[zozbot234]

> To start off: Datalog is distinguished from traditional SQL in its focus on heavily-recursive reachability-based reasoning.

This was historically true but SQL has CTE's and recursive CTE's these days, and even some extra syntactic sugar for reachability query's. And of course (given the former) "inference" and "deduction" over data are just a CREATE VIEW statement away. This is the "you know all of the rules up-front" part; a VIEW is just a query that you do know upfront. Most uses of "deductive" databases are just for querying within some in-memory database, which is not really playing in the same league as a fully general RDBMS. This is not intended to dismiss the work in OP, of course; if anything, its applicability need not be restricted to a tiny niche of software intended for specific application areas, and can be quite a bit broader than that.

[kmicinski]

Right--but CTEs are orders-of-magnitude slower than Datalog, to the point that they are not seriously worth considering for any modern application where Datalog would be used. As you said, Datalog is less expressive than SQL's ability to create new queries in an ad-hoc way: knowing all queries within the fixed point enables much more efficient compilation.