--; ) {
+      dom[i].init(home, domsize);
+      for (ViewValues<View> vv(x[i]); vv(); ++vv)
+        dom[i].set(static_cast<unsigned int>(vv.val()-ts()->min));
+    }
+  }
+
+  template<class View, bool subscribe>
+  forceinline bool
+  Base<View,subscribe>::valid(Tuple t, Domain dom) {
+    for (int i = x.size(); i--; )
+      if (!dom[i].get(static_cast<unsigned int>(t[i]-ts()->min)))
+        return false;
+    return true;
+  }
+#undef GECODE_LAST_TUPLE
+  template<class View, bool subscribe>
+  forceinline Tuple
+  Base<View,subscribe>::find_support(Domain dom, int i, int n) {
+    Tuple l = last(i,n);
+    while ((l != NULL) && !valid(l, dom))
+      l = last_next(i,n);
+    return l;
+  }
+
+
+  template<class View, bool subscribe>
+  forceinline size_t
+  Base<View,subscribe>::dispose(Space& home) {
+    home.ignore(*this,AP_DISPOSE);
+    (void) Propagator::dispose(home);
+    if (subscribe)
+      x.cancel(home,*this,PC_INT_DOM);
+    // take care of last_data
+    unsigned int literals = ts()->domsize*x.size();
+    home.rfree(last_data, sizeof(Tuple*)*literals);
+    (void) tupleSet.~TupleSet();
+    return sizeof(*this);
+  }
+
+}}}
+
+// STATISTICS: int-prop
+

data/ext/libgecode3/vendor/gecode-3.7.3/gecode/int/extensional/basic.hpp

@@ -0,0 +1,148 @@
+/* -*- mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- */
+/*
+ *  Main authors:
+ *     Mikael Lagerkvist <lagerkvist@gecode.org>
+ *
+ *  Contributing authors:
+ *     Christian Schulte <schulte@gecode.org>
+ *
+ *  Copyright:
+ *     Mikael Lagerkvist, 2007
+ *     Christian Schulte, 2008
+ *
+ *  Last modified:
+ *     $Date: 2010-07-15 01:46:18 +1000 (Thu, 15 Jul 2010) $ by $Author: schulte $
+ *     $Revision: 11192 $
+ *
+ *  This file is part of Gecode, the generic constraint
+ *  development environment:
+ *     http://www.gecode.org
+ *
+ *  Permission is hereby granted, free of charge, to any person obtaining
+ *  a copy of this software and associated documentation files (the
+ *  "Software"), to deal in the Software without restriction, including
+ *  without limitation the rights to use, copy, modify, merge, publish,
+ *  distribute, sublicense, and/or sell copies of the Software, and to
+ *  permit persons to whom the Software is furnished to do so, subject to
+ *  the following conditions:
+ *
+ *  The above copyright notice and this permission notice shall be
+ *  included in all copies or substantial portions of the Software.
+ *
+ *  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ *  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ *  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ *  NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ *  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ *  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ *  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+namespace Gecode { namespace Int { namespace Extensional {
+
+  /*
+   * The propagator proper
+   *
+   */
+
+  template<class View, bool shared>
+  forceinline
+  Basic<View,shared>::Basic(Home home, ViewArray<View>& x,
+                            const TupleSet& t)
+    : Base<View>(home,x,t) {
+  }
+
+  template<class View, bool shared>
+  forceinline ExecStatus
+  Basic<View,shared>::post(Home home, ViewArray<View>& x,
+                           const TupleSet& t) {
+    // All variables in the correct domain
+    for (int i = x.size(); i--; ) {
+      GECODE_ME_CHECK(x[i].gq(home, t.min()));
+      GECODE_ME_CHECK(x[i].lq(home, t.max()));
+    }
+    (void) new (home) Basic<View,shared>(home,x,t);
+    return ES_OK;
+  }
+
+  template<class View, bool shared>
+  forceinline
+  Basic<View,shared>::Basic(Space& home, bool share, Basic<View,shared>& p)
+    : Base<View>(home,share,p) {
+  }
+
+  template<class View, bool shared>
+  PropCost
+  Basic<View,shared>::cost(const Space&, const ModEventDelta& med) const {
+    if (View::me(med) == ME_INT_VAL)
+      return PropCost::quadratic(PropCost::HI,x.size());
+    else
+      return PropCost::cubic(PropCost::HI,x.size());
+  }
+
+  template<class View, bool shared>
+  Actor*
+  Basic<View,shared>::copy(Space& home, bool share) {
+    return new (home) Basic<View,shared>(home,share,*this);
+  }
+
+  template<class View, bool shared>
+  ExecStatus
+  Basic<View,shared>::propagate(Space& home, const ModEventDelta&) {
+    // Set up datastructures
+
+    // Bit-sets for amortized O(1) access to domains
+    Region r(home);
+    BitSet* dom = r.alloc<BitSet>(x.size());
+    init_dom(home, dom);
+
+    // Bit-sets for processed values.
