LUCENE-6940: Speed up MUST_NOT clauses.

git-svn-id: https://svn.apache.org/repos/asf/lucene/dev/trunk@1722443 13f79535-47bb-0310-9956-ffa450edef68

lucene/CHANGES.txt

@@ -156,6 +156,9 @@ Optimizations
 * LUCENE-6944: BooleanWeight no longer creates sub-scorers if BS1 is not
   applicable. (Adrien Grand)
 
+* LUCENE-6940: MUST_NOT clauses execute faster, especially when they are sparse.
+  (Adrien Grand)
+
 Bug Fixes
 
 * LUCENE-6918: LRUQueryCache.onDocIdSetEviction is only called when at least

lucene/core/src/java/org/apache/lucene/search/BooleanScorer.java

@@ -37,35 +37,6 @@ final class BooleanScorer extends BulkScorer {
   static final int SET_SIZE = 1 << (SHIFT - 6);
   static final int SET_MASK = SET_SIZE - 1;
 
-  private static BulkScorer disableScoring(final BulkScorer scorer) {
-    return new BulkScorer() {
-
-      @Override
-      public int score(final LeafCollector collector, Bits acceptDocs, int min, int max) throws IOException {
-        final LeafCollector noScoreCollector = new LeafCollector() {
-          FakeScorer fake = new FakeScorer();
-
-          @Override
-          public void setScorer(Scorer scorer) throws IOException {
-            collector.setScorer(fake);
-          }
-
-          @Override
-          public void collect(int doc) throws IOException {
-            fake.doc = doc;
-            collector.collect(doc);
-          }
-        };
-        return scorer.score(noScoreCollector, acceptDocs, min, max);
-      }
-
-      @Override
-      public long cost() {
-        return scorer.cost();
-      }
-    };
-  }
-
   static class Bucket {
     double score;
     int freq;
@@ -193,7 +164,7 @@ final class BooleanScorer extends BulkScorer {
       if (needsScores == false) {
         // OrCollector calls score() all the time so we have to explicitly
         // disable scoring in order to avoid decoding useless norms
-        scorer = disableScoring(scorer);
+        scorer = BooleanWeight.disableScoring(scorer);
       }
       final BulkScorerAndDoc evicted = tail.insertWithOverflow(new BulkScorerAndDoc(scorer));
       if (evicted != null) {

lucene/core/src/java/org/apache/lucene/search/BooleanWeight.java

@@ -28,6 +28,7 @@ import org.apache.lucene.index.LeafReaderContext;
 import org.apache.lucene.index.Term;
 import org.apache.lucene.search.BooleanClause.Occur;
 import org.apache.lucene.search.similarities.Similarity;
+import org.apache.lucene.util.Bits;
 
 /**
  * Expert: the Weight for BooleanQuery, used to
@@ -187,27 +188,45 @@ final class BooleanWeight extends Weight {
     }
   }
 
-  /** Try to build a boolean scorer for this weight. Returns null if {@link BooleanScorer}
-   *  cannot be used. */
-  // pkg-private for forcing use of BooleanScorer in tests
-  BulkScorer booleanScorer(LeafReaderContext context) throws IOException {
-    if (query.getClauses(Occur.MUST).isEmpty() == false
-        || query.getClauses(Occur.FILTER).isEmpty() == false) {
-      // TODO: there are some cases where BooleanScorer
-      // would handle conjunctions faster than
-      // BooleanScorer2...
-      return null;
-    } else if (query.getClauses(Occur.MUST_NOT).isEmpty() == false) {
-      // TODO: there are some cases where BooleanScorer could do this faster
-      return null;
-    }
+  static BulkScorer disableScoring(final BulkScorer scorer) {
+    return new BulkScorer() {
 
+      @Override
+      public int score(final LeafCollector collector, Bits acceptDocs, int min, int max) throws IOException {
+        final LeafCollector noScoreCollector = new LeafCollector() {
+          FakeScorer fake = new FakeScorer();
+
+          @Override
+          public void setScorer(Scorer scorer) throws IOException {
+            collector.setScorer(fake);
+          }
+
+          @Override
+          public void collect(int doc) throws IOException {
+            fake.doc = doc;
+            collector.collect(doc);
+          }
+        };
+        return scorer.score(noScoreCollector, acceptDocs, min, max);
+      }
+
+      @Override
+      public long cost() {
+        return scorer.cost();
+      }
+    };
+  }
+
+  // Return a BulkScorer for the optional clauses only,
+  // or null if it is not applicable
+  // pkg-private for forcing use of BooleanScorer in tests
+  BulkScorer optionalBulkScorer(LeafReaderContext context) throws IOException {
     List<BulkScorer> optional = new ArrayList<BulkScorer>();
     Iterator<BooleanClause> cIter = query.iterator();
     for (Weight w  : weights) {
       BooleanClause c =  cIter.next();
       if (c.getOccur() != Occur.SHOULD) {
-        throw new AssertionError();
+        continue;
       }
       BulkScorer subScorer = w.bulkScorer(context);
 
