LUCENE-3622: fix naming issues, IDV->DV, ValueType -> DV.Type, move all DV user-facing api into .index, TypePromoter pkg-private, etc

git-svn-id: https://svn.apache.org/repos/asf/lucene/dev/branches/lucene3622@1212506 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Robert Muir 2011-12-09 16:43:45 +00:00
parent 2698e09434
commit 4387ff6161
76 changed files with 1379 additions and 1445 deletions

View File

@ -497,23 +497,23 @@ New features
bytes in RAM. (Mike McCandless)
* LUCENE-3108, LUCENE-2935, LUCENE-2168, LUCENE-1231: Changes from
IndexDocValues (ColumnStrideFields):
DocValues (ColumnStrideFields):
- IndexWriter now supports typesafe dense per-document values stored in
a column like storage. IndexDocValues are stored on a per-document
a column like storage. DocValues are stored on a per-document
basis where each documents field can hold exactly one value of a given
type. IndexDocValues are provided via Fieldable and can be used in
type. DocValues are provided via Fieldable and can be used in
conjunction with stored and indexed values.
- IndexDocValues provides an entirely RAM resident document id to value
- DocValues provides an entirely RAM resident document id to value
mapping per field as well as a DocIdSetIterator based disk-resident
sequential access API relying on filesystem-caches.
- Both APIs are exposed via IndexReader and the Codec / Flex API allowing
expert users to integrate customized IndexDocValues reader and writer
expert users to integrate customized DocValues reader and writer
implementations by extending existing Codecs.
- IndexDocValues provides implementations for primitive datatypes like int,
- DocValues provides implementations for primitive datatypes like int,
long, float, double and arrays of byte. Byte based implementations further
provide storage variants like straight or dereferenced stored bytes, fixed
and variable length bytes as well as index time sorted based on

View File

@ -40,13 +40,13 @@ import org.apache.lucene.index.DocsEnum;
import org.apache.lucene.index.FieldInvertState;
import org.apache.lucene.index.Fields;
import org.apache.lucene.index.FieldsEnum;
import org.apache.lucene.index.PerDocValues;
import org.apache.lucene.index.IndexReader.AtomicReaderContext;
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.OrdTermState;
import org.apache.lucene.index.StoredFieldVisitor;
import org.apache.lucene.index.Term;
import org.apache.lucene.index.TermState;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.index.Terms;
import org.apache.lucene.index.TermsEnum;
import org.apache.lucene.search.Collector;

View File

@ -28,8 +28,8 @@ import org.apache.lucene.index.FieldInfo;
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.IndexableField;
import org.apache.lucene.index.IndexableFieldType;
import org.apache.lucene.index.values.PerDocFieldValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.PerDocFieldValues;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.util.BytesRef;
/** Defers actually loading a field's value until you ask
@ -166,7 +166,7 @@ public class LazyDocument {
}
@Override
public ValueType docValuesType() {
public DocValues.Type docValuesType() {
if (num == 0) {
return getDocument().getField(name).docValuesType();
} else {

View File

@ -20,8 +20,9 @@ import java.io.Reader;
import java.util.Comparator;
import org.apache.lucene.index.IndexableFieldType;
import org.apache.lucene.index.values.PerDocFieldValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.PerDocFieldValues;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.Type; // javadocs
import org.apache.lucene.util.BytesRef;
/**
@ -31,14 +32,14 @@ import org.apache.lucene.util.BytesRef;
* example usage, adding an int value:
*
* <pre>
* document.add(new IndexDocValuesField(name).setInt(value));
* document.add(new DocValuesField(name).setInt(value));
* </pre>
*
* For optimal performance, re-use the <code>DocValuesField</code> and
* {@link Document} instance for more than one document:
*
* <pre>
* IndexDocValuesField field = new IndexDocValuesField(name);
* DocValuesField field = new DocValuesField(name);
* Document document = new Document();
* document.add(field);
*
@ -56,7 +57,7 @@ import org.apache.lucene.util.BytesRef;
* when creating the field:
*
* <pre>
* IndexDocValuesField field = new IndexDocValuesField(name, StringField.TYPE_STORED);
* DocValuesField field = new DocValuesField(name, StringField.TYPE_STORED);
* Document document = new Document();
* document.add(field);
* for(all documents) {
@ -68,27 +69,26 @@ import org.apache.lucene.util.BytesRef;
* </pre>
*
* */
// TODO: maybe rename to DocValuesField?
public class IndexDocValuesField extends Field implements PerDocFieldValues {
public class DocValuesField extends Field implements PerDocFieldValues {
protected BytesRef bytes;
protected double doubleValue;
protected long longValue;
protected ValueType type;
protected DocValues.Type type;
protected Comparator<BytesRef> bytesComparator;
/**
* Creates a new {@link IndexDocValuesField} with the given name.
* Creates a new {@link DocValuesField} with the given name.
*/
public IndexDocValuesField(String name) {
public DocValuesField(String name) {
this(name, new FieldType());
}
public IndexDocValuesField(String name, IndexableFieldType type) {
public DocValuesField(String name, IndexableFieldType type) {
this(name, type, null);
}
public IndexDocValuesField(String name, IndexableFieldType type, String value) {
public DocValuesField(String name, IndexableFieldType type, String value) {
super(name, type);
fieldsData = value;
}
@ -99,9 +99,9 @@ public class IndexDocValuesField extends Field implements PerDocFieldValues {
}
/**
* Sets the given <code>long</code> value and sets the field's {@link ValueType} to
* {@link ValueType#VAR_INTS} unless already set. If you want to change the
* default type use {@link #setDocValuesType(ValueType)}.
* Sets the given <code>long</code> value and sets the field's {@link Type} to
* {@link Type#VAR_INTS} unless already set. If you want to change the
* default type use {@link #setDocValuesType(DocValues.Type)}.
*/
public void setInt(long value) {
setInt(value, false);
@ -113,20 +113,20 @@ public class IndexDocValuesField extends Field implements PerDocFieldValues {
* @param value
* the value to set
* @param fixed
* if <code>true</code> {@link ValueType#FIXED_INTS_64} is used
* otherwise {@link ValueType#VAR_INTS}
* if <code>true</code> {@link Type#FIXED_INTS_64} is used
* otherwise {@link Type#VAR_INTS}
*/
public void setInt(long value, boolean fixed) {
if (type == null) {
type = fixed ? ValueType.FIXED_INTS_64 : ValueType.VAR_INTS;
type = fixed ? DocValues.Type.FIXED_INTS_64 : DocValues.Type.VAR_INTS;
}
longValue = value;
}
/**
* Sets the given <code>int</code> value and sets the field's {@link ValueType} to
* {@link ValueType#VAR_INTS} unless already set. If you want to change the
* default type use {@link #setDocValuesType(ValueType)}.
* Sets the given <code>int</code> value and sets the field's {@link Type} to
* {@link Type#VAR_INTS} unless already set. If you want to change the
* default type use {@link #setDocValuesType(DocValues.Type)}.
*/
public void setInt(int value) {
setInt(value, false);
@ -138,20 +138,20 @@ public class IndexDocValuesField extends Field implements PerDocFieldValues {
* @param value
* the value to set
* @param fixed
* if <code>true</code> {@link ValueType#FIXED_INTS_32} is used
* otherwise {@link ValueType#VAR_INTS}
* if <code>true</code> {@link Type#FIXED_INTS_32} is used
* otherwise {@link Type#VAR_INTS}
*/
public void setInt(int value, boolean fixed) {
if (type == null) {
type = fixed ? ValueType.FIXED_INTS_32 : ValueType.VAR_INTS;
type = fixed ? DocValues.Type.FIXED_INTS_32 : DocValues.Type.VAR_INTS;
}
longValue = value;
}
/**
* Sets the given <code>short</code> value and sets the field's {@link ValueType} to
* {@link ValueType#VAR_INTS} unless already set. If you want to change the
* default type use {@link #setDocValuesType(ValueType)}.
* Sets the given <code>short</code> value and sets the field's {@link Type} to
* {@link Type#VAR_INTS} unless already set. If you want to change the
* default type use {@link #setDocValuesType(DocValues.Type)}.
*/
public void setInt(short value) {
setInt(value, false);
@ -163,20 +163,20 @@ public class IndexDocValuesField extends Field implements PerDocFieldValues {
* @param value
* the value to set
* @param fixed
* if <code>true</code> {@link ValueType#FIXED_INTS_16} is used
* otherwise {@link ValueType#VAR_INTS}
* if <code>true</code> {@link Type#FIXED_INTS_16} is used
* otherwise {@link Type#VAR_INTS}
*/
public void setInt(short value, boolean fixed) {
if (type == null) {
type = fixed ? ValueType.FIXED_INTS_16 : ValueType.VAR_INTS;
type = fixed ? DocValues.Type.FIXED_INTS_16 : DocValues.Type.VAR_INTS;
}
longValue = value;
}
/**
* Sets the given <code>byte</code> value and sets the field's {@link ValueType} to
* {@link ValueType#VAR_INTS} unless already set. If you want to change the
* default type use {@link #setDocValuesType(ValueType)}.
* Sets the given <code>byte</code> value and sets the field's {@link Type} to
* {@link Type#VAR_INTS} unless already set. If you want to change the
* default type use {@link #setDocValuesType(DocValues.Type)}.
*/
public void setInt(byte value) {
setInt(value, false);
@ -188,59 +188,59 @@ public class IndexDocValuesField extends Field implements PerDocFieldValues {
* @param value
* the value to set
* @param fixed
* if <code>true</code> {@link ValueType#FIXED_INTS_8} is used
* otherwise {@link ValueType#VAR_INTS}
* if <code>true</code> {@link Type#FIXED_INTS_8} is used
* otherwise {@link Type#VAR_INTS}
*/
public void setInt(byte value, boolean fixed) {
if (type == null) {
type = fixed ? ValueType.FIXED_INTS_8 : ValueType.VAR_INTS;
type = fixed ? DocValues.Type.FIXED_INTS_8 : DocValues.Type.VAR_INTS;
}
longValue = value;
}
/**
* Sets the given <code>float</code> value and sets the field's {@link ValueType}
* to {@link ValueType#FLOAT_32} unless already set. If you want to
* change the type use {@link #setDocValuesType(ValueType)}.
* Sets the given <code>float</code> value and sets the field's {@link Type}
* to {@link Type#FLOAT_32} unless already set. If you want to
* change the type use {@link #setDocValuesType(DocValues.Type)}.
*/
public void setFloat(float value) {
if (type == null) {
type = ValueType.FLOAT_32;
type = DocValues.Type.FLOAT_32;
}
doubleValue = value;
}
/**
* Sets the given <code>double</code> value and sets the field's {@link ValueType}
* to {@link ValueType#FLOAT_64} unless already set. If you want to
* change the default type use {@link #setDocValuesType(ValueType)}.
* Sets the given <code>double</code> value and sets the field's {@link Type}
* to {@link Type#FLOAT_64} unless already set. If you want to
* change the default type use {@link #setDocValuesType(DocValues.Type)}.
*/
public void setFloat(double value) {
if (type == null) {
type = ValueType.FLOAT_64;
type = DocValues.Type.FLOAT_64;
}
doubleValue = value;
}
/**
* Sets the given {@link BytesRef} value and the field's {@link ValueType}. The
* Sets the given {@link BytesRef} value and the field's {@link Type}. The
* comparator for this field is set to <code>null</code>. If a
* <code>null</code> comparator is set the default comparator for the given
* {@link ValueType} is used.
* {@link Type} is used.
*/
public void setBytes(BytesRef value, ValueType type) {
public void setBytes(BytesRef value, DocValues.Type type) {
setBytes(value, type, null);
}
/**
* Sets the given {@link BytesRef} value, the field's {@link ValueType} and the
* Sets the given {@link BytesRef} value, the field's {@link Type} and the
* field's comparator. If the {@link Comparator} is set to <code>null</code>
* the default for the given {@link ValueType} is used instead.
* the default for the given {@link Type} is used instead.
*
* @throws IllegalArgumentException
* if the value or the type are null
*/
public void setBytes(BytesRef value, ValueType type, Comparator<BytesRef> comp) {
public void setBytes(BytesRef value, DocValues.Type type, Comparator<BytesRef> comp) {
if (value == null) {
throw new IllegalArgumentException("value must not be null");
}
@ -283,16 +283,16 @@ public class IndexDocValuesField extends Field implements PerDocFieldValues {
/**
* Sets the {@link BytesRef} comparator for this field. If the field has a
* numeric {@link ValueType} the comparator will be ignored.
* numeric {@link Type} the comparator will be ignored.
*/
public void setBytesComparator(Comparator<BytesRef> comp) {
this.bytesComparator = comp;
}
/**
* Sets the {@link ValueType} for this field.
* Sets the {@link Type} for this field.
*/
public void setDocValuesType(ValueType type) {
public void setDocValuesType(DocValues.Type type) {
if (type == null) {
throw new IllegalArgumentException("Type must not be null");
}
@ -307,7 +307,7 @@ public class IndexDocValuesField extends Field implements PerDocFieldValues {
}
@Override
public ValueType docValuesType() {
public DocValues.Type docValuesType() {
return type;
}
@ -348,21 +348,21 @@ public class IndexDocValuesField extends Field implements PerDocFieldValues {
default:
throw new IllegalArgumentException("unknown type: " + type);
}
return "<" + name() + ": IndexDocValuesField " + value + ">";
return "<" + name() + ": DocValuesField " + value + ">";
}
/**
* Returns an IndexDocValuesField holding the value from
* Returns an DocValuesField holding the value from
* the provided string field, as the specified type. The
* incoming field must have a string value. The name, {@link
* FieldType} and string value are carried over from the
* incoming Field.
*/
public static IndexDocValuesField build(Field field, ValueType type) {
if (field instanceof IndexDocValuesField) {
return (IndexDocValuesField) field;
public static DocValuesField build(Field field, DocValues.Type type) {
if (field instanceof DocValuesField) {
return (DocValuesField) field;
}
final IndexDocValuesField valField = new IndexDocValuesField(field.name(), field.fieldType(), field.stringValue());
final DocValuesField valField = new DocValuesField(field.name(), field.fieldType(), field.stringValue());
switch (type) {
case BYTES_FIXED_DEREF:
case BYTES_FIXED_STRAIGHT:

View File

@ -25,10 +25,10 @@ import org.apache.lucene.analysis.Analyzer;
import org.apache.lucene.analysis.TokenStream;
import org.apache.lucene.analysis.tokenattributes.CharTermAttribute;
import org.apache.lucene.analysis.tokenattributes.OffsetAttribute;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.IndexableFieldType;
import org.apache.lucene.index.IndexableField;
import org.apache.lucene.index.values.PerDocFieldValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.PerDocFieldValues;
import org.apache.lucene.util.BytesRef;
/**
@ -302,7 +302,7 @@ public class Field implements IndexableField {
}
@Override
public ValueType docValuesType() {
public DocValues.Type docValuesType() {
return null;
}

View File

@ -25,6 +25,7 @@ import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
import org.apache.lucene.document.Document;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.codecs.Codec;
import java.io.File;
import java.io.IOException;
@ -38,9 +39,6 @@ import java.util.List;
import java.util.Map;
import org.apache.lucene.index.codecs.BlockTreeTermsReader;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.store.FSDirectory;
import org.apache.lucene.util.Bits;
import org.apache.lucene.util.BytesRef;
@ -1133,7 +1131,7 @@ public class CheckIndex {
if (fieldInfo.hasDocValues()) {
status.totalValueFields++;
final PerDocValues perDocValues = reader.perDocValues();
final IndexDocValues docValues = perDocValues.docValues(fieldInfo.name);
final DocValues docValues = perDocValues.docValues(fieldInfo.name);
if (docValues == null) {
continue;
}

View File

@ -33,7 +33,6 @@ import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.Lock;
import org.apache.lucene.store.LockObtainFailedException;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.util.Bits;
import org.apache.lucene.util.BytesRef;
import org.apache.lucene.util.InfoStream;

View File

@ -30,8 +30,7 @@ import org.apache.lucene.index.codecs.DocValuesConsumer;
import org.apache.lucene.index.codecs.DocValuesFormat;
import org.apache.lucene.index.codecs.FieldInfosWriter;
import org.apache.lucene.index.codecs.PerDocConsumer;
import org.apache.lucene.index.values.PerDocFieldValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.util.ArrayUtil;
import org.apache.lucene.util.IOUtils;
@ -311,12 +310,12 @@ final class DocFieldProcessor extends DocConsumer {
final private Map<String, DocValuesConsumerAndDocID> docValues = new HashMap<String, DocValuesConsumerAndDocID>();
final private Map<Integer, PerDocConsumer> perDocConsumers = new HashMap<Integer, PerDocConsumer>();
DocValuesConsumer docValuesConsumer(ValueType valueType, DocState docState, FieldInfo fieldInfo)
DocValuesConsumer docValuesConsumer(DocValues.Type valueType, DocState docState, FieldInfo fieldInfo)
throws IOException {
DocValuesConsumerAndDocID docValuesConsumerAndDocID = docValues.get(fieldInfo.name);
if (docValuesConsumerAndDocID != null) {
if (docState.docID == docValuesConsumerAndDocID.docID) {
throw new IllegalArgumentException("IndexDocValuesField \"" + fieldInfo.name + "\" appears more than once in this document (only one value is allowed, per field)");
throw new IllegalArgumentException("DocValuesField \"" + fieldInfo.name + "\" appears more than once in this document (only one value is allowed, per field)");
}
assert docValuesConsumerAndDocID.docID < docState.docID;
docValuesConsumerAndDocID.docID = docState.docID;

View File

@ -0,0 +1,687 @@
package org.apache.lucene.index;
/**
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
import java.io.Closeable;
import java.io.IOException;
import java.util.Comparator;
import org.apache.lucene.document.DocValuesField;
import org.apache.lucene.index.codecs.DocValuesFormat;
import org.apache.lucene.util.BytesRef;
import org.apache.lucene.util.packed.PackedInts;
/**
* {@link DocValues} provides a dense per-document typed storage for fast
* value access based on the lucene internal document id. {@link DocValues}
* exposes two distinct APIs:
* <ul>
* <li>via {@link #getSource()} providing RAM resident random access</li>
* <li>via {@link #getDirectSource()} providing on disk random access</li>
* </ul> {@link DocValues} are exposed via
* {@link IndexReader#perDocValues()} on a per-segment basis. For best
* performance {@link DocValues} should be consumed per-segment just like
* IndexReader.
* <p>
* {@link DocValues} are fully integrated into the {@link DocValuesFormat} API.
*
* @see Type for limitations and default implementation documentation
* @see DocValuesField for adding values to the index
* @see DocValuesFormat#docsConsumer(org.apache.lucene.index.PerDocWriteState) for
* customization
* @lucene.experimental
*/
public abstract class DocValues implements Closeable {
public static final DocValues[] EMPTY_ARRAY = new DocValues[0];
private volatile SourceCache cache = new SourceCache.DirectSourceCache();
private final Object cacheLock = new Object();
/**
* Loads a new {@link Source} instance for this {@link DocValues} field
* instance. Source instances returned from this method are not cached. It is
* the callers responsibility to maintain the instance and release its
* resources once the source is not needed anymore.
* <p>
* For managed {@link Source} instances see {@link #getSource()}.
*
* @see #getSource()
* @see #setCache(SourceCache)
*/
public abstract Source load() throws IOException;
/**
* Returns a {@link Source} instance through the current {@link SourceCache}.
* Iff no {@link Source} has been loaded into the cache so far the source will
* be loaded through {@link #load()} and passed to the {@link SourceCache}.
* The caller of this method should not close the obtained {@link Source}
* instance unless it is not needed for the rest of its life time.
* <p>
* {@link Source} instances obtained from this method are closed / released
* from the cache once this {@link DocValues} instance is closed by the
* {@link IndexReader}, {@link Fields} or {@link FieldsEnum} the
* {@link DocValues} was created from.
*/
public Source getSource() throws IOException {
return cache.load(this);
}
/**
* Returns a disk resident {@link Source} instance. Direct Sources are not
* cached in the {@link SourceCache} and should not be shared between threads.
*/
public abstract Source getDirectSource() throws IOException;
/**
* Returns the {@link Type} of this {@link DocValues} instance
*/
public abstract Type type();
/**
* Closes this {@link DocValues} instance. This method should only be called
* by the creator of this {@link DocValues} instance. API users should not
* close {@link DocValues} instances.
*/
public void close() throws IOException {
cache.close(this);
}
/**
* Returns the size per value in bytes or <code>-1</code> iff size per value
* is variable.
*
* @return the size per value in bytes or <code>-1</code> iff size per value
* is variable.
*/
public int getValueSize() {
return -1;
}
/**
* Sets the {@link SourceCache} used by this {@link DocValues} instance. This
* method should be called before {@link #load()} is called. All {@link Source} instances in the currently used cache will be closed
* before the new cache is installed.
* <p>
* Note: All instances previously obtained from {@link #load()} will be lost.
*
* @throws IllegalArgumentException
* if the given cache is <code>null</code>
*
*/
public void setCache(SourceCache cache) {
if (cache == null)
throw new IllegalArgumentException("cache must not be null");
synchronized (cacheLock) {
SourceCache toClose = this.cache;
this.cache = cache;
toClose.close(this);
}
}
/**
* Source of per document values like long, double or {@link BytesRef}
* depending on the {@link DocValues} fields {@link Type}. Source
* implementations provide random access semantics similar to array lookups
* <p>
* @see DocValues#getSource()
* @see DocValues#getDirectSource()
*/
public static abstract class Source {
protected final Type type;
protected Source(Type type) {
this.type = type;
}
/**
* Returns a <tt>long</tt> for the given document id or throws an
* {@link UnsupportedOperationException} if this source doesn't support
* <tt>long</tt> values.
*
* @throws UnsupportedOperationException
* if this source doesn't support <tt>long</tt> values.
*/
public long getInt(int docID) {
throw new UnsupportedOperationException("ints are not supported");
}
/**
* Returns a <tt>double</tt> for the given document id or throws an
* {@link UnsupportedOperationException} if this source doesn't support
* <tt>double</tt> values.
*
* @throws UnsupportedOperationException
* if this source doesn't support <tt>double</tt> values.
*/
public double getFloat(int docID) {
throw new UnsupportedOperationException("floats are not supported");
}
/**
* Returns a {@link BytesRef} for the given document id or throws an
* {@link UnsupportedOperationException} if this source doesn't support
* <tt>byte[]</tt> values.
* @throws IOException
*
* @throws UnsupportedOperationException
* if this source doesn't support <tt>byte[]</tt> values.
*/
public BytesRef getBytes(int docID, BytesRef ref) {
throw new UnsupportedOperationException("bytes are not supported");
}
/**
* Returns the {@link Type} of this source.
*
* @return the {@link Type} of this source.
*/
public Type type() {
return type;
}
/**
* Returns <code>true</code> iff this {@link Source} exposes an array via
* {@link #getArray()} otherwise <code>false</code>.
*
* @return <code>true</code> iff this {@link Source} exposes an array via
* {@link #getArray()} otherwise <code>false</code>.
*/
public boolean hasArray() {
return false;
}
/**
* Returns the internal array representation iff this {@link Source} uses an
* array as its inner representation, otherwise <code>UOE</code>.
*/
public Object getArray() {
throw new UnsupportedOperationException("getArray is not supported");
}
/**
* If this {@link Source} is sorted this method will return an instance of
* {@link SortedSource} otherwise <code>UOE</code>
*/
public SortedSource asSortedSource() {
throw new UnsupportedOperationException("asSortedSource is not supported");
}
}
/**
* A sorted variant of {@link Source} for <tt>byte[]</tt> values per document.
* <p>
*/
public static abstract class SortedSource extends Source {
private final Comparator<BytesRef> comparator;
protected SortedSource(Type type, Comparator<BytesRef> comparator) {
super(type);
this.comparator = comparator;
}
@Override
public BytesRef getBytes(int docID, BytesRef bytesRef) {
final int ord = ord(docID);
if (ord < 0) {
bytesRef.length = 0;
} else {
getByOrd(ord , bytesRef);
}
return bytesRef;
}
/**
* Returns ord for specified docID. Ord is dense, ie, starts at 0, then increments by 1
* for the next (as defined by {@link Comparator} value.
*/
public abstract int ord(int docID);
/** Returns value for specified ord. */
public abstract BytesRef getByOrd(int ord, BytesRef bytesRef);
/** Return true if it's safe to call {@link
* #getDocToOrd}. */
public boolean hasPackedDocToOrd() {
return false;
}
/**
* Returns the PackedInts.Reader impl that maps document to ord.
*/
public abstract PackedInts.Reader getDocToOrd();
/**
* Returns the comparator used to order the BytesRefs.
*/
public Comparator<BytesRef> getComparator() {
return comparator;
}
/**
* Performs a lookup by value.
*
* @param value
* the value to look up
* @param spare
* a spare {@link BytesRef} instance used to compare internal
* values to the given value. Must not be <code>null</code>
* @return the given values ordinal if found or otherwise
* <code>(-(ord)-1)</code>, defined as the ordinal of the first
* element that is greater than the given value. This guarantees
* that the return value will always be &gt;= 0 if the given value
* is found.
*/
public int getByValue(BytesRef value, BytesRef spare) {
return binarySearch(value, spare, 0, getValueCount() - 1);
}
private int binarySearch(BytesRef b, BytesRef bytesRef, int low,
int high) {
int mid = 0;
while (low <= high) {
mid = (low + high) >>> 1;
getByOrd(mid, bytesRef);
final int cmp = comparator.compare(bytesRef, b);
if (cmp < 0) {
low = mid + 1;
} else if (cmp > 0) {
high = mid - 1;
} else {
return mid;
}
}
assert comparator.compare(bytesRef, b) != 0;
return -(low + 1);
}
@Override
public SortedSource asSortedSource() {
return this;
}
/**
* Returns the number of unique values in this sorted source
*/
public abstract int getValueCount();
}
/** Returns a Source that always returns default (missing)
* values for all documents. */
public static Source getDefaultSource(final Type type) {
return new Source(type) {
@Override
public long getInt(int docID) {
return 0;
}
@Override
public double getFloat(int docID) {
return 0.0;
}
@Override
public BytesRef getBytes(int docID, BytesRef ref) {
ref.length = 0;
return ref;
}
};
}
/** Returns a SortedSource that always returns default (missing)
* values for all documents. */
public static SortedSource getDefaultSortedSource(final Type type, final int size) {
final PackedInts.Reader docToOrd = new PackedInts.Reader() {
@Override
public long get(int index) {
return 0;
}
@Override
public int getBitsPerValue() {
return 0;
}
@Override
public int size() {
return size;
}
@Override
public boolean hasArray() {
return false;
}
@Override
public Object getArray() {
return null;
}
};
return new SortedSource(type, BytesRef.getUTF8SortedAsUnicodeComparator()) {
@Override
public BytesRef getBytes(int docID, BytesRef ref) {
ref.length = 0;
return ref;
}
@Override
public int ord(int docID) {
return 0;
}
@Override
public BytesRef getByOrd(int ord, BytesRef bytesRef) {
assert ord == 0;
bytesRef.length = 0;
return bytesRef;
}
@Override
public boolean hasPackedDocToOrd() {
return true;
}
@Override
public PackedInts.Reader getDocToOrd() {
return docToOrd;
}
@Override
public int getByValue(BytesRef value, BytesRef spare) {
if (value.length == 0) {
return 0;
} else {
return -1;
}
}
@Override
public int getValueCount() {
return 1;
}
};
}
/**
* <code>Type</code> specifies the {@link DocValues} type for a
* certain field. A <code>Type</code> only defines the data type for a field
* while the actual implementation used to encode and decode the values depends
* on the the {@link DocValuesFormat#docsConsumer} and {@link DocValuesFormat#docsProducer} methods.
*
* @lucene.experimental
*/
public static enum Type {
/**
* A variable bit signed integer value. By default this type uses
* {@link PackedInts} to compress the values, as an offset
* from the minimum value, as long as the value range
* fits into 2<sup>63</sup>-1. Otherwise,
* the default implementation falls back to fixed size 64bit
* integers ({@link #FIXED_INTS_64}).
* <p>
* NOTE: this type uses <tt>0</tt> as the default value without any
* distinction between provided <tt>0</tt> values during indexing. All
* documents without an explicit value will use <tt>0</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
VAR_INTS,
/**
* A 8 bit signed integer value. {@link Source} instances of
* this type return a <tt>byte</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0</tt> as the default value without any
* distinction between provided <tt>0</tt> values during indexing. All
* documents without an explicit value will use <tt>0</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FIXED_INTS_8,
/**
* A 16 bit signed integer value. {@link Source} instances of
* this type return a <tt>short</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0</tt> as the default value without any
* distinction between provided <tt>0</tt> values during indexing. All
* documents without an explicit value will use <tt>0</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FIXED_INTS_16,
/**
* A 32 bit signed integer value. {@link Source} instances of
* this type return a <tt>int</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0</tt> as the default value without any
* distinction between provided <tt>0</tt> values during indexing. All
* documents without an explicit value will use <tt>0</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FIXED_INTS_32,
/**
* A 64 bit signed integer value. {@link Source} instances of
* this type return a <tt>long</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0</tt> as the default value without any
* distinction between provided <tt>0</tt> values during indexing. All
* documents without an explicit value will use <tt>0</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FIXED_INTS_64,
/**
* A 32 bit floating point value. By default there is no compression
* applied. To fit custom float values into less than 32bit either a custom
* implementation is needed or values must be encoded into a
* {@link #BYTES_FIXED_STRAIGHT} type. {@link Source} instances of
* this type return a <tt>float</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0.0f</tt> as the default value without any
* distinction between provided <tt>0.0f</tt> values during indexing. All
* documents without an explicit value will use <tt>0.0f</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FLOAT_32,
/**
*
* A 64 bit floating point value. By default there is no compression
* applied. To fit custom float values into less than 64bit either a custom
* implementation is needed or values must be encoded into a
* {@link #BYTES_FIXED_STRAIGHT} type. {@link Source} instances of
* this type return a <tt>double</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0.0d</tt> as the default value without any
* distinction between provided <tt>0.0d</tt> values during indexing. All
* documents without an explicit value will use <tt>0.0d</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FLOAT_64,
// TODO(simonw): -- shouldn't lucene decide/detect straight vs
// deref, as well fixed vs var?
/**
* A fixed length straight byte[]. All values added to
* such a field must be of the same length. All bytes are stored sequentially
* for fast offset access.
* <p>
* NOTE: this type uses <tt>0 byte</tt> filled byte[] based on the length of the first seen
* value as the default value without any distinction between explicitly
* provided values during indexing. All documents without an explicit value
* will use the default instead.Custom default values must be assigned explicitly.
* </p>
*/
BYTES_FIXED_STRAIGHT,
/**
* A fixed length dereferenced byte[] variant. Fields with
* this type only store distinct byte values and store an additional offset
* pointer per document to dereference the shared byte[].
* Use this type if your documents may share the same byte[].
* <p>
* NOTE: Fields of this type will not store values for documents without and
* explicitly provided value. If a documents value is accessed while no
* explicit value is stored the returned {@link BytesRef} will be a 0-length
* reference. Custom default values must be assigned explicitly.
* </p>
*/
BYTES_FIXED_DEREF,
/**
* Variable length straight stored byte[] variant. All bytes are
* stored sequentially for compactness. Usage of this type via the
* disk-resident API might yield performance degradation since no additional
* index is used to advance by more than one document value at a time.
* <p>
* NOTE: Fields of this type will not store values for documents without an
* explicitly provided value. If a documents value is accessed while no
* explicit value is stored the returned {@link BytesRef} will be a 0-length
* byte[] reference. Custom default values must be assigned explicitly.
* </p>
*/
BYTES_VAR_STRAIGHT,
/**
* A variable length dereferenced byte[]. Just like
* {@link #BYTES_FIXED_DEREF}, but allowing each
* document's value to be a different length.
* <p>
* NOTE: Fields of this type will not store values for documents without and
* explicitly provided value. If a documents value is accessed while no
* explicit value is stored the returned {@link BytesRef} will be a 0-length
* reference. Custom default values must be assigned explicitly.
* </p>
*/
BYTES_VAR_DEREF,
/**
* A variable length pre-sorted byte[] variant. Just like
* {@link #BYTES_FIXED_SORTED}, but allowing each
* document's value to be a different length.
* <p>
* NOTE: Fields of this type will not store values for documents without and
* explicitly provided value. If a documents value is accessed while no
* explicit value is stored the returned {@link BytesRef} will be a 0-length
* reference.Custom default values must be assigned explicitly.
* </p>
*
* @see SortedSource
*/
BYTES_VAR_SORTED,
/**
* A fixed length pre-sorted byte[] variant. Fields with this type only
* store distinct byte values and store an additional offset pointer per
* document to dereference the shared byte[]. The stored
* byte[] is presorted, by default by unsigned byte order,
* and allows access via document id, ordinal and by-value.
* Use this type if your documents may share the same byte[].
* <p>
* NOTE: Fields of this type will not store values for documents without and
* explicitly provided value. If a documents value is accessed while no
* explicit value is stored the returned {@link BytesRef} will be a 0-length
* reference. Custom default values must be assigned
* explicitly.
* </p>
*
* @see SortedSource
*/
BYTES_FIXED_SORTED
}
/**
* Abstract base class for {@link DocValues} {@link Source} cache.
* <p>
* {@link Source} instances loaded via {@link DocValues#load()} are entirely memory resident
* and need to be maintained by the caller. Each call to
* {@link DocValues#load()} will cause an entire reload of
* the underlying data. Source instances obtained from
* {@link DocValues#getSource()} and {@link DocValues#getSource()}
* respectively are maintained by a {@link SourceCache} that is closed (
* {@link #close(DocValues)}) once the {@link IndexReader} that created the
* {@link DocValues} instance is closed.
* <p>
* Unless {@link Source} instances are managed by another entity it is
* recommended to use the cached variants to obtain a source instance.
* <p>
* Implementation of this API must be thread-safe.
*
* @see DocValues#setCache(SourceCache)
* @see DocValues#getSource()
*
* @lucene.experimental
*/
public static abstract class SourceCache {
/**
* Atomically loads a {@link Source} into the cache from the given
* {@link DocValues} and returns it iff no other {@link Source} has already
* been cached. Otherwise the cached source is returned.
* <p>
* This method will not return <code>null</code>
*/
public abstract Source load(DocValues values) throws IOException;
/**
* Atomically invalidates the cached {@link Source}
* instances if any and empties the cache.
*/
public abstract void invalidate(DocValues values);
/**
* Atomically closes the cache and frees all resources.
*/
public synchronized void close(DocValues values) {
invalidate(values);
}
/**
* Simple per {@link DocValues} instance cache implementation that holds a
* {@link Source} a member variable.
* <p>
* If a {@link DirectSourceCache} instance is closed or invalidated the cached
* reference are simply set to <code>null</code>
*/
public static final class DirectSourceCache extends SourceCache {
private Source ref;
public synchronized Source load(DocValues values) throws IOException {
if (ref == null) {
ref = values.load();
}
return ref;
}
public synchronized void invalidate(DocValues values) {
ref = null;
}
}
}
}

