LUCENE-4498: pulse docFreq=1 in 4.1 codec

git-svn-id: https://svn.apache.org/repos/asf/lucene/dev/trunk@1401284 13f79535-47bb-0310-9956-ffa450edef68

lucene/CHANGES.txt

@@ -108,6 +108,10 @@ Optimizations
 * LUCENE-4497: Don't write PosVIntCount to the positions file in 
   Lucene41PostingsFormat, as its always totalTermFreq % BLOCK_SIZE. (Robert Muir)
 
+* LUCENE-4498: In Lucene41PostingsFormat, when a term appears in only one document, 
+  Instead of writing a file pointer to a VIntBlock containing the doc id, just 
+  write the doc id.  (Mike McCandless, Robert Muir)
+
 Build
 
 * LUCENE-4451: Memory leak per unique thread caused by 

lucene/core/src/java/org/apache/lucene/codecs/lucene41/Lucene41PostingsFormat.java

@@ -127,10 +127,10 @@ import org.apache.lucene.util.packed.PackedInts;
  *
  * <ul>
  *   <li>Postings Metadata --&gt; Header, PackedBlockSize</li>
- *   <li>Term Metadata --&gt; DocFPDelta, PosFPDelta?, PosVIntBlockFPDelta?, PayFPDelta?, 
+ *   <li>Term Metadata --&gt; (DocFPDelta|SingletonDocID), PosFPDelta?, PosVIntBlockFPDelta?, PayFPDelta?, 
  *                            SkipFPDelta?</li>
  *   <li>Header, --&gt; {@link CodecUtil#writeHeader CodecHeader}</li>
- *   <li>PackedBlockSize --&gt; {@link DataOutput#writeVInt VInt}</li>
+ *   <li>PackedBlockSize, SingletonDocID --&gt; {@link DataOutput#writeVInt VInt}</li>
  *   <li>DocFPDelta, PosFPDelta, PayFPDelta, PosVIntBlockFPDelta, SkipFPDelta --&gt; {@link DataOutput#writeVLong VLong}</li>
  * </ul>
  * <p>Notes:</p>
@@ -162,6 +162,9 @@ import org.apache.lucene.util.packed.PackedInts;
  *        file. In particular, it is the length of the TermFreq data.
  *        SkipDelta is only stored if DocFreq is not smaller than SkipMinimum
  *        (i.e. 8 in Lucene41PostingsFormat).</li>
+ *    <li>SingletonDocID is an optimization when a term only appears in one document. In this case, instead
+ *        of writing a file pointer to the .doc file (DocFPDelta), and then a VIntBlock at that location, the 
+ *        single document ID is written to the term dictionary.</li>
  * </ul>
  * </dd>
  * </dl>
@@ -277,7 +280,7 @@ import org.apache.lucene.util.packed.PackedInts;
  *   <li>VIntBlock --&gt; &lt;PositionDelta[, PayloadLength?], PayloadData?, 
  *                        OffsetDelta?, OffsetLength?&gt;<sup>PosVIntCount</sup>
  *   <li>PackedPosDeltaBlock --&gt; {@link PackedInts PackedInts}</li>
- *   <li>PosVIntCount, PositionDelta, OffsetDelta, OffsetLength --&gt; 
+ *   <li>PositionDelta, OffsetDelta, OffsetLength --&gt; 
  *       {@link DataOutput#writeVInt VInt}</li>
  *   <li>PayloadData --&gt; {@link DataOutput#writeByte byte}<sup>PayLength</sup></li>
  * </ul>

lucene/core/src/java/org/apache/lucene/codecs/lucene41/Lucene41PostingsReader.java

@@ -148,6 +148,9 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
     long payStartFP;
     long skipOffset;
     long lastPosBlockOffset;
+    // docid when there is a single pulsed posting, otherwise -1
+    // freq is always implicitly totalTermFreq in this case.
+    int singletonDocID;
 