+    BitSet* has_support = r.alloc<BitSet>(x.size());
+    for (int i = x.size(); i--; )
+      has_support[i].init(home, ts()->domsize);
+
+
+    // Values to prune
+    Support::StaticStack<int,Region> nq(r,static_cast<int>(ts()->domsize));
+
+    // Run algorithm
+
+    // Check consistency for each view-value pair
+    for (int i = x.size(); i--; ) {
+      for (ViewValues<View> vv(x[i]); vv(); ++vv) {
+        // Value offset for indexing
+        int val = vv.val() - ts()->min;
+        if (!has_support[i].get(static_cast<unsigned int>(val))) {
+          // Find support for value vv.val() in view
+          Tuple l = find_support(dom, i, val);
+          if (l == NULL) {
+            // No possible supports left
+            nq.push(vv.val());
+          } else {
+            // Mark values as supported
+            // Only forward direction marking is needed since all
+            // previous values have been checked
+            for (int j = i; j--; ) {
+              has_support[j].set(static_cast<unsigned int>(l[j]- ts()->min));
+              assert(has_support[j].get(l[j] - ts()->min));
+            }
+          }
+        }
+      }
+
+      // Prune values for x[i] which do not have support anymore
+      while (!nq.empty())
+        GECODE_ME_CHECK(x[i].nq(home,nq.pop()));
+    }
+
+    for (int i = x.size(); i--; )
+      if (!x[i].assigned())
+        return shared ? ES_NOFIX : ES_FIX;
+    return home.ES_SUBSUMED(*this);
+  }
+
+}}}
+
+// STATISTICS: int-prop
+

data/ext/libgecode3/vendor/gecode-3.7.3/gecode/int/extensional/dfa.cpp

@@ -0,0 +1,535 @@
+/* -*- mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- */
+/*
+ *  Main authors:
+ *     Christian Schulte <schulte@gecode.org>
+ *
+ *  Copyright:
+ *     Christian Schulte, 2004
+ *
+ *  Last modified:
+ *     $Date: 2009-09-30 04:18:44 +1000 (Wed, 30 Sep 2009) $ by $Author: schulte $
+ *     $Revision: 9773 $
+ *
+ *  This file is part of Gecode, the generic constraint
+ *  development environment:
+ *     http://www.gecode.org
+ *
+ *  Permission is hereby granted, free of charge, to any person obtaining
+ *  a copy of this software and associated documentation files (the
+ *  "Software"), to deal in the Software without restriction, including
+ *  without limitation the rights to use, copy, modify, merge, publish,
+ *  distribute, sublicense, and/or sell copies of the Software, and to
+ *  permit persons to whom the Software is furnished to do so, subject to
+ *  the following conditions:
+ *
+ *  The above copyright notice and this permission notice shall be
+ *  included in all copies or substantial portions of the Software.
