/** Seeks to largest term that's <= target. */
protected void doSeekFloor() throws IOException {
// TODO: possibly caller could/should provide common
// prefix length? ie this work may be redundant if
// caller is in fact intersecting against its own
// automaton
// System.out.println("FE: seek floor upto=" + upto);
// Save CPU by starting at the end of the shared prefix
// b/w our current term & the target:
rewindPrefix();
// System.out.println("FE: after rewind upto=" + upto);
FST.Arc<T> arc = getArc(upto);
int targetLabel = getTargetLabel();
// System.out.println("FE: init targetLabel=" + targetLabel);
// Now scan forward, matching the new suffix of the target
while (true) {
// System.out.println(" cycle upto=" + upto + " arc.label=" + arc.label + " (" + (char)
// arc.label + ") targetLabel=" + targetLabel + " isLast?=" + arc.isLast());
if (arc.bytesPerArc != 0 && arc.label != FST.END_LABEL) {
// Arcs are fixed array -- use binary search to find
// the target.
final FST<T>.BytesReader in = fst.getBytesReader(0);
int low = arc.arcIdx;
int high = arc.numArcs - 1;
int mid = 0;
// System.out.println("do arc array low=" + low + " high=" + high + " targetLabel=" +
// targetLabel);
boolean found = false;
while (low <= high) {
mid = (low + high) >>> 1;
in.pos = arc.posArcsStart - arc.bytesPerArc * mid - 1;
final int midLabel = fst.readLabel(in);
final int cmp = midLabel - targetLabel;
// System.out.println(" cycle low=" + low + " high=" + high + " mid=" + mid + "
// midLabel=" + midLabel + " cmp=" + cmp);
if (cmp < 0) low = mid + 1;
else if (cmp > 0) high = mid - 1;
else {
found = true;
break;
}
}
// NOTE: this code is dup'd w/ the code below (in
// the outer else clause):
if (found) {
// Match -- recurse
// System.out.println(" match! arcIdx=" + mid);
arc.arcIdx = mid - 1;
fst.readNextRealArc(arc, in);
assert arc.arcIdx == mid;
assert arc.label == targetLabel
: "arc.label=" + arc.label + " vs targetLabel=" + targetLabel + " mid=" + mid;
output[upto] = fst.outputs.add(output[upto - 1], arc.output);
if (targetLabel == FST.END_LABEL) {
return;
}
setCurrentLabel(arc.label);
incr();
arc = fst.readFirstTargetArc(arc, getArc(upto));
targetLabel = getTargetLabel();
continue;
} else if (high == -1) {
// System.out.println(" before first");
// Very first arc is after our target
// TODO: if each arc could somehow read the arc just
// before, we can save this re-scan. The ceil case
// doesn't need this because it reads the next arc
// instead:
while (true) {
// First, walk backwards until we find a first arc
// that's before our target label:
fst.readFirstTargetArc(getArc(upto - 1), arc);
if (arc.label < targetLabel) {
// Then, scan forwards to the arc just before
// the targetLabel:
while (!arc.isLast() && fst.readNextArcLabel(arc) < targetLabel) {
fst.readNextArc(arc);
}
pushLast();
return;
}
upto--;
if (upto == 0) {
return;
}
targetLabel = getTargetLabel();
arc = getArc(upto);
}
} else {
// There is a floor arc:
arc.arcIdx = (low > high ? high : low) - 1;
// System.out.println(" hasFloor arcIdx=" + (arc.arcIdx+1));
fst.readNextRealArc(arc, in);
assert arc.isLast() || fst.readNextArcLabel(arc) > targetLabel;
assert arc.label < targetLabel;
pushLast();
return;
}
} else {
if (arc.label == targetLabel) {
// Match -- recurse
output[upto] = fst.outputs.add(output[upto - 1], arc.output);
if (targetLabel == FST.END_LABEL) {
return;
}
setCurrentLabel(arc.label);
incr();
arc = fst.readFirstTargetArc(arc, getArc(upto));
targetLabel = getTargetLabel();
} else if (arc.label > targetLabel) {
// TODO: if each arc could somehow read the arc just
// before, we can save this re-scan. The ceil case
// doesn't need this because it reads the next arc
// instead:
while (true) {
// First, walk backwards until we find a first arc
// that's before our target label:
fst.readFirstTargetArc(getArc(upto - 1), arc);
if (arc.label < targetLabel) {
// Then, scan forwards to the arc just before
// the targetLabel:
while (!arc.isLast() && fst.readNextArcLabel(arc) < targetLabel) {
fst.readNextArc(arc);
}
pushLast();
return;
}
upto--;
if (upto == 0) {
return;
}
targetLabel = getTargetLabel();
arc = getArc(upto);
}
} else if (!arc.isLast()) {
// System.out.println(" check next label=" + fst.readNextArcLabel(arc) + " (" + (char)
// fst.readNextArcLabel(arc) + ")");
if (fst.readNextArcLabel(arc) > targetLabel) {
pushLast();
return;
} else {
// keep scanning
fst.readNextArc(arc);
}
} else {
pushLast();
return;
}
}
}
}
/** Seeks to smallest term that's >= target. */
protected void doSeekCeil() throws IOException {
// System.out.println(" advance len=" + target.length + " curlen=" + current.length);
// TODO: possibly caller could/should provide common
// prefix length? ie this work may be redundant if
// caller is in fact intersecting against its own
// automaton
// System.out.println("FE.seekCeil upto=" + upto);
// Save time by starting at the end of the shared prefix
// b/w our current term & the target:
rewindPrefix();
// System.out.println(" after rewind upto=" + upto);
FST.Arc<T> arc = getArc(upto);
int targetLabel = getTargetLabel();
// System.out.println(" init targetLabel=" + targetLabel);
// Now scan forward, matching the new suffix of the target
while (true) {
// System.out.println(" cycle upto=" + upto + " arc.label=" + arc.label + " (" + (char)
// arc.label + ") vs targetLabel=" + targetLabel);
if (arc.bytesPerArc != 0 && arc.label != -1) {
// Arcs are fixed array -- use binary search to find
// the target.
