/// <summary>
/// Minimizes the given automaton using Hopcroft's algorithm.
/// </summary>
public static void MinimizeHopcroft(Automaton a)
{
a.Determinize();
if (a.Initial.numTransitions == 1)
{
Transition t = a.Initial.TransitionsArray[0];
if (t.To == a.Initial && t.Min_Renamed == Character.MIN_CODE_POINT && t.Max_Renamed == Character.MAX_CODE_POINT)
{
return;
}
}
a.Totalize();
// initialize data structures
int[] sigma = a.StartPoints;
State[] states = a.NumberedStates;
int sigmaLen = sigma.Length, statesLen = states.Length;
List <State>[,] reverse = new List <State> [statesLen, sigmaLen];
HashSet <State>[] partition = new HashSet <State> [statesLen];
List <State>[] splitblock = new List <State> [statesLen];
int[] block = new int[statesLen];
StateList[,] active = new StateList[statesLen, sigmaLen];
StateListNode[,] active2 = new StateListNode[statesLen, sigmaLen];
LinkedList <IntPair> pending = new LinkedList <IntPair>();
BitArray pending2 = new BitArray(sigmaLen * statesLen);
BitArray split = new BitArray(statesLen), refine = new BitArray(statesLen), refine2 = new BitArray(statesLen);
for (int q = 0; q < statesLen; q++)
{
splitblock[q] = new List <State>();
partition[q] = new HashSet <State>();
for (int x = 0; x < sigmaLen; x++)
{
active[q, x] = new StateList();
}
}
// find initial partition and reverse edges
for (int q = 0; q < statesLen; q++)
{
State qq = states[q];
int j = qq.accept ? 0 : 1;
partition[j].Add(qq);
block[q] = j;
for (int x = 0; x < sigmaLen; x++)
{
//List<State>[] r = reverse[qq.Step(sigma[x]).number];
var r = qq.Step(sigma[x]).number;
if (reverse[r, x] == null)
{
reverse[r, x] = new List <State>();
}
reverse[r, x].Add(qq);
}
}
// initialize active sets
for (int j = 0; j <= 1; j++)
{
for (int x = 0; x < sigmaLen; x++)
{
foreach (State qq in partition[j])
{
if (reverse[qq.number, x] != null)
{
active2[qq.number, x] = active[j, x].Add(qq);
}
}
}
}
// initialize pending
for (int x = 0; x < sigmaLen; x++)
{
int j = (active[0, x].Size <= active[1, x].Size) ? 0 : 1;
pending.AddLast(new IntPair(j, x));
pending2.Set(x * statesLen + j, true);
}
// process pending until fixed point
int k = 2;
while (pending.Count > 0)
{
IntPair ip = pending.First.Value;
pending.RemoveFirst();
int p = ip.N1;
int x = ip.N2;
pending2.Set(x * statesLen + p, false);
// find states that need to be split off their blocks
for (StateListNode m = active[p, x].First; m != null; m = m.Next)
{
List <State> r = reverse[m.q.number, x];
if (r != null)
{
foreach (State s in r)
{
int i = s.number;
if (!split.Get(i))
{
split.Set(i, true);
int j = block[i];
splitblock[j].Add(s);
if (!refine2.Get(j))
{
refine2.Set(j, true);
refine.Set(j, true);
}
}
}
}
}
// refine blocks
for (int j = Number.NextSetBit(refine, 0); j >= 0; j = Number.NextSetBit(refine, j + 1))
{
List <State> sb = splitblock[j];
if (sb.Count < partition[j].Count)
{
HashSet <State> b1 = partition[j];
HashSet <State> b2 = partition[k];
foreach (State s in sb)
{
b1.Remove(s);
b2.Add(s);
block[s.number] = k;
for (int c = 0; c < sigmaLen; c++)
{
StateListNode sn = active2[s.number, c];
if (sn != null && sn.Sl == active[j, c])
{
sn.Remove();
active2[s.number, c] = active[k, c].Add(s);
}
}
}
// update pending
for (int c = 0; c < sigmaLen; c++)
{
int aj = active[j, c].Size, ak = active[k, c].Size, ofs = c * statesLen;
if (!pending2.Get(ofs + j) && 0 < aj && aj <= ak)
{
pending2.Set(ofs + j, true);
pending.AddLast(new IntPair(j, c));
}
else
{
pending2.Set(ofs + k, true);
