// TODO: maybe we could add bulk-add method to
// Builder? Takes FST and unions it w/ current
// FST.
private void append(Builder<BytesRef> builder, FST<BytesRef> subIndex) throws IOException {
final BytesRefFSTEnum<BytesRef> subIndexEnum = new BytesRefFSTEnum<BytesRef>(subIndex);
BytesRefFSTEnum.InputOutput<BytesRef> indexEnt;
while ((indexEnt = subIndexEnum.next()) != null) {
// if (DEBUG) {
// System.out.println(" add sub=" + indexEnt.input + " " + indexEnt.input + " output="
// + indexEnt.output);
// }
builder.add(Util.toIntsRef(indexEnt.input, scratchIntsRef), indexEnt.output);
}
}
@Override
public void finishTerm(BytesRef text, TermStats stats) throws IOException {
assert stats.docFreq > 0;
// if (DEBUG) System.out.println("BTTW.finishTerm term=" + fieldInfo.name + ":" +
// toString(text) + " seg=" + segment + " df=" + stats.docFreq);
blockBuilder.add(Util.toIntsRef(text, scratchIntsRef), noOutputs.getNoOutput());
pending.add(new PendingTerm(BytesRef.deepCopyOf(text), stats));
postingsWriter.finishTerm(stats);
numTerms++;
}
@Override
public void add(BytesRef text, TermStats stats, long termsFilePointer) throws IOException {
if (text.length == 0) {
// We already added empty string in ctor
assert termsFilePointer == startTermsFilePointer;
return;
}
final int lengthSave = text.length;
text.length = indexedTermPrefixLength(lastTerm, text);
try {
fstBuilder.add(Util.toIntsRef(text, scratchIntsRef), termsFilePointer);
} finally {
text.length = lengthSave;
}
lastTerm.copyBytes(text);
}
@Override
public void build(TermFreqIterator iterator) throws IOException {
BytesRef scratch = new BytesRef();
TermFreqIterator iter =
new WFSTTermFreqIteratorWrapper(iterator, BytesRef.getUTF8SortedAsUnicodeComparator());
IntsRef scratchInts = new IntsRef();
BytesRef previous = null;
PositiveIntOutputs outputs = PositiveIntOutputs.getSingleton(true);
Builder<Long> builder = new Builder<Long>(FST.INPUT_TYPE.BYTE1, outputs);
while ((scratch = iter.next()) != null) {
long cost = iter.weight();
if (previous == null) {
previous = new BytesRef();
} else if (scratch.equals(previous)) {
continue; // for duplicate suggestions, the best weight is actually
// added
}
Util.toIntsRef(scratch, scratchInts);
builder.add(scratchInts, cost);
previous.copyBytes(scratch);
}
fst = builder.finish();
}
private void loadTerms() throws IOException {
PositiveIntOutputs posIntOutputs = PositiveIntOutputs.getSingleton(false);
final Builder<PairOutputs.Pair<Long, PairOutputs.Pair<Long, Long>>> b;
final PairOutputs<Long, Long> outputsInner =
new PairOutputs<Long, Long>(posIntOutputs, posIntOutputs);
final PairOutputs<Long, PairOutputs.Pair<Long, Long>> outputs =
new PairOutputs<Long, PairOutputs.Pair<Long, Long>>(posIntOutputs, outputsInner);
b =
new Builder<PairOutputs.Pair<Long, PairOutputs.Pair<Long, Long>>>(
FST.INPUT_TYPE.BYTE1, outputs);
IndexInput in = (IndexInput) SimpleTextFieldsReader.this.in.clone();
in.seek(termsStart);
final BytesRef lastTerm = new BytesRef(10);
long lastDocsStart = -1;
int docFreq = 0;
long totalTermFreq = 0;
OpenBitSet visitedDocs = new OpenBitSet();
final IntsRef scratchIntsRef = new IntsRef();
while (true) {
SimpleTextUtil.readLine(in, scratch);
if (scratch.equals(END) || StringHelper.startsWith(scratch, FIELD)) {
if (lastDocsStart != -1) {
b.add(
Util.toIntsRef(lastTerm, scratchIntsRef),
outputs.newPair(
lastDocsStart, outputsInner.newPair((long) docFreq, totalTermFreq)));