View File

@ -1,6 +1,6 @@
package org.apache.lucene.index;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues;
/**
* Licensed to the Apache Software Foundation (ASF) under one or more
@ -25,7 +25,7 @@ public final class FieldInfo {
public final int number;
public boolean isIndexed;
private ValueType docValues;
private DocValues.Type docValues;
// true if term vector for this field should be stored
@ -57,7 +57,7 @@ public final class FieldInfo {
*/
public FieldInfo(String name, boolean isIndexed, int number, boolean storeTermVector,
boolean storePositionWithTermVector, boolean storeOffsetWithTermVector,
boolean omitNorms, boolean storePayloads, IndexOptions indexOptions, ValueType docValues) {
boolean omitNorms, boolean storePayloads, IndexOptions indexOptions, DocValues.Type docValues) {
this.name = name;
this.isIndexed = isIndexed;
this.number = number;
@ -119,13 +119,13 @@ public final class FieldInfo {
assert this.indexOptions == IndexOptions.DOCS_AND_FREQS_AND_POSITIONS || !this.storePayloads;
}
void setDocValuesType(ValueType v) {
void setDocValuesType(DocValues.Type v) {
if (docValues == null) {
docValues = v;
}
}
public void resetDocValuesType(ValueType v) {
public void resetDocValuesType(DocValues.Type v) {
if (docValues != null) {
docValues = v;
}
@ -135,7 +135,7 @@ public final class FieldInfo {
return docValues != null;
}
public ValueType getDocValuesType() {
public DocValues.Type getDocValuesType() {
return docValues;
}

View File

@ -25,7 +25,7 @@ import java.util.SortedMap;
import java.util.TreeMap;
import org.apache.lucene.index.FieldInfo.IndexOptions;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues;
/** Access to the Field Info file that describes document fields and whether or
* not they are indexed. Each segment has a separate Field Info file. Objects
@ -299,7 +299,7 @@ public final class FieldInfos implements Iterable<FieldInfo> {
*/
synchronized public FieldInfo addOrUpdate(String name, boolean isIndexed, boolean storeTermVector,
boolean storePositionWithTermVector, boolean storeOffsetWithTermVector,
boolean omitNorms, boolean storePayloads, IndexOptions indexOptions, ValueType docValues) {
boolean omitNorms, boolean storePayloads, IndexOptions indexOptions, DocValues.Type docValues) {
return addOrUpdateInternal(name, -1, isIndexed, storeTermVector, storePositionWithTermVector,
storeOffsetWithTermVector, omitNorms, storePayloads, indexOptions, docValues);
}
@ -322,7 +322,7 @@ public final class FieldInfos implements Iterable<FieldInfo> {
synchronized private FieldInfo addOrUpdateInternal(String name, int preferredFieldNumber, boolean isIndexed,
boolean storeTermVector, boolean storePositionWithTermVector, boolean storeOffsetWithTermVector,
boolean omitNorms, boolean storePayloads, IndexOptions indexOptions, ValueType docValues) {
boolean omitNorms, boolean storePayloads, IndexOptions indexOptions, DocValues.Type docValues) {
if (globalFieldNumbers == null) {
throw new IllegalStateException("FieldInfos are read-only, create a new instance with a global field map to make modifications to FieldInfos");
}
@ -351,7 +351,7 @@ public final class FieldInfos implements Iterable<FieldInfo> {
*/
private FieldInfo addInternal(String name, int fieldNumber, boolean isIndexed,
boolean storeTermVector, boolean storePositionWithTermVector,
boolean storeOffsetWithTermVector, boolean omitNorms, boolean storePayloads, IndexOptions indexOptions, ValueType docValuesType) {
boolean storeOffsetWithTermVector, boolean omitNorms, boolean storePayloads, IndexOptions indexOptions, DocValues.Type docValuesType) {
// don't check modifiable here since we use that to initially build up FIs
if (globalFieldNumbers != null) {
globalFieldNumbers.setIfNotSet(fieldNumber, name);

View File

@ -17,7 +17,6 @@ package org.apache.lucene.index;
* limitations under the License.
*/
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.store.Directory;
import org.apache.lucene.util.Bits;
import org.apache.lucene.util.BytesRef;

View File

@ -28,8 +28,6 @@ import java.util.concurrent.atomic.AtomicInteger;
import org.apache.lucene.document.Document;
import org.apache.lucene.document.DocumentStoredFieldVisitor;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.search.FieldCache; // javadocs
import org.apache.lucene.search.SearcherManager; // javadocs
import org.apache.lucene.store.*;
@ -1148,7 +1146,7 @@ public abstract class IndexReader implements Cloneable,Closeable {
throw new UnsupportedOperationException("This reader does not support this method.");
}
public final IndexDocValues docValues(String field) throws IOException {
public final DocValues docValues(String field) throws IOException {
ensureOpen();
final PerDocValues perDoc = perDocValues();
if (perDoc == null) {

View File

@ -23,8 +23,7 @@ import java.io.Reader;
import org.apache.lucene.analysis.Analyzer;
import org.apache.lucene.analysis.TokenStream;
import org.apache.lucene.document.NumericField;
import org.apache.lucene.index.values.PerDocFieldValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.util.BytesRef;
// TODO: how to handle versioning here...?
@ -80,7 +79,7 @@ public interface IndexableField {
public PerDocFieldValues docValues();
/** DocValues type; only used if docValues is non-null */
public ValueType docValuesType();
public DocValues.Type docValuesType();
/**
* Creates the TokenStream used for indexing this field. If appropriate,

View File

@ -1,4 +1,4 @@
package org.apache.lucene.index.values;
package org.apache.lucene.index;
/**
* Licensed to the Apache Software Foundation (ASF) under one or more
@ -24,20 +24,20 @@ import org.apache.lucene.util.ReaderUtil;
/**
* A wrapper for compound IndexReader providing access to per segment
* {@link IndexDocValues}
* {@link DocValues}
*
* @lucene.experimental
* @lucene.internal
*/
public class MultiIndexDocValues extends IndexDocValues {
public class MultiDocValues extends DocValues {
public static class DocValuesIndex {
public final static DocValuesIndex[] EMPTY_ARRAY = new DocValuesIndex[0];
final int start;
final int length;
final IndexDocValues docValues;
final DocValues docValues;
public DocValuesIndex(IndexDocValues docValues, int start, int length) {
public DocValuesIndex(DocValues docValues, int start, int length) {
this.docValues = docValues;
this.start = start;
this.length = length;
@ -46,15 +46,15 @@ public class MultiIndexDocValues extends IndexDocValues {
private DocValuesIndex[] docValuesIdx;
private int[] starts;
private ValueType type;
private Type type;
private int valueSize;
public MultiIndexDocValues() {
public MultiDocValues() {
starts = new int[0];
docValuesIdx = new DocValuesIndex[0];
}
public MultiIndexDocValues(DocValuesIndex[] docValuesIdx) {
public MultiDocValues(DocValuesIndex[] docValuesIdx) {
reset(docValuesIdx);
}
@ -63,7 +63,7 @@ public class MultiIndexDocValues extends IndexDocValues {
return new MultiSource(docValuesIdx, starts, false);
}
public IndexDocValues reset(DocValuesIndex[] docValuesIdx) {
public DocValues reset(DocValuesIndex[] docValuesIdx) {
final int[] start = new int[docValuesIdx.length];
TypePromoter promoter = TypePromoter.getIdentityPromoter();
for (int i = 0; i < docValuesIdx.length; i++) {
@ -86,11 +86,11 @@ public class MultiIndexDocValues extends IndexDocValues {
return this;
}
public static class EmptyDocValues extends IndexDocValues {
public static class EmptyDocValues extends DocValues {
final int maxDoc;
final Source emptySource;
public EmptyDocValues(int maxDoc, ValueType type) {
public EmptyDocValues(int maxDoc, Type type) {
this.maxDoc = maxDoc;
this.emptySource = new EmptySource(type);
}
@ -101,7 +101,7 @@ public class MultiIndexDocValues extends IndexDocValues {
}
@Override
public ValueType type() {
public Type type() {
return emptySource.type();
}
@ -169,10 +169,10 @@ public class MultiIndexDocValues extends IndexDocValues {
}
}
// TODO: this is dup of IndexDocValues.getDefaultSource()?
// TODO: this is dup of DocValues.getDefaultSource()?
private static class EmptySource extends Source {
public EmptySource(ValueType type) {
public EmptySource(Type type) {
super(type);
}
@ -195,7 +195,7 @@ public class MultiIndexDocValues extends IndexDocValues {
}
@Override
public ValueType type() {
public Type type() {
return type;
}

View File

@ -17,7 +17,6 @@ package org.apache.lucene.index;
* limitations under the License.
*/
import org.apache.lucene.index.values.MultiIndexDocValues;
import org.apache.lucene.util.PriorityQueue;
import org.apache.lucene.util.ReaderUtil;
@ -45,7 +44,7 @@ public final class MultiFieldsEnum extends FieldsEnum {
// Re-used TermsEnum
private final MultiTermsEnum terms;
private final MultiIndexDocValues docValues;
private final MultiDocValues docValues;
private final Fields fields;
@ -57,7 +56,7 @@ public final class MultiFieldsEnum extends FieldsEnum {
this.fields = fields;
terms = new MultiTermsEnum(subSlices);
queue = new FieldMergeQueue(subs.length);
docValues = new MultiIndexDocValues();
docValues = new MultiDocValues();
top = new FieldsEnumWithSlice[subs.length];
List<FieldsEnumWithSlice> enumWithSlices = new ArrayList<FieldsEnumWithSlice>();

View File

@ -17,17 +17,12 @@ package org.apache.lucene.index;
*/
import java.io.IOException;
import java.util.ArrayList;
import java.util.Collection;
import java.util.List;
import java.util.Map;
import java.util.TreeSet;
import java.util.concurrent.ConcurrentHashMap;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.MultiIndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.values.MultiIndexDocValues.DocValuesIndex;
import org.apache.lucene.index.MultiDocValues.DocValuesIndex;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.util.IOUtils;
import org.apache.lucene.util.ReaderUtil;
import org.apache.lucene.util.ReaderUtil.Gather;
@ -48,7 +43,7 @@ import org.apache.lucene.util.ReaderUtil.Gather;
public class MultiPerDocValues extends PerDocValues {
private final PerDocValues[] subs;
private final ReaderUtil.Slice[] subSlices;
private final Map<String, IndexDocValues> docValues = new ConcurrentHashMap<String, IndexDocValues>();
private final Map<String, DocValues> docValues = new ConcurrentHashMap<String, DocValues>();
public MultiPerDocValues(PerDocValues[] subs, ReaderUtil.Slice[] subSlices) {
this.subs = subs;
@ -106,40 +101,40 @@ public class MultiPerDocValues extends PerDocValues {
return perDocValues;
}
public IndexDocValues docValues(String field) throws IOException {
IndexDocValues result = docValues.get(field);
public DocValues docValues(String field) throws IOException {
DocValues result = docValues.get(field);
if (result == null) {
// Lazy init: first time this field is requested, we
// create & add to docValues:
final List<MultiIndexDocValues.DocValuesIndex> docValuesIndex = new ArrayList<MultiIndexDocValues.DocValuesIndex>();
final List<MultiDocValues.DocValuesIndex> docValuesIndex = new ArrayList<MultiDocValues.DocValuesIndex>();
int docsUpto = 0;
ValueType type = null;
DocValues.Type type = null;
// Gather all sub-readers that share this field
for (int i = 0; i < subs.length; i++) {
IndexDocValues values = subs[i].docValues(field);
DocValues values = subs[i].docValues(field);
final int start = subSlices[i].start;
final int length = subSlices[i].length;
if (values != null) {
if (docsUpto != start) {
type = values.type();
docValuesIndex.add(new MultiIndexDocValues.DocValuesIndex(
new MultiIndexDocValues.EmptyDocValues(start, type), docsUpto, start
docValuesIndex.add(new MultiDocValues.DocValuesIndex(
new MultiDocValues.EmptyDocValues(start, type), docsUpto, start
- docsUpto));
}
docValuesIndex.add(new MultiIndexDocValues.DocValuesIndex(values, start,
docValuesIndex.add(new MultiDocValues.DocValuesIndex(values, start,
length));
docsUpto = start + length;
} else if (i + 1 == subs.length && !docValuesIndex.isEmpty()) {
docValuesIndex.add(new MultiIndexDocValues.DocValuesIndex(
new MultiIndexDocValues.EmptyDocValues(start, type), docsUpto, start
docValuesIndex.add(new MultiDocValues.DocValuesIndex(
new MultiDocValues.EmptyDocValues(start, type), docsUpto, start
- docsUpto));
}
}
if (docValuesIndex.isEmpty()) {
return null;
}
result = new MultiIndexDocValues(
result = new MultiDocValues(
docValuesIndex.toArray(DocValuesIndex.EMPTY_ARRAY));
docValues.put(field, result);
}

View File

@ -22,7 +22,6 @@ import java.util.Collection;
import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.util.Bits;
import org.apache.lucene.util.BytesRef;
import org.apache.lucene.util.ReaderUtil;

View File

@ -17,8 +17,6 @@ package org.apache.lucene.index;
* limitations under the License.
*/
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.util.Bits;
import org.apache.lucene.util.BytesRef;
import org.apache.lucene.util.MapBackedSet;
@ -472,7 +470,7 @@ public class ParallelReader extends IndexReader {
// Single instance of this, per ParallelReader instance
private static final class ParallelPerDocs extends PerDocValues {
final TreeMap<String,IndexDocValues> fields = new TreeMap<String,IndexDocValues>();
final TreeMap<String,DocValues> fields = new TreeMap<String,DocValues>();
void addField(String field, IndexReader r) throws IOException {
PerDocValues perDocs = MultiPerDocValues.getPerDocs(r);
@ -487,7 +485,7 @@ public class ParallelReader extends IndexReader {
}
@Override
public IndexDocValues docValues(String field) throws IOException {
public DocValues docValues(String field) throws IOException {
return fields.get(field);
}
}

View File

@ -1,4 +1,4 @@
package org.apache.lucene.index.values;
package org.apache.lucene.index;
/**
* Licensed to the Apache Software Foundation (ASF) under one or more
@ -18,13 +18,15 @@ package org.apache.lucene.index.values;
*/
import java.util.Comparator;
import org.apache.lucene.document.IndexDocValuesField;
import org.apache.lucene.index.codecs.DocValuesConsumer;
import org.apache.lucene.document.DocValuesField;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.Type; // javadocs
import org.apache.lucene.index.codecs.DocValuesConsumer; // javadocs
import org.apache.lucene.util.BytesRef;
/**
* Per document and field values consumed by {@link DocValuesConsumer}.
* @see IndexDocValuesField
* @see DocValuesField
*
* @lucene.experimental
*/
@ -46,19 +48,19 @@ public interface PerDocFieldValues {
public void setFloat(double value);
/**
* Sets the given {@link BytesRef} value and the field's {@link ValueType}. The
* Sets the given {@link BytesRef} value and the field's {@link Type}. The
* comparator for this field is set to <code>null</code>. If a
* <code>null</code> comparator is set the default comparator for the given
* {@link ValueType} is used.
* {@link Type} is used.
*/
public void setBytes(BytesRef value, ValueType type);
public void setBytes(BytesRef value, DocValues.Type type);
/**
* Sets the given {@link BytesRef} value, the field's {@link ValueType} and the
* Sets the given {@link BytesRef} value, the field's {@link Type} and the
* field's comparator. If the {@link Comparator} is set to <code>null</code>
* the default for the given {@link ValueType} is used instead.
* the default for the given {@link Type} is used instead.
*/
public void setBytes(BytesRef value, ValueType type, Comparator<BytesRef> comp);
public void setBytes(BytesRef value, DocValues.Type type, Comparator<BytesRef> comp);
/**
* Returns the set {@link BytesRef} or <code>null</code> if not set.
@ -82,17 +84,17 @@ public interface PerDocFieldValues {
/**
* Sets the {@link BytesRef} comparator for this field. If the field has a
* numeric {@link ValueType} the comparator will be ignored.
* numeric {@link Type} the comparator will be ignored.
*/
public void setBytesComparator(Comparator<BytesRef> comp);
/**
* Sets the {@link ValueType}
* Sets the {@link Type}
*/
public void setDocValuesType(ValueType type);
public void setDocValuesType(DocValues.Type type);
/**
* Returns the {@link ValueType}
* Returns the {@link Type}
*/
public ValueType docValuesType();
public DocValues.Type docValuesType();
}

View File

@ -1,4 +1,4 @@
package org.apache.lucene.index.codecs;
package org.apache.lucene.index;
/**
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
@ -18,7 +18,8 @@ package org.apache.lucene.index.codecs;
import java.io.Closeable;
import java.io.IOException;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.codecs.PerDocConsumer;
import org.apache.lucene.index.codecs.PostingsFormat;
/**
* Abstract API that provides access to one or more per-document storage
@ -34,15 +35,15 @@ import org.apache.lucene.index.values.IndexDocValues;
*/
public abstract class PerDocValues implements Closeable {
/**
* Returns {@link IndexDocValues} for the current field.
* Returns {@link DocValues} for the current field.
*
* @param field
* the field name
* @return the {@link IndexDocValues} for this field or <code>null</code> if not
* @return the {@link DocValues} for this field or <code>null</code> if not
* applicable.
* @throws IOException
*/
public abstract IndexDocValues docValues(String field) throws IOException;
public abstract DocValues docValues(String field) throws IOException;
public static final PerDocValues[] EMPTY_ARRAY = new PerDocValues[0];
}

View File

@ -25,7 +25,6 @@ import org.apache.lucene.index.codecs.NormsReader;
import org.apache.lucene.index.codecs.PostingsFormat;
import org.apache.lucene.index.codecs.FieldsProducer;
import org.apache.lucene.index.codecs.StoredFieldsReader;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.index.codecs.TermVectorsReader;
import org.apache.lucene.store.CompoundFileDirectory;
import org.apache.lucene.store.Directory;

View File

@ -26,6 +26,7 @@ import java.util.Map;
import org.apache.lucene.index.FieldInfo.IndexOptions;
import org.apache.lucene.index.IndexReader.FieldOption;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.codecs.Codec;
import org.apache.lucene.index.codecs.FieldInfosWriter;
import org.apache.lucene.index.codecs.FieldsConsumer;
@ -33,9 +34,6 @@ import org.apache.lucene.index.codecs.NormsWriter;
import org.apache.lucene.index.codecs.StoredFieldsWriter;
import org.apache.lucene.index.codecs.PerDocConsumer;
import org.apache.lucene.index.codecs.TermVectorsWriter;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.TypePromoter;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.util.Bits;
@ -191,7 +189,7 @@ final class SegmentMerger {
// returns an updated typepromoter (tracking type and size) given a previous one,
// and a newly encountered docvalues
private TypePromoter mergeDocValuesType(TypePromoter previous, IndexDocValues docValues) {
private TypePromoter mergeDocValuesType(TypePromoter previous, DocValues docValues) {
TypePromoter incoming = TypePromoter.create(docValues.type(), docValues.getValueSize());
if (previous == null) {
previous = TypePromoter.getIdentityPromoter();
@ -199,7 +197,7 @@ final class SegmentMerger {
TypePromoter promoted = previous.promote(incoming);
if (promoted == null) {
// type is incompatible: promote to BYTES_VAR_STRAIGHT
return TypePromoter.create(ValueType.BYTES_VAR_STRAIGHT, TypePromoter.VAR_TYPE_VALUE_SIZE);
return TypePromoter.create(DocValues.Type.BYTES_VAR_STRAIGHT, TypePromoter.VAR_TYPE_VALUE_SIZE);
} else {
return promoted;
}
@ -237,7 +235,7 @@ final class SegmentMerger {
mergeState.fieldInfos.addOrUpdate(dvNames, false);
for (String dvName : dvNames) {
FieldInfo merged = mergeState.fieldInfos.fieldInfo(dvName);
IndexDocValues docValues = reader.docValues(dvName);
DocValues docValues = reader.docValues(dvName);
merged.setDocValuesType(docValues.type());
TypePromoter previous = docValuesTypes.get(merged);
docValuesTypes.put(merged, mergeDocValuesType(previous, docValues));

View File

@ -28,7 +28,6 @@ import java.util.concurrent.atomic.AtomicInteger;
import org.apache.lucene.store.Directory;
import org.apache.lucene.index.FieldInfo.IndexOptions;
import org.apache.lucene.index.codecs.StoredFieldsReader;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.index.codecs.TermVectorsReader;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.util.BitVector;

View File

@ -26,7 +26,6 @@ import org.apache.lucene.util.ReaderUtil; // javadoc
import org.apache.lucene.index.DirectoryReader; // javadoc
import org.apache.lucene.index.MultiReader; // javadoc
import org.apache.lucene.index.codecs.PerDocValues;
/**
* This class forces a composite reader (eg a {@link