@@ -236,10 +255,50 @@ final class BooleanWeight extends Weight {
     return new BooleanScorer(this, disableCoord, maxCoord, optional, Math.max(1, query.getMinimumNumberShouldMatch()), needsScores);
   }
 
-  @Override
-  public BulkScorer bulkScorer(LeafReaderContext context) throws IOException {
-    final BulkScorer bulkScorer = booleanScorer(context);
-    if (bulkScorer != null) { // BooleanScorer is applicable
+  // Return a BulkScorer for the required clauses only,
+  // or null if it is not applicable
+  private BulkScorer requiredBulkScorer(LeafReaderContext context) throws IOException {
+    BulkScorer scorer = null;
+
+    Iterator<BooleanClause> cIter = query.iterator();
+    for (Weight w  : weights) {
+      BooleanClause c =  cIter.next();
+      if (c.isRequired() == false) {
+        continue;
+      }
+      if (scorer != null) {
+        // we don't have a BulkScorer for conjunctions
+        return null;
+      }
+      scorer = w.bulkScorer(context);
+      if (scorer == null) {
+        // no matches
+        return null;
+      }
+      if (c.isScoring() == false) {
+        if (needsScores) {
+          scorer = disableScoring(scorer);
+        }
+      } else {
+        assert maxCoord == 1;
+      }
+    }
+    return scorer;
+  }
+
+  /** Try to build a boolean scorer for this weight. Returns null if {@link BooleanScorer}
+   *  cannot be used. */
+  BulkScorer booleanScorer(LeafReaderContext context) throws IOException {
+    final int numOptionalClauses = query.getClauses(Occur.SHOULD).size();
+    final int numRequiredClauses = query.getClauses(Occur.MUST).size() + query.getClauses(Occur.FILTER).size();
+    
+    BulkScorer positiveScorer;
+    if (numRequiredClauses == 0) {
+      positiveScorer = optionalBulkScorer(context);
+      if (positiveScorer == null) {
+        return null;
+      }
+
       // TODO: what is the right heuristic here?
       final long costThreshold;
       if (query.getMinimumNumberShouldMatch() <= 1) {
@@ -257,11 +316,59 @@ final class BooleanWeight extends Weight {
         costThreshold = context.reader().maxDoc() / 3;
       }
 
-      if (bulkScorer.cost() > costThreshold) {
-        return bulkScorer;
+      if (positiveScorer.cost() < costThreshold) {
+        return null;
+      }
+
+    } else if (numRequiredClauses == 1
+        && numOptionalClauses == 0
+        && query.getMinimumNumberShouldMatch() == 0) {
+      positiveScorer = requiredBulkScorer(context);
+    } else {
+      // TODO: there are some cases where BooleanScorer
+      // would handle conjunctions faster than
+      // BooleanScorer2...
+      return null;
+    }
+
+    if (positiveScorer == null) {
+      return null;
+    }
+
+    List<Scorer> prohibited = new ArrayList<>();
+    Iterator<BooleanClause> cIter = query.iterator();
+    for (Weight w  : weights) {
+      BooleanClause c =  cIter.next();
+      if (c.isProhibited()) {
+        Scorer scorer = w.scorer(context);
+        if (scorer != null) {
+          prohibited.add(scorer);
+        }
       }
     }
-    return super.bulkScorer(context);
+
+    if (prohibited.isEmpty()) {
+      return positiveScorer;
+    } else {
+      Scorer prohibitedScorer = opt(prohibited, 1, true);
+      if (prohibitedScorer.twoPhaseIterator() != null) {
+        // ReqExclBulkScorer can't deal efficiently with two-phased prohibited clauses
+        return null;
+      }
+      return new ReqExclBulkScorer(positiveScorer, prohibitedScorer.iterator());
+    }
+  }
+
+  @Override
+  public BulkScorer bulkScorer(LeafReaderContext context) throws IOException {
+    final BulkScorer bulkScorer = booleanScorer(context);
+    if (bulkScorer != null) {
+      // bulk scoring is applicable, use it
+      return bulkScorer;
+    } else {
+      // use a Scorer-based impl (BS2)
+      return super.bulkScorer(context);
+    }
   }
 
