@Override
protected long decode(BytesRef scratch, ByteArrayDataInput tmpInput) {
tmpInput.reset(scratch.bytes);
tmpInput.skipBytes(scratch.length - 4); // suggestion + separator
scratch.length -= 5; // sep + long
return tmpInput.readInt();
}
/* Decodes only the term bytes of the next term. If caller then asks for
metadata, ie docFreq, totalTermFreq or pulls a D/&PEnum, we then (lazily)
decode all metadata up to the current term. */
private BytesRef _next() throws IOException {
// System.out.println("BTR._next seg=" + segment + " this=" + this + " termCount=" +
// state.termBlockOrd + " (vs " + blockTermCount + ")");
if (state.termBlockOrd == blockTermCount && !nextBlock()) {
// System.out.println(" eof");
indexIsCurrent = false;
return null;
}
// TODO: cutover to something better for these ints! simple64?
final int suffix = termSuffixesReader.readVInt();
// System.out.println(" suffix=" + suffix);
term.setLength(termBlockPrefix + suffix);
term.grow(term.length());
termSuffixesReader.readBytes(term.bytes(), termBlockPrefix, suffix);
state.termBlockOrd++;
// NOTE: meaningless in the non-ord case
state.ord++;
// System.out.println(" return term=" + fieldInfo.name + ":" + term.utf8ToString() + " " +
// term + " tbOrd=" + state.termBlockOrd);
return term.get();
}
// Interleaves all output tokens onto the futureOutputs:
private void addOutput(BytesRef bytes, int matchInputLength, int matchEndOffset) {
bytesReader.reset(bytes.bytes, bytes.offset, bytes.length);
final int code = bytesReader.readVInt();
final boolean keepOrig = (code & 0x1) == 0;
final int count = code >>> 1;
// System.out.println(" addOutput count=" + count + " keepOrig=" + keepOrig);
for (int outputIDX = 0; outputIDX < count; outputIDX++) {
synonyms.words.get(bytesReader.readVInt(), scratchBytes);
// System.out.println(" outIDX=" + outputIDX + " bytes=" + scratchBytes.length);
scratchChars.copyUTF8Bytes(scratchBytes);
int lastStart = 0;
final int chEnd = lastStart + scratchChars.length();
int outputUpto = nextRead;
for (int chIDX = lastStart; chIDX <= chEnd; chIDX++) {
if (chIDX == chEnd || scratchChars.charAt(chIDX) == SynonymMap.WORD_SEPARATOR) {
final int outputLen = chIDX - lastStart;
// Caller is not allowed to have empty string in
// the output:
assert outputLen > 0 : "output contains empty string: " + scratchChars;
final int endOffset;
final int posLen;
if (chIDX == chEnd && lastStart == 0) {
// This rule had a single output token, so, we set
// this output's endOffset to the current
// endOffset (ie, endOffset of the last input
// token it matched):
endOffset = matchEndOffset;
posLen = keepOrig ? matchInputLength : 1;
} else {
// This rule has more than one output token; we
// can't pick any particular endOffset for this
// case, so, we inherit the endOffset for the
// input token which this output overlaps:
endOffset = -1;
posLen = 1;
}
futureOutputs[outputUpto].add(
scratchChars.chars(), lastStart, outputLen, endOffset, posLen);
// System.out.println(" " + new String(scratchChars.chars, lastStart, outputLen) + "
// outputUpto=" + outputUpto);
lastStart = 1 + chIDX;
// System.out.println(" slot=" + outputUpto + " keepOrig=" + keepOrig);
outputUpto = rollIncr(outputUpto);
assert futureOutputs[outputUpto].posIncr == 1
: "outputUpto=" + outputUpto + " vs nextWrite=" + nextWrite;
}
}
}
int upto = nextRead;
for (int idx = 0; idx < matchInputLength; idx++) {
futureInputs[upto].keepOrig |= keepOrig;
futureInputs[upto].matched = true;
upto = rollIncr(upto);
}
}
/** decodes the payload at the current position */
protected BytesRef decodePayload(BytesRef scratch, ByteArrayDataInput tmpInput) {
tmpInput.reset(scratch.bytes);
tmpInput.skipBytes(scratch.length - 2); // skip to payload size
short payloadLength = tmpInput.readShort(); // read payload size
tmpInput.setPosition(scratch.length - 2 - payloadLength); // setPosition to start of payload
BytesRef payloadScratch = new BytesRef(payloadLength);
tmpInput.readBytes(payloadScratch.bytes, 0, payloadLength); // read payload
payloadScratch.length = payloadLength;
scratch.length -= 2; // payload length info (short)
scratch.length -= payloadLength; // payload
return payloadScratch;
}
/* Does initial decode of next block of terms; this
doesn't actually decode the docFreq, totalTermFreq,
postings details (frq/prx offset, etc.) metadata;
it just loads them as byte[] blobs which are then
decoded on-demand if the metadata is ever requested
for any term in this block. This enables terms-only
intensive consumes (eg certain MTQs, respelling) to
not pay the price of decoding metadata they won't
use. */
private boolean nextBlock() throws IOException {
// TODO: we still lazy-decode the byte[] for each
// term (the suffix), but, if we decoded
// all N terms up front then seeking could do a fast
// bsearch w/in the block...