View File

@ -1,4 +1,4 @@
package org.apache.lucene.index.values;
package org.apache.lucene.index;
/**
* Licensed to the Apache Software Foundation (ASF) under one or more
@ -19,15 +19,17 @@ package org.apache.lucene.index.values;
import java.util.HashMap;
import java.util.Map;
import org.apache.lucene.index.DocValues.Type;
/**
* Type promoter that promotes {@link IndexDocValues} during merge based on
* their {@link ValueType} and {@link #getValueSize()}
* Type promoter that promotes {@link DocValues} during merge based on
* their {@link Type} and {@link #getValueSize()}
*
* @lucene.internal
*/
public class TypePromoter {
class TypePromoter {
private final static Map<Integer, ValueType> FLAGS_MAP = new HashMap<Integer, ValueType>();
private final static Map<Integer, Type> FLAGS_MAP = new HashMap<Integer, Type>();
private static final TypePromoter IDENTITY_PROMOTER = new IdentityTypePromoter();
public static final int VAR_TYPE_VALUE_SIZE = -1;
@ -49,7 +51,7 @@ public class TypePromoter {
private static final int IS_32_BIT = 1 << 10 | 1 << 11;
private static final int IS_64_BIT = 1 << 11;
private final ValueType type;
private final Type type;
private final int flags;
private final int valueSize;
@ -65,7 +67,7 @@ public class TypePromoter {
}
static {
for (ValueType type : ValueType.values()) {
for (Type type : Type.values()) {
TypePromoter create = create(type, VAR_TYPE_VALUE_SIZE);
FLAGS_MAP.put(create.flags, type);
}
@ -75,13 +77,13 @@ public class TypePromoter {
* Creates a new {@link TypePromoter}
*
* @param type
* the {@link ValueType} this promoter represents
* the {@link Type} this promoter represents
* @param flags
* the promoters flags
* @param valueSize
* the value size if {@link #IS_FIXED} or <code>-1</code> otherwise.
*/
protected TypePromoter(ValueType type, int flags, int valueSize) {
protected TypePromoter(Type type, int flags, int valueSize) {
this.type = type;
this.flags = flags;
this.valueSize = valueSize;
@ -117,11 +119,11 @@ public class TypePromoter {
}
/**
* Returns the {@link ValueType} of this {@link TypePromoter}
* Returns the {@link Type} of this {@link TypePromoter}
*
* @return the {@link ValueType} of this {@link TypePromoter}
* @return the {@link Type} of this {@link TypePromoter}
*/
public ValueType type() {
public Type type() {
return type;
}
@ -134,13 +136,13 @@ public class TypePromoter {
* Creates a new {@link TypePromoter} for the given type and size per value.
*
* @param type
* the {@link ValueType} to create the promoter for
* the {@link Type} to create the promoter for
* @param valueSize
* the size per value in bytes or <code>-1</code> iff the types have
* variable length.
* @return a new {@link TypePromoter}
*/
public static TypePromoter create(ValueType type, int valueSize) {
public static TypePromoter create(Type type, int valueSize) {
if (type == null) {
return null;
}

View File

@ -18,10 +18,10 @@ package org.apache.lucene.index.codecs;
*/
import java.io.IOException;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.MergeState;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.PerDocFieldValues;
import org.apache.lucene.index.PerDocFieldValues;
import org.apache.lucene.util.Bits;
/**
@ -70,11 +70,11 @@ public abstract class DocValuesConsumer {
* the state to merge
* @param docValues docValues array containing one instance per reader (
* {@link org.apache.lucene.index.MergeState#readers}) or <code>null</code> if the reader has
* no {@link IndexDocValues} instance.
* no {@link DocValues} instance.
* @throws IOException
* if an {@link IOException} occurs
*/
public void merge(MergeState mergeState, IndexDocValues[] docValues) throws IOException {
public void merge(MergeState mergeState, DocValues[] docValues) throws IOException {
assert mergeState != null;
boolean hasMerged = false;
for(int readerIDX=0;readerIDX<mergeState.readers.size();readerIDX++) {
@ -115,7 +115,7 @@ public abstract class DocValuesConsumer {
* the source reader for this MergeState - merged values should be read from
* this instance
*/
public final IndexDocValues reader;
public final DocValues reader;
/** the absolute docBase for this MergeState within the resulting segment */
public final int docBase;
/** the number of documents in this MergeState */
@ -123,7 +123,7 @@ public abstract class DocValuesConsumer {
/** the not deleted bits for this MergeState */
public final Bits liveDocs;
public SingleSubMergeState(IndexDocValues reader, int docBase, int docCount, Bits liveDocs) {
public SingleSubMergeState(DocValues reader, int docBase, int docCount, Bits liveDocs) {
assert reader != null;
this.reader = reader;
this.docBase = docBase;

View File

@ -20,6 +20,7 @@ package org.apache.lucene.index.codecs;
import java.io.IOException;
import java.util.Set;
import org.apache.lucene.index.PerDocValues;
import org.apache.lucene.index.PerDocWriteState;
import org.apache.lucene.index.SegmentInfo;
import org.apache.lucene.index.SegmentReadState;

View File

@ -26,11 +26,12 @@ import java.util.TreeMap;
import org.apache.lucene.index.FieldInfo;
import org.apache.lucene.index.FieldInfos;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.Type; // javadocs
import org.apache.lucene.index.PerDocValues;
import org.apache.lucene.index.codecs.lucene40.values.Bytes;
import org.apache.lucene.index.codecs.lucene40.values.Floats;
import org.apache.lucene.index.codecs.lucene40.values.Ints;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.util.BytesRef;
@ -39,10 +40,11 @@ import org.apache.lucene.util.BytesRef;
* Abstract base class for PerDocValues implementations
* @lucene.experimental
*/
// TODO: this needs to go under lucene40 codec (its specific to its impl)
public abstract class DocValuesReaderBase extends PerDocValues {
protected abstract void closeInternal(Collection<? extends Closeable> closeables) throws IOException;
protected abstract Map<String, IndexDocValues> docValues();
protected abstract Map<String, DocValues> docValues();
@Override
public void close() throws IOException {
@ -50,7 +52,7 @@ public abstract class DocValuesReaderBase extends PerDocValues {
}
@Override
public IndexDocValues docValues(String field) throws IOException {
public DocValues docValues(String field) throws IOException {
return docValues().get(field);
}
@ -59,10 +61,10 @@ public abstract class DocValuesReaderBase extends PerDocValues {
}
// Only opens files... doesn't actually load any values
protected TreeMap<String, IndexDocValues> load(FieldInfos fieldInfos,
protected TreeMap<String, DocValues> load(FieldInfos fieldInfos,
String segment, int docCount, Directory dir, IOContext context)
throws IOException {
TreeMap<String, IndexDocValues> values = new TreeMap<String, IndexDocValues>();
TreeMap<String, DocValues> values = new TreeMap<String, DocValues>();
boolean success = false;
try {
@ -88,26 +90,26 @@ public abstract class DocValuesReaderBase extends PerDocValues {
}
/**
* Loads a {@link IndexDocValues} instance depending on the given {@link ValueType}.
* Codecs that use different implementations for a certain {@link ValueType} can
* Loads a {@link DocValues} instance depending on the given {@link Type}.
* Codecs that use different implementations for a certain {@link Type} can
* simply override this method and return their custom implementations.
*
* @param docCount
* number of documents in the segment
* @param dir
* the {@link Directory} to load the {@link IndexDocValues} from
* the {@link Directory} to load the {@link DocValues} from
* @param id
* the unique file ID within the segment
* @param type
* the type to load
* @return a {@link IndexDocValues} instance for the given type
* @return a {@link DocValues} instance for the given type
* @throws IOException
* if an {@link IOException} occurs
* @throws IllegalArgumentException
* if the given {@link ValueType} is not supported
* if the given {@link Type} is not supported
*/
protected IndexDocValues loadDocValues(int docCount, Directory dir, String id,
ValueType type, IOContext context) throws IOException {
protected DocValues loadDocValues(int docCount, Directory dir, String id,
DocValues.Type type, IOContext context) throws IOException {
switch (type) {
case FIXED_INTS_16:
case FIXED_INTS_32:

View File

@ -23,7 +23,7 @@ import java.util.Comparator;
import org.apache.lucene.index.FieldInfo;
import org.apache.lucene.index.PerDocWriteState;
import org.apache.lucene.index.codecs.lucene40.values.Writer;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.util.BytesRef;
@ -33,6 +33,7 @@ import org.apache.lucene.util.Counter;
* Abstract base class for PerDocConsumer implementations
* @lucene.experimental
*/
//TODO: this needs to go under lucene40 codec (its specific to its impl)
public abstract class DocValuesWriterBase extends PerDocConsumer {
protected final String segmentName;
protected final String segmentSuffix;
@ -53,7 +54,7 @@ public abstract class DocValuesWriterBase extends PerDocConsumer {
}
@Override
public DocValuesConsumer addValuesField(ValueType valueType, FieldInfo field) throws IOException {
public DocValuesConsumer addValuesField(DocValues.Type valueType, FieldInfo field) throws IOException {
return Writer.create(valueType,
docValuesId(segmentName, field.number),
getDirectory(), getComparator(), bytesUsed, context);

View File

@ -19,9 +19,9 @@ import java.io.Closeable;
import java.io.IOException;
import org.apache.lucene.index.FieldInfo;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.MergeState;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.PerDocValues;
/**
* Abstract API that consumes per document values. Concrete implementations of
@ -35,7 +35,7 @@ import org.apache.lucene.index.values.ValueType;
*/
public abstract class PerDocConsumer implements Closeable{
/** Adds a new DocValuesField */
public abstract DocValuesConsumer addValuesField(ValueType type, FieldInfo field)
public abstract DocValuesConsumer addValuesField(DocValues.Type type, FieldInfo field)
throws IOException;
/**
@ -43,7 +43,7 @@ public abstract class PerDocConsumer implements Closeable{
* into this consumers format.
*/
public void merge(MergeState mergeState) throws IOException {
final IndexDocValues[] docValues = new IndexDocValues[mergeState.readers.size()];
final DocValues[] docValues = new DocValues[mergeState.readers.size()];
final PerDocValues[] perDocValues = new PerDocValues[mergeState.readers.size()];
// pull all PerDocValues
for (int i = 0; i < perDocValues.length; i++) {

View File

@ -20,6 +20,7 @@ package org.apache.lucene.index.codecs.lucene3x;
import java.io.IOException;
import java.util.Set;
import org.apache.lucene.index.PerDocValues;
import org.apache.lucene.index.PerDocWriteState;
import org.apache.lucene.index.SegmentInfo;
import org.apache.lucene.index.SegmentReadState;
@ -29,7 +30,6 @@ import org.apache.lucene.index.codecs.FieldInfosFormat;
import org.apache.lucene.index.codecs.NormsFormat;
import org.apache.lucene.index.codecs.StoredFieldsFormat;
import org.apache.lucene.index.codecs.PerDocConsumer;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.index.codecs.PostingsFormat;
import org.apache.lucene.index.codecs.SegmentInfosFormat;
import org.apache.lucene.index.codecs.TermVectorsFormat;

View File

@ -20,12 +20,12 @@ package org.apache.lucene.index.codecs.lucene40;
import java.io.IOException;
import java.util.Set;
import org.apache.lucene.index.PerDocValues;
import org.apache.lucene.index.PerDocWriteState;
import org.apache.lucene.index.SegmentInfo;
import org.apache.lucene.index.SegmentReadState;
import org.apache.lucene.index.codecs.DocValuesFormat;
import org.apache.lucene.index.codecs.PerDocConsumer;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.store.Directory;
public class Lucene40DocValuesFormat extends DocValuesFormat {

View File

@ -24,10 +24,10 @@ import java.util.Collection;
import java.util.Map;
import java.util.TreeMap;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.IndexFileNames;
import org.apache.lucene.index.SegmentReadState;
import org.apache.lucene.index.codecs.DocValuesReaderBase;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.store.CompoundFileDirectory;
import org.apache.lucene.store.Directory;
import org.apache.lucene.util.IOUtils;
@ -37,12 +37,12 @@ import org.apache.lucene.util.IOUtils;
* @lucene.experimental
*/
public class Lucene40DocValuesProducer extends DocValuesReaderBase {
protected final TreeMap<String,IndexDocValues> docValues;
protected final TreeMap<String,DocValues> docValues;
private final Directory cfs;
/**
* Creates a new {@link Lucene40DocValuesProducer} instance and loads all
* {@link IndexDocValues} instances for this segment and codec.
* {@link DocValues} instances for this segment and codec.
*/
public Lucene40DocValuesProducer(SegmentReadState state) throws IOException {
if (state.fieldInfos.anyDocValuesFields()) {
@ -53,12 +53,12 @@ public class Lucene40DocValuesProducer extends DocValuesReaderBase {
docValues = load(state.fieldInfos, state.segmentInfo.name, state.segmentInfo.docCount, cfs, state.context);
} else {
cfs = null;
docValues = new TreeMap<String,IndexDocValues>();
docValues = new TreeMap<String,DocValues>();
}
}
@Override
protected Map<String,IndexDocValues> docValues() {
protected Map<String,DocValues> docValues() {
return docValues;
}

View File

@ -12,7 +12,7 @@ import org.apache.lucene.index.IndexFormatTooOldException;
import org.apache.lucene.index.SegmentInfo;
import org.apache.lucene.index.FieldInfo.IndexOptions;
import org.apache.lucene.index.codecs.FieldInfosReader;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
@ -95,7 +95,7 @@ public class Lucene40FieldInfosReader extends FieldInfosReader {
hasVectors |= storeTermVector;
hasProx |= isIndexed && indexOptions == IndexOptions.DOCS_AND_FREQS_AND_POSITIONS;
hasFreq |= isIndexed && indexOptions != IndexOptions.DOCS_ONLY;
ValueType docValuesType = null;
DocValues.Type docValuesType = null;
if (format <= Lucene40FieldInfosWriter.FORMAT_FLEX) {
final byte b = input.readByte();
switch(b) {
@ -103,43 +103,43 @@ public class Lucene40FieldInfosReader extends FieldInfosReader {
docValuesType = null;
break;
case 1:
docValuesType = ValueType.VAR_INTS;
docValuesType = DocValues.Type.VAR_INTS;
break;
case 2:
docValuesType = ValueType.FLOAT_32;
docValuesType = DocValues.Type.FLOAT_32;
break;
case 3:
docValuesType = ValueType.FLOAT_64;
docValuesType = DocValues.Type.FLOAT_64;
break;
case 4:
docValuesType = ValueType.BYTES_FIXED_STRAIGHT;
docValuesType = DocValues.Type.BYTES_FIXED_STRAIGHT;
break;
case 5:
docValuesType = ValueType.BYTES_FIXED_DEREF;
docValuesType = DocValues.Type.BYTES_FIXED_DEREF;
break;
case 6:
docValuesType = ValueType.BYTES_VAR_STRAIGHT;
docValuesType = DocValues.Type.BYTES_VAR_STRAIGHT;
break;
case 7:
docValuesType = ValueType.BYTES_VAR_DEREF;
docValuesType = DocValues.Type.BYTES_VAR_DEREF;
break;
case 8:
docValuesType = ValueType.FIXED_INTS_16;
docValuesType = DocValues.Type.FIXED_INTS_16;
break;
case 9:
docValuesType = ValueType.FIXED_INTS_32;
docValuesType = DocValues.Type.FIXED_INTS_32;
break;
case 10:
docValuesType = ValueType.FIXED_INTS_64;
docValuesType = DocValues.Type.FIXED_INTS_64;
break;
case 11:
docValuesType = ValueType.FIXED_INTS_8;
docValuesType = DocValues.Type.FIXED_INTS_8;
break;
case 12:
docValuesType = ValueType.BYTES_FIXED_SORTED;
docValuesType = DocValues.Type.BYTES_FIXED_SORTED;
break;
case 13:
docValuesType = ValueType.BYTES_VAR_SORTED;
docValuesType = DocValues.Type.BYTES_VAR_SORTED;
break;
default:

View File

@ -22,12 +22,12 @@ import java.io.IOException;
import java.util.Comparator;
import java.util.concurrent.atomic.AtomicLong;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.IndexFileNames;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.PerDocFieldValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.values.IndexDocValues.SortedSource;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.index.PerDocFieldValues;
import org.apache.lucene.index.DocValues.SortedSource;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.store.DataOutput;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
@ -150,13 +150,13 @@ public final class Bytes {
}
/**
* Creates a new {@link IndexDocValues} instance that provides either memory
* Creates a new {@link DocValues} instance that provides either memory
* resident or iterative access to a per-document stored <tt>byte[]</tt>
* value. The returned {@link IndexDocValues} instance will be initialized without
* value. The returned {@link DocValues} instance will be initialized without
* consuming a significant amount of memory.
*
* @param dir
* the directory to load the {@link IndexDocValues} from.
* the directory to load the {@link DocValues} from.
* @param id
* the file ID in the {@link Directory} to load the values from.
* @param mode
@ -169,11 +169,11 @@ public final class Bytes {
* @param sortComparator {@link BytesRef} comparator used by sorted variants.
* If <code>null</code> {@link BytesRef#getUTF8SortedAsUnicodeComparator()}
* is used instead
* @return an initialized {@link IndexDocValues} instance.
* @return an initialized {@link DocValues} instance.
* @throws IOException
* if an {@link IOException} occurs
*/
public static IndexDocValues getValues(Directory dir, String id, Mode mode,
public static DocValues getValues(Directory dir, String id, Mode mode,
boolean fixedSize, int maxDoc, Comparator<BytesRef> sortComparator, IOContext context) throws IOException {
if (sortComparator == null) {
sortComparator = BytesRef.getUTF8SortedAsUnicodeComparator();
@ -185,7 +185,7 @@ public final class Bytes {
} else if (mode == Mode.DEREF) {
return new FixedDerefBytesImpl.FixedDerefReader(dir, id, maxDoc, context);
} else if (mode == Mode.SORTED) {
return new FixedSortedBytesImpl.Reader(dir, id, maxDoc, context, ValueType.BYTES_FIXED_SORTED, sortComparator);
return new FixedSortedBytesImpl.Reader(dir, id, maxDoc, context, Type.BYTES_FIXED_SORTED, sortComparator);
}
} else {
if (mode == Mode.STRAIGHT) {
@ -193,7 +193,7 @@ public final class Bytes {
} else if (mode == Mode.DEREF) {
return new VarDerefBytesImpl.VarDerefReader(dir, id, maxDoc, context);
} else if (mode == Mode.SORTED) {
return new VarSortedBytesImpl.Reader(dir, id, maxDoc,context, ValueType.BYTES_VAR_SORTED, sortComparator);
return new VarSortedBytesImpl.Reader(dir, id, maxDoc,context, Type.BYTES_VAR_SORTED, sortComparator);
}
}
@ -211,7 +211,7 @@ public final class Bytes {
protected BytesSourceBase(IndexInput datIn, IndexInput idxIn,
PagedBytes pagedBytes, long bytesToRead, ValueType type) throws IOException {
PagedBytes pagedBytes, long bytesToRead, Type type) throws IOException {
super(type);
assert bytesToRead <= datIn.length() : " file size is less than the expected size diff: "
+ (bytesToRead - datIn.length()) + " pos: " + datIn.getFilePointer();
@ -314,15 +314,15 @@ public final class Bytes {
* Opens all necessary files, but does not read any data in until you call
* {@link #load}.
*/
static abstract class BytesReaderBase extends IndexDocValues {
static abstract class BytesReaderBase extends DocValues {
protected final IndexInput idxIn;
protected final IndexInput datIn;
protected final int version;
protected final String id;
protected final ValueType type;
protected final Type type;
protected BytesReaderBase(Directory dir, String id, String codecName,
int maxVersion, boolean doIndex, IOContext context, ValueType type) throws IOException {
int maxVersion, boolean doIndex, IOContext context, Type type) throws IOException {
IndexInput dataIn = null;
IndexInput indexIn = null;
boolean success = false;
@ -375,7 +375,7 @@ public final class Bytes {
}
@Override
public ValueType type() {
public Type type() {
return type;
}
@ -559,12 +559,12 @@ public final class Bytes {
protected final PagedBytes.Reader data;
protected BytesSortedSourceBase(IndexInput datIn, IndexInput idxIn,
Comparator<BytesRef> comp, long bytesToRead, ValueType type, boolean hasOffsets) throws IOException {
Comparator<BytesRef> comp, long bytesToRead, Type type, boolean hasOffsets) throws IOException {
this(datIn, idxIn, comp, new PagedBytes(PAGED_BYTES_BITS), bytesToRead, type, hasOffsets);
}
protected BytesSortedSourceBase(IndexInput datIn, IndexInput idxIn,
Comparator<BytesRef> comp, PagedBytes pagedBytes, long bytesToRead, ValueType type, boolean hasOffsets)
Comparator<BytesRef> comp, PagedBytes pagedBytes, long bytesToRead, Type type, boolean hasOffsets)
throws IOException {
super(type, comp);
assert bytesToRead <= datIn.length() : " file size is less than the expected size diff: "

View File

@ -19,9 +19,9 @@ package org.apache.lucene.index.codecs.lucene40.values;
import java.io.IOException;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.store.IndexInput;
import org.apache.lucene.util.BytesRef;
@ -35,7 +35,7 @@ public abstract class DirectSource extends Source {
private final ToNumeric toNumeric;
protected final long baseOffset;
public DirectSource(IndexInput input, ValueType type) {
public DirectSource(IndexInput input, Type type) {
super(type);
this.data = input;
baseOffset = input.getFilePointer();

View File

@ -5,8 +5,8 @@ import java.util.Collections;
import java.util.EnumMap;
import java.util.Map;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.store.IndexInput;
import org.apache.lucene.util.BytesRef;
import org.apache.lucene.util.RamUsageEstimator;
@ -31,30 +31,30 @@ import org.apache.lucene.util.RamUsageEstimator;
/**
* @lucene.experimental
*/
abstract class IndexDocValuesArray extends Source {
abstract class DocValuesArray extends Source {
static final Map<ValueType, IndexDocValuesArray> TEMPLATES;
static final Map<Type, DocValuesArray> TEMPLATES;
static {
EnumMap<ValueType, IndexDocValuesArray> templates = new EnumMap<ValueType, IndexDocValuesArray>(
ValueType.class);
templates.put(ValueType.FIXED_INTS_16, new ShortValues());
templates.put(ValueType.FIXED_INTS_32, new IntValues());
templates.put(ValueType.FIXED_INTS_64, new LongValues());
templates.put(ValueType.FIXED_INTS_8, new ByteValues());
templates.put(ValueType.FLOAT_32, new FloatValues());
templates.put(ValueType.FLOAT_64, new DoubleValues());
EnumMap<Type, DocValuesArray> templates = new EnumMap<Type, DocValuesArray>(
Type.class);
templates.put(Type.FIXED_INTS_16, new ShortValues());
templates.put(Type.FIXED_INTS_32, new IntValues());
templates.put(Type.FIXED_INTS_64, new LongValues());
templates.put(Type.FIXED_INTS_8, new ByteValues());
templates.put(Type.FLOAT_32, new FloatValues());
templates.put(Type.FLOAT_64, new DoubleValues());
TEMPLATES = Collections.unmodifiableMap(templates);
}
protected final int bytesPerValue;
IndexDocValuesArray(int bytesPerValue, ValueType type) {
DocValuesArray(int bytesPerValue, Type type) {
super(type);
this.bytesPerValue = bytesPerValue;
}
public abstract IndexDocValuesArray newFromInput(IndexInput input, int numDocs)
public abstract DocValuesArray newFromInput(IndexInput input, int numDocs)
throws IOException;
@Override
@ -70,16 +70,16 @@ abstract class IndexDocValuesArray extends Source {
BytesRefUtils.copyLong(bytesRef, Double.doubleToRawLongBits(value));
}
final static class ByteValues extends IndexDocValuesArray {
final static class ByteValues extends DocValuesArray {
private final byte[] values;
ByteValues() {
super(1, ValueType.FIXED_INTS_8);
super(1, Type.FIXED_INTS_8);
values = new byte[0];
}
private ByteValues(IndexInput input, int numDocs) throws IOException {
super(1, ValueType.FIXED_INTS_8);
super(1, Type.FIXED_INTS_8);
values = new byte[numDocs];
input.readBytes(values, 0, values.length, false);
}
@ -96,7 +96,7 @@ abstract class IndexDocValuesArray extends Source {
}
@Override
public IndexDocValuesArray newFromInput(IndexInput input, int numDocs)
public DocValuesArray newFromInput(IndexInput input, int numDocs)
throws IOException {
return new ByteValues(input, numDocs);
}
@ -107,16 +107,16 @@ abstract class IndexDocValuesArray extends Source {
};
final static class ShortValues extends IndexDocValuesArray {
final static class ShortValues extends DocValuesArray {
private final short[] values;
ShortValues() {
super(RamUsageEstimator.NUM_BYTES_SHORT, ValueType.FIXED_INTS_16);
super(RamUsageEstimator.NUM_BYTES_SHORT, Type.FIXED_INTS_16);
values = new short[0];
}
private ShortValues(IndexInput input, int numDocs) throws IOException {
super(RamUsageEstimator.NUM_BYTES_SHORT, ValueType.FIXED_INTS_16);
super(RamUsageEstimator.NUM_BYTES_SHORT, Type.FIXED_INTS_16);
values = new short[numDocs];
for (int i = 0; i < values.length; i++) {
values[i] = input.readShort();
@ -135,7 +135,7 @@ abstract class IndexDocValuesArray extends Source {
}
@Override
public IndexDocValuesArray newFromInput(IndexInput input, int numDocs)
public DocValuesArray newFromInput(IndexInput input, int numDocs)
throws IOException {
return new ShortValues(input, numDocs);
}
@ -146,16 +146,16 @@ abstract class IndexDocValuesArray extends Source {
};
final static class IntValues extends IndexDocValuesArray {
final static class IntValues extends DocValuesArray {
private final int[] values;
IntValues() {
super(RamUsageEstimator.NUM_BYTES_INT, ValueType.FIXED_INTS_32);
super(RamUsageEstimator.NUM_BYTES_INT, Type.FIXED_INTS_32);
values = new int[0];
}
private IntValues(IndexInput input, int numDocs) throws IOException {
super(RamUsageEstimator.NUM_BYTES_INT, ValueType.FIXED_INTS_32);
super(RamUsageEstimator.NUM_BYTES_INT, Type.FIXED_INTS_32);
values = new int[numDocs];
for (int i = 0; i < values.length; i++) {
values[i] = input.readInt();
@ -174,7 +174,7 @@ abstract class IndexDocValuesArray extends Source {
}
@Override
public IndexDocValuesArray newFromInput(IndexInput input, int numDocs)
public DocValuesArray newFromInput(IndexInput input, int numDocs)
throws IOException {
return new IntValues(input, numDocs);
}
@ -185,16 +185,16 @@ abstract class IndexDocValuesArray extends Source {
};
final static class LongValues extends IndexDocValuesArray {
final static class LongValues extends DocValuesArray {
private final long[] values;
LongValues() {
super(RamUsageEstimator.NUM_BYTES_LONG, ValueType.FIXED_INTS_64);
super(RamUsageEstimator.NUM_BYTES_LONG, Type.FIXED_INTS_64);
values = new long[0];
}
private LongValues(IndexInput input, int numDocs) throws IOException {
super(RamUsageEstimator.NUM_BYTES_LONG, ValueType.FIXED_INTS_64);
super(RamUsageEstimator.NUM_BYTES_LONG, Type.FIXED_INTS_64);
values = new long[numDocs];
for (int i = 0; i < values.length; i++) {
values[i] = input.readLong();
@ -213,23 +213,23 @@ abstract class IndexDocValuesArray extends Source {
}
@Override
public IndexDocValuesArray newFromInput(IndexInput input, int numDocs)
public DocValuesArray newFromInput(IndexInput input, int numDocs)
throws IOException {
return new LongValues(input, numDocs);
}
};
final static class FloatValues extends IndexDocValuesArray {
final static class FloatValues extends DocValuesArray {
private final float[] values;
FloatValues() {
super(RamUsageEstimator.NUM_BYTES_FLOAT, ValueType.FLOAT_32);
super(RamUsageEstimator.NUM_BYTES_FLOAT, Type.FLOAT_32);
values = new float[0];
}
private FloatValues(IndexInput input, int numDocs) throws IOException {
super(RamUsageEstimator.NUM_BYTES_FLOAT, ValueType.FLOAT_32);
super(RamUsageEstimator.NUM_BYTES_FLOAT, Type.FLOAT_32);
values = new float[numDocs];
/*
* we always read BIG_ENDIAN here since the writer serialized plain bytes
@ -258,22 +258,22 @@ abstract class IndexDocValuesArray extends Source {
}
@Override
public IndexDocValuesArray newFromInput(IndexInput input, int numDocs)
public DocValuesArray newFromInput(IndexInput input, int numDocs)
throws IOException {
return new FloatValues(input, numDocs);
}
};
final static class DoubleValues extends IndexDocValuesArray {
final static class DoubleValues extends DocValuesArray {
private final double[] values;
DoubleValues() {
super(RamUsageEstimator.NUM_BYTES_DOUBLE, ValueType.FLOAT_64);
super(RamUsageEstimator.NUM_BYTES_DOUBLE, Type.FLOAT_64);
values = new double[0];
}
private DoubleValues(IndexInput input, int numDocs) throws IOException {
super(RamUsageEstimator.NUM_BYTES_DOUBLE, ValueType.FLOAT_64);
super(RamUsageEstimator.NUM_BYTES_DOUBLE, Type.FLOAT_64);
values = new double[numDocs];
/*
* we always read BIG_ENDIAN here since the writer serialized plain bytes
@ -296,7 +296,7 @@ abstract class IndexDocValuesArray extends Source {
}
@Override
public IndexDocValuesArray newFromInput(IndexInput input, int numDocs)
public DocValuesArray newFromInput(IndexInput input, int numDocs)
throws IOException {
return new DoubleValues(input, numDocs);
}