pending.AddLast(new IntPair(k, c));
}
}
k++;
}
refine2.Set(j, false);
foreach (State s in sb)
{
split.Set(s.number, false);
}
sb.Clear();
}
refine.SetAll(false);
}
// make a new state for each equivalence class, set initial state
State[] newstates = new State[k];
for (int n = 0; n < newstates.Length; n++)
{
State s = new State();
newstates[n] = s;
foreach (State q in partition[n])
{
if (q == a.Initial)
{
a.Initial = s;
}
s.accept = q.accept;
s.number = q.number; // select representative
q.number = n;
}
}
// build transitions and set acceptance
for (int n = 0; n < newstates.Length; n++)
{
State s = newstates[n];
s.accept = states[s.number].accept;
foreach (Transition t in states[s.number].Transitions)
{
s.AddTransition(new Transition(t.Min_Renamed, t.Max_Renamed, newstates[t.To.number]));
}
}
a.ClearNumberedStates();
a.RemoveDeadTransitions();
}
public int[] GetRandomAcceptedString(Random r)
{
IList <int?> soFar = new List <int?>();
if (a.IsSingleton)
{
// accepts only one
var s = a.Singleton;
int charUpto = 0;
while (charUpto < s.Length)
{
int cp = s.CodePointAt(charUpto);
charUpto += Character.CharCount(cp);
soFar.Add(cp);
}
}
else
{
var s = a.initial;
while (true)
{
if (s.accept)
{
if (s.numTransitions == 0)
{
// stop now
break;
}
else
{
if (r.NextBoolean())
{
break;
}
}
}
if (s.numTransitions == 0)
{
throw new Exception("this automaton has dead states");
}
bool cheat = r.NextBoolean();
Transition t;
if (cheat)
{
// pick a transition that we know is the fastest
// path to an accept state
IList <Transition> toAccept = new List <Transition>();
for (int i = 0; i < s.numTransitions; i++)
{
Transition t0 = s.TransitionsArray[i];
if (leadsToAccept.ContainsKey(t0))
{
toAccept.Add(t0);
}
}
if (toAccept.Count == 0)
{
// this is OK -- it means we jumped into a cycle
t = s.TransitionsArray[r.Next(s.numTransitions)];
}
else
{
t = toAccept[r.Next(toAccept.Count)];
}
}
else
{
t = s.TransitionsArray[r.Next(s.numTransitions)];
}
soFar.Add(AutomatonTestUtil.GetRandomCodePoint(r, t));
s = t.to;
}
}
return(ArrayUtil.ToInt32Array(soFar));
}
public ArrivingTransition(State from, Transition t)
{
this.from = from;
this.t = t;
}
/// <summary>
/// Determinizes the given automaton.
/// <para/>
/// Worst case complexity: exponential in number of states.
/// </summary>
public static void Determinize(Automaton a)
{
if (a.IsDeterministic || a.IsSingleton)
{
return;
}
State[] allStates = a.GetNumberedStates();
// subset construction
bool initAccept = a.initial.accept;
int initNumber = a.initial.number;
a.initial = new State();
SortedInt32Set.FrozenInt32Set initialset = new SortedInt32Set.FrozenInt32Set(initNumber, a.initial);
LinkedList <SortedInt32Set.FrozenInt32Set> worklist = new LinkedList <SortedInt32Set.FrozenInt32Set>();
IDictionary <SortedInt32Set.FrozenInt32Set, State> newstate = new Dictionary <SortedInt32Set.FrozenInt32Set, State>();
worklist.AddLast(initialset);
a.initial.accept = initAccept;
newstate[initialset] = a.initial;
int newStateUpto = 0;
State[] newStatesArray = new State[5];
newStatesArray[newStateUpto] = a.initial;
a.initial.number = newStateUpto;
newStateUpto++;
// like Set<Integer,PointTransitions>
PointTransitionSet points = new PointTransitionSet();
// like SortedMap<Integer,Integer>
SortedInt32Set statesSet = new SortedInt32Set(5);
// LUCENENET NOTE: The problem here is almost certainly
// due to the conversion to FrozenIntSet along with its
// differing equality checking.