// System.out.println("BTR.nextBlock() fp=" + in.getFilePointer() + " this=" + this);
state.blockFilePointer = in.getFilePointer();
blockTermCount = in.readVInt();
// System.out.println(" blockTermCount=" + blockTermCount);
if (blockTermCount == 0) {
return false;
}
termBlockPrefix = in.readVInt();
// term suffixes:
int len = in.readVInt();
if (termSuffixes.length < len) {
termSuffixes = new byte[ArrayUtil.oversize(len, 1)];
}
// System.out.println(" termSuffixes len=" + len);
in.readBytes(termSuffixes, 0, len);
termSuffixesReader.reset(termSuffixes, 0, len);
// docFreq, totalTermFreq
len = in.readVInt();
if (docFreqBytes.length < len) {
docFreqBytes = new byte[ArrayUtil.oversize(len, 1)];
}
// System.out.println(" freq bytes len=" + len);
in.readBytes(docFreqBytes, 0, len);
freqReader.reset(docFreqBytes, 0, len);
// metadata
len = in.readVInt();
if (bytes == null) {
bytes = new byte[ArrayUtil.oversize(len, 1)];
bytesReader = new ByteArrayDataInput();
} else if (bytes.length < len) {
bytes = new byte[ArrayUtil.oversize(len, 1)];
}
in.readBytes(bytes, 0, len);
bytesReader.reset(bytes, 0, len);
metaDataUpto = 0;
state.termBlockOrd = 0;
indexIsCurrent = false;
// System.out.println(" indexIsCurrent=" + indexIsCurrent);
return true;
}
@Override
public void setDocument(int docId) {
bytes = values.get(docId);
in.reset(bytes.bytes, bytes.offset, bytes.length);
if (!in.eof()) {
// first value uses vLong on top of zig-zag encoding, then deltas are encoded using vLong
long previousValue = longs[0] = ByteUtils.zigZagDecode(ByteUtils.readVLong(in));
count = 1;
while (!in.eof()) {
longs = ArrayUtil.grow(longs, count + 1);
previousValue = longs[count++] = previousValue + ByteUtils.readVLong(in);
}
} else {
count = 0;
}
}
// Pushes a frame we seek'd to
IDVersionSegmentTermsEnumFrame pushFrame(
FST.Arc<Pair<BytesRef, Long>> arc, Pair<BytesRef, Long> frameData, int length)
throws IOException {
scratchReader.reset(
frameData.output1.bytes, frameData.output1.offset, frameData.output1.length);
final long code = scratchReader.readVLong();
final long fpSeek = code >>> VersionBlockTreeTermsWriter.OUTPUT_FLAGS_NUM_BITS;
final IDVersionSegmentTermsEnumFrame f = getFrame(1 + currentFrame.ord);
f.maxIDVersion = Long.MAX_VALUE - frameData.output2;
f.hasTerms = (code & VersionBlockTreeTermsWriter.OUTPUT_FLAG_HAS_TERMS) != 0;
f.hasTermsOrig = f.hasTerms;
f.isFloor = (code & VersionBlockTreeTermsWriter.OUTPUT_FLAG_IS_FLOOR) != 0;
if (f.isFloor) {
f.setFloorData(scratchReader, frameData.output1);
}
pushFrame(arc, fpSeek, length);
return f;
}
private void decodeMetaData() throws IOException {
// System.out.println("BTR.decodeMetadata mdUpto=" + metaDataUpto + " vs termCount=" +
// state.termBlockOrd + " state=" + state);
if (!seekPending) {
// TODO: cutover to random-access API
// here.... really stupid that we have to decode N
// wasted term metadata just to get to the N+1th
// that we really need...