View File

@ -22,7 +22,7 @@ import java.io.IOException;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesReaderBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesSourceBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.DerefBytesWriterBase;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
@ -71,7 +71,7 @@ class FixedDerefBytesImpl {
private final int size;
private final int numValuesStored;
FixedDerefReader(Directory dir, String id, int maxDoc, IOContext context) throws IOException {
super(dir, id, CODEC_NAME, VERSION_START, true, context, ValueType.BYTES_FIXED_DEREF);
super(dir, id, CODEC_NAME, VERSION_START, true, context, Type.BYTES_FIXED_DEREF);
size = datIn.readInt();
numValuesStored = idxIn.readInt();
}
@ -100,7 +100,7 @@ class FixedDerefBytesImpl {
protected FixedDerefSource(IndexInput datIn, IndexInput idxIn, int size, long numValues) throws IOException {
super(datIn, idxIn, new PagedBytes(PAGED_BYTES_BITS), size * numValues,
ValueType.BYTES_FIXED_DEREF);
Type.BYTES_FIXED_DEREF);
this.size = size;
addresses = PackedInts.getReader(idxIn);
}
@ -117,7 +117,7 @@ class FixedDerefBytesImpl {
private final PackedInts.Reader index;
private final int size;
DirectFixedDerefSource(IndexInput data, IndexInput index, int size, ValueType type)
DirectFixedDerefSource(IndexInput data, IndexInput index, int size, Type type)
throws IOException {
super(data, type);
this.size = size;

View File

@ -21,15 +21,15 @@ import java.io.IOException;
import java.util.Comparator;
import java.util.List;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.SortedSource;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.index.MergeState;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesReaderBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesSortedSourceBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.DerefBytesWriterBase;
import org.apache.lucene.index.codecs.lucene40.values.SortedBytesMergeUtils.MergeContext;
import org.apache.lucene.index.codecs.lucene40.values.SortedBytesMergeUtils.SortedSourceSlice;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.values.IndexDocValues.SortedSource;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
@ -61,11 +61,11 @@ class FixedSortedBytesImpl {
}
@Override
public void merge(MergeState mergeState, IndexDocValues[] docValues)
public void merge(MergeState mergeState, DocValues[] docValues)
throws IOException {
boolean success = false;
try {
final MergeContext ctx = SortedBytesMergeUtils.init(ValueType.BYTES_FIXED_SORTED, docValues, comp, mergeState);
final MergeContext ctx = SortedBytesMergeUtils.init(Type.BYTES_FIXED_SORTED, docValues, comp, mergeState);
List<SortedSourceSlice> slices = SortedBytesMergeUtils.buildSlices(mergeState, docValues, ctx);
final IndexOutput datOut = getOrCreateDataOut();
datOut.writeInt(ctx.sizePerValues);
@ -124,7 +124,7 @@ class FixedSortedBytesImpl {
private final Comparator<BytesRef> comparator;
public Reader(Directory dir, String id, int maxDoc, IOContext context,
ValueType type, Comparator<BytesRef> comparator) throws IOException {
Type type, Comparator<BytesRef> comparator) throws IOException {
super(dir, id, CODEC_NAME, VERSION_START, true, context, type);
size = datIn.readInt();
valueCount = idxIn.readInt();
@ -155,7 +155,7 @@ class FixedSortedBytesImpl {
FixedSortedSource(IndexInput datIn, IndexInput idxIn, int size,
int numValues, Comparator<BytesRef> comp) throws IOException {
super(datIn, idxIn, comp, size * numValues, ValueType.BYTES_FIXED_SORTED,
super(datIn, idxIn, comp, size * numValues, Type.BYTES_FIXED_SORTED,
false);
this.size = size;
this.valueCount = numValues;
@ -181,7 +181,7 @@ class FixedSortedBytesImpl {
private final int valueCount;
DirectFixedSortedSource(IndexInput datIn, IndexInput idxIn, int size,
int valueCount, Comparator<BytesRef> comp, ValueType type)
int valueCount, Comparator<BytesRef> comp, Type type)
throws IOException {
super(type, comp);
docToOrdIndex = PackedInts.getDirectReader(idxIn);

View File

@ -21,12 +21,12 @@ import static org.apache.lucene.util.ByteBlockPool.BYTE_BLOCK_SIZE;
import java.io.IOException;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesReaderBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesSourceBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesWriterBase;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
@ -180,7 +180,7 @@ class FixedStraightBytesImpl {
}
}
protected boolean tryBulkMerge(IndexDocValues docValues) {
protected boolean tryBulkMerge(DocValues docValues) {
return docValues instanceof FixedStraightReader;
}
@ -256,10 +256,10 @@ class FixedStraightBytesImpl {
protected final int maxDoc;
FixedStraightReader(Directory dir, String id, int maxDoc, IOContext context) throws IOException {
this(dir, id, CODEC_NAME, VERSION_CURRENT, maxDoc, context, ValueType.BYTES_FIXED_STRAIGHT);
this(dir, id, CODEC_NAME, VERSION_CURRENT, maxDoc, context, Type.BYTES_FIXED_STRAIGHT);
}
protected FixedStraightReader(Directory dir, String id, String codec, int version, int maxDoc, IOContext context, ValueType type) throws IOException {
protected FixedStraightReader(Directory dir, String id, String codec, int version, int maxDoc, IOContext context, Type type) throws IOException {
super(dir, id, codec, version, false, context, type);
size = datIn.readInt();
this.maxDoc = maxDoc;
@ -292,7 +292,7 @@ class FixedStraightBytesImpl {
private final byte[] data;
public SingleByteSource(IndexInput datIn, int maxDoc) throws IOException {
super(ValueType.BYTES_FIXED_STRAIGHT);
super(Type.BYTES_FIXED_STRAIGHT);
try {
data = new byte[maxDoc];
datIn.readBytes(data, 0, data.length, false);
@ -324,7 +324,7 @@ class FixedStraightBytesImpl {
private final static class FixedStraightSource extends BytesSourceBase {
private final int size;
public FixedStraightSource(IndexInput datIn, int size, int maxDoc, ValueType type)
public FixedStraightSource(IndexInput datIn, int size, int maxDoc, Type type)
throws IOException {
super(datIn, null, new PagedBytes(PAGED_BYTES_BITS), size * maxDoc,
type);
@ -340,7 +340,7 @@ class FixedStraightBytesImpl {
public final static class DirectFixedStraightSource extends DirectSource {
private final int size;
DirectFixedStraightSource(IndexInput input, int size, ValueType type) {
DirectFixedStraightSource(IndexInput input, int size, Type type) {
super(input, type);
this.size = size;
}

View File

@ -18,10 +18,10 @@ package org.apache.lucene.index.codecs.lucene40.values;
*/
import java.io.IOException;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.PerDocFieldValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.PerDocFieldValues;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
@ -45,16 +45,16 @@ public class Floats {
protected static final int VERSION_CURRENT = VERSION_START;
public static Writer getWriter(Directory dir, String id, Counter bytesUsed,
IOContext context, ValueType type) throws IOException {
IOContext context, Type type) throws IOException {
return new FloatsWriter(dir, id, bytesUsed, context, type);
}
public static IndexDocValues getValues(Directory dir, String id, int maxDoc, IOContext context, ValueType type)
public static DocValues getValues(Directory dir, String id, int maxDoc, IOContext context, Type type)
throws IOException {
return new FloatsReader(dir, id, maxDoc, context, type);
}
private static int typeToSize(ValueType type) {
private static int typeToSize(Type type) {
switch (type) {
case FLOAT_32:
return 4;
@ -68,14 +68,14 @@ public class Floats {
final static class FloatsWriter extends FixedStraightBytesImpl.Writer {
private final int size;
private final IndexDocValuesArray template;
private final DocValuesArray template;
public FloatsWriter(Directory dir, String id, Counter bytesUsed,
IOContext context, ValueType type) throws IOException {
IOContext context, Type type) throws IOException {
super(dir, id, CODEC_NAME, VERSION_CURRENT, bytesUsed, context);
size = typeToSize(type);
this.bytesRef = new BytesRef(size);
bytesRef.length = size;
template = IndexDocValuesArray.TEMPLATES.get(type);
template = DocValuesArray.TEMPLATES.get(type);
assert template != null;
}
@ -90,7 +90,7 @@ public class Floats {
}
@Override
protected boolean tryBulkMerge(IndexDocValues docValues) {
protected boolean tryBulkMerge(DocValues docValues) {
// only bulk merge if value type is the same otherwise size differs
return super.tryBulkMerge(docValues) && docValues.type() == template.type();
}
@ -103,11 +103,11 @@ public class Floats {
}
final static class FloatsReader extends FixedStraightBytesImpl.FixedStraightReader {
final IndexDocValuesArray arrayTemplate;
FloatsReader(Directory dir, String id, int maxDoc, IOContext context, ValueType type)
final DocValuesArray arrayTemplate;
FloatsReader(Directory dir, String id, int maxDoc, IOContext context, Type type)
throws IOException {
super(dir, id, CODEC_NAME, VERSION_CURRENT, maxDoc, context, type);
arrayTemplate = IndexDocValuesArray.TEMPLATES.get(type);
arrayTemplate = DocValuesArray.TEMPLATES.get(type);
assert size == 4 || size == 8;
}

View File

@ -19,9 +19,9 @@ package org.apache.lucene.index.codecs.lucene40.values;
import java.io.IOException;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.PerDocFieldValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.index.PerDocFieldValues;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
@ -43,33 +43,33 @@ public final class Ints {
}
public static Writer getWriter(Directory dir, String id, Counter bytesUsed,
ValueType type, IOContext context) throws IOException {
return type == ValueType.VAR_INTS ? new PackedIntValues.PackedIntsWriter(dir, id,
Type type, IOContext context) throws IOException {
return type == Type.VAR_INTS ? new PackedIntValues.PackedIntsWriter(dir, id,
bytesUsed, context) : new IntsWriter(dir, id, bytesUsed, context, type);
}
public static IndexDocValues getValues(Directory dir, String id, int numDocs,
ValueType type, IOContext context) throws IOException {
return type == ValueType.VAR_INTS ? new PackedIntValues.PackedIntsReader(dir, id,
public static DocValues getValues(Directory dir, String id, int numDocs,
Type type, IOContext context) throws IOException {
return type == Type.VAR_INTS ? new PackedIntValues.PackedIntsReader(dir, id,
numDocs, context) : new IntsReader(dir, id, numDocs, context, type);
}
private static ValueType sizeToType(int size) {
private static Type sizeToType(int size) {
switch (size) {
case 1:
return ValueType.FIXED_INTS_8;
return Type.FIXED_INTS_8;
case 2:
return ValueType.FIXED_INTS_16;
return Type.FIXED_INTS_16;
case 4:
return ValueType.FIXED_INTS_32;
return Type.FIXED_INTS_32;
case 8:
return ValueType.FIXED_INTS_64;
return Type.FIXED_INTS_64;
default:
throw new IllegalStateException("illegal size " + size);
}
}
private static int typeToSize(ValueType type) {
private static int typeToSize(Type type) {
switch (type) {
case FIXED_INTS_16:
return 2;
@ -86,20 +86,20 @@ public final class Ints {
static class IntsWriter extends FixedStraightBytesImpl.Writer {
private final IndexDocValuesArray template;
private final DocValuesArray template;
public IntsWriter(Directory dir, String id, Counter bytesUsed,
IOContext context, ValueType valueType) throws IOException {
IOContext context, Type valueType) throws IOException {
this(dir, id, CODEC_NAME, VERSION_CURRENT, bytesUsed, context, valueType);
}
protected IntsWriter(Directory dir, String id, String codecName,
int version, Counter bytesUsed, IOContext context, ValueType valueType) throws IOException {
int version, Counter bytesUsed, IOContext context, Type valueType) throws IOException {
super(dir, id, codecName, version, bytesUsed, context);
size = typeToSize(valueType);
this.bytesRef = new BytesRef(size);
bytesRef.length = size;
template = IndexDocValuesArray.TEMPLATES.get(valueType);
template = DocValuesArray.TEMPLATES.get(valueType);
}
@Override
@ -120,20 +120,20 @@ public final class Ints {
}
@Override
protected boolean tryBulkMerge(IndexDocValues docValues) {
protected boolean tryBulkMerge(DocValues docValues) {
// only bulk merge if value type is the same otherwise size differs
return super.tryBulkMerge(docValues) && docValues.type() == template.type();
}
}
final static class IntsReader extends FixedStraightBytesImpl.FixedStraightReader {
private final IndexDocValuesArray arrayTemplate;
private final DocValuesArray arrayTemplate;
IntsReader(Directory dir, String id, int maxDoc, IOContext context, ValueType type)
IntsReader(Directory dir, String id, int maxDoc, IOContext context, Type type)
throws IOException {
super(dir, id, CODEC_NAME, VERSION_CURRENT, maxDoc,
context, type);
arrayTemplate = IndexDocValuesArray.TEMPLATES.get(type);
arrayTemplate = DocValuesArray.TEMPLATES.get(type);
assert arrayTemplate != null;
assert type == sizeToType(size);
}

View File

@ -18,13 +18,13 @@ package org.apache.lucene.index.codecs.lucene40.values;
*/
import java.io.IOException;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.IndexFileNames;
import org.apache.lucene.index.PerDocFieldValues;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.index.codecs.lucene40.values.FixedStraightBytesImpl.FixedBytesWriterBase;
import org.apache.lucene.index.codecs.lucene40.values.IndexDocValuesArray.LongValues;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.PerDocFieldValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.index.codecs.lucene40.values.DocValuesArray.LongValues;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
@ -160,7 +160,7 @@ class PackedIntValues {
* Opens all necessary files, but does not read any data in until you call
* {@link #load}.
*/
static class PackedIntsReader extends IndexDocValues {
static class PackedIntsReader extends DocValues {
private final IndexInput datIn;
private final byte type;
private final int numDocs;
@ -220,14 +220,14 @@ class PackedIntValues {
@Override
public ValueType type() {
return ValueType.VAR_INTS;
public Type type() {
return Type.VAR_INTS;
}
@Override
public Source getDirectSource() throws IOException {
return values != null ? new FixedStraightBytesImpl.DirectFixedStraightSource((IndexInput) datIn.clone(), 8, ValueType.FIXED_INTS_64) : new PackedIntsSource((IndexInput) datIn.clone(), true);
return values != null ? new FixedStraightBytesImpl.DirectFixedStraightSource((IndexInput) datIn.clone(), 8, Type.FIXED_INTS_64) : new PackedIntsSource((IndexInput) datIn.clone(), true);
}
}
@ -238,7 +238,7 @@ class PackedIntValues {
private final PackedInts.Reader values;
public PackedIntsSource(IndexInput dataIn, boolean direct) throws IOException {
super(ValueType.VAR_INTS);
super(Type.VAR_INTS);
minValue = dataIn.readLong();
defaultValue = dataIn.readLong();
values = direct ? PackedInts.getDirectReader(dataIn) : PackedInts.getReader(dataIn);

View File

@ -22,12 +22,12 @@ import java.util.Collections;
import java.util.Comparator;
import java.util.List;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.MergeState;
import org.apache.lucene.index.DocValues.SortedSource;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.index.MergeState.IndexReaderAndLiveDocs;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.values.IndexDocValues.SortedSource;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.store.IndexOutput;
import org.apache.lucene.util.ArrayUtil;
import org.apache.lucene.util.BytesRef;
@ -43,11 +43,11 @@ final class SortedBytesMergeUtils {
// no instance
}
static MergeContext init(ValueType type, IndexDocValues[] docValues,
static MergeContext init(Type type, DocValues[] docValues,
Comparator<BytesRef> comp, MergeState mergeState) {
int size = -1;
if (type == ValueType.BYTES_FIXED_SORTED) {
for (IndexDocValues indexDocValues : docValues) {
if (type == Type.BYTES_FIXED_SORTED) {
for (DocValues indexDocValues : docValues) {
if (indexDocValues != null) {
size = indexDocValues.getValueSize();
break;
@ -62,13 +62,13 @@ final class SortedBytesMergeUtils {
private final Comparator<BytesRef> comp;
private final BytesRef missingValue = new BytesRef();
final int sizePerValues; // -1 if var length
final ValueType type;
final Type type;
final int[] docToEntry;
long[] offsets; // if non-null #mergeRecords collects byte offsets here
public MergeContext(Comparator<BytesRef> comp, MergeState mergeState,
int size, ValueType type) {
assert type == ValueType.BYTES_FIXED_SORTED || type == ValueType.BYTES_VAR_SORTED;
int size, Type type) {
assert type == Type.BYTES_FIXED_SORTED || type == Type.BYTES_VAR_SORTED;
this.comp = comp;
this.sizePerValues = size;
this.type = type;
@ -81,7 +81,7 @@ final class SortedBytesMergeUtils {
}
static List<SortedSourceSlice> buildSlices(MergeState mergeState,
IndexDocValues[] docValues, MergeContext ctx) throws IOException {
DocValues[] docValues, MergeContext ctx) throws IOException {
final List<SortedSourceSlice> slices = new ArrayList<SortedSourceSlice>();
for (int i = 0; i < docValues.length; i++) {
final SortedSourceSlice nextSlice;

View File

@ -22,7 +22,7 @@ import java.io.IOException;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesReaderBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesSourceBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.DerefBytesWriterBase;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
@ -93,7 +93,7 @@ class VarDerefBytesImpl {
public static class VarDerefReader extends BytesReaderBase {
private final long totalBytes;
VarDerefReader(Directory dir, String id, int maxDoc, IOContext context) throws IOException {
super(dir, id, CODEC_NAME, VERSION_START, true, context, ValueType.BYTES_VAR_DEREF);
super(dir, id, CODEC_NAME, VERSION_START, true, context, Type.BYTES_VAR_DEREF);
totalBytes = idxIn.readLong();
}
@ -115,7 +115,7 @@ class VarDerefBytesImpl {
public VarDerefSource(IndexInput datIn, IndexInput idxIn, long totalBytes)
throws IOException {
super(datIn, idxIn, new PagedBytes(PAGED_BYTES_BITS), totalBytes,
ValueType.BYTES_VAR_DEREF);
Type.BYTES_VAR_DEREF);
addresses = PackedInts.getReader(idxIn);
}
@ -130,7 +130,7 @@ class VarDerefBytesImpl {
final static class DirectVarDerefSource extends DirectSource {
private final PackedInts.Reader index;
DirectVarDerefSource(IndexInput data, IndexInput index, ValueType type)
DirectVarDerefSource(IndexInput data, IndexInput index, Type type)
throws IOException {
super(data, type);
this.index = PackedInts.getDirectReader(index);

View File

@ -21,15 +21,15 @@ import java.io.IOException;
import java.util.Comparator;
import java.util.List;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.SortedSource;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.index.MergeState;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesReaderBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesSortedSourceBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.DerefBytesWriterBase;
import org.apache.lucene.index.codecs.lucene40.values.SortedBytesMergeUtils.MergeContext;
import org.apache.lucene.index.codecs.lucene40.values.SortedBytesMergeUtils.SortedSourceSlice;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.values.IndexDocValues.SortedSource;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
@ -62,11 +62,11 @@ final class VarSortedBytesImpl {
size = 0;
}
@Override
public void merge(MergeState mergeState, IndexDocValues[] docValues)
public void merge(MergeState mergeState, DocValues[] docValues)
throws IOException {
boolean success = false;
try {
MergeContext ctx = SortedBytesMergeUtils.init(ValueType.BYTES_VAR_SORTED, docValues, comp, mergeState);
MergeContext ctx = SortedBytesMergeUtils.init(Type.BYTES_VAR_SORTED, docValues, comp, mergeState);
final List<SortedSourceSlice> slices = SortedBytesMergeUtils.buildSlices(mergeState, docValues, ctx);
IndexOutput datOut = getOrCreateDataOut();
@ -149,14 +149,14 @@ final class VarSortedBytesImpl {
private final Comparator<BytesRef> comparator;
Reader(Directory dir, String id, int maxDoc,
IOContext context, ValueType type, Comparator<BytesRef> comparator)
IOContext context, Type type, Comparator<BytesRef> comparator)
throws IOException {
super(dir, id, CODEC_NAME, VERSION_START, true, context, type);
this.comparator = comparator;
}
@Override
public org.apache.lucene.index.values.IndexDocValues.Source load()
public org.apache.lucene.index.DocValues.Source load()
throws IOException {
return new VarSortedSource(cloneData(), cloneIndex(), comparator);
}
@ -172,7 +172,7 @@ final class VarSortedBytesImpl {
VarSortedSource(IndexInput datIn, IndexInput idxIn,
Comparator<BytesRef> comp) throws IOException {
super(datIn, idxIn, comp, idxIn.readLong(), ValueType.BYTES_VAR_SORTED, true);
super(datIn, idxIn, comp, idxIn.readLong(), Type.BYTES_VAR_SORTED, true);
valueCount = ordToOffsetIndex.size()-1; // the last value here is just a dummy value to get the length of the last value
closeIndexInput();
}
@ -199,7 +199,7 @@ final class VarSortedBytesImpl {
private final int valueCount;
DirectSortedSource(IndexInput datIn, IndexInput idxIn,
Comparator<BytesRef> comparator, ValueType type) throws IOException {
Comparator<BytesRef> comparator, Type type) throws IOException {
super(type, comparator);
idxIn.readLong();
ordToOffsetIndex = PackedInts.getDirectReader(idxIn);

View File

@ -22,7 +22,7 @@ import java.io.IOException;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesReaderBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesSourceBase;
import org.apache.lucene.index.codecs.lucene40.values.Bytes.BytesWriterBase;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
@ -229,7 +229,7 @@ class VarStraightBytesImpl {
private final int maxDoc;
VarStraightReader(Directory dir, String id, int maxDoc, IOContext context) throws IOException {
super(dir, id, CODEC_NAME, VERSION_START, true, context, ValueType.BYTES_VAR_STRAIGHT);
super(dir, id, CODEC_NAME, VERSION_START, true, context, Type.BYTES_VAR_STRAIGHT);
this.maxDoc = maxDoc;
}
@ -250,7 +250,7 @@ class VarStraightBytesImpl {
public VarStraightSource(IndexInput datIn, IndexInput idxIn) throws IOException {
super(datIn, idxIn, new PagedBytes(PAGED_BYTES_BITS), idxIn.readVLong(),
ValueType.BYTES_VAR_STRAIGHT);
Type.BYTES_VAR_STRAIGHT);
addresses = PackedInts.getReader(idxIn);
}
@ -266,7 +266,7 @@ class VarStraightBytesImpl {
private final PackedInts.Reader index;
DirectVarStraightSource(IndexInput data, IndexInput index, ValueType type)
DirectVarStraightSource(IndexInput data, IndexInput index, Type type)
throws IOException {
super(data, type);
index.readVLong();

View File

@ -19,9 +19,9 @@ package org.apache.lucene.index.codecs.lucene40.values;
import java.io.IOException;
import java.util.Comparator;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.index.codecs.DocValuesConsumer;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.util.Bits;
@ -165,20 +165,20 @@ public abstract class Writer extends DocValuesConsumer {
/**
* Factory method to create a {@link Writer} instance for a given type. This
* method returns default implementations for each of the different types
* defined in the {@link ValueType} enumeration.
* defined in the {@link Type} enumeration.
*
* @param type
* the {@link ValueType} to create the {@link Writer} for
* the {@link Type} to create the {@link Writer} for
* @param id
* the file name id used to create files within the writer.
* @param directory
* the {@link Directory} to create the files from.
* @param bytesUsed
* a byte-usage tracking reference
* @return a new {@link Writer} instance for the given {@link ValueType}
* @return a new {@link Writer} instance for the given {@link Type}
* @throws IOException
*/
public static Writer create(ValueType type, String id, Directory directory,
public static Writer create(Type type, String id, Directory directory,
Comparator<BytesRef> comp, Counter bytesUsed, IOContext context) throws IOException {
if (comp == null) {
comp = BytesRef.getUTF8SortedAsUnicodeComparator();

View File

@ -22,9 +22,9 @@ import java.util.Collection;
import java.util.Map;
import java.util.TreeMap;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.SegmentReadState;
import org.apache.lucene.index.codecs.DocValuesReaderBase;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.util.IOUtils;
/**
@ -32,18 +32,18 @@ import org.apache.lucene.util.IOUtils;
* @lucene.experimental
*/
public class SepDocValuesProducer extends DocValuesReaderBase {
private final TreeMap<String, IndexDocValues> docValues;
private final TreeMap<String, DocValues> docValues;
/**
* Creates a new {@link SepDocValuesProducer} instance and loads all
* {@link IndexDocValues} instances for this segment and codec.
* {@link DocValues} instances for this segment and codec.
*/
public SepDocValuesProducer(SegmentReadState state) throws IOException {
docValues = load(state.fieldInfos, state.segmentInfo.name, state.segmentInfo.docCount, state.dir, state.context);
}
@Override
protected Map<String,IndexDocValues> docValues() {
protected Map<String,DocValues> docValues() {
return docValues;
}

View File

@ -26,8 +26,8 @@ import org.apache.lucene.index.FieldInfos;
import org.apache.lucene.index.IndexFileNames;
import org.apache.lucene.index.SegmentInfo;
import org.apache.lucene.index.FieldInfo.IndexOptions;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.codecs.FieldInfosReader;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.IOContext;
import org.apache.lucene.store.IndexInput;
@ -98,12 +98,12 @@ public class SimpleTextFieldInfosReader extends FieldInfosReader {
SimpleTextUtil.readLine(input, scratch);
assert StringHelper.startsWith(scratch, DOCVALUES);
String dvType = readString(DOCVALUES.length, scratch);
final ValueType docValuesType;
final DocValues.Type docValuesType;
if ("false".equals(dvType)) {
docValuesType = null;
} else {
docValuesType = ValueType.valueOf(dvType);
docValuesType = DocValues.Type.valueOf(dvType);
}
SimpleTextUtil.readLine(input, scratch);

