/**
* Sets the enum to operate in linear fashion, as we have found a looping transition at position:
* we set an upper bound and act like a TermRangeQuery for this portion of the term space.
*/
private void setLinear(int position) {
assert linear == false;
int state = runAutomaton.getInitialState();
int maxInterval = 0xff;
for (int i = 0; i < position; i++) {
state = runAutomaton.step(state, seekBytesRef.bytes[i] & 0xff);
assert state >= 0 : "state=" + state;
}
for (int i = 0; i < allTransitions[state].length; i++) {
Transition t = allTransitions[state][i];
if (t.getMin() <= (seekBytesRef.bytes[position] & 0xff)
&& (seekBytesRef.bytes[position] & 0xff) <= t.getMax()) {
maxInterval = t.getMax();
break;
}
}
// 0xff terms don't get the optimization... not worth the trouble.
if (maxInterval != 0xff) maxInterval++;
int length = position + 1; /* position + maxTransition */
if (linearUpperBound.bytes.length < length) linearUpperBound.bytes = new byte[length];
System.arraycopy(seekBytesRef.bytes, 0, linearUpperBound.bytes, 0, position);
linearUpperBound.bytes[position] = (byte) maxInterval;
linearUpperBound.length = length;
linear = true;
}
@Override
protected BytesRef nextSeekTerm(final BytesRef term) throws IOException {
// System.out.println("ATE.nextSeekTerm term=" + term);
if (term == null) {
assert seekBytesRef.length == 0;
// return the empty term, as its valid
if (runAutomaton.isAccept(runAutomaton.getInitialState())) {
return seekBytesRef;
}
} else {
seekBytesRef.copyBytes(term);
}
// seek to the next possible string;
if (nextString()) {
return seekBytesRef; // reposition
} else {
return null; // no more possible strings can match
}
}
/**
* Increments the byte buffer to the next String in binary order after s that will not put the
* machine into a reject state. If such a string does not exist, returns false.
*
* <p>The correctness of this method depends upon the automaton being deterministic, and having no
* transitions to dead states.
*
* @return true if more possible solutions exist for the DFA
*/
private boolean nextString() {
int state;
int pos = 0;
savedStates.grow(seekBytesRef.length + 1);
final int[] states = savedStates.ints;
states[0] = runAutomaton.getInitialState();
while (true) {
curGen++;
linear = false;
// walk the automaton until a character is rejected.
for (state = states[pos]; pos < seekBytesRef.length; pos++) {
visited[state] = curGen;
int nextState = runAutomaton.step(state, seekBytesRef.bytes[pos] & 0xff);
if (nextState == -1) break;
states[pos + 1] = nextState;
// we found a loop, record it for faster enumeration
if (!finite && !linear && visited[nextState] == curGen) {
setLinear(pos);
}
state = nextState;
}
// take the useful portion, and the last non-reject state, and attempt to
// append characters that will match.
if (nextString(state, pos)) {
return true;
} else {
/* no more solutions exist from this useful portion, backtrack */
if ((pos = backtrack(pos)) < 0) /* no more solutions at all */ return false;
final int newState = runAutomaton.step(states[pos], seekBytesRef.bytes[pos] & 0xff);
if (newState >= 0 && runAutomaton.isAccept(newState))
/* String is good to go as-is */
return true;
/* else advance further */
// TODO: paranoia? if we backtrack thru an infinite DFA, the loop detection is important!
// for now, restart from scratch for all infinite DFAs
if (!finite) pos = 0;
}
}
}
/** Load frame for start arc(node) on fst */
Frame loadFirstFrame(Frame frame) throws IOException {
frame.fstArc = fst.getFirstArc(frame.fstArc);
frame.fsaState = fsa.getInitialState();
return frame;
}