// lazily catch up on metadata decode:
final int limit = state.termBlockOrd;
boolean absolute = metaDataUpto == 0;
// TODO: better API would be "jump straight to term=N"???
while (metaDataUpto < limit) {
// System.out.println(" decode mdUpto=" + metaDataUpto);
// TODO: we could make "tiers" of metadata, ie,
// decode docFreq/totalTF but don't decode postings
// metadata; this way caller could get
// docFreq/totalTF w/o paying decode cost for
// postings
// TODO: if docFreq were bulk decoded we could
// just skipN here:
// docFreq, totalTermFreq
state.docFreq = freqReader.readVInt();
// System.out.println(" dF=" + state.docFreq);
if (fieldInfo.getIndexOptions() != IndexOptions.DOCS) {
state.totalTermFreq = state.docFreq + freqReader.readVLong();
// System.out.println(" totTF=" + state.totalTermFreq);
}
// metadata
for (int i = 0; i < longs.length; i++) {
longs[i] = bytesReader.readVLong();
}
postingsReader.decodeTerm(longs, bytesReader, fieldInfo, state, absolute);
metaDataUpto++;
absolute = false;
}
} else {
// System.out.println(" skip! seekPending");
}
}
/** decodes the contexts at the current position */
protected Set<BytesRef> decodeContexts(BytesRef scratch, ByteArrayDataInput tmpInput) {
tmpInput.reset(scratch.bytes);
tmpInput.skipBytes(scratch.length - 2); // skip to context set size
short ctxSetSize = tmpInput.readShort();
scratch.length -= 2;
final Set<BytesRef> contextSet = new HashSet<>();
for (short i = 0; i < ctxSetSize; i++) {
tmpInput.setPosition(scratch.length - 2);
short curContextLength = tmpInput.readShort();
scratch.length -= 2;
tmpInput.setPosition(scratch.length - curContextLength);
BytesRef contextSpare = new BytesRef(curContextLength);
tmpInput.readBytes(contextSpare.bytes, 0, curContextLength);
contextSpare.length = curContextLength;
contextSet.add(contextSpare);
scratch.length -= curContextLength;
}
return contextSet;
}
@SuppressWarnings("unused")
private void printSeekState(PrintStream out) throws IOException {
if (currentFrame == staticFrame) {
out.println(" no prior seek");
} else {
out.println(" prior seek state:");
int ord = 0;
boolean isSeekFrame = true;
while (true) {
IDVersionSegmentTermsEnumFrame f = getFrame(ord);
assert f != null;
final BytesRef prefix = new BytesRef(term.bytes(), 0, f.prefix);
if (f.nextEnt == -1) {
out.println(
" frame "
+ (isSeekFrame ? "(seek)" : "(next)")
+ " ord="
+ ord
+ " fp="
+ f.fp
+ (f.isFloor ? (" (fpOrig=" + f.fpOrig + ")") : "")
+ " prefixLen="
+ f.prefix
+ " prefix="
+ brToString(prefix)
+ (f.nextEnt == -1 ? "" : (" (of " + f.entCount + ")"))
+ " hasTerms="
+ f.hasTerms
+ " isFloor="
+ f.isFloor
+ " code="
+ ((f.fp << VersionBlockTreeTermsWriter.OUTPUT_FLAGS_NUM_BITS)
+ (f.hasTerms ? VersionBlockTreeTermsWriter.OUTPUT_FLAG_HAS_TERMS : 0)
+ (f.isFloor ? VersionBlockTreeTermsWriter.OUTPUT_FLAG_IS_FLOOR : 0))
+ " isLastInFloor="
+ f.isLastInFloor
+ " mdUpto="
+ f.metaDataUpto
+ " tbOrd="
+ f.getTermBlockOrd());
} else {
out.println(
" frame "
+ (isSeekFrame ? "(seek, loaded)" : "(next, loaded)")
+ " ord="
+ ord
+ " fp="
+ f.fp
+ (f.isFloor ? (" (fpOrig=" + f.fpOrig + ")") : "")
+ " prefixLen="
+ f.prefix
+ " prefix="
+ brToString(prefix)
+ " nextEnt="
+ f.nextEnt
+ (f.nextEnt == -1 ? "" : (" (of " + f.entCount + ")"))
+ " hasTerms="
+ f.hasTerms
+ " isFloor="