View File

@ -1,425 +0,0 @@
package org.apache.lucene.index.values;
/**
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
import java.io.Closeable;
import java.io.IOException;
import java.util.Comparator;
import org.apache.lucene.document.IndexDocValuesField;
import org.apache.lucene.index.Fields;
import org.apache.lucene.index.FieldsEnum;
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.codecs.DocValuesFormat;
import org.apache.lucene.util.BytesRef;
import org.apache.lucene.util.packed.PackedInts;
/**
* {@link IndexDocValues} provides a dense per-document typed storage for fast
* value access based on the lucene internal document id. {@link IndexDocValues}
* exposes two distinct APIs:
* <ul>
* <li>via {@link #getSource()} providing RAM resident random access</li>
* <li>via {@link #getDirectSource()} providing on disk random access</li>
* </ul> {@link IndexDocValues} are exposed via
* {@link IndexReader#perDocValues()} on a per-segment basis. For best
* performance {@link IndexDocValues} should be consumed per-segment just like
* IndexReader.
* <p>
* {@link IndexDocValues} are fully integrated into the {@link DocValuesFormat} API.
*
* @see ValueType for limitations and default implementation documentation
* @see IndexDocValuesField for adding values to the index
* @see DocValuesFormat#docsConsumer(org.apache.lucene.index.PerDocWriteState) for
* customization
* @lucene.experimental
*/
public abstract class IndexDocValues implements Closeable {
public static final IndexDocValues[] EMPTY_ARRAY = new IndexDocValues[0];
private volatile SourceCache cache = new SourceCache.DirectSourceCache();
private final Object cacheLock = new Object();
/**
* Loads a new {@link Source} instance for this {@link IndexDocValues} field
* instance. Source instances returned from this method are not cached. It is
* the callers responsibility to maintain the instance and release its
* resources once the source is not needed anymore.
* <p>
* For managed {@link Source} instances see {@link #getSource()}.
*
* @see #getSource()
* @see #setCache(SourceCache)
*/
public abstract Source load() throws IOException;
/**
* Returns a {@link Source} instance through the current {@link SourceCache}.
* Iff no {@link Source} has been loaded into the cache so far the source will
* be loaded through {@link #load()} and passed to the {@link SourceCache}.
* The caller of this method should not close the obtained {@link Source}
* instance unless it is not needed for the rest of its life time.
* <p>
* {@link Source} instances obtained from this method are closed / released
* from the cache once this {@link IndexDocValues} instance is closed by the
* {@link IndexReader}, {@link Fields} or {@link FieldsEnum} the
* {@link IndexDocValues} was created from.
*/
public Source getSource() throws IOException {
return cache.load(this);
}
/**
* Returns a disk resident {@link Source} instance. Direct Sources are not
* cached in the {@link SourceCache} and should not be shared between threads.
*/
public abstract Source getDirectSource() throws IOException;
/**
* Returns the {@link ValueType} of this {@link IndexDocValues} instance
*/
public abstract ValueType type();
/**
* Closes this {@link IndexDocValues} instance. This method should only be called
* by the creator of this {@link IndexDocValues} instance. API users should not
* close {@link IndexDocValues} instances.
*/
public void close() throws IOException {
cache.close(this);
}
/**
* Returns the size per value in bytes or <code>-1</code> iff size per value
* is variable.
*
* @return the size per value in bytes or <code>-1</code> iff size per value
* is variable.
*/
public int getValueSize() {
return -1;
}
/**
* Sets the {@link SourceCache} used by this {@link IndexDocValues} instance. This
* method should be called before {@link #load()} is called. All {@link Source} instances in the currently used cache will be closed
* before the new cache is installed.
* <p>
* Note: All instances previously obtained from {@link #load()} will be lost.
*
* @throws IllegalArgumentException
* if the given cache is <code>null</code>
*
*/
public void setCache(SourceCache cache) {
if (cache == null)
throw new IllegalArgumentException("cache must not be null");
synchronized (cacheLock) {
SourceCache toClose = this.cache;
this.cache = cache;
toClose.close(this);
}
}
/**
* Source of per document values like long, double or {@link BytesRef}
* depending on the {@link IndexDocValues} fields {@link ValueType}. Source
* implementations provide random access semantics similar to array lookups
* <p>
* @see IndexDocValues#getSource()
* @see IndexDocValues#getDirectSource()
*/
public static abstract class Source {
protected final ValueType type;
protected Source(ValueType type) {
this.type = type;
}
/**
* Returns a <tt>long</tt> for the given document id or throws an
* {@link UnsupportedOperationException} if this source doesn't support
* <tt>long</tt> values.
*
* @throws UnsupportedOperationException
* if this source doesn't support <tt>long</tt> values.
*/
public long getInt(int docID) {
throw new UnsupportedOperationException("ints are not supported");
}
/**
* Returns a <tt>double</tt> for the given document id or throws an
* {@link UnsupportedOperationException} if this source doesn't support
* <tt>double</tt> values.
*
* @throws UnsupportedOperationException
* if this source doesn't support <tt>double</tt> values.
*/
public double getFloat(int docID) {
throw new UnsupportedOperationException("floats are not supported");
}
/**
* Returns a {@link BytesRef} for the given document id or throws an
* {@link UnsupportedOperationException} if this source doesn't support
* <tt>byte[]</tt> values.
* @throws IOException
*
* @throws UnsupportedOperationException
* if this source doesn't support <tt>byte[]</tt> values.
*/
public BytesRef getBytes(int docID, BytesRef ref) {
throw new UnsupportedOperationException("bytes are not supported");
}
/**
* Returns the {@link ValueType} of this source.
*
* @return the {@link ValueType} of this source.
*/
public ValueType type() {
return type;
}
/**
* Returns <code>true</code> iff this {@link Source} exposes an array via
* {@link #getArray()} otherwise <code>false</code>.
*
* @return <code>true</code> iff this {@link Source} exposes an array via
* {@link #getArray()} otherwise <code>false</code>.
*/
public boolean hasArray() {
return false;
}
/**
* Returns the internal array representation iff this {@link Source} uses an
* array as its inner representation, otherwise <code>UOE</code>.
*/
public Object getArray() {
throw new UnsupportedOperationException("getArray is not supported");
}
/**
* If this {@link Source} is sorted this method will return an instance of
* {@link SortedSource} otherwise <code>UOE</code>
*/
public SortedSource asSortedSource() {
throw new UnsupportedOperationException("asSortedSource is not supported");
}
}
/**
* A sorted variant of {@link Source} for <tt>byte[]</tt> values per document.
* <p>
*/
public static abstract class SortedSource extends Source {
private final Comparator<BytesRef> comparator;
protected SortedSource(ValueType type, Comparator<BytesRef> comparator) {
super(type);
this.comparator = comparator;
}
@Override
public BytesRef getBytes(int docID, BytesRef bytesRef) {
final int ord = ord(docID);
if (ord < 0) {
bytesRef.length = 0;
} else {
getByOrd(ord , bytesRef);
}
return bytesRef;
}
/**
* Returns ord for specified docID. Ord is dense, ie, starts at 0, then increments by 1
* for the next (as defined by {@link Comparator} value.
*/
public abstract int ord(int docID);
/** Returns value for specified ord. */
public abstract BytesRef getByOrd(int ord, BytesRef bytesRef);
/** Return true if it's safe to call {@link
* #getDocToOrd}. */
public boolean hasPackedDocToOrd() {
return false;
}
/**
* Returns the PackedInts.Reader impl that maps document to ord.
*/
public abstract PackedInts.Reader getDocToOrd();
/**
* Returns the comparator used to order the BytesRefs.
*/
public Comparator<BytesRef> getComparator() {
return comparator;
}
/**
* Performs a lookup by value.
*
* @param value
* the value to look up
* @param spare
* a spare {@link BytesRef} instance used to compare internal
* values to the given value. Must not be <code>null</code>
* @return the given values ordinal if found or otherwise
* <code>(-(ord)-1)</code>, defined as the ordinal of the first
* element that is greater than the given value. This guarantees
* that the return value will always be &gt;= 0 if the given value
* is found.
*/
public int getByValue(BytesRef value, BytesRef spare) {
return binarySearch(value, spare, 0, getValueCount() - 1);
}
private int binarySearch(BytesRef b, BytesRef bytesRef, int low,
int high) {
int mid = 0;
while (low <= high) {
mid = (low + high) >>> 1;
getByOrd(mid, bytesRef);
final int cmp = comparator.compare(bytesRef, b);
if (cmp < 0) {
low = mid + 1;
} else if (cmp > 0) {
high = mid - 1;
} else {
return mid;
}
}
assert comparator.compare(bytesRef, b) != 0;
return -(low + 1);
}
@Override
public SortedSource asSortedSource() {
return this;
}
/**
* Returns the number of unique values in this sorted source
*/
public abstract int getValueCount();
}
/** Returns a Source that always returns default (missing)
* values for all documents. */
public static Source getDefaultSource(final ValueType type) {
return new Source(type) {
@Override
public long getInt(int docID) {
return 0;
}
@Override
public double getFloat(int docID) {
return 0.0;
}
@Override
public BytesRef getBytes(int docID, BytesRef ref) {
ref.length = 0;
return ref;
}
};
}
/** Returns a SortedSource that always returns default (missing)
* values for all documents. */
public static SortedSource getDefaultSortedSource(final ValueType type, final int size) {
final PackedInts.Reader docToOrd = new PackedInts.Reader() {
@Override
public long get(int index) {
return 0;
}
@Override
public int getBitsPerValue() {
return 0;
}
@Override
public int size() {
return size;
}
@Override
public boolean hasArray() {
return false;
}
@Override
public Object getArray() {
return null;
}
};
return new SortedSource(type, BytesRef.getUTF8SortedAsUnicodeComparator()) {
@Override
public BytesRef getBytes(int docID, BytesRef ref) {
ref.length = 0;
return ref;
}
@Override
public int ord(int docID) {
return 0;
}
@Override
public BytesRef getByOrd(int ord, BytesRef bytesRef) {
assert ord == 0;
bytesRef.length = 0;
return bytesRef;
}
@Override
public boolean hasPackedDocToOrd() {
return true;
}
@Override
public PackedInts.Reader getDocToOrd() {
return docToOrd;
}
@Override
public int getByValue(BytesRef value, BytesRef spare) {
if (value.length == 0) {
return 0;
} else {
return -1;
}
}
@Override
public int getValueCount() {
return 1;
}
};
}
}

View File

@ -1,93 +0,0 @@
package org.apache.lucene.index.values;
/**
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
import java.io.IOException;
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.values.IndexDocValues.Source;
/**
* Abstract base class for {@link IndexDocValues} {@link Source} cache.
* <p>
* {@link Source} instances loaded via {@link IndexDocValues#load()} are entirely memory resident
* and need to be maintained by the caller. Each call to
* {@link IndexDocValues#load()} will cause an entire reload of
* the underlying data. Source instances obtained from
* {@link IndexDocValues#getSource()} and {@link IndexDocValues#getSource()}
* respectively are maintained by a {@link SourceCache} that is closed (
* {@link #close(IndexDocValues)}) once the {@link IndexReader} that created the
* {@link IndexDocValues} instance is closed.
* <p>
* Unless {@link Source} instances are managed by another entity it is
* recommended to use the cached variants to obtain a source instance.
* <p>
* Implementation of this API must be thread-safe.
*
* @see IndexDocValues#setCache(SourceCache)
* @see IndexDocValues#getSource()
*
* @lucene.experimental
*/
public abstract class SourceCache {
/**
* Atomically loads a {@link Source} into the cache from the given
* {@link IndexDocValues} and returns it iff no other {@link Source} has already
* been cached. Otherwise the cached source is returned.
* <p>
* This method will not return <code>null</code>
*/
public abstract Source load(IndexDocValues values) throws IOException;
/**
* Atomically invalidates the cached {@link Source}
* instances if any and empties the cache.
*/
public abstract void invalidate(IndexDocValues values);
/**
* Atomically closes the cache and frees all resources.
*/
public synchronized void close(IndexDocValues values) {
invalidate(values);
}
/**
* Simple per {@link IndexDocValues} instance cache implementation that holds a
* {@link Source} a member variable.
* <p>
* If a {@link DirectSourceCache} instance is closed or invalidated the cached
* reference are simply set to <code>null</code>
*/
public static final class DirectSourceCache extends SourceCache {
private Source ref;
public synchronized Source load(IndexDocValues values) throws IOException {
if (ref == null) {
ref = values.load();
}
return ref;
}
public synchronized void invalidate(IndexDocValues values) {
ref = null;
}
}
}

View File

@ -1,220 +0,0 @@
package org.apache.lucene.index.values;
/**
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
import org.apache.lucene.index.codecs.DocValuesFormat;
import org.apache.lucene.index.values.IndexDocValues.SortedSource;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.util.BytesRef;
import org.apache.lucene.util.packed.PackedInts;
/**
* <code>ValueType</code> specifies the {@link IndexDocValues} type for a
* certain field. A <code>ValueType</code> only defines the data type for a field
* while the actual implementation used to encode and decode the values depends
* on the the {@link DocValuesFormat#docsConsumer} and {@link DocValuesFormat#docsProducer} methods.
*
* @lucene.experimental
*/
public enum ValueType {
/**
* A variable bit signed integer value. By default this type uses
* {@link PackedInts} to compress the values, as an offset
* from the minimum value, as long as the value range
* fits into 2<sup>63</sup>-1. Otherwise,
* the default implementation falls back to fixed size 64bit
* integers ({@link #FIXED_INTS_64}).
* <p>
* NOTE: this type uses <tt>0</tt> as the default value without any
* distinction between provided <tt>0</tt> values during indexing. All
* documents without an explicit value will use <tt>0</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
VAR_INTS,
/**
* A 8 bit signed integer value. {@link Source} instances of
* this type return a <tt>byte</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0</tt> as the default value without any
* distinction between provided <tt>0</tt> values during indexing. All
* documents without an explicit value will use <tt>0</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FIXED_INTS_8,
/**
* A 16 bit signed integer value. {@link Source} instances of
* this type return a <tt>short</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0</tt> as the default value without any
* distinction between provided <tt>0</tt> values during indexing. All
* documents without an explicit value will use <tt>0</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FIXED_INTS_16,
/**
* A 32 bit signed integer value. {@link Source} instances of
* this type return a <tt>int</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0</tt> as the default value without any
* distinction between provided <tt>0</tt> values during indexing. All
* documents without an explicit value will use <tt>0</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FIXED_INTS_32,
/**
* A 64 bit signed integer value. {@link Source} instances of
* this type return a <tt>long</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0</tt> as the default value without any
* distinction between provided <tt>0</tt> values during indexing. All
* documents without an explicit value will use <tt>0</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FIXED_INTS_64,
/**
* A 32 bit floating point value. By default there is no compression
* applied. To fit custom float values into less than 32bit either a custom
* implementation is needed or values must be encoded into a
* {@link #BYTES_FIXED_STRAIGHT} type. {@link Source} instances of
* this type return a <tt>float</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0.0f</tt> as the default value without any
* distinction between provided <tt>0.0f</tt> values during indexing. All
* documents without an explicit value will use <tt>0.0f</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FLOAT_32,
/**
*
* A 64 bit floating point value. By default there is no compression
* applied. To fit custom float values into less than 64bit either a custom
* implementation is needed or values must be encoded into a
* {@link #BYTES_FIXED_STRAIGHT} type. {@link Source} instances of
* this type return a <tt>double</tt> array from {@link Source#getArray()}
* <p>
* NOTE: this type uses <tt>0.0d</tt> as the default value without any
* distinction between provided <tt>0.0d</tt> values during indexing. All
* documents without an explicit value will use <tt>0.0d</tt> instead.
* Custom default values must be assigned explicitly.
* </p>
*/
FLOAT_64,
// TODO(simonw): -- shouldn't lucene decide/detect straight vs
// deref, as well fixed vs var?
/**
* A fixed length straight byte[]. All values added to
* such a field must be of the same length. All bytes are stored sequentially
* for fast offset access.
* <p>
* NOTE: this type uses <tt>0 byte</tt> filled byte[] based on the length of the first seen
* value as the default value without any distinction between explicitly
* provided values during indexing. All documents without an explicit value
* will use the default instead.Custom default values must be assigned explicitly.
* </p>
*/
BYTES_FIXED_STRAIGHT,
/**
* A fixed length dereferenced byte[] variant. Fields with
* this type only store distinct byte values and store an additional offset
* pointer per document to dereference the shared byte[].
* Use this type if your documents may share the same byte[].
* <p>
* NOTE: Fields of this type will not store values for documents without and
* explicitly provided value. If a documents value is accessed while no
* explicit value is stored the returned {@link BytesRef} will be a 0-length
* reference. Custom default values must be assigned explicitly.
* </p>
*/
BYTES_FIXED_DEREF,
/**
* Variable length straight stored byte[] variant. All bytes are
* stored sequentially for compactness. Usage of this type via the
* disk-resident API might yield performance degradation since no additional
* index is used to advance by more than one document value at a time.
* <p>
* NOTE: Fields of this type will not store values for documents without an
* explicitly provided value. If a documents value is accessed while no
* explicit value is stored the returned {@link BytesRef} will be a 0-length
* byte[] reference. Custom default values must be assigned explicitly.
* </p>
*/
BYTES_VAR_STRAIGHT,
/**
* A variable length dereferenced byte[]. Just like
* {@link #BYTES_FIXED_DEREF}, but allowing each
* document's value to be a different length.
* <p>
* NOTE: Fields of this type will not store values for documents without and
* explicitly provided value. If a documents value is accessed while no
* explicit value is stored the returned {@link BytesRef} will be a 0-length
* reference. Custom default values must be assigned explicitly.
* </p>
*/
BYTES_VAR_DEREF,
/**
* A variable length pre-sorted byte[] variant. Just like
* {@link #BYTES_FIXED_SORTED}, but allowing each
* document's value to be a different length.
* <p>
* NOTE: Fields of this type will not store values for documents without and
* explicitly provided value. If a documents value is accessed while no
* explicit value is stored the returned {@link BytesRef} will be a 0-length
* reference.Custom default values must be assigned explicitly.
* </p>
*
* @see SortedSource
*/
BYTES_VAR_SORTED,
/**
* A fixed length pre-sorted byte[] variant. Fields with this type only
* store distinct byte values and store an additional offset pointer per
* document to dereference the shared byte[]. The stored
* byte[] is presorted, by default by unsigned byte order,
* and allows access via document id, ordinal and by-value.
* Use this type if your documents may share the same byte[].
* <p>
* NOTE: Fields of this type will not store values for documents without and
* explicitly provided value. If a documents value is accessed while no
* explicit value is stored the returned {@link BytesRef} will be a 0-length
* reference. Custom default values must be assigned
* explicitly.
* </p>
*
* @see SortedSource
*/
BYTES_FIXED_SORTED
}

View File

@ -21,11 +21,8 @@ import java.io.IOException;
import java.util.Comparator;
import org.apache.lucene.index.IndexReader.AtomicReaderContext;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.values.IndexDocValues.SortedSource;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.search.FieldCache.ByteParser;
import org.apache.lucene.search.FieldCache.DocTerms;
import org.apache.lucene.search.FieldCache.DocTermsIndex;
@ -360,7 +357,7 @@ public abstract class FieldComparator<T> {
public static final class FloatDocValuesComparator extends FieldComparator<Double> {
private final double[] values;
private final String field;
private Source currentReaderValues;
private DocValues.Source currentReaderValues;
private double bottom;
FloatDocValuesComparator(int numHits, String field) {
@ -400,11 +397,11 @@ public abstract class FieldComparator<T> {
@Override
public FieldComparator setNextReader(AtomicReaderContext context) throws IOException {
final IndexDocValues docValues = context.reader.docValues(field);
final DocValues docValues = context.reader.docValues(field);
if (docValues != null) {
currentReaderValues = docValues.getSource();
} else {
currentReaderValues = IndexDocValues.getDefaultSource(ValueType.FLOAT_64);
currentReaderValues = DocValues.getDefaultSource(DocValues.Type.FLOAT_64);
}
return this;
}
@ -648,7 +645,7 @@ public abstract class FieldComparator<T> {
/** Loads int index values and sorts by ascending value. */
public static final class IntDocValuesComparator extends FieldComparator<Long> {
private final long[] values;
private Source currentReaderValues;
private DocValues.Source currentReaderValues;
private final String field;
private long bottom;
@ -693,11 +690,11 @@ public abstract class FieldComparator<T> {
@Override
public FieldComparator setNextReader(AtomicReaderContext context) throws IOException {
IndexDocValues docValues = context.reader.docValues(field);
DocValues docValues = context.reader.docValues(field);
if (docValues != null) {
currentReaderValues = docValues.getSource();
} else {
currentReaderValues = IndexDocValues.getDefaultSource(ValueType.FIXED_INTS_64);
currentReaderValues = DocValues.getDefaultSource(DocValues.Type.FIXED_INTS_64);
}
return this;
}
@ -1382,7 +1379,7 @@ public abstract class FieldComparator<T> {
/* Current reader's doc ord/values.
@lucene.internal */
SortedSource termsIndex;
DocValues.SortedSource termsIndex;
/* Comparator for comparing by value.
@lucene.internal */
@ -1490,10 +1487,10 @@ public abstract class FieldComparator<T> {
// Used per-segment when bit width of doc->ord is 8:
private final class ByteOrdComparator extends PerSegmentComparator {
private final byte[] readerOrds;
private final SortedSource termsIndex;
private final DocValues.SortedSource termsIndex;
private final int docBase;
public ByteOrdComparator(byte[] readerOrds, SortedSource termsIndex, int docBase) {
public ByteOrdComparator(byte[] readerOrds, DocValues.SortedSource termsIndex, int docBase) {
this.readerOrds = readerOrds;
this.termsIndex = termsIndex;
this.docBase = docBase;
@ -1535,10 +1532,10 @@ public abstract class FieldComparator<T> {
// Used per-segment when bit width of doc->ord is 16:
private final class ShortOrdComparator extends PerSegmentComparator {
private final short[] readerOrds;
private final SortedSource termsIndex;
private final DocValues.SortedSource termsIndex;
private final int docBase;
public ShortOrdComparator(short[] readerOrds, SortedSource termsIndex, int docBase) {
public ShortOrdComparator(short[] readerOrds, DocValues.SortedSource termsIndex, int docBase) {
this.readerOrds = readerOrds;
this.termsIndex = termsIndex;
this.docBase = docBase;
@ -1580,10 +1577,10 @@ public abstract class FieldComparator<T> {
// Used per-segment when bit width of doc->ord is 32:
private final class IntOrdComparator extends PerSegmentComparator {
private final int[] readerOrds;
private final SortedSource termsIndex;
private final DocValues.SortedSource termsIndex;
private final int docBase;
public IntOrdComparator(int[] readerOrds, SortedSource termsIndex, int docBase) {
public IntOrdComparator(int[] readerOrds, DocValues.SortedSource termsIndex, int docBase) {
this.readerOrds = readerOrds;
this.termsIndex = termsIndex;
this.docBase = docBase;
@ -1709,12 +1706,12 @@ public abstract class FieldComparator<T> {
public FieldComparator setNextReader(AtomicReaderContext context) throws IOException {
final int docBase = context.docBase;
final IndexDocValues dv = context.reader.docValues(field);
final DocValues dv = context.reader.docValues(field);
if (dv == null) {
// This may mean entire segment had no docs with
// this DV field; use default field value (empty
// byte[]) in this case:
termsIndex = IndexDocValues.getDefaultSortedSource(ValueType.BYTES_VAR_SORTED, context.reader.maxDoc());
termsIndex = DocValues.getDefaultSortedSource(DocValues.Type.BYTES_VAR_SORTED, context.reader.maxDoc());
} else {
termsIndex = dv.getSource().asSortedSource();
if (termsIndex == null) {
@ -1892,7 +1889,7 @@ public abstract class FieldComparator<T> {
public static final class TermValDocValuesComparator extends FieldComparator<BytesRef> {
private BytesRef[] values;
private Source docTerms;
private DocValues.Source docTerms;
private final String field;
private BytesRef bottom;
private final BytesRef tempBR = new BytesRef();
@ -1925,11 +1922,11 @@ public abstract class FieldComparator<T> {
@Override
public FieldComparator setNextReader(AtomicReaderContext context) throws IOException {
final IndexDocValues dv = context.reader.docValues(field);
final DocValues dv = context.reader.docValues(field);
if (dv != null) {
docTerms = dv.getSource();
} else {
docTerms = IndexDocValues.getDefaultSource(ValueType.BYTES_VAR_DEREF);
docTerms = DocValues.getDefaultSource(DocValues.Type.BYTES_VAR_DEREF);
}
return this;
}

View File

@ -20,7 +20,7 @@ package org.apache.lucene.search.similarities;
import java.io.IOException;
import org.apache.lucene.document.IndexDocValuesField; // javadoc
import org.apache.lucene.document.DocValuesField; // javadoc
import org.apache.lucene.index.FieldInvertState;
import org.apache.lucene.index.IndexReader; // javadoc
import org.apache.lucene.index.IndexReader.AtomicReaderContext;
@ -71,11 +71,11 @@ import org.apache.lucene.util.TermContext;
* <p>
* Because index-time boost is handled entirely at the application level anyway,
* an application can alternatively store the index-time boost separately using an
* {@link IndexDocValuesField}, and access this at query-time with
* {@link DocValuesField}, and access this at query-time with
* {@link IndexReader#docValues(String)}.
* <p>
* Finally, using index-time boosts (either via folding into the normalization byte or
* via IndexDocValues), is an inefficient way to boost the scores of different fields if the
* via DocValues), is an inefficient way to boost the scores of different fields if the
* boost will be the same for every document, instead the Similarity can simply take a constant
* boost parameter <i>C</i>, and the SimilarityProvider can return different instances with
* different boosts depending upon field name.

