/** encodes an entry (bytes+(contexts)+(payload)+weight) to the provided writer */
protected void encode(
ByteSequencesWriter writer,
ByteArrayDataOutput output,
byte[] buffer,
BytesRef spare,
BytesRef payload,
Set<BytesRef> contexts,
long weight)
throws IOException {
int requiredLength = spare.length + 8 + ((hasPayloads) ? 2 + payload.length : 0);
if (hasContexts) {
for (BytesRef ctx : contexts) {
requiredLength += 2 + ctx.length;
}
requiredLength += 2; // for length of contexts
}
if (requiredLength >= buffer.length) {
buffer = ArrayUtil.grow(buffer, requiredLength);
}
output.reset(buffer);
output.writeBytes(spare.bytes, spare.offset, spare.length);
if (hasContexts) {
for (BytesRef ctx : contexts) {
output.writeBytes(ctx.bytes, ctx.offset, ctx.length);
output.writeShort((short) ctx.length);
}
output.writeShort((short) contexts.size());
}
if (hasPayloads) {
output.writeBytes(payload.bytes, payload.offset, payload.length);
output.writeShort((short) payload.length);
}
output.writeLong(weight);
writer.write(buffer, 0, output.getPosition());
}
@Override
protected void encode(
ByteSequencesWriter writer,
ByteArrayDataOutput output,
byte[] buffer,
BytesRef spare,
long weight)
throws IOException {
if (spare.length + 5 >= buffer.length) {
buffer = ArrayUtil.grow(buffer, spare.length + 5);
}
output.reset(buffer);
output.writeBytes(spare.bytes, spare.offset, spare.length);
output.writeByte((byte) 0); // separator: not used, just for sort order
output.writeInt(encodeWeight(weight));
writer.write(buffer, 0, output.getPosition());
}
/** Builds an {@link SynonymMap} and returns it. */
public SynonymMap build() throws IOException {
ByteSequenceOutputs outputs = ByteSequenceOutputs.getSingleton();
// TODO: are we using the best sharing options?
org.apache.lucene.util.fst.Builder<BytesRef> builder =
new org.apache.lucene.util.fst.Builder<>(FST.INPUT_TYPE.BYTE4, outputs);
BytesRefBuilder scratch = new BytesRefBuilder();
ByteArrayDataOutput scratchOutput = new ByteArrayDataOutput();
final Set<Integer> dedupSet;
if (dedup) {
dedupSet = new HashSet<>();
} else {
dedupSet = null;
}
final byte[] spare = new byte[5];
Set<CharsRef> keys = workingSet.keySet();
CharsRef sortedKeys[] = keys.toArray(new CharsRef[keys.size()]);
Arrays.sort(sortedKeys, CharsRef.getUTF16SortedAsUTF8Comparator());
final IntsRefBuilder scratchIntsRef = new IntsRefBuilder();
// System.out.println("fmap.build");
for (int keyIdx = 0; keyIdx < sortedKeys.length; keyIdx++) {
CharsRef input = sortedKeys[keyIdx];
MapEntry output = workingSet.get(input);
int numEntries = output.ords.size();
// output size, assume the worst case
int estimatedSize = 5 + numEntries * 5; // numEntries + one ord for each entry
scratch.grow(estimatedSize);
scratchOutput.reset(scratch.bytes());
// now write our output data:
int count = 0;
for (int i = 0; i < numEntries; i++) {
if (dedupSet != null) {
// box once
final Integer ent = output.ords.get(i);
if (dedupSet.contains(ent)) {
continue;
}
dedupSet.add(ent);
}
scratchOutput.writeVInt(output.ords.get(i));
count++;
}
final int pos = scratchOutput.getPosition();
scratchOutput.writeVInt(count << 1 | (output.includeOrig ? 0 : 1));
final int pos2 = scratchOutput.getPosition();
final int vIntLen = pos2 - pos;
// Move the count + includeOrig to the front of the byte[]:
System.arraycopy(scratch.bytes(), pos, spare, 0, vIntLen);
System.arraycopy(scratch.bytes(), 0, scratch.bytes(), vIntLen, pos);
System.arraycopy(spare, 0, scratch.bytes(), 0, vIntLen);
if (dedupSet != null) {
dedupSet.clear();
}
scratch.setLength(scratchOutput.getPosition());
// System.out.println(" add input=" + input + " output=" + scratch + " offset=" +
// scratch.offset + " length=" + scratch.length + " count=" + count);
builder.add(Util.toUTF32(input, scratchIntsRef), scratch.toBytesRef());
}
FST<BytesRef> fst = builder.finish();
return new SynonymMap(fst, words, maxHorizontalContext);
}
@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();
}
}