while (worklist.Count > 0)
{
SortedInt32Set.FrozenInt32Set s = worklist.First.Value;
worklist.Remove(s);
// Collate all outgoing transitions by min/1+max:
for (int i = 0; i < s.values.Length; i++)
{
State s0 = allStates[s.values[i]];
for (int j = 0; j < s0.numTransitions; j++)
{
points.Add(s0.TransitionsArray[j]);
}
}
if (points.count == 0)
{
// No outgoing transitions -- skip it
continue;
}
points.Sort();
int lastPoint = -1;
int accCount = 0;
State r = s.state;
for (int i = 0; i < points.count; i++)
{
int point = points.points[i].point;
if (statesSet.upto > 0)
{
Debug.Assert(lastPoint != -1);
statesSet.ComputeHash();
State q;
if (!newstate.TryGetValue(statesSet.ToFrozenInt32Set(), out q) || q == null)
{
q = new State();
SortedInt32Set.FrozenInt32Set p = statesSet.Freeze(q);
worklist.AddLast(p);
if (newStateUpto == newStatesArray.Length)
{
State[] newArray = new State[ArrayUtil.Oversize(1 + newStateUpto, RamUsageEstimator.NUM_BYTES_OBJECT_REF)];
Array.Copy(newStatesArray, 0, newArray, 0, newStateUpto);
newStatesArray = newArray;
}
newStatesArray[newStateUpto] = q;
q.number = newStateUpto;
newStateUpto++;
q.accept = accCount > 0;
newstate[p] = q;
}
else
{
Debug.Assert((accCount > 0) == q.accept, "accCount=" + accCount + " vs existing accept=" + q.accept + " states=" + statesSet);
}
r.AddTransition(new Transition(lastPoint, point - 1, q));
}
// process transitions that end on this point
// (closes an overlapping interval)
Transition[] transitions = points.points[i].ends.transitions;
int limit = points.points[i].ends.count;
for (int j = 0; j < limit; j++)
{
Transition t = transitions[j];
int num = t.to.number;
statesSet.Decr(num);
accCount -= t.to.accept ? 1 : 0;
}
points.points[i].ends.count = 0;
// process transitions that start on this point
// (opens a new interval)
transitions = points.points[i].starts.transitions;
limit = points.points[i].starts.count;
for (int j = 0; j < limit; j++)
{
Transition t = transitions[j];
int num = t.to.number;
statesSet.Incr(num);
accCount += t.to.accept ? 1 : 0;
}
lastPoint = point;
points.points[i].starts.count = 0;
}
points.Reset();
Debug.Assert(statesSet.upto == 0, "upto=" + statesSet.upto);
}
a.deterministic = true;
a.SetNumberedStates(newStatesArray, newStateUpto);
}
public void Add(Transition t)
{
Find(t.min).starts.Add(t);
Find(1 + t.max).ends.Add(t);
}
//private static final boolean DEBUG = BlockTreeTermsWriter.DEBUG;
private BytesRef AddTail(int state, BytesRef term, int idx, int leadLabel)
{
// Find biggest transition that's < label
// TODO: use binary search here
Transition maxTransition = null;
foreach (Transition transition in sortedTransitions[state])
{
if (transition.min < leadLabel)
{
maxTransition = transition;
}
}
Debug.Assert(maxTransition != null);
// Append floorLabel
int floorLabel;
if (maxTransition.max > leadLabel - 1)
{
floorLabel = leadLabel - 1;
}
else
{
floorLabel = maxTransition.max;
}
if (idx >= term.Bytes.Length)
{
term.Grow(1 + idx);
}
//if (DEBUG) System.out.println(" add floorLabel=" + (char) floorLabel + " idx=" + idx);
term.Bytes[idx] = (byte)floorLabel;
state = maxTransition.to.Number;
idx++;
// Push down to last accept state
while (true)
{
Transition[] transitions = sortedTransitions[state];
if (transitions.Length == 0)
{
Debug.Assert(RunAutomaton.IsAccept(state));
term.Length = idx;
//if (DEBUG) System.out.println(" return " + term.utf8ToString());
return(term);
}
else
{
// We are pushing "top" -- so get last label of
// last transition:
Debug.Assert(transitions.Length != 0);
Transition lastTransition = transitions[transitions.Length - 1];
if (idx >= term.Bytes.Length)
{
term.Grow(1 + idx);
}
//if (DEBUG) System.out.println(" push maxLabel=" + (char) lastTransition.max + " idx=" + idx);
term.Bytes[idx] = (byte)lastTransition.max;
state = lastTransition.to.Number;
idx++;
}
}
}