View File

@ -24,11 +24,11 @@ import java.util.Random;
import org.apache.lucene.analysis.Analyzer;
import org.apache.lucene.analysis.MockAnalyzer;
import org.apache.lucene.document.IndexDocValuesField;
import org.apache.lucene.document.DocValuesField;
import org.apache.lucene.document.Document;
import org.apache.lucene.index.IndexWriter; // javadoc
import org.apache.lucene.index.codecs.Codec;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.search.Query;
import org.apache.lucene.store.Directory;
import org.apache.lucene.util.BytesRef;
@ -167,12 +167,12 @@ public class RandomIndexWriter implements Closeable {
private void randomPerDocFieldValues(Random random, Document doc) {
ValueType[] values = ValueType.values();
ValueType type = values[random.nextInt(values.length)];
DocValues.Type[] values = DocValues.Type.values();
DocValues.Type type = values[random.nextInt(values.length)];
String name = "random_" + type.name() + "" + docValuesFieldPrefix;
if ("Lucene3x".equals(codec.getName()) || doc.getField(name) != null)
return;
IndexDocValuesField docValuesField = new IndexDocValuesField(name);
DocValuesField docValuesField = new DocValuesField(name);
switch (type) {
case BYTES_FIXED_DEREF:
case BYTES_FIXED_STRAIGHT:

View File

@ -20,12 +20,12 @@ package org.apache.lucene.index.codecs.mocksep;
import java.io.IOException;
import java.util.Set;
import org.apache.lucene.index.PerDocValues;
import org.apache.lucene.index.PerDocWriteState;
import org.apache.lucene.index.SegmentInfo;
import org.apache.lucene.index.SegmentReadState;
import org.apache.lucene.index.codecs.DocValuesFormat;
import org.apache.lucene.index.codecs.PerDocConsumer;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.index.codecs.sep.SepDocValuesConsumer;
import org.apache.lucene.index.codecs.sep.SepDocValuesProducer;
import org.apache.lucene.store.Directory;

View File

@ -27,7 +27,7 @@ import org.apache.lucene.analysis.MockAnalyzer;
import org.apache.lucene.document.Document;
import org.apache.lucene.document.Field;
import org.apache.lucene.document.FieldType;
import org.apache.lucene.document.IndexDocValuesField;
import org.apache.lucene.document.DocValuesField;
import org.apache.lucene.document.StringField;
import org.apache.lucene.document.TextField;
import org.apache.lucene.index.IndexWriterConfig.OpenMode;
@ -47,7 +47,6 @@ import org.apache.lucene.index.codecs.lucene40.Lucene40SegmentInfosFormat;
import org.apache.lucene.index.codecs.lucene40.Lucene40StoredFieldsFormat;
import org.apache.lucene.index.codecs.lucene40.Lucene40TermVectorsFormat;
import org.apache.lucene.index.codecs.pulsing.Pulsing40PostingsFormat;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.search.DocIdSetIterator;
import org.apache.lucene.search.PhraseQuery;
import org.apache.lucene.store.AlreadyClosedException;
@ -1254,7 +1253,7 @@ public class TestAddIndexes extends LuceneTestCase {
RandomIndexWriter w = new RandomIndexWriter(random, d1);
Document doc = new Document();
doc.add(newField("id", "1", StringField.TYPE_STORED));
IndexDocValuesField dv = new IndexDocValuesField("dv");
DocValuesField dv = new DocValuesField("dv");
dv.setInt(1);
doc.add(dv);
w.addDocument(doc);
@ -1265,7 +1264,7 @@ public class TestAddIndexes extends LuceneTestCase {
w = new RandomIndexWriter(random, d2);
doc = new Document();
doc.add(newField("id", "2", StringField.TYPE_STORED));
dv = new IndexDocValuesField("dv");
dv = new DocValuesField("dv");
dv.setInt(2);
doc.add(dv);
w.addDocument(doc);
@ -1285,7 +1284,7 @@ public class TestAddIndexes extends LuceneTestCase {
w.close();
IndexReader sr = getOnlySegmentReader(r3);
assertEquals(2, sr.numDocs());
IndexDocValues docValues = sr.perDocValues().docValues("dv");
DocValues docValues = sr.perDocValues().docValues("dv");
assertNotNull(docValues);
r3.close();
d3.close();

View File

@ -25,10 +25,6 @@ import org.apache.lucene.analysis.Analyzer;
import org.apache.lucene.analysis.MockAnalyzer;
import org.apache.lucene.document.Document;
import org.apache.lucene.index.codecs.Codec;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.search.DocIdSetIterator;
import org.apache.lucene.store.Directory;
import org.apache.lucene.util.Bits;
@ -536,8 +532,8 @@ public class TestDuelingCodecs extends LuceneTestCase {
FieldsEnum fieldsEnum = leftFields.iterator();
String field;
while ((field = fieldsEnum.next()) != null) {
IndexDocValues leftDocValues = leftPerDoc.docValues(field);
IndexDocValues rightDocValues = rightPerDoc.docValues(field);
DocValues leftDocValues = leftPerDoc.docValues(field);
DocValues rightDocValues = rightPerDoc.docValues(field);
if (leftDocValues == null || rightDocValues == null) {
assertNull(info, leftDocValues);
assertNull(info, rightDocValues);
@ -551,8 +547,8 @@ public class TestDuelingCodecs extends LuceneTestCase {
/**
* checks source API
*/
public void assertDocValuesSource(Source left, Source right) throws Exception {
ValueType leftType = left.type();
public void assertDocValuesSource(DocValues.Source left, DocValues.Source right) throws Exception {
DocValues.Type leftType = left.type();
assertEquals(info, leftType, right.type());
switch(leftType) {
case VAR_INTS:

View File

@ -28,8 +28,7 @@ import org.apache.lucene.document.Document;
import org.apache.lucene.document.NumericField.DataType;
import org.apache.lucene.document.NumericField;
import org.apache.lucene.document.StringField;
import org.apache.lucene.index.values.PerDocFieldValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.search.BooleanClause;
import org.apache.lucene.search.BooleanQuery;
import org.apache.lucene.search.IndexSearcher;
@ -162,7 +161,7 @@ public class TestIndexableField extends LuceneTestCase {
}
@Override
public ValueType docValuesType() {
public DocValues.Type docValuesType() {
return null;
}

View File

@ -20,12 +20,14 @@ package org.apache.lucene.index.values;
import java.io.IOException;
import java.util.Comparator;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.SortedSource;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.index.codecs.lucene40.values.Bytes;
import org.apache.lucene.index.codecs.lucene40.values.Floats;
import org.apache.lucene.index.codecs.lucene40.values.Ints;
import org.apache.lucene.index.codecs.lucene40.values.Writer;
import org.apache.lucene.index.values.IndexDocValues.SortedSource;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.store.Directory;
import org.apache.lucene.util.BytesRef;
import org.apache.lucene.util.Counter;
@ -33,6 +35,7 @@ import org.apache.lucene.util.LuceneTestCase;
import org.apache.lucene.util.UnicodeUtil;
import org.apache.lucene.util._TestUtil;
// TODO: some of this should be under lucene40 codec tests? is talking to codec directly?f
public class TestDocValues extends LuceneTestCase {
private static final Comparator<BytesRef> COMP = BytesRef.getUTF8SortedAsUnicodeComparator();
// TODO -- for sorted test, do our own Sort of the
@ -80,12 +83,12 @@ public class TestDocValues extends LuceneTestCase {
w.finish(maxDoc);
assertEquals(0, trackBytes.get());
IndexDocValues r = Bytes.getValues(dir, "test", mode, fixedSize, maxDoc, COMP, newIOContext(random));
DocValues r = Bytes.getValues(dir, "test", mode, fixedSize, maxDoc, COMP, newIOContext(random));
// Verify we can load source twice:
for (int iter = 0; iter < 2; iter++) {
Source s;
IndexDocValues.SortedSource ss;
DocValues.SortedSource ss;
if (mode == Bytes.Mode.SORTED) {
// default is unicode so we can simply pass null here
s = ss = getSortedSource(r);
@ -159,19 +162,19 @@ public class TestDocValues extends LuceneTestCase {
{ Long.MIN_VALUE + 1, 1 }, { -1, Long.MAX_VALUE },
{ Long.MIN_VALUE, -1 }, { 1, Long.MAX_VALUE },
{ -1, Long.MAX_VALUE - 1 }, { Long.MIN_VALUE + 2, 1 }, };
ValueType[] expectedTypes = new ValueType[] { ValueType.FIXED_INTS_64,
ValueType.FIXED_INTS_64, ValueType.FIXED_INTS_64,
ValueType.FIXED_INTS_64, ValueType.VAR_INTS, ValueType.VAR_INTS,
ValueType.VAR_INTS, };
Type[] expectedTypes = new Type[] { Type.FIXED_INTS_64,
Type.FIXED_INTS_64, Type.FIXED_INTS_64,
Type.FIXED_INTS_64, Type.VAR_INTS, Type.VAR_INTS,
Type.VAR_INTS, };
for (int i = 0; i < minMax.length; i++) {
Directory dir = newDirectory();
final Counter trackBytes = Counter.newCounter();
Writer w = Ints.getWriter(dir, "test", trackBytes, ValueType.VAR_INTS, newIOContext(random));
Writer w = Ints.getWriter(dir, "test", trackBytes, Type.VAR_INTS, newIOContext(random));
w.add(0, minMax[i][0]);
w.add(1, minMax[i][1]);
w.finish(2);
assertEquals(0, trackBytes.get());
IndexDocValues r = Ints.getValues(dir, "test", 2, ValueType.VAR_INTS, newIOContext(random));
DocValues r = Ints.getValues(dir, "test", 2, Type.VAR_INTS, newIOContext(random));
Source source = getSource(r);
assertEquals(i + " with min: " + minMax[i][0] + " max: " + minMax[i][1],
expectedTypes[i], source.type());
@ -184,14 +187,14 @@ public class TestDocValues extends LuceneTestCase {
}
public void testVInts() throws IOException {
testInts(ValueType.VAR_INTS, 63);
testInts(Type.VAR_INTS, 63);
}
public void testFixedInts() throws IOException {
testInts(ValueType.FIXED_INTS_64, 63);
testInts(ValueType.FIXED_INTS_32, 31);
testInts(ValueType.FIXED_INTS_16, 15);
testInts(ValueType.FIXED_INTS_8, 7);
testInts(Type.FIXED_INTS_64, 63);
testInts(Type.FIXED_INTS_32, 31);
testInts(Type.FIXED_INTS_16, 15);
testInts(Type.FIXED_INTS_8, 7);
}
@ -199,12 +202,12 @@ public class TestDocValues extends LuceneTestCase {
byte[] sourceArray = new byte[] {1,2,3};
Directory dir = newDirectory();
final Counter trackBytes = Counter.newCounter();
Writer w = Ints.getWriter(dir, "test", trackBytes, ValueType.FIXED_INTS_8, newIOContext(random));
Writer w = Ints.getWriter(dir, "test", trackBytes, Type.FIXED_INTS_8, newIOContext(random));
for (int i = 0; i < sourceArray.length; i++) {
w.add(i, (long) sourceArray[i]);
}
w.finish(sourceArray.length);
IndexDocValues r = Ints.getValues(dir, "test", sourceArray.length, ValueType.FIXED_INTS_8, newIOContext(random));
DocValues r = Ints.getValues(dir, "test", sourceArray.length, Type.FIXED_INTS_8, newIOContext(random));
Source source = r.getSource();
assertTrue(source.hasArray());
byte[] loaded = ((byte[])source.getArray());
@ -220,12 +223,12 @@ public class TestDocValues extends LuceneTestCase {
short[] sourceArray = new short[] {1,2,3};
Directory dir = newDirectory();
final Counter trackBytes = Counter.newCounter();
Writer w = Ints.getWriter(dir, "test", trackBytes, ValueType.FIXED_INTS_16, newIOContext(random));
Writer w = Ints.getWriter(dir, "test", trackBytes, Type.FIXED_INTS_16, newIOContext(random));
for (int i = 0; i < sourceArray.length; i++) {
w.add(i, (long) sourceArray[i]);
}
w.finish(sourceArray.length);
IndexDocValues r = Ints.getValues(dir, "test", sourceArray.length, ValueType.FIXED_INTS_16, newIOContext(random));
DocValues r = Ints.getValues(dir, "test", sourceArray.length, Type.FIXED_INTS_16, newIOContext(random));
Source source = r.getSource();
assertTrue(source.hasArray());
short[] loaded = ((short[])source.getArray());
@ -241,12 +244,12 @@ public class TestDocValues extends LuceneTestCase {
long[] sourceArray = new long[] {1,2,3};
Directory dir = newDirectory();
final Counter trackBytes = Counter.newCounter();
Writer w = Ints.getWriter(dir, "test", trackBytes, ValueType.FIXED_INTS_64, newIOContext(random));
Writer w = Ints.getWriter(dir, "test", trackBytes, Type.FIXED_INTS_64, newIOContext(random));
for (int i = 0; i < sourceArray.length; i++) {
w.add(i, sourceArray[i]);
}
w.finish(sourceArray.length);
IndexDocValues r = Ints.getValues(dir, "test", sourceArray.length, ValueType.FIXED_INTS_64, newIOContext(random));
DocValues r = Ints.getValues(dir, "test", sourceArray.length, Type.FIXED_INTS_64, newIOContext(random));
Source source = r.getSource();
assertTrue(source.hasArray());
long[] loaded = ((long[])source.getArray());
@ -262,12 +265,12 @@ public class TestDocValues extends LuceneTestCase {
int[] sourceArray = new int[] {1,2,3};
Directory dir = newDirectory();
final Counter trackBytes = Counter.newCounter();
Writer w = Ints.getWriter(dir, "test", trackBytes, ValueType.FIXED_INTS_32, newIOContext(random));
Writer w = Ints.getWriter(dir, "test", trackBytes, Type.FIXED_INTS_32, newIOContext(random));
for (int i = 0; i < sourceArray.length; i++) {
w.add(i, (long) sourceArray[i]);
}
w.finish(sourceArray.length);
IndexDocValues r = Ints.getValues(dir, "test", sourceArray.length, ValueType.FIXED_INTS_32, newIOContext(random));
DocValues r = Ints.getValues(dir, "test", sourceArray.length, Type.FIXED_INTS_32, newIOContext(random));
Source source = r.getSource();
assertTrue(source.hasArray());
int[] loaded = ((int[])source.getArray());
@ -283,12 +286,12 @@ public class TestDocValues extends LuceneTestCase {
float[] sourceArray = new float[] {1,2,3};
Directory dir = newDirectory();
final Counter trackBytes = Counter.newCounter();
Writer w = Floats.getWriter(dir, "test", trackBytes, newIOContext(random), ValueType.FLOAT_32);
Writer w = Floats.getWriter(dir, "test", trackBytes, newIOContext(random), Type.FLOAT_32);
for (int i = 0; i < sourceArray.length; i++) {
w.add(i, sourceArray[i]);
}
w.finish(sourceArray.length);
IndexDocValues r = Floats.getValues(dir, "test", 3, newIOContext(random), ValueType.FLOAT_32);
DocValues r = Floats.getValues(dir, "test", 3, newIOContext(random), Type.FLOAT_32);
Source source = r.getSource();
assertTrue(source.hasArray());
float[] loaded = ((float[])source.getArray());
@ -304,12 +307,12 @@ public class TestDocValues extends LuceneTestCase {
double[] sourceArray = new double[] {1,2,3};
Directory dir = newDirectory();
final Counter trackBytes = Counter.newCounter();
Writer w = Floats.getWriter(dir, "test", trackBytes, newIOContext(random), ValueType.FLOAT_64);
Writer w = Floats.getWriter(dir, "test", trackBytes, newIOContext(random), Type.FLOAT_64);
for (int i = 0; i < sourceArray.length; i++) {
w.add(i, sourceArray[i]);
}
w.finish(sourceArray.length);
IndexDocValues r = Floats.getValues(dir, "test", 3, newIOContext(random), ValueType.FLOAT_64);
DocValues r = Floats.getValues(dir, "test", 3, newIOContext(random), Type.FLOAT_64);
Source source = r.getSource();
assertTrue(source.hasArray());
double[] loaded = ((double[])source.getArray());
@ -321,7 +324,7 @@ public class TestDocValues extends LuceneTestCase {
dir.close();
}
private void testInts(ValueType type, int maxBit) throws IOException {
private void testInts(Type type, int maxBit) throws IOException {
long maxV = 1;
final int NUM_VALUES = 333 + random.nextInt(333);
final long[] values = new long[NUM_VALUES];
@ -338,7 +341,7 @@ public class TestDocValues extends LuceneTestCase {
w.finish(NUM_VALUES + additionalDocs);
assertEquals(0, trackBytes.get());
IndexDocValues r = Ints.getValues(dir, "test", NUM_VALUES + additionalDocs, type, newIOContext(random));
DocValues r = Ints.getValues(dir, "test", NUM_VALUES + additionalDocs, type, newIOContext(random));
for (int iter = 0; iter < 2; iter++) {
Source s = getSource(r);
assertEquals(type, s.type());
@ -354,17 +357,17 @@ public class TestDocValues extends LuceneTestCase {
}
public void testFloats4() throws IOException {
runTestFloats(ValueType.FLOAT_32, 0.00001);
runTestFloats(Type.FLOAT_32, 0.00001);
}
private void runTestFloats(ValueType type, double delta) throws IOException {
private void runTestFloats(Type type, double delta) throws IOException {
Directory dir = newDirectory();
final Counter trackBytes = Counter.newCounter();
Writer w = Floats.getWriter(dir, "test", trackBytes, newIOContext(random), type);
final int NUM_VALUES = 777 + random.nextInt(777);;
final double[] values = new double[NUM_VALUES];
for (int i = 0; i < NUM_VALUES; i++) {
final double v = type == ValueType.FLOAT_32 ? random.nextFloat() : random
final double v = type == Type.FLOAT_32 ? random.nextFloat() : random
.nextDouble();
values[i] = v;
w.add(i, v);
@ -373,7 +376,7 @@ public class TestDocValues extends LuceneTestCase {
w.finish(NUM_VALUES + additionalValues);
assertEquals(0, trackBytes.get());
IndexDocValues r = Floats.getValues(dir, "test", NUM_VALUES + additionalValues, newIOContext(random), type);
DocValues r = Floats.getValues(dir, "test", NUM_VALUES + additionalValues, newIOContext(random), type);
for (int iter = 0; iter < 2; iter++) {
Source s = getSource(r);
for (int i = 0; i < NUM_VALUES; i++) {
@ -385,11 +388,11 @@ public class TestDocValues extends LuceneTestCase {
}
public void testFloats8() throws IOException {
runTestFloats(ValueType.FLOAT_64, 0.0);
runTestFloats(Type.FLOAT_64, 0.0);
}
private Source getSource(IndexDocValues values) throws IOException {
private Source getSource(DocValues values) throws IOException {
// getSource uses cache internally
switch(random.nextInt(5)) {
case 3:
@ -403,7 +406,7 @@ public class TestDocValues extends LuceneTestCase {
}
}
private SortedSource getSortedSource(IndexDocValues values) throws IOException {
private SortedSource getSortedSource(DocValues values) throws IOException {
return getSource(values).asSortedSource();
}

View File

@ -27,21 +27,23 @@ import java.util.List;
import org.apache.lucene.analysis.MockAnalyzer;
import org.apache.lucene.document.Document;
import org.apache.lucene.document.Field;
import org.apache.lucene.document.IndexDocValuesField;
import org.apache.lucene.document.DocValuesField;
import org.apache.lucene.document.StringField;
import org.apache.lucene.document.TextField;
import org.apache.lucene.index.CorruptIndexException;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.IndexWriter;
import org.apache.lucene.index.IndexWriterConfig;
import org.apache.lucene.index.LogDocMergePolicy;
import org.apache.lucene.index.LogMergePolicy;
import org.apache.lucene.index.MultiPerDocValues;
import org.apache.lucene.index.PerDocValues;
import org.apache.lucene.index.RandomIndexWriter;
import org.apache.lucene.index.Term;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.index.codecs.Codec;
import org.apache.lucene.index.codecs.PerDocValues;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.search.*;
import org.apache.lucene.store.Directory;
import org.apache.lucene.store.LockObtainFailedException;
@ -76,7 +78,7 @@ public class TestDocValuesIndexing extends LuceneTestCase {
IndexWriter writer = new IndexWriter(dir, writerConfig(false));
for (int i = 0; i < 5; i++) {
Document doc = new Document();
IndexDocValuesField valuesField = new IndexDocValuesField("docId");
DocValuesField valuesField = new DocValuesField("docId");
valuesField.setInt(i);
doc.add(valuesField);
doc.add(new TextField("docId", "" + i));
@ -102,7 +104,7 @@ public class TestDocValuesIndexing extends LuceneTestCase {
TopDocs search = searcher.search(query, 10);
assertEquals(5, search.totalHits);
ScoreDoc[] scoreDocs = search.scoreDocs;
IndexDocValues docValues = MultiPerDocValues.getPerDocs(reader).docValues("docId");
DocValues docValues = MultiPerDocValues.getPerDocs(reader).docValues("docId");
Source source = docValues.getSource();
for (int i = 0; i < scoreDocs.length; i++) {
assertEquals(i, scoreDocs[i].doc);
@ -130,10 +132,10 @@ public class TestDocValuesIndexing extends LuceneTestCase {
public void testAddIndexes() throws IOException {
int valuesPerIndex = 10;
List<ValueType> values = Arrays.asList(ValueType.values());
List<Type> values = Arrays.asList(Type.values());
Collections.shuffle(values, random);
ValueType first = values.get(0);
ValueType second = values.get(1);
Type first = values.get(0);
Type second = values.get(1);
// index first index
Directory d_1 = newDirectory();
IndexWriter w_1 = new IndexWriter(d_1, writerConfig(random.nextBoolean()));
@ -256,11 +258,11 @@ public class TestDocValuesIndexing extends LuceneTestCase {
Directory d = newDirectory();
IndexWriter w = new IndexWriter(d, cfg);
final int numValues = 50 + atLeast(10);
final List<ValueType> numVariantList = new ArrayList<ValueType>(NUMERICS);
final List<Type> numVariantList = new ArrayList<Type>(NUMERICS);
// run in random order to test if fill works correctly during merges
Collections.shuffle(numVariantList, random);
for (ValueType val : numVariantList) {
for (Type val : numVariantList) {
FixedBitSet deleted = indexValues(w, numValues, val, numVariantList,
withDeletions, 7);
List<Closeable> closeables = new ArrayList<Closeable>();
@ -277,7 +279,7 @@ public class TestDocValuesIndexing extends LuceneTestCase {
case FIXED_INTS_32:
case FIXED_INTS_64:
case VAR_INTS: {
IndexDocValues intsReader = getDocValues(r, val.name());
DocValues intsReader = getDocValues(r, val.name());
assertNotNull(intsReader);
Source ints = getSource(intsReader);
@ -298,7 +300,7 @@ public class TestDocValuesIndexing extends LuceneTestCase {
break;
case FLOAT_32:
case FLOAT_64: {
IndexDocValues floatReader = getDocValues(r, val.name());
DocValues floatReader = getDocValues(r, val.name());
assertNotNull(floatReader);
Source floats = getSource(floatReader);
for (int i = 0; i < base; i++) {
@ -333,11 +335,11 @@ public class TestDocValuesIndexing extends LuceneTestCase {
throws CorruptIndexException, LockObtainFailedException, IOException {
final Directory d = newDirectory();
IndexWriter w = new IndexWriter(d, cfg);
final List<ValueType> byteVariantList = new ArrayList<ValueType>(BYTES);
final List<Type> byteVariantList = new ArrayList<Type>(BYTES);
// run in random order to test if fill works correctly during merges
Collections.shuffle(byteVariantList, random);
final int numValues = 50 + atLeast(10);
for (ValueType byteIndexValue : byteVariantList) {
for (Type byteIndexValue : byteVariantList) {
List<Closeable> closeables = new ArrayList<Closeable>();
final int bytesSize = 1 + atLeast(50);
FixedBitSet deleted = indexValues(w, numValues, byteIndexValue,
@ -346,7 +348,7 @@ public class TestDocValuesIndexing extends LuceneTestCase {
assertEquals(0, r.numDeletedDocs());
final int numRemainingValues = numValues - deleted.cardinality();
final int base = r.numDocs() - numRemainingValues;
IndexDocValues bytesReader = getDocValues(r, byteIndexValue.name());
DocValues bytesReader = getDocValues(r, byteIndexValue.name());
assertNotNull("field " + byteIndexValue.name()
+ " returned null reader - maybe merged failed", bytesReader);
Source bytes = getSource(bytesReader);
@ -416,7 +418,7 @@ public class TestDocValuesIndexing extends LuceneTestCase {
d.close();
}
private IndexDocValues getDocValues(IndexReader reader, String field)
private DocValues getDocValues(IndexReader reader, String field)
throws IOException {
boolean singleSeg = reader.getSequentialSubReaders().length == 1;
PerDocValues perDoc = singleSeg ? reader.getSequentialSubReaders()[0].perDocValues()
@ -425,7 +427,7 @@ public class TestDocValuesIndexing extends LuceneTestCase {
case 0:
return perDoc.docValues(field);
case 1:
IndexDocValues docValues = perDoc.docValues(field);
DocValues docValues = perDoc.docValues(field);
if (docValues != null) {
return docValues;
}
@ -436,7 +438,7 @@ public class TestDocValuesIndexing extends LuceneTestCase {
throw new RuntimeException();
}
private Source getSource(IndexDocValues values) throws IOException {
private Source getSource(DocValues values) throws IOException {
// getSource uses cache internally
switch(random.nextInt(5)) {
case 3:
@ -451,24 +453,24 @@ public class TestDocValuesIndexing extends LuceneTestCase {
}
private static EnumSet<ValueType> BYTES = EnumSet.of(ValueType.BYTES_FIXED_DEREF,
ValueType.BYTES_FIXED_STRAIGHT, ValueType.BYTES_VAR_DEREF,
ValueType.BYTES_VAR_STRAIGHT, ValueType.BYTES_FIXED_SORTED, ValueType.BYTES_VAR_SORTED);
private static EnumSet<Type> BYTES = EnumSet.of(Type.BYTES_FIXED_DEREF,
Type.BYTES_FIXED_STRAIGHT, Type.BYTES_VAR_DEREF,
Type.BYTES_VAR_STRAIGHT, Type.BYTES_FIXED_SORTED, Type.BYTES_VAR_SORTED);
private static EnumSet<ValueType> NUMERICS = EnumSet.of(ValueType.VAR_INTS,
ValueType.FIXED_INTS_16, ValueType.FIXED_INTS_32,
ValueType.FIXED_INTS_64,
ValueType.FIXED_INTS_8,
ValueType.FLOAT_32,
ValueType.FLOAT_64);
private static EnumSet<Type> NUMERICS = EnumSet.of(Type.VAR_INTS,
Type.FIXED_INTS_16, Type.FIXED_INTS_32,
Type.FIXED_INTS_64,
Type.FIXED_INTS_8,
Type.FLOAT_32,
Type.FLOAT_64);
private FixedBitSet indexValues(IndexWriter w, int numValues, ValueType value,
List<ValueType> valueVarList, boolean withDeletions, int bytesSize)
private FixedBitSet indexValues(IndexWriter w, int numValues, Type value,
List<Type> valueVarList, boolean withDeletions, int bytesSize)
throws CorruptIndexException, IOException {
final boolean isNumeric = NUMERICS.contains(value);
FixedBitSet deleted = new FixedBitSet(numValues);
Document doc = new Document();
IndexDocValuesField valField = new IndexDocValuesField(value.name());
DocValuesField valField = new DocValuesField(value.name());
doc.add(valField);
final BytesRef bytesRef = new BytesRef();
@ -522,7 +524,7 @@ public class TestDocValuesIndexing extends LuceneTestCase {
if (i % 7 == 0) {
if (withDeletions && random.nextBoolean()) {
ValueType val = valueVarList.get(random.nextInt(1 + valueVarList
Type val = valueVarList.get(random.nextInt(1 + valueVarList
.indexOf(value)));
final int randInt = val == value ? random.nextInt(1 + i) : random
.nextInt(numValues);
@ -545,11 +547,11 @@ public class TestDocValuesIndexing extends LuceneTestCase {
return deleted;
}
public void testMultiValuedIndexDocValuesField() throws Exception {
public void testMultiValuedDocValuesField() throws Exception {
Directory d = newDirectory();
RandomIndexWriter w = new RandomIndexWriter(random, d);
Document doc = new Document();
IndexDocValuesField f = new IndexDocValuesField("field");
DocValuesField f = new DocValuesField("field");
f.setInt(17);
// Index doc values are single-valued so we should not
// be able to add same field more than once:
@ -577,12 +579,12 @@ public class TestDocValuesIndexing extends LuceneTestCase {
Directory d = newDirectory();
RandomIndexWriter w = new RandomIndexWriter(random, d);
Document doc = new Document();
IndexDocValuesField f = new IndexDocValuesField("field");
DocValuesField f = new DocValuesField("field");
f.setInt(17);
// Index doc values are single-valued so we should not
// be able to add same field more than once:
doc.add(f);
IndexDocValuesField f2 = new IndexDocValuesField("field");
DocValuesField f2 = new DocValuesField("field");
f2.setFloat(22.0);
doc.add(f2);
try {