   @Override

lucene/core/src/java/org/apache/lucene/search/ReqExclBulkScorer.java

@@ -0,0 +1,64 @@
+package org.apache.lucene.search;
+
+/*
+ * 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.util.Bits;
+
+final class ReqExclBulkScorer extends BulkScorer {
+
+  private final BulkScorer req;
+  private final DocIdSetIterator excl;
+
+  ReqExclBulkScorer(BulkScorer req, DocIdSetIterator excl) {
+    this.req = req;
+    this.excl = excl;
+  }
+
+  @Override
+  public int score(LeafCollector collector, Bits acceptDocs, int min, int max) throws IOException {
+    int upTo = min;
+    int exclDoc = excl.docID();
+
+    while (upTo < max) {
+      if (exclDoc < upTo) {
+        exclDoc = excl.advance(upTo);
+      }
+      if (exclDoc == upTo) {
+        // upTo is excluded so we can consider that we scored up to upTo+1
+        upTo += 1;
+        exclDoc = excl.nextDoc();
+      } else {
+        upTo = req.score(collector, acceptDocs, upTo, Math.min(exclDoc, max));
+      }
+    }
+
+    if (upTo == max) {
+      upTo = req.score(collector, acceptDocs, upTo, upTo);
+    }
+
+    return upTo;
+  }
+
+  @Override
+  public long cost() {
+    return req.cost();
+  }
+
+}

lucene/core/src/test/org/apache/lucene/search/TestBooleanScorer.java

@@ -196,6 +196,60 @@ public class TestBooleanScorer extends LuceneTestCase {
     dir.close();
   }
 
+  public void testOptimizeProhibitedClauses() throws IOException {
+    Directory dir = newDirectory();
+    RandomIndexWriter w = new RandomIndexWriter(random(), dir);
+    Document doc = new Document();
+    doc.add(new StringField("foo", "bar", Store.NO));
+    doc.add(new StringField("foo", "baz", Store.NO));
+    w.addDocument(doc);
+    doc = new Document();
+    doc.add(new StringField("foo", "baz", Store.NO));
+    w.addDocument(doc);
+    w.forceMerge(1);
+    IndexReader reader = w.getReader();
+    IndexSearcher searcher = new IndexSearcher(reader);
+    searcher.setQueryCache(null); // so that weights are not wrapped
+    final LeafReaderContext ctx = reader.leaves().get(0);
+
+    Query query = new BooleanQuery.Builder()
+      .add(new TermQuery(new Term("foo", "baz")), Occur.SHOULD)
+      .add(new TermQuery(new Term("foo", "bar")), Occur.MUST_NOT)
+      .build();
+    Weight weight = searcher.createNormalizedWeight(query, true);
+    BulkScorer scorer = ((BooleanWeight) weight).booleanScorer(ctx);
+    assertTrue(scorer instanceof ReqExclBulkScorer);
+
+    query = new BooleanQuery.Builder()
+        .add(new TermQuery(new Term("foo", "baz")), Occur.SHOULD)
+        .add(new MatchAllDocsQuery(), Occur.SHOULD)
+        .add(new TermQuery(new Term("foo", "bar")), Occur.MUST_NOT)
+        .build();
+    weight = searcher.createNormalizedWeight(query, true);
+    scorer = ((BooleanWeight) weight).booleanScorer(ctx);
+    assertTrue(scorer instanceof ReqExclBulkScorer);
+
+    query = new BooleanQuery.Builder()
+        .add(new TermQuery(new Term("foo", "baz")), Occur.MUST)
+        .add(new TermQuery(new Term("foo", "bar")), Occur.MUST_NOT)
+        .build();
+    weight = searcher.createNormalizedWeight(query, true);
+    scorer = ((BooleanWeight) weight).booleanScorer(ctx);
+    assertTrue(scorer instanceof ReqExclBulkScorer);
+
+    query = new BooleanQuery.Builder()
+        .add(new TermQuery(new Term("foo", "baz")), Occur.FILTER)
+        .add(new TermQuery(new Term("foo", "bar")), Occur.MUST_NOT)
+        .build();
+    weight = searcher.createNormalizedWeight(query, true);
+    scorer = ((BooleanWeight) weight).booleanScorer(ctx);
+    assertTrue(scorer instanceof ReqExclBulkScorer);
+
+    w.close();
+    reader.close();
+    dir.close();
+  }
+
   public void testSparseClauseOptimization() throws IOException {
     // When some windows have only one scorer that can match, the scorer will
     // directly call the collector in this window

