LUCENE-10159: Fix invalid access in sorted set dv (#389)

We introduced invalid accesses for sorted set doc values in LUCENE-9613. 
However, the issue has been unnoticed because the ordinals in doc values
tests aren't complex enough to use high packed bits, and the 3 padding
bytes make these invalid accesses perfectly fine. To reproduce this
issue, we need to use at least 20 bits per value for the ordinals.

lucene/core/src/java/org/apache/lucene/codecs/lucene90/Lucene90DocValuesProducer.java

@@ -1374,9 +1374,15 @@ final class Lucene90DocValuesProducer extends DocValuesProducer {
 
       int i = 0;
       int count = 0;
+      boolean set = false;
 
       @Override
       public long nextOrd() throws IOException {
+        if (set == false) {
+          set = true;
+          i = 0;
+          count = ords.docValueCount();
+        }
         if (i++ == count) {
           return NO_MORE_ORDS;
         }
@@ -1385,13 +1391,8 @@ final class Lucene90DocValuesProducer extends DocValuesProducer {
 
       @Override
       public boolean advanceExact(int target) throws IOException {
-        if (ords.advanceExact(target)) {
+        set = false;
-          count = ords.docValueCount();
+        return ords.advanceExact(target);
-          i = 0;
-          return true;
-        } else {
-          return false;
-        }
       }
 
       @Override
@@ -1401,18 +1402,14 @@ final class Lucene90DocValuesProducer extends DocValuesProducer {
 
       @Override
       public int nextDoc() throws IOException {
-        int doc = ords.nextDoc();
+        set = false;
-        count = ords.docValueCount();
+        return ords.nextDoc();
-        i = 0;
-        return doc;
       }
 
       @Override
       public int advance(int target) throws IOException {
-        int doc = ords.advance(target);
+        set = false;
-        count = ords.docValueCount();
+        return ords.advance(target);
-        i = 0;
-        return doc;
       }
 
       @Override

lucene/test-framework/src/java/org/apache/lucene/index/BaseDocValuesFormatTestCase.java

@@ -3499,6 +3499,39 @@ public abstract class BaseDocValuesFormatTestCase extends BaseIndexFileFormatTes
         });
   }
 
+  /**
+   * Tests where a DVField uses a high number of packed bits to store its ords. See:
+   * https://issues.apache.org/jira/browse/LUCENE-10159
+   */
+  @Nightly
+  public void testHighOrdsSortedSetDV() throws Exception {
+    Directory dir = newDirectory();
+    IndexWriterConfig iwc = new IndexWriterConfig();
+    iwc.setRAMBufferSizeMB(8 + random().nextInt(64));
+    IndexWriter writer = new IndexWriter(dir, iwc);
+    // many docs with some of them have very high ords
+    int numDocs = 20_000 + random().nextInt(10_000);
+    for (int i = 1; i < numDocs; i++) {
+      final int numOrds;
+      if (random().nextInt(100) <= 5) {
+        numOrds = 1000 + random().nextInt(500);
+      } else {
+        numOrds = random().nextInt(10);
+      }
+      Document doc = new Document();
+      for (int ord = 0; ord < numOrds; ord++) {
+        doc.add(
+            new SortedSetDocValuesField("sorted_set_dv", TestUtil.randomBinaryTerm(random(), 2)));
+      }
+      writer.addDocument(doc);
+    }
+    writer.forceMerge(1, true);
+    try (DirectoryReader reader = DirectoryReader.open(writer)) {
+      TestUtil.checkReader(reader);
+    }
+    IOUtils.close(writer, dir);
+  }
+
   private interface FieldCreator {
     public Field next();
 

lucene/test-framework/src/java/org/apache/lucene/util/TestUtil.java

@@ -1175,7 +1175,11 @@ public final class TestUtil {
 
   /** Returns a random binary term. */
   public static BytesRef randomBinaryTerm(Random r) {
-    int length = r.nextInt(15);
+    return randomBinaryTerm(r, r.nextInt(15));
+  }
+
+  /** Returns a random binary with a given length */
+  public static BytesRef randomBinaryTerm(Random r, int length) {
     BytesRef b = new BytesRef(length);
     r.nextBytes(b.bytes);
     b.length = length;