View File

@ -7,18 +7,20 @@ import java.util.Random;
import org.apache.lucene.analysis.MockAnalyzer;
import org.apache.lucene.document.Document;
import org.apache.lucene.document.Field;
import org.apache.lucene.document.IndexDocValuesField;
import org.apache.lucene.document.DocValuesField;
import org.apache.lucene.document.TextField;
import org.apache.lucene.index.CorruptIndexException;
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.IndexReader.ReaderContext;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.index.IndexWriter;
import org.apache.lucene.index.IndexWriterConfig;
import org.apache.lucene.index.NoMergePolicy;
import org.apache.lucene.index.SlowMultiReaderWrapper;
import org.apache.lucene.index.codecs.Codec;
import org.apache.lucene.index.codecs.lucene40.values.BytesRefUtils;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.store.Directory;
import org.apache.lucene.util.BytesRef;
import org.apache.lucene.util.LuceneTestCase;
@ -47,22 +49,22 @@ public class TestTypePromotion extends LuceneTestCase {
assumeFalse("cannot work with preflex codec", Codec.getDefault().getName().equals("Lucene3x"));
}
private static EnumSet<ValueType> INTEGERS = EnumSet.of(ValueType.VAR_INTS,
ValueType.FIXED_INTS_16, ValueType.FIXED_INTS_32,
ValueType.FIXED_INTS_64, ValueType.FIXED_INTS_8);
private static EnumSet<Type> INTEGERS = EnumSet.of(Type.VAR_INTS,
Type.FIXED_INTS_16, Type.FIXED_INTS_32,
Type.FIXED_INTS_64, Type.FIXED_INTS_8);
private static EnumSet<ValueType> FLOATS = EnumSet.of(ValueType.FLOAT_32,
ValueType.FLOAT_64);
private static EnumSet<Type> FLOATS = EnumSet.of(Type.FLOAT_32,
Type.FLOAT_64);
private static EnumSet<ValueType> UNSORTED_BYTES = EnumSet.of(
ValueType.BYTES_FIXED_DEREF, ValueType.BYTES_FIXED_STRAIGHT,
ValueType.BYTES_VAR_STRAIGHT, ValueType.BYTES_VAR_DEREF);
private static EnumSet<Type> UNSORTED_BYTES = EnumSet.of(
Type.BYTES_FIXED_DEREF, Type.BYTES_FIXED_STRAIGHT,
Type.BYTES_VAR_STRAIGHT, Type.BYTES_VAR_DEREF);
private static EnumSet<ValueType> SORTED_BYTES = EnumSet.of(
ValueType.BYTES_FIXED_SORTED, ValueType.BYTES_VAR_SORTED);
private static EnumSet<Type> SORTED_BYTES = EnumSet.of(
Type.BYTES_FIXED_SORTED, Type.BYTES_VAR_SORTED);
public ValueType randomValueType(EnumSet<ValueType> typeEnum, Random random) {
ValueType[] array = typeEnum.toArray(new ValueType[0]);
public Type randomValueType(EnumSet<Type> typeEnum, Random random) {
Type[] array = typeEnum.toArray(new Type[0]);
return array[random.nextInt(array.length)];
}
@ -70,7 +72,7 @@ public class TestTypePromotion extends LuceneTestCase {
Int, Float, Byte
}
private void runTest(EnumSet<ValueType> types, TestType type)
private void runTest(EnumSet<Type> types, TestType type)
throws CorruptIndexException, IOException {
Directory dir = newDirectory();
IndexWriter writer = new IndexWriter(dir,
@ -79,11 +81,11 @@ public class TestTypePromotion extends LuceneTestCase {
int num_2 = atLeast(200);
int num_3 = atLeast(200);
long[] values = new long[num_1 + num_2 + num_3];
index(writer, new IndexDocValuesField("promote"),
index(writer, new DocValuesField("promote"),
randomValueType(types, random), values, 0, num_1);
writer.commit();
index(writer, new IndexDocValuesField("promote"),
index(writer, new DocValuesField("promote"),
randomValueType(types, random), values, num_1, num_2);
writer.commit();
@ -94,7 +96,7 @@ public class TestTypePromotion extends LuceneTestCase {
Directory dir_2 = newDirectory() ;
IndexWriter writer_2 = new IndexWriter(dir_2,
newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(random)));
index(writer_2, new IndexDocValuesField("promote"),
index(writer_2, new DocValuesField("promote"),
randomValueType(types, random), values, num_1 + num_2, num_3);
writer_2.commit();
writer_2.close();
@ -110,7 +112,7 @@ public class TestTypePromotion extends LuceneTestCase {
}
dir_2.close();
} else {
index(writer, new IndexDocValuesField("promote"),
index(writer, new DocValuesField("promote"),
randomValueType(types, random), values, num_1 + num_2, num_3);
}
@ -127,7 +129,7 @@ public class TestTypePromotion extends LuceneTestCase {
assertEquals(1, reader.getSequentialSubReaders().length);
ReaderContext topReaderContext = reader.getTopReaderContext();
ReaderContext[] children = topReaderContext.children();
IndexDocValues docValues = children[0].reader.docValues("promote");
DocValues docValues = children[0].reader.docValues("promote");
assertEquals(1, children.length);
Source directSource = docValues.getDirectSource();
for (int i = 0; i < values.length; i++) {
@ -172,8 +174,8 @@ public class TestTypePromotion extends LuceneTestCase {
reader.close();
}
public void index(IndexWriter writer, IndexDocValuesField valField,
ValueType valueType, long[] values, int offset, int num)
public void index(IndexWriter writer, DocValuesField valField,
Type valueType, long[] values, int offset, int num)
throws CorruptIndexException, IOException {
BytesRef ref = new BytesRef(new byte[] { 1, 2, 3, 4 });
for (int i = offset; i < offset + num; i++) {
@ -267,7 +269,7 @@ public class TestTypePromotion extends LuceneTestCase {
int num_1 = atLeast(200);
int num_2 = atLeast(200);
long[] values = new long[num_1 + num_2];
index(writer, new IndexDocValuesField("promote"),
index(writer, new DocValuesField("promote"),
randomValueType(INTEGERS, random), values, 0, num_1);
writer.commit();
@ -276,7 +278,7 @@ public class TestTypePromotion extends LuceneTestCase {
Directory dir_2 = newDirectory() ;
IndexWriter writer_2 = new IndexWriter(dir_2,
newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(random)));
index(writer_2, new IndexDocValuesField("promote"),
index(writer_2, new DocValuesField("promote"),
randomValueType(random.nextBoolean() ? UNSORTED_BYTES : SORTED_BYTES, random), values, num_1, num_2);
writer_2.commit();
writer_2.close();
@ -290,7 +292,7 @@ public class TestTypePromotion extends LuceneTestCase {
}
dir_2.close();
} else {
index(writer, new IndexDocValuesField("promote"),
index(writer, new DocValuesField("promote"),
randomValueType(random.nextBoolean() ? UNSORTED_BYTES : SORTED_BYTES, random), values, num_1, num_2);
writer.commit();
}
@ -307,10 +309,10 @@ public class TestTypePromotion extends LuceneTestCase {
assertEquals(1, reader.getSequentialSubReaders().length);
ReaderContext topReaderContext = reader.getTopReaderContext();
ReaderContext[] children = topReaderContext.children();
IndexDocValues docValues = children[0].reader.docValues("promote");
DocValues docValues = children[0].reader.docValues("promote");
assertNotNull(docValues);
assertValues(TestType.Byte, dir, values);
assertEquals(ValueType.BYTES_VAR_STRAIGHT, docValues.type());
assertEquals(Type.BYTES_VAR_STRAIGHT, docValues.type());
reader.close();
dir.close();
}

View File

@ -21,15 +21,15 @@ import java.io.IOException;
import org.apache.lucene.document.Document;
import org.apache.lucene.document.Field;
import org.apache.lucene.document.IndexDocValuesField;
import org.apache.lucene.document.DocValuesField;
import org.apache.lucene.document.TextField;
import org.apache.lucene.index.FieldInvertState;
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.RandomIndexWriter;
import org.apache.lucene.index.Term;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.IndexReader.AtomicReaderContext;
import org.apache.lucene.index.codecs.Codec;
import org.apache.lucene.index.values.IndexDocValues.Source;
import org.apache.lucene.search.similarities.DefaultSimilarityProvider;
import org.apache.lucene.search.similarities.Similarity;
import org.apache.lucene.search.similarities.SimilarityProvider;
@ -48,7 +48,7 @@ public class TestDocValuesScoring extends LuceneTestCase {
private static final float SCORE_EPSILON = 0.001f; /* for comparing floats */
public void testSimple() throws Exception {
assumeFalse("PreFlex codec cannot work with IndexDocValues!",
assumeFalse("PreFlex codec cannot work with DocValues!",
"Lucene3x".equals(Codec.getDefault().getName()));
Directory dir = newDirectory();
@ -56,7 +56,7 @@ public class TestDocValuesScoring extends LuceneTestCase {
Document doc = new Document();
Field field = newField("foo", "", TextField.TYPE_UNSTORED);
doc.add(field);
IndexDocValuesField dvField = new IndexDocValuesField("foo_boost");
DocValuesField dvField = new DocValuesField("foo_boost");
doc.add(dvField);
Field field2 = newField("bar", "", TextField.TYPE_UNSTORED);
doc.add(field2);

View File

@ -28,7 +28,7 @@ import org.apache.lucene.analysis.MockAnalyzer;
import org.apache.lucene.document.Document;
import org.apache.lucene.document.Field;
import org.apache.lucene.document.FieldType;
import org.apache.lucene.document.IndexDocValuesField;
import org.apache.lucene.document.DocValuesField;
import org.apache.lucene.document.StringField;
import org.apache.lucene.document.TextField;
import org.apache.lucene.index.CorruptIndexException;
@ -42,7 +42,7 @@ import org.apache.lucene.index.RandomIndexWriter;
import org.apache.lucene.index.SlowMultiReaderWrapper;
import org.apache.lucene.index.Term;
import org.apache.lucene.index.codecs.Codec;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues;
import org.apache.lucene.search.BooleanClause.Occur;
import org.apache.lucene.search.FieldValueHitQueue.Entry;
import org.apache.lucene.store.Directory;
@ -121,18 +121,18 @@ public class TestSort extends LuceneTestCase {
dirs.add(indexStore);
RandomIndexWriter writer = new RandomIndexWriter(random, indexStore, newIndexWriterConfig(TEST_VERSION_CURRENT, new MockAnalyzer(random)).setMergePolicy(newLogMergePolicy()));
final ValueType stringDVType;
final DocValues.Type stringDVType;
if (dvStringSorted) {
// Index sorted
stringDVType = random.nextBoolean() ? ValueType.BYTES_VAR_SORTED : ValueType.BYTES_FIXED_SORTED;
stringDVType = random.nextBoolean() ? DocValues.Type.BYTES_VAR_SORTED : DocValues.Type.BYTES_FIXED_SORTED;
} else {
// Index non-sorted
if (random.nextBoolean()) {
// Fixed
stringDVType = random.nextBoolean() ? ValueType.BYTES_FIXED_STRAIGHT : ValueType.BYTES_FIXED_DEREF;
stringDVType = random.nextBoolean() ? DocValues.Type.BYTES_FIXED_STRAIGHT : DocValues.Type.BYTES_FIXED_DEREF;
} else {
// Var
stringDVType = random.nextBoolean() ? ValueType.BYTES_VAR_STRAIGHT : ValueType.BYTES_VAR_DEREF;
stringDVType = random.nextBoolean() ? DocValues.Type.BYTES_VAR_STRAIGHT : DocValues.Type.BYTES_VAR_DEREF;
}
}
@ -148,21 +148,21 @@ public class TestSort extends LuceneTestCase {
if (data[i][2] != null) {
Field f = new StringField ("int", data[i][2]);
if (supportsDocValues) {
f = IndexDocValuesField.build(f, ValueType.VAR_INTS);
f = DocValuesField.build(f, DocValues.Type.VAR_INTS);
}
doc.add(f);
}
if (data[i][3] != null) {
Field f = new StringField ("float", data[i][3]);
if (supportsDocValues) {
f = IndexDocValuesField.build(f, ValueType.FLOAT_32);
f = DocValuesField.build(f, DocValues.Type.FLOAT_32);
}
doc.add(f);
}
if (data[i][4] != null) {
Field f = new StringField ("string", data[i][4]);
if (supportsDocValues) {
f = IndexDocValuesField.build(f, stringDVType);
f = DocValuesField.build(f, stringDVType);
}
doc.add(f);
}
@ -172,7 +172,7 @@ public class TestSort extends LuceneTestCase {
if (data[i][8] != null) {
Field f = new StringField ("double", data[i][8]);
if (supportsDocValues) {
f = IndexDocValuesField.build(f, ValueType.FLOAT_64);
f = DocValuesField.build(f, DocValues.Type.FLOAT_64);
}
doc.add(f);
}
@ -219,13 +219,13 @@ public class TestSort extends LuceneTestCase {
//doc.add (new Field ("contents", Integer.toString(i), Field.Store.NO, Field.Index.ANALYZED));
Field f = new StringField("string", num);
if (supportsDocValues) {
f = IndexDocValuesField.build(f, ValueType.BYTES_VAR_SORTED);
f = DocValuesField.build(f, DocValues.Type.BYTES_VAR_SORTED);
}
doc.add (f);
String num2 = getRandomCharString(getRandomNumber(1, 4), 48, 50);
f = new StringField ("string2", num2);
if (supportsDocValues) {
f = IndexDocValuesField.build(f, ValueType.BYTES_VAR_SORTED);
f = DocValuesField.build(f, DocValues.Type.BYTES_VAR_SORTED);
}
doc.add (f);
doc.add (new Field ("tracer2", num2, onlyStored));
@ -238,13 +238,13 @@ public class TestSort extends LuceneTestCase {
//doc.add (new Field ("contents", Integer.toString(i), Field.Store.NO, Field.Index.ANALYZED));
f = new StringField("string_fixed", numFixed);
if (supportsDocValues) {
f = IndexDocValuesField.build(f, ValueType.BYTES_FIXED_SORTED);
f = DocValuesField.build(f, DocValues.Type.BYTES_FIXED_SORTED);
}
doc.add (f);
String num2Fixed = getRandomCharString(fixedLen2, 48, 52);
f = new StringField ("string2_fixed", num2Fixed);
if (supportsDocValues) {
f = IndexDocValuesField.build(f, ValueType.BYTES_FIXED_SORTED);
f = DocValuesField.build(f, DocValues.Type.BYTES_FIXED_SORTED);
}
doc.add (f);
doc.add (new Field ("tracer2_fixed", num2Fixed, onlyStored));

View File

@ -17,4 +17,4 @@ LUCENE-3483: Move Function grouping collectors from Solr to
New features
LUCENE-3496: Support grouping by IndexDocValues. (Martijn van Groningen)
LUCENE-3496: Support grouping by DocValues. (Martijn van Groningen)

View File

@ -17,9 +17,9 @@ package org.apache.lucene.search.grouping.dv;
* limitations under the License.
*/
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.Type; // javadocs
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.search.*;
import org.apache.lucene.search.grouping.AbstractAllGroupHeadsCollector;
import org.apache.lucene.util.BytesRef;
@ -40,13 +40,13 @@ public abstract class DVAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsC
final String groupField;
final boolean diskResident;
final ValueType valueType;
final DocValues.Type valueType;
final BytesRef scratchBytesRef = new BytesRef();
IndexReader.AtomicReaderContext readerContext;
Scorer scorer;
DVAllGroupHeadsCollector(String groupField, ValueType valueType, int numberOfSorts, boolean diskResident) {
DVAllGroupHeadsCollector(String groupField, DocValues.Type valueType, int numberOfSorts, boolean diskResident) {
super(numberOfSorts);
this.groupField = groupField;
this.valueType = valueType;
@ -59,12 +59,12 @@ public abstract class DVAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsC
*
* @param groupField The field to group by
* @param sortWithinGroup The sort within each group
* @param type The {@link ValueType} which is used to select a concrete implementation.
* @param type The {@link Type} which is used to select a concrete implementation.
* @param diskResident Whether the values to group by should be disk resident
* @return an <code>AbstractAllGroupHeadsCollector</code> instance based on the supplied arguments
* @throws IOException If I/O related errors occur
*/
public static AbstractAllGroupHeadsCollector create(String groupField, Sort sortWithinGroup, ValueType type, boolean diskResident) throws IOException {
public static AbstractAllGroupHeadsCollector create(String groupField, Sort sortWithinGroup, DocValues.Type type, boolean diskResident) throws IOException {
switch (type) {
case VAR_INTS:
case FIXED_INTS_8:
@ -126,8 +126,8 @@ public abstract class DVAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsC
public void setNextReader(IndexReader.AtomicReaderContext readerContext) throws IOException {
this.readerContext = readerContext;
final IndexDocValues dv = readerContext.reader.docValues(groupField);
final IndexDocValues.Source dvSource;
final DocValues dv = readerContext.reader.docValues(groupField);
final DocValues.Source dvSource;
if (dv != null) {
dvSource = diskResident ? dv.getDirectSource() : dv.getSource();
} else {
@ -141,14 +141,14 @@ public abstract class DVAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsC
*
* @param source The idv source to be used by concrete implementations
*/
protected abstract void setDocValuesSources(IndexDocValues.Source source);
protected abstract void setDocValuesSources(DocValues.Source source);
/**
* @return The default source when no doc values are available.
* @param readerContext The current reader context
*/
protected IndexDocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return IndexDocValues.getDefaultSource(valueType);
protected DocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return DocValues.getDefaultSource(valueType);
}
// A general impl that works for any group sort.
@ -157,7 +157,7 @@ public abstract class DVAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsC
private final Sort sortWithinGroup;
private final Map<Comparable, GroupHead> groups;
GeneralAllGroupHeadsCollector(String groupField, ValueType valueType, Sort sortWithinGroup, boolean diskResident) throws IOException {
GeneralAllGroupHeadsCollector(String groupField, DocValues.Type valueType, Sort sortWithinGroup, boolean diskResident) throws IOException {
super(groupField, valueType, sortWithinGroup.getSort().length, diskResident);
this.sortWithinGroup = sortWithinGroup;
groups = new HashMap<Comparable, GroupHead>();
@ -211,9 +211,9 @@ public abstract class DVAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsC
static class SortedBR extends GeneralAllGroupHeadsCollector {
private IndexDocValues.SortedSource source;
private DocValues.SortedSource source;
SortedBR(String groupField, ValueType valueType, Sort sortWithinGroup, boolean diskResident) throws IOException {
SortedBR(String groupField, DocValues.Type valueType, Sort sortWithinGroup, boolean diskResident) throws IOException {
super(groupField, valueType, sortWithinGroup, diskResident);
}
@ -225,21 +225,21 @@ public abstract class DVAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsC
return BytesRef.deepCopyOf((BytesRef) value);
}
protected void setDocValuesSources(IndexDocValues.Source source) {
protected void setDocValuesSources(DocValues.Source source) {
this.source = source.asSortedSource();
}
@Override
protected IndexDocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return IndexDocValues.getDefaultSortedSource(valueType, readerContext.reader.maxDoc());
protected DocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return DocValues.getDefaultSortedSource(valueType, readerContext.reader.maxDoc());
}
}
static class BR extends GeneralAllGroupHeadsCollector {
private IndexDocValues.Source source;
private DocValues.Source source;
BR(String groupField, ValueType valueType, Sort sortWithinGroup, boolean diskResident) throws IOException {
BR(String groupField, DocValues.Type valueType, Sort sortWithinGroup, boolean diskResident) throws IOException {
super(groupField, valueType, sortWithinGroup, diskResident);
}
@ -251,7 +251,7 @@ public abstract class DVAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsC
return BytesRef.deepCopyOf((BytesRef) value);
}
protected void setDocValuesSources(IndexDocValues.Source source) {
protected void setDocValuesSources(DocValues.Source source) {
this.source = source;
}
@ -259,9 +259,9 @@ public abstract class DVAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsC
static class Lng extends GeneralAllGroupHeadsCollector {
private IndexDocValues.Source source;
private DocValues.Source source;
Lng(String groupField, ValueType valueType, Sort sortWithinGroup, boolean diskResident) throws IOException {
Lng(String groupField, DocValues.Type valueType, Sort sortWithinGroup, boolean diskResident) throws IOException {
super(groupField, valueType, sortWithinGroup, diskResident);
}
@ -273,16 +273,16 @@ public abstract class DVAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsC
return value;
}
protected void setDocValuesSources(IndexDocValues.Source source) {
protected void setDocValuesSources(DocValues.Source source) {
this.source = source;
}
}
static class Dbl extends GeneralAllGroupHeadsCollector {
private IndexDocValues.Source source;
private DocValues.Source source;
Dbl(String groupField, ValueType valueType, Sort sortWithinGroup, boolean diskResident) throws IOException {
Dbl(String groupField, DocValues.Type valueType, Sort sortWithinGroup, boolean diskResident) throws IOException {
super(groupField, valueType, sortWithinGroup, diskResident);
}
@ -294,7 +294,7 @@ public abstract class DVAllGroupHeadsCollector<GH extends AbstractAllGroupHeadsC
return value;
}
protected void setDocValuesSources(IndexDocValues.Source source) {
protected void setDocValuesSources(DocValues.Source source) {
this.source = source;
}

View File

@ -17,9 +17,9 @@ package org.apache.lucene.search.grouping.dv;
* limitations under the License.
*/
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.Type; // javadocs
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.search.grouping.AbstractAllGroupsCollector;
import org.apache.lucene.search.grouping.SentinelIntSet;
import org.apache.lucene.util.BytesRef;
@ -29,7 +29,7 @@ import java.util.*;
/**
* Implementation of {@link AbstractAllGroupsCollector} that groups documents based on
* {@link IndexDocValues} fields.
* {@link DocValues} fields.
*
* @lucene.experimental
*/
@ -39,20 +39,20 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
/**
* Expert: Constructs a {@link DVAllGroupsCollector}.
* Selects and constructs the most optimal all groups collector implementation for grouping by {@link IndexDocValues}.
* Selects and constructs the most optimal all groups collector implementation for grouping by {@link DocValues}.
*
*
* @param groupField The field to group by
* @param type The {@link ValueType} which is used to select a concrete implementation.
* @param type The {@link Type} which is used to select a concrete implementation.
* @param diskResident Whether the values to group by should be disk resident
* @param initialSize The initial allocation size of the
* internal int set and group list
* which should roughly match the total
* number of expected unique groups. Be aware that the
* heap usage is 4 bytes * initialSize. Not all concrete implementions use this!
* @return the most optimal all groups collector implementation for grouping by {@link IndexDocValues}
* @return the most optimal all groups collector implementation for grouping by {@link DocValues}
*/
public static DVAllGroupsCollector create(String groupField, ValueType type, boolean diskResident, int initialSize) {
public static DVAllGroupsCollector create(String groupField, DocValues.Type type, boolean diskResident, int initialSize) {
switch (type) {
case VAR_INTS:
case FIXED_INTS_8:
@ -78,25 +78,25 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
/**
* Constructs a {@link DVAllGroupsCollector}.
* Selects and constructs the most optimal all groups collector implementation for grouping by {@link IndexDocValues}.
* Selects and constructs the most optimal all groups collector implementation for grouping by {@link DocValues}.
* If implementations require an initial allocation size then this will be set to 128.
*
*
* @param groupField The field to group by
* @param type The {@link ValueType} which is used to select a concrete implementation.
* @param type The {@link Type} which is used to select a concrete implementation.
* @param diskResident Wether the values to group by should be disk resident
* @return the most optimal all groups collector implementation for grouping by {@link IndexDocValues}
* @return the most optimal all groups collector implementation for grouping by {@link DocValues}
*/
public static DVAllGroupsCollector create(String groupField, ValueType type, boolean diskResident) {
public static DVAllGroupsCollector create(String groupField, DocValues.Type type, boolean diskResident) {
return create(groupField, type, diskResident, DEFAULT_INITIAL_SIZE);
}
final String groupField;
final ValueType valueType;
final DocValues.Type valueType;
final boolean diskResident;
final Collection<GROUP_VALUE_TYPE> groups;
DVAllGroupsCollector(String groupField, ValueType valueType, boolean diskResident, Collection<GROUP_VALUE_TYPE> groups) {
DVAllGroupsCollector(String groupField, DocValues.Type valueType, boolean diskResident, Collection<GROUP_VALUE_TYPE> groups) {
this.groupField = groupField;
this.valueType = valueType;
this.diskResident = diskResident;
@ -105,8 +105,8 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
@Override
public void setNextReader(IndexReader.AtomicReaderContext readerContext) throws IOException {
final IndexDocValues dv = readerContext.reader.docValues(groupField);
final IndexDocValues.Source dvSource;
final DocValues dv = readerContext.reader.docValues(groupField);
final DocValues.Source dvSource;
if (dv != null) {
dvSource = diskResident ? dv.getDirectSource() : dv.getSource();
} else {
@ -121,21 +121,21 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
* @param source The idv source to be used by concrete implementations
* @param readerContext The current reader context
*/
protected abstract void setDocValuesSources(IndexDocValues.Source source, IndexReader.AtomicReaderContext readerContext);
protected abstract void setDocValuesSources(DocValues.Source source, IndexReader.AtomicReaderContext readerContext);
/**
* @return The default source when no doc values are available.
* @param readerContext The current reader context
*/
protected IndexDocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return IndexDocValues.getDefaultSource(valueType);
protected DocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return DocValues.getDefaultSource(valueType);
}
static class Lng extends DVAllGroupsCollector<Long> {
private IndexDocValues.Source source;
private DocValues.Source source;
Lng(String groupField, ValueType valueType, boolean diskResident) {
Lng(String groupField, DocValues.Type valueType, boolean diskResident) {
super(groupField, valueType, diskResident, new TreeSet<Long>());
}
@ -150,7 +150,7 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
return groups;
}
protected void setDocValuesSources(IndexDocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
protected void setDocValuesSources(DocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
this.source = source;
}
@ -158,9 +158,9 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
static class Dbl extends DVAllGroupsCollector<Double> {
private IndexDocValues.Source source;
private DocValues.Source source;
Dbl(String groupField, ValueType valueType, boolean diskResident) {
Dbl(String groupField, DocValues.Type valueType, boolean diskResident) {
super(groupField, valueType, diskResident, new TreeSet<Double>());
}
@ -175,7 +175,7 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
return groups;
}
protected void setDocValuesSources(IndexDocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
protected void setDocValuesSources(DocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
this.source = source;
}
@ -185,9 +185,9 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
private final BytesRef spare = new BytesRef();
private IndexDocValues.Source source;
private DocValues.Source source;
BR(String groupField, ValueType valueType, boolean diskResident) {
BR(String groupField, DocValues.Type valueType, boolean diskResident) {
super(groupField, valueType, diskResident, new TreeSet<BytesRef>());
}
@ -202,7 +202,7 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
return groups;
}
protected void setDocValuesSources(IndexDocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
protected void setDocValuesSources(DocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
this.source = source;
}
@ -213,9 +213,9 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
private final SentinelIntSet ordSet;
private final BytesRef spare = new BytesRef();
private IndexDocValues.SortedSource source;
private DocValues.SortedSource source;
SortedBR(String groupField, ValueType valueType, boolean diskResident, int initialSize) {
SortedBR(String groupField, DocValues.Type valueType, boolean diskResident, int initialSize) {
super(groupField, valueType, diskResident, new ArrayList<BytesRef>(initialSize));
ordSet = new SentinelIntSet(initialSize, -1);
}
@ -233,7 +233,7 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
return groups;
}
protected void setDocValuesSources(IndexDocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
protected void setDocValuesSources(DocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
this.source = source.asSortedSource();
ordSet.clear();
@ -246,8 +246,8 @@ public abstract class DVAllGroupsCollector<GROUP_VALUE_TYPE> extends AbstractAll
}
@Override
protected IndexDocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return IndexDocValues.getDefaultSortedSource(valueType, readerContext.reader.maxDoc());
protected DocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return DocValues.getDefaultSortedSource(valueType, readerContext.reader.maxDoc());
}
}

