in lucene/suggest/src/java/org/apache/lucene/search/suggest/analyzing/AnalyzingSuggester.java [389:597]
public void build(InputIterator iterator) throws IOException {
if (iterator.hasContexts()) {
throw new IllegalArgumentException("this suggester doesn't support contexts");
}
hasPayloads = iterator.hasPayloads();
OfflineSorter sorter =
new OfflineSorter(tempDir, tempFileNamePrefix, new AnalyzingComparator(hasPayloads));
IndexOutput tempInput =
tempDir.createTempOutput(tempFileNamePrefix, "input", IOContext.DEFAULT);
OfflineSorter.ByteSequencesWriter writer = new OfflineSorter.ByteSequencesWriter(tempInput);
OfflineSorter.ByteSequencesReader reader = null;
BytesRefBuilder scratch = new BytesRefBuilder();
TokenStreamToAutomaton ts2a = getTokenStreamToAutomaton();
String tempSortedFileName = null;
long newCount = 0;
byte[] buffer = new byte[8];
try {
ByteArrayDataOutput output = new ByteArrayDataOutput(buffer);
for (BytesRef surfaceForm; (surfaceForm = iterator.next()) != null; ) {
LimitedFiniteStringsIterator finiteStrings =
new LimitedFiniteStringsIterator(toAutomaton(surfaceForm, ts2a), maxGraphExpansions);
for (IntsRef string; (string = finiteStrings.next()) != null; newCount++) {
Util.toBytesRef(string, scratch);
// length of the analyzed text (FST input)
if (scratch.length() > Short.MAX_VALUE - 2) {
throw new IllegalArgumentException(
"cannot handle analyzed forms > "
+ (Short.MAX_VALUE - 2)
+ " in length (got "
+ scratch.length()
+ ")");
}
short analyzedLength = (short) scratch.length();
// compute the required length:
// analyzed sequence + weight (4) + surface + analyzedLength (short)
int requiredLength = analyzedLength + 4 + surfaceForm.length + 2;
BytesRef payload;
if (hasPayloads) {
if (surfaceForm.length > (Short.MAX_VALUE - 2)) {
throw new IllegalArgumentException(
"cannot handle surface form > "
+ (Short.MAX_VALUE - 2)
+ " in length (got "
+ surfaceForm.length
+ ")");
}
payload = iterator.payload();
// payload + surfaceLength (short)
requiredLength += payload.length + 2;
} else {
payload = null;
}
buffer = ArrayUtil.growNoCopy(buffer, requiredLength);
output.reset(buffer);
output.writeShort(analyzedLength);
output.writeBytes(scratch.bytes(), 0, scratch.length());
output.writeInt(encodeWeight(iterator.weight()));
if (hasPayloads) {
for (int i = 0; i < surfaceForm.length; i++) {
if (surfaceForm.bytes[i] == PAYLOAD_SEP) {
throw new IllegalArgumentException(
"surface form cannot contain unit separator character U+001F; this character is reserved");
}
}
output.writeShort((short) surfaceForm.length);
output.writeBytes(surfaceForm.bytes, surfaceForm.offset, surfaceForm.length);
output.writeBytes(payload.bytes, payload.offset, payload.length);
} else {
output.writeBytes(surfaceForm.bytes, surfaceForm.offset, surfaceForm.length);
}
assert output.getPosition() == requiredLength
: output.getPosition() + " vs " + requiredLength;
writer.write(buffer, 0, output.getPosition());
}
maxAnalyzedPathsForOneInput = Math.max(maxAnalyzedPathsForOneInput, finiteStrings.size());
}
CodecUtil.writeFooter(tempInput);
writer.close();
// Sort all input/output pairs (required by FST.Builder):
tempSortedFileName = sorter.sort(tempInput.getName());
// Free disk space:
tempDir.deleteFile(tempInput.getName());
reader =
new OfflineSorter.ByteSequencesReader(
tempDir.openChecksumInput(tempSortedFileName), tempSortedFileName);
PairOutputs<Long, BytesRef> outputs =
new PairOutputs<>(PositiveIntOutputs.getSingleton(), ByteSequenceOutputs.getSingleton());
FSTCompiler<Pair<Long, BytesRef>> fstCompiler =
new FSTCompiler.Builder<>(FST.INPUT_TYPE.BYTE1, outputs).build();
// Build FST:
BytesRefBuilder previousAnalyzed = null;
BytesRefBuilder analyzed = new BytesRefBuilder();
BytesRef surface = new BytesRef();
IntsRefBuilder scratchInts = new IntsRefBuilder();
ByteArrayDataInput input = new ByteArrayDataInput();
// Used to remove duplicate surface forms (but we
// still index the hightest-weight one). We clear
// this when we see a new analyzed form, so it cannot
// grow unbounded (at most 256 entries):
Set<BytesRef> seenSurfaceForms = new HashSet<>();
int dedup = 0;
while (true) {
BytesRef bytes = reader.next();
if (bytes == null) {
break;
}
input.reset(bytes.bytes, bytes.offset, bytes.length);
short analyzedLength = input.readShort();
analyzed.growNoCopy(analyzedLength + 2);
input.readBytes(analyzed.bytes(), 0, analyzedLength);
analyzed.setLength(analyzedLength);
long cost = input.readInt();
surface.bytes = bytes.bytes;
if (hasPayloads) {
surface.length = input.readShort();
surface.offset = input.getPosition();
} else {
surface.offset = input.getPosition();
surface.length = bytes.length - surface.offset;
}
if (previousAnalyzed == null) {
previousAnalyzed = new BytesRefBuilder();
previousAnalyzed.copyBytes(analyzed.get());
seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
} else if (analyzed.get().equals(previousAnalyzed.get())) {
dedup++;
if (dedup >= maxSurfaceFormsPerAnalyzedForm) {
// More than maxSurfaceFormsPerAnalyzedForm
// dups: skip the rest:
continue;
}
if (seenSurfaceForms.contains(surface)) {
continue;
}
seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
} else {
dedup = 0;
previousAnalyzed.copyBytes(analyzed);
seenSurfaceForms.clear();
seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
}
// TODO: I think we can avoid the extra 2 bytes when
// there is no dup (dedup==0), but we'd have to fix
// the exactFirst logic ... which would be sort of
// hairy because we'd need to special case the two
// (dup/not dup)...
// NOTE: must be byte 0 so we sort before whatever
// is next
analyzed.append((byte) 0);
analyzed.append((byte) dedup);
Util.toIntsRef(analyzed.get(), scratchInts);
// System.out.println("ADD: " + scratchInts + " -> " + cost + ": " +
// surface.utf8ToString());
if (!hasPayloads) {
fstCompiler.add(scratchInts.get(), outputs.newPair(cost, BytesRef.deepCopyOf(surface)));
} else {
int payloadOffset = input.getPosition() + surface.length;
int payloadLength = bytes.length - payloadOffset;
BytesRef br = new BytesRef(surface.length + 1 + payloadLength);
System.arraycopy(surface.bytes, surface.offset, br.bytes, 0, surface.length);
br.bytes[surface.length] = PAYLOAD_SEP;
System.arraycopy(bytes.bytes, payloadOffset, br.bytes, surface.length + 1, payloadLength);
br.length = br.bytes.length;
fstCompiler.add(scratchInts.get(), outputs.newPair(cost, br));
}
}
fst = FST.fromFSTReader(fstCompiler.compile(), fstCompiler.getFSTReader());
count = newCount;
// Util.dotToFile(fst, "/tmp/suggest.dot");
} finally {
IOUtils.closeWhileHandlingException(reader, writer);
IOUtils.deleteFilesIgnoringExceptions(tempDir, tempInput.getName(), tempSortedFileName);
}
}