You can clone with
HTTPS or Subversion.
Currently get_all_rows() API returns values from get_all_column_ids() that is implemented in inverted_index_storage.
inverted_index_storage::get_all_column_ids uses column2norm_, which is not updated on clear_row.
Isn't it better to use row IDs that comes from orig_->get_all_row_ids() for inverted_index?
From discussion in meeting on 2014-02-17:
This issue affects not only to inverted_index, but also to bit_index_storage, which is used by minhash and lsh.
I have a fixed version in df8254f (fix-659 branch). However this depends on pull-request #685.
Since I merged #685, please send a pull-request with fix-659 branch so that I can also review it.
free memory for erased rows in inverted_index (backport #659)