View File

@ -17,9 +17,9 @@ package org.apache.lucene.search.grouping.dv;
* limitations under the License.
*/
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.Type; // javadocs
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.search.Sort;
import org.apache.lucene.search.grouping.AbstractFirstPassGroupingCollector;
import org.apache.lucene.util.BytesRef;
@ -35,9 +35,9 @@ public abstract class DVFirstPassGroupingCollector<GROUP_VALUE_TYPE> extends Abs
final String groupField;
final boolean diskResident;
final ValueType valueType;
final DocValues.Type valueType;
public static DVFirstPassGroupingCollector create(Sort groupSort, int topNGroups, String groupField, ValueType type, boolean diskResident) throws IOException {
public static DVFirstPassGroupingCollector create(Sort groupSort, int topNGroups, String groupField, DocValues.Type type, boolean diskResident) throws IOException {
switch (type) {
case VAR_INTS:
case FIXED_INTS_8:
@ -61,7 +61,7 @@ public abstract class DVFirstPassGroupingCollector<GROUP_VALUE_TYPE> extends Abs
}
}
DVFirstPassGroupingCollector(Sort groupSort, int topNGroups, String groupField, boolean diskResident, ValueType valueType) throws IOException {
DVFirstPassGroupingCollector(Sort groupSort, int topNGroups, String groupField, boolean diskResident, DocValues.Type valueType) throws IOException {
super(groupSort, topNGroups);
this.groupField = groupField;
this.diskResident = diskResident;
@ -72,8 +72,8 @@ public abstract class DVFirstPassGroupingCollector<GROUP_VALUE_TYPE> extends Abs
public void setNextReader(IndexReader.AtomicReaderContext readerContext) throws IOException {
super.setNextReader(readerContext);
final IndexDocValues dv = readerContext.reader.docValues(groupField);
final IndexDocValues.Source dvSource;
final DocValues dv = readerContext.reader.docValues(groupField);
final DocValues.Source dvSource;
if (dv != null) {
dvSource = diskResident ? dv.getDirectSource() : dv.getSource();
} else {
@ -87,21 +87,21 @@ public abstract class DVFirstPassGroupingCollector<GROUP_VALUE_TYPE> extends Abs
*
* @param source The idv source to be used by concrete implementations
*/
protected abstract void setDocValuesSources(IndexDocValues.Source source);
protected abstract void setDocValuesSources(DocValues.Source source);
/**
* @return The default source when no doc values are available.
* @param readerContext The current reader context
*/
protected IndexDocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return IndexDocValues.getDefaultSource(valueType);
protected DocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return DocValues.getDefaultSource(valueType);
}
static class Lng extends DVFirstPassGroupingCollector<Long> {
private IndexDocValues.Source source;
private DocValues.Source source;
Lng(Sort groupSort, int topNGroups, String groupField, boolean diskResident, ValueType type) throws IOException {
Lng(Sort groupSort, int topNGroups, String groupField, boolean diskResident, DocValues.Type type) throws IOException {
super(groupSort, topNGroups, groupField, diskResident, type);
}
@ -113,16 +113,16 @@ public abstract class DVFirstPassGroupingCollector<GROUP_VALUE_TYPE> extends Abs
return groupValue;
}
protected void setDocValuesSources(IndexDocValues.Source source) {
protected void setDocValuesSources(DocValues.Source source) {
this.source = source;
}
}
static class Dbl extends DVFirstPassGroupingCollector<Double> {
private IndexDocValues.Source source;
private DocValues.Source source;
Dbl(Sort groupSort, int topNGroups, String groupField, boolean diskResident, ValueType type) throws IOException {
Dbl(Sort groupSort, int topNGroups, String groupField, boolean diskResident, DocValues.Type type) throws IOException {
super(groupSort, topNGroups, groupField, diskResident, type);
}
@ -134,17 +134,17 @@ public abstract class DVFirstPassGroupingCollector<GROUP_VALUE_TYPE> extends Abs
return groupValue;
}
protected void setDocValuesSources(IndexDocValues.Source source) {
protected void setDocValuesSources(DocValues.Source source) {
this.source = source;
}
}
static class BR extends DVFirstPassGroupingCollector<BytesRef> {
private IndexDocValues.Source source;
private DocValues.Source source;
private final BytesRef spare = new BytesRef();
BR(Sort groupSort, int topNGroups, String groupField, boolean diskResident, ValueType type) throws IOException {
BR(Sort groupSort, int topNGroups, String groupField, boolean diskResident, DocValues.Type type) throws IOException {
super(groupSort, topNGroups, groupField, diskResident, type);
}
@ -162,17 +162,17 @@ public abstract class DVFirstPassGroupingCollector<GROUP_VALUE_TYPE> extends Abs
}
@Override
protected void setDocValuesSources(IndexDocValues.Source source) {
protected void setDocValuesSources(DocValues.Source source) {
this.source = source;
}
}
static class SortedBR extends DVFirstPassGroupingCollector<BytesRef> {
private IndexDocValues.SortedSource sortedSource;
private DocValues.SortedSource sortedSource;
private final BytesRef spare = new BytesRef();
SortedBR(Sort groupSort, int topNGroups, String groupField, boolean diskResident, ValueType type) throws IOException {
SortedBR(Sort groupSort, int topNGroups, String groupField, boolean diskResident, DocValues.Type type) throws IOException {
super(groupSort, topNGroups, groupField, diskResident, type);
}
@ -192,13 +192,13 @@ public abstract class DVFirstPassGroupingCollector<GROUP_VALUE_TYPE> extends Abs
}
@Override
protected void setDocValuesSources(IndexDocValues.Source source) {
protected void setDocValuesSources(DocValues.Source source) {
this.sortedSource = source.asSortedSource();
}
@Override
protected IndexDocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return IndexDocValues.getDefaultSortedSource(valueType, readerContext.reader.maxDoc());
protected DocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return DocValues.getDefaultSortedSource(valueType, readerContext.reader.maxDoc());
}
}

View File

@ -17,9 +17,9 @@ package org.apache.lucene.search.grouping.dv;
* limitations under the License.
*/
import org.apache.lucene.index.DocValues;
import org.apache.lucene.index.DocValues.Type; // javadocs
import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.search.Sort;
import org.apache.lucene.search.grouping.AbstractSecondPassGroupingCollector;
import org.apache.lucene.search.grouping.SearchGroup;
@ -38,11 +38,11 @@ public abstract class DVSecondPassGroupingCollector<GROUP_VALUE> extends Abstrac
/**
* Constructs a {@link DVSecondPassGroupingCollector}.
* Selects and constructs the most optimal second pass collector implementation for grouping by {@link IndexDocValues}.
* Selects and constructs the most optimal second pass collector implementation for grouping by {@link DocValues}.
*
* @param groupField The field to group by
* @param diskResident Whether the values to group by should be disk resident
* @param type The {@link org.apache.lucene.index.values.ValueType} which is used to select a concrete implementation.
* @param type The {@link Type} which is used to select a concrete implementation.
* @param searchGroups The groups from the first phase search
* @param groupSort The sort used for the groups
* @param withinGroupSort The sort used for documents inside a group
@ -50,13 +50,13 @@ public abstract class DVSecondPassGroupingCollector<GROUP_VALUE> extends Abstrac
* @param getScores Whether to include scores for the documents inside a group
* @param getMaxScores Whether to keep track of the higest score per group
* @param fillSortFields Whether to include the sort values
* @return the most optimal second pass collector implementation for grouping by {@link IndexDocValues}
* @return the most optimal second pass collector implementation for grouping by {@link DocValues}
* @throws IOException If I/O related errors occur
*/
@SuppressWarnings("unchecked")
public static DVSecondPassGroupingCollector create(String groupField,
boolean diskResident,
ValueType type,
DocValues.Type type,
Collection<SearchGroup> searchGroups,
Sort groupSort,
Sort withinGroupSort,
@ -92,10 +92,10 @@ public abstract class DVSecondPassGroupingCollector<GROUP_VALUE> extends Abstrac
}
final String groupField;
final ValueType valueType;
final DocValues.Type valueType;
final boolean diskResident;
DVSecondPassGroupingCollector(String groupField, ValueType valueType, boolean diskResident, Collection<SearchGroup<GROUP_VALUE>> searchGroups, Sort groupSort, Sort withinGroupSort, int maxDocsPerGroup, boolean getScores, boolean getMaxScores, boolean fillSortFields) throws IOException {
DVSecondPassGroupingCollector(String groupField, DocValues.Type valueType, boolean diskResident, Collection<SearchGroup<GROUP_VALUE>> searchGroups, Sort groupSort, Sort withinGroupSort, int maxDocsPerGroup, boolean getScores, boolean getMaxScores, boolean fillSortFields) throws IOException {
super(searchGroups, groupSort, withinGroupSort, maxDocsPerGroup, getScores, getMaxScores, fillSortFields);
this.groupField = groupField;
this.valueType = valueType;
@ -106,8 +106,8 @@ public abstract class DVSecondPassGroupingCollector<GROUP_VALUE> extends Abstrac
public void setNextReader(IndexReader.AtomicReaderContext readerContext) throws IOException {
super.setNextReader(readerContext);
final IndexDocValues dv = readerContext.reader.docValues(groupField);
final IndexDocValues.Source dvSource;
final DocValues dv = readerContext.reader.docValues(groupField);
final DocValues.Source dvSource;
if (dv != null) {
dvSource = diskResident ? dv.getDirectSource() : dv.getSource();
} else {
@ -122,21 +122,21 @@ public abstract class DVSecondPassGroupingCollector<GROUP_VALUE> extends Abstrac
* @param source The idv source to be used by concrete implementations
* @param readerContext The current reader context
*/
protected abstract void setDocValuesSources(IndexDocValues.Source source, IndexReader.AtomicReaderContext readerContext);
protected abstract void setDocValuesSources(DocValues.Source source, IndexReader.AtomicReaderContext readerContext);
/**
* @return The default source when no doc values are available.
* @param readerContext The current reader context
*/
protected IndexDocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return IndexDocValues.getDefaultSource(valueType);
protected DocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return DocValues.getDefaultSource(valueType);
}
static class Lng extends DVSecondPassGroupingCollector<Long> {
private IndexDocValues.Source source;
private DocValues.Source source;
Lng(String groupField, ValueType valueType, boolean diskResident, Collection<SearchGroup<Long>> searchGroups, Sort groupSort, Sort withinGroupSort, int maxDocsPerGroup, boolean getScores, boolean getMaxScores, boolean fillSortFields) throws IOException {
Lng(String groupField, DocValues.Type valueType, boolean diskResident, Collection<SearchGroup<Long>> searchGroups, Sort groupSort, Sort withinGroupSort, int maxDocsPerGroup, boolean getScores, boolean getMaxScores, boolean fillSortFields) throws IOException {
super(groupField, valueType, diskResident, searchGroups, groupSort, withinGroupSort, maxDocsPerGroup, getScores, getMaxScores, fillSortFields);
}
@ -144,16 +144,16 @@ public abstract class DVSecondPassGroupingCollector<GROUP_VALUE> extends Abstrac
return groupMap.get(source.getInt(doc));
}
protected void setDocValuesSources(IndexDocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
protected void setDocValuesSources(DocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
this.source = source;
}
}
static class Dbl extends DVSecondPassGroupingCollector<Double> {
private IndexDocValues.Source source;
private DocValues.Source source;
Dbl(String groupField, ValueType valueType, boolean diskResident, Collection<SearchGroup<Double>> searchGroups, Sort groupSort, Sort withinGroupSort, int maxDocsPerGroup, boolean getScores, boolean getMaxScores, boolean fillSortFields) throws IOException {
Dbl(String groupField, DocValues.Type valueType, boolean diskResident, Collection<SearchGroup<Double>> searchGroups, Sort groupSort, Sort withinGroupSort, int maxDocsPerGroup, boolean getScores, boolean getMaxScores, boolean fillSortFields) throws IOException {
super(groupField, valueType, diskResident, searchGroups, groupSort, withinGroupSort, maxDocsPerGroup, getScores, getMaxScores, fillSortFields);
}
@ -161,17 +161,17 @@ public abstract class DVSecondPassGroupingCollector<GROUP_VALUE> extends Abstrac
return groupMap.get(source.getFloat(doc));
}
protected void setDocValuesSources(IndexDocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
protected void setDocValuesSources(DocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
this.source = source;
}
}
static class BR extends DVSecondPassGroupingCollector<BytesRef> {
private IndexDocValues.Source source;
private DocValues.Source source;
private final BytesRef spare = new BytesRef();
BR(String groupField, ValueType valueType, boolean diskResident, Collection<SearchGroup<BytesRef>> searchGroups, Sort groupSort, Sort withinGroupSort, int maxDocsPerGroup, boolean getScores, boolean getMaxScores, boolean fillSortFields) throws IOException {
BR(String groupField, DocValues.Type valueType, boolean diskResident, Collection<SearchGroup<BytesRef>> searchGroups, Sort groupSort, Sort withinGroupSort, int maxDocsPerGroup, boolean getScores, boolean getMaxScores, boolean fillSortFields) throws IOException {
super(groupField, valueType, diskResident, searchGroups, groupSort, withinGroupSort, maxDocsPerGroup, getScores, getMaxScores, fillSortFields);
}
@ -180,7 +180,7 @@ public abstract class DVSecondPassGroupingCollector<GROUP_VALUE> extends Abstrac
}
@Override
protected void setDocValuesSources(IndexDocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
protected void setDocValuesSources(DocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
this.source = source;
}
@ -188,12 +188,12 @@ public abstract class DVSecondPassGroupingCollector<GROUP_VALUE> extends Abstrac
static class SortedBR extends DVSecondPassGroupingCollector<BytesRef> {
private IndexDocValues.SortedSource source;
private DocValues.SortedSource source;
private final BytesRef spare = new BytesRef();
private final SentinelIntSet ordSet;
@SuppressWarnings("unchecked")
SortedBR(String groupField, ValueType valueType, boolean diskResident, Collection<SearchGroup<BytesRef>> searchGroups, Sort groupSort, Sort withinGroupSort, int maxDocsPerGroup, boolean getScores, boolean getMaxScores, boolean fillSortFields) throws IOException {
SortedBR(String groupField, DocValues.Type valueType, boolean diskResident, Collection<SearchGroup<BytesRef>> searchGroups, Sort groupSort, Sort withinGroupSort, int maxDocsPerGroup, boolean getScores, boolean getMaxScores, boolean fillSortFields) throws IOException {
super(groupField, valueType, diskResident, searchGroups, groupSort, withinGroupSort, maxDocsPerGroup, getScores, getMaxScores, fillSortFields);
ordSet = new SentinelIntSet(groupMap.size(), -1);
groupDocs = (SearchGroupDocs<BytesRef>[]) new SearchGroupDocs[ordSet.keys.length];
@ -209,7 +209,7 @@ public abstract class DVSecondPassGroupingCollector<GROUP_VALUE> extends Abstrac
}
@Override
protected void setDocValuesSources(IndexDocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
protected void setDocValuesSources(DocValues.Source source, IndexReader.AtomicReaderContext readerContext) {
this.source = source.asSortedSource();
ordSet.clear();
@ -222,8 +222,8 @@ public abstract class DVSecondPassGroupingCollector<GROUP_VALUE> extends Abstrac
}
@Override
protected IndexDocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return IndexDocValues.getDefaultSortedSource(valueType, readerContext.reader.maxDoc());
protected DocValues.Source getDefaultSource(IndexReader.AtomicReaderContext readerContext) {
return DocValues.getDefaultSortedSource(valueType, readerContext.reader.maxDoc());
}
}

View File

@ -186,14 +186,14 @@ fields, <code>FieldCache</code>, etc.).
</p>
<p>
There are also IndexDocValues based implementations available for the group collectors. There are factory methods
available for creating idv based instances. A typical example using idv based grouping collectors:
There are also DocValues based implementations available for the group collectors. There are factory methods
available for creating dv based instances. A typical example using dv based grouping collectors:
</p>
<pre class="prettyprint">
boolean diskResident = true; // Whether values should fetched directly from disk by passing the Java heap space.
AbstractFirstPassGroupingCollector c1 = DVFirstPassGroupingCollector.create(
groupSort, groupOffset+topNGroups, "author", ValueType.BYTES_VAR_SORTED, diskResident
groupSort, groupOffset+topNGroups, "author", DocValues.Type.BYTES_VAR_SORTED, diskResident
);
s.search(new TermQuery(new Term("content", searchTerm)), c1);
@ -209,7 +209,7 @@ fields, <code>FieldCache</code>, etc.).
boolean getMaxScores = true;
boolean fillFields = true;
AbstractSecondPassGroupingCollector&lt;BytesRef&gt; c2 = DVSecondPassGroupingCollector.create(
"author", diskResident, ValueType.BYTES_VAR_SORTED, topGroups, groupSort, docSort,
"author", diskResident, DocValues.Type.BYTES_VAR_SORTED, topGroups, groupSort, docSort,
docOffset+docsPerGroup, getScores, getMaxScores, fillFields
);

View File

@ -23,7 +23,7 @@ import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.RandomIndexWriter;
import org.apache.lucene.index.SlowMultiReaderWrapper;
import org.apache.lucene.index.Term;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.queries.function.ValueSource;
import org.apache.lucene.queries.function.valuesource.BytesRefFieldSource;
import org.apache.lucene.search.*;
@ -41,8 +41,8 @@ import java.util.*;
public class AllGroupHeadsCollectorTest extends LuceneTestCase {
private static final ValueType[] vts = new ValueType[]{
ValueType.BYTES_VAR_DEREF, ValueType.BYTES_VAR_STRAIGHT, ValueType.BYTES_VAR_SORTED
private static final Type[] vts = new Type[]{
Type.BYTES_VAR_DEREF, Type.BYTES_VAR_STRAIGHT, Type.BYTES_VAR_SORTED
};
public void testBasic() throws Exception {
@ -54,7 +54,7 @@ public class AllGroupHeadsCollectorTest extends LuceneTestCase {
newIndexWriterConfig(TEST_VERSION_CURRENT,
new MockAnalyzer(random)).setMergePolicy(newLogMergePolicy()));
boolean canUseIDV = !"Lucene3x".equals(w.w.getConfig().getCodec().getName());
ValueType valueType = vts[random.nextInt(vts.length)];
Type valueType = vts[random.nextInt(vts.length)];
// 0
Document doc = new Document();
@ -203,15 +203,15 @@ public class AllGroupHeadsCollectorTest extends LuceneTestCase {
new MockAnalyzer(random)));
boolean preFlex = "Lucene3x".equals(w.w.getConfig().getCodec().getName());
boolean canUseIDV = !preFlex;
ValueType valueType = vts[random.nextInt(vts.length)];
Type valueType = vts[random.nextInt(vts.length)];
Document doc = new Document();
Document docNoGroup = new Document();
Field group = newField("group", "", StringField.TYPE_UNSTORED);
doc.add(group);
IndexDocValuesField valuesField = null;
DocValuesField valuesField = null;
if (canUseIDV) {
valuesField = new IndexDocValuesField("group");
valuesField = new DocValuesField("group");
doc.add(valuesField);
}
Field sort1 = newField("sort1", "", StringField.TYPE_UNSTORED);
@ -506,7 +506,7 @@ public class AllGroupHeadsCollectorTest extends LuceneTestCase {
};
}
private AbstractAllGroupHeadsCollector createRandomCollector(String groupField, Sort sortWithinGroup, boolean canUseIDV, ValueType valueType) throws IOException {
private AbstractAllGroupHeadsCollector createRandomCollector(String groupField, Sort sortWithinGroup, boolean canUseIDV, Type valueType) throws IOException {
AbstractAllGroupHeadsCollector collector;
if (random.nextBoolean()) {
ValueSource vs = new BytesRefFieldSource(groupField);
@ -525,10 +525,10 @@ public class AllGroupHeadsCollectorTest extends LuceneTestCase {
return collector;
}
private void addGroupField(Document doc, String groupField, String value, boolean canUseIDV, ValueType valueType) {
private void addGroupField(Document doc, String groupField, String value, boolean canUseIDV, Type valueType) {
doc.add(new Field(groupField, value, TextField.TYPE_STORED));
if (canUseIDV) {
IndexDocValuesField valuesField = new IndexDocValuesField(groupField);
DocValuesField valuesField = new DocValuesField(groupField);
valuesField.setBytes(new BytesRef(value), valueType);
doc.add(valuesField);
}

View File

@ -21,7 +21,7 @@ import org.apache.lucene.analysis.MockAnalyzer;
import org.apache.lucene.document.*;
import org.apache.lucene.index.RandomIndexWriter;
import org.apache.lucene.index.Term;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.queries.function.ValueSource;
import org.apache.lucene.queries.function.valuesource.BytesRefFieldSource;
import org.apache.lucene.search.IndexSearcher;
@ -123,8 +123,8 @@ public class AllGroupsCollectorTest extends LuceneTestCase {
private void addGroupField(Document doc, String groupField, String value, boolean canUseIDV) {
doc.add(new Field(groupField, value, TextField.TYPE_STORED));
if (canUseIDV) {
IndexDocValuesField valuesField = new IndexDocValuesField(groupField);
valuesField.setBytes(new BytesRef(value), ValueType.BYTES_VAR_SORTED);
DocValuesField valuesField = new DocValuesField(groupField);
valuesField.setBytes(new BytesRef(value), Type.BYTES_VAR_SORTED);
doc.add(valuesField);
}
}
@ -133,7 +133,7 @@ public class AllGroupsCollectorTest extends LuceneTestCase {
AbstractAllGroupsCollector selected;
if (random.nextBoolean() && canUseIDV) {
boolean diskResident = random.nextBoolean();
selected = DVAllGroupsCollector.create(groupField, ValueType.BYTES_VAR_SORTED, diskResident);
selected = DVAllGroupsCollector.create(groupField, Type.BYTES_VAR_SORTED, diskResident);
} else if (random.nextBoolean()) {
selected = new TermAllGroupsCollector(groupField);
} else {

View File

@ -24,7 +24,7 @@ import org.apache.lucene.index.IndexReader;
import org.apache.lucene.index.RandomIndexWriter;
import org.apache.lucene.index.SlowMultiReaderWrapper;
import org.apache.lucene.index.Term;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.queries.function.ValueSource;
import org.apache.lucene.queries.function.valuesource.BytesRefFieldSource;
import org.apache.lucene.search.*;
@ -171,8 +171,8 @@ public class TestGrouping extends LuceneTestCase {
private void addGroupField(Document doc, String groupField, String value, boolean canUseIDV) {
doc.add(new Field(groupField, value, TextField.TYPE_STORED));
if (canUseIDV) {
IndexDocValuesField valuesField = new IndexDocValuesField(groupField);
valuesField.setBytes(new BytesRef(value), ValueType.BYTES_VAR_SORTED);
DocValuesField valuesField = new DocValuesField(groupField);
valuesField.setBytes(new BytesRef(value), Type.BYTES_VAR_SORTED);
doc.add(valuesField);
}
}
@ -181,7 +181,7 @@ public class TestGrouping extends LuceneTestCase {
AbstractFirstPassGroupingCollector selected;
if (canUseIDV && random.nextBoolean()) {
boolean diskResident = random.nextBoolean();
selected = DVFirstPassGroupingCollector.create(groupSort, topDocs, groupField, ValueType.BYTES_VAR_SORTED, diskResident);
selected = DVFirstPassGroupingCollector.create(groupSort, topDocs, groupField, Type.BYTES_VAR_SORTED, diskResident);
} else if (random.nextBoolean()) {
ValueSource vs = new BytesRefFieldSource(groupField);
selected = new FunctionFirstPassGroupingCollector(vs, new HashMap(), groupSort, topDocs);
@ -197,7 +197,7 @@ public class TestGrouping extends LuceneTestCase {
private AbstractFirstPassGroupingCollector createFirstPassCollector(String groupField, Sort groupSort, int topDocs, AbstractFirstPassGroupingCollector firstPassGroupingCollector) throws IOException {
if (DVFirstPassGroupingCollector.class.isAssignableFrom(firstPassGroupingCollector.getClass())) {
boolean diskResident = random.nextBoolean();
return DVFirstPassGroupingCollector.create(groupSort, topDocs, groupField, ValueType.BYTES_VAR_SORTED, diskResident);
return DVFirstPassGroupingCollector.create(groupSort, topDocs, groupField, Type.BYTES_VAR_SORTED, diskResident);
} else if (TermFirstPassGroupingCollector.class.isAssignableFrom(firstPassGroupingCollector.getClass())) {
ValueSource vs = new BytesRefFieldSource(groupField);
return new FunctionFirstPassGroupingCollector(vs, new HashMap(), groupSort, topDocs);
@ -220,7 +220,7 @@ public class TestGrouping extends LuceneTestCase {
if (DVFirstPassGroupingCollector.class.isAssignableFrom(firstPassGroupingCollector.getClass())) {
boolean diskResident = random.nextBoolean();
Collection<SearchGroup> searchGroups = firstPassGroupingCollector.getTopGroups(groupOffset, fillSortFields);
return DVSecondPassGroupingCollector.create(groupField, diskResident, ValueType.BYTES_VAR_SORTED, searchGroups, groupSort, sortWithinGroup, maxDocsPerGroup, getScores, getMaxScores, fillSortFields);
return DVSecondPassGroupingCollector.create(groupField, diskResident, Type.BYTES_VAR_SORTED, searchGroups, groupSort, sortWithinGroup, maxDocsPerGroup, getScores, getMaxScores, fillSortFields);
} else if (TermFirstPassGroupingCollector.class.isAssignableFrom(firstPassGroupingCollector.getClass())) {
Collection<SearchGroup<BytesRef>> searchGroups = firstPassGroupingCollector.getTopGroups(groupOffset, fillSortFields);
return new TermSecondPassGroupingCollector(groupField, searchGroups, groupSort, sortWithinGroup, maxDocsPerGroup , getScores, getMaxScores, fillSortFields);
@ -244,7 +244,7 @@ public class TestGrouping extends LuceneTestCase {
boolean fillSortFields) throws IOException {
if (DVFirstPassGroupingCollector.class.isAssignableFrom(firstPassGroupingCollector.getClass())) {
boolean diskResident = random.nextBoolean();
return DVSecondPassGroupingCollector.create(groupField, diskResident, ValueType.BYTES_VAR_SORTED, (Collection) searchGroups, groupSort, sortWithinGroup, maxDocsPerGroup, getScores, getMaxScores, fillSortFields);
return DVSecondPassGroupingCollector.create(groupField, diskResident, Type.BYTES_VAR_SORTED, (Collection) searchGroups, groupSort, sortWithinGroup, maxDocsPerGroup, getScores, getMaxScores, fillSortFields);
} else if (firstPassGroupingCollector.getClass().isAssignableFrom(TermFirstPassGroupingCollector.class)) {
return new TermSecondPassGroupingCollector(groupField, searchGroups, groupSort, sortWithinGroup, maxDocsPerGroup , getScores, getMaxScores, fillSortFields);
} else {
@ -274,7 +274,7 @@ public class TestGrouping extends LuceneTestCase {
return new TermAllGroupsCollector(groupField);
} else if (firstPassGroupingCollector.getClass().isAssignableFrom(DVFirstPassGroupingCollector.class)) {
boolean diskResident = random.nextBoolean();
return DVAllGroupsCollector.create(groupField, ValueType.BYTES_VAR_SORTED, diskResident);
return DVAllGroupsCollector.create(groupField, Type.BYTES_VAR_SORTED, diskResident);
} else {
ValueSource vs = new BytesRefFieldSource(groupField);
return new FunctionAllGroupsCollector(vs, new HashMap());
@ -705,7 +705,7 @@ public class TestGrouping extends LuceneTestCase {
Document doc = new Document();
Document docNoGroup = new Document();
IndexDocValuesField idvGroupField = new IndexDocValuesField("group");
DocValuesField idvGroupField = new DocValuesField("group");
if (canUseIDV) {
doc.add(idvGroupField);
}
@ -747,7 +747,7 @@ public class TestGrouping extends LuceneTestCase {
if (groupDoc.group != null) {
group.setValue(groupDoc.group.utf8ToString());
if (canUseIDV) {
idvGroupField.setBytes(BytesRef.deepCopyOf(groupDoc.group), ValueType.BYTES_VAR_SORTED);
idvGroupField.setBytes(BytesRef.deepCopyOf(groupDoc.group), Type.BYTES_VAR_SORTED);
}
}
sort1.setValue(groupDoc.sort1.utf8ToString());

View File

@ -19,16 +19,16 @@ package org.apache.lucene.queries.function.valuesource;
import java.io.IOException;
import java.util.Map;
import org.apache.lucene.index.DocValues.Source;
import org.apache.lucene.index.DocValues.Type;
import org.apache.lucene.index.IndexReader.AtomicReaderContext;
import org.apache.lucene.index.values.IndexDocValues;
import org.apache.lucene.index.values.ValueType;
import org.apache.lucene.queries.function.DocValues;
import org.apache.lucene.queries.function.ValueSource;
/**
* Expert: obtains numeric field values from a {@link IndexDocValues} field.
* Expert: obtains numeric field values from a {@link DocValues} field.
* This {@link ValueSource} is compatible with all numerical
* {@link IndexDocValues}
* {@link DocValues}
*
* @lucene.experimental
*
@ -43,9 +43,9 @@ public class NumericIndexDocValueSource extends ValueSource {
@Override
public DocValues getValues(Map context, AtomicReaderContext readerContext) throws IOException {
final IndexDocValues.Source source = readerContext.reader.docValues(field)
final Source source = readerContext.reader.docValues(field)
.getSource();
ValueType type = source.type();
Type type = source.type();
switch (type) {
case FLOAT_32:
case FLOAT_64: