// we don't actually write a .fdx-like index, instead we read the
// stored fields file in entirety up-front and save the offsets
// so we can seek to the documents later.
private void readIndex(int size) throws IOException {
ChecksumIndexInput input = new BufferedChecksumIndexInput(in);
offsets = new long[size];
int upto = 0;
while (!scratch.get().equals(END)) {
SimpleTextUtil.readLine(input, scratch);
if (StringHelper.startsWith(scratch.get(), DOC)) {
offsets[upto] = input.getFilePointer();
upto++;
}
}
SimpleTextUtil.checkFooter(input);
assert upto == offsets.length;
}
private void readField(BytesRef type, FieldInfo fieldInfo, StoredFieldVisitor visitor)
throws IOException {
readLine();
assert StringHelper.startsWith(scratch.get(), VALUE);
if (type == TYPE_STRING) {
byte[] bytes = new byte[scratch.length() - VALUE.length];
System.arraycopy(scratch.bytes(), VALUE.length, bytes, 0, bytes.length);
visitor.stringField(fieldInfo, bytes);
} else if (type == TYPE_BINARY) {
byte[] copy = new byte[scratch.length() - VALUE.length];
System.arraycopy(scratch.bytes(), VALUE.length, copy, 0, copy.length);
visitor.binaryField(fieldInfo, copy);
} else if (type == TYPE_INT) {
scratchUTF16.copyUTF8Bytes(scratch.bytes(), VALUE.length, scratch.length() - VALUE.length);
visitor.intField(fieldInfo, Integer.parseInt(scratchUTF16.toString()));
} else if (type == TYPE_LONG) {
scratchUTF16.copyUTF8Bytes(scratch.bytes(), VALUE.length, scratch.length() - VALUE.length);
visitor.longField(fieldInfo, Long.parseLong(scratchUTF16.toString()));
} else if (type == TYPE_FLOAT) {
scratchUTF16.copyUTF8Bytes(scratch.bytes(), VALUE.length, scratch.length() - VALUE.length);
visitor.floatField(fieldInfo, Float.parseFloat(scratchUTF16.toString()));
} else if (type == TYPE_DOUBLE) {
scratchUTF16.copyUTF8Bytes(scratch.bytes(), VALUE.length, scratch.length() - VALUE.length);
visitor.doubleField(fieldInfo, Double.parseDouble(scratchUTF16.toString()));
}
}
@Override
public BytesRef next() throws IOException {
boolean success = false;
if (done) {
return null;
}
try {
ByteArrayDataInput input = new ByteArrayDataInput();
if (reader.read(scratch)) {
final BytesRef bytes = scratch.get();
weight = decode(bytes, input);
if (hasPayloads) {
payload = decodePayload(bytes, input);
}
if (hasContexts) {
contexts = decodeContexts(bytes, input);
}
success = true;
return bytes;
}
close();
success = done = true;
return null;
} finally {
if (!success) {
done = true;
close();
}
}
}
@Override
public BytesRef indexedValueForSearch(Object value) {
long longValue = NumericUtils.doubleToSortableLong(parseDoubleValue(value));
BytesRefBuilder bytesRef = new BytesRefBuilder();
NumericUtils.longToPrefixCoded(longValue, 0, bytesRef); // 0 because of exact match
return bytesRef.get();
}
/* 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();
}
@Override
public BytesRef indexedValueForSearch(Object value) {
int intValue = NumericUtils.floatToSortableInt(parseValue(value));
BytesRefBuilder bytesRef = new BytesRefBuilder();
NumericUtils.intToPrefixCoded(intValue, 0, bytesRef); // 0 because of exact match
return bytesRef.get();
}
@Override
public BytesRef indexedValueForSearch(Object value) {
BytesRefBuilder bytesRef = new BytesRefBuilder();
LegacyNumericUtils.intToPrefixCoded(
parseValue(value), 0, bytesRef); // 0 because of exact match
return bytesRef.get();
}
public void testRandomReads() throws IOException {
int length = randomIntBetween(10, scaledRandomIntBetween(PAGE_SIZE * 2, PAGE_SIZE * 20));
BytesReference pbr = newBytesReference(length);
StreamInput streamInput = pbr.streamInput();
BytesRefBuilder target = new BytesRefBuilder();
while (target.length() < pbr.length()) {
switch (randomIntBetween(0, 10)) {
case 6:
case 5:
target.append(new BytesRef(new byte[] {streamInput.readByte()}));
break;
case 4:
case 3:
BytesRef bytesRef =
streamInput.readBytesRef(scaledRandomIntBetween(1, pbr.length() - target.length()));
target.append(bytesRef);
break;
default:
byte[] buffer = new byte[scaledRandomIntBetween(1, pbr.length() - target.length())];
