Critical Description
Recent versions of Lucene introduced IndexReader.document(int, FieldSelector) method. When using IndexWriter.addIndexes(IndexReader[]) Lucene now uses that method from IndexReader instead of the old one IndexReader.document(int).
Unfortunately, this new method is not overriden in IndexSorter, which leads to a subtle corruption of sorted indexes - while the indexed fields are sorted properly, the values from stored fields are not sorted and remain in the sorted index in the original order. This means that in a sorted index the values of indexed fields and stored fields are completely out of sync, which later results in incorrect documents being retrieved from segments.