+ f.isFloor
+ " code="
+ ((f.fp << VersionBlockTreeTermsWriter.OUTPUT_FLAGS_NUM_BITS)
+ (f.hasTerms ? VersionBlockTreeTermsWriter.OUTPUT_FLAG_HAS_TERMS : 0)
+ (f.isFloor ? VersionBlockTreeTermsWriter.OUTPUT_FLAG_IS_FLOOR : 0))
+ " lastSubFP="
+ f.lastSubFP
+ " isLastInFloor="
+ f.isLastInFloor
+ " mdUpto="
+ f.metaDataUpto
+ " tbOrd="
+ f.getTermBlockOrd());
}
if (fr.index != null) {
assert !isSeekFrame || f.arc != null : "isSeekFrame=" + isSeekFrame + " f.arc=" + f.arc;
if (f.prefix > 0 && isSeekFrame && f.arc.label != (term.byteAt(f.prefix - 1) & 0xFF)) {
out.println(
" broken seek state: arc.label="
+ (char) f.arc.label
+ " vs term byte="
+ (char) (term.byteAt(f.prefix - 1) & 0xFF));
throw new RuntimeException("seek state is broken");
}
Pair<BytesRef, Long> output = Util.get(fr.index, prefix);
if (output == null) {
out.println(" broken seek state: prefix is not final in index");
throw new RuntimeException("seek state is broken");
} else if (isSeekFrame && !f.isFloor) {
final ByteArrayDataInput reader =
new ByteArrayDataInput(
output.output1.bytes, output.output1.offset, output.output1.length);
final long codeOrig = reader.readVLong();
final long code =
(f.fp << VersionBlockTreeTermsWriter.OUTPUT_FLAGS_NUM_BITS)
| (f.hasTerms ? VersionBlockTreeTermsWriter.OUTPUT_FLAG_HAS_TERMS : 0)
| (f.isFloor ? VersionBlockTreeTermsWriter.OUTPUT_FLAG_IS_FLOOR : 0);
if (codeOrig != code) {
out.println(
" broken seek state: output code="
+ codeOrig
+ " doesn't match frame code="
+ code);
throw new RuntimeException("seek state is broken");
}
}
}
if (f == currentFrame) {
break;
}
if (f.prefix == validIndexPrefix) {
isSeekFrame = false;
}
ord++;
}
}
}
/** decodes the weight at the current position */
protected long decode(BytesRef scratch, ByteArrayDataInput tmpInput) {
tmpInput.reset(scratch.bytes);
tmpInput.skipBytes(scratch.length - 8); // suggestion
scratch.length -= 8; // long
return tmpInput.readLong();
}
@Override
public void build(TermFreqIterator tfit) throws IOException {
if (tfit instanceof TermFreqPayloadIterator) {
throw new IllegalArgumentException("this suggester doesn't support payloads");
}
File tempInput =
File.createTempFile(
FSTCompletionLookup.class.getSimpleName(), ".input", Sort.defaultTempDir());
File tempSorted =
File.createTempFile(
FSTCompletionLookup.class.getSimpleName(), ".sorted", Sort.defaultTempDir());
Sort.ByteSequencesWriter writer = new Sort.ByteSequencesWriter(tempInput);
Sort.ByteSequencesReader reader = null;
ExternalRefSorter sorter = null;
// Push floats up front before sequences to sort them. For now, assume they are non-negative.
// If negative floats are allowed some trickery needs to be done to find their byte order.
boolean success = false;
try {
byte[] buffer = new byte[0];
ByteArrayDataOutput output = new ByteArrayDataOutput(buffer);
BytesRef spare;
while ((spare = tfit.next()) != null) {
if (spare.length + 4 >= buffer.length) {
buffer = ArrayUtil.grow(buffer, spare.length + 4);
}
output.reset(buffer);
output.writeInt(encodeWeight(tfit.weight()));
output.writeBytes(spare.bytes, spare.offset, spare.length);
writer.write(buffer, 0, output.getPosition());
}
writer.close();
// We don't know the distribution of scores and we need to bucket them, so we'll sort
// and divide into equal buckets.
