public override void Finish(long termsFilePointer)
{
// write primary terms dict offsets
packedIndexStart = outerInstance.m_output.GetFilePointer();
PackedInt32s.Writer w = PackedInt32s.GetWriter(outerInstance.m_output, numIndexTerms, PackedInt32s.BitsRequired(termsFilePointer), PackedInt32s.DEFAULT);
// relative to our indexStart
long upto = 0;
for (int i = 0; i < numIndexTerms; i++)
{
upto += termsPointerDeltas[i];
w.Add(upto);
}
w.Finish();
packedOffsetsStart = outerInstance.m_output.GetFilePointer();
// write offsets into the byte[] terms
w = PackedInt32s.GetWriter(outerInstance.m_output, 1 + numIndexTerms, PackedInt32s.BitsRequired(totTermLength), PackedInt32s.DEFAULT);
upto = 0;
for (int i = 0; i < numIndexTerms; i++)
{
w.Add(upto);
upto += termLengths[i];
}
w.Add(upto);
w.Finish();
// our referrer holds onto us, while other fields are
// being written, so don't tie up this RAM:
termLengths = null;
termsPointerDeltas = null;
}
private void FlushNumTerms(int totalFields)
{
int maxNumTerms = 0;
foreach (DocData dd in pendingDocs)
{
foreach (FieldData fd in dd.fields)
{
maxNumTerms |= fd.numTerms;
}
}
int bitsRequired = PackedInt32s.BitsRequired(maxNumTerms);
vectorsStream.WriteVInt32(bitsRequired);
PackedInt32s.Writer writer = PackedInt32s.GetWriterNoHeader(vectorsStream, PackedInt32s.Format.PACKED, totalFields, bitsRequired, 1);
foreach (DocData dd in pendingDocs)
{
foreach (FieldData fd in dd.fields)
{
writer.Add(fd.numTerms);
}
}
if (Debugging.AssertsEnabled)
{
Debugging.Assert(writer.Ord == totalFields - 1);
}
writer.Finish();
}
private void AddFixedSortedBytesField(FieldInfo field, IndexOutput data, IndexOutput index, IEnumerable <BytesRef> values, IEnumerable <long?> docToOrd, int length)
{
field.PutAttribute(legacyKey, LegacyDocValuesType.BYTES_FIXED_SORTED.ToString());
CodecUtil.WriteHeader(data, Lucene40DocValuesFormat.BYTES_FIXED_SORTED_CODEC_NAME_DAT, Lucene40DocValuesFormat.BYTES_FIXED_SORTED_VERSION_CURRENT);
CodecUtil.WriteHeader(index, Lucene40DocValuesFormat.BYTES_FIXED_SORTED_CODEC_NAME_IDX, Lucene40DocValuesFormat.BYTES_FIXED_SORTED_VERSION_CURRENT);
/* values */
data.WriteInt32(length);
int valueCount = 0;
foreach (BytesRef v in values)
{
data.WriteBytes(v.Bytes, v.Offset, v.Length);
valueCount++;
}
/* ordinals */
index.WriteInt32(valueCount);
int maxDoc = state.SegmentInfo.DocCount;
Debug.Assert(valueCount > 0);
PackedInt32s.Writer w = PackedInt32s.GetWriter(index, maxDoc, PackedInt32s.BitsRequired(valueCount - 1), PackedInt32s.DEFAULT);
foreach (long n in docToOrd)
{
w.Add((long)n);
}
w.Finish();
}
[ExceptionToNetNumericConvention] // LUCENENET: Private API, keeping as-is
private void AddVarIntsField(FieldInfo field, IndexOutput output, IEnumerable <long?> values, long minValue, long maxValue)
{
field.PutAttribute(legacyKey, LegacyDocValuesType.VAR_INTS.ToString());
CodecUtil.WriteHeader(output, Lucene40DocValuesFormat.VAR_INTS_CODEC_NAME, Lucene40DocValuesFormat.VAR_INTS_VERSION_CURRENT);
long delta = maxValue - minValue;
if (delta < 0)
{
// writes longs
output.WriteByte((byte)Lucene40DocValuesFormat.VAR_INTS_FIXED_64);
foreach (long?n in values)
{
output.WriteInt64(n.GetValueOrDefault());
}
}
else
{
// writes packed ints
output.WriteByte((byte)Lucene40DocValuesFormat.VAR_INTS_PACKED);
output.WriteInt64(minValue);
output.WriteInt64(0 - minValue); // default value (representation of 0)
PackedInt32s.Writer writer = PackedInt32s.GetWriter(output, state.SegmentInfo.DocCount, PackedInt32s.BitsRequired(delta), PackedInt32s.DEFAULT);
foreach (long?n in values)
{
writer.Add(n.GetValueOrDefault() - minValue);
}
writer.Finish();
}
}
private void AddVarSortedBytesField(FieldInfo field, IndexOutput data, IndexOutput index, IEnumerable <BytesRef> values, IEnumerable <long?> docToOrd)
{