final FST<T>.BytesReader in = fst.getBytesReader(0);
int low = arc.arcIdx;
int high = arc.numArcs - 1;
int mid = 0;
// System.out.println("do arc array low=" + low + " high=" + high + " targetLabel=" +
// targetLabel);
boolean found = false;
while (low <= high) {
mid = (low + high) >>> 1;
in.pos = arc.posArcsStart - arc.bytesPerArc * mid - 1;
final int midLabel = fst.readLabel(in);
final int cmp = midLabel - targetLabel;
// System.out.println(" cycle low=" + low + " high=" + high + " mid=" + mid + "
// midLabel=" + midLabel + " cmp=" + cmp);
if (cmp < 0) low = mid + 1;
else if (cmp > 0) high = mid - 1;
else {
found = true;
break;
}
}
// NOTE: this code is dup'd w/ the code below (in
// the outer else clause):
if (found) {
// Match
arc.arcIdx = mid - 1;
fst.readNextRealArc(arc, in);
assert arc.arcIdx == mid;
assert arc.label == targetLabel
: "arc.label=" + arc.label + " vs targetLabel=" + targetLabel + " mid=" + mid;
output[upto] = fst.outputs.add(output[upto - 1], arc.output);
if (targetLabel == FST.END_LABEL) {
return;
}
setCurrentLabel(arc.label);
incr();
arc = fst.readFirstTargetArc(arc, getArc(upto));
targetLabel = getTargetLabel();
continue;
} else if (low == arc.numArcs) {
// Dead end
arc.arcIdx = arc.numArcs - 2;
fst.readNextRealArc(arc, in);
assert arc.isLast();
// Dead end (target is after the last arc);
// rollback to last fork then push
upto--;
while (true) {
if (upto == 0) {
return;
}
final FST.Arc<T> prevArc = getArc(upto);
// System.out.println(" rollback upto=" + upto + " arc.label=" + prevArc.label + "
// isLast?=" + prevArc.isLast());
if (!prevArc.isLast()) {
fst.readNextArc(prevArc);
pushFirst();
return;
}
upto--;
}
} else {
arc.arcIdx = (low > high ? low : high) - 1;
fst.readNextRealArc(arc, in);
assert arc.label > targetLabel;
pushFirst();
return;
}
} else {
// Arcs are not array'd -- must do linear scan:
if (arc.label == targetLabel) {
// recurse
output[upto] = fst.outputs.add(output[upto - 1], arc.output);
if (targetLabel == FST.END_LABEL) {
return;
}
setCurrentLabel(arc.label);
incr();
arc = fst.readFirstTargetArc(arc, getArc(upto));
targetLabel = getTargetLabel();
} else if (arc.label > targetLabel) {
pushFirst();
return;
} else if (arc.isLast()) {
// Dead end (target is after the last arc);
// rollback to last fork then push
upto--;
while (true) {
if (upto == 0) {
return;
}
final FST.Arc<T> prevArc = getArc(upto);
// System.out.println(" rollback upto=" + upto + " arc.label=" + prevArc.label + "
// isLast?=" + prevArc.isLast());
if (!prevArc.isLast()) {
fst.readNextArc(prevArc);
pushFirst();
return;
}
upto--;
}
} else {
// keep scanning
// System.out.println(" next scan");
fst.readNextArc(arc);
}
}
}
}
