Thanks, looking forward. If you semantically see data structures as facts derived by a rule, then you get interesting parallels between PL GC and IVM (reachability with refcounts, putback evaluation, or differential dataflow). Souffle always accumulates (non-starter in a persistent db system).

There are three problems to address in Datalog systems with structured data:
- Efficiency (not too many bits, ideally no random insert/access)
- Parallel alloc (multiple threads and distributed)
- Maintenance (essentially GC)
I haven't seen any solution to all. I'm very curious what you do!

Interesting work, thanks. How do you compute the internal unique id that is used? (sorry, I poked around a little in the source code, but couldn't find the relevant code - if you have a pointer that'd be cool)