field.PutAttribute(legacyKey, LegacyDocValuesType.BYTES_VAR_SORTED.ToString());
CodecUtil.WriteHeader(data, Lucene40DocValuesFormat.BYTES_VAR_SORTED_CODEC_NAME_DAT, Lucene40DocValuesFormat.BYTES_VAR_SORTED_VERSION_CURRENT);
CodecUtil.WriteHeader(index, Lucene40DocValuesFormat.BYTES_VAR_SORTED_CODEC_NAME_IDX, Lucene40DocValuesFormat.BYTES_VAR_SORTED_VERSION_CURRENT);
/* values */
long startPos = data.GetFilePointer();
int valueCount = 0;
foreach (BytesRef v in values)
{
data.WriteBytes(v.Bytes, v.Offset, v.Length);
valueCount++;
}
/* addresses */
long maxAddress = data.GetFilePointer() - startPos;
index.WriteInt64(maxAddress);
Debug.Assert(valueCount != int.MaxValue); // unsupported by the 4.0 impl
PackedInt32s.Writer w = PackedInt32s.GetWriter(index, valueCount + 1, PackedInt32s.BitsRequired(maxAddress), PackedInt32s.DEFAULT);
long currentPosition = 0;
foreach (BytesRef v in values)
{
w.Add(currentPosition);
currentPosition += v.Length;
}
// write sentinel
Debug.Assert(currentPosition == maxAddress);
w.Add(currentPosition);
w.Finish();
/* ordinals */
int maxDoc = state.SegmentInfo.DocCount;
Debug.Assert(valueCount > 0);
PackedInt32s.Writer ords = PackedInt32s.GetWriter(index, maxDoc, PackedInt32s.BitsRequired(valueCount - 1), PackedInt32s.DEFAULT);
foreach (long n in docToOrd)
{
ords.Add((long)n);
}
ords.Finish();
}
// NOTE: 4.0 file format docs are crazy/wrong here...
private void AddVarStraightBytesField(FieldInfo field, IndexOutput data, IndexOutput index, IEnumerable <BytesRef> values)
{
field.PutAttribute(legacyKey, LegacyDocValuesType.BYTES_VAR_STRAIGHT.ToString());
CodecUtil.WriteHeader(data, Lucene40DocValuesFormat.BYTES_VAR_STRAIGHT_CODEC_NAME_DAT, Lucene40DocValuesFormat.BYTES_VAR_STRAIGHT_VERSION_CURRENT);
CodecUtil.WriteHeader(index, Lucene40DocValuesFormat.BYTES_VAR_STRAIGHT_CODEC_NAME_IDX, Lucene40DocValuesFormat.BYTES_VAR_STRAIGHT_VERSION_CURRENT);
/* values */
long startPos = data.Position; // LUCENENET specific: Renamed from getFilePointer() to match FileStream
foreach (BytesRef v in values)
{
if (v != null)
{
data.WriteBytes(v.Bytes, v.Offset, v.Length);
}
}
/* addresses */
long maxAddress = data.Position - startPos; // LUCENENET specific: Renamed from getFilePointer() to match FileStream
index.WriteVInt64(maxAddress);
int maxDoc = state.SegmentInfo.DocCount;
if (Debugging.AssertsEnabled)
{
Debugging.Assert(maxDoc != int.MaxValue); // unsupported by the 4.0 impl
}
PackedInt32s.Writer w = PackedInt32s.GetWriter(index, maxDoc + 1, PackedInt32s.BitsRequired(maxAddress), PackedInt32s.DEFAULT);
long currentPosition = 0;
foreach (BytesRef v in values)
{
w.Add(currentPosition);
if (v != null)
{
currentPosition += v.Length;
}
}
// write sentinel
if (Debugging.AssertsEnabled)
{
Debugging.Assert(currentPosition == maxAddress);
}
w.Add(currentPosition);
w.Finish();
}
private void FlushFields(int totalFields, int[] fieldNums)
{
PackedInt32s.Writer writer = PackedInt32s.GetWriterNoHeader(vectorsStream, PackedInt32s.Format.PACKED, totalFields, PackedInt32s.BitsRequired(fieldNums.Length - 1), 1);
foreach (DocData dd in pendingDocs)
{
foreach (FieldData fd in dd.fields)
{
int fieldNumIndex = Array.BinarySearch(fieldNums, fd.fieldNum);
Debug.Assert(fieldNumIndex >= 0);
writer.Add(fieldNumIndex);
}
}
writer.Finish();
}
private void AddFixedDerefBytesField(FieldInfo field, IndexOutput data, IndexOutput index, IEnumerable <BytesRef> values, int length)
{
field.PutAttribute(legacyKey, LegacyDocValuesType.BYTES_FIXED_DEREF.ToString());
CodecUtil.WriteHeader(data, Lucene40DocValuesFormat.BYTES_FIXED_DEREF_CODEC_NAME_DAT, Lucene40DocValuesFormat.BYTES_FIXED_DEREF_VERSION_CURRENT);
CodecUtil.WriteHeader(index, Lucene40DocValuesFormat.BYTES_FIXED_DEREF_CODEC_NAME_IDX, Lucene40DocValuesFormat.BYTES_FIXED_DEREF_VERSION_CURRENT);
// deduplicate
