While looking at TR-107, I found a potential bug on DirectIndexInputStream (actually, observed on a parent method: BitPostingIndexInputStream.print()).

Here is a sample code to trigger the problem:

{code:java}
Document[] sourceDocs = new Document[]{
new FileDocument("doc1", new ByteArrayInputStream("cats dogs horses".getBytes()), new EnglishTokeniser()),
new FileDocument("doc2", new ByteArrayInputStream("chicken cats chicken chicken".getBytes()), new EnglishTokeniser())
};

Collection col = new CollectionDocumentList(sourceDocs, "filename");
Indexer indexer = new BasicIndexer(ApplicationSetup.TERRIER_INDEX_PATH, ApplicationSetup.TERRIER_INDEX_PREFIX);

indexer.createDirectIndex(new Collection[]{col});
indexer.createInvertedIndex();

Index index = Index.createIndex();
BitPostingIndexInputStream bpiis = null;

System.out.println("INVERTED ----------");
bpiis = (BitPostingIndexInputStream) index.getIndexStructureInputStream("inverted");
bpiis.print();

System.out.println("DIRECT ----------");
bpiis = (BitPostingIndexInputStream) index.getIndexStructureInputStream("direct");
bpiis.print();
{code}

And here is the corresponding output:

{quote}
INVERTED ----------
0 (0,1) (1,1) // cats -> doc1, doc2 -> OK
1 (1,3) // chicken -> doc2 -> OK
2 (0,1) // dogs -> doc1 -> OK
3 (0,1) // horses -> doc1 -> OK
DIRECT ----------
0 (0,1) (1,1) (2,1) // doc1 -> cats, chicken, dogs -> NOT OK
1 (2,1) (3,3) // doc2 -> dogs, horses -> NOT OK
{quote}