int offset = scaledRandomIntBetween(0, buffer.length - 1);
int read = streamInput.read(buffer, offset, buffer.length - offset);
target.append(new BytesRef(buffer, offset, read));
break;
}
}
assertEquals(pbr.length(), target.length());
BytesRef targetBytes = target.get();
assertArrayEquals(
pbr.toBytes(),
Arrays.copyOfRange(targetBytes.bytes, targetBytes.offset, targetBytes.length));
}
@Override
public void visitDocument(int n, StoredFieldVisitor visitor) throws IOException {
in.seek(offsets[n]);
while (true) {
readLine();
if (StringHelper.startsWith(scratch.get(), FIELD) == false) {
break;
}
int fieldNumber = parseIntAt(FIELD.length);
FieldInfo fieldInfo = fieldInfos.fieldInfo(fieldNumber);
readLine();
assert StringHelper.startsWith(scratch.get(), NAME);
readLine();
assert StringHelper.startsWith(scratch.get(), TYPE);
final BytesRef type;
if (equalsAt(TYPE_STRING, scratch.get(), TYPE.length)) {
type = TYPE_STRING;
} else if (equalsAt(TYPE_BINARY, scratch.get(), TYPE.length)) {
type = TYPE_BINARY;
} else if (equalsAt(TYPE_INT, scratch.get(), TYPE.length)) {
type = TYPE_INT;
} else if (equalsAt(TYPE_LONG, scratch.get(), TYPE.length)) {
type = TYPE_LONG;
} else if (equalsAt(TYPE_FLOAT, scratch.get(), TYPE.length)) {
type = TYPE_FLOAT;
} else if (equalsAt(TYPE_DOUBLE, scratch.get(), TYPE.length)) {
type = TYPE_DOUBLE;
} else {
throw new RuntimeException("unknown field type");
}
switch (visitor.needsField(fieldInfo)) {
case YES:
readField(type, fieldInfo, visitor);
break;
case NO:
readLine();
assert StringHelper.startsWith(scratch.get(), VALUE);
break;
case STOP:
return;
}
}
}
@Override
public BytesRef next() throws IOException {
if (++curPos < entries.size()) {
entries.get(spare, curPos);
return spare.get();
}
return null;
}
/** Unmarshals a string-based field value. */
protected static Object unmarshalStringSortValue(Object value) {
if (null == value) {
return null;
}
BytesRefBuilder spare = new BytesRefBuilder();
String stringVal = (String) value;
spare.copyChars(stringVal);
return spare.get();
}
@Override
public BytesRef getBytesRef() {
assert valueSize == 64 || valueSize == 32;
if (valueSize == 64) {
NumericUtils.longToPrefixCoded(value, shift, bytes);
} else {
NumericUtils.intToPrefixCoded((int) value, shift, bytes);
}
return bytes.get();
}
@Override
protected void doSetNextReader(LeafReaderContext context) throws IOException {
if (segmentFacetCounts != null) {
segmentResults.add(createSegmentResult());
}
groupFieldTermsIndex = DocValues.getSorted(context.reader(), groupField);
facetFieldTermsIndex = DocValues.getSorted(context.reader(), facetField);
// 1+ to allow for the -1 "not set":
segmentFacetCounts = new int[facetFieldTermsIndex.getValueCount() + 1];
segmentTotalCount = 0;
segmentGroupedFacetHits.clear();
for (GroupedFacetHit groupedFacetHit : groupedFacetHits) {
int facetOrd =
groupedFacetHit.facetValue == null
? -1
: facetFieldTermsIndex.lookupTerm(groupedFacetHit.facetValue);
if (groupedFacetHit.facetValue != null && facetOrd < 0) {
continue;
}
int groupOrd =
groupedFacetHit.groupValue == null
? -1
: groupFieldTermsIndex.lookupTerm(groupedFacetHit.groupValue);
if (groupedFacetHit.groupValue != null && groupOrd < 0) {
continue;
}
int segmentGroupedFacetsIndex =
groupOrd * (facetFieldTermsIndex.getValueCount() + 1) + facetOrd;
segmentGroupedFacetHits.put(segmentGroupedFacetsIndex);
}
if (facetPrefix != null) {
startFacetOrd = facetFieldTermsIndex.lookupTerm(facetPrefix);
if (startFacetOrd < 0) {
// Points to the ord one higher than facetPrefix
startFacetOrd = -startFacetOrd - 1;
}
BytesRefBuilder facetEndPrefix = new BytesRefBuilder();
facetEndPrefix.append(facetPrefix);
facetEndPrefix.append(UnicodeUtil.BIG_TERM);
endFacetOrd = facetFieldTermsIndex.lookupTerm(facetEndPrefix.get());
assert endFacetOrd < 0;
endFacetOrd = -endFacetOrd - 1; // Points to the ord one higher than facetEndPrefix
} else {
startFacetOrd = -1;
endFacetOrd = facetFieldTermsIndex.getValueCount();
}
}
@Override
public void seekExact(long ord) throws IOException {
// TODO: would be better to make this simpler and faster.