JCG.SortedSet <BytesRef> dictionary = new JCG.SortedSet <BytesRef>();
foreach (BytesRef v in values)
{
dictionary.Add(v == null ? new BytesRef() : BytesRef.DeepCopyOf(v));
}
/* values */
data.WriteInt32(length);
foreach (BytesRef v in dictionary)
{
data.WriteBytes(v.Bytes, v.Offset, v.Length);
}
/* ordinals */
int valueCount = dictionary.Count;
if (Debugging.AssertsEnabled)
{
Debugging.Assert(valueCount > 0);
}
index.WriteInt32(valueCount);
int maxDoc = state.SegmentInfo.DocCount;
PackedInt32s.Writer w = PackedInt32s.GetWriter(index, maxDoc, PackedInt32s.BitsRequired(valueCount - 1), PackedInt32s.DEFAULT);
BytesRef brefDummy;
foreach (BytesRef v in values)
{
brefDummy = v;
if (v == null)
{
brefDummy = new BytesRef();
}
//int ord = dictionary.HeadSet(brefDummy).Size();
int ord = dictionary.Count(@ref => @ref.CompareTo(brefDummy) < 0);
w.Add(ord);
}
w.Finish();
}
/// <summary>
/// NOTE: This was saveInts() in Lucene.
/// </summary>
private static void SaveInt32s(int[] values, int length, DataOutput @out)
{
if (Debugging.AssertsEnabled)
{
Debugging.Assert(length > 0);
}
if (length == 1)
{
@out.WriteVInt32(values[0]);
}
else
{
bool allEqual = true;
for (int i = 1; i < length; ++i)
{
if (values[i] != values[0])
{
allEqual = false;
break;
}
}
if (allEqual)
{
@out.WriteVInt32(0);
@out.WriteVInt32(values[0]);
}
else
{
long max = 0;
for (int i = 0; i < length; ++i)
{
max |= (uint)values[i];
}
int bitsRequired = PackedInt32s.BitsRequired(max);
@out.WriteVInt32(bitsRequired);
PackedInt32s.Writer w = PackedInt32s.GetWriterNoHeader(@out, PackedInt32s.Format.PACKED, length, bitsRequired, 1);
for (int i = 0; i < length; ++i)
{
w.Add(values[i]);
}
w.Finish();
}
}
}
private void AddVarDerefBytesField(FieldInfo field, IndexOutput data, IndexOutput index, IEnumerable <BytesRef> values)
{
field.PutAttribute(legacyKey, LegacyDocValuesType.BYTES_VAR_DEREF.ToString());
CodecUtil.WriteHeader(data, Lucene40DocValuesFormat.BYTES_VAR_DEREF_CODEC_NAME_DAT, Lucene40DocValuesFormat.BYTES_VAR_DEREF_VERSION_CURRENT);
CodecUtil.WriteHeader(index, Lucene40DocValuesFormat.BYTES_VAR_DEREF_CODEC_NAME_IDX, Lucene40DocValuesFormat.BYTES_VAR_DEREF_VERSION_CURRENT);
// deduplicate
SortedSet <BytesRef> dictionary = new SortedSet <BytesRef>();
foreach (BytesRef v in values)
{
dictionary.Add(v == null ? new BytesRef() : BytesRef.DeepCopyOf(v));
}
/* values */
long startPosition = data.GetFilePointer();
long currentAddress = 0;
Dictionary <BytesRef, long> valueToAddress = new Dictionary <BytesRef, long>();
foreach (BytesRef v in dictionary)
{
currentAddress = data.GetFilePointer() - startPosition;
valueToAddress[v] = currentAddress;
WriteVShort(data, v.Length);
data.WriteBytes(v.Bytes, v.Offset, v.Length);
}
/* ordinals */
long totalBytes = data.GetFilePointer() - startPosition;
index.WriteInt64(totalBytes);
int maxDoc = state.SegmentInfo.DocCount;
PackedInt32s.Writer w = PackedInt32s.GetWriter(index, maxDoc, PackedInt32s.BitsRequired(currentAddress), PackedInt32s.DEFAULT);
foreach (BytesRef v in values)
{
w.Add(valueToAddress[v == null ? new BytesRef() : v]);
}
w.Finish();
}
/// <summary>
/// Returns a sorted array containing unique field numbers. </summary>
private int[] FlushFieldNums()
{
JCG.SortedSet <int> fieldNums = new JCG.SortedSet <int>();
foreach (DocData dd in pendingDocs)
{
foreach (FieldData fd in dd.fields)
{
fieldNums.Add(fd.fieldNum);
}
}
int numDistinctFields = fieldNums.Count;
if (Debugging.AssertsEnabled)
{
Debugging.Assert(numDistinctFields > 0);
}
int bitsRequired = PackedInt32s.BitsRequired(fieldNums.Max);
int token = (Math.Min(numDistinctFields - 1, 0x07) << 5) | bitsRequired;
vectorsStream.WriteByte((byte)token);
if (numDistinctFields - 1 >= 0x07)
{
vectorsStream.WriteVInt32(numDistinctFields - 1 - 0x07);
}