lucene/core/src/test/org/apache/lucene/search/TestMinShouldMatch2.java

@@ -132,7 +132,7 @@ public class TestMinShouldMatch2 extends LuceneTestCase {
     case SCORER:
       return weight.scorer(reader.getContext());
     case BULK_SCORER:
-      final BulkScorer bulkScorer = weight.booleanScorer(reader.getContext());
+      final BulkScorer bulkScorer = weight.optionalBulkScorer(reader.getContext());
       if (bulkScorer == null) {
         if (weight.scorer(reader.getContext()) != null) {
           throw new AssertionError("BooleanScorer should be applicable for this query");

lucene/core/src/test/org/apache/lucene/search/TestReqExclBulkScorer.java

@@ -0,0 +1,116 @@
+package org.apache.lucene.search;
+
+/*
+ * 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.util.Bits;
+import org.apache.lucene.util.DocIdSetBuilder;
+import org.apache.lucene.util.FixedBitSet;
+import org.apache.lucene.util.LuceneTestCase;
+import org.apache.lucene.util.TestUtil;
+
+public class TestReqExclBulkScorer extends LuceneTestCase {
+
+  public void testRandom() throws IOException {
+    final int iters = atLeast(10);
+    for (int iter = 0; iter < iters; ++iter) {
+      doTestRandom();
+    }
+  }
+
+  public void doTestRandom() throws IOException {
+    final int maxDoc = TestUtil.nextInt(random(), 1, 1000);
+    DocIdSetBuilder reqBuilder = new DocIdSetBuilder(maxDoc);
+    DocIdSetBuilder exclBuilder = new DocIdSetBuilder(maxDoc);
+    final int numIncludedDocs = TestUtil.nextInt(random(), 1, maxDoc);
+    final int numExcludedDocs = TestUtil.nextInt(random(), 1, maxDoc);
+    for (int i = 0; i < numIncludedDocs; ++i) {
+      reqBuilder.add(random().nextInt(maxDoc));
+    }
+    for (int i = 0; i < numExcludedDocs; ++i) {
+      exclBuilder.add(random().nextInt(maxDoc));
+    }
+
+    final DocIdSet req = reqBuilder.build();
+    final DocIdSet excl = exclBuilder.build();
+
+    final BulkScorer reqBulkScorer = new BulkScorer() {
+      final DocIdSetIterator iterator = req.iterator();
+
+      @Override
+      public int score(LeafCollector collector, Bits acceptDocs, int min, int max) throws IOException {
+        int doc = iterator.docID();
+        if (iterator.docID() < min) {
+          doc = iterator.advance(min);
+        }
+        while (doc < max) {
+          if (acceptDocs == null || acceptDocs.get(doc)) {
+            collector.collect(doc);
+          }
+          doc = iterator.nextDoc();
+        }
+        return doc;
+      }
+
+      @Override
+      public long cost() {
+        return iterator.cost();
+      }
+    };
+
+    ReqExclBulkScorer reqExcl = new ReqExclBulkScorer(reqBulkScorer, excl.iterator());
+    final FixedBitSet actualMatches = new FixedBitSet(maxDoc);
+    if (random().nextBoolean()) {
+      reqExcl.score(new LeafCollector() {
+        @Override
+        public void setScorer(Scorer scorer) throws IOException {}
+        
+        @Override
+        public void collect(int doc) throws IOException {
+          actualMatches.set(doc);
+        }
+      }, null);
+    } else {
+      int next = 0;
+      while (next < maxDoc) {
+        final int min = next;
+        final int max = min + random().nextInt(10);
+        next = reqExcl.score(new LeafCollector() {
+          @Override
+          public void setScorer(Scorer scorer) throws IOException {}
+          
+          @Override
+          public void collect(int doc) throws IOException {
+            actualMatches.set(doc);
+          }
+        }, null, min, max);
+        assertTrue(next >= max);
+      }
+    }
+
+    final FixedBitSet expectedMatches = new FixedBitSet(maxDoc);
+    expectedMatches.or(req.iterator());
+    FixedBitSet excludedSet = new FixedBitSet(maxDoc);
+    excludedSet.or(excl.iterator());
+    expectedMatches.andNot(excludedSet);
+
+    assertArrayEquals(expectedMatches.getBits(), actualMatches.getBits());
+  }
+
+}