+ *
+ *  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ *  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ *  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ *  NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ *  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ *  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ *  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <gecode/int.hh>
+
+namespace Gecode { namespace Int { namespace Extensional {
+
+  /**
+   * \brief Sort transition array by input state
+   */
+  class TransByI_State {
+  public:
+    forceinline bool
+    operator ()(const DFA::Transition& x, const DFA::Transition& y) {
+      return x.i_state < y.i_state;
+    }
+    forceinline static void
+    sort(DFA::Transition t[], int n) {
+      TransByI_State tbis;
+      Support::quicksort<DFA::Transition,TransByI_State>(t,n,tbis);
+    }
+  };
+
+  /**
+   * \brief Sort transition array by symbol (value)
+   */
+  class TransBySymbol {
+  public:
+    forceinline bool
+    operator ()(const DFA::Transition& x, const DFA::Transition& y) {
+      return x.symbol < y.symbol;
+    }
+    forceinline static void
+    sort(DFA::Transition t[], int n) {
+      TransBySymbol tbs;
+      Support::quicksort<DFA::Transition,TransBySymbol>(t,n,tbs);
+    }
+  };
+
+  /**
+   * \brief Sort transition array by symbol and then input states
+   */
+  class TransBySymbolI_State {
+  public:
+    forceinline bool
+    operator ()(const DFA::Transition& x, const DFA::Transition& y) {
+      return ((x.symbol < y.symbol) ||
+              ((x.symbol == y.symbol) && (x.i_state < y.i_state)));
+    }
+    forceinline static void
+    sort(DFA::Transition t[], int n) {
+      TransBySymbolI_State tbsi;
+      Support::quicksort<DFA::Transition,TransBySymbolI_State>(t,n,tbsi);
+    }
+  };
+
+  /**
+   * \brief Sort transition array by output state
+   */
+  class TransByO_State {
+  public:
+    forceinline bool
+    operator ()(const DFA::Transition& x, const DFA::Transition& y) {
+      return x.o_state < y.o_state;
+    }
+    forceinline static void
+    sort(DFA::Transition t[], int n) {
+      TransByO_State tbos;
+      Support::quicksort<DFA::Transition,TransByO_State>(t,n,tbos);
+    }
+  };
+
+
+  /**
+   * \brief Stategroup is used to compute a partition of states
+   */
+  class StateGroup {
+  public:
+    int state;
+    int group;
+  };
+
+  /**
+   * \brief Sort groups stated by group and then state
+   */
+  class StateGroupByGroup {
+  public:
+    forceinline bool
+    operator ()(const StateGroup& x, const StateGroup& y) {
+      return ((x.group < y.group) ||
+              ((x.group == y.group) && (x.state < y.state)));
+    }
+    static void
+    sort(StateGroup sg[], int n) {
+      StateGroupByGroup o;
+      Support::quicksort<StateGroup,StateGroupByGroup>(sg,n,o);
+    }
+  };
+
+  /**
+   * \brief %GroupStates is used to index %StateGroup by group
+   */
+  class GroupStates {
+  public:
+    StateGroup* fst;
+    StateGroup* lst;
+  };
+
+  /// Information about states
+  enum StateInfo {
+    SI_NONE       = 0, ///< State is not reachable
+    SI_FROM_START = 1, ///< State is reachable from start state
+    SI_TO_FINAL   = 2, ///< Final state is reachable from state
+    SI_FINAL      = 4  ///< State is final
+  };
+
+}}}
+
+namespace Gecode {
+
+  DFA::DFA(int start, Transition t_spec[], int f_spec[], bool minimize) {
+    using namespace Int;
+    using namespace Extensional;
+    // Compute number of states and transitions
+    int n_states = start;
+    int n_trans  = 0;
+    for (Transition* t = &t_spec[0]; t->i_state >= 0; t++) {
+      n_states = std::max(n_states,t->i_state);
+      n_states = std::max(n_states,t->o_state);
+      n_trans++;
+    }
+    for (int* f = &f_spec[0]; *f >= 0; f++)
+      n_states = std::max(n_states,*f);
+    n_states++;
+
+    // Temporary structure for transitions
+    Transition* trans = heap.alloc<Transition>(n_trans);
+    for (int i = n_trans; i--; )
+      trans[i] = t_spec[i];