// but we dont want to introduce a bug that corrupts our enum state!
bytesReader.setPosition(0);
fst.getFirstArc(firstArc);
IntsRef output = Util.getByOutput(fst, ord, bytesReader, firstArc, scratchArc, scratchInts);
BytesRefBuilder scratchBytes = new BytesRefBuilder();
scratchBytes.clear();
Util.toBytesRef(output, scratchBytes);
// TODO: we could do this lazily, better to try to push into FSTEnum though?
in.seekExact(scratchBytes.get());
}
// NOTE: while it's tempting to make this public, since
// caller's parser likely knows the
// numInput/numOutputWords, sneaky exceptions, much later
// on, will result if these values are wrong; so we always
// recompute ourselves to be safe:
private void add(
CharsRef input,
int numInputWords,
CharsRef output,
int numOutputWords,
boolean includeOrig) {
// first convert to UTF-8
if (numInputWords <= 0) {
throw new IllegalArgumentException("numInputWords must be > 0 (got " + numInputWords + ")");
}
if (input.length <= 0) {
throw new IllegalArgumentException("input.length must be > 0 (got " + input.length + ")");
}
if (numOutputWords <= 0) {
throw new IllegalArgumentException(
"numOutputWords must be > 0 (got " + numOutputWords + ")");
}
if (output.length <= 0) {
throw new IllegalArgumentException("output.length must be > 0 (got " + output.length + ")");
}
assert !hasHoles(input) : "input has holes: " + input;
assert !hasHoles(output) : "output has holes: " + output;
// System.out.println("fmap.add input=" + input + " numInputWords=" + numInputWords + "
// output=" + output + " numOutputWords=" + numOutputWords);
utf8Scratch.copyChars(output.chars, output.offset, output.length);
// lookup in hash
int ord = words.add(utf8Scratch.get());
if (ord < 0) {
// already exists in our hash
ord = (-ord) - 1;
// System.out.println(" output=" + output + " old ord=" + ord);
} else {
// System.out.println(" output=" + output + " new ord=" + ord);
}
MapEntry e = workingSet.get(input);
if (e == null) {
e = new MapEntry();
workingSet.put(
CharsRef.deepCopyOf(input), e); // make a copy, since we will keep around in our map
}
e.ords.add(ord);
e.includeOrig |= includeOrig;
maxHorizontalContext = Math.max(maxHorizontalContext, numInputWords);
maxHorizontalContext = Math.max(maxHorizontalContext, numOutputWords);
}
public static int sortAndDedup(final BytesRefArray bytes, final int[] indices) {
final BytesRefBuilder scratch = new BytesRefBuilder();
final BytesRefBuilder scratch1 = new BytesRefBuilder();
final int numValues = bytes.size();
assert indices.length >= numValues;
if (numValues <= 1) {
return numValues;
}
sort(scratch, scratch1, bytes, indices);
int uniqueCount = 1;
BytesRefBuilder previous = scratch;
BytesRefBuilder current = scratch1;
bytes.get(previous, indices[0]);
for (int i = 1; i < numValues; ++i) {
bytes.get(current, indices[i]);
if (!previous.get().equals(current.get())) {
indices[uniqueCount++] = indices[i];
}
BytesRefBuilder tmp = previous;
previous = current;
current = tmp;
}
return uniqueCount;
}
@Override
public void seekExact(BytesRef target, TermState otherState) {
// if (DEBUG) {
// System.out.println("BTTR.seekExact termState seg=" + segment + " target=" +
// target.utf8ToString() + " " + target + " state=" + otherState);
// }
assert clearEOF();
if (target.compareTo(term.get()) != 0 || !termExists) {
assert otherState != null && otherState instanceof BlockTermState;
currentFrame = staticFrame;
currentFrame.state.copyFrom(otherState);
term.copyBytes(target);
currentFrame.metaDataUpto = currentFrame.getTermBlockOrd();
assert currentFrame.metaDataUpto > 0;
validIndexPrefix = 0;
} else {
// if (DEBUG) {
// System.out.println(" skip seek: already on target state=" + currentFrame.state);
// }
}
}
public static void displayPointRanges(Scanner in) {
double[][] point = getPoints(in, 1);
long hash, hashUpper;
double lon, lat, lonUpper, latUpper;
for (int i = 63; i >= 45; i -= GeoPointField.PRECISION_STEP) {
BytesRefBuilder brb = new BytesRefBuilder();
