msg2046 suggests to change the find() function in roundup/backends/indexer_rdbms.py to decode the words before passing them to the db (implying the wordlist initially contains bytes).  

As in my previous message, in Python 3 the word list should already be a list of unicode strings, but on Python 2 it might be either a list of bytes, or a list of unicode strings.  Decoding the bytes with b2s in Python 2 should be ok, as long as the db can accept unicode.  Tests should be added to ensure that find() works with both bytes and unicode in Python 2 and that it handles non-ascii words correctly on both versions.