sumTotalTermFreq += totalTermFreq;
}
break;
} else if (StringHelper.startsWith(scratch, DOC)) {
docFreq++;
sumDocFreq++;
UnicodeUtil.UTF8toUTF16(
scratch.bytes,
scratch.offset + DOC.length,
scratch.length - DOC.length,
scratchUTF16);
int docID = ArrayUtil.parseInt(scratchUTF16.chars, 0, scratchUTF16.length);
visitedDocs.set(docID);
} else if (StringHelper.startsWith(scratch, FREQ)) {
UnicodeUtil.UTF8toUTF16(
scratch.bytes,
scratch.offset + FREQ.length,
scratch.length - FREQ.length,
scratchUTF16);
totalTermFreq += ArrayUtil.parseInt(scratchUTF16.chars, 0, scratchUTF16.length);
} else if (StringHelper.startsWith(scratch, TERM)) {
if (lastDocsStart != -1) {
b.add(
Util.toIntsRef(lastTerm, scratchIntsRef),
outputs.newPair(
lastDocsStart, outputsInner.newPair((long) docFreq, totalTermFreq)));
}
lastDocsStart = in.getFilePointer();
final int len = scratch.length - TERM.length;
if (len > lastTerm.length) {
lastTerm.grow(len);
}
System.arraycopy(scratch.bytes, TERM.length, lastTerm.bytes, 0, len);
lastTerm.length = len;
docFreq = 0;
sumTotalTermFreq += totalTermFreq;
totalTermFreq = 0;
termCount++;
}
}
docCount = (int) visitedDocs.cardinality();
fst = b.finish();
/*
PrintStream ps = new PrintStream("out.dot");
fst.toDot(ps);
ps.close();
System.out.println("SAVED out.dot");
*/
// System.out.println("FST " + fst.sizeInBytes());
}
public void compileIndex(List<PendingBlock> floorBlocks, RAMOutputStream scratchBytes)
throws IOException {
assert (isFloor && floorBlocks != null && floorBlocks.size() != 0)
|| (!isFloor && floorBlocks == null)
: "isFloor=" + isFloor + " floorBlocks=" + floorBlocks;
assert scratchBytes.getFilePointer() == 0;
// TODO: try writing the leading vLong in MSB order
// (opposite of what Lucene does today), for better
// outputs sharing in the FST
scratchBytes.writeVLong(encodeOutput(fp, hasTerms, isFloor));
if (isFloor) {
scratchBytes.writeVInt(floorBlocks.size());
for (PendingBlock sub : floorBlocks) {
assert sub.floorLeadByte != -1;
// if (DEBUG) {
// System.out.println(" write floorLeadByte=" +
// Integer.toHexString(sub.floorLeadByte&0xff));
// }
scratchBytes.writeByte((byte) sub.floorLeadByte);
assert sub.fp > fp;
scratchBytes.writeVLong((sub.fp - fp) << 1 | (sub.hasTerms ? 1 : 0));
}
}
final ByteSequenceOutputs outputs = ByteSequenceOutputs.getSingleton();
final Builder<BytesRef> indexBuilder =
new Builder<BytesRef>(
FST.INPUT_TYPE.BYTE1,
0,
0,
true,
false,
Integer.MAX_VALUE,
outputs,
null,
false,
PackedInts.COMPACT,
true,
15);
// if (DEBUG) {
// System.out.println(" compile index for prefix=" + prefix);
// }
// indexBuilder.DEBUG = false;
final byte[] bytes = new byte[(int) scratchBytes.getFilePointer()];
assert bytes.length > 0;
scratchBytes.writeTo(bytes, 0);
indexBuilder.add(
Util.toIntsRef(prefix, scratchIntsRef), new BytesRef(bytes, 0, bytes.length));
scratchBytes.reset();
// Copy over index for all sub-blocks
if (subIndices != null) {
for (FST<BytesRef> subIndex : subIndices) {
append(indexBuilder, subIndex);
}
}
if (floorBlocks != null) {
for (PendingBlock sub : floorBlocks) {
if (sub.subIndices != null) {
for (FST<BytesRef> subIndex : sub.subIndices) {
append(indexBuilder, subIndex);
}
}
sub.subIndices = null;
}
}
index = indexBuilder.finish();
subIndices = null;
/*
Writer w = new OutputStreamWriter(new FileOutputStream("out.dot"));
Util.toDot(index, w, false, false);
System.out.println("SAVED to out.dot");
w.close();
*/
}