     // Only used by the "primary" TermState -- clones don't
     // copy this (basically they are "transient"):
@@ -170,6 +173,7 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
       payStartFP = other.payStartFP;
       lastPosBlockOffset = other.lastPosBlockOffset;
       skipOffset = other.skipOffset;
+      singletonDocID = other.singletonDocID;
 
       // Do not copy bytes, bytesReader (else TermState is
       // very heavy, ie drags around the entire block's
@@ -179,7 +183,7 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
 
     @Override
     public String toString() {
-      return super.toString() + " docStartFP=" + docStartFP + " posStartFP=" + posStartFP + " payStartFP=" + payStartFP + " lastPosBlockOffset=" + lastPosBlockOffset;
+      return super.toString() + " docStartFP=" + docStartFP + " posStartFP=" + posStartFP + " payStartFP=" + payStartFP + " lastPosBlockOffset=" + lastPosBlockOffset + " singletonDocID=" + singletonDocID;
     }
   }
 
@@ -223,7 +227,13 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
 
     final DataInput in = termState.bytesReader;
     if (isFirstTerm) {
+      if (termState.docFreq == 1) {
+        termState.singletonDocID = in.readVInt();
+        termState.docStartFP = 0;
+      } else {
+        termState.singletonDocID = -1;
         termState.docStartFP = in.readVLong();
+      }
       if (fieldHasPositions) {
         termState.posStartFP = in.readVLong();
         if (termState.totalTermFreq > BLOCK_SIZE) {
@@ -238,7 +248,12 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
         }
       }
     } else {
+      if (termState.docFreq == 1) {
+        termState.singletonDocID = in.readVInt();
+      } else {
+        termState.singletonDocID = -1;
         termState.docStartFP += in.readVLong();
+      }
       if (fieldHasPositions) {
         termState.posStartFP += in.readVLong();
         if (termState.totalTermFreq > BLOCK_SIZE) {
@@ -327,13 +342,14 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
 
     final IndexInput startDocIn;
 
-    final IndexInput docIn;
+    IndexInput docIn;
     final boolean indexHasFreq;
     final boolean indexHasPos;
     final boolean indexHasOffsets;
     final boolean indexHasPayloads;
 
     private int docFreq;                              // number of docs in this posting list
+    private long totalTermFreq;                       // sum of freqs in this posting list (or docFreq when omitted)
     private int docUpto;                              // how many docs we've read
     private int doc;                                  // doc we last read
     private int accum;                                // accumulator for doc deltas
@@ -354,10 +370,11 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
     private Bits liveDocs;
     
     private boolean needsFreq; // true if the caller actually needs frequencies
+    private int singletonDocID; // docid when there is a single pulsed posting, otherwise -1
 
     public BlockDocsEnum(FieldInfo fieldInfo) throws IOException {
       this.startDocIn = Lucene41PostingsReader.this.docIn;
-      this.docIn = startDocIn.clone();
+      this.docIn = null;
       indexHasFreq = fieldInfo.getIndexOptions().compareTo(IndexOptions.DOCS_AND_FREQS) >= 0;
       indexHasPos = fieldInfo.getIndexOptions().compareTo(IndexOptions.DOCS_AND_FREQS_AND_POSITIONS) >= 0;
       indexHasOffsets = fieldInfo.getIndexOptions().compareTo(IndexOptions.DOCS_AND_FREQS_AND_POSITIONS_AND_OFFSETS) >= 0;
@@ -378,9 +395,17 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
       //   System.out.println("  FPR.reset: termState=" + termState);
       // }
       docFreq = termState.docFreq;
+      totalTermFreq = indexHasFreq ? termState.totalTermFreq : docFreq;
       docTermStartFP = termState.docStartFP;
-      docIn.seek(docTermStartFP);
       skipOffset = termState.skipOffset;
+      singletonDocID = termState.singletonDocID;
+      if (docFreq > 1) {
+        if (docIn == null) {
+          // lazy init
+          docIn = startDocIn.clone();
+        }
+        docIn.seek(docTermStartFP);
+      }
 
