As his original question suggested, and Re: Re: Re: Re: Re: Re: many to many join on text files clarified, the actual dataset has a many to many joins, and where each side is many, he wants every combination to be represented. Which can be done with walking in parallel, but it requires some backtracking logic that can be tricky to get right.

The BTREE solution that I gave is very similar to pre-sorting both and walking in parallel. In particular, a BTREE is an ordered structure which is not completely filled, but is close to it. The details are all handled by DB_File at the C level, and should be reasonably efficient.