+    // Temporary structures for finals
+    int* final = heap.alloc<int>(n_states+1);
+    bool* is_final = heap.alloc<bool>(n_states+1);
+    int n_finals = 0;
+    for (int i = n_states+1; i--; )
+      is_final[i] = false;
+    for (int* f = &f_spec[0]; *f != -1; f++) {
+      is_final[*f]      = true;
+      final[n_finals++] = *f;
+    }
+
+    if (minimize) {
+      // Sort transitions by symbol and i_state
+      TransBySymbolI_State::sort(trans, n_trans);
+      Transition** idx = heap.alloc<Transition*>(n_trans+1);
+      //  idx[i]...idx[i+1]-1 gives where transitions for symbol i start
+      int n_symbols = 0;
+      {
+        int j = 0;
+        while (j < n_trans) {
+          idx[n_symbols++] = &trans[j];
+          int s = trans[j].symbol;
+          while ((j < n_trans) && (s == trans[j].symbol))
+            j++;
+        }
+        idx[n_symbols] = &trans[j];
+        assert(j == n_trans);
+      }
+      // Map states to groups
+      int* s2g = heap.alloc<int>(n_states+1);
+      StateGroup* part = heap.alloc<StateGroup>(n_states+1);
+      GroupStates* g2s = heap.alloc<GroupStates>(n_states+1);
+      // Initialize: final states is group one, all other group zero
+      for (int i = n_states+1; i--; ) {
+        part[i].state = i;
+        part[i].group = is_final[i] ? 1 : 0;
+        s2g[i]        = part[i].group;
+      }
+      // Important: the state n_state is the dead state, conceptually
+      // if there is no transition for a symbol and input state, it is
+      // assumed that there is an implicit transition to n_state
+
+      // Set up the indexing data structure, sort by group
+      StateGroupByGroup::sort(part,n_states+1);
+      int n_groups;
+      if (part[0].group == part[n_states].group) {
+        // No final states, just one group
+        g2s[0].fst = &part[0]; g2s[0].lst = &part[n_states+1];
+        n_groups = 1;
+      } else  {
+        int i = 0;
+        assert(part[0].group == 0);
+        do i++; while (part[i].group == 0);
+        g2s[0].fst = &part[0]; g2s[0].lst = &part[i];
+        g2s[1].fst = &part[i]; g2s[1].lst = &part[n_states+1];
+        n_groups = 2;
+      }
+
+      // Do the refinement
+      {
+        int m_groups;
+        do {
+          m_groups = n_groups;
+          // Iterate over symbols
+          for (int sidx = n_symbols; sidx--; ) {
+            // Iterate over groups
+            for (int g = n_groups; g--; ) {
+              // Ignore singleton groups
+              if (g2s[g].lst-g2s[g].fst > 1) {
+                // Apply transitions to group states
+                // This exploits that both transitions as well as
+                // stategroups are sorted by (input) state
+                Transition* t     = idx[sidx];
+                Transition* t_lst = idx[sidx+1];
+                for (StateGroup* sg = g2s[g].fst; sg<g2s[g].lst; sg++) {
+                  while ((t < t_lst) && (t->i_state < sg->state))
+                    t++;
+                  // Compute group resulting from transition
+                  if ((t < t_lst) && (t->i_state == sg->state))
+                    sg->group = s2g[t->o_state];
+                  else
+                    sg->group = s2g[n_states]; // Go to dead state
+                }
+                // Sort group by groups from transitions
+                StateGroupByGroup::sort(g2s[g].fst,
+                                        static_cast<int>(g2s[g].lst-g2s[g].fst));
+                // Group must be split?
+                if (g2s[g].fst->group != (g2s[g].lst-1)->group) {
+                  // Skip first group
+                  StateGroup* sg = g2s[g].fst+1;
+                  while ((sg-1)->group == sg->group)
+                    sg++;
+                  // Start splitting
+                  StateGroup* lst = g2s[g].lst;
+                  g2s[g].lst = sg;
+                  while (sg < lst) {
+                    s2g[sg->state] = n_groups;
+                    g2s[n_groups].fst  = sg++;
+                    while ((sg < lst) && ((sg-1)->group == sg->group)) {
