[LUCENE-1374] Merging of compressed string Fields may hit NPE

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.4
• Fix Version/s: 2.4
• Component/s: core/index
• Lucene Fields: New
• Assignee: Michael McCandless
• Reporter: Michael McCandless
• Created: 03/Sep/08 10:14
• Updated: 28/Aug/22 11:52
• Resolved: 03/Sep/08 14:00

Description

This bug was introduced with LUCENE-1219 (only present on 2.4).

The bug happens when merging compressed string fields, but only if bulk-merging code does not apply because the FieldInfos for the segment being merged are not congruent. This test shows the bug:

  public void testMergeCompressedFields() throws IOException {
    File indexDir = new File(System.getProperty("tempDir"), "mergecompressedfields");
    Directory dir = FSDirectory.getDirectory(indexDir);
    try {
      for(int i=0;i<5;i++) {
        // Must make a new writer & doc each time, w/
        // different fields, so bulk merge of stored fields
        // cannot run:
        IndexWriter w = new IndexWriter(dir, new WhitespaceAnalyzer(), i==0, IndexWriter.MaxFieldLength.UNLIMITED);
        w.setMergeFactor(5);
        w.setMergeScheduler(new SerialMergeScheduler());
        Document doc = new Document();
        doc.add(new Field("test1", "this is some data that will be compressed this this this", Field.Store.COMPRESS, Field.Index.NO));
        doc.add(new Field("test2", new byte[20], Field.Store.COMPRESS));
        doc.add(new Field("field" + i, "random field", Field.Store.NO, Field.Index.TOKENIZED));
        w.addDocument(doc);
        w.close();
      }

      byte[] cmp = new byte[20];

      IndexReader r = IndexReader.open(dir);
      for(int i=0;i<5;i++) {
        Document doc = r.document(i);
        assertEquals("this is some data that will be compressed this this this", doc.getField("test1").stringValue());
        byte[] b = doc.getField("test2").binaryValue();
        assertTrue(Arrays.equals(b, cmp));
      }
    } finally {
      dir.close();
      _TestUtil.rmDir(indexDir);
    }
  }

It's because in FieldsReader, when we load a field "for merge" we create a FieldForMerge instance which subsequently does not return the right values for getBinary

{Value,Length,Offset}

.