NumericUtils.longToPrefixCoded(
GeoUtils.mortonHash(point[0][LON_INDEX], point[0][LAT_INDEX]), i, brb);
BytesRef br = brb.get();
hash = NumericUtils.prefixCodedToLong(br);
hashUpper = hash | ((1L << i) - 1);
lon = GeoUtils.mortonUnhashLon(hash);
lat = GeoUtils.mortonUnhashLat(hash);
lonUpper = GeoUtils.mortonUnhashLon(hashUpper);
latUpper = GeoUtils.mortonUnhashLat(hashUpper);
System.out.println(
i + ": " + br + " " + hash + " (" + lon + "," + lat + ")" + " : " + "(" + lonUpper + ","
+ latUpper + ")");
}
}
@Override
public Fields get(int doc) throws IOException {
SortedMap<String, SimpleTVTerms> fields = new TreeMap<>();
in.seek(offsets[doc]);
readLine();
assert StringHelper.startsWith(scratch.get(), NUMFIELDS);
int numFields = parseIntAt(NUMFIELDS.length);
if (numFields == 0) {
return null; // no vectors for this doc
}
for (int i = 0; i < numFields; i++) {
readLine();
assert StringHelper.startsWith(scratch.get(), FIELD);
// skip fieldNumber:
parseIntAt(FIELD.length);
readLine();
assert StringHelper.startsWith(scratch.get(), FIELDNAME);
String fieldName = readString(FIELDNAME.length, scratch);
readLine();
assert StringHelper.startsWith(scratch.get(), FIELDPOSITIONS);
boolean positions = Boolean.parseBoolean(readString(FIELDPOSITIONS.length, scratch));
readLine();
assert StringHelper.startsWith(scratch.get(), FIELDOFFSETS);
boolean offsets = Boolean.parseBoolean(readString(FIELDOFFSETS.length, scratch));
readLine();
assert StringHelper.startsWith(scratch.get(), FIELDPAYLOADS);
boolean payloads = Boolean.parseBoolean(readString(FIELDPAYLOADS.length, scratch));
readLine();
assert StringHelper.startsWith(scratch.get(), FIELDTERMCOUNT);
int termCount = parseIntAt(FIELDTERMCOUNT.length);
SimpleTVTerms terms = new SimpleTVTerms(offsets, positions, payloads);
fields.put(fieldName, terms);
BytesRefBuilder term = new BytesRefBuilder();
for (int j = 0; j < termCount; j++) {
readLine();
assert StringHelper.startsWith(scratch.get(), TERMTEXT);
int termLength = scratch.length() - TERMTEXT.length;
term.grow(termLength);
term.setLength(termLength);
System.arraycopy(scratch.bytes(), TERMTEXT.length, term.bytes(), 0, termLength);
SimpleTVPostings postings = new SimpleTVPostings();
terms.terms.put(term.toBytesRef(), postings);
readLine();
assert StringHelper.startsWith(scratch.get(), TERMFREQ);
postings.freq = parseIntAt(TERMFREQ.length);
if (positions || offsets) {
if (positions) {
postings.positions = new int[postings.freq];
if (payloads) {
postings.payloads = new BytesRef[postings.freq];
}
}
if (offsets) {
postings.startOffsets = new int[postings.freq];
postings.endOffsets = new int[postings.freq];
}
for (int k = 0; k < postings.freq; k++) {
if (positions) {
readLine();
assert StringHelper.startsWith(scratch.get(), POSITION);
postings.positions[k] = parseIntAt(POSITION.length);
if (payloads) {
readLine();
assert StringHelper.startsWith(scratch.get(), PAYLOAD);
if (scratch.length() - PAYLOAD.length == 0) {
postings.payloads[k] = null;
} else {
byte payloadBytes[] = new byte[scratch.length() - PAYLOAD.length];
System.arraycopy(
scratch.bytes(), PAYLOAD.length, payloadBytes, 0, payloadBytes.length);
postings.payloads[k] = new BytesRef(payloadBytes);
}
}
}
if (offsets) {
readLine();
assert StringHelper.startsWith(scratch.get(), STARTOFFSET);
postings.startOffsets[k] = parseIntAt(STARTOFFSET.length);
readLine();
assert StringHelper.startsWith(scratch.get(), ENDOFFSET);
postings.endOffsets[k] = parseIntAt(ENDOFFSET.length);
}
}
}
}
}
return new SimpleTVFields(fields);
}
@Override
public BytesRef term() {
assert !eof;
return term.get();
}
/* 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. */
@Override
public BytesRef next() throws IOException {
if (in == null) {
// Fresh TermsEnum; seek to first term:
final FST.Arc<Pair<BytesRef, Long>> arc;
if (fr.index != null) {
arc = fr.index.getFirstArc(arcs[0]);
// Empty string prefix must have an output in the index!