+                      s2g[sg->state] = n_groups; sg++;
+                    }
+                    g2s[n_groups++].lst = sg;
+                  }
+                }
+              }
+            }
+          }
+        } while (n_groups != m_groups);
+        // New start state
+        start = s2g[start];
+        // Compute new final states
+        n_finals = 0;
+        for (int g = n_groups; g--; )
+          for (StateGroup* sg = g2s[g].fst; sg < g2s[g].lst; sg++)
+            if (is_final[sg->state]) {
+              final[n_finals++] = g;
+              break;
+            }
+        // Compute representatives
+        int* s2r = heap.alloc<int>(n_states+1);
+        for (int i = n_states+1; i--; )
+          s2r[i] = -1;
+        for (int g = n_groups; g--; )
+          s2r[g2s[g].fst->state] = g;
+        // Clean transitions
+        int j = 0;
+        for (int i = 0; i<n_trans; i++)
+          if (s2r[trans[i].i_state] != -1) {
+            trans[j].i_state = s2g[trans[i].i_state];
+            trans[j].symbol  = trans[i].symbol;
+            trans[j].o_state = s2g[trans[i].o_state];
+            j++;
+          }
+        n_trans  = j;
+        n_states = n_groups;
+        heap.free<int>(s2r,n_states+1);
+      }
+      heap.free<GroupStates>(g2s,n_states+1);
+      heap.free<StateGroup>(part,n_states+1);
+      heap.free<int>(s2g,n_states+1);
+      heap.free<Transition*>(idx,n_trans+1);
+    }
+
+    // Do a reachability analysis for all states starting from start state
+    Gecode::Support::StaticStack<int,Heap> visit(heap,n_states);
+    int* state = heap.alloc<int>(n_states);
+    for (int i=n_states; i--; )
+      state[i] = SI_NONE;
+
+    Transition** idx = heap.alloc<Transition*>(n_states+1);
+    {
+      // Sort all transitions according to i_state and create index structure
+      //  idx[i]...idx[i+1]-1 gives where transitions for state i start
+      TransByI_State::sort(trans, n_trans);
+      {
+        int j = 0;
+        for (int i=0; i<n_states; i++) {
+          idx[i] = &trans[j];
+          while ((j < n_trans) && (i == trans[j].i_state))
+            j++;
+        }
+        idx[n_states] = &trans[j];
+        assert(j == n_trans);
+      }
+
+      state[start] = SI_FROM_START;
+      visit.push(start);
+      while (!visit.empty()) {
+        int s = visit.pop();
+        for (Transition* t = idx[s]; t < idx[s+1]; t++)
+          if (!(state[t->o_state] & SI_FROM_START)) {
+            state[t->o_state] |= SI_FROM_START;
+            visit.push(t->o_state);
+          }
+      }
+    }
+
+    // Do a reachability analysis for all states to a final state
+    {
+      // Sort all transitions according to o_state and create index structure
+      //  idx[i]...idx[i+1]-1 gives where transitions for state i start
+      TransByO_State::sort(trans, n_trans);
+      {
+        int j = 0;
+        for (int i=0; i<n_states; i++) {
+          idx[i] = &trans[j];
+          while ((j < n_trans) && (i == trans[j].o_state))
+            j++;
+        }
+        idx[n_states] = &trans[j];
+        assert(j == n_trans);
+      }
+
+      for (int i = n_finals; i--; ) {
+        state[final[i]] |= (SI_TO_FINAL | SI_FINAL);
+        visit.push(final[i]);
+      }
+      while (!visit.empty()) {
+        int s = visit.pop();
+        for (Transition* t = idx[s]; t < idx[s+1]; t++)
+          if (!(state[t->i_state] & SI_TO_FINAL)) {
+            state[t->i_state] |= SI_TO_FINAL;
+            visit.push(t->i_state);
+          }
+      }
+    }
+    heap.free<Transition*>(idx,n_states+1);
+    heap.free<int>(final,n_states+1);
+    heap.free<bool>(is_final,n_states+1);
+
+    // Now all reachable states are known (also the final ones)
+    int* re = heap.alloc<int>(n_states);
+    for (int i = n_states; i--; )
+      re[i] = -1;
+
+    // Renumber states
+    int m_states = 0;
+    // Start state gets zero
+    re[start] = m_states++;
+
+    // Renumber final states
+    for (int i = n_states; i--; )
+      if ((state[i] == (SI_FINAL | SI_FROM_START | SI_TO_FINAL)) && (re[i] < 0))