       doc = -1;
       this.needsFreq = (flags & DocsEnum.FLAG_FREQS) != 0;
@@ -425,6 +450,9 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
             forUtil.skipBlock(docIn); // skip over freqs
           }
         }
+      } else if (docFreq == 1) {
+        docDeltaBuffer[0] = singletonDocID;
+        freqBuffer[0] = (int) totalTermFreq;
       } else {
         // Read vInts:
         // if (DEBUG) {
@@ -590,7 +618,7 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
 
     final IndexInput startDocIn;
 
-    final IndexInput docIn;
+    IndexInput docIn;
     final IndexInput posIn;
 
     final boolean indexHasOffsets;
@@ -635,10 +663,11 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
     private int nextSkipDoc;
 
     private Bits liveDocs;
+    private int singletonDocID; // docid when there is a single pulsed posting, otherwise -1
     
     public BlockDocsAndPositionsEnum(FieldInfo fieldInfo) throws IOException {
       this.startDocIn = Lucene41PostingsReader.this.docIn;
-      this.docIn = startDocIn.clone();
+      this.docIn = null;
       this.posIn = Lucene41PostingsReader.this.posIn.clone();
       encoded = new byte[MAX_ENCODED_SIZE];
       indexHasOffsets = fieldInfo.getIndexOptions().compareTo(IndexOptions.DOCS_AND_FREQS_AND_POSITIONS_AND_OFFSETS) >= 0;
@@ -660,9 +689,16 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
       docTermStartFP = termState.docStartFP;
       posTermStartFP = termState.posStartFP;
       payTermStartFP = termState.payStartFP;
-      docIn.seek(docTermStartFP);
       skipOffset = termState.skipOffset;
       totalTermFreq = termState.totalTermFreq;
+      singletonDocID = termState.singletonDocID;
+      if (docFreq > 1) {
+        if (docIn == null) {
+          // lazy init
+          docIn = startDocIn.clone();
+        }
+        docIn.seek(docTermStartFP);
+      }
       posPendingFP = posTermStartFP;
       posPendingCount = 0;
       if (termState.totalTermFreq < BLOCK_SIZE) {
@@ -705,6 +741,9 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
         //   System.out.println("    fill freq block from fp=" + docIn.getFilePointer());
         // }
         forUtil.readBlock(docIn, encoded, freqBuffer);
+      } else if (docFreq == 1) {
+        docDeltaBuffer[0] = singletonDocID;
+        freqBuffer[0] = (int) totalTermFreq;
       } else {
         // Read vInts:
         // if (DEBUG) {
@@ -1002,7 +1041,7 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
 
     final IndexInput startDocIn;
 
-    final IndexInput docIn;
+    IndexInput docIn;
     final IndexInput posIn;
     final IndexInput payIn;
     final BytesRef payload;
@@ -1056,10 +1095,11 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
     
     private boolean needsOffsets; // true if we actually need offsets
     private boolean needsPayloads; // true if we actually need payloads
+    private int singletonDocID; // docid when there is a single pulsed posting, otherwise -1
     