assert arc.isFinal();
} else {
arc = null;
}
currentFrame = pushFrame(arc, fr.rootCode, 0);
currentFrame.loadBlock();
}
targetBeforeCurrentLength = currentFrame.ord;
assert !eof;
// if (DEBUG) {
// System.out.println("\nBTTR.next seg=" + segment + " term=" + brToString(term) + "
// termExists?=" + termExists + " field=" + fieldInfo.name + " termBlockOrd=" +
// currentFrame.state.termBlockOrd + " validIndexPrefix=" + validIndexPrefix);
// printSeekState();
// }
if (currentFrame == staticFrame) {
// If seek was previously called and the term was
// cached, or seek(TermState) was called, usually
// caller is just going to pull a D/&PEnum or get
// docFreq, etc. But, if they then call next(),
// this method catches up all internal state so next()
// works properly:
// if (DEBUG) System.out.println(" re-seek to pending term=" + term.utf8ToString() + " " +
// term);
final boolean result = seekExact(term.get());
assert result;
}
// Pop finished blocks
while (currentFrame.nextEnt == currentFrame.entCount) {
if (!currentFrame.isLastInFloor) {
currentFrame.loadNextFloorBlock();
} else {
// if (DEBUG) System.out.println(" pop frame");
if (currentFrame.ord == 0) {
// if (DEBUG) System.out.println(" return null");
assert setEOF();
term.clear();
validIndexPrefix = 0;
currentFrame.rewind();
termExists = false;
return null;
}
final long lastFP = currentFrame.fpOrig;
currentFrame = stack[currentFrame.ord - 1];
if (currentFrame.nextEnt == -1 || currentFrame.lastSubFP != lastFP) {
// We popped into a frame that's not loaded
// yet or not scan'd to the right entry
currentFrame.scanToFloorFrame(term.get());
currentFrame.loadBlock();
currentFrame.scanToSubBlock(lastFP);
}
// Note that the seek state (last seek) has been
// invalidated beyond this depth
validIndexPrefix = Math.min(validIndexPrefix, currentFrame.prefix);
// if (DEBUG) {
// System.out.println(" reset validIndexPrefix=" + validIndexPrefix);
// }
}
}
while (true) {
if (currentFrame.next()) {
// Push to new block:
// if (DEBUG) System.out.println(" push frame");
currentFrame = pushFrame(null, currentFrame.lastSubFP, term.length());
// This is a "next" frame -- even if it's
// floor'd we must pretend it isn't so we don't
// try to scan to the right floor frame:
currentFrame.isFloor = false;
// currentFrame.hasTerms = true;
currentFrame.loadBlock();
} else {
// if (DEBUG) System.out.println(" return term=" + term.utf8ToString() + " " + term + "
// currentFrame.ord=" + currentFrame.ord);
return term.get();
}
}
}
// 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;
}
}
@Override
public BytesRef term() {
return term.get();
}
public static NamedList<Integer> getCounts(
SolrIndexSearcher searcher,
DocSet docs,
String fieldName,
int offset,
int limit,
int mincount,
boolean missing,
String sort,
String prefix,
String contains,
boolean ignoreCase)
throws IOException {
SchemaField schemaField = searcher.getSchema().getField(fieldName);
FieldType ft = schemaField.getType();
NamedList<Integer> res = new NamedList<>();
// TODO: remove multiValuedFieldCache(), check dv type / uninversion type?
