Comment 17 for bug 390563

I believe this is the problem. Specifically, because it is considering 'whole nodes' not 'interesting items', we see all the items in new pages, not just the items that are new in those pages.

        chk_bytes = self.from_repository.chk_bytes
        def _filter_id_to_entry():
            for record, items in chk_map.iter_interesting_nodes(chk_bytes,
                        self._chk_id_roots, uninteresting_root_keys):
                for name, bytes in items:
                    # Note: we don't care about name_utf8, because we are always
                    # rich-root = True
                    _, file_id, revision_id = bytes_to_info(bytes)
                    self._text_keys.add((file_id, revision_id))
                if record is not None:
                    yield record

To me this reinforces the need to push this diff code into chk_map.py, and have a single implementation of it.