     public EverythingEnum(FieldInfo fieldInfo) throws IOException {
       this.startDocIn = Lucene41PostingsReader.this.docIn;
-      this.docIn = startDocIn.clone();
+      this.docIn = null;
       this.posIn = Lucene41PostingsReader.this.posIn.clone();
       this.payIn = Lucene41PostingsReader.this.payIn.clone();
       encoded = new byte[MAX_ENCODED_SIZE];
@@ -1101,9 +1141,16 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
       docTermStartFP = termState.docStartFP;
       posTermStartFP = termState.posStartFP;
       payTermStartFP = termState.payStartFP;
-      docIn.seek(docTermStartFP);
       skipOffset = termState.skipOffset;
       totalTermFreq = termState.totalTermFreq;
+      singletonDocID = termState.singletonDocID;
+      if (docFreq > 1) {
+        if (docIn == null) {
+          // lazy init
+          docIn = startDocIn.clone();
+        }
+        docIn.seek(docTermStartFP);
+      }
       posPendingFP = posTermStartFP;
       payPendingFP = payTermStartFP;
       posPendingCount = 0;
@@ -1150,6 +1197,9 @@ public final class Lucene41PostingsReader extends PostingsReaderBase {
         //   System.out.println("    fill freq block from fp=" + docIn.getFilePointer());
         // }
         forUtil.readBlock(docIn, encoded, freqBuffer);
+      } else if (docFreq == 1) {
+        docDeltaBuffer[0] = singletonDocID;
+        freqBuffer[0] = (int) totalTermFreq;
       } else {
         // if (DEBUG) {
         //   System.out.println("    fill last vInt doc block from fp=" + docIn.getFilePointer());

lucene/core/src/java/org/apache/lucene/codecs/lucene41/Lucene41PostingsWriter.java

@@ -354,13 +354,15 @@ public final class Lucene41PostingsWriter extends PostingsWriterBase {
     public final long payStartFP;
     public final long skipOffset;
     public final long lastPosBlockOffset;
+    public final int singletonDocID;
 
-    public PendingTerm(long docStartFP, long posStartFP, long payStartFP, long skipOffset, long lastPosBlockOffset) {
+    public PendingTerm(long docStartFP, long posStartFP, long payStartFP, long skipOffset, long lastPosBlockOffset, int singletonDocID) {
       this.docStartFP = docStartFP;
       this.posStartFP = posStartFP;
       this.payStartFP = payStartFP;
       this.skipOffset = skipOffset;
       this.lastPosBlockOffset = lastPosBlockOffset;
+      this.singletonDocID = singletonDocID;
     }
   }
 
@@ -385,6 +387,13 @@ public final class Lucene41PostingsWriter extends PostingsWriterBase {
     //   }
     // }
     
+    // docFreq == 1, don't write the single docid/freq to a separate file along with a pointer to it.
+    final int singletonDocID;
+    if (stats.docFreq == 1) {
+      // pulse the singleton docid into the term dictionary, freq is implicitly totalTermFreq
+      singletonDocID = docDeltaBuffer[0];
+    } else {
+      singletonDocID = -1;
       // vInt encode the remaining doc deltas and freqs:
       for(int i=0;i<docBufferUpto;i++) {
         final int docDelta = docDeltaBuffer[i];
@@ -398,6 +407,7 @@ public final class Lucene41PostingsWriter extends PostingsWriterBase {
           docOut.writeVInt(freq);
         }
       }
+    }
 
     final long lastPosBlockOffset;
 
@@ -507,7 +517,7 @@ public final class Lucene41PostingsWriter extends PostingsWriterBase {
     //   System.out.println("  payStartFP=" + payStartFP);
     // }
 
-    pendingTerms.add(new PendingTerm(docTermStartFP, posTermStartFP, payStartFP, skipOffset, lastPosBlockOffset));
+    pendingTerms.add(new PendingTerm(docTermStartFP, posTermStartFP, payStartFP, skipOffset, lastPosBlockOffset, singletonDocID));
     docBufferUpto = 0;
     posBufferUpto = 0;
     lastDocID = 0;
@@ -535,8 +545,12 @@ public final class Lucene41PostingsWriter extends PostingsWriterBase {
     for(int idx=limit-count; idx<limit; idx++) {
       PendingTerm term = pendingTerms.get(idx);
 
+      if (term.singletonDocID == -1) {
         bytesWriter.writeVLong(term.docStartFP - lastDocStartFP);
         lastDocStartFP = term.docStartFP;
+      } else {
+        bytesWriter.writeVInt(term.singletonDocID);
+      }
 
       if (fieldHasPositions) {
         bytesWriter.writeVLong(term.posStartFP - lastPosStartFP);