1 package org.apache.lucene.facet.taxonomy.lucene;
3 import java.io.IOException;
5 import org.apache.lucene.index.CorruptIndexException;
6 import org.apache.lucene.index.IndexReader;
7 import org.apache.lucene.index.Term;
8 import org.apache.lucene.index.TermPositions;
10 import org.apache.lucene.facet.taxonomy.TaxonomyReader;
13 * Licensed to the Apache Software Foundation (ASF) under one or more
14 * contributor license agreements. See the NOTICE file distributed with
15 * this work for additional information regarding copyright ownership.
16 * The ASF licenses this file to You under the Apache License, Version 2.0
17 * (the "License"); you may not use this file except in compliance with
18 * the License. You may obtain a copy of the License at
20 * http://www.apache.org/licenses/LICENSE-2.0
22 * Unless required by applicable law or agreed to in writing, software
23 * distributed under the License is distributed on an "AS IS" BASIS,
24 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
25 * See the License for the specific language governing permissions and
26 * limitations under the License.
29 // getParent() needs to be extremely efficient, to the point that we need
30 // to fetch all the data in advance into memory, and answer these calls
31 // from memory. Currently we use a large integer array, which is
32 // initialized when the taxonomy is opened, and potentially enlarged
33 // when it is refresh()ed.
35 * @lucene.experimental
39 // These arrays are not syncrhonized. Rather, the reference to the array
40 // is volatile, and the only writing operation (refreshPrefetchArrays)
41 // simply creates a new array and replaces the reference. The volatility
42 // of the reference ensures the correct atomic replacement and its
43 // visibility properties (the content of the array is visible when the
44 // new reference is visible).
45 private volatile int prefetchParentOrdinal[] = null;
47 public int[] getArray() {
48 return prefetchParentOrdinal;
52 * refreshPrefetch() refreshes the parent array. Initially, it fills the
53 * array from the positions of an appropriate posting list. If called during
54 * a refresh(), when the arrays already exist, only values for new documents
55 * (those beyond the last one in the array) are read from the positions and
56 * added to the arrays (that are appropriately enlarged). We assume (and
57 * this is indeed a correct assumption in our case) that existing categories
58 * are never modified or deleted.
60 void refresh(IndexReader indexReader) throws IOException {
61 // Note that it is not necessary for us to obtain the read lock.
62 // The reason is that we are only called from refresh() (precluding
63 // another concurrent writer) or from the constructor (when no method
65 // The write lock is also not held during the following code, meaning
66 // that reads *can* happen while this code is running. The "volatile"
67 // property of the prefetchParentOrdinal and prefetchDepth array
68 // references ensure the correct visibility property of the assignment
69 // but other than that, we do *not* guarantee that a reader will not
70 // use an old version of one of these arrays (or both) while a refresh
71 // is going on. But we find this acceptable - until a refresh has
72 // finished, the reader should not expect to see new information
73 // (and the old information is the same in the old and new versions).
75 int num = indexReader.maxDoc();
76 if (prefetchParentOrdinal==null) {
77 prefetchParentOrdinal = new int[num];
78 // Starting Lucene 2.9, following the change LUCENE-1542, we can
79 // no longer reliably read the parent "-1" (see comment in
80 // LuceneTaxonomyWriter.SinglePositionTokenStream). We have no way
81 // to fix this in indexing without breaking backward-compatibility
82 // with existing indexes, so what we'll do instead is just
83 // hard-code the parent of ordinal 0 to be -1, and assume (as is
84 // indeed the case) that no other parent can be -1.
86 prefetchParentOrdinal[0] = TaxonomyReader.INVALID_ORDINAL;
90 first = prefetchParentOrdinal.length;
92 return; // nothing to do - no category was added
94 // In Java 6, we could just do Arrays.copyOf()...
95 int[] newarray = new int[num];
96 System.arraycopy(prefetchParentOrdinal, 0, newarray, 0,
97 prefetchParentOrdinal.length);
98 prefetchParentOrdinal = newarray;
101 // Read the new part of the parents array from the positions:
102 TermPositions positions = indexReader.termPositions(
103 new Term(Consts.FIELD_PAYLOADS, Consts.PAYLOAD_PARENT));
105 if (!positions.skipTo(first) && first < num) {
106 throw new CorruptIndexException("Missing parent data for category " + first);
108 for (int i=first; i<num; i++) {
109 // Note that we know positions.doc() >= i (this is an
110 // invariant kept throughout this loop)
111 if (positions.doc()==i) {
112 if (positions.freq() == 0) { // shouldn't happen
113 throw new CorruptIndexException(
114 "Missing parent data for category "+i);
117 // TODO (Facet): keep a local (non-volatile) copy of the prefetchParentOrdinal
118 // reference, because access to volatile reference is slower (?).
119 // Note: The positions we get here are one less than the position
120 // increment we added originally, so we get here the right numbers:
121 prefetchParentOrdinal[i] = positions.nextPosition();
123 if (!positions.next()) {
125 throw new CorruptIndexException(
126 "Missing parent data for category "+(i+1));
130 } else { // this shouldn't happen
131 throw new CorruptIndexException(
132 "Missing parent data for category "+i);
136 positions.close(); // to be on the safe side.
141 * add() is used in LuceneTaxonomyWriter, not in LuceneTaxonomyReader.
142 * It is only called from a synchronized method, so it is not reentrant,
143 * and also doesn't need to worry about reads happening at the same time.
145 * NOTE: add() and refresh() CANNOT be used together. If you call add(),
146 * this changes the arrays and refresh() can no longer be used.
148 void add(int ordinal, int parentOrdinal) throws IOException {
149 if (ordinal >= prefetchParentOrdinal.length) {
150 // grow the array, if necessary.
151 // In Java 6, we could just do Arrays.copyOf()...
152 int[] newarray = new int[ordinal*2+1];
153 System.arraycopy(prefetchParentOrdinal, 0, newarray, 0,
154 prefetchParentOrdinal.length);
155 prefetchParentOrdinal = newarray;
157 prefetchParentOrdinal[ordinal] = parentOrdinal;