final boolean multiValued = schemaField.multiValued() || ft.multiValuedFieldCache();
final SortedSetDocValues si; // for term lookups only
OrdinalMap ordinalMap = null; // for mapping per-segment ords to global ones
if (multiValued) {
si = searcher.getLeafReader().getSortedSetDocValues(fieldName);
if (si instanceof MultiSortedSetDocValues) {
ordinalMap = ((MultiSortedSetDocValues) si).mapping;
}
} else {
SortedDocValues single = searcher.getLeafReader().getSortedDocValues(fieldName);
si = single == null ? null : DocValues.singleton(single);
if (single instanceof MultiSortedDocValues) {
ordinalMap = ((MultiSortedDocValues) single).mapping;
}
}
if (si == null) {
return finalize(res, searcher, schemaField, docs, -1, missing);
}
if (si.getValueCount() >= Integer.MAX_VALUE) {
throw new UnsupportedOperationException(
"Currently this faceting method is limited to " + Integer.MAX_VALUE + " unique terms");
}
final BytesRefBuilder prefixRef;
if (prefix == null) {
prefixRef = null;
} else if (prefix.length() == 0) {
prefix = null;
prefixRef = null;
} else {
prefixRef = new BytesRefBuilder();
prefixRef.copyChars(prefix);
}
int startTermIndex, endTermIndex;
if (prefix != null) {
startTermIndex = (int) si.lookupTerm(prefixRef.get());
if (startTermIndex < 0) startTermIndex = -startTermIndex - 1;
prefixRef.append(UnicodeUtil.BIG_TERM);
endTermIndex = (int) si.lookupTerm(prefixRef.get());
assert endTermIndex < 0;
endTermIndex = -endTermIndex - 1;
} else {
startTermIndex = -1;
endTermIndex = (int) si.getValueCount();
}
final int nTerms = endTermIndex - startTermIndex;
int missingCount = -1;
final CharsRefBuilder charsRef = new CharsRefBuilder();
if (nTerms > 0 && docs.size() >= mincount) {
// count collection array only needs to be as big as the number of terms we are
// going to collect counts for.
final int[] counts = new int[nTerms];
Filter filter = docs.getTopFilter();
List<LeafReaderContext> leaves = searcher.getTopReaderContext().leaves();
for (int subIndex = 0; subIndex < leaves.size(); subIndex++) {
LeafReaderContext leaf = leaves.get(subIndex);
DocIdSet dis =
filter.getDocIdSet(leaf, null); // solr docsets already exclude any deleted docs
DocIdSetIterator disi = null;
if (dis != null) {
disi = dis.iterator();
}
if (disi != null) {
if (multiValued) {
SortedSetDocValues sub = leaf.reader().getSortedSetDocValues(fieldName);
if (sub == null) {
sub = DocValues.emptySortedSet();
}
final SortedDocValues singleton = DocValues.unwrapSingleton(sub);
if (singleton != null) {
// some codecs may optimize SORTED_SET storage for single-valued fields
accumSingle(counts, startTermIndex, singleton, disi, subIndex, ordinalMap);
} else {
accumMulti(counts, startTermIndex, sub, disi, subIndex, ordinalMap);
}
} else {
SortedDocValues sub = leaf.reader().getSortedDocValues(fieldName);
if (sub == null) {
sub = DocValues.emptySorted();
}
accumSingle(counts, startTermIndex, sub, disi, subIndex, ordinalMap);
}
}
}
if (startTermIndex == -1) {
missingCount = counts[0];
}
// IDEA: we could also maintain a count of "other"... everything that fell outside
// of the top 'N'
int off = offset;
int lim = limit >= 0 ? limit : Integer.MAX_VALUE;
if (sort.equals(FacetParams.FACET_SORT_COUNT)
|| sort.equals(FacetParams.FACET_SORT_COUNT_LEGACY)) {
int maxsize = limit > 0 ? offset + limit : Integer.MAX_VALUE - 1;
maxsize = Math.min(maxsize, nTerms);
LongPriorityQueue queue =
new LongPriorityQueue(Math.min(maxsize, 1000), maxsize, Long.MIN_VALUE);
int min = mincount - 1; // the smallest value in the top 'N' values
for (int i = (startTermIndex == -1) ? 1 : 0; i < nTerms; i++) {
int c = counts[i];
if (contains != null) {
final BytesRef term = si.lookupOrd(startTermIndex + i);
if (!SimpleFacets.contains(term.utf8ToString(), contains, ignoreCase)) {
continue;
}
}
if (c > min) {
// NOTE: we use c>min rather than c>=min as an optimization because we are going in
// index order, so we already know that the keys are ordered. This can be very
// important if a lot of the counts are repeated (like zero counts would be).
// smaller term numbers sort higher, so subtract the term number instead
long pair = (((long) c) << 32) + (Integer.MAX_VALUE - i);
boolean displaced = queue.insert(pair);
if (displaced) min = (int) (queue.top() >>> 32);
}
}
// if we are deep paging, we don't have to order the highest "offset" counts.
int collectCount = Math.max(0, queue.size() - off);
assert collectCount <= lim;
// the start and end indexes of our list "sorted" (starting with the highest value)
int sortedIdxStart = queue.size() - (collectCount - 1);
int sortedIdxEnd = queue.size() + 1;
final long[] sorted = queue.sort(collectCount);
for (int i = sortedIdxStart; i < sortedIdxEnd; i++) {
long pair = sorted[i];
int c = (int) (pair >>> 32);
int tnum = Integer.MAX_VALUE - (int) pair;
final BytesRef term = si.lookupOrd(startTermIndex + tnum);
ft.indexedToReadable(term, charsRef);
res.add(charsRef.toString(), c);
}
} else {
// add results in index order
int i = (startTermIndex == -1) ? 1 : 0;
if (mincount <= 0 && contains == null) {
// if mincount<=0 and we're not examining the values for contains, then
// we won't discard any terms and we know exactly where to start.
i += off;
off = 0;
}
for (; i < nTerms; i++) {
int c = counts[i];
if (c < mincount) continue;
BytesRef term = null;
if (contains != null) {
term = si.lookupOrd(startTermIndex + i);
if (!SimpleFacets.contains(term.utf8ToString(), contains, ignoreCase)) {
continue;
}
}
if (--off >= 0) continue;
if (--lim < 0) break;
if (term == null) {
term = si.lookupOrd(startTermIndex + i);
}
ft.indexedToReadable(term, charsRef);
res.add(charsRef.toString(), c);
}
}
}
return finalize(res, searcher, schemaField, docs, missingCount, missing);
}
@Override
public Directory getCompoundReader(Directory dir, SegmentInfo si, IOContext context)
throws IOException {
String dataFile = IndexFileNames.segmentFileName(si.name, "", DATA_EXTENSION);
final IndexInput in = dir.openInput(dataFile, context);
BytesRefBuilder scratch = new BytesRefBuilder();
// first get to TOC:
DecimalFormat df =
new DecimalFormat(OFFSETPATTERN, DecimalFormatSymbols.getInstance(Locale.ROOT));
long pos = in.length() - TABLEPOS.length - OFFSETPATTERN.length() - 1;
in.seek(pos);
SimpleTextUtil.readLine(in, scratch);
assert StringHelper.startsWith(scratch.get(), TABLEPOS);
long tablePos = -1;
try {
tablePos = df.parse(stripPrefix(scratch, TABLEPOS)).longValue();
} catch (ParseException e) {
throw new CorruptIndexException(
"can't parse CFS trailer, got: " + scratch.get().utf8ToString(), in);
}
// seek to TOC and read it
in.seek(tablePos);
SimpleTextUtil.readLine(in, scratch);
assert StringHelper.startsWith(scratch.get(), TABLE);
int numEntries = Integer.parseInt(stripPrefix(scratch, TABLE));
final String fileNames[] = new String[numEntries];
final long startOffsets[] = new long[numEntries];
final long endOffsets[] = new long[numEntries];
for (int i = 0; i < numEntries; i++) {
SimpleTextUtil.readLine(in, scratch);
assert StringHelper.startsWith(scratch.get(), TABLENAME);
fileNames[i] = si.name + IndexFileNames.stripSegmentName(stripPrefix(scratch, TABLENAME));
if (i > 0) {
// files must be unique and in sorted order
assert fileNames[i].compareTo(fileNames[i - 1]) > 0;
}
SimpleTextUtil.readLine(in, scratch);
assert StringHelper.startsWith(scratch.get(), TABLESTART);
startOffsets[i] = Long.parseLong(stripPrefix(scratch, TABLESTART));
SimpleTextUtil.readLine(in, scratch);
assert StringHelper.startsWith(scratch.get(), TABLEEND);
endOffsets[i] = Long.parseLong(stripPrefix(scratch, TABLEEND));
}
return new Directory() {
private int getIndex(String name) throws IOException {
int index = Arrays.binarySearch(fileNames, name);
if (index < 0) {
throw new FileNotFoundException(
"No sub-file found (fileName="
+ name
+ " files: "
+ Arrays.toString(fileNames)
+ ")");
}
return index;
}
@Override
public String[] listAll() throws IOException {
ensureOpen();
return fileNames.clone();
}
@Override
public long fileLength(String name) throws IOException {
ensureOpen();
int index = getIndex(name);
return endOffsets[index] - startOffsets[index];
}
@Override
public IndexInput openInput(String name, IOContext context) throws IOException {
ensureOpen();
int index = getIndex(name);
return in.slice(name, startOffsets[index], endOffsets[index] - startOffsets[index]);
}
@Override
public void close() throws IOException {
in.close();
}
// write methods: disabled
@Override
public IndexOutput createOutput(String name, IOContext context) {
throw new UnsupportedOperationException();
}
@Override
public void sync(Collection<String> names) {
throw new UnsupportedOperationException();
}
@Override
public void deleteFile(String name) {
throw new UnsupportedOperationException();
}
@Override
public void renameFile(String source, String dest) {
throw new UnsupportedOperationException();
}
@Override
public Lock makeLock(String name) {
throw new UnsupportedOperationException();
}
};
}
@Override
public BytesRef getBytesRef() {
return bytes.get();
}
@Override
protected void doSetNextReader(LeafReaderContext context) throws IOException {
if (segmentFacetCounts != null) {
segmentResults.add(createSegmentResult());
}
groupFieldTermsIndex = DocValues.getSorted(context.reader(), groupField);
facetFieldDocTermOrds = DocValues.getSortedSet(context.reader(), facetField);
facetFieldNumTerms = (int) facetFieldDocTermOrds.getValueCount();
if (facetFieldNumTerms == 0) {
facetOrdTermsEnum = null;
} else {
facetOrdTermsEnum = facetFieldDocTermOrds.termsEnum();
}
// [facetFieldNumTerms() + 1] for all possible facet values and docs not containing facet
// field
segmentFacetCounts = new int[facetFieldNumTerms + 1];
segmentTotalCount = 0;
segmentGroupedFacetHits.clear();
for (GroupedFacetHit groupedFacetHit : groupedFacetHits) {
int groupOrd =
groupedFacetHit.groupValue == null
? -1
: groupFieldTermsIndex.lookupTerm(groupedFacetHit.groupValue);
if (groupedFacetHit.groupValue != null && groupOrd < 0) {
continue;
}
int facetOrd;
if (groupedFacetHit.facetValue != null) {
if (facetOrdTermsEnum == null
|| !facetOrdTermsEnum.seekExact(groupedFacetHit.facetValue)) {
continue;
}
facetOrd = (int) facetOrdTermsEnum.ord();
} else {
facetOrd = facetFieldNumTerms;
}
// (facetFieldDocTermOrds.numTerms() + 1) for all possible facet values and docs not
// containing facet field
int segmentGroupedFacetsIndex = groupOrd * (facetFieldNumTerms + 1) + facetOrd;
segmentGroupedFacetHits.put(segmentGroupedFacetsIndex);
}
if (facetPrefix != null) {
TermsEnum.SeekStatus seekStatus;
if (facetOrdTermsEnum != null) {
seekStatus = facetOrdTermsEnum.seekCeil(facetPrefix);
} else {
seekStatus = TermsEnum.SeekStatus.END;
}
if (seekStatus != TermsEnum.SeekStatus.END) {
startFacetOrd = (int) facetOrdTermsEnum.ord();
} else {
startFacetOrd = 0;
endFacetOrd = 0;
return;
}
BytesRefBuilder facetEndPrefix = new BytesRefBuilder();
facetEndPrefix.append(facetPrefix);
facetEndPrefix.append(UnicodeUtil.BIG_TERM);
seekStatus = facetOrdTermsEnum.seekCeil(facetEndPrefix.get());
if (seekStatus != TermsEnum.SeekStatus.END) {
endFacetOrd = (int) facetOrdTermsEnum.ord();
} else {
endFacetOrd = facetFieldNumTerms; // Don't include null...
}
} else {
startFacetOrd = 0;
endFacetOrd = facetFieldNumTerms + 1;
}
}