PackedInt32s.Writer writer = PackedInt32s.GetWriterNoHeader(vectorsStream, PackedInt32s.Format.PACKED, fieldNums.Count, bitsRequired, 1);
foreach (int fieldNum in fieldNums)
{
writer.Add(fieldNum);
}
writer.Finish();
int[] fns = new int[fieldNums.Count];
int i = 0;
foreach (int key in fieldNums)
{
fns[i++] = key;
}
return(fns);
}
internal virtual void AddNumericField(FieldInfo field, IEnumerable <long?> values, bool optimizeStorage)
{
long count = 0;
long minValue = long.MaxValue;
long maxValue = long.MinValue;
long gcd = 0;
bool missing = false;
// TODO: more efficient?
JCG.HashSet <long> uniqueValues = null;
if (optimizeStorage)
{
uniqueValues = new JCG.HashSet <long>();
foreach (long?nv in values)
{
long v;
if (nv == null)
{
v = 0;
missing = true;
}
else
{
v = nv.Value;
}
if (gcd != 1)
{
if (v < long.MinValue / 2 || v > long.MaxValue / 2)
{
// in that case v - minValue might overflow and make the GCD computation return
// wrong results. Since these extreme values are unlikely, we just discard
// GCD computation for them
gcd = 1;
} // minValue needs to be set first
else if (count != 0)
{
gcd = MathUtil.Gcd(gcd, v - minValue);
}
}
minValue = Math.Min(minValue, v);
maxValue = Math.Max(maxValue, v);
if (uniqueValues != null)
{
if (uniqueValues.Add(v))
{
if (uniqueValues.Count > 256)
{
uniqueValues = null;
}
}
}
++count;
}
}
else
{
foreach (var nv in values)
{
++count;
}
}
long delta = maxValue - minValue;
int format;
if (uniqueValues != null && (delta < 0L || PackedInt32s.BitsRequired(uniqueValues.Count - 1) < PackedInt32s.BitsRequired(delta)) && count <= int.MaxValue)
{
format = TABLE_COMPRESSED;
}
else if (gcd != 0 && gcd != 1)
{
format = GCD_COMPRESSED;
}
else
{
format = DELTA_COMPRESSED;
}
meta.WriteVInt32(field.Number);
meta.WriteByte((byte)Lucene45DocValuesFormat.NUMERIC);
meta.WriteVInt32(format);
if (missing)
{
meta.WriteInt64(data.Position); // LUCENENET specific: Renamed from getFilePointer() to match FileStream
WriteMissingBitset(values);
}
else
{
meta.WriteInt64(-1L);
}
meta.WriteVInt32(PackedInt32s.VERSION_CURRENT);
meta.WriteInt64(data.Position); // LUCENENET specific: Renamed from getFilePointer() to match FileStream
meta.WriteVInt64(count);
meta.WriteVInt32(BLOCK_SIZE);
switch (format)
{
case GCD_COMPRESSED:
meta.WriteInt64(minValue);
meta.WriteInt64(gcd);
BlockPackedWriter quotientWriter = new BlockPackedWriter(data, BLOCK_SIZE);
foreach (long?nv in values)
{
quotientWriter.Add((nv.GetValueOrDefault() - minValue) / gcd);
}
quotientWriter.Finish();
break;
case DELTA_COMPRESSED:
BlockPackedWriter writer = new BlockPackedWriter(data, BLOCK_SIZE);
foreach (long?nv in values)
{
writer.Add(nv.GetValueOrDefault());
}
writer.Finish();
break;
case TABLE_COMPRESSED:
// LUCENENET NOTE: diming an array and then using .CopyTo() for better efficiency than LINQ .ToArray()
long[] decode = new long[uniqueValues.Count];
uniqueValues.CopyTo(decode, 0);
Dictionary <long, int> encode = new Dictionary <long, int>();
meta.WriteVInt32(decode.Length);
for (int i = 0; i < decode.Length; i++)
{
meta.WriteInt64(decode[i]);
encode[decode[i]] = i;
}
int bitsRequired = PackedInt32s.BitsRequired(uniqueValues.Count - 1);
PackedInt32s.Writer ordsWriter = PackedInt32s.GetWriterNoHeader(data, PackedInt32s.Format.PACKED, (int)count, bitsRequired, PackedInt32s.DEFAULT_BUFFER_SIZE);
foreach (long?nv in values)
{
ordsWriter.Add(encode[nv.GetValueOrDefault()]);
}
ordsWriter.Finish();
break;
default:
throw AssertionError.Create();
}
}
internal virtual void AddNumericField(FieldInfo field, IEnumerable <long?> values, bool optimizeStorage)
{
meta.WriteVInt32(field.Number);
meta.WriteByte(MemoryDocValuesProducer.NUMBER);
meta.WriteInt64(data.GetFilePointer());
long minValue = long.MaxValue;
long maxValue = long.MinValue;
long gcd = 0;
bool missing = false;
// TODO: more efficient?
