Terms terms = reader.terms(key.field);
final float acceptableOverheadRatio = ((Float) key.custom).floatValue();
final PagedBytes bytes = new PagedBytes(15);
int startTermsBPV;
final int termCountHardLimit;
if (maxDoc == Integer.MAX_VALUE) {
termCountHardLimit = Integer.MAX_VALUE;
} else {
termCountHardLimit = maxDoc+1;
}
// TODO: use Uninvert?
if (terms != null) {
// Try for coarse estimate for number of bits; this
// should be an underestimate most of the time, which
// is fine -- GrowableWriter will reallocate as needed
long numUniqueTerms = terms.size();
if (numUniqueTerms != -1L) {
if (numUniqueTerms > termCountHardLimit) {
// app is misusing the API (there is more than
// one term per doc); in this case we make best
// effort to load what we can (see LUCENE-2142)
numUniqueTerms = termCountHardLimit;
}
startTermsBPV = PackedInts.bitsRequired(numUniqueTerms);
} else {
startTermsBPV = 1;
}
} else {
startTermsBPV = 1;
}
MonotonicAppendingLongBuffer termOrdToBytesOffset = new MonotonicAppendingLongBuffer();
final GrowableWriter docToTermOrd = new GrowableWriter(startTermsBPV, maxDoc, acceptableOverheadRatio);
int termOrd = 0;
// TODO: use Uninvert?
if (terms != null) {
final TermsEnum termsEnum = terms.iterator(null);
DocsEnum docs = null;
while(true) {
final BytesRef term = termsEnum.next();
if (term == null) {
break;
}
if (termOrd >= termCountHardLimit) {
break;
}
termOrdToBytesOffset.add(bytes.copyUsingLengthPrefix(term));
docs = termsEnum.docs(null, docs, DocsEnum.FLAG_NONE);
while (true) {
final int docID = docs.nextDoc();
if (docID == DocIdSetIterator.NO_MORE_DOCS) {
break;
}
// Store 1+ ord into packed bits
docToTermOrd.set(docID, 1+termOrd);
}
termOrd++;
}
}
termOrdToBytesOffset.freeze();
// maybe an int-only impl?
return new SortedDocValuesImpl(bytes.freeze(true), termOrdToBytesOffset, docToTermOrd.getMutable(), termOrd);
}