SortInfo info = new Sort().sort(tempInput, tempSorted);
tempInput.delete();
FSTCompletionBuilder builder =
new FSTCompletionBuilder(
buckets, sorter = new ExternalRefSorter(new Sort()), sharedTailLength);
final int inputLines = info.lines;
reader = new Sort.ByteSequencesReader(tempSorted);
long line = 0;
int previousBucket = 0;
int previousScore = 0;
ByteArrayDataInput input = new ByteArrayDataInput();
BytesRef tmp1 = new BytesRef();
BytesRef tmp2 = new BytesRef();
while (reader.read(tmp1)) {
input.reset(tmp1.bytes);
int currentScore = input.readInt();
int bucket;
if (line > 0 && currentScore == previousScore) {
bucket = previousBucket;
} else {
bucket = (int) (line * buckets / inputLines);
}
previousScore = currentScore;
previousBucket = bucket;
// Only append the input, discard the weight.
tmp2.bytes = tmp1.bytes;
tmp2.offset = input.getPosition();
tmp2.length = tmp1.length - input.getPosition();
builder.add(tmp2, bucket);
line++;
}
// The two FSTCompletions share the same automaton.
this.higherWeightsCompletion = builder.build();
this.normalCompletion =
new FSTCompletion(higherWeightsCompletion.getFST(), false, exactMatchFirst);
success = true;
} finally {
if (success) IOUtils.close(reader, writer, sorter);
else IOUtils.closeWhileHandlingException(reader, writer, sorter);
tempInput.delete();
tempSorted.delete();
}
}
// TODO: we may want an alternate mode here which is
// "if you are about to return NOT_FOUND I won't use
// the terms data from that"; eg FuzzyTermsEnum will
// (usually) just immediately call seek again if we
// return NOT_FOUND so it's a waste for us to fill in
// the term that was actually NOT_FOUND
@Override
public SeekStatus seekCeil(final BytesRef target) throws IOException {
if (indexEnum == null) {
throw new IllegalStateException("terms index was not loaded");
}
// System.out.println("BTR.seek seg=" + segment + " target=" + fieldInfo.name + ":" +
// target.utf8ToString() + " " + target + " current=" + term().utf8ToString() + " " + term()
// + " indexIsCurrent=" + indexIsCurrent + " didIndexNext=" + didIndexNext + " seekPending="
// + seekPending + " divisor=" + indexReader.getDivisor() + " this=" + this);
if (didIndexNext) {
if (nextIndexTerm == null) {
// System.out.println(" nextIndexTerm=null");
} else {
// System.out.println(" nextIndexTerm=" + nextIndexTerm.utf8ToString());
}
}
boolean doSeek = true;
// See if we can avoid seeking, because target term
// is after current term but before next index term:
if (indexIsCurrent) {
final int cmp = BytesRef.getUTF8SortedAsUnicodeComparator().compare(term.get(), target);
if (cmp == 0) {
// Already at the requested term
return SeekStatus.FOUND;
} else if (cmp < 0) {
// Target term is after current term
if (!didIndexNext) {
if (indexEnum.next() == -1) {
nextIndexTerm = null;
} else {
nextIndexTerm = indexEnum.term();
}
// System.out.println(" now do index next() nextIndexTerm=" + (nextIndexTerm == null
// ? "null" : nextIndexTerm.utf8ToString()));
didIndexNext = true;
}
if (nextIndexTerm == null
|| BytesRef.getUTF8SortedAsUnicodeComparator().compare(target, nextIndexTerm) < 0) {
// Optimization: requested term is within the
// same term block we are now in; skip seeking
// (but do scanning):
doSeek = false;
// System.out.println(" skip seek: nextIndexTerm=" + (nextIndexTerm == null ? "null"
// : nextIndexTerm.utf8ToString()));
}
}
}
if (doSeek) {
// System.out.println(" seek");
// Ask terms index to find biggest indexed term (=
// first term in a block) that's <= our text:
in.seek(indexEnum.seek(target));
boolean result = nextBlock();
// Block must exist since, at least, the indexed term
// is in the block:
assert result;
indexIsCurrent = true;
didIndexNext = false;
if (doOrd) {
state.ord = indexEnum.ord() - 1;
}
term.copyBytes(indexEnum.term());
// System.out.println(" seek: term=" + term.utf8ToString());
} else {
// System.out.println(" skip seek");
if (state.termBlockOrd == blockTermCount && !nextBlock()) {
indexIsCurrent = false;
return SeekStatus.END;
}
}
seekPending = false;
int common = 0;
// Scan within block. We could do this by calling
// _next() and testing the resulting term, but this
// is wasteful. Instead, we first confirm the
// target matches the common prefix of this block,
// and then we scan the term bytes directly from the
// termSuffixesreader's byte[], saving a copy into
// the BytesRef term per term. Only when we return
// do we then copy the bytes into the term.