ISet <long?> uniqueValues = null;
if (optimizeStorage)
{
uniqueValues = new JCG.HashSet <long?>();
long count = 0;
foreach (var nv in values)
{
long v;
if (nv == null)
{
v = 0;
missing = true;
}
else
{
v = nv.Value;
}
if (gcd != 1)
{
if (v < long.MinValue / 2 || v > long.MaxValue / 2)
{
// in that case v - minValue might overflow and make the GCD computation return
// wrong results. Since these extreme values are unlikely, we just discard
// GCD computation for them
gcd = 1;
} // minValue needs to be set first
else if (count != 0)
{
gcd = MathUtil.Gcd(gcd, v - minValue);
}
}
minValue = Math.Min(minValue, v);
maxValue = Math.Max(maxValue, v);
if (uniqueValues != null)
{
if (uniqueValues.Add(v))
{
if (uniqueValues.Count > 256)
{
uniqueValues = null;
}
}
}
++count;
}
if (Debugging.AssertsEnabled)
{
Debugging.Assert(count == maxDoc);
}
}
if (missing)
{
long start = data.GetFilePointer();
WriteMissingBitset(values);
meta.WriteInt64(start);
meta.WriteInt64(data.GetFilePointer() - start);
}
else
{
meta.WriteInt64(-1L);
}
if (uniqueValues != null)
{
// small number of unique values
int bitsPerValue = PackedInt32s.BitsRequired(uniqueValues.Count - 1);
FormatAndBits formatAndBits = PackedInt32s.FastestFormatAndBits(maxDoc, bitsPerValue,
acceptableOverheadRatio);
if (formatAndBits.BitsPerValue == 8 && minValue >= sbyte.MinValue && maxValue <= sbyte.MaxValue)
{
meta.WriteByte(MemoryDocValuesProducer.UNCOMPRESSED); // uncompressed
foreach (var nv in values)
{
data.WriteByte((byte)nv.GetValueOrDefault());
}
}
else
{
meta.WriteByte(MemoryDocValuesProducer.TABLE_COMPRESSED); // table-compressed
long?[] decode = new long?[uniqueValues.Count];
uniqueValues.CopyTo(decode, 0);
var encode = new Dictionary <long?, int?>();
data.WriteVInt32(decode.Length);
for (int i = 0; i < decode.Length; i++)
{
data.WriteInt64(decode[i].Value);
encode[decode[i]] = i;
}
meta.WriteVInt32(PackedInt32s.VERSION_CURRENT);
data.WriteVInt32(formatAndBits.Format.Id);
data.WriteVInt32(formatAndBits.BitsPerValue);
PackedInt32s.Writer writer = PackedInt32s.GetWriterNoHeader(data, formatAndBits.Format, maxDoc,
formatAndBits.BitsPerValue, PackedInt32s.DEFAULT_BUFFER_SIZE);
foreach (var nv in values)
{
var v = encode[nv.GetValueOrDefault()];
writer.Add((long)v);
}
writer.Finish();
}
}
else if (gcd != 0 && gcd != 1)
{
meta.WriteByte(MemoryDocValuesProducer.GCD_COMPRESSED);
meta.WriteVInt32(PackedInt32s.VERSION_CURRENT);
data.WriteInt64(minValue);
data.WriteInt64(gcd);
data.WriteVInt32(MemoryDocValuesProducer.BLOCK_SIZE);
var writer = new BlockPackedWriter(data, MemoryDocValuesProducer.BLOCK_SIZE);
foreach (var nv in values)
{
writer.Add((nv.GetValueOrDefault() - minValue) / gcd);
}
writer.Finish();
}
else
{
meta.WriteByte(MemoryDocValuesProducer.DELTA_COMPRESSED); // delta-compressed
meta.WriteVInt32(PackedInt32s.VERSION_CURRENT);
data.WriteVInt32(MemoryDocValuesProducer.BLOCK_SIZE);
var writer = new BlockPackedWriter(data, MemoryDocValuesProducer.BLOCK_SIZE);
foreach (var nv in values)
{
writer.Add(nv.GetValueOrDefault());
}
writer.Finish();
}
}
private void FlushFlags(int totalFields, int[] fieldNums)
{
// check if fields always have the same flags
bool nonChangingFlags = true;
int[] fieldFlags = new int[fieldNums.Length];
Arrays.Fill(fieldFlags, -1);
bool breakOuterLoop;
foreach (DocData dd in pendingDocs)
{
breakOuterLoop = false;
foreach (FieldData fd in dd.fields)
{
int fieldNumOff = Array.BinarySearch(fieldNums, fd.fieldNum);
if (Debugging.AssertsEnabled)
{
Debugging.Assert(fieldNumOff >= 0);
}
if (fieldFlags[fieldNumOff] == -1)
{
fieldFlags[fieldNumOff] = fd.flags;
}
else if (fieldFlags[fieldNumOff] != fd.flags)
{
nonChangingFlags = false;
breakOuterLoop = true;
}
}
if (breakOuterLoop)
{
break;
}
}
if (nonChangingFlags)
{
// write one flag per field num
vectorsStream.WriteVInt32(0);
PackedInt32s.Writer writer = PackedInt32s.GetWriterNoHeader(vectorsStream, PackedInt32s.Format.PACKED, fieldFlags.Length, FLAGS_BITS, 1);
foreach (int flags in fieldFlags)
{
if (Debugging.AssertsEnabled)
{
Debugging.Assert(flags >= 0);
}
writer.Add(flags);
}
if (Debugging.AssertsEnabled)
{
Debugging.Assert(writer.Ord == fieldFlags.Length - 1);
}
writer.Finish();
}
else
{
// write one flag for every field instance
vectorsStream.WriteVInt32(1);
PackedInt32s.Writer writer = PackedInt32s.GetWriterNoHeader(vectorsStream, PackedInt32s.Format.PACKED, totalFields, FLAGS_BITS, 1);
foreach (DocData dd in pendingDocs)
{
foreach (FieldData fd in dd.fields)
{
writer.Add(fd.flags);
}
}
if (Debugging.AssertsEnabled)
{
Debugging.Assert(writer.Ord == totalFields - 1);
}
writer.Finish();
}
}
private void WriteBlock()
{
if (Debugging.AssertsEnabled)
{
Debugging.Assert(blockChunks > 0);
}
fieldsIndexOut.WriteVInt32(blockChunks);
// The trick here is that we only store the difference from the average start
// pointer or doc base, this helps save bits per value.
// And in order to prevent a few chunks that would be far from the average to
// raise the number of bits per value for all of them, we only encode blocks
// of 1024 chunks at once
// See LUCENE-4512
// doc bases
int avgChunkDocs;
if (blockChunks == 1)
{
avgChunkDocs = 0;
}
else
{
avgChunkDocs = (int)Math.Round((float)(blockDocs - docBaseDeltas[blockChunks - 1]) / (blockChunks - 1));
}
fieldsIndexOut.WriteVInt32(totalDocs - blockDocs); // docBase
fieldsIndexOut.WriteVInt32(avgChunkDocs);
int docBase = 0;
long maxDelta = 0;
for (int i = 0; i < blockChunks; ++i)
{
int delta = docBase - avgChunkDocs * i;
maxDelta |= MoveSignToLowOrderBit(delta);
docBase += docBaseDeltas[i];
}
int bitsPerDocBase = PackedInt32s.BitsRequired(maxDelta);
fieldsIndexOut.WriteVInt32(bitsPerDocBase);
PackedInt32s.Writer writer = PackedInt32s.GetWriterNoHeader(fieldsIndexOut, PackedInt32s.Format.PACKED, blockChunks, bitsPerDocBase, 1);
docBase = 0;
for (int i = 0; i < blockChunks; ++i)
{
long delta = docBase - avgChunkDocs * i;
if (Debugging.AssertsEnabled)
{
Debugging.Assert(PackedInt32s.BitsRequired(MoveSignToLowOrderBit(delta)) <= writer.BitsPerValue);
}
writer.Add(MoveSignToLowOrderBit(delta));
docBase += docBaseDeltas[i];
}
writer.Finish();
// start pointers
fieldsIndexOut.WriteVInt64(firstStartPointer);
long avgChunkSize;
if (blockChunks == 1)
{
avgChunkSize = 0;
}
else
{
avgChunkSize = (maxStartPointer - firstStartPointer) / (blockChunks - 1);
}
fieldsIndexOut.WriteVInt64(avgChunkSize);
long startPointer = 0;
maxDelta = 0;
for (int i = 0; i < blockChunks; ++i)
{
startPointer += startPointerDeltas[i];
long delta = startPointer - avgChunkSize * i;
maxDelta |= MoveSignToLowOrderBit(delta);
}
int bitsPerStartPointer = PackedInt32s.BitsRequired(maxDelta);
fieldsIndexOut.WriteVInt32(bitsPerStartPointer);
writer = PackedInt32s.GetWriterNoHeader(fieldsIndexOut, PackedInt32s.Format.PACKED, blockChunks, bitsPerStartPointer, 1);
startPointer = 0;
for (int i = 0; i < blockChunks; ++i)
{
startPointer += startPointerDeltas[i];
long delta = startPointer - avgChunkSize * i;
if (Debugging.AssertsEnabled)
{
Debugging.Assert(PackedInt32s.BitsRequired(MoveSignToLowOrderBit(delta)) <= writer.BitsPerValue);
}
writer.Add(MoveSignToLowOrderBit(delta));
}
writer.Finish();
}
public override void AddNumericField(FieldInfo field, IEnumerable <long?> values)
{
meta.WriteVInt32(field.Number);
meta.WriteByte((byte)NUMBER);
meta.WriteInt64(data.GetFilePointer());
long minValue = long.MaxValue;
long maxValue = long.MinValue;
long gcd = 0;
// TODO: more efficient?
HashSet <long> uniqueValues = null;
if (true)
{
uniqueValues = new HashSet <long>();
long count = 0;
foreach (long?nv in values)
{
Debug.Assert(nv != null);
long v = nv.Value;
if (gcd != 1)
{
if (v < long.MinValue / 2 || v > long.MaxValue / 2)
{
// in that case v - minValue might overflow and make the GCD computation return
// wrong results. Since these extreme values are unlikely, we just discard
// GCD computation for them
gcd = 1;
} // minValue needs to be set first
else if (count != 0)
{
gcd = MathUtil.Gcd(gcd, v - minValue);
}
}
minValue = Math.Min(minValue, v);
maxValue = Math.Max(maxValue, v);
if (uniqueValues != null)
{
if (uniqueValues.Add(v))
{
if (uniqueValues.Count > 256)
{
uniqueValues = null;
}
}
}
++count;
}
Debug.Assert(count == maxDoc);
}
if (uniqueValues != null)
{
// small number of unique values
int bitsPerValue = PackedInt32s.BitsRequired(uniqueValues.Count - 1);
FormatAndBits formatAndBits = PackedInt32s.FastestFormatAndBits(maxDoc, bitsPerValue, acceptableOverheadRatio);
if (formatAndBits.BitsPerValue == 8 && minValue >= sbyte.MinValue && maxValue <= sbyte.MaxValue)
{
meta.WriteByte((byte)UNCOMPRESSED); // uncompressed
foreach (long?nv in values)
{
data.WriteByte((byte)nv.GetValueOrDefault());
}
}
else
{
meta.WriteByte((byte)TABLE_COMPRESSED); // table-compressed
var decode = uniqueValues.ToArray();
var encode = new Dictionary <long, int>();
data.WriteVInt32(decode.Length);
for (int i = 0; i < decode.Length; i++)
{
data.WriteInt64(decode[i]);
encode[decode[i]] = i;
}
meta.WriteVInt32(PackedInt32s.VERSION_CURRENT);
data.WriteVInt32(formatAndBits.Format.Id);
data.WriteVInt32(formatAndBits.BitsPerValue);
PackedInt32s.Writer writer = PackedInt32s.GetWriterNoHeader(data, formatAndBits.Format, maxDoc, formatAndBits.BitsPerValue, PackedInt32s.DEFAULT_BUFFER_SIZE);
foreach (long?nv in values)
{
writer.Add(encode[nv.GetValueOrDefault()]);
}
writer.Finish();
}
}
else if (gcd != 0 && gcd != 1)
{
meta.WriteByte((byte)GCD_COMPRESSED);
meta.WriteVInt32(PackedInt32s.VERSION_CURRENT);
data.WriteInt64(minValue);
data.WriteInt64(gcd);
data.WriteVInt32(BLOCK_SIZE);
var writer = new BlockPackedWriter(data, BLOCK_SIZE);
foreach (long?nv in values)
{
writer.Add((nv.GetValueOrDefault() - minValue) / gcd);
}
writer.Finish();
}
else
{
meta.WriteByte((byte)DELTA_COMPRESSED); // delta-compressed
meta.WriteVInt32(PackedInt32s.VERSION_CURRENT);
data.WriteVInt32(BLOCK_SIZE);
var writer = new BlockPackedWriter(data, BLOCK_SIZE);
foreach (long?nv in values)
{
writer.Add(nv.GetValueOrDefault());
}
writer.Finish();
}
}
internal virtual void AddNumericField(FieldInfo field, IEnumerable <long?> values, bool optimizeStorage)
{
Meta.WriteVInt32(field.Number);
Meta.WriteByte((byte)Lucene42DocValuesProducer.NUMBER);
Meta.WriteInt64(Data.GetFilePointer());
long minValue = long.MaxValue;
long maxValue = long.MinValue;
long gcd = 0;
// TODO: more efficient?