+        re[i] = m_states++;
+    // If start state is final, final states start from zero, otherwise from one
+    int final_fst = (state[start] & SI_FINAL) ? 0 : 1;
+    int final_lst = m_states;
+    // final_fst...final_lst-1 are the final states
+
+    // Renumber remaining states
+    for (int i = n_states; i--; )
+      if ((state[i] == (SI_FROM_START | SI_TO_FINAL)) && (re[i] < 0))
+        re[i] = m_states++;
+
+    // Count number of remaining transitions
+    int m_trans = 0;
+    for (int i = n_trans; i--; )
+      if ((re[trans[i].i_state] >= 0) && (re[trans[i].o_state] >= 0))
+        m_trans++;
+
+    // All done... Construct the automaton
+    DFAI* d = new DFAI(m_trans);
+    d->n_states  = m_states;
+    d->n_trans   = m_trans;
+    d->final_fst = final_fst;
+    d->final_lst = final_lst;
+    {
+      int j = 0;
+      Transition* r = &d->trans[0];
+      for (int i = 0; i<n_trans; i++)
+        if ((re[trans[i].i_state] >= 0) && (re[trans[i].o_state] >= 0)) {
+          r[j].symbol  = trans[i].symbol;
+          r[j].i_state = re[trans[i].i_state];
+          r[j].o_state = re[trans[i].o_state];
+          j++;
+        }
+      TransBySymbol::sort(r,m_trans);
+    }
+    {
+      // Count number of symbols
+      unsigned int n_symbols = 0;
+      for (int i = 0; i<m_trans; ) {
+        int s = d->trans[i++].symbol;
+        n_symbols++;
+        while ((i<m_trans) && (d->trans[i].symbol == s))
+          i++;
+      }
+      d->n_symbols = n_symbols;
+    }
+    {
+      // Compute maximal degree
+      unsigned int max_degree = 0;
+      unsigned int* deg = heap.alloc<unsigned int>(m_states);
+
+      // Compute in-degree per state
+      for (int i = m_states; i--; )
+        deg[i] = 0;
+      for (int i = m_trans; i--; )
+        deg[d->trans[i].o_state]++;
+      for (int i = m_states; i--; )
+        max_degree = std::max(max_degree,deg[i]);
+
+      // Compute out-degree per state
+      for (int i = m_states; i--; )
+        deg[i] = 0;
+      for (int i = m_trans; i--; )
+        deg[d->trans[i].i_state]++;
+      for (int i = m_states; i--; )
+        max_degree = std::max(max_degree,deg[i]);
+
+      heap.free<unsigned int>(deg,m_states);
+
+      // Compute transitions per symbol
+      {
+        int i=0;
+        while (i < m_trans) {
+          int j=i++;
+          while ((i < m_trans) && 
+                 (d->trans[j].symbol == d->trans[i].symbol))
+            i++;
+          max_degree = std::max(max_degree,static_cast<unsigned int>(i-j));
+        }
+      }
+
+      d->max_degree = max_degree;
+    }
+
+    d->fill();
+    object(d);
+    heap.free<int>(re,n_states);
+    heap.free<int>(state,n_states);
+    heap.free<Transition>(trans,n_trans);
+  }
+
+  SharedHandle::Object*
+  DFA::DFAI::copy(void) const {
+    DFAI* d = new DFAI(n_trans);
+    d->n_states   = n_states;
+    d->n_symbols  = n_symbols;
+    d->n_trans    = n_trans;
+    d->max_degree = max_degree;
+    d->final_fst  = final_fst;
+    d->final_lst  = final_lst;
+    heap.copy<Transition>(&d->trans[0], &trans[0], n_trans);
+    d->fill();
+    return d;
+  }
+
+  void
+  DFA::DFAI::fill(void) {
+    // Compute smallest logarithm larger than n_symbols
+    n_log = 1;
+    while (n_symbols >= static_cast<unsigned int>(1<<n_log))
+      n_log++;
+    // Allocate memory
+    table = heap.alloc<HashEntry>(1<<n_log);
+    // Initialize table
+    for (int i=(1<<n_log); i--; )
+      table[i].fst = table[i].lst = NULL;
+    int mask = (1 << n_log) - 1;
+    // Enter transitions to table
+    for (int i = 0; i<n_trans; ) {
+      int s = trans[i].symbol;
+      Transition* fst = &trans[i];
+      i++;
+      while ((i<n_trans) && (trans[i].symbol == s))
+        i++;
+      Transition* lst = &trans[i];
+      // Enter with linear collision resolution
+      int p = s & mask;
+      while (table[p].fst != NULL)
+        p = (p+1) & mask;
+      table[p].symbol = s;
+      table[p].fst    = fst;
+      table[p].lst    = lst;
+    }
+  }
+
+}
+
+// STATISTICS: int-prop
+