while (true) {
// First, see if target term matches common prefix
// in this block:
if (common < termBlockPrefix) {
final int cmp =
(term.byteAt(common) & 0xFF) - (target.bytes[target.offset + common] & 0xFF);
if (cmp < 0) {
// TODO: maybe we should store common prefix
// in block header? (instead of relying on
// last term of previous block)
// Target's prefix is after the common block
// prefix, so term cannot be in this block
// but it could be in next block. We
// must scan to end-of-block to set common
// prefix for next block:
if (state.termBlockOrd < blockTermCount) {
while (state.termBlockOrd < blockTermCount - 1) {
state.termBlockOrd++;
state.ord++;
termSuffixesReader.skipBytes(termSuffixesReader.readVInt());
}
final int suffix = termSuffixesReader.readVInt();
term.setLength(termBlockPrefix + suffix);
term.grow(term.length());
termSuffixesReader.readBytes(term.bytes(), termBlockPrefix, suffix);
}
state.ord++;
if (!nextBlock()) {
indexIsCurrent = false;
return SeekStatus.END;
}
common = 0;
} else if (cmp > 0) {
// Target's prefix is before the common prefix
// of this block, so we position to start of
// block and return NOT_FOUND:
assert state.termBlockOrd == 0;
final int suffix = termSuffixesReader.readVInt();
term.setLength(termBlockPrefix + suffix);
term.grow(term.length());
termSuffixesReader.readBytes(term.bytes(), termBlockPrefix, suffix);
return SeekStatus.NOT_FOUND;
} else {
common++;
}
continue;
}
// Test every term in this block
while (true) {
state.termBlockOrd++;
state.ord++;
final int suffix = termSuffixesReader.readVInt();
// We know the prefix matches, so just compare the new suffix:
final int termLen = termBlockPrefix + suffix;
int bytePos = termSuffixesReader.getPosition();
boolean next = false;
final int limit = target.offset + (termLen < target.length ? termLen : target.length);
int targetPos = target.offset + termBlockPrefix;
while (targetPos < limit) {
final int cmp = (termSuffixes[bytePos++] & 0xFF) - (target.bytes[targetPos++] & 0xFF);
if (cmp < 0) {
// Current term is still before the target;
// keep scanning
next = true;
break;
} else if (cmp > 0) {
// Done! Current term is after target. Stop
// here, fill in real term, return NOT_FOUND.
term.setLength(termBlockPrefix + suffix);
term.grow(term.length());
termSuffixesReader.readBytes(term.bytes(), termBlockPrefix, suffix);
// System.out.println(" NOT_FOUND");
return SeekStatus.NOT_FOUND;
}
}
if (!next && target.length <= termLen) {
term.setLength(termBlockPrefix + suffix);
term.grow(term.length());
termSuffixesReader.readBytes(term.bytes(), termBlockPrefix, suffix);
if (target.length == termLen) {
// Done! Exact match. Stop here, fill in
// real term, return FOUND.
// System.out.println(" FOUND");
return SeekStatus.FOUND;
} else {
// System.out.println(" NOT_FOUND");
return SeekStatus.NOT_FOUND;
}
}
if (state.termBlockOrd == blockTermCount) {
// Must pre-fill term for next block's common prefix
term.setLength(termBlockPrefix + suffix);
term.grow(term.length());
termSuffixesReader.readBytes(term.bytes(), termBlockPrefix, suffix);
break;
} else {
termSuffixesReader.skipBytes(suffix);
}
}
// The purpose of the terms dict index is to seek
// the enum to the closest index term before the
// term we are looking for. So, we should never
// cross another index term (besides the first
// one) while we are scanning:
assert indexIsCurrent;
if (!nextBlock()) {
// System.out.println(" END");
indexIsCurrent = false;
return SeekStatus.END;
}
common = 0;
}
}