HashSet <long> uniqueValues = null;
if (optimizeStorage)
{
uniqueValues = new HashSet <long>();
long count = 0;
foreach (long?nv in values)
{
// TODO: support this as MemoryDVFormat (and be smart about missing maybe)
long v = nv.GetValueOrDefault();
if (gcd != 1)
{
if (v < long.MinValue / 2 || v > long.MaxValue / 2)
{
// in that case v - minValue might overflow and make the GCD computation return
// wrong results. Since these extreme values are unlikely, we just discard
// GCD computation for them
gcd = 1;
} // minValue needs to be set first
else if (count != 0)
{
gcd = MathUtil.Gcd(gcd, v - minValue);
}
}
minValue = Math.Min(minValue, v);
maxValue = Math.Max(maxValue, v);
if (uniqueValues != null)
{
if (uniqueValues.Add(v))
{
if (uniqueValues.Count > 256)
{
uniqueValues = null;
}
}
}
++count;
}
Debug.Assert(count == MaxDoc);
}
if (uniqueValues != null)
{
// small number of unique values
int bitsPerValue = PackedInt32s.BitsRequired(uniqueValues.Count - 1);
FormatAndBits formatAndBits = PackedInt32s.FastestFormatAndBits(MaxDoc, bitsPerValue, AcceptableOverheadRatio);
if (formatAndBits.BitsPerValue == 8 && minValue >= sbyte.MinValue && maxValue <= sbyte.MaxValue)
{
Meta.WriteByte((byte)Lucene42DocValuesProducer.UNCOMPRESSED); // uncompressed
foreach (long?nv in values)
{
Data.WriteByte((byte)nv.GetValueOrDefault());
}
}
else
{
Meta.WriteByte((byte)Lucene42DocValuesProducer.TABLE_COMPRESSED); // table-compressed
long[] decode = uniqueValues.ToArray(/*new long?[uniqueValues.Count]*/);
var encode = new Dictionary <long, int>();
Data.WriteVInt32(decode.Length);
for (int i = 0; i < decode.Length; i++)
{
Data.WriteInt64(decode[i]);
encode[decode[i]] = i;
}
Meta.WriteVInt32(PackedInt32s.VERSION_CURRENT);
Data.WriteVInt32(formatAndBits.Format.Id);
Data.WriteVInt32(formatAndBits.BitsPerValue);
PackedInt32s.Writer writer = PackedInt32s.GetWriterNoHeader(Data, formatAndBits.Format, MaxDoc, formatAndBits.BitsPerValue, PackedInt32s.DEFAULT_BUFFER_SIZE);
foreach (long?nv in values)
{
writer.Add(encode[nv.GetValueOrDefault()]);
}
writer.Finish();
}
}
else if (gcd != 0 && gcd != 1)
{
Meta.WriteByte((byte)Lucene42DocValuesProducer.GCD_COMPRESSED);
Meta.WriteVInt32(PackedInt32s.VERSION_CURRENT);
Data.WriteInt64(minValue);
Data.WriteInt64(gcd);
Data.WriteVInt32(Lucene42DocValuesProducer.BLOCK_SIZE);
BlockPackedWriter writer = new BlockPackedWriter(Data, Lucene42DocValuesProducer.BLOCK_SIZE);
foreach (long?nv in values)
{
writer.Add((nv.GetValueOrDefault() - minValue) / gcd);
}
writer.Finish();
}
else
{
Meta.WriteByte((byte)Lucene42DocValuesProducer.DELTA_COMPRESSED); // delta-compressed
Meta.WriteVInt32(PackedInt32s.VERSION_CURRENT);
Data.WriteVInt32(Lucene42DocValuesProducer.BLOCK_SIZE);
BlockPackedWriter writer = new BlockPackedWriter(Data, Lucene42DocValuesProducer.BLOCK_SIZE);
foreach (long?nv in values)
{
writer.Add(nv.GetValueOrDefault());
}
writer.Finish();
}
}
