@Override
void newTerm(final int termID) {
assert docState.testPoint("TermVectorsTermsWriterPerField.newTerm start");
TermVectorsPostingsArray postings = (TermVectorsPostingsArray) termsHashPerField.postingsArray;
postings.freqs[termID] = 1;
if (doVectorOffsets) {
int startOffset = fieldState.offset + offsetAttribute.startOffset();
int endOffset = fieldState.offset + offsetAttribute.endOffset();
termsHashPerField.writeVInt(1, startOffset);
termsHashPerField.writeVInt(1, endOffset - startOffset);
postings.lastOffsets[termID] = endOffset;
}
if (doVectorPositions) {
termsHashPerField.writeVInt(0, fieldState.position);
postings.lastPositions[termID] = fieldState.position;
}
}
/**
* Called once per field per document if term vectors are enabled, to write the vectors to
* RAMOutputStream, which is then quickly flushed to the real term vectors files in the Directory.
*/
@Override
void finish() throws IOException {
assert docState.testPoint("TermVectorsTermsWriterPerField.finish start");
final int numPostings = termsHashPerField.bytesHash.size();
final BytesRef flushTerm = perThread.flushTerm;
assert numPostings >= 0;
if (!doVectors || numPostings == 0) return;
if (numPostings > maxNumPostings) maxNumPostings = numPostings;
final IndexOutput tvf = perThread.doc.perDocTvf;
// This is called once, after inverting all occurrences
// of a given field in the doc. At this point we flush
// our hash into the DocWriter.
assert fieldInfo.storeTermVector;
assert perThread.vectorFieldsInOrder(fieldInfo);
perThread.doc.addField(termsHashPerField.fieldInfo.number);
TermVectorsPostingsArray postings = (TermVectorsPostingsArray) termsHashPerField.postingsArray;
// TODO: we may want to make this sort in same order
// as Codec's terms dict?
final int[] termIDs =
termsHashPerField.sortPostings(BytesRef.getUTF8SortedAsUnicodeComparator());
tvf.writeVInt(numPostings);
byte bits = 0x0;
if (doVectorPositions) bits |= TermVectorsReader.STORE_POSITIONS_WITH_TERMVECTOR;
if (doVectorOffsets) bits |= TermVectorsReader.STORE_OFFSET_WITH_TERMVECTOR;
tvf.writeByte(bits);
int lastLen = 0;
byte[] lastBytes = null;
int lastStart = 0;
final ByteSliceReader reader = perThread.vectorSliceReader;
final ByteBlockPool termBytePool = perThread.termsHashPerThread.termBytePool;
for (int j = 0; j < numPostings; j++) {
final int termID = termIDs[j];
final int freq = postings.freqs[termID];
// Get BytesRef
termBytePool.setBytesRef(flushTerm, postings.textStarts[termID]);
// Compute common byte prefix between last term and
// this term
int prefix = 0;
if (j > 0) {
while (prefix < lastLen && prefix < flushTerm.length) {
if (lastBytes[lastStart + prefix] != flushTerm.bytes[flushTerm.offset + prefix]) {
break;
}
prefix++;
}
}
lastLen = flushTerm.length;
lastBytes = flushTerm.bytes;
lastStart = flushTerm.offset;
final int suffix = flushTerm.length - prefix;
tvf.writeVInt(prefix);
tvf.writeVInt(suffix);
tvf.writeBytes(flushTerm.bytes, lastStart + prefix, suffix);
tvf.writeVInt(freq);
if (doVectorPositions) {
termsHashPerField.initReader(reader, termID, 0);
reader.writeTo(tvf);
}
if (doVectorOffsets) {
termsHashPerField.initReader(reader, termID, 1);
reader.writeTo(tvf);
}
}
termsHashPerField.reset();
// NOTE: we clear, per-field, at the thread level,
// because term vectors fully write themselves on each
// field; this saves RAM (eg if large doc has two large
// fields w/ term vectors on) because we recycle/reuse
// all RAM after each field:
perThread.termsHashPerThread.reset(false);
}
